Re: [PATCH/RFC] don't store struct sysent::native_scno member if we aren't using it.

[Denys V. <dvl...@re...>]

On 02/20/2012 11:34 PM, Dmitry V. Levin wrote:
> On Tue, Feb 07, 2012 at 02:14:11PM +0100, Denys Vlasenko wrote:
>> On 02/07/2012 01:17 PM, Dmitry V. Levin wrote:
>>>> I noticed that this structure member is used only if
>>>> we have more than one personality.
>>>
>>> This native_scno thing is not so much used in the code than before.
>>> For example, there are 21 initialized native_scno entries in
>>> linux/i386/syscallent.h:
>>>
>>> 	{ 1,	TP,	sys_exit,		"_exit", SYS_exit }, /* 1 */
>>> 	{ 0,	TP,	sys_fork,		"fork", SYS_fork }, /* 2 */
>>> 	{ 3,	TD,	sys_read,		"read", SYS_read }, /* 3 */
>>> 	{ 3,	TD,	sys_write,		"write", SYS_write }, /* 4 */
>>> 	{ 3,	TP,	sys_waitpid,		"waitpid", SYS_wait4 }, /* 7 */
>>> 	{ 3,	TF|TP,	sys_execve,		"execve", SYS_execve }, /* 11 */
>>> 	{ 2,	TD,	sys_socketcall,		"socketcall", SYS_socketcall }, /* 102 */
>>> 	{ 4,	TP,	sys_wait4,		"wait4", SYS_wait4 }, /* 114 */
>>> 	{ 6,	TI,	sys_ipc,		"ipc", SYS_ipc }, /* 117 */
>>> 	{ 5,	TP,	sys_clone,		"clone", SYS_clone }, /* 120 */
>>> 	{ 3,	0,	sys_sysfs,		"sysfs", SYS_sysfs }, /* 135 */
>>> 	{ 3,	TD,	sys_readv,		"readv", SYS_readv }, /* 145 */
>>> 	{ 3,	TD,	sys_writev,		"writev", SYS_writev }, /* 146 */
>>> 	{ 5,	TD,	sys_pread,		"pread64", SYS_read }, /* 180 */
>>> 	{ 5,	TD,	sys_pwrite,		"pwrite64", SYS_write }, /* 181 */
>>> 	{ 0,	TP,	sys_vfork,		"vfork", SYS_vfork }, /* 190 */
>>> 	{ 5,	TP,	sys_waitid,		"waitid", SYS_waitid }, /* 284 */
>>> 	{ 4,	TN,	sys_send,		"send", SYS_sub_send }, /* 409 */
>>> 	{ 4,	TN,	sys_recv,		"recv", SYS_sub_recv }, /* 410 */
>>> 	{ 6,	TN,	sys_sendto,		"sendto", SYS_sub_sendto }, /* 411 */
>>> 	{ 6,	TN,	sys_recvfrom,		"recvfrom", SYS_sub_recvfrom }, /* 412 */
>>
>> Hmm... in fact it's the only arch which has them....
>
> The only arch that uses this feature is x86-64; other two (x86 and ia64)
> where this code is also compiled have no benefits from it.
>
>>> Most of them are no longer used, and I guess all the rest could be
>>> replaced by sys_func checks.
>
> Here it is, tested on x86 and x86-64, please give it a try:
> http://strace.git.sourceforge.net/git/gitweb.cgi?p=strace/strace;a=shortlog;h=ldv/master

Looks good to me.

-- 
vda