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: <d9c6c19135497054986900f94c8e47410b15316a.1670623111.git.bristot@kernel.org>
Date:   Fri,  9 Dec 2022 23:05:53 +0100
From:   Daniel Bristot de Oliveira <bristot@...nel.org>
To:     Daniel Bristot de Oliveira <bristot@...nel.org>,
        Steven Rostedt <rostedt@...dmis.org>
Cc:     Masami Hiramatsu <mhiramat@...nel.org>,
        Jonathan Corbet <corbet@....net>,
        Juri Lelli <juri.lelli@...hat.com>,
        Clark Williams <williams@...hat.com>,
        Bagas Sanjaya <bagasdotme@...il.com>,
        linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: [PATCH V5 2/3] tracing/osnoise: Add preempt and/or irq disabled options

The osnoise workload runs with preemption and IRQs enabled in such
a way as to allow all sorts of noise to disturb osnoise's execution.
hwlat tracer has a similar workload but works with irq disabled,
allowing only NMIs and the hardware to generate noise.

While thinking about adding an options file to hwlat tracer to
allow the system to panic, and other features I was thinking
to add, like having a tracepoint at each noise detection, it
came to my mind that is easier to make osnoise and also do
hardware latency detection than making hwlat "feature compatible"
with osnoise.

Other points are:
 - osnoise already has an independent cpu file.
 - osnoise has a more intuitive interface, e.g., runtime/period vs.
   window/width (and people often need help remembering what it is).
 - osnoise: tracepoints
 - osnoise stop options
 - osnoise options file itself

Moreover, the user-space side (in rtla) is simplified by reusing the
existing osnoise code.

Finally, people have been asking me about using osnoise for hw latency
detection, and I have to explain that it was sufficient but not
necessary. These options make it sufficient and necessary.

Adding a Suggested-by Clark, as he often asked me about this
possibility.

Cc: Suggested-by: Clark Williams <williams@...hat.com>
Cc: Daniel Bristot de Oliveira <bristot@...nel.org>
Cc: Steven Rostedt <rostedt@...dmis.org>
Cc: Masami Hiramatsu <mhiramat@...nel.org>
Cc: Jonathan Corbet <corbet@....net>
Signed-off-by: Daniel Bristot de Oliveira <bristot@...nel.org>
---
 kernel/trace/trace_osnoise.c | 48 ++++++++++++++++++++++++++++++++----
 1 file changed, 43 insertions(+), 5 deletions(-)

diff --git a/kernel/trace/trace_osnoise.c b/kernel/trace/trace_osnoise.c
index 801eba0b5cf8..4da021d9629c 100644
--- a/kernel/trace/trace_osnoise.c
+++ b/kernel/trace/trace_osnoise.c
@@ -55,10 +55,17 @@ enum osnoise_options_index {
 	OSN_DEFAULTS = 0,
 	OSN_WORKLOAD,
 	OSN_PANIC_ON_STOP,
+	OSN_PREEMPT_DISABLE,
+	OSN_IRQ_DISABLE,
 	OSN_MAX
 };
 
-static const char * const osnoise_options_str[OSN_MAX] = { "DEFAULTS", "OSNOISE_WORKLOAD", "PANIC_ON_STOP" };
+static const char * const osnoise_options_str[OSN_MAX] = {
+							"DEFAULTS",
+							"OSNOISE_WORKLOAD",
+							"PANIC_ON_STOP",
+							"OSNOISE_PREEMPT_DISABLE",
+							"OSNOISE_IRQ_DISABLE" };
 
 #define OSN_DEFAULT_OPTIONS	0x2
 unsigned long osnoise_options	= OSN_DEFAULT_OPTIONS;
@@ -1308,18 +1315,26 @@ static void notify_new_max_latency(u64 latency)
  */
 static int run_osnoise(void)
 {
+	bool disable_irq = test_bit(OSN_IRQ_DISABLE, &osnoise_options);
 	struct osnoise_variables *osn_var = this_cpu_osn_var();
 	u64 start, sample, last_sample;
 	u64 last_int_count, int_count;
 	s64 noise = 0, max_noise = 0;
 	s64 total, last_total = 0;
 	struct osnoise_sample s;
+	bool disable_preemption;
 	unsigned int threshold;
 	u64 runtime, stop_in;
 	u64 sum_noise = 0;
 	int hw_count = 0;
 	int ret = -1;
 
+	/*
+	 * Disabling preemption is only required if IRQs are enabled,
+	 * and the options is set on.
+	 */
+	disable_preemption = !disable_irq && test_bit(OSN_PREEMPT_DISABLE, &osnoise_options);
+
 	/*
 	 * Considers the current thread as the workload.
 	 */
@@ -1335,6 +1350,15 @@ static int run_osnoise(void)
 	 */
 	threshold = tracing_thresh ? : 5000;
 
+	/*
+	 * Apply PREEMPT and IRQ disabled options.
+	 */
+	if (disable_irq)
+		local_irq_disable();
+
+	if (disable_preemption)
+		preempt_disable();
+
 	/*
 	 * Make sure NMIs see sampling first
 	 */
@@ -1422,16 +1446,21 @@ static int run_osnoise(void)
 		 * cond_resched()
 		 */
 		if (IS_ENABLED(CONFIG_PREEMPT_RCU)) {
-			local_irq_disable();
+			if (!disable_irq)
+				local_irq_disable();
+
 			rcu_momentary_dyntick_idle();
-			local_irq_enable();
+
+			if (!disable_irq)
+				local_irq_enable();
 		}
 
 		/*
 		 * For the non-preemptive kernel config: let threads runs, if
-		 * they so wish.
+		 * they so wish, unless set not do to so.
 		 */
-		cond_resched();
+		if (!disable_irq && !disable_preemption)
+			cond_resched();
 
 		last_sample = sample;
 		last_int_count = int_count;
@@ -1450,6 +1479,15 @@ static int run_osnoise(void)
 	 */
 	barrier();
 
+	/*
+	 * Return to the preemptive state.
+	 */
+	if (disable_preemption)
+		preempt_enable();
+
+	if (disable_irq)
+		local_irq_enable();
+
 	/*
 	 * Save noise info.
 	 */
-- 
2.32.0

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ