mxs auart timeout waiting for transmission with hw flow control

DISCUSSION Hector Palacios 2 participants 4 messages Sep 30, 2013 – Oct 1, 2013 TTY layer and serial drivers

Share & tools

Hector Palacios

author

2013-09-30 16:15 UTC #0 lore raw

Hi Uwe,

I saw your patch at http://www.spinics.net/lists/stable/msg12952.html and I believe I'm having a similar issue when not using the port as console.

With hardware flow control enabled transmission fails at low baudrates (9600,38400). Apparently the transmitter closes the port before the data has been really shifted out (or even transferred by the DMA). While the console functions check the status of AUART_STAT_BUSY, the standard functions don't, and the AUART is busy when shutdown is called at low baudrates.

I also see the function uart_wait_until_sent() of serial_core.c getting out with timeout expired. At 9600 baud, this function waits 2 or 4 jiffies. Increasing (a lot) the timeout of this function solves the problem. I monitored the time it takes for TX to be empty and it resulted in sometimes around 200 jiffies (@9600).

The issue can be reproduced using two iMX28 devices (transmitting between two ports on the same platform usually works fine), configuring a low baudrate (9600) and hardware flow enabled, and sending a file of 40K, for example. The last ~200 bytes or so are typically not sent over the line.

Best regards,
--
Hector Palacios

Uwe Kleine-König

2013-09-30 16:20 UTC #1 lore raw

Hi Hector,

Quoted 6 lines · Hector Palacios

I saw your patch at
http://www.spinics.net/lists/stable/msg12952.html and I believe I'm having a similar issue when not using the port as console.
With hardware flow control enabled transmission fails at low baudrates (9600,38400). Apparently the transmitter closes the port

Which kernel are you using? I think there were some other fixes concering flow control some time ago.

Quoted 10 lines

before the data has been really shifted out (or even transferred by the DMA). While the console functions check the status of AUART_STAT_BUSY, the standard functions don't, and the AUART is busy when shutdown is called at low baudrates.
I also see the function uart_wait_until_sent() of serial_core.c getting out with timeout expired. At 9600 baud, this function waits 2 or 4 jiffies. Increasing (a lot) the timeout of this function solves the problem. I monitored the time it takes for TX to be empty and it resulted in sometimes around 200 jiffies (@9600).

Again with flow control and another imx28 on the opposite side? Does it work without flow control? Is CTS busy for some time? In that case a simple timeout isn't appropriate. Using console and flow control on the same port might be hairy, not sure though.

Quoted 5 lines

The issue can be reproduced using two iMX28 devices (transmitting between two ports on the same platform usually works fine), configuring a low baudrate (9600) and hardware flow enabled, and sending a file of 40K, for example. The last ~200 bytes or so are typically not sent over the line.

I currently don't have the time and machines to test, but maybe the freescale guys (added Fabio to Cc, too) might want to test it.

Best regards from Freiburg
Uwe

Signature · 6 lines

Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
--
To unsubscribe from this list: send the line "unsubscribe linux-serial" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Hector Palacios

author

2013-09-30 16:48 UTC #2 lore raw

Hi Uwe,

Quoted 11 lines · Uwe Kleine-König

Hi Hector,
Quoted 6 lines · Hector Palacios
I saw your patch at
http://www.spinics.net/lists/stable/msg12952.html and I believe I'm having a similar issue when not using the port as console.
With hardware flow control enabled transmission fails at low baudrates (9600,38400). Apparently the transmitter closes the port
Which kernel are you using? I think there were some other fixes concering flow control some time ago.

v3.10. The driver is pretty close to current status.

Quoted 10 lines

Quoted 10 lines
before the data has been really shifted out (or even transferred by the DMA). While the console functions check the status of AUART_STAT_BUSY, the standard functions don't, and the AUART is busy when shutdown is called at low baudrates.
I also see the function uart_wait_until_sent() of serial_core.c getting out with timeout expired. At 9600 baud, this function waits 2 or 4 jiffies. Increasing (a lot) the timeout of this function solves the problem. I monitored the time it takes for TX to be empty and it resulted in sometimes around 200 jiffies (@9600).

Quoted 4 lines

Again with flow control and another imx28 on the opposite side? Does it work without flow control? Is CTS busy for some time? In that case a simple timeout isn't appropriate. Using console and flow control on the same port might be hairy, not sure though.

Yes, two imx28 devices with hw flow control enabled. Without flow control it works. CTS is not busy. At these low baudrates the CPU is fast enough to process the received data in time. I'm not using the AUART as console but as a standard port.

Quoted 9 lines

Quoted 5 lines
The issue can be reproduced using two iMX28 devices (transmitting between two ports on the same platform usually works fine), configuring a low baudrate (9600) and hardware flow enabled, and sending a file of 40K, for example. The last ~200 bytes or so are typically not sent over the line.
I currently don't have the time and machines to test, but maybe the freescale guys (added Fabio to Cc, too) might want to test it.
Best regards from Freiburg

I wish I could drop by Freiburg one of these days! Love the city (and the beer).

Best regards,
--
Hector Palacios
--
To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Hector Palacios

author

2013-10-01 15:37 UTC #3 lore raw

Hello,

Quoted 10 lines · Hector Palacios

Hi Uwe,
Quoted 7 lines · Uwe Kleine-König
Hi Hector,
Quoted 4 lines · Hector Palacios
I saw your patch at
http://www.spinics.net/lists/stable/msg12952.html and I believe I'm having a similar issue when not using the port as console.

Quoted 17 lines

Quoted 4 lines
Quoted 2 lines
With hardware flow control enabled transmission fails at low baudrates (9600,38400). Apparently the transmitter closes the port
Which kernel are you using? I think there were some other fixes concering flow control some time ago.
v3.10. The driver is pretty close to current status.
Quoted 10 lines
Quoted 10 lines
before the data has been really shifted out (or even transferred by the DMA). While the console functions check the status of AUART_STAT_BUSY, the standard functions don't, and the AUART is busy when shutdown is called at low baudrates.
I also see the function uart_wait_until_sent() of serial_core.c getting out with timeout expired. At 9600 baud, this function waits 2 or 4 jiffies. Increasing (a lot) the timeout of this function solves the problem. I monitored the time it takes for TX to be empty and it resulted in sometimes around 200 jiffies (@9600).

The problem is that the shutdown callback does not wait for the DMA buffer to flush. I will submit a patch that fixes it.

Quoted 6 lines

Quoted 4 lines
Again with flow control and another imx28 on the opposite side? Does it work without flow control? Is CTS busy for some time? In that case a simple timeout isn't appropriate. Using console and flow control on the same port might be hairy, not sure though.
Yes, two imx28 devices with hw flow control enabled. Without flow control it works.

The reason why it works without hardware flow control is that in this driver the DMA is only activated when hardware flow control is enabled.

By the way, where did the spinlocks go in this driver?
Ocasionally I saw unhandled NULL pointer exception in uart_write_wakeup(). Maybe my patch will make it even hardware to hit, but still...