On 01/02/2023 13:21, Luben Tuikov wrote:
> Hi Guilherme,
>
> Since setting sched->ready to false, seems to be taking place in, directly
> amdgpu_ring_fini()
> and in amdgpu_fence_driver_sw_fini() indirectly as that function calls
> drm_sched_fini()
> which sets it to false, we seem to have two
;>>>> Hi Piccoli,
>>>>>
>>>>> Please ignore my request of full dmesg log. I can reproduce the issue and
>>>>> get the same failure callstack by returning early with an error code
>>>>> prior to amdgpu_device_init_schedulers.
c: amd-...@lists.freedesktop.org; ker...@gpiccoli.net; Pan, Xinhui ;
dri-devel@lists.freedesktop.org; Tuikov, Luben ; Limonciello, Mario
; kernel-...@igalia.com; Deucher, Alexander
Subject: Re: [PATCH] drm/amdgpu/fence: Fix oops due to non-matching drm_sched
init/fini
Am 31.01.23 um 10:17 schrieb Chen, Guchu
tian
> > Sent: Tuesday, January 31, 2023 6:59 PM
> > To: Chen, Guchun ; Alex Deucher
> > ; Guilherme G. Piccoli
> > Cc: amd-...@lists.freedesktop.org; ker...@gpiccoli.net; Pan, Xinhui
> > ; dri-devel@lists.freedesktop.org; Tuikov, Luben
> > ; Limonciello, M
amd-...@lists.freedesktop.org; ker...@gpiccoli.net; Pan, Xinhui ;
dri-devel@lists.freedesktop.org; Tuikov, Luben ; Limonciello, Mario
; kernel-...@igalia.com; Deucher, Alexander
Subject: Re: [PATCH] drm/amdgpu/fence: Fix oops due to non-matching drm_sched
init/fini
Am 31.01.23 um 10:17 schrieb C
On Tue, Jan 31, 2023 at 1:23 PM Guilherme G. Piccoli
wrote:
>
> On 31/01/2023 14:52, Alex Deucher wrote:
> > [...]
> >> (b) We can't use sched.ready, which would make sense...but amdgpu
> >> overrides its meaning, the driver manipulates this value for its own
> >> purposes of tracking ring init,
On 31/01/2023 14:52, Alex Deucher wrote:
> [...]
>> (b) We can't use sched.ready, which would make sense...but amdgpu
>> overrides its meaning, the driver manipulates this value for its own
>> purposes of tracking ring init, or something like that.
>>
>> This is the tangential topic: what should
On Tue, Jan 31, 2023 at 9:32 AM Guilherme G. Piccoli
wrote:
>
> On 31/01/2023 10:58, Chen, Guchun wrote:
> > Hi Christian,
> >
> > Do you think if it makes sense that we can set 'ring->sched.ready' to be
> > true in each ring init, even if before executing/setting up drm_sched_init
> > in
> > Cc: amd-...@lists.freedesktop.org; ker...@gpiccoli.net; Pan, Xinhui
> > ; dri-devel@lists.freedesktop.org; Tuikov, Luben
> > ; Limonciello, Mario ;
> > kernel-...@igalia.com; Deucher, Alexander ;
> > Koenig, Christian
> > Subject: RE: [PATCH] drm/amdgpu/
On 31/01/2023 10:58, Chen, Guchun wrote:
> Hi Christian,
>
> Do you think if it makes sense that we can set 'ring->sched.ready' to be true
> in each ring init, even if before executing/setting up drm_sched_init in
> amdgpu_device_init_schedulers? As 'ready' is a member of gpu scheduler
>
her, Alexander
Subject: Re: [PATCH] drm/amdgpu/fence: Fix oops due to non-matching drm_sched
init/fini
Am 31.01.23 um 10:17 schrieb Chen, Guchun:
> Hi Piccoli,
>
> Please ignore my request of full dmesg log. I can reproduce the issue and get
> the same failure callstack by returning earl
Subject: RE: [PATCH] drm/amdgpu/fence: Fix oops due to non-matching drm_sched
init/fini
Hi Piccoli,
I agree with Alex's point, using ring->sched.name for such check is not a good
way. BTW, can you please attach a full dmesg long in bad case to help me
understand more?
Regards,
Guc
Re: [PATCH] drm/amdgpu/fence: Fix oops due to non-matching drm_sched
init/fini
On Mon, Jan 30, 2023 at 4:51 PM Guilherme G. Piccoli
wrote:
>
> + Luben
>
> (sorry, missed that in the first submission).
>
> On 30/01/2023 18:45, Guilherme G. Piccoli wrote:
> > Currently amdgpu
AM
To: Guilherme G. Piccoli
Cc: amd-...@lists.freedesktop.org; ker...@gpiccoli.net; Chen, Guchun
; Pan, Xinhui ;
dri-devel@lists.freedesktop.org; Tuikov, Luben ;
Limonciello, Mario ; kernel-...@igalia.com; Deucher,
Alexander ; Koenig, Christian
Subject: Re: [PATCH] drm/amdgpu/fence: Fix oops due to
On Mon, Jan 30, 2023 at 4:51 PM Guilherme G. Piccoli
wrote:
>
> + Luben
>
> (sorry, missed that in the first submission).
>
> On 30/01/2023 18:45, Guilherme G. Piccoli wrote:
> > Currently amdgpu calls drm_sched_fini() from the fence driver sw fini
> > routine - such function is expected to be
+ Luben
(sorry, missed that in the first submission).
On 30/01/2023 18:45, Guilherme G. Piccoli wrote:
> Currently amdgpu calls drm_sched_fini() from the fence driver sw fini
> routine - such function is expected to be called only after the
> respective init function - drm_sched_init() - was
Currently amdgpu calls drm_sched_fini() from the fence driver sw fini
routine - such function is expected to be called only after the
respective init function - drm_sched_init() - was executed successfully.
Happens that we faced a driver probe failure in the Steam Deck
recently, and the function
17 matches
Mail list logo