Skip to contentSkip to conversation

ARM: add serial.h and set BASE_BAUD to 0

PATCH In review Rafał Miłecki 3 participants 3 messages Sep 20, 2018 – Sep 21, 2018

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: add serial.h and set BASE_BAUD to 0

From: Rafa? Mi?ecki <[email protected]>

For years arm has been using serial.h from asm-generic which sets BASE_BAUD value to the (1843200 / 16). This is incorrect as:
1) This value obviously isn't correct for all devices
2) There are no device specific serial.h with CONFIG_ARCH_MULTIPLATFORM

That value breaks early serial console e.g. on Broadcom Northstar boards (CONFIG_ARCH_BCM_5301X) which need BASE_BAUD = (99532800 / 16).

The problem remained unnoticed until 4.14 and the commit 31cb9a8575ca0 ("earlycon: initialise baud field of earlycon device structure").

Above commit started filling "baud" in the "struct earlycon_device" (it seems to be a correct behavior) and changed behavior of the 8250/16550 early console driver. Starting with 4.14 kernel's 8250_early.c sets UART divisor if both: "uartclk" and "baud" are set. Having wrong value for BASE_BAUD (and so "uartclk") results in broken console setup.

Solution to that has been introduced in commit 0ff3ab701963 ("serial:
8250_early: Only set divisor if valid clk & baud") and requires setting BASE_BAUD (and so "uartclk") to 0.

A change similar to this action has been pushed for MIPS in the commit c8ec2041f549 ("MIPS: Add custom serial.h with BASE_BAUD override for generic kernel"). Another platform using 0 for BASE_BAUD is s390.

Fixes 31cb9a8575ca ("earlycon: initialise baud field of earlycon device structure")
Cc stable at vger.kernel.org # v4.14+

---
This problem has been originally reported by me in the e-mail:
ns16550 earlycon baud broken on BCM5301X since commit 31cb9a8575ca0 ("earlycon: initialise baud field of earlycon device structure")
https://www.spinics.net/lists/arm-kernel/msg644295.html

---
 arch/arm/include/asm/serial.h | 8 ++++++++
 1 file changed, 8 insertions(+)
 create mode 100644 arch/arm/include/asm/serial.h
arch/arm/include/asm/serial.h new file +8 -0
new file mode 100644
index 000000000000..e12f262290ad
@@ -0,0 +1,8 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_ARM_SERIAL_H
+#define _ASM_ARM_SERIAL_H
+
+/* ARM kernels support multiple devices so there isn't a single valid value. */
+#define BASE_BAUD 0
+
+#endif /* _ASM_ARM_SERIAL_H */
Signature · 1 line
2.13.7
#1 lore raw
Quoted 6 lines · Rafa? Mi?ecki

From: Rafa? Mi?ecki <[email protected]>

For years arm has been using serial.h from asm-generic which sets BASE_BAUD value to the (1843200 / 16). This is incorrect as:
1) This value obviously isn't correct for all devices
2) There are no device specific serial.h with CONFIG_ARCH_MULTIPLATFORM

This goes back a long way. The BASE_BAUD definition is the base baud for 8250 compatible UARTs _only_, no others.

However, as of 182ead3e418a ("earlycon: Remove hardcoded port->uartclk initialization in of_setup_earlycon"), port->uartclk is no longer initialised using BASE_BAUD. As acknowledged in 814453adea7d ("earlycon: Initialize port->uartclk based on clock-frequency property") the initialisation using BASE_BAUD was bogus, and there is now the clock-frequency DT property which should be present to set this up.

Now, setting BASE_BAUD to zero as per your patch will break the 8250 serial driver - this relies on BASE_BAUD being set to the current value, and yes, ARM hardware that uses this will break. So this is not a solution.

The only solution is that BASE_BAUD must not be abused - it must not be used by non-8250 hardware, and thankfully there are already solutions in the kernel (such as clock-frequency) to allow the clock rate to be specified.

Signature · 3 lines
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 13.8Mbps down 630kbps up
According to speedtest.net: 13Mbps down 490kbps up
#2 lore raw
Quoted 12 lines · Russell King - ARM Linux
Quoted 7 lines · Rafa? Mi?ecki

From: Rafa? Mi?ecki <[email protected]>

For years arm has been using serial.h from asm-generic which sets BASE_BAUD value to the (1843200 / 16). This is incorrect as:
1) This value obviously isn't correct for all devices
2) There are no device specific serial.h with CONFIG_ARCH_MULTIPLATFORM

This goes back a long way. The BASE_BAUD definition is the base baud for 8250 compatible UARTs _only_, no others.

Quoted 7 lines

However, as of 182ead3e418a ("earlycon: Remove hardcoded port->uartclk initialization in of_setup_earlycon"), port->uartclk is no longer initialised using BASE_BAUD. As acknowledged in 814453adea7d ("earlycon: Initialize port->uartclk based on clock-frequency property")
the initialisation using BASE_BAUD was bogus, and there is now the clock-frequency DT property which should be present to set this up.

Thanks a lot for looking at that. I was not aware of the commit 182ead3e418a ("earlycon: Remove hardcoded port->uartclk initialization in of_setup_earlycon"), too bad it wasn't marked for stable.

I simply assumed that
port->uartclk = BASE_BAUD * 16;
is correct and that has poisoned all my further reasoning.

I've just confirmed that backporting 182ead3e418a to the 4.14 fixes the problem for me (it results in the same 8250_early.c behavior as setting BASE_BAUD to 0).

I'm planning to backport both:
182ead3e418a ("earlycon: Remove hardcoded port->uartclk initialization in of_setup_earlycon")
814453adea7d ("earlycon: Initialize port->uartclk based on clock-frequency property")
to 4.14+ as they fix regression triggered (not to say "caused") by the commit 31cb9a8575ca0 ("earlycon: initialise baud field of earlycon device structure").

Please DROP my serial.h patch and thanks again.

Quoted 9 lines

Now, setting BASE_BAUD to zero as per your patch will break the 8250 serial driver - this relies on BASE_BAUD being set to the current value, and yes, ARM hardware that uses this will break. So this is not a solution.

The only solution is that BASE_BAUD must not be abused - it must not be used by non-8250 hardware, and thankfully there are already solutions in the kernel (such as clock-frequency) to allow the clock rate to be specified.

Somehow my serial console kept working with BASE_BAUD set to 0. I'm not sure why.