| Message ID | 20220705142619.4135905-5-catalin.marinas@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm64: mte: Fix racing on MTE tag initialisation | expand |
On Tue, Jul 5, 2022 at 7:26 AM Catalin Marinas <catalin.marinas@arm.com> wrote: > > Initialising the tags and setting PG_mte_tagged flag for a page can race > between multiple set_pte_at() on shared pages or setting the stage 2 pte > via user_mem_abort(). Introduce a new PG_mte_lock flag as PG_arch_3 and > set it before attempting page initialisation. Given that PG_mte_tagged > is never cleared for a page, consider setting this flag to mean page > unlocked and wait on this bit with acquire semantics if the page is > locked: > > - try_page_mte_tagging() - lock the page for tagging, return true if it > can be tagged, false if already tagged. No acquire semantics if it > returns true (PG_mte_tagged not set) as there is no serialisation with > a previous set_page_mte_tagged(). > > - set_page_mte_tagged() - set PG_mte_tagged with release semantics. > > The two-bit locking is based on Peter Colingbourne's idea. nit: Collingbourne (two l's). > > Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> > Cc: Will Deacon <will@kernel.org> > Cc: Marc Zyngier <maz@kernel.org> > Cc: Steven Price <steven.price@arm.com> > Cc: Peter Collingbourne <pcc@google.com> > --- > arch/arm64/include/asm/mte.h | 32 ++++++++++++++++++++++++++++++++ > arch/arm64/include/asm/pgtable.h | 1 + > arch/arm64/kernel/cpufeature.c | 2 +- > arch/arm64/kernel/mte.c | 7 +++++-- > arch/arm64/kvm/guest.c | 16 ++++++++++------ > arch/arm64/kvm/mmu.c | 2 +- > arch/arm64/mm/copypage.c | 2 ++ > arch/arm64/mm/fault.c | 2 ++ > arch/arm64/mm/mteswap.c | 3 +++ > 9 files changed, 57 insertions(+), 10 deletions(-) > > diff --git a/arch/arm64/include/asm/mte.h b/arch/arm64/include/asm/mte.h > index c69218c56980..29712fc9df8c 100644 > --- a/arch/arm64/include/asm/mte.h > +++ b/arch/arm64/include/asm/mte.h > @@ -36,6 +36,8 @@ void mte_free_tag_storage(char *storage); > > /* track which pages have valid allocation tags */ > #define PG_mte_tagged PG_arch_2 > +/* simple lock to avoid multiple threads tagging the same page */ > +#define PG_mte_lock PG_arch_3 > > static inline void set_page_mte_tagged(struct page *page) > { > @@ -60,6 +62,32 @@ static inline bool page_mte_tagged(struct page *page) > return ret; > } > > +/* > + * Lock the page for tagging and return 'true' if the page can be tagged, > + * 'false' if already tagged. PG_mte_tagged is never cleared and therefore the > + * locking only happens once for page initialisation. > + * > + * The page MTE lock state: > + * > + * Locked: PG_mte_lock && !PG_mte_tagged > + * Unlocked: !PG_mte_lock || PG_mte_tagged > + * > + * Acquire semantics only if the page is tagged (returning 'false'). > + */ > +static inline bool try_page_mte_tagging(struct page *page) > +{ > + if (!test_and_set_bit(PG_mte_lock, &page->flags)) > + return !page_mte_tagged(page); Since all callers of set_page_mte_tagged() are now dominated by a call to try_page_mte_tagging() and PG_mte_lock is never cleared I think we can't end up in the state where !PG_mte_lock && PG_mte_tagged. So I think this can be simplified to "return true;". I can still boot VMs with MTE enabled after making my suggested change. > + > + /* > + * The tags are being initialised, wait for the PG_mte_tagged flag to I think at this point the tags are either being initialized or have already been initialized, so the comment isn't quite right. Peter > + * be set. > + */ > + smp_cond_load_acquire(&page->flags, VAL & (1UL << PG_mte_tagged)); > + > + return false; > +} > + > void mte_zero_clear_page_tags(void *addr); > void mte_sync_tags(pte_t old_pte, pte_t pte); > void mte_copy_page_tags(void *kto, const void *kfrom); > @@ -84,6 +112,10 @@ static inline bool page_mte_tagged(struct page *page) > { > return false; > } > +static inline bool try_page_mte_tagging(struct page *page) > +{ > + return false; > +} > static inline void mte_zero_clear_page_tags(void *addr) > { > } > diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h > index 08823669db0a..ce2dc72f64f4 100644 > --- a/arch/arm64/include/asm/pgtable.h > +++ b/arch/arm64/include/asm/pgtable.h > @@ -1033,6 +1033,7 @@ static inline void arch_swap_invalidate_area(int type) > #define __HAVE_ARCH_SWAP_RESTORE > static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio) > { > + /* mte_restore_tags() takes the PG_mte_lock */ > if (system_supports_mte() && mte_restore_tags(entry, &folio->page)) > set_page_mte_tagged(&folio->page); > } > diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c > index 4478e5774580..c07dd7916517 100644 > --- a/arch/arm64/kernel/cpufeature.c > +++ b/arch/arm64/kernel/cpufeature.c > @@ -1964,7 +1964,7 @@ static void cpu_enable_mte(struct arm64_cpu_capabilities const *cap) > * Clear the tags in the zero page. This needs to be done via the > * linear map which has the Tagged attribute. > */ > - if (!page_mte_tagged(ZERO_PAGE(0))) { > + if (try_page_mte_tagging(ZERO_PAGE(0))) { > mte_clear_page_tags(lm_alias(empty_zero_page)); > set_page_mte_tagged(ZERO_PAGE(0)); > } > diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c > index 67e82ce4c285..54d284a1e0a7 100644 > --- a/arch/arm64/kernel/mte.c > +++ b/arch/arm64/kernel/mte.c > @@ -41,6 +41,7 @@ static void mte_sync_page_tags(struct page *page, pte_t old_pte, > if (check_swap && is_swap_pte(old_pte)) { > swp_entry_t entry = pte_to_swp_entry(old_pte); > > + /* mte_restore_tags() takes the PG_mte_lock */ > if (!non_swap_entry(entry) && mte_restore_tags(entry, page)) { > set_page_mte_tagged(page); > return; > @@ -59,8 +60,10 @@ static void mte_sync_page_tags(struct page *page, pte_t old_pte, > * the new page->flags are visible before the tags were updated. > */ > smp_wmb(); > - mte_clear_page_tags(page_address(page)); > - set_page_mte_tagged(page); > + if (try_page_mte_tagging(page)) { > + mte_clear_page_tags(page_address(page)); > + set_page_mte_tagged(page); > + } > } > > void mte_sync_tags(pte_t old_pte, pte_t pte) > diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c > index 3b04e69006b4..059b38e7a9e8 100644 > --- a/arch/arm64/kvm/guest.c > +++ b/arch/arm64/kvm/guest.c > @@ -1067,15 +1067,19 @@ long kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm, > clear_user(tags, MTE_GRANULES_PER_PAGE); > kvm_release_pfn_clean(pfn); > } else { > + /* > + * Only locking to serialise with a concurrent > + * set_pte_at() in the VMM but still overriding the > + * tags, hence ignoring the return value. > + */ > + try_page_mte_tagging(page); > num_tags = mte_copy_tags_from_user(maddr, tags, > MTE_GRANULES_PER_PAGE); > > - /* > - * Set the flag after checking the write > - * completed fully > - */ > - if (num_tags == MTE_GRANULES_PER_PAGE) > - set_page_mte_tagged(page); > + /* uaccess failed, don't leave stale tags */ > + if (num_tags != MTE_GRANULES_PER_PAGE) > + mte_clear_page_tags(page); > + set_page_mte_tagged(page); > > kvm_release_pfn_dirty(pfn); > } > diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c > index 35850f17ae08..fdd46089f260 100644 > --- a/arch/arm64/kvm/mmu.c > +++ b/arch/arm64/kvm/mmu.c > @@ -1066,7 +1066,7 @@ static void sanitise_mte_tags(struct kvm *kvm, kvm_pfn_t pfn, > return; > > for (i = 0; i < nr_pages; i++, page++) { > - if (!page_mte_tagged(page)) { > + if (try_page_mte_tagging(page)) { > mte_clear_page_tags(page_address(page)); > set_page_mte_tagged(page); > } > diff --git a/arch/arm64/mm/copypage.c b/arch/arm64/mm/copypage.c > index f36d796f1bce..9c73bc020894 100644 > --- a/arch/arm64/mm/copypage.c > +++ b/arch/arm64/mm/copypage.c > @@ -31,6 +31,8 @@ void copy_highpage(struct page *to, struct page *from) > * the new page->flags are visible before the tags were updated. > */ > smp_wmb(); > + /* It's a new page, shouldn't have been tagged yet */ > + WARN_ON_ONCE(!try_page_mte_tagging(to)); > mte_copy_page_tags(kto, kfrom); > set_page_mte_tagged(to); > } > diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c > index 147fe28d3fbe..bd28d6bd9286 100644 > --- a/arch/arm64/mm/fault.c > +++ b/arch/arm64/mm/fault.c > @@ -926,6 +926,8 @@ struct page *alloc_zeroed_user_highpage_movable(struct vm_area_struct *vma, > > void tag_clear_highpage(struct page *page) > { > + /* Newly allocated page, shouldn't have been tagged yet */ > + WARN_ON_ONCE(!try_page_mte_tagging(page)); > mte_zero_clear_page_tags(page_address(page)); > page_kasan_tag_reset(page); > set_page_mte_tagged(page); > diff --git a/arch/arm64/mm/mteswap.c b/arch/arm64/mm/mteswap.c > index 9d3a8cf388fc..aec76a4423e9 100644 > --- a/arch/arm64/mm/mteswap.c > +++ b/arch/arm64/mm/mteswap.c > @@ -62,6 +62,9 @@ bool mte_restore_tags(swp_entry_t entry, struct page *page) > * the new page->flags are visible before the tags were updated. > */ > smp_wmb(); > + /* racing tag restoring? */ > + if (!try_page_mte_tagging(page)) > + return false; > mte_restore_page_tags(page_address(page), tags); > > return true;
On Fri, Jul 08, 2022 at 04:11:59PM -0700, Peter Collingbourne wrote: > On Tue, Jul 5, 2022 at 7:26 AM Catalin Marinas <catalin.marinas@arm.com> wrote: > > @@ -60,6 +62,32 @@ static inline bool page_mte_tagged(struct page *page) > > return ret; > > } > > > > +/* > > + * Lock the page for tagging and return 'true' if the page can be tagged, > > + * 'false' if already tagged. PG_mte_tagged is never cleared and therefore the > > + * locking only happens once for page initialisation. > > + * > > + * The page MTE lock state: > > + * > > + * Locked: PG_mte_lock && !PG_mte_tagged > > + * Unlocked: !PG_mte_lock || PG_mte_tagged > > + * > > + * Acquire semantics only if the page is tagged (returning 'false'). > > + */ > > +static inline bool try_page_mte_tagging(struct page *page) > > +{ > > + if (!test_and_set_bit(PG_mte_lock, &page->flags)) > > + return !page_mte_tagged(page); > > Since all callers of set_page_mte_tagged() are now dominated by a call > to try_page_mte_tagging() and PG_mte_lock is never cleared I think we > can't end up in the state where !PG_mte_lock && PG_mte_tagged. So I > think this can be simplified to "return true;". I can still boot VMs > with MTE enabled after making my suggested change. Correct. Not sure why I complicated this since the "Unlocked" description above states that try_page_mte_tagging() should return "unlocked" if !PG_mte_lock, so no need for the PG_mte_tagged check. > > + > > + /* > > + * The tags are being initialised, wait for the PG_mte_tagged flag to > > I think at this point the tags are either being initialized or have > already been initialized, so the comment isn't quite right. Yeah, they may have been initialised already by the time we got here and smp_cond_load_acquire() would just return immediately. I was too lazy to write all the use-cases here.
diff --git a/arch/arm64/include/asm/mte.h b/arch/arm64/include/asm/mte.h index c69218c56980..29712fc9df8c 100644 --- a/arch/arm64/include/asm/mte.h +++ b/arch/arm64/include/asm/mte.h @@ -36,6 +36,8 @@ void mte_free_tag_storage(char *storage); /* track which pages have valid allocation tags */ #define PG_mte_tagged PG_arch_2 +/* simple lock to avoid multiple threads tagging the same page */ +#define PG_mte_lock PG_arch_3 static inline void set_page_mte_tagged(struct page *page) { @@ -60,6 +62,32 @@ static inline bool page_mte_tagged(struct page *page) return ret; } +/* + * Lock the page for tagging and return 'true' if the page can be tagged, + * 'false' if already tagged. PG_mte_tagged is never cleared and therefore the + * locking only happens once for page initialisation. + * + * The page MTE lock state: + * + * Locked: PG_mte_lock && !PG_mte_tagged + * Unlocked: !PG_mte_lock || PG_mte_tagged + * + * Acquire semantics only if the page is tagged (returning 'false'). + */ +static inline bool try_page_mte_tagging(struct page *page) +{ + if (!test_and_set_bit(PG_mte_lock, &page->flags)) + return !page_mte_tagged(page); + + /* + * The tags are being initialised, wait for the PG_mte_tagged flag to + * be set. + */ + smp_cond_load_acquire(&page->flags, VAL & (1UL << PG_mte_tagged)); + + return false; +} + void mte_zero_clear_page_tags(void *addr); void mte_sync_tags(pte_t old_pte, pte_t pte); void mte_copy_page_tags(void *kto, const void *kfrom); @@ -84,6 +112,10 @@ static inline bool page_mte_tagged(struct page *page) { return false; } +static inline bool try_page_mte_tagging(struct page *page) +{ + return false; +} static inline void mte_zero_clear_page_tags(void *addr) { } diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h index 08823669db0a..ce2dc72f64f4 100644 --- a/arch/arm64/include/asm/pgtable.h +++ b/arch/arm64/include/asm/pgtable.h @@ -1033,6 +1033,7 @@ static inline void arch_swap_invalidate_area(int type) #define __HAVE_ARCH_SWAP_RESTORE static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio) { + /* mte_restore_tags() takes the PG_mte_lock */ if (system_supports_mte() && mte_restore_tags(entry, &folio->page)) set_page_mte_tagged(&folio->page); } diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c index 4478e5774580..c07dd7916517 100644 --- a/arch/arm64/kernel/cpufeature.c +++ b/arch/arm64/kernel/cpufeature.c @@ -1964,7 +1964,7 @@ static void cpu_enable_mte(struct arm64_cpu_capabilities const *cap) * Clear the tags in the zero page. This needs to be done via the * linear map which has the Tagged attribute. */ - if (!page_mte_tagged(ZERO_PAGE(0))) { + if (try_page_mte_tagging(ZERO_PAGE(0))) { mte_clear_page_tags(lm_alias(empty_zero_page)); set_page_mte_tagged(ZERO_PAGE(0)); } diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c index 67e82ce4c285..54d284a1e0a7 100644 --- a/arch/arm64/kernel/mte.c +++ b/arch/arm64/kernel/mte.c @@ -41,6 +41,7 @@ static void mte_sync_page_tags(struct page *page, pte_t old_pte, if (check_swap && is_swap_pte(old_pte)) { swp_entry_t entry = pte_to_swp_entry(old_pte); + /* mte_restore_tags() takes the PG_mte_lock */ if (!non_swap_entry(entry) && mte_restore_tags(entry, page)) { set_page_mte_tagged(page); return; @@ -59,8 +60,10 @@ static void mte_sync_page_tags(struct page *page, pte_t old_pte, * the new page->flags are visible before the tags were updated. */ smp_wmb(); - mte_clear_page_tags(page_address(page)); - set_page_mte_tagged(page); + if (try_page_mte_tagging(page)) { + mte_clear_page_tags(page_address(page)); + set_page_mte_tagged(page); + } } void mte_sync_tags(pte_t old_pte, pte_t pte) diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c index 3b04e69006b4..059b38e7a9e8 100644 --- a/arch/arm64/kvm/guest.c +++ b/arch/arm64/kvm/guest.c @@ -1067,15 +1067,19 @@ long kvm_vm_ioctl_mte_copy_tags(struct kvm *kvm, clear_user(tags, MTE_GRANULES_PER_PAGE); kvm_release_pfn_clean(pfn); } else { + /* + * Only locking to serialise with a concurrent + * set_pte_at() in the VMM but still overriding the + * tags, hence ignoring the return value. + */ + try_page_mte_tagging(page); num_tags = mte_copy_tags_from_user(maddr, tags, MTE_GRANULES_PER_PAGE); - /* - * Set the flag after checking the write - * completed fully - */ - if (num_tags == MTE_GRANULES_PER_PAGE) - set_page_mte_tagged(page); + /* uaccess failed, don't leave stale tags */ + if (num_tags != MTE_GRANULES_PER_PAGE) + mte_clear_page_tags(page); + set_page_mte_tagged(page); kvm_release_pfn_dirty(pfn); } diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c index 35850f17ae08..fdd46089f260 100644 --- a/arch/arm64/kvm/mmu.c +++ b/arch/arm64/kvm/mmu.c @@ -1066,7 +1066,7 @@ static void sanitise_mte_tags(struct kvm *kvm, kvm_pfn_t pfn, return; for (i = 0; i < nr_pages; i++, page++) { - if (!page_mte_tagged(page)) { + if (try_page_mte_tagging(page)) { mte_clear_page_tags(page_address(page)); set_page_mte_tagged(page); } diff --git a/arch/arm64/mm/copypage.c b/arch/arm64/mm/copypage.c index f36d796f1bce..9c73bc020894 100644 --- a/arch/arm64/mm/copypage.c +++ b/arch/arm64/mm/copypage.c @@ -31,6 +31,8 @@ void copy_highpage(struct page *to, struct page *from) * the new page->flags are visible before the tags were updated. */ smp_wmb(); + /* It's a new page, shouldn't have been tagged yet */ + WARN_ON_ONCE(!try_page_mte_tagging(to)); mte_copy_page_tags(kto, kfrom); set_page_mte_tagged(to); } diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c index 147fe28d3fbe..bd28d6bd9286 100644 --- a/arch/arm64/mm/fault.c +++ b/arch/arm64/mm/fault.c @@ -926,6 +926,8 @@ struct page *alloc_zeroed_user_highpage_movable(struct vm_area_struct *vma, void tag_clear_highpage(struct page *page) { + /* Newly allocated page, shouldn't have been tagged yet */ + WARN_ON_ONCE(!try_page_mte_tagging(page)); mte_zero_clear_page_tags(page_address(page)); page_kasan_tag_reset(page); set_page_mte_tagged(page); diff --git a/arch/arm64/mm/mteswap.c b/arch/arm64/mm/mteswap.c index 9d3a8cf388fc..aec76a4423e9 100644 --- a/arch/arm64/mm/mteswap.c +++ b/arch/arm64/mm/mteswap.c @@ -62,6 +62,9 @@ bool mte_restore_tags(swp_entry_t entry, struct page *page) * the new page->flags are visible before the tags were updated. */ smp_wmb(); + /* racing tag restoring? */ + if (!try_page_mte_tagging(page)) + return false; mte_restore_page_tags(page_address(page), tags); return true;
Initialising the tags and setting PG_mte_tagged flag for a page can race between multiple set_pte_at() on shared pages or setting the stage 2 pte via user_mem_abort(). Introduce a new PG_mte_lock flag as PG_arch_3 and set it before attempting page initialisation. Given that PG_mte_tagged is never cleared for a page, consider setting this flag to mean page unlocked and wait on this bit with acquire semantics if the page is locked: - try_page_mte_tagging() - lock the page for tagging, return true if it can be tagged, false if already tagged. No acquire semantics if it returns true (PG_mte_tagged not set) as there is no serialisation with a previous set_page_mte_tagged(). - set_page_mte_tagged() - set PG_mte_tagged with release semantics. The two-bit locking is based on Peter Colingbourne's idea. Signed-off-by: Catalin Marinas <catalin.marinas@arm.com> Cc: Will Deacon <will@kernel.org> Cc: Marc Zyngier <maz@kernel.org> Cc: Steven Price <steven.price@arm.com> Cc: Peter Collingbourne <pcc@google.com> --- arch/arm64/include/asm/mte.h | 32 ++++++++++++++++++++++++++++++++ arch/arm64/include/asm/pgtable.h | 1 + arch/arm64/kernel/cpufeature.c | 2 +- arch/arm64/kernel/mte.c | 7 +++++-- arch/arm64/kvm/guest.c | 16 ++++++++++------ arch/arm64/kvm/mmu.c | 2 +- arch/arm64/mm/copypage.c | 2 ++ arch/arm64/mm/fault.c | 2 ++ arch/arm64/mm/mteswap.c | 3 +++ 9 files changed, 57 insertions(+), 10 deletions(-)