| Message ID | 1394192571-11056-5-git-send-email-pbonzini@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
2014-03-07 12:42+0100, Paolo Bonzini: > When not running in guest-debug mode (i.e. the guest controls the debug > registers, having to take an exit for each DR access is a waste of time. > If the guest gets into a state where each context switch causes DR to be > saved and restored, this can take away as much as 40% of the execution > time from the guest. > > If the guest is running with vcpu->arch.db == vcpu->arch.eff_db, we > can let it write freely to the debug registers and reload them on the > next exit. We still need to exit on the first access, so that the > KVM_DEBUGREG_WONT_EXIT flag is set in switch_db_regs; after that, further > accesses to the debug registers will not cause a vmexit. > > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- > arch/x86/kvm/vmx.c | 37 ++++++++++++++++++++++++++++++++++++- > 1 file changed, 36 insertions(+), 1 deletion(-) > > diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c > index 06e4ec877a1c..e377522408b1 100644 > --- a/arch/x86/kvm/vmx.c > +++ b/arch/x86/kvm/vmx.c > @@ -2843,7 +2843,7 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) > vmx_capability.ept, vmx_capability.vpid); > } > > - min = 0; > + min = VM_EXIT_SAVE_DEBUG_CONTROLS; > #ifdef CONFIG_X86_64 > min |= VM_EXIT_HOST_ADDR_SPACE_SIZE; > #endif > @@ -5113,6 +5113,22 @@ static int handle_dr(struct kvm_vcpu *vcpu) > } > } > > + if (vcpu->guest_debug == 0) { > + u32 cpu_based_vm_exec_control; > + > + cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); > + cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING; > + vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); vmcs_clear_bits() covers exactly this use-case. (Barring the explicit bit-width.) > + > + /* > + * No more DR vmexits; force a reload of the debug registers > + * and reenter on this instruction. The next vmexit will > + * retrieve the full state of the debug registers. > + */ > + vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; > + return 1; > + } > + We could make the code slighly uglier and move the functional part of this block before the previous one, so it would do both things in one exit. (Exception handler will likely access DR too.) > exit_qualification = vmcs_readl(EXIT_QUALIFICATION); > dr = exit_qualification & DEBUG_REG_ACCESS_NUM; > reg = DEBUG_REG_ACCESS_REG(exit_qualification); > @@ -5139,6 +5155,24 @@ static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val) > { > } > > +static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) > +{ > + u32 cpu_based_vm_exec_control; > + > + get_debugreg(vcpu->arch.db[0], 0); > + get_debugreg(vcpu->arch.db[1], 1); > + get_debugreg(vcpu->arch.db[2], 2); > + get_debugreg(vcpu->arch.db[3], 3); > + get_debugreg(vcpu->arch.dr6, 6); > + vcpu->arch.dr7 = vmcs_readl(GUEST_DR7); > + > + vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; > + > + cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); > + cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING; > + vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); (I'll be calling this sneaky later on.) > +} > + > static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val) > { > vmcs_writel(GUEST_DR7, val); > @@ -8590,6 +8624,7 @@ static struct kvm_x86_ops vmx_x86_ops = { > .get_dr6 = vmx_get_dr6, > .set_dr6 = vmx_set_dr6, > .set_dr7 = vmx_set_dr7, > + .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs, > .cache_reg = vmx_cache_reg, > .get_rflags = vmx_get_rflags, > .set_rflags = vmx_set_rflags, > -- > 1.8.3.1 > > > -- > 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/ -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Il 09/03/2014 19:26, Radim Kr?má? ha scritto: > > + > > + cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); > > + cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING; > > + vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); > > vmcs_clear_bits() covers exactly this use-case. > (Barring the explicit bit-width.) Good idea. > > + > > + /* > > + * No more DR vmexits; force a reload of the debug registers > > + * and reenter on this instruction. The next vmexit will > > + * retrieve the full state of the debug registers. > > + */ > > + vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; > > + return 1; > > + } > > + > > We could make the code slighly uglier and move the functional part of > this block before the previous one, so it would do both things in one > exit. I considered this, but decided that it's unlikely for emulation to be faster than hardware---especially on those AMD CPUs that lack decode assists (and it's good for VMX and SVM code to look as similar as possible). > (Exception handler will likely access DR too.) Which exception handler? Paolo -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
2014-03-09 21:12+0100, Paolo Bonzini: > Il 09/03/2014 19:26, Radim Kr?má? ha scritto: > >> + > >> + /* > >> + * No more DR vmexits; force a reload of the debug registers > >> + * and reenter on this instruction. The next vmexit will > >> + * retrieve the full state of the debug registers. > >> + */ > >> + vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; > >> + return 1; > >> + } > >> + > > > >We could make the code slighly uglier and move the functional part of > >this block before the previous one, so it would do both things in one > >exit. > > I considered this, but decided that it's unlikely for emulation to > be faster than hardware---especially on those AMD CPUs that lack > decode assists (and it's good for VMX and SVM code to look as > similar as possible). I meant to move it before the block that exits if there is the 'exception on access' bit set in cr7, so we wouldn't exit again right away on the actual access, which is quite likely. (Exiting without emulation was a great move.) > >(Exception handler will likely access DR too.) > > Which exception handler? For #DB. (Pure guesswork, I haven't seen any of them.) -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Il 10/03/2014 13:17, Radim Kr?má? ha scritto: >> > I considered this, but decided that it's unlikely for emulation to >> > be faster than hardware---especially on those AMD CPUs that lack >> > decode assists (and it's good for VMX and SVM code to look as >> > similar as possible). > I meant to move it before the block that exits if there is the > 'exception on access' bit set in cr7, so we wouldn't exit again right > away on the actual access, which is quite likely. > (Exiting without emulation was a great move.) > Ah yes, for the GD bit. I don't know, it's quite unlikely that anyone will use it... It's a feeble attempt at allowing virtualization of the debug registers without VMX. Paolo -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c index 06e4ec877a1c..e377522408b1 100644 --- a/arch/x86/kvm/vmx.c +++ b/arch/x86/kvm/vmx.c @@ -2843,7 +2843,7 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf) vmx_capability.ept, vmx_capability.vpid); } - min = 0; + min = VM_EXIT_SAVE_DEBUG_CONTROLS; #ifdef CONFIG_X86_64 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE; #endif @@ -5113,6 +5113,22 @@ static int handle_dr(struct kvm_vcpu *vcpu) } } + if (vcpu->guest_debug == 0) { + u32 cpu_based_vm_exec_control; + + cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); + cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING; + vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); + + /* + * No more DR vmexits; force a reload of the debug registers + * and reenter on this instruction. The next vmexit will + * retrieve the full state of the debug registers. + */ + vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT; + return 1; + } + exit_qualification = vmcs_readl(EXIT_QUALIFICATION); dr = exit_qualification & DEBUG_REG_ACCESS_NUM; reg = DEBUG_REG_ACCESS_REG(exit_qualification); @@ -5139,6 +5155,24 @@ static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val) { } +static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) +{ + u32 cpu_based_vm_exec_control; + + get_debugreg(vcpu->arch.db[0], 0); + get_debugreg(vcpu->arch.db[1], 1); + get_debugreg(vcpu->arch.db[2], 2); + get_debugreg(vcpu->arch.db[3], 3); + get_debugreg(vcpu->arch.dr6, 6); + vcpu->arch.dr7 = vmcs_readl(GUEST_DR7); + + vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; + + cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); + cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING; + vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control); +} + static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val) { vmcs_writel(GUEST_DR7, val); @@ -8590,6 +8624,7 @@ static struct kvm_x86_ops vmx_x86_ops = { .get_dr6 = vmx_get_dr6, .set_dr6 = vmx_set_dr6, .set_dr7 = vmx_set_dr7, + .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs, .cache_reg = vmx_cache_reg, .get_rflags = vmx_get_rflags, .set_rflags = vmx_set_rflags,
When not running in guest-debug mode (i.e. the guest controls the debug registers, having to take an exit for each DR access is a waste of time. If the guest gets into a state where each context switch causes DR to be saved and restored, this can take away as much as 40% of the execution time from the guest. If the guest is running with vcpu->arch.db == vcpu->arch.eff_db, we can let it write freely to the debug registers and reload them on the next exit. We still need to exit on the first access, so that the KVM_DEBUGREG_WONT_EXIT flag is set in switch_db_regs; after that, further accesses to the debug registers will not cause a vmexit. Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> --- arch/x86/kvm/vmx.c | 37 ++++++++++++++++++++++++++++++++++++- 1 file changed, 36 insertions(+), 1 deletion(-)