Re: Re: [PATCH v2 2/2] KVM: x86: use x86_get_freq to get freq for kvmclock

[Peter Zijlstra]

On Thu, Dec 02, 2021 at 09:19:25AM +0200, Maxim Levitsky wrote:
> On Thu, 2021-12-02 at 13:26 +0800, zhenwei pi wrote:

> Note that on my Zen2 machine (3970X), aperf/mperf returns current cpu freqency,

Correct, and it computes it over a random period of history. IOW, it's a
random number generator.

> 1. It sucks that on AMD, the TSC frequency is calibrated from other 
> clocksources like PIT/HPET, since the result is not exact and varies
> from boot to boot.

CPUID.15h is supposed to tell us the actual frequency; except even Intel
find it very hard to actually put the right (or any, really) number in
there :/ Bribe your friendly AMD engineer with beers or something.

> 2. In the guest on AMD, we mark the TSC as unsynchronized always due to the code
> in unsynchronized_tsc, unless invariant tsc is used in guest cpuid,
> which is IMHO not fair to AMD as we don't do this for  Intel cpus.
> (look at unsynchronized_tsc function)

Possibly we could treat >= Zen similar to Intel there. Also that comment
there is hillarious, it talks about multi-socket and then tests
num_possible_cpus(). Clearly that code hasn't been touched in like
forever.

> 3. I wish the kernel would export the tsc frequency it found to userspace
> somewhere in /sys or /proc, as this would be very useful for userspace applications.
> Currently it can only be found in dmesg if I am not mistaken..
> I don't mind if such frequency would only be exported if the TSC is stable,
> always running, not affected by CPUfreq, etc.

Perf exposes it, it's not really convenient if you're not using perf,
but it can be found there.


---
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index 2e076a459a0c..09da2935534a 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -29,6 +29,7 @@
 #include <asm/intel-family.h>
 #include <asm/i8259.h>
 #include <asm/uv/uv.h>
+#include <asm/topology.h>
 
 unsigned int __read_mostly cpu_khz;	/* TSC clocks / usec, not used here */
 EXPORT_SYMBOL(cpu_khz);
@@ -1221,9 +1222,20 @@ int unsynchronized_tsc(void)
 	 * Intel systems are normally all synchronized.
 	 * Exceptions must mark TSC as unstable:
 	 */
-	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
+	switch (boot_cpu_data.x86_vendor) {
+	case X86_VENDOR_INTEL:
+		/* Really only Core and later */
+		break;
+
+	case X86_VENDOR_AMD:
+	case X86_VENDOR_HYGON:
+		if (boot_cpu_data.x86 >= 0x17) /* >= Zen */
+			break;
+		fallthrough;
+
+	default:
 		/* assume multi socket systems are not synchronized: */
-		if (num_possible_cpus() > 1)
+		if (topology_max_packages() > 1)
 			return 1;
 	}