by Stephen Boyd

[permalink] [raw]

Subject: Re: [PATCH v4 2/2] iio: Aspeed ADC

On 04/06, Rick Altherr wrote:
> On Wed, Apr 5, 2017 at 6:10 PM, Stephen Boyd <[email protected]> wrote:
> > On 04/05, Rick Altherr wrote:
> >> On Wed, Apr 5, 2017 at 1:27 PM, Stephen Boyd <[email protected]> wrote:
> >> > On 03/23, Rick Altherr wrote:
> >> >> +
> >> >> +static int aspeed_adc_probe(struct platform_device *pdev)
> >> >> +{
> >> >> + struct iio_dev *indio_dev;
> >> >> + struct aspeed_adc_data *data;
> >> >> + const struct aspeed_adc_model_data *model_data;
> >> >> + struct resource *res;
> >> >> + const char *clk_parent_name;
> >> >> + int ret;
> >> >> + u32 adc_engine_control_reg_val;
> >> >> +
> >> >> + indio_dev = devm_iio_device_alloc(&pdev->dev, sizeof(*data));
> >> >> + if (!indio_dev)
> >> >> + return -ENOMEM;
> >> >> +
> >> >> + data = iio_priv(indio_dev);
> >> >> + data->dev = &pdev->dev;
> >> >> +
> >> >> + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> >> >> + data->base = devm_ioremap_resource(&pdev->dev, res);
> >> >> + if (IS_ERR(data->base))
> >> >> + return PTR_ERR(data->base);
> >> >> +
> >> >> + /* Register ADC clock prescaler with source specified by device tree. */
> >> >> + spin_lock_init(&data->clk_lock);
> >> >> + clk_parent_name = of_clk_get_parent_name(pdev->dev.of_node, 0);
> >> >
> >> > What if the parent clk is not registered yet? Or if we're not
> >> > always using DT in this driver? Put another way, this code is
> >> > fragile. But I guess it probably works well enough for now so no
> >> > big deal, just pointing out my fear.
> >>
> >> I'm not terribly worried about not using DT for this driver as it is
> >> for an ARM SoC's built-in ADC which is only supported via DT. I take
> >> your point though. What's the right way to do this? Use clk_get() to
> >> request by name and clock-names in the DT?
> >
> > Yes that will work. When we add probe defer to clk_get() things
> > will work better. Presumably the clocks property already exists
> > for this device so that of_clk_get_parent_name() works so it's
> > not a binding change?
>
> The bindings include clocks but not clock-names. In this case,
> clk_register_divider() only has variants that take a parent clock
> name. I can't see what I'd do with the result of clk_get(). If the
> bindings provide clock-names, I can provide a fixed string for the
> parent name instead of relying on of_clk_get_parent_name(). The
> removes an explicit driver dependency on DT. I'm unclear if it
> resolves your concerns about registration ordering.

We have __clk_get_name() as a helper in clk-provider.h to assist
drivers in finding parent names of clks they get through
clk_get(). Eventually we want to get rid of that function, but
using it at least helps us find it later.

Aside: This is a problem we've had for a while. You're an
innocent bystander here. The usage of strings for clk tree
description is not working well and we really need to move away
from it. So far we have no solution so all I can say is sorry you
tripped up on this.

>
> >
> >>
> >> >
> >> >> +
> >> >> + data->clk_prescaler = clk_hw_register_divider(
> >> >> + &pdev->dev, "prescaler", clk_parent_name, 0,
> >> >> + data->base + ASPEED_REG_CLOCK_CONTROL,
> >> >> + 17, 15, 0, &data->clk_lock);
> >> >> + if (IS_ERR(data->clk_prescaler))
> >> >> + return PTR_ERR(data->clk_prescaler);
> >> >> +
> >> >> + /*
> >> >> + * Register ADC clock scaler downstream from the prescaler. Allow rate
> >> >> + * setting to adjust the prescaler as well.
> >> >> + */
> >> >> + data->clk_scaler = clk_hw_register_divider(
> >> >> + &pdev->dev, "scaler", "prescaler",
> >> >> + CLK_SET_RATE_PARENT,
> >> >> + data->base + ASPEED_REG_CLOCK_CONTROL,
> >> >> + 0, 10, 0, &data->clk_lock);
> >> >> + if (IS_ERR(data->clk_scaler)) {
> >> >> + ret = PTR_ERR(data->clk_scaler);
> >> >> + goto scaler_error;
> >> >> + }
> >> >> +
> >> >> + /* Start all channels in normal mode. */
> >> >> + clk_prepare_enable(data->clk_scaler->clk);
> >> >
> >> > Eventually we'd like to get rid of hw->clk pointer so that users
> >> > have to call some sort of clk_get() API and then we get warm
> >> > fuzzies from knowing who is consuming a clk structure. Can you
> >> > change this to register a clk provider and call clk_get()? I
> >> > think a device that references itself should be OK in DT still,
> >> > and would properly reflect what's going on.
> >>
> >> Do you mean call of_clk_add_hw_provider with of_clk_hw_simple_get or
> >> something else? I'm still wrapping my head around the distinction
> >> between clk, clk_hw, and a clk provider.
> >>
> >
> > Yes. Unless you don't want to expose these details in DT because
> > it's all internal to the device?
> >
>
> There's no reason to expose the scaler or prescaler in DT. These are
> clocks that internal to the ADC and have no way for other devices to
> use them. I only used the clock framework to avoid reinventing the
> wheel on calculating divider values for a desired clock rate.
>
> > In that case we need to go merge that patch on the clk list to
> > have clk_hw_create_clk() or something like that, so we can turn a
> > clk_hw structure into a clk pointer without directly referencing
> > the clk member of clk_hw.
>
> This makes the most sense to me.
>

Ok. Let me see what I can do to merge that patch now. I should be
able to put it on its own branch that could be pulled into the
tree for this driver. Or a merge window can pass and things fixed
then.

--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project