Re: DM3730 Bluetooth Performance differences between SERIAL_8250_OMAP vs SERIAL_OMAP

[Adam Ford]

On Fri, Oct 4, 2019 at 9:08 AM Adam Ford <aford173@xxxxxxxxx> wrote:
>
> On Fri, Oct 4, 2019 at 7:27 AM Yegor Yefremov
> <yegorslists@xxxxxxxxxxxxxx> wrote:
> >
> > Hi Adam,
> >
> > On Fri, Oct 4, 2019 at 12:39 PM Adam Ford <aford173@xxxxxxxxx> wrote:
> > >
> > > On Fri, Oct 4, 2019 at 5:02 AM Adam Ford <aford173@xxxxxxxxx> wrote:
> > > >
> > > > I am running Kernel 5.3.2 trying to troubleshoot some intermittent
> > > > Bluetooth issues, and I think I have narrowed it down to the serial
> > > > driver in use.
> > >
> > > I should have also noted that it's using UART2 with CTS and RTS on the
> > > DM3730 (omap3630) and its configured with a baud rate of 3M.
> > > I tried slowing it to 115200, but that didn't help.  I tried disabling
> > > the DMA hooks from the device tree, and that didn't help.
> > >
> > > > By default, omap2plus_defconfig enables both SERIAL_8250_OMAP and
> > > > SERIAL_OMAP.  I have my console device configured as  ttyS0, and all
> > > > appears fine.  When I enable Bluetooth, however, I get intermittent
> > > > errors on an DM3730 / OMAP3630.
> > > >
> > > > Using the 8250 driver for Blueotooth I get intermittent frame errors
> > > > and data loss.
> > > >
> > > > Scanning ...
> > > > [   28.482452] Bluetooth: hci0: Frame reassembly failed (-84)
> > > > [   36.162170] Bluetooth: hci0: Frame reassembly failed (-84)
> > > >         F4:4E:FC:C9:2F:57       BluJax
> > > > # l2ping F4:4E:FC:C9:2F:57
> > > > Ping: F4:4E:FC:C9:2F:57 from 00:18:30:49:7D:63 (data size 44) ...
> > > > 44 bytes from F4:4E:FC:C9:2F:57 id 0 time 8.27ms
> > > > no response from F4:4E:FC:C9:2F:57: id 1
> > > > ^C2 sent, 1 received, 50% loss
> > > >
> > > > (after a fairly long hang, I hit control-c)
> > > >
> > > > However, disabling the 8250 driver and using the only SERIAL_OMAP and
> > > > the console routed to ttyO0, the Bluetooth works well, so I believe it
> > > > to be a serial driver issue and not a Bluetooth error.
> > > >
> > > > # hcitool scan
> > > > Scanning ...
> > > >         F4:4E:FC:C9:2F:57       BluJax
> > > > ^C
> > > > # l2ping F4:4E:FC:C9:2F:57
> > > > Ping: F4:4E:FC:C9:2F:57 from 00:18:30:49:7D:63 (data size 44) ...
> > > > 44 bytes from F4:4E:FC:C9:2F:57 id 0 time 6.90ms
> > > > ...
> > > > 44 bytes from F4:4E:FC:C9:2F:57 id 14 time 28.29ms
> > > > ^C15 sent, 15 received, 0% loss
> > > > #
> > > >
> > > > 0% loss and regular, repeatable communication without any Frame
> > > > reassembly errors.
> > > >
> > >
> > > I tried disabling SERIAL_OMAP and using only SERIAL_8250_OMAP, but
> > > that didn't help.  Because the issue goes away when I disable
> > > SERIAL_8250_OMAP, I am wondering if something is either being
> > > misconfigured or some IRQ or DMA integration is missing that may be
> > > present with the older SERIAL_OMAP driver.
> > >
> > > > Any suggestions on how to troubleshoot or what might cause the
> > > > difference between the two drivers?
> >
> > Can it be related to this issue [1]? Can you confirm that 5.2 is
> > working as expected with the 8250 driver?
> >
> > [1] https://marc.info/?l=linux-serial&m=156965039008649&w=2
>
> I reverted the whole 8250 directory to d99482673f95 ("serial:
> mctrl_gpio: Check if GPIO property exisits before requesting it") and
> it is somewhat better, but it's not as good as the stock OMAP serial
> driver.  I get some frame errors and eventually, I get some timeouts,
> but it's not as bad.  I'll try to implement the RTS and CTS as gpio
> pins and change the device tree accordingly.  It might shed some light
> on the situation.

I tried to manually setup RTS and CTS pins as GPIO, but that didn't
work, so I changed it back.

It looks like the clocking is correct.  I don't know enough about the
DMA or the IRQ to know if it's working correctly.

I was wondering if the problem is in the handshaking or not.
I added " uart-has-rtscts;" to by uart node thinking it might help,
but it did not.

>
8250_omap.c has some checks to see if we can enable autoRTS:

if (termios->c_cflag & CRTSCTS && up->port.flags & UPF_HARD_FLOW &&
    !up->gpios) {
     /* Enable AUTOCTS (autoRTS is enabled when RTS is raised) */
     ...
}

Based on this, I would expect up->gpios to always be zero if we want
auto RTS CTS.

I threw some debug code into the serial driver to look at the status
of the various flags that go into setting up auto RTS/CTS.

[   13.837005] termios->c_cflag & CRTSCTS = 80000000
[   13.841888] up->port.flags & UPF_HARD_FLOW = 300000
[   13.846801] up->gpios = ce3f3cc0
[   17.166595] termios->c_cflag & CRTSCTS = 0
[   17.170745] up->port.flags & UPF_HARD_FLOW = 300000
[   17.175781] up->gpios = ce3f3d40

I don't know which port is which, but I only have two active uarts
(console and Bluetooth)
The Bluetooth port should use handshaking and the console should not.

>From the look of the dump, up->gpios is never 0, so the last
evaluation would always be false unless I am misinterpreting
something.

I tried to modify the check to remove !up->gpios check, but that
didn't help either, but it made it a little better.


> adam
> >
> > Cheers,
> > Yegor