Re: [PATCH] mmc: sdhci-pci-gli: Add runtime PM for GL9763E

[Ben Chuang]

Hi,

On Wed, Mar 2, 2022 at 6:04 PM Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
>
> On Wed, 2 Mar 2022 at 03:10, Ben Chuang <benchuanggli@xxxxxxxxx> wrote:
> >
> > Hi,
> >
> > On Tue, Mar 1, 2022 at 6:05 PM Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
> > >
> > > On 28/02/2022 19:03, Ulf Hansson wrote:
> > > > On Fri, 25 Feb 2022 at 13:56, Ben Chuang <benchuanggli@xxxxxxxxx> wrote:
> > > >>
> > > >> From: Ben Chuang <ben.chuang@xxxxxxxxxxxxxxxxxxx>
> > > >>
> > > >> Add runtime PM for GL9763E and disable PLL in runtime suspend. So power
> > > >> gated of upstream port can be enabled.
> > > >>
> > > >> Signed-off-by: Ben Chuang <ben.chuang@xxxxxxxxxxxxxxxxxxx>
> > > >> Tested-by: Kevin Chang <kevin.chang@xxxxxxxxxxxxxxxxxx>
> > > >> ---
> > > >>  drivers/mmc/host/sdhci-pci-gli.c | 54 ++++++++++++++++++++++++++++++++
> > > >>  1 file changed, 54 insertions(+)
> > > >>
> > > >> diff --git a/drivers/mmc/host/sdhci-pci-gli.c b/drivers/mmc/host/sdhci-pci-gli.c
> > > >> index 97035d77c18c..cf99b6af792d 100644
> > > >> --- a/drivers/mmc/host/sdhci-pci-gli.c
> > > >> +++ b/drivers/mmc/host/sdhci-pci-gli.c
> > > >> @@ -873,6 +873,55 @@ static void gli_set_gl9763e(struct sdhci_pci_slot *slot)
> > > >>         pci_write_config_dword(pdev, PCIE_GLI_9763E_VHS, value);
> > > >>  }
> > > >>
> > > >> +#ifdef CONFIG_PM
> > > >> +static int gl9763e_runtime_suspend(struct sdhci_pci_chip *chip)
> > > >> +{
> > > >> +       struct sdhci_pci_slot *slot = chip->slots[0];
> > > >> +       struct sdhci_host *host = slot->host;
> > > >> +       u16 clock;
> > > >> +
> > > >> +       clock = sdhci_readw(host, SDHCI_CLOCK_CONTROL);
> > > >> +       clock &= ~(SDHCI_CLOCK_PLL_EN | SDHCI_CLOCK_CARD_EN);
> > > >> +       sdhci_writew(host, clock, SDHCI_CLOCK_CONTROL);
> > > >> +
> > > >> +       return 0;
> > > >> +}
> > > >> +
> > > >> +static int gl9763e_runtime_resume(struct sdhci_pci_chip *chip)
> > > >> +{
> > > >> +       struct sdhci_pci_slot *slot = chip->slots[0];
> > > >> +       struct sdhci_host *host = slot->host;
> > > >> +       ktime_t timeout;
> > > >> +       u16 clock;
> > > >> +
> > > >> +       clock = sdhci_readw(host, SDHCI_CLOCK_CONTROL);
> > > >> +
> > > >> +       clock |= SDHCI_CLOCK_PLL_EN;
> > > >> +       clock &= ~SDHCI_CLOCK_INT_STABLE;
> > > >> +       sdhci_writew(host, clock, SDHCI_CLOCK_CONTROL);
> > > >> +
> > > >> +       timeout = ktime_add_ms(ktime_get(), 150);
> > > >> +       while (1) {
> > > >> +               bool timedout = ktime_after(ktime_get(), timeout);
> > > >> +
> > > >> +               clock = sdhci_readw(host, SDHCI_CLOCK_CONTROL);
> > > >> +               if (clock & SDHCI_CLOCK_INT_STABLE)
> > > >> +                       break;
> > > >> +               if (timedout) {
> > > >> +                       pr_err("%s: PLL clock never stabilised.\n",
> > > >> +                              mmc_hostname(host->mmc));
> > > >> +                       sdhci_dumpregs(host);
> > > >> +                       break;
> > > >> +               }
> > > >> +               udelay(10);
> > > >> +       }
> > >
> > > Could use something like read_poll_timeout() here e.g.
> > >
> > >         if (read_poll_timeout(sdhci_readw, clk, (clk & SDHCI_CLOCK_INT_STABLE),
> > >                               1000, 150000, false, host, SDHCI_CLOCK_CONTROL)) {
> > >                 pr_err("%s: PLL clock never stabilised.\n",
> > >                        mmc_hostname(host->mmc));
> > >                 sdhci_dumpregs(host);
> > >         }
> > >
> > >
> >
> > Thanks for the tip. I will prepare the next patch.
> >
> > > >> +       clock |= SDHCI_CLOCK_CARD_EN;
> > > >> +       sdhci_writew(host, clock, SDHCI_CLOCK_CONTROL);
> > > >> +
> > > >> +       return 0;
> > > >> +}
> > > >
> > > > Both functions above look very similar to what sdhci_set_clock() does.
> > > > Can you use that, rather than open coding the above?
> > > >
> >
> > The codes turn on/off the clock but it doesn't change the clock.
> > Using sdhci_set_clock()  needs to store the clock value because it
> > clears the clock.
>
> Well, in that case at least you can call sdhci_enable_clk() from
> gl9763e_runtime_resume(), rather than open coding it (the code looks
> like a direct copy of that code). All you need is to give the
> sdhci_enable_clk() the correct "clk" as in-parameter.
>

I want to explain a little more,
gl9763e is used as a boot disk controller.
In gl9763e_runtime_suspend(), gl9763e still needs the internal clock
to keep states.
So gl9763e_runtime_suspend() only clears SDHCI_CLOCK_PLL_EN and
SDHCI_CLOCK_CARD_EN.
The SDHCI_CLOCK_INT_EN is still one (1) In runtime suspend,

If using sdhci_enable_clk()  in gl9763e_runtime_resume(), sdhci_enable_clk()
only fills one (1) to this bit (SDHCI_CLOCK_INT_EN) again, it cannot
cause internal stable bit
(SDHCI_CLOCK_INT_STABLE) to one for gl9763e in the first while loop.

> >
> > > > Other than that, I would appreciate it if Adrian could have a look at
> > > > this too. For example, I wonder if perhaps
> > > > sdhci_runtime_suspend|resume_host() should be called in these paths
> > > > too.
> > >
> > > Assuming the host controller does not lose state information, it should be fine.
> > >
> >
> > Yes, the host always has aux. power and keep state.
> >
> > Thank you both for your comments.
>
> Okay, thanks for confirming!
>
> [...]
>
> Kind regards
> Uffe

Best regards,
Ben