Re: [PATCH v3 1/2] mm/vmalloc: add vmalloc_no_huge

[Nicholas Piggin]

Excerpts from Claudio Imbrenda's message of June 11, 2021 1:42 am:
> The recent patches to add support for hugepage vmalloc mappings added a
> flag for __vmalloc_node_range to allow to request small pages.
> This flag is not accessible when calling vmalloc, the only option is to
> call directly __vmalloc_node_range, which is not exported.
> 
> This means that a module can't vmalloc memory with small pages.
> 
> Case in point: KVM on s390x needs to vmalloc a large area, and it needs
> to be mapped with small pages, because of a hardware limitation.
> 
> This patch adds the function vmalloc_no_huge, which works like vmalloc,
> but it is guaranteed to always back the mapping using small pages. This
> function is exported, therefore it is usable by modules.
> 
> Signed-off-by: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Nicholas Piggin <npiggin@xxxxxxxxx>
> Cc: Uladzislau Rezki (Sony) <urezki@xxxxxxxxx>
> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> Cc: David Rientjes <rientjes@xxxxxxxxxx>
> Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>
> ---
>  include/linux/vmalloc.h |  1 +
>  mm/vmalloc.c            | 16 ++++++++++++++++
>  2 files changed, 17 insertions(+)
> 
> diff --git a/include/linux/vmalloc.h b/include/linux/vmalloc.h
> index 4d668abb6391..bfaaf0b6fa76 100644
> --- a/include/linux/vmalloc.h
> +++ b/include/linux/vmalloc.h
> @@ -135,6 +135,7 @@ extern void *__vmalloc_node_range(unsigned long size, unsigned long align,
>  			const void *caller);
>  void *__vmalloc_node(unsigned long size, unsigned long align, gfp_t gfp_mask,
>  		int node, const void *caller);
> +void *vmalloc_no_huge(unsigned long size);
>  
>  extern void vfree(const void *addr);
>  extern void vfree_atomic(const void *addr);
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index a13ac524f6ff..296a2fcc3fbe 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -2998,6 +2998,22 @@ void *vmalloc(unsigned long size)
>  }
>  EXPORT_SYMBOL(vmalloc);
>  
> +/**
> + * vmalloc_no_huge - allocate virtually contiguous memory using small pages
> + * @size:    allocation size
> + *
> + * Allocate enough non-huge pages to cover @size from the page level
> + * allocator and map them into contiguous kernel virtual space.
> + *
> + * Return: pointer to the allocated memory or %NULL on error
> + */
> +void *vmalloc_no_huge(unsigned long size)
> +{
> +	return __vmalloc_node_range(size, 1, VMALLOC_START, VMALLOC_END, GFP_KERNEL, PAGE_KERNEL,
> +				    VM_NO_HUGE_VMAP, NUMA_NO_NODE, __builtin_return_address(0));
> +}
> +EXPORT_SYMBOL(vmalloc_no_huge);

At some point if the combination of flags becomes too much we will need a 
different strategy. A vmalloc API with (size, align, gfp_t, vm_flags, 
node) args would help 3/6 of the existing non-arch callers too. And one 
more if you had a prot parameter or _exec variant.

But for now I'm okay with this.

Acked-by: Nicholas Piggin <npiggin@xxxxxxxxx>

Thanks,
Nick