Re: Dell XPS13: MCE (Hardware Error) reported

[Borislav Petkov]

On Thu, Jan 05, 2017 at 03:28:00PM -0800, Raj, Ashok wrote:
> After looking at the code, seems like these events are logged as MCE's
> but are really picked from real lvt thermal event interrupts.  via a fake
> bank 128 for MCE_THERMAL. These are not really HW MCE's, but fake ones created 
> and logged as mcelog entries. (arch/x86/kernel/cpu/mcheck/therm_throt.c)

Right, we've done that since forever but I do think that it confuses
people. This thread case-in-point. I mean, we already scream:

	pr_crit("CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",

to dmesg, why do we have to log a fake MCE too?!

Hell, we even log an MCE when things go back to normal:

        if (old_event) {
                if (event == THERMAL_THROTTLING_EVENT)
                        pr_info("CPU%d: %s temperature/speed normal\n", this_cpu,
                                level == CORE_LEVEL ? "Core" : "Package");
                return 1;

And Alexander's log shows exactly that:

[ 6338.170924] CPU1: Core temperature above threshold, cpu clock throttled (total events = 21068)
[ 6338.170925] CPU5: Core temperature above threshold, cpu clock throttled (total events = 21068)
[ 6338.170928] CPU7: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170931] CPU4: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170932] CPU0: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170933] CPU6: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170935] CPU2: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170936] CPU3: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170937] CPU5: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170945] CPU1: Package temperature above threshold, cpu clock throttled (total events = 22842)
[ 6338.170947] mce_notify_irq: 1 callbacks suppressed
[ 6338.170948] mce: [Hardware Error]: Machine check events logged				<--- new event
[ 6338.171917] CPU1: Core temperature/speed normal
[ 6338.171918] CPU5: Core temperature/speed normal
[ 6338.171920] CPU4: Package temperature/speed normal
[ 6338.171920] CPU0: Package temperature/speed normal
[ 6338.171922] CPU2: Package temperature/speed normal
[ 6338.171923] CPU6: Package temperature/speed normal
[ 6338.171924] CPU3: Package temperature/speed normal
[ 6338.171925] CPU7: Package temperature/speed normal
[ 6338.171927] CPU5: Package temperature/speed normal
[ 6338.171929] CPU1: Package temperature/speed normal
[ 6338.171930] mce: [Hardware Error]: Machine check events logged				<--- old event

Oh, and it's not like the user can do anything - there's a thermald
which is supposed to deal with all that. Which is not really
trouble-free too, TBH. What happens if that thing dies? Fried CPU?

So I say we should rip out that mce_log_therm_throt_event() and never
ever handle thermal events with MCEs. It is a bad idea.

-- 
Regards/Gruss,
    Boris.

SUSE Linux GmbH, GF: Felix ImendÃrffer, Jane Smithard, Graham Norton, HRB 21284 (AG NÃrnberg)
--