[parisc-linux] [PATCH] timer_interrupt and gettimeoffset.

[John David Anglin]

> On Mon, 2006-09-04 at 12:57 -0400, John David Anglin wrote:
> > There needs to be a check to ensure that the code has enough time
> > to set cr16 with the new compare time before the counter passes this
> > value.  If it misses, the counter will wrap.  This probably doesn't
> > need to be a full half tick if it's desireable to avoid negative
> > offsets.
> 
> Exactly!

The loop in the last timer patch that I sent, as compiled for a 32-bit 
c3k kernel, is four instructions.  Using a 500 us half tick time and
assuming a slow 50 MHz cpu, I estimate we should be able to handle
latencies up to about 6 seconds.  The maximum latency time for faster
cpus are proportionately longer.  That should be enough to handle most
circumstances except for an uninitialized tick value.  Another loop
could be added with a larger decrement if we need to handle larger
latencies.

I'd hoped that improving the handling of timer interrupts might
fix the behavior of expect, but it doesn't.  If I run the binutils
ld testsuite under load, it usually hangs at the end of the test:

dave     28731 28706  4 14:56 pts/1    00:00:25 expect -- /usr/share/dejagnu/runtest.exp --tool ld --srcdir /home/dave/gnu/binutils-2.16.91/src/ld/testsuite CC=gcc -L/home/dave/gnu/binutils-2.16.91/objdir/./ld CFLAGS=-g -O2 CXX=c++ -L/home/dave/gnu/binutils-2.16.91/objdir/./ld CXXFLAGS=-g -O2 CC_FOR_HOST=gcc CFLAGS_FOR_HOST=-g -O2 OFILES=ldgram.o ldlex.o lexsup.o ldlang.o mri.o ldctor.o ldmain.o ldwrite.o ldexp.o  ldemul.o ldver.o ldmisc.o ldfile.o ldcref.o ehppalinux.o  BFDLIB=../bfd/.libs/libbfd.a LIBIBERTY=../libiberty/libiberty.a  LIBS=
dave     28751 28731  0 14:56 pts/1    00:00:00 [expect] <defunct>

Dave
-- 
J. David Anglin                                  dave.anglin at nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)