[PATCH 3.2] crypto: twofish-x86_64-3way - blacklist pentium4 and atom

[PATCH 3.2] crypto: twofish-x86_64-3way - blacklist pentium4 and atom

[Jussi Kivilinna]

Performance of twofish-x86_64-3way on Intel Pentium 4 and Atom is lower than
of twofish-x86_64 module. So blacklist these CPUs.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
---
 arch/x86/crypto/twofish_glue_3way.c |   47 +++++++++++++++++++++++++++++++++++
 1 files changed, 47 insertions(+), 0 deletions(-)

diff --git a/arch/x86/crypto/twofish_glue_3way.c b/arch/x86/crypto/twofish_glue_3way.c
index 5ede9c4..4897b6b 100644
--- a/arch/x86/crypto/twofish_glue_3way.c
+++ b/arch/x86/crypto/twofish_glue_3way.c
@@ -25,6 +25,7 @@
  *
  */
 
+#include <asm/processor.h>
 #include <linux/crypto.h>
 #include <linux/init.h>
 #include <linux/module.h>
@@ -432,10 +433,56 @@ static struct crypto_alg blk_ctr_alg = {
 	},
 };
 
+static bool is_blacklisted_cpu(void)
+{
+	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
+		return false;
+
+	if (boot_cpu_data.x86 == 0x06 &&
+		(boot_cpu_data.x86_model == 0x1c ||
+		 boot_cpu_data.x86_model == 0x26 ||
+		 boot_cpu_data.x86_model == 0x36)) {
+		/*
+		 * On Atom, twofish-3way is slower than original assembler
+		 * implementation. Twofish-3way trades off some performance in
+		 * storing blocks in 64bit registers to allow three blocks to
+		 * be processed parallel. Parallel operation then allows gaining
+		 * more performance than was trade off, on out-of-order CPUs.
+		 * However Atom does not benefit from this parallellism and
+		 * should be blacklisted.
+		 */
+		return true;
+	}
+
+	if (boot_cpu_data.x86 == 0x0f) {
+		/*
+		 * On Pentium 4, twofish-3way is slower than original assembler
+		 * implementation because excessive uses of 64bit rotate and
+		 * left-shifts (which are really slow on P4) needed to store and
+		 * handle 128bit block in two 64bit registers.
+		 */
+		return true;
+	}
+
+	return false;
+}
+
+static int force;
+module_param(force, int, 0);
+MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
+
 int __init init(void)
 {
 	int err;
 
+	if (!force && is_blacklisted_cpu()) {
+		printk(KERN_INFO
+			"twofish-x86_64-3way: performance on this CPU "
+			"would be suboptimal: disabling "
+			"twofish-x86_64-3way.\n");
+		return -ENODEV;
+	}
+
 	err = crypto_register_alg(&blk_ecb_alg);
 	if (err)
 		goto ecb_err;

Re: [PATCH 3.2] crypto: twofish-x86_64-3way - blacklist pentium4 and atom

[Herbert Xu]

On Sat, Dec 03, 2011 at 04:10:04PM +0200, Jussi Kivilinna wrote:
> Performance of twofish-x86_64-3way on Intel Pentium 4 and Atom is lower than
> of twofish-x86_64 module. So blacklist these CPUs.
> 
> Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>

Please send patches based on cryptodev instead.  If necessary I
can then rebase them to crypto.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

[PATCH 1/2] crypto: twofish-x86_64-3way - blacklist pentium4 and atom

[Jussi Kivilinna]

Performance of twofish-x86_64-3way on Intel Pentium 4 and Atom is lower than
of twofish-x86_64 module. So blacklist these CPUs.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
---
 arch/x86/crypto/twofish_glue_3way.c |   47 +++++++++++++++++++++++++++++++++++
 1 files changed, 47 insertions(+), 0 deletions(-)

diff --git a/arch/x86/crypto/twofish_glue_3way.c b/arch/x86/crypto/twofish_glue_3way.c
index 7fee8c1..0afd134 100644
--- a/arch/x86/crypto/twofish_glue_3way.c
+++ b/arch/x86/crypto/twofish_glue_3way.c
@@ -25,6 +25,7 @@
  *
  */
 
+#include <asm/processor.h>
 #include <linux/crypto.h>
 #include <linux/init.h>
 #include <linux/module.h>
@@ -637,10 +638,56 @@ static struct crypto_alg blk_xts_alg = {
 	},
 };
 
+static bool is_blacklisted_cpu(void)
+{
+	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
+		return false;
+
+	if (boot_cpu_data.x86 == 0x06 &&
+		(boot_cpu_data.x86_model == 0x1c ||
+		 boot_cpu_data.x86_model == 0x26 ||
+		 boot_cpu_data.x86_model == 0x36)) {
+		/*
+		 * On Atom, twofish-3way is slower than original assembler
+		 * implementation. Twofish-3way trades off some performance in
+		 * storing blocks in 64bit registers to allow three blocks to
+		 * be processed parallel. Parallel operation then allows gaining
+		 * more performance than was trade off, on out-of-order CPUs.
+		 * However Atom does not benefit from this parallellism and
+		 * should be blacklisted.
+		 */
+		return true;
+	}
+
+	if (boot_cpu_data.x86 == 0x0f) {
+		/*
+		 * On Pentium 4, twofish-3way is slower than original assembler
+		 * implementation because excessive uses of 64bit rotate and
+		 * left-shifts (which are really slow on P4) needed to store and
+		 * handle 128bit block in two 64bit registers.
+		 */
+		return true;
+	}
+
+	return false;
+}
+
+static int force;
+module_param(force, int, 0);
+MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
+
 int __init init(void)
 {
 	int err;
 
+	if (!force && is_blacklisted_cpu()) {
+		printk(KERN_INFO
+			"twofish-x86_64-3way: performance on this CPU "
+			"would be suboptimal: disabling "
+			"twofish-x86_64-3way.\n");
+		return -ENODEV;
+	}
+
 	err = crypto_register_alg(&blk_ecb_alg);
 	if (err)
 		goto ecb_err;

Re: [PATCH 1/2] crypto: twofish-x86_64-3way - blacklist pentium4 and atom

[Herbert Xu]

On Tue, Dec 20, 2011 at 12:20:16PM +0200, Jussi Kivilinna wrote:
> Performance of twofish-x86_64-3way on Intel Pentium 4 and Atom is lower than
> of twofish-x86_64 module. So blacklist these CPUs.
> 
> Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>

Both patches applied.  Thanks!
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

[PATCH 2/2] crypto: blowfish-x86_64 - blacklist Pentium 4

[Jussi Kivilinna]

Implementation in blowfish-x86_64 uses 64bit rotations which are slow on P4,
making blowfish-x86_64 slower than generic C implementation. Therefore
blacklist P4.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
---
 arch/x86/crypto/blowfish_glue.c |   30 ++++++++++++++++++++++++++++++
 1 files changed, 30 insertions(+), 0 deletions(-)

diff --git a/arch/x86/crypto/blowfish_glue.c b/arch/x86/crypto/blowfish_glue.c
index b05aa16..2970110 100644
--- a/arch/x86/crypto/blowfish_glue.c
+++ b/arch/x86/crypto/blowfish_glue.c
@@ -25,6 +25,7 @@
  *
  */
 
+#include <asm/processor.h>
 #include <crypto/blowfish.h>
 #include <linux/crypto.h>
 #include <linux/init.h>
@@ -446,10 +447,39 @@ static struct crypto_alg blk_ctr_alg = {
 	},
 };
 
+static bool is_blacklisted_cpu(void)
+{
+	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
+		return false;
+
+	if (boot_cpu_data.x86 == 0x0f) {
+		/*
+		 * On Pentium 4, blowfish-x86_64 is slower than generic C
+		 * implementation because use of 64bit rotates (which are really
+		 * slow on P4). Therefore blacklist P4s.
+		 */
+		return true;
+	}
+
+	return false;
+}
+
+static int force;
+module_param(force, int, 0);
+MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
+
 static int __init init(void)
 {
 	int err;
 
+	if (!force && is_blacklisted_cpu()) {
+		printk(KERN_INFO
+			"blowfish-x86_64: performance on this CPU "
+			"would be suboptimal: disabling "
+			"blowfish-x86_64.\n");
+		return -ENODEV;
+	}
+
 	err = crypto_register_alg(&bf_alg);
 	if (err)
 		goto bf_err;