Re: [PATCH] drm/radeon: fix potential racing issue due to mmap_lock

[Christian König]

Am 13.11.22 um 13:42 schrieb Dawei Li:
> Both find_vma() and get_user_pages() need explicit protection of
> mmap lock, fix them by mmap_lock and get_user_pages_fast().

NAK, the MM read lock should already be taken when we reach this function.

Could be that this is buggy and the function is called without holding 
the lock, but trying to grab it while holding the reservation lock is 
also forbidden.

Christian.

> Fixes: ddd00e33e17a ("drm/radeon: add userptr flag to limit it to anonymous memory v2")
> Fixes: f72a113a71ab ("drm/radeon: add userptr support v8")
> Signed-off-by: Dawei Li <set_pte_at@xxxxxxxxxxx>
> ---
>   drivers/gpu/drm/radeon/radeon_ttm.c | 6 ++++--
>   1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon_ttm.c b/drivers/gpu/drm/radeon/radeon_ttm.c
> index d33fec488713..741ea64b9402 100644
> --- a/drivers/gpu/drm/radeon/radeon_ttm.c
> +++ b/drivers/gpu/drm/radeon/radeon_ttm.c
> @@ -351,7 +351,10 @@ static int radeon_ttm_tt_pin_userptr(struct ttm_device *bdev, struct ttm_tt *ttm
>   		   to prevent problems with writeback */
>   		unsigned long end = gtt->userptr + (u64)ttm->num_pages * PAGE_SIZE;
>   		struct vm_area_struct *vma;
> +
> +		mmap_read_lock(gtt->usermm);
>   		vma = find_vma(gtt->usermm, gtt->userptr);
> +		mmap_read_unlock(gtt->usermm);
>   		if (!vma || vma->vm_file || vma->vm_end < end)
>   			return -EPERM;
>   	}
> @@ -361,8 +364,7 @@ static int radeon_ttm_tt_pin_userptr(struct ttm_device *bdev, struct ttm_tt *ttm
>   		uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE;
>   		struct page **pages = ttm->pages + pinned;
>   
> -		r = get_user_pages(userptr, num_pages, write ? FOLL_WRITE : 0,
> -				   pages, NULL);
> +		r = get_user_pages_fast(userptr, num_pages, write ? FOLL_WRITE : 0, pages);
>   		if (r < 0)
>   			goto release_pages;
>