Re: [PATCH v2] fpga: zynq-fpga: Enable pm_runtime (suspend, resume)

From: Moritz Fischer
Date: Mon Nov 23 2015 - 17:02:40 EST


Hi Mike,

thanks for your feedback. I put what I think you suggested inline below.

On Thu, Nov 19, 2015 at 11:25 PM, Mike Looijmans
<mike.looijmans@xxxxxxxx> wrote:
> On 19-11-15 23:07, Moritz Fischer wrote:

>> @@ -457,19 +457,26 @@ static int zynq_fpga_probe(struct platform_device
>> *pdev)
>> return err;
>> }
>>
>> + pm_runtime_get_noresume(&pdev->dev);
>> + pm_runtime_set_active(&pdev->dev);
>> + pm_runtime_enable(&pdev->dev);
>> +
>> /* unlock the device */
>> zynq_fpga_write(priv, UNLOCK_OFFSET, UNLOCK_MASK);
>>
>> - clk_disable(priv->clk);
>>
>> err = fpga_mgr_register(dev, "Xilinx Zynq FPGA Manager",
>> &zynq_fpga_ops, priv);
>> if (err) {
>> dev_err(dev, "unable to register FPGA manager");
>> - clk_unprepare(priv->clk);
>> + clk_disable_unprepare(priv->clk);
- clk_disable_unprepare(priv->clk);
>> + pm_runtime_put_noidle(&pdev->dev);
>> + pm_runtime_disable(&pdev->dev);
>> return err;
>> }
>>
>> + pm_runtime_put(&pdev->dev);
>> +
>> return 0;
>> }
>>
>> @@ -483,11 +490,45 @@ static int zynq_fpga_remove(struct platform_device
>> *pdev)
>>
>> fpga_mgr_unregister(&pdev->dev);
>>
>> - clk_unprepare(priv->clk);
>> + pm_runtime_get_sync(&pdev->dev);
>> + clk_disable_unprepare(priv->clk);
- clk_disable_unprepare(priv->clk);

>> + pm_runtime_put_noidle(&pdev->dev);
>> + pm_runtime_disable(&pdev->dev);
>>
>> return 0;
>> }
>>
>> +static int __maybe_unused zynq_fpga_runtime_suspend(struct device *dev)
>> +{
>> + struct zynq_fpga_priv *priv;
>> + struct fpga_manager *mgr;
>> +
>> + mgr = dev_get_drvdata(dev);
>> + priv = mgr->priv;
>> +
>> + clk_disable(priv->clk);

- clk_disable(priv->clk)
+ clk_disable_unprepare(priv->clk)
>
>
> From what I understand, this call is done in a sleepable context, so you can
> use clk_disable_unprepare here (and its counterpart in resume). And remove
> the prepare at probe time and unprepare at removal.
>
> Not all clocks can implement atomic enable/disable, for example I2C and SPI
> controlled clocks only implement the prepare/unprepare routines.
>
> I guess the "clk" here will always be a SOC provided one, so it won't make
> any difference for the Zynq, but someone is likely to some day copy/paste
> this driver and use it for some externally connected FPGA instead.

To clarify. Is the above / below what you suggested?

>> +static int __maybe_unused zynq_fpga_runtime_resume(struct device *dev)
>> +{
>> + struct zynq_fpga_priv *priv;
>> + struct fpga_manager *mgr;
>> +
>> + mgr = dev_get_drvdata(dev);
>> + priv = mgr->priv;
>> +
>> + clk_enable(priv->clk);

- clk_enable(priv->clk)
+ clk_prepare_enable(priv->clk)
>> +
>> + return 0;
>> +}
>> +

Cheers,

Moritz
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/