rg “^\W+trace_”
arch_timer.c
190: trace_kvm_get_timer_map(vcpu->vcpu_id, map);
348: trace_kvm_timer_hrtimer_expire(ctx);
453: trace_kvm_timer_update_irq(vcpu->vcpu_id, timer_irq(timer_ctx),
470: trace_kvm_timer_emulate(ctx, should_fire);
566: trace_kvm_timer_save_state(ctx);
655: trace_kvm_timer_restore_state(ctx);
mmio.c
136: trace_kvm_mmio(KVM_TRACE_MMIO_READ, len, run->mmio.phys_addr,
169: trace_kvm_mmio_nisv(*vcpu_pc(vcpu), kvm_vcpu_get_esr(vcpu),
201: trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, len, fault_ipa, &data);
207: trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, len,
arm.c
1221: trace_kvm_entry(*vcpu_pc(vcpu));
1280: trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1380: trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
mmu.c
1773: trace_kvm_access_fault(fault_ipa);
1847: trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_esr(vcpu),
2294: trace_kvm_set_way_flush(*vcpu_pc(vcpu),
2317: trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled);
sys_regs.c
4309: trace_kvm_sys_access(*vcpu_pc(vcpu), params, r);
4817: trace_kvm_handle_sys_reg(esr);
guest.c
921: trace_kvm_set_guest_debug(vcpu, dbg->control);
emulate-nested.c
2587: trace_kvm_forward_sysreg_trap(vcpu, sysreg, is_read);
2696: trace_kvm_nested_eret(vcpu, elr, spsr);
2712: trace_kvm_inject_nested_exception(vcpu, esr_el2, type);
handle_exit.c
40: trace_kvm_hvc_arm64(*vcpu_pc(vcpu), vcpu_get_reg(vcpu, 0),
139: trace_kvm_wfx_arm64(*vcpu_pc(vcpu), true);
142: trace_kvm_wfx_arm64(*vcpu_pc(vcpu), false);
vgic/vgic.c
446: trace_vgic_update_irq_pending(vcpu ? vcpu->vcpu_idx : 0, intid, level);
aarch64 kvm_stat 观测
在 6.16 的环境中测试，其结果为:
kvm_get_timer_map 1032963 18.3 9838
kvm_timer_update_irq 697115 12.3 6620
vgic_update_irq_pending 697115 12.3 6620
kvm_timer_save_state 676580 12.0 6467
kvm_timer_restore_state 676580 12.0 6467
kvm_exit 477754 8.4 3691
kvm_exit(WFx) 413413 86.5 3228
kvm_exit(SYS64) 33750 7.1 234
kvm_exit(UNKNOWN) 28805 6.0 212
kvm_exit(DABT_LOW) 1755 0.4 16
kvm_exit(HVC64) 18 0.0 0
kvm_exit(IABT_LOW) 1 0.0 0
kvm_entry 477754 8.4 3690
kvm_wfx_arm64 413414 7.3 3228
kvm_vcpu_wakeup 338921 6.0 3228
kvm_halt_poll_ns 92889 1.6 613
kvm_sys_access 33750 0.6 234
kvm_handle_sys_reg 33750 0.6 234
kvm_mmio 1875 0.0 16
kvm_guest_fault 1756 0.0 16
kvm_userspace_exit 1484 0.0 15
kvm_userspace_exit(MMIO) 1484 100.0 15
kvm_unmap_hva_range 366 0.0 2
kvm_set_irq 46 0.0 0
kvm_hvc_arm64 18 0.0 0
Total 5654130 50980
但是在 4.19 的环境测试发现结果为这个:
kvm statistics - summary
Event Total %Total CurAvg/s
kvm_arm_set_dreg32 476260 16.5 8838
kvm_get_timer_map 337009 11.7 6392
kvm_exit 238130 8.3 4419
kvm_exit(UNKNOWN) 124281 52.2 2374
kvm_exit(WFx) 112549 47.3 2028
kvm_exit(SYS64) 501 0.2 12
kvm_exit(DABT_LOW) 723 0.3 6
kvm_exit(IABT_LOW) 63 0.0 0
kvm_exit(HVC64) 5 0.0 0
kvm_entry 238130 8.3 4419
kvm_arm_clear_debug 238130 8.3 4419
kvm_arm_setup_debug 238130 8.3 4419
kvm_timer_update_irq 227856 7.9 4362
vgic_update_irq_pending 227856 7.9 4362
kvm_timer_restore_state 219878 7.6 4079
kvm_timer_save_state 219878 7.6 4079
kvm_wfx_arm64 112550 3.9 2028
kvm_vcpu_wakeup 109141 3.8 2027
kvm_halt_poll_ns 806 0.0 21
kvm_sys_access 501 0.0 12
kvm_handle_sys_reg 501 0.0 12
kvm_guest_fault 786 0.0 6
kvm_mmio 443 0.0 6
kvm_userspace_exit 264 0.0 5
kvm_userspace_exit(MMIO) 264 100.0 5
kvm_hvc_arm64 5 0.0 0
kvm_set_irq 4 0.0 0
Total 2886258 53903
也就是多出来了很多:
kvm_arm_set_dreg32 476260 16.5 8838
kvm_arm_clear_debug 238130 8.3 4419
kvm_arm_setup_debug 238130 8.3 4419
高版本只是把对应的 tracepoint 给移除掉了而已，具体的代码在 arch/arm64/kvm/debug.c
每次上下文切换都是有 debug 寄存器的恢复，所以可以 trace 到特别多的东西。
解决一下这些问题:
- kvm_exit 的 reason 都是意思?
- 既然都是硬件发送中断了，为什么还有这么多时钟中断相关的调用
kvm_get_timer_map 1032963 18.3 9838 kvm_timer_update_irq 697115 12.3 6620 vgic_update_irq_pending 697115 12.3 6620 kvm_timer_save_state 676580 12.0 6467 kvm_timer_restore_state 676580 12.0 6467
本站所有文章转发 CSDN 将按侵权追究法律责任，其它情况随意。