IRC logs for #openrisc Thursday, 2012-07-19

-!- Jia` is now known as Jia10:03
derRichardhmm, uclibs libpthread is really nasty. especially in the openrisc case where you have no cmpxchg13:00
derRichardthis while (testandset(spinlock)) stuff in __pthread_acquire is utter crap.13:00
juliusbderRichard: what's up with the cmpxchg stuff? dose that usually rely on hardware semaphore support?14:02
derRichardjuliusb: if an arch does not support a cmpxchg like instruction you have to go into the kernel to do locking. all this sched_yield(), nanosleep() stuff what ulibc (and dietlibc) does in this case is crap14:04
derRichardAFAIK arm has a cmpxchg system call for this. glibc uses it if no cmpxchg is available14:04
juliusbthat's probably a good suggestion14:10
juliusbARM does have a swap instruction, though, but it may not be usable in user mode14:11
derRichardjuliusb: not all arm socs support it14:11
derRichardbbl, lunch14:11
juliusbwell, they don't have it on the new stuff anyway (they prefer the exclusive accesses)14:11
juliusbbut it sounds like a good idea to go into the kernel for it14:12
derRichardjuliusb: going into the kernel is the only way to fix the issue. glibc does so for very good reasons14:49
derRichardi guess dielibc and ulibc never noticed the problem because you need more then a trivial test case to spot the bug14:50
stekernah, I think I've found out what made clang/llvm-compiled linux bug out yesterday18:37
stekernlooks like I've made functions that don't need a frame-pointer threat the fp (r2) with too little respect18:38
stekerni.e. it uses it as a temp-reg but doesn't save/restore it18:38
stekernfixing that made it progress further, but it's still crashing20:40

Generated by 2.15.2 by Marius Gedminas - find it at!