lkml.org 
[lkml]   [2022]   [Jul]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [RFC PATCH 17/26] hugetlb: update follow_hugetlb_page to support HGM
On Tue, Jul 19, 2022 at 3:48 AM manish.mishra <manish.mishra@nutanix.com> wrote:
>
>
> On 24/06/22 11:06 pm, James Houghton wrote:
> > This enables support for GUP, and it is needed for the KVM demand paging
> > self-test to work.
> >
> > One important change here is that, before, we never needed to grab the
> > i_mmap_sem, but now, to prevent someone from collapsing the page tables
> > out from under us, we grab it for reading when doing high-granularity PT
> > walks.
> >
> > Signed-off-by: James Houghton <jthoughton@google.com>
> > ---
> > mm/hugetlb.c | 70 ++++++++++++++++++++++++++++++++++++++++++----------
> > 1 file changed, 57 insertions(+), 13 deletions(-)
> >
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index f9c7daa6c090..aadfcee947cf 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -6298,14 +6298,18 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > unsigned long vaddr = *position;
> > unsigned long remainder = *nr_pages;
> > struct hstate *h = hstate_vma(vma);
> > + struct address_space *mapping = vma->vm_file->f_mapping;
> > int err = -EFAULT, refs;
> > + bool has_i_mmap_sem = false;
> >
> > while (vaddr < vma->vm_end && remainder) {
> > pte_t *pte;
> > spinlock_t *ptl = NULL;
> > bool unshare = false;
> > int absent;
> > + unsigned long pages_per_hpte;
> > struct page *page;
> > + struct hugetlb_pte hpte;
> >
> > /*
> > * If we have a pending SIGKILL, don't keep faulting pages and
> > @@ -6325,9 +6329,23 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > */
> > pte = huge_pte_offset(mm, vaddr & huge_page_mask(h),
> > huge_page_size(h));
> > - if (pte)
> > - ptl = huge_pte_lock(h, mm, pte);
> > - absent = !pte || huge_pte_none(huge_ptep_get(pte));
> > + if (pte) {
> > + hugetlb_pte_populate(&hpte, pte, huge_page_shift(h));
> > + if (hugetlb_hgm_enabled(vma)) {
> > + BUG_ON(has_i_mmap_sem);
>
> Just thinking can we do without i_mmap_lock_read in most cases. Like earlier
>
> this function was good without i_mmap_lock_read doing almost everything
>
> which is happening now?

We need something to prevent the page tables from being rearranged
while we're walking them. In this RFC, I used the i_mmap_lock. I'm
going to change it, probably to a per-VMA lock (or maybe a per-hpage
lock. I'm trying to figure out if a system with PTLs/hugetlb_pte_lock
could work too :)).

>
> > + i_mmap_lock_read(mapping);
> > + /*
> > + * Need to hold the mapping semaphore for
> > + * reading to do a HGM walk.
> > + */
> > + has_i_mmap_sem = true;
> > + hugetlb_walk_to(mm, &hpte, vaddr, PAGE_SIZE,
> > + /*stop_at_none=*/true);
> > + }
> > + ptl = hugetlb_pte_lock(mm, &hpte);
> > + }
> > +
> > + absent = !pte || hugetlb_pte_none(&hpte);
> >
> > /*
> > * When coredumping, it suits get_dump_page if we just return
> > @@ -6338,8 +6356,13 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > */
> > if (absent && (flags & FOLL_DUMP) &&
> > !hugetlbfs_pagecache_present(h, vma, vaddr)) {
> > - if (pte)
> > + if (pte) {
> > + if (has_i_mmap_sem) {
> > + i_mmap_unlock_read(mapping);
> > + has_i_mmap_sem = false;
> > + }
> > spin_unlock(ptl);
> > + }
> > remainder = 0;
> > break;
> > }
> > @@ -6359,8 +6382,13 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > vm_fault_t ret;
> > unsigned int fault_flags = 0;
> >
> > - if (pte)
> > + if (pte) {
> > + if (has_i_mmap_sem) {
> > + i_mmap_unlock_read(mapping);
> > + has_i_mmap_sem = false;
> > + }
> > spin_unlock(ptl);
> > + }
> > if (flags & FOLL_WRITE)
> > fault_flags |= FAULT_FLAG_WRITE;
> > else if (unshare)
> > @@ -6403,8 +6431,11 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > continue;
> > }
> >
> > - pfn_offset = (vaddr & ~huge_page_mask(h)) >> PAGE_SHIFT;
> > - page = pte_page(huge_ptep_get(pte));
> > + pfn_offset = (vaddr & ~hugetlb_pte_mask(&hpte)) >> PAGE_SHIFT;
> > + page = pte_page(hugetlb_ptep_get(&hpte));
> > + pages_per_hpte = hugetlb_pte_size(&hpte) / PAGE_SIZE;
> > + if (hugetlb_hgm_enabled(vma))
> > + page = compound_head(page);
> >
> > VM_BUG_ON_PAGE((flags & FOLL_PIN) && PageAnon(page) &&
> > !PageAnonExclusive(page), page);
> > @@ -6414,17 +6445,21 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > * and skip the same_page loop below.
> > */
> > if (!pages && !vmas && !pfn_offset &&
> > - (vaddr + huge_page_size(h) < vma->vm_end) &&
> > - (remainder >= pages_per_huge_page(h))) {
> > - vaddr += huge_page_size(h);
> > - remainder -= pages_per_huge_page(h);
> > - i += pages_per_huge_page(h);
> > + (vaddr + pages_per_hpte < vma->vm_end) &&
> > + (remainder >= pages_per_hpte)) {
> > + vaddr += pages_per_hpte;
> > + remainder -= pages_per_hpte;
> > + i += pages_per_hpte;
> > spin_unlock(ptl);
> > + if (has_i_mmap_sem) {
> > + has_i_mmap_sem = false;
> > + i_mmap_unlock_read(mapping);
> > + }
> > continue;
> > }
> >
> > /* vaddr may not be aligned to PAGE_SIZE */
> > - refs = min3(pages_per_huge_page(h) - pfn_offset, remainder,
> > + refs = min3(pages_per_hpte - pfn_offset, remainder,
> > (vma->vm_end - ALIGN_DOWN(vaddr, PAGE_SIZE)) >> PAGE_SHIFT);
> >
> > if (pages || vmas)
> > @@ -6447,6 +6482,10 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > if (WARN_ON_ONCE(!try_grab_folio(pages[i], refs,
> > flags))) {
> > spin_unlock(ptl);
> > + if (has_i_mmap_sem) {
> > + has_i_mmap_sem = false;
> > + i_mmap_unlock_read(mapping);
> > + }
> > remainder = 0;
> > err = -ENOMEM;
> > break;
> > @@ -6458,8 +6497,13 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> > i += refs;
> >
> > spin_unlock(ptl);
> > + if (has_i_mmap_sem) {
> > + has_i_mmap_sem = false;
> > + i_mmap_unlock_read(mapping);
> > + }
> > }
> > *nr_pages = remainder;
> > + BUG_ON(has_i_mmap_sem);
> > /*
> > * setting position is actually required only if remainder is
> > * not zero but it's faster not to add a "if (remainder)"
>
> Thanks
>
> Manish Mishra
>

\
 
 \ /
  Last update: 2022-07-19 18:20    [W:0.122 / U:0.624 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site