x86, sched: Bail out of frequency invariance if base frequency is unknown
authorGiovanni Gherdovich <ggherdovich@suse.cz>
Thu, 16 Apr 2020 05:47:42 +0000 (07:47 +0200)
committerPeter Zijlstra <peterz@infradead.org>
Wed, 22 Apr 2020 21:10:13 +0000 (23:10 +0200)
Some hypervisors such as VMWare ESXi 5.5 advertise support for
X86_FEATURE_APERFMPERF but then fill all MSR's with zeroes. In particular,
MSR_PLATFORM_INFO set to zero tricks the code that wants to know the base
clock frequency of the CPU (highest non-turbo frequency), producing a
division by zero when computing the ratio turbo_freq/base_freq necessary
for frequency invariant accounting.

It is to be noted that even if MSR_PLATFORM_INFO contained the appropriate
data, APERF and MPERF are constantly zero on ESXi 5.5, thus freq-invariance
couldn't be done in principle (not that it would make a lot of sense in a
VM anyway). The real problem is advertising X86_FEATURE_APERFMPERF. This
appears to be fixed in more recent versions: ESXi 6.7 doesn't advertise
that feature.

Fixes: 1567c3e3467c ("x86, sched: Add support for frequency invariance")
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200416054745.740-2-ggherdovich@suse.cz
arch/x86/kernel/smpboot.c

index fe3ab96..3a318ec 100644 (file)
@@ -1985,6 +1985,15 @@ static bool intel_set_max_freq_ratio(void)
        return false;
 
 out:
+       /*
+        * Some hypervisors advertise X86_FEATURE_APERFMPERF
+        * but then fill all MSR's with zeroes.
+        */
+       if (!base_freq) {
+               pr_debug("Couldn't determine cpu base frequency, necessary for scale-invariant accounting.\n");
+               return false;
+       }
+
        arch_turbo_freq_ratio = div_u64(turbo_freq * SCHED_CAPACITY_SCALE,
                                        base_freq);
        arch_set_max_freq_ratio(turbo_disabled());