[5.14-rc5] etnaviv-gpu 130000.gpu: recover hung GPU!

[Wadim Egorov]

On 8/19/21 3:37 PM, Lucas Stach wrote:
> Am Donnerstag, dem 19.08.2021 um 15:07 +0200 schrieb Wadim Egorov:
>> Hello Lucas,
>>
>> On 8/19/21 1:14 PM, Lucas Stach wrote:
>>> Hi Wadim,
>>>
>>> Am Donnerstag, dem 19.08.2021 um 12:09 +0200 schrieb Wadim Egorov:
>>>> Hi,
>>>>
>>>> we have noticed "recover hung GPU" messages with the latest Linux on a i.MX6
>>>> Solo. The GPU recover message appears after some time running a Qt application
>>>> with QT_QPA_PLATFORM=eglfs and Mesa 21.0.3.
>>>> So far there are no problems with the GUI. Nevertheless, we want to understand
>>>> the cause of the problem.
>>>>
>>>> Below is the output from the viv-unpack tool of a coredump we've saved.
>>>> Unfortunately, I am not familiar with the whole DRM/GPU framework and don't
>>>> really know how to proceed and provide more relevant debug information.
>>>>
>>>> Is this a known problem?
>>>>
>>> Is this something that started showing up after the kernel update, or
>>> are there other system component changes involved, like a Mesa update?
>>>   From the hang location (inside a user cmdstream) a Mesa issue sounds
>>> more likely than a kernel issue.
>> We have already encountered this Issue with older components, Mesa 19.0.8 and
>> Linux v4.19. So I can't really say that this issue started showing up after a
>> specific component change.
>>
>> With the older components the system broke the complete Qt GUI. After the
>> recover message triggered, the GUI was totally visually distorted. This seems to
>> be fixed with newer versions.
>>
>> Since you think it is more likely related to mesa, do you think it makes sense
>> to report this problem in the mesa bug tracker?
>>
> Yes, I think that makes sense. Please open an issue in the freedesktop
> gitlab and attach the complete devcoredump. Since you might not be able
> to add a label in gitlab, please indicate when you opened the issue, so
> I can add the label to avoid the issue getting lost.

I've just opened the Issue and attached the devcoredump file.

   https://gitlab.freedesktop.org/mesa/mesa/-/issues/5275

Regards,
Wadim


>
> Regards,
> Lucas
>
>> Regards,
>> Wadim
>>
>>> Regards,
>>> Lucas
>>>
>>>> Regards,
>>>> Wadim
>>>>
>>>>
>>>> === Register dump
>>>> 0000000c = 00000051
>>>> 00000000 = 00040100
>>>> 00000004 = 7ffffff6 Idle: FE- DE+ PE+ SH- PA+ SE+ RA+ TX+ VG+ IM+ FP+ TS+
>>>> 00000008 = 00002200
>>>> 00000014 = ffffffff
>>>> 00000018 = 0c010000
>>>> 0000001c = e02864ad
>>>> 00000020 = 00000880
>>>> 00000024 = 00005106
>>>> 00000028 = 20110511
>>>> 0000002c = 01012500
>>>> 00000034 = c9f99eff
>>>> 00000038 = c9f99eff
>>>> 00000070 = 00000000
>>>> 00000100 = 00140021
>>>> 00000104 = 00000010
>>>> 00000108 = 000000e4
>>>> 0000010c = 00000000
>>>> 00000400 = 2a200000
>>>> 00000404 = 2a200000
>>>> 00000408 = 2a200000
>>>> 0000040c = 2a200000
>>>> 00000410 = 2a200000
>>>> 00000414 = 3e000000
>>>> 00000418 = 28000000
>>>> 0000041c = 28000000
>>>> 00000420 = 28000000
>>>> 00000424 = 28000000
>>>> 00000428 = 28000000
>>>> 0000042c = 2a230000
>>>> 00000480 = 00000088
>>>> 0000065c = 00000001
>>>> 00000660 = 00000804 Cmd: [adr1 DMA: idle Fetch: valid] Req idle Cal idle
>>>> 00000664 = 02082650 Command DMA address
>>>> 00000668 = 08010380 FE fetched word 0
>>>> 0000066c = 00000003 FE fetched word 1
>>>> 00000670 = 00000000
>>>> === Buffers
>>>>     Num Name  IOVA     Size
>>>>       0 reg   00000000 00000128      296
>>>>       1 mmu   00000000 00200000  2097152
>>>>       2 ring  02080000 00001000     4096
>>>> *  3 cmd   02082000 00000a08     2568
>>>>       4 bomap 00000000 00006300    25344
>>>>       5 bo    80a07000 00003000    12288
>>>>       6 bo    80003000 00003000    12288
>>>>       7 bo    80006000 00100000  1048576
>>>>       8 bo    80a0a000 00001000     4096
>>>>       9 bo    80107000 00280000  2621440
>>>>      10 bo    80a0b000 00280000  2621440
>>>>      11 bo    80106000 00001000     4096
>>>>      12 bo    80607000 00400000  4194304
>>>>      13 bo    02700000 00258000  2457600
>>>>
>>>> Checking MMU entries... ok
>>>>
>