Re: [PATCH] i2c: tegra: Allocate DMA memory for DMA engine

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Oct 20, 2022 at 04:14:20PM +0100, Robin Murphy wrote:
> On 2022-10-20 15:39, Thierry Reding wrote:
> > From: Thierry Reding <treding@xxxxxxxxxx>
> > 
> > When the I2C controllers are running in DMA mode, it is the DMA engine
> > that performs the memory accesses rather than the I2C controller. Pass
> > the DMA engine's struct device pointer to the DMA API to make sure the
> > correct DMA operations are used.
> > 
> > This fixes an issue where the DMA engine's SMMU stream ID needs to be
> > misleadingly set for the I2C controllers in device tree.
> > 
> > Suggested-by: Robin Murphy <robin.murphy@xxxxxxx>
> > Signed-off-by: Thierry Reding <treding@xxxxxxxxxx>
> > ---
> >   drivers/i2c/busses/i2c-tegra.c | 16 ++++++++++------
> >   1 file changed, 10 insertions(+), 6 deletions(-)
> > 
> > diff --git a/drivers/i2c/busses/i2c-tegra.c b/drivers/i2c/busses/i2c-tegra.c
> > index 954022c04cc4..3869c258a529 100644
> > --- a/drivers/i2c/busses/i2c-tegra.c
> > +++ b/drivers/i2c/busses/i2c-tegra.c
> > @@ -284,6 +284,7 @@ struct tegra_i2c_dev {
> >   	struct dma_chan *tx_dma_chan;
> >   	struct dma_chan *rx_dma_chan;
> >   	unsigned int dma_buf_size;
> > +	struct device *dma_dev;
> >   	dma_addr_t dma_phys;
> >   	void *dma_buf;
> > @@ -420,7 +421,7 @@ static int tegra_i2c_dma_submit(struct tegra_i2c_dev *i2c_dev, size_t len)
> >   static void tegra_i2c_release_dma(struct tegra_i2c_dev *i2c_dev)
> >   {
> >   	if (i2c_dev->dma_buf) {
> > -		dma_free_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> > +		dma_free_coherent(i2c_dev->dma_dev, i2c_dev->dma_buf_size,
> >   				  i2c_dev->dma_buf, i2c_dev->dma_phys);
> >   		i2c_dev->dma_buf = NULL;
> >   	}
> > @@ -472,10 +473,13 @@ static int tegra_i2c_init_dma(struct tegra_i2c_dev *i2c_dev)
> >   	i2c_dev->tx_dma_chan = chan;
> > +	WARN_ON(i2c_dev->tx_dma_chan->device != i2c_dev->rx_dma_chan->device);
> > +	i2c_dev->dma_dev = chan->device->dev;
> > +
> >   	i2c_dev->dma_buf_size = i2c_dev->hw->quirks->max_write_len +
> >   				I2C_PACKET_HEADER_SIZE;
> > -	dma_buf = dma_alloc_coherent(i2c_dev->dev, i2c_dev->dma_buf_size,
> > +	dma_buf = dma_alloc_coherent(i2c_dev->dma_dev, i2c_dev->dma_buf_size,
> >   				     &dma_phys, GFP_KERNEL | __GFP_NOWARN);
> >   	if (!dma_buf) {
> >   		dev_err(i2c_dev->dev, "failed to allocate DMA buffer\n");
> 
> That much is definitely better, but as mentioned previously all the syncs
> below look completely bogus either way, so should really be removed rather
> than "fixed". If it's necessary to ensure ordering of CPU accesses to the
> buffer relative to the DMA transfer itself, where that isn't already
> implicit in some readl()/writel() involved in starting and stopping the DMA
> channel, then dma_wmb()/dma_rmb() are the tools for that job.

Yes, I've had those comments in the back of my mind and wanted to get
this out separately because this is needed to redress an issue that
we've introduced in the DT files for these devices, where now all I2C
controllers are sorted into the same IOMMU group as the DMA controller,
which doesn't make any sense as they are not themselves performing any
DMA accesses. I'll look at the synchronization bits next.

I suspect that you're right about the sync calls all being redundant. If
dma_alloc_coherent() memory is always guaranteed to be uncached (unless
perhaps if dma-coherent is specified for the DMA device), then I don't
see why they should be needed.

Thierry

> 
> Thanks,
> Robin.
> 
> > @@ -1272,7 +1276,7 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
> >   	if (i2c_dev->dma_mode) {
> >   		if (i2c_dev->msg_read) {
> > -			dma_sync_single_for_device(i2c_dev->dev,
> > +			dma_sync_single_for_device(i2c_dev->dma_dev,
> >   						   i2c_dev->dma_phys,
> >   						   xfer_size, DMA_FROM_DEVICE);
> > @@ -1280,7 +1284,7 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
> >   			if (err)
> >   				return err;
> >   		} else {
> > -			dma_sync_single_for_cpu(i2c_dev->dev,
> > +			dma_sync_single_for_cpu(i2c_dev->dma_dev,
> >   						i2c_dev->dma_phys,
> >   						xfer_size, DMA_TO_DEVICE);
> >   		}
> > @@ -1293,7 +1297,7 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
> >   			memcpy(i2c_dev->dma_buf + I2C_PACKET_HEADER_SIZE,
> >   			       msg->buf, msg->len);
> > -			dma_sync_single_for_device(i2c_dev->dev,
> > +			dma_sync_single_for_device(i2c_dev->dma_dev,
> >   						   i2c_dev->dma_phys,
> >   						   xfer_size, DMA_TO_DEVICE);
> > @@ -1344,7 +1348,7 @@ static int tegra_i2c_xfer_msg(struct tegra_i2c_dev *i2c_dev,
> >   		}
> >   		if (i2c_dev->msg_read && i2c_dev->msg_err == I2C_ERR_NONE) {
> > -			dma_sync_single_for_cpu(i2c_dev->dev,
> > +			dma_sync_single_for_cpu(i2c_dev->dma_dev,
> >   						i2c_dev->dma_phys,
> >   						xfer_size, DMA_FROM_DEVICE);

Attachment: signature.asc
Description: PGP signature


[Index of Archives]     [Linux GPIO]     [Linux SPI]     [Linux Hardward Monitoring]     [LM Sensors]     [Linux USB Devel]     [Linux Media]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux