DMA transfers may be slow if the peer device is slow (eg. a UART controller, a SPI controller...). The completion worker is started very early compared to the actual draining speed, leading to low speed devices suffering from a full 1s delay before noticing the transfers are done, further delaying the execution of their callback.
1s seems overly large for a polling delay, reduce it to arbitrarily 100us which is the minimum amount of time to transfer a byte (and see some progress on the residue variable) on a 100kHz bus. Based on this first measure, the next sleep will be much closer to what is actually needed for the transfer to complete.
The 1 second polling delay involves that any device driver with a (very standard) 1 second timeout will error out indicating a transfer error, whereas the data has actually been transferred correctly.
Fixes: 25dcb5dd7b7c ("dmaengine: ti: New driver for K3 UDMA") Cc: stable@vger.kernel.org # 5.7 Signed-off-by: Miquel Raynal miquel.raynal@bootlin.com --- This issue has been observed by playing with the DMA controller on Beagle Play.
The patch will not apply before v5.7 and probably do not need to be backported before anyway. --- drivers/dma/ti/k3-udma.c | 17 ++++++++--------- 1 file changed, 8 insertions(+), 9 deletions(-)
diff --git a/drivers/dma/ti/k3-udma.c b/drivers/dma/ti/k3-udma.c index aa2dc762140f6eee334f4506a592e72600ae9834..b2059baed1b2ffc81c10feca797c763e2a04a357 100644 --- a/drivers/dma/ti/k3-udma.c +++ b/drivers/dma/ti/k3-udma.c @@ -1115,19 +1115,18 @@ static void udma_check_tx_completion(struct work_struct *work) time_diff = ktime_sub(uc->tx_drain.tstamp, time_diff) + 1; residue_diff -= uc->tx_drain.residue; + /* + * Try to guess when we should check next time by + * calculating rate at which data is being drained at + * the peer device. Slow devices might have not yet + * started, showing no progress. Use an arbitrary delay + * in this case. + */ if (residue_diff) { - /* - * Try to guess when we should check - * next time by calculating rate at - * which data is being drained at the - * peer device - */ delay = (time_diff / residue_diff) * uc->tx_drain.residue; } else { - /* No progress, check again in 1 second */ - schedule_delayed_work(&uc->tx_drain.work, HZ); - break; + delay = 100000; }
spin_unlock_irqrestore(&uc->vc.lock, flags);