[ARM] 2947/1: copy template with new memcpy/memmove
Patch from Nicolas Pitre
This patch provides a new implementation for optimized memory copy
functions on ARM. It is made of two levels: a template that consists of
the core copy code and separate files that define macros to be used with
the core code depending on the type of copy needed. This allows for best
performances while sharing the same core for implementing memcpy(),
copy_from_user() and copy_to_user() for instance.
Two reasons for this work:
1) the current copy_to_user/copy_from_user implementation assumes no
task switch will ever occur in the middle of each copied page making
it completely unsafe with CONFIG_PREEMPT=y.
2) current copy implementations are measurably suboptimal and optimizing
different implementations separately is a pain and more opportunities
for bugs.
The reason for (1) is the fact that copy inside user pages are performed
with the ldm instruction which has no mean for testing user protections
and could possibly race with process preemption bypassing the COW mechanism
for example. This is a longstanding issue that we said ought to be fixed
for about two years now. The solution is to substitute those ldm insns
with a series of ldrt or strt insns to enforce user memory protection.
At least on StrongARM and XScale cores the ldm is not faster than the
equivalent ldr/str insns with a warm i-cache so there is no measurable
performance degradation with that change. The fact that the copy code is
a template makes it pretty easy to reuse the same core code as for memcpy
and benefit from the same performance optimizations.
Now (2) is best demonstrated with actual throughput measurements.
First, here is a summary of memcopy tests performed on a StrongARM core:
PTR alignment buffer size kernel version this version
------------------------------------------------------------
aligned 32 59.73 107.43
unaligned 32 61.31 74.72
aligned 100 132.47 136.15
unaligned 100 103.84 123.76
aligned 4096 130.67 130.80
unaligned 4096 130.68 130.64
aligned 1048576 68.03 68.18
unaligned 1048576 68.03 68.18
The buffer size is in bytes and the measured speed in MB/s. The copy
was performed repeatedly with given buffer and throughput averaged over
3 seconds.
Here we can see that the current kernel version has a higher entry cost
that shows up with small buffers. As buffer size grows both implementation
converge to the same throughput.
Now here's the exact same test performed on an XScale core (PXA255):
PTR alignment buffer size kernel version this version
------------------------------------------------------------
aligned 32 46.99 77.58
unaligned 32 53.61 59.59
aligned 100 107.19 136.59
unaligned 100 83.61 97.58
aligned 4096 129.13 129.98
unaligned 4096 128.36 128.53
aligned 1048576 53.76 59.41
unaligned 1048576 33.67 56.96
Again we can see the entry setup cost being higher for the current kernel
before getting to the main copy loop. Then throughput results converge
as long as the buffer remains in the cache. Then the 1MB case shows more
differences probably due to better pld placement and/or less instruction
interlocks in this proposed implementation.
Disclaimer: The PXA system was running with slower clocks than the
StrongARM system so trying to infer any conclusion by comparing those
separate sets of results side by side would be completely inappropriate.
So... What this patch does is to replace both memcpy and memmove with
an implementation based on the provided copy code template. The memmove
code is kept separate since it is used only if the memory areas involved
do overlap in which case the code is a transposition of the template but
with the copy occurring in the opposite direction (trying to fit that
mode into the template turned it into a mess not worth it for memmove
alone). And obviously both memcpy and memmove were tested with all kinds
of pointer alignments and buffer sizes to exercise all code paths for
correctness.
The next patch will provide the now trivial replacement implementation
copy_to_user and copy_from_user.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
diff --git a/arch/arm/lib/memcpy.S b/arch/arm/lib/memcpy.S
index f5a593c..7e71d67 100644
--- a/arch/arm/lib/memcpy.S
+++ b/arch/arm/lib/memcpy.S
@@ -1,393 +1,59 @@
/*
* linux/arch/arm/lib/memcpy.S
*
- * Copyright (C) 1995-1999 Russell King
+ * Author: Nicolas Pitre
+ * Created: Sep 28, 2005
+ * Copyright: MontaVista Software, Inc.
*
- * 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.
- *
- * ASM optimised string functions
+ * 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.
*/
+
#include <linux/linkage.h>
#include <asm/assembler.h>
- .text
+ .macro ldr1w ptr reg abort
+ ldr \reg, [\ptr], #4
+ .endm
-#define ENTER \
- mov ip,sp ;\
- stmfd sp!,{r0,r4-r9,fp,ip,lr,pc} ;\
- sub fp,ip,#4
+ .macro ldr4w ptr reg1 reg2 reg3 reg4 abort
+ ldmia \ptr!, {\reg1, \reg2, \reg3, \reg4}
+ .endm
-#define EXIT \
- LOADREGS(ea, fp, {r0, r4 - r9, fp, sp, pc})
+ .macro ldr8w ptr reg1 reg2 reg3 reg4 reg5 reg6 reg7 reg8 abort
+ ldmia \ptr!, {\reg1, \reg2, \reg3, \reg4, \reg5, \reg6, \reg7, \reg8}
+ .endm
-#define EXITEQ \
- LOADREGS(eqea, fp, {r0, r4 - r9, fp, sp, pc})
+ .macro ldr1b ptr reg cond=al abort
+ ldr\cond\()b \reg, [\ptr], #1
+ .endm
-/*
- * Prototype: void memcpy(void *to,const void *from,unsigned long n);
- */
+ .macro str1w ptr reg abort
+ str \reg, [\ptr], #4
+ .endm
+
+ .macro str8w ptr reg1 reg2 reg3 reg4 reg5 reg6 reg7 reg8 abort
+ stmia \ptr!, {\reg1, \reg2, \reg3, \reg4, \reg5, \reg6, \reg7, \reg8}
+ .endm
+
+ .macro str1b ptr reg cond=al abort
+ str\cond\()b \reg, [\ptr], #1
+ .endm
+
+ .macro enter reg1 reg2
+ stmdb sp!, {r0, \reg1, \reg2}
+ .endm
+
+ .macro exit reg1 reg2
+ ldmfd sp!, {r0, \reg1, \reg2}
+ .endm
+
+ .text
+
+/* Prototype: void *memcpy(void *dest, const void *src, size_t n); */
+
ENTRY(memcpy)
-ENTRY(memmove)
- ENTER
- cmp r1, r0
- bcc 23f
- subs r2, r2, #4
- blt 6f
- PLD( pld [r1, #0] )
- ands ip, r0, #3
- bne 7f
- ands ip, r1, #3
- bne 8f
-1: subs r2, r2, #8
- blt 5f
- subs r2, r2, #20
- blt 4f
- PLD( pld [r1, #28] )
- PLD( subs r2, r2, #64 )
- PLD( blt 3f )
-2: PLD( pld [r1, #60] )
- PLD( pld [r1, #92] )
- ldmia r1!, {r3 - r9, ip}
- subs r2, r2, #32
- stmgeia r0!, {r3 - r9, ip}
- ldmgeia r1!, {r3 - r9, ip}
- subges r2, r2, #32
- stmia r0!, {r3 - r9, ip}
- bge 2b
-3: PLD( ldmia r1!, {r3 - r9, ip} )
- PLD( adds r2, r2, #32 )
- PLD( stmgeia r0!, {r3 - r9, ip} )
- PLD( ldmgeia r1!, {r3 - r9, ip} )
- PLD( subges r2, r2, #32 )
- PLD( stmia r0!, {r3 - r9, ip} )
-4: cmn r2, #16
- ldmgeia r1!, {r3 - r6}
- subge r2, r2, #16
- stmgeia r0!, {r3 - r6}
- adds r2, r2, #20
- ldmgeia r1!, {r3 - r5}
- subge r2, r2, #12
- stmgeia r0!, {r3 - r5}
-5: adds r2, r2, #8
- blt 6f
- subs r2, r2, #4
- ldrlt r3, [r1], #4
- ldmgeia r1!, {r4, r5}
- subge r2, r2, #4
- strlt r3, [r0], #4
- stmgeia r0!, {r4, r5}
-
-6: adds r2, r2, #4
- EXITEQ
- cmp r2, #2
- ldrb r3, [r1], #1
- ldrgeb r4, [r1], #1
- ldrgtb r5, [r1], #1
- strb r3, [r0], #1
- strgeb r4, [r0], #1
- strgtb r5, [r0], #1
- EXIT
-
-7: rsb ip, ip, #4
- cmp ip, #2
- ldrb r3, [r1], #1
- ldrgeb r4, [r1], #1
- ldrgtb r5, [r1], #1
- strb r3, [r0], #1
- strgeb r4, [r0], #1
- strgtb r5, [r0], #1
- subs r2, r2, ip
- blt 6b
- ands ip, r1, #3
- beq 1b
-
-8: bic r1, r1, #3
- ldr r7, [r1], #4
- cmp ip, #2
- bgt 18f
- beq 13f
- cmp r2, #12
- blt 11f
- PLD( pld [r1, #12] )
- sub r2, r2, #12
- PLD( subs r2, r2, #32 )
- PLD( blt 10f )
- PLD( pld [r1, #28] )
-9: PLD( pld [r1, #44] )
-10: mov r3, r7, pull #8
- ldmia r1!, {r4 - r7}
- subs r2, r2, #16
- orr r3, r3, r4, push #24
- mov r4, r4, pull #8
- orr r4, r4, r5, push #24
- mov r5, r5, pull #8
- orr r5, r5, r6, push #24
- mov r6, r6, pull #8
- orr r6, r6, r7, push #24
- stmia r0!, {r3 - r6}
- bge 9b
- PLD( cmn r2, #32 )
- PLD( bge 10b )
- PLD( add r2, r2, #32 )
- adds r2, r2, #12
- blt 12f
-11: mov r3, r7, pull #8
- ldr r7, [r1], #4
- subs r2, r2, #4
- orr r3, r3, r7, push #24
- str r3, [r0], #4
- bge 11b
-12: sub r1, r1, #3
- b 6b
-
-13: cmp r2, #12
- blt 16f
- PLD( pld [r1, #12] )
- sub r2, r2, #12
- PLD( subs r2, r2, #32 )
- PLD( blt 15f )
- PLD( pld [r1, #28] )
-14: PLD( pld [r1, #44] )
-15: mov r3, r7, pull #16
- ldmia r1!, {r4 - r7}
- subs r2, r2, #16
- orr r3, r3, r4, push #16
- mov r4, r4, pull #16
- orr r4, r4, r5, push #16
- mov r5, r5, pull #16
- orr r5, r5, r6, push #16
- mov r6, r6, pull #16
- orr r6, r6, r7, push #16
- stmia r0!, {r3 - r6}
- bge 14b
- PLD( cmn r2, #32 )
- PLD( bge 15b )
- PLD( add r2, r2, #32 )
- adds r2, r2, #12
- blt 17f
-16: mov r3, r7, pull #16
- ldr r7, [r1], #4
- subs r2, r2, #4
- orr r3, r3, r7, push #16
- str r3, [r0], #4
- bge 16b
-17: sub r1, r1, #2
- b 6b
-
-18: cmp r2, #12
- blt 21f
- PLD( pld [r1, #12] )
- sub r2, r2, #12
- PLD( subs r2, r2, #32 )
- PLD( blt 20f )
- PLD( pld [r1, #28] )
-19: PLD( pld [r1, #44] )
-20: mov r3, r7, pull #24
- ldmia r1!, {r4 - r7}
- subs r2, r2, #16
- orr r3, r3, r4, push #8
- mov r4, r4, pull #24
- orr r4, r4, r5, push #8
- mov r5, r5, pull #24
- orr r5, r5, r6, push #8
- mov r6, r6, pull #24
- orr r6, r6, r7, push #8
- stmia r0!, {r3 - r6}
- bge 19b
- PLD( cmn r2, #32 )
- PLD( bge 20b )
- PLD( add r2, r2, #32 )
- adds r2, r2, #12
- blt 22f
-21: mov r3, r7, pull #24
- ldr r7, [r1], #4
- subs r2, r2, #4
- orr r3, r3, r7, push #8
- str r3, [r0], #4
- bge 21b
-22: sub r1, r1, #1
- b 6b
-
-
-23: add r1, r1, r2
- add r0, r0, r2
- subs r2, r2, #4
- blt 29f
- PLD( pld [r1, #-4] )
- ands ip, r0, #3
- bne 30f
- ands ip, r1, #3
- bne 31f
-
-24: subs r2, r2, #8
- blt 28f
- subs r2, r2, #20
- blt 27f
- PLD( pld [r1, #-32] )
- PLD( subs r2, r2, #64 )
- PLD( blt 26f )
-25: PLD( pld [r1, #-64] )
- PLD( pld [r1, #-96] )
- ldmdb r1!, {r3 - r9, ip}
- subs r2, r2, #32
- stmgedb r0!, {r3 - r9, ip}
- ldmgedb r1!, {r3 - r9, ip}
- subges r2, r2, #32
- stmdb r0!, {r3 - r9, ip}
- bge 25b
-26: PLD( ldmdb r1!, {r3 - r9, ip} )
- PLD( adds r2, r2, #32 )
- PLD( stmgedb r0!, {r3 - r9, ip} )
- PLD( ldmgedb r1!, {r3 - r9, ip} )
- PLD( subges r2, r2, #32 )
- PLD( stmdb r0!, {r3 - r9, ip} )
-27: cmn r2, #16
- ldmgedb r1!, {r3 - r6}
- subge r2, r2, #16
- stmgedb r0!, {r3 - r6}
- adds r2, r2, #20
- ldmgedb r1!, {r3 - r5}
- subge r2, r2, #12
- stmgedb r0!, {r3 - r5}
-28: adds r2, r2, #8
- blt 29f
- subs r2, r2, #4
- ldrlt r3, [r1, #-4]!
- ldmgedb r1!, {r4, r5}
- subge r2, r2, #4
- strlt r3, [r0, #-4]!
- stmgedb r0!, {r4, r5}
-
-29: adds r2, r2, #4
- EXITEQ
- cmp r2, #2
- ldrb r3, [r1, #-1]!
- ldrgeb r4, [r1, #-1]!
- ldrgtb r5, [r1, #-1]!
- strb r3, [r0, #-1]!
- strgeb r4, [r0, #-1]!
- strgtb r5, [r0, #-1]!
- EXIT
-
-30: cmp ip, #2
- ldrb r3, [r1, #-1]!
- ldrgeb r4, [r1, #-1]!
- ldrgtb r5, [r1, #-1]!
- strb r3, [r0, #-1]!
- strgeb r4, [r0, #-1]!
- strgtb r5, [r0, #-1]!
- subs r2, r2, ip
- blt 29b
- ands ip, r1, #3
- beq 24b
-
-31: bic r1, r1, #3
- ldr r3, [r1], #0
- cmp ip, #2
- blt 41f
- beq 36f
- cmp r2, #12
- blt 34f
- PLD( pld [r1, #-16] )
- sub r2, r2, #12
- PLD( subs r2, r2, #32 )
- PLD( blt 33f )
- PLD( pld [r1, #-32] )
-32: PLD( pld [r1, #-48] )
-33: mov r7, r3, push #8
- ldmdb r1!, {r3, r4, r5, r6}
- subs r2, r2, #16
- orr r7, r7, r6, pull #24
- mov r6, r6, push #8
- orr r6, r6, r5, pull #24
- mov r5, r5, push #8
- orr r5, r5, r4, pull #24
- mov r4, r4, push #8
- orr r4, r4, r3, pull #24
- stmdb r0!, {r4, r5, r6, r7}
- bge 32b
- PLD( cmn r2, #32 )
- PLD( bge 33b )
- PLD( add r2, r2, #32 )
- adds r2, r2, #12
- blt 35f
-34: mov ip, r3, push #8
- ldr r3, [r1, #-4]!
- subs r2, r2, #4
- orr ip, ip, r3, pull #24
- str ip, [r0, #-4]!
- bge 34b
-35: add r1, r1, #3
- b 29b
-
-36: cmp r2, #12
- blt 39f
- PLD( pld [r1, #-16] )
- sub r2, r2, #12
- PLD( subs r2, r2, #32 )
- PLD( blt 38f )
- PLD( pld [r1, #-32] )
-37: PLD( pld [r1, #-48] )
-38: mov r7, r3, push #16
- ldmdb r1!, {r3, r4, r5, r6}
- subs r2, r2, #16
- orr r7, r7, r6, pull #16
- mov r6, r6, push #16
- orr r6, r6, r5, pull #16
- mov r5, r5, push #16
- orr r5, r5, r4, pull #16
- mov r4, r4, push #16
- orr r4, r4, r3, pull #16
- stmdb r0!, {r4, r5, r6, r7}
- bge 37b
- PLD( cmn r2, #32 )
- PLD( bge 38b )
- PLD( add r2, r2, #32 )
- adds r2, r2, #12
- blt 40f
-39: mov ip, r3, push #16
- ldr r3, [r1, #-4]!
- subs r2, r2, #4
- orr ip, ip, r3, pull #16
- str ip, [r0, #-4]!
- bge 39b
-40: add r1, r1, #2
- b 29b
-
-41: cmp r2, #12
- blt 44f
- PLD( pld [r1, #-16] )
- sub r2, r2, #12
- PLD( subs r2, r2, #32 )
- PLD( blt 43f )
- PLD( pld [r1, #-32] )
-42: PLD( pld [r1, #-48] )
-43: mov r7, r3, push #24
- ldmdb r1!, {r3, r4, r5, r6}
- subs r2, r2, #16
- orr r7, r7, r6, pull #8
- mov r6, r6, push #24
- orr r6, r6, r5, pull #8
- mov r5, r5, push #24
- orr r5, r5, r4, pull #8
- mov r4, r4, push #24
- orr r4, r4, r3, pull #8
- stmdb r0!, {r4, r5, r6, r7}
- bge 42b
- PLD( cmn r2, #32 )
- PLD( bge 43b )
- PLD( add r2, r2, #32 )
- adds r2, r2, #12
- blt 45f
-44: mov ip, r3, push #24
- ldr r3, [r1, #-4]!
- subs r2, r2, #4
- orr ip, ip, r3, pull #8
- str ip, [r0, #-4]!
- bge 44b
-45: add r1, r1, #1
- b 29b
+#include "copy_template.S"