[PATCH] drivers: i2c-core: Add trace events for i2c_master_send()/i2c_master_recv()

[PATCH] drivers: i2c-core: Add trace events for i2c_master_send()/i2c_master_recv()

[Dimitris Papastamos]

Add basic trace events for i2c_master_send() and i2c_master_recv() so we
can observe the I2C activity without filling up the dmesg ring buffer.  It
also makes it easy to enable and disable tracing of the i2c-core.

Signed-off-by: Dimitris Papastamos <dp-yzvPICuk2AATkU/dhu1WVueM+bqZidxxQQ4Iyu8u01E@public.gmane.org>
---
 drivers/i2c/i2c-core.c     |    4 +++
 include/trace/events/i2c.h |   55 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 59 insertions(+), 0 deletions(-)
 create mode 100644 include/trace/events/i2c.h

diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
index f0bd5bc..2c2bb45 100644
--- a/drivers/i2c/i2c-core.c
+++ b/drivers/i2c/i2c-core.c
@@ -42,6 +42,8 @@
 
 #include "i2c-core.h"
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/i2c.h>
 
 /* core_lock protects i2c_adapter_idr, and guarantees
    that device detection, deletion of detected devices, and attach_adapter
@@ -1366,6 +1368,7 @@ int i2c_master_send(const struct i2c_client *client, const char *buf, int count)
 	msg.buf = (char *)buf;
 
 	ret = i2c_transfer(adap, &msg, 1);
+	trace_i2c_master_send(&msg, ret);
 
 	/* If everything went ok (i.e. 1 msg transmitted), return #bytes
 	   transmitted, else error code. */
@@ -1394,6 +1397,7 @@ int i2c_master_recv(const struct i2c_client *client, char *buf, int count)
 	msg.buf = buf;
 
 	ret = i2c_transfer(adap, &msg, 1);
+	trace_i2c_master_recv(&msg, ret);
 
 	/* If everything went ok (i.e. 1 msg transmitted), return #bytes
 	   transmitted, else error code. */
diff --git a/include/trace/events/i2c.h b/include/trace/events/i2c.h
new file mode 100644
index 0000000..cdf64e8
--- /dev/null
+++ b/include/trace/events/i2c.h
@@ -0,0 +1,55 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM i2c
+
+#if !defined(_TRACE_I2C_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_I2C_H
+
+#include <linux/ktime.h>
+#include <linux/tracepoint.h>
+
+struct i2c_msg;
+
+DECLARE_EVENT_CLASS(i2c_master,
+
+	TP_PROTO(struct i2c_msg *msg, int ret),
+
+	TP_ARGS(msg, ret),
+
+	TP_STRUCT__entry(
+		__field(	u16,	addr	)
+		__field(	u16,	len	)
+		__field(	u16,	flags	)
+		__field(	int,	ret	)
+	),
+
+	TP_fast_assign(
+		__entry->addr = msg->addr;
+		__entry->len = msg->len;
+		__entry->flags = msg->flags;
+		__entry->ret = ret;
+	),
+
+	TP_printk("addr=%x len=%x flags=%x ret=%d", (u16)__entry->addr,
+		  (u16)__entry->len, (u16)__entry->flags, (int)__entry->ret)
+);
+
+DEFINE_EVENT(i2c_master, i2c_master_send,
+
+	TP_PROTO(struct i2c_msg *msg, int ret),
+
+	TP_ARGS(msg, ret)
+
+);
+
+DEFINE_EVENT(i2c_master, i2c_master_recv,
+
+	TP_PROTO(struct i2c_msg *msg, int ret),
+
+	TP_ARGS(msg, ret)
+
+);
+
+#endif /* _TRACE_I2C_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
-- 
1.7.3.5

Re: [PATCH] drivers: i2c-core: Add trace events for i2c_master_send()/i2c_master_recv()

[Jean Delvare]

On Thu, 27 Jan 2011 11:01:06 +0000, Dimitris Papastamos wrote:
> Add basic trace events for i2c_master_send() and i2c_master_recv() so we
> can observe the I2C activity without filling up the dmesg ring buffer.  It
> also makes it easy to enable and disable tracing of the i2c-core.

This doesn't make sense. If you want to trace something, trace
i2c_transfer(). i2c_master_send() and i2c_master_recv() are only helpers
for the most simple cases, so tracing only them is pointless.

Note that even with this, you'll miss calls to i2c_smbus_xfer(), which
is the other access point to I2C/SMBus controllers. So I guess you want
to trace it as well.

-- 
Jean Delvare

Re: [PATCH] drivers: i2c-core: Add trace events for i2c_master_send()/i2c_master_recv()

[Mark Brown]

On Thu, Jan 27, 2011 at 01:33:21PM +0100, Jean Delvare wrote:
> On Thu, 27 Jan 2011 11:01:06 +0000, Dimitris Papastamos wrote:

> > Add basic trace events for i2c_master_send() and i2c_master_recv() so we
> > can observe the I2C activity without filling up the dmesg ring buffer.  It
> > also makes it easy to enable and disable tracing of the i2c-core.

> This doesn't make sense. If you want to trace something, trace
> i2c_transfer(). i2c_master_send() and i2c_master_recv() are only helpers
> for the most simple cases, so tracing only them is pointless.

> Note that even with this, you'll miss calls to i2c_smbus_xfer(), which
> is the other access point to I2C/SMBus controllers. So I guess you want
> to trace it as well.

Might also be interesting to do start/stop tracing - I2C is relatively
slow and causes scheduling.  Dunno if it's worth it, though you could
always turn on only one of the events if it was too boring to have both.