Received-SPF: pass (google.com: domain of linux-kernel+bounces-153168-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
From: Jonathan Haslam <jonathan.haslam@gmail.com>
To: linux-trace-kernel@vger.kernel.org,
	mhiramat@kernel.org
Cc: jonathan.haslam@gmail.com,
	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>,
	Jiri Olsa <jolsa@kernel.org>,
	Ian Rogers <irogers@google.com>,
	Adrian Hunter <adrian.hunter@intel.com>,
	linux-perf-users@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: [PATCH v2] uprobes: reduce contention on uprobes_tree access
Date: Mon, 22 Apr 2024 03:23:05 -0700
Message-ID: <20240422102306.6026-1-jonathan.haslam@gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

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)

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