Re: [PATCH v1] tty: serial: imx: enable IDDMAEN for the last tail data

[Lucas Stach]

Am Donnerstag, den 20.09.2018, 10:33 +0000 schrieb Andy Duan:
> From: Lucas Stach <l.stach@xxxxxxxxxxxxxx> Sent: 2018å9æ20æ 17:40
> > Am Donnerstag, den 20.09.2018, 08:39 +0000 schrieb Robin Gong:
> > > > -----Original Message-----
> > > > From: Uwe Kleine-KÃnig <u.kleine-koenig@xxxxxxxxxxxxxx>
> > > > Sent: 2018å9æ20æ 15:55
> > > > To: Robin Gong <yibin.gong@xxxxxxx>
> > > > Cc: jslaby@xxxxxxxx; Andy Duan <fugang.duan@xxxxxxx>;
> > > > linux-serial@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; dl-
> > > > linux-imx <linux-imx@xxxxxxx>; kernel@xxxxxxxxxxxxxx
> > > > Subject: Re: [PATCH v1] tty: serial: imx: enable IDDMAEN for
> > > > the
> > > > last tail data
> > > > 
> > > > Hello,
> > > > 
> > > > On Thu, Sep 20, 2018 at 11:26:00PM +0800, Robin Gong wrote:
> > > > > enable IDDMAEN in UCR4 to let sdma script has the chance to
> > > > > detect
> > > > > the idle status and transfer the last tail data with the
> > > > > interrupt
> > > > > notifying uart driver.Otherwise, the last dma done interrupt
> > > > > of
> > > > > the tail data in rxfifo whose size is less than watermark may
> > > > > never be received by uart driver.
> > > > 
> > > > Is this a theoretic issue? Or does it fix a real problem?
> > > > 
> > > > If the former I'd say UCR1_ATDMAEN being set should prevent
> > > > this
> > > > stall.
> > > 
> > > It's a real fix, you will see 'timeout We read 0 byte' with the
> > > attached stress test running on i.mx7d or i.mx6 if no this patch
> > > applied.
> > > ' while [ 1 ]; do /unit_tests/UART/mxc_uart_stress_test.out
> > > /dev/ttymxc5 2000000 D L 100 100 L; done &'
> > > Please make sure dma enabled on ttymxcX for your test, and sdma
> > > firmware setup in your rootfs /lib/firmware/imx/sdma/sdma-
> > > imx6q.bin
> > 
> > or
> > > /lib/firmware/imx/sdma/sdma- imx7d.bin.
> > > You can get sdma firmware from below:
> > > 
> > 
> > https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2F
> > gi
> > t
> > > .kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Ffirmware%2Flin
> > 
> > ux-&amp
> > > ;data=02%7C01%7Cfugang.duan%40nxp.com%7C213076f6d1114148b5
> > 
> > ce08d61edcfb
> > > 
> > 
> > 88%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C636730331
> > 814804545&amp
> > > ;sdata=d0uZMjLc%2FrIae3Wd6PWtX2gEGlvZcUjPBVqKK9550jg%3D&amp
> > 
> > ;reserved=0
> > > firmware.git/tree/imx/sdma
> > 
> > Please see commit 905c0decad28402aa166975023fb88c8f62f93c8 on
> > why using the idle detect together with with SDMA is wrong. We
> > don't
> > want to re- introduce the broken behavior.
> > 
> 
> The patch just want to use idle timer to stop SDMA transfer using the
> new ram script, whose action is not the same as rom script.
> Sdma rom script how to stop sdma transfer if only one byte in rx FIFO
> that trigger SDMA request ?

Right, and using the idle detect doesn't work with the ROM script as
this doesn't read the remaining bytes in the FIFO when the idle detect
condition is hit.

Instead with the ROM script the aging timer is used to terminate the
DMA transfer. So if 1 byte is sitting in the FIFO below the watermark
level, the serial core will trigger a DMA request when the aging time
is hit. The ROM script checks the aging status and reads the remaining
bytes from the FIFO, then terminates the transfer. This is all
documented in the reference manual.

I fail to see why this shouldn't work with the RAM script. If it
doesn't work with the RAM firmware, this script is violating the
interface between the serial core and the script in an incompatible way
and needs fixing.

Regards,
Lucas