cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16

cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16

[Michael Walle]

Hi,

Since the next-20220526 I'm getting the following error during boot:
cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16

E.g. on the following arm64 board:
https://lavalab.kontron.com/scheduler/job/121315

But initially I've noticed it on an arm board, which isn't
in kernelci yet:
https://pastebin.com/raw/JY2nFJjT
(Ignore the subsequent i2c kernel oops, that seems to be
a bug in the xdma driver).

next-20220524 was still alright (not sure what has happened
to next-20220525):
https://lavalab.kontron.com/scheduler/job/120496

Bisect of mm-everything..v5.18-rc4 is currently running but
takes a while. Maybe anyone has an idea what goes wrong here?

-michael

Re: cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16

[Andrew Morton]

On Thu, 26 May 2022 17:06:08 +0200 Michael Walle <michael@walle.cc> wrote:

> Hi,
> 
> Since the next-20220526 I'm getting the following error during boot:
> cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16
> 
> E.g. on the following arm64 board:
> https://lavalab.kontron.com/scheduler/job/121315
> 
> But initially I've noticed it on an arm board, which isn't
> in kernelci yet:
> https://pastebin.com/raw/JY2nFJjT
> (Ignore the subsequent i2c kernel oops, that seems to be
> a bug in the xdma driver).
> 
> next-20220524 was still alright (not sure what has happened
> to next-20220525):
> https://lavalab.kontron.com/scheduler/job/120496
> 
> Bisect of mm-everything..v5.18-rc4 is currently running but
> takes a while. Maybe anyone has an idea what goes wrong here?

Please try reverting 88ee134320b8311 ("mm: fix a potential infinite
loop in start_isolate_page_range()").

Is a real bummer.  That was v11 of the patchset, it was in -next for a
month.  The day before the merge window opens, bug turns up and it
seems that the attempt to fix the bug added a new one.  Hopefully Zi Yan
will be able to plug this one soon.

Re: cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16

[Zi Yan]

[-- Attachment #1: Type: text/plain, Size: 1507 bytes --]

On 26 May 2022, at 13:15, Andrew Morton wrote:

> On Thu, 26 May 2022 17:06:08 +0200 Michael Walle <michael@walle.cc> wrote:
>
>> Hi,
>>
>> Since the next-20220526 I'm getting the following error during boot:
>> cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16
>>
>> E.g. on the following arm64 board:
>> https://lavalab.kontron.com/scheduler/job/121315
>>
>> But initially I've noticed it on an arm board, which isn't
>> in kernelci yet:
>> https://pastebin.com/raw/JY2nFJjT
>> (Ignore the subsequent i2c kernel oops, that seems to be
>> a bug in the xdma driver).
>>
>> next-20220524 was still alright (not sure what has happened
>> to next-20220525):
>> https://lavalab.kontron.com/scheduler/job/120496
>>
>> Bisect of mm-everything..v5.18-rc4 is currently running but
>> takes a while. Maybe anyone has an idea what goes wrong here?
>
> Please try reverting 88ee134320b8311 ("mm: fix a potential infinite
> loop in start_isolate_page_range()").
>
> Is a real bummer.  That was v11 of the patchset, it was in -next for a
> month.  The day before the merge window opens, bug turns up and it
> seems that the attempt to fix the bug added a new one.  Hopefully Zi Yan
> will be able to plug this one soon.

Hi Michael,

The patch from this link should fix it: https://lore.kernel.org/linux-mm/8C74B0BA-3C69-46EE-8C14-0CE6787D0B65@nvidia.com/. Please give it a try.


Hi Andrew,

I will send a formal patchset soon.

--
Best Regards,
Yan, Zi

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 854 bytes --]

Re: cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16

[Michael Walle]

Hi,

Am 2022-05-26 22:15, schrieb Zi Yan:
> On 26 May 2022, at 13:15, Andrew Morton wrote:
> 
>> On Thu, 26 May 2022 17:06:08 +0200 Michael Walle <michael@walle.cc> 
>> wrote:
>> 
>>> Hi,
>>> 
>>> Since the next-20220526 I'm getting the following error during boot:
>>> cma: cma_alloc: reserved: alloc failed, req-size: 128 pages, ret: -16
>>> 
>>> E.g. on the following arm64 board:
>>> https://lavalab.kontron.com/scheduler/job/121315
>>> 
>>> But initially I've noticed it on an arm board, which isn't
>>> in kernelci yet:
>>> https://pastebin.com/raw/JY2nFJjT
>>> (Ignore the subsequent i2c kernel oops, that seems to be
>>> a bug in the xdma driver).
>>> 
>>> next-20220524 was still alright (not sure what has happened
>>> to next-20220525):
>>> https://lavalab.kontron.com/scheduler/job/120496
>>> 
>>> Bisect of mm-everything..v5.18-rc4 is currently running but
>>> takes a while. Maybe anyone has an idea what goes wrong here?
>> 
>> Please try reverting 88ee134320b8311 ("mm: fix a potential infinite
>> loop in start_isolate_page_range()").

Thanks for the quick reply. I can confirm this is the culprit.

>> Is a real bummer.  That was v11 of the patchset, it was in -next for a
>> month.  The day before the merge window opens, bug turns up and it
>> seems that the attempt to fix the bug added a new one.  Hopefully Zi 
>> Yan
>> will be able to plug this one soon.
> 
> Hi Michael,
> 
> The patch from this link should fix it:
> https://lore.kernel.org/linux-mm/8C74B0BA-3C69-46EE-8C14-0CE6787D0B65@nvidia.com/.
> Please give it a try.

I can also confirm this is fixing it. You can add my
Tested-by: Michael Walle <michael@walle.cc>

-michael