Re: [PATCH v5 3.1.0-rc4-tip 12/26] Uprobes: Handle breakpointand Singlestep

[Oleg Nesterov]

On 09/26, Peter Zijlstra wrote:
>
> On Mon, 2011-09-26 at 21:31 +0530, Srikar Dronamraju wrote:
> > * Peter Zijlstra <peterz@xxxxxxxxxxxxx> [2011-09-26 15:59:13]:
> >
> > > On Tue, 2011-09-20 at 17:32 +0530, Srikar Dronamraju wrote:
> > > > 						Hence provide some extra
> > > > + * time (by way of synchronize_sched() for breakpoint hit threads to acquire
> > > > + * the uprobes_treelock before the uprobe is removed from the rbtree.
> > >
> > > 'Some extra time' doesn't make me all warm an fuzzy inside, but instead
> > > screams we fudge around a race condition.
> >
> > The extra time provided is sufficient to avoid the race. So will modify
> > it to mean "sufficient" instead of "some".
> >
> > Would that suffice?
>
> Much better, for extra point, explain why its sufficient as well ;-)

+1 ;)

I can't understand why synchronize_sched helps. In fact it is very
possible I simply misunderstood the problem, I'll appreciate if you
can explain.

Just for example. Suppose that uprobe_notify_resume() sleeps in
down_read(mmap_sem). In this case synchronize_sched() can return
even before it takes this sem, how this can help the subsequent
find_uprobe() ? Or that task  can be simply preempted before.

Or I missed the point completely?

Oleg.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/