Re: [PATCH v2] rcu: Remove redundant rcu_read_lock/unlock() in spin_lock critical sections

On 2025-09-15 21:47:29 [+0800], pengdonglin wrote:
> From: pengdonglin <[email protected]>
> 
> Per Documentation/RCU/rcu_dereference.rst [1], since Linux 4.20's RCU
> consolidation [2][3], RCU read-side critical sections include:
>   - Explicit rcu_read_lock()
>   - BH/interrupt/preemption-disabling regions
>   - Spinlock critical sections (including CONFIG_PREEMPT_RT kernels [4])
> 
> Thus, explicit rcu_read_lock()/unlock() calls within spin_lock*() regions are redundant.
> This patch removes them, simplifying locking semantics while preserving RCU protection.
> 
> [1] https://elixir.bootlin.com/linux/v6.17-rc5/source/Documentation/RCU/rcu_dereference.rst#L407
> [2] https://lore.kernel.org/lkml/[email protected]/
> [3] https://lwn.net/Articles/777036/
> [4] https://lore.kernel.org/lkml/6435833a-bdcb-4114-b29d-28b7f436d47d@paulmck-laptop/

What about something like this:

  Since commit a8bb74acd8efe ("rcu: Consolidate RCU-sched update-side
  function definitions") there is no difference between rcu_read_lock(),
  rcu_read_lock_bh() and rcu_read_lock_sched() in terms of RCU read
  section and the relevant grace period. That means that spin_lock(),
  which implies rcu_read_lock_sched(), also implies rcu_read_lock().

  There is no need no explicitly start a RCU read section if one has
  already been started implicitly by spin_lock().

  Simplify the code and remove the inner rcu_read_lock() invocation.


The description above should make it clear what:
- the intention is
- the proposed solution to it and why it is correct.

You can't send a patch like this. You need to split it at the very least
by subsystem. The networking bits need to follow to follow for instance
   Documentation/process/maintainer-netdev.rst

and so on.

Sebastian