Re: [PATCH V2 2/2] cpufreq: dt: Refactor initialization to handle probe deferral properly

From: Viresh Kumar <viresh.kumar@linaro.org>
To: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: ulf.hansson@linaro.org, "Rafael J. Wysocki" <rjw@rjwysocki.net>,
	Liam Girdwood <lgirdwood@gmail.com>,
	Mark Brown <broonie@kernel.org>,
	linux-pm@vger.kernel.org,
	Vincent Guittot <vincent.guittot@linaro.org>,
	Stephen Boyd <sboyd@kernel.org>, Nishanth Menon <nm@ti.com>,
	nks@flawful.org, georgi.djakov@linaro.org,
	Stephan Gerhold <stephan@gerhold.net>,
	linux-kernel@vger.kernel.org,
	'Linux Samsung SOC' <linux-samsung-soc@vger.kernel.org>,
	Krzysztof Kozlowski <krzk@kernel.org>
Subject: Re: [PATCH V2 2/2] cpufreq: dt: Refactor initialization to handle probe deferral properly
Date: Tue, 1 Sep 2020 15:15:08 +0530	[thread overview]
Message-ID: <20200901094508.4sgyfv3yj575wlzp@vireshk-i7> (raw)
In-Reply-To: <00a87bad-f750-b08c-4ccb-545b90dd87fc@samsung.com>

On 01-09-20, 10:57, Marek Szyprowski wrote:
> This patch landed in linux-next about a week ago. It introduces a 
> following warning on Samsung Exnyos3250 SoC:
> 
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 1000000000, volt: 1150000, enabled: 1. New: freq: 1000000000, volt: 
> 1150000, enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 900000000, volt: 1112500, enabled: 1. New: freq: 900000000, volt: 
> 1112500, enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 800000000, volt: 1075000, enabled: 1. New: freq: 800000000, volt: 
> 1075000, enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 700000000, volt: 1037500, enabled: 1. New: freq: 700000000, volt: 
> 1037500, enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 600000000, volt: 1000000, enabled: 1. New: freq: 600000000, volt: 
> 1000000, enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 500000000, volt: 962500, enabled: 1. New: freq: 500000000, volt: 962500, 
> enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 400000000, volt: 925000, enabled: 1. New: freq: 400000000, volt: 925000, 
> enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 300000000, volt: 887500, enabled: 1. New: freq: 300000000, volt: 887500, 
> enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 200000000, volt: 850000, enabled: 1. New: freq: 200000000, volt: 850000, 
> enabled: 1
> cpu cpu1: _opp_is_duplicate: duplicate OPPs detected. Existing: freq: 
> 100000000, volt: 850000, enabled: 1. New: freq: 100000000, volt: 850000, 
> enabled: 1
> 
> I've checked a bit and this is related to the fact that Exynos3250 SoC 
> use OPP-v1 table. Is this intentional? It is not a problem to convert it 
> to OPP-v2 and mark OPP table as shared, but this is a kind of a regression.

It took me 20 minutes for me to see "where has my patch gone" :(

I wrote a small patch for that to work without any issues, but not
sure how I missed or abandoned it. Anyway, here is the diff again and
I will send it out again once you confirm it fixes the issue. Can you
please also test your driver as a module with multiple insertion/removals ?

diff --git a/drivers/opp/of.c b/drivers/opp/of.c
index 5dac8bffd68c..e72753be7dc7 100644
--- a/drivers/opp/of.c
+++ b/drivers/opp/of.c
@@ -905,6 +905,16 @@ static int _of_add_opp_table_v1(struct device *dev, struct opp_table *opp_table)
        const __be32 *val;
        int nr, ret = 0;
 
+       mutex_lock(&opp_table->lock);
+       if (opp_table->parsed_static_opps) {
+               opp_table->parsed_static_opps++;
+               mutex_unlock(&opp_table->lock);
+               return 0;
+       }
+
+       opp_table->parsed_static_opps = 1;
+       mutex_unlock(&opp_table->lock);
+
        prop = of_find_property(dev->of_node, "operating-points", NULL);
        if (!prop)
                return -ENODEV;
@@ -921,10 +931,6 @@ static int _of_add_opp_table_v1(struct device *dev, struct opp_table *opp_table)
                return -EINVAL;
        }
 
-       mutex_lock(&opp_table->lock);
-       opp_table->parsed_static_opps = 1;
-       mutex_unlock(&opp_table->lock);
-
        val = prop->value;
        while (nr) {
                unsigned long freq = be32_to_cpup(val++) * 1000;

-- 
viresh
