[PATCH v2] drm/amdgpu: check a user-provided number of BOs in list

[Marek Olšák]

USHRT_MAX seems too low. Traces for workstation apps create 20-30k BOs,
which is not very far from the limit. RADV doesn't suballocate BOs. Neither
GL nor VK has a ilmit on the number of BOs that can be created. The
hypothetical maximum number of BOs that can be allocated on a GPU with 32GB
of addressable memory is 8 million.

Marek

On Mon, Apr 28, 2025 at 10:53 AM Christian König <christian.koenig at amd.com>
wrote:

> On 4/24/25 15:40, Alex Deucher wrote:
> > On Wed, Apr 23, 2025 at 10:29 AM Christian König
> > <christian.koenig at amd.com> wrote:
> >>
> >> On 4/22/25 18:26, Deucher, Alexander wrote:
> >>> [Public]
> >>>
> >>>> -----Original Message-----
> >>>> From: Alex Deucher <alexdeucher at gmail.com>
> >>>> Sent: Tuesday, April 22, 2025 9:46 AM
> >>>> To: Koenig, Christian <Christian.Koenig at amd.com>
> >>>> Cc: Denis Arefev <arefev at swemel.ru>; Deucher, Alexander
> >>>> <Alexander.Deucher at amd.com>; David Airlie <airlied at gmail.com>;
> Simona Vetter
> >>>> <simona at ffwll.ch>; Andrey Grodzovsky <andrey.grodzovsky at amd.com>;
> >>>> Chunming Zhou <david1.zhou at amd.com>; amd-gfx at lists.freedesktop.org;
> dri-
> >>>> devel at lists.freedesktop.org; linux-kernel at vger.kernel.org; lvc-
> >>>> project at linuxtesting.org; stable at vger.kernel.org
> >>>> Subject: Re: [PATCH v2] drm/amdgpu: check a user-provided number of
> BOs in list
> >>>>
> >>>> Applied.  Thanks!
> >>>
> >>> This change beaks the following IGT tests:
> >>>
> >>> igt at amdgpu/amd_vcn at vcn-decoder-create-decode-destroy
> @vcn-decoder-create
> >>> igt at amdgpu/amd_vcn at vcn-decoder-create-decode-destroy
> @vcn-decoder-decode
> >>> igt at amdgpu/amd_vcn at vcn-decoder-create-decode-destroy
> @vcn-decoder-destroy
> >>> igt at amdgpu/amd_jpeg_dec at amdgpu_cs_jpeg_decode
> >>> igt at amdgpu/amd_cs_nop at cs-nops-with-nop-compute0
> @cs-nop-with-nop-compute0
> >>> igt at amdgpu/amd_cs_nop at cs-nops-with-sync-compute0
> @cs-nop-with-sync-compute0
> >>> igt at amdgpu/amd_cs_nop at cs-nops-with-fork-compute0
> @cs-nop-with-fork-compute0
> >>> igt at amdgpu/amd_cs_nop at cs-nops-with-sync-fork-compute0
> @cs-nop-with-sync-fork-compute0
> >>> igt at amdgpu/amd_basic at userptr-with-ip-dma@userptr
> >>> igt at amdgpu/amd_basic at cs-compute-with-ip-compute@cs-compute
> >>> igt at amdgpu/amd_basic at cs-sdma-with-ip-dma@cs-sdma
> >>> igt at amdgpu/amd_basic at eviction-test-with-ip-dma@eviction_test
> >>> igt at amdgpu/amd_cp_dma_misc at gtt_to_vram-amdgpu_hw_ip_compute0
> >>> igt at amdgpu/amd_cp_dma_misc at vram_to_gtt-amdgpu_hw_ip_compute0
> >>> igt at amdgpu/amd_cp_dma_misc at vram_to_vram-amdgpu_hw_ip_compute0
> >>
> >>
> >> Could it be that we used BO list with zero entries for those?
> >
> > Yes.  Dropping the 0 check fixed them.  E.g.,
> >
> > +       if (in->bo_number > USHRT_MAX)
> > +               return -EINVAL;
>
>
> Feel free to keep my rb on that version as well.
>
> Christian.
>
> >
> > Alex
> >
> >>
> >> Christian.
> >>
> >>>
> >>> Alex
> >>>
> >>>>
> >>>> On Tue, Apr 22, 2025 at 5:13 AM Koenig, Christian <
> Christian.Koenig at amd.com>
> >>>> wrote:
> >>>>>
> >>>>> [AMD Official Use Only - AMD Internal Distribution Only]
> >>>>>
> >>>>> Reviewed-by: Christian König <christian.koenig at amd.com>
> >>>>>
> >>>>> ________________________________________
> >>>>> Von: Denis Arefev <arefev at swemel.ru>
> >>>>> Gesendet: Freitag, 18. April 2025 10:31
> >>>>> An: Deucher, Alexander
> >>>>> Cc: Koenig, Christian; David Airlie; Simona Vetter; Andrey
> Grodzovsky;
> >>>>> Chunming Zhou; amd-gfx at lists.freedesktop.org;
> >>>>> dri-devel at lists.freedesktop.org; linux-kernel at vger.kernel.org;
> >>>>> lvc-project at linuxtesting.org; stable at vger.kernel.org
> >>>>> Betreff: [PATCH v2] drm/amdgpu: check a user-provided number of BOs
> in
> >>>>> list
> >>>>>
> >>>>> The user can set any value to the variable ‘bo_number’, via the ioctl
> >>>>> command DRM_IOCTL_AMDGPU_BO_LIST. This will affect the arithmetic
> >>>>> expression ‘in->bo_number * in->bo_info_size’, which is prone to
> >>>>> overflow. Add a valid value check.
> >>>>>
> >>>>> Found by Linux Verification Center (linuxtesting.org) with SVACE.
> >>>>>
> >>>>> Fixes: 964d0fbf6301 ("drm/amdgpu: Allow to create BO lists in CS
> ioctl
> >>>>> v3")
> >>>>> Cc: stable at vger.kernel.org
> >>>>> Signed-off-by: Denis Arefev <arefev at swemel.ru>
> >>>>> ---
> >>>>> V1 -> V2:
> >>>>> Set a reasonable limit 'USHRT_MAX' for 'bo_number' it as Christian
> >>>>> König <christian.koenig at amd.com> suggested
> >>>>>
> >>>>>  drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 3 +++
> >>>>>  1 file changed, 3 insertions(+)
> >>>>>
> >>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> >>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> >>>>> index 702f6610d024..85f7ee1e085d 100644
> >>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> >>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c
> >>>>> @@ -189,6 +189,9 @@ int amdgpu_bo_create_list_entry_array(struct
> >>>> drm_amdgpu_bo_list_in *in,
> >>>>>         struct drm_amdgpu_bo_list_entry *info;
> >>>>>         int r;
> >>>>>
> >>>>> +       if (!in->bo_number || in->bo_number > USHRT_MAX)
> >>>>> +               return -EINVAL;
> >>>>> +
> >>>>>         info = kvmalloc_array(in->bo_number, info_size, GFP_KERNEL);
> >>>>>         if (!info)
> >>>>>                 return -ENOMEM;
> >>>>> --
> >>>>> 2.43.0
> >>>>>
> >>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20250428/604bfef6/attachment-0001.htm>