Re: [PATCH v4 07/10] powerpc/mm: Use generic_hugetlb_get_unmapped_area()

[Catalin Marinas <catalin.marinas@arm.com>]

On Thu, Dec 16, 2021 at 05:13:47PM +0000, Christophe Leroy wrote:
> Le 09/12/2021 à 11:02, Nicholas Piggin a écrit :
> > Excerpts from Christophe Leroy's message of December 9, 2021 3:18 am:
> >> Use the generic version of arch_hugetlb_get_unmapped_area()
> >> which is now available at all time.
> >>
> >> Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
> >> ---
> >>   arch/powerpc/include/asm/book3s/64/hugetlb.h |  4 --
> >>   arch/powerpc/mm/book3s64/radix_hugetlbpage.c | 55 --------------------
> >>   arch/powerpc/mm/hugetlbpage.c                |  4 +-
> >>   3 files changed, 1 insertion(+), 62 deletions(-)
> >>
> >> diff --git a/arch/powerpc/include/asm/book3s/64/hugetlb.h b/arch/powerpc/include/asm/book3s/64/hugetlb.h
> >> index 12e150e615b7..b37a28f62cf6 100644
> >> --- a/arch/powerpc/include/asm/book3s/64/hugetlb.h
> >> +++ b/arch/powerpc/include/asm/book3s/64/hugetlb.h
> >> @@ -8,10 +8,6 @@
> >>    */
> >>   void radix__flush_hugetlb_page(struct vm_area_struct *vma, unsigned long vmaddr);
> >>   void radix__local_flush_hugetlb_page(struct vm_area_struct *vma, unsigned long vmaddr);
> >> -extern unsigned long
> >> -radix__hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
> >> -				unsigned long len, unsigned long pgoff,
> >> -				unsigned long flags);
> >>   
> >>   extern void radix__huge_ptep_modify_prot_commit(struct vm_area_struct *vma,
> >>   						unsigned long addr, pte_t *ptep,
> >> diff --git a/arch/powerpc/mm/book3s64/radix_hugetlbpage.c b/arch/powerpc/mm/book3s64/radix_hugetlbpage.c
> >> index 23d3e08911d3..d2fb776febb4 100644
> >> --- a/arch/powerpc/mm/book3s64/radix_hugetlbpage.c
> >> +++ b/arch/powerpc/mm/book3s64/radix_hugetlbpage.c
> >> @@ -41,61 +41,6 @@ void radix__flush_hugetlb_tlb_range(struct vm_area_struct *vma, unsigned long st
> >>   		radix__flush_tlb_range_psize(vma->vm_mm, start, end, psize);
> >>   }
> >>   
> >> -/*
> >> - * A vairant of hugetlb_get_unmapped_area doing topdown search
> >> - * FIXME!! should we do as x86 does or non hugetlb area does ?
> >> - * ie, use topdown or not based on mmap_is_legacy check ?
> >> - */
> >> -unsigned long
> >> -radix__hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
> >> -				unsigned long len, unsigned long pgoff,
> >> -				unsigned long flags)
> >> -{
> >> -	struct mm_struct *mm = current->mm;
> >> -	struct vm_area_struct *vma;
> >> -	struct hstate *h = hstate_file(file);
> >> -	int fixed = (flags & MAP_FIXED);
> >> -	unsigned long high_limit;
> >> -	struct vm_unmapped_area_info info;
> >> -
> >> -	high_limit = DEFAULT_MAP_WINDOW;
> >> -	if (addr >= high_limit || (fixed && (addr + len > high_limit)))
> >> -		high_limit = TASK_SIZE;
> > 
> > I wonder if generic hugetlb_get_unmapped_area needs to have the
> > arch_get_mmap_end() added.
> > 
> > arm64 has arch_get_mmap_end() and !HAVE_ARCH_HUGETLB_UNMAPPED_AREA so
> > it looks like it has broken large address hint logic for hugetlbfs
> > mappings? x86-64 defines their own and does the same hinting for
> > normal and hugetlbfs mmap.
> > 
> > If we had that and defied arch_get_mmap_end(), then this patch should
> > work.
> > 
> 
> As far as I can see, hugetlb_get_unmapped_area() variants used to be 
> very similar to get_unmapped_area() until commit 1be7107fbe18 ("mm: 
> larger stack guard gap, between vmas") and commit f6795053dac8 ("mm: 
> mmap: Allow for "high" userspace addresses")
> 
> I see no reason why those changes couldn't apply to 
> hugetlb_get_unmapped_area() as well.
> 
> Need to know what ARM64 think about it thought. Will, Catalin, any opinion ?

I think we should have fixed hugetlb_get_unmapped_area() as well when we
added support for 52-bit VA. The reason for commit f6795053dac8 was to
prevent normal mmap() from returning addresses above 48-bit by default
as some user-space had hard assumptions about this.

It's a slight ABI change if you do this for hugetlb_get_unmapped_area()
but I doubt anyone would notice. It's more likely that the current
behaviour would cause issues, so I'd rather have them consistent.

-- 
Catalin