On Mon, Apr 28, 2025 at 6:26 PM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
>
> On Fri, Mar 21, 2025, Jim Mattson wrote:
> > +#include <fcntl.h>
> > +#include <limits.h>
> > +#include <pthread.h>
> > +#include <sched.h>
> > +#include <stdbool.h>
> > +#include <stdio.h>
> > +#include <stdint.h>
> > +#include <unistd.h>
> > +#include <asm/msr-index.h>
> > +
> > +#include "kvm_util.h"
> > +#include "processor.h"
> > +#include "test_util.h"
> > +
> > +#define NUM_ITERATIONS 100
> > +
> > +static void pin_thread(int cpu)
> > +{
> > + cpu_set_t cpuset;
> > + int rc;
> > +
> > + CPU_ZERO(&cpuset);
> > + CPU_SET(cpu, &cpuset);
> > +
> > + rc = pthread_setaffinity_np(pthread_self(), sizeof(cpuset), &cpuset);
> > + TEST_ASSERT(rc == 0, "%s: Can't set thread affinity", __func__);
>
> Heh, you copy-pasted this from hardware_disable_test.c, didn't you? :-)
Probably.
> Would it make sense to turn this into a generic API that takes care of the entire
> sched_getcpu() => pthread_setaffinity_np()? E.g. kvm_pin_task_to_current_cpu().
> I suspect there are other (potential) tests that don't care about what CPU they
> run on, so long as the test is pinned.
Sure.
> > +}
> > +
> > +static int open_dev_msr(int cpu)
> > +{
> > + char path[PATH_MAX];
> > + int msr_fd;
> > +
> > + snprintf(path, sizeof(path), "/dev/cpu/%d/msr", cpu);
> > + msr_fd = open(path, O_RDONLY);
> > + __TEST_REQUIRE(msr_fd >= 0, "Can't open %s for read", path);
>
> Please use open_path_or_exit().
TIL.
> Hmm, and I'm planning on posting a small series to add a variant that takes an
> ENOENT message, and spits out a (hopefully) helpful message for the EACCES case.
> It would be nice to have this one spit out something like "Is msk.ko loaded?",
> but I would say don't worry about trying to coordinate anything. Worst case
> scenario we can add a help message when the dust settles.
>
> > + return msr_fd;
> > +}
> > +
> > +static uint64_t read_dev_msr(int msr_fd, uint32_t msr)
> > +{
> > + uint64_t data;
> > + ssize_t rc;
> > +
> > + rc = pread(msr_fd, &data, sizeof(data), msr);
> > + TEST_ASSERT(rc == sizeof(data), "Read of MSR 0x%x failed", msr);
> > +
> > + return data;
> > +}
> > +
> > +static void guest_code(void)
> > +{
> > + int i;
> > +
> > + for (i = 0; i < NUM_ITERATIONS; i++) {
> > + uint64_t aperf = rdmsr(MSR_IA32_APERF);
> > + uint64_t mperf = rdmsr(MSR_IA32_MPERF);
> > +
> > + GUEST_SYNC2(aperf, mperf);
>
> Does the test generate multiple RDMSR per MSR if you do:
>
> GUEST_SYNC2(rdmsr(MSR_IA32_APERF), rdmsr(MSR_IA32_MPERF));
>
> If the code generation comes out
I'll have to check.
>
> > + }
> > +
> > + GUEST_DONE();
> > +}
> > +
> > +static bool kvm_can_disable_aperfmperf_exits(struct kvm_vm *vm)
> > +{
> > + int flags = vm_check_cap(vm, KVM_CAP_X86_DISABLE_EXITS);
> > +
> > + return flags & KVM_X86_DISABLE_EXITS_APERFMPERF;
> > +}
>
> Please don't add one-off helpers like this, especially when they're the condition
> for TEST_REQUIRE(). I *want* the gory details if the test is skipped, so that I
> don't have to go look at the source code to figure out what's missing.
>
> And it's literally more code.
Okay.
> > +
> > +int main(int argc, char *argv[])
> > +{
> > + uint64_t host_aperf_before, host_mperf_before;
> > + int cpu = sched_getcpu();
> > + struct kvm_vcpu *vcpu;
> > + struct kvm_vm *vm;
> > + int msr_fd;
> > + int i;
> > +
> > + pin_thread(cpu);
> > +
> > + msr_fd = open_dev_msr(cpu);
> > +
> > + /*
> > + * This test requires a non-standard VM initialization, because
> > + * KVM_ENABLE_CAP cannot be used on a VM file descriptor after
> > + * a VCPU has been created.
>
> Hrm, we should really sort this out. Every test that needs to enable a capability
> is having to copy+paste this pattern. I don't love the idea of expanding
> __vm_create_with_one_vcpu(), but there's gotta be a solution that isn't horrible,
> and anything is better than endly copy paste.
This is all your fault, I believe. But, I'll see what I can do.
> > + */
> > + vm = vm_create(1);
> > +
> > + TEST_REQUIRE(kvm_can_disable_aperfmperf_exits(vm));
>
> TEST_REQUIRE(vm_check_cap(vm, KVM_CAP_X86_DISABLE_EXITS) &
> KVM_X86_DISABLE_EXITS_APERFMPERF);
> > +
> > + vm_enable_cap(vm, KVM_CAP_X86_DISABLE_EXITS,
> > + KVM_X86_DISABLE_EXITS_APERFMPERF);
> > +
> > + vcpu = vm_vcpu_add(vm, 0, guest_code);
> > +
> > + host_aperf_before = read_dev_msr(msr_fd, MSR_IA32_APERF);
> > + host_mperf_before = read_dev_msr(msr_fd, MSR_IA32_MPERF);
> > +
> > + for (i = 0; i < NUM_ITERATIONS; i++) {
> > + uint64_t host_aperf_after, host_mperf_after;
> > + uint64_t guest_aperf, guest_mperf;
> > + struct ucall uc;
> > +
> > + vcpu_run(vcpu);
> > + TEST_ASSERT_KVM_EXIT_REASON(vcpu, KVM_EXIT_IO);
> > +
> > + switch (get_ucall(vcpu, &uc)) {
> > + case UCALL_DONE:
> > + break;
> > + case UCALL_ABORT:
> > + REPORT_GUEST_ASSERT(uc);
> > + case UCALL_SYNC:
> > + guest_aperf = uc.args[0];
> > + guest_mperf = uc.args[1];
> > +
> > + host_aperf_after = read_dev_msr(msr_fd, MSR_IA32_APERF);
> > + host_mperf_after = read_dev_msr(msr_fd, MSR_IA32_MPERF);
> > +
> > + TEST_ASSERT(host_aperf_before < guest_aperf,
> > + "APERF: host_before (%lu) >= guest (%lu)",
> > + host_aperf_before, guest_aperf);
>
> Honest question, is decimal really better than hex for these?
They are just numbers, so any base should be fine. I guess it depends
on which base you're most comfortable with. I could add a command-line
parameter.
> > + TEST_ASSERT(guest_aperf < host_aperf_after,
> > + "APERF: guest (%lu) >= host_after (%lu)",
> > + guest_aperf, host_aperf_after);
> > + TEST_ASSERT(host_mperf_before < guest_mperf,
> > + "MPERF: host_before (%lu) >= guest (%lu)",
> > + host_mperf_before, guest_mperf);
> > + TEST_ASSERT(guest_mperf < host_mperf_after,
> > + "MPERF: guest (%lu) >= host_after (%lu)",
> > + guest_mperf, host_mperf_after);
> > +
> > + host_aperf_before = host_aperf_after;
> > + host_mperf_before = host_mperf_after;
> > +
> > + break;
> > + }
> > + }
> > +
> > + TEST_ASSERT_EQ(i, NUM_ITERATIONS);
>
> Why?
I think this was leftover from a version where it was possible to
break out of the loop early. I'll get rid of it.
V4 this week, I hope.
