On Thu, 4 Nov 2021 12:30:57 +0000
<Eugen.Hristev@xxxxxxxxxxxxx> wrote:
> On 10/28/21 5:39 PM, Eugen Hristev - M18282 wrote:
> > On 10/28/21 5:34 PM, Jonathan Cameron wrote:
> >> On Tue, 19 Oct 2021 10:29:28 +0200
> >> Lars-Peter Clausen <lars@xxxxxxxxxx> wrote:
> >>
> >>> The at91-sama5d2 driver calls `to_platform_device()` on a struct device
> >>> that is part of a IIO device. This is incorrect since
> >>> `to_platform_device()` must only be called on a struct device that is part
> >>> of a platform device.
> >>>
> >>> The code still works by accident because non of the struct platform_device
> >>> specific fields are accessed.
> >>>
> >>> Refactor the code a bit so that it behaves identically, but does not use
> >>> the incorrect cast. This avoids accidentally adding undefined behavior in
> >>> the future by assuming the `struct platform_device` is actually valid.
> >>>
> >>> Signed-off-by: Lars-Peter Clausen <lars@xxxxxxxxxx>
> >>
> >> This makes me nervous for the reason you give below.
> >> Looking for a response from Eugen or someone else with access to the device
> >> before I apply this one.
> >
> > Hi,
> >
> > I will take some time to test this on my setup. Thanks for the patch,
> > and sorry for the delays !
> >
> > Eugen
>
> Hello Jonathan and Lars,
>
> Sorry for the delay,
>
> I have applied the patches in my tree and tested basic DMA transfers
> (sanity checks) on my board.
> It looks to be fine. Have not found any weird or non functioning behavior.
>
> You can add my
> Tested-by: Eugen Hristev <eugen.hristev@xxxxxxxxxxxxx>
I've queued this up now, but hope Lars can address the questions below.
Thanks,
Jonathan
>
> About the query below,
> >
> >>
> >> Thanks,
> >>
> >> Jonathan
> >>
> >>> ---
> >>> The code is equivalent to before, but I'm a bit confused how this works.
> >>> We call dma_request_chan() on the IIO device's struct device. Which should
> >>> not yield any results.
> >>>
> >>> Eugen can you check if/why this works and see if a follow up patch using
> >>> the right struct device (the platform_device's) to request the DMA channel
> >>> makes sense?
>
> I do not fully understand the problem yet. Why should dma_request_chan
> on the iio dev should not yield any results? the dma channel is
> allocated and it worked so far.
>
> I tried to following: save the pdev pointer in the state struct, and
> then use dma_request_chan on the pdev->dev . It still works and I could
> not find any difference in functionality in basic sanity checks.
>
> I believe that the dma_request_chan wants a valid struct device. If it's
> iio's dev or platform device's dev, it doesn't seem to matter at the
> first glance.
>
> Could you please detail a bit how it should be used ?
>
> Thanks,
> Eugen
>
> >>> ---
> >>> drivers/iio/adc/at91-sama5d2_adc.c | 34 ++++++++++++++----------------
> >>> 1 file changed, 16 insertions(+), 18 deletions(-)
> >>>
> >>> diff --git a/drivers/iio/adc/at91-sama5d2_adc.c b/drivers/iio/adc/at91-sama5d2_adc.c
> >>> index 4c922ef634f8..3841e7b6c81d 100644
> >>> --- a/drivers/iio/adc/at91-sama5d2_adc.c
> >>> +++ b/drivers/iio/adc/at91-sama5d2_adc.c
> >>> @@ -1661,10 +1661,9 @@ static int at91_adc_write_raw(struct iio_dev *indio_dev,
> >>> }
> >>> }
> >>>
> >>> -static void at91_adc_dma_init(struct platform_device *pdev)
> >>> +static void at91_adc_dma_init(struct at91_adc_state *st)
> >>> {
> >>> - struct iio_dev *indio_dev = platform_get_drvdata(pdev);
> >>> - struct at91_adc_state *st = iio_priv(indio_dev);
> >>> + struct device *dev = &st->indio_dev->dev;
> >>> struct dma_slave_config config = {0};
> >>> /* we have 2 bytes for each channel */
> >>> unsigned int sample_size = st->soc_info.platform->nr_channels * 2;
> >>> @@ -1679,9 +1678,9 @@ static void at91_adc_dma_init(struct platform_device *pdev)
> >>> if (st->dma_st.dma_chan)
> >>> return;
> >>>
> >>> - st->dma_st.dma_chan = dma_request_chan(&pdev->dev, "rx");
> >>> + st->dma_st.dma_chan = dma_request_chan(dev, "rx");
> >>> if (IS_ERR(st->dma_st.dma_chan)) {
> >>> - dev_info(&pdev->dev, "can't get DMA channel\n");
> >>> + dev_info(dev, "can't get DMA channel\n");
> >>> st->dma_st.dma_chan = NULL;
> >>> goto dma_exit;
> >>> }
> >>> @@ -1691,7 +1690,7 @@ static void at91_adc_dma_init(struct platform_device *pdev)
> >>> &st->dma_st.rx_dma_buf,
> >>> GFP_KERNEL);
> >>> if (!st->dma_st.rx_buf) {
> >>> - dev_info(&pdev->dev, "can't allocate coherent DMA area\n");
> >>> + dev_info(dev, "can't allocate coherent DMA area\n");
> >>> goto dma_chan_disable;
> >>> }
> >>>
> >>> @@ -1704,11 +1703,11 @@ static void at91_adc_dma_init(struct platform_device *pdev)
> >>> config.dst_maxburst = 1;
> >>>
> >>> if (dmaengine_slave_config(st->dma_st.dma_chan, &config)) {
> >>> - dev_info(&pdev->dev, "can't configure DMA slave\n");
> >>> + dev_info(dev, "can't configure DMA slave\n");
> >>> goto dma_free_area;
> >>> }
> >>>
> >>> - dev_info(&pdev->dev, "using %s for rx DMA transfers\n",
> >>> + dev_info(dev, "using %s for rx DMA transfers\n",
> >>> dma_chan_name(st->dma_st.dma_chan));
> >>>
> >>> return;
> >>> @@ -1720,13 +1719,12 @@ static void at91_adc_dma_init(struct platform_device *pdev)
> >>> dma_release_channel(st->dma_st.dma_chan);
> >>> st->dma_st.dma_chan = NULL;
> >>> dma_exit:
> >>> - dev_info(&pdev->dev, "continuing without DMA support\n");
> >>> + dev_info(dev, "continuing without DMA support\n");
> >>> }
> >>>
> >>> -static void at91_adc_dma_disable(struct platform_device *pdev)
> >>> +static void at91_adc_dma_disable(struct at91_adc_state *st)
> >>> {
> >>> - struct iio_dev *indio_dev = platform_get_drvdata(pdev);
> >>> - struct at91_adc_state *st = iio_priv(indio_dev);
> >>> + struct device *dev = &st->indio_dev->dev;
> >>> /* we have 2 bytes for each channel */
> >>> unsigned int sample_size = st->soc_info.platform->nr_channels * 2;
> >>> unsigned int pages = DIV_ROUND_UP(AT91_HWFIFO_MAX_SIZE *
> >>> @@ -1744,7 +1742,7 @@ static void at91_adc_dma_disable(struct platform_device *pdev)
> >>> dma_release_channel(st->dma_st.dma_chan);
> >>> st->dma_st.dma_chan = NULL;
> >>>
> >>> - dev_info(&pdev->dev, "continuing without DMA support\n");
> >>> + dev_info(dev, "continuing without DMA support\n");
> >>> }
> >>>
> >>> static int at91_adc_set_watermark(struct iio_dev *indio_dev, unsigned int val)
> >>> @@ -1770,9 +1768,9 @@ static int at91_adc_set_watermark(struct iio_dev *indio_dev, unsigned int val)
> >>> */
> >>>
> >>> if (val == 1)
> >>> - at91_adc_dma_disable(to_platform_device(&indio_dev->dev));
> >>> + at91_adc_dma_disable(st);
> >>> else if (val > 1)
> >>> - at91_adc_dma_init(to_platform_device(&indio_dev->dev));
> >>> + at91_adc_dma_init(st);
> >>>
> >>> /*
> >>> * We can start the DMA only after setting the watermark and
> >>> @@ -1780,7 +1778,7 @@ static int at91_adc_set_watermark(struct iio_dev *indio_dev, unsigned int val)
> >>> */
> >>> ret = at91_adc_buffer_prepare(indio_dev);
> >>> if (ret)
> >>> - at91_adc_dma_disable(to_platform_device(&indio_dev->dev));
> >>> + at91_adc_dma_disable(st);
> >>>
> >>> return ret;
> >>> }
> >>> @@ -2077,7 +2075,7 @@ static int at91_adc_probe(struct platform_device *pdev)
> >>> return 0;
> >>>
> >>> dma_disable:
> >>> - at91_adc_dma_disable(pdev);
> >>> + at91_adc_dma_disable(st);
> >>> per_clk_disable_unprepare:
> >>> clk_disable_unprepare(st->per_clk);
> >>> vref_disable:
> >>> @@ -2094,7 +2092,7 @@ static int at91_adc_remove(struct platform_device *pdev)
> >>>
> >>> iio_device_unregister(indio_dev);
> >>>
> >>> - at91_adc_dma_disable(pdev);
> >>> + at91_adc_dma_disable(st);
> >>>
> >>> clk_disable_unprepare(st->per_clk);
> >>>
> >>
> >
>
