i915 ttm_tt shmem backend

[Christian König]

Am 09.09.21 um 18:56 schrieb Matthew Auld:
> On Thu, 9 Sept 2021 at 17:43, Koenig, Christian
> <Christian.Koenig at amd.com> wrote:
>> Hi Matthew,
>>
>> this doesn't work, I've already tried something similar.
>>
>> TTM uses the reverse lookup functionality when migrating BOs between system and device memory. And that doesn't seem to work with pages from a shmem file.
> Hmm, what do you mean by reverse lookup functionality? Could you
> please point out where that is in the TTM code?

When TTM moves a buffer it must make sure that the buffer is not 
accessed by the CPU while moving it.

For this the standard reverse lockup functionality of the Linux kernel 
is used to figure out in which page tables a page is mapped and mark 
those as invalid. Accessing the buffer object will then cause a page 
fault which in turn waits for the buffer move to finish.

But when you back the pages in a TT object with pages from a shmemfile 
this reverse lockup functionality doesn't work for some reason. I 
couldn't figure out what exactly was going wrong here and didn't looked 
deeper, I assumed it's because of not setting up page->mapping and 
page->index correctly. Thomas or Daniel might know more.

Apart from that your approach sounds like pretty much what I tried as well.

Regards,
Christian.

>
>> Regards,
>> Christian.
>>
>> ________________________________
>> Von: Matthew Auld <matthew.william.auld at gmail.com>
>> Gesendet: Donnerstag, 9. September 2021 16:56
>> An: Christian König <ckoenig.leichtzumerken at gmail.com>; Koenig, Christian <Christian.Koenig at amd.com>
>> Cc: Thomas Hellström <thomas.hellstrom at linux.intel.com>; ML dri-devel <dri-devel at lists.freedesktop.org>
>> Betreff: i915 ttm_tt shmem backend
>>
>> Hi Christian,
>>
>> We are looking into using shmem as a ttm_tt backend in i915 for cached
>> system memory objects. We would also like to make such objects visible
>> to the i915-gem shrinker, so that they may be swapped out or discarded
>> when under memory pressure.
>>
>> One idea for handling this is roughly something like:
>> - Add a new TTM_PAGE_FLAG_SHMEM flag, or similar.
>> - Skip the ttm_pages_allocated accounting on such objects, similar to
>> how FLAG_SG is already handled.
>> - Skip all the page->mapping and page->index related bits, like in
>> tt_add_mapping, since it looks like these are set and used by shmem.
>> Not sure what functionally this might break, but looks like it's maybe
>> only driver specific?
>> - Skip calling into ttm_bo_swap_out/in and just have
>> ttm_populate/unpopulate handle this directly for such objects.
>> - Make such objects visible to the i915-gem shrinker.
>>
>> Does this approach look acceptable?