to submit your changes. Please apply the two fixes I pointed out
below in a separate patch.
Thanks,
Felix
On 2019-06-20 0:15, Kuehling, Felix wrote:
> On 2019-06-19 20:53, Kim, Jonathan wrote:
>> v4: fixed kzalloc error check and modified df func init to return error code
> This c
On 2019-06-19 20:53, Kim, Jonathan wrote:
> v4: fixed kzalloc error check and modified df func init to return error code
This comment isn't applicable any more because there is no more kzalloc.
Maybe remove the review version history and instead update the
description of what this commit
One more possible simplification below.
On 2019-06-19 6:50 p.m., Kim, Jonathan wrote:
> v4: fixed kzalloc error check and modified df func init to return error code
>
> v3: fixed cleanup by adding fini to free up adev df config counters
>
> v2: simplified by removing xgmi references in function
Found three more problems. With those fixed, this patch is Reviewed-by:
Felix Kuehling
On 2019-06-19 6:50 p.m., Kim, Jonathan wrote:
> adding perf event counters
>
> Change-Id: Ib8800c2d652e3e788872bd721458ac5422859b62
> Signed-off-by: Jonathan Kim
> ---
> drivers/gpu/drm/amd/amdgpu/Makefile
Thanks.
To add to what Kent said, this is for providing per-process information
in rocm-smi or other tools. It should allow us to do something like "ps"
or "top" for compute processes running on the GPU. Currently it's only
the PASID. We'll add per-process memory usage and probably user mode
On 2019-06-18 1:37, Christoph Hellwig wrote:
> On Mon, Jun 17, 2019 at 09:45:09PM -0300, Jason Gunthorpe wrote:
>> Am I looking at the wrong thing? Looks like it calls it through a work
>> queue should should be OK..
> Yes, it calls it through a work queue. I guess that is fine because
> it needs
On 2019-06-18 17:34, Kim, Jonathan wrote:
> v2: fixed missing break in switch statement
>
> add pmu counters
>
> Change-Id: I1aca271fd12cabce0ccfc076f771cde2d4cadd54
> Signed-off-by: Jonathan Kim
> ---
> drivers/gpu/drm/amd/amdgpu/Makefile| 2 +-
>
Sorry, I caught another problem with error handling. See below.
On 2019-06-18 15:24, Kim, Jonathan wrote:
> v3: fixed cleanup by adding fini to free up adev df config counters
>
> v2: simplified by removing xgmi references in function names and moving to
> generic df function names. fixed issue
On 2019-06-14 9:52 p.m., Yang, Philip wrote:
> Under memory pressure, hmm_range_fault may return error code -ENOMEM
> or -EBUSY, change pr_info to pr_debug to remove unnecessary kernel log
> message because we will retry restore again.
>
> Call get_user_pages_done if TTM get user pages failed will
On 2019-06-17 3:28 p.m., Christian König wrote:
> Am 17.06.19 um 21:15 schrieb Kuehling, Felix:
>> Looks good to me. One cosmetic comment inline. With that fixed this
>> patch is Reviewed-by: Felix Kuehling
>>
>> On 2019-06-14 12:51 p.m., StDenis, Tom wrote:
>>
Looks good to me. One cosmetic comment inline. With that fixed this
patch is Reviewed-by: Felix Kuehling
On 2019-06-14 12:51 p.m., StDenis, Tom wrote:
> On 32-bit hosts mem->num_pages is 32-bits and can overflow
> when shifted. Add a cast to avoid this.
>
> Signed-off-by: Tom St Denis
> ---
On 2019-06-13 8:59, Yang, Philip wrote:
> On 2019-06-13 4:54 a.m., Koenig, Christian wrote:
>> Am 12.06.19 um 23:13 schrieb Yang, Philip:
>>> On 2019-06-12 3:28 p.m., Christian König wrote:
Am 12.06.19 um 17:13 schrieb Yang, Philip:
> TTM create two zones, kernel zone and dma32 zone for
On 2019-06-11 9:20, Zeng, Oak wrote:
> Call reservation_object_reserve_shared to reserve
> space for shared fence. Otherwise it will trigger
> BUG_ON condition in reservation_object_add_shared_fence.
>
> Change-Id: Ib0fae16247e33ee68f95bffa723451c0cd619344
> Signed-off-by: Oak Zeng
Reviewed-by:
On 2019-06-12 4:08, StDenis, Tom wrote:
> (v2): Return 0 and set mem->mm_node to NULL.
> (v3): Use atomic64_add_return instead.
>
> Signed-off-by: Tom St Denis
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 17 -
> 1 file changed, 12 insertions(+), 5 deletions(-)
>
>
[+Philip]
Hi Jason,
I'm out of the office this week.
Hi Philip, can you give this a go? Not sure how much you've been
following this patch series review. Message or call me on Skype to
discuss any questions.
Thanks,
Felix
On 2019-06-11 12:48, Jason Gunthorpe wrote:
> On Thu, Jun 06, 2019
TTM itself has some logic for need_dma32 and TTM_PAGE_FLAG_DMA32. I
believe that should already handle this. need_dma32 is passed from
amdgpu to ttm_bo_device_init to bdev->need_dma32. ttm_tt_create
translates that to page_flags |= TTM_PAGE_FLAG_DMA32 and passes that to
On 2019-06-10 13:48, Zeng, Oak wrote:
> Move HSA_CAP_ATS_PRESENT initialization logic from kfd iommu codes to
> kfd topology codes. This removes kfd_iommu_device_init's dependency
> on kfd_topology_add_device. Also remove duplicate code setting the
> same.
>
> Change-Id:
On 2019-06-06 5:51 p.m., Zeng, Oak wrote:
> dqm is referenced in function kfd_toplogy_add_device.
> Move dqm initialization up to avoid NULL pointer reference.
This addresses a pretty unlikely race condition where someone looks at
/sys/kernel/debug/kfd/hqds during the device initialization.
We
On 2019-06-06 5:40 p.m., Zeng, Oak wrote:
> Introduce a new mqd allocation interface and split the original
> init_mqd function into two functions: allocate_mqd and init_mqd.
> Also renamed uninit_mqd to free_mqd. This is preparation work to
> fix a circular lock dependency.
>
> Change-Id:
Patches 5 and 6 are Reviewed-by: Felix Kuehling
On 2019-06-06 2:25 p.m., Zeng, Oak wrote:
> SDMA queue allocation requires the dqm lock at it modify
> the global dqm members. Move up the dqm_lock so sdma
> queue allocation is enclosed in the critical section. Move
> mqd allocation out of
On 2019-06-06 2:25 p.m., Zeng, Oak wrote:
> Introduce a new mqd allocation interface and split the original
> init_mqd function into two functions: allocate_mqd and init_mqd.
> Also renamed uninit_mqd to free_mqd. This is preparation work to
> fix a circular lock dependency.
>
> Change-Id:
[resent with correct address for Alex]
On 2019-06-06 11:11 a.m., Jason Gunthorpe wrote:
> On Fri, May 10, 2019 at 07:53:21PM +0000, Kuehling, Felix wrote:
>> These problems were found in AMD-internal testing as we're working on
>> adopting HMM. They are rebased against glisse/h
On 2019-06-06 11:11 a.m., Jason Gunthorpe wrote:
> On Fri, May 10, 2019 at 07:53:21PM +0000, Kuehling, Felix wrote:
>> These problems were found in AMD-internal testing as we're working on
>> adopting HMM. They are rebased against glisse/hmm-5.2-v3. We'd like to get
>> them
I think the simpler way to fix this, is to restructure
create_queue_cpsch similar to the nocpsch version where we allocate the
MQD early and take the DQM lock right after that. That way you don't
need locked and unlocked variants of allocate_sdma_queue and
deallocate_sdma_queue.
Regards,
This patch looks good to me, but it'll probably change a little if you
implement my suggestions for patch 4.
Regards,
Felix
On 2019-06-05 12:06 p.m., Zeng, Oak wrote:
> The idea to break the circular lock dependency is to move allocate_mqd
> out of dqm lock protection. See callstack #1
Patches 1-3 are Reviewed-by: Felix Kuehling
On 2019-06-05 12:06 p.m., Zeng, Oak wrote:
> This is prepare work to fix a circular lock dependency.
> No logic change
>
> Change-Id: I4e0ee918260e7780de972dd71f4ce787b4f6dde9
> Signed-off-by: Oak Zeng
> ---
>
Nice. Surprisingly this actually resulted in slightly fewer lines of
code. Some suggestions:
Let's also rename uninit_mqd to free_mqd, as it is the counterpart to
allocate_mqd, not init_mqd.
And unless I'm missing something, init_mqd and update_mqd cannot fail
now that allocate_mqd has been
On 2019-06-05 9:56, Michel Dänzer wrote:
> On 2019-06-05 1:24 p.m., Christian König wrote:
>> Am 04.06.19 um 21:03 schrieb Zeng, Oak:
>>> From: amd-gfx On Behalf Of
>>> Kuehling, Felix
>>> On 2019-06-04 11:23, Christian König wrote:
[snip]
>>> --
On 2019-06-04 16:13, Yang, Philip wrote:
> HMM provides new APIs and helps in kernel 5.2-rc1 to simplify driver
> path. The old hmm APIs are deprecated and will be removed in future.
>
> Below are changes in driver:
>
> 1. Change hmm_vma_fault to hmm_range_register and hmm_range_fault which
>
On 2019-06-04 16:15, Alex Deucher wrote:
> On Tue, Jun 4, 2019 at 4:12 PM Kuehling, Felix wrote:
>> There haven't been any objections to these changes. Does anyone want to
>> add their Acked-by before I submit?
> Series is:
> Acked-by: Alex Deucher
>
> I wonder if the
There haven't been any objections to these changes. Does anyone want to
add their Acked-by before I submit?
Thanks,
Felix
On 2019-05-28 18:38, Kuehling, Felix wrote:
> New feature: queue priorities
>
> The eviction state logic change is preparation for some debugger support
> w
On 2019-06-04 11:23, Christian König wrote:
> Since we now keep BOs on the LRU we need to make sure
> that they are removed when they are pinned.
>
> Signed-off-by: Christian König
> ---
> include/drm/ttm/ttm_bo_driver.h | 14 ++
> 1 file changed, 6 insertions(+), 8 deletions(-)
>
[+Jerome]
On 2019-06-03 7:20 p.m., Yang, Philip wrote:
>
> On 2019-06-03 5:02 p.m., Kuehling, Felix wrote:
>> On 2019-06-03 2:44 p.m., Yang, Philip wrote:
>>> HMM provides new APIs and helps in kernel 5.2-rc1 to simplify driver
>>> path. The old hmm APIs are
allocate_vmid, allocate_hqd and allocate_sdma_queue all work on data in
the DQM structure. So it seems these need to be protected by the DQM
lock. allocate_doorbell doesn't need the DQM lock because its data
structures are in the process_device structure, which is protected by
the process
On 2019-06-03 1:51 p.m., Zeng, Oak wrote:
> This is prepare work to fix a circular lock dependency.
> No logic change
>
> Change-Id: I4e0ee918260e7780de972dd71f4ce787b4f6dde9
> Signed-off-by: Oak Zeng
> ---
> .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 171
> +++--
>
On 2019-06-03 2:44 p.m., Yang, Philip wrote:
> HMM provides new APIs and helps in kernel 5.2-rc1 to simplify driver
> path. The old hmm APIs are deprecated and will be removed in future.
>
> Below are changes in driver:
>
> 1. Change hmm_vma_fault to hmm_range_register and hmm_range_fault which
>
On 2019-05-31 3:09 p.m., Yang, Philip wrote:
> HMM provides new APIs and helps in kernel 5.2-rc1 to simplify driver
> path. The old hmm APIs are deprecated and will be removed in future.
>
> Below are changes in driver:
>
> 1. Change hmm_vma_fault to hmm_range_register and hmm_range_fault which
>
On 2019-05-31 5:32 p.m., Yang, Philip wrote:
>
> On 2019-05-31 3:42 p.m., Kuehling, Felix wrote:
>> On 2019-05-31 1:28 p.m., Yang, Philip wrote:
>>> On 2019-05-30 6:36 p.m., Kuehling, Felix wrote:
>>>>>
>>>>> #if IS_ENABLED
On 2019-05-31 5:31 p.m., Kuehling, Felix wrote:
> On 2019-05-31 5:19 p.m., Zeng, Oak wrote:
>> SDMA queue allocation requires the dqm lock as it modify
>> the global dqm members. Introduce functions to allocate/deallocate
>> in locked/unlocked circumst
On 2019-05-31 5:19 p.m., Zeng, Oak wrote:
> This is prepare work to fix a circular lock dependency.
> No logic change
>
> Change-Id: I4e0ee918260e7780de972dd71f4ce787b4f6dde9
> Signed-off-by: Oak Zeng
> ---
> .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 172
> -
>
On 2019-05-31 5:19 p.m., Zeng, Oak wrote:
> SDMA queue allocation requires the dqm lock as it modify
> the global dqm members. Introduce functions to allocate/deallocate
> in locked/unlocked circumstance.
>
> Change-Id: Id3084524c5f65d9629b12cf6b4862a7516945cb1
> Signed-off-by: Oak Zeng
> ---
>
On 2019-05-31 4:53 p.m., Dave Airlie wrote:
> On Sat, 1 Jun 2019 at 06:04, Kuehling, Felix wrote:
>> On 2019-05-30 11:13 p.m., Dave Airlie wrote:
>>> On Sat, 25 May 2019 at 05:48, Kuehling, Felix
>>> wrote:
>>>> On 2019-05-23 6:41 p.m., Zeng, Oak wrote
On 2019-05-30 11:13 p.m., Dave Airlie wrote:
> On Sat, 25 May 2019 at 05:48, Kuehling, Felix wrote:
>> On 2019-05-23 6:41 p.m., Zeng, Oak wrote:
>>> Add a new kfd ioctl to allocate queue GWS. Queue
>>> GWS is released on queue destroy.
>>>
>>> Change-
On 2019-05-31 1:28 p.m., Yang, Philip wrote:
>
> On 2019-05-30 6:36 p.m., Kuehling, Felix wrote:
>>>
>>> #if IS_ENABLED(CONFIG_DRM_AMDGPU_USERPTR)
>>> - if (gtt->ranges &&
>>> - ttm->pages[0] == hmm_pfn_to_page(>ranges[0],
This is a nice simplification. See a few comments inline.
On 2019-05-30 10:41 a.m., Yang, Philip wrote:
> HMM provides new APIs and helps in kernel 5.2-rc1 to simplify driver
> path. The old hmm APIs are deprecated and will be removed in future.
>
> Below are changes in driver:
>
> 1. Change
On 2019-05-29 2:32 a.m., Xiao, Jack wrote:
> Since amdgpu has always requested PCIE atomics, kfd don't
> need duplicated PCIE atomics enablement. Referring to amdgpu
> request result is enough.
>
> Signed-off-by: Jack Xiao
This patch is Reviewed-by: Felix Kuehling
> ---
>
On 2019-05-29 11:07 a.m., Colin King wrote:
> From: Colin Ian King
>
> The pointer dev is set to null yet it is being dereferenced when
> checking dev->dqm->sched_policy. Fix this by performing the check
> on dev->dqm->sched_policy after dev has been assigned and null
> checked. Also remove the
On 2019-05-28 3:02 p.m., Tejun Heo wrote:
> Hello,
>
> On Fri, May 17, 2019 at 08:12:17PM +, Kuehling, Felix wrote:
>> Patches 1,2,4 will be submitted through amd-staging-drm-next. Patch 3
>> goes through the cgroup tree. Patch 4 depends on patch 3. So submitting
>>
From: Philip Yang
While the page is migrating by NUMA balancing, HMM failed to detect this
condition and still return the old page. Application will use the new
page migrated, but driver pass the old page physical address to GPU, this
crash the application later.
Use pte_protnone(pte) to
These are two important HMM bug fixes to fix the HMM-based userptr
implementation. They are alread staged in MMOTS:
https://www.ozlabs.org/~akpm/mmots/broken-out/
Kuehling, Felix (1):
mm/hmm.c: only set FAULT_FLAG_ALLOW_RETRY for non-blocking
Philip Yang (1):
mm/hmm.c: support automatic NUMA
From: "Kuehling, Felix"
Don't set this flag by default in hmm_vma_do_fault. It is set
conditionally just a few lines below. Setting it unconditionally
can lead to handle_mm_fault doing a non-blocking fault, returning
-EBUSY and unlocking mmap_sem unexpectedly.
Link: http://lkml.ke
Always mark evicted queues with q->properties.is_evicted = true, even
queues that are inactive for other reason. This simplifies maintaining
the eviction state as it doesn't require updating is_evicted when other
queue activation conditions change.
On the other hand, we now need to check those
From: Jay Cornwall
Ported from gfx8, no changes in register setup.
Signed-off-by: Jay Cornwall
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 10 +++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git
New feature: queue priorities
The eviction state logic change is preparation for some debugger support
we're working on but haven't settled on the final ABI yet.
Felix Kuehling (1):
drm/amdkfd: Simplify eviction state logic
Jay Cornwall (1):
drm/amdkfd: Implement queue priority controls for
From: ozeng
Translate queue priority into pipe priority and write to MQDs.
The priority values are used to perform queue and pipe arbitration.
Signed-off-by: Oak Zeng
Reviewed-by: Felix Kuehling
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | 20
On 2019-05-28 2:28 p.m., Zeng, Oak wrote:
> The idea to break the circular lock dependency is, unlock dqm
> temporarily before calling init_mqd in call stack #1 (see below)
>
> [ 513.604034] ==
> [ 513.604205] WARNING: possible circular locking
On 2019-05-27 7:51 a.m., Christian König wrote:
> That idea sounds sane to me as well.
>
> By the way, do we somewhere signal to userspace if atomics are
> supported or not?
Yes. KFD topology (a flag in the iolink) provides that information to
user mode.
Regards,
Felix
>
> I mean would be
is nothing easier to evict.
ROCm applications like to use lots of memory. So it probably makes sense
for us to stop removing our BOs from the LRU as well while we
mass-validate our BOs in amdgpu_amdkfd_gpuvm_restore_process_bos.
Regards,
Felix
>
> Christian.
>
> Am 22.05.19 um 21:43 sc
On 2019-05-23 6:41 p.m., Zeng, Oak wrote:
> Add a new kfd ioctl to allocate queue GWS. Queue
> GWS is released on queue destroy.
>
> Change-Id: I60153c26a577992ad873e4292e759e5c3d5bbd15
> Signed-off-by: Oak Zeng
Reviewed-by: Felix Kuehling
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |
On 2019-05-23 6:41 p.m., Zeng, Oak wrote:
> Add functions in process queue manager to
> set/unset queue gws. Also set process's number
> of gws used. Currently only one queue in
> process can use and use all gws.
>
> Change-Id: I03e480c8692db3eabfc3a188cce8904d5d962ab7
> Signed-off-by: Oak Zeng
>
On 2019-05-24 3:12 p.m., Kim, Jonathan wrote:
> add pmu counters to monitor amdgpu device performance
>
> Change-Id: I8449f4ea824c411ee24a5b783ac066189b9de08e
> Signed-off-by: Jonathan Kim
> ---
> drivers/gpu/drm/amd/amdgpu/Makefile| 2 +-
>
Hi Oak,
I'm not sure why this is part of the GWS patch series. It's unrelated to
GWS. Anyway, see one comment inline.
On 2019-05-23 6:41 p.m., Zeng, Oak wrote:
> TTM doesn't support CPU mapping of sg type bo (under which
> mmio bo is created). Switch mmaping of mmio page to kfd
> device file.
>
On 2019-05-23 2:41 p.m., Zeng, Oak wrote:
> Add a new kfd ioctl to allocate queue GWS. Queue
> GWS is released on queue destroy.
>
> Change-Id: I60153c26a577992ad873e4292e759e5c3d5bbd15
> Signed-off-by: Oak Zeng
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 30
On 2019-05-23 2:41 p.m., Zeng, Oak wrote:
> Add functions in process queue manager to
> set queue gws. Also set process's number
> of gws used. Currently only one queue in
> process can use and use all gws.
>
> Change-Id: I03e480c8692db3eabfc3a188cce8904d5d962ab7
> Signed-off-by: Oak Zeng
> ---
>
On 2019-05-22 11:51 a.m., Zeng, Oak wrote:
> GWS bo is shared between all kfd processes. Add function to add gws
> to kfd process's bo list so gws can be evicted from and restored
> for process.
>
> Change-Id: I75d74cfdadb7075ff8b2b68634e205deb73dc1ea
> Signed-off-by: Oak Zeng
Reviewed-by: Felix
On 2019-05-22 11:51 a.m., Zeng, Oak wrote:
> Add a new kfd ioctl to allocate queue GWS. Queue
> GWS is released on queue destroy.
>
> Change-Id: I60153c26a577992ad873e4292e759e5c3d5bbd15
> Signed-off-by: Oak Zeng
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 39
>
On 2019-05-22 11:51 a.m., Zeng, Oak wrote:
> Add functions in process queue manager to
> set/get queue gws. Also set process's number
> of gws used. Currently only one queue in
> process can use and use all gws.
>
> Change-Id: I03e480c8692db3eabfc3a188cce8904d5d962ab7
> Signed-off-by: Oak Zeng
>
Can you explain how this avoids OOM situations? When is it safe to leave
a reserved BO on the LRU list? Could we do the same thing in
amdgpu_amdkfd_gpuvm.c? And if we did, what would be the expected side
effects or consequences?
Thanks,
Felix
On 2019-05-22 8:59 a.m., Christian König wrote:
On 2019-05-15 4:37 p.m., Zeng, Oak wrote:
> Add amdgpu_amdkfd interface to get num_gws and add num_gws
> to /sys/class/kfd/kfd/topology/nodes/x/properties. Only report
> num_gws if MEC FW support GWS barriers. Currently it is
> determined by a environment variable which will be replaced
It's not
On 2019-05-15 4:37 p.m., Zeng, Oak wrote:
> Add functions in process queue manager to
> set/get queue gws. Also set process's number
> of gws used. Currently only one queue in
> process can use and use all gws.
>
> Change-Id: I03e480c8692db3eabfc3a188cce8904d5d962ab7
> Signed-off-by: Oak Zeng
>
On 2019-05-15 4:37 p.m., Zeng, Oak wrote:
> [CAUTION: External Email]
>
> Add a new kfd ioctl to allocate queue GWS. Queue
> GWS is released on queue destroy.
>
> Change-Id: I60153c26a577992ad873e4292e759e5c3d5bbd15
> Signed-off-by: Oak Zeng
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
On 2019-05-15 4:37 p.m., Zeng, Oak wrote:
> GWS bo is shared between all kfd processes. Add function to add gws
> to kfd process's bo list so gws can be evicted from and restored
> for process.
>
> Change-Id: I75d74cfdadb7075ff8b2b68634e205deb73dc1ea
> Signed-off-by: Oak Zeng
> ---
>
On 2019-05-15 4:37 p.m., Zeng, Oak wrote:
> Add functions in process queue manager to
> set/get queue gws. Also set process's number
> of gws used. Currently only one queue in
> process can use and use all gws.
>
> Change-Id: I03e480c8692db3eabfc3a188cce8904d5d962ab7
> Signed-off-by: Oak Zeng
>
Patches 1,2,4 will be submitted through amd-staging-drm-next. Patch 3
goes through the cgroup tree. Patch 4 depends on patch 3. So submitting
patch 4 will need to wait until we rebase amd-staging-drm-next on a new
enough kernel release that includes patch 3.
Patch 1 and 2 could be submitted
Hi Daniel,
On 2019-05-12 9:44 p.m., Daniel Kasak wrote:
> [CAUTION: External Email]
> Hi all. I had version 2.2.0 of the ROCM stack running on a 5.0.x and
> 5.1.0 kernel. Things were going great with various boinc GPU tasks.
> But there is a setiathome GPU task which reliably gives me a hard
>
On 2019-05-13 5:27 p.m., Andrew Morton wrote:
> [CAUTION: External Email]
>
> On Fri, 10 May 2019 19:53:23 + "Kuehling, Felix"
> wrote:
>
>> From: Philip Yang
>>
>> While the page is migrating by NUMA balancing, HMM failed to detect this
> which did not have a clear line of sight for 5.2 either.
When was that? I saw "Use HMM for userptr" in Dave's 5.2-rc1 pull
request to Linus.
Regards,
Felix
>
> Alex
>
> *From:* amd-gfx on
[Fixed Alex's email address, sorry for getting it wrong first]
On 2019-05-13 3:49 p.m., Jerome Glisse wrote:
> [CAUTION: External Email]
>
> Andrew can we get this 2 fixes line up for 5.2 ?
>
> On Mon, May 13, 2019 at 07:36:44PM +0000, Kuehling, Felix wrote:
>> Hi Jerome,
7:53:24PM +, Kuehling, Felix wrote:
>> Don't set this flag by default in hmm_vma_do_fault. It is set
>> conditionally just a few lines below. Setting it unconditionally
>> can lead to handle_mm_fault doing a non-blocking fault, returning
>> -EBUSY and unlocking mmap_s
On 2019-05-13 12:03 p.m., Zeng, Oak wrote:
> Hi Felix,
>
> See comments inline [Oak]
>
> Hi Kent, there is one FYI embedded, so be careful when you merge this change
> back to kfd-staging branch.
>
> Regards,
> Oak
>
> -----Original Message-
> From: Kue
On 2019-05-10 12:01 p.m., Zeng, Oak wrote:
> Add a new kfd ioctl to allocate queue GWS. Queue
> GWS is released on queue destroy.
>
> Change-Id: I60153c26a577992ad873e4292e759e5c3d5bbd15
> Signed-off-by: Oak Zeng
> ---
> drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 45
>
The subject for this one should start with drm/amdgpu: because it's a
change in amdgpu, not KFD.
See two more comments inline.
Regards,
Felix
On 2019-05-10 12:01 p.m., Zeng, Oak wrote:
> [CAUTION: External Email]
>
> GWS bo is shared between all kfd processes. Add function to add gws
> to
From: Philip Yang
While the page is migrating by NUMA balancing, HMM failed to detect this
condition and still return the old page. Application will use the new
page migrated, but driver pass the old page physical address to GPU,
this crash the application later.
Use pte_protnone(pte) to return
Don't set this flag by default in hmm_vma_do_fault. It is set
conditionally just a few lines below. Setting it unconditionally
can lead to handle_mm_fault doing a non-blocking fault, returning
-EBUSY and unlocking mmap_sem unexpectedly.
Signed-off-by: Felix Kuehling
---
mm/hmm.c | 2 +-
1 file
These problems were found in AMD-internal testing as we're working on
adopting HMM. They are rebased against glisse/hmm-5.2-v3. We'd like to get
them applied to a mainline Linux kernel as well as drm-next and
amd-staging-drm-next sooner rather than later.
Currently the HMM in amd-staging-drm-next
Use unsigned long for number of pages.
Check that pfns are valid after hmm_vma_fault. If they are not,
return an error instead of continuing with invalid page pointers and
PTEs.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 22 ++
1 file
This patch untag user pointers in
> amdgpu_gem_userptr_ioctl() for the GEM case and in amdgpu_amdkfd_gpuvm_
> alloc_memory_of_gpu() for the KFD case. This also makes sure that an
> untagged pointer is passed to amdgpu_ttm_tt_get_user_pages(), which uses
> it for vma lookups.
>
> Suggested-by: Kuehling,
On 2019-05-06 12:30 p.m., Andrey Konovalov wrote:
> [CAUTION: External Email]
>
> This patch is a part of a series that extends arm64 kernel ABI to allow to
> pass tagged user pointers (with the top byte set to something else other
> than 0x00) as syscall arguments.
>
> In
On 2019-05-06 7:24 a.m., Christian König wrote:
> [CAUTION: External Email]
>
> We are getting a dma-buf implementation completely separate from drm prime,
> so rename the files now and cleanup the code a bit.
>
> No functional change.
>
> Signed-off-by: Christian König
Acked-by: Felix Kuehling
Need to reserve space for the shared eviction fence when initializing
a KFD VM.
Signed-off-by: Felix Kuehling
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 4
1 file changed, 4 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
On 2019-04-30 10:48 a.m., Trigger Huang wrote:
> [CAUTION: External Email]
>
> amdgpu_vm_make_compute is used to turn a GFX VM into a compute VM,
> the prerequisite is this VM is clean. Let's check if some page tables
> are already filled , while not check if some mapping is already made.
>
>
Thank you Kent! The series is Reviewed-by: Felix Kuehling
On 2019-05-01 8:31 a.m., Russell, Kent wrote:
> [CAUTION: External Email]
>
> Fix some spacing issues, log output, uses of !=NULL/==NULL, unneeded
> extra lines and clean up a declaration from =1 to =true for clarity
>
> Change-Id:
and in
> amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu() for the KFD case.
>
> Suggested-by: Kuehling, Felix
> Signed-off-by: Andrey Konovalov
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 2 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 ++
> drivers/gpu/drm/a
On 2019-04-30 9:25 a.m., Andrey Konovalov wrote:
> [CAUTION: External Email]
>
> This patch is a part of a series that extends arm64 kernel ABI to allow to
> pass tagged user pointers (with the top byte set to something else other
> than 0x00) as syscall arguments.
>
> radeon_ttm_tt_pin_userptr()
On 2019-04-30 1:03 p.m., Koenig, Christian wrote:
> Am 30.04.19 um 17:36 schrieb Kuehling, Felix:
>> On 2019-04-30 5:32 a.m., Christian König wrote:
>>> [CAUTION: External Email]
>>>
>>> Am 30.04.19 um 01:16 schrieb Kuehling, Felix:
>>>> On 2019-
On 2019-04-30 5:32 a.m., Christian König wrote:
> [CAUTION: External Email]
>
> Am 30.04.19 um 01:16 schrieb Kuehling, Felix:
>> On 2019-04-29 8:34 a.m., Christian König wrote:
>>> Am 28.04.19 um 09:44 schrieb Kuehling, Felix:
>>>> From: Kent Russell
>
I remember a past discussion to change the CSA allocation/mapping scheme
to avoid this issue in the first place. Can adding the CSA to the VM be
delayed a little to a point after the VM gets converted to a compute VM?
Maybe the first command submission?
Regards,
Felix
On 2019-04-28 6:25
,
Felix
On 2019-04-28 3:44 a.m., Kuehling, Felix wrote:
> Assorted KFD changes that have been accumulating on amd-kfd-staging. New
> features and fixes included:
> * Support for VegaM
> * Support for systems with multiple PCI domains
> * New SDMA queue type that's optimized for XGMI lin
On 2019-04-29 8:34 a.m., Christian König wrote:
> Am 28.04.19 um 09:44 schrieb Kuehling, Felix:
>> From: Kent Russell
>>
>> GTT size is currently limited to the minimum of VRAM size or 3/4 of
>> system memory. This severely limits the quanitity of system memory
From: Jay Cornwall
When MEM_VIOL is asserted the context save handler rewinds the
program counter. This is incorrect for any source of the exception.
MEM_VIOL may be raised in normal operation by out-of-bounds access
to LDS or GDS and does not require special handling.
Remove PC adjustment when
201 - 300 of 545 matches
Mail list logo