Re: [PATCH 7/7] KVM: selftests: Test Intel counters' bit width emulation

[Sean Christopherson <seanjc@google.com>]

On Thu, Mar 23, 2023, Like Xu wrote:
> +static uint64_t test_ctrs_bit_width_setup(struct kvm_vcpu *vcpu,
> +					  uint8_t bit_width,
> +					  uint64_t perf_cap,
> +					  uint32_t msr_base)
> +{
> +	struct kvm_cpuid_entry2 *entry;
> +	bool fw_wr = perf_cap & PMU_CAP_FW_WRITES;
> +	uint64_t kvm_width;
> +	uint64_t value;
> +
> +	entry = vcpu_get_cpuid_entry(vcpu, 0xa);
> +	if (msr_base != MSR_CORE_PERF_FIXED_CTR0) {
> +		kvm_width = kvm_gp_ctr_bit_width();
> +		entry->eax = (entry->eax & ~GP_WIDTH_MASK) |
> +			(bit_width << GP_WIDTH_OFS_BIT);
> +	} else {
> +		kvm_width = kvm_fixed_ctr_bit_width();
> +		entry->edx = (entry->edx & ~FIXED_WIDTH_MASK) |
> +			(bit_width << FIXED_WIDTH_OFS_BIT);
> +	}
> +	TEST_REQUIRE(kvm_width > 31);

Unfortunately, using TEST_REQUIRE() in a subtest is generally a bad idea.  This
will skip _all_ tests if the requirement isn't met.  That might be a signal that
the test is doing too much, i.e. should be split into multiple tests.  Unlike KUT,
selftests are more geared towards lots of small tests, not a handful of massive
tests.