Hi Niklas,
Thank you for the patch.
On Thursday 21 January 2016 15:01:33 Niklas Söderlund wrote:
> Enable slave transfers to devices behind IPMMU:s by mapping the slave
> addresses using the dma-mapping API.
>
> Signed-off-by: Niklas Söderlund <niklas.soderlund+renesas@xxxxxxxxxxxx>
> ---
> drivers/dma/sh/rcar-dmac.c | 59 ++++++++++++++++++++++++++++++++++++++++----
> 1 file changed, 54 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/dma/sh/rcar-dmac.c b/drivers/dma/sh/rcar-dmac.c
> index 743873c..f3ec8a6 100644
> --- a/drivers/dma/sh/rcar-dmac.c
> +++ b/drivers/dma/sh/rcar-dmac.c
> @@ -1106,21 +1106,70 @@ rcar_dmac_prep_dma_cyclic(struct dma_chan *chan,
> dma_addr_t buf_addr, return desc;
> }
>
> +static int __rcar_dmac_device_config(struct dma_chan *chan,
I'd call this rcar_dmac_set_slave_addr (or something similar) to make the
purpose of the function more explicit.
> + struct rcar_dmac_chan_slave *slave,
> + phys_addr_t addr, size_t size,
> + enum dma_data_direction dir,
> + struct dma_attrs *attrs)
> +{
> + struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
You can pass rchan to the function directly and avoid this cast.
> + struct page *page;
> + size_t offset;
> +
> + /* unmap old */
The drivers capitalizes the first word of each sentence in comments and ends
the sentences with a full stop. Could you please use the same style for
consistency.
> + if (slave->slave_addr) {
In theory 0 is a valid address. There's unfortunately no standard way to set
an address to an invalid value that can be later checked (DMA_ERROR_CODE isn't
defined on all architectures and dma_mapping_error() has no set counterpart).
I would use xfer_size instead of slave_addr to check whether the page has been
mapped.
> + dma_unmap_page_attrs(chan->device->dev, slave->slave_addr,
> + slave->xfer_size, dir, attrs);
> + slave->slave_addr = 0;
> + slave->xfer_size = 0;
> + }
> +
> + /* map new */
> + if (addr) {
> + /* phys_to_page not available on all platforms */
> + page = pfn_to_page(addr >> PAGE_SHIFT);
> + offset = addr - page_to_phys(page);
How about just
offset = addr & ~PAGE_MASK;
It seems simpler to me.
> + slave->slave_addr = dma_map_page_attrs(chan->device->dev, page,
I'm still worried about passing a page pointer computed from an address that
has no associated page structure. This issue can probably be addressed
separately from this series, but I'd like to see at least a comment that
mentions the problem here.
> + offset, size, dir, attrs);
> +
> + if (dma_mapping_error(chan->device->dev, slave->slave_addr)) {
> + dev_err(chan->device->dev,
> + "chan%u: failed to map %zx@%pap",
> + rchan->index, size, &addr);
> + return -EIO;
> + }
> +
> + slave->xfer_size = size;
> + }
> +
> + return 0;
> +
> +}
> +
> static int rcar_dmac_device_config(struct dma_chan *chan,
> struct dma_slave_config *cfg)
> {
> struct rcar_dmac_chan *rchan = to_rcar_dmac_chan(chan);
> + struct dma_attrs attrs;
> + int ret;
>
> /*
> * We could lock this, but you shouldn't be configuring the
> * channel, while using it...
> */
> - rchan->src.slave_addr = cfg->src_addr;
> - rchan->dst.slave_addr = cfg->dst_addr;
> - rchan->src.xfer_size = cfg->src_addr_width;
> - rchan->dst.xfer_size = cfg->dst_addr_width;
>
> - return 0;
> + init_dma_attrs(&attrs);
> + dma_set_attr(DMA_ATTR_NO_KERNEL_MAPPING, &attrs);
> + dma_set_attr(DMA_ATTR_SKIP_CPU_SYNC, &attrs);
> +
> + ret = __rcar_dmac_device_config(chan, &rchan->src, cfg->src_addr,
You're passing a dma_addr_t while the function expects a phys_addr_t. As you
mentioned in the cover letter this has been discussed before without reaching
any clear conclusion, we need to come to an agreement.
> + cfg->src_addr_width, DMA_FROM_DEVICE, &attrs);
> + if (!ret)
> + return ret;
> +
> + ret = __rcar_dmac_device_config(chan, &rchan->dst, cfg->dst_addr,
> + cfg->dst_addr_width, DMA_TO_DEVICE, &attrs);
> + return ret;
> }
>
> static int rcar_dmac_chan_terminate_all(struct dma_chan *chan)
--
Regards,
Laurent Pinchart
