Re: [PATCH 3/6] dmaengine: mxs: add the power management functions

[Sascha Hauer]

On Thu, Jan 16, 2020 at 10:36:33AM -0600, Han Xu wrote:
> On Wed, Jan 15, 2020 at 2:03 AM Sascha Hauer <s.hauer@xxxxxxxxxxxxxx> wrote:
> >
> > On Wed, Jan 15, 2020 at 05:44:00AM +0800, Han Xu wrote:
> > > add the power management functions and leverage the runtime pm for
> > > system suspend/resume
> > >
> > > Signed-off-by: Han Xu <han.xu@xxxxxxx>
> > > ---
> > >  drivers/dma/mxs-dma.c | 97 +++++++++++++++++++++++++++++++++++++++----
> > >  1 file changed, 90 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/drivers/dma/mxs-dma.c b/drivers/dma/mxs-dma.c
> > > index b458f06f9067..251492c5ea58 100644
> > > --- a/drivers/dma/mxs-dma.c
> > > +++ b/drivers/dma/mxs-dma.c
> > > @@ -25,6 +25,7 @@
> > >  #include <linux/of_dma.h>
> > >  #include <linux/list.h>
> > >  #include <linux/dma/mxs-dma.h>
> > > +#include <linux/pm_runtime.h>
> > >
> > >  #include <asm/irq.h>
> > >
> > > @@ -39,6 +40,8 @@
> > >  #define dma_is_apbh(mxs_dma) ((mxs_dma)->type == MXS_DMA_APBH)
> > >  #define apbh_is_old(mxs_dma) ((mxs_dma)->dev_id == IMX23_DMA)
> > >
> > > +#define MXS_DMA_RPM_TIMEOUT 50 /* ms */
> > > +
> > >  #define HW_APBHX_CTRL0                               0x000
> > >  #define BM_APBH_CTRL0_APB_BURST8_EN          (1 << 29)
> > >  #define BM_APBH_CTRL0_APB_BURST_EN           (1 << 28)
> > > @@ -416,6 +419,7 @@ static int mxs_dma_alloc_chan_resources(struct dma_chan *chan)
> > >  {
> > >       struct mxs_dma_chan *mxs_chan = to_mxs_dma_chan(chan);
> > >       struct mxs_dma_engine *mxs_dma = mxs_chan->mxs_dma;
> > > +     struct device *dev = &mxs_dma->pdev->dev;
> > >       int ret;
> > >
> > >       mxs_chan->ccw = dma_alloc_coherent(mxs_dma->dma_device.dev,
> > > @@ -431,9 +435,11 @@ static int mxs_dma_alloc_chan_resources(struct dma_chan *chan)
> > >       if (ret)
> > >               goto err_irq;
> > >
> > > -     ret = clk_prepare_enable(mxs_dma->clk);
> > > -     if (ret)
> > > +     ret = pm_runtime_get_sync(dev);
> > > +     if (ret < 0) {
> > > +             dev_err(dev, "Failed to enable clock\n");
> > >               goto err_clk;
> >
> > From looking at other DMA drivers I know we are in good company here,
> > but I think this is wrong. Doing pm_runtime_get_sync() in
> > alloc_chan_resources() and going to autosuspend in free_chan_resources()
> > effectively disables runtime_pm as clients normally acquire their
> > channels during driver probe and release them only in driver remove.
> 
> Thanks for the comments.
> That's why I moved acquire_dma_resource from the probe to
> runtime_resume in the gpmi driver, this change won't disable the
> runtime_pm function and the incremental counter always balanced.

Yes, that's what I've written a few lines further down:

> 
> >
> > In the next patch you release the DMA channels in the GPMI nand drivers
> > runtime_suspend hook just to somehow trigger the runtime_suspend of the
> > DMA driver.

And I consider doing this a crude hack. Here is what I suggested doing
instead:

> >
> > What you should do instead is to make sure the hook runtime_pm to the
> > DMA drivers activity phases, like for example the pl330 driver does.
> > Then you wouldn't have to care about manually putting the DMA driver into
> > suspend from the GPMI NAND driver.

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |