Re: [Intel-gfx] [PATCH v4 01/14] drm/ttm: stop calling tt_swapin in vm_access

["Christian König" <christian.koenig@amd.com>]

Am 21.09.21 um 13:37 schrieb Thomas Hellström:
> HI, Christian,
> On Tue, 2021-09-21 at 13:28 +0200, Christian König wrote:
>> Am 21.09.21 um 13:01 schrieb Matthew Auld:
>>> In commit:
>>>
>>> commit 09ac4fcb3f255e9225967c75f5893325c116cdbe
>>> Author: Felix Kuehling <Felix.Kuehling@amd.com>
>>> Date:   Thu Jul 13 17:01:16 2017 -0400
>>>
>>>       drm/ttm: Implement vm_operations_struct.access v2
>>>
>>> we added the vm_access hook, where we also directly call tt_swapin
>>> for
>>> some reason. If something is swapped-out then the ttm_tt must also
>>> be
>>> unpopulated, and since access_kmap should also call tt_populate, if
>>> needed, then swapping-in will already be handled there.
>> Sounds like you completely misunderstand what that is good for.
>>
>> This is for debugger attaching to a process and peek/poke into the
>> VMA
>> and completely unrelated to kmap.
> I think what Matthew is saying is that there is a fallthrough to
> TTM_PL_TT which calls
>
> ttm_bo_vm_access_kmap

Ah, good point. Now that makes much more sense.

>
> which calls
>
> ttm_tt_populate().
>
> So from my pow, unless there are other concerns, this is
>
> Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

In that case the patch is Reviewed-by: Christian König 
<christian.koenig@amd.com> as well.

When this is gone we can also make ttm_tt_swapin() static since that 
here is the only user.

And BTW replacing the switch/case with a check for use_tt in the 
resource_manager is probably a good idea as well.

Regards,
Christian.

>
>
>>> If anything, calling tt_swapin directly here would likely always
>>> fail
>>> since the tt->pages won't yet be populated, or worse since the tt-
>>>> pages
>>> array is never actually cleared in unpopulate this might lead to a
>>> nasty
>>> uaf.
>> That's indeed true, but we just need to unconditionally call
>> ttm_tt_populate() here instead.
>>
>> Regards,
>> Christian.
>>
>>> Fixes: 09ac4fcb3f25 ("drm/ttm: Implement
>>> vm_operations_struct.access v2")
>>> Signed-off-by: Matthew Auld <matthew.auld@intel.com>
>>> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>>> Cc: Christian König <christian.koenig@amd.com>
>>> ---
>>>    drivers/gpu/drm/ttm/ttm_bo_vm.c | 5 -----
>>>    1 file changed, 5 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c
>>> b/drivers/gpu/drm/ttm/ttm_bo_vm.c
>>> index f56be5bc0861..5b9b7fd01a69 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
>>> @@ -519,11 +519,6 @@ int ttm_bo_vm_access(struct vm_area_struct
>>> *vma, unsigned long addr,
>>>    
>>>          switch (bo->resource->mem_type) {
>>>          case TTM_PL_SYSTEM:
>>> -               if (unlikely(bo->ttm->page_flags &
>>> TTM_PAGE_FLAG_SWAPPED)) {
>>> -                       ret = ttm_tt_swapin(bo->ttm);
>>> -                       if (unlikely(ret != 0))
>>> -                               return ret;
>>> -               }
>>>                  fallthrough;
>>>          case TTM_PL_TT:
>>>                  ret = ttm_bo_vm_access_kmap(bo, offset, buf, len,
>>> write);
>