[parisc-linux] nanosleep

[parisc-linux] nanosleep

[Berthold Gunreben]

Hi List,    
    
I tried to compile glibc2.3.1, and have some problems with the threading support there.     
When using perl with the new glibc on a C240, things seem to work without problems. On    
a 715/33, I get (probably) endless loops with lots of sched_yield() calls.    
    
In the Internet, I found the following site:    
http://sources.redhat.com/ml/libc-alpha/2000-02/msg00057.html   
where Andrea Arcangeli fixed a kernel problem with nanosleep on kernel 2.2 and 2.3, and  
also provided a testprogram. Here are the results of my test:  
  
C240: (2.4.19-pa20)  
# ./nanotest & sleep 4; killall nanotest  
[1] 8455  
userspace             15 sec 778496030 nsec  
kernel                15 sec 800000000 nsec  
[1]+  Done                    ./nanotest  
  
715/33: (2.4.17-32)  
# ./nanotest & sleep 4; killall nanotest  
[1] 1840  
userspace             15 sec 422651030 nsec  
kernel                15 sec 450000000 nsec  
[1]+  Done                    ./nanotest  
  
Intel based for reference: (2.4.19)  
# ./nanotest & sleep 4 ; killall nanotest  
[1] 19648  
userspace             15 sec 984455030 nsec  
kernel                15 sec 984477030 nsec  
[1]+  Done                    ./nanotest  
  
Is there a bugfix missing for parisc? 
 
Berthold 
________________________________________________________________
Keine verlorenen Lotto-Quittungen, keine vergessenen Gewinne mehr! 
Beim WEB.DE Lottoservice: http://tippen2.web.de/?x=13

Re: [parisc-linux] nanosleep

[Carlos O'Donell]

> Hi List,         
> I tried to compile glibc2.3.1, and have some problems with the threading support there.     
> When using perl with the new glibc on a C240, things seem to work without problems. On    
> a 715/33, I get (probably) endless loops with lots of sched_yield() calls.    
     
glibc 2.3.1 from debian CVS or upstream?

c.

Re: [parisc-linux] nanosleep

[b.gunreben]

Carlos O'Donell wrote:
> 
> > Hi List,
> > I tried to compile glibc2.3.1, and have some problems with the threading support there.
> > When using perl with the new glibc on a C240, things seem to work without problems. On
> > a 715/33, I get (probably) endless loops with lots of sched_yield() calls.
> 
> glibc 2.3.1 from debian CVS or upstream?

This is upstream with hppa patches from the debian tree. 

The report was probably bad. In the mail thread I mentioned in my
previous mail, the nanosleep in the kernel was fixed because of
threading problems. If I understand it correctly, threading will always
have problems without a realy working nanosleep.

Berthold

Re: [parisc-linux] nanosleep

[Carlos O'Donell]

> > glibc 2.3.1 from debian CVS or upstream?
> 
> This is upstream with hppa patches from the debian tree. 

That's currently broke, and I've been slotting in my time to fix this.
Currently rebuilding glibc :}
 
> The report was probably bad. In the mail thread I mentioned in my
> previous mail, the nanosleep in the kernel was fixed because of
> threading problems. If I understand it correctly, threading will always
> have problems without a realy working nanosleep.
 
Yes. You'll see it's use sprinkled in linuxthreads/* and in particular 
spinlock.c

c.