On 17:55 Mon 05 Feb , Robin Murphy wrote:
> On 2024-02-04 6:59 am, Andrea della Porta wrote:
> > From: Phil Elwell <phil@xxxxxxxxxxxxxxx>
> >
> > Unless the DMA mask is set wider than 32 bits, DMA mapping will use a
> > bounce buffer.
> >
> > Signed-off-by: Phil Elwell <phil@xxxxxxxxxxxxxxx>
> > ---
> > drivers/dma/bcm2835-dma.c | 18 +++++++++++++++---
> > 1 file changed, 15 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/dma/bcm2835-dma.c b/drivers/dma/bcm2835-dma.c
> > index 36bad198b655..237dcdb8d726 100644
> > --- a/drivers/dma/bcm2835-dma.c
> > +++ b/drivers/dma/bcm2835-dma.c
> > @@ -39,6 +39,7 @@
> > #define BCM2711_DMA_MEMCPY_CHAN 14
> > struct bcm2835_dma_cfg_data {
> > + u64 dma_mask;
> > u32 chan_40bit_mask;
> > };
> > @@ -308,10 +309,12 @@ DEFINE_SPINLOCK(memcpy_lock);
> > static const struct bcm2835_dma_cfg_data bcm2835_dma_cfg = {
> > .chan_40bit_mask = 0,
> > + .dma_mask = DMA_BIT_MASK(32),
> > };
> > static const struct bcm2835_dma_cfg_data bcm2711_dma_cfg = {
> > .chan_40bit_mask = BIT(11) | BIT(12) | BIT(13) | BIT(14),
> > + .dma_mask = DMA_BIT_MASK(36),
> > };
> > static inline size_t bcm2835_dma_max_frame_length(struct bcm2835_chan *c)
> > @@ -1263,6 +1266,8 @@ static struct dma_chan *bcm2835_dma_xlate(struct of_phandle_args *spec,
> > static int bcm2835_dma_probe(struct platform_device *pdev)
> > {
> > + const struct bcm2835_dma_cfg_data *cfg_data;
> > + const struct of_device_id *of_id;
> > struct bcm2835_dmadev *od;
> > struct resource *res;
> > void __iomem *base;
> > @@ -1272,13 +1277,20 @@ static int bcm2835_dma_probe(struct platform_device *pdev)
> > int irq_flags;
> > uint32_t chans_available;
> > char chan_name[BCM2835_DMA_CHAN_NAME_SIZE];
> > - const struct of_device_id *of_id;
> > int chan_count, chan_start, chan_end;
> > + of_id = of_match_node(bcm2835_dma_of_match, pdev->dev.of_node);
> > + if (!of_id) {
> > + dev_err(&pdev->dev, "Failed to match compatible string\n");
> > + return -EINVAL;
> > + }
> > +
> > + cfg_data = of_id->data;
>
> We've had of_device_get_match_data() for nearly 9 years now, and even a
> generic device_get_match_data() for 6 ;)
>
> > +
> > if (!pdev->dev.dma_mask)
> > pdev->dev.dma_mask = &pdev->dev.coherent_dma_mask;
>
> [ Passing nit: that also really shouldn't be there, especially since
> cdfee5623290 ]
>
> > - rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
> > + rc = dma_set_mask_and_coherent(&pdev->dev, cfg_data->dma_mask);
>
> Wait, does chan_40bit_mask mean that you still have some channels which
> *can't* address this full mask? If so this can't work properly. You may well
> need to redesign a bit further to have a separate DMA device for each
> channel such they can each have different masks.
>
It seems that the original intention here was to create a device for each value of dma_mask in
hw descriptors. That is, for 2711 which has 32 and 40 bit channels, the DT should look something
like this:
dma: dma-controller@7e007000 {
interrupts = <...>;
brcm,dma-channel-mask = <0x7f5>;
compatible = "brcm,bcm2835-dma";
interrupt-names = "...";
reg = <0x7e007000 0xb00>;
#dma-cells = <0x01>;
};
dma40: dma-controller@7e007b00 {
interrupts = <...>;
brcm,dma-channel-mask = <0x3000>;
compatible = "brcm,bcm2711-dma";
interrupt-names = "...";
reg = <0x00 0x7e007b00 0x00 0x400>;
#dma-cells = <0x01>;
};
Two devices dma0 and dma1 will be created, each one serving a different mask and the call
to dma_set_mask_and_coherent(..., dma_mask) on the specific device will be consistent. Please
note that of course "brcm,dma-channel-mask" from DT only refers to what channels are available
to be used in the kernel, while dma_mask parameter of the aforementioned dma_set_mask_and_coherent()
call is the addressing mask enforced by the driver, and its the same for each specific device
(dma0 or dma1).
Many thanks,
Andrea
