2021-07-31 08:14:49

by Tuo Li

Subject: [PATCH] drm/amdgpu: fix possible null-pointer dereference in amdgpu_ttm_tt_unpopulate()

The variable ttm is assigned to the variable gtt, and the variable gtt
is checked in:
if (gtt && gtt->userptr)

This indicates that both ttm and gtt can be NULL.
If so, a null-pointer dereference will occur:
if (ttm->page_flags & TTM_PAGE_FLAG_SG)

Also, some null-pointer dereferences will occur in the function
ttm_pool_free() which is called in:
return ttm_pool_free(&adev->mman.bdev.pool, ttm);

To fix these possible null-pointer dereferences, the function returns
when ttm is NULL.

Reported-by: TOTE Robot <[email protected]>
Signed-off-by: Tuo Li <[email protected]>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 3a55f08e00e1..0216ca085f11 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1146,7 +1146,10 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
struct amdgpu_ttm_tt *gtt = (void *)ttm;
struct amdgpu_device *adev;

- if (gtt && gtt->userptr) {
+ if (ttm == NULL)
+ return;
+
+ if (gtt->userptr) {
amdgpu_ttm_tt_set_user_pages(ttm, NULL);
kfree(ttm->sg);
ttm->sg = NULL;
--
2.25.1

2021-08-01 17:24:52

by Christian König

Subject: Re: [PATCH] drm/amdgpu: fix possible null-pointer dereference in amdgpu_ttm_tt_unpopulate()

Am 31.07.21 um 10:13 schrieb Tuo Li:
> The variable ttm is assigned to the variable gtt, and the variable gtt
> is checked in:
> if (gtt && gtt->userptr)
>
> This indicates that both ttm and gtt can be NULL.
> If so, a null-pointer dereference will occur:
> if (ttm->page_flags & TTM_PAGE_FLAG_SG)
>
> Also, some null-pointer dereferences will occur in the function
> ttm_pool_free() which is called in:
> return ttm_pool_free(&adev->mman.bdev.pool, ttm);
>
> To fix these possible null-pointer dereferences, the function returns
> when ttm is NULL.

NAK, same as with the other patch.

The ttm object is mandatory, asking the driver to destroy a ttm object
which doesn't exists makes no sense at all and is a bug in the upper layer.

The NULL check is just a leftover from when the gtt and ttm objects
where distinct. Please remove that one instead.

BTW: Bonus points for changing the (void *) cast into a much cleaner
container_of().

Thanks,
Christian.

>
> Reported-by: TOTE Robot <[email protected]>
> Signed-off-by: Tuo Li <[email protected]>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 3a55f08e00e1..0216ca085f11 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -1146,7 +1146,10 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
> struct amdgpu_ttm_tt *gtt = (void *)ttm;
> struct amdgpu_device *adev;
>
> - if (gtt && gtt->userptr) {
> + if (ttm == NULL)
> + return;
> +
> + if (gtt->userptr) {
> amdgpu_ttm_tt_set_user_pages(ttm, NULL);
> kfree(ttm->sg);
> ttm->sg = NULL;