Received-SPF: pass (google.com: domain of linux-kernel+bounces-153611-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Date: Tue, 23 Apr 2024 00:12:48 +0900
From: Masami Hiramatsu (Google) <mhiramat@kernel.org>
To: Jiri Olsa <olsajiri@gmail.com>
Cc: Jonathan Haslam <jonathan.haslam@gmail.com>,
 linux-trace-kernel@vger.kernel.org, mhiramat@kernel.org, andrii@kernel.org,
 bpf@vger.kernel.org, rostedt@goodmis.org, Peter Zijlstra
 <peterz@infradead.org>, Ingo Molnar <mingo@redhat.com>, Arnaldo Carvalho de
 Melo <acme@kernel.org>, Namhyung Kim <namhyung@kernel.org>, Mark Rutland
 <mark.rutland@arm.com>, Alexander Shishkin
 <alexander.shishkin@linux.intel.com>, Ian Rogers <irogers@google.com>,
 Adrian Hunter <adrian.hunter@intel.com>, linux-perf-users@vger.kernel.org,
 linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2] uprobes: reduce contention on uprobes_tree access
Message-Id: <20240423001248.161933b7646e1170f106400c@kernel.org>
In-Reply-To: <ZiZMdLIK55q3EvMP@krava>
References: <20240422102306.6026-1-jonathan.haslam@gmail.com>
	<ZiZMdLIK55q3EvMP@krava>
Precedence: bulk
Mime-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

On Mon, 22 Apr 2024 13:39:32 +0200
Jiri Olsa <olsajiri@gmail.com> wrote:

> On Mon, Apr 22, 2024 at 03:23:05AM -0700, Jonathan Haslam wrote:
> > Active uprobes are stored in an RB tree and accesses to this tree are
> > dominated by read operations. Currently these accesses are serialized by
> > a spinlock but this leads to enormous contention when large numbers of
> > threads are executing active probes.
> > 
> > This patch converts the spinlock used to serialize access to the
> > uprobes_tree RB tree into a reader-writer spinlock. This lock type
> > aligns naturally with the overwhelmingly read-only nature of the tree
> > usage here. Although the addition of reader-writer spinlocks are
> > discouraged [0], this fix is proposed as an interim solution while an
> > RCU based approach is implemented (that work is in a nascent form). This
> > fix also has the benefit of being trivial, self contained and therefore
> > simple to backport.
> > 
> > We have used a uprobe benchmark from the BPF selftests [1] to estimate
> > the improvements. Each block of results below show 1 line per execution
> > of the benchmark ("the "Summary" line) and each line is a run with one
> > more thread added - a thread is a "producer". The lines are edited to
> > remove extraneous output.
> > 
> > The tests were executed with this driver script:
> > 
> > for num_threads in {1..20}
> > do
> >   sudo ./bench -a -p $num_threads trig-uprobe-nop | grep Summary
> > done
> > 
> > SPINLOCK (BEFORE)
> > ==================
> > Summary: hits    1.396 ± 0.007M/s (  1.396M/prod)
> > Summary: hits    1.656 ± 0.016M/s (  0.828M/prod)
> > Summary: hits    2.246 ± 0.008M/s (  0.749M/prod)
> > Summary: hits    2.114 ± 0.010M/s (  0.529M/prod)
> > Summary: hits    2.013 ± 0.009M/s (  0.403M/prod)
> > Summary: hits    1.753 ± 0.008M/s (  0.292M/prod)
> > Summary: hits    1.847 ± 0.001M/s (  0.264M/prod)
> > Summary: hits    1.889 ± 0.001M/s (  0.236M/prod)
> > Summary: hits    1.833 ± 0.006M/s (  0.204M/prod)
> > Summary: hits    1.900 ± 0.003M/s (  0.190M/prod)
> > Summary: hits    1.918 ± 0.006M/s (  0.174M/prod)
> > Summary: hits    1.925 ± 0.002M/s (  0.160M/prod)
> > Summary: hits    1.837 ± 0.001M/s (  0.141M/prod)
> > Summary: hits    1.898 ± 0.001M/s (  0.136M/prod)
> > Summary: hits    1.799 ± 0.016M/s (  0.120M/prod)
> > Summary: hits    1.850 ± 0.005M/s (  0.109M/prod)
> > Summary: hits    1.816 ± 0.002M/s (  0.101M/prod)
> > Summary: hits    1.787 ± 0.001M/s (  0.094M/prod)
> > Summary: hits    1.764 ± 0.002M/s (  0.088M/prod)
> > 
> > RW SPINLOCK (AFTER)
> > ===================
> > Summary: hits    1.444 ± 0.020M/s (  1.444M/prod)
> > Summary: hits    2.279 ± 0.011M/s (  1.139M/prod)
> > Summary: hits    3.422 ± 0.014M/s (  1.141M/prod)
> > Summary: hits    3.565 ± 0.017M/s (  0.891M/prod)
> > Summary: hits    2.671 ± 0.013M/s (  0.534M/prod)
> > Summary: hits    2.409 ± 0.005M/s (  0.401M/prod)
> > Summary: hits    2.485 ± 0.008M/s (  0.355M/prod)
> > Summary: hits    2.496 ± 0.003M/s (  0.312M/prod)
> > Summary: hits    2.585 ± 0.002M/s (  0.287M/prod)
> > Summary: hits    2.908 ± 0.011M/s (  0.291M/prod)
> > Summary: hits    2.346 ± 0.016M/s (  0.213M/prod)
> > Summary: hits    2.804 ± 0.004M/s (  0.234M/prod)
> > Summary: hits    2.556 ± 0.001M/s (  0.197M/prod)
> > Summary: hits    2.754 ± 0.004M/s (  0.197M/prod)
> > Summary: hits    2.482 ± 0.002M/s (  0.165M/prod)
> > Summary: hits    2.412 ± 0.005M/s (  0.151M/prod)
> > Summary: hits    2.710 ± 0.003M/s (  0.159M/prod)
> > Summary: hits    2.826 ± 0.005M/s (  0.157M/prod)
> > Summary: hits    2.718 ± 0.001M/s (  0.143M/prod)
> > Summary: hits    2.844 ± 0.006M/s (  0.142M/prod)
> 
> nice, I'm assuming Masami will take this one.. in any case:
> 
> Acked-by: Jiri Olsa <jolsa@kernel.org>

Thanks Jiri!

This looks good to me too.
Let me pick this for probes/for-next.

Thank you,

> 
> thanks,
> jirka
> 
> > 
> > The numbers in parenthesis give averaged throughput per thread which is
> > of greatest interest here as a measure of scalability. Improvements are
> > in the order of 22 - 68% with this particular benchmark (mean = 43%).
> > 
> > V2:
> >  - Updated commit message to include benchmark results.
> > 
> > [0] https://docs.kernel.org/locking/spinlocks.html
> > [1] https://github.com/torvalds/linux/blob/master/tools/testing/selftests/bpf/benchs/bench_trigger.c
> > 
> > Signed-off-by: Jonathan Haslam <jonathan.haslam@gmail.com>
> > ---
> >  kernel/events/uprobes.c | 22 +++++++++++-----------
> >  1 file changed, 11 insertions(+), 11 deletions(-)
> > 
> > diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> > index e4834d23e1d1..8ae0eefc3a34 100644
> > --- a/kernel/events/uprobes.c
> > +++ b/kernel/events/uprobes.c
> > @@ -39,7 +39,7 @@ static struct rb_root uprobes_tree = RB_ROOT;
> >   */
> >  #define no_uprobe_events()	RB_EMPTY_ROOT(&uprobes_tree)
> >  
> > -static DEFINE_SPINLOCK(uprobes_treelock);	/* serialize rbtree access */
> > +static DEFINE_RWLOCK(uprobes_treelock);	/* serialize rbtree access */
> >  
> >  #define UPROBES_HASH_SZ	13
> >  /* serialize uprobe->pending_list */
> > @@ -669,9 +669,9 @@ static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
> >  {
> >  	struct uprobe *uprobe;
> >  
> > -	spin_lock(&uprobes_treelock);
> > +	read_lock(&uprobes_treelock);
> >  	uprobe = __find_uprobe(inode, offset);
> > -	spin_unlock(&uprobes_treelock);
> > +	read_unlock(&uprobes_treelock);
> >  
> >  	return uprobe;
> >  }
> > @@ -701,9 +701,9 @@ static struct uprobe *insert_uprobe(struct uprobe *uprobe)
> >  {
> >  	struct uprobe *u;
> >  
> > -	spin_lock(&uprobes_treelock);
> > +	write_lock(&uprobes_treelock);
> >  	u = __insert_uprobe(uprobe);
> > -	spin_unlock(&uprobes_treelock);
> > +	write_unlock(&uprobes_treelock);
> >  
> >  	return u;
> >  }
> > @@ -935,9 +935,9 @@ static void delete_uprobe(struct uprobe *uprobe)
> >  	if (WARN_ON(!uprobe_is_active(uprobe)))
> >  		return;
> >  
> > -	spin_lock(&uprobes_treelock);
> > +	write_lock(&uprobes_treelock);
> >  	rb_erase(&uprobe->rb_node, &uprobes_tree);
> > -	spin_unlock(&uprobes_treelock);
> > +	write_unlock(&uprobes_treelock);
> >  	RB_CLEAR_NODE(&uprobe->rb_node); /* for uprobe_is_active() */
> >  	put_uprobe(uprobe);
> >  }
> > @@ -1298,7 +1298,7 @@ static void build_probe_list(struct inode *inode,
> >  	min = vaddr_to_offset(vma, start);
> >  	max = min + (end - start) - 1;
> >  
> > -	spin_lock(&uprobes_treelock);
> > +	read_lock(&uprobes_treelock);
> >  	n = find_node_in_range(inode, min, max);
> >  	if (n) {
> >  		for (t = n; t; t = rb_prev(t)) {
> > @@ -1316,7 +1316,7 @@ static void build_probe_list(struct inode *inode,
> >  			get_uprobe(u);
> >  		}
> >  	}
> > -	spin_unlock(&uprobes_treelock);
> > +	read_unlock(&uprobes_treelock);
> >  }
> >  
> >  /* @vma contains reference counter, not the probed instruction. */
> > @@ -1407,9 +1407,9 @@ vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long e
> >  	min = vaddr_to_offset(vma, start);
> >  	max = min + (end - start) - 1;
> >  
> > -	spin_lock(&uprobes_treelock);
> > +	read_lock(&uprobes_treelock);
> >  	n = find_node_in_range(inode, min, max);
> > -	spin_unlock(&uprobes_treelock);
> > +	read_unlock(&uprobes_treelock);
> >  
> >  	return !!n;
> >  }
> > -- 
> > 2.43.0
> > 
> 


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>