linux-kernel - [RFC PATCH 6/7] crypto: x86/ghash

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20221006223151.22159-7-elliott@hpe.com>
Date:   Thu,  6 Oct 2022 17:31:50 -0500
From:   Robert Elliott <elliott@....com>
To:     herbert@...dor.apana.org.au, davem@...emloft.net,
        tim.c.chen@...ux.intel.com, linux-crypto@...r.kernel.org,
        linux-kernel@...r.kernel.org
Cc:     Robert Elliott <elliott@....com>
Subject: [RFC PATCH 6/7] crypto: x86/ghash - limit FPU preemption

As done by the ECB and CBC helpers in arch/x86/crypt/ecb_cbc_helpers.h,
limit the number of bytes processed between kernel_fpu_begin() and
kernel_fpu_end() calls.

Those functions call preempt_disable() and preempt_enable(), so
the CPU core is unavailable for scheduling while running, leading to:
    rcu: INFO: rcu_preempt detected expedited stalls on CPUs/tasks: {12-... } 22 jiffies s: 277 root: 0x1/.

Fixes: 0e1227d356e9 ("crypto: ghash - Add PCLMULQDQ accelerated implementation")
Suggested-by: Herbert Xu <herbert@...dor.apana.org.au>
Signed-off-by: Robert Elliott <elliott@....com>
---
 arch/x86/crypto/ghash-clmulni-intel_glue.c | 27 ++++++++++++++++------
 1 file changed, 20 insertions(+), 7 deletions(-)

diff --git a/arch/x86/crypto/ghash-clmulni-intel_glue.c b/arch/x86/crypto/ghash-clmulni-intel_glue.c
index b25730c5b267..b9a3e2187f5b 100644
--- a/arch/x86/crypto/ghash-clmulni-intel_glue.c
+++ b/arch/x86/crypto/ghash-clmulni-intel_glue.c
@@ -25,6 +25,8 @@
 #define GHASH_BLOCK_SIZE	16
 #define GHASH_DIGEST_SIZE	16
 
+#define FPU_BYTES 4096U /* avoid kernel_fpu_begin/end scheduler/rcu stalls */
+
 void clmul_ghash_mul(char *dst, const u128 *shash);
 
 void clmul_ghash_update(char *dst, const char *src, unsigned int srclen,
@@ -81,10 +83,11 @@ static int ghash_update(struct shash_desc *desc,
 	struct ghash_desc_ctx *dctx = shash_desc_ctx(desc);
 	struct ghash_ctx *ctx = crypto_shash_ctx(desc->tfm);
 	u8 *dst = dctx->buffer;
+	unsigned int fpulen;
 
 	if (dctx->bytes) {
 		int n = min(srclen, dctx->bytes);
-		u8 *pos = dst + (GHASH_BLOCK_SIZE - dctx->bytes);
+		u8 *pos = dst + GHASH_BLOCK_SIZE - dctx->bytes;
 
 		dctx->bytes -= n;
 		srclen -= n;
@@ -99,13 +102,23 @@ static int ghash_update(struct shash_desc *desc,
 		}
 	}
 
-	kernel_fpu_begin();
-	clmul_ghash_update(dst, src, srclen, &ctx->shash);
-	kernel_fpu_end();
+	while (srclen >= GHASH_BLOCK_SIZE) {
+		fpulen = min(srclen, FPU_BYTES);
+
+		kernel_fpu_begin();
+		while (fpulen >= GHASH_BLOCK_SIZE) {
+			int n = min_t(unsigned int, fpulen, GHASH_BLOCK_SIZE);
+
+			clmul_ghash_update(dst, src, n, &ctx->shash);
+
+			srclen -= n;
+			fpulen -= n;
+			src += n;
+		}
+		kernel_fpu_end();
+	}
 
-	if (srclen & 0xf) {
-		src += srclen - (srclen & 0xf);
-		srclen &= 0xf;
+	if (srclen) {
 		dctx->bytes = GHASH_BLOCK_SIZE - srclen;
 		while (srclen--)
 			*dst++ ^= *src++;
-- 
2.37.3