Hi Tim, On Thu, Sep 21, 2023 at 09:57:11AM +0000, Tim van der Staaij | Zign wrote: > dev_warn internally acquires the lock that is already held when > sdma_update_channel_loop is called. Therefore it is acquired twice and > this is detected as a deadlock. Temporarily release the lock while > logging to avoid this. > > Signed-off-by: Tim van der Staaij <tim.vanderstaaij@xxxxxxxxxxxxx> > Link: https://lore.kernel.org/all/AM0PR08MB308979EC3A8A53AE6E2D3408802CA@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/ > --- > drivers/dma/imx-sdma.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c > index 51012bd39900..3a7cd783a567 100644 > --- a/drivers/dma/imx-sdma.c > +++ b/drivers/dma/imx-sdma.c > @@ -904,7 +904,10 @@ static void sdma_update_channel_loop(struct sdma_channel *sdmac) > * owned buffer is available (i.e. BD_DONE was set too late). > */ > if (sdmac->desc && !is_sdma_channel_enabled(sdmac->sdma, sdmac->channel)) { > + spin_unlock(&sdmac->vc.lock); > dev_warn(sdmac->sdma->dev, "restart cyclic channel %d\n", sdmac->channel); > + spin_lock(&sdmac->vc.lock); This is strange. Why and how does dev_warn() call back into the SDMA driver? We shouldn't merge this without having a clue what exactly goes wrong here. Please provide the corresponding lockdep output. Sascha -- Pengutronix e.K. | | Steuerwalder Str. 21 | http://www.pengutronix.de/ | 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
