Re: [PATCH bpf-next 2/2] bpf: fix bpf_prog_pack build HPAGE_PMD_SIZE

[Eric Dumazet <eric.dumazet@gmail.com>]

On 2/8/22 14:05, Song Liu wrote:
> Fix build with CONFIG_TRANSPARENT_HUGEPAGE=n with BPF_PROG_PACK_SIZE as
> PAGE_SIZE.
>
> Fixes: 57631054fae6 ("bpf: Introduce bpf_prog_pack allocator")
> Reported-by: kernel test robot <lkp@intel.com>
> Signed-off-by: Song Liu <song@kernel.org>
> ---
>   kernel/bpf/core.c | 4 ++++
>   1 file changed, 4 insertions(+)
>
> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> index 306aa63fa58e..9519264ab1ee 100644
> --- a/kernel/bpf/core.c
> +++ b/kernel/bpf/core.c
> @@ -814,7 +814,11 @@ int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,
>    * allocator. The prog_pack allocator uses HPAGE_PMD_SIZE page (2MB on x86)
>    * to host BPF programs.
>    */
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>   #define BPF_PROG_PACK_SIZE	HPAGE_PMD_SIZE
> +#else
> +#define BPF_PROG_PACK_SIZE	PAGE_SIZE
> +#endif
>   #define BPF_PROG_CHUNK_SHIFT	6
>   #define BPF_PROG_CHUNK_SIZE	(1 << BPF_PROG_CHUNK_SHIFT)
>   #define BPF_PROG_CHUNK_MASK	(~(BPF_PROG_CHUNK_SIZE - 1))

BTW, I do not understand with module_alloc(HPAGE_PMD_SIZE) would 
necessarily allocate a huge page.

I am pretty sure it does not on x86_64 and dual socket host (NUMA)

It seems you need to multiply this by num_online_nodes()  or change the 
way __vmalloc_node_range()

works, because it currently does:

     if (vmap_allow_huge && !(vm_flags & VM_NO_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
          * their allocations due to apply_to_page_range not
          * 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)
             shift = PMD_SHIFT;
         else
             shift = arch_vmap_pte_supported_shift(size_per_node);