On 26/02/2019 14:58, Jan Beulich wrote: >>>> On 26.02.19 at 14:25, <andrew.coop...@citrix.com> wrote: >> On 26/02/2019 13:12, Igor Druzhinin wrote: >> >>> @@ -323,6 +327,10 @@ static void setup_p6_watchdog(unsigned counter) >>> unsigned int evntsel; >>> >>> nmi_perfctr_msr = MSR_P6_PERFCTR(0); >>> + if ( current_cpu_data.cpuid_level >= 0xa ) >>> + nmi_p6_event_width = MASK_EXTR(cpuid_eax(0xa), >>> P6_EVENT_WIDTH_MASK); >>> + if ( !nmi_p6_event_width ) >>> + nmi_p6_event_width = P6_EVENT_WIDTH_DEFAULT; >> >> This is called on each cpu, but writes to a shared variable. This >> entire block wants to be something like: >> >> if ( !nmi_p6_event_width ) >> nmi_p6_event_width = (current_cpu_data.cpuid_level >= 0xa) ? >> MASK_EXTR(cpuid_eax(0xa), P6_EVENT_WIDTH_MASK) : P6_EVENT_WIDTH_DEFAULT; >> >> (suitably wrapped) so it gets filled only once. > > Plus the low and high capping of the value read, as requested in > reply to v1.
Isn't it the counter clipping you requested (which was implemented in v2)? What is the point in high capping? Is it safe to continue if the value read is lower than P6_EVENT_WIDTH_DEFAULT or should I panic here? Obviously, I'd need to clip the counter in that case as the rest of the code wouldn't work otherwise. Igor _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
