linux-kernel - Re: [PATCH 1/3] pwm: pca9685: Switch to atomic API

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20201119160013.GA217674@workstation.tuxnet>
Date:   Thu, 19 Nov 2020 17:00:13 +0100
From:   Clemens Gruber <clemens.gruber@...ruber.com>
To:     Sven Van Asbroeck <thesven73@...il.com>
Cc:     linux-pwm@...r.kernel.org,
        Thierry Reding <thierry.reding@...il.com>,
        Uwe Kleine-König 
        <u.kleine-koenig@...gutronix.de>, Lee Jones <lee.jones@...aro.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Mika Westerberg <mika.westerberg@...ux.intel.com>,
        David Jander <david@...tonic.nl>
Subject: Re: [PATCH 1/3] pwm: pca9685: Switch to atomic API

On Thu, Nov 19, 2020 at 09:58:26AM -0500, Sven Van Asbroeck wrote:
> On Thu, Nov 19, 2020 at 5:00 AM Clemens Gruber
> <clemens.gruber@...ruber.com> wrote:
> >
> > > You appear to mix cached and uncached uses of prescale,
> > > is there a need for this? If not, perhaps pick one and use
> > > it consistently?
> >
> > Yes, sticking to the cached value is probably the way to go.
> >
> 
> I would suggest going one step further, and turn on the cache in
> regmap, i.e. .cache_type = REGCACHE_RBTREE, then:
> - no need to cache pca->prescale explicitly, you can just read it with
>   regmap_read() every time, and it won't result in bus activity.
>   then you can eliminate pca->prescale, which simplifies the driver.
> - pca9685_pwm_get_state() no longer results in bus reads, every regmap_read()
>   is cached, this is extremely efficient.
> - pca9685_pwm_apply() and pca9685_pwm_gpio_set() now only does bus writes if
>   registers actually change, i.e. calling pwm_apply() multiple times in a row
>   with the same parameters, writes the registers only once.

Interesting, I will look into that.

> 
> We can do this safely because this chip never actively writes to its
> registers (as far as I know).

I think so too.

> 
> But maybe that's a suggestion for a follow-up patch...
> 
> > > Also, if the prescale register contains an invalid value
> > > during probe(), e.g. 0x00 or 0x01, would it make sense
> > > to explicitly overwrite it with a valid setting?
> >
> > As long as it is overwritten with a correct setting when the PWM is used
> > for the first time, it should be OK?
> 
> I'm not sure. Consider the following scenario:
> - prescale register is invalid at probe, say it contains 0x02
> - user calls pwm_apply() but with an invalid period, which results
>   in a calculated prescale value of 0x02
> - pca9685_pwm_apply() skips prescale setup because prescale does not
>   change, returns OK(0)
> - user believes setup was ok, actually it's broken...

Makes sense. I will write the default prescale setting in case we read
an invalid one from the register.

> 
> Also, some people use this chip exclusively as a gpiochip, in that
> case the prescale register is never touched. So an invalid prescale
> at probe is never corrected.
> 
> Speaking of the gpiochip side, would it make sense to call
> pca9685_pwm_full_on()/_off() in pca9685_pwm_gpio_set() too?

Yes, I think so. Would be cleaner and we avoid setting all registers to
0 when the GPIO is disabled.

--

One thing I noticed: The driver currently assumes that it comes out of
POR in "active" state (comment at end of probe and PM calls).
However, the SLEEP bit is set by default / after POR.

Do you agree with the following solution?
1) In .probe: call pm_runtime_set_suspended() instead of _set_active()
   (If CONFIG_PM is enabled, the SLEEP bit will be cleared in .resume)
2) If !CONFIG_PM: Clear the SLEEP bit in .probe

Thanks,
Clemens