Re: [PATCH] [arm64/tlb] Fix mmu notifiers for range-based invalidates

On Wed, Mar 05, 2025 at 06:48:19PM +0000, Catalin Marinas wrote:

On Tue, Mar 04, 2025 at 12:51:27AM -0800, Piotr Jaroszynski wrote:

quoted

Update the __flush_tlb_range_op macro not to modify its parameters as
these are unexepcted semantics. In practice, this fixes the call to
mmu_notifier_arch_invalidate_secondary_tlbs() in
__flush_tlb_range_nosync() to use the correct range instead of an empty
range with start=end. The empty range was (un)lucky as it results in
taking the invalidate-all path that doesn't cause correctness issues,
but can certainly result in suboptimal perf.

This has been broken since commit 6bbd42e2df8f ("mmu_notifiers: call
invalidate_range() when invalidating TLBs") when the call to the
notifiers was added to __flush_tlb_range(). It predates the addition of
the __flush_tlb_range_op() macro from commit 360839027a6e ("arm64: tlb:
Refactor the core flush algorithm of __flush_tlb_range") that made the
bug hard to spot.

That's the problem with macros.

Yep, that's why I missed it when adding the notifier call. Anyway:

Reviewed-by: Alistair Popple <apopple@nvidia.com>

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>

Will, do you want to take this as a fix? It's only a performance
regression, though you never know how it breaks the callers of the macro
at some point.

quoted

Fixes: 6bbd42e2df8f ("mmu_notifiers: call invalidate_range() when invalidating TLBs")

Signed-off-by: Piotr Jaroszynski <redacted>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Raghavendra Rao Ananta <redacted>
Cc: SeongJae Park <sj@kernel.org>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Nicolin Chen <redacted>
Cc: <redacted>
Cc: <redacted>
Cc: <redacted>
Cc: <redacted>
Cc: <redacted>
---
 arch/arm64/include/asm/tlbflush.h | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/arch/arm64/include/asm/tlbflush.h b/arch/arm64/include/asm/tlbflush.h
index bc94e036a26b..8104aee4f9a0 100644
--- a/arch/arm64/include/asm/tlbflush.h
+++ b/arch/arm64/include/asm/tlbflush.h

@@ -396,33 +396,35 @@ static inline void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch)
 #define __flush_tlb_range_op(op, start, pages, stride,			\
 				asid, tlb_level, tlbi_user, lpa2)	\
 do {									\
+	typeof(start) __flush_start = start;				\
+	typeof(pages) __flush_pages = pages;				\
 	int num = 0;							\
 	int scale = 3;							\
 	int shift = lpa2 ? 16 : PAGE_SHIFT;				\
 	unsigned long addr;						\
 									\
-	while (pages > 0) {						\
+	while (__flush_pages > 0) {					\
 		if (!system_supports_tlb_range() ||			\
-		    pages == 1 ||					\
-		    (lpa2 && start != ALIGN(start, SZ_64K))) {		\
-			addr = __TLBI_VADDR(start, asid);		\
+		    __flush_pages == 1 ||				\
+		    (lpa2 && __flush_start != ALIGN(__flush_start, SZ_64K))) {	\
+			addr = __TLBI_VADDR(__flush_start, asid);	\
 			__tlbi_level(op, addr, tlb_level);		\
 			if (tlbi_user)					\
 				__tlbi_user_level(op, addr, tlb_level);	\
-			start += stride;				\
-			pages -= stride >> PAGE_SHIFT;			\
+			__flush_start += stride;			\
+			__flush_pages -= stride >> PAGE_SHIFT;		\
 			continue;					\
 		}							\
 									\
-		num = __TLBI_RANGE_NUM(pages, scale);			\
+		num = __TLBI_RANGE_NUM(__flush_pages, scale);		\
 		if (num >= 0) {						\
-			addr = __TLBI_VADDR_RANGE(start >> shift, asid, \
+			addr = __TLBI_VADDR_RANGE(__flush_start >> shift, asid, \
 						scale, num, tlb_level);	\
 			__tlbi(r##op, addr);				\
 			if (tlbi_user)					\
 				__tlbi_user(r##op, addr);		\
-			start += __TLBI_RANGE_PAGES(num, scale) << PAGE_SHIFT; \
-			pages -= __TLBI_RANGE_PAGES(num, scale);	\
+			__flush_start += __TLBI_RANGE_PAGES(num, scale) << PAGE_SHIFT; \
+			__flush_pages -= __TLBI_RANGE_PAGES(num, scale);\
 		}							\
 		scale--;						\
 	}								\

base-commit: 99fa936e8e4f117d62f229003c9799686f74cebc
-- 
2.22.1.7.gac84d6e93c.dirty

-- 
Catalin