[U-Boot] [BUG] display_options() fails on Raspberry Pi 3B+

[U-Boot] [BUG] display_options() fails on Raspberry Pi 3B+

[Heinrich Schuchardt]

Hello Matthias,

Tom reported:

Since patch
143256b353c4d2ad1cf5e5e63b38410188b35e49
fdt: update bcm283x device tree sources to Linux 5.1-rc6 state

the console no longer shows the initial message, e.g. "U-Boot
2020.01-rc1-00159-g73b6e6ad25-dirty". This leads to a failure in
automated testing.

To find out what is happening I applied the following code change:

diff --git a/drivers/serial/serial_bcm283x_mu.c
b/drivers/serial/serial_bcm283x_mu.c
index bd1d89ec83..395c49c60a 100644
--- a/drivers/serial/serial_bcm283x_mu.c
+++ b/drivers/serial/serial_bcm283x_mu.c
@@ -171,8 +171,11 @@ static int
bcm283x_mu_serial_ofdata_to_platdata(struct udevice *dev)
         fdt_addr_t addr;

         /* Don't spawn the device if it's not muxed */
-       if (!bcm283x_is_serial_muxed())
+       if (!bcm283x_is_serial_muxed()) {
+               sprintf((void *)0x02400000, "muxed4 false\n");
                 return -ENODEV;
+       }
+       sprintf((void *)0x02400010, "muxed5 true");

         addr = devfdt_get_addr(dev);
         if (addr == FDT_ADDR_T_NONE)
diff --git a/lib/display_options.c b/lib/display_options.c
index ec16d75e0e..9d44f49731 100644
--- a/lib/display_options.c
+++ b/lib/display_options.c
@@ -45,6 +45,7 @@ int display_options(void)

         display_options_get_banner(true, buf, sizeof(buf));
         printf("%s", buf);
+       sprintf((void *)0x02400009, "disp5options");

         return 0;
  }


And see:

02400000: 6d 75 78 65 64 34 20 66 61 64 69 73 70 35 6f 70    muxed4
fadisp5op
02400010: 6d 75 78 65 64 35 20 74 72 75 65 00 05 07 06 01    muxed5
true.....

bcm283x_mu_serial_ofdata_to_platdata() is obviously called twice.

The first call is in serial_init(). At that time the GPIOs are not
correctly set up to use the serial port. So display_options() fails.
After relocation another call call occurs after the GPIOs are set up and
this call succeeds.

Do you have an idea how to proceed?

Best regards

Heinrich

[U-Boot] [BUG] display_options() fails on Raspberry Pi 3B+

[Matthias Brugger]

Hi Heinrich,

adding Alex just in case he knows better, as he wrote the initial code.

On 06/11/2019 01:57, Heinrich Schuchardt wrote:
> Hello Matthias,
> 
> Tom reported:
> 
> Since patch
> 143256b353c4d2ad1cf5e5e63b38410188b35e49
> fdt: update bcm283x device tree sources to Linux 5.1-rc6 state
> 
> the console no longer shows the initial message, e.g. "U-Boot
> 2020.01-rc1-00159-g73b6e6ad25-dirty". This leads to a failure in
> automated testing.
> 
> To find out what is happening I applied the following code change:
> 
> diff --git a/drivers/serial/serial_bcm283x_mu.c
> b/drivers/serial/serial_bcm283x_mu.c
> index bd1d89ec83..395c49c60a 100644
> --- a/drivers/serial/serial_bcm283x_mu.c
> +++ b/drivers/serial/serial_bcm283x_mu.c
> @@ -171,8 +171,11 @@ static int
> bcm283x_mu_serial_ofdata_to_platdata(struct udevice *dev)
>         fdt_addr_t addr;
> 
>         /* Don't spawn the device if it's not muxed */
> -       if (!bcm283x_is_serial_muxed())
> +       if (!bcm283x_is_serial_muxed()) {
> +               sprintf((void *)0x02400000, "muxed4 false\n");
>                 return -ENODEV;
> +       }
> +       sprintf((void *)0x02400010, "muxed5 true");
> 
>         addr = devfdt_get_addr(dev);
>         if (addr == FDT_ADDR_T_NONE)
> diff --git a/lib/display_options.c b/lib/display_options.c
> index ec16d75e0e..9d44f49731 100644
> --- a/lib/display_options.c
> +++ b/lib/display_options.c
> @@ -45,6 +45,7 @@ int display_options(void)
> 
>         display_options_get_banner(true, buf, sizeof(buf));
>         printf("%s", buf);
> +       sprintf((void *)0x02400009, "disp5options");
> 
>         return 0;
>  }
> 
> 
> And see:
> 
> 02400000: 6d 75 78 65 64 34 20 66 61 64 69 73 70 35 6f 70    muxed4
> fadisp5op
> 02400010: 6d 75 78 65 64 35 20 74 72 75 65 00 05 07 06 01    muxed5
> true.....
> 
> bcm283x_mu_serial_ofdata_to_platdata() is obviously called twice.
> 
> The first call is in serial_init(). At that time the GPIOs are not
> correctly set up to use the serial port. So display_options() fails.
> After relocation another call call occurs after the GPIOs are set up and
> this call succeeds.
> 
> Do you have an idea how to proceed?
> 

After digging a bit, my guess is: in the mentioned devicetree commit you change
the stdout-path from serial0 (pl011) to serial1 (aux-uart).

My guess is, that the FW sets the pin mux to ALT0 which leads
bcm283x_is_serial_muxed() to fail.

I wasn't able to find where the pinctrl is initialized. Shouldn't that happen
before we call serial_init()?

Regards,
Matthias

[U-Boot] [BUG] display_options() fails on Raspberry Pi 3B+

[Matthias Brugger]

Hi Heinrich and all,

On 07/11/2019 12:44, Matthias Brugger wrote:
> Hi Heinrich,
> 
> adding Alex just in case he knows better, as he wrote the initial code.
> 
> On 06/11/2019 01:57, Heinrich Schuchardt wrote:
>> Hello Matthias,
>>
>> Tom reported:
>>
>> Since patch
>> 143256b353c4d2ad1cf5e5e63b38410188b35e49
>> fdt: update bcm283x device tree sources to Linux 5.1-rc6 state
>>
>> the console no longer shows the initial message, e.g. "U-Boot
>> 2020.01-rc1-00159-g73b6e6ad25-dirty". This leads to a failure in
>> automated testing.
>>
>> To find out what is happening I applied the following code change:
>>
>> diff --git a/drivers/serial/serial_bcm283x_mu.c
>> b/drivers/serial/serial_bcm283x_mu.c
>> index bd1d89ec83..395c49c60a 100644
>> --- a/drivers/serial/serial_bcm283x_mu.c
>> +++ b/drivers/serial/serial_bcm283x_mu.c
>> @@ -171,8 +171,11 @@ static int
>> bcm283x_mu_serial_ofdata_to_platdata(struct udevice *dev)
>>         fdt_addr_t addr;
>>
>>         /* Don't spawn the device if it's not muxed */
>> -       if (!bcm283x_is_serial_muxed())
>> +       if (!bcm283x_is_serial_muxed()) {
>> +               sprintf((void *)0x02400000, "muxed4 false\n");
>>                 return -ENODEV;
>> +       }
>> +       sprintf((void *)0x02400010, "muxed5 true");
>>
>>         addr = devfdt_get_addr(dev);
>>         if (addr == FDT_ADDR_T_NONE)
>> diff --git a/lib/display_options.c b/lib/display_options.c
>> index ec16d75e0e..9d44f49731 100644
>> --- a/lib/display_options.c
>> +++ b/lib/display_options.c
>> @@ -45,6 +45,7 @@ int display_options(void)
>>
>>         display_options_get_banner(true, buf, sizeof(buf));
>>         printf("%s", buf);
>> +       sprintf((void *)0x02400009, "disp5options");
>>
>>         return 0;
>>  }
>>
>>
>> And see:
>>
>> 02400000: 6d 75 78 65 64 34 20 66 61 64 69 73 70 35 6f 70    muxed4
>> fadisp5op
>> 02400010: 6d 75 78 65 64 35 20 74 72 75 65 00 05 07 06 01    muxed5
>> true.....
>>
>> bcm283x_mu_serial_ofdata_to_platdata() is obviously called twice.
>>
>> The first call is in serial_init(). At that time the GPIOs are not
>> correctly set up to use the serial port. So display_options() fails.
>> After relocation another call call occurs after the GPIOs are set up and
>> this call succeeds.
>>
>> Do you have an idea how to proceed?
>>
> 
> After digging a bit, my guess is: in the mentioned devicetree commit you change
> the stdout-path from serial0 (pl011) to serial1 (aux-uart).
> 
> My guess is, that the FW sets the pin mux to ALT0 which leads
> bcm283x_is_serial_muxed() to fail.
> 
> I wasn't able to find where the pinctrl is initialized. Shouldn't that happen
> before we call serial_init()?
> 

That was not correct. I found the issue. The patch deletes the reference to
bcm283x-uboot.dtsi which sets  soc, uart[01] and pinctrl nodes as dm-pre-reloc.

I'll send a patch for that.

Regards,
Matthias