Skip to contentSkip to conversation

serial: 8250: probe all 16550A variants by default

PATCH In review Vladimir Oltean 4 participants 7 messages May 25, 2020 – May 26, 2020 TTY layer and serial drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 serial: 8250: probe all 16550A variants by default

From: Vladimir Oltean <[email protected]>

On NXP T1040, the UART is typically detected as 16550A_FSL64. After said patch, it gets detected as plain 16550A and the Linux console is completely garbled and missing characters.

So clearly, introducing the SERIAL_8250_16550A_VARIANTS config option has broken many existing users because it has changed the default behavior. Restore that by adding a 'default y' to this option. Users who care about 20 ms shorter boot time can always disable it, but stop wasting many debugging hours for people who don't care all that much.

Fixes dc56ecb81a0a ("serial: 8250: Support disabling mdelay-filled probes of 16550A variants")
---
 drivers/tty/serial/8250/Kconfig | 1 +
 1 file changed, 1 insertion(+)
drivers/tty/serial/8250/Kconfig +1 -0
index af0688156dd0..89c7ecb55619 100644
@@ -63,6 +63,7 @@ config SERIAL_8250_PNP
 config SERIAL_8250_16550A_VARIANTS
 	bool "Support for variants of the 16550A serial port"
 	depends on SERIAL_8250
+	default y
 	help
 	  The 8250 driver can probe for many variants of the venerable 16550A
 	  serial port. Doing so takes additional time at boot.
Signature · 1 line
2.25.1
#1 lore raw
Quoted 11 lines · Vladimir Oltean

From: Vladimir Oltean <[email protected]>

On NXP T1040, the UART is typically detected as 16550A_FSL64. After said patch, it gets detected as plain 16550A and the Linux console is completely garbled and missing characters.

So clearly, introducing the SERIAL_8250_16550A_VARIANTS config option has broken many existing users because it has changed the default behavior. Restore that by adding a 'default y' to this option. Users who care about 20 ms shorter boot time can always disable it, but stop wasting many debugging hours for people who don't care all that much.

While this sounds like a good enough fix of the regression, the proper one seems to have a separate quirk driver which avoids auto probe and uses compatible strings or whatever that board can supply. 8250_port historically has a burden of old and buggy hardware, that's why the above mentioned patch exposed that.

That said, I have no objections, but had to note about possible approaches.

Quoted 21 lines
Fixes dc56ecb81a0a ("serial: 8250: Support disabling mdelay-filled probes of 16550A variants")
---
 drivers/tty/serial/8250/Kconfig | 1 +
 1 file changed, 1 insertion(+)
drivers/tty/serial/8250/Kconfig +1 -0
index af0688156dd0..89c7ecb55619 100644
@@ -63,6 +63,7 @@ config SERIAL_8250_PNP
 config SERIAL_8250_16550A_VARIANTS
 	bool "Support for variants of the 16550A serial port"
 	depends on SERIAL_8250
+	default y
 	help
 	  The 8250 driver can probe for many variants of the venerable 16550A
 	  serial port. Doing so takes additional time at boot.
Signature · 1 line
2.25.1
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#2 lore raw
In reply to Vladimir Oltean
Quoted 3 lines · Vladimir Oltean

On NXP T1040, the UART is typically detected as 16550A_FSL64. After said patch, it gets detected as plain 16550A and the Linux console is completely garbled and missing characters.

Interesting that there's *new* powerpc hardware that needs these variants. I based the patch on the fact that, on x86 at least, hardware using these variants hasn't been made for a long time.

In the hopes of preserving at least part of the benefit of the patch, could you please change it to default y if !X86_64?

Quoted 18 lines
 drivers/tty/serial/8250/Kconfig | 1 +
 1 file changed, 1 insertion(+)
drivers/tty/serial/8250/Kconfig +1 -0
index af0688156dd0..89c7ecb55619 100644
@@ -63,6 +63,7 @@ config SERIAL_8250_PNP
 config SERIAL_8250_16550A_VARIANTS
 	bool "Support for variants of the 16550A serial port"
 	depends on SERIAL_8250
+	default y
 	help
 	  The 8250 driver can probe for many variants of the venerable 16550A
 	  serial port. Doing so takes additional time at boot.
Signature · 1 line
2.25.1
#3 lore raw

Hi Josh,

Quoted 6 lines · Josh Triplett
Quoted 3 lines · Vladimir Oltean

On NXP T1040, the UART is typically detected as 16550A_FSL64. After said patch, it gets detected as plain 16550A and the Linux console is completely garbled and missing characters.

Quoted 7 lines

Interesting that there's *new* powerpc hardware that needs these variants. I based the patch on the fact that, on x86 at least, hardware using these variants hasn't been made for a long time.

In the hopes of preserving at least part of the benefit of the patch, could you please change it to default y if !X86_64?

Why don't you add CONFIG_SERIAL_8250_16550A_VARIANTS=n in x86_64_defconfig?

Quoted 18 lines
Quoted 16 lines
 drivers/tty/serial/8250/Kconfig | 1 +
 1 file changed, 1 insertion(+)
drivers/tty/serial/8250/Kconfig +1 -1
index af0688156dd0..89c7ecb55619 100644
@@ -63,6 +63,7 @@ config SERIAL_8250_PNP
 config SERIAL_8250_16550A_VARIANTS
      bool "Support for variants of the 16550A serial port"
      depends on SERIAL_8250
+     default y
      help
        The 8250 driver can probe for many variants of the venerable 16550A
        serial port. Doing so takes additional time at boot.
--
Quoted 2 lines

2.25.1

Thanks,
-Vladimir

#4 lore raw
Quoted 13 lines · Vladimir Oltean

Hi Josh,

Quoted 10 lines · Josh Triplett
Quoted 3 lines · Vladimir Oltean

On NXP T1040, the UART is typically detected as 16550A_FSL64. After said patch, it gets detected as plain 16550A and the Linux console is completely garbled and missing characters.

Interesting that there's *new* powerpc hardware that needs these variants. I based the patch on the fact that, on x86 at least, hardware using these variants hasn't been made for a long time.

Quoted 5 lines
Quoted 3 lines

In the hopes of preserving at least part of the benefit of the patch, could you please change it to default y if !X86_64?

Why don't you add CONFIG_SERIAL_8250_16550A_VARIANTS=n in x86_64_defconfig?

In general, it seems preferable to me when the defconfig files differ less from the defaults encoded in Kconfig.

You're proposing a change to the default; could you please include one or the other additional change in your patch to preserve the behavior on x86_64?

Quoted 21 lines
Quoted 18 lines
Quoted 16 lines
 drivers/tty/serial/8250/Kconfig | 1 +
 1 file changed, 1 insertion(+)
drivers/tty/serial/8250/Kconfig +1 -1
index af0688156dd0..89c7ecb55619 100644
@@ -63,6 +63,7 @@ config SERIAL_8250_PNP
 config SERIAL_8250_16550A_VARIANTS
      bool "Support for variants of the 16550A serial port"
      depends on SERIAL_8250
+     default y
      help
        The 8250 driver can probe for many variants of the venerable 16550A
        serial port. Doing so takes additional time at boot.
--
Quoted 2 lines

2.25.1

Thanks,
-Vladimir

#5 lore raw
In reply to Josh Triplett

This change breaks all my devices: OMAP-L138 (davinci based), LS1021A, T1040, Marvell (kirkwood2 based). Only enabling VARIANTS on all my devices fix the issue.

Quoted 7 lines · Josh Triplett
Quoted 3 lines · Vladimir Oltean

On NXP T1040, the UART is typically detected as 16550A_FSL64. After said patch, it gets detected as plain 16550A and the Linux console is completely garbled and missing characters.

Interesting that there's*new* powerpc hardware that needs these variants. I based the patch on the fact that, on x86 at least, hardware using these variants hasn't been made for a long time.

#6 lore raw
Quoted 3 lines · Maxim Kochetkov

This change breaks all my devices: OMAP-L138 (davinci based), LS1021A, T1040, Marvell (kirkwood2 based). Only enabling VARIANTS on all my devices fix the issue.

Preparing a patch right now, with the appropriate Fixes tag so it should end up on any kernel that has the original patch.

- Josh Triplett