Am 17.06.24 um 17:35 schrieb Xi Ruoyao:
On Mon, 2024-06-17 at 22:30 +0800, Icenowy Zheng wrote:
Two consecutive writes to the same bus address are perfectly legal
from
the PCIe specification and can happen all the time, even without this
specific hw workaround.
Yes I know it, and I am not from
Am 17.06.24 um 18:09 schrieb Icenowy Zheng:
BTW is there any operation that could be taken to examine this specific
workaround?
Is there any case possible to reproduce?
No idea, I mean that's for GFX7/8 which was released between 2013 and 2017.
My educated guess is that you could create a tes
Am 18.06.24 um 11:11 schrieb Pavel Machek:
Hi!
[ Upstream commit a0cf36546cc24ae1c95d72253c7795d4d2fc77aa ]
The pointer parent may be NULLed by the function amdgpu_vm_pt_parent.
To make the code more robust, check the pointer parent.
If this can happen, it should not WARN().
If this can not
Am 18.06.24 um 17:23 schrieb Pierre-Eric Pelloux-Prayer:
Waking up a device can take multiple seconds, so if it's not
going to be used we might as well not resume it.
The safest default behavior for all ioctls is to resume the GPU,
so this change allows specific ioctls to opt-out of generic
runt
I would try to avoid that.
Putting everything into amdgpu_info_device was a mistake only done
because people assumed that IOCTLs on Linux are to expensive to query
all information separately.
We should rather have distinct IOCTLs for each value because that is way
more flexible and we won't
Am 20.06.24 um 02:30 schrieb Yunxiang Li:
Before, every time fdinfo is queried we try to lock all the BOs in the
VM and calculate memory usage from scratch. This works okay if the
fdinfo is rarely read and the VMs don't have a ton of BOs. If either of
these conditions is not true, we get a massiv
Am 20.06.24 um 15:06 schrieb Pierre-Eric Pelloux-Prayer:
Le 19/06/2024 à 11:26, Christian König a écrit :
Am 18.06.24 um 17:23 schrieb Pierre-Eric Pelloux-Prayer:
Waking up a device can take multiple seconds, so if it's not
going to be used we might as well not resume it.
The safest de
Am 20.06.24 um 16:30 schrieb Li, Yunxiang (Teddy):
[Public]
+ dma_resv_lock(bo->tbo.base.resv, NULL);
Why do you grab the BO lock to update the stats? That doesn't seem to make
any sense.
+ update_stats = !(bo->flags & AMDGPU_GEM_WAS_EXPORTED);
+ if (update_stats)
+ amdgpu_bo
Am 20.06.24 um 16:44 schrieb Thomas Zimmermann:
Prepares for using ttm_bo_vmap() and ttm_bo_vunmap() in amdgpu. Both
require the caller to hold the GEM reservation lock, which is not the
case while releasing a buffer object. Hence, push a possible call to
unmap out from the buffer-object release
Am 20.06.24 um 15:55 schrieb Danijel Slivka:
Clearing the IH_RB_W/RPTR during interrupts disable is not clearing
the RB_OVERFLOW bit.
Adding workaround to clear the wptr when enabling interrupts in case
RB_OVERFLOW bit is set.
Signed-off-by: Danijel Slivka
---
drivers/gpu/drm/amd/amdgpu/ih_v6
Am 21.06.24 um 09:16 schrieb Thomas Zimmermann:
Hi
Am 20.06.24 um 16:30 schrieb Hoi Pok Wu:
Dear Thomas,
Thank you for testing my patch. The dev->dev_private is indeed the
problem.
However, most of the functions that uses dev->dev_private is passing
drm_device as parameter, and then uses de
Am 21.06.24 um 09:32 schrieb Thomas Zimmermann:
Hi
Am 20.06.24 um 17:50 schrieb Christian König:
Am 20.06.24 um 16:44 schrieb Thomas Zimmermann:
Prepares for using ttm_bo_vmap() and ttm_bo_vunmap() in amdgpu. Both
require the caller to hold the GEM reservation lock, which is not the
case
Am 21.06.24 um 11:24 schrieb Bob Zhou:
Fix the unchecked return value warning reported by Coverity,
so add error handling.
That seems to be completely superfluous. The only case when
amdgpu_ring_alloc() returns an error is when we try to allocate more
than the maximum submission size.
And i
This reverts commit 358c258a816baed4c6997b59c2117578a1360498.
Jerome actually pointed out why that stuff doesn't work in 2016:
https://lists.freedesktop.org/archives/dri-devel/2016-March/103062.html
Unfortunately the revert somehow got lost.
Signed-off-by: Christian König
Am 24.06.24 um 08:58 schrieb Danijel Slivka:
Why:
Setting IH_RB_WPTR register to 0 will not clear the RB_OVERFLOW bit
if RB_ENABLE is not set.
How to fix:
Set WPTR_OVERFLOW_CLEAR bit after RB_ENABLE bit is set.
The RB_ENABLE bit is required to be set, together with
WPTR_OVERFLOW_ENABLE bit so th
Am 24.06.24 um 11:13 schrieb Jane Jian:
[WHY]
sriov has the higher bit violation when flushing tlb
[HOW]
normalize the registers to keep lower 16-bit(dword aligned) to aviod higher bit
violation
RLCG will mask xcd out and always assume it's accessing its own xcd
[TODO]
later will add the norma
Am 24.06.24 um 08:34 schrieb Lazar, Lijo:
On 6/24/2024 12:01 PM, Vignesh Chander wrote:
correctly handle the case when trylock fails when gpu is
about to be reset by dropping the request instead of using mmio
Signed-off-by: Vignesh Chander
Reviewed-by: Lijo Lazar
Thanks,
Lijo
---
drive
Am 24.06.24 um 11:52 schrieb Lazar, Lijo:
On 6/24/2024 3:08 PM, Christian König wrote:
Am 24.06.24 um 08:34 schrieb Lazar, Lijo:
On 6/24/2024 12:01 PM, Vignesh Chander wrote:
correctly handle the case when trylock fails when gpu is
about to be reset by dropping the request instead of using
ions.
Regards,
Christian.
Thanks,
Jane
-Original Message-
From: Christian König
Sent: Monday, June 24, 2024 5:35 PM
To: Jian, Jane ; Lazar, Lijo ; Chang, HaiJun
; Zhao, Victor
Cc: amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH] drm/amdgpu: normalize registers as local xcc to read/write
eg_id,
tmp))
+ return -ETIMEDOUT;
+ } else {
+ WREG32_NO_KIQ(ih_regs->ih_rb_cntl, tmp);
+ }
The last write can be merged with the one below to eventually enable the
interrupt. So this chunk here can be droppe
Am 24.06.24 um 11:13 schrieb Jane Jian:
[WHY]
sriov has the higher bit violation when flushing tlb
[HOW]
normalize the registers to keep lower 16-bit(dword aligned) to aviod higher bit
violation
RLCG will mask xcd out and always assume it's accessing its own xcd
[TODO]
later will add the norma
Am 24.06.24 um 13:57 schrieb Lazar, Lijo:
On 6/24/2024 5:19 PM, Christian König wrote:
Am 24.06.24 um 11:52 schrieb Lazar, Lijo:
On 6/24/2024 3:08 PM, Christian König wrote:
Am 24.06.24 um 08:34 schrieb Lazar, Lijo:
On 6/24/2024 12:01 PM, Vignesh Chander wrote:
correctly handle the case
Am 24.06.24 um 14:24 schrieb Lazar, Lijo:
On 6/24/2024 5:31 PM, Christian König wrote:
Am 24.06.24 um 13:57 schrieb Lazar, Lijo:
On 6/24/2024 5:19 PM, Christian König wrote:
Am 24.06.24 um 11:52 schrieb Lazar, Lijo:
On 6/24/2024 3:08 PM, Christian König wrote:
Am 24.06.24 um 08:34 schrieb
al Message-----
From: Christian König
Sent: Monday, June 24, 2024 7:58 PM
To: Jian, Jane ; Lazar, Lijo ; Chang, HaiJun
; Zhao, Victor
Cc: amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH] drm/amdgpu: normalize registers as local xcc to read/write
under sriov in TLB flush
Am 24.06.24 um 11:13 sc
Am 25.06.24 um 14:41 schrieb Pierre-Eric Pelloux-Prayer:
The call to radeon_vm_clear_freed might clear bo_va->bo, so
we have to check it before dereferencing it.
Signed-off-by: Pierre-Eric Pelloux-Prayer
---
drivers/gpu/drm/radeon/radeon_gem.c | 11 ++-
1 file changed, 10 insertions(
Jian
Lijo has a better technical understanding of the background than I do,
so he should have the last word.
But this looks like I would have expected it to look like so feel free
to add Acked-by: Christian König .
Regards,
Christian.
---
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
Am 25.06.24 um 19:44 schrieb Alex Deucher:
On Tue, Jun 25, 2024 at 10:32 AM Pierre-Eric Pelloux-Prayer
wrote:
The call to radeon_vm_clear_freed might clear bo_va->bo, so
we have to check it before dereferencing it.
Signed-off-by: Pierre-Eric Pelloux-Prayer
Acked-by: Alex Deucher
Should I
Am 01.07.24 um 19:14 schrieb Lucas Stach:
This exposes a accumulated GPU active time per client via the
fdinfo infrastructure.
Signed-off-by: Lucas Stach
Acked-by: Christian König
Sorry that I couldn't find time to finalize and upstream that patch set
myself.
Regards,
Chri
Am 02.07.24 um 10:26 schrieb Sunil Khatri:
During ip dump in gfx10 the index variable is reused but is
not reinitialized to 0 and this causes the index calculation
to be wrong and access out of bound access.
Acked-by: Christian König for the series.
Regards,
Christian.
Signed-off-by
Am 02.07.24 um 10:42 schrieb Tvrtko Ursulin:
Hi,
I few questions below.
On 01/07/2024 18:14, Lucas Stach wrote:
From: Christian König
Multiple drivers came up with the requirement to measure how
much runtime each entity accumulated on the HW.
A previous attempt of accounting this had to
like /* Only allow a single BO list to avoid memory leak. */
With that fixed Reviewed-by: Christian König
Regards,
Christian.
+ if (p->bo_list)
+ return -EINVAL;
+
r = amdgpu_bo_create_list_entry_array(data, &info);
if (r)
return r;
Am 02.07.24 um 20:20 schrieb Alex Deucher:
+ dri-devel
On Tue, Jul 2, 2024 at 1:40 PM Rajneesh Bhardwaj
wrote:
Otherwise the nid is always passed as 0 during memory reclaim so
make TTM shrinkers NUMA aware.
Signed-off-by: Rajneesh Bhardwaj
---
drivers/gpu/drm/ttm/ttm_pool.c | 2 +-
1 f
Am 02.07.24 um 23:54 schrieb Bhardwaj, Rajneesh:
[AMD Official Use Only - AMD Internal Distribution Only]
-Original Message-
From: Koenig, Christian
Sent: Tuesday, July 2, 2024 2:25 PM
To: Alex Deucher ; Bhardwaj, Rajneesh
; Maling list - DRI developers
Cc: amd-gfx@lists.freedesktop.
Error handling is to resolve or at least mitigate problems caused by the
user or the hardware.
When the driver tries to allocate more than allowed from the ring then
that's a driver bug and not something for error handling.
So the error handling here doesn't seem to make sense in the first pl
Am 03.07.24 um 11:01 schrieb Bob Zhou:
To avoid null pointer dereference reported by Coverity,
so add null pointer check for the return of amdgpu_device_get_gang().
NAK, that's complete nonsense in the first place.
The pointer is guaranteed to be never NULL or otherwise the logic would
have c
Well Coverity probably can't know that and will keep complaining.
Maybe we can add some extra code to point out that old can never be NULL
here?
Regards,
Christian.
Am 03.07.24 um 13:06 schrieb Zhou, Bob:
[AMD Official Use Only - AMD Internal Distribution Only]
Hi Christian
Thanks for your
Am 03.07.24 um 14:48 schrieb Pierre-Eric Pelloux-Prayer:
Le 02/07/2024 à 15:35, Christian König a écrit :
Am 02.07.24 um 15:23 schrieb Pierre-Eric Pelloux-Prayer:
Before this commit, only submits with both a BO_HANDLES chunk and a
'bo_list_handle' would be rejected (by amdgpu_cs_
Am 03.07.24 um 21:36 schrieb Alex Deucher:
Switch to fist last for consistency.
Signed-off-by: Alex Deucher
Cc: Xinhui Pan
Ah, that's probably my fault. I copy&pasted that from a mail.
Reviewed-by: Christian König
---
MAINTAINERS | 2 +-
1 file changed, 1 insertion(+), 1
Am 03.07.24 um 21:36 schrieb Alex Deucher:
Evan is no longer maintaining powerplay and swsmu.
Add Kenneth Feng as his replacement.
Signed-off-by: Alex Deucher
Cc: Kenneth Feng
Acked-by: Christian König
---
MAINTAINERS | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a
)
- Set the Alignment to a default value if the callback doesn't exist.
- Add the callback to amdgpu_gmc_funcs.
Signed-off-by: Arunpravin Paneer Selvam
Acked-by: Christian König for the series.
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h | 3 ++
drivers/gpu/drm/amd/a
Am 03.07.24 um 19:49 schrieb Sunil Khatri:
We need IH ring1 for handling the pagefault
interrupts which over flow in default
ring for specific usecases.
Enable ring1 allows software to redirect
high interrupts to ring1 from default IH
ring.
Signed-off-by: Sunil Khatri
Reviewed-by: Christian
Am 04.07.24 um 06:58 schrieb Hoi Pok Wu:
Thanks a lot for your help Thomas.
On Wed, Jul 3, 2024 at 4:52 AM Thomas Zimmermann wrote:
Hi
Am 30.06.24 um 18:59 schrieb Wu Hoi Pok:
.load and drm_dev_alloc are deprecated. These patch series aims to
remove them.
v3: Both v1 and v2 sucks. v3 improv
mmon mistakes of all times :)
Reviewed-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index 9224fc6418e4..518b10f190ec 100644
-
Am 08.07.24 um 21:04 schrieb Alex Deucher:
Use the dev_info/err variants so we get per device logging
in multi-GPU cases.
Signed-off-by: Alex Deucher
Reviewed-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 21 +++--
1 file changed, 11 insertions
Am 10.07.24 um 02:31 schrieb jiadong@amd.com:
From: Jiadong Zhu
The job's embedded fence is dma_fence which shall not be conversed
to amdgpu_fence.
Good catch.
The start timestamp shall be saved on job for
hw_fence.
But NAK to that approach. Why do we need the start time here in the
Am 10.07.24 um 09:54 schrieb Zhu, Jiadong:
[AMD Official Use Only - AMD Internal Distribution Only]
-Original Message-
From: Christian König
Sent: Wednesday, July 10, 2024 3:17 PM
To: Zhu, Jiadong ; amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH v2] drm/amdgpu: set start timestamp
Am 10.07.24 um 12:15 schrieb Zhu, Jiadong:
[AMD Official Use Only - AMD Internal Distribution Only]
-Original Message-
From: Christian König
Sent: Wednesday, July 10, 2024 5:27 PM
To: Zhu, Jiadong ; amd-gfx@lists.freedesktop.org;
Deucher, Alexander
Subject: Re: [PATCH v2] drm/amdgpu
Am 11.07.24 um 03:31 schrieb Zhu, Jiadong:
[AMD Official Use Only - AMD Internal Distribution Only]
-Original Message-
From: Christian König
Sent: Wednesday, July 10, 2024 8:46 PM
To: Zhu, Jiadong ; amd-gfx@lists.freedesktop.org;
Deucher, Alexander
Subject: Re: [PATCH v2] drm/amdgpu
Am 11.07.24 um 07:37 schrieb jiadong@amd.com:
From: Jiadong Zhu
The job's embedded fence is dma_fence which shall not be conversed
to amdgpu_fence. The start timestamp shall be saved on job for
hw_fence.
Again big NAK to adding the start time to the job.
Regards,
Christian.
v2: optimi
Yeah, completely agree. This patch doesn't really make sense.
Please explain why you would want to do this?
Regards,
Christian.
Am 11.07.24 um 13:56 schrieb Huang, Trigger:
[AMD Official Use Only - AMD Internal Distribution Only]
This patch seems to be wrong.
Quite a lot of preparations have
Am 11.07.24 um 20:17 schrieb Tvrtko Ursulin:
From: Tvrtko Ursulin
From the department of questionable optimisations today we have a minor
improvement to how padding / filling the rings with nops is done.
Having noticed that typically 200+ nops per submission are filled into the
ring, using a
been synced to move notify, not
the move action.
Thanks
River
-Original Message-
From: Christian König
Sent: Thursday, July 11, 2024 8:39 PM
To: Huang, Trigger ; YuanShang Mao (River)
; amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH] drm/amdgpu: Mark amdgpu_bo as invalid after moved
because compute shader is still accessing the
"invalidated" BO.
I am not familiar with amdgpu_vm_bo state machine, so I don’t know if it is an
code error or an design error.
Thanks
River
-Original Message-
From: YuanShang Mao (River)
Sent: Friday, July 12, 2024 10:55 AM
To:
Am 12.07.24 um 11:14 schrieb Tvrtko Ursulin:
On 12/07/2024 08:33, Christian König wrote:
Am 11.07.24 um 20:17 schrieb Tvrtko Ursulin:
From: Tvrtko Ursulin
From the department of questionable optimisations today we have a
minor
improvement to how padding / filling the rings with nops is
compaction. (/proc/sys/vm/compact_memory)
Note: On certain distros such as RHEL, the proactive compaction is
disabled. (https://tinyurl.com/4f32f7rs)
Cc: Dave Airlie
Cc: Vlastimil Babka
Cc: Daniel Vetter
Reviewed-by: Christian König
Signed-off-by: Rajneesh Bhardwaj
---
drivers/gpu/drm/ttm
en compute and graphics. So
graphics and compute submissions at the same time are possible. @Christian,
this is a concequence of using libdrm and insisting that each process uses only
a single VM per GPU.
Regards,
Felix
On 2024-07-12 3:39, Christian König wrote:
Hi River,
well that isn
Am 15.07.24 um 16:47 schrieb Jane Jian:
From: Lijo Lazar
JPEG v4.0.3 doesn't support HDP flush when RRMT is enabled. Instead,
mmsch fw will do the flush.
Signed-off-by: Lijo Lazar
Signed-off-by: Jane Jian
---
drivers/gpu/drm/amd/amdgpu/jpeg_v4_0_3.c | 9 +
1 file changed, 9 ins
Am 15.07.24 um 17:08 schrieb Lazar, Lijo:
On 7/15/2024 8:28 PM, Christian König wrote:
Am 15.07.24 um 16:47 schrieb Jane Jian:
From: Lijo Lazar
JPEG v4.0.3 doesn't support HDP flush when RRMT is enabled. Instead,
mmsch fw will do the flush.
Signed-off-by: Lijo Lazar
Signed-off-by:
- Set the Alignment to a default value if the callback doesn't
exist.
- Add the callback to amdgpu_gmc_funcs.
v6:
- Fix checkpatch error reported by Intel CI.
Signed-off-by: Arunpravin Paneer Selvam
Acked-by: Alex Deucher
Acked-by: Christian König
G_CONTIGUOUS) &&
(amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(12, 0, 0) ||
amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(12, 0, 1))) {
}
Regards,
Arun.
On 7/17/2024 2:38 PM, Christian König wrote:
Well that approach was discussed before and seemed to be to complicated.
But I to
ic hw generation.
Signed-off-by: Arunpravin Paneer Selvam
Acked-by: Alex Deucher
Acked-by: Christian König
Reviewed-by: Frank Min
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h | 6 +++
drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 39 +++-
drivers/gpu/drm/amd/amdgpu/gmc_
The series is
Reviewed-by: Felix Kuehling
I only skimmed over it and will probably find something to complain on
later.
But we need to get this out of the door, so feel free to add Acked-by:
Christian König to the series for now.
Thanks,
Christian.
Philip Yang (9):
drm/amdkfd
Am 19.07.24 um 11:19 schrieb ZhenGuo Yin:
[Why]
Page table of compute VM in the VRAM will lost after gpu reset.
VRAM won't be restored since compute VM has no shadows.
[How]
Use higher 32-bit of vm->generation to record a vram_lost_counter.
Reset the VM state machine when the counter is not equa
Am 19.07.24 um 11:36 schrieb Yin, ZhenGuo (Chris):
[AMD Official Use Only - AMD Internal Distribution Only]
Hi, Christian
Why loosing VRAM would result in the vm entity to become invalid?
I think only if there has a fence error appeared(like a pending SDMA job got
timedout or cancelled), then
: ac4eb83ab255 ("drm/sched: select new rq even if there is only one
v3")
References: 981b04d96856 ("drm/sched: improve docs around drm_sched_entity")
Cc: Christian König
Cc: Alex Deucher
Cc: Luben Tuikov
Cc: Matthew Brost
Cc: Daniel Vetter
Cc: amd-gfx@lists.freedesktop.or
Am 19.07.24 um 15:02 schrieb Christian König:
Am 19.07.24 um 11:47 schrieb Tvrtko Ursulin:
From: Tvrtko Ursulin
Long time ago in commit b3ac17667f11 ("drm/scheduler: rework entity
creation") a change was made which prevented priority changes for
entities
with only one assigned
Am 19.07.24 um 11:16 schrieb Jack Xiao:
wait memory room until enough before writing mes packets
to avoid ring buffer overflow.
Signed-off-by: Jack Xiao
---
drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 18 ++
drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 18 ++
2 file
x27;t
overflow the logs with them.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 30 +++---
1 file changed, 9 insertions(+), 21 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
k we rather need to increase the MES ring size instead./
Unfortunately, it doesn't work. I guess mes firmware has limitation.
Regards,
Jack
----
*From:* Christian König
*Sent:* Friday, 19 July 2024 23:44
*To:* Xiao
a bit.
Signed-off-by: Christian König
---
.../drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c| 4 +--
drivers/gpu/drm/etnaviv/etnaviv_sched.c | 4 +--
drivers/gpu/drm/imagination/pvr_queue.c | 4 +--
drivers/gpu/drm/lima
11 command submission
git bisect bad e356d321d0240663a09b139fa3658ddbca163e27
# first bad commit: [e356d321d0240663a09b139fa3658ddbca163e27]
drm/amdgpu: cleanup MES11 command submission
Author: Christian König
Date: Fri May 31 10:56:00 2024 +0200
drm/amdgpu: cleanup MES11 command s
d more waiting time.
Regards,
Jack
----
*From:* Christian König
*Sent:* Monday, 22 July 2024 16:20
*To:* Xiao, Jack ; amd-gfx@lists.freedesktop.org
; Deucher, Alexander
*Subject:* Re: [PATCH] drm/amdgpu/mes: fix mes ring buffer o
Am 17.07.24 um 22:37 schrieb Alex Deucher:
It should work the same for compute as well as gfx.
Signed-off-by: Alex Deucher
Reviewed-by: Christian König for the whole
series.
---
drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/gpu/drm
Am 17.07.24 um 22:38 schrieb Alex Deucher:
Need to handle the interrupt enables for all pipes.
v2: fix indexing (Jessie)
Signed-off-by: Alex Deucher
Acked-by: Christian König for the whole series.
---
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 130 +
1 file
Am 17.07.24 um 22:40 schrieb Alex Deucher:
From: Jesse Zhang
For the bad opcode case, it will cause CP/ME hang.
The firmware will prevent the ME side from hanging by raising a bad opcode
interrupt.
And the driver needs to perform a vmid reset when receiving the interrupt.
v2: update irq namin
Am 22.07.24 um 15:52 schrieb Tvrtko Ursulin:
On 19/07/2024 16:18, Christian König wrote:
Am 19.07.24 um 15:02 schrieb Christian König:
Am 19.07.24 um 11:47 schrieb Tvrtko Ursulin:
From: Tvrtko Ursulin
Long time ago in commit b3ac17667f11 ("drm/scheduler: rework entity
creation"
Am 22.07.24 um 16:43 schrieb Tvrtko Ursulin:
On 22/07/2024 15:06, Christian König wrote:
Am 22.07.24 um 15:52 schrieb Tvrtko Ursulin:
On 19/07/2024 16:18, Christian König wrote:
Am 19.07.24 um 15:02 schrieb Christian König:
Am 19.07.24 um 11:47 schrieb Tvrtko Ursulin:
From: Tvrtko Ursulin
Am 23.07.24 um 05:05 schrieb ZhenGuo Yin:
[Why]
Page table of compute VM in the VRAM will lost after gpu reset.
VRAM won't be restored since compute VM has no shadows.
[How]
Use higher 32-bit of vm->generation to record a vram_lost_counter.
Reset the VM state machine when vm->genertaion is not e
Am 23.07.24 um 10:27 schrieb Jack Xiao:
Only allow API_NUMBER_OF_COMMAND_MAX packet in mes ring buffer,
refine the code for maximum packet execution.
Signed-off-by: Jack Xiao
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 2 ++
drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 2 +-
drivers/gpu/dr
valid, but let me comment on the patch
itself.
Regards,
Christian.
Best,
Zhenguo
Cloud-GPU Core team, SRDC
-Original Message-
From: Christian König
Sent: Tuesday, July 23, 2024 3:13 PM
To: Yin, ZhenGuo (Chris) ; amd-gfx@lists.freedesktop.org
Cc: Koenig, Christian
Subject: Re: [PATCH v
Am 23.07.24 um 05:05 schrieb ZhenGuo Yin:
[Why]
Page table of compute VM in the VRAM will lost after gpu reset.
VRAM won't be restored since compute VM has no shadows.
[How]
Use higher 32-bit of vm->generation to record a vram_lost_counter.
Reset the VM state machine when vm->genertaion is not e
,7 @@ uint64_t amdgpu_vm_generation(struct amdgpu_device *adev,
struct amdgpu_vm *vm)
if (!vm)
return result;
- result += vm->generation;
+ result += (vm->generation & 0xULL);
Please use the lower_32_bits() macro here.
With that fixed the patch is Revi
Am 24.07.24 um 09:02 schrieb Dave Airlie:
From: Dave Airlie
This is a partial revert of drm/amdgpu: Modify the contiguous flags behaviour.
This broke VCN AV1 decoding on radv video on GFX11.
On VCN4 only the first VCN block has AV1 decode support, so the kernel has
a hacky heurisitic to work
Am 24.07.24 um 09:06 schrieb Tim Huang:
This is to address the Coverity explicit null dereferenced warnings
by NULL returns from amdgpu_mes_ctx_get_offs* but without follow-up
Checks. Meanwhile refactor the code to keep only one *_get_gpu/cpu_addr.
Well nice that you are looking into that, but
Otherwise we won't get correct access to the IB.
Signed-off-by: Christian König
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 15 +++
1 file changed, 7 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
Am 24.07.24 um 10:16 schrieb Tvrtko Ursulin:
[SNIP]
Absolutely.
Absolutely good and absolutely me, or absolutely you? :)
You, I don't even have time to finish all the stuff I already started :/
These are the TODO points and their opens:
- Adjust amdgpu_ctx_set_entity_
_start alignment with min chunk_size
>> - use range_overflows()
>>
>> Signed-off-by: Arunpravin Paneer Selvam
mailto:arunpravin.paneersel...@amd.com>>
>> Acked-by: Alex Deucher mailto:alexander.deuc...@amd.com>>
>> Acked-by: Chr
Am 24.07.24 um 09:37 schrieb Jiapeng Chong:
Use existing swap() function rather than duplicating its implementation.
./drivers/gpu/drm/amd/display/dc/dml2/dml21/src/dml2_pmo/dml2_pmo_dcn3.c:17:29-30:
WARNING opportunity for swap().
Reported-by: Abaci Robot
Closes: https://bugzilla.openanolis.
Am 24.07.24 um 20:43 schrieb vitaly.pros...@amd.com:
From: Vitaly Prosyak
The current implementation of drm_sched_start uses a hardcoded -ECANCELED to
dispose of a job when
the parent/hw fence is NULL. This results in drm_sched_job_done being called
with -ECANCELED for
each job with a NULL pa
Otherwise we won't get correct access to the IB.
v2: keep setting AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS to avoid problems in
the VRAM backend.
Signed-off-by: Christian König
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3501
Fixes: e362b7c8f8c7 ("drm/amdgpu: Modify the
Am 25.07.24 um 20:09 schrieb Nikita Zhandarovich:
Several cs track offsets (such as 'track->db_s_read_offset')
either are initialized with or plainly take big enough values that,
once shifted 8 bits left, may be hit with integer overflow if the
resulting values end up going over u32 limit.
Some
I strongly suggest to revert that again. See my other mail.
Christian.
Am 25.07.24 um 22:59 schrieb Alex Deucher:
Applied. Thanks!
Alex
On Thu, Jul 25, 2024 at 2:20 PM Nikita Zhandarovich
wrote:
Several cs track offsets (such as 'track->db_s_read_offset')
either are initialized with or pla
ssage,
drop the new function again and update all callers,
apply the errno also to scheduler fences with hw fences
Signed-off-by: Jesse Zhang
Signed-off-by: Vitaly Prosyak
Signed-off-by: Christian König
Cc: Alex Deucher
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.
Am 26.07.24 um 14:52 schrieb Alex Deucher:
On Fri, Jul 26, 2024 at 3:05 AM Christian König
wrote:
Am 25.07.24 um 20:09 schrieb Nikita Zhandarovich:
Several cs track offsets (such as 'track->db_s_read_offset')
either are initialized with or plainly take big enough values that,
o
Am 29.07.24 um 12:42 schrieb Michel Dänzer:
On 2024-07-26 17:25, Alex Deucher wrote:
On Fri, Jul 26, 2024 at 9:50 AM Alex Deucher wrote:
This adds allocation latency, but aligns better with user
expectations. The latency should improve with the drm buddy
clearing patches that Arun has been wo
Am 29.07.24 um 19:26 schrieb Nikita Zhandarovich:
Hi,
On 7/29/24 02:23, Christian König wrote:
Am 26.07.24 um 14:52 schrieb Alex Deucher:
On Fri, Jul 26, 2024 at 3:05 AM Christian König
wrote:
Am 25.07.24 um 20:09 schrieb Nikita Zhandarovich:
Several cs track offsets (such as '
Am 29.07.24 um 20:04 schrieb Christian König:
Am 29.07.24 um 19:26 schrieb Nikita Zhandarovich:
Hi,
On 7/29/24 02:23, Christian König wrote:
Am 26.07.24 um 14:52 schrieb Alex Deucher:
On Fri, Jul 26, 2024 at 3:05 AM Christian König
wrote:
Am 25.07.24 um 20:09 schrieb Nikita Zhandarovich
Am 25.07.24 um 20:19 schrieb Xiaogang.Chen:
From: Xiaogang Chen
When app unmap vm ranges(munmap) kfd/svm starts drain pending page fault and
not handle any incoming pages fault of this process until a deferred work item
got executed by default system wq. The time period of "not handle page faul
Am 26.07.24 um 14:30 schrieb Matthew Brost:
On Fri, Jul 26, 2024 at 09:55:50AM +0200, Christian König wrote:
The current implementation of drm_sched_start uses a hardcoded
-ECANCELED to dispose of a job when the parent/hw fence is NULL.
This results in drm_sched_job_done being called with
Am 26.07.24 um 16:21 schrieb Daniel Vetter:
On Fri, Jul 26, 2024 at 09:55:50AM +0200, Christian König wrote:
The current implementation of drm_sched_start uses a hardcoded
-ECANCELED to dispose of a job when the parent/hw fence is NULL.
This results in drm_sched_job_done being called with
301 - 400 of 10574 matches
Mail list logo