| Message ID | 20180621115822.20058-1-andrea.merello@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
> -----Original Message----- > From: dmaengine-owner@vger.kernel.org [mailto:dmaengine- > owner@vger.kernel.org] On Behalf Of Andrea Merello > Sent: Thursday, June 21, 2018 5:28 PM > To: vkoul@kernel.org; dan.j.williams@intel.com; Michal Simek > <michals@xilinx.com>; Appana Durga Kedareswara Rao > <appanad@xilinx.com>; dmaengine@vger.kernel.org > Cc: linux-arm-kernel@lists.infradead.org; linux-kernel@vger.kernel.org; > Andrea Merello <andrea.merello@gmail.com> > Subject: [PATCH v2 1/5] dmaengine: xilinx_dma: in axidma slave_sg and > dma_cyclic mode align split descriptors > > Whenever a single or cyclic transaction is prepared, the driver > could eventually split it over several SG descriptors in order > to deal with the HW maximum transfer length. > > This could end up in DMA operations starting from a misaligned > address. This seems fatal for the HW if DRE is not enabled. > > This patch eventually adjusts the transfer size in order to make sure > all operations start from an aligned address. > > Signed-off-by: Andrea Merello <andrea.merello@gmail.com> > --- > Changes in v2: > - don't introduce copy_mask field, rather rely on already-esistent > copy_align field. Suggested by Radhey Shyam Pandey > - reword title > --- > drivers/dma/xilinx/xilinx_dma.c | 22 ++++++++++++++++------ > 1 file changed, 16 insertions(+), 6 deletions(-) > > diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c > index 27b523530c4a..22d7a6b85e65 100644 > --- a/drivers/dma/xilinx/xilinx_dma.c > +++ b/drivers/dma/xilinx/xilinx_dma.c > @@ -1789,10 +1789,15 @@ static struct dma_async_tx_descriptor > *xilinx_dma_prep_slave_sg( > > /* > * Calculate the maximum number of bytes to transfer, > - * making sure it is less than the hw limit > + * making sure it is less than the hw limit and that > + * the next chunck start address is aligned /s/chunck/chunk . Same for later occurrence. > */ > - copy = min_t(size_t, sg_dma_len(sg) - sg_used, > - XILINX_DMA_MAX_TRANS_LEN); > + copy = sg_dma_len(sg) - sg_used; > + if (copy > XILINX_DMA_MAX_TRANS_LEN && > + chan->xdev->common.copy_align) > + copy = > rounddown(XILINX_DMA_MAX_TRANS_LEN, > + (1 << chan->xdev- > >common.copy_align)); > + If DRE is not enabled (copy_align=0) we are copying entire sg_dma_len which is not correct i.e more than XILINX_DMA_MAX_TRANS_LEN. > hw = &segment->hw; > > /* Fill in the descriptor */ > @@ -1894,10 +1899,15 @@ static struct dma_async_tx_descriptor > *xilinx_dma_prep_dma_cyclic( > > /* > * Calculate the maximum number of bytes to transfer, > - * making sure it is less than the hw limit > + * making sure it is less than the hw limit and that > + * the next chunck start address is aligned > */ > - copy = min_t(size_t, period_len - sg_used, > - XILINX_DMA_MAX_TRANS_LEN); > + copy = period_len - sg_used; > + if (copy > XILINX_DMA_MAX_TRANS_LEN && > + chan->xdev->common.copy_align) > + copy = > rounddown(XILINX_DMA_MAX_TRANS_LEN, > + (1 << chan->xdev- > >common.copy_align)); > + > hw = &segment->hw; > xilinx_axidma_buf(chan, hw, buf_addr, sg_used, > period_len * i); > -- > 2.17.1 > > -- > To unsubscribe from this list: send the line "unsubscribe dmaengine" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c index 27b523530c4a..22d7a6b85e65 100644 --- a/drivers/dma/xilinx/xilinx_dma.c +++ b/drivers/dma/xilinx/xilinx_dma.c @@ -1789,10 +1789,15 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_slave_sg( /* * Calculate the maximum number of bytes to transfer, - * making sure it is less than the hw limit + * making sure it is less than the hw limit and that + * the next chunck start address is aligned */ - copy = min_t(size_t, sg_dma_len(sg) - sg_used, - XILINX_DMA_MAX_TRANS_LEN); + copy = sg_dma_len(sg) - sg_used; + if (copy > XILINX_DMA_MAX_TRANS_LEN && + chan->xdev->common.copy_align) + copy = rounddown(XILINX_DMA_MAX_TRANS_LEN, + (1 << chan->xdev->common.copy_align)); + hw = &segment->hw; /* Fill in the descriptor */ @@ -1894,10 +1899,15 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_dma_cyclic( /* * Calculate the maximum number of bytes to transfer, - * making sure it is less than the hw limit + * making sure it is less than the hw limit and that + * the next chunck start address is aligned */ - copy = min_t(size_t, period_len - sg_used, - XILINX_DMA_MAX_TRANS_LEN); + copy = period_len - sg_used; + if (copy > XILINX_DMA_MAX_TRANS_LEN && + chan->xdev->common.copy_align) + copy = rounddown(XILINX_DMA_MAX_TRANS_LEN, + (1 << chan->xdev->common.copy_align)); + hw = &segment->hw; xilinx_axidma_buf(chan, hw, buf_addr, sg_used, period_len * i);
Whenever a single or cyclic transaction is prepared, the driver could eventually split it over several SG descriptors in order to deal with the HW maximum transfer length. This could end up in DMA operations starting from a misaligned address. This seems fatal for the HW if DRE is not enabled. This patch eventually adjusts the transfer size in order to make sure all operations start from an aligned address. Signed-off-by: Andrea Merello <andrea.merello@gmail.com> --- Changes in v2: - don't introduce copy_mask field, rather rely on already-esistent copy_align field. Suggested by Radhey Shyam Pandey - reword title --- drivers/dma/xilinx/xilinx_dma.c | 22 ++++++++++++++++------ 1 file changed, 16 insertions(+), 6 deletions(-)