Hi Huoqing,
Your patch is technically correct. However, I don't think it fixes any
actual bug, and it changes a code path that has no performance
implications. Therefore I would just leave it as it is.
Regards,
Felix
Am 2021-07-20 um 2:34 a.m. schrieb Cai Huoqing:
> no need to get error code
Am 2021-07-23 um 6:46 p.m. schrieb Sierra Guiza, Alejandro (Alex):
>
> On 7/17/2021 2:54 PM, Sierra Guiza, Alejandro (Alex) wrote:
>>
>> On 7/16/2021 5:14 PM, Felix Kuehling wrote:
>>> Am 2021-07-16 um 11:07 a.m. schrieb Theodore Y. Ts'o:
>>>> On
Am 2021-07-28 um 7:45 p.m. schrieb Sierra Guiza, Alejandro (Alex):
>
> On 7/22/2021 12:26 PM, Jason Gunthorpe wrote:
>> On Thu, Jul 22, 2021 at 11:59:17AM -0500, Sierra Guiza, Alejandro
>> (Alex) wrote:
>>> On 7/22/2021 7:23 AM, Jason Gunthorpe wrote:
On Sat, Jul 17, 2021 at 02:21:32PM -0500,
On 2021-04-07 7:25 a.m., Christian König wrote:
+ /*
+ * Don't verify access for KFD BOs. They don't have a GEM
+ * object associated with them.
+ */
+ if (bo->kfd_bo)
+ goto out;
Who does the access verification now?
This is somewhat confusing.
I took this check as-is
On 2021-04-07 3:34 p.m., Felix Kuehling wrote:
On 2021-04-07 7:25 a.m., Christian König wrote:
+ /*
+ * Don't verify access for KFD BOs. They don't have a GEM
+ * object associated with them.
+ */
+ if (bo->kfd_bo)
+ goto out;
Who does the access ver
ROCm user mode has acquired VMs from DRM file descriptors for as long
as it supported the upstream KFD. Legacy code to support older versions
of ROCm is not needed any more.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 4 --
.../gpu/drm/amd/amdgpu
DRM allows access automatically when it creates a GEM handle for a BO.
KFD BOs don't have GEM handles, so KFD needs to manage access manually.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 3 ++-
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu needs the drm_priv to allow mmap
to access the BO through the corresponding file descriptor.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 14 ++--
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 69
This shortcut is no longer needed with access managed progerly by KFD.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 7 ---
1 file changed, 7 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index
Am 2021-04-08 um 11:02 a.m. schrieb Jason Gunthorpe:
> On Mon, Apr 05, 2021 at 09:45:55PM -0400, Felix Kuehling wrote:
>> Rebased on upstream. Dropped already upstream patch
>> "drm/amdgpu: reserve fence slot to update page table".
>>
>> Added more fixes:
This should have been fixed by this commit in amd-staging-drm-next:
https://lore.kernel.org/patchwork/patch/1392368/
commit b8aff1f3a0b3d8434f8ccf5d3017137c29aca77b
Author: Felix Kuehling
Date: Mon Mar 8 22:15:42 2021 -0500
drm/amdkfd: fix build error with AMD_IOMMU_V2=m
Using
This name is more fitting, especially for the changes coming next to
support multi-GPU systems with proper DMA mappings. Cleaned up the code
and renamed some related functions and variables to improve readability.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
For now they all reference the same BO. For correct DMA mappings they will
refer to different BOs per-GPU.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 22 ++-
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/amd
comments in this patch series where I'm
hoping for some expert advice. Patches 8 and 9 are some related fixes
in TTM and amdgpu_ttm. I'm pretty sure patch 9 is not the right way to
do this.
Felix Kuehling (9):
drm/amdgpu: Rename kfd_bo_va_list to kfd_mem_attachment
drm/amdgpu:
amdgpu_amdkfd_gpuvm_free_memory_of_gpu.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 103 --
1 file changed, 48 insertions(+), 55 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
b/drivers/gpu/drm/amd/amdgpu
DMA map kfd_mem_attachments in update_gpuvm_pte. This function is called
with the BO and page tables reserved, so we can safely update the DMA
mapping.
DMA unmap when a BO is unmapped from a GPU and before updating mappings
in restore workers.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd
Add BO-type specific helpers functions to DMA-map and unmap
kfd_mem_attachments. Implement this functionality for userptrs by creating
one SG BO per GPU and filling it with a DMA mapping of the pages from the
original mem->bo.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amd
This is needed to avoid deadlocks with DMA buf import in the next patch.
Also move PT/PD validation out of kfd_mem_attach, that way the caller
can bo this unconditionally.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 72 +++
1 file changed
Pages in SG BOs were not allocated by TTM. So don't count them against
TTM's pages limit.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/ttm/ttm_tt.c | 27 ++-
1 file changed, 18 insertions(+), 9 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/d
Use DMABufs with dynamic attachment to DMA-map GTT BOs on other GPUs.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 2 +
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 74 ++-
2 files changed, 75 insertions(+), 1 deletion(-)
diff --git a
ind instead?
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 13 +
1 file changed, 13 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 936b3cfdde55..257750921eed 100644
--- a/drivers/gpu/
comments in this patch series where I'm
hoping for some expert advice. Patches 8 and 9 are some related fixes
in TTM and amdgpu_ttm. I'm pretty sure patch 9 is not the right way to
do this.
Felix Kuehling (9):
drm/amdgpu: Rename kfd_bo_va_list to kfd_mem_attachment
drm/amdgpu:
For now they all reference the same BO. For correct DMA mappings they will
refer to different BOs per-GPU.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 22 ++-
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/amd
This name is more fitting, especially for the changes coming next to
support multi-GPU systems with proper DMA mappings. Cleaned up the code
and renamed some related functions and variables to improve readability.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
amdgpu_amdkfd_gpuvm_free_memory_of_gpu.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 103 --
1 file changed, 48 insertions(+), 55 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
b/drivers/gpu/drm/amd/amdgpu
Pages in SG BOs were not allocated by TTM. So don't count them against
TTM's pages limit.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/ttm/ttm_tt.c | 27 ++-
1 file changed, 18 insertions(+), 9 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/d
ind instead?
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 13 +
1 file changed, 13 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 936b3cfdde55..257750921eed 100644
--- a/drivers/gpu/
Add BO-type specific helpers functions to DMA-map and unmap
kfd_mem_attachments. Implement this functionality for userptrs by creating
one SG BO per GPU and filling it with a DMA mapping of the pages from the
original mem->bo.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amd
Use DMABufs with dynamic attachment to DMA-map GTT BOs on other GPUs.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 2 +
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 74 ++-
2 files changed, 75 insertions(+), 1 deletion(-)
diff --git a
This is needed to avoid deadlocks with DMA buf import in the next patch.
Also move PT/PD validation out of kfd_mem_attach, that way the caller
can bo this unconditionally.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 72 +++
1 file changed
DMA map kfd_mem_attachments in update_gpuvm_pte. This function is called
with the BO and page tables reserved, so we can safely update the DMA
mapping.
DMA unmap when a BO is unmapped from a GPU and before updating mappings
in restore workers.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd
Sorry for the spam. I mis-spelled the amd-gfx list on the To: line of
this patch series. Please ignore this and see the patch series I sent a
minute later.
Regards,
Felix
On 2021-04-14 2:46 a.m., Felix Kuehling wrote:
This patch series fixes DMA-mappings of system memory (GTT and userptr
Am 2021-04-14 um 3:44 a.m. schrieb Thomas Zimmermann:
> Hi
>
> Am 07.04.21 um 21:49 schrieb Felix Kuehling:
>> On 2021-04-07 3:34 p.m., Felix Kuehling wrote:
>>> On 2021-04-07 7:25 a.m., Christian König wrote:
>>>>>>>> + /*
>>>>>&g
Am 2021-04-14 um 8:25 a.m. schrieb Daniel Vetter:
>> Sorry I though that this would be obvious :)
>>
>> I've already pushed the patch in the morning, but going to keep that in
>> mind for the next time.
> I'll keep reminding you to pls elaborate more in commit messages, it's
> coming up every once
Am 2021-04-14 um 3:33 a.m. schrieb Christian König:
> Am 14.04.21 um 08:46 schrieb Felix Kuehling:
>> amdgpu_ttm_tt_unpopulate can be called during bo_destroy. The
>> dmabuf->resv
>> must not be held by the caller or dma_buf_detach will deadlock. This is
>> proba
Am 2021-04-14 um 11:37 a.m. schrieb philip yang:
>
>
> On 2021-04-07 7:12 p.m., Felix Kuehling wrote:
>> DRM allows access automatically when it creates a GEM handle for a BO.
>> KFD BOs don't have GEM handles, so KFD needs to manage access manually.
>
> After readi
Am 2021-04-14 um 11:21 a.m. schrieb philip yang:
>
>
> On 2021-04-07 7:12 p.m., Felix Kuehling wrote:
>> amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu needs the drm_priv to allow mmap
>> to access the BO through the corresponding file descriptor.
>>
>> Signed-off-by: F
fence
drm/amdgpu: add param bit flag to create SVM BOs
drm/amdgpu: svm bo enable_signal call condition
drm/amdgpu: add svm_bo eviction to enable_signal cb
Felix Kuehling (13):
drm/amdkfd: map svm range to GPUs
drm/amdkfd: svm range eviction and restore
drm/amdgpu: Enable retry f
: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_priv.h| 11 +++
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 10 ++
2 files changed, 21 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
From: Philip Yang
Use HMM to get system memory pages address, which will be used to
map to GPUs or migrate to vram.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 116 ++-
drivers/gpu
index into bitmap.
Signed-off-by: Philip Yang
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/Makefile | 3 +-
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 17 +
drivers/gpu/drm/amd/amdkfd/kfd_priv.h| 8 +
drivers
From: Philip Yang
Get the intersection of attributes over all memory in the given
range
Signed-off-by: Philip Yang
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 164 +++
1 file changed
From: Philip Yang
Move the HMM get pages function from amdgpu_ttm and to amdgpu_mn. This
common function will be used by new svm APIs.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c | 83
s the
mmap_write_lock to guarantee that it has an up-to-date svm_range_list.
Signed-off-by: Philip Yang
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 +
drivers/gpu/drm/amd/amdkfd/kfd_svm.c
From: Philip Yang
For larger range allocation, if hmm_range_fault return -EBUSY, set retry
timeout based on 1 second for every 512MB, this is safe timeout value.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c
retries per KFD process.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 28 +++
include/uapi/linux/kfd_ioctl.h | 43 +++-
2 files changed, 70 insertions(+), 1 deletion
flushes any deferred work to make sure it restores
an up-to-date svm_range_list.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_priv.h| 2 +
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 1 +
drivers/gpu/drm/amd/amdkfd
From: Philip Yang
Register vram memory as MEMORY_DEVICE_PRIVATE type resource, to
allocate vram backing pages for page migration.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/Kconfig | 2 +
drivers/gpu/drm/amd
RETRY_DISABLE setting. However, processes not relying on recoverable
page faults can work with RETRY enabled. This means XNACK off is always
available as a fallback so we can use the same mode on all GPUs in a
process.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
From: Philip Yang
It will be used by kfd to map svm range to GPU, because svm range does
not have amdgpu_bo and bo_va, cannot use amdgpu_bo_update interface, use
amdgpu vm update interface directly.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
. Restore work wait for migration is finished, then update GPUs page
table mapping to new vram pages, resume process queues
If migrate_vma_setup failed to collect all ram pages of range, retry 3
times until success to start migration.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off
mapping exist, then process exit will
not have unmap callback for this prange to free prange and svm bo. Free
outstanding pranges from svms list before process is freed in
svm_range_list_fini.
Signed-off-by: Philip Yang
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix
same kernel buffer function to
fill in gart table mapping, so this is serialized with memory copy by
sdma job submit. We only need wait for the last memory copy sdma fence
for larger buffer migration.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 94 ++--
1 file changed, 75 insertions(+), 19 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index b12eb0
From: Alex Sierra
Use SVM API to restore page tables when retry fault and
compute context are enabled.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 20 +++-
1 file changed, 15 insertions
Page table restore implementation in SVM API. This is called from
the fault handler at amdgpu_vm. To update page tables through
the page fault retry IH.
Signed-off-by: Alex Sierra
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd
attributes that can be used to set or get one or multiple attributes.
Signed-off-by: Philip Yang
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 12 ++
drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c | 4
Use amdgpu_vm_bo_update_mapping to update GPU page table to map or unmap
svm range system memory pages address to GPUs.
Signed-off-by: Philip Yang
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 473
GPUVM faults.
Signed-off-by: Alex Sierra
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 6 +-
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 77 +++-
drivers/gpu/drm/amd/amdkfd/kfd_svm.h
do an eviction. Instead it will discard
the memory held by the BO. This is needed for HMM migration to user
mode system memory pages.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 14 ++
1 file change
This is needed to allow per-process XNACK mode selection in the SQ when
booting with XNACK off by default.
Signed-off-by: Felix Kuehling
Reviewed-by: Philip Yang
Tested-by: Alex Sierra
---
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 ++-
drivers/gpu/drm/amd/amdgpu/gfxhub_v1_0.c | 8
finished, then update GPU page
table mapping to system memory, and resume process queues
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 302 ++-
drivers/gpu/drm/amd/amdkfd/kfd_migrate.h
VRAM space
is available.
Here, all the ttm_evict calls are synchronous, this guarantees
that each eviction has completed and the fence has signaled before
it returns.
Signed-off-by: Alex Sierra
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu
From: Alex Sierra
By default this timestamp is 32 bit counter. It gets
overflowed in around 10 minutes.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Philip Yang
---
drivers/gpu/drm/amd/amdgpu/vega10_ih.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers
From: Alex Sierra
Add CREATE_SVM_BO define bit for SVM BOs.
Another define flag was moved to concentrate these
KFD type flags in one include file.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 7
regular BO evictions. This also include modifications to set the
reference at the fence creation call.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 4 +++-
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c
than AMDGPU_SVM_RANGE_RETRY_FAULT_PENDING, that means the
retry fault is from another GPU, then continue to handle retry fault
recover.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 17 +
drivers
From: Alex Sierra
Add to amdgpu_amdkfd_fence.enable_signal callback, support
for svm_bo fence eviction.
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_fence.c | 11 ---
1 file changed, 8 insertions
for visible options.
Reviewed-by: Philip Yang
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/Kconfig | 15 ++--
drivers/gpu/drm/amd/amdkfd/Makefile | 9 ---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 7 ++
drivers/gpu/drm/amd/amdkfd/kfd_migrate.h | 17 +++
With xnack on, GPU vm fault handler decide the best restore location,
then migrate range to the best restore location and update GPU mapping
to recover the GPU vm fault.
Signed-off-by: Philip Yang
Signed-off-by: Alex Sierra
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
gpu to migrate to system memory,
then use sdma of destination gpu to migrate from system memory to gpu.
Print out gpuid or gpuidx in debug messages.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 47
esses.
CoherentHostAccess property added to HSA_MEMORYPROPERTY, the value match
HSA_MEMORYPROPERTY defined in Thunk spec:
CoherentHostAccess: whether or not device memory can be coherently
accessed by the host CPU.
Signed-off-by: Philip Yang
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kue
amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu needs the drm_priv to allow mmap
to access the BO through the corresponding file descriptor. The VM can
also be extracted from drm_priv, so drm_priv can replace the vm parameter
in the kfd2kgd interface.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm
This shortcut is no longer needed with access managed properly by KFD.
Signed-off-by: Felix Kuehling
Reviewed-by: Philip Yang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 7 ---
1 file changed, 7 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
b/drivers/gpu/drm/amd
ccess
manually. Use drm_vma_node_allow to allow user mode to mmap BOs allocated
with kfd_ioctl_alloc_memory_of_gpu through the DRM render node that was
used in the kfd_ioctl_acquire_vm call for the same GPU.
Signed-off-by: Felix Kuehling
Acked-by: Christian König
---
drivers/gpu/drm/amd/a
Am 2021-04-15 um 3:41 a.m. schrieb Christian König:
>
>
> Am 14.04.21 um 17:15 schrieb Felix Kuehling:
>> Am 2021-04-14 um 3:33 a.m. schrieb Christian König:
>>> Am 14.04.21 um 08:46 schrieb Felix Kuehling:
>>>> amdgpu_ttm_tt_unpopulate can be called duri
Am 2021-04-20 um 4:51 a.m. schrieb Daniel Vetter:
>>> Whole series is Reviewed-by: Christian König
>> Thanks a lot. If I'm not mistaken, the patches at [1] need to go in first.
>> So it could take a a bit until this lands.
>>
>> Otherwise, this series could go through the same tree as [1] if nouv
The dmabuf attachment should be updated by moving the SG BO to DOMAIN_CPU
and back to DOMAIN_GTT. This does not necessarily invoke the
populate/unpopulate callbacks. Do this in backend_bind/unbind instead.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 3
For now they all reference the same BO. For correct DMA mappings they will
refer to different BOs per-GPU.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 22 ++-
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm/amd
Add BO-type specific helpers functions to DMA-map and unmap
kfd_mem_attachments. Implement this functionality for userptrs by creating
one SG BO per GPU and filling it with a DMA mapping of the pages from the
original mem->bo.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amd
Enable the Intel IOMMU driver in the rock-dbg_defconfig. This enables
testing of DMA mappings on systems with an Intel IOMMU.
Signed-off-by: Felix Kuehling
---
arch/x86/configs/rock-dbg_defconfig | 11 ++-
1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/arch/x86/configs
DMA map kfd_mem_attachments in update_gpuvm_pte. This function is called
with the BO and page tables reserved, so we can safely update the DMA
mapping.
DMA unmap when a BO is unmapped from a GPU and before updating mappings
in restore workers.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd
created for DMA mappings
cause many tests fail because TTM incorrectly thinks it's out of memory.
Felix Kuehling (10):
rock-dbg_defconfig: Enable Intel IOMMU
drm/amdgpu: Rename kfd_bo_va_list to kfd_mem_attachment
drm/amdgpu: Keep a bo-reference per-attachment
drm/amdgpu: Simplify
amdgpu_amdkfd_gpuvm_free_memory_of_gpu.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 103 --
1 file changed, 48 insertions(+), 55 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
b/drivers/gpu/drm/amd/amdgpu
Use DMABufs with dynamic attachment to DMA-map GTT BOs on other GPUs.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 2 +
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 76 ++-
2 files changed, 77 insertions(+), 1 deletion(-)
diff --git a
Pages in SG BOs were not allocated by TTM. So don't count them against
TTM's pages limit.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/ttm/ttm_tt.c | 27 ++-
1 file changed, 18 insertions(+), 9 deletions(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/d
This name is more fitting, especially for the changes coming next to
support multi-GPU systems with proper DMA mappings. Cleaned up the code
and renamed some related functions and variables to improve readability.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
This is needed to avoid deadlocks with DMA buf import in the next patch.
Also move PT/PD validation out of kfd_mem_attach, that way the caller
can bo this unconditionally.
Signed-off-by: Felix Kuehling
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 75 +++
1 file changed
Am 2021-04-22 um 8:31 a.m. schrieb Colin King:
> From: Colin Ian King
>
> Currently the call to kfd_process_gpuidx_from_gpuid is returning an
> int value and this is being assigned to a uint32_t variable gpuidx
> and this is being checked for a negative error return which is always
> going to be f
Am 2021-04-22 um 8:44 a.m. schrieb Colin King:
> From: Colin Ian King
>
> The variable r is being initialized with a value that is never read
> and it is being updated later with a new value. The initialization is
> redundant and can be removed.
>
> Addresses-Coverity: ("Unused value")
> Signed-of
Am 2021-04-22 um 9:33 p.m. schrieb Zeng, Oak:
> Regards,
> Oak
>
>
>
> On 2021-04-21, 9:31 PM, "amd-gfx on behalf of Felix Kuehling"
>
> wrote:
>
> Do AQL queue double-mapping with a single attach call. That will make it
> easier to create
Am 2021-04-23 um 5:39 p.m. schrieb Souptick Joarder:
> Kernel test robot throws below warning ->
>
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c:125:5: warning:
>>> no previous prototype for 'kgd_arcturus_hqd_sdma_load'
>>> [-Wmissing-prototypes]
> 125 | int kgd_arcturus_hqd_sdma_loa
Am 2021-04-24 um 12:50 p.m. schrieb Fabio M. De Francesco:
> Fixed a kernel-doc error in the documentation of a function.
>
> Signed-off-by: Fabio M. De Francesco
I fixed the headline prefix to match our usual convention (drm/amdkfd:
...) and applied the patch to amd-staging-drm-next.
Thanks,
Am 2021-04-24 um 5:40 a.m. schrieb Souptick Joarder:
> Kernel test robot throws below warning ->
>
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c:125:5: warning:
>>> no previous prototype for 'kgd_arcturus_hqd_sdma_load'
>>> [-Wmissing-prototypes]
> 125 | int kgd_arcturus_hqd_sdma_loa
On 2021-04-26 8:15, Nirmoy wrote:
> Reviewed-by: Nirmoy Das
>
> On 4/26/21 2:13 PM, Colin King wrote:
>> From: Colin Ian King
>>
>> There is a spelling mistake in a pr_debug message. Fix it.
>>
>> Signed-off-by: Colin Ian King
Applied to amd-staging-drm-next.
Thanks,
Felix
>> ---
>> driv
ptr in iommu. Maybe you map them in iommu not during
> memory attachment time?
>
> Also see a nit-pick inline
>
> Regards,
> Oak
>
>
>
> On 2021-04-21, 9:31 PM, "dri-devel on behalf of Felix Kuehling"
>
> wrote:
>
> Add BO-type speci
Am 2021-04-26 um 8:23 p.m. schrieb Zeng, Oak:
> Regards,
> Oak
>
>
>
> On 2021-04-21, 9:31 PM, "dri-devel on behalf of Felix Kuehling"
>
> wrote:
>
> DMA map kfd_mem_attachments in update_gpuvm_pte. This function is called
> with the BO
Am 2021-04-26 um 8:35 p.m. schrieb Zeng, Oak:
> Regards,
> Oak
>
>
>
> On 2021-04-21, 9:31 PM, "amd-gfx on behalf of Felix Kuehling"
>
> wrote:
>
> Use DMABufs with dynamic attachment to DMA-map GTT BOs on other GPUs.
>
> Signed-off-by
>
> > On 2021-04-21, 9:31 PM, "amd-gfx on behalf of Felix Kuehling"
>
> wrote:
> >
> > Use DMABufs with dynamic attachment to DMA-map GTT BOs on other
> GPUs.
> >
> > Signed-off-by: Felix Kuehling
>
ttm_bo_swapout returns a non-0 value on success. Don't treat that as an
error in ttm_tt_populate.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/ttm/ttm_tt.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/ttm/ttm_tt.c b/drivers/gpu/drm/ttm/ttm_tt.c
1 - 100 of 900 matches
Mail list logo
