Re: [PATCH 3/9] drm/amdkfd: Make IOMMUv2 code conditional

On Wed, Feb 7, 2018 at 2:30 AM, Felix Kuehling wrote: > On 2018-02-06 03:53 AM, Oded Gabbay wrote: >> On Mon, Feb 5, 2018 at 9:00 PM, Christian König >> wrote: >>> Looks good to me on first glance. >>> >>> You probably don't mind that I'm

Re: [PATCH 0/5] prevent OOM triggered by TTM

Hi, Roger, On 02/06/2018 10:04 AM, Roger He wrote: currently ttm code has no any allocation limit. So it allows pages allocatation unlimited until OOM. Because if swap space is full of swapped pages and then system memory will be filled up with ttm pages. and then any memory allocation request

[PATCH 03/25] drm/amdgpu: Fix header file dependencies

Signed-off-by: Felix Kuehling Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 1 + 2 files changed, 2 insertions(+) diff --git

[PATCH 22/25] drm/amdkfd: Add TC flush on VMID deallocation for Hawaii

On GFX7 the CP does not perform a TC flush when queues are unmapped. To avoid TC eviction from accessing an invalid VMID, flush it explicitly before releasing a VMID. v2: * Fix unnecessary list_for_each_entry_safe Signed-off-by: Amber Lin Signed-off-by: Felix Kuehling

[PATCH 08/25] drm/amdgpu: add amdgpu_sync_clone

Cloning a sync object is useful for waiting for a sync object without locking the original structure indefinitely, blocking other threads. Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c | 35

[PATCH 17/25] uapi: Fix type used in ioctl parameter structures

Use __u32 and __u64 instead of POSIX types that may not be defined in user mode builds. Signed-off-by: Felix Kuehling --- include/uapi/linux/kfd_ioctl.h | 8 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/include/uapi/linux/kfd_ioctl.h

[PATCH 12/25] drm/amdkfd: Use per-device sched_policy

This was missed in a previous commit that made the scheduler policy a per-device setting. Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git

[PATCH 13/25] drm/amdkfd: Remove unaligned memory access

From: Harish Kasiviswanathan Unaligned atomic operations can cause problems on some CPU architectures. Use simpler bitmask operations instead. Atomic bit manipulations are not necessary since dqm->lock is held during these operations. Signed-off-by: Harish

[PATCH 20/25] drm/amdkfd: Add per-process IDR for buffer handles

Also used for cleaning up on process termination. Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_priv.h| 11 ++ drivers/gpu/drm/amd/amdkfd/kfd_process.c | 66 2 files changed, 77 insertions(+) diff --git

[PATCH 25/25] drm/amdkfd: Add module option for testing large-BAR functionality

Simulate large-BAR system by exporting only visible memory. This limits the amount of available VRAM to the size of the BAR, but enables CPU access to VRAM. Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 5 +

[PATCH 15/25] drm/amdkfd: Add GPUVM virtual address space to PDD

Create/destroy the GPUVM context during PDD creation/destruction. Get VM page table base and program it during process registration (HWS) or VMID allocation (non-HWS). v2: * Used dev instead of pdd->dev in kfd_flush_tlb Signed-off-by: Felix Kuehling ---

[PATCH 24/25] drm/amdkfd: Kmap event page for dGPUs

The events page must be accessible in user mode by the GPU and CPU as well as in kernel mode by the CPU. On dGPUs user mode virtual addresses are managed by the Thunk's GPU memory allocation code. Therefore we can't allocate the memory in kernel mode like we do on APUs. But KFD still needs to map

[PATCH 23/25] drm/amdkfd: Add ioctls for GPUVM memory management

v2: * Fix error handling after kfd_bind_process_to_device in kfd_ioctl_map_memory_to_gpu Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_chardev.c| 329 drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 8 +

[PATCH 01/25] drm/amdgpu: remove useless BUG_ONs

Dereferencing NULL pointers will cause a BUG anyway. No need to do an explicit check. Signed-off-by: Felix Kuehling Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c| 6 --

[PATCH 16/25] drm/amdkfd: Implement KFD process eviction/restore

When the TTM memory manager in KGD evicts BOs, all user mode queues potentially accessing these BOs must be evicted temporarily. Once user mode queues are evicted, the eviction fence is signaled, allowing the migration of the BO to proceed. A delayed worker is scheduled to restore all the BOs

[PATCH 09/25] drm/amdgpu: Add GPUVM memory management functions for KFD

v2: * Removed unused flags from struct kgd_mem * Updated some comments * Added a check to unmap_memory_from_gpu whether BO was mapped Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdgpu/Makefile |1 + drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c

[PATCH 04/25] drm/amdgpu: Fix wrong mask in get_atc_vmid_pasid_mapping_pasid

Signed-off-by: Felix Kuehling Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git

[PATCH 00/25] Add KFD GPUVM support for dGPUs v2

Changes since v1: * Rebased * Several fixes * Conditional IOMMU change carried over from previous patch series and updated AMDGPU: Patches 1-5 are minor cleanups and fixes Patches 6-10 add and implement KFD->KGD interfaces for GPUVM AMDKFD: Patch 11: Make IOMMU conditional (carried over from

[PATCH 14/25] drm/amdkfd: Populate DRM render device minor

From: Oak Zeng Populate DRM render device minor in kfd topology Signed-off-by: Oak Zeng Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 4 drivers/gpu/drm/amd/amdkfd/kfd_topology.h | 1 + 2 files

[PATCH 07/25] drm/amdgpu: Update kgd2kfd_shared_resources for dGPU support

Add GPUVM size and DRM render node. Also add function to query the VMID mask to avoid hard-coding it in multiple places later. v2: * Cut off GPUVM size at the VA hole Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 20

[PATCH 05/25] drm/amdgpu: Remove unused kfd2kgd interface

Signed-off-by: Felix Kuehling Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 9 - drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 10 -- drivers/gpu/drm/amd/include/kgd_kfd_interface.h

[PATCH 11/25] drm/amdkfd: Centralize IOMMUv2 code and make it conditional

dGPUs work without IOMMUv2. Make IOMMUv2 initialization dependent on ASIC information. Also allow building KFD without IOMMUv2 support. This is still useful for dGPUs and prepares for enabling KFD on architectures that don't support AMD IOMMUv2. v2: * Centralize IOMMUv2 code to avoid #ifdefs in

[PATCH 10/25] drm/amdgpu: Add submit IB function for KFD

This can be used for flushing caches when not using the HWS. Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c| 55 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 4 ++

[PATCH 18/25] drm/amdkfd: Remove limit on number of GPUs

Currently the number of GPUs is limited by aperture placement options available on GFX7 and GFX8 hardware. This limitation is not necessary. Scratch and LDS represent per-work-item and per-work-group storage respectively. Different work-items and work-groups use the same virtual address to access

[PATCH 19/25] drm/amdkfd: Aperture setup for dGPUs

Set up the GPUVM aperture for SVM (shared virtual memory) that allows sharing a part of virtual address space between GPUs and CPUs. Report the size of the the GPUVM size supported by KGD accurately. The low part of the GPUVM aperture is reserved for kernel use. This is for kernel-allocated

[PATCH 21/25] drm/amdkfd: Allocate CWSR trap handler memory for dGPUs

Add helpers for allocating GPUVM memory in kernel mode and use them to allocate memory for the CWSR trap handler. v2: * Use dev instead of pdd->dev in kfd_process_free_gpuvm Signed-off-by: Felix Kuehling --- drivers/gpu/drm/amd/amdkfd/kfd_process.c | 125

[PATCH 06/25] drm/amdgpu: Add KFD eviction fence

This fence is used by KFD to keep memory resident while user mode queues are enabled. Trying to evict memory will trigger the enable_signaling callback, which starts a KFD eviction, which involves preempting user mode queues before signaling the fence. There is one such fence per process. v2: *

[PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

Add common smu_soc_asic_init function to emulate the sillicon post sequence Change-Id: I6ff04e1199d1ebdbdb31d0e7e8ca3c240c61ab3a Signed-off-by: Shaoyun Liu --- drivers/gpu/drm/amd/amdgpu/Makefile| 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu.h| 2 ++

RE: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

> -Original Message- > From: Liu, Shaoyun > Sent: Tuesday, February 6, 2018 5:42 PM > To: Deucher, Alexander > Cc: amd-gfx list ; Bridgman, John > > Subject: RE: [PATCH] drm/amdgpu: Add place holder for

Re: [PATCH] drm/amdgpu: Avoid get vram info from atom bios on emulation mode

Alternatively, you could also just return 0 in amdgpu_atomfirmware_get_vram_width() if emu == 1. Either way: Reviewed-by: Alex Deucher From: amd-gfx on behalf of Shaoyun Liu

RE: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

Thanks . Can I add Review-By you for this patch ? Shaoyun.liu -Original Message- From: Deucher, Alexander Sent: Tuesday, February 06, 2018 5:37 PM To: Liu, Shaoyun; Alex Deucher Cc: amd-gfx list; Bridgman, John Subject: RE: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on

RE: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

> -Original Message- > From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf > Of Liu, Shaoyun > Sent: Tuesday, February 6, 2018 5:26 PM > To: Alex Deucher > Cc: amd-gfx list ; Bridgman, John >

[PATCH] drm/amdgpu: Avoid get vram info from atom bios on emulation mode

Change-Id: I10030e83d63b2c27bfd16b64178e4f07c13addfa Signed-off-by: Shaoyun Liu --- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c

RE: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

You are right , I try to use the emu_soc_asic_init()as a entry point for any family . For driver to get into amdgpu_atom_asic_init() function call , it depends on other logic like read bios and check some register etc . On emulation mode , we can not get bios image so it will requires

Re: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

On Tue, Feb 6, 2018 at 4:55 PM, Liu, Shaoyun wrote: > Ye, I try to put the asic specific register sequences for emulation in a > separate emu_soc.c file, I think it's suggested by John or you ? The asic > specific code will be keep in separate bring up branch , but a

Re: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

Yes, I made that suggestion in order to make it easy for us to replace the massive function we use in emulation with a stub function for upstream. I figured if we had that function in a separate file it would be cleaner, but maybe that is not the case. I was thinking that we could replace the

RE: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

Ye, I try to put the asic specific register sequences for emulation in a separate emu_soc.c file, I think it's suggested by John or you ? The asic specific code will be keep in separate bring up branch , but a common place to add them seems not bad to me . Shaoyun.liu -Original

Re: [PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

On Tue, Feb 6, 2018 at 4:41 PM, Shaoyun Liu wrote: > Change-Id: I6ff04e1199d1ebdbdb31d0e7e8ca3c240c61ab3a > Signed-off-by: Shaoyun Liu What is the purpose of this? If it's just as a hint the the driver writer as to where to add the emulation register

[PATCH] drm/amdgpu: Add place holder for soc15 asic init on emulation

Change-Id: I6ff04e1199d1ebdbdb31d0e7e8ca3c240c61ab3a Signed-off-by: Shaoyun Liu --- drivers/gpu/drm/amd/amdgpu/Makefile | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 ++ drivers/gpu/drm/amd/amdgpu/emu_soc.c | 33 +

Re: [PATCH 1/3] drm/amdgpu: add mmhub powergating set by smu

Yes. SMU only provides enabling message. Thanks, Eric On 2018-02-06 03:50 PM, Deucher, Alexander wrote: -Original Message- From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf Of Eric Huang Sent: Tuesday, February 6, 2018 3:41 PM To: amd-gfx@lists.freedesktop.org Cc:

RE: [PATCH 3/3] drm/amd/powerplay: implement set_mmhub_powergating_by_smu for Raven

> -Original Message- > From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf > Of Eric Huang > Sent: Tuesday, February 6, 2018 3:41 PM > To: amd-gfx@lists.freedesktop.org > Cc: Huang, JinHuiEric > Subject: [PATCH 3/3] drm/amd/powerplay: implement

RE: [PATCH 2/3] drm/amd/powerplay: add mmhub powergating by smu in powerplay

> -Original Message- > From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf > Of Eric Huang > Sent: Tuesday, February 6, 2018 3:41 PM > To: amd-gfx@lists.freedesktop.org > Cc: Huang, JinHuiEric > Subject: [PATCH 2/3] drm/amd/powerplay: add mmhub

RE: [PATCH 1/3] drm/amdgpu: add mmhub powergating set by smu

> -Original Message- > From: amd-gfx [mailto:amd-gfx-boun...@lists.freedesktop.org] On Behalf > Of Eric Huang > Sent: Tuesday, February 6, 2018 3:41 PM > To: amd-gfx@lists.freedesktop.org > Cc: Huang, JinHuiEric > Subject: [PATCH 1/3] drm/amdgpu: add mmhub

[PATCH 1/3] drm/amdgpu: add mmhub powergating set by smu

PSP will disable legacy mmhub PG setting that is programming registers, and drive will use SMC message to set it up. Signed-off-by: Eric Huang --- drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h| 4 drivers/gpu/drm/amd/amdgpu/mmhub_v1_0.c| 3 +++

[PATCH 3/3] drm/amd/powerplay: implement set_mmhub_powergating_by_smu for Raven

Signed-off-by: Eric Huang --- drivers/gpu/drm/amd/powerplay/hwmgr/rv_hwmgr.c | 6 ++ drivers/gpu/drm/amd/powerplay/inc/rv_ppsmc.h | 3 ++- 2 files changed, 8 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/amd/powerplay/hwmgr/rv_hwmgr.c

[PATCH 2/3] drm/amd/powerplay: add mmhub powergating by smu in powerplay

Signed-off-by: Eric Huang --- drivers/gpu/drm/amd/powerplay/amd_powerplay.c | 22 ++ drivers/gpu/drm/amd/powerplay/inc/hwmgr.h | 1 + 2 files changed, 23 insertions(+) diff --git a/drivers/gpu/drm/amd/powerplay/amd_powerplay.c

Re: [PATCH] drm/amdgpu: Fix none-powerplay issue when load driver on emulation mode

On Tue, Feb 6, 2018 at 2:34 PM, Shaoyun Liu wrote: > On emulation mode , driver will be loaded with powerplay disabled > > Change-Id: I22c343d990f9a306b033728a1cb560c1e8a4677f > Signed-off-by: Shaoyun Liu Reviewed-by: Alex Deucher

[PATCH] drm/amdgpu: Fix none-powerplay issue when load driver on emulation mode

On emulation mode , driver will be loaded with powerplay disabled Change-Id: I22c343d990f9a306b033728a1cb560c1e8a4677f Signed-off-by: Shaoyun Liu --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 8 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 3 +++

Re: [PATCH] drm/amd/powerplay: delete some dead code

On Tue, Feb 6, 2018 at 7:35 AM, Dan Carpenter wrote: > We deleted some code in e154162ef75d ("drm/amd/powerplay: refine pp code > for raven") but there were a few related bits that were missed. Let's > delete them as well. > > Signed-off-by: Dan Carpenter

Re: [PATCH] drm/amd/powerplay: Zero out power data in AMDGPU_PP_SENSOR_GPU_POWER readout

Reviewed-by: Alex Deucher From: amd-gfx on behalf of Tom St Denis Sent: Tuesday, February 6, 2018 2:09 PM To: amd-gfx@lists.freedesktop.org Cc: StDenis, Tom Subject:

[PATCH] drm/amd/powerplay: Zero out power data in AMDGPU_PP_SENSOR_GPU_POWER readout

Since 12 of the 16 bytes are not initialized with anything let's ensure they're sensibly zeroed out otherwise debugfs callers will read back garbage (because they assume debugfs wrote sensible data back...) Signed-off-by: Tom St Denis ---

Re: [PATCH] drm/amdgpu: Fix no irq process when evict vram

Yintian, Any plans to commit this? Alex On Wed, Dec 13, 2017 at 11:36 AM, Deucher, Alexander wrote: > This seems like a good thing to do in general even with Monk's fix. > > > Acked-by: Alex Deucher > > >

RE: [PATCH 0/5] prevent OOM triggered by TTM

Move the new call out of ttm_mem_global_reserve() and into ttm_page_alloc.c or ttm_page_alloc_dma.c (but keep it in ttm_memory.c). ttm_mem_global_reserve() is called for each page allocated and si_mem_available() is a bit to heavy for that. Good idea! Agree with you completely,

Re: [PATCH 0/5] prevent OOM triggered by TTM

Nice work, but a few comments. First of all you need to reorder the patches. Adding the exceptions to the restrictions should come first, then the restriction itself. Otherwise we might break a setup in between the patches and that is bad for bisecting. Then make all values configurable,

[PATCH 5/5] drm/ttm: add input parameter allow_allo_anyway for ttm_bo_evict_mm

if true for it, allocate TTM pages regardless of zone global memory account limit. that is for another special case: suspend. doesn't care the zone global memory account limit for this case. Signed-off-by: Roger He --- drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 2 +-

[PATCH 4/5] drm/ttm: add bit flag TTM_OPT_FLAG_ALLOW_ALLOC_ANYWAY

set TTM_OPT_FLAG_ALLOW_ALLOC_ANYWAY when we are servicing for page fault routine. for ttm_mem_global_reserve if in page fault routine, allow the gtt pages reservation always. because page fault routing already grabbed system memory and the allowance of this exception is harmless. Otherwise, it

[PATCH 3/5] drm/ttm: use bit flag to replace allow_reserved_eviction in ttm_operation_ctx

for saving memory and more bit flag can be used in future Signed-off-by: Roger He --- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 ++-- drivers/gpu/drm/ttm/ttm_bo.c | 3 ++- include/drm/ttm/ttm_bo_api.h

[PATCH 1/5] drm/ttm: check if the free swap space is under limit 256MB

for avoid OOM. if free swap space is less than 256MB, reject the TTM page allocation. Otherwise, swap space will be full of swapped pages and then system memory will be filled up with ttm pages. and then any memory allocation request will trigger OOM. to cover two cases: a. If total swap space >

[PATCH 2/5] drm/ttm: keep original behavior except with flag no_retry

set the no_retry flag in struct ttm_mem_global and init it after ttm_mem_global_init Signed-off-by: Roger He --- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 8 +--- drivers/gpu/drm/ttm/ttm_memory.c| 3 +++ include/drm/ttm/ttm_memory.h| 3 +++ 3 files

[PATCH 0/5] prevent OOM triggered by TTM

currently ttm code has no any allocation limit. So it allows pages allocatation unlimited until OOM. Because if swap space is full of swapped pages and then system memory will be filled up with ttm pages. and then any memory allocation request will trigger OOM. the following patches is for

Re: [PATCH 3/9] drm/amdkfd: Make IOMMUv2 code conditional

On Mon, Feb 5, 2018 at 9:00 PM, Christian König wrote: > Looks good to me on first glance. > > You probably don't mind that I'm going to pull a good part of that into > amdgpu as next step? > That indeed looks better then the first approach. Felix, I've applied

Re: [PATCH 7/9] drm/amdkfd: Add dGPU support to kernel_queue_init

On Wed, Jan 31, 2018 at 6:27 PM, Felix Kuehling wrote: > On 2018-01-31 10:29 AM, Oded Gabbay wrote: >> On Wed, Jan 31, 2018 at 5:23 PM, Deucher, Alexander >> wrote: >>> >>> From: amd-gfx