diff mbox series

[RFC,14/14] mm: conditional check of pfn in pte_flush_type

Message ID 20220718120212.3180-15-namit@vmware.com (mailing list archive)
State New
Headers show
Series [RFC,01/14] userfaultfd: set dirty and young on writeprotect | expand

Commit Message

Nadav Amit July 18, 2022, 12:02 p.m. UTC 
From: Nadav Amit <namit@vmware.com>

Checking whether PFNs in two PTEs are the same takes surprisingly large
number of instructions. Yet in fact, in most cases the caller to
pte_flush_type() already knows if the PFN was changed. For instance,
mprotect() does not change the PFN, but only modifies the protection
flags.

Add argument to pte_flush_type() to indicate whether the PFN should be
checked. Keep checking it in mm-debug to see if some caller was wrong to
assume the PFN is the same.

Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Nick Piggin <npiggin@gmail.com>
Signed-off-by: Nadav Amit <namit@vmware.com>
---
 arch/x86/include/asm/tlbflush.h | 14 ++++++++++----
 include/asm-generic/tlb.h       |  6 ++++--
 mm/huge_memory.c                |  2 +-
 mm/mprotect.c                   |  2 +-
 mm/rmap.c                       |  2 +-
 5 files changed, 17 insertions(+), 9 deletions(-)
diff mbox series

Patch

diff --git a/arch/x86/include/asm/tlbflush.h b/arch/x86/include/asm/tlbflush.h
index 58c95e36b098..50349861fdc9 100644
--- a/arch/x86/include/asm/tlbflush.h
+++ b/arch/x86/include/asm/tlbflush.h
@@ -340,14 +340,17 @@  static inline enum pte_flush_type pte_flags_flush_type(unsigned long oldflags,
  * whether a strict or relaxed TLB flush is need. It should only be used on
  * userspace PTEs.
  */
-static inline enum pte_flush_type pte_flush_type(pte_t oldpte, pte_t newpte)
+static inline enum pte_flush_type pte_flush_type(pte_t oldpte, pte_t newpte,
+						   bool check_pfn)
 {
 	/* !PRESENT -> * ; no need for flush */
 	if (!(pte_flags(oldpte) & _PAGE_PRESENT))
 		return PTE_FLUSH_NONE;
 
 	/* PFN changed ; needs flush */
-	if (pte_pfn(oldpte) != pte_pfn(newpte))
+	if (!check_pfn)
+		VM_BUG_ON(pte_pfn(oldpte) != pte_pfn(newpte));
+	else if (pte_pfn(oldpte) != pte_pfn(newpte))
 		return PTE_FLUSH_STRICT;
 
 	/*
@@ -363,14 +366,17 @@  static inline enum pte_flush_type pte_flush_type(pte_t oldpte, pte_t newpte)
  * huge_pmd_flush_type() checks whether permissions were demoted and require a
  * flush. It should only be used for userspace huge PMDs.
  */
-static inline enum pte_flush_type huge_pmd_flush_type(pmd_t oldpmd, pmd_t newpmd)
+static inline enum pte_flush_type huge_pmd_flush_type(pmd_t oldpmd, pmd_t newpmd,
+						      bool check_pfn)
 {
 	/* !PRESENT -> * ; no need for flush */
 	if (!(pmd_flags(oldpmd) & _PAGE_PRESENT))
 		return PTE_FLUSH_NONE;
 
 	/* PFN changed ; needs flush */
-	if (pmd_pfn(oldpmd) != pmd_pfn(newpmd))
+	if (!check_pfn)
+		VM_BUG_ON(pmd_pfn(oldpmd) != pmd_pfn(newpmd));
+	else if (pmd_pfn(oldpmd) != pmd_pfn(newpmd))
 		return PTE_FLUSH_STRICT;
 
 	/*
diff --git a/include/asm-generic/tlb.h b/include/asm-generic/tlb.h
index 07b3eb8caf63..aee9da6cc5d5 100644
--- a/include/asm-generic/tlb.h
+++ b/include/asm-generic/tlb.h
@@ -677,14 +677,16 @@  static inline void tlb_flush_p4d_range(struct mmu_gather *tlb,
 #endif
 
 #ifndef pte_flush_type
-static inline struct pte_flush_type pte_flush_type(pte_t oldpte, pte_t newpte)
+static inline struct pte_flush_type pte_flush_type(pte_t oldpte, pte_t newpte,
+						   bool check_pfn)
 {
 	return PTE_FLUSH_STRICT;
 }
 #endif
 
 #ifndef huge_pmd_flush_type
-static inline bool huge_pmd_flush_type(pmd_t oldpmd, pmd_t newpmd)
+static inline bool huge_pmd_flush_type(pmd_t oldpmd, pmd_t newpmd,
+				       bool check_pfn)
 {
 	return PTE_FLUSH_STRICT;
 }
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index b32b7da0f6f7..92a7b3ca317f 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1818,7 +1818,7 @@  int change_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 
 	flush_type = PTE_FLUSH_STRICT;
 	if (!tlb->strict)
-		flush_type = huge_pmd_flush_type(oldpmd, entry);
+		flush_type = huge_pmd_flush_type(oldpmd, entry, false);
 	if (flush_type != PTE_FLUSH_NONE)
 		tlb_flush_pmd_range(tlb, addr, HPAGE_PMD_SIZE,
 				    flush_type == PTE_FLUSH_STRICT);
diff --git a/mm/mprotect.c b/mm/mprotect.c
index cf775f6c8c08..78081d7f4edf 100644
--- a/mm/mprotect.c
+++ b/mm/mprotect.c
@@ -204,7 +204,7 @@  static unsigned long change_pte_range(struct mmu_gather *tlb,
 
 			flush_type = PTE_FLUSH_STRICT;
 			if (!tlb->strict)
-				flush_type = pte_flush_type(oldpte, ptent);
+				flush_type = pte_flush_type(oldpte, ptent, false);
 			if (flush_type != PTE_FLUSH_NONE)
 				tlb_flush_pte_range(tlb, addr, PAGE_SIZE,
 						flush_type == PTE_FLUSH_STRICT);
diff --git a/mm/rmap.c b/mm/rmap.c
index 62f4b2a4f067..63261619b607 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -974,7 +974,7 @@  static int page_vma_mkclean_one(struct page_vma_mapped_walk *pvmw)
 			entry = pte_wrprotect(oldpte);
 			entry = pte_mkclean(entry);
 
-			if (pte_flush_type(oldpte, entry) != PTE_FLUSH_NONE ||
+			if (pte_flush_type(oldpte, entry, false) != PTE_FLUSH_NONE ||
 			    mm_tlb_flush_pending(vma->vm_mm))
 				flush_tlb_page(vma, address);