Re: DM3730 Bluetooth Performance differences between SERIAL_8250_OMAP vs SERIAL_OMAP

From: Adam Ford
Date: Sat Oct 05 2019 - 21:24:30 EST


On Fri, Oct 4, 2019 at 10:45 PM Adam Ford <aford173@xxxxxxxxx> wrote:
>
> On Fri, Oct 4, 2019 at 11:51 AM Adam Ford <aford173@xxxxxxxxx> wrote:
> >
> > 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.
>
> I have it working now on the 5.3.y branch, but I had to do 4 things.
>
> 1) Disable the older OMAP serial driver
> 2) Apply [1] patch
> 3) Disable 8250 DMA support
> 4) Remove checks for !up->gpios in the 8250-omap.c code.
>
> With all those changes, I am able to successfully and repeatedly scan
> and l2ping Bluetooth devices. Without any one of those, I get frame
> errors or lost packets.
>
> I wonder if we can dump the legacy omap serial driver once the 8250
> driver is working better. That would eliminate at least one of the 4.
> I am not sure what's up with the DMA, and I've already stated some
> thoughts on item 4 in a previous e-mail.
>
> I think the [1] patch is probably going to help fix some other issues
> and should probably be backported, but by itself it isn't enough to
> fix the Bluetooth problem on the OMAP3630.
>
> [1] - https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/tty/serial/8250/8250_omap.c?h=v5.4-rc1&id=627a545c6bb0c7de09208e6546f5111290477261
>
In the latest 'master' branch of 5.4-RC1 (43b815c6a8e7), I have
confirmed that by removing the DMA and forcing the checks for
!up->gpios to be true to keep the hardware flow control working, I can
repeatably make Bluetooth work.

I am not sure what's broken in the DMA, but the check for !up->gpios
is never false, so it never actually does the code inside to address
handshaking

index c68e2b3a1634..884fcba61303 100644
--- a/drivers/tty/serial/8250/8250_omap.c
+++ b/drivers/tty/serial/8250/8250_omap.c
@@ -141,7 +141,7 @@ static void omap8250_set_mctrl(struct uart_port
*port, unsigned int mctrl)

serial8250_do_set_mctrl(port, mctrl);

- if (!up->gpios) {
+ if (1 /* !up->gpios */) {
/*
* Turn off autoRTS if RTS is lowered and restore autoRTS
* setting if RTS is raised
@@ -455,8 +455,7 @@ static void omap_8250_set_termios(struct uart_port *port,
priv->efr = 0;
up->port.status &= ~(UPSTAT_AUTOCTS | UPSTAT_AUTORTS | UPSTAT_AUTOXOFF);

- if (termios->c_cflag & CRTSCTS && up->port.flags & UPF_HARD_FLOW &&
- !up->gpios) {
+ if (termios->c_cflag & CRTSCTS && up->port.flags & UPF_HARD_FLOW) {
/* Enable AUTOCTS (autoRTS is enabled when RTS is raised) */
up->port.status |= UPSTAT_AUTOCTS | UPSTAT_AUTORTS;
priv->efr |= UART_EFR_CTS;

I know this is wrong to do what I did, but whatever is setting gpios
is clearly not setting it to NULL, so I am guessing these checks here
to be something else or something needs to be able to set these to
NULL when there are no manually configured gpio pints for cts/rts.

adam

> adam
> >
> >
> > > adam
> > > >
> > > > Cheers,
> > > > Yegor