From 0d6526f82c3cdefcca47f73f5fc08dc6f335eac6 Mon Sep 17 00:00:00 2001 From: Tim Chen Date: Mon, 21 Sep 2026 17:37:22 -0700 Subject: sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling bug alb_break_llc() decides whether to break LLC preference during active load balance. It does so by testing that every runnable fair task on the source rq prefers its LLC: env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable But the two counters cover different sets. nr_pref_llc_running is updated in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued, so it follows queued tasks. h_nr_runnable is updated in set_delayed()/ clear_delayed() and drops delay-dequeued tasks. So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks, alb_break_llc() returns false, and active balance is free to pull a task off its preferred LLC. Active balance only moves runnable tasks, and this is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB skips the per-task test in can_migrate_task(). The runnable set is the one we want. Fix it on the counter side. A task should be counted in nr_pref_llc_running exactly while it is both queued on its preferred LLC (pref_llc_queued) and runnable (!sched_delayed). Define that membership once in task_pref_llc_runnable(), and adjust the counter only through pref_llc_running_inc()/pref_llc_running_dec() from the four sites that change either input: account_llc_enqueue(), account_llc_dequeue(), set_delayed() and clear_delayed(). Gating every update on the same predicate keeps the delay, wake and dequeue paths from double-counting or underflowing; see the comments at those sites for the ordering. nr_llc_running and sd->llc_counts are not touched and stay on queued semantics. Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC") Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/ Reported-by: Zhan Xusheng Suggested-by: Chen Yu Signed-off-by: Tim Chen Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Kayra Cizmeci Cc: # v7.2.x Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com --- arch/x86/kvm/debugfs.c | 196 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 196 insertions(+) create mode 100644 arch/x86/kvm/debugfs.c (limited to 'arch/x86/kvm/debugfs.c') diff --git a/arch/x86/kvm/debugfs.c b/arch/x86/kvm/debugfs.c new file mode 100644 index 000000000..0074a56e4 --- /dev/null +++ b/arch/x86/kvm/debugfs.c @@ -0,0 +1,196 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Kernel-based Virtual Machine driver for Linux + * + * Copyright 2016 Red Hat, Inc. and/or its affiliates. + */ +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include +#include +#include "lapic.h" +#include "mmu.h" +#include "mmu/mmu_internal.h" + +static int vcpu_get_timer_advance_ns(void *data, u64 *val) +{ + struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data; + *val = vcpu->arch.apic->lapic_timer.timer_advance_ns; + return 0; +} + +DEFINE_SIMPLE_ATTRIBUTE(vcpu_timer_advance_ns_fops, vcpu_get_timer_advance_ns, NULL, "%llu\n"); + +static int vcpu_get_guest_mode(void *data, u64 *val) +{ + struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data; + *val = vcpu->stat.guest_mode; + return 0; +} + +DEFINE_SIMPLE_ATTRIBUTE(vcpu_guest_mode_fops, vcpu_get_guest_mode, NULL, "%lld\n"); + +static int vcpu_get_tsc_offset(void *data, u64 *val) +{ + struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data; + *val = vcpu->arch.tsc_offset; + return 0; +} + +DEFINE_SIMPLE_ATTRIBUTE(vcpu_tsc_offset_fops, vcpu_get_tsc_offset, NULL, "%lld\n"); + +static int vcpu_get_tsc_scaling_ratio(void *data, u64 *val) +{ + struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data; + *val = vcpu->arch.tsc_scaling_ratio; + return 0; +} + +DEFINE_SIMPLE_ATTRIBUTE(vcpu_tsc_scaling_fops, vcpu_get_tsc_scaling_ratio, NULL, "%llu\n"); + +static int vcpu_get_tsc_scaling_frac_bits(void *data, u64 *val) +{ + *val = kvm_caps.tsc_scaling_ratio_frac_bits; + return 0; +} + +DEFINE_SIMPLE_ATTRIBUTE(vcpu_tsc_scaling_frac_fops, vcpu_get_tsc_scaling_frac_bits, NULL, "%llu\n"); + +void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry) +{ + debugfs_create_file("guest_mode", 0444, debugfs_dentry, vcpu, + &vcpu_guest_mode_fops); + debugfs_create_file("tsc-offset", 0444, debugfs_dentry, vcpu, + &vcpu_tsc_offset_fops); + + if (lapic_in_kernel(vcpu)) + debugfs_create_file("lapic_timer_advance_ns", 0444, + debugfs_dentry, vcpu, + &vcpu_timer_advance_ns_fops); + + if (kvm_caps.has_tsc_control) { + debugfs_create_file("tsc-scaling-ratio", 0444, + debugfs_dentry, vcpu, + &vcpu_tsc_scaling_fops); + debugfs_create_file("tsc-scaling-ratio-frac-bits", 0444, + debugfs_dentry, vcpu, + &vcpu_tsc_scaling_frac_fops); + } +} + +/* + * This covers statistics <1024 (11=log(1024)+1), which should be enough to + * cover RMAP_RECYCLE_THRESHOLD. + */ +#define RMAP_LOG_SIZE 11 + +static const char *kvm_lpage_str[KVM_NR_PAGE_SIZES] = { "4K", "2M", "1G" }; + +static int kvm_mmu_rmaps_stat_show(struct seq_file *m, void *v) +{ + struct kvm_rmap_head *rmap; + struct kvm *kvm = m->private; + struct kvm_memory_slot *slot; + struct kvm_memslots *slots; + unsigned int lpage_size, index; + /* Still small enough to be on the stack */ + unsigned int *log[KVM_NR_PAGE_SIZES], *cur; + int i, j, k, l, ret; + + if (!kvm_memslots_have_rmaps(kvm)) + return 0; + + ret = -ENOMEM; + memset(log, 0, sizeof(log)); + for (i = 0; i < KVM_NR_PAGE_SIZES; i++) { + log[i] = kcalloc(RMAP_LOG_SIZE, sizeof(unsigned int), GFP_KERNEL); + if (!log[i]) + goto out; + } + + mutex_lock(&kvm->slots_lock); + write_lock(&kvm->mmu_lock); + + for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) { + int bkt; + + slots = __kvm_memslots(kvm, i); + kvm_for_each_memslot(slot, bkt, slots) + for (k = 0; k < KVM_NR_PAGE_SIZES; k++) { + rmap = slot->arch.rmap[k]; + lpage_size = kvm_mmu_slot_lpages(slot, k + 1); + cur = log[k]; + for (l = 0; l < lpage_size; l++) { + index = fls(pte_list_count(&rmap[l])); + if (WARN_ON_ONCE(index >= RMAP_LOG_SIZE)) + index = RMAP_LOG_SIZE - 1; + cur[index]++; + } + } + } + + write_unlock(&kvm->mmu_lock); + mutex_unlock(&kvm->slots_lock); + + /* index=0 counts no rmap; index=1 counts 1 rmap */ + seq_printf(m, "Rmap_Count:\t0\t1\t"); + for (i = 2; i < RMAP_LOG_SIZE; i++) { + j = 1 << (i - 1); + k = (1 << i) - 1; + seq_printf(m, "%d-%d\t", j, k); + } + seq_printf(m, "\n"); + + for (i = 0; i < KVM_NR_PAGE_SIZES; i++) { + seq_printf(m, "Level=%s:\t", kvm_lpage_str[i]); + cur = log[i]; + for (j = 0; j < RMAP_LOG_SIZE; j++) + seq_printf(m, "%d\t", cur[j]); + seq_printf(m, "\n"); + } + + ret = 0; +out: + for (i = 0; i < KVM_NR_PAGE_SIZES; i++) + kfree(log[i]); + + return ret; +} + +static int kvm_mmu_rmaps_stat_open(struct inode *inode, struct file *file) +{ + struct kvm *kvm = inode->i_private; + int r; + + if (!kvm_get_kvm_safe(kvm)) + return -ENOENT; + + r = single_open(file, kvm_mmu_rmaps_stat_show, kvm); + if (r < 0) + kvm_put_kvm(kvm); + + return r; +} + +static int kvm_mmu_rmaps_stat_release(struct inode *inode, struct file *file) +{ + struct kvm *kvm = inode->i_private; + + kvm_put_kvm(kvm); + + return single_release(inode, file); +} + +static const struct file_operations mmu_rmaps_stat_fops = { + .owner = THIS_MODULE, + .open = kvm_mmu_rmaps_stat_open, + .read = seq_read, + .llseek = seq_lseek, + .release = kvm_mmu_rmaps_stat_release, +}; + +void kvm_arch_create_vm_debugfs(struct kvm *kvm) +{ + debugfs_create_file("mmu_rmaps_stat", 0644, kvm->debugfs_dentry, kvm, + &mmu_rmaps_stat_fops); +} -- cgit v1.3.1