On 04/11/2016 10:10 PM, Peter Hurley wrote: > On 04/11/2016 11:31 AM, Sebastian Andrzej Siewior wrote: >> On 04/11/2016 07:53 PM, Peter Hurley wrote: >>> On 04/11/2016 01:18 AM, John Ogness wrote: >>>> On 2016-04-05, Peter Hurley <peter@xxxxxxxxxxxxxxxxxx> wrote: >>>>> On 03/31/2016 01:41 AM, John Ogness wrote: >>>>>> It has been observed that the TX-DMA can stall >>>>> >>>>> Does this happen on any other OMAP part besides am335x? >>>>> I looked back over the LKML history of this and didn't see >>>>> any other design implicated in this problem. >>>> >>>> I just ran the tests again using 4.6-rc2. I am able to reproduce the >>>> dma-tx stall with am335x/edma and dra7/sdma. >>> >>> I thought we already established sdma was not to be used since >>> the hardware does not actually support pausing without data loss. >> >> This workaround was not invented for sdma but for edma (with am335x). > > According to John above, dra7/sdma requires this workaround. It was reported by Frans Klaver against am335x http://lkml.kernel.org/r/20140908183353.GB4686@ci00147.xsens-tech.local and I managed to reproduce this with his yocto image on dra7 and am335x: http://lkml.kernel.org/r/20140921204100.GA10111@xxxxxxxxxxxxx >>> http://www.spinics.net/lists/linux-serial/msg18503.html >> >> This could be fixed. See >> http://www.spinics.net/lists/linux-serial/msg18517.html >> http://www.spinics.net/lists/linux-serial/msg18531.html >> >> rmk was fine with it from what I read. So what is missing is just >> refurbish the patch (update the comment according to rmk replay) and >> then we could re-enable DMA again. > > That's hardly all that is required. well it would enable pause of RX transfers. TX would not work (unless TI's HW people can confirm that it will). > 1. edma pause returns an error if the descriptor has already been retired > when a pause is attempted. This makes distinguishing between reporting an > error for unsupported feature indistinguishable from a transient dma > error that can simply be logged. > 2. The question of a spurious uart interrupt with every dma transaction > on am335x is still unanswered. This is correct. If I remember correctly, the Intel people see the same thing and I *think* John told me that the Intel manual says that RDI should be disabled if DMA is used. > 3. Handling XON/XOFF transmit is mandatory; I don't see a way to do that > without pause/resume. Yes, not doing XON/XOFF with DMA is not good. Using hardware flow control is one workaround but the user has no chance of knowing that XON/XOFF has been silently disabled. You could send the x_char after TX transfer completed. After all you need to ensure that you have some space in the TX-FIFO. However if you send a 4KiB of data you might want to send x_char rather sooner than later. I *think* even with pause the hardware will complete the last burst before stopping but is probably better than waiting for the 4KiB to complete. > 4. Since virt-dma uses tasklets which since 3.8 are no longer serviced > in a timely manner, rx dma is unreliable, since it's often kicked out > to regular interrupts. Is this only the delay in omap_dma_callback() (which you don't have !cyclic) or something else? omap_dma_issue_pending() seems to program the transfer right away. Oh now I see the same thing in edma_completion_handler(). Okay but this affects now everyone that relies on low latency? > 5. omap dma maintenance is not keeping up with baseline dma. John switched to cylic mode so he was not effected very much non-pause problem. > IOW, omap dma has turned into one big tangle of workarounds. Most of them are hardware shortcomings. I think disabling RX-DMA due to missing pause in omap-dma is the only workaround that could be avoided if the driver would be changed. > Let's start with making a list of which TI designs need which workarounds. > > *am335x* I am not sure if the limitations are based on the DMA engine or the > - requires write to tx fifo to trigger tx dma (ie. OMAP_DMA_TX_KICK > workaround necessitating completely different tx dma completion handler) This one for instance I don't see on BeagleBoard xM / omap36xx and DRA7x. Both (not affected) use SDMA instead EDMA. It would be interesting to see if DRA7x is affected once it uses EDMA. > - requires rx dma already queued before UART data ready interrupt > (ie., necessitates completely different irq handler and rx dma completion > handler) true. But is this something that would work for others, too? > - hangs changing some unknown register if tx dma in progress > (ie., this termios change workaround) I think some registers are the baud-rate registers which pause engine. > - generates spurious uart interrupt for every rx dma transaction > (ie., necessitates acking every UART interrupt, even UART_IIR_NO_INT) > _Even with this workaround_, it still generates spurious interrupt warning > which shuts off interrupts for several ms while logging the error > message to the console, virtually guaranteeing lost data. as I wrote in my other email I think RDI should be disabled with DMA according the Intel manual and I *think* someone here reported that they see the same problem. > Can any TI design use the baseline 8250 tx dma transaction flow without > workarounds? I know the am335x can't; any others? Am335x. Has edma and so has dm814x. According to the code, dm814x based HW does not need it, can this be confirmed? Sekhar, TOny? > Can any TI design use the baseline 8250 rx dma transaction flow without > workarounds? Again, I know the am335x can't; any others? Is dra7 out? Because that one needs to enqueue RX transfers asap. And omap36xx (aka BeagleBoard-xm) as well. I don't kown anything about later SoCs (like am437x and so on) but I would assume so. Sebastian -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html