Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932585AbZLOEHi (ORCPT ); Mon, 14 Dec 2009 23:07:38 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932506AbZLOEHa (ORCPT ); Mon, 14 Dec 2009 23:07:30 -0500 Received: from mx1.redhat.com ([209.132.183.28]:12878 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932451AbZLOEHZ (ORCPT ); Mon, 14 Dec 2009 23:07:25 -0500 From: Zachary Amsden To: kvm@vger.kernel.org Cc: Zachary Amsden , Avi Kivity , Marcelo Tosatti , Joerg Roedel , linux-kernel@vger.kernel.org, Dor Laor Subject: [PATCH RFC: kvm tsc virtualization 14/20] Move TSC cpu vars to a struct Date: Mon, 14 Dec 2009 18:08:41 -1000 Message-Id: <1260850127-9766-15-git-send-email-zamsden@redhat.com> In-Reply-To: <1260850127-9766-14-git-send-email-zamsden@redhat.com> References: <1260850127-9766-1-git-send-email-zamsden@redhat.com> <1260850127-9766-2-git-send-email-zamsden@redhat.com> <1260850127-9766-3-git-send-email-zamsden@redhat.com> <1260850127-9766-4-git-send-email-zamsden@redhat.com> <1260850127-9766-5-git-send-email-zamsden@redhat.com> <1260850127-9766-6-git-send-email-zamsden@redhat.com> <1260850127-9766-7-git-send-email-zamsden@redhat.com> <1260850127-9766-8-git-send-email-zamsden@redhat.com> <1260850127-9766-9-git-send-email-zamsden@redhat.com> <1260850127-9766-10-git-send-email-zamsden@redhat.com> <1260850127-9766-11-git-send-email-zamsden@redhat.com> <1260850127-9766-12-git-send-email-zamsden@redhat.com> <1260850127-9766-13-git-send-email-zamsden@redhat.com> <1260850127-9766-14-git-send-email-zamsden@redhat.com> Organization: Frobozz Magic Timekeeping Company Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 9246 Lines: 264 There's enough of these floating around randomly to justify using a structure for keeping them in one place. Signed-off-by: Zachary Amsden --- arch/x86/kvm/x86.c | 110 ++++++++++++++++++++++++++++++---------------------- 1 files changed, 64 insertions(+), 46 deletions(-) diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 8bd362b..7bc20ac 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -738,19 +738,24 @@ static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info * } DEFINE_SPINLOCK(kvm_tsc_lock); -static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); -static DEFINE_PER_CPU(unsigned long, cpu_tsc_multiplier); -static DEFINE_PER_CPU(int, cpu_tsc_shift); -static DEFINE_PER_CPU(s64, cpu_tsc_offset); -static DEFINE_PER_CPU(u64, cpu_tsc_measure_base); -static DEFINE_PER_CPU(atomic_t, cpu_tsc_synchronized); -static DEFINE_PER_CPU(int, cpu_tsc_generation); +struct cpu_tsc_vars +{ + unsigned long tsc_khz; + unsigned long tsc_multiplier; + unsigned int tsc_shift; + int tsc_generation; + s64 tsc_offset; + u64 tsc_measure_base; + atomic_t tsc_synchronized; +}; +static DEFINE_PER_CPU(struct cpu_tsc_vars, cpu_tsc_vars); + static int tsc_base_cpu = -1; static unsigned long ref_tsc_khz; static inline int cpu_is_tsc_synchronized(int cpu) { - return (atomic_read(&per_cpu(cpu_tsc_synchronized, cpu)) != 0); + return (atomic_read(&per_cpu(cpu_tsc_vars, cpu).tsc_synchronized) != 0); } static inline unsigned long div_precise(unsigned long hi, unsigned long lo, @@ -808,12 +813,12 @@ static inline unsigned long mult_precise(unsigned long val, unsigned long mult, return bot; } -static inline u64 compute_ref_tsc(int cpu) +static inline u64 compute_ref_tsc(void) { - u64 tsc = native_read_tsc() - per_cpu(cpu_tsc_measure_base, cpu); - tsc = mult_precise(tsc, per_cpu(cpu_tsc_multiplier, cpu), - per_cpu(cpu_tsc_shift, cpu)); - return tsc + per_cpu(cpu_tsc_offset, cpu); + struct cpu_tsc_vars *cv = &__get_cpu_var(cpu_tsc_vars); + u64 tsc = native_read_tsc() - cv->tsc_measure_base; + tsc = mult_precise(tsc, cv->tsc_multiplier, cv->tsc_shift); + return tsc + cv->tsc_offset; } /* @@ -824,15 +829,17 @@ static inline u64 compute_ref_tsc(int cpu) u64 kvm_get_ref_tsc(void) { int cpu, gen; + struct cpu_tsc_vars *cv; u64 tsc; cpu = get_cpu(); + cv = &per_cpu(cpu_tsc_vars, cpu); again: - gen = per_cpu(cpu_tsc_generation, cpu); + gen = cv->tsc_generation; smp_rmb(); - tsc = compute_ref_tsc(cpu); + tsc = compute_ref_tsc(); smp_rmb(); - if (unlikely(gen != per_cpu(cpu_tsc_generation, cpu))) + if (unlikely(gen != cv->tsc_generation)) goto again; put_cpu(); return tsc; @@ -889,7 +896,7 @@ static void sync_tsc_helper(int measure_cpu, s64 *delta, atomic_t *ready) /* wait */; } native_cpuid(&junk, &junk, &junk, &junk); - tsc = compute_ref_tsc(cpu); + tsc = compute_ref_tsc(); rdtsc_barrier(); if (cpu == measure_cpu) { while (!atomic_read(ready)) @@ -972,17 +979,18 @@ static void kvm_sync_tsc(void *cpup) unsigned long flags; s64 *delta1, *delta2; static atomic_t ready ____cacheline_aligned = ATOMIC_INIT(1); + struct cpu_tsc_vars *cv = &per_cpu(cpu_tsc_vars, new_cpu); BUG_ON(tsc_base_cpu == -1); - pr_debug("%s: IN, cpu = %d, freq = %ldkHz, tsc_base_cpu = %d\n", __func__, raw_smp_processor_id(), per_cpu(cpu_tsc_khz, raw_smp_processor_id()) , tsc_base_cpu); local_irq_save(flags); if (raw_smp_processor_id() == new_cpu) { - per_cpu(cpu_tsc_measure_base, new_cpu) = native_read_tsc(); - per_cpu(cpu_tsc_offset, new_cpu) = 0; - compute_best_multiplier(ref_tsc_khz, - per_cpu(cpu_tsc_khz, new_cpu), - &per_cpu(cpu_tsc_multiplier, new_cpu), - &per_cpu(cpu_tsc_shift, new_cpu)); + cv->tsc_measure_base = native_read_tsc(); + cv->tsc_offset = 0; + compute_best_multiplier(ref_tsc_khz, cv->tsc_khz, + &cv->tsc_multiplier, &cv->tsc_shift); + pr_debug("%s: IN, cpu = %d, freq = %ldkHz, measure_base = %lld," + " tsc_base_cpu = %d\n", __func__, new_cpu, cv->tsc_khz, + cv->tsc_measure_base, tsc_base_cpu); } delta1 = per_cpu(delta_array, tsc_base_cpu).delta; delta2 = per_cpu(delta_array, new_cpu).delta; @@ -1004,22 +1012,32 @@ static void kvm_sync_tsc(void *cpup) accumulator -= average_samples(&delta2[2], SYNC_TRIES-3); accumulator /= 2; - per_cpu(cpu_tsc_offset, new_cpu) = accumulator; - ++per_cpu(cpu_tsc_generation, new_cpu); - atomic_set(&per_cpu(cpu_tsc_synchronized, new_cpu), 1); - pr_debug("%s: OUT, cpu = %d, cpu_tsc_offset = %lld, cpu_tsc_multiplier=%ld, cpu_tsc_shift=%d\n", __func__, raw_smp_processor_id(), per_cpu(cpu_tsc_offset, new_cpu), per_cpu(cpu_tsc_multiplier, new_cpu), per_cpu(cpu_tsc_shift, new_cpu)); + cv->tsc_offset = accumulator; + smp_wmb(); + ++cv->tsc_generation; + atomic_set(&cv->tsc_synchronized, 1); + ++kvm_stats.tsc_resync; + pr_debug("%s: OUT, cpu = %d, delta = %lld, cpu_tsc_offset = " + "%lld, cpu_tsc_multiplier=%ld, cpu_tsc_shift=%d\n", + __func__, new_cpu, accumulator, cv->tsc_offset, + cv->tsc_multiplier, cv->tsc_shift); } local_irq_restore(flags); } static void kvm_do_sync_tsc(int cpu) { + BUG_ON(tsc_base_cpu == -1); spin_lock(&kvm_tsc_lock); /* tsc_base_cpu can change without tsc_lock, so recheck */ if (unlikely(cpu == tsc_base_cpu)) goto out_unlock; + /* + * We're calling a co-routine; if we're one of the called CPUs, we + * must call ourselves last. + */ if (raw_smp_processor_id() != tsc_base_cpu) { smp_call_function_single(tsc_base_cpu, kvm_sync_tsc, (void *)&cpu, 0); @@ -1046,12 +1064,12 @@ static void kvm_write_guest_time(struct kvm_vcpu *v) if ((!vcpu->time_page)) return; - this_tsc_khz = get_cpu_var(cpu_tsc_khz); + this_tsc_khz = get_cpu_var(cpu_tsc_vars).tsc_khz; if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) { kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock); vcpu->hv_clock_tsc_khz = this_tsc_khz; } - put_cpu_var(cpu_tsc_khz); + put_cpu_var(cpu_tsc_vars); /* Keep irq disabled to prevent changes to the clock */ local_irq_save(flags); @@ -3572,18 +3590,16 @@ static void resync_user(struct work_struct *work) static void resync(void *info) { - int cpu; + struct cpu_tsc_vars *cv = &__get_cpu_var(cpu_tsc_vars); u64 tsc; /* Fixup our own values to stay in sync with the reference */ - cpu = raw_smp_processor_id(); - tsc = compute_ref_tsc(cpu); - per_cpu(cpu_tsc_measure_base, cpu) = native_read_tsc(); - per_cpu(cpu_tsc_offset, cpu) = tsc; - compute_best_multiplier(ref_tsc_khz, per_cpu(cpu_tsc_khz, cpu), - &per_cpu(cpu_tsc_multiplier, cpu), - &per_cpu(cpu_tsc_shift, cpu)); - atomic_set(&per_cpu(cpu_tsc_synchronized, cpu), 1); + tsc = compute_ref_tsc(); + cv->tsc_measure_base = native_read_tsc(); + cv->tsc_offset = tsc; + compute_best_multiplier(ref_tsc_khz, cv->tsc_khz, &cv->tsc_multiplier, + &cv->tsc_shift); + atomic_set(&cv->tsc_synchronized, 1); /* Then, get everybody else on board */ if (!work_scheduled) { @@ -3599,6 +3615,7 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va struct kvm *kvm; struct kvm_vcpu *vcpu; int i, send_ipi = 0; + struct cpu_tsc_vars *cv = &per_cpu(cpu_tsc_vars, freq->cpu); /* * There is no way to precisely know the TSC value at which time the @@ -3612,7 +3629,7 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va */ if (val == CPUFREQ_PRECHANGE) { get_online_cpus(); - atomic_set(&per_cpu(cpu_tsc_synchronized, freq->cpu), 0); + atomic_set(&cv->tsc_synchronized, 0); spin_lock(&kvm_lock); list_for_each_entry(kvm, &vm_list, vm_list) { kvm_for_each_vcpu(i, vcpu, kvm) { @@ -3642,7 +3659,7 @@ static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long va * Just use the vm_tsc_lock for a mutex. */ spin_lock(&kvm_tsc_lock); - per_cpu(cpu_tsc_khz, freq->cpu) = freq->new; + cv->tsc_khz = freq->new; spin_unlock(&kvm_tsc_lock); return 0; @@ -3691,12 +3708,13 @@ static int kvm_x86_cpu_hotplug(struct notifier_block *notifier, case CPU_DYING: case CPU_UP_CANCELED: - atomic_set(&per_cpu(cpu_tsc_synchronized, cpu), 0); + atomic_set(&per_cpu(cpu_tsc_vars, cpu).tsc_synchronized, 0); break; case CPU_ONLINE: if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) - per_cpu(cpu_tsc_khz, cpu) = cpufreq_quick_get(cpu); + per_cpu(cpu_tsc_vars, cpu).tsc_khz = + cpufreq_quick_get(cpu); kvm_do_sync_tsc(cpu); break; } @@ -3767,13 +3785,13 @@ static void kvm_timer_init(void) if (!khz) khz = tsc_khz; spin_lock(&kvm_tsc_lock); - if (!per_cpu(cpu_tsc_khz, cpu)) - per_cpu(cpu_tsc_khz, cpu) = khz; + if (!per_cpu(cpu_tsc_vars, cpu).tsc_khz) + per_cpu(cpu_tsc_vars, cpu).tsc_khz = khz; spin_unlock(&kvm_tsc_lock); } } else { for_each_possible_cpu(cpu) - per_cpu(cpu_tsc_khz, cpu) = tsc_khz; + per_cpu(cpu_tsc_vars, cpu).tsc_khz = tsc_khz; } tsc_base_cpu = get_cpu(); resync(NULL); -- 1.6.5.2 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/