[Letux-kernel] LX20 prototype is booting

[H. Nikolaus Schaller]

> Am 06.05.2025 um 10:50 schrieb H. Nikolaus Schaller <hns at goldelico.com>:
> 
>> 
>> NOTE 1: git diff v6.1.28 sudomaker-6.1.28 arch/mips/kernel/cpu-probe.c is not empty!
> 
> There are three sudomaker-patches modifying this file:
> 
> https://git.goldelico.com/?p=letux-kernel.git;a=history;f=arch/mips/kernel/cpu-probe.c;h=49282a6aa8c88b7f129ffb3e7bf4163cb4be3993;hb=39077e7113a1bc29f316a786e13a57e9df92a1eb

Now, I have tried to find out why the 6.15-rc5 stalls with SYS_HAS_CPU_MIPS32_R2.
It happens at the same location as in 6.1.28-sudomaker.

So the above mentioned changes are not relevant at all.

Some more checks show boot with SYS_HAS_CPU_MIPS32_R2 hangs here:

	BUG_ON(current_cpu_type() != c->cputype);
	https://elixir.bootlin.com/linux/v6.15-rc5/source/arch/mips/kernel/cpu-probe.c#L1907

Ah, that finally was the correct hint.

There is a non-upstream patch for cpu_probe_ingenic() to set c->cputype = CPU_XBURST2:

	https://git.goldelico.com/?p=letux-kernel.git;a=commitdiff;h=a67c8db0a394fc

This also adds CPU_XBURST2 to __get_cpu_type() - but only for CONFIG_SYS_HAS_CPU_MIPS32_R5!
And if it is missing this ends in the unreachable() call in __get_cpu_type().

So we have found the reason why it does not work with CONFIG_SYS_HAS_CPU_MIPS32_R2.

But should it or not? To be compatible with elf-loading of Debian Jessie MIPSEL code?

Indeed this hack (but is it correct?) makes the sudomaker-6.1.28 kernel boot a little further:

diff --git a/arch/mips/include/asm/cpu-type.h b/arch/mips/include/asm/cpu-type.h
index a412895a8b4f1..24f26f6b271ec 100644
--- a/arch/mips/include/asm/cpu-type.h
+++ b/arch/mips/include/asm/cpu-type.h
@@ -39,6 +39,7 @@ static inline int __pure __get_cpu_type(const int cpu_type)
     defined(CONFIG_SYS_HAS_CPU_MIPS32_R2)
        case CPU_4KEC:
        case CPU_XBURST:
+       case CPU_XBURST2:
 #endif
   #ifdef CONFIG_SYS_HAS_CPU_MIPS32_R2
diff --git a/arch/mips/ingenic/Kconfig b/arch/mips/ingenic/Kconfig
index 22940d2c2a4ac..1f80372ce9a62 100644
--- a/arch/mips/ingenic/Kconfig
+++ b/arch/mips/ingenic/Kconfig
@@ -148,7 +148,7 @@ config MACH_X1830
 config MACH_X2000
        bool
        select GENERIC_CLOCKEVENTS_BROADCAST if SMP
-       select SYS_HAS_CPU_MIPS32_R5
+       select SYS_HAS_CPU_MIPS32_R2
        select SYS_SUPPORTS_HIGHMEM
        select SYS_SUPPORTS_SMP
        select MIPS_O32_FP64_SUPPORT

At least this explains why the X2000 is so special and all other devices we support
are not affected. The X2000 is the only XBurst2 and MIPS32_R5 system have.

I think I should now better focus on fixing the hardware and build some more prototypes.
There is only a major problem to be found on one of the current prototpyes - if it needs
a general fix.

And, I have assigned some I2C and GPIO pins to the headers. In a ways that it should be
useful, but certainly not optimal. A big issue is the 1.8V I/O level on some GPIO ports.
Which means I would have to add expensive bidirectional level shifters to get the RasPi
3.3V levels. But let's discuss when the next version is finished and kernel is booting...

BR,
Nikolaus