On 4/23/21 3:19 AM, Adrian Larumbe wrote:
This was required as an answer to a very unusual race condition, whereby a
thread waiting on a completion signaled in a callback triggered by
dmaengine_desc_callback_invoke might immediately attempt to release the
DMA channel whilst the channel lock was still free. This would cause the
remaining descriptors to be deallocated, and xilinx_dma_chan_desc_cleanup
to perform an invalid memory access when attempting to traverse the rest
of the channel's done_list.
Now, when releasing a DMA channel, it will wait until the tasklet has
finished.
Signed-off-by: Adrian Larumbe <adrian.martinezlarumbe@xxxxxxxxxx>
Hi,
Patch looks good, but basically the same got already applied a few weeks
ago.
See
https://git.kernel.org/pub/scm/linux/kernel/git/vkoul/dmaengine.git/commit/?h=next&id=50db2050faf854cbaf4b6557a7a8ca21bff302ae
- Lars
---
drivers/dma/xilinx/xilinx_dma.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index 40c6cf8bf0e6..d4bad999e7f9 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c
@@ -942,6 +942,13 @@ static void xilinx_dma_free_chan_resources(struct dma_chan *dchan)
}
+static void xilinx_dma_synchronize(struct dma_chan *dchan)
+{
+ struct xilinx_dma_chan *chan = to_xilinx_chan(dchan);
+
+ tasklet_kill(&chan->tasklet);
+}
+
/**
* xilinx_dma_get_residue - Compute residue for a given descriptor
* @chan: Driver specific dma channel
@@ -3235,6 +3242,7 @@ static int xilinx_dma_probe(struct platform_device *pdev)
xdev->common.device_tx_status = xilinx_dma_tx_status;
xdev->common.device_issue_pending = xilinx_dma_issue_pending;
xdev->common.device_config = xilinx_dma_slave_config;
+ xdev->common.device_synchronize = xilinx_dma_synchronize;
if (xdev->dma_config->dmatype == XDMA_TYPE_AXIDMA) {
dma_cap_set(DMA_CYCLIC, xdev->common.cap_mask);
xdev->common.device_prep_slave_sg = xilinx_dma_prep_slave_sg;
