As I understand, if we want to implement the logic in bo_add/rmv
function , I need following conditions are all to be true for a valid
XGMI request.
1. check the adev and the bo_adev are different .
This is correct on rocm implementation .(
amdgpu_amdkfd_gpuvm_map_memory_to_gpu pass
On 2019-03-18 5:58 p.m., Paul Menzel wrote:
>
> Dear Harry,
>
... snip ...
>>
>> Michel, do you know if this is supposed to work with
>> xf86-video-amdgpu? When I've tried it before I didn't have any luck but
>> didn't have time to look into it.
>
> Sorry, what is your question. With the
> See here:
> #if IS_ENABLED(CONFIG_AGP)
> if (rdev->flags & RADEON_IS_AGP) {
> return ttm_agp_tt_populate(ttm, ctx);
> }
> #endif
>
This code appears to be after the potential location of NULL pointer dereference
that I pointed out. Please see,
```
patch#2 and patch#4 are Ached-by: Chunming Zhou
patch#1, #3, #5~#8 are Reviewed-by: Chunming Zhou
On 2019年03月19日 20:44, Christian König wrote:
And remove the existing code when it is unused.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 89
these two enumerated types are same for now. both of them might change in the
future.
I have not used clang, but would .block_id = (int)head->block fix your
warning? If such change is acceptable, I can make one then.
Thanks
xinhui
-Original Message-
From: Nathan Chancellor
Sent:
syncobj wait/signal operation is appending in command submission.
v2: separate to two kinds in/out_deps functions
v3: fix checking for timeline syncobj
Signed-off-by: Chunming Zhou
Cc: Tobias Hector
Cc: Jason Ekstrand
Cc: Dave Airlie
Cc: Chris Wilson
Cc: Lionel Landwerlin
---
we need to import/export timeline point.
v2: unify to one transfer ioctl
Signed-off-by: Chunming Zhou
Cc: Lionel Landwerlin
---
drivers/gpu/drm/drm_internal.h | 2 +
drivers/gpu/drm/drm_ioctl.c| 2 +
drivers/gpu/drm/drm_syncobj.c | 74 ++
1. clear cmd buffer
2. make amdgpu_memcpy_dispatch_test static
3. tab/space fix
Change-Id: Idf55f8881f66458b585092eccb55b6042520e4ad
Signed-off-by: Flora Cui
---
tests/amdgpu/basic_tests.c | 20
1 file changed, 12 insertions(+), 8 deletions(-)
diff --git
On 2019年03月19日 19:54, Lionel Landwerlin wrote:
On 15/03/2019 12:09, Chunming Zhou wrote:
v2: individually allocate chain array, since chain node is free
independently.
v3: all existing points must be already signaled before cpu perform
signal operation,
so add check condition for that.
From: Christian König
Implement finding the right timeline point in drm_syncobj_find_fence.
v2: return -EINVAL when the point is not submitted yet.
v3: fix reference counting bug, add flags handling as well
v4: add timeout for find fence
Signed-off-by: Christian König
Cc: Lionel Landwerlin
user mode can query timeline payload.
v2: check return value of copy_to_user
v3: handle querying entry by entry
v4: rebase on new chain container, simplify interface
v5: query last signaled timeline point, not last point.
v6: add unorder point check
Signed-off-by: Chunming Zhou
Cc: Tobias Hector
From: Christian König
Lockless container implementation similar to a dma_fence_array, but with
only two elements per node and automatic garbage collection.
v2: properly document dma_fence_chain_for_each, add dma_fence_chain_find_seqno,
drop prev reference during garbage collection if it's
Signed-off-by: Chunming Zhou
---
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 8a0732088640..4d8db87048d3 100644
---
points array is one-to-one match with syncobjs array.
v2:
add seperate ioctl for timeline point wait, otherwise break uapi.
v3:
userspace can specify two kinds waits::
a. Wait for time point to be completed.
b. and wait for time point to become available
v4:
rebase
v5:
add comment for
v2: individually allocate chain array, since chain node is free independently.
v3: all existing points must be already signaled before cpu perform signal
operation,
so add check condition for that.
v4: remove v3 change and add checking to prevent out-of-order
Signed-off-by: Chunming Zhou
From: Christian König
Use the dma_fence_chain object to create a timeline of fence objects
instead of just replacing the existing fence.
v2: rebase and cleanup
v3: fix garbage collection parameters
v4: add unorder point check, print a warn calltrace
Signed-off-by: Christian König
Cc: Lionel
I'm pleased to announce the 19.0.1 release of xf86-video-amdgpu, the
Xorg driver for AMD Radeon GPUs supported by the amdgpu kernel driver.
This release supports xserver versions 1.13-1.20.
This is a minor follow-up release with only one change:
* Add support for RandR output tile properties,
Hi Christian,
Thank you very much for your reply. I'm a little confused here so I really
appreciate if you could clarify it more.
For example, I don't understand why the condition of function
`radeon_ttm_tt_to_gtt` returning NULL is the argument being an AGP ttm. Based on
its definition, it
On 2019/3/18 14:15, Shaobo He wrote:
Hello everyone,
My name is Shaobo He and I am a graduate student at University of Utah. I am
using a static analysis tool to search for null pointer dereferences and came
across a potentially invalid memory accesses in the file
... or the backend methods is not `radeon_backend_func`.
That's the case when it is an AGP backend.
Moreover, could you point out the check of such case before the
offending code?
See here:
#if IS_ENABLED(CONFIG_AGP)
if (rdev->flags & RADEON_IS_AGP) {
return
Dear Mario,
On 18.03.19 18:19, Mario Kleiner wrote:
In VRR mode, proper vblank/pageflip timestamps can only be computed
after the display scanout position has left front-porch. Therefore
delay calls to drm_crtc_handle_vblank(), and thereby calls to
drm_update_vblank_count() and pageflip event
On 15/03/2019 12:09, Chunming Zhou wrote:
From: Christian König
Lockless container implementation similar to a dma_fence_array, but with
only two elements per node and automatic garbage collection.
v2: properly document dma_fence_chain_for_each, add dma_fence_chain_find_seqno,
drop prev
On 15/03/2019 12:09, Chunming Zhou wrote:
v2: individually allocate chain array, since chain node is free independently.
v3: all existing points must be already signaled before cpu perform signal
operation,
so add check condition for that.
Signed-off-by: Chunming Zhou
---
We discussed a few different approaches before settling on this one.
Maybe it needs some more background. XGMI links are quite power hungry.
Being able to power them down improves performance for power-limited
workloads that don't need XGMI. In machine learning, pretty much all
workloads are
On 3/19/2019 8:49 AM, Christian König wrote:
> Yeah, all that is perfectly fine.
>
> The problem is Shaoyun didn't put this into the mapping code, but
> rather into the VM state machine. So this won't work at all (the
> counter and increment/decrement unbalanced and multiple times).
We tried to
This reverts commit c9115f8904eef0f880d3b4f8306f553b1bb1c532.
Adding this to the mapping is complete nonsense and the whole
implementation looks racy. This patch wasn't thoughtfully reviewed
and should be reverted for now.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h
Am 19.03.19 um 13:27 schrieb Lionel Landwerlin:
On 15/03/2019 12:09, Chunming Zhou wrote:
[SNIP]
+/**
+ * struct dma_fence_chain - fence to represent an node of a fence chain
+ * @base: fence base class
+ * @lock: spinlock for fence handling
+ * @prev: previous fence of the chain
+ *
Allocating 16KB was way to much, just use 2KB as a start for now.
Signed-off-by: Christian König
---
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
Move the update parameter into the VM header and rename them.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 75 ++
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 45
2 files changed, 60 insertions(+), 60 deletions(-)
diff --git
Clean that up a bit and allow to always have the DMA addresses around.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 21 +
1 file changed, 9 insertions(+), 12 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
Not needed any more.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 5 -
1 file changed, 5 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index f5c25c0ae367..8348804c46cd 100644
---
And remove the existing code when it is unused.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 229 +
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 13 --
2 files changed, 6 insertions(+), 236 deletions(-)
diff --git
And remove the existing code when it is unused.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 74 +-
1 file changed, 14 insertions(+), 60 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
Separate out all functions for SDMA and CPU based page table
updates into separate backends.
This way we can keep most of the complexity of those from the
core VM code.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/Makefile | 3 +-
And remove the existing code when it is unused.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 89 +-
1 file changed, 32 insertions(+), 57 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
Yeah, all that is perfectly fine.
The problem is Shaoyun didn't put this into the mapping code, but rather
into the VM state machine. So this won't work at all (the counter and
increment/decrement unbalanced and multiple times).
The correct place to add this is
On 3/18/19 1:19 PM, Mario Kleiner wrote:
> We want vblank counts and timestamps of flip completion as sent
> in pageflip completion events to be consistent with the vblank
> count and timestamp of the vblank of flip completion, like in non
> VRR mode.
>
> In VRR mode, drm_update_vblank_count() -
On 3/18/19 1:19 PM, Mario Kleiner wrote:
> In VRR mode, proper vblank/pageflip timestamps can only be computed
> after the display scanout position has left front-porch. Therefore
> delay calls to drm_crtc_handle_vblank(), and thereby calls to
> drm_update_vblank_count() and pageflip event
Hi Shaobo,
that question came up a couple of times now. And the answer is: No,
there can't be a NULL pointer dereference.
The function radeon_ttm_tt_to_gtt returns NULL only when it is an AGP
ttm structure, and that case is checked right before the offending code.
Unfortunately I don't see
Thanks Felix .
We did consider to put the logic into bo_add/bo_rmv, but Felix pointed
out the object can be migrate from FB to system memory after allocation
. I also think of put the logic inside amdgpu_vm_bo_update_mapping ,
but seems that function prefer to take the dma address already
On Tue, Mar 19, 2019 at 9:31 AM Huang Rui wrote:
>
> RLC #53815 ucode has the noise issue on 4k playback while gfxoff enabled.
>
> Signed-off-by: Huang Rui
Acked-by: Alex Deucher
> ---
> drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git
On 3/19/19 9:23 AM, Kazlauskas, Nicholas wrote:
> On 3/18/19 1:19 PM, Mario Kleiner wrote:
>> In VRR mode, proper vblank/pageflip timestamps can only be computed
>> after the display scanout position has left front-porch. Therefore
>> delay calls to drm_crtc_handle_vblank(), and thereby calls to
RLC #53815 ucode has the noise issue on 4k playback while gfxoff enabled.
Signed-off-by: Huang Rui
---
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 6b48d4c..3765d97
I just put a Radeon 9600XT into an Athlon XP machine. I do not know if the card
is good or not, but 2D dumb framebufer works fine with radeon module. However,
init of the acceleration engine fails. I compiled my own kernel (v5.0 from git)
with UBSAN (no other debug yet) and got a cascade of UBSAN
I am seeing the following warning on at least 2 different servers with onboard
radeon graphics.
It was not there with 5.0.
[4.764838] [drm] radeon kernel modesetting enabled.
[4.765334] radeon :01:01.0: vgaarb: deactivate vga console
[4.767746] Console: switching to colour dummy
On Tue, Mar 19, 2019 at 12:26 AM Chengming Gui wrote:
>
> Max Link Width's full mask is 0x3f,
> and it's highest bit express X16.
>
> Signed-off-by: Chengming Gui
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 12 ++--
> 1 file changed, 2 insertions(+), 10 deletions(-)
>
> diff
I'm pleased to announce the 19.0.1 release of xf86-video-ati, the Xorg
driver for ATI/AMD Radeon GPUs supported by the radeon kernel driver.
This release supports xserver versions 1.13-1.20.
This is a bug-fix release, with only three changes:
* Fixes for two regressions which crept into the
Am 19.03.19 um 16:42 schrieb Liu, Shaoyun:
> Thanks Felix .
>
> We did consider to put the logic into bo_add/bo_rmv, but Felix pointed
> out the object can be migrate from FB to system memory after allocation
> .
Yeah, I also considered that and that is actually a really good argument.
But
48 matches
Mail list logo
