KVM: X86: Don't reset dr6 unconditionally when the vcpu being scheduled out
PATCH 1/1 KVM: X86: Don't reset dr6 unconditionally when the vcpu being scheduled out
From: Lai Jiangshan <[email protected]>
The commit efdab992813fb ("KVM: x86: fix escape of guest dr6 to the host") fixed a bug. It did a great job and reset dr6 unconditionally when the vcpu being scheduled out since the linux kernel only activates breakpoints in rescheduling (perf events).
But writing to debug registers is slow, and it can be shown in perf results sometimes even neither the host nor the guest activate breakpoints.
It'd be better to reset it conditionally and this patch moves the code of reseting dr6 to the path of VM-exit where we can check the related conditions. And the comment is also updated.
---
And even in the future, the linux kernel might activate breakpoint in interrupts (rather than in rescheduling only), the host would not be confused by the stale dr6 after this patch. The possible future author who would implement it wouldn't need to care about how the kvm mananges debug registers since it sticks to the host's expectations.
Another solution is changing breakpoint.c and make the linux kernel always reset dr6 in activating breakpoints. So that dr6 is allowed to be stale when host breakpoints is not enabled and we don't need to reset dr6 in this case. But it would be no harm to reset it even in both places and killing stale states is good in programing.
arch/x86/kvm/x86.c | 29 +++++++++++++++++++++-------- 1 file changed, 21 insertions(+), 8 deletions(-)
arch/x86/kvm/x86.c +21 -8
@@ -4310,12 +4310,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)static_call (kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc =rdtsc (); -/* -* If userspace has set any breakpoints or watchpoints, dr6 is restored -* on every vmexit, but if not, we might have a stale dr6 from the -* guest. do_debug expects dr6 to be cleared after it runs, do the same. -*/ -set_debugreg (0 ,6 ); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
@@ -9375,6 +9369,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)fastpath_t exit_fastpath;bool req_immediate_exit = false; +bool reset_dr6 = false;/* Forbid vmenter if vcpu dirty ring is soft-full */ if (unlikely(vcpu->kvm->dirty_ring_size &&
@@ -9601,6 +9596,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)set_debugreg (vcpu->arch.eff_db[3 ],3 );set_debugreg (vcpu->arch.dr6,6 ); vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD ; + reset_dr6 = true; }else if (unlikely (hw_breakpoint_active ())) {set_debugreg (0 ,7 ); }
@@ -9631,17 +9627,34 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)kvm_update_dr0123 (vcpu);kvm_update_dr7 (vcpu); vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD ; + reset_dr6 = true; }/* * If the guest has used debug registers, at least dr7 * will be disabled while returning to the host. +* +* If we have active breakpoints in the host, restore the old state. +* * If we don't have active breakpoints in the host, we don't -* care about the messed up debug address registers. But if -* we have some of them active, restore the old state. +* care about the messed up debug address registers but dr6 +* which is expected to be cleared normally. Otherwise we might +* leak a stale dr6 from the guest and confuse the host since +* neither the host reset dr6 on activating next breakpoints nor +* the hardware clear it on dilivering #DB. The Intel SDM says: +* +* Certain debug exceptions may clear bits 0-3. The remaining +* contents of the DR6 register are never cleared by the +* processor. To avoid confusion in identifying debug +* exceptions, debug handlers should clear the register before +* returning to the interrupted task. +* +* Keep it simple and live up to expectations: clear DR6 immediately. */ if (hw_breakpoint_active ())hw_breakpoint_restore (); +else if (unlikely (reset_dr6)) +set_debugreg (DR6_RESERVED ,6 ); vcpu->arch.last_vmentry_cpu = vcpu->cpu; vcpu->arch.last_guest_tsc =kvm_read_l1_tsc (vcpu,rdtsc ());
Signature · 1 line
2.19.1.6.gb485710b
Sean Christopherson
maintainerQuoted 14 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
The commit
efdab992813fb("KVM: x86: fix escape of guest dr6 to the host") fixed a bug. It did a great job and reset dr6 unconditionally when the vcpu being scheduled out since the linux kernel only activates breakpoints in rescheduling (perf events).But writing to debug registers is slow, and it can be shown in perf results sometimes even neither the host nor the guest activate breakpoints.
It'd be better to reset it conditionally and this patch moves the code of reseting dr6 to the path of VM-exit where we can check the related conditions. And the comment is also updated.
Quoted 7 lines
Signed-off-by Lai Jiangshan <[email protected]>---
And even in the future, the linux kernel might activate breakpoint in interrupts (rather than in rescheduling only), the host would not be confused by the stale dr6 after this patch. The possible future author who would implement it wouldn't need to care about how the kvm mananges debug registers since it sticks to the host's expectations.
Eh, I don't think this is a valid argument. KGBD already manipulates breakpoints in NMI context, activating breakpoints from interrupt context would absolutely require a full audit of the kernel.
Quoted 5 lines
Another solution is changing
breakpoint.cand make the linux kernel always reset dr6 in activating breakpoints. So that dr6 is allowed to be stale when host breakpoints is not enabled and we don't need to reset dr6 in this case. But it would be no harm to reset it even in both places and killing stale states is good in programing.
Hmm, other than further penalizing guests that enable debug breakpoints.
What about adding a arch.switch_db_regs flag to note that DR6 is loaded with a guest value and keeping the reset code in kvm_arch_vcpu_put()?
Quoted 29 lines
arch/x86/kvm/x86.c | 29 +++++++++++++++++++++-------- 1 file changed, 21 insertions(+), 8 deletions(-)arch/x86/kvm/x86.c +1
-6@@ -4310,12 +4310,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)static_call (kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc =rdtsc (); -/* -* If userspace has set any breakpoints or watchpoints, dr6 is restored -* on every vmexit, but if not, we might have a stale dr6 from the -* guest. do_debug expects dr6 to be cleared after it runs, do the same. -*/ -set_debugreg (0 ,6 ); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,@@ -9375,6 +9369,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)fastpath_t exit_fastpath;bool req_immediate_exit = false; +bool reset_dr6 = false;/* Forbid vmenter if vcpu dirty ring is soft-full */ if (unlikely(vcpu->kvm->dirty_ring_size &&
arch/x86/kvm/x86.c static int vcpu_enter_guest(struct kvm_vcpu *vcpu) · line 9597
Quoted 2 lines
arch/x86/kvm/x86.c +0
-0@@ -9601,6 +9596,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)set_debugreg (vcpu->arch.eff_db[3 ],3 );set_debugreg (vcpu->arch.dr6,6 );
Not directly related to this patch, but why does KVM_DEBUGREG_RELOAD exist? Commit ae561edeb421 ("KVM: x86: DR0-DR3 are not clear on reset") added it to ensure DR0-3 are fresh when they're modified through non-standard paths, but I don't see any reason why the new values _must_ be loaded into hardware. eff_db needs to be updated, but I don't see why hardware DRs need to be updated unless hardware breakpoints are active or DR exiting is disabled, and in those cases updating hardware is handled by KVM_DEBUGREG_WONT_EXIT and KVM_DEBUGREG_BP_ENABLED.
Am I missing something?
Quoted 45 lines
arch/x86/kvm/x86.c +20
-2vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD ; + reset_dr6 = true; }else if (unlikely (hw_breakpoint_active ())) {set_debugreg (0 ,7 ); }@@ -9631,17 +9627,34 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)kvm_update_dr0123 (vcpu);kvm_update_dr7 (vcpu); vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD ; + reset_dr6 = true; }/* * If the guest has used debug registers, at least dr7 * will be disabled while returning to the host. +* +* If we have active breakpoints in the host, restore the old state. +* * If we don't have active breakpoints in the host, we don't -* care about the messed up debug address registers. But if -* we have some of them active, restore the old state. +* care about the messed up debug address registers but dr6 +* which is expected to be cleared normally. Otherwise we might +* leak a stale dr6 from the guest and confuse the host since +* neither the host reset dr6 on activating next breakpoints nor +* the hardware clear it on dilivering #DB. The Intel SDM says: +* +* Certain debug exceptions may clear bits 0-3. The remaining +* contents of the DR6 register are never cleared by the +* processor. To avoid confusion in identifying debug +* exceptions, debug handlers should clear the register before +* returning to the interrupted task. +* +* Keep it simple and live up to expectations: clear DR6 immediately. */ if (hw_breakpoint_active ())hw_breakpoint_restore (); +else if (unlikely (reset_dr6)) +set_debugreg (DR6_RESERVED ,6 ); vcpu->arch.last_vmentry_cpu = vcpu->cpu; vcpu->arch.last_guest_tsc =kvm_read_l1_tsc (vcpu,rdtsc ());Signature · 1 line
2.19.1.6.gb485710b
PATCH v2 1/3 KVM: X86: Remove unneeded KVM_DEBUGREG_RELOAD
From: Lai Jiangshan <[email protected]>
Commit ae561edeb421 ("KVM: x86: DR0-DR3 are not clear on reset") added code to ensure eff_db are updated when they're modified through non-standard paths.
But there is no reason to also update hardware DRs unless hardware breakpoints are active or DR exiting is disabled, and in those cases updating hardware is handled by KVM_DEBUGREG_WONT_EXIT and KVM_DEBUGREG_BP_ENABLED.
KVM_DEBUGREG_RELOAD just causes unnecesarry load of hardware DRs and is better to be removed.
--- arch/x86/include/asm/kvm_host.h | 1 - arch/x86/kvm/x86.c | 3 --- 2 files changed, 4 deletions(-)
arch/x86/include/asm/kvm_host.h +0 -1
@@ -522,7 +522,6 @@ struct kvm_pmu_ops;enum {KVM_DEBUGREG_BP_ENABLED =1 ,KVM_DEBUGREG_WONT_EXIT =2 , -KVM_DEBUGREG_RELOAD =4 , };struct kvm_mtrr_range {
arch/x86/kvm/x86.c +0 -3
@@ -1180,7 +1180,6 @@ static void kvm_update_dr0123(struct kvm_vcpu *vcpu)if (!(vcpu->guest_debug &KVM_GUESTDBG_USE_HW_BP )) {for (i =0 ; i <KVM_NR_DB_REGS ; i++) vcpu->arch.eff_db[i] = vcpu->arch.db[i]; - vcpu->arch.switch_db_regs |=KVM_DEBUGREG_RELOAD ; } }
@@ -9600,7 +9599,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)set_debugreg (vcpu->arch.eff_db[2 ],2 );set_debugreg (vcpu->arch.eff_db[3 ],3 );set_debugreg (vcpu->arch.dr6,6 ); - vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD ; }else if (unlikely (hw_breakpoint_active ())) {set_debugreg (0 ,7 ); }
@@ -9630,7 +9628,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)static_call (kvm_x86_sync_dirty_debug_regs)(vcpu);kvm_update_dr0123 (vcpu);kvm_update_dr7 (vcpu); - vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD ; }/*
Signature · 1 line
2.19.1.6.gb485710b
PATCH v2 2/3 KVM: X86: Set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT
From: Lai Jiangshan <[email protected]>
Commit c77fb5fe6f03 ("KVM: x86: Allow the guest to run with dirty debug registers") allows the guest accessing to DRs without exiting when KVM_DEBUGREG_WONT_EXIT and we need to ensure that they are synchronized on entry to the guest---including DR6 that was not synced before the commit.
But the commit sets the hardware DR6 not only when KVM_DEBUGREG_WONT_EXIT, but also when KVM_DEBUGREG_BP_ENABLED. The second case is unnecessary and just leads to a more case which leaks stale DR6 to the host which has to be resolved by unconditionally reseting DR6 in kvm_arch_vcpu_put().
We'd better to set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT, so that we can fine-grain control the cases when we need to reset it which is done in later patch.
--- arch/x86/kvm/x86.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/kvm/x86.c +3 -1
@@ -9598,7 +9598,9 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)set_debugreg (vcpu->arch.eff_db[1 ],1 );set_debugreg (vcpu->arch.eff_db[2 ],2 );set_debugreg (vcpu->arch.eff_db[3 ],3 ); -set_debugreg (vcpu->arch.dr6,6 ); +/* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */ +if (vcpu->arch.switch_db_regs &KVM_DEBUGREG_WONT_EXIT ) +set_debugreg (vcpu->arch.dr6,6 ); }else if (unlikely (hw_breakpoint_active ())) {set_debugreg (0 ,7 ); }
Signature · 1 line
2.19.1.6.gb485710b
Paolo Bonzini
maintainerRe: [PATCH V2 2/3] KVM: X86: Set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 27 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
Commit
c77fb5fe6f03("KVM: x86: Allow the guest to run with dirty debug registers") allows the guest accessing to DRs without exiting whenKVM_DEBUGREG_WONT_EXITand we need to ensure that they are synchronized on entry to the guest---including DR6 that was not synced before the commit.But the commit sets the hardware DR6 not only when
KVM_DEBUGREG_WONT_EXIT, but also whenKVM_DEBUGREG_BP_ENABLED. The second case is unnecessary and just leads to a more case which leaks stale DR6 to the host which has to be resolved by unconditionally reseting DR6 inkvm_arch_vcpu_put().We'd better to set the hardware DR6 only when
KVM_DEBUGREG_WONT_EXIT, so that we can fine-grain control the cases when we need to reset it which is done in later patch.Signed-off-by Lai Jiangshan <[email protected]>--- arch/x86/kvm/x86.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)arch/x86/kvm/x86.c +0
-0
arch/x86/kvm/x86.c static int vcpu_enter_guest(struct kvm_vcpu *vcpu) · line 9606
Quoted 10 lines
arch/x86/kvm/x86.c +3
-1@@ -9598,7 +9598,9 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)set_debugreg (vcpu->arch.eff_db[1 ],1 );set_debugreg (vcpu->arch.eff_db[2 ],2 );set_debugreg (vcpu->arch.eff_db[3 ],3 ); -set_debugreg (vcpu->arch.dr6,6 ); +/* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */ +if (vcpu->arch.switch_db_regs &KVM_DEBUGREG_WONT_EXIT ) +set_debugreg (vcpu->arch.dr6,6 ); }else if (unlikely (hw_breakpoint_active ())) {set_debugreg (0 ,7 ); }
Even better, this should be moved to vmx.c's vcpu_enter_guest. This matches the handling in svm.c:
/*
* Run with all-zero DR6 unless needed, so that we can get the exact cause
* of a #DB.
*/
if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
svm_set_dr6(svm, vcpu->arch.dr6);
else
svm_set_dr6(svm, DR6_ACTIVE_LOW);That is,
KVM: X86: Set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT
Commit c77fb5fe6f03 ("KVM: x86: Allow the guest to run with dirty debug registers") allows the guest accessing to DRs without exiting when KVM_DEBUGREG_WONT_EXIT and we need to ensure that they are synchronized on entry to the guest---including DR6 that was not synced before the commit.
But the commit sets the hardware DR6 not only when KVM_DEBUGREG_WONT_EXIT, but also when KVM_DEBUGREG_BP_ENABLED. The second case is unnecessary and just leads to a more case which leaks stale DR6 to the host which has to be resolved by unconditionally reseting DR6 in kvm_arch_vcpu_put().
Even if KVM_DEBUGREG_WONT_EXIT, however, setting the host DR6 only matters on VMX because SVM always uses the DR6 value from the VMCB. So move this line to vmx.c and make it conditional on KVM_DEBUGREG_WONT_EXIT.
arch/x86/kvm/vmx/vmx.c +4 -0
@@ -6625,6 +6625,10 @@ static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu) vmx->loaded_vmcs->host_state.cr4 = cr4; } +/* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */ +if (vcpu->arch.switch_db_regs &KVM_DEBUGREG_WONT_EXIT ) +set_debugreg (vcpu->arch.dr6,6 ); +/* When single-stepping over STI and MOV SS, we must clear the * corresponding interruptibility bits in the guest state. Otherwise * vmentry fails as it then expects bit 14 (BS) in pending debug
arch/x86/kvm/x86.c +0 -1
@@ -9597,7 +9597,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)set_debugreg (vcpu->arch.eff_db[1 ],1 );set_debugreg (vcpu->arch.eff_db[2 ],2 );set_debugreg (vcpu->arch.eff_db[3 ],3 ); -set_debugreg (vcpu->arch.dr6,6 ); }else if (unlikely (hw_breakpoint_active ())) {set_debugreg (0 ,7 ); }
Paolo
Re: [PATCH V2 2/3] KVM: X86: Set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 72 lines · Paolo Bonzini
Quoted 27 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
Commit
c77fb5fe6f03("KVM: x86: Allow the guest to run with dirty debug registers") allows the guest accessing to DRs without exiting whenKVM_DEBUGREG_WONT_EXITand we need to ensure that they are synchronized on entry to the guest---including DR6 that was not synced before the commit.But the commit sets the hardware DR6 not only when
KVM_DEBUGREG_WONT_EXIT, but also when KVM_DEBUGREG_BP_ENABLED. The second case is unnecessary and just leads to a more case which leaks stale DR6 to the host which has to be resolved by unconditionally reseting DR6 inkvm_arch_vcpu_put().We'd better to set the hardware DR6 only when
KVM_DEBUGREG_WONT_EXIT, so that we can fine-grain control the cases when we need to reset it which is done in later patch.Signed-off-by Lai Jiangshan <[email protected]>---
arch/x86/kvm/x86.c| 4 +++-1 file changed, 3 insertions(+), 1 deletion(-)arch/x86/kvm/x86.c +0
-0arch/x86/kvm/x86.c static int vcpu_enter_guest(struct kvm_vcpu *vcpu) · line 9606
Quoted 10 lines
arch/x86/kvm/x86.c +3
-1@@ -9598,7 +9598,9 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu) set_debugreg(vcpu->arch.eff_db[1], 1); set_debugreg(vcpu->arch.eff_db[2], 2); set_debugreg(vcpu->arch.eff_db[3], 3); - set_debugreg(vcpu->arch.dr6, 6); + /* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */ + if (vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT) + set_debugreg(vcpu->arch.dr6, 6); } else if (unlikely(hw_breakpoint_active())) { set_debugreg(0, 7); }Even better, this should be moved to vmx.c's vcpu_enter_guest. This matches the handling in
svm.c:/*
* Run with all-zero DR6 unless needed, so that we can get the exact cause
* of a #DB.
*/
if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))svm_set_dr6(svm, vcpu->arch.dr6);
else
svm_set_dr6(svm, DR6_ACTIVE_LOW);That is,
KVM: X86: Set the hardware DR6 only when
KVM_DEBUGREG_WONT_EXITCommitc77fb5fe6f03("KVM: x86: Allow the guest to run with dirty debug registers") allows the guest accessing to DRs without exiting whenKVM_DEBUGREG_WONT_EXITand we need to ensure that they are synchronized on entry to the guest---including DR6 that was not synced before the commit. But the commit sets the hardware DR6 not only whenKVM_DEBUGREG_WONT_EXIT, but also when KVM_DEBUGREG_BP_ENABLED. The second case is unnecessary and just leads to a more case which leaks stale DR6 to the host which has to be resolved by unconditionally reseting DR6 inkvm_arch_vcpu_put(). Even ifKVM_DEBUGREG_WONT_EXIT, however, setting the host DR6 only matters on VMX because SVM always uses the DR6 value from the VMCB. So move this line tovmx.cand make it conditional onKVM_DEBUGREG_WONT_EXIT.Signed-off-by Paolo Bonzini <[email protected]>arch/x86/kvm/vmx/vmx.c +0
-0
arch/x86/kvm/vmx/vmx.c static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu) · line 6630
Quoted 6 lines
arch/x86/kvm/vmx/vmx.c +3
-0@@ -6625,6 +6625,10 @@ static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu) vmx->loaded_vmcs->host_state.cr4 = cr4; } + /* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */ + if (vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT) + set_debugreg(vcpu->arch.dr6, 6);
I also noticed the related code in svm.c, but I refrained myself to add a new branch in vmx_vcpu_run(). But after I see you put the code of resetting dr6 in vmx_sync_dirty_debug_regs(), the whole solution is much clean and better.
And if any chance you are also concern about the additional branch, could you add a new callback to set dr6 and call the callback from x86.c when KVM_DEBUGREG_WONT_EXIT.
The possible implementation of the callback:
for vmx: set_debugreg(vcpu->arch.dr6, 6);
for svm: svm_set_dr6(svm, vcpu->arch.dr6);
and always do svm_set_dr6(svm, DR6_ACTIVE_LOW); at the end of the svm_handle_exit().
Thanks
Lai
Quoted 18 lines
arch/x86/kvm/vmx/vmx.c +1
-0+ /* When single-stepping over STI and MOV SS, we must clear the* corresponding interruptibility bits in the guest state. Otherwise * vmentry fails as it then expects bit 14 (BS) in pending debug arch/x86/kvm/x86.c +0
-1@@ -9597,7 +9597,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu) set_debugreg(vcpu->arch.eff_db[1], 1); set_debugreg(vcpu->arch.eff_db[2], 2); set_debugreg(vcpu->arch.eff_db[3], 3); - set_debugreg(vcpu->arch.dr6, 6); } else if (unlikely(hw_breakpoint_active())) { set_debugreg(0, 7); }Paolo
Paolo Bonzini
maintainerRe: [PATCH V2 2/3] KVM: X86: Set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 19 lines · Lai Jiangshan
arch/x86/kvm/vmx/vmx.c static fastpath_t vmx_vcpu_run(struct kvm_vcpu · line 6631
Quoted 12 lines
arch/x86/kvm/vmx/vmx.c +3
-0@@ -6625,6 +6625,10 @@ static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu) vmx->loaded_vmcs->host_state.cr4 = cr4; } + /* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */ + if (vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT) + set_debugreg(vcpu->arch.dr6, 6);I also noticed the related code in
svm.c, but I refrained myself to add a new branch in vmx_vcpu_run(). But after I see you put the code of resetting dr6 invmx_sync_dirty_debug_regs(), the whole solution is much clean and better.
Quoted 3 lines
And if any chance you are also concern about the additional branch, could you add a new callback to set dr6 and call the callback from
x86.cwhenKVM_DEBUGREG_WONT_EXIT.
The extra branch should be well predicted, and the idea you sketched below would cause DR6 to be marked uselessly as dirty in SVM, so I think this is cleaner. Let's add an "unlikely" around it too.
Paolo
Quoted 28 lines
The possible implementation of the callback:
for vmx:set_debugreg(vcpu->arch.dr6, 6);
for svm:svm_set_dr6(svm, vcpu->arch.dr6);
and always dosvm_set_dr6(svm, DR6_ACTIVE_LOW); at the end of thesvm_handle_exit().Thanks
LaiQuoted 18 lines
arch/x86/kvm/vmx/vmx.c +1
-0+ /* When single-stepping over STI and MOV SS, we must clear the* corresponding interruptibility bits in the guest state. Otherwise * vmentry fails as it then expects bit 14 (BS) in pending debug
arch/x86/kvm/x86.c +0
-1@@ -9597,7 +9597,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu) set_debugreg(vcpu->arch.eff_db[1], 1); set_debugreg(vcpu->arch.eff_db[2], 2); set_debugreg(vcpu->arch.eff_db[3], 3); - set_debugreg(vcpu->arch.dr6, 6); } else if (unlikely(hw_breakpoint_active())) { set_debugreg(0, 7); }Paolo
Re: [PATCH V2 2/3] KVM: X86: Set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 19 lines · Paolo Bonzini
Quoted 18 lines · Lai Jiangshan
arch/x86/kvm/vmx/vmx.c static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu) · line 6630
Quoted 11 lines
arch/x86/kvm/vmx/vmx.c +3
-0@@ -6625,6 +6625,10 @@ static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu) vmx->loaded_vmcs->host_state.cr4 = cr4; } + /* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */ + if (vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT) + set_debugreg(vcpu->arch.dr6, 6);I also noticed the related code in
svm.c, but I refrained myself to add a new branch in vmx_vcpu_run(). But after I see you put the code of resetting dr6 invmx_sync_dirty_debug_regs(), the whole solution is much clean and better.
Quoted 5 lines
Quoted 3 lines
And if any chance you are also concern about the additional branch, could you add a new callback to set dr6 and call the callback from
x86.cwhenKVM_DEBUGREG_WONT_EXIT.The extra branch should be well predicted, and the idea you sketched below would cause DR6 to be marked uselessly as dirty in SVM, so I think this is cleaner. Let's add an "unlikely" around it too.
I'm OK with it. But I don't think the sketched idea would cause DR6 to be marked uselessly as dirty in SVM. It doesn't mark it dirty if the value is unchanged, and the value is always DR6_ACTIVE_LOW except when it just clears KVM_DEBUGREG_WONT_EXIT.
Quoted 31 lines
Paolo
Quoted 28 lines
The possible implementation of the callback:
for vmx:set_debugreg(vcpu->arch.dr6, 6);
for svm:svm_set_dr6(svm, vcpu->arch.dr6);
and always dosvm_set_dr6(svm, DR6_ACTIVE_LOW); at the end of thesvm_handle_exit().Thanks
LaiQuoted 18 lines
arch/x86/kvm/vmx/vmx.c +1
-0+ /* When single-stepping over STI and MOV SS, we must clear the* corresponding interruptibility bits in the guest state. Otherwise * vmentry fails as it then expects bit 14 (BS) in pending debug arch/x86/kvm/x86.c +0
-1@@ -9597,7 +9597,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu) set_debugreg(vcpu->arch.eff_db[1], 1); set_debugreg(vcpu->arch.eff_db[2], 2); set_debugreg(vcpu->arch.eff_db[3], 3); - set_debugreg(vcpu->arch.dr6, 6); } else if (unlikely(hw_breakpoint_active())) { set_debugreg(0, 7); }Paolo
Paolo Bonzini
maintainerRe: [PATCH V2 2/3] KVM: X86: Set the hardware DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 1 line · Lai Jiangshan
I'm OK with it. But I don't think the sketched idea would cause DR6 to be marked uselessly as dirty in SVM. It doesn't mark it dirty if the value is unchanged, and the value is always
DR6_ACTIVE_LOWexcept when it just clearsKVM_DEBUGREG_WONT_EXIT.
It would be marked dirty if it is not DR6_ACTIVE_LOW, because it would be set first to DR6_ACTIVE_LOW in svm_handle_exit and then set back to the guest value on the next entry.
Paolo
PATCH v2 3/3 KVM: X86: Reset DR6 only when KVM_DEBUGREG_WONT_EXIT
In reply to Lai JiangshanFrom: Lai Jiangshan <[email protected]>
The commit efdab992813fb ("KVM: x86: fix escape of guest dr6 to the host") fixed a bug by reseting DR6 unconditionally when the vcpu being scheduled out.
But writing to debug registers is slow, and it can be shown in perf results sometimes even neither the host nor the guest activate breakpoints.
It'd be better to reset it conditionally and this patch moves the code of reseting DR6 to the path of VM-exit and only reset it when KVM_DEBUGREG_WONT_EXIT which is the only case that DR6 is guest value.
--- arch/x86/kvm/x86.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
arch/x86/kvm/x86.c +2 -6
@@ -4309,12 +4309,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)static_call (kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc =rdtsc (); -/* -* If userspace has set any breakpoints or watchpoints, dr6 is restored -* on every vmexit, but if not, we might have a stale dr6 from the -* guest. do_debug expects dr6 to be cleared after it runs, do the same. -*/ -set_debugreg (0 ,6 ); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
@@ -9630,6 +9624,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)static_call (kvm_x86_sync_dirty_debug_regs)(vcpu);kvm_update_dr0123 (vcpu);kvm_update_dr7 (vcpu); +/* Reset Dr6 which is guest value. */ +set_debugreg (DR6_RESERVED ,6 ); }/*
Signature · 1 line
2.19.1.6.gb485710b
Paolo Bonzini
maintainerRe: [PATCH V2 3/3] KVM: X86: Reset DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 35 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
The commit
efdab992813fb("KVM: x86: fix escape of guest dr6 to the host") fixed a bug by reseting DR6 unconditionally when the vcpu being scheduled out.But writing to debug registers is slow, and it can be shown in perf results sometimes even neither the host nor the guest activate breakpoints.
It'd be better to reset it conditionally and this patch moves the code of reseting DR6 to the path of VM-exit and only reset it when
KVM_DEBUGREG_WONT_EXITwhich is the only case that DR6 is guest value.Signed-off-by Lai Jiangshan <[email protected]>--- arch/x86/kvm/x86.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)arch/x86/kvm/x86.c +0
-6@@ -4309,12 +4309,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)static_call (kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc =rdtsc (); -/* -* If userspace has set any breakpoints or watchpoints, dr6 is restored -* on every vmexit, but if not, we might have a stale dr6 from the -* guest. do_debug expects dr6 to be cleared after it runs, do the same. -*/ -set_debugreg (0 ,6 ); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
arch/x86/kvm/x86.c static int vcpu_enter_guest(struct kvm_vcpu *vcpu) · line 9627
Quoted 4 lines
arch/x86/kvm/x86.c +1
-0@@ -9630,6 +9624,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)static_call (kvm_x86_sync_dirty_debug_regs)(vcpu);kvm_update_dr0123 (vcpu);kvm_update_dr7 (vcpu); +/* Reset Dr6 which is guest value. */
Better keep the rationale from the original comment,
/* * do_debug expects dr6 to be cleared after it runs, so do * not leave the guest value in the host DR6. */
Paolo
Quoted 5 lines
arch/x86/kvm/x86.c +1
-0+set_debugreg (DR6_RESERVED ,6 ); }/*
Paolo Bonzini
maintainerRe: [PATCH V2 3/3] KVM: X86: Reset DR6 only when KVM_DEBUGREG_WONT_EXIT
In reply to Lai JiangshanQuoted 36 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
The commit
efdab992813fb("KVM: x86: fix escape of guest dr6 to the host") fixed a bug by reseting DR6 unconditionally when the vcpu being scheduled out.But writing to debug registers is slow, and it can be shown in perf results sometimes even neither the host nor the guest activate breakpoints.
It'd be better to reset it conditionally and this patch moves the code of reseting DR6 to the path of VM-exit and only reset it when
KVM_DEBUGREG_WONT_EXITwhich is the only case that DR6 is guest value.Signed-off-by Lai Jiangshan <[email protected]>--- arch/x86/kvm/x86.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)arch/x86/kvm/x86.c +0
-6@@ -4309,12 +4309,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)static_call (kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc =rdtsc (); -/* -* If userspace has set any breakpoints or watchpoints, dr6 is restored -* on every vmexit, but if not, we might have a stale dr6 from the -* guest. do_debug expects dr6 to be cleared after it runs, do the same. -*/ -set_debugreg (0 ,6 ); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
arch/x86/kvm/x86.c static int vcpu_enter_guest(struct kvm_vcpu *vcpu) · line 9631
Quoted 8 lines
arch/x86/kvm/x86.c +2
-0@@ -9630,6 +9624,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)static_call (kvm_x86_sync_dirty_debug_regs)(vcpu);kvm_update_dr0123 (vcpu);kvm_update_dr7 (vcpu); +/* Reset Dr6 which is guest value. */ +set_debugreg (DR6_RESERVED ,6 ); }/*
... and this should also be done exclusively for VMX, in vmx_sync_dirty_debug_regs:
KVM: VMX: Reset DR6 only when KVM_DEBUGREG_WONT_EXIT
The commit efdab992813fb ("KVM: x86: fix escape of guest dr6 to the host") fixed a bug by resetting DR6 unconditionally when the vcpu being scheduled out.
But writing to debug registers is slow, and it can be visible in perf results sometimes, even if neither the host nor the guest activate breakpoints.
Since KVM_DEBUGREG_WONT_EXIT on Intel processors is the only case where DR6 gets the guest value, and it never happens at all on SVM, the register can be cleared in vmx.c right after reading it.
arch/x86/kvm/vmx/vmx.c +6 -0
@@ -5110,6 +5110,12 @@ static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT ;exec_controls_setbit (to_vmx (vcpu),CPU_BASED_MOV_DR_EXITING ); + +/* +* do_debug expects dr6 to be cleared after it runs, avoid that it sees +* a stale dr6 from the guest. +*/ +set_debugreg (DR6_RESERVED ,6 ); }static void vmx_set_dr7 (struct kvm_vcpu *vcpu,unsigned long val)
arch/x86/kvm/x86.c +0 -6
@@ -4313,12 +4313,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)static_call (kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc =rdtsc (); -/* -* If userspace has set any breakpoints or watchpoints, dr6 is restored -* on every vmexit, but if not, we might have a stale dr6 from the -* guest. do_debug expects dr6 to be cleared after it runs, do the same. -*/ -set_debugreg (0 ,6 ); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
Paolo
Re: [PATCH V2 3/3] KVM: X86: Reset DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 61 lines · Paolo Bonzini
Quoted 33 lines · Lai Jiangshan
From: Lai Jiangshan <[email protected]>
The commit
efdab992813fb("KVM: x86: fix escape of guest dr6 to the host") fixed a bug by reseting DR6 unconditionally when the vcpu being scheduled out.But writing to debug registers is slow, and it can be shown in perf results sometimes even neither the host nor the guest activate breakpoints.
It'd be better to reset it conditionally and this patch moves the code of reseting DR6 to the path of VM-exit and only reset it when
KVM_DEBUGREG_WONT_EXITwhich is the only case that DR6 is guest value.Signed-off-by Lai Jiangshan <[email protected]>---
arch/x86/kvm/x86.c| 8 ++------1 file changed, 2 insertions(+), 6 deletions(-)arch/x86/kvm/x86.c +0
-6@@ -4309,12 +4309,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) static_call(kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc = rdtsc(); - /* - * If userspace has set any breakpoints or watchpoints, dr6 is restored - * on every vmexit, but if not, we might have a stale dr6 from the - * guest. do_debug expects dr6 to be cleared after it runs, do the same. - */ - set_debugreg(0, 6); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,arch/x86/kvm/x86.c static int vcpu_enter_guest(struct kvm_vcpu *vcpu) · line 9631
Quoted 8 lines
arch/x86/kvm/x86.c +2
-0@@ -9630,6 +9624,8 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu) static_call(kvm_x86_sync_dirty_debug_regs)(vcpu); kvm_update_dr0123(vcpu); kvm_update_dr7(vcpu); + /* Reset Dr6 which is guest value. */ + set_debugreg(DR6_RESERVED, 6); } /*... and this should also be done exclusively for VMX, in
vmx_sync_dirty_debug_regs:KVM: VMX: Reset DR6 only when
KVM_DEBUGREG_WONT_EXIT
The commitefdab992813fb("KVM: x86: fix escape of guest dr6 to the host") fixed a bug by resetting DR6 unconditionally when the vcpu being scheduled out. But writing to debug registers is slow, and it can be visible in perf results sometimes, even if neither the host nor the guest activate breakpoints. SinceKVM_DEBUGREG_WONT_EXITon Intel processors is the only case where DR6 gets the guest value, and it never happens at all on SVM, the register can be cleared invmx.cright after reading it.Reported-by Lai Jiangshan <[email protected]>Signed-off-by Paolo Bonzini <[email protected]>arch/x86/kvm/vmx/vmx.c +0
-0
arch/x86/kvm/vmx/vmx.c static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) · line 5117
Quoted 8 lines
arch/x86/kvm/vmx/vmx.c +5
-0@@ -5110,6 +5110,12 @@ static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu) vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT; exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING); + + /* +* do_debug expects dr6 to be cleared after it runs, avoid that it sees +* a stale dr6 from the guest. +*/
do_debug() is renamed. Maybe you can use "The host kernel #DB handler".
Quoted 24 lines
arch/x86/kvm/vmx/vmx.c +1
-0+ set_debugreg(DR6_RESERVED, 6); } static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)arch/x86/kvm/x86.c +0
-6@@ -4313,12 +4313,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) static_call(kvm_x86_vcpu_put)(vcpu); vcpu->arch.last_host_tsc = rdtsc(); - /* - * If userspace has set any breakpoints or watchpoints, dr6 is restored - * on every vmexit, but if not, we might have a stale dr6 from the - * guest. do_debug expects dr6 to be cleared after it runs, do the same. - */ - set_debugreg(0, 6); } static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,Paolo
Paolo Bonzini
maintainerRe: [PATCH V2 3/3] KVM: X86: Reset DR6 only when KVM_DEBUGREG_WONT_EXIT
Quoted 8 lines · Lai Jiangshan
Quoted 5 lines
+ *
do_debugexpects dr6 to be cleared after it runs, avoid that it sees
+ * a stale dr6 from the guest.
+ */
do_debug()is renamed. Maybe you can use "The host kernel #DB handler".
Or exc_debug.
Paolo
Paolo Bonzini
maintainerQuoted 7 lines · Sean Christopherson
Not directly related to this patch, but why does
KVM_DEBUGREG_RELOADexist? Commitae561edeb421("KVM: x86: DR0-DR3 are not clear on reset") added it to ensure DR0-3 are fresh when they're modified through non-standard paths, but I don't see any reason why the newvalues_must_be loaded into hardware.eff_dbneeds to be updated, but I don't see why hardware DRs need to be updated unless hardware breakpoints are active or DR exiting is disabled, and in those cases updating hardware is handled byKVM_DEBUGREG_WONT_EXITandKVM_DEBUGREG_BP_ENABLED.
The original implementation of KVM_DEBUGREG_WONT_EXIT (by yours truly) had a bug where it did not call kvm_update_dr7 and thus KVM_DEBUGREG_BP_ENABLED was not set correctly. I agree that commit 70e4da7a8ff6 ("KVM: x86: fix root cause for missed hardware breakpoints") should have gotten rid of KVM_DEBUGREG_RELOAD altogether.
Paolo