2018-07-11 19:56:46

by Alexandre Belloni

[permalink] [raw]
Subject: [PATCH] ARM: early_printk: respect keep parameter

Parse and use the keep parameter of earlyprintk to avoid having the kernel
output stop with:

Console: colour dummy device 80x30
console [tty0] enabled
bootconsole [earlycon0] disabled

This happens because once tty0, the virtual terminal console is registered,
it is selected as the default console, disregarding an existing stdout-path
DT property and the early console is shutdown before the console driver is
probed.

Signed-off-by: Alexandre Belloni <[email protected]>
---
arch/arm/kernel/early_printk.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/arch/arm/kernel/early_printk.c b/arch/arm/kernel/early_printk.c
index 9257736ec9fa..bbb10150b11a 100644
--- a/arch/arm/kernel/early_printk.c
+++ b/arch/arm/kernel/early_printk.c
@@ -42,6 +42,9 @@ static struct console early_console_dev = {

static int __init setup_early_printk(char *buf)
{
+ if (buf && strstr(buf, "keep"))
+ early_console_dev.flags &= ~CON_BOOT;
+
early_console = &early_console_dev;
register_console(&early_console_dev);
return 0;
--
2.18.0



2018-07-17 20:18:30

by Nicolas Pitre

[permalink] [raw]
Subject: Re: [PATCH] ARM: early_printk: respect keep parameter

On Wed, 11 Jul 2018, Alexandre Belloni wrote:

> Parse and use the keep parameter of earlyprintk to avoid having the kernel
> output stop with:
>
> Console: colour dummy device 80x30
> console [tty0] enabled
> bootconsole [earlycon0] disabled
>
> This happens because once tty0, the virtual terminal console is registered,
> it is selected as the default console, disregarding an existing stdout-path
> DT property and the early console is shutdown before the console driver is
> probed.
>
> Signed-off-by: Alexandre Belloni <[email protected]>

Why can't you have a real console driver that takes over at that point
which respects whatever DT property you give it?

The early_printk code is very raw and inefficient, and it shouldn't be
used for anything but the debugging of early boot hangs.

> ---
> arch/arm/kernel/early_printk.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/arch/arm/kernel/early_printk.c b/arch/arm/kernel/early_printk.c
> index 9257736ec9fa..bbb10150b11a 100644
> --- a/arch/arm/kernel/early_printk.c
> +++ b/arch/arm/kernel/early_printk.c
> @@ -42,6 +42,9 @@ static struct console early_console_dev = {
>
> static int __init setup_early_printk(char *buf)
> {
> + if (buf && strstr(buf, "keep"))
> + early_console_dev.flags &= ~CON_BOOT;
> +
> early_console = &early_console_dev;
> register_console(&early_console_dev);
> return 0;
> --
> 2.18.0
>
>


2018-07-17 20:33:54

by Alexandre Belloni

[permalink] [raw]
Subject: Re: [PATCH] ARM: early_printk: respect keep parameter

On 17/07/2018 16:16:44-0400, Nicolas Pitre wrote:
> On Wed, 11 Jul 2018, Alexandre Belloni wrote:
>
> > Parse and use the keep parameter of earlyprintk to avoid having the kernel
> > output stop with:
> >
> > Console: colour dummy device 80x30
> > console [tty0] enabled
> > bootconsole [earlycon0] disabled
> >
> > This happens because once tty0, the virtual terminal console is registered,
> > it is selected as the default console, disregarding an existing stdout-path
> > DT property and the early console is shutdown before the console driver is
> > probed.
> >
> > Signed-off-by: Alexandre Belloni <[email protected]>
>
> Why can't you have a real console driver that takes over at that point
> which respects whatever DT property you give it?
>

See Russell's analysis here:

https://www.spinics.net/lists/linux-serial/msg25363.html


> The early_printk code is very raw and inefficient, and it shouldn't be
> used for anything but the debugging of early boot hangs.
>

The issue I had is that he pinctrl driver wasn't probing so the real
console driver could never probe. I think this fits the use case.

Also, atmel_serial doesn't yet have earlycon support.

> > ---
> > arch/arm/kernel/early_printk.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/arch/arm/kernel/early_printk.c b/arch/arm/kernel/early_printk.c
> > index 9257736ec9fa..bbb10150b11a 100644
> > --- a/arch/arm/kernel/early_printk.c
> > +++ b/arch/arm/kernel/early_printk.c
> > @@ -42,6 +42,9 @@ static struct console early_console_dev = {
> >
> > static int __init setup_early_printk(char *buf)
> > {
> > + if (buf && strstr(buf, "keep"))
> > + early_console_dev.flags &= ~CON_BOOT;
> > +
> > early_console = &early_console_dev;
> > register_console(&early_console_dev);
> > return 0;
> > --
> > 2.18.0
> >
> >
>

--
Alexandre Belloni, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
https://bootlin.com

2018-07-17 20:51:17

by Nicolas Pitre

[permalink] [raw]
Subject: Re: [PATCH] ARM: early_printk: respect keep parameter

On Tue, 17 Jul 2018, Alexandre Belloni wrote:

> On 17/07/2018 16:16:44-0400, Nicolas Pitre wrote:
> > On Wed, 11 Jul 2018, Alexandre Belloni wrote:
> >
> > > Parse and use the keep parameter of earlyprintk to avoid having the kernel
> > > output stop with:
> > >
> > > Console: colour dummy device 80x30
> > > console [tty0] enabled
> > > bootconsole [earlycon0] disabled
> > >
> > > This happens because once tty0, the virtual terminal console is registered,
> > > it is selected as the default console, disregarding an existing stdout-path
> > > DT property and the early console is shutdown before the console driver is
> > > probed.
> > >
> > > Signed-off-by: Alexandre Belloni <[email protected]>
> >
> > Why can't you have a real console driver that takes over at that point
> > which respects whatever DT property you give it?
> >
>
> See Russell's analysis here:
>
> https://www.spinics.net/lists/linux-serial/msg25363.html

That is a good analysis, but not necessarily a good argument for this
particular patch.

> > The early_printk code is very raw and inefficient, and it shouldn't be
> > used for anything but the debugging of early boot hangs.
> >
>
> The issue I had is that he pinctrl driver wasn't probing so the real
> console driver could never probe. I think this fits the use case.

Yes, that certainly fits. This is not what your commit log message is
saying though.

> Also, atmel_serial doesn't yet have earlycon support.

Fair enough, and probably something you should mention too.

> > > ---
> > > arch/arm/kernel/early_printk.c | 3 +++
> > > 1 file changed, 3 insertions(+)
> > >
> > > diff --git a/arch/arm/kernel/early_printk.c b/arch/arm/kernel/early_printk.c
> > > index 9257736ec9fa..bbb10150b11a 100644
> > > --- a/arch/arm/kernel/early_printk.c
> > > +++ b/arch/arm/kernel/early_printk.c
> > > @@ -42,6 +42,9 @@ static struct console early_console_dev = {
> > >
> > > static int __init setup_early_printk(char *buf)
> > > {
> > > + if (buf && strstr(buf, "keep"))
> > > + early_console_dev.flags &= ~CON_BOOT;
> > > +
> > > early_console = &early_console_dev;
> > > register_console(&early_console_dev);
> > > return 0;
> > > --
> > > 2.18.0
> > >
> > >
> >
>
> --
> Alexandre Belloni, Bootlin (formerly Free Electrons)
> Embedded Linux and Kernel engineering
> https://bootlin.com
>