crypto: crc32c - Optimize CRC32C calculation with PCLMULQDQ instruction

This patch adds the crc_pcl function that calculates CRC32C checksum using the
PCLMULQDQ instruction on processors that support this feature. This will
provide speedup over using CRC32 instruction only.
The usage of PCLMULQDQ necessitate the invocation of kernel_fpu_begin and
kernel_fpu_end and incur some overhead.  So the new crc_pcl function is only
invoked for buffer size of 512 bytes or more.  Larger sized
buffers will expect to see greater speedup.  This feature is best used coupled
with eager_fpu which reduces the kernel_fpu_begin/end overhead.  For
buffer size of 1K the speedup is around 1.6x and for buffer size greater than
4K, the speedup is around 3x compared to original implementation in crc32c-intel
module. Test was performed on Sandy Bridge based platform with constant frequency
set for cpu.

A white paper detailing the algorithm can be found here:
http://download.intel.com/design/intarch/papers/323405.pdfSigned-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

arch/x86/crypto/Makefile

@@ -48,3 +48,4 @@ aesni-intel-y := aesni-intel_asm.o aesni-intel_glue.o fpu.o
 ghash-clmulni-intel-y := ghash-clmulni-intel_asm.o ghash-clmulni-intel_glue.o
 sha1-ssse3-y := sha1_ssse3_asm.o sha1_ssse3_glue.o
 crc32c-intel-y := crc32c-intel_glue.o
+crc32c-intel-$(CONFIG_CRYPTO_CRC32C_X86_64) += crc32c-pcl-intel-asm_64.o

arch/x86/crypto/crc32c-intel_glue.c

@@ -32,6 +32,8 @@
 
 #include <asm/cpufeature.h>
 #include <asm/cpu_device_id.h>
+#include <asm/i387.h>
+#include <asm/fpu-internal.h>
 
 #define CHKSUM_BLOCK_SIZE	1
 #define CHKSUM_DIGEST_SIZE	4
@@ -44,6 +46,31 @@
 #define REX_PRE
 #endif
 
+#ifdef CONFIG_X86_64
+/*
+ * use carryless multiply version of crc32c when buffer
+ * size is >= 512 (when eager fpu is enabled) or
+ * >= 1024 (when eager fpu is disabled) to account
+ * for fpu state save/restore overhead.
+ */
+#define CRC32C_PCL_BREAKEVEN_EAGERFPU	512
+#define CRC32C_PCL_BREAKEVEN_NOEAGERFPU	1024
+
+asmlinkage unsigned int crc_pcl(const u8 *buffer, int len,
+				unsigned int crc_init);
+static int crc32c_pcl_breakeven = CRC32C_PCL_BREAKEVEN_EAGERFPU;
+#if defined(X86_FEATURE_EAGER_FPU)
+#define set_pcl_breakeven_point()					\
+do {									\
+	if (!use_eager_fpu())						\
+		crc32c_pcl_breakeven = CRC32C_PCL_BREAKEVEN_NOEAGERFPU;	\
+} while (0)
+#else
+#define set_pcl_breakeven_point()					\
+	(crc32c_pcl_breakeven = CRC32C_PCL_BREAKEVEN_NOEAGERFPU)
+#endif
+#endif /* CONFIG_X86_64 */
+
 static u32 crc32c_intel_le_hw_byte(u32 crc, unsigned char const *data, size_t length)
 {
 	while (length--) {
@@ -154,6 +181,52 @@ static int crc32c_intel_cra_init(struct crypto_tfm *tfm)
 	return 0;
 }
 
+#ifdef CONFIG_X86_64
+static int crc32c_pcl_intel_update(struct shash_desc *desc, const u8 *data,
+			       unsigned int len)
+{
+	u32 *crcp = shash_desc_ctx(desc);
+
+	/*
+	 * use faster PCL version if datasize is large enough to
+	 * overcome kernel fpu state save/restore overhead
+	 */
+	if (len >= crc32c_pcl_breakeven && irq_fpu_usable()) {
+		kernel_fpu_begin();
+		*crcp = crc_pcl(data, len, *crcp);
+		kernel_fpu_end();
+	} else
+		*crcp = crc32c_intel_le_hw(*crcp, data, len);
+	return 0;
+}
+
+static int __crc32c_pcl_intel_finup(u32 *crcp, const u8 *data, unsigned int len,
+				u8 *out)
+{
+	if (len >= crc32c_pcl_breakeven && irq_fpu_usable()) {
+		kernel_fpu_begin();
+		*(__le32 *)out = ~cpu_to_le32(crc_pcl(data, len, *crcp));
+		kernel_fpu_end();
+	} else
+		*(__le32 *)out =
+			~cpu_to_le32(crc32c_intel_le_hw(*crcp, data, len));
+	return 0;
+}
+
+static int crc32c_pcl_intel_finup(struct shash_desc *desc, const u8 *data,
+			      unsigned int len, u8 *out)
+{
+	return __crc32c_pcl_intel_finup(shash_desc_ctx(desc), data, len, out);
+}
+
+static int crc32c_pcl_intel_digest(struct shash_desc *desc, const u8 *data,
+			       unsigned int len, u8 *out)
+{
+	return __crc32c_pcl_intel_finup(crypto_shash_ctx(desc->tfm), data, len,
+				    out);
+}
+#endif /* CONFIG_X86_64 */
+
 static struct shash_alg alg = {
 	.setkey			=	crc32c_intel_setkey,
 	.init			=	crc32c_intel_init,
@@ -184,6 +257,14 @@ static int __init crc32c_intel_mod_init(void)
 {
 	if (!x86_match_cpu(crc32c_cpu_id))
 		return -ENODEV;
+#ifdef CONFIG_X86_64
+	if (cpu_has_pclmulqdq) {
+		alg.update = crc32c_pcl_intel_update;
+		alg.finup = crc32c_pcl_intel_finup;
+		alg.digest = crc32c_pcl_intel_digest;
+		set_pcl_breakeven_point();
+	}
+#endif
 	return crypto_register_shash(&alg);
 }
 

crypto/Kconfig

@@ -324,9 +324,19 @@ config CRYPTO_CRC32C
 	  by iSCSI for header and data digests and by others.
 	  See Castagnoli93.  Module will be crc32c.
 
+config CRYPTO_CRC32C_X86_64
+	bool
+	depends on X86 && 64BIT
+	select CRYPTO_HASH
+	help
+	  In Intel processor with SSE4.2 supported, the processor will
+	  support CRC32C calculation using hardware accelerated CRC32
+	  instruction optimized with PCLMULQDQ instruction when available.
+
 config CRYPTO_CRC32C_INTEL
 	tristate "CRC32c INTEL hardware acceleration"
 	depends on X86
+	select CRYPTO_CRC32C_X86_64 if 64BIT
 	select CRYPTO_HASH
 	help
 	  In Intel processor with SSE4.2 supported, the processor will