On 22/9/2022 1:47 pm, Like Xu wrote:
Why this new request? It's not in the Intel-specific version of these
function that you elide below.
Perhaps you could split up the semantic changes from the simple renamings?
The creation of perf_event is delayed to the last step,
https://lore.kernel.org/kvm/20220831085328.45489-5-likexu@xxxxxxxxxxx/
Based on the new code base, no semantic changes I assume.
Sorry, I think we do need one more clear commit like this:
---
From e08b2d03a652e5ec226d8907c7648bff57f31d3b Mon Sep 17 00:00:00 2001
From: Like Xu <likexu@xxxxxxxxxxx>
Date: Thu, 22 Sep 2022 14:15:18 +0800
Subject: [PATCH] KVM: x86/pmu: Rewrite reprogram_counters() to improve
performance
Before using pmu->reprogram_pmi, the test for a valid pmc is always
applied. This part of the redundancy could be removed by setting the
counters' bitmask directly, and furthermore triggering KVM_REQ_PMU
only once to save more cycles.
Signed-off-by: Like Xu <likexu@xxxxxxxxxxx>
---
arch/x86/kvm/pmu.h | 12 ++++--------
1 file changed, 4 insertions(+), 8 deletions(-)
diff --git a/arch/x86/kvm/pmu.h b/arch/x86/kvm/pmu.h
index fb8040854f4d..9b8e74ccd18a 100644
--- a/arch/x86/kvm/pmu.h
+++ b/arch/x86/kvm/pmu.h
@@ -204,15 +204,11 @@ static inline bool pmc_is_enabled_globally(struct kvm_pmc
*pmc)
return test_bit(pmc->idx, (unsigned long *)&pmu->global_ctrl);
}
-static void reprogram_counters(struct kvm_pmu *pmu, u64 diff)
+static inline void reprogram_counters(struct kvm_pmu *pmu, u64 diff)
{
- int bit;
- struct kvm_pmc *pmc;
-
- for_each_set_bit(bit, (unsigned long *)&diff, X86_PMC_IDX_MAX) {
- pmc = intel_pmc_idx_to_pmc(pmu, bit);
- if (pmc)
- kvm_pmu_request_counter_reprogam(pmc);
+ if (diff) {
+ pmu->reprogram_pmi |= diff;
+ kvm_make_request(KVM_REQ_PMU, pmu_to_vcpu(pmu));
}
}
--
2.37.3