On 2021-01-13 2:13 p.m., Christian König wrote:
> The only flag we really need is __GFP_NOMEMALLOC, highmem depends on
> dma32 and moveable/compound should never be set in the first place.
> 
> Signed-off-by: Christian König <christian.koenig@amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_pool.c | 11 ++++++-----
>   1 file changed, 6 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ttm/ttm_pool.c b/drivers/gpu/drm/ttm/ttm_pool.c
> index 8cd776adc592..11e0313db0ea 100644
> --- a/drivers/gpu/drm/ttm/ttm_pool.c
> +++ b/drivers/gpu/drm/ttm/ttm_pool.c
> @@ -79,12 +79,13 @@ static struct page *ttm_pool_alloc_page(struct ttm_pool *pool, gfp_t gfp_flags,
>   	struct page *p;
>   	void *vaddr;
>   
> -	if (order) {
> -		gfp_flags |= GFP_TRANSHUGE_LIGHT | __GFP_NORETRY |
> +	/* Don't set the __GFP_COMP flag for higher order allocations.
> +	 * Mapping pages directly into an userspace process and calling
> +	 * put_page() on a TTM allocated page is illegal.
> +	 */
> +	if (order)
> +		gfp_flags |= __GFP_NOMEMALLOC | __GFP_NORETRY |
>   			__GFP_KSWAPD_RECLAIM;
> -		gfp_flags &= ~__GFP_MOVABLE;
> -		gfp_flags &= ~__GFP_COMP;
> -	}
>   
>   	if (!pool->use_dma_alloc) {
>   		p = alloc_pages(gfp_flags, order);
> 

I picked up this change today, and got the attached splat while running 
piglit.

scripts/faddr2line drivers/gpu/drm/ttm/ttm.ko ttm_pool_alloc+0x2e4/0x5e0

gives:

ttm_pool_alloc+0x2e4/0x5e0:
  alloc_pages at /home/daenzer/src/linux-git/linux/./include/linux/gfp.h:547

(inlined by) ttm_pool_alloc_page at 
/home/daenzer/src/linux-git/linux/drivers/gpu/drm//ttm/ttm_pool.c:91

(inlined by) ttm_pool_alloc at 
/home/daenzer/src/linux-git/linux/drivers/gpu/drm//ttm/ttm_pool.c:398



I suspect we need __GFP_NOWARN as well to avoid these splats.


-- 
Earthling Michel Dänzer               |               https://redhat.com
Libre software enthusiast             |             Mesa and X developer