From: Evgenii Stepanov <eugenis@google.com> To: Andrey Ryabinin <ryabinin.a.a@gmail.com>, Alexander Potapenko <glider@google.com>, Andrey Konovalov <andreyknvl@gmail.com>, Dmitry Vyukov <dvyukov@google.com>, Catalin Marinas <catalin.marinas@arm.com>, Will Deacon <will@kernel.org>, Steven Price <steven.price@arm.com>, Peter Collingbourne <pcc@google.com>, Evgenii Stepanov <eugenis@google.com>, kasan-dev@googlegroups.com, linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org Subject: [PATCH v5] kasan: speed up mte_set_mem_tag_range Date: Thu, 20 May 2021 18:00:23 -0700 [thread overview] Message-ID: <20210521010023.3244784-1-eugenis@google.com> (raw) Use DC GVA / DC GZVA to speed up KASan memory tagging in HW tags mode. The first cacheline is always tagged using STG/STZG even if the address is cacheline-aligned, as benchmarks show it is faster than a conditional branch. Signed-off-by: Evgenii Stepanov <eugenis@google.com> Co-developed-by: Peter Collingbourne <pcc@google.com> Signed-off-by: Peter Collingbourne <pcc@google.com> Reviewed-by: Catalin Marinas <catalin.marinas@arm.com> --- Changelog since v1: - Added Co-developed-by. Changelog since v2: - Added Signed-off-by. Changelog since v3: - Move the implementation back to C with a bit of inline asm. Changelog since v3: - Fixed coding style issues. arch/arm64/include/asm/mte-kasan.h | 93 +++++++++++++++++++++--------- 1 file changed, 67 insertions(+), 26 deletions(-) diff --git a/arch/arm64/include/asm/mte-kasan.h b/arch/arm64/include/asm/mte-kasan.h index ddd4d17cf9a0..d952352bd008 100644 --- a/arch/arm64/include/asm/mte-kasan.h +++ b/arch/arm64/include/asm/mte-kasan.h @@ -48,43 +48,84 @@ static inline u8 mte_get_random_tag(void) return mte_get_ptr_tag(addr); } +static inline u64 __stg_post(u64 p) +{ + asm volatile(__MTE_PREAMBLE "stg %0, [%0], #16" + : "+r"(p) + : + : "memory"); + return p; +} + +static inline u64 __stzg_post(u64 p) +{ + asm volatile(__MTE_PREAMBLE "stzg %0, [%0], #16" + : "+r"(p) + : + : "memory"); + return p; +} + +static inline void __dc_gva(u64 p) +{ + asm volatile(__MTE_PREAMBLE "dc gva, %0" : : "r"(p) : "memory"); +} + +static inline void __dc_gzva(u64 p) +{ + asm volatile(__MTE_PREAMBLE "dc gzva, %0" : : "r"(p) : "memory"); +} + /* * Assign allocation tags for a region of memory based on the pointer tag. * Note: The address must be non-NULL and MTE_GRANULE_SIZE aligned and - * size must be non-zero and MTE_GRANULE_SIZE aligned. + * size must be MTE_GRANULE_SIZE aligned. */ -static inline void mte_set_mem_tag_range(void *addr, size_t size, - u8 tag, bool init) +static inline void mte_set_mem_tag_range(void *addr, size_t size, u8 tag, + bool init) { - u64 curr, end; + u64 curr, mask, dczid_bs, end1, end2, end3; - if (!size) - return; + /* Read DC G(Z)VA block size from the system register. */ + dczid_bs = 4ul << (read_cpuid(DCZID_EL0) & 0xf); curr = (u64)__tag_set(addr, tag); - end = curr + size; + mask = dczid_bs - 1; + /* STG/STZG up to the end of the first block. */ + end1 = curr | mask; + end3 = curr + size; + /* DC GVA / GZVA in [end1, end2) */ + end2 = end3 & ~mask; /* - * 'asm volatile' is required to prevent the compiler to move - * the statement outside of the loop. + * The following code uses STG on the first DC GVA block even if the + * start address is aligned - it appears to be faster than an alignment + * check + conditional branch. Also, if the range size is at least 2 DC + * GVA blocks, the first two loops can use post-condition to save one + * branch each. */ - if (init) { - do { - asm volatile(__MTE_PREAMBLE "stzg %0, [%0]" - : - : "r" (curr) - : "memory"); - curr += MTE_GRANULE_SIZE; - } while (curr != end); - } else { - do { - asm volatile(__MTE_PREAMBLE "stg %0, [%0]" - : - : "r" (curr) - : "memory"); - curr += MTE_GRANULE_SIZE; - } while (curr != end); - } +#define SET_MEMTAG_RANGE(stg_post, dc_gva) \ + do { \ + if (size >= 2 * dczid_bs) { \ + do { \ + curr = stg_post(curr); \ + } while (curr < end1); \ + \ + do { \ + dc_gva(curr); \ + curr += dczid_bs; \ + } while (curr < end2); \ + } \ + \ + while (curr < end3) \ + curr = stg_post(curr); \ + } while (0) + + if (init) + SET_MEMTAG_RANGE(__stzg_post, __dc_gzva); + else + SET_MEMTAG_RANGE(__stg_post, __dc_gva); +#undef SET_MEMTAG_RANGE } void mte_enable_kernel_sync(void); -- 2.31.1.818.g46aad6cb9e-goog
WARNING: multiple messages have this Message-ID (diff)
From: Evgenii Stepanov <eugenis@google.com> To: Andrey Ryabinin <ryabinin.a.a@gmail.com>, Alexander Potapenko <glider@google.com>, Andrey Konovalov <andreyknvl@gmail.com>, Dmitry Vyukov <dvyukov@google.com>, Catalin Marinas <catalin.marinas@arm.com>, Will Deacon <will@kernel.org>, Steven Price <steven.price@arm.com>, Peter Collingbourne <pcc@google.com>, Evgenii Stepanov <eugenis@google.com>, kasan-dev@googlegroups.com, linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org Subject: [PATCH v5] kasan: speed up mte_set_mem_tag_range Date: Thu, 20 May 2021 18:00:23 -0700 [thread overview] Message-ID: <20210521010023.3244784-1-eugenis@google.com> (raw) Use DC GVA / DC GZVA to speed up KASan memory tagging in HW tags mode. The first cacheline is always tagged using STG/STZG even if the address is cacheline-aligned, as benchmarks show it is faster than a conditional branch. Signed-off-by: Evgenii Stepanov <eugenis@google.com> Co-developed-by: Peter Collingbourne <pcc@google.com> Signed-off-by: Peter Collingbourne <pcc@google.com> Reviewed-by: Catalin Marinas <catalin.marinas@arm.com> --- Changelog since v1: - Added Co-developed-by. Changelog since v2: - Added Signed-off-by. Changelog since v3: - Move the implementation back to C with a bit of inline asm. Changelog since v3: - Fixed coding style issues. arch/arm64/include/asm/mte-kasan.h | 93 +++++++++++++++++++++--------- 1 file changed, 67 insertions(+), 26 deletions(-) diff --git a/arch/arm64/include/asm/mte-kasan.h b/arch/arm64/include/asm/mte-kasan.h index ddd4d17cf9a0..d952352bd008 100644 --- a/arch/arm64/include/asm/mte-kasan.h +++ b/arch/arm64/include/asm/mte-kasan.h @@ -48,43 +48,84 @@ static inline u8 mte_get_random_tag(void) return mte_get_ptr_tag(addr); } +static inline u64 __stg_post(u64 p) +{ + asm volatile(__MTE_PREAMBLE "stg %0, [%0], #16" + : "+r"(p) + : + : "memory"); + return p; +} + +static inline u64 __stzg_post(u64 p) +{ + asm volatile(__MTE_PREAMBLE "stzg %0, [%0], #16" + : "+r"(p) + : + : "memory"); + return p; +} + +static inline void __dc_gva(u64 p) +{ + asm volatile(__MTE_PREAMBLE "dc gva, %0" : : "r"(p) : "memory"); +} + +static inline void __dc_gzva(u64 p) +{ + asm volatile(__MTE_PREAMBLE "dc gzva, %0" : : "r"(p) : "memory"); +} + /* * Assign allocation tags for a region of memory based on the pointer tag. * Note: The address must be non-NULL and MTE_GRANULE_SIZE aligned and - * size must be non-zero and MTE_GRANULE_SIZE aligned. + * size must be MTE_GRANULE_SIZE aligned. */ -static inline void mte_set_mem_tag_range(void *addr, size_t size, - u8 tag, bool init) +static inline void mte_set_mem_tag_range(void *addr, size_t size, u8 tag, + bool init) { - u64 curr, end; + u64 curr, mask, dczid_bs, end1, end2, end3; - if (!size) - return; + /* Read DC G(Z)VA block size from the system register. */ + dczid_bs = 4ul << (read_cpuid(DCZID_EL0) & 0xf); curr = (u64)__tag_set(addr, tag); - end = curr + size; + mask = dczid_bs - 1; + /* STG/STZG up to the end of the first block. */ + end1 = curr | mask; + end3 = curr + size; + /* DC GVA / GZVA in [end1, end2) */ + end2 = end3 & ~mask; /* - * 'asm volatile' is required to prevent the compiler to move - * the statement outside of the loop. + * The following code uses STG on the first DC GVA block even if the + * start address is aligned - it appears to be faster than an alignment + * check + conditional branch. Also, if the range size is at least 2 DC + * GVA blocks, the first two loops can use post-condition to save one + * branch each. */ - if (init) { - do { - asm volatile(__MTE_PREAMBLE "stzg %0, [%0]" - : - : "r" (curr) - : "memory"); - curr += MTE_GRANULE_SIZE; - } while (curr != end); - } else { - do { - asm volatile(__MTE_PREAMBLE "stg %0, [%0]" - : - : "r" (curr) - : "memory"); - curr += MTE_GRANULE_SIZE; - } while (curr != end); - } +#define SET_MEMTAG_RANGE(stg_post, dc_gva) \ + do { \ + if (size >= 2 * dczid_bs) { \ + do { \ + curr = stg_post(curr); \ + } while (curr < end1); \ + \ + do { \ + dc_gva(curr); \ + curr += dczid_bs; \ + } while (curr < end2); \ + } \ + \ + while (curr < end3) \ + curr = stg_post(curr); \ + } while (0) + + if (init) + SET_MEMTAG_RANGE(__stzg_post, __dc_gzva); + else + SET_MEMTAG_RANGE(__stg_post, __dc_gva); +#undef SET_MEMTAG_RANGE } void mte_enable_kernel_sync(void); -- 2.31.1.818.g46aad6cb9e-goog _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
next reply other threads:[~2021-05-21 1:00 UTC|newest] Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top 2021-05-21 1:00 Evgenii Stepanov [this message] 2021-05-21 1:00 ` [PATCH v5] kasan: speed up mte_set_mem_tag_range Evgenii Stepanov 2021-05-25 18:58 ` Will Deacon 2021-05-25 18:58 ` Will Deacon
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20210521010023.3244784-1-eugenis@google.com \ --to=eugenis@google.com \ --cc=andreyknvl@gmail.com \ --cc=catalin.marinas@arm.com \ --cc=dvyukov@google.com \ --cc=glider@google.com \ --cc=kasan-dev@googlegroups.com \ --cc=linux-arm-kernel@lists.infradead.org \ --cc=linux-kernel@vger.kernel.org \ --cc=pcc@google.com \ --cc=ryabinin.a.a@gmail.com \ --cc=steven.price@arm.com \ --cc=will@kernel.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.