[PATCH/RFC] tty: serial_core: Move uart_console() check after console registration
diff mbox

Message ID 1475163064-20399-1-git-send-email-geert+renesas@glider.be
State Under Review
Delegated to: Geert Uytterhoeven
Headers show

Commit Message

Geert Uytterhoeven Sept. 29, 2016, 3:31 p.m. UTC
The port->console flag is always false, as uart_console() is called
before the serial console has been registered.

Hence for a serial port used as the console, uart_tty_port_shutdown()
will still be called when userspace closes the port, powering it down.
This will lead to a system lock up when the serial console driver writes
to the serial port's registers.

To fix this, move the setting of port->console after the call to
uart_configure_port(), which registers the serial console.

Fixes: 761ed4a94582ab29 ("tty: serial_core: convert uart_close to use tty_port_close")
Reported-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Reported-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
RFC because of the comment "If this port is a console, then the spinlock
is already initialised", and the pre-existing code calling
uart_console() before uart_configure_port().

Can be reproduced on Renesas boards:
  1. With systemd: hangs during boot,
  2. Without systemd: "telinit n" to switch to a runlevel that doesn't
     run a getty on ttySC0, and trigger kernel output.
---
 drivers/tty/serial/serial_core.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Comments

Rob Herring Oct. 3, 2016, 6:25 p.m. UTC | #1
On Thu, Sep 29, 2016 at 10:31 AM, Geert Uytterhoeven
<geert+renesas@glider.be> wrote:
> The port->console flag is always false, as uart_console() is called
> before the serial console has been registered.

I'm not seeing how that is. Everything uart_console() depends on
(port->cons, port->cons->index, and port->line) should be set already.
Maybe you pass in -1 for index which gets changed to 0 is the only
thing I can see. Is doing that valid if you have multiple ports?

> Hence for a serial port used as the console, uart_tty_port_shutdown()
> will still be called when userspace closes the port, powering it down.
> This will lead to a system lock up when the serial console driver writes
> to the serial port's registers.
>
> To fix this, move the setting of port->console after the call to
> uart_configure_port(), which registers the serial console.
>
> Fixes: 761ed4a94582ab29 ("tty: serial_core: convert uart_close to use tty_port_close")
> Reported-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
> Reported-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
> RFC because of the comment "If this port is a console, then the spinlock
> is already initialised", and the pre-existing code calling
> uart_console() before uart_configure_port().

So the spinlock is initialized twice which is probably harmless? If
the spinlock was initialized by the console, then the index would
probably not be -1.

Rob
Geert Uytterhoeven Oct. 3, 2016, 7:42 p.m. UTC | #2
Hi Rob,

On Mon, Oct 3, 2016 at 8:25 PM, Rob Herring <robh@kernel.org> wrote:
> On Thu, Sep 29, 2016 at 10:31 AM, Geert Uytterhoeven
> <geert+renesas@glider.be> wrote:
>> The port->console flag is always false, as uart_console() is called
>> before the serial console has been registered.
>
> I'm not seeing how that is. Everything uart_console() depends on
> (port->cons, port->cons->index, and port->line) should be set already.
> Maybe you pass in -1 for index which gets changed to 0 is the only
> thing I can see. Is doing that valid if you have multiple ports?

Isn't .index always initialized to -1, and set to the actual line number when
the console is registered? That's how I remember it from when
register_console() was introduced (in 2.1)...

>> Hence for a serial port used as the console, uart_tty_port_shutdown()
>> will still be called when userspace closes the port, powering it down.
>> This will lead to a system lock up when the serial console driver writes
>> to the serial port's registers.
>>
>> To fix this, move the setting of port->console after the call to
>> uart_configure_port(), which registers the serial console.
>>
>> Fixes: 761ed4a94582ab29 ("tty: serial_core: convert uart_close to use tty_port_close")
>> Reported-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
>> Reported-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
>> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
>> ---
>> RFC because of the comment "If this port is a console, then the spinlock
>> is already initialised", and the pre-existing code calling
>> uart_console() before uart_configure_port().
>
> So the spinlock is initialized twice which is probably harmless? If
> the spinlock was initialized by the console, then the index would
> probably not be -1.

Shouldn't a double initialization cause a warning, with DEBUG_SPINLOCK=y
(which I have enabled)?

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
Rob Herring Oct. 3, 2016, 8:30 p.m. UTC | #3
On Mon, Oct 3, 2016 at 2:42 PM, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> Hi Rob,
>
> On Mon, Oct 3, 2016 at 8:25 PM, Rob Herring <robh@kernel.org> wrote:
>> On Thu, Sep 29, 2016 at 10:31 AM, Geert Uytterhoeven
>> <geert+renesas@glider.be> wrote:
>>> The port->console flag is always false, as uart_console() is called
>>> before the serial console has been registered.
>>
>> I'm not seeing how that is. Everything uart_console() depends on
>> (port->cons, port->cons->index, and port->line) should be set already.
>> Maybe you pass in -1 for index which gets changed to 0 is the only
>> thing I can see. Is doing that valid if you have multiple ports?
>
> Isn't .index always initialized to -1, and set to the actual line number when
> the console is registered? That's how I remember it from when
> register_console() was introduced (in 2.1)...

Okay, the index is also set and gets marked enabled when specified on
the command line (or DT).

>>> Hence for a serial port used as the console, uart_tty_port_shutdown()
>>> will still be called when userspace closes the port, powering it down.
>>> This will lead to a system lock up when the serial console driver writes
>>> to the serial port's registers.
>>>
>>> To fix this, move the setting of port->console after the call to
>>> uart_configure_port(), which registers the serial console.
>>>
>>> Fixes: 761ed4a94582ab29 ("tty: serial_core: convert uart_close to use tty_port_close")
>>> Reported-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
>>> Reported-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
>>> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
>>> ---
>>> RFC because of the comment "If this port is a console, then the spinlock
>>> is already initialised", and the pre-existing code calling
>>> uart_console() before uart_configure_port().
>>
>> So the spinlock is initialized twice which is probably harmless? If
>> the spinlock was initialized by the console, then the index would
>> probably not be -1.
>
> Shouldn't a double initialization cause a warning, with DEBUG_SPINLOCK=y
> (which I have enabled)?

I though only missing init caused warnings.

Anyway, we can't really re-order things as uart_configure_port uses
the spinlock. This has existed this way for some time and is all an
independent of my change and your fix. So for your fix:

Acked-by: Rob Herring <robh@kernel.org>

Rob
Geert Uytterhoeven Oct. 5, 2016, 7:50 a.m. UTC | #4
Hi Rob,

On Mon, Oct 3, 2016 at 10:30 PM, Rob Herring <robh@kernel.org> wrote:
> On Mon, Oct 3, 2016 at 2:42 PM, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
>> On Mon, Oct 3, 2016 at 8:25 PM, Rob Herring <robh@kernel.org> wrote:
>>> On Thu, Sep 29, 2016 at 10:31 AM, Geert Uytterhoeven
>>> <geert+renesas@glider.be> wrote:
>>>> The port->console flag is always false, as uart_console() is called
>>>> before the serial console has been registered.
>>>
>>> I'm not seeing how that is. Everything uart_console() depends on
>>> (port->cons, port->cons->index, and port->line) should be set already.
>>> Maybe you pass in -1 for index which gets changed to 0 is the only
>>> thing I can see. Is doing that valid if you have multiple ports?
>>
>> Isn't .index always initialized to -1, and set to the actual line number when
>> the console is registered? That's how I remember it from when
>> register_console() was introduced (in 2.1)...
>
> Okay, the index is also set and gets marked enabled when specified on
> the command line (or DT).

I'm using DT and stdout-path. Where is the index set?

>>>> Hence for a serial port used as the console, uart_tty_port_shutdown()
>>>> will still be called when userspace closes the port, powering it down.
>>>> This will lead to a system lock up when the serial console driver writes
>>>> to the serial port's registers.
>>>>
>>>> To fix this, move the setting of port->console after the call to
>>>> uart_configure_port(), which registers the serial console.
>>>>
>>>> Fixes: 761ed4a94582ab29 ("tty: serial_core: convert uart_close to use tty_port_close")
>>>> Reported-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
>>>> Reported-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
>>>> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
>>>> ---
>>>> RFC because of the comment "If this port is a console, then the spinlock
>>>> is already initialised", and the pre-existing code calling
>>>> uart_console() before uart_configure_port().
>>>
>>> So the spinlock is initialized twice which is probably harmless? If
>>> the spinlock was initialized by the console, then the index would
>>> probably not be -1.
>>
>> Shouldn't a double initialization cause a warning, with DEBUG_SPINLOCK=y
>> (which I have enabled)?
>
> I though only missing init caused warnings.
>
> Anyway, we can't really re-order things as uart_configure_port uses
> the spinlock. This has existed this way for some time and is all an
> independent of my change and your fix. So for your fix:
>
> Acked-by: Rob Herring <robh@kernel.org>

Thanks!

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Patch
diff mbox

diff --git a/drivers/tty/serial/serial_core.c b/drivers/tty/serial/serial_core.c
index 6e4f63627479db8d..ce8899c13af3d97f 100644
--- a/drivers/tty/serial/serial_core.c
+++ b/drivers/tty/serial/serial_core.c
@@ -2746,8 +2746,6 @@  int uart_add_one_port(struct uart_driver *drv, struct uart_port *uport)
 	uport->cons = drv->cons;
 	uport->minor = drv->tty_driver->minor_start + uport->line;
 
-	port->console = uart_console(uport);
-
 	/*
 	 * If this port is a console, then the spinlock is already
 	 * initialised.
@@ -2761,6 +2759,8 @@  int uart_add_one_port(struct uart_driver *drv, struct uart_port *uport)
 
 	uart_configure_port(drv, state, uport);
 
+	port->console = uart_console(uport);
+
 	num_groups = 2;
 	if (uport->attr_group)
 		num_groups++;