IRC channel logs


back to list of logs

***Server sets mode: +nt
<damo221>youpi: nice job with interrupts, i will test your changes later tonight
<damo221>as im at work
<damo22>youpi: im getting this printed once per second on average:
<damo22>irq handler 11: release dead delivery 0 unacked irqs
<damo22>after i run rump twice
<damo22>(once and then stop, then again)
<damo22>is there an off by one bug that tries to release even though there are zero unacked irqs?
<damo22>or did i screw up something in pci-userspace in rump?
<youpi>ah, that's because the gnumach code doesn't clean yet until an actual irq happens. It's harmless, just noisy. I'll fix that
<damo22>but it somehow prevents rump from working again the second time
<damo22>with my old patchset i was able to run my test rump program many times
<damo22>it opens the block device and dumps first sector
<youpi>it's surprising that you can't run a second time
<youpi>I did test that with netdde, it was working fine
<youpi>where is your rump code?
<damo22>its all in incubator
<damo22>i added some patches from debian for gnumach on top of origin/master-user_level_drivers-debian
<youpi>I don't see intr_register/enable calls in there
<youpi>the rump branch in the incubator, right?
<damo22>yeah but it refers to external code from github
<damo22>i didnt want to include all of it
<youpi>well, that's what I need to see, the code that manages irqs :)
<damo22>let me find it
<damo22>im on that branch
<youpi>ok, the intr_register/enable code looks ok
<youpi>what actually happens on the second run?
<youpi>does it just not get interrupts?
<youpi>I guess you should add printfs in the gnumach user-level interrupts code to make sure what is happening
<damo22>how do i build a debug kernel without debian packaging
<damo22>does mach_print work without that?
<damo22>ok thats why then
<damo22>what about using s = splhi() and splx(s) instead of cli/sti in device/intr.c
<damo22>is that appropriate? wont it restore the previous flags
<youpi>that'd probably be better yes
<youpi>the people who wrote the code probably didn't know about spl etc.
<damo22>something is still fishy with interrupts, or theres a bug in rump
<damo22>maybe i should revert my changes to the netbsd source that limits the timeouts
<damo22>maybe we just fixed everything
<damo22>arggh i cant make rump link with pciaccess anymore
<damo22>"LDFLAGS += -lpciaccess" doesnt work in the
<damo22>LDFLAGS += -Wl,--no-as-needed -lpciaccess -Wl,--as-needed
<damo22>irq fired: 11
<damo22>task /hurd/rumpdisk(2128) deallocating a bogus port 107, probably a bug
<damo22>thats the last 3 lines of output from mounting a disk
<youpi>more printfs would probably be useful
<youpi>damo22: not only using splhigh/splx instead of cli/sti in intr.c is better, it actually is now needed
<youpi>previously the code wasn't calling code that uses splhigh/splx, but now it is
<youpi>and thus we need proper nesting
<youpi>I have pushed updated branches, I'll upload a fixed package
<youpi>I have been high by the issue on buildds
<youpi>I have fixed another issue with crashed netdde
<youpi>see latest push, and I'll upload a fixed debian package again.