[v6,2/8] mm: vmalloc: don't account for number of nodes for HUGE_VMAP allocations

Commit Message

Mike Rapoport Oct. 16, 2024, 12:24 p.m. UTC

From: "Mike Rapoport (Microsoft)" <[e-mail]>

vmalloc allocations with VM_ALLOW_HUGE_VMAP that do not explicitly
specify node ID will use huge pages only if size_per_node is larger than
a huge page.
Still the actual allocated memory is not distributed between nodes and
there is no advantage in such approach.
On the contrary, BPF allocates SZ_2M * num_possible_nodes() for each
new bpf_prog_pack, while it could do with a single huge page per pack.

Don't account for number of nodes for VM_ALLOW_HUGE_VMAP with
NUMA_NO_NODE and use huge pages whenever the requested allocation size
is larger than a huge page.

Signed-off-by: Mike Rapoport (Microsoft) <[e-mail]>
Reviewed-by: Christoph Hellwig <[e-mail]>
---
 mm/vmalloc.c | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)

Comments

Uladzislau Rezki Oct. 16, 2024, 5:21 p.m. UTC | #1

On Wed, Oct 16, 2024 at 03:24:18PM +0300, Mike Rapoport wrote:
> From: "Mike Rapoport (Microsoft)" <[e-mail]>
> 
> vmalloc allocations with VM_ALLOW_HUGE_VMAP that do not explicitly
> specify node ID will use huge pages only if size_per_node is larger than
> a huge page.
> Still the actual allocated memory is not distributed between nodes and
> there is no advantage in such approach.
> On the contrary, BPF allocates SZ_2M * num_possible_nodes() for each
> new bpf_prog_pack, while it could do with a single huge page per pack.
> 
> Don't account for number of nodes for VM_ALLOW_HUGE_VMAP with
> NUMA_NO_NODE and use huge pages whenever the requested allocation size
> is larger than a huge page.
> 
> Signed-off-by: Mike Rapoport (Microsoft) <[e-mail]>
> Reviewed-by: Christoph Hellwig <[e-mail]>
> ---
>  mm/vmalloc.c | 9 ++-------
>  1 file changed, 2 insertions(+), 7 deletions(-)
> 
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 634162271c00..86b2344d7461 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -3763,8 +3763,6 @@ void *__vmalloc_node_range_noprof(unsigned long size, unsigned long align,
>  	}
>  
>  	if (vmap_allow_huge && (vm_flags & VM_ALLOW_HUGE_VMAP)) {
> -		unsigned long size_per_node;
> -
>  		/*
>  		 * Try huge pages. Only try for PAGE_KERNEL allocations,
>  		 * others like modules don't yet expect huge pages in
> @@ -3772,13 +3770,10 @@ void *__vmalloc_node_range_noprof(unsigned long size, unsigned long align,
>  		 * supporting them.
>  		 */
>  
> -		size_per_node = size;
> -		if (node == NUMA_NO_NODE)
> -			size_per_node /= num_online_nodes();
> -		if (arch_vmap_pmd_supported(prot) && size_per_node >= PMD_SIZE)
> +		if (arch_vmap_pmd_supported(prot) && size >= PMD_SIZE)
>  			shift = PMD_SHIFT;
>  		else
> -			shift = arch_vmap_pte_supported_shift(size_per_node);
> +			shift = arch_vmap_pte_supported_shift(size);
>  
>  		align = max(real_align, 1UL << shift);
>  		size = ALIGN(real_size, 1UL << shift);
>
Looking at this place, i see that an overwriting a "size" approach seems as
something that is a bit hard to follow. Below we have following code:

<snip>
...
again:
	area = __get_vm_area_node(real_size, align, shift, VM_ALLOC |
	  VM_UNINITIALIZED | vm_flags, start, end, node,
	  gfp_mask, caller);
...
<snip>

where we pass a "real_size", whereas there is only one place in the
__vmalloc_node_range_noprof() function where a "size" is used. It is
in the end of function:

<snip>
...
	size = PAGE_ALIGN(size);
	if (!(vm_flags & VM_DEFER_KMEMLEAK))
		kmemleak_vmalloc(area, size, gfp_mask);

	return area->addr;
<snip>

As fro this patch:

Reviewed-by: Uladzislau Rezki (Sony) <[e-mail]>

--
Uladzislau Rezki

Luis Chamberlain Oct. 19, 2024, 10:19 p.m. UTC | #2

On Wed, Oct 16, 2024 at 03:24:18PM +0300, Mike Rapoport wrote:
> From: "Mike Rapoport (Microsoft)" <[e-mail]>
> 
> vmalloc allocations with VM_ALLOW_HUGE_VMAP that do not explicitly
> specify node ID will use huge pages only if size_per_node is larger than
> a huge page.
> Still the actual allocated memory is not distributed between nodes and
> there is no advantage in such approach.
> On the contrary, BPF allocates SZ_2M * num_possible_nodes() for each
> new bpf_prog_pack, while it could do with a single huge page per pack.
> 
> Don't account for number of nodes for VM_ALLOW_HUGE_VMAP with
> NUMA_NO_NODE and use huge pages whenever the requested allocation size
> is larger than a huge page.
> 
> Signed-off-by: Mike Rapoport (Microsoft) <[e-mail]>
> Reviewed-by: Christoph Hellwig <[e-mail]>

Reviewed-by: Luis Chamberlain <[e-mail]>

  Luis

Patch

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 634162271c00..86b2344d7461 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -3763,8 +3763,6 @@  void *__vmalloc_node_range_noprof(unsigned long size, unsigned long align,
 	}
 
 	if (vmap_allow_huge && (vm_flags & VM_ALLOW_HUGE_VMAP)) {
-		unsigned long size_per_node;
-
 		/*
 		 * Try huge pages. Only try for PAGE_KERNEL allocations,
 		 * others like modules don't yet expect huge pages in
@@ -3772,13 +3770,10 @@  void *__vmalloc_node_range_noprof(unsigned long size, unsigned long align,
 		 * supporting them.
 		 */
 
-		size_per_node = size;
-		if (node == NUMA_NO_NODE)
-			size_per_node /= num_online_nodes();
-		if (arch_vmap_pmd_supported(prot) && size_per_node >= PMD_SIZE)
+		if (arch_vmap_pmd_supported(prot) && size >= PMD_SIZE)
 			shift = PMD_SHIFT;
 		else
-			shift = arch_vmap_pte_supported_shift(size_per_node);
+			shift = arch_vmap_pte_supported_shift(size);
 
 		align = max(real_align, 1UL << shift);
 		size = ALIGN(real_size, 1UL << shift);