ucode: use wrmsr_early_safe() for early CPU microcode update

[kib]

Am I right that this version would avoid loading ucode on AP is BSP early load failed?

sys/x86/x86/ucode.c

108

133

Why not use wrmsr_early_safe() there as well?

[kib]

sys/x86/x86/ucode.c

280

This should be only used for UEFI boot ATM if (efi_boot) {...}. If really wanted, I can make it work with bios boot as well, by reusing system gdt.

[glebius]

In D52643#1202525, @kib wrote:

Am I right that this version would avoid loading ucode on AP is BSP early load failed?

Yes, if attempt on BSP sets ucode_error to LOAD_FAILED, then ucode_load_ap() will skip.

sys/x86/x86/ucode.c

133

I don't want to touch Intel with this patch, cause I can't test it.

[glebius]

sys/x86/x86/ucode.c

280

Yes, I kept this in mind, but forgot. I'd probably not change it here, rather in ucode_load_bsp() I will make the function argument efi_boot ? EARLY : UNSAFE.

[glebius]

• Fix typo in printf.
• ucode_bsp_load() sets how with efi_boot ? EARLY : UNSAFE.

[kib]

sys/x86/x86/ucode.c

108

Please fix this.

[glebius]

Make it compilable on i386.

[kib]

sys/x86/x86/ucode.c

133

So it will be tested in wild.

450

efi_boot check is no longer neccessary

[stas]

sys/x86/x86/ucode.c

133

I'm with kib@ on this. Let's do the same for Intel and get someone to test (although I don't see why this would not work?)

[glebius]

• Update Intel
• Remove efi_boot check