On Tue, Jun 7, 2022 at 12:33 PM Zach O'Keefe <zokeefe@xxxxxxxxxx> wrote:
>
> On Tue, Jun 7, 2022 at 9:01 AM Zach O'Keefe <zokeefe@xxxxxxxxxx> wrote:
> >
> > On Mon, Jun 6, 2022 at 1:46 PM Yang Shi <shy828301@xxxxxxxxx> wrote:
> > >
> > > On Fri, Jun 3, 2022 at 5:40 PM Zach O'Keefe <zokeefe@xxxxxxxxxx> wrote:
> > > >
> > > > When scanning an anon pmd to see if it's eligible for collapse, return
> > > > SCAN_PMD_MAPPED if the pmd already maps a THP. Note that
> > > > SCAN_PMD_MAPPED is different from SCAN_PAGE_COMPOUND used in the
> > > > file-collapse path, since the latter might identify pte-mapped compound
> > > > pages. This is required by MADV_COLLAPSE which necessarily needs to
> > > > know what hugepage-aligned/sized regions are already pmd-mapped.
> > > >
> > > > Signed-off-by: Zach O'Keefe <zokeefe@xxxxxxxxxx>
> > > > ---
> > > > include/trace/events/huge_memory.h | 1 +
> > > > mm/internal.h | 1 +
> > > > mm/khugepaged.c | 32 ++++++++++++++++++++++++++----
> > > > mm/rmap.c | 15 ++++++++++++--
> > > > 4 files changed, 43 insertions(+), 6 deletions(-)
> > > >
> > > > diff --git a/include/trace/events/huge_memory.h b/include/trace/events/huge_memory.h
> > > > index d651f3437367..55392bf30a03 100644
> > > > --- a/include/trace/events/huge_memory.h
> > > > +++ b/include/trace/events/huge_memory.h
> > > > @@ -11,6 +11,7 @@
> > > > EM( SCAN_FAIL, "failed") \
> > > > EM( SCAN_SUCCEED, "succeeded") \
> > > > EM( SCAN_PMD_NULL, "pmd_null") \
> > > > + EM( SCAN_PMD_MAPPED, "page_pmd_mapped") \
> > > > EM( SCAN_EXCEED_NONE_PTE, "exceed_none_pte") \
> > > > EM( SCAN_EXCEED_SWAP_PTE, "exceed_swap_pte") \
> > > > EM( SCAN_EXCEED_SHARED_PTE, "exceed_shared_pte") \
> > > > diff --git a/mm/internal.h b/mm/internal.h
> > > > index 6e14749ad1e5..f768c7fae668 100644
> > > > --- a/mm/internal.h
> > > > +++ b/mm/internal.h
> > > > @@ -188,6 +188,7 @@ extern void reclaim_throttle(pg_data_t *pgdat, enum vmscan_throttle_state reason
> > > > /*
> > > > * in mm/rmap.c:
> > > > */
> > > > +pmd_t *mm_find_pmd_raw(struct mm_struct *mm, unsigned long address);
> > > > extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
> > > >
> > > > /*
> > > > diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> > > > index cc3d6fb446d5..7a914ca19e96 100644
> > > > --- a/mm/khugepaged.c
> > > > +++ b/mm/khugepaged.c
> > > > @@ -28,6 +28,7 @@ enum scan_result {
> > > > SCAN_FAIL,
> > > > SCAN_SUCCEED,
> > > > SCAN_PMD_NULL,
> > > > + SCAN_PMD_MAPPED,
> > > > SCAN_EXCEED_NONE_PTE,
> > > > SCAN_EXCEED_SWAP_PTE,
> > > > SCAN_EXCEED_SHARED_PTE,
> > > > @@ -901,6 +902,31 @@ static int hugepage_vma_revalidate(struct mm_struct *mm, unsigned long address,
> > > > return 0;
> > > > }
> > > >
> > > > +static int find_pmd_or_thp_or_none(struct mm_struct *mm,
> > > > + unsigned long address,
> > > > + pmd_t **pmd)
> > > > +{
> > > > + pmd_t pmde;
> > > > +
> > > > + *pmd = mm_find_pmd_raw(mm, address);
> > > > + if (!*pmd)
> > > > + return SCAN_PMD_NULL;
> > > > +
> > > > + pmde = pmd_read_atomic(*pmd);
> > > > +
> > > > +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> > > > + /* See comments in pmd_none_or_trans_huge_or_clear_bad() */
> > > > + barrier();
> > > > +#endif
> > > > + if (!pmd_present(pmde))
> > > > + return SCAN_PMD_NULL;
> > > > + if (pmd_trans_huge(pmde))
> > > > + return SCAN_PMD_MAPPED;
> > > > + if (pmd_bad(pmde))
> > > > + return SCAN_FAIL;
> > >
> > > khugepaged doesn't handle pmd_bad before, IIRC it may just return
> > > SCAN_SUCCEED if everything else is good? It is fine to add it, but it
> > > may be better to return SCAN_PMD_NULL?
> >
> > Correct, pmd_bad() wasn't handled before. I actually don't know how a
> > bad pmd might arise in the wild (would love to actually know this),
> > but I don't see the check hurting (might be overly convervative
> > though). Conversely, I'm not sure where things go astray currently if
> > the pmd is bad. Guess it depends in what way the flags are mutated.
> > Returning SCAN_PMD_NULL SGTM.
> >
> > >
> > > > + return SCAN_SUCCEED;
> > > > +}
> > > > +
> > > > /*
> > > > * Bring missing pages in from swap, to complete THP collapse.
> > > > * Only done if khugepaged_scan_pmd believes it is worthwhile.
> > > > @@ -1146,11 +1172,9 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
> > > >
> > > > VM_BUG_ON(address & ~HPAGE_PMD_MASK);
> > > >
> > > > - pmd = mm_find_pmd(mm, address);
> > > > - if (!pmd) {
> > > > - result = SCAN_PMD_NULL;
> > > > + result = find_pmd_or_thp_or_none(mm, address, &pmd);
> > > > + if (result != SCAN_SUCCEED)
> > >
> > > There are a couple of other callsites for mm_find_pmd(), you may need
> > > to change all of them to find_pmd_or_thp_or_none() for MADV_COLLAPSE
> > > since khugepaged may collapse the area before MADV_COLLAPSE
> > > reacquiring mmap_lock IIUC and MADV_COLLAPSE does care this case. It
> > > is fine w/o MADV_COLLAPSE since khupaged doesn't care if it is PMD
> > > mapped or not.
> >
> > Ya, I was just questioning the same thing after responding above - at
> > least w.r.t whether the pmd_bad() also needs to be in these callsites
> > (check for pmd mapping, as you mention, I think is definitely
> > necessary). Thanks for catching this!
> >
> > > So it may be better to move this patch right before MADV_COLLAPSE is introduced?
> >
> > I think this should be ok - I'll give it a try at least.
> >
> > Again, thank you for taking the time to thoroughly review this.
> >
> > Best,
> > Zach
> >
> > > > goto out;
> > > > - }
> > > >
> > > > memset(khugepaged_node_load, 0, sizeof(khugepaged_node_load));
> > > > pte = pte_offset_map_lock(mm, pmd, address, &ptl);
> > > > diff --git a/mm/rmap.c b/mm/rmap.c
> > > > index 04fac1af870b..c9979c6ad7a1 100644
> > > > --- a/mm/rmap.c
> > > > +++ b/mm/rmap.c
> > > > @@ -767,13 +767,12 @@ unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
> > > > return vma_address(page, vma);
> > > > }
> > > >
> > > > -pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
> > > > +pmd_t *mm_find_pmd_raw(struct mm_struct *mm, unsigned long address)
> > >
> > > May be better to have some notes for mm_find_pmd_raw() and mm_find_pmd().
> > >
>
> Agreed. Looking over this code again, there are only 3 users of mm_find_pmd():
>
> 1) khugepaged
> 2) ksm (replace_page())
> 3) split_huge_pmd_address()
>
> Once khugepaged codepaths care about THP-pmds, ksm is the only
> remaining user that really wants a pte-mapping pmd.
>
> I've gone and consolidated the open-coded code in
> split_huge_pmd_address() to use the mm_find_pmd_raw().
>
> I've also done a name switch:
>
> mm_find_pmd() -> mm_find_pte_pmd()
> mm_find_pmd_raw() -> mm_find_pmd()
If ksm is the only user of *current* mm_find_pmd(), I think you should
be able to open code it w/o introducing mm_find_pte_pmd() and revert
mm_find_pmd() to its *old* behavior.
>
> This basically reverts mm_find_pmd() to its pre commit f72e7dcdd252
> ("mm: let mm_find_pmd fix buggy race with THP fault")
> behavior, and special cases (what will be, after MADV_COLLAPSE file
> support) the only remaining callsite which *doesn't* care about
> THP-pmds (ksm). The naming here is a little more meaningful than
> "*raw", and IMHO more readable.
>
>
> > > > {
> > > > pgd_t *pgd;
> > > > p4d_t *p4d;
> > > > pud_t *pud;
> > > > pmd_t *pmd = NULL;
> > > > - pmd_t pmde;
> > > >
> > > > pgd = pgd_offset(mm, address);
> > > > if (!pgd_present(*pgd))
> > > > @@ -788,6 +787,18 @@ pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
> > > > goto out;
> > > >
> > > > pmd = pmd_offset(pud, address);
> > > > +out:
> > > > + return pmd;
> > > > +}
> > > > +
> > > > +pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
> > > > +{
> > > > + pmd_t pmde;
> > > > + pmd_t *pmd;
> > > > +
> > > > + pmd = mm_find_pmd_raw(mm, address);
> > > > + if (!pmd)
> > > > + goto out;
> > > > /*
> > > > * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
> > > > * without holding anon_vma lock for write. So when looking for a
> > > > --
> > > > 2.36.1.255.ge46751e96f-goog
> > > >
