[PATCH v2] watchdog: Add tracing events for the most usual watchdog events

[PATCH v2] watchdog: Add tracing events for the most usual watchdog events

[Uwe Kleine-König]

To simplify debugging which process touches a watchdog and when, add
tracing events for .start(), .set_timeout(), .ping() and .stop().

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
---
Changes since (implict) v1 sent with Message-Id:
20220930144935.3373366-1-u.kleine-koenig@pengutronix.de:

 - Make use of DECLARE_EVENT_CLASS as {requ,sugg}ested by Steven.

Best regards
Uwe

 MAINTAINERS                      |  1 +
 drivers/watchdog/watchdog_core.c |  4 ++
 drivers/watchdog/watchdog_dev.c  | 12 +++++-
 include/trace/events/watchdog.h  | 66 ++++++++++++++++++++++++++++++++
 4 files changed, 81 insertions(+), 2 deletions(-)
 create mode 100644 include/trace/events/watchdog.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 72b9654f764c..507de599116b 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -21890,6 +21890,7 @@ F:	Documentation/watchdog/
 F:	drivers/watchdog/
 F:	include/linux/watchdog.h
 F:	include/uapi/linux/watchdog.h
+F:	include/trace/events/watchdog.h
 
 WHISKEYCOVE PMIC GPIO DRIVER
 M:	Kuppuswamy Sathyanarayanan <sathyanarayanan.kuppuswamy@linux.intel.com>
diff --git a/drivers/watchdog/watchdog_core.c b/drivers/watchdog/watchdog_core.c
index 3fe8a7edc252..c777a612d932 100644
--- a/drivers/watchdog/watchdog_core.c
+++ b/drivers/watchdog/watchdog_core.c
@@ -38,6 +38,9 @@
 
 #include "watchdog_core.h"	/* For watchdog_dev_register/... */
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/watchdog.h>
+
 static DEFINE_IDA(watchdog_ida);
 
 static int stop_on_reboot = -1;
@@ -163,6 +166,7 @@ static int watchdog_reboot_notifier(struct notifier_block *nb,
 			int ret;
 
 			ret = wdd->ops->stop(wdd);
+			trace_watchdog_stop(wdd, ret);
 			if (ret)
 				return NOTIFY_BAD;
 		}
diff --git a/drivers/watchdog/watchdog_dev.c b/drivers/watchdog/watchdog_dev.c
index 54903f3c851e..3323309c913c 100644
--- a/drivers/watchdog/watchdog_dev.c
+++ b/drivers/watchdog/watchdog_dev.c
@@ -47,6 +47,8 @@
 #include "watchdog_core.h"
 #include "watchdog_pretimeout.h"
 
+#include <trace/events/watchdog.h>
+
 /* the dev_t structure to store the dynamically allocated watchdog devices */
 static dev_t watchdog_devt;
 /* Reference to watchdog device behind /dev/watchdog */
@@ -157,10 +159,13 @@ static int __watchdog_ping(struct watchdog_device *wdd)
 
 	wd_data->last_hw_keepalive = now;
 
-	if (wdd->ops->ping)
+	if (wdd->ops->ping) {
 		err = wdd->ops->ping(wdd);  /* ping the watchdog */
-	else
+		trace_watchdog_ping(wdd, err);
+	} else {
 		err = wdd->ops->start(wdd); /* restart watchdog */
+		trace_watchdog_start(wdd, err);
+	}
 
 	if (err == 0)
 		watchdog_hrtimer_pretimeout_start(wdd);
@@ -259,6 +264,7 @@ static int watchdog_start(struct watchdog_device *wdd)
 		}
 	} else {
 		err = wdd->ops->start(wdd);
+		trace_watchdog_start(wdd, err);
 		if (err == 0) {
 			set_bit(WDOG_ACTIVE, &wdd->status);
 			wd_data->last_keepalive = started_at;
@@ -297,6 +303,7 @@ static int watchdog_stop(struct watchdog_device *wdd)
 	if (wdd->ops->stop) {
 		clear_bit(WDOG_HW_RUNNING, &wdd->status);
 		err = wdd->ops->stop(wdd);
+		trace_watchdog_stop(wdd, err);
 	} else {
 		set_bit(WDOG_HW_RUNNING, &wdd->status);
 	}
@@ -369,6 +376,7 @@ static int watchdog_set_timeout(struct watchdog_device *wdd,
 
 	if (wdd->ops->set_timeout) {
 		err = wdd->ops->set_timeout(wdd, timeout);
+		trace_watchdog_set_timeout(wdd, timeout, err);
 	} else {
 		wdd->timeout = timeout;
 		/* Disable pretimeout if it doesn't fit the new timeout */
diff --git a/include/trace/events/watchdog.h b/include/trace/events/watchdog.h
new file mode 100644
index 000000000000..77802d0d8c09
--- /dev/null
+++ b/include/trace/events/watchdog.h
@@ -0,0 +1,66 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM watchdog
+
+#if !defined(_TRACE_WATCHDOG_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_WATCHDOG_H
+
+#include <linux/watchdog.h>
+#include <linux/tracepoint.h>
+
+DECLARE_EVENT_CLASS(watchdog_template,
+
+	TP_PROTO(struct watchdog_device *wdd, int err),
+
+	TP_ARGS(wdd, err),
+
+	TP_STRUCT__entry(
+		__field(int, id)
+		__field(int, err)
+	),
+
+	TP_fast_assign(
+		__entry->id = wdd->id;
+		__entry->err = err;
+	),
+
+	TP_printk("watchdog%d err=%d", __entry->id, __entry->err)
+);
+
+DEFINE_EVENT(watchdog_template, watchdog_start,
+	TP_PROTO(struct watchdog_device *wdd, int err),
+	TP_ARGS(wdd, err));
+
+TRACE_EVENT(watchdog_set_timeout,
+
+	TP_PROTO(struct watchdog_device *wdd, unsigned int timeout, int err),
+
+	TP_ARGS(wdd, timeout, err),
+
+	TP_STRUCT__entry(
+		__field(int, id)
+		__field(unsigned int, timeout)
+		__field(int, err)
+	),
+
+	TP_fast_assign(
+		__entry->id = wdd->id;
+		__entry->timeout = timeout;
+		__entry->err = err;
+	),
+
+	TP_printk("watchdog%d timeout=%u err=%d", __entry->id, __entry->timeout, __entry->err)
+);
+
+DEFINE_EVENT(watchdog_template, watchdog_ping,
+	TP_PROTO(struct watchdog_device *wdd, int err),
+	TP_ARGS(wdd, err));
+
+DEFINE_EVENT(watchdog_template, watchdog_stop,
+	TP_PROTO(struct watchdog_device *wdd, int err),
+	TP_ARGS(wdd, err));
+
+#endif /* !defined(_TRACE_WATCHDOG_H) || defined(TRACE_HEADER_MULTI_READ) */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>

base-commit: 4fe89d07dcc2804c8b562f6c7896a45643d34b2f
-- 
2.37.2

Re: [PATCH v2] watchdog: Add tracing events for the most usual watchdog events

[Steven Rostedt]

On Tue,  4 Oct 2022 11:19:49 +0200
Uwe Kleine-König <u.kleine-koenig@pengutronix.de> wrote:

> +DEFINE_EVENT(watchdog_template, watchdog_start,
> +	TP_PROTO(struct watchdog_device *wdd, int err),
> +	TP_ARGS(wdd, err));
> +
> +TRACE_EVENT(watchdog_set_timeout,
> +
> +	TP_PROTO(struct watchdog_device *wdd, unsigned int timeout, int err),
> +
> +	TP_ARGS(wdd, timeout, err),
> +
> +	TP_STRUCT__entry(
> +		__field(int, id)
> +		__field(unsigned int, timeout)
> +		__field(int, err)
> +	),
> +
> +	TP_fast_assign(
> +		__entry->id = wdd->id;
> +		__entry->timeout = timeout;
> +		__entry->err = err;
> +	),
> +
> +	TP_printk("watchdog%d timeout=%u err=%d", __entry->id, __entry->timeout, __entry->err)
> +);

Nit, but I would probably put the above TRACE_EVENT() below the two
DEFINE_EVENT()s below. That way we have all the DEFINE_EVENT()s for a
specific DECLARE_EVENT_CLASS() together. Otherwise people may get confused.

-- Steve



> +
> +DEFINE_EVENT(watchdog_template, watchdog_ping,
> +	TP_PROTO(struct watchdog_device *wdd, int err),
> +	TP_ARGS(wdd, err));
> +
> +DEFINE_EVENT(watchdog_template, watchdog_stop,
> +	TP_PROTO(struct watchdog_device *wdd, int err),
> +	TP_ARGS(wdd, err));
> +
> +#endif /* !defined(_TRACE_WATCHDOG_H) || defined(TRACE_HEADER_MULTI_READ) */
> +

Re: [PATCH v2] watchdog: Add tracing events for the most usual watchdog events

[Uwe Kleine-König]

[-- Attachment #1: Type: text/plain, Size: 1597 bytes --]

On Tue, Oct 04, 2022 at 06:51:46PM -0400, Steven Rostedt wrote:
> On Tue,  4 Oct 2022 11:19:49 +0200
> Uwe Kleine-König <u.kleine-koenig@pengutronix.de> wrote:
> 
> > +DEFINE_EVENT(watchdog_template, watchdog_start,
> > +	TP_PROTO(struct watchdog_device *wdd, int err),
> > +	TP_ARGS(wdd, err));
> > +
> > +TRACE_EVENT(watchdog_set_timeout,
> > +
> > +	TP_PROTO(struct watchdog_device *wdd, unsigned int timeout, int err),
> > +
> > +	TP_ARGS(wdd, timeout, err),
> > +
> > +	TP_STRUCT__entry(
> > +		__field(int, id)
> > +		__field(unsigned int, timeout)
> > +		__field(int, err)
> > +	),
> > +
> > +	TP_fast_assign(
> > +		__entry->id = wdd->id;
> > +		__entry->timeout = timeout;
> > +		__entry->err = err;
> > +	),
> > +
> > +	TP_printk("watchdog%d timeout=%u err=%d", __entry->id, __entry->timeout, __entry->err)
> > +);
> 
> Nit, but I would probably put the above TRACE_EVENT() below the two
> DEFINE_EVENT()s below. That way we have all the DEFINE_EVENT()s for a
> specific DECLARE_EVENT_CLASS() together. Otherwise people may get confused.

I thought about that, too. The argument for the order I chose is that
having start at the start and stop at the end is also intuitive.

But I don't care much and would let the watchdog guys decide what they
prefer.

@Wim+Guenter: Feel free to reorder at application time or ask for a v3
if this v2 doesn't fit your preference.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v2] watchdog: Add tracing events for the most usual watchdog events

[Guenter Roeck]

On Wed, Oct 05, 2022 at 09:19:46AM +0200, Uwe Kleine-König wrote:
> On Tue, Oct 04, 2022 at 06:51:46PM -0400, Steven Rostedt wrote:
> > On Tue,  4 Oct 2022 11:19:49 +0200
> > Uwe Kleine-König <u.kleine-koenig@pengutronix.de> wrote:
> > 
> > > +DEFINE_EVENT(watchdog_template, watchdog_start,
> > > +	TP_PROTO(struct watchdog_device *wdd, int err),
> > > +	TP_ARGS(wdd, err));
> > > +
> > > +TRACE_EVENT(watchdog_set_timeout,
> > > +
> > > +	TP_PROTO(struct watchdog_device *wdd, unsigned int timeout, int err),
> > > +
> > > +	TP_ARGS(wdd, timeout, err),
> > > +
> > > +	TP_STRUCT__entry(
> > > +		__field(int, id)
> > > +		__field(unsigned int, timeout)
> > > +		__field(int, err)
> > > +	),
> > > +
> > > +	TP_fast_assign(
> > > +		__entry->id = wdd->id;
> > > +		__entry->timeout = timeout;
> > > +		__entry->err = err;
> > > +	),
> > > +
> > > +	TP_printk("watchdog%d timeout=%u err=%d", __entry->id, __entry->timeout, __entry->err)
> > > +);
> > 
> > Nit, but I would probably put the above TRACE_EVENT() below the two
> > DEFINE_EVENT()s below. That way we have all the DEFINE_EVENT()s for a
> > specific DECLARE_EVENT_CLASS() together. Otherwise people may get confused.
> 
> I thought about that, too. The argument for the order I chose is that
> having start at the start and stop at the end is also intuitive.
> 
> But I don't care much and would let the watchdog guys decide what they
> prefer.
> 
> @Wim+Guenter: Feel free to reorder at application time or ask for a v3
> if this v2 doesn't fit your preference.

For my part I would prefer a version with Steven's Reviewed-by: tag,
whatever it is.

Guenter

Re: [PATCH v2] watchdog: Add tracing events for the most usual watchdog events

[Steven Rostedt]

On Wed, 5 Oct 2022 12:39:24 -0700
Guenter Roeck <linux@roeck-us.net> wrote:

> > > Nit, but I would probably put the above TRACE_EVENT() below the two
> > > DEFINE_EVENT()s below. That way we have all the DEFINE_EVENT()s for a
> > > specific DECLARE_EVENT_CLASS() together. Otherwise people may get confused.  
> > 
> > I thought about that, too. The argument for the order I chose is that
> > having start at the start and stop at the end is also intuitive.
> > 
> > But I don't care much and would let the watchdog guys decide what they
> > prefer.
> > 
> > @Wim+Guenter: Feel free to reorder at application time or ask for a v3
> > if this v2 doesn't fit your preference.  
> 
> For my part I would prefer a version with Steven's Reviewed-by: tag,
> whatever it is.

I much rather have the DEFINE_EVENTS followed by the DECLARE_EVENT_CLASS()
as that's what most people look for.

For start and stop being together, I believe that will not trip many people
up, where as the DEFINE_EVENTS() scattering will.

I'm OK if maintainers of the code are fine with the scattering, but it will
break precedence. 

-- Steve

Re: [PATCH v2] watchdog: Add tracing events for the most usual watchdog events

[Guenter Roeck]

On Wed, Oct 05, 2022 at 05:11:36PM -0400, Steven Rostedt wrote:
> On Wed, 5 Oct 2022 12:39:24 -0700
> Guenter Roeck <linux@roeck-us.net> wrote:
> 
> > > > Nit, but I would probably put the above TRACE_EVENT() below the two
> > > > DEFINE_EVENT()s below. That way we have all the DEFINE_EVENT()s for a
> > > > specific DECLARE_EVENT_CLASS() together. Otherwise people may get confused.  
> > > 
> > > I thought about that, too. The argument for the order I chose is that
> > > having start at the start and stop at the end is also intuitive.
> > > 
> > > But I don't care much and would let the watchdog guys decide what they
> > > prefer.
> > > 
> > > @Wim+Guenter: Feel free to reorder at application time or ask for a v3
> > > if this v2 doesn't fit your preference.  
> > 
> > For my part I would prefer a version with Steven's Reviewed-by: tag,
> > whatever it is.
> 
> I much rather have the DEFINE_EVENTS followed by the DECLARE_EVENT_CLASS()
> as that's what most people look for.
> 
> For start and stop being together, I believe that will not trip many people
> up, where as the DEFINE_EVENTS() scattering will.
> 

I agree. Uwe, please send v3 with those changes.

Thanks,
Guenter