[lkml]   [2013]   [Feb]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: Debugging Thinkpad T430s occasional suspend failure.
2013/2/15 Linus Torvalds <>:
> On Fri, Feb 15, 2013 at 9:44 AM, Paul E. McKenney
> <> wrote:
>> This commit was designed to increase the probability of hitting the
>> races described in These races result
>> in deadlocks involving the runqueue lock (and perhaps also the priority
>> inheritance locks). And yes, I most certainly should have described
>> this in the commit message. :-(
> Ugh. That particular race seems to be because the softirq handling is
> just crazy, and does the "wakeup_softirqd()" form interrupt context,
> Because it claims to do it from softirq context, which is pure
> garbage. It's not actually in softirq context.
> The whole hardirq -> softirq transition seems stupid. I'm sure I made
> some serious mistake in cleaning it up, and there's probably some
> missed tracepoint (or perhaps screwed-up lockdep annotation), but I
> think the hardirq -> softirq preempt thing shoudl be done as an atomic
> preempt downgrade, so that we never have a window of "uhhuh, another
> interrupt can come in between and see us as being in neither). And the
> wakeup_softirqd should be done without playing with preempt count at
> all.
> Something like this ENTIRELY UNTESTED patch.
> Note: I doubt this patch affects Dave's issue at all, I just started
> looking at that do_softirq code when I read your bug explanation.
> Adding random people for kernel/softirq.c to the participants list.
> Comments about the patch? Do note that it's entirely untested, so
> consider it more a RFD than a real patch.. It looks like it adds a lot
> of lines, but most of it is for comments and simplification of the
> logic.

preempt_value_in_interrupt() looks buggy in your patch: it makes
invoke_softirq() returning if (val & HARDIRQ_MASK). But that's always
true since you have moved further the sub_preempt_count(IRQ_EXIT)

 \ /
  Last update: 2013-02-17 16:21    [W:0.099 / U:0.088 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site