mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Joel Fernandes <joelagnelf@nvidia.com>
To: linux-kernel@vger.kernel.org,
	"Davidlohr Bueso" <dave@stgolabs.net>,
	"Paul E. McKenney" <paulmck@kernel.org>,
	"Josh Triplett" <josh@joshtriplett.org>,
	"Frederic Weisbecker" <frederic@kernel.org>,
	"Neeraj Upadhyay" <neeraj.upadhyay@kernel.org>,
	"Joel Fernandes" <joelagnelf@nvidia.com>,
	"Boqun Feng" <boqun.feng@gmail.com>,
	"Uladzislau Rezki" <urezki@gmail.com>,
	"Steven Rostedt" <rostedt@goodmis.org>,
	"Mathieu Desnoyers" <mathieu.desnoyers@efficios.com>,
	"Lai Jiangshan" <jiangshanlai@gmail.com>,
	Zqiang <qiang.zhang@linux.dev>, "Miguel Ojeda" <ojeda@kernel.org>,
	"Alex Gaynor" <alex.gaynor@gmail.com>,
	"Gary Guo" <gary@garyguo.net>,
	"Björn Roy Baron" <bjorn3_gh@protonmail.com>,
	"Benno Lossin" <benno.lossin@proton.me>,
	"Andreas Hindborg" <a.hindborg@kernel.org>,
	"Alice Ryhl" <aliceryhl@google.com>,
	"Trevor Gross" <tmgross@umich.edu>,
	"Danilo Krummrich" <dakr@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>,
	rcu@vger.kernel.org, Joel Fernandes <joel@joelfernandes.org>,
	rust-for-linux@vger.kernel.org
Subject: [PATCH v2] refscale: Add tests for local_irq_disable() vs local_interrupt_disable()
Date: Thu, 19 Jun 2025 13:53:35 -0400	[thread overview]
Message-ID: <20250619175335.2905836-1-joelagnelf@nvidia.com> (raw)

Add two new refscale test cases to compare the performance of
traditional local_irq_disable()/local_irq_enable() with the newer
local_interrupt_disable()/local_interrupt_enable() APIs.

The local_interrupt_disable()/local_interrupt_enable() APIs are
introduced to provide a Rust-compatible interface for interrupt
control, as mentioned in:
https://lore.kernel.org/all/20240527222254.565881-1-lyude@redhat.com/

The 2 new tests are "local_interrupt" for the new API and "local_irq" test
for the traditional one. This allows direct performance comparison
between the two approaches.

Test results on x86 with 4 readers, 5 runs, 10000 loops:

local_irq (traditional API):
  Run 1: 1.306 ns
  Run 2: 1.306 ns
  Run 3: 1.305 ns
  Run 4: 1.307 ns
  Run 5: 1.085 ns
  Average: ~1.26 ns per operation

local_interrupt (new API):
  Run 1: 4.594 ns
  Run 2: 4.201 ns
  Run 3: 4.428 ns
  Run 4: 4.905 ns
  Run 5: 4.566 ns
  Average: ~4.54 ns per operation

The results show higher overhead with local_interrupt_disable()/enable()
possibly coming from the additional state tracking.

To run the module, modprobe refscale scale_type=local_irq (or local_interrupt).

Cc: Lyude Paul <lyude@redhat.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: rcu@vger.kernel.org
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
---
 kernel/rcu/refscale.c | 73 ++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 69 insertions(+), 4 deletions(-)

diff --git a/kernel/rcu/refscale.c b/kernel/rcu/refscale.c
index f11a7c2af778..ac6e2391d672 100644
--- a/kernel/rcu/refscale.c
+++ b/kernel/rcu/refscale.c
@@ -71,7 +71,7 @@ MODULE_AUTHOR("Joel Fernandes (Google) <joel@joelfernandes.org>");
 
 static char *scale_type = "rcu";
 module_param(scale_type, charp, 0444);
-MODULE_PARM_DESC(scale_type, "Type of test (rcu, srcu, refcnt, rwsem, rwlock.");
+MODULE_PARM_DESC(scale_type, "Type of test (rcu, srcu, refcnt, rwsem, rwlock, local_interrupt, local_irq.");
 
 torture_param(int, verbose, 0, "Enable verbose debugging printk()s");
 torture_param(int, verbose_batched, 0, "Batch verbose debugging printk()s");
@@ -524,6 +524,62 @@ static const struct ref_scale_ops lock_irq_ops = {
 	.name		= "lock-irq"
 };
 
+// IRQ disable/enable tests using interrupt_disable/enable.
+static void ref_local_interrupt_section(const int nloops)
+{
+	int i;
+
+	for (i = nloops; i >= 0; i--) {
+		local_interrupt_disable();
+		local_interrupt_enable();
+	}
+}
+
+static void ref_local_interrupt_delay_section(const int nloops, const int udl, const int ndl)
+{
+	int i;
+
+	for (i = nloops; i >= 0; i--) {
+		local_interrupt_disable();
+		un_delay(udl, ndl);
+		local_interrupt_enable();
+	}
+}
+
+static const struct ref_scale_ops local_interrupt_ops = {
+	.readsection	= ref_local_interrupt_section,
+	.delaysection	= ref_local_interrupt_delay_section,
+	.name		= "local_interrupt"
+};
+
+// IRQ disable/enable tests using local_irq_disable/enable.
+static void ref_local_irq_section(const int nloops)
+{
+	int i;
+
+	for (i = nloops; i >= 0; i--) {
+		local_irq_disable();
+		local_irq_enable();
+	}
+}
+
+static void ref_local_irq_delay_section(const int nloops, const int udl, const int ndl)
+{
+	int i;
+
+	for (i = nloops; i >= 0; i--) {
+		local_irq_disable();
+		un_delay(udl, ndl);
+		local_irq_enable();
+	}
+}
+
+static const struct ref_scale_ops local_irq_ops = {
+	.readsection	= ref_local_irq_section,
+	.delaysection	= ref_local_irq_delay_section,
+	.name		= "local_irq"
+};
+
 // Definitions acquire-release.
 static DEFINE_PER_CPU(unsigned long, test_acqrel);
 
@@ -956,13 +1012,22 @@ ref_scale_reader(void *arg)
 			rcu_scale_one_reader();
 	// Also keep interrupts disabled.  This also has the effect
 	// of preventing entries into slow path for rcu_read_unlock().
-	local_irq_save(flags);
+	// Exception: for IRQ ops, use preempt_disable instead since we need
+	// to test actual IRQ disable/enable performance.
+	if (cur_ops == &local_interrupt_ops || cur_ops == &local_irq_ops)
+		preempt_disable();
+	else
+		local_irq_save(flags);
 	start = ktime_get_mono_fast_ns();
 
 	rcu_scale_one_reader();
 
 	duration = ktime_get_mono_fast_ns() - start;
-	local_irq_restore(flags);
+
+	if (cur_ops == &local_interrupt_ops || cur_ops == &local_irq_ops)
+		preempt_enable();
+	else
+		local_irq_restore(flags);
 
 	rt->last_duration_ns = WARN_ON_ONCE(duration < 0) ? 0 : duration;
 	// To reduce runtime-skew noise, do maintain-load invocations until
@@ -1194,7 +1259,7 @@ ref_scale_init(void)
 	int firsterr = 0;
 	static const struct ref_scale_ops *scale_ops[] = {
 		&rcu_ops, &srcu_ops, &srcu_fast_ops, &srcu_lite_ops, RCU_TRACE_OPS RCU_TASKS_OPS
-		&refcnt_ops, &rwlock_ops, &rwsem_ops, &lock_ops, &lock_irq_ops,
+		&refcnt_ops, &rwlock_ops, &rwsem_ops, &lock_ops, &lock_irq_ops, &local_interrupt_ops, &local_irq_ops,
 		&acqrel_ops, &sched_clock_ops, &clock_ops, &jiffies_ops,
 		&typesafe_ref_ops, &typesafe_lock_ops, &typesafe_seqlock_ops,
 	};
-- 
2.43.0


             reply	other threads:[~2025-06-19 17:53 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-06-19 17:53 Joel Fernandes [this message]
2025-06-20  4:46 ` kernel test robot

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20250619175335.2905836-1-joelagnelf@nvidia.com \
    --to=joelagnelf@nvidia.com \
    --cc=a.hindborg@kernel.org \
    --cc=alex.gaynor@gmail.com \
    --cc=aliceryhl@google.com \
    --cc=benno.lossin@proton.me \
    --cc=bjorn3_gh@protonmail.com \
    --cc=boqun.feng@gmail.com \
    --cc=dakr@kernel.org \
    --cc=dave@stgolabs.net \
    --cc=frederic@kernel.org \
    --cc=gary@garyguo.net \
    --cc=jiangshanlai@gmail.com \
    --cc=joel@joelfernandes.org \
    --cc=josh@joshtriplett.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=lyude@redhat.com \
    --cc=mathieu.desnoyers@efficios.com \
    --cc=neeraj.upadhyay@kernel.org \
    --cc=ojeda@kernel.org \
    --cc=paulmck@kernel.org \
    --cc=qiang.zhang@linux.dev \
    --cc=rcu@vger.kernel.org \
    --cc=rostedt@goodmis.org \
    --cc=rust-for-linux@vger.kernel.org \
    --cc=tmgross@umich.edu \
    --cc=urezki@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®