RE: [PATCH v1] spi: dw: Enable Autosuspend delay for SPI DesignWare

[Vaidya, Mahesh R]

> -----Original Message-----
> From: Serge Semin <fancer.lancer@xxxxxxxxx>
> Sent: Sunday, September 5, 2021 8:49 PM
> To: Vaidya, Mahesh R <mahesh.r.vaidya@xxxxxxxxx>
> Cc: Serge Semin <Sergey.Semin@xxxxxxxxxxxxxxxxxxxx>; broonie@xxxxxxxxxx;
> linux-spi@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> mgross@xxxxxxxxxxxxxxx; Pan, Kris <kris.pan@xxxxxxxxx>; Zhou, Furong
> <furong.zhou@xxxxxxxxx>; Sangannavar, Mallikarjunappa
> <mallikarjunappa.sangannavar@xxxxxxxxx>; Raja Subramanian, Lakshmi Bai
> <lakshmi.bai.raja.subramanian@xxxxxxxxx>
> Subject: Re: [PATCH v1] spi: dw: Enable Autosuspend delay for SPI
> DesignWare
> 
> On Sat, Sep 04, 2021 at 05:27:58AM +0800, mahesh.r.vaidya@xxxxxxxxx
> wrote:
> > From: Mahesh R Vaidya <mahesh.r.vaidya@xxxxxxxxx>
> >
> > Enable and set Autosuspend delay for SPI DesignWare driver.
> > The number 1000 ms for the autosuspend delay was picked a bit
> > arbitrarily, so if someone has measurements showing a better value we
> > could easily change this.
> >
> > Signed-off-by: Mahesh R Vaidya <mahesh.r.vaidya@xxxxxxxxx>
> > ---
> >  drivers/spi/spi-dw-mmio.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> >
> > diff --git a/drivers/spi/spi-dw-mmio.c b/drivers/spi/spi-dw-mmio.c
> > index 3379720cfcb8..8b588ce9c16b 100644
> > --- a/drivers/spi/spi-dw-mmio.c
> > +++ b/drivers/spi/spi-dw-mmio.c
> > @@ -33,6 +33,8 @@ struct dw_spi_mmio {
> >  	struct reset_control *rstc;
> >  };
> >
> > +#define SPI_DW_DEFAULT_AUTOSUSP_VAL		1000
> > +
> >  #define MSCC_CPU_SYSTEM_CTRL_GENERAL_CTRL	0x24
> >  #define OCELOT_IF_SI_OWNER_OFFSET		4
> >  #define JAGUAR2_IF_SI_OWNER_OFFSET		6
> > @@ -309,6 +311,10 @@ static int dw_spi_mmio_probe(struct
> platform_device *pdev)
> >  			goto out;
> >  	}
> >
> 
> > +	/* Set initial autosuspend default delay value and enable */
> > +	pm_runtime_set_autosuspend_delay(&pdev->dev,
> SPI_DW_DEFAULT_AUTOSUSP_VAL);
> > +	pm_runtime_use_autosuspend(&pdev->dev);
> > +	pm_runtime_set_active(&pdev->dev);
> 
> Seems reasonable, but doesn't it need to be reverted in the remove callback?
> Like it's done in the spi-imx.c driver for instance by calling the
> pm_runtime_dont_use_autosuspend() method.
> 

I initially referred to the below two commits which enable the autosuspend feature for spi-rockchip and spi-sun6i, hence had followed the same.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=940f3bbf3dacd6e31e482a10e64c96e69b00dded
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=ae0f18bef30d0e76dd62be46c59b24c75f175092
But, yes, I Agree with you, right way would be to handle the reverted calls in the remove part of the code. I will modify the change accordingly in the next patch.

> >  	pm_runtime_enable(&pdev->dev);
> >
> >  	ret = dw_spi_add_host(&pdev->dev, dws); @@ -319,6 +325,7 @@
> static
> > int dw_spi_mmio_probe(struct platform_device *pdev)
> >  	return 0;
> >
> >  out:
> 
> > +	pm_runtime_put_noidle(&pdev->dev);
> 
> This doesn't seem like related to the subject, thus needs to be submitted as a
> separate patch with fixes tag if it fixes some another problem. Additionally it
> isn't balanced with the pm_runtime_get*() method invocation. Am I missing
> something?
> 
> -Sergey
> 

I checked this further and pm_runtime_put_noidle() seems to be redundant in my use case. Hence, I will remove this in the next patch.

Regards,
Mahesh

> >  	pm_runtime_disable(&pdev->dev);
> >  	clk_disable_unprepare(dwsmmio->pclk);
> >  out_clk:
> > --
> > 2.17.1
> >