Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934099AbcDLONJ (ORCPT ); Tue, 12 Apr 2016 10:13:09 -0400 Received: from down.free-electrons.com ([37.187.137.238]:38932 "EHLO mail.free-electrons.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S932855AbcDLONG (ORCPT ); Tue, 12 Apr 2016 10:13:06 -0400 Date: Tue, 12 Apr 2016 16:13:01 +0200 From: Boris Brezillon To: Thierry Reding Cc: linux-pwm@vger.kernel.org, Mike Turquette , Stephen Boyd , linux-clk@vger.kernel.org, Mark Brown , Liam Girdwood , Kamil Debski , lm-sensors@lm-sensors.org, Jean Delvare , Guenter Roeck , Dmitry Torokhov , linux-input@vger.kernel.org, Bryan Wu , Richard Purdie , Jacek Anaszewski , linux-leds@vger.kernel.org, Maxime Ripard , Chen-Yu Tsai , linux-sunxi@googlegroups.com, Joachim Eastwood , Thomas Petazzoni , Heiko Stuebner , linux-rockchip@lists.infradead.org, Jingoo Han , Lee Jones , linux-fbdev@vger.kernel.org, Jean-Christophe Plagniol-Villard , Tomi Valkeinen , Robert Jarzmik , Alexandre Belloni , Kukjin Kim , Krzysztof Kozlowski , linux-samsung-soc@vger.kernel.org, intel-gfx@lists.freedesktop.org, Daniel Vetter , Jani Nikula , Jonathan Corbet , linux-doc@vger.kernel.org, David Airlie , Daniel Vetter , dri-devel@lists.freedesktop.org, linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org, Hartley Sweeten , Ryan Mallon , Alexander Shiyan , Milo Kim Subject: Re: [PATCH v5 15/46] pwm: introduce the pwm_state concept Message-ID: <20160412161301.30b0dc4f@bbrezillon> In-Reply-To: <20160412140546.GS18882@ulmo.ba.sec> References: <1459368249-13241-1-git-send-email-boris.brezillon@free-electrons.com> <1459368249-13241-16-git-send-email-boris.brezillon@free-electrons.com> <20160412114904.GM18882@ulmo.ba.sec> <20160412141718.5fe4cf24@bbrezillon> <20160412122141.GP18882@ulmo.ba.sec> <20160412144508.2ee181fe@bbrezillon> <20160412131118.GQ18882@ulmo.ba.sec> <20160412152644.45ff517a@bbrezillon> <20160412140546.GS18882@ulmo.ba.sec> X-Mailer: Claws Mail 3.12.0 (GTK+ 2.24.28; x86_64-pc-linux-gnu) MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 6357 Lines: 127 On Tue, 12 Apr 2016 16:05:46 +0200 Thierry Reding wrote: > On Tue, Apr 12, 2016 at 03:26:44PM +0200, Boris Brezillon wrote: > > On Tue, 12 Apr 2016 15:11:18 +0200 > > Thierry Reding wrote: > > > > > On Tue, Apr 12, 2016 at 02:45:08PM +0200, Boris Brezillon wrote: > > > > On Tue, 12 Apr 2016 14:21:41 +0200 > > > > Thierry Reding wrote: > > > > > > > > > On Tue, Apr 12, 2016 at 02:17:18PM +0200, Boris Brezillon wrote: > > > > > > On Tue, 12 Apr 2016 13:49:04 +0200 > > > > > > Thierry Reding wrote: > > > > > > > > > > > > > On Wed, Mar 30, 2016 at 10:03:38PM +0200, Boris Brezillon wrote: > > > > > > > > The PWM state, represented by its period, duty_cycle and polarity, > > > > > > > > is currently directly stored in the PWM device. > > > > > > > > Declare a pwm_state structure embedding those field so that we can later > > > > > > > > use this struct to atomically update all the PWM parameters at once. > > > > > > > > > > > > > > > > All pwm_get_xxx() helpers are now implemented as wrappers around > > > > > > > > pwm_get_state(). > > > > > > > > > > > > > > > > Signed-off-by: Boris Brezillon > > > > > > > > --- > > > > > > > > drivers/pwm/core.c | 8 ++++---- > > > > > > > > include/linux/pwm.h | 54 +++++++++++++++++++++++++++++++++++++++++------------ > > > > > > > > 2 files changed, 46 insertions(+), 16 deletions(-) > > > > > > > > > > > > > > > > diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c > > > > > > > > index 6433059..f3f91e7 100644 > > > > > > > > --- a/drivers/pwm/core.c > > > > > > > > +++ b/drivers/pwm/core.c > > > > > > > > @@ -268,7 +268,7 @@ int pwmchip_add_with_polarity(struct pwm_chip *chip, > > > > > > > > pwm->chip = chip; > > > > > > > > pwm->pwm = chip->base + i; > > > > > > > > pwm->hwpwm = i; > > > > > > > > - pwm->polarity = polarity; > > > > > > > > + pwm->state.polarity = polarity; > > > > > > > > > > > > > > Would this not more correctly be assigned to pwm->args.polarity? After > > > > > > > all this is setting up the "initial" state, much like DT or the lookup > > > > > > > tables would for duty cycle and period. > > > > > > > > > > > > Yes, I wasn't sure about the pwm_add_with_polarity() meaning. To me, > > > > > > all the reference info should be extracted from DT, PWM lookup table or > > > > > > driver specific ->request() implementation, but I can definitely > > > > > > initialize the args.polarity here too. > > > > > > > > > > > > Should I keep the pwm->state.polarity assignment (to set the initial > > > > > > polarity when the driver does not support hardware readout)? > > > > > > > > > > Wouldn't this work automatically as part of the pwm_apply_args() helper > > > > > if we extended it with this setting? > > > > > > > > Well, as you explained in you answer to patch 5, pwm_apply_args() > > > > should be called on a per-request basis (each time a PWM device is > > > > requested), while the initial polarity setting should only be applied > > > > when registering the PWM chip (and its devices). After that, the > > > > framework takes care of keeping the PWM state in sync with the hardware > > > > state. > > > > > > > > Let's take a real (though a bit unusual) example. Say you have a single > > > > PWM device referenced by two different users. Only one user can be > > > > enabled at a time, but each of them has its own reference config > > > > (different polarity, different period). > > > > > > > > User1 calls pwm_get() and applies its own polarity and period. Then > > > > user1 is unregistered and release the PWM device, leaving the polarity > > > > and period untouched. > > > > > > > > User2 is registered and request the same PWM device, but user2 is > > > > smarter and tries to extract the current PWM state before adapting the > > > > config according to pwm_args. If you just reset pwm->state.polarity > > > > each time pwm_apply_args() is called (and you suggested to call it as > > > > part of the request procedure), then this means the PWM state is no > > > > longer in sync with the hardware state. > > > > > > In that case neither will be the period or duty cycle. Essentially this > > > gets us back to square one where we need to decide how to handle current > > > state vs. initial arguments. > > > > That's not true. Now we clearly differentiate the reference config > > (content of pwm_args which is only a subset of what you'll find in > > pwm_state) and the PWM state (represented by pwm_state). > > > > We should be safe as long as we keep those 2 elements as 2 orthogonal > > concepts: > > - pwm_args is supposed to give some hint to the PWM user to help him > > configure it's PWM appropriately > > - pwm_state is here to reflect the real PWM state, and apply new > > configs > > > > > > > > But I don't think this is really going to be an issue because this is > > > all moot until we've moved over to the atomic API, at which point this > > > is all going to go away anyway. > > > > As stated in my answer to patch 5, I think I misunderstood your > > suggestion. pwm_apply_args() is supposed to adjust the PWM config to > > match the period and polarity specified in pwm_args, right? > > > > If that's the case, my question is, should we really call this function > > each time a new user requests a PWM instead of letting those users call > > the function on-demand (not all users want to adapt the current PWM > > config to the pwm_args, some may just want to apply a completely new > > config). > > I think we're still talking past each other. I didn't mean for this to > be a proper part of the API. Like you said the struct pwm_args doesn't > contain enough data to construct a complete state and apply it. > > What I was suggesting is to factor out the individual calls to the > various pwm_set_*() functions into a single call. So we wouldn't be > changing semantics, just refactoring to make it easier to get rid of > again in one of the subsequent patches. > > That is, pwm_apply_args() would go away again within this very series, > at the same point that you're currently removing the pwm_set_*() calls. Okay, eventually got it :). -- Boris Brezillon, Free Electrons Embedded Linux and Kernel engineering http://free-electrons.com