linux-kernel - Re: [PATCH] fs/buffer.c: disable per-CPU buffer

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <ZM2PvQJd7kRyWnAZ@tpad>
Date:   Fri, 4 Aug 2023 20:54:37 -0300
From:   Marcelo Tosatti <mtosatti@...hat.com>
To:     Frederic Weisbecker <frederic@...nel.org>
Cc:     linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
        Alexander Viro <viro@...iv.linux.org.uk>,
        Christian Brauner <brauner@...nel.org>,
        Matthew Wilcox <willy@...radead.org>,
        Christoph Hellwig <hch@....de>, Jens Axboe <axboe@...nel.dk>,
        Dave Chinner <david@...morbit.com>,
        Valentin Schneider <vschneid@...hat.com>,
        Leonardo Bras <leobras@...hat.com>,
        Yair Podemsky <ypodemsk@...hat.com>, P J P <ppandit@...hat.com>
Subject: Re: [PATCH] fs/buffer.c: disable per-CPU buffer_head cache for
 isolated CPUs

On Sat, Aug 05, 2023 at 12:03:59AM +0200, Frederic Weisbecker wrote:
> On Tue, Jun 27, 2023 at 05:08:15PM -0300, Marcelo Tosatti wrote:
> > 
> > For certain types of applications (for example PLC software or
> > RAN processing), upon occurrence of an event, it is necessary to
> > complete a certain task in a maximum amount of time (deadline).
> > 
> > One way to express this requirement is with a pair of numbers,
> > deadline time and execution time, where:
> > 
> >         * deadline time: length of time between event and deadline.
> >         * execution time: length of time it takes for processing of event
> >                           to occur on a particular hardware platform
> >                           (uninterrupted).
> > 
> > The particular values depend on use-case. For the case
> > where the realtime application executes in a virtualized
> > guest, an IPI which must be serviced in the host will cause
> > the following sequence of events:
> > 
> >         1) VM-exit
> >         2) execution of IPI (and function call)
> >         3) VM-entry
> > 
> > Which causes an excess of 50us latency as observed by cyclictest
> > (this violates the latency requirement of vRAN application with 1ms TTI,
> > for example).
> > 
> > invalidate_bh_lrus calls an IPI on each CPU that has non empty
> > per-CPU cache:
> > 
> >         on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1);
> > 
> > The performance when using the per-CPU LRU cache is as follows:
> > 
> >  42 ns per __find_get_block
> >  68 ns per __find_get_block_slow
> > 
> > Given that the main use cases for latency sensitive applications
> > do not involve block I/O (data necessary for program operation is 
> > locked in RAM), disable per-CPU buffer_head caches for isolated CPUs.

Hi Frederic,

> So what happens if they ever do I/O then? Like if they need to do
> some prep work before entering an isolated critical section?

Then instead of going through the per-CPU LRU buffer_head cache
(__find_get_block), isolated CPUs will work as if their per-CPU
cache is always empty, going through the slowpath 
(__find_get_block_slow). The algorithm is:

/*
 * Perform a pagecache lookup for the matching buffer.  If it's there, refresh
 * it in the LRU and mark it as accessed.  If it is not present then return
 * NULL
 */
struct buffer_head *
__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
{
        struct buffer_head *bh = lookup_bh_lru(bdev, block, size);

        if (bh == NULL) {
                /* __find_get_block_slow will mark the page accessed */
                bh = __find_get_block_slow(bdev, block);
                if (bh)
                        bh_lru_install(bh);
        } else
                touch_buffer(bh);

        return bh;
}
EXPORT_SYMBOL(__find_get_block);

I think the performance difference between the per-CPU LRU cache
VS __find_get_block_slow was much more significant when the cache 
was introduced. Nowadays its only 26ns (moreover modern filesystems 
do not use buffer_head's).

> Thanks.

Thank you for the review.

> > 
> > Signed-off-by: Marcelo Tosatti <mtosatti@...hat.com>
> > 
> > diff --git a/fs/buffer.c b/fs/buffer.c
> > index a7fc561758b1..49e9160ce100 100644
> > --- a/fs/buffer.c
> > +++ b/fs/buffer.c
> > @@ -49,6 +49,7 @@
> >  #include <trace/events/block.h>
> >  #include <linux/fscrypt.h>
> >  #include <linux/fsverity.h>
> > +#include <linux/sched/isolation.h>
> >  
> >  #include "internal.h"
> >  
> > @@ -1289,7 +1290,7 @@ static void bh_lru_install(struct buffer_head *bh)
> >  	 * failing page migration.
> >  	 * Skip putting upcoming bh into bh_lru until migration is done.
> >  	 */
> > -	if (lru_cache_disabled()) {
> > +	if (lru_cache_disabled() || cpu_is_isolated(smp_processor_id())) {
> >  		bh_lru_unlock();
> >  		return;
> >  	}
> > @@ -1319,6 +1320,10 @@ lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
> >  
> >  	check_irqs_on();
> >  	bh_lru_lock();
> > +	if (cpu_is_isolated(smp_processor_id())) {
> > +		bh_lru_unlock();
> > +		return NULL;
> > +	}
> >  	for (i = 0; i < BH_LRU_SIZE; i++) {
> >  		struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
> >  
> > 
> 
>