Re: [PATCH 16/16] rcu/tree: Use delayed work instead of hrtimer to refill the cache

[Paul E. McKenney]

On Thu, Oct 29, 2020 at 09:22:41PM +0100, Uladzislau Rezki wrote:
> On Thu, Oct 29, 2020 at 09:13:42PM +0100, Uladzislau Rezki wrote:
> > On Thu, Oct 29, 2020 at 12:47:24PM -0700, Paul E. McKenney wrote:
> > > On Thu, Oct 29, 2020 at 05:50:19PM +0100, Uladzislau Rezki (Sony) wrote:
> > > > A CONFIG_PREEMPT_COUNT is unconditionally enabled, thus a page
> > > > can be obtained directly from a kvfree_rcu() path. To distinguish
> > > > that and take a decision the preemptable() macro is used when it
> > > > is save to enter allocator.
> > > > 
> > > > It means that refilling a cache is not important from timing point
> > > > of view. Switch to a delayed work, so the actual work is queued from
> > > > the timer interrupt with 1 jiffy delay. An immediate placing a task
> > > > on a current CPU can lead to rq->lock double lock. That is why a
> > > > delayed method is in place.
> > > > 
> > > > Signed-off-by: Uladzislau Rezki (Sony) <urezki@xxxxxxxxx>
> > > 
> > > Thank you, Uladzislau!
> > > 
> > > I applied this on top of v5.10-rc1 and got the following from the
> > > single-CPU builds:
> > > 
> > >   SYNC    include/config/auto.conf.cmd
> > >   DESCEND  objtool
> > >   CC      kernel/bounds.s
> > >   CALL    scripts/atomic/check-atomics.sh
> > >   UPD     include/generated/bounds.h
> > >   CC      arch/x86/kernel/asm-offsets.s
> > > In file included from ./include/asm-generic/atomic-instrumented.h:20:0,
> > >                  from ./include/linux/atomic.h:82,
> > >                  from ./include/linux/crypto.h:15,
> > >                  from arch/x86/kernel/asm-offsets.c:9:
> > > ./include/linux/pagemap.h: In function ‘__page_cache_add_speculative’:
> > > ./include/linux/build_bug.h:30:34: error: called object is not a function or function pointer
> > >  #define BUILD_BUG_ON_INVALID(e) ((void)(sizeof((__force long)(e))))
> > >                                  ~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > ./include/linux/mmdebug.h:45:25: note: in expansion of macro ‘BUILD_BUG_ON_INVALID’
> > >  #define VM_BUG_ON(cond) BUILD_BUG_ON_INVALID(cond)
> > >                          ^~~~~~~~~~~~~~~~~~~~
> > > ./include/linux/pagemap.h:207:2: note: in expansion of macro ‘VM_BUG_ON’
> > >   VM_BUG_ON(preemptible())
> > >   ^~~~~~~~~
> > > scripts/Makefile.build:117: recipe for target 'arch/x86/kernel/asm-offsets.s' failed
> > > make[1]: *** [arch/x86/kernel/asm-offsets.s] Error 1
> > > Makefile:1199: recipe for target 'prepare0' failed
> > > make: *** [prepare0] Error 2
> > > 
> > > I vaguely recall something like this showing up in the previous series
> > > and that we did something or another to address it.  Could you please
> > > check against the old series at -rcu branch dev.2020.10.22a?  (I verified
> > > that the old series does run correctly in the single-CPU scenarios.)
> > > 
> > I see the same build error. Will double check if we have similar in the
> > previous series also. It looks like the error is caused by the Thomas series.
> > 
> > Will check!
> > 
> OK. Found it:
> 
> urezki@pc638:~/data/coding/linux.git$ git diff
> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> index cbfbe2bcca75..7dd3523093db 100644
> --- a/include/linux/pagemap.h
> +++ b/include/linux/pagemap.h
> @@ -204,7 +204,7 @@ void release_pages(struct page **pages, int nr);
>  static inline int __page_cache_add_speculative(struct page *page, int count)
>  {
>  #ifdef CONFIG_TINY_RCU
> -       VM_BUG_ON(preemptible())
> +       VM_BUG_ON(preemptible());
>         /*
>          * Preempt must be disabled here - we rely on rcu_read_lock doing
>          * this for us.
> urezki@pc638:~/data/coding/linux.git$
> 
> I guess we had some extra patch that fixes a kernel compilation for !SMP
> case. Will check dev.2020.10.22a.

I guess I am blind today!

And yes, that semicolon is in the corresponding commit in -tip.

So, an important safety tip:  When forward porting, start from the
commits that have been tested rather than from the original series
of patches.  ;-)

							Thanx, Paul