| Message ID | 20231214073912.1938330-2-sunnanyong@huawei.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | A Solution to Re-enable hugetlb vmemmap optimize on Arm | expand |
On 2023/12/14 15:39, Nanyong Sun wrote: > Add pmd/pte update and tlb flush helper function to update page > table. This refactoring patch is designed to facilitate each > architecture to implement its own special logic in preparation > for the arm64 architecture to follow the necessary break-before-make > sequence when updating page tables. > > Signed-off-by: Nanyong Sun <sunnanyong@huawei.com> > --- > mm/hugetlb_vmemmap.c | 55 ++++++++++++++++++++++++++++++++++---------- > 1 file changed, 43 insertions(+), 12 deletions(-) > > diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c > index 87818ee7f01d..49e8b351def3 100644 > --- a/mm/hugetlb_vmemmap.c > +++ b/mm/hugetlb_vmemmap.c > @@ -45,6 +45,37 @@ struct vmemmap_remap_walk { > unsigned long flags; > }; > > +#ifndef vmemmap_update_pmd > +static inline void vmemmap_update_pmd(unsigned long start, > + pmd_t *pmd, pte_t *pgtable) > +{ > + pmd_populate_kernel(&init_mm, pmd, pgtable); > +} > +#endif > + > +#ifndef vmemmap_update_pte > +static inline void vmemmap_update_pte(unsigned long addr, > + pte_t *pte, pte_t entry) > +{ > + set_pte_at(&init_mm, addr, pte, entry); > +} > +#endif > + > +#ifndef flush_tlb_vmemmap_all > +static inline void flush_tlb_vmemmap_all(void) > +{ > + flush_tlb_all(); > +} > +#endif > + > +#ifndef flush_tlb_vmemmap_range > +static inline void flush_tlb_vmemmap_range(unsigned long start, > + unsigned long end) > +{ > + flush_tlb_kernel_range(start, end); > +} > +#endif I'd like to rename both tlb-flush helpers to vmemmap_flush_tlb_all/range since other helpers all are prefixed with "vmemmap". It'll be more consistent for me. Otherwise LGTM. Thanks. Reviewed-by: Muchun Song <songmuchun@bytedance.com> > + > static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start, bool flush) > { > pmd_t __pmd; > @@ -87,9 +118,9 @@ static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start, bool flush) > > /* Make pte visible before pmd. See comment in pmd_install(). */ > smp_wmb(); > - pmd_populate_kernel(&init_mm, pmd, pgtable); > + vmemmap_update_pmd(start, pmd, pgtable); > if (flush) > - flush_tlb_kernel_range(start, start + PMD_SIZE); > + flush_tlb_vmemmap_range(start, start + PMD_SIZE); > } else { > pte_free_kernel(&init_mm, pgtable); > } > @@ -217,7 +248,7 @@ static int vmemmap_remap_range(unsigned long start, unsigned long end, > } while (pgd++, addr = next, addr != end); > > if (walk->remap_pte && !(walk->flags & VMEMMAP_REMAP_NO_TLB_FLUSH)) > - flush_tlb_kernel_range(start, end); > + flush_tlb_vmemmap_range(start, end); > > return 0; > } > @@ -263,15 +294,15 @@ static void vmemmap_remap_pte(pte_t *pte, unsigned long addr, > > /* > * Makes sure that preceding stores to the page contents from > - * vmemmap_remap_free() become visible before the set_pte_at() > - * write. > + * vmemmap_remap_free() become visible before the > + * vmemmap_update_pte() write. > */ > smp_wmb(); > } > > entry = mk_pte(walk->reuse_page, pgprot); > list_add(&page->lru, walk->vmemmap_pages); > - set_pte_at(&init_mm, addr, pte, entry); > + vmemmap_update_pte(addr, pte, entry); > } > > /* > @@ -310,10 +341,10 @@ static void vmemmap_restore_pte(pte_t *pte, unsigned long addr, > > /* > * Makes sure that preceding stores to the page contents become visible > - * before the set_pte_at() write. > + * before the vmemmap_update_pte() write. > */ > smp_wmb(); > - set_pte_at(&init_mm, addr, pte, mk_pte(page, pgprot)); > + vmemmap_update_pte(addr, pte, mk_pte(page, pgprot)); > } > > /** > @@ -576,7 +607,7 @@ long hugetlb_vmemmap_restore_folios(const struct hstate *h, > } > > if (restored) > - flush_tlb_all(); > + flush_tlb_vmemmap_all(); > if (!ret) > ret = restored; > return ret; > @@ -744,7 +775,7 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l > break; > } > > - flush_tlb_all(); > + flush_tlb_vmemmap_all(); > > list_for_each_entry(folio, folio_list, lru) { > int ret = __hugetlb_vmemmap_optimize_folio(h, folio, > @@ -760,7 +791,7 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l > * allowing more vmemmap remaps to occur. > */ > if (ret == -ENOMEM && !list_empty(&vmemmap_pages)) { > - flush_tlb_all(); > + flush_tlb_vmemmap_all(); > free_vmemmap_page_list(&vmemmap_pages); > INIT_LIST_HEAD(&vmemmap_pages); > __hugetlb_vmemmap_optimize_folio(h, folio, > @@ -769,7 +800,7 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l > } > } > > - flush_tlb_all(); > + flush_tlb_vmemmap_all(); > free_vmemmap_page_list(&vmemmap_pages); > } >
On 2023/12/14 15:39, Nanyong Sun wrote: > Add pmd/pte update and tlb flush helper function to update page > table. This refactoring patch is designed to facilitate each > architecture to implement its own special logic in preparation > for the arm64 architecture to follow the necessary break-before-make > sequence when updating page tables. > > Signed-off-by: Nanyong Sun <sunnanyong@huawei.com> > --- > mm/hugetlb_vmemmap.c | 55 ++++++++++++++++++++++++++++++++++---------- > 1 file changed, 43 insertions(+), 12 deletions(-) > > diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c > index 87818ee7f01d..49e8b351def3 100644 > --- a/mm/hugetlb_vmemmap.c > +++ b/mm/hugetlb_vmemmap.c > @@ -45,6 +45,37 @@ struct vmemmap_remap_walk { > unsigned long flags; > }; > > +#ifndef vmemmap_update_pmd > +static inline void vmemmap_update_pmd(unsigned long start, > + pmd_t *pmd, pte_t *pgtable) pgtable -> ptep > +{ > + pmd_populate_kernel(&init_mm, pmd, pgtable); > +} > +#endif > + > +#ifndef vmemmap_update_pte > +static inline void vmemmap_update_pte(unsigned long addr, > + pte_t *pte, pte_t entry) pte -> ptep entry -> pte > +{ > + set_pte_at(&init_mm, addr, pte, entry); > +} > +#endif
On 2023/12/15 11:36, Muchun Song wrote: > > > On 2023/12/14 15:39, Nanyong Sun wrote: >> Add pmd/pte update and tlb flush helper function to update page >> table. This refactoring patch is designed to facilitate each >> architecture to implement its own special logic in preparation >> for the arm64 architecture to follow the necessary break-before-make >> sequence when updating page tables. >> >> Signed-off-by: Nanyong Sun <sunnanyong@huawei.com> >> --- >> mm/hugetlb_vmemmap.c | 55 ++++++++++++++++++++++++++++++++++---------- >> 1 file changed, 43 insertions(+), 12 deletions(-) >> >> diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c >> index 87818ee7f01d..49e8b351def3 100644 >> --- a/mm/hugetlb_vmemmap.c >> +++ b/mm/hugetlb_vmemmap.c >> @@ -45,6 +45,37 @@ struct vmemmap_remap_walk { >> unsigned long flags; >> }; >> +#ifndef vmemmap_update_pmd >> +static inline void vmemmap_update_pmd(unsigned long start, >> + pmd_t *pmd, pte_t *pgtable) >> +{ >> + pmd_populate_kernel(&init_mm, pmd, pgtable); >> +} >> +#endif >> + >> +#ifndef vmemmap_update_pte >> +static inline void vmemmap_update_pte(unsigned long addr, >> + pte_t *pte, pte_t entry) >> +{ >> + set_pte_at(&init_mm, addr, pte, entry); >> +} >> +#endif >> + >> +#ifndef flush_tlb_vmemmap_all >> +static inline void flush_tlb_vmemmap_all(void) >> +{ >> + flush_tlb_all(); >> +} >> +#endif >> + >> +#ifndef flush_tlb_vmemmap_range >> +static inline void flush_tlb_vmemmap_range(unsigned long start, >> + unsigned long end) >> +{ >> + flush_tlb_kernel_range(start, end); >> +} >> +#endif > > I'd like to rename both tlb-flush helpers to vmemmap_flush_tlb_all/range > since other helpers all are prefixed with "vmemmap". It'll be more > consistent for me. > > Otherwise LGTM. Thanks. > > Reviewed-by: Muchun Song <songmuchun@bytedance.com> Hi Muchun， Thank you for your review on this patch set, I'll fix them and send out the v2 version later. > >> + >> static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start, >> bool flush) >> { >> pmd_t __pmd; >> @@ -87,9 +118,9 @@ static int split_vmemmap_huge_pmd(pmd_t *pmd, >> unsigned long start, bool flush) >> /* Make pte visible before pmd. See comment in >> pmd_install(). */ >> smp_wmb(); >> - pmd_populate_kernel(&init_mm, pmd, pgtable); >> + vmemmap_update_pmd(start, pmd, pgtable); >> if (flush) >> - flush_tlb_kernel_range(start, start + PMD_SIZE); >> + flush_tlb_vmemmap_range(start, start + PMD_SIZE); >> } else { >> pte_free_kernel(&init_mm, pgtable); >> } >> @@ -217,7 +248,7 @@ static int vmemmap_remap_range(unsigned long >> start, unsigned long end, >> } while (pgd++, addr = next, addr != end); >> if (walk->remap_pte && !(walk->flags & >> VMEMMAP_REMAP_NO_TLB_FLUSH)) >> - flush_tlb_kernel_range(start, end); >> + flush_tlb_vmemmap_range(start, end); >> return 0; >> } >> @@ -263,15 +294,15 @@ static void vmemmap_remap_pte(pte_t *pte, >> unsigned long addr, >> /* >> * Makes sure that preceding stores to the page contents from >> - * vmemmap_remap_free() become visible before the set_pte_at() >> - * write. >> + * vmemmap_remap_free() become visible before the >> + * vmemmap_update_pte() write. >> */ >> smp_wmb(); >> } >> entry = mk_pte(walk->reuse_page, pgprot); >> list_add(&page->lru, walk->vmemmap_pages); >> - set_pte_at(&init_mm, addr, pte, entry); >> + vmemmap_update_pte(addr, pte, entry); >> } >> /* >> @@ -310,10 +341,10 @@ static void vmemmap_restore_pte(pte_t *pte, >> unsigned long addr, >> /* >> * Makes sure that preceding stores to the page contents become >> visible >> - * before the set_pte_at() write. >> + * before the vmemmap_update_pte() write. >> */ >> smp_wmb(); >> - set_pte_at(&init_mm, addr, pte, mk_pte(page, pgprot)); >> + vmemmap_update_pte(addr, pte, mk_pte(page, pgprot)); >> } >> /** >> @@ -576,7 +607,7 @@ long hugetlb_vmemmap_restore_folios(const struct >> hstate *h, >> } >> if (restored) >> - flush_tlb_all(); >> + flush_tlb_vmemmap_all(); >> if (!ret) >> ret = restored; >> return ret; >> @@ -744,7 +775,7 @@ void hugetlb_vmemmap_optimize_folios(struct >> hstate *h, struct list_head *folio_l >> break; >> } >> - flush_tlb_all(); >> + flush_tlb_vmemmap_all(); >> list_for_each_entry(folio, folio_list, lru) { >> int ret = __hugetlb_vmemmap_optimize_folio(h, folio, >> @@ -760,7 +791,7 @@ void hugetlb_vmemmap_optimize_folios(struct >> hstate *h, struct list_head *folio_l >> * allowing more vmemmap remaps to occur. >> */ >> if (ret == -ENOMEM && !list_empty(&vmemmap_pages)) { >> - flush_tlb_all(); >> + flush_tlb_vmemmap_all(); >> free_vmemmap_page_list(&vmemmap_pages); >> INIT_LIST_HEAD(&vmemmap_pages); >> __hugetlb_vmemmap_optimize_folio(h, folio, >> @@ -769,7 +800,7 @@ void hugetlb_vmemmap_optimize_folios(struct >> hstate *h, struct list_head *folio_l >> } >> } >> - flush_tlb_all(); >> + flush_tlb_vmemmap_all(); >> free_vmemmap_page_list(&vmemmap_pages); >> } > > .
On 2023/12/15 14:16, Kefeng Wang wrote: > > > On 2023/12/14 15:39, Nanyong Sun wrote: >> Add pmd/pte update and tlb flush helper function to update page >> table. This refactoring patch is designed to facilitate each >> architecture to implement its own special logic in preparation >> for the arm64 architecture to follow the necessary break-before-make >> sequence when updating page tables. >> >> Signed-off-by: Nanyong Sun <sunnanyong@huawei.com> >> --- >> mm/hugetlb_vmemmap.c | 55 ++++++++++++++++++++++++++++++++++---------- >> 1 file changed, 43 insertions(+), 12 deletions(-) >> >> diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c >> index 87818ee7f01d..49e8b351def3 100644 >> --- a/mm/hugetlb_vmemmap.c >> +++ b/mm/hugetlb_vmemmap.c >> @@ -45,6 +45,37 @@ struct vmemmap_remap_walk { >> unsigned long flags; >> }; >> +#ifndef vmemmap_update_pmd >> +static inline void vmemmap_update_pmd(unsigned long start, >> + pmd_t *pmd, pte_t *pgtable) > > pgtable -> ptep Hi Kefeng， Thank you for your review on this patch set, I'll fix them and send out the v2 version later. > >> +{ >> + pmd_populate_kernel(&init_mm, pmd, pgtable); >> +} >> +#endif >> + >> +#ifndef vmemmap_update_pte >> +static inline void vmemmap_update_pte(unsigned long addr, >> + pte_t *pte, pte_t entry) > > pte -> ptep > entry -> pte > >> +{ >> + set_pte_at(&init_mm, addr, pte, entry); >> +} >> +#endif > > .
diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c index 87818ee7f01d..49e8b351def3 100644 --- a/mm/hugetlb_vmemmap.c +++ b/mm/hugetlb_vmemmap.c @@ -45,6 +45,37 @@ struct vmemmap_remap_walk { unsigned long flags; }; +#ifndef vmemmap_update_pmd +static inline void vmemmap_update_pmd(unsigned long start, + pmd_t *pmd, pte_t *pgtable) +{ + pmd_populate_kernel(&init_mm, pmd, pgtable); +} +#endif + +#ifndef vmemmap_update_pte +static inline void vmemmap_update_pte(unsigned long addr, + pte_t *pte, pte_t entry) +{ + set_pte_at(&init_mm, addr, pte, entry); +} +#endif + +#ifndef flush_tlb_vmemmap_all +static inline void flush_tlb_vmemmap_all(void) +{ + flush_tlb_all(); +} +#endif + +#ifndef flush_tlb_vmemmap_range +static inline void flush_tlb_vmemmap_range(unsigned long start, + unsigned long end) +{ + flush_tlb_kernel_range(start, end); +} +#endif + static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start, bool flush) { pmd_t __pmd; @@ -87,9 +118,9 @@ static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start, bool flush) /* Make pte visible before pmd. See comment in pmd_install(). */ smp_wmb(); - pmd_populate_kernel(&init_mm, pmd, pgtable); + vmemmap_update_pmd(start, pmd, pgtable); if (flush) - flush_tlb_kernel_range(start, start + PMD_SIZE); + flush_tlb_vmemmap_range(start, start + PMD_SIZE); } else { pte_free_kernel(&init_mm, pgtable); } @@ -217,7 +248,7 @@ static int vmemmap_remap_range(unsigned long start, unsigned long end, } while (pgd++, addr = next, addr != end); if (walk->remap_pte && !(walk->flags & VMEMMAP_REMAP_NO_TLB_FLUSH)) - flush_tlb_kernel_range(start, end); + flush_tlb_vmemmap_range(start, end); return 0; } @@ -263,15 +294,15 @@ static void vmemmap_remap_pte(pte_t *pte, unsigned long addr, /* * Makes sure that preceding stores to the page contents from - * vmemmap_remap_free() become visible before the set_pte_at() - * write. + * vmemmap_remap_free() become visible before the + * vmemmap_update_pte() write. */ smp_wmb(); } entry = mk_pte(walk->reuse_page, pgprot); list_add(&page->lru, walk->vmemmap_pages); - set_pte_at(&init_mm, addr, pte, entry); + vmemmap_update_pte(addr, pte, entry); } /* @@ -310,10 +341,10 @@ static void vmemmap_restore_pte(pte_t *pte, unsigned long addr, /* * Makes sure that preceding stores to the page contents become visible - * before the set_pte_at() write. + * before the vmemmap_update_pte() write. */ smp_wmb(); - set_pte_at(&init_mm, addr, pte, mk_pte(page, pgprot)); + vmemmap_update_pte(addr, pte, mk_pte(page, pgprot)); } /** @@ -576,7 +607,7 @@ long hugetlb_vmemmap_restore_folios(const struct hstate *h, } if (restored) - flush_tlb_all(); + flush_tlb_vmemmap_all(); if (!ret) ret = restored; return ret; @@ -744,7 +775,7 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l break; } - flush_tlb_all(); + flush_tlb_vmemmap_all(); list_for_each_entry(folio, folio_list, lru) { int ret = __hugetlb_vmemmap_optimize_folio(h, folio, @@ -760,7 +791,7 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l * allowing more vmemmap remaps to occur. */ if (ret == -ENOMEM && !list_empty(&vmemmap_pages)) { - flush_tlb_all(); + flush_tlb_vmemmap_all(); free_vmemmap_page_list(&vmemmap_pages); INIT_LIST_HEAD(&vmemmap_pages); __hugetlb_vmemmap_optimize_folio(h, folio, @@ -769,7 +800,7 @@ void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_l } } - flush_tlb_all(); + flush_tlb_vmemmap_all(); free_vmemmap_page_list(&vmemmap_pages); }
Add pmd/pte update and tlb flush helper function to update page table. This refactoring patch is designed to facilitate each architecture to implement its own special logic in preparation for the arm64 architecture to follow the necessary break-before-make sequence when updating page tables. Signed-off-by: Nanyong Sun <sunnanyong@huawei.com> --- mm/hugetlb_vmemmap.c | 55 ++++++++++++++++++++++++++++++++++---------- 1 file changed, 43 insertions(+), 12 deletions(-)