On 5/15/2025 8:30 PM, Christian König wrote:
Just to exercise the functionality.
Signed-off-by: Christian König
---
drivers/gpu/drm/scheduler/tests/tests_basic.c | 59 ++-
1 file changed, 58 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/scheduler/tests/tests_ba
Reviewed-by: Arvind Yadav
On 5/15/2025 6:49 PM, Liang, Prike wrote:
[Public]
[Public]
I haven't cleaned up the userq resource destroy at postclose callback in my
last patch, so here please remove the duplicated useq destroy. With that, the
change in the patch is
Reviewed-by: Prike Liang
R
Reviewed-by: Arvind Yadav
On 5/14/2025 9:02 PM, Alex Deucher wrote:
On Sun, May 11, 2025 at 10:21 PM Prike Liang wrote:
Lock and refer to the eviction fence before the eviction fence
schedules work queue tries to signal it.
Suggested-by: Christian König
Signed-off-by: Prike Liang
Acked-by
These log messages are overwhelming during the boot process.
Acked-by: Arvind Yadav
~arvind
On 5/13/2025 7:05 PM, Harry Wentland wrote:
On 2025-05-12 23:20, Wayne Lin wrote:
It's expected that we'll encounter temporary exceptions
during aux transactions. Adjust logging from d
Reviewed-by: Arvind Yadav
On 5/9/2025 4:48 PM, Christian König wrote:
On 5/9/25 11:21, Jesse.Zhang wrote:
This resolves a deadlock between user queue management and GPU reset
paths by enforcing consistent lock ordering.
The deadlock occurred when:
1. Process exit path (amdgpu_userq_mgr_fini)
On 5/8/2025 2:52 PM, Christian König wrote:
On 5/8/25 07:04, Yadav, Arvind wrote:
On 5/8/2025 12:36 AM, Alex Deucher wrote:
On Wed, May 7, 2025 at 2:38 PM Arvind Yadav wrote:
Switch cancel_delayed_work() to cancel_delayed_work_sync() to ensure
the delayed work has finished executing before
On 5/8/2025 12:36 AM, Alex Deucher wrote:
On Wed, May 7, 2025 at 2:38 PM Arvind Yadav wrote:
Switch cancel_delayed_work() to cancel_delayed_work_sync() to ensure
the delayed work has finished executing before proceeding with
resource cleanup. This prevents a potential use-after-free or
NULL d
Reviewed-by: Arvind Yadav
On 5/7/2025 3:05 PM, Khatri, Sunil wrote:
[AMD Official Use Only - AMD Internal Distribution Only]
@Yadav, Arvind <mailto:arvind.ya...@amd.com>
-Original Message-
From: Sunil Khatri
Sent: Wednesday, May 7, 2025 3:00 PM
To: amd-gfx@lists.freedeskt
I also thought for previous patch but else was doing that. We can use
something like this.
just alternative solution.
if (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(10, 0, 0) ||
(!AMDGPU_IS_MMHUB0(i) && !AMDGPU_IS_MMHUB1(i)))
id_mgr->num_ids = adev->vm_manager.first_kfd_vm
Reviewed-by: Arvind Yadav
On 4/29/2025 11:20 PM, Alex Deucher wrote:
When kernel queues are disabled, all GC vmids are available
for the scheduler. MM vmids are still managed by the driver
so make all 16 available.
Also fix gmc 10 vs 11 mix up in
commit 1f61fc28b939 ("drm/amdgpu/mes: make mor
On 4/29/2025 9:38 PM, Christian König wrote:
On 4/29/25 17:24, Arvind Yadav wrote:
Introduce a new `enum dma_fence_unwrap_mode` to support multiple modes
of fence unwrapping and merging behavior. This lays the groundwork for
more flexible and efficient fence operations.
Add a new macro `dma_f
Reviewed-by: Arvind Yadav
On 4/26/2025 12:04 AM, Alex Deucher wrote:
Add the missing locking.
Fixes: 94976e7e5ede ("drm/amdgpu/userq: add helpers to start/stop scheduling")
Signed-off-by: Alex Deucher
---
drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 4
1 file changed, 4 insertions(+)
Reviewed-by: Arvind Yadav
On 4/26/2025 12:04 AM, Alex Deucher wrote:
Add the missing locking.
Fixes: 73e12e98ec0c ("drm/amdgpu/userq: add suspend and resume helpers")
Signed-off-by: Alex Deucher
---
drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 4
1 file changed, 4 insertions(+)
diff -
This is problem for TLB flush. We should not do this changes. Here we
are utilizing DMA_RESV_USAGE_BOOKKEEP due to the TLB flush fence
associated with the page table (PT). We are ensuring that no page
directory (PD) or page table (PT) should be free before flush and ttm bo
release and delete bo
Reviewed-by: Arvind Yadav
On 4/24/2025 2:08 PM, Pierre-Eric Pelloux-Prayer wrote:
Log fences using the same format for coherency.
Signed-off-by: Pierre-Eric Pelloux-Prayer
Reviewed-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h | 22 ++
1 file change
Reviewed-by: Arvind Yadav
On 4/24/2025 2:08 PM, Pierre-Eric Pelloux-Prayer wrote:
Its only purpose was for trace events, but jobs can already be
uniquely identified using their fence.
The downside of using the fence is that it's only available
after 'drm_sched_job_arm' was called which is true
existing code and not just add a new function.
sure, I will not add new function and try to re-use existing.
Thankyou,
~arvind
Regards,
Christian.
Regards
Sunil Khatri
-Original Message-
From: Yadav, Arvind
Sent: Wednesday, April 23, 2025 4:18 PM
To: Koenig, Christian ; Deucher
Reviewed-by: Arvind Yadav
On 4/22/2025 7:06 PM, Christian König wrote:
Am 22.04.25 um 15:23 schrieb Prike Liang:
This debug dump will help on debugging the evf attached gem obj fence
related issue.
Signed-off-by: Prike Liang
Reviewed-by: Christian König for both patches.
---
drivers/gp
Reviewed-by: Arvind Yadav
On 4/22/2025 6:53 PM, Prike Liang wrote:
The evf fence name can clearly identify the userq usage.
Signed-off-by: Prike Liang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_eviction_fence.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/
On 4/22/2025 2:57 PM, Christian König wrote:
Am 22.04.25 um 11:14 schrieb Liang, Prike:
[Public]
-Original Message-
From: Christian König
Sent: Thursday, April 17, 2025 3:40 PM
To: Liang, Prike ; Koenig, Christian
; amd-gfx@lists.freedesktop.org
Cc: Deucher, Alexander
Subject: Re:
Reviewed-by:Arvind Yadav
On 4/21/2025 6:01 PM, Khatri, Sunil wrote:
[AMD Official Use Only - AMD Internal Distribution Only]
-Original Message-
From: Sunil Khatri
Sent: Monday, April 21, 2025 5:51 PM
To: Deucher, Alexander ; Koenig, Christian
Cc: amd-gfx@lists.freedesktop.org; Khatr
On 4/15/2025 7:53 PM, Christian König wrote:
Am 15.04.25 um 16:18 schrieb Arvind Yadav:
Mesa passes shared bo, fence syncobj to userq_ioctl.
There can be duplicates here or some fences that are old.
This patch is remove duplicates fence and only keep
the most recent fence for each context.
Cc
Reviewed-by:Arvind Yadav
On 4/12/2025 8:09 PM, Dan Carpenter wrote:
1) Checkpatch complains if we print an error message for kzalloc()
failure. The kzalloc() failure already has it's own error messages
built in. Also this allocation is small enough that it is guaranteed
to succeed
Reviewed-by:Reviewed-by:Arvind Yadav
On 4/12/2025 8:09 PM, Dan Carpenter wrote:
The goto frees "fence_drv" so this is a double free bug. There is no
need to call amdgpu_seq64_free(adev, fence_drv->va) since the seq64
allocation failed so change the goto to goto free_fence_drv. Also
propagate
Alex,
This is v2 of 2/2 patch. Please review this.
~arvind
On 4/10/2025 8:27 PM, Arvind Yadav wrote:
This patch is load usermode queue based on FW support for gfx12.
CP Ucode FW Vesion: [PFP = 2840, ME = 2780, MEC = 3050, MES = 123]
v2: Addressed review comments from Alex
- Just check the
Please change this also instead of 'goto free_fence_drv' just return err.
fence_drv = kzalloc(sizeof(*fence_drv), GFP_KERNEL);
if (!fence_drv) {
DRM_ERROR("Failed to allocate memory for fence driver\n");
r = -ENOMEM;
goto free_fence_
On 4/10/2025 8:50 PM, Alex Deucher wrote:
On Thu, Apr 10, 2025 at 10:57 AM Arvind Yadav wrote:
This patch is load usermode queue based on FW support for gfx11.
CP Ucode FW version: [PFP = 2530, ME = 2390, MEC = 2600, MES = 120]
v2: Addressed review comments from Alex.
- Just check the f
On 4/10/2025 7:01 PM, Alex Deucher wrote:
On Thu, Apr 10, 2025 at 7:48 AM Arvind Yadav wrote:
This patch is load usermode queue based on FW support for gfx12.
CP Ucode FW Vesion: [PFP = 2840, ME = 2780, MEC = 2600, MES = 123]
Cc: Alex Deucher
Cc: Christian Koenig
Cc: Shashank Sharma
Cc: S
On 2/27/2025 9:12 PM, Christian König wrote:
No, no that's wrong.
That timeline_syncobj is NULL is not an error. In other words when
args->vm_timeline_syncobj_out == 0 then amdgpu_gem_update_timeline_node()
should just set timeline_syncobj=NULL and return 0.
The error happens only if either
On 2/27/2025 7:55 PM, Christian König wrote:
Am 18.02.25 um 15:53 schrieb Arvind Yadav:
Encountering a taint issue during the unloading of gpu_sched
due to the fence not being released/put. In this context,
amdgpu_vm_clear_freed is responsible for creating a job to
update the page table (PT).
ed to
the caller, and the caller must keep one reference at least until this fence is
no longer needed.
Thanks
River
-Original Message-
From: amd-gfx On Behalf Of Yadav, Arvind
Sent: Friday, February 14, 2025 7:42 PM
To: Koenig, Christian ; Ma, Le ;
amd-gfx@lists.freedesktop.org;
On 2/14/2025 4:08 PM, Christian König wrote:
Adding Arvind, please make sure to keep him in the loop.
Am 14.02.25 um 11:07 schrieb Le Ma:
On systems with CONFIG_SLUB_DEBUG enabled, the memleak like below
will show up explicitly during driver unloading if created bo without
drm_timeline object
On 10/4/2023 10:29 PM, Felix Kuehling wrote:
On 2023-10-04 12:16, Arvind Yadav wrote:
This patch is to align the absolute doorbell offset
based on the doorbell's size. So that doorbell offset
will be aligned for both 32 bit and 64 bit.
v2:
- Addressed the review comment from Felix.
v3:
- Add
Adding felix.kuehl...@amd.com for review.
Thanks
~Arvind
On 9/27/2023 9:46 PM, Arvind Yadav wrote:
This patch is to adjust the absolute doorbell offset
against the doorbell id considering the doorbell
size of 32/64 bit.
Cc: Christian Koenig
Cc: Alex Deucher
Signed-off-by: Shashank Sharma
Si
Adding felix.kuehl...@amd.com for review.
Thanks
~Arvind
On 9/27/2023 9:46 PM, Arvind Yadav wrote:
On older chips, the absolute doorbell offset within
the doorbell page is based on the queue ID.
KFD is using queue ID and doorbell size to get an
absolute doorbell offset in userspace.
This patch
On 8/28/2023 9:13 PM, Helen Mae Koike Fornazier wrote:
On Monday, August 28, 2023 09:26 -03, Arvind Yadav wrote:
AMDGPU SOCs supports dynamic workload based power profiles, which can
provide fine-tuned performance for a particular type of workload.
This patch series adds an interface to set/
On 8/22/2023 6:16 PM, Lazar, Lijo wrote:
On 8/22/2023 5:41 PM, Yadav, Arvind wrote:
Hi Lijo,
The *_set function will set the GPU power profile and the *_put
function will schedule the
smu_delayed_work task after 100ms delay. This smu_delayed_work task
will clear a GPU
power profile if
On 8/22/2023 6:24 PM, Lazar, Lijo wrote:
On 8/22/2023 5:52 PM, Yadav, Arvind wrote:
On 8/22/2023 12:01 PM, Lazar, Lijo wrote:
On 8/21/2023 12:17 PM, Arvind Yadav wrote:
This patch adds a suspend function that will clear the GPU
power profile before going into suspend state.
v2:
- Add
On 8/22/2023 11:55 AM, Lazar, Lijo wrote:
On 8/21/2023 12:17 PM, Arvind Yadav wrote:
This patch adds a function which will change the GPU
power profile based on a submitted job. This can optimize
the power performance when the workload is on.
v2:
- Splitting workload_profile_set and workloa
On 8/22/2023 12:01 PM, Lazar, Lijo wrote:
On 8/21/2023 12:17 PM, Arvind Yadav wrote:
This patch adds a suspend function that will clear the GPU
power profile before going into suspend state.
v2:
- Add the new suspend function based on review comment.
Cc: Shashank Sharma
Cc: Christian Koen
Hi Lijo,
The *_set function will set the GPU power profile and the *_put function
will schedule the
smu_delayed_work task after 100ms delay. This smu_delayed_work task will
clear a GPU
power profile if any new jobs are not scheduled within 100 ms. But if
any new job comes within 100ms
then t
On 8/21/2023 11:40 PM, Alex Deucher wrote:
On Mon, Aug 21, 2023 at 1:54 PM Yadav, Arvind wrote:
On 8/21/2023 9:52 PM, Alex Deucher wrote:
On Mon, Aug 21, 2023 at 2:55 AM Arvind Yadav wrote:
This patch adds a function which will change the GPU
power profile based on a submitted job. This
On 8/21/2023 11:36 PM, Alex Deucher wrote:
On Mon, Aug 21, 2023 at 2:55 AM Arvind Yadav wrote:
This patch adds a function which will change the GPU
power profile based on a submitted job. This can optimize
the power performance when the workload is on.
v2:
- Splitting workload_profile_set an
On 8/21/2023 9:52 PM, Alex Deucher wrote:
On Mon, Aug 21, 2023 at 2:55 AM Arvind Yadav wrote:
This patch adds a function which will change the GPU
power profile based on a submitted job. This can optimize
the power performance when the workload is on.
v2:
- Splitting workload_profile_set and
On 8/21/2023 7:09 PM, Shashank Sharma wrote:
On 21/08/2023 08:47, Arvind Yadav wrote:
This patch adds a function which will clear the GPU
power profile after job finished.
This is how it works:
- schedular will set the GPU power profile based on ring_type.
- Schedular will clear the GPU Powe
On 8/21/2023 7:24 PM, Shashank Sharma wrote:
On 21/08/2023 15:35, Yadav, Arvind wrote:
On 8/21/2023 6:36 PM, Shashank Sharma wrote:
Hey Arvind,
On 21/08/2023 08:47, Arvind Yadav wrote:
The'struct amdgpu_smu_workload' initialization/cleanup
functions is added by this
On 8/21/2023 7:16 PM, Shashank Sharma wrote:
On 21/08/2023 08:47, Arvind Yadav wrote:
This patch is to switch the GPU workload profile based
on the submitted job. The workload profile is reset to
default when the job is done.
Cc: Shashank Sharma
Cc: Christian Koenig
Cc: Alex Deucher
Signe
On 8/21/2023 7:13 PM, Shashank Sharma wrote:
On 21/08/2023 08:47, Arvind Yadav wrote:
This patch adds a suspend function that will clear the GPU
power profile before going into suspend state.
v2:
- Add the new suspend function based on review comment.
Cc: Shashank Sharma
Cc: Christian Koen
On 8/21/2023 6:36 PM, Shashank Sharma wrote:
Hey Arvind,
On 21/08/2023 08:47, Arvind Yadav wrote:
The'struct amdgpu_smu_workload' initialization/cleanup
functions is added by this patch.
v2:
- Splitting big patch into separate patches.
- Added new fini function.
Cc: Shashank Sharma
Cc: Chr
On 8/14/2023 8:28 PM, Shashank Sharma wrote:
Hey Arvind,
On 14/08/2023 09:34, Arvind Yadav wrote:
This patch adds a function which will allow to
change the GPU power profile based on a submitted job.
This can optimize the power performance when the
workload is on.
Cc: Shashank Sharma
Cc: Ch
On 8/14/2023 9:35 PM, Shashank Sharma wrote:
Ah, Thanks for pointing that out Alex.
@Arvind, please refer to the patch
(https://patchwork.freedesktop.org/patch/504854/?series=109060&rev=4)
in previous series of SMU workload hints with UAPI (here:
https://patchwork.freedesktop.org/series/109
On 8/14/2023 8:03 PM, Alex Deucher wrote:
On Mon, Aug 14, 2023 at 3:35 AM Arvind Yadav wrote:
This patch adds a function which will allow to
change the GPU power profile based on a submitted job.
This can optimize the power performance when the
workload is on.
A few minor comments inline bel
On 8/14/2023 5:35 PM, Christian König wrote:
Am 14.08.23 um 09:34 schrieb Arvind Yadav:
This patch adds a function which will allow to
change the GPU power profile based on a submitted job.
This can optimize the power performance when the
workload is on.
Cc: Shashank Sharma
Cc: Christian Koe
On 3/31/2023 2:55 AM, Alex Deucher wrote:
On Wed, Mar 29, 2023 at 12:05 PM Shashank Sharma
wrote:
From: Arvind Yadav
This patch:
- adds some new parameters defined for the gfx usermode queues
use cases in the v11_mqd_struct.
- sets those parametes with the respective allocated gpu contex
On 1/4/2023 8:05 PM, Felix Kuehling wrote:
Am 2023-01-04 um 04:23 schrieb Shashank Sharma:
On 04/01/2023 10:17, Christian König wrote:
Am 04.01.23 um 10:13 schrieb Shashank Sharma:
On 04/01/2023 10:10, Christian König wrote:
Am 04.01.23 um 07:21 schrieb Yadav, Arvind:
On 1/4/2023 12:07
On 1/4/2023 12:07 AM, Felix Kuehling wrote:
Am 2023-01-03 um 04:36 schrieb Shashank Sharma:
/*MQD struct for usermode Queue*/
+struct amdgpu_usermode_queue_mqd
This is specific to GC 11. Every IP and version will have its own MQD
format. That should live in the IP specific code, not the gen
56 matches
Mail list logo
