diff mbox

earlycon issues in -next with amba-pl011 updates

Message ID 55C94247.8020701@hurleysoftware.com (mailing list archive)
State New, archived
Headers show

Commit Message

Peter Hurley Aug. 11, 2015, 12:31 a.m. UTC
Hi Leif,

On 08/10/2015 07:23 PM, Leif Lindholm wrote:
> Hi all,
> 
> The kernelci.org bot picked up a complete boot failure (no output past
> UEFI stub) with next-20150806 and Tyler bisected it down to somewhere
> in
> 8cd90e5 uart: pl011: Add support to ZTE ZX296702 uart
> 09dcc7d uart: pl011: Improve LCRH register access decision
> 2c096a9 uart: pl011: Introduce register look up table
> 7b753f3 uart: pl011: Introduce register accessor
> 
> The issue only appears with earlycon on command line, for pl011
> consoles.
> 
> Some investigation shows that the cause lies with
> commit 7b753f318d14 ("uart: pl011: Introduce register accessor")
> and
> commit 2c096a9eedc6 ("uart: pl011: Introduce register look up table")
> 
> Specifically, the changes to pl011_putc() are incorrect:
> The new pl011_ accessors take a (struct uart_amba_port *) input, but
> pl011_putc() directly uses the incoming (struct uart_port *) for this.
> 
> Apart from ending up with an unintended/incorrect UART base address,
> the introduction of the lookup table for register offsets also means
> the accessors try to dereference (struct uart_amba_port *)->reg_lut.


Thanks for the bug report and bisect, and apologies for the breakage;
I should have caught that on review.

Would you please test the patch below and see if that resolves the
problem for you?

Regards,
Peter Hurley

PS - _NOT_ even compile-tested, sorry.

--- >% ---
Subject: [PATCH] serial: pl011: Fix earlycon register LUT breakage

Commit 7b753f318d14 ("uart: pl011: Introduce register accessor")
mistakenly used the register LUT i/o accessors for the pl011
earlycon. Since the port has not been probed at earlycon time,
the struct uart_amba_port (and register LUTs) are uninitialized.

Use direct register addressing for pl011 earlycon; other h/w supported
by the amba-pl011 driver should declare an alternate earlycon.

Cc: Jun Nie <jun.nie@linaro.org>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
---
 drivers/tty/serial/amba-pl011.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

Comments

Leif Lindholm Aug. 11, 2015, 10:07 a.m. UTC | #1
Hi Peter,

On Mon, Aug 10, 2015 at 08:31:03PM -0400, Peter Hurley wrote:
> > The issue only appears with earlycon on command line, for pl011
> > consoles.
> > 
> > Some investigation shows that the cause lies with
> > commit 7b753f318d14 ("uart: pl011: Introduce register accessor")
> > and
> > commit 2c096a9eedc6 ("uart: pl011: Introduce register look up table")
> > 
> > Specifically, the changes to pl011_putc() are incorrect:
> > The new pl011_ accessors take a (struct uart_amba_port *) input, but
> > pl011_putc() directly uses the incoming (struct uart_port *) for this.
> > 
> > Apart from ending up with an unintended/incorrect UART base address,
> > the introduction of the lookup table for register offsets also means
> > the accessors try to dereference (struct uart_amba_port *)->reg_lut.
> 
> 
> Thanks for the bug report and bisect, and apologies for the breakage;
> I should have caught that on review.

I wouldn't have found it if I didn't have the boot failure to track
down, and I'm _not_ going to tell you how long I spent doing that.
 
> Would you please test the patch below and see if that resolves the
> problem for you?

Yeah, that's a more straightforward fix.

> PS - _NOT_ even compile-tested, sorry.
> 
> --- >% ---
> Subject: [PATCH] serial: pl011: Fix earlycon register LUT breakage
> 
> Commit 7b753f318d14 ("uart: pl011: Introduce register accessor")
> mistakenly used the register LUT i/o accessors for the pl011
> earlycon. Since the port has not been probed at earlycon time,
> the struct uart_amba_port (and register LUTs) are uninitialized.
> 
> Use direct register addressing for pl011 earlycon; other h/w supported
> by the amba-pl011 driver should declare an alternate earlycon.
> 
> Cc: Jun Nie <jun.nie@linaro.org>
> Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
> ---
>  drivers/tty/serial/amba-pl011.c | 9 +++------
>  1 file changed, 3 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
> index 2af09ab..fd54991 100644
> --- a/drivers/tty/serial/amba-pl011.c
> +++ b/drivers/tty/serial/amba-pl011.c
> @@ -2348,13 +2348,10 @@ static struct console amba_console = {
>  
>  static void pl011_putc(struct uart_port *port, int c)
>  {
> -	struct uart_amba_port *uap =
> -	    container_of(port, struct uart_amba_port, port);
> -
> -	while (pl011_readw(uap, REG_FR) & UART01x_FR_TXFF)
> +	while (readl(port->membase + UART01x_FR) & UART01x_FR_TXFF)
>  		;
> -	pl011_writeb(uap, c, REG_DR);
> -	while (pl011_readw(uap, REG_FR) & uap->fr_busy)
> +	writeb(c, port->membase + UART01x_DR);
> +	while (readl(port->membase + UART01x_FR) & UART01x_FR_BUSY)
>  		;
>  }
>  
> -- 
> 2.5.0

It works, but builds with:
drivers/tty/serial/amba-pl011.c:329:13: warning: ‘pl011_writeb’
defined but not used [-Wunused-function]
 static void pl011_writeb(struct uart_amba_port *uap, u8 val, int
 index)
             ^
I was dithering about whether having _relaxed accessors in post boot
code and plain ones in earlycon was an issue, but I guess it doesn't
matter much.

/
    Leif
Russell King - ARM Linux Aug. 11, 2015, 12:40 p.m. UTC | #2
On Mon, Aug 10, 2015 at 08:31:03PM -0400, Peter Hurley wrote:
> diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
> index 2af09ab..fd54991 100644
> --- a/drivers/tty/serial/amba-pl011.c
> +++ b/drivers/tty/serial/amba-pl011.c
> @@ -2348,13 +2348,10 @@ static struct console amba_console = {
>  
>  static void pl011_putc(struct uart_port *port, int c)
>  {
> -	struct uart_amba_port *uap =
> -	    container_of(port, struct uart_amba_port, port);
> -
> -	while (pl011_readw(uap, REG_FR) & UART01x_FR_TXFF)
> +	while (readl(port->membase + UART01x_FR) & UART01x_FR_TXFF)
>  		;
> -	pl011_writeb(uap, c, REG_DR);
> -	while (pl011_readw(uap, REG_FR) & uap->fr_busy)
> +	writeb(c, port->membase + UART01x_DR);
> +	while (readl(port->membase + UART01x_FR) & UART01x_FR_BUSY)
>  		;

These loops are wrong.  Kernel coding style is that loops like this
will use cpu_relax() in them, just like the driver used to.  This
was introduced by:

commit 0d3c673e7881e691991b2a4745bd4f149603baa2
Author: Rob Herring <robh@kernel.org>
Date:   Fri Apr 18 17:19:57 2014 -0500

    tty/serial: pl011: add generic earlycon support

Also, the above readl()s should be readl_relaxed(), we don't need
barriered reads or to hit the L2 cache on those reads in the above code.
Peter Hurley Aug. 17, 2015, 4:04 p.m. UTC | #3
On 08/11/2015 08:40 AM, Russell King - ARM Linux wrote:
> On Mon, Aug 10, 2015 at 08:31:03PM -0400, Peter Hurley wrote:
>> diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
>> index 2af09ab..fd54991 100644
>> --- a/drivers/tty/serial/amba-pl011.c
>> +++ b/drivers/tty/serial/amba-pl011.c
>> @@ -2348,13 +2348,10 @@ static struct console amba_console = {
>>  
>>  static void pl011_putc(struct uart_port *port, int c)
>>  {
>> -	struct uart_amba_port *uap =
>> -	    container_of(port, struct uart_amba_port, port);
>> -
>> -	while (pl011_readw(uap, REG_FR) & UART01x_FR_TXFF)
>> +	while (readl(port->membase + UART01x_FR) & UART01x_FR_TXFF)
>>  		;
>> -	pl011_writeb(uap, c, REG_DR);
>> -	while (pl011_readw(uap, REG_FR) & uap->fr_busy)
>> +	writeb(c, port->membase + UART01x_DR);
>> +	while (readl(port->membase + UART01x_FR) & UART01x_FR_BUSY)
>>  		;
> 
> These loops are wrong.  Kernel coding style is that loops like this
> will use cpu_relax() in them, just like the driver used to.  This
> was introduced by:
> 
> commit 0d3c673e7881e691991b2a4745bd4f149603baa2
> Author: Rob Herring <robh@kernel.org>
> Date:   Fri Apr 18 17:19:57 2014 -0500
> 
>     tty/serial: pl011: add generic earlycon support
> 
> Also, the above readl()s should be readl_relaxed(), we don't need
> barriered reads or to hit the L2 cache on those reads in the above code.

This patch just reverts pl011 earlycon back to the existing mainline
state.

The amba-pl011 driver has many places where cpu_relax() is not used,
so that seems like a good patch for -next during the 4.3-rc cycle.

Similarly for read_relaxed() use in earlycon.

Regards,
Peter Hurley
diff mbox

Patch

diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
index 2af09ab..fd54991 100644
--- a/drivers/tty/serial/amba-pl011.c
+++ b/drivers/tty/serial/amba-pl011.c
@@ -2348,13 +2348,10 @@  static struct console amba_console = {
 
 static void pl011_putc(struct uart_port *port, int c)
 {
-	struct uart_amba_port *uap =
-	    container_of(port, struct uart_amba_port, port);
-
-	while (pl011_readw(uap, REG_FR) & UART01x_FR_TXFF)
+	while (readl(port->membase + UART01x_FR) & UART01x_FR_TXFF)
 		;
-	pl011_writeb(uap, c, REG_DR);
-	while (pl011_readw(uap, REG_FR) & uap->fr_busy)
+	writeb(c, port->membase + UART01x_DR);
+	while (readl(port->membase + UART01x_FR) & UART01x_FR_BUSY)
 		;
 }