crypto: powerpc - Add POWER8 optimised crc32c

Use the vector polynomial multiply-sum instructions in POWER8 to
speed up crc32c.

This is just over 41x faster than the slice-by-8 method that it
replaces. Measurements on a 4.1 GHz POWER8 show it sustaining
52 GiB/sec.

A simple btrfs write performance test:

    dd if=/dev/zero of=/mnt/tmpfile bs=1M count=4096
    sync

is over 3.7x faster.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

arch/powerpc/crypto/Makefile

@@ -9,9 +9,11 @@ obj-$(CONFIG_CRYPTO_MD5_PPC) += md5-ppc.o
 obj-$(CONFIG_CRYPTO_SHA1_PPC) += sha1-powerpc.o
 obj-$(CONFIG_CRYPTO_SHA1_PPC_SPE) += sha1-ppc-spe.o
 obj-$(CONFIG_CRYPTO_SHA256_PPC_SPE) += sha256-ppc-spe.o
+obj-$(CONFIG_CRYPT_CRC32C_VPMSUM) += crc32c-vpmsum.o
 
 aes-ppc-spe-y := aes-spe-core.o aes-spe-keys.o aes-tab-4k.o aes-spe-modes.o aes-spe-glue.o
 md5-ppc-y := md5-asm.o md5-glue.o
 sha1-powerpc-y := sha1-powerpc-asm.o sha1.o
 sha1-ppc-spe-y := sha1-spe-asm.o sha1-spe-glue.o
 sha256-ppc-spe-y := sha256-spe-asm.o sha256-spe-glue.o
+crc32c-vpmsum-y := crc32c-vpmsum_asm.o crc32c-vpmsum_glue.o

arch/powerpc/crypto/crc32c-vpmsum_glue.c

+#include <linux/crc32.h>
+#include <crypto/internal/hash.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/kernel.h>
+#include <asm/switch_to.h>
+
+#define CHKSUM_BLOCK_SIZE	1
+#define CHKSUM_DIGEST_SIZE	4
+
+#define VMX_ALIGN		16
+#define VMX_ALIGN_MASK		(VMX_ALIGN-1)
+
+#define VECTOR_BREAKPOINT	512
+
+u32 __crc32c_vpmsum(u32 crc, unsigned char const *p, size_t len);
+
+static u32 crc32c_vpmsum(u32 crc, unsigned char const *p, size_t len)
+{
+	unsigned int prealign;
+	unsigned int tail;
+
+	if (len < (VECTOR_BREAKPOINT + VMX_ALIGN) || in_interrupt())
+		return __crc32c_le(crc, p, len);
+
+	if ((unsigned long)p & VMX_ALIGN_MASK) {
+		prealign = VMX_ALIGN - ((unsigned long)p & VMX_ALIGN_MASK);
+		crc = __crc32c_le(crc, p, prealign);
+		len -= prealign;
+		p += prealign;
+	}
+
+	if (len & ~VMX_ALIGN_MASK) {
+		pagefault_disable();
+		enable_kernel_altivec();
+		crc = __crc32c_vpmsum(crc, p, len & ~VMX_ALIGN_MASK);
+		pagefault_enable();
+	}
+
+	tail = len & VMX_ALIGN_MASK;
+	if (tail) {
+		p += len & ~VMX_ALIGN_MASK;
+		crc = __crc32c_le(crc, p, tail);
+	}
+
+	return crc;
+}
+
+static int crc32c_vpmsum_cra_init(struct crypto_tfm *tfm)
+{
+	u32 *key = crypto_tfm_ctx(tfm);
+
+	*key = 0;
+
+	return 0;
+}
+
+/*
+ * Setting the seed allows arbitrary accumulators and flexible XOR policy
+ * If your algorithm starts with ~0, then XOR with ~0 before you set
+ * the seed.
+ */
+static int crc32c_vpmsum_setkey(struct crypto_shash *hash, const u8 *key,
+			       unsigned int keylen)
+{
+	u32 *mctx = crypto_shash_ctx(hash);
+
+	if (keylen != sizeof(u32)) {
+		crypto_shash_set_flags(hash, CRYPTO_TFM_RES_BAD_KEY_LEN);
+		return -EINVAL;
+	}
+	*mctx = le32_to_cpup((__le32 *)key);
+	return 0;
+}
+
+static int crc32c_vpmsum_init(struct shash_desc *desc)
+{
+	u32 *mctx = crypto_shash_ctx(desc->tfm);
+	u32 *crcp = shash_desc_ctx(desc);
+
+	*crcp = *mctx;
+
+	return 0;
+}
+
+static int crc32c_vpmsum_update(struct shash_desc *desc, const u8 *data,
+			       unsigned int len)
+{
+	u32 *crcp = shash_desc_ctx(desc);
+
+	*crcp = crc32c_vpmsum(*crcp, data, len);
+
+	return 0;
+}
+
+static int __crc32c_vpmsum_finup(u32 *crcp, const u8 *data, unsigned int len,
+				u8 *out)
+{
+	*(__le32 *)out = ~cpu_to_le32(crc32c_vpmsum(*crcp, data, len));
+
+	return 0;
+}
+
+static int crc32c_vpmsum_finup(struct shash_desc *desc, const u8 *data,
+			      unsigned int len, u8 *out)
+{
+	return __crc32c_vpmsum_finup(shash_desc_ctx(desc), data, len, out);
+}
+
+static int crc32c_vpmsum_final(struct shash_desc *desc, u8 *out)
+{
+	u32 *crcp = shash_desc_ctx(desc);
+
+	*(__le32 *)out = ~cpu_to_le32p(crcp);
+
+	return 0;
+}
+
+static int crc32c_vpmsum_digest(struct shash_desc *desc, const u8 *data,
+			       unsigned int len, u8 *out)
+{
+	return __crc32c_vpmsum_finup(crypto_shash_ctx(desc->tfm), data, len,
+				     out);
+}
+
+static struct shash_alg alg = {
+	.setkey		= crc32c_vpmsum_setkey,
+	.init		= crc32c_vpmsum_init,
+	.update		= crc32c_vpmsum_update,
+	.final		= crc32c_vpmsum_final,
+	.finup		= crc32c_vpmsum_finup,
+	.digest		= crc32c_vpmsum_digest,
+	.descsize	= sizeof(u32),
+	.digestsize	= CHKSUM_DIGEST_SIZE,
+	.base		= {
+		.cra_name		= "crc32c",
+		.cra_driver_name	= "crc32c-vpmsum",
+		.cra_priority		= 200,
+		.cra_blocksize		= CHKSUM_BLOCK_SIZE,
+		.cra_ctxsize		= sizeof(u32),
+		.cra_module		= THIS_MODULE,
+		.cra_init		= crc32c_vpmsum_cra_init,
+	}
+};
+
+static int __init crc32c_vpmsum_mod_init(void)
+{
+	if (!cpu_has_feature(CPU_FTR_ARCH_207S))
+		return -ENODEV;
+
+	return crypto_register_shash(&alg);
+}
+
+static void __exit crc32c_vpmsum_mod_fini(void)
+{
+	crypto_unregister_shash(&alg);
+}
+
+module_init(crc32c_vpmsum_mod_init);
+module_exit(crc32c_vpmsum_mod_fini);
+
+MODULE_AUTHOR("Anton Blanchard <anton@samba.org>");
+MODULE_DESCRIPTION("CRC32C using vector polynomial multiply-sum instructions");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_CRYPTO("crc32c");
+MODULE_ALIAS_CRYPTO("crc32c-vpmsum");

arch/powerpc/include/asm/ppc-opcode.h

@@ -174,6 +174,8 @@
 #define PPC_INST_MFSPR_DSCR_USER_MASK	0xfc1fffff
 #define PPC_INST_MTSPR_DSCR_USER	0x7c0303a6
 #define PPC_INST_MTSPR_DSCR_USER_MASK	0xfc1fffff
+#define PPC_INST_MFVSRD			0x7c000066
+#define PPC_INST_MTVSRD			0x7c000166
 #define PPC_INST_SLBFEE			0x7c0007a7
 
 #define PPC_INST_STRING			0x7c00042a
@@ -188,6 +190,8 @@
 #define PPC_INST_WAIT			0x7c00007c
 #define PPC_INST_TLBIVAX		0x7c000624
 #define PPC_INST_TLBSRX_DOT		0x7c0006a5
+#define PPC_INST_VPMSUMW		0x10000488
+#define PPC_INST_VPMSUMD		0x100004c8
 #define PPC_INST_XXLOR			0xf0000510
 #define PPC_INST_XXSWAPD		0xf0000250
 #define PPC_INST_XVCPSGNDP		0xf0000780
@@ -359,6 +363,14 @@
 					       VSX_XX1((s), a, b))
 #define LXVD2X(s, a, b)		stringify_in_c(.long PPC_INST_LXVD2X | \
 					       VSX_XX1((s), a, b))
+#define MFVRD(a, t)		stringify_in_c(.long PPC_INST_MFVSRD | \
+					       VSX_XX1((t)+32, a, R0))
+#define MTVRD(t, a)		stringify_in_c(.long PPC_INST_MTVSRD | \
+					       VSX_XX1((t)+32, a, R0))
+#define VPMSUMW(t, a, b)	stringify_in_c(.long PPC_INST_VPMSUMW | \
+					       VSX_XX3((t), a, b))
+#define VPMSUMD(t, a, b)	stringify_in_c(.long PPC_INST_VPMSUMD | \
+					       VSX_XX3((t), a, b))
 #define XXLOR(t, a, b)		stringify_in_c(.long PPC_INST_XXLOR | \
 					       VSX_XX3((t), a, b))
 #define XXSWAPD(t, a)		stringify_in_c(.long PPC_INST_XXSWAPD | \

crypto/Kconfig

@@ -437,6 +437,17 @@ config CRYPTO_CRC32C_INTEL
 	  gain performance compared with software implementation.
 	  Module will be crc32c-intel.
 
+config CRYPT_CRC32C_VPMSUM
+	tristate "CRC32c CRC algorithm (powerpc64)"
+	depends on PPC64
+	select CRYPTO_HASH
+	select CRC32
+	help
+	  CRC32c algorithm implemented using vector polynomial multiply-sum
+	  (vpmsum) instructions, introduced in POWER8. Enable on POWER8
+	  and newer processors for improved performance.
+
+
 config CRYPTO_CRC32C_SPARC64
 	tristate "CRC32c CRC algorithm (SPARC64)"
 	depends on SPARC64