diff mbox series

[v4] mmu_gather: move tlb flush for VM_PFNMAP/VM_MIXEDMAP vmas into free_pgtables()

Message ID 20250127195321.35779-1-roman.gushchin@linux.dev (mailing list archive)
State New
Headers show
Series [v4] mmu_gather: move tlb flush for VM_PFNMAP/VM_MIXEDMAP vmas into free_pgtables() | expand

Commit Message

Roman Gushchin Jan. 27, 2025, 7:53 p.m. UTC
Commit b67fbebd4cf9 ("mmu_gather: Force tlb-flush VM_PFNMAP vmas")
added a forced tlbflush to tlb_vma_end(), which is required to avoid a
race between munmap() and unmap_mapping_range(). However it added some
overhead to other paths where tlb_vma_end() is used, but vmas are not
removed, e.g. madvise(MADV_DONTNEED).

Fix this by moving the tlb flush out of tlb_end_vma() into
free_pgtables(), somewhat similar to the stable version of the
original commit: e.g. stable commit 895428ee124a ("mm: Force TLB flush
for PFNMAP mappings before unlink_file_vma()").

Note, that if tlb->fullmm is set, no flush is required, as the whole
mm is about to be destroyed.

Signed-off-by: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Jann Horn <jannh@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will@kernel.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Nick Piggin <npiggin@gmail.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: linux-arch@vger.kernel.org
Cc: linux-mm@kvack.org

---

v4:
  - naming/comments update (by Peter Z.)
  - check vma->vma->vm_flags in tlb_free_vma() (by Peter Z.)

v3:
  - added initialization of vma_pfn in __tlb_reset_range() (by Hugh D.)

v2:
  - moved vma_pfn flag handling into tlb.h (by Peter Z.)
  - added comments (by Peter Z.)
  - fixed the vma_pfn flag setting (by Hugh D.)
---
 include/asm-generic/tlb.h | 49 +++++++++++++++++++++++++++++++--------
 mm/memory.c               |  2 ++
 2 files changed, 41 insertions(+), 10 deletions(-)

Comments

Hugh Dickins Jan. 27, 2025, 8:55 p.m. UTC | #1
On Mon, 27 Jan 2025, Roman Gushchin wrote:

> Commit b67fbebd4cf9 ("mmu_gather: Force tlb-flush VM_PFNMAP vmas")
> added a forced tlbflush to tlb_vma_end(), which is required to avoid a
> race between munmap() and unmap_mapping_range(). However it added some
> overhead to other paths where tlb_vma_end() is used, but vmas are not
> removed, e.g. madvise(MADV_DONTNEED).
> 
> Fix this by moving the tlb flush out of tlb_end_vma() into
> free_pgtables(), somewhat similar to the stable version of the
> original commit: e.g. stable commit 895428ee124a ("mm: Force TLB flush
> for PFNMAP mappings before unlink_file_vma()").
> 
> Note, that if tlb->fullmm is set, no flush is required, as the whole
> mm is about to be destroyed.
> 
> Signed-off-by: Roman Gushchin <roman.gushchin@linux.dev>
> Cc: Jann Horn <jannh@google.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Will Deacon <will@kernel.org>
> Cc: "Aneesh Kumar K.V" <aneesh.kumar@kernel.org>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Nick Piggin <npiggin@gmail.com>
> Cc: Hugh Dickins <hughd@google.com>
> Cc: linux-arch@vger.kernel.org
> Cc: linux-mm@kvack.org
> 
> ---
> 
> v4:
>   - naming/comments update (by Peter Z.)
>   - check vma->vma->vm_flags in tlb_free_vma() (by Peter Z.)

Let me just put on record: you were absolutely right not to extend to
this the Ack I gave to v3, this v4 is silly (tlb_free_vma() and its
multiple calls, necessary only because of the unnecessary extra test);
but I don't see it as doing any actual damage, so I'll stop short of
NAKking it.

Hugh
diff mbox series

Patch

diff --git a/include/asm-generic/tlb.h b/include/asm-generic/tlb.h
index e402aef79c93..dd673ec59893 100644
--- a/include/asm-generic/tlb.h
+++ b/include/asm-generic/tlb.h
@@ -58,6 +58,11 @@ 
  *    Defaults to flushing at tlb_end_vma() to reset the range; helps when
  *    there's large holes between the VMAs.
  *
+ *  - tlb_free_vma()
+ *
+ *    tlb_free_vma() marks the start of unlinking the vma and freeing
+ *    page-tables.
+ *
  *  - tlb_remove_table()
  *
  *    tlb_remove_table() is the basic primitive to free page-table directories
@@ -400,7 +405,10 @@  static inline void __tlb_reset_range(struct mmu_gather *tlb)
 	 * Do not reset mmu_gather::vma_* fields here, we do not
 	 * call into tlb_start_vma() again to set them if there is an
 	 * intermediate flush.
+	 *
+	 * Except for vma_pfn, that only cares if there's pending TLBI.
 	 */
+	tlb->vma_pfn = 0;
 }
 
 #ifdef CONFIG_MMU_GATHER_NO_RANGE
@@ -465,7 +473,12 @@  tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma)
 	 */
 	tlb->vma_huge = is_vm_hugetlb_page(vma);
 	tlb->vma_exec = !!(vma->vm_flags & VM_EXEC);
-	tlb->vma_pfn  = !!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP));
+
+	/*
+	 * Track if there's at least one VM_PFNMAP/VM_MIXEDMAP vma
+	 * in the tracked range, see tlb_free_vma().
+	 */
+	tlb->vma_pfn |= !!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP));
 }
 
 static inline void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
@@ -564,23 +577,39 @@  static inline void tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *
 }
 
 static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
+{
+	if (tlb->fullmm || IS_ENABLED(CONFIG_MMU_GATHER_MERGE_VMAS))
+		return;
+
+	/*
+	 * Do a TLB flush and reset the range at VMA boundaries; this avoids
+	 * the ranges growing with the unused space between consecutive VMAs,
+	 * but also the mmu_gather::vma_* flags from tlb_start_vma() rely on
+	 * this.
+	 */
+	tlb_flush_mmu_tlbonly(tlb);
+}
+
+static inline void tlb_free_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
 {
 	if (tlb->fullmm)
 		return;
 
 	/*
 	 * VM_PFNMAP is more fragile because the core mm will not track the
-	 * page mapcount -- there might not be page-frames for these PFNs after
-	 * all. Force flush TLBs for such ranges to avoid munmap() vs
-	 * unmap_mapping_range() races.
+	 * page mapcount -- there might not be page-frames for these PFNs
+	 * after all.
+	 *
+	 * Specifically() there is a race between munmap() and
+	 * unmap_mapping_range(), where munmap() will unlink the VMA, such
+	 * that unmap_mapping_range() will no longer observe the VMA and
+	 * no-op, without observing the TLBI, returning prematurely.
+	 *
+	 * So if we're about to unlink such a VMA, and we have pending
+	 * TLBI for such a vma, flush things now.
 	 */
-	if (tlb->vma_pfn || !IS_ENABLED(CONFIG_MMU_GATHER_MERGE_VMAS)) {
-		/*
-		 * Do a TLB flush and reset the range at VMA boundaries; this avoids
-		 * the ranges growing with the unused space between consecutive VMAs.
-		 */
+	if ((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) && tlb->vma_pfn)
 		tlb_flush_mmu_tlbonly(tlb);
-	}
 }
 
 /*
diff --git a/mm/memory.c b/mm/memory.c
index 539c0f7c6d54..4ea5e286c68f 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -378,6 +378,7 @@  void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas,
 		if (unlikely(xa_is_zero(next)))
 			next = NULL;
 
+		tlb_free_vma(tlb, vma);
 		/*
 		 * Hide vma from rmap and truncate_pagecache before freeing
 		 * pgtables
@@ -403,6 +404,7 @@  void free_pgtables(struct mmu_gather *tlb, struct ma_state *mas,
 				next = mas_find(mas, ceiling - 1);
 				if (unlikely(xa_is_zero(next)))
 					next = NULL;
+				tlb_free_vma(tlb, vma);
 				if (mm_wr_locked)
 					vma_start_write(vma);
 				unlink_anon_vmas(vma);