On Thu, 22 Apr 2004, Simon L. Nielsen wrote: > > > > Are you running a main line kernel with debug.mpsafenet turned on? > > > > > > Main line kernel (I assume you mean no special patches), but > > > debug.mpsafenet is turned off. > > > > > > [root_at_soekris:~] sysctl debug.mpsafenet > > > debug.mpsafenet: 0 > > > > In the main line kernel with debug.mpsafenet turned off, INTR_MPSAFE is > > supposed to be ignored for network interrupts, resulting in inbound > > network stack code running with Giant held. Is there any chance that the > > interrupt thread is being shared with another device? dmesg output would > > Possibly, I'm not really sure how to check that. > > > be useful, along with some devinfo output. It could be there's a problem > > resulting in Giant not being held under those circumstances, or that Giant > > is being dropped somewhere it shouldn't be. > > OK, my kernel foo is not sufficient know if that is the case, but here > is dmesg and outut from devinfo output : Indeed, if_sis is showed as GIANT-LOCKED, which means the interrupt handler is running with Giant. However, on re-reading your stack trace, it's not actually the if_sis interrupt: it's the callout/timeout that if_sis schedules, and it's marked as unconditionally CALLOUT_MPSAFE. Try this patch: Index: if_sis.c =================================================================== RCS file: /home/ncvs/src/sys/pci/if_sis.c,v retrieving revision 1.96 diff -u -r1.96 if_sis.c --- if_sis.c 17 Mar 2004 17:50:53 -0000 1.96 +++ if_sis.c 22 Apr 2004 22:17:41 -0000 _at__at_ -1234,7 +1234,10 _at__at_ } sc->sis_unit = unit; - callout_init(&sc->sis_stat_ch, CALLOUT_MPSAFE); + if (debug_mpsafenet) + callout_init(&sc->sis_stat_ch, CALLOUT_MPSAFE); + else + callout_init(&sc->sis_stat_ch, 0); bcopy(eaddr, (char *)&sc->arpcom.ac_enaddr, ETHER_ADDR_LEN); /*Received on Thu Apr 22 2004 - 13:19:21 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:52 UTC