Re: [PATCH v6 19/21] timer: Implement the hierarchical pull model

[Sebastian Siewior]

On 2023-05-15 12:50:25 [+0200], Anna-Maria Behnsen wrote:
> The part will be replaced by:
> 
> ---8<----
> 
> 	local_irq_disable();
> 	timer_lock_remote_bases(cpu);

Do you s/raw_spin_lock_irq/raw_spin_lock/ in timer_lock_remote_bases()?
Otherwise you disable interrupts twice. Not wrong but not needed.

> 	raw_spin_lock(&tmc->lock);
> 
> 	if (!tmc->online || !tmc->idle) {
> 		timer_unlock_remote_bases(cpu);
> 		goto unlock;
> 	} else {
> 	        fetch_next_timer_interrupt_remote(jif, now, &tevt, cpu);
> 	}
> 
> 	timer_unlock_remote_bases(cpu);

The locking, unlocking and fetch_next_timer_interrupt_remote() is only
used here. So it if it wouldn't be for tmc-> then you could do
everything timer.c in once place…

> 	-> do the preparation and the walk
> 
> unlock:
> 	raw_spin_unlock_irq(&tmc->lock);
> 
> ---8<----
> 
> As menitoned in the reply last week to Frederics objections regarding the
> locking asymmetry, I would like to keep it to make the locking region of
> timer base locks as small as possible and to prevent holding up to five
> locks during the walk.

This looks okay. I wouldn't suggest to hold the timer_base::lock or
tmigr_cpu::lock longer than needed. Both can be acquired independently
cross CPU and have explicit locking order. So it is fine.

> Thanks,
> 
> 	Anna-Maria

Sebastian