[PATCH 0/2] clk: Rate Request Tracing

[PATCH 0/2] clk: Rate Request Tracing

[Maxime Ripard]

Hi,

Here's a couple of patches to enable the tracing of clk_rate_requests as they
are submitted to round/determine_rate.

Let me know what you think,
Maxime

To: Michael Turquette <mturquette@baylibre.com>
To: Stephen Boyd <sboyd@kernel.org>
To: Steven Rostedt <rostedt@goodmis.org>
To: Masami Hiramatsu <mhiramat@kernel.org>
Cc: linux-clk@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Maxime Ripard <maxime@cerno.tech>

---
Maxime Ripard (2):
      clk: Store clk_core for clk_rate_request
      clk: Add trace events for rate requests

 drivers/clk/clk.c            | 32 ++++++++++++++++++++++++++++++++
 include/linux/clk-provider.h |  2 ++
 include/trace/events/clk.h   | 43 +++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 77 insertions(+)
---
base-commit: 69a14e4a9d5a0e1ed73063e0d439c3cb33ce9829
change-id: 20221018-clk-rate-request-tracing-74e80b4534d1

Best regards,
-- 
Maxime Ripard <maxime@cerno.tech>

[PATCH 1/2] clk: Store clk_core for clk_rate_request

[Maxime Ripard]

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
---
 drivers/clk/clk.c            | 1 +
 include/linux/clk-provider.h | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index 11c41d987ff4..bc6f964cb676 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -1468,6 +1468,7 @@ static void clk_core_init_rate_req(struct clk_core * const core,
 	if (!core)
 		return;
 
+	req->core = core;
 	req->rate = rate;
 	clk_core_get_boundaries(core, &req->min_rate, &req->max_rate);
 
diff --git a/include/linux/clk-provider.h b/include/linux/clk-provider.h
index 267cd06b54a0..842e72a5348f 100644
--- a/include/linux/clk-provider.h
+++ b/include/linux/clk-provider.h
@@ -44,6 +44,7 @@ struct dentry;
  *
  * Should be initialized by calling clk_hw_init_rate_request().
  *
+ * @core: 		Pointer to the struct clk_core affected by this request
  * @rate:		Requested clock rate. This field will be adjusted by
  *			clock drivers according to hardware capabilities.
  * @min_rate:		Minimum rate imposed by clk users.
@@ -55,6 +56,7 @@ struct dentry;
  *
  */
 struct clk_rate_request {
+	struct clk_core *core;
 	unsigned long rate;
 	unsigned long min_rate;
 	unsigned long max_rate;

-- 
b4 0.11.0-dev-7da52

Re: [PATCH 1/2] clk: Store clk_core for clk_rate_request

[Stephen Boyd]

Quoting Maxime Ripard (2022-10-18 06:56:41)
> Signed-off-by: Maxime Ripard <maxime@cerno.tech>

Why?

[PATCH 2/2] clk: Add trace events for rate requests

[Maxime Ripard]

It is currently fairly difficult to follow what clk_rate_request are
issued, and how they have been modified once done.

Indeed, there's multiple paths that can be taken, some functions are
recursive and will just forward the request to its parent, etc.

Adding a lot of debug prints is just not very convenient, so let's add
trace events for the clock requests, one before they are submitted and
one after they are returned.

That way we can simply toggle the tracing on without modifying the
kernel code and without affecting performances or the kernel logs too
much.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
---
 drivers/clk/clk.c          | 31 +++++++++++++++++++++++++++++++
 include/trace/events/clk.h | 43 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 74 insertions(+)

diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index bc6f964cb676..343c50e7e214 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -603,10 +603,15 @@ int clk_mux_determine_rate_flags(struct clk_hw *hw,
 			}
 
 			clk_core_forward_rate_req(core, req, parent, &parent_req, req->rate);
+
+			trace_clk_rate_request_start(&parent_req);
+
 			ret = clk_core_round_rate_nolock(parent, &parent_req);
 			if (ret)
 				return ret;
 
+			trace_clk_rate_request_done(&parent_req);
+
 			best = parent_req.rate;
 		} else if (parent) {
 			best = clk_core_get_rate_nolock(parent);
@@ -630,10 +635,15 @@ int clk_mux_determine_rate_flags(struct clk_hw *hw,
 			struct clk_rate_request parent_req;
 
 			clk_core_forward_rate_req(core, req, parent, &parent_req, req->rate);
+
+			trace_clk_rate_request_start(&parent_req);
+
 			ret = clk_core_round_rate_nolock(parent, &parent_req);
 			if (ret)
 				continue;
 
+			trace_clk_rate_request_done(&parent_req);
+
 			parent_rate = parent_req.rate;
 		} else {
 			parent_rate = clk_core_get_rate_nolock(parent);
@@ -1551,10 +1561,15 @@ static int clk_core_round_rate_nolock(struct clk_core *core,
 		struct clk_rate_request parent_req;
 
 		clk_core_forward_rate_req(core, req, core->parent, &parent_req, req->rate);
+
+		trace_clk_rate_request_start(&parent_req);
+
 		ret = clk_core_round_rate_nolock(core->parent, &parent_req);
 		if (ret)
 			return ret;
 
+		trace_clk_rate_request_done(&parent_req);
+
 		req->best_parent_rate = parent_req.rate;
 		req->rate = parent_req.rate;
 
@@ -1605,10 +1620,14 @@ unsigned long clk_hw_round_rate(struct clk_hw *hw, unsigned long rate)
 
 	clk_core_init_rate_req(hw->core, &req, rate);
 
+	trace_clk_rate_request_start(&req);
+
 	ret = clk_core_round_rate_nolock(hw->core, &req);
 	if (ret)
 		return 0;
 
+	trace_clk_rate_request_done(&req);
+
 	return req.rate;
 }
 EXPORT_SYMBOL_GPL(clk_hw_round_rate);
@@ -1637,8 +1656,12 @@ long clk_round_rate(struct clk *clk, unsigned long rate)
 
 	clk_core_init_rate_req(clk->core, &req, rate);
 
+	trace_clk_rate_request_start(&req);
+
 	ret = clk_core_round_rate_nolock(clk->core, &req);
 
+	trace_clk_rate_request_done(&req);
+
 	if (clk->exclusive_count)
 		clk_core_rate_protect(clk->core);
 
@@ -2130,10 +2153,14 @@ static struct clk_core *clk_calc_new_rates(struct clk_core *core,
 
 		clk_core_init_rate_req(core, &req, rate);
 
+		trace_clk_rate_request_start(&req);
+
 		ret = clk_core_determine_round_nolock(core, &req);
 		if (ret < 0)
 			return NULL;
 
+		trace_clk_rate_request_done(&req);
+
 		best_parent_rate = req.best_parent_rate;
 		new_rate = req.rate;
 		parent = req.best_parent_hw ? req.best_parent_hw->core : NULL;
@@ -2329,8 +2356,12 @@ static unsigned long clk_core_req_round_rate_nolock(struct clk_core *core,
 
 	clk_core_init_rate_req(core, &req, req_rate);
 
+	trace_clk_rate_request_start(&req);
+
 	ret = clk_core_round_rate_nolock(core, &req);
 
+	trace_clk_rate_request_done(&req);
+
 	/* restore the protection */
 	clk_core_rate_restore_protect(core, cnt);
 
diff --git a/include/trace/events/clk.h b/include/trace/events/clk.h
index e19edc63ee95..b19f5a549219 100644
--- a/include/trace/events/clk.h
+++ b/include/trace/events/clk.h
@@ -264,6 +264,49 @@ DEFINE_EVENT(clk_duty_cycle, clk_set_duty_cycle_complete,
 	TP_ARGS(core, duty)
 );
 
+DECLARE_EVENT_CLASS(clk_rate_request,
+
+	TP_PROTO(struct clk_rate_request *req),
+
+	TP_ARGS(req),
+
+	TP_STRUCT__entry(
+		__string(        name, req->core ? req->core->name : "none")
+		__field(unsigned long,           min                       )
+		__field(unsigned long,           max                       )
+		__string(       pname, req->best_parent_hw ? clk_hw_get_name(req->best_parent_hw) : "none" )
+		__field(unsigned long,           prate                     )
+	),
+
+	TP_fast_assign(
+		__assign_str(name, req->core ? req->core->name : "none");
+		__assign_str(pname, req->best_parent_hw ? clk_hw_get_name(req->best_parent_hw) : "none");
+		__entry->min = req->min_rate;
+		__entry->max = req->max_rate;
+		__entry->prate = req->best_parent_rate;
+	),
+
+	TP_printk("%s min %lu max %lu, parent %s (%lu)", __get_str(name),
+		  (unsigned long)__entry->min,
+		  (unsigned long)__entry->max,
+		  __get_str(pname),
+		  (unsigned long)__entry->prate)
+);
+
+DEFINE_EVENT(clk_rate_request, clk_rate_request_start,
+
+	TP_PROTO(struct clk_rate_request *req),
+
+	TP_ARGS(req)
+);
+
+DEFINE_EVENT(clk_rate_request, clk_rate_request_done,
+
+	TP_PROTO(struct clk_rate_request *req),
+
+	TP_ARGS(req)
+);
+
 #endif /* _TRACE_CLK_H */
 
 /* This part must be outside protection */

-- 
b4 0.11.0-dev-7da52

Re: [PATCH 2/2] clk: Add trace events for rate requests

[Steven Rostedt]

On Tue, 18 Oct 2022 15:56:42 +0200
Maxime Ripard <maxime@cerno.tech> wrote:

> It is currently fairly difficult to follow what clk_rate_request are
> issued, and how they have been modified once done.
> 
> Indeed, there's multiple paths that can be taken, some functions are
> recursive and will just forward the request to its parent, etc.
> 
> Adding a lot of debug prints is just not very convenient, so let's add
> trace events for the clock requests, one before they are submitted and
> one after they are returned.
> 
> That way we can simply toggle the tracing on without modifying the
> kernel code and without affecting performances or the kernel logs too
> much.
> 
> Signed-off-by: Maxime Ripard <maxime@cerno.tech>

From a tracing point of view:

Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>

-- Steve

> ---
>  drivers/clk/clk.c          | 31 +++++++++++++++++++++++++++++++
>  include/trace/events/clk.h | 43 +++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 74 insertions(+)

Re: [PATCH 2/2] clk: Add trace events for rate requests

[Steven Rostedt]

On Tue, 18 Oct 2022 15:56:42 +0200
Maxime Ripard <maxime@cerno.tech> wrote:

I know I reviewed this but looking at this again, I noticed:


> +DECLARE_EVENT_CLASS(clk_rate_request,
> +
> +	TP_PROTO(struct clk_rate_request *req),
> +
> +	TP_ARGS(req),
> +
> +	TP_STRUCT__entry(
> +		__string(        name, req->core ? req->core->name : "none")
> +		__field(unsigned long,           min                       )
> +		__field(unsigned long,           max                       )
> +		__string(       pname, req->best_parent_hw ? clk_hw_get_name(req->best_parent_hw) : "none" )

It may be best to move the two __string() declarations together. The reason
is that dynamic strings (which __string() is) uses 4 bytes embedded in the
first part of the event. Two bytes for offset, where the dynamic string
exists, and two bytes for the strings length.

On 64 bit machines the above has:

	__string()		4 bytes
	__field(unsigned long)	8 bytes
	__field(unsigned long)	8 bytes
	__string()		4 bytes

and then another unsigned long field below, which is another 8 bytes.

> +		__field(unsigned long,           prate                     )
> +	),
> +
> +	

As compilers tend to use word alignment, the above turns into:

	__string()		4 bytes

	__PADDING__		4 bytes

	__field(unsigned long)	8 bytes
	__field(unsigned long)	8 bytes
	__string()		4 bytes

	__PADDING__		4 bytes

	__field(unsigned long)	8 bytes


Where there will be 8 bytes of padding in that event that wastes precious
ring buffer space. By changing the event to:

	TP_STRUCT__entry(
		__string(        name, req->core ? req->core->name : "none")
		__string(       pname, req->best_parent_hw ? clk_hw_get_name(req->best_parent_hw) : "none" )
		__field(unsigned long,           min                       )
		__field(unsigned long,           max                       )
		__field(unsigned long,           prate                     )
	),

It will turn the size into:

	__string()		4 bytes
	__string()		4 bytes
	__field(unsigned long)	8 bytes
	__field(unsigned long)	8 bytes
	__field(unsigned long)	8 bytes

With no padding and no wasted space.

I would suggest changing this.

-- Steve