Re: [PATCH 2/2] x86/microcode/intel: Extend BDW late-loading with platform id and LLC check

[Jia Zhang]

å 2018/1/15 äå7:48, Henrique de Moraes Holschuh åé:
> On Mon, 15 Jan 2018, Jia Zhang wrote:
>> For more details, see erratum BDF90 in document #334165 (Intel Xeon
>> Processor E7-8800/4800 v4 Product Family Specification Update) from
>> September 2017.
> 
> For the record, this erratum may well affect some E5v4 as well.
> Anything with a LLC/core ratio >= 2.5 is potentially affected as far as
> I could tell when I took a serious look at it months ago (based only on
> crash reports and public information).
> 
> It would be safer to just blacklist by sig == 0x406f1, revision <
> 0x0b00021, and LLC/core ratio >= 2.5, ignoring platform IDs.

By the way, I have another BDW processor with 40MB LLC and 16 cores.
2.5MB (40MB/16) is safe.

Thanks,
Jia

> 
>>  	/*
>>  	 * Late loading on model 79 with microcode revision less than 0x0b000021
>> -	 * may result in a system hang. This behavior is documented in item
>> -	 * BDF90, #334165 (Intel Xeon Processor E7-8800/4800 v4 Product Family).
>> +	 * and LLC size per core bigger than 2.5MB may result in a system hang.
>> +	 * This behavior is documented in item BDF90, #334165 (Intel Xeon
>> +	 * Processor E7-8800/4800 v4 Product Family).
>>  	 */
>>  	if (c->x86 == 6 &&
>>  	    c->x86_model == INTEL_FAM6_BROADWELL_X &&
>>  	    c->x86_mask == 0x01 &&
>> +	    llc_size_per_core(c) > 2621440 &&
>> +	    c->platform_id == 0xef &&
>>  	    c->microcode < 0x0b000021) {
>>  		pr_err_once("Erratum BDF90: late loading with revision < 0x0b000021 (0x%x) disabled.\n", c->microcode);
>>  		pr_err_once("Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
> 
> The c->platform_id test looks wrong.  The processor will only have a
> single bit set, it is the microcode update that has more than a single
> bit set.
> 
> And do you really want 0xef?  That is everyhing the public available
> microcode updates can be applied to in the first place, so even a
> corrected test would be useless (it would always match) unless you
> actually expect to find never-seen-in-the-wild platform mask 0x10?
>