Re: [PATCH] x86/microcode: Fix crashes on early 486 CPUs due to usage of 'cpuid'.

[Kevin Koster <lkml@ombertech.com>]

On Sun, 6 Apr 2025 21:02:53 +0200
Borislav Petkov <bp@alien8.de> wrote:

> On Sun, Apr 06, 2025 at 05:46:33PM +1000, Kevin Koster wrote:
> > But if I move "cpuid_1_eax = native_cpuid_eax(1);" to after the
> > microcode_loader_disabled() check in amd.c, like in the first
> > patch, it boots fine!
> 
> Yeah, I noticed that too last night. Here's a hopefully better
> version after I ran it on my 32-bit Atom - I don't have your old rust

There is the 86box emulator which Eric Voirin suggested before for
reproducing this issue, although that adds the uncertainty of whether
the emulator matches real hardware behaviour.

> and maybe you should simply throw it in the garbage - that thing is
> probably not worth the electricity it uses to power up... :-)

Well my testing modern Linux on 486s was originally prompted by people
on the Tiny Core Linux forums finding compatibility issues with old
PCs. But I like to know Linux really works on the hardware it's built
for, and I'm not much better, writing this now on a Pentium 1.

[snip]
> diff --git a/arch/x86/kernel/cpu/microcode/amd.c
> b/arch/x86/kernel/cpu/microcode/amd.c index
> b61028cf5c8a..dda7f0d409e9 100644
> --- a/arch/x86/kernel/cpu/microcode/amd.c
> +++ b/arch/x86/kernel/cpu/microcode/amd.c
> @@ -1099,7 +1099,7 @@ static int __init save_microcode_in_initrd(void)
>  	enum ucode_state ret;
>  	struct cpio_data cp;
>  
> -	if (dis_ucode_ldr || c->x86_vendor != X86_VENDOR_AMD || c->x86 < 0x10)
> +	if (microcode_loader_disabled() || c->x86_vendor != X86_VENDOR_AMD || c->x86 < 0x10)
		return 0;

Still fails unless the native_cpuid_eax(1) call is moved under here. After that
change, it boots fine.