2015-05-20 19:06:07

by Shailendra Verma

[permalink] [raw]
Subject: [PATCH] Input:evdev - Fix to avoid the execution of same instruction in every iteration of for loop.

Here in for loop the instruction "len / sizeof(compat_long_t)" is used
as a terminating condition which is executing (being computed) in every
iteration of for loop. Below is the armv7-a architecture assembly code with
similar for loop having the same instruction in condition.Note that the
instruction "lsr r3, r3, #2" i.e Logical Shift Right on behalf of
"len / sizeof(compat_long_t)" is getting executed every time the loop
iterates.

.syntax unified
.arch armv7-a
.eabi_attribute 27, 3
.fpu vfpv3-d16
.eabi_attribute 20, 1
.eabi_attribute 21, 1
.eabi_attribute 23, 3
.eabi_attribute 24, 1
.eabi_attribute 25, 1
.eabi_attribute 26, 2
.eabi_attribute 30, 6
.eabi_attribute 34, 1
.eabi_attribute 18, 4
.thumb
.file "test.c"
.section .rodata
.align 2
.LC0:
.ascii "Shail\000"
.text
.align 2
.global main
.thumb
.thumb_func
.type main, %function
main:
@ args = 0, pretend = 0, frame = 8
@ frame_needed = 1, uses_anonymous_args = 0
push {r7, lr}
sub sp, sp, #8
add r7, sp, #0
mov r3, #1000
str r3, [r7, #4]
mov r3, #0
str r3, [r7, #0]
b .L2
.L3:
movw r3, #:lower16:.LC0
movt r3, #:upper16:.LC0
mov r0, r3
bl printf
ldr r3, [r7, #0]
add r3, r3, #1
str r3, [r7, #0]
.L2:
ldr r2, [r7, #0]
ldr r3, [r7, #4]
lsr r3, r3, #2 -------> Instruction "len/sizeof(compat_long_t)"
cmp r2, r3 will execute in every loop iteration.
bcc .L3
mov r3, #0
mov r0, r3
add r7, r7, #8
mov sp, r7
pop {r7, pc}
.size main, .-main
.ident "GCC: (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3"
.section .note.GNU-stack,"",%progbits

This fix will avoid the execution( computation) of same instruction in
every for loop iteration and will perform the instruction computation
before the for loop. Below is the armv7-a assembly code after changes.
In this we can note that the "lsr r3, r3, #2" i.e. Logical Shift Right
instruction on behalf of "len / sizeof(compat_long_t)" is getting
executed only once before the loop.

.syntax unified
.arch armv7-a
.eabi_attribute 27, 3
.fpu vfpv3-d16
.eabi_attribute 20, 1
.eabi_attribute 21, 1
.eabi_attribute 23, 3
.eabi_attribute 24, 1
.eabi_attribute 25, 1
.eabi_attribute 26, 2
.eabi_attribute 30, 6
.eabi_attribute 34, 1
.eabi_attribute 18, 4
.thumb
.file "test.c"
.section .rodata
.align 2
.LC0:
.ascii "Shail\000"
.text
.align 2
.global main
.thumb
.thumb_func
.type main, %function
main:
@ args = 0, pretend = 0, frame = 8
@ frame_needed = 1, uses_anonymous_args = 0
push {r7, lr}
sub sp, sp, #8
add r7, sp, #0
mov r3, #1000
str r3, [r7, #4]
ldr r3, [r7, #4]
lsr r3, r3, #2 ------->Instruction "len/sizeof(compat_long_t)"
str r3, [r7, #4] will execute only once before for loop.
mov r3, #0
str r3, [r7, #0]
b .L2
.L3:
movw r3, #:lower16:.LC0
movt r3, #:upper16:.LC0
mov r0, r3
bl printf
ldr r3, [r7, #0]
add r3, r3, #1
str r3, [r7, #0]
.L2:
ldr r2, [r7, #0]
ldr r3, [r7, #4]
cmp r2, r3
blt .L3
mov r3, #0
mov r0, r3
add r7, r7, #8
mov sp, r7
pop {r7, pc}
.size main, .-main
.ident "GCC: (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3"
.section .note.GNU-stack,"",%progbits

Thus the fix will save duplicate instruction execution in every
iteration of for loop.

Signed-off-by: Shailendra Verma <[email protected]>
---
drivers/input/evdev.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/input/evdev.c b/drivers/input/evdev.c
index dcc588e..bea576f 100644
--- a/drivers/input/evdev.c
+++ b/drivers/input/evdev.c
@@ -618,7 +618,9 @@ static int bits_to_user(unsigned long *bits, unsigned int maxbit,
if (len > maxlen)
len = maxlen;

- for (i = 0; i < len / sizeof(compat_long_t); i++)
+ maxlen = len / sizeof(compat_long_t);
+
+ for (i = 0; i < maxlen; i++)
if (copy_to_user((compat_long_t __user *) p + i,
(compat_long_t *) bits +
i + 1 - ((i % 2) << 1),
--
1.7.9.5


2015-05-20 19:36:51

by Dmitry Torokhov

[permalink] [raw]
Subject: Re: [PATCH] Input:evdev - Fix to avoid the execution of same instruction in every iteration of for loop.

Hi Shailendra,

On Thu, May 21, 2015 at 12:35:46AM +0530, Shailendra Verma wrote:
> Here in for loop the instruction "len / sizeof(compat_long_t)" is used
> as a terminating condition which is executing (being computed) in every
> iteration of for loop. Below is the armv7-a architecture assembly code with
> similar for loop having the same instruction in condition.Note that the
> instruction "lsr r3, r3, #2" i.e Logical Shift Right on behalf of
> "len / sizeof(compat_long_t)" is getting executed every time the loop
> iterates.

No, it is not. We compile kernel with -O2 and compiler is smart enough
to optimize such constructs.

Thanks.

--
Dmitry