Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 19 May 2022 10:03:48 -0400
From:   Steven Rostedt <rostedt@goodmis.org>
To:     Vasily Averin <vvs@openvz.org>
Cc:     YoPOhRctb8wwbmY5@carbon, Shakeel Butt <shakeelb@google.com>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Vlastimil Babka <vbabka@suse.cz>,
        Matthew Wilcox <willy@infradead.org>,
        Hyeonggon Yoo <42.hyeyoo@gmail.com>,
        Muchun Song <songmuchun@bytedance.com>, kernel@openvz.org,
        linux-kernel@vger.kernel.org, Ingo Molnar <mingo@redhat.com>,
        Andrew Morton <akpm@linux-foundation.org>, linux-mm@kvack.org,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        David Rientjes <rientjes@google.com>,
        Pekka Enberg <penberg@kernel.org>,
        Christoph Lameter <cl@linux.com>,
        Michal Hocko <mhocko@suse.com>
Subject: Re: [PATCH v3] tracing: add 'accounted' entry into output of
 allocation tracepoints
Message-ID: <20220519100348.101d027d@gandalf.local.home>
In-Reply-To: <b728f944-e3ae-cdb6-5f02-2fb21466b2fb@openvz.org>
References: <f6625cd8-90f9-6d48-50f6-7bb052bf479f@openvz.org>
        <20220518160447.20a7b96f@gandalf.local.home>
        <b728f944-e3ae-cdb6-5f02-2fb21466b2fb@openvz.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Thu, 19 May 2022 14:35:46 +0300
Vasily Averin <vvs@openvz.org> wrote:

> >> @@ -33,42 +35,46 @@ DECLARE_EVENT_CLASS(kmem_alloc,
> >>  		__entry->bytes_req	= bytes_req;
> >>  		__entry->bytes_alloc	= bytes_alloc;
> >>  		__entry->gfp_flags	= (__force unsigned long)gfp_flags;
> >> +		__entry->accounted	= (gfp_flags & __GFP_ACCOUNT) ||
> >> +					  (s && s->flags & SLAB_ACCOUNT);  
> > 
> > Now you could make this even faster in the fast path and save just the
> > s->flags.
> > 
> > 	__entry->sflags = s ? s->flags : 0;
> >   
> >>  	),
> >>  
> >> -	TP_printk("call_site=%pS ptr=%p bytes_req=%zu bytes_alloc=%zu gfp_flags=%s",
> >> +	TP_printk("call_site=%pS ptr=%p bytes_req=%zu bytes_alloc=%zu gfp_flags=%s accounted=%s",
> >>  		(void *)__entry->call_site,
> >>  		__entry->ptr,
> >>  		__entry->bytes_req,
> >>  		__entry->bytes_alloc,
> >> -		show_gfp_flags(__entry->gfp_flags))
> >> +		show_gfp_flags(__entry->gfp_flags),
> >> +		__entry->accounted ? "true" : "false")  
> > 
> > And then have: "accounted=%s":
> > 
> > 		(__entry->gfp_flags & __GFP_ACCOUNT) ||
> > 		(__entry->sflags & SLAB_ACCOUNT) ? "true" : "false"  
> 
> Unfortunately this returns back sparse warnings about bitwise gfp_t and slab_flags_t casts.
> Could you please explain why your variant is faster?

Micro-optimization, grant you, but it is faster because it moves some of
the logic into the slow path (the read side), and takes it out of the fast
path (the write side).

The idea of tracing is to squeeze out every cycle we can to keep the
tracing overhead down.

But it's really up to you if you need that. I'm not going to let this be a
blocker. This is more of an FYI than anything else.

-- Steve