Re: Firefox consumes 100% CPU while polling

[Francois Tigeot <ftigeot@xxxxxxxxxxxx>]

On Sun, Oct 31, 2010 at 06:44:43PM +0100, Matthias Schmidt wrote:
> 
> I compiled a new version of Firefox from pkgsrc (3.6.11) and after some
> time Firefox consumes 100% CPU even if only one empty tab is open.
> 
> The machine runs
> 
> DragonFly hades 2.9-DEVELOPMENT DragonFly v2.9.0.90.g20134-DEVELOPMENT
> #0: Sun Oct 31 08:56:28 CET 2010
> matthias@hades:/usr/obj/usr/src/sys/T42  i386
> 
> and is a UP machine.
> 
> Precisely the CPU is consumed by the xulrunner process.  Truss revealed
> the following repeating pattern:
> 
> poll(0x2b332ab8,0x3,0xffffffff)                  = 209 (0xd1)
[...]

faxgetty from the comms/hylafax package also takes 100% CPU without any
reason

It is also always repeating the same pattern [kdump -R output]:

0.000024 CALL  gettimeofday(0x7ffffffff850,0)
0.000011 RET   gettimeofday 0
0.000011 CALL  gettimeofday(0x7ffffffff810,0)
0.000010 RET   gettimeofday 0
0.000010 CALL  select(0x4,0x7ffffffffa00,0x7ffffffff980,0x7ffffffff900,0x595c80)
0.000016 RET   select 1
0.000022 CALL  gettimeofday(0x7ffffffff810,0)
0.000011 RET   gettimeofday 0
0.000010 CALL  read(0x1,0x7ffffffff030,0x7ff)
0.000012 GIO   fd 1 read 0 bytes ""
0.000009 RET   read 0

A quick grep at the source told me select() should be waiting at
least 1 or 2 seconds :

  struct timeval tv;
  tv.tv_sec = 2;
  tv.tv_usec = 0;

But it returns immediately (only 16 microseconds from the call).

System is an Atom (4 threads) running 2.8-RELEASE/x86_64

-- 
Francois Tigeot