[v3] PM / OPP: optimize dev_pm_opp_set_rate() performance a bit

Commit Message

Jisheng Zhang July 25, 2016, 6:11 a.m. UTC

In dev_pm_opp_set_rate(), _find_opp_table() is called 4 times: once by
_get_opp_clk(), once by dev_pm_opp_set_rate() itself, and twice by
dev_pm_opp_find_freq_ceil(). If there are several opp_tables in the
system, three times of opp table finding is a big waste. This patch
reduced the call of _find_opp_table() to twice.

Signed-off-by: Jisheng Zhang <[e-mail]>
---
Since v2:
 - fix gcc warnings

Since v1:
 - avoid opp_table pointer usage after the rcu_read_unlock() as pointed
   out by Viresh, thus we can only reduced the call of _find_opp_table
   to twice.

 drivers/base/power/opp/core.c | 31 +++++++++++++++++++------------
 1 file changed, 19 insertions(+), 12 deletions(-)

Comments

Viresh Kumar July 25, 2016, 10:34 p.m. UTC | #1

On 25-07-16, 14:11, Jisheng Zhang wrote:
> In dev_pm_opp_set_rate(), _find_opp_table() is called 4 times: once by
> _get_opp_clk(), once by dev_pm_opp_set_rate() itself, and twice by
> dev_pm_opp_find_freq_ceil(). If there are several opp_tables in the
> system, three times of opp table finding is a big waste. This patch
> reduced the call of _find_opp_table() to twice.
> 
> Signed-off-by: Jisheng Zhang <[e-mail]>
> ---
> Since v2:
>  - fix gcc warnings
> 
> Since v1:
>  - avoid opp_table pointer usage after the rcu_read_unlock() as pointed
>    out by Viresh, thus we can only reduced the call of _find_opp_table
>    to twice.
> 
>  drivers/base/power/opp/core.c | 31 +++++++++++++++++++------------
>  1 file changed, 19 insertions(+), 12 deletions(-)

Acked-by: Viresh Kumar <[e-mail]>

Patch

diff --git a/drivers/base/power/opp/core.c b/drivers/base/power/opp/core.c
index 7c04c87..df0c709 100644
--- a/drivers/base/power/opp/core.c
+++ b/drivers/base/power/opp/core.c
@@ -402,6 +402,22 @@  struct dev_pm_opp *dev_pm_opp_find_freq_exact(struct device *dev,
 }
 EXPORT_SYMBOL_GPL(dev_pm_opp_find_freq_exact);
 
+static noinline struct dev_pm_opp *_find_freq_ceil(struct opp_table *opp_table,
+						   unsigned long *freq)
+{
+	struct dev_pm_opp *temp_opp, *opp = ERR_PTR(-ERANGE);
+
+	list_for_each_entry_rcu(temp_opp, &opp_table->opp_list, node) {
+		if (temp_opp->available && temp_opp->rate >= *freq) {
+			opp = temp_opp;
+			*freq = opp->rate;
+			break;
+		}
+	}
+
+	return opp;
+}
+
 /**
  * dev_pm_opp_find_freq_ceil() - Search for an rounded ceil freq
  * @dev:	device for which we do this operation
@@ -427,7 +443,6 @@  struct dev_pm_opp *dev_pm_opp_find_freq_ceil(struct device *dev,
 					     unsigned long *freq)
 {
 	struct opp_table *opp_table;
-	struct dev_pm_opp *temp_opp, *opp = ERR_PTR(-ERANGE);
 
 	opp_rcu_lockdep_assert();
 
@@ -440,15 +455,7 @@  struct dev_pm_opp *dev_pm_opp_find_freq_ceil(struct device *dev,
 	if (IS_ERR(opp_table))
 		return ERR_CAST(opp_table);
 
-	list_for_each_entry_rcu(temp_opp, &opp_table->opp_list, node) {
-		if (temp_opp->available && temp_opp->rate >= *freq) {
-			opp = temp_opp;
-			*freq = opp->rate;
-			break;
-		}
-	}
-
-	return opp;
+	return _find_freq_ceil(opp_table, freq);
 }
 EXPORT_SYMBOL_GPL(dev_pm_opp_find_freq_ceil);
 
@@ -612,7 +619,7 @@  int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq)
 		return PTR_ERR(opp_table);
 	}
 
-	old_opp = dev_pm_opp_find_freq_ceil(dev, &old_freq);
+	old_opp = _find_freq_ceil(opp_table, &old_freq);
 	if (!IS_ERR(old_opp)) {
 		ou_volt = old_opp->u_volt;
 		ou_volt_min = old_opp->u_volt_min;
@@ -622,7 +629,7 @@  int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq)
 			__func__, old_freq, PTR_ERR(old_opp));
 	}
 
-	opp = dev_pm_opp_find_freq_ceil(dev, &freq);
+	opp = _find_freq_ceil(opp_table, &freq);
 	if (IS_ERR(opp)) {
 		ret = PTR_ERR(opp);
 		dev_err(dev, "%s: failed to find OPP for freq %lu (%d)\n",