Re: [RFC 6/7] mm/vmalloc: Support non-blocking GFP flags in __vmalloc_area_node()

[Michal Hocko]

On Fri 04-07-25 17:25:36, Uladzislau Rezki wrote:
> This patch makes __vmalloc_area_node() to correctly handle non-blocking
> allocation requests, such as GFP_ATOMIC and GFP_NOWAIT. Main changes:
> 
> - nested_gfp flag follows the same non-blocking constraints
>   as the primary gfp_mask, ensuring consistency and avoiding
>   sleeping allocations in atomic contexts.
> 
> - if blocking is not allowed, __GFP_NOFAIL is forcibly cleared
>   and warning is issued if it was set, since __GFP_NOFAIL is
>   incompatible with non-blocking contexts;
> 
> - Add a __GFP_HIGHMEM to gfp_mask only for blocking requests
>   if there are no DMA constraints.
> 
> - in non-blocking mode we use memalloc_noreclaim_save/restore()
>   to prevent reclaim related operations that may sleep while
>   setting up page tables or mapping pages.
> 
> This is particularly important for page table allocations that
> internally use GFP_PGTABLE_KERNEL, which may sleep unless such
> scope restrictions are applied. For example:
> 
> <snip>
>     #define GFP_PGTABLE_KERNEL (GFP_KERNEL | __GFP_ZERO)
> 
>     __pte_alloc_kernel()
>         pte_alloc_one_kernel(&init_mm);
>             pagetable_alloc_noprof(GFP_PGTABLE_KERNEL & ~__GFP_HIGHMEM, 0);
> <snip>
> 
> Signed-off-by: Uladzislau Rezki (Sony) <urezki@xxxxxxxxx>
> ---
>  mm/vmalloc.c | 30 +++++++++++++++++++++++++-----
>  1 file changed, 25 insertions(+), 5 deletions(-)
> 
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 2eaff0575a9e..fe1699e01e02 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -3711,7 +3711,7 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
>  				 pgprot_t prot, unsigned int page_shift,
>  				 int node)
>  {
> -	const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
> +	gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
>  	bool nofail = gfp_mask & __GFP_NOFAIL;
>  	unsigned long addr = (unsigned long)area->addr;
>  	unsigned long size = get_vm_area_size(area);
> @@ -3719,12 +3719,28 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
>  	unsigned int nr_small_pages = size >> PAGE_SHIFT;
>  	unsigned int page_order;
>  	unsigned int flags;
> +	bool noblock;
>  	int ret;
>  
>  	array_size = (unsigned long)nr_small_pages * sizeof(struct page *);
> +	noblock = !gfpflags_allow_blocking(gfp_mask);
>  
> -	if (!(gfp_mask & (GFP_DMA | GFP_DMA32)))
> -		gfp_mask |= __GFP_HIGHMEM;
> +	if (noblock) {
> +		/* __GFP_NOFAIL is incompatible with non-blocking contexts. */
> +		WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
> +		gfp_mask &= ~__GFP_NOFAIL;

Btw. we already ignore GFP_NOFAIL for atomic allocations and warn about
that at the page allocator level (__alloc_pages_slowpath)

What we can do though is to add a pr_warn + dump_stack for request with
size that would require (in the worst case) page tables allocation
larger than a portion of min_free_kbytes (to scale with different memory
sizes). That should be plenty for any reasonable non blocking vmalloc.
We would have means to catch abusers in that way.

-- 
Michal Hocko
SUSE Labs