Messages in this thread | | | Date | Sun, 18 May 1997 21:06:24 -0700 | From | Jeffrey Wiegley <> | Subject | why does SMP in 2.1.38 lockup uniprocessor machines? |
| |
This is sort of like 1 trousers vs. 20 trousers I guess...
In a fit of anxiousness I moved from 2.0.30 to 2.1.38. Obtained all the new version of the necessary stuff as outlined in Doc.../Changes. I have a simple Pentium 166 machine and a simple make mrproper; make config; make dep; make clean; make zImage resulted in a kernel which would boot fine, seemed to run fine except for one thing.
v2.1.38 seems to lockup solid when running PPP. (Or at least the combination of PPP, X-Windows and Netscape 4.0) Also, local network stuff seemed to be slow or useless (such as the apache server on my machine wouldn't server anything)
I moved back to 2.0.30 and then had a brain drizzle this morning. After commenting out the SMP=1 define in the v2.1.38 makefile (and I also applied the 2.1.39 patch) it compiles fine and appears to run without any problems at all.
So in summary:
v2.1.38 with SMP defined locks up my uniprocessor machine, v2.1.39 without SMP appears to run great.
Thats the info, now my question is:
Why does the SMP stuff break a uniprocessor machine? Why can't a uniprocessor machine be treated like a lame SMP machine where the number of CPUs is 1?
I would understand some changes in the boot stuff to detect the number of CPUs available but after you know that why can't the other kernel stuff be written to be independent of the actual number of CPUs present?
- Jeff
| |