Message-ID: <aLmp5ILqhiWYJrw5@tom-desktop>
Date: Thu, 4 Sep 2025 17:01:56 +0200
From: Tommaso Merciai <tommaso.merciai.xr@...renesas.com>
To: Geert Uytterhoeven <geert@...ux-m68k.org>
Cc: tomm.merciai@...il.com, linux-renesas-soc@...r.kernel.org,
	biju.das.jz@...renesas.com,
	Michael Turquette <mturquette@...libre.com>,
	Stephen Boyd <sboyd@...nel.org>, Vinod Koul <vkoul@...nel.org>,
	Philipp Zabel <p.zabel@...gutronix.de>,
	Wolfram Sang <wsa+renesas@...g-engineering.com>,
	Lad Prabhakar <prabhakar.mahadev-lad.rj@...renesas.com>,
	Fabrizio Castro <fabrizio.castro.jz@...esas.com>,
	Uwe Kleine-König <u.kleine-koenig@...libre.com>,
	linux-clk@...r.kernel.org, linux-kernel@...r.kernel.org,
	dmaengine@...r.kernel.org
Subject: Re: [PATCH v2 5/5] dmaengine: sh: rz-dmac: Add runtime PM support

Hi Geert,
Thanks for your comments!

(And sorry for the delay :) )

On Wed, Sep 03, 2025 at 02:17:53PM +0200, Geert Uytterhoeven wrote:
> Hi Tommaso,
> 
> Thanks for your patch!
> 
> I don't understand why you included this patch in a series with clock
> patches. AFAIUC, there is no dependency.  Am I missing something?

I was working on pm support for RZ DMAC when I wrote previous clk
patches, sorry. For that I've included also this one. :'(

> 
> On Wed, 3 Sept 2025 at 10:28, Tommaso Merciai
> <tommaso.merciai.xr@...renesas.com> wrote:
> > Enable runtime power management in the rz-dmac driver by adding suspend and
> > resume callbacks. This ensures the driver can correctly assert and deassert
> 
> This is not really what this patch does: the Runtime PM-related changes
> just hide^Wmove reset handling into the runtime callbacks.

Agreed.

> 
> > the reset control and manage power state transitions during suspend and
> > resume. Adding runtime PM support allows the DMA controller to reduce power
> 
> (I assume) This patch does fix resuming from _system_ suspend.
> 
> > consumption when idle and maintain correct operation across system sleep
> > states, addressing the previous lack of dynamic power management in the
> > driver.
> 
> The driver still does not do dynamic power management: you still call
> pm_runtime_resume_and_get() from the driver's probe() .callback, and
> call pm_runtime_put() only from the .remove() callback, so the device
> is powered all the time.
> To implement dynamic power management, you have to change that,
> and call pm_runtime_resume_and_get() and pm_runtime_put() from the
> .device_alloc_chan_resources() resp. .device_free_chan_resources()
> callbacks (see e.g. drivers/dma/sh/rcar-dmac.c).

Thanks for the hints!
So following your hints we need to:

 - call pm_runtime_get_sync() from rz_dmac_alloc_chan_resources()
 - call pm_runtime_put() from rz_dmac_free_chan_resources()

With that then we can remove pm_runtime_put() from the remove function
and add this at the end of the probe function.

> 
> > Signed-off-by: Tommaso Merciai <tommaso.merciai.xr@...renesas.com>
> 
> > --- a/drivers/dma/sh/rz-dmac.c
> > +++ b/drivers/dma/sh/rz-dmac.c
> > @@ -437,6 +437,17 @@ static int rz_dmac_xfer_desc(struct rz_dmac_chan *chan)
> >   * DMA engine operations
> >   */
> >
> > +static void rz_dmac_chan_init_all(struct rz_dmac *dmac)
> > +{
> > +       unsigned int i;
> > +
> > +       rz_dmac_writel(dmac, DCTRL_DEFAULT, CHANNEL_0_7_COMMON_BASE + DCTRL);
> > +       rz_dmac_writel(dmac, DCTRL_DEFAULT, CHANNEL_8_15_COMMON_BASE + DCTRL);
> > +
> > +       for (i = 0; i < dmac->n_channels; i++)
> > +               rz_dmac_ch_writel(&dmac->channels[i], CHCTRL_DEFAULT, CHCTRL, 1);
> > +}
> > +
> >  static int rz_dmac_alloc_chan_resources(struct dma_chan *chan)
> >  {
> >         struct rz_dmac_chan *channel = to_rz_dmac_chan(chan);
> > @@ -970,10 +981,6 @@ static int rz_dmac_probe(struct platform_device *pdev)
> >                 goto err_pm_disable;
> >         }
> >
> > -       ret = reset_control_deassert(dmac->rstc);
> > -       if (ret)
> > -               goto err_pm_runtime_put;
> > -
> >         for (i = 0; i < dmac->n_channels; i++) {
> >                 ret = rz_dmac_chan_probe(dmac, &dmac->channels[i], i);
> >                 if (ret < 0)
> > @@ -1028,8 +1035,6 @@ static int rz_dmac_probe(struct platform_device *pdev)
> >                                   channel->lmdesc.base_dma);
> >         }
> >
> > -       reset_control_assert(dmac->rstc);
> > -err_pm_runtime_put:
> >         pm_runtime_put(&pdev->dev);
> >  err_pm_disable:
> >         pm_runtime_disable(&pdev->dev);
> > @@ -1052,13 +1057,50 @@ static void rz_dmac_remove(struct platform_device *pdev)
> >                                   channel->lmdesc.base,
> >                                   channel->lmdesc.base_dma);
> >         }
> > -       reset_control_assert(dmac->rstc);
> >         pm_runtime_put(&pdev->dev);
> >         pm_runtime_disable(&pdev->dev);
> >
> >         platform_device_put(dmac->icu.pdev);
> >  }
> >
> > +static int rz_dmac_runtime_suspend(struct device *dev)
> > +{
> > +       struct rz_dmac *dmac = dev_get_drvdata(dev);
> > +
> > +       return reset_control_assert(dmac->rstc);
> 
> Do you really want to reset the device (and thus loose register state)
> each and every time the device is runtime-suspended?  For now it doesn't
> matter much, but once you implement real dynamic power management,
> it does.
> I think the reset handling should be moved to the system suspend/resume
> callbacks.

Agreed. With above changes maybe we can move all into
NOIRQ_SYSTEM_SLEEP_PM_OPS(rz_dmac_suspend, rz_dmac_resume)
With your suggested changes I'm not sure if pm_runtime_ops are really needed.


Thanks & Regards,
Tommaso

> 
> > +}
> > +
> > +static int rz_dmac_runtime_resume(struct device *dev)
> > +{
> > +       struct rz_dmac *dmac = dev_get_drvdata(dev);
> > +
> > +       return reset_control_deassert(dmac->rstc);
> 
> Shouldn't this reinitialize some registers?
> For now that indeed doesn't matter, as reset is only deasserted
> from .probe(), before any register initialization.
> 
> > +}
> > +
> > +static int rz_dmac_resume(struct device *dev)
> > +{
> > +       struct rz_dmac *dmac = dev_get_drvdata(dev);
> > +       int ret;
> > +
> > +       ret = pm_runtime_force_resume(dev);
> > +       if (ret)
> > +               return ret;
> > +
> > +       rz_dmac_chan_init_all(dmac);
> > +
> > +       return 0;
> > +}
> > +
> > +static const struct dev_pm_ops rz_dmac_pm_ops = {
> > +       /*
> > +        * TODO for system sleep/resume:
> > +        *   - Wait for the current transfer to complete and stop the device,
> > +        *   - Resume transfers, if any.
> > +        */
> > +       NOIRQ_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, rz_dmac_resume)
> > +       RUNTIME_PM_OPS(rz_dmac_runtime_suspend, rz_dmac_runtime_resume, NULL)
> > +};
> > +
> >  static const struct of_device_id of_rz_dmac_match[] = {
> >         { .compatible = "renesas,r9a09g057-dmac", },
> >         { .compatible = "renesas,rz-dmac", },
> > @@ -1068,6 +1110,7 @@ MODULE_DEVICE_TABLE(of, of_rz_dmac_match);
> >
> >  static struct platform_driver rz_dmac_driver = {
> >         .driver         = {
> > +               .pm     = pm_ptr(&rz_dmac_pm_ops),
> >                 .name   = "rz-dmac",
> >                 .of_match_table = of_rz_dmac_match,
> >         },
> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> -- 
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@...ux-m68k.org
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds

Powered by blists - more mailing lists