A strange thing with yesterday's head..

[Gary Jennejohn]

On Thu, 28 Jan 2010 10:30:37 -0800
Randall Stewart <rrs at lakerest.net> wrote:

> All:
> 
> I just found a very strange thing with yesterdays head.
> 
> The program
> 
> http://www.freebsd.org/~rrs/my_thr.c
> 
> I compile it:
> 
> cc -g -o my_thr my_thr.c /usr/lib/libthr.a -lpthread
> 
> Now when you run this on a 2 core 64 bit machine running X11 with  
> yesterday
> AM's current. The machine appears to lock completely. It really has  
> not give
> it time and it will chug along.. but the mouse disappears etc... as if  
> its locked.
> 
> It takes quite some time to complete .. and every now and then a  
> hickup will occur
> and you get a slight response..
> 
> Now I took this code home and ran it on my 4core AMD (2.8Gig with real  
> AMD cores). And
> it did NOT do the same.. but ran like you would expect it to.
> 
> I then took the same code running on an identical 8.0 release machine  
> and ran it
> and it worked like you would expect..
> 
> 
> It looks like some change in the scheduler in head is not doing good  
> things.
> 
> Note my 4 core is behind the 2 core.. so it may not be core count  
> related.. when
> I get off work today and get home I will do a update and see if the 4  
> core starts
> behaving badly too...
> 

Are you using SCHED_4BSD?  Seems to me some changes were made to it
very recently.

I tried this on my AMD64 X2 using SCHED_ULE and it worked OK.  But I'm
not running HEAD, because I'm using the svn tree with soft-updates
journaling from a few days ago.

---
Gary Jennejohn