[PATCH 6/6] arm64: lib: Implement optimized string length routines

[zhichang.yuan@linaro.org (zhichang.yuan at linaro.org)]

From: "zhichang.yuan" <zhichang.yuan@linaro.org>

This patch, based on Linaro's Cortex Strings library, adds
an assembly optimized strlen() and strnlen() functions.

Signed-off-by: Zhichang Yuan <zhichang.yuan@linaro.org>
Signed-off-by: Deepak Saxena <dsaxena@linaro.org>
---
 arch/arm64/include/asm/string.h |    6 ++
 arch/arm64/kernel/arm64ksyms.c  |    2 +
 arch/arm64/lib/Makefile         |    3 +-
 arch/arm64/lib/strlen.S         |  131 ++++++++++++++++++++++++++++
 arch/arm64/lib/strnlen.S        |  179 +++++++++++++++++++++++++++++++++++++++
 5 files changed, 320 insertions(+), 1 deletion(-)
 create mode 100644 arch/arm64/lib/strlen.S
 create mode 100644 arch/arm64/lib/strnlen.S

diff --git a/arch/arm64/include/asm/string.h b/arch/arm64/include/asm/string.h
index 6133f49..64d2d48 100644
--- a/arch/arm64/include/asm/string.h
+++ b/arch/arm64/include/asm/string.h
@@ -28,6 +28,12 @@ extern int strcmp(const char *, const char *);
 #define __HAVE_ARCH_STRNCMP
 extern int strncmp(const char *, const char *, __kernel_size_t);
 
+#define __HAVE_ARCH_STRLEN
+extern __kernel_size_t strlen(const char *);
+
+#define __HAVE_ARCH_STRNLEN
+extern __kernel_size_t strnlen(const char *, __kernel_size_t);
+
 #define __HAVE_ARCH_MEMCPY
 extern void *memcpy(void *, const void *, __kernel_size_t);
 
diff --git a/arch/arm64/kernel/arm64ksyms.c b/arch/arm64/kernel/arm64ksyms.c
index d61fa39..c71b435 100644
--- a/arch/arm64/kernel/arm64ksyms.c
+++ b/arch/arm64/kernel/arm64ksyms.c
@@ -49,6 +49,8 @@ EXPORT_SYMBOL(strchr);
 EXPORT_SYMBOL(strrchr);
 EXPORT_SYMBOL(strcmp);
 EXPORT_SYMBOL(strncmp);
+EXPORT_SYMBOL(strlen);
+EXPORT_SYMBOL(strnlen);
 EXPORT_SYMBOL(memset);
 EXPORT_SYMBOL(memcpy);
 EXPORT_SYMBOL(memmove);
diff --git a/arch/arm64/lib/Makefile b/arch/arm64/lib/Makefile
index 5ded21f..e64a94c 100644
--- a/arch/arm64/lib/Makefile
+++ b/arch/arm64/lib/Makefile
@@ -3,4 +3,5 @@ lib-y		:= bitops.o delay.o					\
 		   copy_from_user.o copy_to_user.o copy_in_user.o	\
 		   copy_page.o clear_page.o				\
 		   memchr.o memcpy.o memmove.o memset.o memcmp.o	\
-		   strchr.o strrchr.o strcmp.o strncmp.o
+		   strchr.o strrchr.o strcmp.o strncmp.o		\
+		   strlen.o strnlen.o
diff --git a/arch/arm64/lib/strlen.S b/arch/arm64/lib/strlen.S
new file mode 100644
index 0000000..302f6dd
--- /dev/null
+++ b/arch/arm64/lib/strlen.S
@@ -0,0 +1,131 @@
+/*
+ * Copyright (C) 2013 ARM Ltd.
+ * Copyright (C) 2013 Linaro.
+ *
+ * This code is based on glibc cortex strings work originally authored by Linaro
+ * and re-licensed under GPLv2 for the Linux kernel. The original code can
+ * be found @
+ *
+ * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
+ * files/head:/src/aarch64/
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+
+/*
+ * calculate the length of a string
+ *
+ * Parameters:
+ *	x0 - const string pointer
+ * Returns:
+ *	x0 - the return length of specific string
+ */
+
+/* Arguments and results.  */
+#define srcin		x0
+#define len		x0
+
+/* Locals and temporaries.  */
+#define src		x1
+#define data1		x2
+#define data2		x3
+#define data2a		x4
+#define has_nul1	x5
+#define has_nul2	x6
+#define tmp1		x7
+#define tmp2		x8
+#define tmp3		x9
+#define tmp4		x10
+#define zeroones	x11
+#define pos		x12
+
+#define REP8_01 0x0101010101010101
+#define REP8_7f 0x7f7f7f7f7f7f7f7f
+#define REP8_80 0x8080808080808080
+
+ENTRY(strlen)
+	mov	zeroones, #REP8_01
+	bic	src, srcin, #15
+	ands	tmp1, srcin, #15
+	b.ne	.Lmisaligned
+	/*
+	* NUL detection works on the principle that (X - 1) & (~X) & 0x80
+	* (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
+	* can be done in parallel across the entire word.
+	*/
+	/*
+	* The inner loop deals with two Dwords at a time. This has a
+	* slightly higher start-up cost, but we should win quite quickly,
+	* especially on cores with a high number of issue slots per
+	* cycle, as we get much better parallelism out of the operations.
+	*/
+.Lloop:
+	ldp	data1, data2, [src], #16
+.Lrealigned:
+	sub	tmp1, data1, zeroones
+	orr	tmp2, data1, #REP8_7f
+	sub	tmp3, data2, zeroones
+	orr	tmp4, data2, #REP8_7f
+	bic	has_nul1, tmp1, tmp2
+	bics	has_nul2, tmp3, tmp4
+	ccmp	has_nul1, #0, #0, eq	/* NZCV = 0000  */
+	b.eq	.Lloop
+
+	sub	len, src, srcin
+	cbz	has_nul1, .Lnul_in_data2
+#ifdef __ARM64EB__
+	mov	data2, data1
+#endif
+	sub	len, len, #8
+	mov	has_nul2, has_nul1
+.Lnul_in_data2:
+#ifdef __ARM64EB__
+	/*
+	* For big-endian, carry propagation (if the final byte in the
+	* string is 0x01) means we cannot use has_nul directly.  The
+	* easiest way to get the correct byte is to byte-swap the data
+	* and calculate the syndrome a second time.
+	*/
+	rev	data2, data2
+	sub	tmp1, data2, zeroones
+	orr	tmp2, data2, #REP8_7f
+	bic	has_nul2, tmp1, tmp2
+#endif
+	sub	len, len, #8
+	rev	has_nul2, has_nul2
+	clz	pos, has_nul2
+	add	len, len, pos, lsr #3		/* Bits to bytes.  */
+	ret
+
+.Lmisaligned:
+	cmp	tmp1, #8
+	neg	tmp1, tmp1
+	ldp	data1, data2, [src], #16
+	lsl	tmp1, tmp1, #3		/* Bytes beyond alignment -> bits.  */
+	mov	tmp2, #~0
+#ifdef __ARM64EB__
+	/* Big-endian.  Early bytes are at MSB.  */
+	lsl	tmp2, tmp2, tmp1	/* Shift (tmp1 & 63).  */
+#else
+	/* Little-endian.  Early bytes are at LSB.  */
+	lsr	tmp2, tmp2, tmp1	/* Shift (tmp1 & 63).  */
+#endif
+	orr	data1, data1, tmp2
+	orr	data2a, data2, tmp2
+	csinv	data1, data1, xzr, le
+	csel	data2, data2, data2a, le
+	b	.Lrealigned
+ENDPROC(strlen)
diff --git a/arch/arm64/lib/strnlen.S b/arch/arm64/lib/strnlen.S
new file mode 100644
index 0000000..2293f7b
--- /dev/null
+++ b/arch/arm64/lib/strnlen.S
@@ -0,0 +1,179 @@
+/*
+ * Copyright (C) 2013 ARM Ltd.
+ * Copyright (C) 2013 Linaro.
+ *
+ * This code is based on glibc cortex strings work originally authored by Linaro
+ * and re-licensed under GPLv2 for the Linux kernel. The original code can
+ * be found @
+ *
+ * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
+ * files/head:/src/aarch64/
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+
+/*
+ * determine the length of a fixed-size string
+ *
+ * Parameters:
+ *	x0 - const string pointer
+ *	x1 - maximal string length
+ * Returns:
+ *	x0 - the return length of specific string
+ */
+
+/* Arguments and results.  */
+#define srcin		x0
+#define len		x0
+#define limit		x1
+
+/* Locals and temporaries.  */
+#define src		x2
+#define data1		x3
+#define data2		x4
+#define data2a		x5
+#define has_nul1	x6
+#define has_nul2	x7
+#define tmp1		x8
+#define tmp2		x9
+#define tmp3		x10
+#define tmp4		x11
+#define zeroones	x12
+#define pos		x13
+#define limit_wd	x14
+
+#define REP8_01 0x0101010101010101
+#define REP8_7f 0x7f7f7f7f7f7f7f7f
+#define REP8_80 0x8080808080808080
+
+ENTRY(strnlen)
+	cbz	limit, .Lhit_limit
+	mov	zeroones, #REP8_01
+	bic	src, srcin, #15
+	ands	tmp1, srcin, #15
+	b.ne	.Lmisaligned
+	/* Calculate the number of full and partial words -1.  */
+	sub	limit_wd, limit, #1 /* Limit != 0, so no underflow.  */
+	lsr	limit_wd, limit_wd, #4  /* Convert to Qwords.  */
+
+	/*
+	* NUL detection works on the principle that (X - 1) & (~X) & 0x80
+	* (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
+	* can be done in parallel across the entire word.
+	*/
+	/*
+	* The inner loop deals with two Dwords at a time.  This has a
+	* slightly higher start-up cost, but we should win quite quickly,
+	* especially on cores with a high number of issue slots per
+	* cycle, as we get much better parallelism out of the operations.
+	*/
+.Lloop:
+	ldp	data1, data2, [src], #16
+.Lrealigned:
+	sub	tmp1, data1, zeroones
+	orr	tmp2, data1, #REP8_7f
+	sub	tmp3, data2, zeroones
+	orr	tmp4, data2, #REP8_7f
+	bic	has_nul1, tmp1, tmp2
+	bic	has_nul2, tmp3, tmp4
+	subs	limit_wd, limit_wd, #1
+	orr	tmp1, has_nul1, has_nul2
+	ccmp	tmp1, #0, #0, pl    /* NZCV = 0000  */
+	b.eq	.Lloop
+
+	cbz	tmp1, .Lhit_limit   /* No null in final Qword.  */
+
+	/*
+	* We know there's a null in the final Qword. The easiest thing
+	* to do now is work out the length of the string and return
+	* MIN (len, limit).
+	*/
+	sub	len, src, srcin
+	cbz	has_nul1, .Lnul_in_data2
+#ifdef __ARM64EB__
+	mov	data2, data1
+#endif
+	sub	len, len, #8
+	mov	has_nul2, has_nul1
+.Lnul_in_data2:
+#ifdef __ARM64EB__
+	/*
+	* For big-endian, carry propagation (if the final byte in the
+	* string is 0x01) means we cannot use has_nul directly.  The
+	* easiest way to get the correct byte is to byte-swap the data
+	* and calculate the syndrome a second time.
+	*/
+	rev	data2, data2
+	sub	tmp1, data2, zeroones
+	orr	tmp2, data2, #REP8_7f
+	bic	has_nul2, tmp1, tmp2
+#endif
+	sub	len, len, #8
+	rev	has_nul2, has_nul2
+	clz	pos, has_nul2
+	add	len, len, pos, lsr #3       /* Bits to bytes.  */
+	cmp	len, limit
+	csel	len, len, limit, ls     /* Return the lower value.  */
+	ret
+
+.Lmisaligned:
+	/*
+	* Deal with a partial first word.
+	* We're doing two things in parallel here;
+	* 1) Calculate the number of words (but avoiding overflow if
+	* limit is near ULONG_MAX) - to do this we need to work out
+	* limit + tmp1 - 1 as a 65-bit value before shifting it;
+	* 2) Load and mask the initial data words - we force the bytes
+	* before the ones we are interested in to 0xff - this ensures
+	* early bytes will not hit any zero detection.
+	*/
+	/*
+	* this operation need some cycles,
+	* some other independent instructions can following it.
+	*/
+	ldp	data1, data2, [src], #16
+
+	sub	limit_wd, limit, #1
+	and	tmp3, limit_wd, #15
+	lsr	limit_wd, limit_wd, #4
+
+	add	tmp3, tmp3, tmp1
+	add	limit_wd, limit_wd, tmp3, lsr #4
+
+	neg	tmp4, tmp1
+	lsl	tmp4, tmp4, #3  /* Bytes beyond alignment -> bits.  */
+
+	mov	tmp2, #~0
+#ifdef __ARM64EB__
+	/* Big-endian.  Early bytes are at MSB.  */
+	lsl	tmp2, tmp2, tmp4	/* Shift (tmp1 & 63).  */
+#else
+	/* Little-endian.  Early bytes are at LSB.  */
+	lsr	tmp2, tmp2, tmp4	/* Shift (tmp1 & 63).  */
+#endif
+	cmp	tmp1, #8
+
+	orr	data1, data1, tmp2
+	orr	data2a, data2, tmp2
+
+	csinv	data1, data1, xzr, le
+	csel	data2, data2, data2a, le
+	b	.Lrealigned
+
+.Lhit_limit:
+	mov	len, limit
+	ret
+ENDPROC(strnlen)
-- 
1.7.9.5