Dear Fabio, On Mon, 22 Jun 2015 08:33:08 +0200 David Jander <david@xxxxxxxxxxx> wrote: > Dear Fabio, > > Thanks for replying. > > On Sat, 20 Jun 2015 15:05:16 -0300 > Fabio Estevam <festevam@xxxxxxxxx> wrote: > > > On Thu, Jun 18, 2015 at 6:47 AM, David Jander <david@xxxxxxxxxxx> wrote: > > > > > > Hi > > > > > > I am having the imx.c serial port driver hang in imx_shutdown(). > > > I am running 4.1-rc8 on an i.MX6Q board. > > > If the user-space application calls close() on the serial device, > > > imx_shutdown() hangs in his line: > > > > > > ret = wait_event_interruptible(sport->dma_wait, > > > !sport->dma_is_rxing && !sport->dma_is_txing); > > > > > > wait_event_interruptible() does not return unless I receive some more > > > data. When entering wait_event_interruptible, sport->dma_is_rxing == 1 > > > and sport->dma_is_txing == 0. > > > From that moment onwards, dma_rx_callback() is not called unless I > > > receive more data, so sport->dma_is_rxing never gets cleared. > > > While bisecting the problem, I could identify the following commit as the > > > one that broke the driver for me: > > > > > > 068500e serial: imx: initialized DMA w/o HW flow enabled > > > > > > Obviously the problem is not in this commit, but probably in the DMA > > > code. I am using this driver on an i.MX6Q and have tried sdma firmware > > > version 1.1 and 3.1 so far. The problem appears the same with both. > > > > Could you also try it without loading any external SDMA firmware? > > Just tried and indeed, the problem is gone when using the ROM firmware. > But isn't the external firmware supposed to patch other bugs in the ROM > firmware? What are the consequences of using the ROM firmware? I am afraid this was a little too soon. There are still problems with DMA, even using the ROM firmware, it is just that the close() system call is not hanging anymore. I am considering to propose reversal of '068500e serial: imx: initialized DMA w/o HW flow enabled', because I suspect that SDMA support in this driver is just broken. I base this conclusion on the following test results: Test 1: Incoming data on imx serial port at 115200 baud in bursts (communication frames). Test 2: Outgoing data bursts (TX) with single-byte answers (RX). - imx.c and SDMA firmware version 3.1 or 1.1: The DMA interrupt in imx.c is not fired on the 1-byte reply and system call close() hangs on Test 2. I also see sporadic corruption in Test 1, where it is always the last byte of a frame that gets lost. - imx.c and SDMA firmware ROM version (no external firmware loaded): Sporadic RX data corruption. Single bytes get dropped or corrupted anywhere in the middle of a frame on Test 1. - imx.c with patch 068500e reverted (i.e. no DMA support): Both Test 1 and Test 2 work without any errors. All of this is tested on 4.1-rc8. Conclusion: (S)DMA support in drivers/tty/serial/imx.c is broken at least for i.MX6. So, I guess either we attempt to fix SDMA support in imx.c, or we revert commit 068500e for the time being. Best regards, -- David Jander Protonic Holland. -- To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
