On 18.05.2010 14:48, Wolfgang Grandegger wrote: > On 05/18/2010 02:19 PM, Oliver Hartkopp wrote: >> With shrinked lists ... >> >> Hi Wolfgang, all, >> >> Dave suggests to have the 'expensive' locking all the time - even on >> uniprocessors, where we did not had any problems so far. >> >> What do you think about this change? > > Well, it's not really dramatic if we do it on any system.
Fine - will repost an CONFIG_SMP unconditional patch then. >> I would like to have a common opinion here, before discussing it on netdev-ML >> again ... > > Would using spin_lock/spin_unlock be sufficient? No, it won't as we would be able to block the spin_locked region from userspace/softirq context at sending, which would cause the hard-irq of the rx-interrupt to wait for the lock. Will send an updated patch later this evening. Thanks, Oliver _______________________________________________ Socketcan-core mailing list [email protected] https://lists.berlios.de/mailman/listinfo/socketcan-core
