Re: [PATCH] genirq: add mutex and rcu locking to irq_desc_tree

[Thomas Gleixner]

On Tue, 31 Oct 2017, Masahiro Yamada wrote:

> Add a mutex to prevent concurrency on the updater side of the
> irq_desc radix tree.

The callers of insert/delete are already serialized by sparse_irq_lock. SO
why would we need yet another mutex?

> Add rcu_read_lock/unlock to the reader side so that lifetimes of
> leaf pointers of the radix tree are correctly managed.

That rcu protection is only needed for callers which have special life time
requirements and they have that already. For the bulk of the callers rcu
read lock is not required unless I'm missing something.

Thanks,

	tglx