svn commit: r231989 - in head: lib/libthr/thread sys/kern sys/sys

[David Xu]

On 2012/2/24 6:43, Jilles Tjoelker wrote:
> On Thu, Feb 23, 2012 at 08:07:49AM +0800, David Xu wrote:
>> On 2012/2/23 7:42, Jilles Tjoelker wrote:
>>> On Wed, Feb 22, 2012 at 03:22:50AM +0000, David Xu wrote:
>>>> Author: davidxu
>>>> Date: Wed Feb 22 03:22:49 2012
>>>> New Revision: 231989
>>>> URL: http://svn.freebsd.org/changeset/base/231989
>>>> Log:
>>>>     Use unused fourth argument of umtx_op to pass flags to kernel for operation
>>>>     UMTX_OP_WAIT. Upper 16bits is enough to hold a clock id, and lower
>>>>     16bits is used to pass flags. The change saves a clock_gettime() syscall
>>>>     from libthr.
>>>> Modified:
>>>>     head/lib/libthr/thread/thr_umtx.c
>>>>     head/sys/kern/kern_umtx.c
>>>>     head/sys/sys/umtx.h
>>>>
>>>> Modified: head/lib/libthr/thread/thr_umtx.c
>>>> ==============================================================================
>>>> --- head/lib/libthr/thread/thr_umtx.c	Wed Feb 22 01:50:13 2012	(r231988)
>>>> +++ head/lib/libthr/thread/thr_umtx.c	Wed Feb 22 03:22:49 2012	(r231989)
>>>> @@ -200,20 +200,10 @@ int
>>>> +	       	abstime != NULL ? (void *)(uintptr_t)((clockid<<   16) | UMTX_WAIT_ABSTIME) : 0,
>>> Please check that this shift does not lose any information (i.e.,
>>> clockid>= 0&&   clockid<= 0xFFFF) before doing it.
>>> Implementing clock_getcpuclockid() will require clockids greater than
>>> 65535 because such clockids contain a process id.
>> In which document said the clock id includes a process id ? I think the
>> clock_getcpuclockid()
>> is a simplified version of getrusage().
> clock_getcpuclockid() lets you find a clock ID for the CPU time clock of
> a process. Usage could be like this:
>
> int error;
> pid_t pid;
> clockid_t clk;
> struct timespec ts;
>
> error = clock_getcpuclockid(pid,&clk);
> if (error == 0)
> 	if (clock_gettime(clk,&ts) == 0)
> 		printf("%lu.%09lu\n", (unsigned long)ts.tv_sec,
> 		    (unsigned long)ts.tv_usec);
>
> One way to implement it is to reserve a PID_MAX sized subrange of
> clockid_t for this purpose.
>
> Similarly, pthread_getcpuclockid() lets you find a clock ID for the CPU
> time clock of a thread.
So the clock ID may be dynamically allocated in theory ?
> These seem mostly useful for clock_getres(), clock_gettime() and
> possibly timer_create(). Besides, sem_timedwait() doesn't even let you
> specify a clock ID.
>
Yes, sem_timedwait does not specify a clock ID,  it is inconsistent with
pthread condition variable, it is not surprising, it is not first time 
that the POSIX
has a bug.