Tegra's cpufreq driver was maintaining requested target frequencies in an array: target_cpu_speed. And then finally setting the highest requested freq in the core. This was probably done because both cores share clock line and logically we want to set both cores to the max frequency requested..
But this wasn't required to be done in individual CPUFreq drivers, its already taken care of by CPUFreq governors. They evaluate load for all CPUs and finally call target only for the frequency corresponding to max load.
So, get rid of this stuff from Tegra's cpufreq driver.
Signed-off-by: Viresh Kumar viresh.kumar@linaro.org --- Hi Stephen,
Its only build tested and depends on lots of stuff that I have already sent for cpufreq core and its drivers. All of that is pushed here: https://git.linaro.org/gitweb?p=people/vireshk/linux.git%3Ba=shortlog%3Bh=re...
And only Tegra+cpufreq-core patches are pushed here (only 13 patches): https://git.linaro.org/gitweb?p=people/vireshk/linux.git%3Ba=shortlog%3Bh=re...
You can probably try cpufreq-next-tegra branch for testing on some real hardware.
drivers/cpufreq/tegra-cpufreq.c | 35 ++++++----------------------------- 1 file changed, 6 insertions(+), 29 deletions(-)
diff --git a/drivers/cpufreq/tegra-cpufreq.c b/drivers/cpufreq/tegra-cpufreq.c index b376b67..3f25ab6 100644 --- a/drivers/cpufreq/tegra-cpufreq.c +++ b/drivers/cpufreq/tegra-cpufreq.c @@ -47,7 +47,6 @@ static struct clk *pll_x_clk; static struct clk *pll_p_clk; static struct clk *emc_clk;
-static unsigned long target_cpu_speed[NUM_CPUS]; static DEFINE_MUTEX(tegra_cpu_lock); static bool is_suspended;
@@ -103,9 +102,6 @@ static int tegra_update_cpu_speed(struct cpufreq_policy *policy, { int ret = 0;
- if (tegra_getspeed(0) == rate) - return ret; - /* * Vote on memory bus frequency based on cpu frequency * This sets the minimum frequency, display or avp may request higher @@ -125,35 +121,16 @@ static int tegra_update_cpu_speed(struct cpufreq_policy *policy, return ret; }
-static unsigned long tegra_cpu_highest_speed(void) -{ - unsigned long rate = 0; - int i; - - for_each_online_cpu(i) - rate = max(rate, target_cpu_speed[i]); - return rate; -} - static int tegra_target(struct cpufreq_policy *policy, unsigned int index) { - unsigned int freq; - int ret = 0; + int ret = -EBUSY;
mutex_lock(&tegra_cpu_lock);
- if (is_suspended) { - ret = -EBUSY; - goto out; - } - - freq = freq_table[index].frequency; + if (!is_suspended) + ret = tegra_update_cpu_speed(policy, + freq_table[index].frequency);
- target_cpu_speed[policy->cpu] = freq; - - ret = tegra_update_cpu_speed(policy, tegra_cpu_highest_speed()); - -out: mutex_unlock(&tegra_cpu_lock); return ret; } @@ -167,7 +144,8 @@ static int tegra_pm_notify(struct notifier_block *nb, unsigned long event, is_suspended = true; pr_info("Tegra cpufreq suspend: setting frequency to %d kHz\n", freq_table[0].frequency); - tegra_update_cpu_speed(policy, freq_table[0].frequency); + if (tegra_getspeed(0) != freq_table[0].frequency) + tegra_update_cpu_speed(policy, freq_table[0].frequency); cpufreq_cpu_put(policy); } else if (event == PM_POST_SUSPEND) { is_suspended = false; @@ -190,7 +168,6 @@ static int tegra_cpu_init(struct cpufreq_policy *policy) clk_prepare_enable(cpu_clk);
cpufreq_table_validate_and_show(policy, freq_table); - target_cpu_speed[policy->cpu] = tegra_getspeed(policy->cpu);
/* FIXME: what's the actual transition time? */ policy->cpuinfo.transition_latency = 300 * 1000;
On 08/21/2013 04:31 AM, Viresh Kumar wrote:
Tegra's cpufreq driver was maintaining requested target frequencies in an array: target_cpu_speed. And then finally setting the highest requested freq in the core. This was probably done because both cores share clock line and logically we want to set both cores to the max frequency requested..
But this wasn't required to be done in individual CPUFreq drivers, its already taken care of by CPUFreq governors. They evaluate load for all CPUs and finally call target only for the frequency corresponding to max load.
So, get rid of this stuff from Tegra's cpufreq driver.
Signed-off-by: Viresh Kumar viresh.kumar@linaro.org
Hi Stephen,
Its only build tested and depends on lots of stuff that I have already sent for cpufreq core and its drivers. All of that is pushed here: https://git.linaro.org/gitweb?p=people/vireshk/linux.git%3Ba=shortlog%3Bh=re...
And only Tegra+cpufreq-core patches are pushed here (only 13 patches): https://git.linaro.org/gitweb?p=people/vireshk/linux.git%3Ba=shortlog%3Bh=re...
You can probably try cpufreq-next-tegra branch for testing on some real hardware.
Acked-by: Stephen Warren swarren@nvidia.com
I did test your branch on a Tegra20 and Tegra30 board without issues. But recall that our cpufreq driver doesn't actually get initialized since the conversion of Tegra to the common clock framework, so I haven't really tested the cpufreq changes, except to ensure that nothing in those branches breaks other basic functionality.
linaro-kernel@lists.linaro.org