tty problems in recent head?

[Peter Wemm]

On Fri, Nov 28, 2008 at 8:27 PM, Giorgos Keramidas
<keramida at ceid.upatras.gr> wrote:
> On Fri, 28 Nov 2008 09:06:50 +0200, Giorgos Keramidas <keramida at ceid.upatras.gr> wrote:
>> On Fri, 28 Nov 2008 08:55:15 +0200, Giorgos Keramidas <keramida at ceid.upatras.gr> wrote:
>>> I just restored my laptop after a bit of 'fun' with a broken disk, and
>>> rebuilt all my ports.  Something in head/ @ svn rev 185370 seems to
>>> cause problems to screen & xterm.
>>>
>>> Exiting an xterm window causes xterm processes to be stuck in 'RUN' and
>>> consume a lot of CPU:
>>>
>>>   PID USERNAME    THR PRI NICE   SIZE    RES STATE  C    TIME    CPU  COMMAND
>>> 11211 keramida      1 106    0  7624K  4360K CPU0    0   0:46 51.86%  xterm
>>> 11169 keramida      1 106    0  7624K  4504K RUN     1   1:12 49.66%  xterm
>>> 11201 keramida      1 106    0  7624K  4360K RUN     1   0:47 49.07%  xterm
>>> 11180 keramida      1 106    0  7624K  4360K RUN     1   1:07 48.88%  xterm
>>> ...
>>
>> Nevermind.  This seems to be a problem only with xterm processes started
>> under XFCE4.  Running under startx and plain 'twm' doesn't have the same
>> problem, so I'll have to look a bit more into this...
>
> The xterm processes that get stuck seem to be spinning near line 1854 of
> sched_ule.c.  Running `info threads' on a live kernel after xterm starts
> spinning on a CPU shows:
>
>  129 Thread 100174 (PID=97493: xterm)  sched_switch (td=0xc72fad80,
>        newtd=0xc7245000, flags=519) at /usr/src/sys/kern/sched_ule.c:1854
>
> Since this part of sched_ule.c hasn't changed in a while
>
>    REV CHANGE     AUTHOR
>   ---------------------------------------------------------------------------------------------------
>   1848 171482       jeff               cpu_switch(td, newtd, mtx);
>   1849 171482       jeff               /*
>   1850 171482       jeff                * We may return from cpu_switch on a different cpu.  However,
>   1851 171482       jeff                * we always return with td_lock pointing to the current cpu's
>   1852 171482       jeff                * run queue lock.
>   1853 171482       jeff                */
>   1854 171482       jeff               cpuid = PCPU_GET(cpuid);
>   1855 171482       jeff               tdq = TDQ_CPU(cpuid);
>   1856 174629       jeff               lock_profile_obtain_lock_success(
>   1857 174629       jeff                   &TDQ_LOCKPTR(tdq)->lock_object, 0, 0, __FILE__, __LINE__);
>   1858 145256     jkoshy #ifdef        HWPMC_HOOKS
>   1859 145256     jkoshy               if (PMC_PROC_IS_USING_PMCS(td->td_proc))
>   1860 145256     jkoshy                       PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_IN);
>   1861 145256     jkoshy #endif
>
> any ideas why PCPU_GET() might spin like this?

It isn't.  The 'info' command is misleading you.  It is merely the
next instruction after returning from cpu_switch().  Something is
effectively in a yield loop.

-- 
Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com; KI6FJV
"All of this is for nothing if we don't go to the stars" - JMS/B5
"If Java had true garbage collection, most programs would delete
themselves upon execution." -- Robert Sewell