Date:   Mon, 29 Aug 2016 17:45:39 +0530
From:   Pramod Gurav <pramod.gurav@...aro.org>
To:     Stephen Boyd <sboyd@...eaurora.org>
Cc:     "open list:ARM/QUALCOMM SUPPORT" <linux-arm-msm@...r.kernel.org>,
        "open list:SERIAL DRIVERS" <linux-serial@...r.kernel.org>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Jiri Slaby <jslaby@...e.com>,
        Andy Gross <andy.gross@...aro.org>,
        David Brown <david.brown@...aro.org>,
        Ulf Hansson <ulf.hansson@...aro.org>,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] tty: serial: msm: Add runtime PM and system sleep support

Hi Stephen,

Thanks for having a look.

On 26 August 2016 at 04:20, Stephen Boyd <sboyd@...eaurora.org> wrote:
> On 06/17, Pramod Gurav wrote:
>> @@ -1220,12 +1293,26 @@ static void msm_power(struct uart_port *port, unsigned int state,
>>
>>       switch (state) {
>>       case 0:
>> -             clk_prepare_enable(msm_port->clk);
>> -             clk_prepare_enable(msm_port->pclk);
>> +             /*
>> +              * UART clk must be kept enabled to
>> +              * avoid losing received character
>> +              */
>
> Don't we have a wakeup irq? Two wire interfaces probably don't
> work though (like the debug uart).
I am not aware of wakeup irq for UART.

>
>> +             if (clk_prepare_enable(msm_port->clk))
>> +                     return;
>> +             if (clk_prepare(msm_port->pclk)) {
>> +                     clk_disable_unprepare(msm_port->clk);
>> +                     return;
>> +             }
>> +             if (pm_runtime_get_sync(port->dev) < 0) {
>> +                     clk_unprepare(msm_port->pclk);
>> +                     clk_disable_unprepare(msm_port->clk);
>
> I guess that's why we gate the interface clk and not the core clk
> during runtime PM? core clk goes off and then device is basically
> suspended unless it can wakeup with an irq.
Yes. With core clock disabled we cant get RX working as we dont have
any wakeup mechanism after which we could carry out RX.

>
>> +                     return;
>> +             }
>>               break;
>>       case 3:
>> +             pm_runtime_put(port->dev);
>> +             clk_unprepare(msm_port->pclk);
>>               clk_disable_unprepare(msm_port->clk);
>> -             clk_disable_unprepare(msm_port->pclk);
>>               break;
>>       default:
>>               pr_err("msm_serial: Unknown PM state %d\n", state);
>> @@ -1465,7 +1552,11 @@ static void msm_console_write(struct console *co, const char *s,
>>       port = msm_get_port_from_line(co->index);
>>       msm_port = UART_TO_MSM(port);
>>
>> +     if (pm_runtime_get_sync(port->dev) < 0)
>> +             return;
>>       __msm_console_write(port, s, count, msm_port->is_uartdm);
>> +     pm_runtime_mark_last_busy(port->dev);
>> +     pm_runtime_put_autosuspend(port->dev);
>
> Hmm ok, perhaps we should differentiate runtime PM for devices
> that use the console and ones that are being used for other
> things? I would guess that console can only turn off the
> interface clk while idle, but the non-console devices could turn
> off everything at runtime and rely on some out of band signaling
> to wakeup when something comes over the rx wire?
I will see if there is any way to wakeup the UART like you are saying.

>
>>  }
>>
>>  static int __init msm_console_setup(struct console *co, char *options)
>> @@ -1484,7 +1575,7 @@ static int __init msm_console_setup(struct console *co, char *options)
>>       if (unlikely(!port->membase))
>>               return -ENXIO;
>>
>> -     msm_init_clock(port);
>> +     msm_serial_set_mnd_regs(port);
>>
>>       if (options)
>>               uart_parse_options(options, &baud, &parity, &bits, &flow);
>
> Doesn't uart_set_options() go and touch hardware registers during
> termios settings? The clks are no longer enabled here though so I
> hope this isn't relying on the fact that the clks are enabled in
> the bootloader?
The clocks are enabled in serial_core with call to uart_change_pm()
just before console_setup and hence this should be okay.

>
>> @@ -1627,6 +1718,12 @@ static int msm_serial_probe(struct platform_device *pdev)
>>
>>       platform_set_drvdata(pdev, port);
>>
>> +     pm_runtime_use_autosuspend(&pdev->dev);
>> +     pm_runtime_set_autosuspend_delay(&pdev->dev, 500);
>> +     pm_runtime_irq_safe(&pdev->dev);
>
> So this means irqs are always disabled while runtime PM
> callbacks are run....
Because we are accessing UART registers in IRQ handler.
>
>> +     pm_runtime_enable(&pdev->dev);
>> +     pm_runtime_set_suspended(&pdev->dev);
>> +
>>       return uart_add_one_port(&msm_uart_driver, port);
>>  }
>>
>> @@ -1645,12 +1743,67 @@ static const struct of_device_id msm_match_table[] = {
>>       {}
>>  };
>>
>> +#ifdef CONFIG_PM
>> +static int msm_serial_runtime_suspend(struct device *dev)
>> +{
>> +     struct uart_port *port = dev_get_drvdata(dev);
>> +     struct msm_port *msm_port = UART_TO_MSM(port);
>> +
>> +     if (msm_port->is_uartdm)
>> +             clk_disable(msm_port->pclk);
>
> ... so we can't unprepare clks here. That's unfortunate because
> clks that are ancestors of these clks will be kept prepared and
> that could lead to things like PLLs being kept enabled, etc.
Yes. clk_prepare/unprepare may sleep and we want to avoid that in runtime PM.
This is all we can do in runtime PM. suspend will achieve full
resource release though.

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

Powered by blists - more mailing lists