Re: [PATCH] x86: Unbreak early processor microcode loading

[Daniel J Blueman <daniel@numascale.com>]

On 04/03/2015 00:38, Borislav Petkov wrote:
> On Tue, Mar 03, 2015 at 11:10:44PM +0800, Daniel J Blueman wrote:
>> The changes in 871b72dd "x86: microcode: use smp_call_function_single instead
>> of set_cpus_allowed, cleanup of synchronization logic" introduced a check
>> that prevents built-in microcode from being loaded before init starts.
>>
>> Conditionalise it on early microcode loading, so we get the expected behaviour
>> when early microcode loading is enabled, and when it is not. This has potential
>> importance as BIOSes often don't load the current microcode.
>
> ... probably because they don't have it. Which is also the main reason
> for the existence of this microcode loader btw :)
>
>>
>> Signed-off-by: Daniel J Blueman <daniel@numascale.com>
>> ---
>>   arch/x86/kernel/cpu/microcode/core.c | 2 ++
>>   1 file changed, 2 insertions(+)
>>
>> diff --git a/arch/x86/kernel/cpu/microcode/core.c b/arch/x86/kernel/cpu/microcode/core.c
>> index 36a8361..fa7f9fc 100644
>> --- a/arch/x86/kernel/cpu/microcode/core.c
>> +++ b/arch/x86/kernel/cpu/microcode/core.c
>> @@ -391,9 +391,11 @@ static enum ucode_state microcode_init_cpu(int cpu, bool refresh_fw)
>>   	if (collect_cpu_info(cpu))
>>   		return UCODE_ERROR;
>>
>> +#if !defined(CONFIG_MICROCODE_AMD_EARLY) && !defined(CONFIG_MICROCODE_INTEL_EARLY)
>>   	/* --dimm. Trigger a delayed update? */
>>   	if (system_state != SYSTEM_RUNNING)
>>   		return UCODE_NFOUND;
>> +#endif
>
> Ok, let me try to understand this correctly: where is this microcode
> built in, into the kernel?
>
> If yes, you should consider enabling the early loading
> method and build in the microcode into the initrd, see
> Documentation/x86/early-microcode.txt
>
> This is the preferred method as we're applying the microcode much
> earlier.
>
> Back to you.

Yes, it's built into the kernel with config:

CONFIG_FIRMWARE_IN_KERNEL=y
CONFIG_EXTRA_FIRMWARE="amd-ucode/microcode_amd.bin 
amd-ucode/microcode_amd_fam15h.bin"
CONFIG_EXTRA_FIRMWARE_DIR="../firmware"

That's as some customer and in-house environments we use are 
initramfs-less and some we don't have direct control over the initramfs.

I don't see why built-in microcode loading shouldn't work, so I guess 
the question is, why was that 'system_state .. RUNNING' check introduced?

If just a cleanup and loading built-in microcode early was overlooked, 
it may be reasonable to conditionalise the check like so.

Thanks,
   Daniel