02.10.2019 02:35, Chanwoo Choi пишет:
> Hi,
>
> On 19. 8. 12. 오전 6:23, Dmitry Osipenko wrote:
>> There are cases where unnecessary ACTMON interrupts could be avoided,
>> like when one memory client device requests higher clock rate than the
>> other or when clock rate is manually limited using sysfs devfreq
>> parameters. These cases could be avoided by tuning upper watermark or
>> disabling hardware events when min/max boosting thresholds are reached.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@xxxxxxxxx>
>> ---
>> drivers/devfreq/tegra30-devfreq.c | 87 ++++++++++++++++++++++++++++---
>> 1 file changed, 80 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/devfreq/tegra30-devfreq.c b/drivers/devfreq/tegra30-devfreq.c
>> index 43d50b4366dd..a2623de56d20 100644
>> --- a/drivers/devfreq/tegra30-devfreq.c
>> +++ b/drivers/devfreq/tegra30-devfreq.c
>> @@ -312,7 +312,8 @@ static void tegra_actmon_get_lower_upper(struct tegra_devfreq *tegra,
>> }
>>
>> static void tegra_devfreq_update_avg_wmark(struct tegra_devfreq *tegra,
>> - struct tegra_devfreq_device *dev)
>> + struct tegra_devfreq_device *dev,
>> + unsigned long freq)
>> {
>> unsigned long avg_dependency_freq, lower, upper;
>>
>> @@ -320,6 +321,22 @@ static void tegra_devfreq_update_avg_wmark(struct tegra_devfreq *tegra,
>>
>> avg_dependency_freq = tegra_actmon_dev_avg_dependency_freq(tegra, dev);
>>
>> + /*
>> + * If cumulative EMC frequency selection (MCALL / min_freq) is
>> + * higher than the device's, then there is no need to set upper
>> + * watermark to a lower value because it will result in unnecessary
>> + * upper interrupts.
>> + *
>> + * Note that average watermarks are also updated after EMC
>> + * clock rate change, hence if clock rate goes down, then the
>> + * watermarks will be set in accordance to the new rate after
>> + * changing the rate. There are other ways to achieve the same
>> + * result, but this one is probably the least churning, although
>> + * it may look a bit convoluted.
>> + */
>> + if (freq * ACTMON_SAMPLING_PERIOD > upper)
>> + upper = freq * ACTMON_SAMPLING_PERIOD;
>> +
>> /*
>> * We want to get interrupts when MCCPU client crosses the
>> * dependency threshold in order to take into / out of account
>> @@ -361,7 +378,18 @@ static void tegra_devfreq_update_wmark(struct tegra_devfreq *tegra,
>> tegra_actmon_get_lower_upper(tegra, dev, freq - 1, &lower, &upper);
>>
>> delta = do_percent(upper - lower, dev->config->boost_up_threshold);
>> - device_writel(dev, lower + delta, ACTMON_DEV_UPPER_WMARK);
>
>
> Also, this patch edits the added codes on front patch.
> This code was added on patch5 and then delete it on this patch.
> If it is not necessary, you can remove it on patch5 by refactoring.
>
>> +
>> + /*
>> + * The memory events count could go a bit higher than the maximum
>> + * defined by the OPPs, hence make the upper watermark infinitely
>> + * high to avoid unnecessary upper interrupts in that case.
>> + */
>> + if (freq == tegra->max_freq)
>> + upper = ULONG_MAX;
>> + else
>> + upper = lower + delta;
>> +
>> + device_writel(dev, upper, ACTMON_DEV_UPPER_WMARK);
>
> I think that the changes of tegra_devfreq_update_avg_wmark() on this patch
> can be merged to patch5.
Okay, I'll revisit these parts of tegra_devfreq_update_avg_wmark() and will move them to
patch5 if there won't be any major obstacles.
