lkml.org 
[lkml]   [2017]   [Jun]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH RFC tip/core/rcu 1/2] srcu: Allow use of Tiny/Tree SRCU from both process and interrupt context
On Tue, Jun 06, 2017 at 02:01:54PM +0200, Paolo Bonzini wrote:
>
>
> On 06/06/2017 13:09, Peter Zijlstra wrote:
> >> diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> >> index 36e1f82faed1..681bf6bc04a5 100644
> >> --- a/kernel/rcu/srcutiny.c
> >> +++ b/kernel/rcu/srcutiny.c
> >> @@ -35,8 +35,8 @@
> >>
> >> static int init_srcu_struct_fields(struct srcu_struct *sp)
> >> {
> >> - sp->srcu_lock_nesting[0] = 0;
> >> - sp->srcu_lock_nesting[1] = 0;
> >> + atomic_set(&sp->srcu_lock_nesting[0], 0);
> >> + atomic_set(&sp->srcu_lock_nesting[1], 0);
> >> init_swait_queue_head(&sp->srcu_wq);
> >> sp->srcu_gp_seq = 0;
> >> rcu_segcblist_init(&sp->srcu_cblist);
> >> @@ -86,7 +86,8 @@ EXPORT_SYMBOL_GPL(init_srcu_struct);
> >> */
> >> void cleanup_srcu_struct(struct srcu_struct *sp)
> >> {
> >> - WARN_ON(sp->srcu_lock_nesting[0] || sp->srcu_lock_nesting[1]);
> >> + WARN_ON(atomic_read(&sp->srcu_lock_nesting[0]) ||
> >> + atomic_read(&sp->srcu_lock_nesting[1]));
> >> flush_work(&sp->srcu_work);
> >> WARN_ON(rcu_seq_state(sp->srcu_gp_seq));
> >> WARN_ON(sp->srcu_gp_running);
> >> @@ -97,7 +98,7 @@ EXPORT_SYMBOL_GPL(cleanup_srcu_struct);
> >>
> >> /*
> >> * Counts the new reader in the appropriate per-CPU element of the
> >> - * srcu_struct. Must be called from process context.
> >> + * srcu_struct.
> >> * Returns an index that must be passed to the matching srcu_read_unlock().
> >> */
> >> int __srcu_read_lock(struct srcu_struct *sp)
> >> @@ -105,21 +106,19 @@ int __srcu_read_lock(struct srcu_struct *sp)
> >> int idx;
> >>
> >> idx = READ_ONCE(sp->srcu_idx);
> >> - WRITE_ONCE(sp->srcu_lock_nesting[idx], sp->srcu_lock_nesting[idx] + 1);
> >> + atomic_inc(&sp->srcu_lock_nesting[idx]);
> >> return idx;
> >> }
> >> EXPORT_SYMBOL_GPL(__srcu_read_lock);
> >>
> >> /*
> >> * Removes the count for the old reader from the appropriate element of
> >> - * the srcu_struct. Must be called from process context.
> >> + * the srcu_struct.
> >> */
> >> void __srcu_read_unlock(struct srcu_struct *sp, int idx)
> >> {
> >> - int newval = sp->srcu_lock_nesting[idx] - 1;
> >> -
> >> - WRITE_ONCE(sp->srcu_lock_nesting[idx], newval);
> >> - if (!newval && READ_ONCE(sp->srcu_gp_waiting))
> >> + if (atomic_dec_return_relaxed(&sp->srcu_lock_nesting[idx]) == 0 &&
> >> + READ_ONCE(sp->srcu_gp_waiting))
> >> swake_up(&sp->srcu_wq);
> >> }
> >> EXPORT_SYMBOL_GPL(__srcu_read_unlock);
> >> @@ -148,7 +147,7 @@ void srcu_drive_gp(struct work_struct *wp)
> >> idx = sp->srcu_idx;
> >> WRITE_ONCE(sp->srcu_idx, !sp->srcu_idx);
> >> WRITE_ONCE(sp->srcu_gp_waiting, true); /* srcu_read_unlock() wakes! */
> >> - swait_event(sp->srcu_wq, !READ_ONCE(sp->srcu_lock_nesting[idx]));
> >> + swait_event(sp->srcu_wq, !atomic_read(&sp->srcu_lock_nesting[idx]));
> >> WRITE_ONCE(sp->srcu_gp_waiting, false); /* srcu_read_unlock() cheap. */
> >> rcu_seq_end(&sp->srcu_gp_seq);
> >
> > I'm not entirely sure this is actually needed. TINY_SRCU is !PREEMPT &&
> > !SMP. So that means all we need is to be safe from IRQs.
> >
> > Now, do we (want) support things like:
> >
> > <IRQ>
> > srcu_read_lock();
> > </IRQ>
> >
> > srcu_read_lock();
> >
> > srcu_read_unlock();
> >
> > <IRQ>
> > srcu_read_unlock();
> > </IRC>
> >
> >
> > _OR_
> >
> > do we already (or want to) mandate that SRCU usage in IRQs must be
> > balanced? That is, if it is used from IRQ context it must do an equal
> > amount of srcu_read_lock() and srcu_read_unlock()s?
> >
> > Because if we have the balance requirement (as we do for
> > preempt_disable()) then even on load-store architectures the current
> > code should be sufficient (since if an interrupt does as many dec's as
> > it does inc's, the actual value will not change over an interrupt, and
> > our load from before the interrupt is still valid).
>
> Good point! So the srcutiny part should not be necessary. I'll reply
> to the other email now.

Good analysis, Peter!

So the only part of this patch that is needed is the changes to the
comments, right? ;-)

Thanx, Paul

\
 
 \ /
  Last update: 2017-06-12 00:47    [W:1.304 / U:0.288 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site