dm: uninitialize static variables
Subsystems (2)
PATCH 1/1 dm: uninitialize static variables
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
--- dm.c | 4 ++-- kcopyd.c | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/md/dm.c +2 -2
@@ -20,8 +20,8 @@static const char *_name =DM_NAME ; -static unsigned int major =0 ; -static unsigned int _major =0 ; +static unsigned int major;/* = 0 */ +static unsigned int _major;/* = 0 */ /* * One of these is allocated per bio.
drivers/md/kcopyd.c +1 -1
@@ -588,7 +588,7 @@ static void client_del(struct kcopyd_cli }static DECLARE_MUTEX (kcopyd_init_lock); -static int kcopyd_clients =0 ; +static int kcopyd_clients;/* = 0 */ static int kcopyd_init (void ) {
_______________________________________________
Kernel-janitors mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/kernel-janitors
Hi,
Quoted 3 lines
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
if you want that why not explicitely put them in .bss? something like:
drivers/md/dm.c line 23
Quoted 2 lines
drivers/md/dm.c +1
-1-static unsigned int major =0 ; +static unsigned int major;/* = 0 */
static unsigned int major = 0 __attribute__ ((section(".bss"));
Even when crt should take care of initializing bss to zero, so I think that you're trying to save some cpu cycles
my $1
Signature · 2 lines
Carlos Manuel Duclos Vergara http://www.toolchains.com/personal/blog
* "Carlos Manuel Duclos Vergara" <[email protected]>:
Quoted 11 lines
Quoted 3 lines
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
if you want that why not explicitely put them in .bss? something like:
Review
Quoted 2 lines
(quoted diff) +1
-1-static unsigned int major = 0; +static unsigned int major; /* = 0 */static unsigned int major = 0
__attribute__((section(".bss"));
I personally like the idea of explicitly expressing this intention, but I didn't manage to achieve this together with initialization. I tried
static unsigned int __attribute__((__section__(".bss"))) major = 0; static unsigned int __attribute__((__section__(".bss"))) _major = 0;
but they appear not to end up in .bss nevertheless, compared with the uninitialized version:
dm-mod.ko2.txt 2005-12-05 18:39:52.000000000 +0100 renamed from dm-mod.ko1.txt 2005-12-05 18:40:00.000000000 +0100 +1 -1
@@ -27,7 +27,7 @@ [22] .rel.init.data REL 00000000 00b174 000028 8 29 21 4 [23] .gnu.linkonce.this_module PROGBITS 00000000 0087c0 000160 0 WA 0 0 32 [24] .rel.gnu.linkonce.this_module REL 00000000 00b19c 000010 8 29 23 4 -[25] .bss NOBITS 00000000 008920 00049c 0 WA 0 0 32 +[25] .bss NOBITS 00000000 008920 000498 0 WA 0 0 32 [26] .comment PROGBITS 00000000 008920 00022e 0 0 0 1 [27] .note.GNU-stack PROGBITS 00000000 008b4e 000000 0 0 0 1 [28] .shstrtab STRTAB 00000000 008b4e 00011e 0 0 0 1
If you could tell me how to achieve this in a fancy way which also works, I'm all ears. Maybe the janitorial steering committee might also like the idea enough to adjust their current TODO list.
( http://janitor.kernelnewbies.org/TODO - 10th task in 6th section)
Cheers,
n.
_______________________________________________
Kernel-janitors mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/kernel-janitors
Quoted 2 lines
static unsigned int __attribute__((__section__(".bss"))) major = 0; static unsigned int __attribute__((__section__(".bss")))
_major= 0;
it should be:
static unsigned int major __attribute__ ((section(".bss"))) = 0;
Quoted 3 lines
If you could tell me how to achieve this in a fancy way which also works, I'm all ears. Maybe the janitorial steering committee might also like the idea enough to adjust their current TODO list.
yup, maybe it's worth of a discussion on the list
Signature · 2 lines
Carlos Manuel Duclos Vergara http://www.toolchains.com/personal/blog
Quoted 8 lines · Nicolas Kaiser
* "Carlos Manuel Duclos Vergara" <[email protected]>:
Quoted 7 lines
Quoted 3 lines
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
if you want that why not explicitely put them in .bss? something like:
Quoted 7 lines
Quoted 4 lines
Review
Quoted 2 lines
(quoted diff) +1
-1-static unsigned int major = 0; +static unsigned int major; /* = 0 */static unsigned int major = 0
__attribute__((section(".bss"));I personally like the idea of explicitly expressing this intention, but I didn't manage to achieve this together with initialization.
Not only you use gcc'ism, you also make code needlessly verbose.
_______________________________________________
Kernel-janitors mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/kernel-janitors
In reply to Nicolas Kaiser
Quoted 8 lines
* "Carlos Manuel Duclos Vergara" <[email protected]>:
Quoted 7 lines
Quoted 3 lines
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
if you want that why not explicitely put them in .bss? something like:
Quoted 8 lines
Quoted 4 lines
Review
Quoted 2 lines
(quoted diff) +1
-1-static unsigned int major = 0; +static unsigned int major; /* = 0 */static unsigned int major = 0
__attribute__((section(".bss"));I personally like the idea of explicitly expressing this intention, but I didn't manage to achieve this together with initialization. I tried
From http://janitor.kernelnewbies.org/TODO:
- uninitialize static variables initialized to 0, to make it go to the .bss, instead of .data.
__initdata inside functions has to be initialized, even if to 0 though.
Therefor I would go with
static unsigned int major; /* = 0 */
Let the compiler make the decision, it will make the correct one. The comment is just there to indicate that we haven't forgotten to initialise it. As soon as you explicitly initialise you force the compiler to place it somewhere other than bss.
Signature · 2 lines
There are 10 kinds of people in the world, those who understand binary, and those who don't.
Quoted 5 lines · Nicolas Kaiser
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
dm.c | 4 ++-- kcopyd.c | 2 +-
diffstat -p1 ;-)
drivers/md/dm.c line 24
Quoted 10 lines
drivers/md/dm.c +2
-2@@ -20,8 +20,8 @@static const char *_name =DM_NAME ; -static unsigned int major =0 ; -static unsigned int _major =0 ; +static unsigned int major;/* = 0 */ +static unsigned int _major;/* = 0 */
Drop the comments. major goes into .bss. .bss is initialized to 0.
drivers/md/kcopyd.c static void client_del(struct kcopyd_cli · line 591
Quoted 5 lines
drivers/md/kcopyd.c +1
-1-static int kcopyd_clients =0 ; +static int kcopyd_clients;/* = 0 */
_______________________________________________
Kernel-janitors mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/kernel-janitors
PATCH 1/1 dm: uninitialize static variables v2
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
--- drivers/md/dm.c | 4 ++-- drivers/md/kcopyd.c | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/md/dm.c +2 -2
@@ -20,8 +20,8 @@static const char *_name =DM_NAME ; -static unsigned int major =0 ; -static unsigned int _major =0 ; +static unsigned int major; +static unsigned int _major;/* * One of these is allocated per bio.
drivers/md/kcopyd.c +1 -1
@@ -588,7 +588,7 @@ static void client_del(struct kcopyd_cli }static DECLARE_MUTEX (kcopyd_init_lock); -static int kcopyd_clients =0 ; +static int kcopyd_clients;static int kcopyd_init (void ) {
_______________________________________________
Kernel-janitors mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/kernel-janitors
Quoted 27 lines · Jaco Kroon
In reply to Nicolas Kaiser
Quoted 13 lines
* "Carlos Manuel Duclos Vergara" <[email protected]>:
Quoted 11 lines
Quoted 3 lines
uninitialize static variables initialized to 0, to make them go to .bss instead of .data.
if you want that why not explicitely put them in .bss? something like:
drivers/md/dm.c line 23
Quoted 2 lines
drivers/md/dm.c +1
-1-static unsigned int major =0 ; +static unsigned int major;/* = 0 */ static unsigned int major = 0
__attribute__((section(".bss"));Quoted 5 lines
I personally like the idea of explicitly expressing this intention, but I didn't manage to achieve this together with initialization. I tried
- uninitialize static variables initialized to 0, to make it go to the .bss,
instead of .data.
__initdatainside functions has to be initialized, even if to 0
though.
Quoted 9 lines
Therefor I would go with
static unsigned int major; /* = 0 */
Let the compiler make the decision, it will make the correct one. The comment is just there to indicate that we haven't forgotten to initialise it. As soon as you explicitly initialise you force the compiler to place it somewhere other than bss.
The comment only indicates that you purposely did not initialize the variable. However, it does not indicate *why* you chose to do so like the section attribute would. What about:
static unsigned int major; /* __attribute__ ((section(".bss")) */
or
static unsigned int major; /* section(".bss") */
which is more explicit about your intentions, lets the compiler ultimately choose, and is not tied to gcc.
Quoted 8 lines
Signature · 7 lines
-- There are 10 kinds of people in the world, those who understand binary, and those who don't. _______________________________________________ Kernel-janitors mailing list [email protected] https://lists.osdl.org/mailman/listinfo/kernel-janitors
--
Jesse Millan
CNS Unix/Windows Team
Portland State University
Phone: (503) 725-3285
Homepage: www.system-calls.com
grep --recursive --ignore-case 'SHOULD WORK' /usr/src/linux/* | wc
_______________________________________________
Kernel-janitors mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/kernel-janitors