This is a note to let you know that I've just added the patch titled dmaengine: edma: Align the memcpy acnt array size with the transfer to the 4.9-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary The filename of the patch is: dmaengine-edma-align-the-memcpy-acnt-array-size-with-the-transfer.patch and it can be found in the queue-4.9 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable@xxxxxxxxxxxxxxx> know about it. >From 87a2f622cc6446c7d09ac655b7b9b04886f16a4c Mon Sep 17 00:00:00 2001 From: Peter Ujfalusi <peter.ujfalusi@xxxxxx> Date: Mon, 18 Sep 2017 11:16:26 +0300 Subject: dmaengine: edma: Align the memcpy acnt array size with the transfer From: Peter Ujfalusi <peter.ujfalusi@xxxxxx> commit 87a2f622cc6446c7d09ac655b7b9b04886f16a4c upstream. Memory to Memory transfers does not have any special alignment needs regarding to acnt array size, but if one of the areas are in memory mapped regions (like PCIe memory), we need to make sure that the acnt array size is aligned with the mem copy parameters. Before "dmaengine: edma: Optimize memcpy operation" change the memcpy was set up in a different way: acnt == number of bytes in a word based on __ffs((src | dest | len), bcnt and ccnt for looping the necessary number of words to comlete the trasnfer. Instead of reverting the commit we can fix it to make sure that the ACNT size is aligned to the traswnfer. Fixes: df6694f80365a (dmaengine: edma: Optimize memcpy operation) Signed-off-by: Peter Ujfalusi <peter.ujfalusi@xxxxxx> Signed-off-by: Vinod Koul <vinod.koul@xxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- drivers/dma/edma.c | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-) --- a/drivers/dma/edma.c +++ b/drivers/dma/edma.c @@ -1143,11 +1143,24 @@ static struct dma_async_tx_descriptor *e struct edma_desc *edesc; struct device *dev = chan->device->dev; struct edma_chan *echan = to_edma_chan(chan); - unsigned int width, pset_len; + unsigned int width, pset_len, array_size; if (unlikely(!echan || !len)) return NULL; + /* Align the array size (acnt block) with the transfer properties */ + switch (__ffs((src | dest | len))) { + case 0: + array_size = SZ_32K - 1; + break; + case 1: + array_size = SZ_32K - 2; + break; + default: + array_size = SZ_32K - 4; + break; + } + if (len < SZ_64K) { /* * Transfer size less than 64K can be handled with one paRAM @@ -1169,7 +1182,7 @@ static struct dma_async_tx_descriptor *e * When the full_length is multibple of 32767 one slot can be * used to complete the transfer. */ - width = SZ_32K - 1; + width = array_size; pset_len = rounddown(len, width); /* One slot is enough for lengths multiple of (SZ_32K -1) */ if (unlikely(pset_len == len)) @@ -1217,7 +1230,7 @@ static struct dma_async_tx_descriptor *e } dest += pset_len; src += pset_len; - pset_len = width = len % (SZ_32K - 1); + pset_len = width = len % array_size; ret = edma_config_pset(chan, &edesc->pset[1], src, dest, 1, width, pset_len, DMA_MEM_TO_MEM); Patches currently in stable-queue which might be from peter.ujfalusi@xxxxxx are queue-4.9/dmaengine-edma-align-the-memcpy-acnt-array-size-with-the-transfer.patch queue-4.9/dmaengine-ti-dma-crossbar-fix-possible-race-condition-with-dma_inuse.patch