Re: [RFC PATCH 3/8] perf: Init/fini PMU specific data

[Peter Zijlstra]

On Mon, Dec 02, 2019 at 03:44:34PM -0500, Liang, Kan wrote:

> It's not in context switch. I will use the normal spinlock to instead.

Mutex would make even more sense. And we already have a per-task
perf_event_mutex.

Also, I don't think you need tasklist_lock here, if you set the state
before the iteration, any new clone()s will observe the state and
allocate the storage themselves. Then all you need is RCU iteration of
the tasklist.