[PATCH] drm/amdgpu: clear set_q_mode_offs when VM changed

[Christian König]

If this solution here works then stick with that for now.

At least I still hope to switch to user queues sometimes soon and that 
would completely avoid this problem here in the first place.

Regards,
Christian.


Am 02.04.24 um 10:35 schrieb Yin, ZhenGuo (Chris):
> [AMD Official Use Only - General]
>
> Yeah, I agree.  But I have no idea which place is better.
>
> Best,
> Zhenguo
> Cloud-GPU Core team, SRDC
>
> -----Original Message-----
> From: Koenig, Christian <Christian.Koenig at amd.com>
> Sent: Tuesday, April 2, 2024 4:14 PM
> To: Yin, ZhenGuo (Chris) <ZhenGuo.Yin at amd.com>; amd-gfx at lists.freedesktop.org
> Cc: Deucher, Alexander <Alexander.Deucher at amd.com>
> Subject: Re: [PATCH] drm/amdgpu: clear set_q_mode_offs when VM changed
>
> Am 02.04.24 um 08:37 schrieb ZhenGuo Yin:
>> [Why]
>> set_q_mode_offs don't get cleared after GPU reset, nexting SET_Q_MODE
>> packet to init shadow memory will be skiped, hence there has a page fault.
>>
>> [How]
>> VM flush is needed after GPU reset, clear set_q_mode_offs when
>> emitting VM flush.
>>
>> Fixes: 8dad9c062355 ("drm/amdgpu: workaround to avoid SET_Q_MODE
>> packets v2")
>> Signed-off-by: ZhenGuo Yin <zhenguo.yin at amd.com>
> Good catch, yeah it probably doesn't make much sense to execute this after a VM flush.
>
> Reviewed-by: Christian König <christian.koenig at amd.com>
>
>> ---
>>    drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 1 +
>>    1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
>> b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
>> index 7a906318e451..c11c6299711e 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
>> @@ -5465,6 +5465,7 @@ static void gfx_v11_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
>>        /* Make sure that we can't skip the SET_Q_MODE packets when the VM
>>         * changed in any way.
>>         */
>> +     ring->set_q_mode_offs = 0;
>>        ring->set_q_mode_ptr = NULL;
>>    }
>>