[PATCH v2] drm/radeon: Do not use deprecated drm log API

[Nirmoy Das]

Replace deprecated DRM_DEBUG_KMS_RATELIMITED() and DRM_ERROR()
with proper APIs.

v2: replace pr_err with dev_err(Alex).

Cc: Alex Deucher 
Cc: Christian König 

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/radeon/radeon_dp_auxch.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_dp_auxch.c 
b/drivers/gpu/drm/radeon/radeon_dp_auxch.c
index 69379b95146e..1e5b6baf76a1 100644
--- a/drivers/gpu/drm/radeon/radeon_dp_auxch.c
+++ b/drivers/gpu/drm/radeon/radeon_dp_auxch.c
@@ -158,7 +158,7 @@ radeon_dp_aux_transfer_native(struct drm_dp_aux *aux, 
struct drm_dp_aux_msg *msg
} while (retry_count++ < 1000);
 
if (retry_count >= 1000) {
-   DRM_ERROR("auxch hw never signalled completion, error %08x\n", 
tmp);
+   dev_err(rdev->dev, "auxch hw never signalled completion, error 
%08x\n", tmp);
ret = -EIO;
goto done;
}
@@ -168,8 +168,7 @@ radeon_dp_aux_transfer_native(struct drm_dp_aux *aux, 
struct drm_dp_aux_msg *msg
goto done;
}
if (tmp & AUX_RX_ERROR_FLAGS) {
-   DRM_DEBUG_KMS_RATELIMITED("dp_aux_ch flags not zero: %08x\n",
- tmp);
+   drm_dbg_kms_ratelimited(dev, "dp_aux_ch flags not zero: 
%08x\n", tmp);
ret = -EIO;
goto done;
}
-- 
2.39.0

[PATCH 2/2] drm_print: Remove deprecated DRM_DEBUG_KMS_RATELIMITED()

[Nirmoy Das]

There are no current users of DRM_DEBUG_KMS_RATELIMITED()
so remove it.

Cc: Maarten Lankhorst 
Cc: Maxime Ripard 
Cc: Thomas Zimmermann 
Cc: David Airlie 
Cc: Daniel Vetter 
Cc: Sam Ravnborg 

Signed-off-by: Nirmoy Das 
Reviewed-by: Sam Ravnborg 
---
 include/drm/drm_print.h | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h
index a44fb7ef257f..c3753da97c4e 100644
--- a/include/drm/drm_print.h
+++ b/include/drm/drm_print.h
@@ -605,9 +605,6 @@ void __drm_err(const char *format, ...);
 #define drm_dbg_kms_ratelimited(drm, fmt, ...) \
__DRM_DEFINE_DBG_RATELIMITED(KMS, drm, fmt, ## __VA_ARGS__)
 
-/* NOTE: this is deprecated in favor of drm_dbg_kms_ratelimited(NULL, ...). */
-#define DRM_DEBUG_KMS_RATELIMITED(fmt, ...) drm_dbg_kms_ratelimited(NULL, fmt, 
## __VA_ARGS__)
-
 /*
  * struct drm_device based WARNs
  *
-- 
2.39.0

[PATCH 1/2] drm/radeon: Do not use deprecated drm log API

[Nirmoy Das]

Replace deprecated DRM_DEBUG_KMS_RATELIMITED() and DRM_ERROR()
with proper APIs.

Cc: Alex Deucher 
Cc: Christian König 

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/radeon/radeon_dp_auxch.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_dp_auxch.c 
b/drivers/gpu/drm/radeon/radeon_dp_auxch.c
index 69379b95146e..76ce66efb5f8 100644
--- a/drivers/gpu/drm/radeon/radeon_dp_auxch.c
+++ b/drivers/gpu/drm/radeon/radeon_dp_auxch.c
@@ -158,7 +158,7 @@ radeon_dp_aux_transfer_native(struct drm_dp_aux *aux, 
struct drm_dp_aux_msg *msg
} while (retry_count++ < 1000);
 
if (retry_count >= 1000) {
-   DRM_ERROR("auxch hw never signalled completion, error %08x\n", 
tmp);
+   pr_err("auxch hw never signalled completion, error %08x\n", 
tmp);
ret = -EIO;
goto done;
}
@@ -168,8 +168,7 @@ radeon_dp_aux_transfer_native(struct drm_dp_aux *aux, 
struct drm_dp_aux_msg *msg
goto done;
}
if (tmp & AUX_RX_ERROR_FLAGS) {
-   DRM_DEBUG_KMS_RATELIMITED("dp_aux_ch flags not zero: %08x\n",
- tmp);
+   drm_dbg_kms_ratelimited(dev, "dp_aux_ch flags not zero: 
%08x\n", tmp);
ret = -EIO;
goto done;
}
-- 
2.39.0

[PATCH 2/2] drm_print: Remove deprecated DRM_DEBUG_KMS_RATELIMITED()

[Nirmoy Das]

There are no current users of DRM_DEBUG_KMS_RATELIMITED()
so remove it.

Cc: Maarten Lankhorst 
Cc: Maxime Ripard 
Cc: Thomas Zimmermann 
Cc: David Airlie 
Cc: Daniel Vetter 
Cc: Sam Ravnborg 

Signed-off-by: Nirmoy Das 
---
 include/drm/drm_print.h | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h
index a44fb7ef257f..c3753da97c4e 100644
--- a/include/drm/drm_print.h
+++ b/include/drm/drm_print.h
@@ -605,9 +605,6 @@ void __drm_err(const char *format, ...);
 #define drm_dbg_kms_ratelimited(drm, fmt, ...) \
__DRM_DEFINE_DBG_RATELIMITED(KMS, drm, fmt, ## __VA_ARGS__)
 
-/* NOTE: this is deprecated in favor of drm_dbg_kms_ratelimited(NULL, ...). */
-#define DRM_DEBUG_KMS_RATELIMITED(fmt, ...) drm_dbg_kms_ratelimited(NULL, fmt, 
## __VA_ARGS__)
-
 /*
  * struct drm_device based WARNs
  *
-- 
2.39.0

[PATCH 1/1] drm/amdgpu: move bo_va ref counting to internal funcs

[Nirmoy Das]

GEM code should not deal with struct amdgpu_bo_va's ref_count.
Move ref counting to amdgpu_vm.c.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 11 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 38 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  |  2 ++
 3 files changed, 42 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 4a11a2f4fa73..691f0a879c90 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -176,12 +176,9 @@ static int amdgpu_gem_object_open(struct drm_gem_object 
*obj,
if (r)
return r;
 
-   bo_va = amdgpu_vm_bo_find(vm, abo);
-   if (!bo_va) {
-   bo_va = amdgpu_vm_bo_add(adev, vm, abo);
-   } else {
-   ++bo_va->ref_count;
-   }
+   if (!amdgpu_vm_bo_get(vm, abo))
+   amdgpu_vm_bo_add(adev, vm, abo);
+
amdgpu_bo_unreserve(abo);
return 0;
 }
@@ -218,7 +215,7 @@ static void amdgpu_gem_object_close(struct drm_gem_object 
*obj,
return;
}
bo_va = amdgpu_vm_bo_find(vm, bo);
-   if (!bo_va || --bo_va->ref_count)
+   if (!bo_va)
goto out_unlock;
 
amdgpu_vm_bo_rmv(adev, bo_va);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index b23cb463b106..9d60de6a6697 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -1290,16 +1290,49 @@ struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm 
*vm,
   struct amdgpu_bo *bo)
 {
struct amdgpu_vm_bo_base *base;
+   struct amdgpu_bo_va *bo_va = NULL;
 
for (base = bo->vm_bo; base; base = base->next) {
if (base->vm != vm)
continue;
 
-   return container_of(base, struct amdgpu_bo_va, base);
+   bo_va = container_of(base, struct amdgpu_bo_va, base);
}
-   return NULL;
+
+   if (bo_va && bo_va->ref_count <= 0)
+   return NULL;
+
+   return bo_va;
 }
 
+/**
+ * amdgpu_vm_bo_get - find the bo_va for a specific vm & bo and increase
+ * the ref_count
+ *
+ * @vm: requested vm
+ * @bo: requested buffer object
+ *
+ * Find @bo inside the requested vm.
+ * Search inside the @bos vm list for the requested vm
+ * Returns the found bo_va with +1 ref_count or NULL if none is found
+ *
+ * Object has to be reserved!
+ *
+ * Returns:
+ * Found bo_va or NULL.
+ */
+struct amdgpu_bo_va *amdgpu_vm_bo_get(struct amdgpu_vm *vm,
+ struct amdgpu_bo *bo)
+{
+   struct amdgpu_bo_va *bo_va = amdgpu_vm_bo_find(vm, bo);
+
+   if (bo_va)
+   ++bo_va->ref_count;
+
+   return bo_va;
+}
+
+
 /**
  * amdgpu_vm_map_gart - Resolve gart mapping of addr
  *
@@ -2704,6 +2737,7 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
if (bo && bo_va->is_xgmi)
amdgpu_xgmi_set_pstate(adev, AMDGPU_XGMI_PSTATE_MIN);
 
+   --bo_va->ref_count;
kfree(bo_va);
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 85fcfb8c5efd..6d936fb1b934 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -415,6 +415,8 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
 uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr);
 struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm,
   struct amdgpu_bo *bo);
+struct amdgpu_bo_va *amdgpu_vm_bo_get(struct amdgpu_vm *vm,
+  struct amdgpu_bo *bo);
 struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
  struct amdgpu_vm *vm,
  struct amdgpu_bo *bo);
-- 
2.33.1

[PATCH REBASED 4/4] drm/amdgpu: recover gart table at resume

[Nirmoy Das]

Get rid off pin/unpin of gart BO at resume/suspend and
instead pin only once and try to recover gart content
at resume time. This is much more stable in case there
is OOM situation at 2nd call to amdgpu_device_evict_resources()
while evicting GART table.

v3: remove gart recovery from other places
v2: pin gart at amdgpu_gart_table_vram_alloc()

Reviewed-by: Christian König 
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 11 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c   | 84 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  |  3 +-
 7 files changed, 11 insertions(+), 99 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 58b9a5176082..a89e08e44fea 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4004,16 +4004,11 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
if (!adev->in_s0ix)
amdgpu_amdkfd_suspend(adev, adev->in_runpm);
 
-   /* First evict vram memory */
amdgpu_device_evict_resources(adev);
 
amdgpu_fence_driver_hw_fini(adev);
 
amdgpu_device_ip_suspend_phase2(adev);
-   /* This second call to evict device resources is to evict
-* the gart page table using the CPU.
-*/
-   amdgpu_device_evict_resources(adev);
 
return 0;
 }
@@ -4356,8 +4351,6 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,
goto error;
 
amdgpu_virt_init_data_exchange(adev);
-   /* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);
 
r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4677,10 +4670,6 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}
 
-   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
-   if (r)
-   goto out;
-
r = amdgpu_device_fw_loading(tmp_adev);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index d3e4203f6217..645950a653a0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -114,80 +114,12 @@ void amdgpu_gart_dummy_page_fini(struct amdgpu_device 
*adev)
  */
 int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
 {
-   int r;
-
-   if (adev->gart.bo == NULL) {
-   struct amdgpu_bo_param bp;
-
-   memset(, 0, sizeof(bp));
-   bp.size = adev->gart.table_size;
-   bp.byte_align = PAGE_SIZE;
-   bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-   AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = NULL;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >gart.bo);
-   if (r) {
-   return r;
-   }
-   }
-   return 0;
-}
-
-/**
- * amdgpu_gart_table_vram_pin - pin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Pin the GART page table in vram so it will not be moved
- * by the memory manager (pcie r4xx, r5xx+).  These asics require the
- * gart table to be in video memory.
- * Returns 0 for success, error for failure.
- */
-int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev)
-{
-   int r;
-
-   r = amdgpu_bo_reserve(adev->gart.bo, false);
-   if (unlikely(r != 0))
-   return r;
-   r = amdgpu_bo_pin(adev->gart.bo, AMDGPU_GEM_DOMAIN_VRAM);
-   if (r) {
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-   }
-   r = amdgpu_bo_kmap(adev->gart.bo, >gart.ptr);
-   if (r)
-   amdgpu_bo_unpin(adev->gart.bo);
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-}
-
-/**
- * amdgpu_gart_table_vram_unpin - unpin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Unpin the GART page table in vram (pcie r4xx, r5xx+).
- * These asics require the gart table to be in video memory.
- */
-void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev)
-{
-   int r;
+   if (adev->gart.bo != NULL)
+   return 0;
 
-   if (adev->gart.bo == NULL) {
-   return;
-   }
-   r = amdgpu_bo_re

[PATCH 2/4] drm/amdkfd: remove unused function

[Nirmoy Das]

Remove unused amdgpu_amdkfd_get_vram_usage()

CC: felix.kuehl...@amd.com

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 7 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 1 -
 2 files changed, 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 776a947b45df..6ca1db3c243f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -514,13 +514,6 @@ int amdgpu_amdkfd_get_dmabuf_info(struct amdgpu_device 
*adev, int dma_buf_fd,
return r;
 }

-uint64_t amdgpu_amdkfd_get_vram_usage(struct amdgpu_device *adev)
-{
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
-
-   return amdgpu_vram_mgr_usage(vram_man);
-}
-
 uint8_t amdgpu_amdkfd_get_xgmi_hops_count(struct amdgpu_device *dst,
  struct amdgpu_device *src)
 {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 61f899e54fd5..ac841ae8f5cc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -223,7 +223,6 @@ int amdgpu_amdkfd_get_dmabuf_info(struct amdgpu_device 
*adev, int dma_buf_fd,
  uint64_t *bo_size, void *metadata_buffer,
  size_t buffer_size, uint32_t *metadata_size,
  uint32_t *flags);
-uint64_t amdgpu_amdkfd_get_vram_usage(struct amdgpu_device *adev);
 uint8_t amdgpu_amdkfd_get_xgmi_hops_count(struct amdgpu_device *dst,
  struct amdgpu_device *src);
 int amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(struct amdgpu_device *dst,
--
2.33.1

[PATCH REBASED 3/4] drm/amdgpu: do not pass ttm_resource_manager to vram_mgr

[Nirmoy Das]

Do not allow exported amdgpu_vram_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call vram_mgr functions.

v2: pass adev's vram_mgr instead of adev

Reviewed-by: Christian König 
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c   |  5 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c  | 10 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c  |  6 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h  |  8 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c |  5 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 40 
 6 files changed, 30 insertions(+), 44 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 76fe5b71e35d..7e745164a624 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -298,7 +298,6 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
 {
s64 time_us, increment_us;
u64 free_vram, total_vram, used_vram;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
/* Allow a maximum of 200 accumulated ms. This is basically per-IB
 * throttling.
 *
@@ -315,7 +314,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
}
 
total_vram = adev->gmc.real_vram_size - 
atomic64_read(>vram_pin_size);
-   used_vram = amdgpu_vram_mgr_usage(vram_man);
+   used_vram = amdgpu_vram_mgr_usage(>mman.vram_mgr);
free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;
 
spin_lock(>mm_stats.lock);
@@ -362,7 +361,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
if (!amdgpu_gmc_vram_full_visible(>gmc)) {
u64 total_vis_vram = adev->gmc.visible_vram_size;
u64 used_vis_vram =
- amdgpu_vram_mgr_vis_usage(vram_man);
+ amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);
 
if (used_vis_vram < total_vis_vram) {
u64 free_vis_vram = total_vis_vram - used_vis_vram;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 763de822afa1..289521aafb79 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -672,10 +672,10 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = atomic64_read(>num_vram_cpu_page_faults);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VRAM_USAGE:
-   ui64 = amdgpu_vram_mgr_usage(ttm_manager_type(>mman.bdev, 
TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VIS_VRAM_USAGE:
-   ui64 = 
amdgpu_vram_mgr_vis_usage(ttm_manager_type(>mman.bdev, TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
ui64 = amdgpu_gtt_mgr_usage(>mman.gtt_mgr);
@@ -709,8 +709,6 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
}
case AMDGPU_INFO_MEMORY: {
struct drm_amdgpu_memory_info mem;
-   struct ttm_resource_manager *vram_man =
-   ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
struct ttm_resource_manager *gtt_man =
ttm_manager_type(>mman.bdev, TTM_PL_TT);
memset(, 0, sizeof(mem));
@@ -719,7 +717,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
atomic64_read(>vram_pin_size) -
AMDGPU_VM_RESERVED_VRAM;
mem.vram.heap_usage =
-   amdgpu_vram_mgr_usage(vram_man);
+   amdgpu_vram_mgr_usage(>mman.vram_mgr);
mem.vram.max_allocation = mem.vram.usable_heap_size * 3 / 4;
 
mem.cpu_accessible_vram.total_heap_size =
@@ -729,7 +727,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
atomic64_read(>visible_pin_size),
mem.vram.usable_heap_size);
mem.cpu_accessible_vram.heap_usage =
-   amdgpu_vram_mgr_vis_usage(vram_man);
+   amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);
mem.cpu_accessible_vram.max_allocation =
mem.cpu_accessible_vram.usable_heap_size * 3 / 4;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c

[PATCH REBASED 1/4] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr

[Nirmoy Das]

Do not allow exported amdgpu_gtt_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call gtt_mgr functions.

v4: remove unused adev.
v3: upcast mgr from ttm resopurce manager instead of
getting it from adev.
v2: pass adev's gtt_mgr instead of adev.

Reviewed-by: Christian König 
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 22 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h |  4 ++--
 4 files changed, 16 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 3aab187520c6..58b9a5176082 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4357,7 +4357,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,
 
amdgpu_virt_init_data_exchange(adev);
/* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev, TTM_PL_TT));
+   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);
 
r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4677,7 +4677,7 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}
 
-   r = 
amdgpu_gtt_mgr_recover(ttm_manager_type(_adev->mman.bdev, TTM_PL_TT));
+   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
if (r)
goto out;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index c18f16b3be9c..9151950e0cc3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -77,10 +77,8 @@ static ssize_t amdgpu_mem_info_gtt_used_show(struct device 
*dev,
 {
struct drm_device *ddev = dev_get_drvdata(dev);
struct amdgpu_device *adev = drm_to_adev(ddev);
-   struct ttm_resource_manager *man;
 
-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(man));
+   return sysfs_emit(buf, "%llu\n", 
amdgpu_gtt_mgr_usage(>mman.gtt_mgr));
 }
 
 static DEVICE_ATTR(mem_info_gtt_total, S_IRUGO,
@@ -206,14 +204,15 @@ static void amdgpu_gtt_mgr_del(struct 
ttm_resource_manager *man,
 /**
  * amdgpu_gtt_mgr_usage - return usage of GTT domain
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Return how many bytes are used in the GTT domain
  */
-uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager *man)
+uint64_t amdgpu_gtt_mgr_usage(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   s64 result = man->size - atomic64_read(>available);
+   s64 result;
+
+   result = mgr->manager.size - atomic64_read(>available);
 
return (result > 0 ? result : 0) * PAGE_SIZE;
 }
@@ -221,16 +220,15 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
 /**
  * amdgpu_gtt_mgr_recover - re-init gart
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Re-init the gart for each known BO in the GTT.
  */
-int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
+int amdgpu_gtt_mgr_recover(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   struct amdgpu_device *adev;
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
+   struct amdgpu_device *adev;
int r = 0;
 
adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
@@ -267,7 +265,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,
 
drm_printf(printer, "man size:%llu pages, gtt available:%lld pages, 
usage:%lluMB\n",
   man->size, (u64)atomic64_read(>available),
-  amdgpu_gtt_mgr_usage(man) >> 20);
+  amdgpu_gtt_mgr_usage(mgr) >> 20);
 }
 
 static const struct ttm_resource_manager_func amdgpu_gtt_mgr_func = {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 651c7abfde03..763de822afa1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -678,7 +678,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = 
amdgpu_vram_mgr_vis_usage(ttm_manager_type(>mman.bdev, TTM_PL_VRAM));
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
-   ui64 = amdgpu_gtt_mgr_usage(ttm_manager_type

[RFC PATCH 2/3] drm/amdgpu: add HIQ eng_sel to KIQ packets

[Nirmoy Das]

Allow KIQ to map/unmap HIQ MQD as well.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 14 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  2 ++
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   |  4 ++--
 4 files changed, 20 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 5b8cb76e35a0..053a1119ebfe 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -1010,3 +1010,17 @@ void amdgpu_gfx_state_change_set(struct amdgpu_device 
*adev, enum gfx_change_sta
(adev)->powerplay.pp_handle, state));
mutex_unlock(>pm.mutex);
 }
+
+int amdgpu_kiq_get_eng_num(struct amdgpu_ring *ring)
+{
+
+   switch (ring->funcs->type) {
+   case AMDGPU_RING_TYPE_GFX:
+   return 4;
+   case AMDGPU_RING_TYPE_HIQ:
+   return 1;
+   default:
+   return 0;
+   }
+
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index 4d9c91f4400d..88d942b1ef08 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -373,6 +373,8 @@ static inline u32 amdgpu_gfx_create_bitmask(u32 bit_width)
return (u32)((1ULL << bit_width) - 1);
 }
 
+int amdgpu_kiq_get_eng_num(struct amdgpu_ring *ring);
+
 int amdgpu_gfx_scratch_get(struct amdgpu_device *adev, uint32_t *reg);
 void amdgpu_gfx_scratch_free(struct amdgpu_device *adev, uint32_t reg);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c 
b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
index 90a834dc4008..538130c453a6 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
@@ -3633,7 +3633,7 @@ static void gfx10_kiq_unmap_queues(struct amdgpu_ring 
*kiq_ring,
   enum amdgpu_unmap_queues_action action,
   u64 gpu_addr, u64 seq)
 {
-   uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+   uint32_t eng_sel = amdgpu_kiq_get_eng_num(ring);
 
amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
@@ -3660,7 +3660,7 @@ static void gfx10_kiq_query_status(struct amdgpu_ring 
*kiq_ring,
   u64 addr,
   u64 seq)
 {
-   uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+   uint32_t eng_sel = amdgpu_kiq_get_eng_num(ring);
 
amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
amdgpu_ring_write(kiq_ring,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c 
b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 7f944bb11298..2b29e42bde62 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -847,7 +847,7 @@ static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring 
*kiq_ring,
struct amdgpu_device *adev = kiq_ring->adev;
uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
-   uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+   uint32_t eng_sel = amdgpu_kiq_get_eng_num(ring);
 
amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
@@ -877,7 +877,7 @@ static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring 
*kiq_ring,
   enum amdgpu_unmap_queues_action action,
   u64 gpu_addr, u64 seq)
 {
-   uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+   uint32_t eng_sel = amdgpu_kiq_get_eng_num(ring);
 
amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
-- 
2.31.1

[RFC PATCH 3/3] drm/amdgpu: enable HIQ in amdgpu without kfd

[Nirmoy Das]

There is a HW bug which prevents CP to read secure buffers
with HIQ being configured and mapped using KIQ. KFD already
does this for amdgpu but when kfd is not enabled amdgpu
should that for itself.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 14 -
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 77 
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 80 +
 3 files changed, 170 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 053a1119ebfe..837f76550242 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -519,7 +519,7 @@ int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
AMDGPU_GEM_DOMAIN_VRAM, 
>mqd_obj,
>mqd_gpu_addr, 
>mqd_ptr);
if (r) {
-   dev_warn(adev->dev, "failed to create ring mqd ob 
(%d)", r);
+   dev_warn(adev->dev, "failed to create KIQ ring mqd ob 
(%d)", r);
return r;
}
 
@@ -569,6 +569,18 @@ int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
}
}
 
+   /* create MQD for HIQ */
+   ring = >gfx.hiq.ring;
+   if (!ring->mqd_obj) {
+   r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE,
+   AMDGPU_GEM_DOMAIN_VRAM, 
>mqd_obj,
+   >mqd_gpu_addr, 
>mqd_ptr);
+   if (r) {
+   dev_warn(adev->dev, "failed to create HIQ ring mqd ob 
(%d)", r);
+   return r;
+   }
+   }
+
return 0;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c 
b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
index 538130c453a6..9532f013128f 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
@@ -4794,6 +4794,7 @@ static int gfx_v10_0_sw_init(void *handle)
 {
int i, j, k, r, ring_id = 0;
struct amdgpu_kiq *kiq;
+   struct amdgpu_hiq *hiq;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
switch (adev->ip_versions[GC_HWIP][0]) {
@@ -4923,6 +4924,18 @@ static int gfx_v10_0_sw_init(void *handle)
if (r)
return r;
 
+   if (!adev->kfd.dev) {
+   r = amdgpu_gfx_hiq_init(adev, GFX10_MEC_HPD_SIZE);
+   if (r) {
+   DRM_ERROR("Failed to init HIQ BOs!\n");
+   return r;
+   }
+
+   hiq = >gfx.hiq;
+   r = amdgpu_gfx_hiq_init_ring(adev, >ring, >irq);
+   if (r)
+   return r;
+   }
r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v10_compute_mqd));
if (r)
return r;
@@ -7215,6 +7228,54 @@ static int gfx_v10_0_kcq_resume(struct amdgpu_device 
*adev)
return r;
 }
 
+static int gfx_v10_0_hiq_init_queue(struct amdgpu_ring *ring)
+{
+   struct amdgpu_device *adev = ring->adev;
+   struct v10_compute_mqd *mqd = ring->mqd_ptr;
+
+
+   if (amdgpu_in_reset(adev)) {
+   /* reset ring buffer */
+   ring->wptr = 0;
+   amdgpu_ring_clear_ring(ring);
+
+   } else {
+   memset((void *)mqd, 0, sizeof(*mqd));
+   mutex_lock(>srbm_mutex);
+   nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
+   gfx_v10_0_compute_mqd_init(ring);
+   nv_grbm_select(adev, 0, 0, 0, 0);
+   mutex_unlock(>srbm_mutex);
+   }
+
+   return 0;
+}
+
+static int gfx_v10_0_hiq_resume(struct amdgpu_device *adev)
+{
+   struct amdgpu_ring *ring;
+   int r;
+
+   ring = >gfx.hiq.ring;
+
+   r = amdgpu_bo_reserve(ring->mqd_obj, false);
+   if (unlikely(r != 0))
+   return r;
+
+   r = amdgpu_bo_kmap(ring->mqd_obj, (void **)>mqd_ptr);
+   if (unlikely(r != 0))
+   return r;
+
+   gfx_v10_0_hiq_init_queue(ring);
+   amdgpu_bo_kunmap(ring->mqd_obj);
+   ring->mqd_ptr = NULL;
+   amdgpu_bo_unreserve(ring->mqd_obj);
+   ring->sched.ready = true;
+
+   amdgpu_gfx_enable_hiq(adev);
+   return 0;
+}
+
 static int gfx_v10_0_cp_resume(struct amdgpu_device *adev)
 {
int r, i;
@@ -7252,6 +7313,12 @@ static int gfx_v10_0_cp_resume(struct amdgpu_device 
*adev)
return r;
}
 
+   if (!adev->kfd.dev) {
+   r = gfx_v10_0_hiq_resume(adev);
+   if (r)
+   return r;
+   }
+
for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
ring

[RFC PATCH 1/3] drm/amdgpu: add HIQ ring to amdgpu

[Nirmoy Das]

Add HIQ ring structs and functions that will map HIQ
using KIQ.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h |   1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c  | 142 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h  |  24 
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h |   3 +-
 4 files changed, 169 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h
index 89e6ad30396f..2d9295adac06 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h
@@ -40,6 +40,7 @@ struct amdgpu_doorbell {
  */
 struct amdgpu_doorbell_index {
uint32_t kiq;
+   uint32_t hiq;
uint32_t mec_ring0;
uint32_t mec_ring1;
uint32_t mec_ring2;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 1916ec84dd71..5b8cb76e35a0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -256,6 +256,148 @@ void amdgpu_gfx_graphics_queue_acquire(struct 
amdgpu_device *adev)
bitmap_weight(adev->gfx.me.queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
 }
 
+int amdgpu_gfx_hiq_acquire(struct amdgpu_device *adev, struct amdgpu_ring 
*ring)
+{
+   int queue_bit;
+   int mec, pipe, queue;
+
+   queue_bit = adev->gfx.mec.num_mec
+   * adev->gfx.mec.num_pipe_per_mec
+   * adev->gfx.mec.num_queue_per_pipe;
+
+   while (queue_bit-- >= 0) {
+   if (test_bit(queue_bit, adev->gfx.mec.queue_bitmap))
+   continue;
+
+   amdgpu_queue_mask_bit_to_mec_queue(adev, queue_bit, , 
, );
+
+   if (mec == 1 && pipe > 1)
+   continue;
+
+   ring->me = mec + 1;
+   ring->pipe = pipe;
+   ring->queue = queue;
+
+   return 0;
+   }
+
+   dev_err(adev->dev, "Failed to find a queue for HIQ\n");
+   return -EINVAL;
+}
+
+int amdgpu_gfx_hiq_init_ring(struct amdgpu_device *adev,
+struct amdgpu_ring *ring,
+struct amdgpu_irq_src *irq)
+{
+   struct amdgpu_hiq *hiq = >gfx.hiq;
+   int r = 0;
+
+   ring->adev = NULL;
+   ring->ring_obj = NULL;
+   ring->use_doorbell = true;
+   ring->doorbell_index = adev->doorbell_index.hiq;
+
+   r = amdgpu_gfx_hiq_acquire(adev, ring);
+   if (r)
+   return r;
+
+   ring->eop_gpu_addr = hiq->eop_gpu_addr;
+   ring->no_scheduler = true;
+   sprintf(ring->name, "hiq_%d.%d.%d", ring->me, ring->pipe, ring->queue);
+   r = amdgpu_ring_init(adev, ring, 1024, irq, 
AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP,
+AMDGPU_RING_PRIO_DEFAULT, NULL);
+   if (r)
+   dev_warn(adev->dev, "(%d) failed to init hiq ring\n", r);
+
+   return r;
+}
+
+void amdgpu_gfx_hiq_free_ring(struct amdgpu_ring *ring)
+{
+   amdgpu_ring_fini(ring);
+}
+
+void amdgpu_gfx_hiq_init_ring_fini(struct amdgpu_device *adev)
+{
+   struct amdgpu_hiq *hiq = >gfx.hiq;
+
+   amdgpu_bo_free_kernel(>eop_obj, >eop_gpu_addr, NULL);
+}
+
+int amdgpu_gfx_hiq_init(struct amdgpu_device *adev,
+   unsigned hpd_size)
+{
+   int r;
+   u32 *hpd;
+   struct amdgpu_hiq *hiq = >gfx.hiq;
+
+   r = amdgpu_bo_create_kernel(adev, hpd_size, PAGE_SIZE,
+   AMDGPU_GEM_DOMAIN_GTT, >eop_obj,
+   >eop_gpu_addr, (void **));
+   if (r) {
+   dev_warn(adev->dev, "failed to create HIQ bo (%d).\n", r);
+   return r;
+   }
+
+   memset(hpd, 0, hpd_size);
+
+   r = amdgpu_bo_reserve(hiq->eop_obj, true);
+   if (unlikely(r != 0))
+   dev_warn(adev->dev, "(%d) reserve hiq eop bo failed\n", r);
+   amdgpu_bo_kunmap(hiq->eop_obj);
+   amdgpu_bo_unreserve(hiq->eop_obj);
+
+   return 0;
+}
+
+int amdgpu_gfx_disable_hiq(struct amdgpu_device *adev)
+{
+   struct amdgpu_kiq *kiq = >gfx.kiq;
+   struct amdgpu_ring *kiq_ring = >ring;
+   int r;
+
+   if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
+   return -EINVAL;
+
+   spin_lock(>gfx.kiq.ring_lock);
+   if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
+   spin_unlock(>gfx.kiq.ring_lock);
+   return -ENOMEM;
+   }
+
+   kiq->pmf->kiq_unmap_queues(kiq_ring, >gfx.kiq.ring, RESET_QUEUES,
+  0, 0);
+   r = amdgpu_ring_test_helper(kiq_ring);
+   spin_unlock(>gfx.kiq.ring_lock);
+
+   return r;
+}
+
+int amdgpu_gfx_enable_hiq(struct amdg

[PATCH 1/1] drm/amdgpu: return early on error while setting bar0 memtype

[Nirmoy Das]

We set WC memtype for aper_base but don't check return value
of arch_io_reserve_memtype_wc(). Be more defensive and return
early on error.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 7 ++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 073ba2af0b9c..6b25982a9077 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1032,9 +1032,14 @@ int amdgpu_bo_init(struct amdgpu_device *adev)
/* On A+A platform, VRAM can be mapped as WB */
if (!adev->gmc.xgmi.connected_to_cpu) {
/* reserve PAT memory space to WC for VRAM */
-   arch_io_reserve_memtype_wc(adev->gmc.aper_base,
+   int r = arch_io_reserve_memtype_wc(adev->gmc.aper_base,
adev->gmc.aper_size);
 
+   if (r) {
+   DRM_ERROR("Unable to set WC memtype for the aperture 
base\n");
+   return r;
+   }
+
/* Add an MTRR for the VRAM */
adev->gmc.vram_mtrr = arch_phys_wc_add(adev->gmc.aper_base,
adev->gmc.aper_size);
-- 
2.33.1

[PATCH 1/1] drm/amdgpu: remove unnecessary checks

[Nirmoy Das]

amdgpu_ttm_backend_bind() only needed for TTM_PL_TT
and AMDGPU_PL_PREEMPT.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 5 -
 1 file changed, 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index d784f8d3a834..eb872fc4ad92 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -914,11 +914,6 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
 ttm->num_pages, bo_mem, ttm);
}
 
-   if (bo_mem->mem_type == AMDGPU_PL_GDS ||
-   bo_mem->mem_type == AMDGPU_PL_GWS ||
-   bo_mem->mem_type == AMDGPU_PL_OA)
-   return -EINVAL;
-
if (bo_mem->mem_type != TTM_PL_TT ||
!amdgpu_gtt_mgr_has_gart_addr(bo_mem)) {
gtt->offset = AMDGPU_BO_INVALID_OFFSET;
-- 
2.32.0

[PATCH 3/3] drm/amdgpu: recover gart table at resume

[Nirmoy Das]

Get rid off pin/unpin of gart BO at resume/suspend and
instead pin only once and try to recover gart content
at resume time. This is much more stable in case there
is OOM situation at 2nd call to amdgpu_device_evict_resources()
while evicting GART table.

v3: remove gart recovery from other places
v2: pin gart at amdgpu_gart_table_vram_alloc()
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 11 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c   | 84 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  |  3 +-
 7 files changed, 11 insertions(+), 99 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 2b53d86aebac..f0c70e9d37fb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3935,16 +3935,11 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
if (!adev->in_s0ix)
amdgpu_amdkfd_suspend(adev, adev->in_runpm);
 
-   /* First evict vram memory */
amdgpu_device_evict_resources(adev);
 
amdgpu_fence_driver_hw_fini(adev);
 
amdgpu_device_ip_suspend_phase2(adev);
-   /* This second call to evict device resources is to evict
-* the gart page table using the CPU.
-*/
-   amdgpu_device_evict_resources(adev);
 
return 0;
 }
@@ -4286,8 +4281,6 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,
goto error;
 
amdgpu_virt_init_data_exchange(adev);
-   /* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);
 
r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,10 +4597,6 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}
 
-   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
-   if (r)
-   goto out;
-
r = amdgpu_device_fw_loading(tmp_adev);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index d3e4203f6217..645950a653a0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -114,80 +114,12 @@ void amdgpu_gart_dummy_page_fini(struct amdgpu_device 
*adev)
  */
 int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
 {
-   int r;
-
-   if (adev->gart.bo == NULL) {
-   struct amdgpu_bo_param bp;
-
-   memset(, 0, sizeof(bp));
-   bp.size = adev->gart.table_size;
-   bp.byte_align = PAGE_SIZE;
-   bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-   AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = NULL;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >gart.bo);
-   if (r) {
-   return r;
-   }
-   }
-   return 0;
-}
-
-/**
- * amdgpu_gart_table_vram_pin - pin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Pin the GART page table in vram so it will not be moved
- * by the memory manager (pcie r4xx, r5xx+).  These asics require the
- * gart table to be in video memory.
- * Returns 0 for success, error for failure.
- */
-int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev)
-{
-   int r;
-
-   r = amdgpu_bo_reserve(adev->gart.bo, false);
-   if (unlikely(r != 0))
-   return r;
-   r = amdgpu_bo_pin(adev->gart.bo, AMDGPU_GEM_DOMAIN_VRAM);
-   if (r) {
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-   }
-   r = amdgpu_bo_kmap(adev->gart.bo, >gart.ptr);
-   if (r)
-   amdgpu_bo_unpin(adev->gart.bo);
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-}
-
-/**
- * amdgpu_gart_table_vram_unpin - unpin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Unpin the GART page table in vram (pcie r4xx, r5xx+).
- * These asics require the gart table to be in video memory.
- */
-void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev)
-{
-   int r;
+   if (adev->gart.bo != NULL)
+   return 0;
 
-   if (adev->gart.bo == NULL) {
-   return;
-   }
-   r = amdgpu_bo_reserve

[PATCH 2/3] drm/amdgpu: do not pass ttm_resource_manager to vram_mgr

[Nirmoy Das]

Do not allow exported amdgpu_vram_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call vram_mgr functions.

v2: pass adev's vram_mgr instead of adev
Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c   |  3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c   |  5 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c  | 10 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c  |  6 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h  |  8 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c |  5 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 40 
 7 files changed, 31 insertions(+), 46 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 7077f21f0021..df818e145d9a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -531,9 +531,8 @@ int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int 
dma_buf_fd,
 uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd)
 {
struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);

-   return amdgpu_vram_mgr_usage(vram_man);
+   return amdgpu_vram_mgr_usage(>mman.vram_mgr);
 }

 uint64_t amdgpu_amdkfd_get_hive_id(struct kgd_dev *kgd)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 76fe5b71e35d..7e745164a624 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -298,7 +298,6 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
 {
s64 time_us, increment_us;
u64 free_vram, total_vram, used_vram;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
/* Allow a maximum of 200 accumulated ms. This is basically per-IB
 * throttling.
 *
@@ -315,7 +314,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
}

total_vram = adev->gmc.real_vram_size - 
atomic64_read(>vram_pin_size);
-   used_vram = amdgpu_vram_mgr_usage(vram_man);
+   used_vram = amdgpu_vram_mgr_usage(>mman.vram_mgr);
free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;

spin_lock(>mm_stats.lock);
@@ -362,7 +361,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
if (!amdgpu_gmc_vram_full_visible(>gmc)) {
u64 total_vis_vram = adev->gmc.visible_vram_size;
u64 used_vis_vram =
- amdgpu_vram_mgr_vis_usage(vram_man);
+ amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);

if (used_vis_vram < total_vis_vram) {
u64 free_vis_vram = total_vis_vram - used_vis_vram;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 603ce32db5c5..b426e03ad630 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -672,10 +672,10 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = atomic64_read(>num_vram_cpu_page_faults);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VRAM_USAGE:
-   ui64 = amdgpu_vram_mgr_usage(ttm_manager_type(>mman.bdev, 
TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VIS_VRAM_USAGE:
-   ui64 = 
amdgpu_vram_mgr_vis_usage(ttm_manager_type(>mman.bdev, TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
ui64 = amdgpu_gtt_mgr_usage(>mman.gtt_mgr);
@@ -709,8 +709,6 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
}
case AMDGPU_INFO_MEMORY: {
struct drm_amdgpu_memory_info mem;
-   struct ttm_resource_manager *vram_man =
-   ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
struct ttm_resource_manager *gtt_man =
ttm_manager_type(>mman.bdev, TTM_PL_TT);
memset(, 0, sizeof(mem));
@@ -719,7 +717,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
atomic64_read(>vram_pin_size) -
AMDGPU_VM_RESERVED_VRAM;
mem.vram.heap_usage =
-   amdgpu_vram_mgr_usage(vram_man);
+   amdgpu_vram_mgr_usage(>mman.vram_mgr);
 

[PATCH v4 3/3] drm/amdgpu: recover gart table at resume

[Nirmoy Das]

Get rid off pin/unpin of gart BO at resume/suspend and
instead pin only once and try to recover gart content
at resume time. This is much more stable in case there
is OOM situation at 2nd call to amdgpu_device_evict_resources()
while evicting GART table.

v4: return amdgpu_bo_create_kernel() directly without checking
its return value.
v3: remove gart recovery from other places
v2: pin gart at amdgpu_gart_table_vram_alloc()
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 11 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c   | 82 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  |  3 +-
 7 files changed, 11 insertions(+), 97 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 2b53d86aebac..f0c70e9d37fb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3935,16 +3935,11 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
if (!adev->in_s0ix)
amdgpu_amdkfd_suspend(adev, adev->in_runpm);

-   /* First evict vram memory */
amdgpu_device_evict_resources(adev);

amdgpu_fence_driver_hw_fini(adev);

amdgpu_device_ip_suspend_phase2(adev);
-   /* This second call to evict device resources is to evict
-* the gart page table using the CPU.
-*/
-   amdgpu_device_evict_resources(adev);

return 0;
 }
@@ -4286,8 +4281,6 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,
goto error;

amdgpu_virt_init_data_exchange(adev);
-   /* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);

r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,10 +4597,6 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}

-   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
-   if (r)
-   goto out;
-
r = amdgpu_device_fw_loading(tmp_adev);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index d3e4203f6217..3525f87dc1af 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -116,78 +116,12 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device 
*adev)
 {
int r;

-   if (adev->gart.bo == NULL) {
-   struct amdgpu_bo_param bp;
-
-   memset(, 0, sizeof(bp));
-   bp.size = adev->gart.table_size;
-   bp.byte_align = PAGE_SIZE;
-   bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-   AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = NULL;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >gart.bo);
-   if (r) {
-   return r;
-   }
-   }
-   return 0;
-}
-
-/**
- * amdgpu_gart_table_vram_pin - pin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Pin the GART page table in vram so it will not be moved
- * by the memory manager (pcie r4xx, r5xx+).  These asics require the
- * gart table to be in video memory.
- * Returns 0 for success, error for failure.
- */
-int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev)
-{
-   int r;
-
-   r = amdgpu_bo_reserve(adev->gart.bo, false);
-   if (unlikely(r != 0))
-   return r;
-   r = amdgpu_bo_pin(adev->gart.bo, AMDGPU_GEM_DOMAIN_VRAM);
-   if (r) {
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-   }
-   r = amdgpu_bo_kmap(adev->gart.bo, >gart.ptr);
-   if (r)
-   amdgpu_bo_unpin(adev->gart.bo);
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-}
-
-/**
- * amdgpu_gart_table_vram_unpin - unpin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Unpin the GART page table in vram (pcie r4xx, r5xx+).
- * These asics require the gart table to be in video memory.
- */
-void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev)
-{
-   int r;
+   if (adev->gart.bo != NULL)
+   return 0;

-   if (adev->gart.bo == NULL) {
-   return;
-   }
-   r = amdgpu_bo_reserve

[PATCH v4 1/3] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr

[Nirmoy Das]

Do not allow exported amdgpu_gtt_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call gtt_mgr functions.

v4: remove unused adev.
v3: upcast mgr from ttm resopurce manager instead of
getting it from adev.
v2: pass adev's gtt_mgr instead of adev.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 22 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h |  4 ++--
 4 files changed, 16 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 41ce86244144..2b53d86aebac 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4287,7 +4287,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,

amdgpu_virt_init_data_exchange(adev);
/* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev, TTM_PL_TT));
+   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);

r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,7 +4604,7 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}

-   r = 
amdgpu_gtt_mgr_recover(ttm_manager_type(_adev->mman.bdev, TTM_PL_TT));
+   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
if (r)
goto out;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index c18f16b3be9c..9151950e0cc3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -77,10 +77,8 @@ static ssize_t amdgpu_mem_info_gtt_used_show(struct device 
*dev,
 {
struct drm_device *ddev = dev_get_drvdata(dev);
struct amdgpu_device *adev = drm_to_adev(ddev);
-   struct ttm_resource_manager *man;

-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(man));
+   return sysfs_emit(buf, "%llu\n", 
amdgpu_gtt_mgr_usage(>mman.gtt_mgr));
 }

 static DEVICE_ATTR(mem_info_gtt_total, S_IRUGO,
@@ -206,14 +204,15 @@ static void amdgpu_gtt_mgr_del(struct 
ttm_resource_manager *man,
 /**
  * amdgpu_gtt_mgr_usage - return usage of GTT domain
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Return how many bytes are used in the GTT domain
  */
-uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager *man)
+uint64_t amdgpu_gtt_mgr_usage(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   s64 result = man->size - atomic64_read(>available);
+   s64 result;
+
+   result = mgr->manager.size - atomic64_read(>available);

return (result > 0 ? result : 0) * PAGE_SIZE;
 }
@@ -221,16 +220,15 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
 /**
  * amdgpu_gtt_mgr_recover - re-init gart
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Re-init the gart for each known BO in the GTT.
  */
-int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
+int amdgpu_gtt_mgr_recover(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   struct amdgpu_device *adev;
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
+   struct amdgpu_device *adev;
int r = 0;

adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
@@ -267,7 +265,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,

drm_printf(printer, "man size:%llu pages, gtt available:%lld pages, 
usage:%lluMB\n",
   man->size, (u64)atomic64_read(>available),
-  amdgpu_gtt_mgr_usage(man) >> 20);
+  amdgpu_gtt_mgr_usage(mgr) >> 20);
 }

 static const struct ttm_resource_manager_func amdgpu_gtt_mgr_func = {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index d2955ea4a62b..603ce32db5c5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -678,7 +678,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = 
amdgpu_vram_mgr_vis_usage(ttm_manager_type(>mman.bdev, TTM_PL_VRAM));
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
-   ui64 = amdgpu_gtt_mgr_usage(ttm_manager_type(>mman.bdev, 
TTM_PL_TT));
+   ui64 = amdgp

[PATCH v2 2/3] drm/amdgpu: do not pass ttm_resource_manager to vram_mgr

[Nirmoy Das]

Do not allow exported amdgpu_vram_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call vram_mgr functions.

v2: pass adev's vram_mgr instead of adev

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c   |  3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c   |  5 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c  | 10 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c  |  6 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h  |  8 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c |  5 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 40 
 7 files changed, 31 insertions(+), 46 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 7077f21f0021..df818e145d9a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -531,9 +531,8 @@ int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int 
dma_buf_fd,
 uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd)
 {
struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);

-   return amdgpu_vram_mgr_usage(vram_man);
+   return amdgpu_vram_mgr_usage(>mman.vram_mgr);
 }

 uint64_t amdgpu_amdkfd_get_hive_id(struct kgd_dev *kgd)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 76fe5b71e35d..7e745164a624 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -298,7 +298,6 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
 {
s64 time_us, increment_us;
u64 free_vram, total_vram, used_vram;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
/* Allow a maximum of 200 accumulated ms. This is basically per-IB
 * throttling.
 *
@@ -315,7 +314,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
}

total_vram = adev->gmc.real_vram_size - 
atomic64_read(>vram_pin_size);
-   used_vram = amdgpu_vram_mgr_usage(vram_man);
+   used_vram = amdgpu_vram_mgr_usage(>mman.vram_mgr);
free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;

spin_lock(>mm_stats.lock);
@@ -362,7 +361,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
if (!amdgpu_gmc_vram_full_visible(>gmc)) {
u64 total_vis_vram = adev->gmc.visible_vram_size;
u64 used_vis_vram =
- amdgpu_vram_mgr_vis_usage(vram_man);
+ amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);

if (used_vis_vram < total_vis_vram) {
u64 free_vis_vram = total_vis_vram - used_vis_vram;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 603ce32db5c5..b426e03ad630 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -672,10 +672,10 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = atomic64_read(>num_vram_cpu_page_faults);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VRAM_USAGE:
-   ui64 = amdgpu_vram_mgr_usage(ttm_manager_type(>mman.bdev, 
TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VIS_VRAM_USAGE:
-   ui64 = 
amdgpu_vram_mgr_vis_usage(ttm_manager_type(>mman.bdev, TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
ui64 = amdgpu_gtt_mgr_usage(>mman.gtt_mgr);
@@ -709,8 +709,6 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
}
case AMDGPU_INFO_MEMORY: {
struct drm_amdgpu_memory_info mem;
-   struct ttm_resource_manager *vram_man =
-   ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
struct ttm_resource_manager *gtt_man =
ttm_manager_type(>mman.bdev, TTM_PL_TT);
memset(, 0, sizeof(mem));
@@ -719,7 +717,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
atomic64_read(>vram_pin_size) -
AMDGPU_VM_RESERVED_VRAM;
mem.vram.heap_usage =
-   amdgpu_vram_mgr_usage(vram_man);
+   amdgpu_vram_mgr_usage(>mman.vram_mgr);
 

[PATCH v3 3/3] drm/amdgpu: recover gart table at resume

[Nirmoy Das]

Get rid off pin/unpin of gart BO at resume/suspend and
instead pin only once and try to recover gart content
at resume time. This is much more stable in case there
is OOM situation at 2nd call to amdgpu_device_evict_resources()
while evicting GART table.

v3: remove gart recovery from other places
v2: pin gart at amdgpu_gart_table_vram_alloc()

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 11 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c   | 80 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  |  3 +-
 7 files changed, 12 insertions(+), 94 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 2b53d86aebac..f0c70e9d37fb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3935,16 +3935,11 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
if (!adev->in_s0ix)
amdgpu_amdkfd_suspend(adev, adev->in_runpm);

-   /* First evict vram memory */
amdgpu_device_evict_resources(adev);

amdgpu_fence_driver_hw_fini(adev);

amdgpu_device_ip_suspend_phase2(adev);
-   /* This second call to evict device resources is to evict
-* the gart page table using the CPU.
-*/
-   amdgpu_device_evict_resources(adev);

return 0;
 }
@@ -4286,8 +4281,6 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,
goto error;

amdgpu_virt_init_data_exchange(adev);
-   /* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);

r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,10 +4597,6 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}

-   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
-   if (r)
-   goto out;
-
r = amdgpu_device_fw_loading(tmp_adev);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index d3e4203f6217..679eec122bb5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -116,78 +116,16 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device 
*adev)
 {
int r;

-   if (adev->gart.bo == NULL) {
-   struct amdgpu_bo_param bp;
-
-   memset(, 0, sizeof(bp));
-   bp.size = adev->gart.table_size;
-   bp.byte_align = PAGE_SIZE;
-   bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-   AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = NULL;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >gart.bo);
-   if (r) {
-   return r;
-   }
-   }
-   return 0;
-}
-
-/**
- * amdgpu_gart_table_vram_pin - pin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Pin the GART page table in vram so it will not be moved
- * by the memory manager (pcie r4xx, r5xx+).  These asics require the
- * gart table to be in video memory.
- * Returns 0 for success, error for failure.
- */
-int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev)
-{
-   int r;
+   if (adev->gart.bo != NULL)
+   return 0;

-   r = amdgpu_bo_reserve(adev->gart.bo, false);
-   if (unlikely(r != 0))
-   return r;
-   r = amdgpu_bo_pin(adev->gart.bo, AMDGPU_GEM_DOMAIN_VRAM);
+   r = amdgpu_bo_create_kernel(adev,  adev->gart.table_size, PAGE_SIZE,
+   AMDGPU_GEM_DOMAIN_VRAM, >gart.bo,
+   NULL, (void *)>gart.ptr);
if (r) {
-   amdgpu_bo_unreserve(adev->gart.bo);
return r;
}
-   r = amdgpu_bo_kmap(adev->gart.bo, >gart.ptr);
-   if (r)
-   amdgpu_bo_unpin(adev->gart.bo);
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-}
-
-/**
- * amdgpu_gart_table_vram_unpin - unpin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Unpin the GART page table in vram (pcie r4xx, r5xx+).
- * These asics require the gart table to be in video memory.
- */
-void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev)

[PATCH v3 1/3] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr

[Nirmoy Das]

Do not allow exported amdgpu_gtt_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call gtt_mgr functions.

v3: upcast mgr from ttm resopurce manager instead of
getting it from adev.
v2: pass adev's gtt_mgr instead of adev

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 23 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h |  4 ++--
 4 files changed, 17 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 41ce86244144..2b53d86aebac 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4287,7 +4287,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,

amdgpu_virt_init_data_exchange(adev);
/* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev, TTM_PL_TT));
+   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);

r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,7 +4604,7 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}

-   r = 
amdgpu_gtt_mgr_recover(ttm_manager_type(_adev->mman.bdev, TTM_PL_TT));
+   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
if (r)
goto out;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index c18f16b3be9c..62e14efa61fe 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -77,10 +77,8 @@ static ssize_t amdgpu_mem_info_gtt_used_show(struct device 
*dev,
 {
struct drm_device *ddev = dev_get_drvdata(dev);
struct amdgpu_device *adev = drm_to_adev(ddev);
-   struct ttm_resource_manager *man;

-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(man));
+   return sysfs_emit(buf, "%llu\n", 
amdgpu_gtt_mgr_usage(>mman.gtt_mgr));
 }

 static DEVICE_ATTR(mem_info_gtt_total, S_IRUGO,
@@ -206,14 +204,15 @@ static void amdgpu_gtt_mgr_del(struct 
ttm_resource_manager *man,
 /**
  * amdgpu_gtt_mgr_usage - return usage of GTT domain
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Return how many bytes are used in the GTT domain
  */
-uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager *man)
+uint64_t amdgpu_gtt_mgr_usage(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   s64 result = man->size - atomic64_read(>available);
+   s64 result;
+
+   result = mgr->manager.size - atomic64_read(>available);

return (result > 0 ? result : 0) * PAGE_SIZE;
 }
@@ -221,16 +220,15 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
 /**
  * amdgpu_gtt_mgr_recover - re-init gart
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Re-init the gart for each known BO in the GTT.
  */
-int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
+int amdgpu_gtt_mgr_recover(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   struct amdgpu_device *adev;
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
+   struct amdgpu_device *adev;
int r = 0;

adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
@@ -260,6 +258,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,
 struct drm_printer *printer)
 {
struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
+   struct amdgpu_device *adev = container_of(mgr, typeof(*adev), 
mman.gtt_mgr);

spin_lock(>lock);
drm_mm_print(>mm, printer);
@@ -267,7 +266,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,

drm_printf(printer, "man size:%llu pages, gtt available:%lld pages, 
usage:%lluMB\n",
   man->size, (u64)atomic64_read(>available),
-  amdgpu_gtt_mgr_usage(man) >> 20);
+  amdgpu_gtt_mgr_usage(mgr) >> 20);
 }

 static const struct ttm_resource_manager_func amdgpu_gtt_mgr_func = {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index d2955ea4a62b..603ce32db5c5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -678,7 +678,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_fi

[PATCH v2 2/3] drm/amdgpu: do not pass ttm_resource_manager to vram_mgr

[Nirmoy Das]

Do not allow exported amdgpu_vram_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call vram_mgr functions.

v2: pass adev's vram_mgr instead of adev
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c   |  3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c   |  5 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c  | 10 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c  |  6 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h  |  8 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c |  5 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 38 
 7 files changed, 30 insertions(+), 45 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 7077f21f0021..df818e145d9a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -531,9 +531,8 @@ int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int 
dma_buf_fd,
 uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd)
 {
struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);

-   return amdgpu_vram_mgr_usage(vram_man);
+   return amdgpu_vram_mgr_usage(>mman.vram_mgr);
 }

 uint64_t amdgpu_amdkfd_get_hive_id(struct kgd_dev *kgd)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 76fe5b71e35d..7e745164a624 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -298,7 +298,6 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
 {
s64 time_us, increment_us;
u64 free_vram, total_vram, used_vram;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
/* Allow a maximum of 200 accumulated ms. This is basically per-IB
 * throttling.
 *
@@ -315,7 +314,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
}

total_vram = adev->gmc.real_vram_size - 
atomic64_read(>vram_pin_size);
-   used_vram = amdgpu_vram_mgr_usage(vram_man);
+   used_vram = amdgpu_vram_mgr_usage(>mman.vram_mgr);
free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;

spin_lock(>mm_stats.lock);
@@ -362,7 +361,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
if (!amdgpu_gmc_vram_full_visible(>gmc)) {
u64 total_vis_vram = adev->gmc.visible_vram_size;
u64 used_vis_vram =
- amdgpu_vram_mgr_vis_usage(vram_man);
+ amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);

if (used_vis_vram < total_vis_vram) {
u64 free_vis_vram = total_vis_vram - used_vis_vram;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 603ce32db5c5..b426e03ad630 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -672,10 +672,10 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = atomic64_read(>num_vram_cpu_page_faults);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VRAM_USAGE:
-   ui64 = amdgpu_vram_mgr_usage(ttm_manager_type(>mman.bdev, 
TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VIS_VRAM_USAGE:
-   ui64 = 
amdgpu_vram_mgr_vis_usage(ttm_manager_type(>mman.bdev, TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_vis_usage(>mman.vram_mgr);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
ui64 = amdgpu_gtt_mgr_usage(>mman.gtt_mgr);
@@ -709,8 +709,6 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
}
case AMDGPU_INFO_MEMORY: {
struct drm_amdgpu_memory_info mem;
-   struct ttm_resource_manager *vram_man =
-   ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
struct ttm_resource_manager *gtt_man =
ttm_manager_type(>mman.bdev, TTM_PL_TT);
memset(, 0, sizeof(mem));
@@ -719,7 +717,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
atomic64_read(>vram_pin_size) -
AMDGPU_VM_RESERVED_VRAM;
mem.vram.heap_usage =
-   amdgpu_vram_mgr_usage(vram_man);
+   amdgpu_vram_mgr_usage(>mman.vram_mgr);
 

[PATCH v3 3/3] drm/amdgpu: recover gart table at resume

[Nirmoy Das]

Get rid off pin/unpin of gart BO at resume/suspend and
instead pin only once and try to recover gart content
at resume time. This is much more stable in case there
is OOM situation at 2nd call to amdgpu_device_evict_resources()
while evicting GART table.

v3: remove gart recovery from other places
v2: pin gart at amdgpu_gart_table_vram_alloc()
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 11 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c   | 80 ++
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  |  3 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  |  3 +-
 7 files changed, 12 insertions(+), 94 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 2b53d86aebac..f0c70e9d37fb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3935,16 +3935,11 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
if (!adev->in_s0ix)
amdgpu_amdkfd_suspend(adev, adev->in_runpm);

-   /* First evict vram memory */
amdgpu_device_evict_resources(adev);

amdgpu_fence_driver_hw_fini(adev);

amdgpu_device_ip_suspend_phase2(adev);
-   /* This second call to evict device resources is to evict
-* the gart page table using the CPU.
-*/
-   amdgpu_device_evict_resources(adev);

return 0;
 }
@@ -4286,8 +4281,6 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,
goto error;

amdgpu_virt_init_data_exchange(adev);
-   /* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);

r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,10 +4597,6 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}

-   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
-   if (r)
-   goto out;
-
r = amdgpu_device_fw_loading(tmp_adev);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index d3e4203f6217..679eec122bb5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -116,78 +116,16 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device 
*adev)
 {
int r;

-   if (adev->gart.bo == NULL) {
-   struct amdgpu_bo_param bp;
-
-   memset(, 0, sizeof(bp));
-   bp.size = adev->gart.table_size;
-   bp.byte_align = PAGE_SIZE;
-   bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-   AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = NULL;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >gart.bo);
-   if (r) {
-   return r;
-   }
-   }
-   return 0;
-}
-
-/**
- * amdgpu_gart_table_vram_pin - pin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Pin the GART page table in vram so it will not be moved
- * by the memory manager (pcie r4xx, r5xx+).  These asics require the
- * gart table to be in video memory.
- * Returns 0 for success, error for failure.
- */
-int amdgpu_gart_table_vram_pin(struct amdgpu_device *adev)
-{
-   int r;
+   if (adev->gart.bo != NULL)
+   return 0;

-   r = amdgpu_bo_reserve(adev->gart.bo, false);
-   if (unlikely(r != 0))
-   return r;
-   r = amdgpu_bo_pin(adev->gart.bo, AMDGPU_GEM_DOMAIN_VRAM);
+   r = amdgpu_bo_create_kernel(adev,  adev->gart.table_size, PAGE_SIZE,
+   AMDGPU_GEM_DOMAIN_VRAM, >gart.bo,
+   NULL, (void *)>gart.ptr);
if (r) {
-   amdgpu_bo_unreserve(adev->gart.bo);
return r;
}
-   r = amdgpu_bo_kmap(adev->gart.bo, >gart.ptr);
-   if (r)
-   amdgpu_bo_unpin(adev->gart.bo);
-   amdgpu_bo_unreserve(adev->gart.bo);
-   return r;
-}
-
-/**
- * amdgpu_gart_table_vram_unpin - unpin gart page table in vram
- *
- * @adev: amdgpu_device pointer
- *
- * Unpin the GART page table in vram (pcie r4xx, r5xx+).
- * These asics require the gart table to be in video memory.
- */
-void amdgpu_gart_table_vram_unpin(struct amdgpu_device *adev)

[PATCH v2 1/3] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr

[Nirmoy Das]

Do not allow exported amdgpu_gtt_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call gtt_mgr functions.

v2: pass adev's gtt_mgr instead of adev

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 23 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  4 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h |  4 ++--
 4 files changed, 17 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 41ce86244144..2b53d86aebac 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4287,7 +4287,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,

amdgpu_virt_init_data_exchange(adev);
/* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev, TTM_PL_TT));
+   amdgpu_gtt_mgr_recover(>mman.gtt_mgr);

r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,7 +4604,7 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}

-   r = 
amdgpu_gtt_mgr_recover(ttm_manager_type(_adev->mman.bdev, TTM_PL_TT));
+   r = 
amdgpu_gtt_mgr_recover(_adev->mman.gtt_mgr);
if (r)
goto out;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index c18f16b3be9c..e429f2df73be 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -77,10 +77,8 @@ static ssize_t amdgpu_mem_info_gtt_used_show(struct device 
*dev,
 {
struct drm_device *ddev = dev_get_drvdata(dev);
struct amdgpu_device *adev = drm_to_adev(ddev);
-   struct ttm_resource_manager *man;

-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(man));
+   return sysfs_emit(buf, "%llu\n", 
amdgpu_gtt_mgr_usage(>mman.gtt_mgr));
 }

 static DEVICE_ATTR(mem_info_gtt_total, S_IRUGO,
@@ -206,14 +204,15 @@ static void amdgpu_gtt_mgr_del(struct 
ttm_resource_manager *man,
 /**
  * amdgpu_gtt_mgr_usage - return usage of GTT domain
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Return how many bytes are used in the GTT domain
  */
-uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager *man)
+uint64_t amdgpu_gtt_mgr_usage(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   s64 result = man->size - atomic64_read(>available);
+   s64 result;
+
+   result = mgr->manager.size - atomic64_read(>available);

return (result > 0 ? result : 0) * PAGE_SIZE;
 }
@@ -221,16 +220,15 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
 /**
  * amdgpu_gtt_mgr_recover - re-init gart
  *
- * @man: TTM memory type manager
+ * @mgr: amdgpu_gtt_mgr pointer
  *
  * Re-init the gart for each known BO in the GTT.
  */
-int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
+int amdgpu_gtt_mgr_recover(struct amdgpu_gtt_mgr *mgr)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   struct amdgpu_device *adev;
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
+   struct amdgpu_device *adev;
int r = 0;

adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
@@ -260,6 +258,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,
 struct drm_printer *printer)
 {
struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
+   struct amdgpu_device *adev = container_of(mgr, typeof(*adev), 
mman.gtt_mgr);

spin_lock(>lock);
drm_mm_print(>mm, printer);
@@ -267,7 +266,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,

drm_printf(printer, "man size:%llu pages, gtt available:%lld pages, 
usage:%lluMB\n",
   man->size, (u64)atomic64_read(>available),
-  amdgpu_gtt_mgr_usage(man) >> 20);
+  amdgpu_gtt_mgr_usage(>mman.gtt_mgr) >> 20);
 }

 static const struct ttm_resource_manager_func amdgpu_gtt_mgr_func = {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index d2955ea4a62b..603ce32db5c5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -678,7 +678,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = 
amdgpu_vram_mgr_vis_usage(

[PATCH 1/3] drm/amdgpu: do not pass ttm_resource_manager to gtt_mgr

[Nirmoy Das]

Do not allow exported amdgpu_gtt_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call gtt_mgr functions.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  4 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 31 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  4 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h |  4 +--
 4 files changed, 24 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 41ce86244144..5807df52031c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4287,7 +4287,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device 
*adev,
 
amdgpu_virt_init_data_exchange(adev);
/* we need recover gart prior to run SMC/CP/SDMA resume */
-   amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev, TTM_PL_TT));
+   amdgpu_gtt_mgr_recover(adev);
 
r = amdgpu_device_fw_loading(adev);
if (r)
@@ -4604,7 +4604,7 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
amdgpu_inc_vram_lost(tmp_adev);
}
 
-   r = 
amdgpu_gtt_mgr_recover(ttm_manager_type(_adev->mman.bdev, TTM_PL_TT));
+   r = amdgpu_gtt_mgr_recover(tmp_adev);
if (r)
goto out;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index c18f16b3be9c..5e41f8ef743a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -77,10 +77,8 @@ static ssize_t amdgpu_mem_info_gtt_used_show(struct device 
*dev,
 {
struct drm_device *ddev = dev_get_drvdata(dev);
struct amdgpu_device *adev = drm_to_adev(ddev);
-   struct ttm_resource_manager *man;
 
-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(man));
+   return sysfs_emit(buf, "%llu\n", amdgpu_gtt_mgr_usage(adev));
 }
 
 static DEVICE_ATTR(mem_info_gtt_total, S_IRUGO,
@@ -206,14 +204,19 @@ static void amdgpu_gtt_mgr_del(struct 
ttm_resource_manager *man,
 /**
  * amdgpu_gtt_mgr_usage - return usage of GTT domain
  *
- * @man: TTM memory type manager
+ * @adev: amdgpu_device pointer
  *
  * Return how many bytes are used in the GTT domain
  */
-uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager *man)
+uint64_t amdgpu_gtt_mgr_usage(struct amdgpu_device *adev)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   s64 result = man->size - atomic64_read(>available);
+   struct ttm_resource_manager *man;
+   struct amdgpu_gtt_mgr *mgr;
+   s64 result;
+
+   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
+   mgr = to_gtt_mgr(man);
+   result = man->size - atomic64_read(>available);
 
return (result > 0 ? result : 0) * PAGE_SIZE;
 }
@@ -221,19 +224,20 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
 /**
  * amdgpu_gtt_mgr_recover - re-init gart
  *
- * @man: TTM memory type manager
+ * @adev: amdgpu_device pointer
  *
  * Re-init the gart for each known BO in the GTT.
  */
-int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
+int amdgpu_gtt_mgr_recover(struct amdgpu_device *adev)
 {
-   struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
-   struct amdgpu_device *adev;
+   struct ttm_resource_manager *man;
+   struct amdgpu_gtt_mgr *mgr;
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
int r = 0;
 
-   adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
+   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
+   mgr = to_gtt_mgr(man);
spin_lock(>lock);
drm_mm_for_each_node(mm_node, >mm) {
node = container_of(mm_node, typeof(*node), base.mm_nodes[0]);
@@ -260,6 +264,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,
 struct drm_printer *printer)
 {
struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
+   struct amdgpu_device *adev = container_of(mgr, typeof(*adev), 
mman.gtt_mgr);
 
spin_lock(>lock);
drm_mm_print(>mm, printer);
@@ -267,7 +272,7 @@ static void amdgpu_gtt_mgr_debug(struct 
ttm_resource_manager *man,
 
drm_printf(printer, "man size:%llu pages, gtt available:%lld pages, 
usage:%lluMB\n",
   man->size, (u64)atomic64_read(>available),
-  amdgpu_gtt_mgr_usage(man) >> 20);
+  amdgpu_gtt_mgr_usage(adev) >> 20);
 }
 
 static const struct ttm_resource_manager_func amdgpu_gtt_mgr_func = {
diff --git 

[PATCH v2 3/3] drm/amdgpu: recover gart table at resume

[Nirmoy Das]

Get rid off pin/unpin of gart BO at resume/suspend and
instead pin only once and try to recover gart content
at resume time. This is much more stable in case there
is OOM situation at 2nd call to amdgpu_device_evict_resources()
while evicting GART table.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  4 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c   | 42 --
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c |  9 ++---
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c  | 10 +++---
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  | 10 +++---
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  |  9 ++---
 6 files changed, 45 insertions(+), 39 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 5807df52031c..f69e613805db 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3941,10 +3941,6 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
amdgpu_fence_driver_hw_fini(adev);

amdgpu_device_ip_suspend_phase2(adev);
-   /* This second call to evict device resources is to evict
-* the gart page table using the CPU.
-*/
-   amdgpu_device_evict_resources(adev);

return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index d3e4203f6217..97a9f61fa106 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -107,33 +107,37 @@ void amdgpu_gart_dummy_page_fini(struct amdgpu_device 
*adev)
  *
  * @adev: amdgpu_device pointer
  *
- * Allocate video memory for GART page table
+ * Allocate and pin video memory for GART page table
  * (pcie r4xx, r5xx+).  These asics require the
  * gart table to be in video memory.
  * Returns 0 for success, error for failure.
  */
 int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
 {
+   struct amdgpu_bo_param bp;
int r;

-   if (adev->gart.bo == NULL) {
-   struct amdgpu_bo_param bp;
-
-   memset(, 0, sizeof(bp));
-   bp.size = adev->gart.table_size;
-   bp.byte_align = PAGE_SIZE;
-   bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
-   AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = NULL;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >gart.bo);
-   if (r) {
-   return r;
-   }
-   }
+   if (adev->gart.bo != NULL)
+   return 0;
+
+   memset(, 0, sizeof(bp));
+   bp.size = adev->gart.table_size;
+   bp.byte_align = PAGE_SIZE;
+   bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+   bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+   AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+   bp.type = ttm_bo_type_kernel;
+   bp.resv = NULL;
+   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+
+   r = amdgpu_bo_create(adev, , >gart.bo);
+   if (r)
+   return r;
+
+   r = amdgpu_gart_table_vram_pin(adev);
+   if (r)
+   return r;
+
return 0;
 }

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
index 3ec5ff5a6dbe..75d584e1b0e9 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
@@ -992,9 +992,11 @@ static int gmc_v10_0_gart_enable(struct amdgpu_device 
*adev)
return -EINVAL;
}

-   r = amdgpu_gart_table_vram_pin(adev);
-   if (r)
-   return r;
+   if (adev->in_suspend) {
+   r = amdgpu_gtt_mgr_recover(adev);
+   if (r)
+   return r;
+   }

r = adev->gfxhub.funcs->gart_enable(adev);
if (r)
@@ -1062,7 +1064,6 @@ static void gmc_v10_0_gart_disable(struct amdgpu_device 
*adev)
 {
adev->gfxhub.funcs->gart_disable(adev);
adev->mmhub.funcs->gart_disable(adev);
-   amdgpu_gart_table_vram_unpin(adev);
 }

 static int gmc_v10_0_hw_fini(void *handle)
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
index 0a50fdaced7e..02e90d9443c1 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
@@ -620,9 +620,12 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev)
dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
return -EINVAL;
}
-   r = amdgpu_gart_table_vram_pin(adev);
-   if (r)
-   return r;
+
+   if (adev->in_suspend) {
+   r = amdgpu_gtt_mgr_recover(adev);
+   if (r)
+   return r;
+   }

table_addr = amdgpu_bo_gpu_

[PATCH 2/3] drm/amdgpu: do not pass ttm_resource_manager to vram_mgr

[Nirmoy Das]

Do not allow exported amdgpu_vram_mgr_*() to accept
any ttm_resource_manager pointer. Also there is no need
to force other module to call a ttm function just to
eventually call vram_mgr functions.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c   |  3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c   |  5 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c  | 10 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c  |  6 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h  |  8 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c |  5 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 54 
 7 files changed, 49 insertions(+), 42 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 7077f21f0021..4837c579a787 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -531,9 +531,8 @@ int amdgpu_amdkfd_get_dmabuf_info(struct kgd_dev *kgd, int 
dma_buf_fd,
 uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd)
 {
struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
 
-   return amdgpu_vram_mgr_usage(vram_man);
+   return amdgpu_vram_mgr_usage(adev);
 }
 
 uint64_t amdgpu_amdkfd_get_hive_id(struct kgd_dev *kgd)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 76fe5b71e35d..f4084ca8b614 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -298,7 +298,6 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
 {
s64 time_us, increment_us;
u64 free_vram, total_vram, used_vram;
-   struct ttm_resource_manager *vram_man = 
ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
/* Allow a maximum of 200 accumulated ms. This is basically per-IB
 * throttling.
 *
@@ -315,7 +314,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
}
 
total_vram = adev->gmc.real_vram_size - 
atomic64_read(>vram_pin_size);
-   used_vram = amdgpu_vram_mgr_usage(vram_man);
+   used_vram = amdgpu_vram_mgr_usage(adev);
free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram;
 
spin_lock(>mm_stats.lock);
@@ -362,7 +361,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct 
amdgpu_device *adev,
if (!amdgpu_gmc_vram_full_visible(>gmc)) {
u64 total_vis_vram = adev->gmc.visible_vram_size;
u64 used_vis_vram =
- amdgpu_vram_mgr_vis_usage(vram_man);
+ amdgpu_vram_mgr_vis_usage(adev);
 
if (used_vis_vram < total_vis_vram) {
u64 free_vis_vram = total_vis_vram - used_vis_vram;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index b9b38f70e416..34674ccabd67 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -672,10 +672,10 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
ui64 = atomic64_read(>num_vram_cpu_page_faults);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VRAM_USAGE:
-   ui64 = amdgpu_vram_mgr_usage(ttm_manager_type(>mman.bdev, 
TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_usage(adev);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_VIS_VRAM_USAGE:
-   ui64 = 
amdgpu_vram_mgr_vis_usage(ttm_manager_type(>mman.bdev, TTM_PL_VRAM));
+   ui64 = amdgpu_vram_mgr_vis_usage(adev);
return copy_to_user(out, , min(size, 8u)) ? -EFAULT : 0;
case AMDGPU_INFO_GTT_USAGE:
ui64 = amdgpu_gtt_mgr_usage(adev);
@@ -709,8 +709,6 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
}
case AMDGPU_INFO_MEMORY: {
struct drm_amdgpu_memory_info mem;
-   struct ttm_resource_manager *vram_man =
-   ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
struct ttm_resource_manager *gtt_man =
ttm_manager_type(>mman.bdev, TTM_PL_TT);
memset(, 0, sizeof(mem));
@@ -719,7 +717,7 @@ int amdgpu_info_ioctl(struct drm_device *dev, void *data, 
struct drm_file *filp)
atomic64_read(>vram_pin_size) -
AMDGPU_VM_RESERVED_VRAM;
mem.vram.heap_usage =
-   amdgpu_vram_mgr_usage(vram_man);
+   amdgpu_vram_mgr_usage(adev);
mem.vram.max_allocation = mem.vram.usable_heap_size * 3 / 4;
 
mem.cpu_accessible_vram.total_heap_size =
@@ -729,7 +727,7 @@

[PATCH 1/1] drm/amdgpu: recover gart table at resume

[Nirmoy Das]

Get rid off pin/unpin and evict and swap back gart
page table which should make things less likely to break.

Also remove 2nd call to amdgpu_device_evict_resources()
as we don't need it.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  5 -
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 16 
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c  | 17 +
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c  | 16 
 4 files changed, 37 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 41ce86244144..22ff229ab981 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3941,11 +3941,6 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
amdgpu_fence_driver_hw_fini(adev);
 
amdgpu_device_ip_suspend_phase2(adev);
-   /* This second call to evict device resources is to evict
-* the gart page table using the CPU.
-*/
-   amdgpu_device_evict_resources(adev);
-
return 0;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
index 3ec5ff5a6dbe..18e3f3c5aae6 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
@@ -992,9 +992,16 @@ static int gmc_v10_0_gart_enable(struct amdgpu_device 
*adev)
return -EINVAL;
}
 
-   r = amdgpu_gart_table_vram_pin(adev);
-   if (r)
-   return r;
+   if (!adev->in_suspend) {
+   r = amdgpu_gart_table_vram_pin(adev);
+   if (r)
+   return r;
+   } else {
+   r = amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev,
+   TTM_PL_TT));
+   if (r)
+   return r;
+   }
 
r = adev->gfxhub.funcs->gart_enable(adev);
if (r)
@@ -1062,7 +1069,8 @@ static void gmc_v10_0_gart_disable(struct amdgpu_device 
*adev)
 {
adev->gfxhub.funcs->gart_disable(adev);
adev->mmhub.funcs->gart_disable(adev);
-   amdgpu_gart_table_vram_unpin(adev);
+   if (!adev->in_suspend)
+   amdgpu_gart_table_vram_unpin(adev);
 }
 
 static int gmc_v10_0_hw_fini(void *handle)
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
index 492ebed2915b..0ef50ad3d7d5 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
@@ -837,9 +837,17 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
return -EINVAL;
}
-   r = amdgpu_gart_table_vram_pin(adev);
-   if (r)
-   return r;
+
+   if (!adev->in_suspend) {
+   r = amdgpu_gart_table_vram_pin(adev);
+   if (r)
+   return r;
+   } else {
+   r = amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev,
+   TTM_PL_TT));
+   if (r)
+   return r;
+   }
 
table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
 
@@ -992,7 +1000,8 @@ static void gmc_v8_0_gart_disable(struct amdgpu_device 
*adev)
tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
WREG32(mmVM_L2_CNTL, tmp);
WREG32(mmVM_L2_CNTL2, 0);
-   amdgpu_gart_table_vram_unpin(adev);
+   if (!adev->in_suspend)
+   amdgpu_gart_table_vram_unpin(adev);
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index cb82404df534..1bbcefd53974 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -1714,9 +1714,16 @@ static int gmc_v9_0_gart_enable(struct amdgpu_device 
*adev)
return -EINVAL;
}
 
-   r = amdgpu_gart_table_vram_pin(adev);
-   if (r)
-   return r;
+   if (!adev->in_suspend) {
+   r = amdgpu_gart_table_vram_pin(adev);
+   if (r)
+   return r;
+   } else {
+   r = amdgpu_gtt_mgr_recover(ttm_manager_type(>mman.bdev,
+   TTM_PL_TT));
+   if (r)
+   return r;
+   }
 
r = adev->gfxhub.funcs->gart_enable(adev);
if (r)
@@ -1793,7 +1800,8 @@ static void gmc_v9_0_gart_disable(struct amdgpu_device 
*adev)
 {
adev->gfxhub.funcs->gart_disable(adev);
adev->mmhub.funcs->gart_disable(adev);
-   amdgpu_gart_table_vram_unpin(adev);
+   if (!adev->in_suspend)
+   amdgpu_gart_table_vram_unpin(adev);
 }
 
 static int gmc_v9_0_hw_fini(void *handle)
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: fix BO leak after successful move test

[Nirmoy Das]

GTT BO cleanup code is with in the test for loop and
we would skip cleaning up GTT BO on success.

Reported-by: zhang 
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_test.c | 25 
 1 file changed, 12 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
index 909d830b513e..5fe7ff680c29 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
@@ -35,6 +35,7 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
struct amdgpu_bo *vram_obj = NULL;
struct amdgpu_bo **gtt_obj = NULL;
struct amdgpu_bo_param bp;
+   struct dma_fence *fence = NULL;
uint64_t gart_addr, vram_addr;
unsigned n, size;
int i, r;
@@ -82,7 +83,6 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
void *gtt_map, *vram_map;
void **gart_start, **gart_end;
void **vram_start, **vram_end;
-   struct dma_fence *fence = NULL;
 
bp.domain = AMDGPU_GEM_DOMAIN_GTT;
r = amdgpu_bo_create(adev, , gtt_obj + i);
@@ -212,24 +212,23 @@ static void amdgpu_do_test_moves(struct amdgpu_device 
*adev)
 
DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 
0x%llx\n",
 gart_addr - adev->gmc.gart_start);
-   continue;
+   }
 
+   --i;
 out_lclean_unpin:
-   amdgpu_bo_unpin(gtt_obj[i]);
+   amdgpu_bo_unpin(gtt_obj[i]);
 out_lclean_unres:
-   amdgpu_bo_unreserve(gtt_obj[i]);
+   amdgpu_bo_unreserve(gtt_obj[i]);
 out_lclean_unref:
-   amdgpu_bo_unref(_obj[i]);
+   amdgpu_bo_unref(_obj[i]);
 out_lclean:
-   for (--i; i >= 0; --i) {
-   amdgpu_bo_unpin(gtt_obj[i]);
-   amdgpu_bo_unreserve(gtt_obj[i]);
-   amdgpu_bo_unref(_obj[i]);
-   }
-   if (fence)
-   dma_fence_put(fence);
-   break;
+   for (--i; i >= 0; --i) {
+   amdgpu_bo_unpin(gtt_obj[i]);
+   amdgpu_bo_unreserve(gtt_obj[i]);
+   amdgpu_bo_unref(_obj[i]);
}
+   if (fence)
+   dma_fence_put(fence);
 
amdgpu_bo_unpin(vram_obj);
 out_unres:
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: release gtt bo after each move test

[Nirmoy Das]

When gart size is < gtt size this test will fail with
-ENOMEM as we are not freeing gtt bo after each move test.
This is generally not an issue when gart size >= gtt size.

Reported-by: zhang 
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_test.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
index 909d830b513e..0cf2a560d673 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
@@ -212,7 +212,6 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 
DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 
0x%llx\n",
 gart_addr - adev->gmc.gart_start);
-   continue;
 
 out_lclean_unpin:
amdgpu_bo_unpin(gtt_obj[i]);
@@ -220,6 +219,7 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
amdgpu_bo_unreserve(gtt_obj[i]);
 out_lclean_unref:
amdgpu_bo_unref(_obj[i]);
+   continue;
 out_lclean:
for (--i; i >= 0; --i) {
amdgpu_bo_unpin(gtt_obj[i]);
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: release gtt bo after each move test

[Nirmoy Das]

When gart size is < gtt size this test will fail with
-ENOMEM as we are not freeing gtt bo after each move test.
This is generally not an issue when gart size >= gtt size.

Reported-by: zhang 
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_test.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
index 909d830b513e..0cf2a560d673 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
@@ -212,7 +212,6 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
 
DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 
0x%llx\n",
 gart_addr - adev->gmc.gart_start);
-   continue;
 
 out_lclean_unpin:
amdgpu_bo_unpin(gtt_obj[i]);
@@ -220,6 +219,7 @@ static void amdgpu_do_test_moves(struct amdgpu_device *adev)
amdgpu_bo_unreserve(gtt_obj[i]);
 out_lclean_unref:
amdgpu_bo_unref(_obj[i]);
+   continue;
 out_lclean:
for (--i; i >= 0; --i) {
amdgpu_bo_unpin(gtt_obj[i]);
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: unify BO evicting method in amdgpu_ttm

[Nirmoy Das]

Unify BO evicting functionality for possible memory
types in amdgpu_ttm.c.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c |  8 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  | 30 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 30 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h |  1 +
 6 files changed, 58 insertions(+), 35 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 5497e2d31d1a..164d6a9e9fbb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1328,7 +1328,7 @@ static int amdgpu_debugfs_evict_vram(void *data, u64 *val)
return r;
}
 
-   *val = amdgpu_bo_evict_vram(adev);
+   *val = amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM);
 
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
@@ -1341,17 +1341,15 @@ static int amdgpu_debugfs_evict_gtt(void *data, u64 
*val)
 {
struct amdgpu_device *adev = (struct amdgpu_device *)data;
struct drm_device *dev = adev_to_drm(adev);
-   struct ttm_resource_manager *man;
int r;
 
r = pm_runtime_get_sync(dev->dev);
if (r < 0) {
-   pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+   pm_runtime_put_autosuspend(dev->dev);
return r;
}
 
-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   *val = ttm_resource_manager_evict_all(>mman.bdev, man);
+   *val = amdgpu_ttm_evict_resources(adev, TTM_PL_TT);
 
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 57638fe9cfc2..032deca4cea2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3880,6 +3880,25 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev)
 
 }
 
+/**
+ * amdgpu_device_evict_resources - evict device resources
+ * @adev: amdgpu device object
+ *
+ * Evicts all ttm device resources(vram BOs, gart table) from the lru list
+ * of the vram memory type. Mainly used for evicting device resources
+ * at suspend time.
+ *
+ */
+void amdgpu_device_evict_resources(struct amdgpu_device *adev)
+{
+   /* No need to evict vram on APUs for suspend to ram */
+   if (adev->in_s3 && (adev->flags & AMD_IS_APU))
+   return;
+
+   if (amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM))
+   DRM_WARN("evicting device resources failed\n");
+
+}
 
 /*
  * Suspend & resume.
@@ -3920,17 +3939,16 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
if (!adev->in_s0ix)
amdgpu_amdkfd_suspend(adev, adev->in_runpm);
 
-   /* evict vram memory */
-   amdgpu_bo_evict_vram(adev);
+   /* First evict vram memory */
+   amdgpu_device_evict_resources(adev);
 
amdgpu_fence_driver_hw_fini(adev);
 
amdgpu_device_ip_suspend_phase2(adev);
-   /* evict remaining vram memory
-* This second call to evict vram is to evict the gart page table
-* using the CPU.
+   /* This second call to evict device resources is to evict
+* the gart page table using the CPU.
 */
-   amdgpu_bo_evict_vram(adev);
+   amdgpu_device_evict_resources(adev);
 
return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 4ec904f36ceb..073ba2af0b9c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1004,29 +1004,6 @@ void amdgpu_bo_unpin(struct amdgpu_bo *bo)
}
 }
 
-/**
- * amdgpu_bo_evict_vram - evict VRAM buffers
- * @adev: amdgpu device object
- *
- * Evicts all VRAM buffers on the lru list of the memory type.
- * Mainly used for evicting vram at suspend time.
- *
- * Returns:
- * 0 for success or a negative error code on failure.
- */
-int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
-{
-   struct ttm_resource_manager *man;
-
-   if (adev->in_s3 && (adev->flags & AMD_IS_APU)) {
-   /* No need to evict vram on APUs for suspend to ram */
-   return 0;
-   }
-
-   man = ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
-   return ttm_resource_manager_evict_all(>mman.bdev, man);
-}
-
 static const char *amdgpu_vram_names[] = {
"UNKNOWN",
"GDDR1",
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 8ff61bad4138..d787e0e89e0b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/d

[PATCH 1/1] drm/amdgpu: unify BO evicting method in amdgpu_ttm

[Nirmoy Das]

Unify BO evicting functionality for possible memory
types in amdgpu_ttm.c and remove corresponding function
from amdgpu_object.c.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c |  8 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |  4 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 23 
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 30 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h |  1 +
 6 files changed, 36 insertions(+), 31 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 5497e2d31d1a..22f3de29d783 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1328,7 +1328,7 @@ static int amdgpu_debugfs_evict_vram(void *data, u64 *val)
return r;
}
 
-   *val = amdgpu_bo_evict_vram(adev);
+   *val = amdgpu_bo_evict_memory(adev, TTM_PL_VRAM);
 
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
@@ -1341,17 +1341,15 @@ static int amdgpu_debugfs_evict_gtt(void *data, u64 
*val)
 {
struct amdgpu_device *adev = (struct amdgpu_device *)data;
struct drm_device *dev = adev_to_drm(adev);
-   struct ttm_resource_manager *man;
int r;
 
r = pm_runtime_get_sync(dev->dev);
if (r < 0) {
-   pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+   pm_runtime_put_autosuspend(dev->dev);
return r;
}
 
-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   *val = ttm_resource_manager_evict_all(>mman.bdev, man);
+   *val = amdgpu_bo_evict_memory(adev, TTM_PL_TT);
 
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 57638fe9cfc2..c441ebe9da11 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3921,7 +3921,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
amdgpu_amdkfd_suspend(adev, adev->in_runpm);
 
/* evict vram memory */
-   amdgpu_bo_evict_vram(adev);
+   amdgpu_bo_evict_memory(adev, TTM_PL_VRAM);
 
amdgpu_fence_driver_hw_fini(adev);
 
@@ -3930,7 +3930,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool 
fbcon)
 * This second call to evict vram is to evict the gart page table
 * using the CPU.
 */
-   amdgpu_bo_evict_vram(adev);
+   amdgpu_bo_evict_memory(adev, TTM_PL_VRAM);
 
return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 4ec904f36ceb..073ba2af0b9c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1004,29 +1004,6 @@ void amdgpu_bo_unpin(struct amdgpu_bo *bo)
}
 }
 
-/**
- * amdgpu_bo_evict_vram - evict VRAM buffers
- * @adev: amdgpu device object
- *
- * Evicts all VRAM buffers on the lru list of the memory type.
- * Mainly used for evicting vram at suspend time.
- *
- * Returns:
- * 0 for success or a negative error code on failure.
- */
-int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
-{
-   struct ttm_resource_manager *man;
-
-   if (adev->in_s3 && (adev->flags & AMD_IS_APU)) {
-   /* No need to evict vram on APUs for suspend to ram */
-   return 0;
-   }
-
-   man = ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
-   return ttm_resource_manager_evict_all(>mman.bdev, man);
-}
-
 static const char *amdgpu_vram_names[] = {
"UNKNOWN",
"GDDR1",
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 8ff61bad4138..d787e0e89e0b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -305,7 +305,6 @@ int amdgpu_bo_pin(struct amdgpu_bo *bo, u32 domain);
 int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
 u64 min_offset, u64 max_offset);
 void amdgpu_bo_unpin(struct amdgpu_bo *bo);
-int amdgpu_bo_evict_vram(struct amdgpu_device *adev);
 int amdgpu_bo_init(struct amdgpu_device *adev);
 void amdgpu_bo_fini(struct amdgpu_device *adev);
 int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index e2896ac2c9ce..545b4bdeae07 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -2034,6 +2034,36 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
return r;
 }
 
+/**
+ * amdgpu_bo_evict_memory - evict memory buffers
+ * @adev: am

[PATCH 1/1] drm/amdgpu: add and use amdgpu_bo_evict_gtt

[Nirmoy Das]

Unify BO evicting functionality for VRAM and TT memory
types in amdgpu_object.c. Use amdgpu_bo_evict_gtt()
for evicting gtt memory similar to how we do that
for amdgpu_debugfs_evict_vram().

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c |  6 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c  | 52 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h  |  1 +
 3 files changed, 50 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 5497e2d31d1a..67045983d63d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1341,17 +1341,15 @@ static int amdgpu_debugfs_evict_gtt(void *data, u64 
*val)
 {
struct amdgpu_device *adev = (struct amdgpu_device *)data;
struct drm_device *dev = adev_to_drm(adev);
-   struct ttm_resource_manager *man;
int r;
 
r = pm_runtime_get_sync(dev->dev);
if (r < 0) {
-   pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+   pm_runtime_put_autosuspend(dev->dev);
return r;
}
 
-   man = ttm_manager_type(>mman.bdev, TTM_PL_TT);
-   *val = ttm_resource_manager_evict_all(>mman.bdev, man);
+   *val = amdgpu_bo_evict_gtt(adev);
 
pm_runtime_mark_last_busy(dev->dev);
pm_runtime_put_autosuspend(dev->dev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 4ec904f36ceb..3b8c9cf44d74 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1005,10 +1005,37 @@ void amdgpu_bo_unpin(struct amdgpu_bo *bo)
 }
 
 /**
- * amdgpu_bo_evict_vram - evict VRAM buffers
+ * amdgpu_bo_evict_memory - evict memory buffers
  * @adev: amdgpu device object
+ * @mem_type: evicted BO's memory type
  *
- * Evicts all VRAM buffers on the lru list of the memory type.
+ * Evicts all @mem_type buffers on the lru list of the memory type.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+static int amdgpu_bo_evict_memory(struct amdgpu_device *adev, int mem_type)
+{
+   struct ttm_resource_manager *man;
+
+   switch (mem_type) {
+   case TTM_PL_VRAM:
+   case TTM_PL_TT:
+   man = ttm_manager_type(>mman.bdev, mem_type);
+   break;
+   default:
+   DRM_ERROR("Trying to evict invalid memory type\n");
+   return -EINVAL;
+   }
+
+   return ttm_resource_manager_evict_all(>mman.bdev, man);
+}
+
+/**
+ * amdgpu_bo_evict_vram - evict vram buffers
+ * @adev: amdgpu device object
+ *
+ * Evicts all vram buffers on the lru list of the memory type.
  * Mainly used for evicting vram at suspend time.
  *
  * Returns:
@@ -1016,17 +1043,32 @@ void amdgpu_bo_unpin(struct amdgpu_bo *bo)
  */
 int amdgpu_bo_evict_vram(struct amdgpu_device *adev)
 {
-   struct ttm_resource_manager *man;
 
if (adev->in_s3 && (adev->flags & AMD_IS_APU)) {
/* No need to evict vram on APUs for suspend to ram */
return 0;
}
 
-   man = ttm_manager_type(>mman.bdev, TTM_PL_VRAM);
-   return ttm_resource_manager_evict_all(>mman.bdev, man);
+   return amdgpu_bo_evict_memory(adev, TTM_PL_VRAM);
+}
+
+/**
+ * amdgpu_bo_evict_gtt - evict gtt buffers
+ * @adev: amdgpu device object
+ *
+ * Evicts all gtt buffers on the lru list of the memory type.
+ * Mainly used for evicting gtt buffers through debugfs.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+
+int amdgpu_bo_evict_gtt(struct amdgpu_device *adev)
+{
+   return amdgpu_bo_evict_memory(adev, TTM_PL_TT);
 }
 
+
 static const char *amdgpu_vram_names[] = {
"UNKNOWN",
"GDDR1",
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 8ff61bad4138..5e9b7710b8e3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -306,6 +306,7 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 
domain,
 u64 min_offset, u64 max_offset);
 void amdgpu_bo_unpin(struct amdgpu_bo *bo);
 int amdgpu_bo_evict_vram(struct amdgpu_device *adev);
+int amdgpu_bo_evict_gtt(struct amdgpu_device *adev);
 int amdgpu_bo_init(struct amdgpu_device *adev);
 void amdgpu_bo_fini(struct amdgpu_device *adev);
 int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: return early if debugfs is not initialized

[Nirmoy Das]

Check first if debugfs is initialized before creating
amdgpu debugfs files.

References: https://gitlab.freedesktop.org/drm/amd/-/issues/1686
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 6611b3c7c149..5497e2d31d1a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1617,6 +1617,9 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
struct dentry *ent;
int r, i;
 
+   if (!debugfs_initialized())
+   return 0;
+
ent = debugfs_create_file("amdgpu_preempt_ib", 0600, root, adev,
  _ib_preempt);
if (IS_ERR(ent)) {
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: ignore -EPERM error from debugfs

[Nirmoy Das]

Debugfs core APIs will throw -EPERM when user disables debugfs
using CONFIG_DEBUG_FS_ALLOW_NONE or with kernel param. We shouldn't
see that as an error. Also validate drm root dentry before creating
amdgpu debugfs files.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 10 ++
 1 file changed, 10 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 6611b3c7c149..d786072e918b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1617,6 +1617,16 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
struct dentry *ent;
int r, i;
 
+   if (IS_ERR(root)) {
+   /* When debugfs is disabled we get -EPERM which is not an
+* error as this is user controllable.
+*/
+   if (PTR_ERR(root) == -EPERM)
+   return 0;
+
+   return PTR_ERR(ent);
+   }
+
ent = debugfs_create_file("amdgpu_preempt_ib", 0600, root, adev,
  _ib_preempt);
if (IS_ERR(ent)) {
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: init debugfs drm driver callback

[Nirmoy Das]

drm_dev_register() will try to init driver's debugfs using
drm_driver.debugfs_init call back function. Use that callback
also for amdgpu to intialize debugfs.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 10 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c |  7 +++
 3 files changed, 9 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 6611b3c7c149..3076742f8f85 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1611,8 +1611,9 @@ DEFINE_DEBUGFS_ATTRIBUTE(fops_ib_preempt, NULL,
 DEFINE_DEBUGFS_ATTRIBUTE(fops_sclk_set, NULL,
amdgpu_debugfs_sclk_set, "%llu\n");
 
-int amdgpu_debugfs_init(struct amdgpu_device *adev)
+void amdgpu_debugfs_init(struct drm_minor *minor)
 {
+   struct amdgpu_device *adev = drm_to_adev(minor->dev);
struct dentry *root = adev_to_drm(adev)->primary->debugfs_root;
struct dentry *ent;
int r, i;
@@ -1621,14 +1622,14 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
  _ib_preempt);
if (IS_ERR(ent)) {
DRM_ERROR("unable to create amdgpu_preempt_ib debugsfs file\n");
-   return PTR_ERR(ent);
+   return;
}
 
ent = debugfs_create_file("amdgpu_force_sclk", 0200, root, adev,
  _sclk_set);
if (IS_ERR(ent)) {
DRM_ERROR("unable to create amdgpu_set_sclk debugsfs file\n");
-   return PTR_ERR(ent);
+   return;
}
 
/* Register debugfs entries for amdgpu_ttm */
@@ -1682,11 +1683,10 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
debugfs_create_blob("amdgpu_discovery", 0444, root,
>debugfs_discovery_blob);
 
-   return 0;
 }
 
 #else
-int amdgpu_debugfs_init(struct amdgpu_device *adev)
+void amdgpu_debugfs_init(struct drm_minor *minor)
 {
return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
index 371a6f0deb29..06b68e16e35d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.h
@@ -27,7 +27,7 @@
  */
 
 int amdgpu_debugfs_regs_init(struct amdgpu_device *adev);
-int amdgpu_debugfs_init(struct amdgpu_device *adev);
+void amdgpu_debugfs_init(struct drm_minor *minor);
 void amdgpu_debugfs_fini(struct amdgpu_device *adev);
 void amdgpu_debugfs_fence_init(struct amdgpu_device *adev);
 void amdgpu_debugfs_firmware_init(struct amdgpu_device *adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index df83b1f438b6..ceda650895db 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -2012,10 +2012,6 @@ static int amdgpu_pci_probe(struct pci_dev *pdev,
drm_fbdev_generic_setup(adev_to_drm(adev), 32);
}
 
-   ret = amdgpu_debugfs_init(adev);
-   if (ret)
-   DRM_ERROR("Creating debugfs files failed (%d).\n", ret);
-
return 0;
 
 err_pci:
@@ -2479,6 +2475,9 @@ static const struct drm_driver amdgpu_kms_driver = {
.dumb_map_offset = amdgpu_mode_dumb_mmap,
.fops = _driver_kms_fops,
.release = _driver_release_kms,
+#if defined(CONFIG_DEBUG_FS)
+   .debugfs_init = amdgpu_debugfs_init,
+#endif
 
.prime_handle_to_fd = drm_gem_prime_handle_to_fd,
.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
-- 
2.32.0

[PATCH 1/1] drm/radeon: pass drm dev radeon_agp_head_init directly

[Nirmoy Das]

Pass drm dev directly as rdev->ddev gets initialized later on
at radeon_device_init().

Bug: https://bugzilla.kernel.org/show_bug.cgi?id=214375
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/radeon/radeon_kms.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/radeon/radeon_kms.c 
b/drivers/gpu/drm/radeon/radeon_kms.c
index 0473583dcdac..482fb0ae6cb5 100644
--- a/drivers/gpu/drm/radeon/radeon_kms.c
+++ b/drivers/gpu/drm/radeon/radeon_kms.c
@@ -119,7 +119,7 @@ int radeon_driver_load_kms(struct drm_device *dev, unsigned 
long flags)
 #endif
 
if (pci_find_capability(pdev, PCI_CAP_ID_AGP))
-   rdev->agp = radeon_agp_head_init(rdev->ddev);
+   rdev->agp = radeon_agp_head_init(dev);
if (rdev->agp) {
rdev->agp->agp_mtrr = arch_phys_wc_add(
rdev->agp->agp_info.aper_base,
-- 
2.32.0

[PATCH v2 1/2] drm/amdgpu: use IS_ERR for debugfs APIs

[Nirmoy Das]

debugfs APIs returns encoded error so use
IS_ERR for checking return value.

v2: return PTR_ERR(ent)

References: https://gitlab.freedesktop.org/drm/amd/-/issues/1686
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 10 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c|  4 ++--
 2 files changed, 6 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index d256215ab2c7..60f46a4b0144 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1696,20 +1696,18 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
struct dentry *ent;
int r, i;

-
-
ent = debugfs_create_file("amdgpu_preempt_ib", 0600, root, adev,
  _ib_preempt);
-   if (!ent) {
+   if (IS_ERR(ent)) {
DRM_ERROR("unable to create amdgpu_preempt_ib debugsfs file\n");
-   return -EIO;
+   return PTR_ERR(ent);
}

ent = debugfs_create_file("amdgpu_force_sclk", 0200, root, adev,
  _sclk_set);
-   if (!ent) {
+   if (IS_ERR(ent)) {
DRM_ERROR("unable to create amdgpu_set_sclk debugsfs file\n");
-   return -EIO;
+   return PTR_ERR(ent);
}

/* Register debugfs entries for amdgpu_ttm */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 7b634a1517f9..0554576d3695 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -428,8 +428,8 @@ int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
ent = debugfs_create_file(name,
  S_IFREG | S_IRUGO, root,
  ring, _debugfs_ring_fops);
-   if (!ent)
-   return -ENOMEM;
+   if (IS_ERR(ent))
+   return PTR_ERR(ent);

i_size_write(ent->d_inode, ring->ring_size + 12);
ring->ent = ent;
--
2.32.0

[PATCH 2/2] drm/amdgpu: cleanup debugfs for amdgpu rings

[Nirmoy Das]

Use debugfs_create_file_size API for creating ring debugfs
file, also cleanup surrounding code.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c |  4 +---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c| 18 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h| 10 ++
 3 files changed, 9 insertions(+), 23 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 60f46a4b0144..97d88f3e1c4c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1734,9 +1734,7 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
if (!ring)
continue;
 
-   if (amdgpu_debugfs_ring_init(adev, ring)) {
-   DRM_ERROR("Failed to register debugfs file for rings 
!\n");
-   }
+   amdgpu_debugfs_ring_init(adev, ring);
}
 
amdgpu_ras_debugfs_create_all(adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 0554576d3695..ab2351ba9574 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -415,26 +415,20 @@ static const struct file_operations 
amdgpu_debugfs_ring_fops = {
 
 #endif
 
-int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
-struct amdgpu_ring *ring)
+void amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
+ struct amdgpu_ring *ring)
 {
 #if defined(CONFIG_DEBUG_FS)
struct drm_minor *minor = adev_to_drm(adev)->primary;
-   struct dentry *ent, *root = minor->debugfs_root;
+   struct dentry *root = minor->debugfs_root;
char name[32];
 
sprintf(name, "amdgpu_ring_%s", ring->name);
+   debugfs_create_file_size(name, S_IFREG | S_IRUGO, root, ring,
+_debugfs_ring_fops,
+ring->ring_size + 12);
 
-   ent = debugfs_create_file(name,
- S_IFREG | S_IRUGO, root,
- ring, _debugfs_ring_fops);
-   if (IS_ERR(ent))
-   return PTR_ERR(ent);
-
-   i_size_write(ent->d_inode, ring->ring_size + 12);
-   ring->ent = ent;
 #endif
-   return 0;
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index 88d80eb3fea1..4d380e79752c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -253,10 +253,6 @@ struct amdgpu_ring {
boolhas_compute_vm_bug;
boolno_scheduler;
int hw_prio;
-
-#if defined(CONFIG_DEBUG_FS)
-   struct dentry *ent;
-#endif
 };
 
 #define amdgpu_ring_parse_cs(r, p, ib) ((r)->funcs->parse_cs((p), (ib)))
@@ -356,8 +352,6 @@ static inline void amdgpu_ring_write_multiple(struct 
amdgpu_ring *ring,
 
 int amdgpu_ring_test_helper(struct amdgpu_ring *ring);
 
-int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
-struct amdgpu_ring *ring);
-void amdgpu_debugfs_ring_fini(struct amdgpu_ring *ring);
-
+void amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
+ struct amdgpu_ring *ring);
 #endif
-- 
2.32.0

[PATCH 2/2] drm/amdgpu: cleanup debugfs for amdgpu rings

[Nirmoy Das]

Use debugfs_create_file_size API for creating ring debugfs
file, also cleanup surrounding code.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c |  4 +---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c| 16 +---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h|  8 +---
 3 files changed, 7 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index 077f9baf74fe..dee56ab19a8f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1734,9 +1734,7 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
if (!ring)
continue;
 
-   if (amdgpu_debugfs_ring_init(adev, ring)) {
-   DRM_ERROR("Failed to register debugfs file for rings 
!\n");
-   }
+   amdgpu_debugfs_ring_init(adev, ring);
}
 
amdgpu_ras_debugfs_create_all(adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index f40753e1a60d..968521d80514 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -415,26 +415,20 @@ static const struct file_operations 
amdgpu_debugfs_ring_fops = {
 
 #endif
 
-int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
+void amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
 struct amdgpu_ring *ring)
 {
 #if defined(CONFIG_DEBUG_FS)
struct drm_minor *minor = adev_to_drm(adev)->primary;
-   struct dentry *ent, *root = minor->debugfs_root;
+   struct dentry *root = minor->debugfs_root;
char name[32];
 
sprintf(name, "amdgpu_ring_%s", ring->name);
 
-   ent = debugfs_create_file(name,
- S_IFREG | S_IRUGO, root,
- ring, _debugfs_ring_fops);
-   if (IS_ERR(ent))
-   return -ENOMEM;
-
-   i_size_write(ent->d_inode, ring->ring_size + 12);
-   ring->ent = ent;
+   debugfs_create_file_size(name, S_IFREG | S_IRUGO, root, ring,
+_debugfs_ring_fops,
+ring->ring_size + 12);
 #endif
-   return 0;
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index 88d80eb3fea1..c29fbce0a5b4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -253,10 +253,6 @@ struct amdgpu_ring {
boolhas_compute_vm_bug;
boolno_scheduler;
int hw_prio;
-
-#if defined(CONFIG_DEBUG_FS)
-   struct dentry *ent;
-#endif
 };
 
 #define amdgpu_ring_parse_cs(r, p, ib) ((r)->funcs->parse_cs((p), (ib)))
@@ -356,8 +352,6 @@ static inline void amdgpu_ring_write_multiple(struct 
amdgpu_ring *ring,
 
 int amdgpu_ring_test_helper(struct amdgpu_ring *ring);
 
-int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
+void amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
 struct amdgpu_ring *ring);
-void amdgpu_debugfs_ring_fini(struct amdgpu_ring *ring);
-
 #endif
-- 
2.32.0

[PATCH 1/2] drm/amdgpu: use IS_ERR for debugfs APIs

[Nirmoy Das]

debugfs APIs returns encoded error so use
IS_ERR for checking return value.

References: https://gitlab.freedesktop.org/drm/amd/-/issues/1686
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 6 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c| 2 +-
 2 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index d256215ab2c7..077f9baf74fe 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1696,18 +1696,16 @@ int amdgpu_debugfs_init(struct amdgpu_device *adev)
struct dentry *ent;
int r, i;
 
-
-
ent = debugfs_create_file("amdgpu_preempt_ib", 0600, root, adev,
  _ib_preempt);
-   if (!ent) {
+   if (IS_ERR(ent)) {
DRM_ERROR("unable to create amdgpu_preempt_ib debugsfs file\n");
return -EIO;
}
 
ent = debugfs_create_file("amdgpu_force_sclk", 0200, root, adev,
  _sclk_set);
-   if (!ent) {
+   if (IS_ERR(ent)) {
DRM_ERROR("unable to create amdgpu_set_sclk debugsfs file\n");
return -EIO;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 7b634a1517f9..f40753e1a60d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -428,7 +428,7 @@ int amdgpu_debugfs_ring_init(struct amdgpu_device *adev,
ent = debugfs_create_file(name,
  S_IFREG | S_IRUGO, root,
  ring, _debugfs_ring_fops);
-   if (!ent)
+   if (IS_ERR(ent))
return -ENOMEM;
 
i_size_write(ent->d_inode, ring->ring_size + 12);
-- 
2.32.0

[PATCH v3 1/1] drm/amdgpu: detach ring priority from gfx priority

[Nirmoy Das]

Currently AMDGPU_RING_PRIO_MAX is redefinition of a
max gfx hwip priority, this won't work well when we will
have a hwip with different set of priorities than gfx.
Also, HW ring priorities are different from ring priorities.

Create a global enum for ring priority levels which each
HWIP can use to define its own priority levels.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  | 2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h  | 7 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 9 +++--
 3 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index c88c5c6c54a2..0d1928260650 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -109,7 +109,7 @@ static int amdgpu_ctx_priority_permit(struct drm_file *filp,
return -EACCES;
 }

-static enum gfx_pipe_priority amdgpu_ctx_prio_to_compute_prio(int32_t prio)
+static enum amdgpu_gfx_pipe_priority amdgpu_ctx_prio_to_compute_prio(int32_t 
prio)
 {
switch (prio) {
case AMDGPU_CTX_PRIORITY_HIGH:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index d43fe2ed8116..f851196c83a5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -42,10 +42,9 @@
 #define AMDGPU_MAX_GFX_QUEUES KGD_MAX_QUEUES
 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES

-enum gfx_pipe_priority {
-   AMDGPU_GFX_PIPE_PRIO_NORMAL = 1,
-   AMDGPU_GFX_PIPE_PRIO_HIGH,
-   AMDGPU_GFX_PIPE_PRIO_MAX
+enum amdgpu_gfx_pipe_priority {
+   AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1,
+   AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2
 };

 /* Argument for PPSMC_MSG_GpuChangeState */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index e713d31619fe..88d80eb3fea1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -36,8 +36,13 @@
 #define AMDGPU_MAX_VCE_RINGS   3
 #define AMDGPU_MAX_UVD_ENC_RINGS   2

-#define AMDGPU_RING_PRIO_DEFAULT   1
-#define AMDGPU_RING_PRIO_MAX   AMDGPU_GFX_PIPE_PRIO_MAX
+enum amdgpu_ring_priority_level {
+   AMDGPU_RING_PRIO_0,
+   AMDGPU_RING_PRIO_1,
+   AMDGPU_RING_PRIO_DEFAULT = 1,
+   AMDGPU_RING_PRIO_2,
+   AMDGPU_RING_PRIO_MAX
+};

 /* some special values for the owner field */
 #define AMDGPU_FENCE_OWNER_UNDEFINED   ((void *)0ul)
--
2.32.0

[PATCH v2 1/1] drm/amdgpu: detach ring priority from gfx priority

[Nirmoy Das]

Currently AMDGPU_RING_PRIO_MAX is redefinition of a
max gfx hwip priority, this won't work well when we will
have a hwip with different set of priorities than gfx.
Also, HW ring priorities are different from ring priorities.

Create a global enum for ring priority levels which each
HWIP can use to define its own priority levels.

Signed-off-by: Nirmoy Das 
Reviewed-by: Lijo Lazar 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h  | 5 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 9 +++--
 2 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index d43fe2ed8116..7f747a4291f3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -43,9 +43,8 @@
 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES

 enum gfx_pipe_priority {
-   AMDGPU_GFX_PIPE_PRIO_NORMAL = 1,
-   AMDGPU_GFX_PIPE_PRIO_HIGH,
-   AMDGPU_GFX_PIPE_PRIO_MAX
+   AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1,
+   AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2
 };

 /* Argument for PPSMC_MSG_GpuChangeState */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index e713d31619fe..88d80eb3fea1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -36,8 +36,13 @@
 #define AMDGPU_MAX_VCE_RINGS   3
 #define AMDGPU_MAX_UVD_ENC_RINGS   2

-#define AMDGPU_RING_PRIO_DEFAULT   1
-#define AMDGPU_RING_PRIO_MAX   AMDGPU_GFX_PIPE_PRIO_MAX
+enum amdgpu_ring_priority_level {
+   AMDGPU_RING_PRIO_0,
+   AMDGPU_RING_PRIO_1,
+   AMDGPU_RING_PRIO_DEFAULT = 1,
+   AMDGPU_RING_PRIO_2,
+   AMDGPU_RING_PRIO_MAX
+};

 /* some special values for the owner field */
 #define AMDGPU_FENCE_OWNER_UNDEFINED   ((void *)0ul)
--
2.32.0

[PATCH 1/1] drm/amdgpu: detach ring priority from gfx priority

[Nirmoy Das]

Currently AMDGPU_RING_PRIO_MAX is redefinition of a
max gfx hwip priority, this won't work well when we will
have a hwip with different set of priorities than gfx.
Also, HW ring priorities are different from ring priorities.

Create a global enum for ring priority levels which each
HWIP can use to define its own priority levels.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h  |  6 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 10 --
 2 files changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index d43fe2ed8116..937320293029 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -43,9 +43,9 @@
 #define AMDGPU_MAX_COMPUTE_QUEUES KGD_MAX_QUEUES
 
 enum gfx_pipe_priority {
-   AMDGPU_GFX_PIPE_PRIO_NORMAL = 1,
-   AMDGPU_GFX_PIPE_PRIO_HIGH,
-   AMDGPU_GFX_PIPE_PRIO_MAX
+   AMDGPU_GFX_PIPE_PRIO_NORMAL = AMDGPU_RING_PRIO_1,
+   AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2,
+   AMDGPU_GFX_PIPE_PRIO_MAX = AMDGPU_RING_PRIO_3
 };
 
 /* Argument for PPSMC_MSG_GpuChangeState */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
index e713d31619fe..85541005c1ad 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
@@ -36,8 +36,14 @@
 #define AMDGPU_MAX_VCE_RINGS   3
 #define AMDGPU_MAX_UVD_ENC_RINGS   2
 
-#define AMDGPU_RING_PRIO_DEFAULT   1
-#define AMDGPU_RING_PRIO_MAX   AMDGPU_GFX_PIPE_PRIO_MAX
+enum amdgpu_ring_priority_level {
+   AMDGPU_RING_PRIO_0,
+   AMDGPU_RING_PRIO_1,
+   AMDGPU_RING_PRIO_DEFAULT = 1,
+   AMDGPU_RING_PRIO_2,
+   AMDGPU_RING_PRIO_3,
+   AMDGPU_RING_PRIO_MAX
+};
 
 /* some special values for the owner field */
 #define AMDGPU_FENCE_OWNER_UNDEFINED   ((void *)0ul)
-- 
2.32.0

[PATCH 1/1] drm/amdgpu: rework context priority handling

[Nirmoy Das]

To get a hardware queue priority for a context, we are currently
mapping AMDGPU_CTX_PRIORITY_* to DRM_SCHED_PRIORITY_* and then
to hardware queue priority, which is not the right way to do that
as DRM_SCHED_PRIORITY_* is software scheduler's priority and it is
independent from a hardware queue priority.

Use userspace provided context priority, AMDGPU_CTX_PRIORITY_* to
map a context to proper hardware queue priority.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c   | 127 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h   |   8 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c |  44 ++--
 3 files changed, 105 insertions(+), 74 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index e7a010b7ca1f..c88c5c6c54a2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -43,14 +43,61 @@ const unsigned int 
amdgpu_ctx_num_entities[AMDGPU_HW_IP_NUM] = {
[AMDGPU_HW_IP_VCN_JPEG] =   1,
 };
 
+bool amdgpu_ctx_priority_is_valid(int32_t ctx_prio)
+{
+   switch (ctx_prio) {
+   case AMDGPU_CTX_PRIORITY_UNSET:
+   case AMDGPU_CTX_PRIORITY_VERY_LOW:
+   case AMDGPU_CTX_PRIORITY_LOW:
+   case AMDGPU_CTX_PRIORITY_NORMAL:
+   case AMDGPU_CTX_PRIORITY_HIGH:
+   case AMDGPU_CTX_PRIORITY_VERY_HIGH:
+   return true;
+   default:
+   return false;
+   }
+}
+
+static enum drm_sched_priority
+amdgpu_ctx_to_drm_sched_prio(int32_t ctx_prio)
+{
+   switch (ctx_prio) {
+   case AMDGPU_CTX_PRIORITY_UNSET:
+   return DRM_SCHED_PRIORITY_UNSET;
+
+   case AMDGPU_CTX_PRIORITY_VERY_LOW:
+   return DRM_SCHED_PRIORITY_MIN;
+
+   case AMDGPU_CTX_PRIORITY_LOW:
+   return DRM_SCHED_PRIORITY_MIN;
+
+   case AMDGPU_CTX_PRIORITY_NORMAL:
+   return DRM_SCHED_PRIORITY_NORMAL;
+
+   case AMDGPU_CTX_PRIORITY_HIGH:
+   return DRM_SCHED_PRIORITY_HIGH;
+
+   case AMDGPU_CTX_PRIORITY_VERY_HIGH:
+   return DRM_SCHED_PRIORITY_HIGH;
+
+   /* This should not happen as we sanitized userspace provided priority
+* already, WARN if this happens.
+*/
+   default:
+   WARN(1, "Invalid context priority %d\n", ctx_prio);
+   return DRM_SCHED_PRIORITY_NORMAL;
+   }
+
+}
+
 static int amdgpu_ctx_priority_permit(struct drm_file *filp,
- enum drm_sched_priority priority)
+ int32_t priority)
 {
-   if (priority < 0 || priority >= DRM_SCHED_PRIORITY_COUNT)
+   if (!amdgpu_ctx_priority_is_valid(priority))
return -EINVAL;
 
/* NORMAL and below are accessible by everyone */
-   if (priority <= DRM_SCHED_PRIORITY_NORMAL)
+   if (priority <= AMDGPU_CTX_PRIORITY_NORMAL)
return 0;
 
if (capable(CAP_SYS_NICE))
@@ -62,26 +109,35 @@ static int amdgpu_ctx_priority_permit(struct drm_file 
*filp,
return -EACCES;
 }
 
-static enum gfx_pipe_priority amdgpu_ctx_sched_prio_to_compute_prio(enum 
drm_sched_priority prio)
+static enum gfx_pipe_priority amdgpu_ctx_prio_to_compute_prio(int32_t prio)
 {
switch (prio) {
-   case DRM_SCHED_PRIORITY_HIGH:
-   case DRM_SCHED_PRIORITY_KERNEL:
+   case AMDGPU_CTX_PRIORITY_HIGH:
+   case AMDGPU_CTX_PRIORITY_VERY_HIGH:
return AMDGPU_GFX_PIPE_PRIO_HIGH;
default:
return AMDGPU_GFX_PIPE_PRIO_NORMAL;
}
 }
 
-static unsigned int amdgpu_ctx_prio_sched_to_hw(struct amdgpu_device *adev,
-enum drm_sched_priority prio,
-u32 hw_ip)
+static unsigned int amdgpu_ctx_get_hw_prio(struct amdgpu_ctx *ctx, u32 hw_ip)
 {
+   struct amdgpu_device *adev = ctx->adev;
+   int32_t ctx_prio;
unsigned int hw_prio;
 
-   hw_prio = (hw_ip == AMDGPU_HW_IP_COMPUTE) ?
-   amdgpu_ctx_sched_prio_to_compute_prio(prio) :
-   AMDGPU_RING_PRIO_DEFAULT;
+   ctx_prio = (ctx->override_priority == AMDGPU_CTX_PRIORITY_UNSET) ?
+   ctx->init_priority : ctx->override_priority;
+
+   switch (hw_ip) {
+   case AMDGPU_HW_IP_COMPUTE:
+   hw_prio = amdgpu_ctx_prio_to_compute_prio(ctx_prio);
+   break;
+   default:
+   hw_prio = AMDGPU_RING_PRIO_DEFAULT;
+   break;
+   }
+
hw_ip = array_index_nospec(hw_ip, AMDGPU_HW_IP_NUM);
if (adev->gpu_sched[hw_ip][hw_prio].num_scheds == 0)
hw_prio = AMDGPU_RING_PRIO_DEFAULT;
@@ -89,15 +145,17 @@ static unsigned int amdgpu_ctx_prio_sched_to_hw(struct 
amdgpu_device *adev,
return hw_prio;
 }
 
+
 static int amdgpu_ctx_init_entity(struct amdgpu_ctx *ctx, u32 hw_ip,
-   

[PATCH 2/2] drm/amdgpu: separate out vm pasid assignment

[Nirmoy Das]

Use new helper function amdgpu_vm_set_pasid() to
assign vm pasid value. This also ensures that we don't free
a pasid from vm code as pasids are allocated somewhere else.

Signed-off-by: Nirmoy Das 
Acked-by: Felix Kuehling 
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c  | 13 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c   | 10 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c| 28 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h|  4 +--
 4 files changed, 26 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index f96598279593..3a2ac7f66bbd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1287,11 +1287,22 @@ int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct 
kgd_dev *kgd,
if (avm->process_info)
return -EINVAL;
 
+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid.
+*/
+   if (avm->pasid) {
+   amdgpu_pasid_free(avm->pasid);
+   amdgpu_vm_set_pasid(adev, avm, 0);
+   }
+
/* Convert VM into a compute VM */
-   ret = amdgpu_vm_make_compute(adev, avm, pasid);
+   ret = amdgpu_vm_make_compute(adev, avm);
if (ret)
return ret;
 
+   ret = amdgpu_vm_set_pasid(adev, avm, pasid);
+   if (ret)
+   return ret;
/* Initialize KFD part of the VM and process info */
ret = init_kfd_vm(avm, process_info, ef);
if (ret)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index cbb932f97355..66bf8b0c56bb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -1178,10 +1178,14 @@ int amdgpu_driver_open_kms(struct drm_device *dev, 
struct drm_file *file_priv)
pasid = 0;
}
 
-   r = amdgpu_vm_init(adev, >vm, pasid);
+   r = amdgpu_vm_init(adev, >vm);
if (r)
goto error_pasid;
 
+   r = amdgpu_vm_set_pasid(adev, >vm, pasid);
+   if (r)
+   goto error_vm;
+
fpriv->prt_va = amdgpu_vm_bo_add(adev, >vm, NULL);
if (!fpriv->prt_va) {
r = -ENOMEM;
@@ -1209,8 +1213,10 @@ int amdgpu_driver_open_kms(struct drm_device *dev, 
struct drm_file *file_priv)
amdgpu_vm_fini(adev, >vm);
 
 error_pasid:
-   if (pasid)
+   if (pasid) {
amdgpu_pasid_free(pasid);
+   amdgpu_vm_set_pasid(adev, >vm, 0);
+   }
 
kfree(fpriv);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 0f82df5bfa7a..565c8c59c995 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2899,14 +2899,13 @@ long amdgpu_vm_wait_idle(struct amdgpu_vm *vm, long 
timeout)
  *
  * @adev: amdgpu_device pointer
  * @vm: requested vm
- * @pasid: Process address space identifier
  *
  * Init @vm fields.
  *
  * Returns:
  * 0 for success, error for failure.
  */
-int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, u32 pasid)
+int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
struct amdgpu_bo *root_bo;
struct amdgpu_bo_vm *root;
@@ -2980,10 +2979,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   r = amdgpu_vm_set_pasid(adev, vm, pasid);
-   if (r)
-   goto error_free_root;
-
INIT_KFIFO(vm->faults);
 
return 0;
@@ -3039,7 +3034,6 @@ static int amdgpu_vm_check_clean_reserved(struct 
amdgpu_device *adev,
  *
  * @adev: amdgpu_device pointer
  * @vm: requested vm
- * @pasid: pasid to use
  *
  * This only works on GFX VMs that don't have any BOs added and no
  * page tables allocated yet.
@@ -3047,7 +3041,6 @@ static int amdgpu_vm_check_clean_reserved(struct 
amdgpu_device *adev,
  * Changes the following VM parameters:
  * - use_cpu_for_update
  * - pte_supports_ats
- * - pasid (old PASID is released, because compute manages its own PASIDs)
  *
  * Reinitializes the page directory to reflect the changed ATS
  * setting.
@@ -3055,8 +3048,7 @@ static int amdgpu_vm_check_clean_reserved(struct 
amdgpu_device *adev,
  * Returns:
  * 0 for success, -errno for errors.
  */
-int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-  u32 pasid)
+int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
bool pte_support_ats = (adev->asic_type == CHIP_RAVEN);
int r;
@@ -3070,16 +3062,6 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (r)
goto unreserve_bo;
 
-   /* Free the original amdgpu allocated pasid,
-   

[PATCH 1/2] drm/amdgpu: use xarray for storing pasid in vm

[Nirmoy Das]

Replace idr with xarray as we actually need hash functionality.
Cleanup code related to vm pasid by adding helper function.

Signed-off-by: Nirmoy Das 
Acked-by: Felix Kuehling 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 149 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |   6 +-
 2 files changed, 73 insertions(+), 82 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63975bda8e76..0f82df5bfa7a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -87,6 +87,46 @@ struct amdgpu_prt_cb {
struct dma_fence_cb cb;
 };
 
+/**
+ * amdgpu_vm_set_pasid - manage pasid and vm ptr mapping
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: amdgpu_vm pointer
+ * @pasid: the pasid the VM is using on this GPU
+ *
+ * Set the pasid this VM is using on this GPU, can also be used to remove the
+ * pasid by passing in zero.
+ *
+ */
+int amdgpu_vm_set_pasid(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+   u32 pasid)
+{
+   int r;
+
+   if (vm->pasid == pasid)
+   return 0;
+
+   if (vm->pasid) {
+   r = xa_err(xa_erase_irq(>vm_manager.pasids, vm->pasid));
+   if (r < 0)
+   return r;
+
+   vm->pasid = 0;
+   }
+
+   if (pasid) {
+   r = xa_err(xa_store_irq(>vm_manager.pasids, pasid, vm,
+   GFP_KERNEL));
+   if (r < 0)
+   return r;
+
+   vm->pasid = pasid;
+   }
+
+
+   return 0;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -2940,18 +2980,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   if (r < 0)
-   goto error_free_root;
-
-   vm->pasid = pasid;
-   }
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto error_free_root;
 
INIT_KFIFO(vm->faults);
 
@@ -3039,18 +3070,15 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (r)
goto unreserve_bo;
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid.
+*/
+   if (vm->pasid)
+   amdgpu_pasid_free(vm->pasid);
 
-   if (r == -ENOSPC)
-   goto unreserve_bo;
-   r = 0;
-   }
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto unreserve_bo;
 
/* Check if PD needs to be reinitialized and do it before
 * changing any other state, in case it fails.
@@ -3061,7 +3089,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
   to_amdgpu_bo_vm(vm->root.bo),
   false);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
}
 
/* Update VM state */
@@ -3078,7 +3106,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = amdgpu_bo_sync_wait(vm->root.bo,
AMDGPU_FENCE_OWNER_UNDEFINED, true);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
 
vm->update_funcs = _vm_cpu_funcs;
} else {
@@ -3088,36 +3116,13 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
vm->last_update = NULL;
vm->is_compute_context = true;
 
-   if (vm->pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, vm->pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
-   /* Free the original amdgpu allocated pasid
-* Will be replaced with kfd allocated pasid
-*/
-   amdgpu_pasid_free(vm->

[PATCH 2/2] drm/amdgpu: separate out vm pasid assignment

[Nirmoy Das]

Use new helper function amdgpu_vm_set_pasid() to
assign vm pasid value. This also ensures that we don't free
a pasid from vm code as pasids are allocated somewhere else.

Signed-off-by: Nirmoy Das 
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c  | 13 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c   | 10 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c| 28 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h|  4 +--
 4 files changed, 26 insertions(+), 29 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index f96598279593..3a2ac7f66bbd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1287,11 +1287,22 @@ int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct 
kgd_dev *kgd,
if (avm->process_info)
return -EINVAL;
 
+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid.
+*/
+   if (avm->pasid) {
+   amdgpu_pasid_free(avm->pasid);
+   amdgpu_vm_set_pasid(adev, avm, 0);
+   }
+
/* Convert VM into a compute VM */
-   ret = amdgpu_vm_make_compute(adev, avm, pasid);
+   ret = amdgpu_vm_make_compute(adev, avm);
if (ret)
return ret;
 
+   ret = amdgpu_vm_set_pasid(adev, avm, pasid);
+   if (ret)
+   return ret;
/* Initialize KFD part of the VM and process info */
ret = init_kfd_vm(avm, process_info, ef);
if (ret)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index cbb932f97355..66bf8b0c56bb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -1178,10 +1178,14 @@ int amdgpu_driver_open_kms(struct drm_device *dev, 
struct drm_file *file_priv)
pasid = 0;
}
 
-   r = amdgpu_vm_init(adev, >vm, pasid);
+   r = amdgpu_vm_init(adev, >vm);
if (r)
goto error_pasid;
 
+   r = amdgpu_vm_set_pasid(adev, >vm, pasid);
+   if (r)
+   goto error_vm;
+
fpriv->prt_va = amdgpu_vm_bo_add(adev, >vm, NULL);
if (!fpriv->prt_va) {
r = -ENOMEM;
@@ -1209,8 +1213,10 @@ int amdgpu_driver_open_kms(struct drm_device *dev, 
struct drm_file *file_priv)
amdgpu_vm_fini(adev, >vm);
 
 error_pasid:
-   if (pasid)
+   if (pasid) {
amdgpu_pasid_free(pasid);
+   amdgpu_vm_set_pasid(adev, >vm, 0);
+   }
 
kfree(fpriv);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index fd92ff27931a..1dccede6387e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2899,14 +2899,13 @@ long amdgpu_vm_wait_idle(struct amdgpu_vm *vm, long 
timeout)
  *
  * @adev: amdgpu_device pointer
  * @vm: requested vm
- * @pasid: Process address space identifier
  *
  * Init @vm fields.
  *
  * Returns:
  * 0 for success, error for failure.
  */
-int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, u32 pasid)
+int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
struct amdgpu_bo *root_bo;
struct amdgpu_bo_vm *root;
@@ -2980,10 +2979,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   r = amdgpu_vm_set_pasid(adev, vm, pasid);
-   if (r)
-   goto error_free_root;
-
INIT_KFIFO(vm->faults);
 
return 0;
@@ -3039,7 +3034,6 @@ static int amdgpu_vm_check_clean_reserved(struct 
amdgpu_device *adev,
  *
  * @adev: amdgpu_device pointer
  * @vm: requested vm
- * @pasid: pasid to use
  *
  * This only works on GFX VMs that don't have any BOs added and no
  * page tables allocated yet.
@@ -3047,7 +3041,6 @@ static int amdgpu_vm_check_clean_reserved(struct 
amdgpu_device *adev,
  * Changes the following VM parameters:
  * - use_cpu_for_update
  * - pte_supports_ats
- * - pasid (old PASID is released, because compute manages its own PASIDs)
  *
  * Reinitializes the page directory to reflect the changed ATS
  * setting.
@@ -3055,8 +3048,7 @@ static int amdgpu_vm_check_clean_reserved(struct 
amdgpu_device *adev,
  * Returns:
  * 0 for success, -errno for errors.
  */
-int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-  u32 pasid)
+int amdgpu_vm_make_compute(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
bool pte_support_ats = (adev->asic_type == CHIP_RAVEN);
int r;
@@ -3070,16 +3062,6 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (r)
goto unreserve_bo;
 
-   /* Free the original amdgpu allocated pasid,
-*

[PATCH 1/2] drm/amdgpu: use xarray for storing pasid in vm

[Nirmoy Das]

Replace idr with xarray as we actually need hash functionality.
Cleanup code related to vm pasid by adding helper function.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 149 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |   6 +-
 2 files changed, 73 insertions(+), 82 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63975bda8e76..fd92ff27931a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -87,6 +87,46 @@ struct amdgpu_prt_cb {
struct dma_fence_cb cb;
 };
 
+/**
+ * amdgpu_vm_set_pasid - manage pasid and vm ptr mapping
+ *
+ * @adev: amdgpu_device pointer
+ * @vm: amdgpu_vm pointer
+ * @pasid: requested pasid
+ *
+ * Each pasid associats with a vm pointer. This function can be use to
+ * create a new pasid,vm association or to remove an existing one. To remove an
+ * existing pasid,vm association, pass 0 as @pasid.
+ */
+int amdgpu_vm_set_pasid(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+   unsigned long pasid)
+{
+   int r;
+
+   if (vm->pasid == pasid)
+   return 0;
+
+   if (vm->pasid) {
+   r = xa_err(xa_erase_irq(>vm_manager.pasids, vm->pasid));
+   if (r < 0)
+   return r;
+
+   vm->pasid = 0;
+   }
+
+   if (pasid) {
+   r = xa_err(xa_store_irq(>vm_manager.pasids, pasid, vm,
+   GFP_KERNEL));
+   if (r < 0)
+   return r;
+
+   vm->pasid = pasid;
+   }
+
+
+   return 0;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -2940,18 +2980,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   if (r < 0)
-   goto error_free_root;
-
-   vm->pasid = pasid;
-   }
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto error_free_root;
 
INIT_KFIFO(vm->faults);
 
@@ -3039,18 +3070,15 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (r)
goto unreserve_bo;
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid.
+*/
+   if (vm->pasid)
+   amdgpu_pasid_free(vm->pasid);
 
-   if (r == -ENOSPC)
-   goto unreserve_bo;
-   r = 0;
-   }
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto unreserve_bo;
 
/* Check if PD needs to be reinitialized and do it before
 * changing any other state, in case it fails.
@@ -3061,7 +3089,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
   to_amdgpu_bo_vm(vm->root.bo),
   false);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
}
 
/* Update VM state */
@@ -3078,7 +3106,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = amdgpu_bo_sync_wait(vm->root.bo,
AMDGPU_FENCE_OWNER_UNDEFINED, true);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
 
vm->update_funcs = _vm_cpu_funcs;
} else {
@@ -3088,36 +3116,13 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
vm->last_update = NULL;
vm->is_compute_context = true;
 
-   if (vm->pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, vm->pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
-   /* Free the original amdgpu allocated pasid
-* Will be replaced with kfd allocated pasid
- 

[PATCH v3 1/1] drm/amdgpu: return early for non-TTM_PL_TT type BOs

[Nirmoy Das]

Return early for non-TTM_PL_TT BOs so that we don't pass
wrong pointer to amdgpu_gtt_mgr_has_gart_addr() which assumes
ttm_resource argument to be TTM_PL_TT type BO's.

v3: remove extra braces.
v2: merge if-conditions.

Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index b46726e47bce..7a82e7b9e18e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -926,7 +926,8 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
bo_mem->mem_type == AMDGPU_PL_OA)
return -EINVAL;

-   if (!amdgpu_gtt_mgr_has_gart_addr(bo_mem)) {
+   if (bo_mem->mem_type != TTM_PL_TT ||
+   !amdgpu_gtt_mgr_has_gart_addr(bo_mem)) {
gtt->offset = AMDGPU_BO_INVALID_OFFSET;
return 0;
}
--
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 2/2] drm/amdgpu: free pasid early before converting a vm

[Nirmoy Das]

VM code should not be responsible for freeing pasid as pasid
gets allocated outside of VM code, before initializing a vm.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 8 
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 6 --
 2 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index f96598279593..5a9006aabec7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1287,6 +1287,14 @@ int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct 
kgd_dev *kgd,
if (avm->process_info)
return -EINVAL;

+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid.
+*/
+   if (avm->pasid) {
+   amdgpu_pasid_free(avm->pasid);
+   amdgpu_vm_set_pasid(adev, avm, 0);
+   }
+
/* Convert VM into a compute VM */
ret = amdgpu_vm_make_compute(adev, avm, pasid);
if (ret)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 9b0e8a9d7f86..0bd9575b0872 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -3057,12 +3057,6 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (r)
goto unreserve_bo;

-   /* Free the original amdgpu allocated pasid,
-* will be replaced with kfd allocated pasid.
-*/
-   if (vm->pasid)
-   amdgpu_pasid_free(vm->pasid);
-
r = amdgpu_vm_set_pasid(adev, vm, pasid);
if (r)
goto unreserve_bo;
--
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/2] drm/amdgpu: use xarray for storing pasid in vm

[Nirmoy Das]

Replace idr with xarray as we actually need hash functionality.
Cleanup code related to vm pasid by adding helper function.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 136 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |   6 +-
 2 files changed, 60 insertions(+), 82 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63975bda8e76..9b0e8a9d7f86 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -87,6 +87,33 @@ struct amdgpu_prt_cb {
struct dma_fence_cb cb;
 };
 
+int amdgpu_vm_set_pasid(struct amdgpu_device *adev, struct amdgpu_vm *vm,
+   unsigned long pasid)
+{
+   unsigned long flags;
+   int r;
+
+   if (vm->pasid == pasid)
+   return 0;
+
+   if (vm->pasid) {
+   r = xa_err(xa_erase_irq(>vm_manager.pasids, vm->pasid));
+   if (r < 0)
+   return r;
+   }
+
+   if (pasid) {
+   r = xa_err(xa_store_irq(>vm_manager.pasids, pasid, vm,
+   GFP_KERNEL));
+   if (r < 0)
+   return r;
+   }
+
+   vm->pasid = pasid;
+
+   return 0;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -2940,18 +2967,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   if (r < 0)
-   goto error_free_root;
-
-   vm->pasid = pasid;
-   }
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto error_free_root;
 
INIT_KFIFO(vm->faults);
 
@@ -3039,18 +3057,15 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (r)
goto unreserve_bo;
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid.
+*/
+   if (vm->pasid)
+   amdgpu_pasid_free(vm->pasid);
 
-   if (r == -ENOSPC)
-   goto unreserve_bo;
-   r = 0;
-   }
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto unreserve_bo;
 
/* Check if PD needs to be reinitialized and do it before
 * changing any other state, in case it fails.
@@ -3061,7 +3076,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
   to_amdgpu_bo_vm(vm->root.bo),
   false);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
}
 
/* Update VM state */
@@ -3078,7 +3093,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = amdgpu_bo_sync_wait(vm->root.bo,
AMDGPU_FENCE_OWNER_UNDEFINED, true);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
 
vm->update_funcs = _vm_cpu_funcs;
} else {
@@ -3088,36 +3103,13 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
vm->last_update = NULL;
vm->is_compute_context = true;
 
-   if (vm->pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, vm->pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
-   /* Free the original amdgpu allocated pasid
-* Will be replaced with kfd allocated pasid
-*/
-   amdgpu_pasid_free(vm->pasid);
-   vm->pasid = 0;
-   }
-
/* Free the shadow bo for compute VM */
amdgpu_bo_unref(_amdgpu_bo_vm(vm->root.bo)->shadow);
 
-   if (pasid)
-   vm->pasid = pasid;
-
goto unreserve_bo;
 
-free_idr:
-   if (pasid) {
-   unsigned long flags;
-
-

[PATCH 2/2] drm/amdgpu: free pasid early before converting a vm

[Nirmoy Das]

VM code should not be responsible for freeing pasid as pasid
gets allocated outside of VM code, before initializing a vm.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 8 
 1 file changed, 8 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index f96598279593..5a9006aabec7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1287,6 +1287,14 @@ int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct 
kgd_dev *kgd,
if (avm->process_info)
return -EINVAL;
 
+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid.
+*/
+   if (avm->pasid) {
+   amdgpu_pasid_free(avm->pasid);
+   amdgpu_vm_set_pasid(adev, avm, 0);
+   }
+
/* Convert VM into a compute VM */
ret = amdgpu_vm_make_compute(adev, avm, pasid);
if (ret)
-- 
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 1/2] drm/amdgpu: return early for non-TTM_PL_TT type BOs

[Nirmoy Das]

Return early for non-TTM_PL_TT BOs so that we don't pass
wrong pointer to amdgpu_gtt_mgr_has_gart_addr() which assumes
ttm_resource argument to be TTM_PL_TT type BO's.

v2: merge if-conditions

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index b46726e47bce..208bc3ec1aff 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -926,7 +926,8 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
bo_mem->mem_type == AMDGPU_PL_OA)
return -EINVAL;

-   if (!amdgpu_gtt_mgr_has_gart_addr(bo_mem)) {
+   if ((bo_mem->mem_type != TTM_PL_TT) ||
+   !amdgpu_gtt_mgr_has_gart_addr(bo_mem)) {
gtt->offset = AMDGPU_BO_INVALID_OFFSET;
return 0;
}
--
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/2] drm/amdgpu: raise error on incorrect mem_type

[Nirmoy Das]

Be more defensive and raise error on wrong mem_type
argument in amdgpu_gtt_mgr_has_gart_addr().

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 6 +-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 543000304a1c..0b0fa87b115c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -107,8 +107,12 @@ const struct attribute_group amdgpu_gtt_mgr_attr_group = {
  */
 bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_resource *res)
 {
-   struct amdgpu_gtt_node *node = to_amdgpu_gtt_node(res);
+   struct amdgpu_gtt_node *node;
+
+   if (WARN_ON(res->mem_type != TTM_PL_TT))
+   return false;
 
+   node = to_amdgpu_gtt_node(res);
return drm_mm_node_allocated(>base.mm_nodes[0]);
 }
 
-- 
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/2] drm/amdgpu: raise error on incorrect mem_type

[Nirmoy Das]

Be more defensive and raise error on wrong mem_type
argument in amdgpu_gtt_mgr_has_gart_addr().

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 6 +-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 543000304a1c..0b0fa87b115c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -107,8 +107,12 @@ const struct attribute_group amdgpu_gtt_mgr_attr_group = {
  */
 bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_resource *res)
 {
-   struct amdgpu_gtt_node *node = to_amdgpu_gtt_node(res);
+   struct amdgpu_gtt_node *node;
+
+   if (WARN_ON(res->mem_type != TTM_PL_TT))
+   return false;
 
+   node = to_amdgpu_gtt_node(res);
return drm_mm_node_allocated(>base.mm_nodes[0]);
 }
 
-- 
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/2] drm/amdgpu: return early for preempt type BOs

[Nirmoy Das]

Return early for AMDGPU_PL_PREEMPT BOs so that we don't pass
wrong pointer to amdgpu_gtt_mgr_has_gart_addr() which assumes
ttm_resource argument to be TTM_PL_TT type BO's.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 5 +
 1 file changed, 5 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index b46726e47bce..3df06772a425 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -926,6 +926,11 @@ static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
bo_mem->mem_type == AMDGPU_PL_OA)
return -EINVAL;
 
+   if (bo_mem->mem_type == AMDGPU_PL_PREEMPT) {
+   gtt->offset = AMDGPU_BO_INVALID_OFFSET;
+   return 0;
+   }
+
if (!amdgpu_gtt_mgr_has_gart_addr(bo_mem)) {
gtt->offset = AMDGPU_BO_INVALID_OFFSET;
return 0;
-- 
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/2] drm/amdgpu: use xarray for storing pasid in vm

[Nirmoy Das]

Replace idr with xarray as we actually need hash functionality.
Cleanup code related to vm pasid by adding helper function.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 134 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |   3 +-
 2 files changed, 60 insertions(+), 77 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index be841d62a1d4..e047e56a4be2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -87,6 +87,39 @@ struct amdgpu_prt_cb {
struct dma_fence_cb cb;
 };
 
+static int amdgpu_vm_set_pasid(struct amdgpu_device *adev,
+  struct amdgpu_vm *vm,
+  unsigned long pasid)
+{
+   unsigned long flags;
+   int r;
+
+   if (pasid) {
+   xa_lock_irqsave(>vm_manager.pasids, flags);
+   r = xa_err(__xa_store(>vm_manager.pasids, pasid, vm,
+ GFP_ATOMIC));
+   xa_unlock_irqrestore(>vm_manager.pasids, flags);
+   if (r < 0)
+   return r;
+   } else {
+   unsigned long index;
+   struct amdgpu_vm *res;
+
+   xa_lock_irqsave(>vm_manager.pasids, flags);
+   xa_for_each(>vm_manager.pasids, index, res) {
+   if (res == vm) {
+   __xa_erase(>vm_manager.pasids, index);
+   break;
+   }
+   }
+   xa_unlock_irqrestore(>vm_manager.pasids, flags);
+   }
+
+   vm->pasid = pasid;
+
+   return 0;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -2940,18 +2973,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   if (r < 0)
-   goto error_free_root;
-
-   vm->pasid = pasid;
-   }
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto error_free_root;
 
INIT_KFIFO(vm->faults);
 
@@ -3039,18 +3063,11 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (r)
goto unreserve_bo;
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
-   if (r == -ENOSPC)
-   goto unreserve_bo;
-   r = 0;
-   }
+   /* remove previous {pasid:vm} entry first */
+   r = amdgpu_vm_set_pasid(adev, vm, 0);
+   r = amdgpu_vm_set_pasid(adev, vm, pasid);
+   if (r)
+   goto unreserve_bo;
 
/* Check if PD needs to be reinitialized and do it before
 * changing any other state, in case it fails.
@@ -3061,7 +3078,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
   to_amdgpu_bo_vm(vm->root.bo),
   false);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
}
 
/* Update VM state */
@@ -3078,7 +3095,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = amdgpu_bo_sync_wait(vm->root.bo,
AMDGPU_FENCE_OWNER_UNDEFINED, true);
if (r)
-   goto free_idr;
+   goto free_pasid_entry;
 
vm->update_funcs = _vm_cpu_funcs;
} else {
@@ -3088,31 +3105,14 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
vm->last_update = NULL;
vm->is_compute_context = true;
 
-   if (vm->pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, vm->pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   vm->pasid = 0;
-   }
-
/* Free the shadow bo for compute VM */
amdgpu_bo_unref(_amdgpu_bo_vm(vm->root.bo)->shadow);
 
-   if (pasid

[PATCH 1/2] drm/amdgpu: free pasid early before converting a vm

[Nirmoy Das]

VM code should not be responsible for freeing pasid as pasid
gets allocated outside of VM code, before initializing a vm.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 6 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 5 -
 2 files changed, 6 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index f96598279593..42e22b1fdfee 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1287,6 +1287,12 @@ int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct 
kgd_dev *kgd,
if (avm->process_info)
return -EINVAL;
 
+   /* Free the original amdgpu allocated pasid,
+* will be replaced with kfd allocated pasid
+*/
+   if (avm->pasid)
+   amdgpu_pasid_free(avm->pasid);
+
/* Convert VM into a compute VM */
ret = amdgpu_vm_make_compute(adev, avm, pasid);
if (ret)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63975bda8e76..be841d62a1d4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -3094,11 +3094,6 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
spin_lock_irqsave(>vm_manager.pasid_lock, flags);
idr_remove(>vm_manager.pasid_idr, vm->pasid);
spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
-   /* Free the original amdgpu allocated pasid
-* Will be replaced with kfd allocated pasid
-*/
-   amdgpu_pasid_free(vm->pasid);
vm->pasid = 0;
}
 
-- 
2.32.0

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amdgpu: cleanup pasid handling

[Nirmoy Das]

Cleanup code related to vm pasid by adding helper functions.
Also replace idr with xarray as we actually need hash functionality.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 130 -
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |   3 +-
 2 files changed, 62 insertions(+), 71 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63975bda8e76..abba1e2de264 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -87,6 +87,45 @@ struct amdgpu_prt_cb {
struct dma_fence_cb cb;
 };
 
+static int amdgpu_vm_pasid_insert(struct amdgpu_device *adev,
+ struct amdgpu_vm *vm,
+ unsigned long pasid,
+ unsigned int *vm_pasid)
+{
+   unsigned long flags;
+   int r;
+
+   if (!pasid)
+   return 0;
+
+   xa_lock_irqsave(>vm_manager.pasids, flags);
+   r = xa_err(__xa_store(>vm_manager.pasids, pasid, vm, GFP_ATOMIC));
+   xa_unlock_irqrestore(>vm_manager.pasids, flags);
+   if (r < 0)
+   return r;
+   if (vm_pasid)
+   *vm_pasid = pasid;
+
+   return 0;
+}
+
+static void amdgpu_vm_pasid_remove(struct amdgpu_device *adev,
+  unsigned long pasid,
+  unsigned int *vm_pasid)
+{
+   unsigned long flags;
+
+   if (!pasid)
+   return;
+
+   xa_lock_irqsave(>vm_manager.pasids, flags);
+   __xa_erase(>vm_manager.pasids, pasid);
+   xa_unlock_irqrestore(>vm_manager.pasids, flags);
+
+   if (vm_pasid)
+   *vm_pasid = 0;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -2940,18 +2979,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   if (r < 0)
-   goto error_free_root;
-
-   vm->pasid = pasid;
-   }
+   r = amdgpu_vm_pasid_insert(adev, vm, pasid, >pasid);
+   if (r)
+   goto error_free_root;
 
INIT_KFIFO(vm->faults);
 
@@ -3038,19 +3068,9 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = amdgpu_vm_check_clean_reserved(adev, vm);
if (r)
goto unreserve_bo;
-
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
-   if (r == -ENOSPC)
-   goto unreserve_bo;
-   r = 0;
-   }
+   r = amdgpu_vm_pasid_insert(adev, vm, pasid, NULL);
+   if (r)
+   goto unreserve_bo;
 
/* Check if PD needs to be reinitialized and do it before
 * changing any other state, in case it fails.
@@ -3089,35 +3109,23 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
vm->is_compute_context = true;
 
if (vm->pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, vm->pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
/* Free the original amdgpu allocated pasid
 * Will be replaced with kfd allocated pasid
 */
amdgpu_pasid_free(vm->pasid);
-   vm->pasid = 0;
+   amdgpu_vm_pasid_remove(adev, vm->pasid, >pasid);
}
 
/* Free the shadow bo for compute VM */
amdgpu_bo_unref(_amdgpu_bo_vm(vm->root.bo)->shadow);
-
if (pasid)
vm->pasid = pasid;
 
goto unreserve_bo;
 
 free_idr:
-   if (pasid) {
-   unsigned long flags;
+   amdgpu_vm_pasid_remove(adev, pasid, NULL);
 
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   }
 unreserve_bo:
amdgpu_bo_unreserve(vm->root.bo);
return r;
@@ -3133,14 +3141,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
str

[PATCH 1/1] drm/amdgpu: add helper function for vm pasid

[Nirmoy Das]

Cleanup code related to vm pasid by adding helper functions.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 105 -
 1 file changed, 50 insertions(+), 55 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63975bda8e76..6e476b173cbb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -87,6 +87,46 @@ struct amdgpu_prt_cb {
struct dma_fence_cb cb;
 };

+static int amdgpu_vm_pasid_alloc(struct amdgpu_device *adev,
+struct amdgpu_vm *vm,
+unsigned int pasid,
+unsigned int *vm_pasid)
+{
+   unsigned long flags;
+   int r;
+
+   if (!pasid)
+   return 0;
+
+   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
+   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
+ GFP_ATOMIC);
+   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+   if (r < 0)
+   return r;
+   if (vm_pasid)
+   *vm_pasid = pasid;
+
+   return 0;
+}
+
+static void amdgpu_vm_pasid_remove(struct amdgpu_device *adev,
+  unsigned int pasid,
+  unsigned int *vm_pasid)
+{
+   unsigned long flags;
+
+   if (!pasid)
+   return;
+
+   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
+   idr_remove(>vm_manager.pasid_idr, pasid);
+   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+
+   if (vm_pasid)
+   *vm_pasid = 0;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -2940,18 +2980,8 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)

amdgpu_bo_unreserve(vm->root.bo);

-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   if (r < 0)
-   goto error_free_root;
-
-   vm->pasid = pasid;
-   }
+   if (amdgpu_vm_pasid_alloc(adev, vm, pasid, >pasid))
+   goto error_free_root;

INIT_KFIFO(vm->faults);

@@ -3038,19 +3068,11 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = amdgpu_vm_check_clean_reserved(adev, vm);
if (r)
goto unreserve_bo;
+   r = amdgpu_vm_pasid_alloc(adev, vm, pasid, NULL);
+   if (r ==  -ENOSPC)
+   goto unreserve_bo;

-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
- GFP_ATOMIC);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
-   if (r == -ENOSPC)
-   goto unreserve_bo;
-   r = 0;
-   }
+   r = 0;

/* Check if PD needs to be reinitialized and do it before
 * changing any other state, in case it fails.
@@ -3089,35 +3111,23 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
vm->is_compute_context = true;

if (vm->pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, vm->pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-
/* Free the original amdgpu allocated pasid
 * Will be replaced with kfd allocated pasid
 */
amdgpu_pasid_free(vm->pasid);
-   vm->pasid = 0;
+   amdgpu_vm_pasid_remove(adev, vm->pasid, >pasid);
}

/* Free the shadow bo for compute VM */
amdgpu_bo_unref(_amdgpu_bo_vm(vm->root.bo)->shadow);
-
if (pasid)
vm->pasid = pasid;

goto unreserve_bo;

 free_idr:
-   if (pasid) {
-   unsigned long flags;
+   amdgpu_vm_pasid_remove(adev, pasid, NULL);

-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   idr_remove(>vm_manager.pasid_idr, pasid);
-   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
-   }
 unreserve_bo:
amdgpu_bo_unreserve(vm->root.bo);
return r;
@@ -3133,14 +3143,7 @@ int amdgpu_vm_make_compute(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
  */
 void amdgpu_vm_release_compute(struct am

[PATCH 1/1] drm/amdgpu: add helper function for vm pasid

[Nirmoy Das]

Cleanup code related to vm pasid by adding helper function.
This reduces lots code duplication.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c |  17 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 176 
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  |   2 +-
 3 files changed, 96 insertions(+), 99 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index cbb932f97355..27851fb0e25b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -1149,7 +1149,7 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct 
drm_file *file_priv)
 {
struct amdgpu_device *adev = drm_to_adev(dev);
struct amdgpu_fpriv *fpriv;
-   int r, pasid;
+   int r;
 
/* Ensure IB tests are run on ring */
flush_delayed_work(>delayed_init_work);
@@ -1172,15 +1172,9 @@ int amdgpu_driver_open_kms(struct drm_device *dev, 
struct drm_file *file_priv)
goto out_suspend;
}
 
-   pasid = amdgpu_pasid_alloc(16);
-   if (pasid < 0) {
-   dev_warn(adev->dev, "No more PASIDs available!");
-   pasid = 0;
-   }
-
-   r = amdgpu_vm_init(adev, >vm, pasid);
+   r = amdgpu_vm_init(adev, >vm);
if (r)
-   goto error_pasid;
+   goto free_fpriv;
 
fpriv->prt_va = amdgpu_vm_bo_add(adev, >vm, NULL);
if (!fpriv->prt_va) {
@@ -1208,10 +1202,7 @@ int amdgpu_driver_open_kms(struct drm_device *dev, 
struct drm_file *file_priv)
 error_vm:
amdgpu_vm_fini(adev, >vm);
 
-error_pasid:
-   if (pasid)
-   amdgpu_pasid_free(pasid);
-
+free_fpriv:
kfree(fpriv);
 
 out_suspend:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 63975bda8e76..562c2c48a3a3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -87,6 +87,69 @@ struct amdgpu_prt_cb {
struct dma_fence_cb cb;
 };
 
+static int amdgpu_vm_pasid_alloc(struct amdgpu_device *adev,
+struct amdgpu_vm *vm, unsigned int pasid)
+{
+   unsigned long flags;
+   int r;
+
+   if (!pasid)
+   return 0;
+
+   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
+   r = idr_alloc(>vm_manager.pasid_idr, vm, pasid, pasid + 1,
+ GFP_ATOMIC);
+   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+   if (r < 0)
+   return r;
+
+   vm->pasid = pasid;
+   return 0;
+}
+static void amdgpu_vm_pasid_remove_id(struct amdgpu_device *adev,
+ unsigned int pasid)
+{
+   unsigned long flags;
+
+   if (!pasid)
+   return;
+
+   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
+   idr_remove(>vm_manager.pasid_idr, pasid);
+   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+
+}
+
+static void amdgpu_vm_pasid_remove(struct amdgpu_device *adev,
+  struct amdgpu_vm *vm)
+{
+   amdgpu_vm_pasid_remove_id(adev, vm->pasid);
+   vm->pasid = 0;
+}
+
+static void amdgpu_vm_pasid_free(struct amdgpu_device *adev,
+struct amdgpu_vm *vm)
+{
+   if (!vm->pasid)
+   return;
+
+   amdgpu_pasid_free(vm->pasid);
+   amdgpu_vm_pasid_remove(adev, vm);
+}
+
+static struct amdgpu_vm *amdgpu_vm_pasid_find(struct amdgpu_device *adev,
+ unsigned int pasid)
+{
+   struct amdgpu_vm *vm;
+   unsigned long flags;
+
+   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
+   vm = idr_find(>vm_manager.pasid_idr, pasid);
+   spin_unlock_irqrestore(>vm_manager.pasid_lock, flags);
+
+   return vm;
+}
+
 /*
  * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
  * happens while holding this lock anywhere to prevent deadlocks when
@@ -2859,17 +2922,17 @@ long amdgpu_vm_wait_idle(struct amdgpu_vm *vm, long 
timeout)
  *
  * @adev: amdgpu_device pointer
  * @vm: requested vm
- * @pasid: Process address space identifier
  *
  * Init @vm fields.
  *
  * Returns:
  * 0 for success, error for failure.
  */
-int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, u32 pasid)
+int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 {
struct amdgpu_bo *root_bo;
struct amdgpu_bo_vm *root;
+   unsigned int pasid;
int r, i;
 
vm->va = RB_ROOT_CACHED;
@@ -2940,19 +3003,15 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct 
amdgpu_vm *vm, u32 pasid)
 
amdgpu_bo_unreserve(vm->root.bo);
 
-   if (pasid) {
-   unsigned long flags;
-
-   spin_lock_irqsave(>vm_manager.pasid_lock, flags);
-   r = idr_

[PATCH v3 1/2] drm/amdgpu: parameterize ttm BO destroy callback

[Nirmoy Das]

Make provision to pass different ttm BO destroy callback
while creating a amdgpu_bo.

v3: remove unnecessary amdgpu_bo_destroy_base.
v2: remove whitespace.
call amdgpu_bo_destroy_base() at the end for cleaner code.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 41 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  3 +-
 2 files changed, 32 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 9092ac12a270..ea54fd739c41 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -75,9 +75,7 @@ static void amdgpu_bo_subtract_pin_size(struct amdgpu_bo *bo)

 static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
 {
-   struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
-   struct amdgpu_bo_user *ubo;

if (bo->tbo.pin_count > 0)
amdgpu_bo_subtract_pin_size(bo);
@@ -87,20 +85,33 @@ static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
if (bo->tbo.base.import_attach)
drm_prime_gem_destroy(>tbo.base, bo->tbo.sg);
drm_gem_object_release(>tbo.base);
+   amdgpu_bo_unref(>parent);
+   kvfree(bo);
+}
+
+static void amdgpu_bo_user_destroy(struct ttm_buffer_object *tbo)
+{
+   struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+   struct amdgpu_bo_user *ubo;
+
+   ubo = to_amdgpu_bo_user(bo);
+   kfree(ubo->metadata);
+   amdgpu_bo_destroy(tbo);
+}
+
+static void amdgpu_bo_vm_destroy(struct ttm_buffer_object *tbo)
+{
+   struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
+   struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+
/* in case amdgpu_device_recover_vram got NULL of bo->parent */
if (!list_empty(>shadow_list)) {
mutex_lock(>shadow_list_lock);
list_del_init(>shadow_list);
mutex_unlock(>shadow_list_lock);
}
-   amdgpu_bo_unref(>parent);
-
-   if (bo->tbo.type != ttm_bo_type_kernel) {
-   ubo = to_amdgpu_bo_user(bo);
-   kfree(ubo->metadata);
-   }

-   kvfree(bo);
+   amdgpu_bo_destroy(tbo);
 }

 /**
@@ -115,8 +126,11 @@ static void amdgpu_bo_destroy(struct ttm_buffer_object 
*tbo)
  */
 bool amdgpu_bo_is_amdgpu_bo(struct ttm_buffer_object *bo)
 {
-   if (bo->destroy == _bo_destroy)
+   if (bo->destroy == _bo_destroy ||
+   bo->destroy == _bo_user_destroy ||
+   bo->destroy == _bo_vm_destroy)
return true;
+
return false;
 }

@@ -592,9 +606,12 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
if (bp->type == ttm_bo_type_kernel)
bo->tbo.priority = 1;

+   if (!bp->destroy)
+   bp->destroy = _bo_destroy;
+
r = ttm_bo_init_reserved(>mman.bdev, >tbo, size, bp->type,
 >placement, page_align, ,  NULL,
-bp->resv, _bo_destroy);
+bp->resv, bp->destroy);
if (unlikely(r != 0))
return r;

@@ -658,6 +675,7 @@ int amdgpu_bo_create_user(struct amdgpu_device *adev,
int r;

bp->bo_ptr_size = sizeof(struct amdgpu_bo_user);
+   bp->destroy = _bo_user_destroy;
r = amdgpu_bo_create(adev, bp, _ptr);
if (r)
return r;
@@ -689,6 +707,7 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
 * num of amdgpu_vm_pt entries.
 */
BUG_ON(bp->bo_ptr_size < sizeof(struct amdgpu_bo_vm));
+   bp->destroy = _bo_vm_destroy;
r = amdgpu_bo_create(adev, bp, _ptr);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index e36b84516b4e..a8c702634e1b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -55,7 +55,8 @@ struct amdgpu_bo_param {
u64 flags;
enum ttm_bo_typetype;
boolno_wait_gpu;
-   struct dma_resv *resv;
+   struct dma_resv *resv;
+   void(*destroy)(struct ttm_buffer_object 
*bo);
 };

 /* bo virtual addresses in a vm */
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/2] drm/amdgpu: move shadow_list to amdgpu_bo_vm

[Nirmoy Das]

Move shadow_list to struct amdgpu_bo_vm as shadow BOs
are part of PT/PD BOs.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  5 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 14 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  6 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c |  2 +-
 4 files changed, 14 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index f2636f411a25..3f51b142fc83 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4063,6 +4063,7 @@ static int amdgpu_device_recover_vram(struct 
amdgpu_device *adev)
 {
struct dma_fence *fence = NULL, *next = NULL;
struct amdgpu_bo *shadow;
+   struct amdgpu_bo_vm *vmbo;
long r = 1, tmo;
 
if (amdgpu_sriov_runtime(adev))
@@ -4072,8 +4073,8 @@ static int amdgpu_device_recover_vram(struct 
amdgpu_device *adev)
 
dev_info(adev->dev, "recover vram bo from shadow start\n");
mutex_lock(>shadow_list_lock);
-   list_for_each_entry(shadow, >shadow_list, shadow_list) {
-
+   list_for_each_entry(vmbo, >shadow_list, shadow_list) {
+   shadow = >bo;
/* No need to recover an evicted BO */
if (shadow->tbo.mem.mem_type != TTM_PL_TT ||
shadow->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET ||
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index ea54fd739c41..ea339eaac399 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -103,11 +103,13 @@ static void amdgpu_bo_vm_destroy(struct ttm_buffer_object 
*tbo)
 {
struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+   struct amdgpu_bo_vm *vmbo;
 
+   vmbo = to_amdgpu_bo_vm(bo);
/* in case amdgpu_device_recover_vram got NULL of bo->parent */
-   if (!list_empty(>shadow_list)) {
+   if (!list_empty(>shadow_list)) {
mutex_lock(>shadow_list_lock);
-   list_del_init(>shadow_list);
+   list_del_init(>shadow_list);
mutex_unlock(>shadow_list_lock);
}
 
@@ -583,7 +585,6 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
if (bo == NULL)
return -ENOMEM;
drm_gem_private_object_init(adev_to_drm(adev), >tbo.base, size);
-   INIT_LIST_HEAD(>shadow_list);
bo->vm_bo = NULL;
bo->preferred_domains = bp->preferred_domain ? bp->preferred_domain :
bp->domain;
@@ -713,6 +714,7 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
return r;
 
*vmbo_ptr = to_amdgpu_bo_vm(bo_ptr);
+   INIT_LIST_HEAD(&(*vmbo_ptr)->shadow_list);
return r;
 }
 
@@ -757,12 +759,12 @@ int amdgpu_bo_validate(struct amdgpu_bo *bo)
  *
  * Insert a BO to the shadow list.
  */
-void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo)
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo_vm *vmbo)
 {
-   struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+   struct amdgpu_device *adev = amdgpu_ttm_adev(vmbo->bo.tbo.bdev);
 
mutex_lock(>shadow_list_lock);
-   list_add_tail(>shadow_list, >shadow_list);
+   list_add_tail(>shadow_list, >shadow_list);
mutex_unlock(>shadow_list_lock);
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index a8c702634e1b..18cb2f28e4de 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -109,9 +109,6 @@ struct amdgpu_bo {
 #ifdef CONFIG_MMU_NOTIFIER
struct mmu_interval_notifiernotifier;
 #endif
-
-   struct list_headshadow_list;
-
struct kgd_mem  *kfd_bo;
 };
 
@@ -127,6 +124,7 @@ struct amdgpu_bo_user {
 struct amdgpu_bo_vm {
struct amdgpu_bobo;
struct amdgpu_bo*shadow;
+   struct list_headshadow_list;
struct amdgpu_vm_bo_baseentries[];
 };
 
@@ -333,7 +331,7 @@ u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
 int amdgpu_bo_validate(struct amdgpu_bo *bo);
 void amdgpu_bo_get_memory(struct amdgpu_bo *bo, uint64_t *vram_mem,
uint64_t *gtt_mem, uint64_t *cpu_mem);
-void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo);
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo_vm *vmbo);
 int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow,
 struct dma_fence **fence);
 uint32_t amdgpu_bo_get_preferred_pin_domain(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_

[PATCH 1/1] drm/amdkfd: remove unused variable

[Nirmoy Das]

Remove it.

CC: jonathan@amd.com
CC: felix.kuehl...@amd.com
Fixes: d7b132507384c("drm/amdkfd: fix circular locking on get_wave_state")
Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c 
b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index e6366b408420..539212039876 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -997,7 +997,7 @@ static int start_nocpsch(struct device_queue_manager *dqm)
 {
pr_info("SW scheduler is used");
init_interrupts(dqm);
-   
+
if (dqm->dev->device_info->asic_family == CHIP_HAWAII)
return pm_init(>packets, dqm);
dqm->sched_running = true;
@@ -1674,7 +1674,6 @@ static int get_wave_state(struct device_queue_manager 
*dqm,
  u32 *save_area_used_size)
 {
struct mqd_manager *mqd_mgr;
-   int r;
 
dqm_lock(dqm);
 
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/2] drm/amdgpu: move shadow_list to amdgpu_bo_vm

[Nirmoy Das]

Move shadow_list to struct amdgpu_bo_vm as shadow BOs
are part of PT/PD BOs.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  5 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 14 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  6 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c |  2 +-
 4 files changed, 14 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index f2636f411a25..3f51b142fc83 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4063,6 +4063,7 @@ static int amdgpu_device_recover_vram(struct 
amdgpu_device *adev)
 {
struct dma_fence *fence = NULL, *next = NULL;
struct amdgpu_bo *shadow;
+   struct amdgpu_bo_vm *vmbo;
long r = 1, tmo;
 
if (amdgpu_sriov_runtime(adev))
@@ -4072,8 +4073,8 @@ static int amdgpu_device_recover_vram(struct 
amdgpu_device *adev)
 
dev_info(adev->dev, "recover vram bo from shadow start\n");
mutex_lock(>shadow_list_lock);
-   list_for_each_entry(shadow, >shadow_list, shadow_list) {
-
+   list_for_each_entry(vmbo, >shadow_list, shadow_list) {
+   shadow = >bo;
/* No need to recover an evicted BO */
if (shadow->tbo.mem.mem_type != TTM_PL_TT ||
shadow->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET ||
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 8473d153650f..7dbf711e41ce 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -108,11 +108,13 @@ static void amdgpu_bo_vm_destroy(struct ttm_buffer_object 
*tbo)
 {
struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+   struct amdgpu_bo_vm *vmbo;
 
+   vmbo = to_amdgpu_bo_vm(bo);
/* in case amdgpu_device_recover_vram got NULL of bo->parent */
-   if (!list_empty(>shadow_list)) {
+   if (!list_empty(>shadow_list)) {
mutex_lock(>shadow_list_lock);
-   list_del_init(>shadow_list);
+   list_del_init(>shadow_list);
mutex_unlock(>shadow_list_lock);
}
 
@@ -588,7 +590,6 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
if (bo == NULL)
return -ENOMEM;
drm_gem_private_object_init(adev_to_drm(adev), >tbo.base, size);
-   INIT_LIST_HEAD(>shadow_list);
bo->vm_bo = NULL;
bo->preferred_domains = bp->preferred_domain ? bp->preferred_domain :
bp->domain;
@@ -718,6 +719,7 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
return r;
 
*vmbo_ptr = to_amdgpu_bo_vm(bo_ptr);
+   INIT_LIST_HEAD(&(*vmbo_ptr)->shadow_list);
return r;
 }
 
@@ -762,12 +764,12 @@ int amdgpu_bo_validate(struct amdgpu_bo *bo)
  *
  * Insert a BO to the shadow list.
  */
-void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo)
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo_vm *vmbo)
 {
-   struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+   struct amdgpu_device *adev = amdgpu_ttm_adev(vmbo->bo.tbo.bdev);
 
mutex_lock(>shadow_list_lock);
-   list_add_tail(>shadow_list, >shadow_list);
+   list_add_tail(>shadow_list, >shadow_list);
mutex_unlock(>shadow_list_lock);
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index a8c702634e1b..18cb2f28e4de 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -109,9 +109,6 @@ struct amdgpu_bo {
 #ifdef CONFIG_MMU_NOTIFIER
struct mmu_interval_notifiernotifier;
 #endif
-
-   struct list_headshadow_list;
-
struct kgd_mem  *kfd_bo;
 };
 
@@ -127,6 +124,7 @@ struct amdgpu_bo_user {
 struct amdgpu_bo_vm {
struct amdgpu_bobo;
struct amdgpu_bo*shadow;
+   struct list_headshadow_list;
struct amdgpu_vm_bo_baseentries[];
 };
 
@@ -333,7 +331,7 @@ u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
 int amdgpu_bo_validate(struct amdgpu_bo *bo);
 void amdgpu_bo_get_memory(struct amdgpu_bo *bo, uint64_t *vram_mem,
uint64_t *gtt_mem, uint64_t *cpu_mem);
-void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo);
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo_vm *vmbo);
 int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow,
 struct dma_fence **fence);
 uint32_t amdgpu_bo_get_preferred_pin_domain(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_

[PATCH v2 1/2] drm/amdgpu: parameterize ttm BO destroy callback

[Nirmoy Das]

Make provision to pass different ttm BO destroy callback
while creating a amdgpu_bo.

v2: remove whitespace.
call amdgpu_bo_destroy_base() at the end for cleaner code.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 48 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  3 +-
 2 files changed, 38 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 9092ac12a270..8473d153650f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -73,11 +73,9 @@ static void amdgpu_bo_subtract_pin_size(struct amdgpu_bo *bo)
}
 }

-static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
+static void amdgpu_bo_destroy_base(struct ttm_buffer_object *tbo)
 {
-   struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
-   struct amdgpu_bo_user *ubo;

if (bo->tbo.pin_count > 0)
amdgpu_bo_subtract_pin_size(bo);
@@ -87,20 +85,38 @@ static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
if (bo->tbo.base.import_attach)
drm_prime_gem_destroy(>tbo.base, bo->tbo.sg);
drm_gem_object_release(>tbo.base);
+   amdgpu_bo_unref(>parent);
+   kvfree(bo);
+}
+
+static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
+{
+   amdgpu_bo_destroy_base(tbo);
+}
+
+static void amdgpu_bo_user_destroy(struct ttm_buffer_object *tbo)
+{
+   struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+   struct amdgpu_bo_user *ubo;
+
+   ubo = to_amdgpu_bo_user(bo);
+   kfree(ubo->metadata);
+   amdgpu_bo_destroy_base(tbo);
+}
+
+static void amdgpu_bo_vm_destroy(struct ttm_buffer_object *tbo)
+{
+   struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
+   struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+
/* in case amdgpu_device_recover_vram got NULL of bo->parent */
if (!list_empty(>shadow_list)) {
mutex_lock(>shadow_list_lock);
list_del_init(>shadow_list);
mutex_unlock(>shadow_list_lock);
}
-   amdgpu_bo_unref(>parent);
-
-   if (bo->tbo.type != ttm_bo_type_kernel) {
-   ubo = to_amdgpu_bo_user(bo);
-   kfree(ubo->metadata);
-   }

-   kvfree(bo);
+   amdgpu_bo_destroy_base(tbo);
 }

 /**
@@ -115,8 +131,11 @@ static void amdgpu_bo_destroy(struct ttm_buffer_object 
*tbo)
  */
 bool amdgpu_bo_is_amdgpu_bo(struct ttm_buffer_object *bo)
 {
-   if (bo->destroy == _bo_destroy)
+   if (bo->destroy == _bo_destroy ||
+   bo->destroy == _bo_user_destroy ||
+   bo->destroy == _bo_vm_destroy)
return true;
+
return false;
 }

@@ -592,9 +611,12 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
if (bp->type == ttm_bo_type_kernel)
bo->tbo.priority = 1;

+   if (!bp->destroy)
+   bp->destroy = _bo_destroy;
+
r = ttm_bo_init_reserved(>mman.bdev, >tbo, size, bp->type,
 >placement, page_align, ,  NULL,
-bp->resv, _bo_destroy);
+bp->resv, bp->destroy);
if (unlikely(r != 0))
return r;

@@ -658,6 +680,7 @@ int amdgpu_bo_create_user(struct amdgpu_device *adev,
int r;

bp->bo_ptr_size = sizeof(struct amdgpu_bo_user);
+   bp->destroy = _bo_user_destroy;
r = amdgpu_bo_create(adev, bp, _ptr);
if (r)
return r;
@@ -689,6 +712,7 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
 * num of amdgpu_vm_pt entries.
 */
BUG_ON(bp->bo_ptr_size < sizeof(struct amdgpu_bo_vm));
+   bp->destroy = _bo_vm_destroy;
r = amdgpu_bo_create(adev, bp, _ptr);
if (r)
return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index e36b84516b4e..a8c702634e1b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -55,7 +55,8 @@ struct amdgpu_bo_param {
u64 flags;
enum ttm_bo_typetype;
boolno_wait_gpu;
-   struct dma_resv *resv;
+   struct dma_resv *resv;
+   void(*destroy)(struct ttm_buffer_object 
*bo);
 };

 /* bo virtual addresses in a vm */
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/3] drm/amdgpu: use separate ttm destroy callback

[Nirmoy Das]

Use different ttm destroy callback for different type of amdgpu BO.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 47 --
 1 file changed, 35 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 71a65525eac4..d97c20346a8c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -73,11 +73,9 @@ static void amdgpu_bo_subtract_pin_size(struct amdgpu_bo *bo)
}
 }
 
-static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
+static void amdgpu_bo_destroy_base(struct ttm_buffer_object *tbo)
 {
-   struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
-   struct amdgpu_bo_user *ubo;
 
if (bo->tbo.pin_count > 0)
amdgpu_bo_subtract_pin_size(bo);
@@ -87,18 +85,40 @@ static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
if (bo->tbo.base.import_attach)
drm_prime_gem_destroy(>tbo.base, bo->tbo.sg);
drm_gem_object_release(>tbo.base);
+   amdgpu_bo_unref(>parent);
+}
+
+static void amdgpu_bo_destroy(struct ttm_buffer_object *tbo)
+{
+   struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+
+   amdgpu_bo_destroy_base(tbo);
+   kvfree(bo);
+}
+
+static void amdgpu_bo_user_destroy(struct ttm_buffer_object *tbo)
+{
+   struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+   struct amdgpu_bo_user *ubo;
+
+   amdgpu_bo_destroy_base(tbo);
+   ubo = to_amdgpu_bo_user(bo);
+   kfree(ubo->metadata);
+   kvfree(bo);
+}
+
+static void amdgpu_bo_vm_destroy(struct ttm_buffer_object *tbo)
+{
+   struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
+   struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+
+   amdgpu_bo_destroy_base(tbo);
/* in case amdgpu_device_recover_vram got NULL of bo->parent */
if (!list_empty(>shadow_list)) {
mutex_lock(>shadow_list_lock);
list_del_init(>shadow_list);
mutex_unlock(>shadow_list_lock);
}
-   amdgpu_bo_unref(>parent);
-
-   if (bo->tbo.type != ttm_bo_type_kernel) {
-   ubo = to_amdgpu_bo_user(bo);
-   kfree(ubo->metadata);
-   }
 
kvfree(bo);
 }
@@ -115,8 +135,11 @@ static void amdgpu_bo_destroy(struct ttm_buffer_object 
*tbo)
  */
 bool amdgpu_bo_is_amdgpu_bo(struct ttm_buffer_object *bo)
 {
-   if (bo->destroy == _bo_destroy)
+   if (bo->destroy == _bo_destroy ||
+   bo->destroy == _bo_user_destroy ||
+   bo->destroy == _bo_vm_destroy)
return true;
+
return false;
 }
 
@@ -678,7 +701,7 @@ int amdgpu_bo_create_user(struct amdgpu_device *adev,
int r;
 
bp->bo_ptr_size = sizeof(struct amdgpu_bo_user);
-   r = amdgpu_bo_create(adev, bp, _ptr);
+   r = amdgpu_bo_do_create(adev, bp, _ptr, _bo_user_destroy);
if (r)
return r;
 
@@ -709,7 +732,7 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
 * num of amdgpu_vm_pt entries.
 */
BUG_ON(bp->bo_ptr_size < sizeof(struct amdgpu_bo_vm));
-   r = amdgpu_bo_create(adev, bp, _ptr);
+   r = amdgpu_bo_do_create(adev, bp, _ptr, _bo_vm_destroy);
if (r)
return r;
 
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/3] drm/amdgpu: move shadow_list to amdgpu_bo_vm

[Nirmoy Das]

Move shadow_list to struct amdgpu_bo_vm as shadow BOs
are part of PT/PD BOs.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  5 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 13 +++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  6 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c |  2 +-
 4 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index f2636f411a25..3f51b142fc83 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4063,6 +4063,7 @@ static int amdgpu_device_recover_vram(struct 
amdgpu_device *adev)
 {
struct dma_fence *fence = NULL, *next = NULL;
struct amdgpu_bo *shadow;
+   struct amdgpu_bo_vm *vmbo;
long r = 1, tmo;
 
if (amdgpu_sriov_runtime(adev))
@@ -4072,8 +4073,8 @@ static int amdgpu_device_recover_vram(struct 
amdgpu_device *adev)
 
dev_info(adev->dev, "recover vram bo from shadow start\n");
mutex_lock(>shadow_list_lock);
-   list_for_each_entry(shadow, >shadow_list, shadow_list) {
-
+   list_for_each_entry(vmbo, >shadow_list, shadow_list) {
+   shadow = >bo;
/* No need to recover an evicted BO */
if (shadow->tbo.mem.mem_type != TTM_PL_TT ||
shadow->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET ||
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index d97c20346a8c..cb543e8abe35 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -111,12 +111,13 @@ static void amdgpu_bo_vm_destroy(struct ttm_buffer_object 
*tbo)
 {
struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
struct amdgpu_bo *bo = ttm_to_amdgpu_bo(tbo);
+   struct amdgpu_bo_vm *vmbo = to_amdgpu_bo_vm(bo);
 
amdgpu_bo_destroy_base(tbo);
/* in case amdgpu_device_recover_vram got NULL of bo->parent */
-   if (!list_empty(>shadow_list)) {
+   if (!list_empty(>shadow_list)) {
mutex_lock(>shadow_list_lock);
-   list_del_init(>shadow_list);
+   list_del_init(>shadow_list);
mutex_unlock(>shadow_list_lock);
}
 
@@ -594,7 +595,6 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev,
if (bo == NULL)
return -ENOMEM;
drm_gem_private_object_init(adev_to_drm(adev), >tbo.base, size);
-   INIT_LIST_HEAD(>shadow_list);
bo->vm_bo = NULL;
bo->preferred_domains = bp->preferred_domain ? bp->preferred_domain :
bp->domain;
@@ -737,6 +737,7 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
return r;
 
*vmbo_ptr = to_amdgpu_bo_vm(bo_ptr);
+   INIT_LIST_HEAD(&(*vmbo_ptr)->shadow_list);
return r;
 }
 
@@ -781,12 +782,12 @@ int amdgpu_bo_validate(struct amdgpu_bo *bo)
  *
  * Insert a BO to the shadow list.
  */
-void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo)
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo_vm *vmbo)
 {
-   struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+   struct amdgpu_device *adev = amdgpu_ttm_adev(vmbo->bo.tbo.bdev);
 
mutex_lock(>shadow_list_lock);
-   list_add_tail(>shadow_list, >shadow_list);
+   list_add_tail(>shadow_list, >shadow_list);
mutex_unlock(>shadow_list_lock);
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index e36b84516b4e..a08a34351f12 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -108,9 +108,6 @@ struct amdgpu_bo {
 #ifdef CONFIG_MMU_NOTIFIER
struct mmu_interval_notifiernotifier;
 #endif
-
-   struct list_headshadow_list;
-
struct kgd_mem  *kfd_bo;
 };
 
@@ -126,6 +123,7 @@ struct amdgpu_bo_user {
 struct amdgpu_bo_vm {
struct amdgpu_bobo;
struct amdgpu_bo*shadow;
+   struct list_headshadow_list;
struct amdgpu_vm_bo_baseentries[];
 };
 
@@ -332,7 +330,7 @@ u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
 int amdgpu_bo_validate(struct amdgpu_bo *bo);
 void amdgpu_bo_get_memory(struct amdgpu_bo *bo, uint64_t *vram_mem,
uint64_t *gtt_mem, uint64_t *cpu_mem);
-void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo);
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo_vm *vmbo);
 int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow,
 struct dma_fence **fence);
 uint32_t amdgpu_bo_get_preferred_pin_domain(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/driv

[PATCH 1/3] drm/amdgpu: parameterize ttm BO destroy callback

[Nirmoy Das]

Make provision to pass different ttm BO destroy callback
while creating a amdgpu_bo.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 28 ++
 1 file changed, 24 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 9092ac12a270..71a65525eac4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -522,15 +522,17 @@ bool amdgpu_bo_support_uswc(u64 bo_flags)
  * @adev: amdgpu device object
  * @bp: parameters to be used for the buffer object
  * @bo_ptr: pointer to the buffer object pointer
+ * @destroy: ttm bo destroy callback
  *
- * Creates an _bo buffer object.
+ * Creates an _bo buffer object with a specified ttm bo destroy 
callback.
  *
  * Returns:
  * 0 for success or a negative error code on failure.
  */
-int amdgpu_bo_create(struct amdgpu_device *adev,
+static int amdgpu_bo_do_create(struct amdgpu_device *adev,
   struct amdgpu_bo_param *bp,
-  struct amdgpu_bo **bo_ptr)
+  struct amdgpu_bo **bo_ptr,
+  void (*destroy)(struct ttm_buffer_object *))
 {
struct ttm_operation_ctx ctx = {
.interruptible = (bp->type != ttm_bo_type_kernel),
@@ -594,7 +596,7 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
 
r = ttm_bo_init_reserved(>mman.bdev, >tbo, size, bp->type,
 >placement, page_align, ,  NULL,
-bp->resv, _bo_destroy);
+bp->resv, destroy);
if (unlikely(r != 0))
return r;
 
@@ -638,6 +640,24 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
return r;
 }
 
+/**
+ * amdgpu_bo_create - create an _bo buffer object
+ * @adev: amdgpu device object
+ * @bp: parameters to be used for the buffer object
+ * @bo_ptr: pointer to the buffer object pointer
+ *
+ * Creates an _bo buffer object.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+int amdgpu_bo_create(struct amdgpu_device *adev,
+  struct amdgpu_bo_param *bp,
+  struct amdgpu_bo **bo_ptr)
+{
+   return amdgpu_bo_do_create(adev, bp, bo_ptr, _bo_destroy);
+}
+
 /**
  * amdgpu_bo_create_user - create an _bo_user buffer object
  * @adev: amdgpu device object
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 1/1] drm/amdgpu: remove amdgpu_vm_pt

[Nirmoy Das]

Page table entries are now in embedded in VM BO, so
we do not need struct amdgpu_vm_pt. This patch replaces
struct amdgpu_vm_pt with struct amdgpu_vm_bo_base.

v2: change "!(cursor->level < AMDGPU_VM_PTB)" --> "(cursor->level == 
AMDGPU_VM_PTB)"

Signed-off-by: Nirmoy Das 
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c  |  26 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c|   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c   |   4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c   |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c|   4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c   |  12 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c   |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h|   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c| 164 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h|   9 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c   |   2 +-
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c  |   2 +-
 12 files changed, 105 insertions(+), 126 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index fb6bcc386de1..f96598279593 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -356,7 +356,7 @@ static int amdgpu_amdkfd_validate_vm_bo(void *_unused, 
struct amdgpu_bo *bo)
  */
 static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
 {
-   struct amdgpu_bo *pd = vm->root.base.bo;
+   struct amdgpu_bo *pd = vm->root.bo;
struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
int ret;

@@ -372,7 +372,7 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
return ret;
}

-   vm->pd_phys_addr = amdgpu_gmc_pd_addr(vm->root.base.bo);
+   vm->pd_phys_addr = amdgpu_gmc_pd_addr(vm->root.bo);

if (vm->use_cpu_for_update) {
ret = amdgpu_bo_kmap(pd, NULL);
@@ -387,7 +387,7 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)

 static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync)
 {
-   struct amdgpu_bo *pd = vm->root.base.bo;
+   struct amdgpu_bo *pd = vm->root.bo;
struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
int ret;

@@ -1153,7 +1153,7 @@ static int process_sync_pds_resv(struct 
amdkfd_process_info *process_info,

list_for_each_entry(peer_vm, _info->vm_list_head,
vm_list_node) {
-   struct amdgpu_bo *pd = peer_vm->root.base.bo;
+   struct amdgpu_bo *pd = peer_vm->root.bo;

ret = amdgpu_sync_resv(NULL, sync, pd->tbo.base.resv,
   AMDGPU_SYNC_NE_OWNER,
@@ -1220,7 +1220,7 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void 
**process_info,
vm->process_info = *process_info;

/* Validate page directory and attach eviction fence */
-   ret = amdgpu_bo_reserve(vm->root.base.bo, true);
+   ret = amdgpu_bo_reserve(vm->root.bo, true);
if (ret)
goto reserve_pd_fail;
ret = vm_validate_pt_pd_bos(vm);
@@ -1228,16 +1228,16 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void 
**process_info,
pr_err("validate_pt_pd_bos() failed\n");
goto validate_pd_fail;
}
-   ret = amdgpu_bo_sync_wait(vm->root.base.bo,
+   ret = amdgpu_bo_sync_wait(vm->root.bo,
  AMDGPU_FENCE_OWNER_KFD, false);
if (ret)
goto wait_pd_fail;
-   ret = dma_resv_reserve_shared(vm->root.base.bo->tbo.base.resv, 1);
+   ret = dma_resv_reserve_shared(vm->root.bo->tbo.base.resv, 1);
if (ret)
goto reserve_shared_fail;
-   amdgpu_bo_fence(vm->root.base.bo,
+   amdgpu_bo_fence(vm->root.bo,
>process_info->eviction_fence->base, true);
-   amdgpu_bo_unreserve(vm->root.base.bo);
+   amdgpu_bo_unreserve(vm->root.bo);

/* Update process info */
mutex_lock(>process_info->lock);
@@ -1251,7 +1251,7 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void 
**process_info,
 reserve_shared_fail:
 wait_pd_fail:
 validate_pd_fail:
-   amdgpu_bo_unreserve(vm->root.base.bo);
+   amdgpu_bo_unreserve(vm->root.bo);
 reserve_pd_fail:
vm->process_info = NULL;
if (info) {
@@ -1306,7 +1306,7 @@ void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device 
*adev,
struct amdgpu_vm *vm)
 {
struct amdkfd_process_info *process_info = vm->process_info;
-   struct amdgpu_bo *pd = vm->root.base.bo;
+   struct amdgpu_bo *pd = vm->root.bo;

if (!process_info)
return;
@@ -1362,7 +1362,7 @@ void amdgpu_amdkfd_gpuvm_release_process_vm(struct 
kgd_dev *kgd, void *drm_pri

[PATCH 1/1] drm/amdgpu: remove amdgpu_vm_pt

[Nirmoy Das]

Page table entries are now in embedded in VM BO, so
we do not need struct amdgpu_vm_pt. This patch replaces
struct amdgpu_vm_pt with struct amdgpu_vm_bo_base.

Signed-off-by: Nirmoy Das 
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c  |  26 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c|   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c   |   4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c   |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c|   4 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c   |  12 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c   |   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h|   2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c| 164 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h|   9 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c   |   2 +-
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c  |   2 +-
 12 files changed, 105 insertions(+), 126 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index fb6bcc386de1..f96598279593 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -356,7 +356,7 @@ static int amdgpu_amdkfd_validate_vm_bo(void *_unused, 
struct amdgpu_bo *bo)
  */
 static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
 {
-   struct amdgpu_bo *pd = vm->root.base.bo;
+   struct amdgpu_bo *pd = vm->root.bo;
struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
int ret;
 
@@ -372,7 +372,7 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
return ret;
}
 
-   vm->pd_phys_addr = amdgpu_gmc_pd_addr(vm->root.base.bo);
+   vm->pd_phys_addr = amdgpu_gmc_pd_addr(vm->root.bo);
 
if (vm->use_cpu_for_update) {
ret = amdgpu_bo_kmap(pd, NULL);
@@ -387,7 +387,7 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
 
 static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync)
 {
-   struct amdgpu_bo *pd = vm->root.base.bo;
+   struct amdgpu_bo *pd = vm->root.bo;
struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
int ret;
 
@@ -1153,7 +1153,7 @@ static int process_sync_pds_resv(struct 
amdkfd_process_info *process_info,
 
list_for_each_entry(peer_vm, _info->vm_list_head,
vm_list_node) {
-   struct amdgpu_bo *pd = peer_vm->root.base.bo;
+   struct amdgpu_bo *pd = peer_vm->root.bo;
 
ret = amdgpu_sync_resv(NULL, sync, pd->tbo.base.resv,
   AMDGPU_SYNC_NE_OWNER,
@@ -1220,7 +1220,7 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void 
**process_info,
vm->process_info = *process_info;
 
/* Validate page directory and attach eviction fence */
-   ret = amdgpu_bo_reserve(vm->root.base.bo, true);
+   ret = amdgpu_bo_reserve(vm->root.bo, true);
if (ret)
goto reserve_pd_fail;
ret = vm_validate_pt_pd_bos(vm);
@@ -1228,16 +1228,16 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void 
**process_info,
pr_err("validate_pt_pd_bos() failed\n");
goto validate_pd_fail;
}
-   ret = amdgpu_bo_sync_wait(vm->root.base.bo,
+   ret = amdgpu_bo_sync_wait(vm->root.bo,
  AMDGPU_FENCE_OWNER_KFD, false);
if (ret)
goto wait_pd_fail;
-   ret = dma_resv_reserve_shared(vm->root.base.bo->tbo.base.resv, 1);
+   ret = dma_resv_reserve_shared(vm->root.bo->tbo.base.resv, 1);
if (ret)
goto reserve_shared_fail;
-   amdgpu_bo_fence(vm->root.base.bo,
+   amdgpu_bo_fence(vm->root.bo,
>process_info->eviction_fence->base, true);
-   amdgpu_bo_unreserve(vm->root.base.bo);
+   amdgpu_bo_unreserve(vm->root.bo);
 
/* Update process info */
mutex_lock(>process_info->lock);
@@ -1251,7 +1251,7 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void 
**process_info,
 reserve_shared_fail:
 wait_pd_fail:
 validate_pd_fail:
-   amdgpu_bo_unreserve(vm->root.base.bo);
+   amdgpu_bo_unreserve(vm->root.bo);
 reserve_pd_fail:
vm->process_info = NULL;
if (info) {
@@ -1306,7 +1306,7 @@ void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device 
*adev,
struct amdgpu_vm *vm)
 {
struct amdkfd_process_info *process_info = vm->process_info;
-   struct amdgpu_bo *pd = vm->root.base.bo;
+   struct amdgpu_bo *pd = vm->root.bo;
 
if (!process_info)
return;
@@ -1362,7 +1362,7 @@ void amdgpu_amdkfd_gpuvm_release_process_vm(struct 
kgd_dev *kgd, void *drm_priv)
 uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv)
 {
struct amdgpu_vm

[PATCH v3 1/1] drm/amdkfd: use allowed domain for vmbo validation

[Nirmoy Das]

Fixes handling when page tables are in system memory.

v3: remove struct amdgpu_vm_parser.
v2: remove unwanted variable.
change amdgpu_amdkfd_validate instead of amdgpu_amdkfd_bo_validate.

Signed-off-by: Nirmoy Das 
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c  | 21 ---
 1 file changed, 4 insertions(+), 17 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index d6cb7cf76623..016815b7a773 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -47,12 +47,6 @@ static struct {
spinlock_t mem_limit_lock;
 } kfd_mem_limit;

-/* Struct used for amdgpu_amdkfd_bo_validate */
-struct amdgpu_vm_parser {
-   uint32_tdomain;
-   boolwait;
-};
-
 static const char * const domain_bit_to_string[] = {
"CPU",
"GTT",
@@ -348,11 +342,9 @@ static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, 
uint32_t domain,
return ret;
 }

-static int amdgpu_amdkfd_validate(void *param, struct amdgpu_bo *bo)
+static int amdgpu_amdkfd_validate_vm_bo(void *_unused, struct amdgpu_bo *bo)
 {
-   struct amdgpu_vm_parser *p = param;
-
-   return amdgpu_amdkfd_bo_validate(bo, p->domain, p->wait);
+   return amdgpu_amdkfd_bo_validate(bo, bo->allowed_domains, false);
 }

 /* vm_validate_pt_pd_bos - Validate page table and directory BOs
@@ -366,20 +358,15 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
 {
struct amdgpu_bo *pd = vm->root.base.bo;
struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
-   struct amdgpu_vm_parser param;
int ret;

-   param.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   param.wait = false;
-
-   ret = amdgpu_vm_validate_pt_bos(adev, vm, amdgpu_amdkfd_validate,
-   );
+   ret = amdgpu_vm_validate_pt_bos(adev, vm, amdgpu_amdkfd_validate_vm_bo, 
NULL);
if (ret) {
pr_err("failed to validate PT BOs\n");
return ret;
}

-   ret = amdgpu_amdkfd_validate(, pd);
+   ret = amdgpu_amdkfd_validate_vm_bo(NULL, pd);
if (ret) {
pr_err("failed to validate PD\n");
return ret;
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 1/1] drm/amdkfd: use allowed domain for vmbo validation

[Nirmoy Das]

Fixes handling when page tables are in system memory.

v2: remove unwanted variable.
change amdgpu_amdkfd_validate instead of amdgpu_amdkfd_bo_validate.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 15 ---
 1 file changed, 4 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index d6cb7cf76623..021f25085760 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -348,11 +348,9 @@ static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, 
uint32_t domain,
return ret;
 }

-static int amdgpu_amdkfd_validate(void *param, struct amdgpu_bo *bo)
+static int amdgpu_amdkfd_validate_vm_bo(void *_unused, struct amdgpu_bo *bo)
 {
-   struct amdgpu_vm_parser *p = param;
-
-   return amdgpu_amdkfd_bo_validate(bo, p->domain, p->wait);
+   return amdgpu_amdkfd_bo_validate(bo, bo->allowed_domains, false);
 }

 /* vm_validate_pt_pd_bos - Validate page table and directory BOs
@@ -366,20 +364,15 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
 {
struct amdgpu_bo *pd = vm->root.base.bo;
struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
-   struct amdgpu_vm_parser param;
int ret;

-   param.domain = AMDGPU_GEM_DOMAIN_VRAM;
-   param.wait = false;
-
-   ret = amdgpu_vm_validate_pt_bos(adev, vm, amdgpu_amdkfd_validate,
-   );
+   ret = amdgpu_vm_validate_pt_bos(adev, vm, amdgpu_amdkfd_validate_vm_bo, 
NULL);
if (ret) {
pr_err("failed to validate PT BOs\n");
return ret;
}

-   ret = amdgpu_amdkfd_validate(, pd);
+   ret = amdgpu_amdkfd_validate_vm_bo(NULL, pd);
if (ret) {
pr_err("failed to validate PD\n");
return ret;
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amdgpu: fix shadow bo skip condition

[Nirmoy Das]

Create shadow BOs only for no-compute VM context and only for dGPU.
The existing if-condition would create shadow bo for compute context
on dGPU which not what we wanted.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 74c7d8b21c98..e2146caf6866 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -908,7 +908,7 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
return r;
 
bo = &(*vmbo)->bo;
-   if (vm->is_compute_context && (adev->flags & AMD_IS_APU)) {
+   if (vm->is_compute_context || (adev->flags & AMD_IS_APU)) {
(*vmbo)->shadow = NULL;
return 0;
}
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amdkfd: use allowed domain for vmbo validation

[Nirmoy Das]

Remove fixed domain and use BO's allowed domain for PT/PD
BO validation.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index d6cb7cf76623..a45eadfbf417 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -336,7 +336,7 @@ static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, 
uint32_t domain,
 "Called with userptr BO"))
return -EINVAL;
 
-   amdgpu_bo_placement_from_domain(bo, domain);
+   amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
 
ret = ttm_bo_validate(>tbo, >placement, );
if (ret)
@@ -369,7 +369,6 @@ static int vm_validate_pt_pd_bos(struct amdgpu_vm *vm)
struct amdgpu_vm_parser param;
int ret;
 
-   param.domain = AMDGPU_GEM_DOMAIN_VRAM;
param.wait = false;
 
ret = amdgpu_vm_validate_pt_bos(adev, vm, amdgpu_amdkfd_validate,
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amd/display: fix coding style

[Nirmoy Das]

Fixes: 60d198ca66ecf778 ("drm/amd/display: WARN_ON cleanups")

Signed-off-by: Nirmoy Das 
---
Hi Alex,

Can you please squash this with the "Fixes" patch?

Thanks,
Nirmoy

 drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c 
b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index 09bbec4dab7c..56abc27b78f7 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -315,9 +315,8 @@ get_crtc_by_otg_inst(struct amdgpu_device *adev,
struct drm_crtc *crtc;
struct amdgpu_crtc *amdgpu_crtc;

-   if (WARN_ON(otg_inst == -1)) {
+   if (WARN_ON(otg_inst == -1))
return adev->mode_info.crtcs[0];
-   }

list_for_each_entry(crtc, >mode_config.crtc_list, head) {
amdgpu_crtc = to_amdgpu_crtc(crtc);
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v4 1/1] drm/amdgpu: flush gart changes after all BO recovery

[Nirmoy Das]

Don't flush gart changes after recovering each BO instead
do it after recovering all the BOs. Flishing gart also needed
for amdgpu_ttm_alloc_gart().

v4: use container_of to retrieve adev struct.
v3: rename amdgpu_gart_tlb_flush() -> amdgpu_gart_invalidate_tlb().
v2: abstract out gart tlb flushing logic to amdgpu_gart.c

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c| 22 ++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h|  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c |  4 
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  1 +
 4 files changed, 21 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index 5562b5c90c03..35cc8009ac7a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -312,8 +312,6 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t 
offset,
 int pages, struct page **pagelist, dma_addr_t *dma_addr,
 uint64_t flags)
 {
-   int r, i;
-
if (!adev->gart.ready) {
WARN(1, "trying to bind memory to uninitialized GART !\n");
return -EINVAL;
@@ -322,16 +320,26 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t 
offset,
if (!adev->gart.ptr)
return 0;

-   r = amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
-   adev->gart.ptr);
-   if (r)
-   return r;
+   return amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
+  adev->gart.ptr);
+}
+
+/**
+ * amdgpu_gart_invalidate_tlb - invalidate gart TLB
+ *
+ * @adev: amdgpu device driver pointer
+ *
+ * Invalidate gart TLB which can be use as a way to flush gart changes
+ *
+ */
+void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev)
+{
+   int i;

mb();
amdgpu_asic_flush_hdp(adev, NULL);
for (i = 0; i < adev->num_vmhubs; i++)
amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
-   return 0;
 }

 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
index a25fe97b0196..e104022197ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
@@ -66,5 +66,5 @@ int amdgpu_gart_map(struct amdgpu_device *adev, uint64_t 
offset,
 int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
 int pages, struct page **pagelist,
 dma_addr_t *dma_addr, uint64_t flags);
-
+void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev);
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 8860545344c7..a1cd775fd61c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -206,10 +206,12 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
 int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
 {
struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
+   struct amdgpu_device *adev;
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
int r = 0;

+   adev = container_of(mgr, typeof(*adev), mman.gtt_mgr);
spin_lock(>lock);
drm_mm_for_each_node(mm_node, >mm) {
node = container_of(mm_node, struct amdgpu_gtt_node, node);
@@ -219,6 +221,8 @@ int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
}
spin_unlock(>lock);

+   amdgpu_gart_invalidate_tlb(adev);
+
return r;
 }

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c0aef327292a..786650a4a493 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1014,6 +1014,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
return r;
}

+   amdgpu_gart_invalidate_tlb(adev);
ttm_resource_free(bo, >mem);
bo->mem = tmp;
}
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v3 1/1] drm/amdgpu: flush gart changes after all BO recovery

[Nirmoy Das]

Don't flush gart changes after recovering each BO instead
do it after recovering all the BOs. Flishing gart also needed
for amdgpu_ttm_alloc_gart().

v3: rename amdgpu_gart_tlb_flush() -> amdgpu_gart_invalidate_tlb()
v2: abstract out gart tlb flushing logic to amdgpu_gart.c

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c| 20 +++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h|  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c |  6 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  1 +
 4 files changed, 23 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index 5562b5c90c03..992b59c29dd7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -322,16 +322,26 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t 
offset,
if (!adev->gart.ptr)
return 0;

-   r = amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
-   adev->gart.ptr);
-   if (r)
-   return r;
+   return amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
+  adev->gart.ptr);
+}
+
+/**
+ * amdgpu_gart_invalidate_tlb - invalidate gart TLB
+ *
+ * @adev: amdgpu device driver pointer
+ *
+ * Invalidate gart TLB which can be use as a way to flush gart changes
+ *
+ */
+void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev)
+{
+   int i;

mb();
amdgpu_asic_flush_hdp(adev, NULL);
for (i = 0; i < adev->num_vmhubs; i++)
amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
-   return 0;
 }

 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
index a25fe97b0196..e104022197ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
@@ -66,5 +66,5 @@ int amdgpu_gart_map(struct amdgpu_device *adev, uint64_t 
offset,
 int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
 int pages, struct page **pagelist,
 dma_addr_t *dma_addr, uint64_t flags);
-
+void amdgpu_gart_invalidate_tlb(struct amdgpu_device *adev);
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 8860545344c7..b01bc2346082 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -205,6 +205,7 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
  */
 int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
 {
+   struct amdgpu_device *adev = NULL;
struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
@@ -216,9 +217,14 @@ int amdgpu_gtt_mgr_recover(struct ttm_resource_manager 
*man)
r = amdgpu_ttm_recover_gart(node->tbo);
if (r)
break;
+   if (!adev)
+   adev = amdgpu_ttm_adev(node->tbo->bdev);
}
spin_unlock(>lock);

+   if (adev)
+   amdgpu_gart_invalidate_tlb(adev);
+
return r;
 }

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c0aef327292a..786650a4a493 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1014,6 +1014,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
return r;
}

+   amdgpu_gart_invalidate_tlb(adev);
ttm_resource_free(bo, >mem);
bo->mem = tmp;
}
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 1/1] drm/amdgpu: cleanup gart tlb flush logic

[Nirmoy Das]

Don't flush gpu tlb after recovering each BO instead
do it after receovering all the BOs.

v2: abstract out gart tlb flushing logic to amdgpu_gart.c

Signed-off-by: Nirmoy Das 
---
Is there a better way to get adev in amdgpu_gtt_mgr_recover()?

 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c| 20 +++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h|  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c |  6 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c |  1 +
 4 files changed, 23 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index 5562b5c90c03..e2059f7ed639 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -322,16 +322,26 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t 
offset,
if (!adev->gart.ptr)
return 0;

-   r = amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
-   adev->gart.ptr);
-   if (r)
-   return r;
+   return amdgpu_gart_map(adev, offset, pages, dma_addr, flags,
+  adev->gart.ptr);
+}
+
+/**
+ * amdgpu_gart_tlb_flush - flush gart TLB
+ *
+ * @adev: amdgpu device driver pointer
+ *
+ * Flush TLB of gart page table.
+ *
+ */
+void amdgpu_gart_tlb_flush(struct amdgpu_device *adev)
+{
+   int i;

mb();
amdgpu_asic_flush_hdp(adev, NULL);
for (i = 0; i < adev->num_vmhubs; i++)
amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
-   return 0;
 }

 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
index a25fe97b0196..c853b70a24cf 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.h
@@ -66,5 +66,5 @@ int amdgpu_gart_map(struct amdgpu_device *adev, uint64_t 
offset,
 int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t offset,
 int pages, struct page **pagelist,
 dma_addr_t *dma_addr, uint64_t flags);
-
+void amdgpu_gart_tlb_flush(struct amdgpu_device *adev);
 #endif
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
index 8860545344c7..b61a54f6d95d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c
@@ -205,6 +205,7 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_resource_manager 
*man)
  */
 int amdgpu_gtt_mgr_recover(struct ttm_resource_manager *man)
 {
+   struct amdgpu_device *adev = NULL;
struct amdgpu_gtt_mgr *mgr = to_gtt_mgr(man);
struct amdgpu_gtt_node *node;
struct drm_mm_node *mm_node;
@@ -216,9 +217,14 @@ int amdgpu_gtt_mgr_recover(struct ttm_resource_manager 
*man)
r = amdgpu_ttm_recover_gart(node->tbo);
if (r)
break;
+   if (!adev)
+   adev = amdgpu_ttm_adev(node->tbo->bdev);
}
spin_unlock(>lock);

+   if (adev)
+   amdgpu_gart_tlb_flush(adev);
+
return r;
 }

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c0aef327292a..5e514759d319 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1014,6 +1014,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
return r;
}

+   amdgpu_gart_tlb_flush(adev);
ttm_resource_free(bo, >mem);
bo->mem = tmp;
}
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amdgpu: flush gpu tlb after a gart allocation

[Nirmoy Das]

Flush gpu tlb in amdgpu_ttm_alloc_gart(). Also
don't flush gpu tlb after recovering each BO.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c | 2 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  | 5 -
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
index 5562b5c90c03..68d41063e120 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c
@@ -329,8 +329,6 @@ int amdgpu_gart_bind(struct amdgpu_device *adev, uint64_t 
offset,
 
mb();
amdgpu_asic_flush_hdp(adev, NULL);
-   for (i = 0; i < adev->num_vmhubs; i++)
-   amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
return 0;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c0aef327292a..e68b5dab84a4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -977,7 +977,7 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
struct ttm_placement placement;
struct ttm_place placements;
uint64_t addr, flags;
-   int r;
+   int r, i;
 
if (bo->mem.start != AMDGPU_BO_INVALID_OFFSET)
return 0;
@@ -1014,6 +1014,9 @@ int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo)
return r;
}
 
+   for (i = 0; i < adev->num_vmhubs; i++)
+   amdgpu_gmc_flush_gpu_tlb(adev, 0, i, 0);
+
ttm_resource_free(bo, >mem);
bo->mem = tmp;
}
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 6/6] drm/amdgpu: do not allocate entries separately

[Nirmoy Das]

Allocate PD/PT entries while allocating VM BOs and use that
instead of allocating those entries separately.

v2: create a new var for num entries.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 34 +++---
 1 file changed, 20 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 223c63342ecd..7e478ffb7fdf 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -877,6 +877,7 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
struct amdgpu_bo *bo;
struct dma_resv *resv;
int r;
+   unsigned int num_entries;

memset(, 0, sizeof(bp));

@@ -886,7 +887,14 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
bp.domain = amdgpu_bo_get_preferred_pin_domain(adev, bp.domain);
bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo_vm);
+
+   if (level < AMDGPU_VM_PTB)
+   num_entries = amdgpu_vm_num_entries(adev, level);
+   else
+   num_entries = 0;
+
+   bp.bo_ptr_size = struct_size((*vmbo), entries, num_entries);
+
if (vm->use_cpu_for_update)
bp.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;

@@ -957,19 +965,14 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
struct amdgpu_bo_vm *pt;
int r;

-   if (cursor->level < AMDGPU_VM_PTB && !entry->entries) {
-   unsigned num_entries;
-
-   num_entries = amdgpu_vm_num_entries(adev, cursor->level);
-   entry->entries = kvmalloc_array(num_entries,
-   sizeof(*entry->entries),
-   GFP_KERNEL | __GFP_ZERO);
-   if (!entry->entries)
-   return -ENOMEM;
-   }
-
-   if (entry->base.bo)
+   if (entry->base.bo) {
+   if (cursor->level < AMDGPU_VM_PTB)
+   entry->entries =
+   to_amdgpu_bo_vm(entry->base.bo)->entries;
+   else
+   entry->entries = NULL;
return 0;
+   }

r = amdgpu_vm_pt_create(adev, vm, cursor->level, immediate, );
if (r)
@@ -981,6 +984,10 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
pt_bo = >bo;
pt_bo->parent = amdgpu_bo_ref(cursor->parent->base.bo);
amdgpu_vm_bo_base_init(>base, vm, pt_bo);
+   if (cursor->level < AMDGPU_VM_PTB)
+   entry->entries = pt->entries;
+   else
+   entry->entries = NULL;

r = amdgpu_vm_clear_bo(adev, vm, pt, immediate);
if (r)
@@ -1010,7 +1017,6 @@ static void amdgpu_vm_free_table(struct amdgpu_vm_pt 
*entry)
amdgpu_bo_unref();
amdgpu_bo_unref(>base.bo);
}
-   kvfree(entry->entries);
entry->entries = NULL;
 }

--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 5/6] drm/amdgpu: remove unused code

[Nirmoy Das]

Remove unused code related to shadow BO.

v2: removing shadow bo ptr from base class.

Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 29 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  6 -
 2 files changed, 35 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 89ba612a5080..15cee49f11e2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -638,35 +638,6 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
return r;
 }

-int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-   unsigned long size,
-   struct amdgpu_bo *bo)
-{
-   struct amdgpu_bo_param bp;
-   int r;
-
-   if (bo->shadow)
-   return 0;
-
-   memset(, 0, sizeof(bp));
-   bp.size = size;
-   bp.domain = AMDGPU_GEM_DOMAIN_GTT;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = bo->tbo.base.resv;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >shadow);
-   if (!r) {
-   bo->shadow->parent = amdgpu_bo_ref(bo);
-   mutex_lock(>shadow_list_lock);
-   list_add_tail(>shadow->shadow_list, >shadow_list);
-   mutex_unlock(>shadow_list_lock);
-   }
-
-   return r;
-}
-
 /**
  * amdgpu_bo_create_user - create an _bo_user buffer object
  * @adev: amdgpu device object
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 9afccf6c66f2..fa75251148be 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -104,9 +104,6 @@ struct amdgpu_bo {
struct amdgpu_vm_bo_base*vm_bo;
/* Constant after initialization */
struct amdgpu_bo*parent;
-   struct amdgpu_bo*shadow;
-
-

 #ifdef CONFIG_MMU_NOTIFIER
struct mmu_interval_notifiernotifier;
@@ -300,9 +297,6 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
struct amdgpu_bo_vm **ubo_ptr);
 void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
   void **cpu_addr);
-int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-   unsigned long size,
-   struct amdgpu_bo *bo);
 int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr);
 void *amdgpu_bo_kptr(struct amdgpu_bo *bo);
 void amdgpu_bo_kunmap(struct amdgpu_bo *bo);
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v4 4/6] drm/amdgpu: switch to amdgpu_bo_vm for vm code

[Nirmoy Das]

The subclass, amdgpu_bo_vm is intended for PT/PD BOs which are also
shadowed, so switch to amdgpu_bo_vm BO for PT/PD BOs.

v4: update amdgpu_vm_update_funcs to accept amdgpu_bo_vm.
v3: simplify code.
check also if shadow bo exist instead of checking bo only type.
v2: squash three related patches.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 123 
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h  |   5 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_cpu.c  |  14 +--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c |  19 +--
 4 files changed, 96 insertions(+), 65 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 6bc7566cc193..223c63342ecd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -652,15 +652,15 @@ void amdgpu_vm_move_to_lru_tail(struct amdgpu_device 
*adev,
spin_lock(>mman.bdev.lru_lock);
list_for_each_entry(bo_base, >idle, vm_status) {
struct amdgpu_bo *bo = bo_base->bo;
+   struct amdgpu_bo *shadow = amdgpu_bo_shadowed(bo);

if (!bo->parent)
continue;

ttm_bo_move_to_lru_tail(>tbo, >tbo.mem,
>lru_bulk_move);
-   if (bo->shadow)
-   ttm_bo_move_to_lru_tail(>shadow->tbo,
-   >shadow->tbo.mem,
+   if (shadow)
+   ttm_bo_move_to_lru_tail(>tbo, >tbo.mem,
>lru_bulk_move);
}
spin_unlock(>mman.bdev.lru_lock);
@@ -692,12 +692,13 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,

list_for_each_entry_safe(bo_base, tmp, >evicted, vm_status) {
struct amdgpu_bo *bo = bo_base->bo;
+   struct amdgpu_bo *shadow = amdgpu_bo_shadowed(bo);

r = validate(param, bo);
if (r)
return r;
-   if (bo->shadow) {
-   r = validate(param, bo->shadow);
+   if (shadow) {
+   r = validate(param, shadow);
if (r)
return r;
}
@@ -705,7 +706,7 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
if (bo->tbo.type != ttm_bo_type_kernel) {
amdgpu_vm_bo_moved(bo_base);
} else {
-   vm->update_funcs->map_table(bo);
+   vm->update_funcs->map_table(to_amdgpu_bo_vm(bo));
amdgpu_vm_bo_relocated(bo_base);
}
}
@@ -737,7 +738,7 @@ bool amdgpu_vm_ready(struct amdgpu_vm *vm)
  *
  * @adev: amdgpu_device pointer
  * @vm: VM to clear BO from
- * @bo: BO to clear
+ * @vmbo: BO to clear
  * @immediate: use an immediate update
  *
  * Root PD needs to be reserved when calling this.
@@ -747,13 +748,14 @@ bool amdgpu_vm_ready(struct amdgpu_vm *vm)
  */
 static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
  struct amdgpu_vm *vm,
- struct amdgpu_bo *bo,
+ struct amdgpu_bo_vm *vmbo,
  bool immediate)
 {
struct ttm_operation_ctx ctx = { true, false };
unsigned level = adev->vm_manager.root_level;
struct amdgpu_vm_update_params params;
-   struct amdgpu_bo *ancestor = bo;
+   struct amdgpu_bo *ancestor = >bo;
+   struct amdgpu_bo *bo = >bo;
unsigned entries, ats_entries;
uint64_t addr;
int r;
@@ -793,14 +795,15 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
if (r)
return r;

-   if (bo->shadow) {
-   r = ttm_bo_validate(>shadow->tbo, >shadow->placement,
-   );
+   if (vmbo->shadow) {
+   struct amdgpu_bo *shadow = vmbo->shadow;
+
+   r = ttm_bo_validate(>tbo, >placement, );
if (r)
return r;
}

-   r = vm->update_funcs->map_table(bo);
+   r = vm->update_funcs->map_table(vmbo);
if (r)
return r;

@@ -824,7 +827,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
amdgpu_gmc_get_vm_pde(adev, level, , );
}

-   r = vm->update_funcs->update(, bo, addr, 0, ats_entries,
+   r = vm->update_funcs->update(, vmbo, addr, 0, 
ats_entries,
 value, flags);
if (r)
return r;
@@ -847,7 +850,7 @@ static int amdgpu_vm_clear_bo(struc

[PATCH 3/6] drm/admgpu: add two shadow BO helper functions

[Nirmoy Das]

Add amdgpu_bo_add_to_shadow_list() to handle shadow list
additions and amdgpu_bo_shadowed() to check if a BO is shadowed.

Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 16 
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 17 +
 2 files changed, 33 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 2e5426ab24a8..89ba612a5080 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -760,6 +760,22 @@ int amdgpu_bo_validate(struct amdgpu_bo *bo)
return r;
 }
 
+/**
+ * amdgpu_bo_add_to_shadow_list - add a BO to the shadow list
+ *
+ * @bo: BO that will be inserted into the shadow list
+ *
+ * Insert a BO to the shadow list.
+ */
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo)
+{
+   struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+
+   mutex_lock(>shadow_list_lock);
+   list_add_tail(>shadow_list, >shadow_list);
+   mutex_unlock(>shadow_list_lock);
+}
+
 /**
  * amdgpu_bo_restore_shadow - restore an _bo shadow
  *
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index a7fbf5f7051e..9afccf6c66f2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -259,6 +259,22 @@ static inline bool amdgpu_bo_encrypted(struct amdgpu_bo 
*bo)
return bo->flags & AMDGPU_GEM_CREATE_ENCRYPTED;
 }
 
+/**
+ * amdgpu_bo_shadowed - check if the BO is shadowed
+ *
+ * @bo: BO to be tested.
+ *
+ * Returns:
+ * NULL if not shadowed or else return a BO pointer.
+ */
+static inline struct amdgpu_bo *amdgpu_bo_shadowed(struct amdgpu_bo *bo)
+{
+   if (bo->tbo.type == ttm_bo_type_kernel)
+   return to_amdgpu_bo_vm(bo)->shadow;
+
+   return NULL;
+}
+
 bool amdgpu_bo_is_amdgpu_bo(struct ttm_buffer_object *bo);
 void amdgpu_bo_placement_from_domain(struct amdgpu_bo *abo, u32 domain);
 
@@ -322,6 +338,7 @@ u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
 int amdgpu_bo_validate(struct amdgpu_bo *bo);
 void amdgpu_bo_get_memory(struct amdgpu_bo *bo, uint64_t *vram_mem,
uint64_t *gtt_mem, uint64_t *cpu_mem);
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo);
 int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow,
 struct dma_fence **fence);
 uint32_t amdgpu_bo_get_preferred_pin_domain(struct amdgpu_device *adev,
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 2/6] drm/amdgpu: move shadow bo validation to VM code

[Nirmoy Das]

Do the shadow bo validation in the VM code as
VM code knows/owns shadow BOs.

v2: Fix a typo.

Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 23 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c |  5 +
 2 files changed, 9 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 90136f9dedd6..f6a8f0c5a52f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -396,10 +396,10 @@ void amdgpu_cs_report_moved_bytes(struct amdgpu_device 
*adev, u64 num_bytes,
spin_unlock(>mm_stats.lock);
 }

-static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
-struct amdgpu_bo *bo)
+static int amdgpu_cs_bo_validate(void *param, struct amdgpu_bo *bo)
 {
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+   struct amdgpu_cs_parser *p = param;
struct ttm_operation_ctx ctx = {
.interruptible = true,
.no_wait_gpu = false,
@@ -451,21 +451,6 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser 
*p,
return r;
 }

-static int amdgpu_cs_validate(void *param, struct amdgpu_bo *bo)
-{
-   struct amdgpu_cs_parser *p = param;
-   int r;
-
-   r = amdgpu_cs_bo_validate(p, bo);
-   if (r)
-   return r;
-
-   if (bo->shadow)
-   r = amdgpu_cs_bo_validate(p, bo->shadow);
-
-   return r;
-}
-
 static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
struct list_head *validated)
 {
@@ -493,7 +478,7 @@ static int amdgpu_cs_list_validate(struct amdgpu_cs_parser 
*p,
 lobj->user_pages);
}

-   r = amdgpu_cs_validate(p, bo);
+   r = amdgpu_cs_bo_validate(p, bo);
if (r)
return r;

@@ -593,7 +578,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
p->bytes_moved_vis = 0;

r = amdgpu_vm_validate_pt_bos(p->adev, >vm,
- amdgpu_cs_validate, p);
+ amdgpu_cs_bo_validate, p);
if (r) {
DRM_ERROR("amdgpu_vm_validate_pt_bos() failed.\n");
goto error_validate;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index da155c276c51..6bc7566cc193 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -696,6 +696,11 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = validate(param, bo);
if (r)
return r;
+   if (bo->shadow) {
+   r = validate(param, bo->shadow);
+   if (r)
+   return r;
+   }

if (bo->tbo.type != ttm_bo_type_kernel) {
amdgpu_vm_bo_moved(bo_base);
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/6] drm/amdgpu: add amdgpu_bo_vm bo type

[Nirmoy Das]

Add new BO subclass that will be used by amdgpu vm code.

Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 32 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 10 +++
 2 files changed, 42 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index e9f8701fd046..2e5426ab24a8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -694,6 +694,38 @@ int amdgpu_bo_create_user(struct amdgpu_device *adev,
*ubo_ptr = to_amdgpu_bo_user(bo_ptr);
return r;
 }
+
+/**
+ * amdgpu_bo_create_vm - create an _bo_vm buffer object
+ * @adev: amdgpu device object
+ * @bp: parameters to be used for the buffer object
+ * @vmbo_ptr: pointer to the buffer object pointer
+ *
+ * Create a BO to be for GPUVM.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+
+int amdgpu_bo_create_vm(struct amdgpu_device *adev,
+   struct amdgpu_bo_param *bp,
+   struct amdgpu_bo_vm **vmbo_ptr)
+{
+   struct amdgpu_bo *bo_ptr;
+   int r;
+
+   /* bo_ptr_size will be determined by the caller and it depends on
+* num of amdgpu_vm_pt entries.
+*/
+   BUG_ON(bp->bo_ptr_size < sizeof(struct amdgpu_bo_vm));
+   r = amdgpu_bo_create(adev, bp, _ptr);
+   if (r)
+   return r;
+
+   *vmbo_ptr = to_amdgpu_bo_vm(bo_ptr);
+   return r;
+}
+
 /**
  * amdgpu_bo_validate - validate an _bo buffer object
  * @bo: pointer to the buffer object
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 11480c5a2716..a7fbf5f7051e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -44,6 +44,7 @@
 #define AMDGPU_AMDKFD_CREATE_SVM_BO(1ULL << 62)
 
 #define to_amdgpu_bo_user(abo) container_of((abo), struct amdgpu_bo_user, bo)
+#define to_amdgpu_bo_vm(abo) container_of((abo), struct amdgpu_bo_vm, bo)
 
 struct amdgpu_bo_param {
unsigned long   size;
@@ -125,6 +126,12 @@ struct amdgpu_bo_user {
 
 };
 
+struct amdgpu_bo_vm {
+   struct amdgpu_bobo;
+   struct amdgpu_bo*shadow;
+   struct amdgpu_vm_pt entries[];
+};
+
 static inline struct amdgpu_bo *ttm_to_amdgpu_bo(struct ttm_buffer_object *tbo)
 {
return container_of(tbo, struct amdgpu_bo, tbo);
@@ -272,6 +279,9 @@ int amdgpu_bo_create_kernel_at(struct amdgpu_device *adev,
 int amdgpu_bo_create_user(struct amdgpu_device *adev,
  struct amdgpu_bo_param *bp,
  struct amdgpu_bo_user **ubo_ptr);
+int amdgpu_bo_create_vm(struct amdgpu_device *adev,
+   struct amdgpu_bo_param *bp,
+   struct amdgpu_bo_vm **ubo_ptr);
 void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
   void **cpu_addr);
 int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amdgpu: convert gtt_window_lock to a spinlock

[Nirmoy Das]

amdgpu_device_gpu_recover() will eventually call
gmc_v10_0_flush_gpu_tlb() with holding a spinlock, which
puts the process in atomic context. Sleeping in atomic context
is not allowed so convert gtt_window_lock into a spinlock.

BUG: sleeping function called from invalid context at kernel/locking/mutex.c:935
in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 2662, name: cat
Call Trace:
 dump_stack+0x8b/0xb0
 ___might_sleep.cold+0xb6/0xc6
 ? gmc_v10_0_flush_gpu_tlb+0x72/0x2c0 [amdgpu]
 __mutex_lock+0x45/0x820
 ? amdgpu_device_skip_hw_access+0x3e/0x70 [amdgpu]
 gmc_v10_0_flush_gpu_tlb+0x72/0x2c0 [amdgpu]
 ? amdgpu_device_skip_hw_access+0x3e/0x70 [amdgpu]
 amdgpu_gart_bind+0x7a/0xc0 [amdgpu]
 amdgpu_ttm_gart_bind+0x7d/0xd0 [amdgpu]
 ? amdgpu_ttm_recover_gart+0x2e/0x70 [amdgpu]
 amdgpu_gtt_mgr_recover+0x4e/0x70 [amdgpu]
 amdgpu_do_asic_reset.cold+0x90/0x1c4 [amdgpu]
 ? amdgpu_device_pre_asic_reset+0xa8/0x250 [amdgpu]
 amdgpu_device_gpu_recover.cold+0x7bd/0xa23 [amdgpu]
 ? lock_acquired+0x210/0x390
 gpu_recover_get+0x29/0x60 [amdgpu]
 simple_attr_read+0x69/0xf0
 debugfs_attr_read+0x40/0x60
 full_proxy_read+0x56/0x80
 vfs_read+0xd1/0x1d0
 ksys_read+0x68/0xe0
 do_syscall_64+0x33/0x80
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  6 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h  |  2 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c   | 10 +-
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c |  4 ++--
 4 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c0aef327292a..4cb8fd193b6c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -333,7 +333,7 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,
amdgpu_res_first(src->mem, src->offset, size, _mm);
amdgpu_res_first(dst->mem, dst->offset, size, _mm);
 
-   mutex_lock(>mman.gtt_window_lock);
+   spin_lock(>mman.gtt_window_lock);
while (src_mm.remaining) {
uint32_t src_page_offset = src_mm.start & ~PAGE_MASK;
uint32_t dst_page_offset = dst_mm.start & ~PAGE_MASK;
@@ -373,7 +373,7 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,
amdgpu_res_next(_mm, cur_size);
}
 error:
-   mutex_unlock(>mman.gtt_window_lock);
+   spin_unlock(>mman.gtt_window_lock);
if (f)
*f = dma_fence_get(fence);
dma_fence_put(fence);
@@ -1661,7 +1661,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
int r;
u64 vis_vram_limit;
 
-   mutex_init(>mman.gtt_window_lock);
+   spin_lock_init(>mman.gtt_window_lock);
 
/* No others user of address space so set it to 0 */
r = ttm_device_init(>mman.bdev, _bo_driver, adev->dev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 2877a924086f..afd905dc337b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -70,7 +70,7 @@ struct amdgpu_mman {
struct amdgpu_ring  *buffer_funcs_ring;
boolbuffer_funcs_enabled;
 
-   struct mutexgtt_window_lock;
+   spinlock_t  gtt_window_lock;
/* Scheduler entity for buffer moves */
struct drm_sched_entity entity;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
index ceab5ef50790..f4ce3eb4d7e4 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
@@ -341,11 +341,11 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device 
*adev, uint32_t vmid,
return;
}
 
-   mutex_lock(>mman.gtt_window_lock);
+   spin_lock(>mman.gtt_window_lock);
 
if (vmhub == AMDGPU_MMHUB_0) {
gmc_v10_0_flush_vm_hub(adev, vmid, AMDGPU_MMHUB_0, 0);
-   mutex_unlock(>mman.gtt_window_lock);
+   spin_unlock(>mman.gtt_window_lock);
return;
}
 
@@ -356,7 +356,7 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device 
*adev, uint32_t vmid,
amdgpu_in_reset(adev) ||
ring->sched.ready == false) {
gmc_v10_0_flush_vm_hub(adev, vmid, AMDGPU_GFXHUB_0, 0);
-   mutex_unlock(>mman.gtt_window_lock);
+   spin_unlock(>mman.gtt_window_lock);
return;
}
 
@@ -379,7 +379,7 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device 
*adev, uint32_t vmid,
if (r)
goto error_submit;
 
-   mutex_unlock(>mman.gtt_window_lock);
+   spin_unlock(>mman.gtt_window_lock);
 
dma_fence_wait(fence, false);
dma_fence_put

[PATCH 1/1] drm/amd/display: WARN_ON cleanups

[Nirmoy Das]

Uuse WARN_ON() inside if-condition when possible.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 11 ---
 1 file changed, 4 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c 
b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index f2ed51484980..736eb537a8e4 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -315,8 +315,7 @@ get_crtc_by_otg_inst(struct amdgpu_device *adev,
struct drm_crtc *crtc;
struct amdgpu_crtc *amdgpu_crtc;
 
-   if (otg_inst == -1) {
-   WARN_ON(1);
+   if (WARN_ON(otg_inst == -1)) {
return adev->mode_info.crtcs[0];
}
 
@@ -397,8 +396,7 @@ static void dm_pflip_high_irq(void *interrupt_params)
e = amdgpu_crtc->event;
amdgpu_crtc->event = NULL;
 
-   if (!e)
-   WARN_ON(1);
+   WARN_ON(!e);
 
vrr_active = amdgpu_dm_vrr_active_irq(amdgpu_crtc);
 
@@ -6698,9 +6696,8 @@ static int dm_crtc_helper_atomic_check(struct drm_crtc 
*crtc,
 
dm_update_crtc_active_planes(crtc, crtc_state);
 
-   if (unlikely(!dm_crtc_state->stream &&
-modeset_required(crtc_state, NULL, 
dm_crtc_state->stream))) {
-   WARN_ON(1);
+   if (WARN_ON(unlikely(!dm_crtc_state->stream &&
+modeset_required(crtc_state, NULL, 
dm_crtc_state->stream {
return ret;
}
 
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 1/1] drm/amd/display: do not dereference on NULL

[Nirmoy Das]

WARN_ON() will not kill the process so the second WARN_ON()
will try to dereference NULL pointer.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 5 -
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c 
b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
index 1134cf6eac82..f2ed51484980 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
@@ -9108,7 +9108,10 @@ static void amdgpu_dm_atomic_commit_tail(struct 
drm_atomic_state *state)
}
 
status = dc_stream_get_status(dm_new_crtc_state->stream);
-   WARN_ON(!status);
+
+   if (WARN_ON(!status))
+   continue;
+
WARN_ON(!status->plane_count);
 
/*
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 5/6] drm/amdgpu: remove unused code

[Nirmoy Das]

Remove unused code related to shadow BO.

v2: removing shadow bo ptr from base class.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 29 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  6 -
 2 files changed, 35 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index a63b450cd603..db9c64836556 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -638,35 +638,6 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
return r;
 }

-int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-   unsigned long size,
-   struct amdgpu_bo *bo)
-{
-   struct amdgpu_bo_param bp;
-   int r;
-
-   if (bo->shadow)
-   return 0;
-
-   memset(, 0, sizeof(bp));
-   bp.size = size;
-   bp.domain = AMDGPU_GEM_DOMAIN_GTT;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = bo->tbo.base.resv;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >shadow);
-   if (!r) {
-   bo->shadow->parent = amdgpu_bo_ref(bo);
-   mutex_lock(>shadow_list_lock);
-   list_add_tail(>shadow->shadow_list, >shadow_list);
-   mutex_unlock(>shadow_list_lock);
-   }
-
-   return r;
-}
-
 /**
  * amdgpu_bo_create_user - create an _bo_user buffer object
  * @adev: amdgpu device object
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 9afccf6c66f2..fa75251148be 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -104,9 +104,6 @@ struct amdgpu_bo {
struct amdgpu_vm_bo_base*vm_bo;
/* Constant after initialization */
struct amdgpu_bo*parent;
-   struct amdgpu_bo*shadow;
-
-

 #ifdef CONFIG_MMU_NOTIFIER
struct mmu_interval_notifiernotifier;
@@ -300,9 +297,6 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
struct amdgpu_bo_vm **ubo_ptr);
 void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
   void **cpu_addr);
-int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-   unsigned long size,
-   struct amdgpu_bo *bo);
 int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr);
 void *amdgpu_bo_kptr(struct amdgpu_bo *bo);
 void amdgpu_bo_kunmap(struct amdgpu_bo *bo);
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/6] drm/admgpu: add two shadow BO helper functions

[Nirmoy Das]

Add amdgpu_bo_add_to_shadow_list() to handle shadow list
additions and amdgpu_bo_shadowed() to check if a BO is shadowed.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 16 
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 17 +
 2 files changed, 33 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 6870cc297ae6..a63b450cd603 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -760,6 +760,22 @@ int amdgpu_bo_validate(struct amdgpu_bo *bo)
return r;
 }
 
+/**
+ * amdgpu_bo_add_to_shadow_list - add a BO to the shadow list
+ *
+ * @bo: BO that will be inserted into the shadow list
+ *
+ * Insert a BO to the shadow list.
+ */
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo)
+{
+   struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+
+   mutex_lock(>shadow_list_lock);
+   list_add_tail(>shadow_list, >shadow_list);
+   mutex_unlock(>shadow_list_lock);
+}
+
 /**
  * amdgpu_bo_restore_shadow - restore an _bo shadow
  *
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index a7fbf5f7051e..9afccf6c66f2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -259,6 +259,22 @@ static inline bool amdgpu_bo_encrypted(struct amdgpu_bo 
*bo)
return bo->flags & AMDGPU_GEM_CREATE_ENCRYPTED;
 }
 
+/**
+ * amdgpu_bo_shadowed - check if the BO is shadowed
+ *
+ * @bo: BO to be tested.
+ *
+ * Returns:
+ * NULL if not shadowed or else return a BO pointer.
+ */
+static inline struct amdgpu_bo *amdgpu_bo_shadowed(struct amdgpu_bo *bo)
+{
+   if (bo->tbo.type == ttm_bo_type_kernel)
+   return to_amdgpu_bo_vm(bo)->shadow;
+
+   return NULL;
+}
+
 bool amdgpu_bo_is_amdgpu_bo(struct ttm_buffer_object *bo);
 void amdgpu_bo_placement_from_domain(struct amdgpu_bo *abo, u32 domain);
 
@@ -322,6 +338,7 @@ u64 amdgpu_bo_gpu_offset_no_check(struct amdgpu_bo *bo);
 int amdgpu_bo_validate(struct amdgpu_bo *bo);
 void amdgpu_bo_get_memory(struct amdgpu_bo *bo, uint64_t *vram_mem,
uint64_t *gtt_mem, uint64_t *cpu_mem);
+void amdgpu_bo_add_to_shadow_list(struct amdgpu_bo *bo);
 int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow,
 struct dma_fence **fence);
 uint32_t amdgpu_bo_get_preferred_pin_domain(struct amdgpu_device *adev,
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 2/6] drm/amdgpu: move shadow bo validation to VM code

[Nirmoy Das]

Do the shadow bo validation in the VM code as
VM code knows/owns shadow BOs.

v2: Fix a typo.

Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 23 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c |  5 +
 2 files changed, 9 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 90136f9dedd6..f6a8f0c5a52f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -396,10 +396,10 @@ void amdgpu_cs_report_moved_bytes(struct amdgpu_device 
*adev, u64 num_bytes,
spin_unlock(>mm_stats.lock);
 }

-static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
-struct amdgpu_bo *bo)
+static int amdgpu_cs_bo_validate(void *param, struct amdgpu_bo *bo)
 {
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+   struct amdgpu_cs_parser *p = param;
struct ttm_operation_ctx ctx = {
.interruptible = true,
.no_wait_gpu = false,
@@ -451,21 +451,6 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser 
*p,
return r;
 }

-static int amdgpu_cs_validate(void *param, struct amdgpu_bo *bo)
-{
-   struct amdgpu_cs_parser *p = param;
-   int r;
-
-   r = amdgpu_cs_bo_validate(p, bo);
-   if (r)
-   return r;
-
-   if (bo->shadow)
-   r = amdgpu_cs_bo_validate(p, bo->shadow);
-
-   return r;
-}
-
 static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
struct list_head *validated)
 {
@@ -493,7 +478,7 @@ static int amdgpu_cs_list_validate(struct amdgpu_cs_parser 
*p,
 lobj->user_pages);
}

-   r = amdgpu_cs_validate(p, bo);
+   r = amdgpu_cs_bo_validate(p, bo);
if (r)
return r;

@@ -593,7 +578,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
p->bytes_moved_vis = 0;

r = amdgpu_vm_validate_pt_bos(p->adev, >vm,
- amdgpu_cs_validate, p);
+ amdgpu_cs_bo_validate, p);
if (r) {
DRM_ERROR("amdgpu_vm_validate_pt_bos() failed.\n");
goto error_validate;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index da155c276c51..6bc7566cc193 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -696,6 +696,11 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = validate(param, bo);
if (r)
return r;
+   if (bo->shadow) {
+   r = validate(param, bo->shadow);
+   if (r)
+   return r;
+   }

if (bo->tbo.type != ttm_bo_type_kernel) {
amdgpu_vm_bo_moved(bo_base);
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 6/6] drm/amdgpu: do not allocate entries separately

[Nirmoy Das]

Allocate PD/PT entries while allocating VM BOs and use that
instead of allocating those entries separately.

v2: create a new var for num entries.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 34 +++---
 1 file changed, 20 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index d723873df765..70df453863cb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -876,6 +876,7 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
struct amdgpu_bo *bo;
struct dma_resv *resv;
int r;
+   unsigned int num_entries;

memset(, 0, sizeof(bp));

@@ -885,7 +886,14 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
bp.domain = amdgpu_bo_get_preferred_pin_domain(adev, bp.domain);
bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo_vm);
+
+   if (level < AMDGPU_VM_PTB)
+   num_entries = amdgpu_vm_num_entries(adev, level);
+   else
+   num_entries = 0;
+
+   bp.bo_ptr_size = struct_size((*vmbo), entries, num_entries);
+
if (vm->use_cpu_for_update)
bp.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;

@@ -956,19 +964,14 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
struct amdgpu_bo_vm *pt;
int r;

-   if (cursor->level < AMDGPU_VM_PTB && !entry->entries) {
-   unsigned num_entries;
-
-   num_entries = amdgpu_vm_num_entries(adev, cursor->level);
-   entry->entries = kvmalloc_array(num_entries,
-   sizeof(*entry->entries),
-   GFP_KERNEL | __GFP_ZERO);
-   if (!entry->entries)
-   return -ENOMEM;
-   }
-
-   if (entry->base.bo)
+   if (entry->base.bo) {
+   if (cursor->level < AMDGPU_VM_PTB)
+   entry->entries =
+   to_amdgpu_bo_vm(entry->base.bo)->entries;
+   else
+   entry->entries = NULL;
return 0;
+   }

r = amdgpu_vm_pt_create(adev, vm, cursor->level, immediate, );
if (r)
@@ -980,6 +983,10 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
pt_bo = >bo;
pt_bo->parent = amdgpu_bo_ref(cursor->parent->base.bo);
amdgpu_vm_bo_base_init(>base, vm, pt_bo);
+   if (cursor->level < AMDGPU_VM_PTB)
+   entry->entries = pt->entries;
+   else
+   entry->entries = NULL;

r = amdgpu_vm_clear_bo(adev, vm, pt_bo, immediate);
if (r)
@@ -1009,7 +1016,6 @@ static void amdgpu_vm_free_table(struct amdgpu_vm_pt 
*entry)
amdgpu_bo_unref();
amdgpu_bo_unref(>base.bo);
}
-   kvfree(entry->entries);
entry->entries = NULL;
 }

--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v3 4/6] drm/amdgpu: switch to amdgpu_bo_vm for vm code

[Nirmoy Das]

The subclass, amdgpu_bo_vm is intended for PT/PD BOs which are also
shadowed, so switch to amdgpu_bo_vm BO for PT/PD BOs.

v3: simplify code.
check also if shadow bo exist, instead of checking only bo's type.
v2: squash three related patches.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c  | 93 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 18 ++--
 2 files changed, 68 insertions(+), 43 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 6bc7566cc193..d723873df765 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -652,15 +652,15 @@ void amdgpu_vm_move_to_lru_tail(struct amdgpu_device 
*adev,
spin_lock(>mman.bdev.lru_lock);
list_for_each_entry(bo_base, >idle, vm_status) {
struct amdgpu_bo *bo = bo_base->bo;
+   struct amdgpu_bo *shadow = amdgpu_bo_shadowed(bo);

if (!bo->parent)
continue;

ttm_bo_move_to_lru_tail(>tbo, >tbo.mem,
>lru_bulk_move);
-   if (bo->shadow)
-   ttm_bo_move_to_lru_tail(>shadow->tbo,
-   >shadow->tbo.mem,
+   if (shadow)
+   ttm_bo_move_to_lru_tail(>tbo, >tbo.mem,
>lru_bulk_move);
}
spin_unlock(>mman.bdev.lru_lock);
@@ -692,12 +692,13 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,

list_for_each_entry_safe(bo_base, tmp, >evicted, vm_status) {
struct amdgpu_bo *bo = bo_base->bo;
+   struct amdgpu_bo *shadow = amdgpu_bo_shadowed(bo);

r = validate(param, bo);
if (r)
return r;
-   if (bo->shadow) {
-   r = validate(param, bo->shadow);
+   if (shadow) {
+   r = validate(param, shadow);
if (r)
return r;
}
@@ -754,6 +755,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
unsigned level = adev->vm_manager.root_level;
struct amdgpu_vm_update_params params;
struct amdgpu_bo *ancestor = bo;
+   struct amdgpu_bo *shadow;
unsigned entries, ats_entries;
uint64_t addr;
int r;
@@ -793,9 +795,9 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
if (r)
return r;

-   if (bo->shadow) {
-   r = ttm_bo_validate(>shadow->tbo, >shadow->placement,
-   );
+   shadow = amdgpu_bo_shadowed(bo);
+   if (shadow) {
+   r = ttm_bo_validate(>tbo, >placement, );
if (r)
return r;
}
@@ -863,14 +865,16 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev,
  * @vm: requesting vm
  * @level: the page table level
  * @immediate: use a immediate update
- * @bo: pointer to the buffer object pointer
+ * @vmbo: pointer to the buffer object pointer
  */
 static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
   struct amdgpu_vm *vm,
   int level, bool immediate,
-  struct amdgpu_bo **bo)
+  struct amdgpu_bo_vm **vmbo)
 {
struct amdgpu_bo_param bp;
+   struct amdgpu_bo *bo;
+   struct dma_resv *resv;
int r;

memset(, 0, sizeof(bp));
@@ -881,7 +885,7 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
bp.domain = amdgpu_bo_get_preferred_pin_domain(adev, bp.domain);
bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
+   bp.bo_ptr_size = sizeof(struct amdgpu_bo_vm);
if (vm->use_cpu_for_update)
bp.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;

@@ -890,26 +894,41 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
if (vm->root.base.bo)
bp.resv = vm->root.base.bo->tbo.base.resv;

-   r = amdgpu_bo_create(adev, , bo);
+   r = amdgpu_bo_create_vm(adev, , vmbo);
if (r)
return r;

-   if (vm->is_compute_context && (adev->flags & AMD_IS_APU))
+   bo = &(*vmbo)->bo;
+   if (vm->is_compute_context && (adev->flags & AMD_IS_APU)) {
+   (*vmbo)->shadow = NULL;
return 0;
+   }

if (!bp.resv)
-   WARN_ON(dma_resv_lock((*bo)->tbo.base.resv,
+   WARN_ON(dma_resv_lock(bo->tbo.base.resv,
   

[PATCH 1/6] drm/amdgpu: add amdgpu_bo_vm bo type

[Nirmoy Das]

Add new BO subclass that will be used by amdgpu vm code.

Signed-off-by: Nirmoy Das 
Reviewed-by: Christian König 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 32 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 10 +++
 2 files changed, 42 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 3f85ba8222ef..6870cc297ae6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -694,6 +694,38 @@ int amdgpu_bo_create_user(struct amdgpu_device *adev,
*ubo_ptr = to_amdgpu_bo_user(bo_ptr);
return r;
 }
+
+/**
+ * amdgpu_bo_create_vm - create an _bo_vm buffer object
+ * @adev: amdgpu device object
+ * @bp: parameters to be used for the buffer object
+ * @vmbo_ptr: pointer to the buffer object pointer
+ *
+ * Create a BO to be for GPUVM.
+ *
+ * Returns:
+ * 0 for success or a negative error code on failure.
+ */
+
+int amdgpu_bo_create_vm(struct amdgpu_device *adev,
+   struct amdgpu_bo_param *bp,
+   struct amdgpu_bo_vm **vmbo_ptr)
+{
+   struct amdgpu_bo *bo_ptr;
+   int r;
+
+   /* bo_ptr_size will be determined by the caller and it depends on
+* num of amdgpu_vm_pt entries.
+*/
+   BUG_ON(bp->bo_ptr_size < sizeof(struct amdgpu_bo_vm));
+   r = amdgpu_bo_create(adev, bp, _ptr);
+   if (r)
+   return r;
+
+   *vmbo_ptr = to_amdgpu_bo_vm(bo_ptr);
+   return r;
+}
+
 /**
  * amdgpu_bo_validate - validate an _bo buffer object
  * @bo: pointer to the buffer object
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 11480c5a2716..a7fbf5f7051e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -44,6 +44,7 @@
 #define AMDGPU_AMDKFD_CREATE_SVM_BO(1ULL << 62)
 
 #define to_amdgpu_bo_user(abo) container_of((abo), struct amdgpu_bo_user, bo)
+#define to_amdgpu_bo_vm(abo) container_of((abo), struct amdgpu_bo_vm, bo)
 
 struct amdgpu_bo_param {
unsigned long   size;
@@ -125,6 +126,12 @@ struct amdgpu_bo_user {
 
 };
 
+struct amdgpu_bo_vm {
+   struct amdgpu_bobo;
+   struct amdgpu_bo*shadow;
+   struct amdgpu_vm_pt entries[];
+};
+
 static inline struct amdgpu_bo *ttm_to_amdgpu_bo(struct ttm_buffer_object *tbo)
 {
return container_of(tbo, struct amdgpu_bo, tbo);
@@ -272,6 +279,9 @@ int amdgpu_bo_create_kernel_at(struct amdgpu_device *adev,
 int amdgpu_bo_create_user(struct amdgpu_device *adev,
  struct amdgpu_bo_param *bp,
  struct amdgpu_bo_user **ubo_ptr);
+int amdgpu_bo_create_vm(struct amdgpu_device *adev,
+   struct amdgpu_bo_param *bp,
+   struct amdgpu_bo_vm **ubo_ptr);
 void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
   void **cpu_addr);
 int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 5/5] drm/amdgpu: do not allocate entries separately

[Nirmoy Das]

Allocate PD/PT entries while allocating VM BOs and use that
instead of allocating those entries separately.

v2: create a new var for num entries.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 34 +++---
 1 file changed, 20 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 80d50e6d75f9..de9dd882ecdc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -876,6 +876,7 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
struct amdgpu_bo *shadow_bo;
struct dma_resv *resv;
int r;
+   unsigned int num_entries;
 
memset(, 0, sizeof(bp));
 
@@ -885,7 +886,14 @@ static int amdgpu_vm_pt_create(struct amdgpu_device *adev,
bp.domain = amdgpu_bo_get_preferred_pin_domain(adev, bp.domain);
bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo_vm);
+
+   if (level < AMDGPU_VM_PTB)
+   num_entries = amdgpu_vm_num_entries(adev, level);
+   else
+   num_entries = 0;
+
+   bp.bo_ptr_size = struct_size((*vmbo), entries, num_entries);
+
if (vm->use_cpu_for_update)
bp.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
 
@@ -958,19 +966,14 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
struct amdgpu_bo_vm *pt;
int r;
 
-   if (cursor->level < AMDGPU_VM_PTB && !entry->entries) {
-   unsigned num_entries;
-
-   num_entries = amdgpu_vm_num_entries(adev, cursor->level);
-   entry->entries = kvmalloc_array(num_entries,
-   sizeof(*entry->entries),
-   GFP_KERNEL | __GFP_ZERO);
-   if (!entry->entries)
-   return -ENOMEM;
-   }
-
-   if (entry->base.bo)
+   if (entry->base.bo) {
+   if (cursor->level < AMDGPU_VM_PTB)
+   entry->entries =
+   to_amdgpu_bo_vm(entry->base.bo)->entries;
+   else
+   entry->entries = NULL;
return 0;
+   }
 
r = amdgpu_vm_pt_create(adev, vm, cursor->level, immediate, );
if (r)
@@ -982,6 +985,10 @@ static int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
pt_bo = >bo;
pt_bo->parent = amdgpu_bo_ref(cursor->parent->base.bo);
amdgpu_vm_bo_base_init(>base, vm, pt_bo);
+   if (cursor->level < AMDGPU_VM_PTB)
+   entry->entries = pt->entries;
+   else
+   entry->entries = NULL;
 
r = amdgpu_vm_clear_bo(adev, vm, pt_bo, immediate);
if (r)
@@ -1009,7 +1016,6 @@ static void amdgpu_vm_free_table(struct amdgpu_vm_pt 
*entry)

amdgpu_bo_unref(_amdgpu_bo_vm(entry->base.bo)->shadow);
amdgpu_bo_unref(>base.bo);
}
-   kvfree(entry->entries);
entry->entries = NULL;
 }
 
-- 
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 4/5] drm/amdgpu: remove unused code

[Nirmoy Das]

Remove unused code related to shadow BO.

v2: removing shadow bo ptr from base class.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 29 --
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.h |  6 -
 2 files changed, 35 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 6870cc297ae6..7930b7d9a3b1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -638,35 +638,6 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
return r;
 }

-int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-   unsigned long size,
-   struct amdgpu_bo *bo)
-{
-   struct amdgpu_bo_param bp;
-   int r;
-
-   if (bo->shadow)
-   return 0;
-
-   memset(, 0, sizeof(bp));
-   bp.size = size;
-   bp.domain = AMDGPU_GEM_DOMAIN_GTT;
-   bp.flags = AMDGPU_GEM_CREATE_CPU_GTT_USWC;
-   bp.type = ttm_bo_type_kernel;
-   bp.resv = bo->tbo.base.resv;
-   bp.bo_ptr_size = sizeof(struct amdgpu_bo);
-
-   r = amdgpu_bo_create(adev, , >shadow);
-   if (!r) {
-   bo->shadow->parent = amdgpu_bo_ref(bo);
-   mutex_lock(>shadow_list_lock);
-   list_add_tail(>shadow->shadow_list, >shadow_list);
-   mutex_unlock(>shadow_list_lock);
-   }
-
-   return r;
-}
-
 /**
  * amdgpu_bo_create_user - create an _bo_user buffer object
  * @adev: amdgpu device object
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index a7fbf5f7051e..07478f86f05a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -104,9 +104,6 @@ struct amdgpu_bo {
struct amdgpu_vm_bo_base*vm_bo;
/* Constant after initialization */
struct amdgpu_bo*parent;
-   struct amdgpu_bo*shadow;
-
-

 #ifdef CONFIG_MMU_NOTIFIER
struct mmu_interval_notifiernotifier;
@@ -284,9 +281,6 @@ int amdgpu_bo_create_vm(struct amdgpu_device *adev,
struct amdgpu_bo_vm **ubo_ptr);
 void amdgpu_bo_free_kernel(struct amdgpu_bo **bo, u64 *gpu_addr,
   void **cpu_addr);
-int amdgpu_bo_create_shadow(struct amdgpu_device *adev,
-   unsigned long size,
-   struct amdgpu_bo *bo);
 int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr);
 void *amdgpu_bo_kptr(struct amdgpu_bo *bo);
 void amdgpu_bo_kunmap(struct amdgpu_bo *bo);
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH v2 2/5] drm/amdgpu: move shadow bo validation to VM code

[Nirmoy Das]

Do the shadow bo validation in the VM code as
VM code knows/owns shadow BOs.

v2: Fix a typo.

Signed-off-by: Nirmoy Das 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 23 ---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c |  5 +
 2 files changed, 9 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 90136f9dedd6..f6a8f0c5a52f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -396,10 +396,10 @@ void amdgpu_cs_report_moved_bytes(struct amdgpu_device 
*adev, u64 num_bytes,
spin_unlock(>mm_stats.lock);
 }

-static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser *p,
-struct amdgpu_bo *bo)
+static int amdgpu_cs_bo_validate(void *param, struct amdgpu_bo *bo)
 {
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
+   struct amdgpu_cs_parser *p = param;
struct ttm_operation_ctx ctx = {
.interruptible = true,
.no_wait_gpu = false,
@@ -451,21 +451,6 @@ static int amdgpu_cs_bo_validate(struct amdgpu_cs_parser 
*p,
return r;
 }

-static int amdgpu_cs_validate(void *param, struct amdgpu_bo *bo)
-{
-   struct amdgpu_cs_parser *p = param;
-   int r;
-
-   r = amdgpu_cs_bo_validate(p, bo);
-   if (r)
-   return r;
-
-   if (bo->shadow)
-   r = amdgpu_cs_bo_validate(p, bo->shadow);
-
-   return r;
-}
-
 static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
struct list_head *validated)
 {
@@ -493,7 +478,7 @@ static int amdgpu_cs_list_validate(struct amdgpu_cs_parser 
*p,
 lobj->user_pages);
}

-   r = amdgpu_cs_validate(p, bo);
+   r = amdgpu_cs_bo_validate(p, bo);
if (r)
return r;

@@ -593,7 +578,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
p->bytes_moved_vis = 0;

r = amdgpu_vm_validate_pt_bos(p->adev, >vm,
- amdgpu_cs_validate, p);
+ amdgpu_cs_bo_validate, p);
if (r) {
DRM_ERROR("amdgpu_vm_validate_pt_bos() failed.\n");
goto error_validate;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index da155c276c51..6bc7566cc193 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -696,6 +696,11 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
r = validate(param, bo);
if (r)
return r;
+   if (bo->shadow) {
+   r = validate(param, bo->shadow);
+   if (r)
+   return r;
+   }

if (bo->tbo.type != ttm_bo_type_kernel) {
amdgpu_vm_bo_moved(bo_base);
--
2.31.1

___
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx