On Thu 28 Jun 10:14 PDT 2018, Stephen Boyd wrote: > Quoting Linus Walleij (2018-06-28 07:25:46) > > On Fri, Jun 22, 2018 at 8:29 PM Bjorn Andersson > > <bjorn.andersson@xxxxxxxxxx> wrote: > > > On Fri 22 Jun 10:58 PDT 2018, Bjorn Andersson wrote: > > > > On Mon 18 Jun 13:52 PDT 2018, Stephen Boyd wrote: > > > > > > > > > We rely on devices to use pinmuxing configurations in DT to select the > > > > > GPIO function (function 0) if they're going to use the gpio in GPIO > > > > > mode. Let's simplify things for driver authors by implementing > > > > > gpio_request_enable() for this pinctrl driver to mux out the GPIO > > > > > function when the gpio is use from gpiolib. > > > > > > > > > > Cc: Bjorn Andersson <bjorn.andersson@xxxxxxxxxx> > > > > > > > > Reviewed-by: Bjorn Andersson <bjorn.andersson@xxxxxxxxxx> > > (...) > > > While both patch 2 and 3 are convenient ways to get around the annoyance > > > of having to specify a pinmux state both patches then ends up relying on > > > some default pinconf state; which I think is bad. > > What default state are we relying on? The reset state of the pins? I'm > very confused by this statement. These last two patches are making sure > that when a GPIO is requested as a GPIO or IRQ, it's really in GPIO > mode. Yes and this is convenient, as the TLMM is both multiplexor and gpio controller this is probably what people would expect. However looking at the downstream code people don't think this way (i.e. many drivers calls gpio_request() to get some sort of exclusive access to its pins - not to request gpio to be the muxed function). But my concern is related to pinconf, not pinmux - automating pinmuxing doesn't change the fact that the systems integrator should make sure to configure appropriate pull properties on the pins. > If this code is not in place, then we'll allow drivers to request > a GPIO but pinmuxing won't be called to actually mux that pin to GPIO > mode unless they have a DT conf for function = "gpio". That seems > entirely unexpected and easy to mess up. > I do agree with this, but the opposite doesn't feel crystal clear either. > > > > Nothing stops you from setting up a default conf in > > this callback though. > > > > But admittedly this call was added for simpler hardware. > > > > > Further more in situations like i2c-qup (downstream), where the pins are > > > requested as gpios in order to "bitbang" a reset this would mean that > > > the driver has to counter the convenience; by either switching in the > > > default pinmux at the end of probe or postponing the gpio_request() to > > > the invocation of reset and then, after issuing the gpio_release, > > > switching in the default pinmux explicitly again. > > > > That's a bigger problem. If the system is using device and GPIO > > mode orthogonally, it'd be good to leave like this. > > > > Doesn't that driver need to explicitly mux GPIO mode vs. device mode > right now? So having gpio_request() do the muxing to GPIO mode and then > explicit muxing of the pin to device mode would be what we have after > this patch, while before this patch we would have mux to GPIO, request > GPIO (nop), mux to device. We saved an explicit pinmux call? > It's currently possible to call gpio_request() in the i2c driver's probe function and then mux when needed. With this patch you would have to either follow the gpio_request with a mux-to-default or defer the gpio_request until the point where they today would have an explicit mux-to-gpio. Regards, Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
