Re: [OE-core] [PATCH 01/31] core-image-weston: bump QB_MEM to 1G in qemuarm64

["Alexander Kanavin" <alex.kanavin@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2042 bytes --]

Sadly, I can confirm: weston swelling to double size on qemuarm64 is caused
by this recent change in glibc:

+EXTRA_OECONF_append_aarch64 = " --enable-memory-tagging"

You can verify with this innocent looking program. Run 'free' while it is
running and compare with and without tagging:

main()
{
   for (int i=0; i<768; i++)
       calloc(1,262368);
   sleep(200);
}

I'll queue a revert.

Alex

On Wed, 16 Jun 2021 at 16:08, Alexander Kanavin via lists.openembedded.org
<alex.kanavin=gmail.com@lists.openembedded.org> wrote:

> It seems that mesa, for some reason, allocates a lot more RAM in qemuarm64
> (150 Mb more or so), than it does in qemux86_64. I'll investigate.
>
> Alex
>
> On Wed, 16 Jun 2021 at 12:06, Alexander Kanavin via lists.openembedded.org
> <alex.kanavin=gmail.com@lists.openembedded.org> wrote:
>
>> I'd like to take a deeper look at this, please hold.
>>
>> Alex
>>
>> On Tue, 15 Jun 2021 at 10:12, Alexander Kanavin via
>> lists.openembedded.org <alex.kanavin=gmail.com@lists.openembedded.org>
>> wrote:
>>
>>> weston tests start additional compositors, other tests are also
>>> heavy on RAM, weston itself takes 256M with opengl enabled,
>>> so 512M causes OOM errors on arm64.
>>>
>>> Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com>
>>> ---
>>>  meta/recipes-graphics/images/core-image-weston.bb | 1 +
>>>  1 file changed, 1 insertion(+)
>>>
>>> diff --git a/meta/recipes-graphics/images/core-image-weston.bb
>>> b/meta/recipes-graphics/images/core-image-weston.bb
>>> index 62305cc1ce..e332e19368 100644
>>> --- a/meta/recipes-graphics/images/core-image-weston.bb
>>> +++ b/meta/recipes-graphics/images/core-image-weston.bb
>>> @@ -10,3 +10,4 @@ CORE_IMAGE_BASE_INSTALL += "gtk+3-demo"
>>>  CORE_IMAGE_BASE_INSTALL += "${@bb.utils.contains('DISTRO_FEATURES',
>>> 'x11', 'weston-xwayland matchbox-terminal', '', d)}"
>>>
>>>  QB_MEM = "-m 512"
>>> +QB_MEM_qemuarm64 = "-m 1024"
>>> --
>>> 2.31.1
>>>
>>>
>>>
>>>
>>>
>>
>>
>>
> 
>
>

[-- Attachment #2: Type: text/html, Size: 3871 bytes --]