Messages in this thread | | | Date | Tue, 27 Oct 2015 17:27:52 -0700 | Subject | Re: [PATCH v2] perf/core: fix RCU issues with cgroup monitoring mode | From | Stephane Eranian <> |
| |
On Tue, Oct 27, 2015 at 5:17 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote: > On Tue, 2015-10-27 at 16:37 -0700, Stephane Eranian wrote: > >> But that macro is already embedded into the task_css_set() macro. >> What we are saying here is that we have another way of ensuring the >> cgroup cannot disappear here, and thus the rcu lockdep detects we >> are not holding the rcu read lock but we know this is okay. We are >> trying to avoid grabbing the rcu read lock when it is not really needed >> to guarantee correct execution. So we are just shutting up the lockdep >> in these particular cases by passing true as the safe argument value. > > Sure, but in lockdep verbs, the following : > > + /* holding ctx->lock, so cgroup access is safe */ > + cgrp = perf_cgroup_from_task(task, true); > > would map to : > > cgrp = perf_cgroup_from_task(task, lockdep_is_held(&ctx->lock)); > > Notice the comment becomes useless. > Ok, this is more explicit for sure and should anything change, we would catch it. I misunderstand what Eric was proposing. I will modify this for V3 then and test again. Thanks.
| |