| Message ID | 1632821269-52969-1-git-send-email-lirongqing@baidu.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | KVM: x86: directly call wbinvd for local cpu when emulate wbinvd | expand |
On 9/28/21 02:27, Li RongQing wrote: > directly call wbinvd for local pCPU, which can avoid ipi for > itself and calling of get_cpu/on_each_cpu_mask/etc. > Why is this an improvement? Trading get_cpu() vs preempt_disable() seems like a negligible difference, and it makes the code more complicated. > In fact, This change reverts commit 2eec73437487 ("KVM: x86: Avoid > issuing wbinvd twice"), since smp_call_function_many is skiping the > local cpu (as description of c2162e13d6e2f), wbinvd is not issued > twice > > and reverts commit c2162e13d6e2f ("KVM: X86: Fix missing local pCPU > when executing wbinvd on all dirty pCPUs") too, which fixed the > previous patch, when revert previous patch, it is not needed. > > Signed-off-by: Li RongQing <lirongqing@baidu.com> > --- > arch/x86/kvm/x86.c | 13 ++++++------- > 1 files changed, 6 insertions(+), 7 deletions(-) > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c > index 28ef141..ee65941 100644 > --- a/arch/x86/kvm/x86.c > +++ b/arch/x86/kvm/x86.c > @@ -6984,15 +6984,14 @@ static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) > return X86EMUL_CONTINUE; > > if (static_call(kvm_x86_has_wbinvd_exit)()) { > - int cpu = get_cpu(); > - > - cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); > - on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask, > + preempt_disable(); > + smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, > wbinvd_ipi, NULL, 1); > - put_cpu(); > + preempt_enable(); > cpumask_clear(vcpu->arch.wbinvd_dirty_mask); > - } else > - wbinvd(); > + } > + > + wbinvd(); > return X86EMUL_CONTINUE; > } > >
> -----邮件原件----- > 发件人: Andy Lutomirski <luto@kernel.org> > 发送时间: 2021年10月5日 6:57 > 收件人: Li,Rongqing <lirongqing@baidu.com>; kvm@vger.kernel.org; > wanpengli@tencent.com; jan.kiszka@siemens.com; x86@kernel.org > 主题: Re: [PATCH] KVM: x86: directly call wbinvd for local cpu when emulate > wbinvd > > On 9/28/21 02:27, Li RongQing wrote: > > directly call wbinvd for local pCPU, which can avoid ipi for itself > > and calling of get_cpu/on_each_cpu_mask/etc. > > > > Why is this an improvement? Trading get_cpu() vs preempt_disable() seems > like a negligible difference, and it makes the code more complicated. > First: to local pCpu, this reduces a ipi to itself, ipi will trigger context switch between irq and thread, it is expensive. Second, preempt_disable/preempt_enable vs get_cpu/ cpumask_set_cpu/put_cpu, the preempt_disable/ preempt_enable is more slight. And this can avoid the atomic cpumask_set_cpu -Li > > In fact, This change reverts commit 2eec73437487 ("KVM: x86: Avoid > > issuing wbinvd twice"), since smp_call_function_many is skiping the > > local cpu (as description of c2162e13d6e2f), wbinvd is not issued > > twice > > > > and reverts commit c2162e13d6e2f ("KVM: X86: Fix missing local pCPU > > when executing wbinvd on all dirty pCPUs") too, which fixed the > > previous patch, when revert previous patch, it is not needed. > > > > Signed-off-by: Li RongQing <lirongqing@baidu.com> > > --- > > arch/x86/kvm/x86.c | 13 ++++++------- > > 1 files changed, 6 insertions(+), 7 deletions(-) > > > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index > > 28ef141..ee65941 100644 > > --- a/arch/x86/kvm/x86.c > > +++ b/arch/x86/kvm/x86.c > > @@ -6984,15 +6984,14 @@ static int kvm_emulate_wbinvd_noskip(struct > kvm_vcpu *vcpu) > > return X86EMUL_CONTINUE; > > > > if (static_call(kvm_x86_has_wbinvd_exit)()) { > > - int cpu = get_cpu(); > > - > > - cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); > > - on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask, > > + preempt_disable(); > > + smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, > > wbinvd_ipi, NULL, 1); > > - put_cpu(); > > + preempt_enable(); > > cpumask_clear(vcpu->arch.wbinvd_dirty_mask); > > - } else > > - wbinvd(); > > + } > > + > > + wbinvd(); > > return X86EMUL_CONTINUE; > > } > > > >
> -----邮件原件----- > 发件人: Li,Rongqing > 发送时间: 2021年10月5日 16:05 > 收件人: 'Andy Lutomirski' <luto@kernel.org>; kvm@vger.kernel.org; > wanpengli@tencent.com; jan.kiszka@siemens.com; x86@kernel.org > 主题: 答复: [PATCH] KVM: x86: directly call wbinvd for local cpu when emulate > wbinvd > > > > > -----邮件原件----- > > 发件人: Andy Lutomirski <luto@kernel.org> > > 发送时间: 2021年10月5日 6:57 > > 收件人: Li,Rongqing <lirongqing@baidu.com>; kvm@vger.kernel.org; > > wanpengli@tencent.com; jan.kiszka@siemens.com; x86@kernel.org > > 主题: Re: [PATCH] KVM: x86: directly call wbinvd for local cpu when > > emulate wbinvd > > > > On 9/28/21 02:27, Li RongQing wrote: > > > directly call wbinvd for local pCPU, which can avoid ipi for itself > > > and calling of get_cpu/on_each_cpu_mask/etc. > > > > > > > Why is this an improvement? Trading get_cpu() vs preempt_disable() > > seems like a negligible difference, and it makes the code more complicated. > > > > First: to local pCpu, this reduces a ipi to itself, ipi will trigger context switch > between irq and thread, it is expensive. > on_each_cpu_mask will not send ipi to local cpu, the callback function will be executed directly for local cpu But this patch is useful still, Using smp_call_function_many and directly calling wbinvd can reduce the unnecessary cpumask_set_cpu and cpumask_test_cpu(), and some duplicate preempt disable -Li > Second, preempt_disable/preempt_enable vs get_cpu/ > cpumask_set_cpu/put_cpu, the preempt_disable/ preempt_enable is more > slight. > > And this can avoid the atomic cpumask_set_cpu > > -Li > > > > > In fact, This change reverts commit 2eec73437487 ("KVM: x86: Avoid > > > issuing wbinvd twice"), since smp_call_function_many is skiping the > > > local cpu (as description of c2162e13d6e2f), wbinvd is not issued > > > twice > > > > > > and reverts commit c2162e13d6e2f ("KVM: X86: Fix missing local pCPU > > > when executing wbinvd on all dirty pCPUs") too, which fixed the > > > previous patch, when revert previous patch, it is not needed. > > > > > > Signed-off-by: Li RongQing <lirongqing@baidu.com> > > > --- > > > arch/x86/kvm/x86.c | 13 ++++++------- > > > 1 files changed, 6 insertions(+), 7 deletions(-) > > > > > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index > > > 28ef141..ee65941 100644 > > > --- a/arch/x86/kvm/x86.c > > > +++ b/arch/x86/kvm/x86.c > > > @@ -6984,15 +6984,14 @@ static int kvm_emulate_wbinvd_noskip(struct > > kvm_vcpu *vcpu) > > > return X86EMUL_CONTINUE; > > > > > > if (static_call(kvm_x86_has_wbinvd_exit)()) { > > > - int cpu = get_cpu(); > > > - > > > - cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); > > > - on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask, > > > + preempt_disable(); > > > + smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, > > > wbinvd_ipi, NULL, 1); > > > - put_cpu(); > > > + preempt_enable(); > > > cpumask_clear(vcpu->arch.wbinvd_dirty_mask); > > > - } else > > > - wbinvd(); > > > + } > > > + > > > + wbinvd(); > > > return X86EMUL_CONTINUE; > > > } > > > > > >
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c index 28ef141..ee65941 100644 --- a/arch/x86/kvm/x86.c +++ b/arch/x86/kvm/x86.c @@ -6984,15 +6984,14 @@ static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) return X86EMUL_CONTINUE; if (static_call(kvm_x86_has_wbinvd_exit)()) { - int cpu = get_cpu(); - - cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); - on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask, + preempt_disable(); + smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, wbinvd_ipi, NULL, 1); - put_cpu(); + preempt_enable(); cpumask_clear(vcpu->arch.wbinvd_dirty_mask); - } else - wbinvd(); + } + + wbinvd(); return X86EMUL_CONTINUE; }
directly call wbinvd for local pCPU, which can avoid ipi for itself and calling of get_cpu/on_each_cpu_mask/etc. In fact, This change reverts commit 2eec73437487 ("KVM: x86: Avoid issuing wbinvd twice"), since smp_call_function_many is skiping the local cpu (as description of c2162e13d6e2f), wbinvd is not issued twice and reverts commit c2162e13d6e2f ("KVM: X86: Fix missing local pCPU when executing wbinvd on all dirty pCPUs") too, which fixed the previous patch, when revert previous patch, it is not needed. Signed-off-by: Li RongQing <lirongqing@baidu.com> --- arch/x86/kvm/x86.c | 13 ++++++------- 1 files changed, 6 insertions(+), 7 deletions(-)