Blob Blame History Raw
From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Fri, 21 Feb 2014 17:24:04 +0100
Subject: crypto: Reduce preempt disabled regions, more algos
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git
Git-commit: 4d6956a1b60ea5495618205ce9005013e541085b
Patch-mainline: Queued in subsystem maintainer repository
References: SLE Realtime Extension

Don Estabrook reported
| kernel: WARNING: CPU: 2 PID: 858 at kernel/sched/core.c:2428 migrate_disable+0xed/0x100()
| kernel: WARNING: CPU: 2 PID: 858 at kernel/sched/core.c:2462 migrate_enable+0x17b/0x200()
| kernel: WARNING: CPU: 3 PID: 865 at kernel/sched/core.c:2428 migrate_disable+0xed/0x100()

and his backtrace showed some crypto functions which looked fine.

The problem is the following sequence:

glue_xts_crypt_128bit()
{
	blkcipher_walk_virt(); /* normal migrate_disable() */

	glue_fpu_begin(); /* get atomic */

	while (nbytes) {
		__glue_xts_crypt_128bit();
		blkcipher_walk_done(); /* with nbytes = 0, migrate_enable()
					* while we are atomic */
	};
	glue_fpu_end() /* no longer atomic */
}

and this is why the counter get out of sync and the warning is printed.
The other problem is that we are non-preemptible between
glue_fpu_begin() and glue_fpu_end() and the latency grows. To fix this,
I shorten the FPU off region and ensure blkcipher_walk_done() is called
with preemption enabled. This might hurt the performance because we now
enable/disable the FPU state more often but we gain lower latency and
the bug is gone.


Reported-by: Don Estabrook <don.estabrook@gmail.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Mike Galbraith <mgalbraith@suse.de>
---
 arch/x86/crypto/cast5_avx_glue.c |   21 +++++++++------------
 arch/x86/crypto/glue_helper.c    |   31 +++++++++++++++----------------
 2 files changed, 24 insertions(+), 28 deletions(-)

--- a/arch/x86/crypto/cast5_avx_glue.c
+++ b/arch/x86/crypto/cast5_avx_glue.c
@@ -59,7 +59,7 @@ static inline void cast5_fpu_end(bool fp
 static int ecb_crypt(struct blkcipher_desc *desc, struct blkcipher_walk *walk,
 		     bool enc)
 {
-	bool fpu_enabled = false;
+	bool fpu_enabled;
 	struct cast5_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
 	const unsigned int bsize = CAST5_BLOCK_SIZE;
 	unsigned int nbytes;
@@ -75,7 +75,7 @@ static int ecb_crypt(struct blkcipher_de
 		u8 *wsrc = walk->src.virt.addr;
 		u8 *wdst = walk->dst.virt.addr;
 
-		fpu_enabled = cast5_fpu_begin(fpu_enabled, nbytes);
+		fpu_enabled = cast5_fpu_begin(false, nbytes);
 
 		/* Process multi-block batch */
 		if (nbytes >= bsize * CAST5_PARALLEL_BLOCKS) {
@@ -103,10 +103,9 @@ static int ecb_crypt(struct blkcipher_de
 		} while (nbytes >= bsize);
 
 done:
+		cast5_fpu_end(fpu_enabled);
 		err = blkcipher_walk_done(desc, walk, nbytes);
 	}
-
-	cast5_fpu_end(fpu_enabled);
 	return err;
 }
 
@@ -227,7 +226,7 @@ static unsigned int __cbc_decrypt(struct
 static int cbc_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
 		       struct scatterlist *src, unsigned int nbytes)
 {
-	bool fpu_enabled = false;
+	bool fpu_enabled;
 	struct blkcipher_walk walk;
 	int err;
 
@@ -236,12 +235,11 @@ static int cbc_decrypt(struct blkcipher_
 	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
 
 	while ((nbytes = walk.nbytes)) {
-		fpu_enabled = cast5_fpu_begin(fpu_enabled, nbytes);
+		fpu_enabled = cast5_fpu_begin(false, nbytes);
 		nbytes = __cbc_decrypt(desc, &walk);
+		cast5_fpu_end(fpu_enabled);
 		err = blkcipher_walk_done(desc, &walk, nbytes);
 	}
-
-	cast5_fpu_end(fpu_enabled);
 	return err;
 }
 
@@ -311,7 +309,7 @@ static unsigned int __ctr_crypt(struct b
 static int ctr_crypt(struct blkcipher_desc *desc, struct scatterlist *dst,
 		     struct scatterlist *src, unsigned int nbytes)
 {
-	bool fpu_enabled = false;
+	bool fpu_enabled;
 	struct blkcipher_walk walk;
 	int err;
 
@@ -320,13 +318,12 @@ static int ctr_crypt(struct blkcipher_de
 	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
 
 	while ((nbytes = walk.nbytes) >= CAST5_BLOCK_SIZE) {
-		fpu_enabled = cast5_fpu_begin(fpu_enabled, nbytes);
+		fpu_enabled = cast5_fpu_begin(false, nbytes);
 		nbytes = __ctr_crypt(desc, &walk);
+		cast5_fpu_end(fpu_enabled);
 		err = blkcipher_walk_done(desc, &walk, nbytes);
 	}
 
-	cast5_fpu_end(fpu_enabled);
-
 	if (walk.nbytes) {
 		ctr_crypt_final(desc, &walk);
 		err = blkcipher_walk_done(desc, &walk, 0);
--- a/arch/x86/crypto/glue_helper.c
+++ b/arch/x86/crypto/glue_helper.c
@@ -40,7 +40,7 @@ static int __glue_ecb_crypt_128bit(const
 	void *ctx = crypto_blkcipher_ctx(desc->tfm);
 	const unsigned int bsize = 128 / 8;
 	unsigned int nbytes, i, func_bytes;
-	bool fpu_enabled = false;
+	bool fpu_enabled;
 	int err;
 
 	err = blkcipher_walk_virt(desc, walk);
@@ -50,7 +50,7 @@ static int __glue_ecb_crypt_128bit(const
 		u8 *wdst = walk->dst.virt.addr;
 
 		fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
-					     desc, fpu_enabled, nbytes);
+					     desc, false, nbytes);
 
 		for (i = 0; i < gctx->num_funcs; i++) {
 			func_bytes = bsize * gctx->funcs[i].num_blocks;
@@ -72,10 +72,10 @@ static int __glue_ecb_crypt_128bit(const
 		}
 
 done:
+		glue_fpu_end(fpu_enabled);
 		err = blkcipher_walk_done(desc, walk, nbytes);
 	}
 
-	glue_fpu_end(fpu_enabled);
 	return err;
 }
 
@@ -195,7 +195,7 @@ int glue_cbc_decrypt_128bit(const struct
 			    struct scatterlist *src, unsigned int nbytes)
 {
 	const unsigned int bsize = 128 / 8;
-	bool fpu_enabled = false;
+	bool fpu_enabled;
 	struct blkcipher_walk walk;
 	int err;
 
@@ -204,12 +204,12 @@ int glue_cbc_decrypt_128bit(const struct
 
 	while ((nbytes = walk.nbytes)) {
 		fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
-					     desc, fpu_enabled, nbytes);
+					     desc, false, nbytes);
 		nbytes = __glue_cbc_decrypt_128bit(gctx, desc, &walk);
+		glue_fpu_end(fpu_enabled);
 		err = blkcipher_walk_done(desc, &walk, nbytes);
 	}
 
-	glue_fpu_end(fpu_enabled);
 	return err;
 }
 EXPORT_SYMBOL_GPL(glue_cbc_decrypt_128bit);
@@ -278,7 +278,7 @@ int glue_ctr_crypt_128bit(const struct c
 			  struct scatterlist *src, unsigned int nbytes)
 {
 	const unsigned int bsize = 128 / 8;
-	bool fpu_enabled = false;
+	bool fpu_enabled;
 	struct blkcipher_walk walk;
 	int err;
 
@@ -287,13 +287,12 @@ int glue_ctr_crypt_128bit(const struct c
 
 	while ((nbytes = walk.nbytes) >= bsize) {
 		fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
-					     desc, fpu_enabled, nbytes);
+					     desc, false, nbytes);
 		nbytes = __glue_ctr_crypt_128bit(gctx, desc, &walk);
+		glue_fpu_end(fpu_enabled);
 		err = blkcipher_walk_done(desc, &walk, nbytes);
 	}
 
-	glue_fpu_end(fpu_enabled);
-
 	if (walk.nbytes) {
 		glue_ctr_crypt_final_128bit(
 			gctx->funcs[gctx->num_funcs - 1].fn_u.ctr, desc, &walk);
@@ -383,7 +382,7 @@ int glue_xts_crypt_128bit(const struct c
 			  void *tweak_ctx, void *crypt_ctx)
 {
 	const unsigned int bsize = 128 / 8;
-	bool fpu_enabled = false;
+	bool fpu_enabled;
 	struct blkcipher_walk walk;
 	int err;
 
@@ -396,21 +395,21 @@ int glue_xts_crypt_128bit(const struct c
 
 	/* set minimum length to bsize, for tweak_fn */
 	fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
-				     desc, fpu_enabled,
+				     desc, false,
 				     nbytes < bsize ? bsize : nbytes);
-
 	/* calculate first value of T */
 	tweak_fn(tweak_ctx, walk.iv, walk.iv);
+	glue_fpu_end(fpu_enabled);
 
 	while (nbytes) {
+		fpu_enabled = glue_fpu_begin(bsize, gctx->fpu_blocks_limit,
+				desc, false, nbytes);
 		nbytes = __glue_xts_crypt_128bit(gctx, crypt_ctx, desc, &walk);
 
+		glue_fpu_end(fpu_enabled);
 		err = blkcipher_walk_done(desc, &walk, nbytes);
 		nbytes = walk.nbytes;
 	}
-
-	glue_fpu_end(fpu_enabled);
-
 	return err;
 }
 EXPORT_SYMBOL_GPL(glue_xts_crypt_128bit);