Re: [PATCH] fs/buffer: Make BH_Uptodate_Lock bit_spin_lock a regular spinlock_t

[Jan Kara]

On Fri 11-10-19 13:25:25, Sebastian Siewior wrote:
> On 2019-08-20 20:01:14 [+0200], Thomas Gleixner wrote:
> > On Tue, 20 Aug 2019, Matthew Wilcox wrote:
> > > On Tue, Aug 20, 2019 at 07:08:18PM +0200, Sebastian Siewior wrote:
> > > > Bit spinlocks are problematic if PREEMPT_RT is enabled, because they
> > > > disable preemption, which is undesired for latency reasons and breaks when
> > > > regular spinlocks are taken within the bit_spinlock locked region because
> > > > regular spinlocks are converted to 'sleeping spinlocks' on RT. So RT
> > > > replaces the bit spinlocks with regular spinlocks to avoid this problem.
> > > > Bit spinlocks are also not covered by lock debugging, e.g. lockdep.
> > > > 
> > > > Substitute the BH_Uptodate_Lock bit spinlock with a regular spinlock.
> > > > 
> > > > Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > > > [bigeasy: remove the wrapper and use always spinlock_t]
> > > 
> > > Uhh ... always grow the buffer_head, even for non-PREEMPT_RT?  Why?
> > 
> > Christoph requested that:
> > 
> >   https://lkml.kernel.org/r/20190802075612.GA20962@xxxxxxxxxxxxx
> 
> What do we do about this one?

I was thinking about this for quite some time. In the end I think the patch
is almost fine but I'd name the lock b_update_lock and put it just after
b_size element in struct buffer_head to use the hole there. That way we
don't grow struct buffer_head.

With some effort, we could even shrink struct buffer_head from 104 bytes
(on x86_64) to 96 bytes but I don't think that effort is worth it (I'd find
it better use of time to actually work on getting rid of buffer heads
completely).

								Honza

-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR