Re: [PATCH] mm/vmalloc: Don't unconditonally use __GFP_HIGHMEM

From: Michal Hocko
Date: Thu Aug 17 2017 - 03:40:11 EST


On Wed 16-08-17 15:07:05, Laura Abbott wrote:
> Commit 19809c2da28a ("mm, vmalloc: use __GFP_HIGHMEM implicitly")
> added use of __GFP_HIGHMEM for allocations. vmalloc_32 may use
> GFP_DMA/GFP_DMA32 which does not play nice with __GFP_HIGHMEM
> and will drigger a BUG in gfp_zone. Only add __GFP_HIGHMEM if

s@drigger@trigger@

> we aren't using GFP_DMA/GFP_DMA32.
>
> Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1482249
> Fixes: 19809c2da28a ("mm, vmalloc: use __GFP_HIGHMEM implicitly")
> Signed-off-by: Laura Abbott <labbott@xxxxxxxxxx>

Sorry about that. My fault! I have completely missed that VM_BUG_ON.
I was double checking that GFP_DMA|__GFP_HIGHMEM works as inteded
which is the case because DMA has always a preference. I have reached
the same conclusion for GFP_DMA32 but now that I am looking at the
GFP_ZONE_TABLE again I was wrong because we would end up in ZONE_DMA
AFAICS because bit 1 wouldn't be set.

Acked-by: Michal Hocko <mhocko@xxxxxxxx>

Thanks!

> ---
> mm/vmalloc.c | 13 ++++++++-----
> 1 file changed, 8 insertions(+), 5 deletions(-)
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 8698c1c86c4d..a47e3894c775 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -1671,7 +1671,10 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
> struct page **pages;
> unsigned int nr_pages, array_size, i;
> const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
> - const gfp_t alloc_mask = gfp_mask | __GFP_HIGHMEM | __GFP_NOWARN;
> + const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
> + const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
> + 0 :
> + __GFP_HIGHMEM;
>
> nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
> array_size = (nr_pages * sizeof(struct page *));
> @@ -1679,7 +1682,7 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
> area->nr_pages = nr_pages;
> /* Please note that the recursion is strictly bounded. */
> if (array_size > PAGE_SIZE) {
> - pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
> + pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
> PAGE_KERNEL, node, area->caller);
> } else {
> pages = kmalloc_node(array_size, nested_gfp, node);
> @@ -1700,9 +1703,9 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
> }
>
> if (node == NUMA_NO_NODE)
> - page = alloc_page(alloc_mask);
> + page = alloc_page(alloc_mask|highmem_mask);
> else
> - page = alloc_pages_node(node, alloc_mask, 0);
> + page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
>
> if (unlikely(!page)) {
> /* Successfully allocated i pages, free them in __vunmap() */
> @@ -1710,7 +1713,7 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
> goto fail;
> }
> area->pages[i] = page;
> - if (gfpflags_allow_blocking(gfp_mask))
> + if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
> cond_resched();
> }
>
> --
> 2.13.0
>

--
Michal Hocko
SUSE Labs