Re: [PATCH V2 2/2] i2c: i2c-qcom-geni: Add shutdown callback for i2c

[rojay]

On 2020-08-21 05:52, Stephen Boyd wrote:
> Quoting Roja Rani Yarubandi (2020-08-20 03:35:22)
> > If the hardware is still accessing memory after SMMU translation
> > is disabled (as part of smmu shutdown callback), then the
> > IOVAs (I/O virtual address) which it was using will go on the bus
> > as the physical addresses which will result in unknown crashes
> > like NoC/interconnect errors.
> >
> > So, implement shutdown callback to i2c driver to unmap DMA mappings
> > during system "reboot" or "shutdown".
> >
> > Signed-off-by: Roja Rani Yarubandi <rojay@xxxxxxxxxxxxxx>
> > ---
>
> I'd still put a Fixes tag because it's fixing the driver when someone
> runs shutdown.

Okay, will add fixes tag.

> > Changes in V2:
> >  - As per Stephen's comments added seperate function for stop 
> > transfer,
> >    fixed minor nitpicks.
> >
> >  drivers/i2c/busses/i2c-qcom-geni.c | 43 
> > ++++++++++++++++++++++++++++++
> >  include/linux/qcom-geni-se.h       |  5 ++++
> >  2 files changed, 48 insertions(+)
> >
> > diff --git a/drivers/i2c/busses/i2c-qcom-geni.c 
> > b/drivers/i2c/busses/i2c-qcom-geni.c
> > index 1fda5c7c2cfc..d07f2f33bb75 100644
> > --- a/drivers/i2c/busses/i2c-qcom-geni.c
> > +++ b/drivers/i2c/busses/i2c-qcom-geni.c
> > @@ -486,6 +486,28 @@ static int geni_i2c_xfer(struct i2c_adapter 
> > *adap,
> >         return ret;
> >  }
> >
> > +static void geni_i2c_stop_xfer(struct geni_i2c_dev *gi2c)
> > +{
> > +       u32 val;
> > +       struct geni_se *se = &gi2c->se;
> > +
> > +       val = readl_relaxed(gi2c->se.base + SE_DMA_DEBUG_REG0);
> > +       if (val & DMA_TX_ACTIVE) {
> > +               geni_i2c_abort_xfer(gi2c);
> > +               gi2c->cur_wr = 0;
> > +               if (gi2c->err)
> > +                       geni_i2c_tx_fsm_rst(gi2c);
> > +               geni_se_tx_dma_unprep(se, gi2c->tx_dma, 
> > gi2c->xfer_len);
> > +       }
> > +       if (val & DMA_RX_ACTIVE) {
> > +               geni_i2c_abort_xfer(gi2c);
> > +               gi2c->cur_rd = 0;
> > +               if (gi2c->err)
> > +                       geni_i2c_rx_fsm_rst(gi2c);
> > +               geni_se_rx_dma_unprep(se, gi2c->rx_dma, 
> > gi2c->xfer_len);
> > +       }
> > +}
> > +
> >  static u32 geni_i2c_func(struct i2c_adapter *adap)
> >  {
> >         return I2C_FUNC_I2C | (I2C_FUNC_SMBUS_EMUL & 
> > ~I2C_FUNC_SMBUS_QUICK);
> > @@ -617,6 +639,26 @@ static int geni_i2c_remove(struct platform_device 
> > *pdev)
> >         return 0;
> >  }
> >
> > +static void geni_i2c_shutdown(struct platform_device *pdev)
> > +{
> > +       int ret;
> > +       u32 dma;
> > +       struct geni_i2c_dev *gi2c = platform_get_drvdata(pdev);
> > +       struct geni_se *se = &gi2c->se;
> > +
> > +       ret = pm_runtime_get_sync(gi2c->se.dev);
> > +       if (ret < 0) {
> > +               dev_err(gi2c->se.dev, "Failed to resume device: %d\n", 
> > ret);
> > +               return;
> > +       }
> > +
> > +       dma = readl_relaxed(se->base + SE_GENI_DMA_MODE_EN);
> > +       if (dma)
> > +               geni_i2c_stop_xfer(gi2c);
>
> Any reason the if (dma) check isn't inside geni_i2c_stop_xfer()?
> Checking for dma and then bailing out early should work and keep the
> logic in one function. Then the pm_runtime_sync() call could go in 
> there
> too and if (!dma) goto out. This assumes that we're going to call
> geni_i2c_stop_xfer() from somewhere else, like if a transfer times out
> or something?

Okay, will do the changes.

> > +
> > +       pm_runtime_put_sync_suspend(gi2c->se.dev);
> > +}
> > +
> >  static int __maybe_unused geni_i2c_runtime_suspend(struct device 
> > *dev)
> >  {
> >         int ret;