The patch titled
Subject: lib/list: tweak LIST_POISON2 for better code generation on x86_64
has been added to the -mm tree. Its filename is
tweak-list_poison2-for-better-code-generation-on-x86_64.patch
This patch should soon appear at
http://ozlabs.org/~akpm/mmots/broken-out/tweak-list_poison2-for-better-code-generation-on-x86_64.patch
and later at
http://ozlabs.org/~akpm/mmotm/broken-out/tweak-list_poison2-for-better-code-generation-on-x86_64.patch
Before you just go and hit "reply", please:
a) Consider who else should be cc'ed
b) Prefer to cc a suitable mailing list as well
c) Ideally: find the original patch on the mailing list and do a
reply-to-all to that, adding suitable additional cc's
*** Remember to use Documentation/process/submit-checklist.rst when testing your code ***
The -mm tree is included into linux-next and is updated
there every 3-4 working days
------------------------------------------------------
From: Alexey Dobriyan <adobriyan@xxxxxxxxx>
Subject: lib/list: tweak LIST_POISON2 for better code generation on x86_64
list_del() poisoning can generate 2 64-bit immediate loads but it also can
generate one 64-bit immediate load and an addition:
48 b8 00 01 00 00 00 00 ad de movabs rax,0xdead000000000100
48 89 47 58 mov QWORD PTR [rdi+0x58],rax
48 05 00 01 00 00 <=====> add rax,0x100
48 89 47 60 mov QWORD PTR [rdi+0x60],rax
However on x86_64 not all constants are equal: those within [-128, 127]
range can be added with shorter "add r64, imm32" instruction:
48 b8 00 01 00 00 00 00 ad de movabs rax,0xdead000000000100
48 89 47 58 mov QWORD PTR [rdi+0x58],rax
48 83 c0 22 <======> add rax,0x22
48 89 47 60 mov QWORD PTR [rdi+0x60],rax
Patch saves 2 bytes per some LIST_POISON2 usage.
(Slightly disappointing) space savings on F29 x86_64 config:
add/remove: 0/0 grow/shrink: 0/2164 up/down: 0/-5184 (-5184)
Function old new delta
zstd_get_workspace 548 546 -2
...
mlx4_delete_all_resources_for_slave 4826 4804 -22
Total: Before=83304131, After=83298947, chg -0.01%
New constants are:
0xdead000000000100
0xdead000000000122
Note: LIST_POISON1 can't be changed to ...11 because something in page
allocator requires low bit unset.
Link: http://lkml.kernel.org/r/20190513191502.GA8492@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@xxxxxxxxx>
Cc: Vasiliy Kulikov <segoon@xxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
include/linux/poison.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
--- a/include/linux/poison.h~tweak-list_poison2-for-better-code-generation-on-x86_64
+++ a/include/linux/poison.h
@@ -21,7 +21,7 @@
* non-initialized list entries.
*/
#define LIST_POISON1 ((void *) 0x100 + POISON_POINTER_DELTA)
-#define LIST_POISON2 ((void *) 0x200 + POISON_POINTER_DELTA)
+#define LIST_POISON2 ((void *) 0x122 + POISON_POINTER_DELTA)
/********** include/linux/timer.h **********/
/*
_
Patches currently in -mm which might be from adobriyan@xxxxxxxxx are
ramfs-support-o_tmpfile.patch
tweak-list_poison2-for-better-code-generation-on-x86_64.patch
