KVM: Restore original behavior of kvm.halt_poll_ns
cover KVM: Restore original behavior of kvm.halt_poll_ns
This series restores the original behavior of the module parameter kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL. This should allow admins to administer VMs not using KVM_CAP_HALT_POLL just as they were before this capability was introduced.
VMs that use KVM_CAP_HALT_POLL can be administered through userspace changes that invoke KVM_CAP_HALT_POLL (i.e. the capability can be invoked multiple times to change the max halt poll time).
If any admin needs a system-wide override of KVM_CAP_HALT_POLL we can add that through a new mechanism (e.g. a new bool module parameter).
Compile-tested only. I want to get feedback from Christian and Yunan if this approach would address their concerns.
David Matlack (3):
KVM: Cap vcpu->halt_poll_ns before halting rather than after
KVM: Avoid re-reading kvm->max_halt_poll_ns during halt-polling
KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 52 ++++++++++++++++++++++++++++++++-------- 2 files changed, 43 insertions(+), 10 deletions(-)
base-commit: d663b8a285986072428a6a145e5994bc275df994
Signature · 1 line
2.38.1.431.g37b22c650d-goog
RFC PATCH 1/3 KVM: Cap vcpu->halt_poll_ns before halting rather than after
Cap vcpu->halt_poll_ns based on the max halt polling time just before halting, rather than after the last halt. This arguably provides better accuracy if an admin disables halt polling in between halts, although the improvement is nominal.
A side-effect of this change is that grow_halt_poll_ns() no longer needs to access vcpu->kvm->max_halt_poll_ns, which will be useful in a future commit where the max halt polling time can come from the module parameter halt_poll_ns instead.
--- virt/kvm/kvm_main.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-)
virt/kvm/kvm_main.c +6 -4
@@ -3385,9 +3385,6 @@ static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)if (val < grow_start) val = grow_start; -if (val > vcpu->kvm->max_halt_poll_ns) - val = vcpu->kvm->max_halt_poll_ns; - vcpu->halt_poll_ns = val; out:trace_kvm_halt_poll_ns_grow (vcpu->vcpu_id, val, old);
@@ -3500,11 +3497,16 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,void kvm_vcpu_halt (struct kvm_vcpu *vcpu) {bool halt_poll_allowed = !kvm_arch_no_poll (vcpu); -bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;ktime_t start, cur, poll_end;bool waited = false; +bool do_halt_poll;u64 halt_ns; +if (vcpu->halt_poll_ns > vcpu->kvm->max_halt_poll_ns) + vcpu->halt_poll_ns = vcpu->kvm->max_halt_poll_ns; + + do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; + start = cur = poll_end =ktime_get ();if (do_halt_poll) {ktime_t stop =ktime_add_ns (start, vcpu->halt_poll_ns);
Signature · 1 line
2.38.1.431.g37b22c650d-goog
Re: [RFC PATCH 1/3] KVM: Cap vcpu->halt_poll_ns before halting rather than after
Am 17.11.22 um 01:16 schrieb David Matlack:
Quoted 11 lines
Cap
vcpu->halt_poll_nsbased on the max halt polling time just before halting, rather than after the last halt. This arguably provides better accuracy if an admin disables halt polling in between halts, although the improvement is nominal.A side-effect of this change is that
grow_halt_poll_ns()no longer needs to accessvcpu->kvm->max_halt_poll_ns, which will be useful in a future commit where the max halt polling time can come from the module parameterhalt_poll_nsinstead.Signed-off-by David Matlack <[email protected]>
Looks sane
Quoted 36 lines
--- virt/kvm/kvm_main.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-)virt/kvm/kvm_main.c +6
-4@@ -3385,9 +3385,6 @@ static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)if (val < grow_start) val = grow_start; -if (val > vcpu->kvm->max_halt_poll_ns) - val = vcpu->kvm->max_halt_poll_ns; - vcpu->halt_poll_ns = val; out:trace_kvm_halt_poll_ns_grow (vcpu->vcpu_id, val, old);@@ -3500,11 +3497,16 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,void kvm_vcpu_halt (struct kvm_vcpu *vcpu) {bool halt_poll_allowed = !kvm_arch_no_poll (vcpu); -bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;ktime_t start, cur, poll_end;bool waited = false; +bool do_halt_poll;u64 halt_ns; +if (vcpu->halt_poll_ns > vcpu->kvm->max_halt_poll_ns) + vcpu->halt_poll_ns = vcpu->kvm->max_halt_poll_ns; + + do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; + start = cur = poll_end =ktime_get ();if (do_halt_poll) {ktime_t stop =ktime_add_ns (start, vcpu->halt_poll_ns);
Re: [RFC PATCH 1/3] KVM: Cap vcpu->halt_poll_ns before halting rather than after
In reply to David MatlackQuoted 37 lines · David Matlack
Cap
vcpu->halt_poll_nsbased on the max halt polling time just before halting, rather than after the last halt. This arguably provides better accuracy if an admin disables halt polling in between halts, although the improvement is nominal.A side-effect of this change is that
grow_halt_poll_ns()no longer needs to accessvcpu->kvm->max_halt_poll_ns, which will be useful in a future commit where the max halt polling time can come from the module parameterhalt_poll_nsinstead.Signed-off-by David Matlack <[email protected]>--- virt/kvm/kvm_main.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-)virt/kvm/kvm_main.c +1
-4@@ -3385,9 +3385,6 @@ static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)if (val < grow_start) val = grow_start; -if (val > vcpu->kvm->max_halt_poll_ns) - val = vcpu->kvm->max_halt_poll_ns; - vcpu->halt_poll_ns = val; out:trace_kvm_halt_poll_ns_grow (vcpu->vcpu_id, val, old);@@ -3500,11 +3497,16 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,void kvm_vcpu_halt (struct kvm_vcpu *vcpu) {bool halt_poll_allowed = !kvm_arch_no_poll (vcpu); -bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;ktime_t start, cur, poll_end;bool waited = false; +bool do_halt_poll;
virt/kvm/kvm_main.c static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, · line 3512
Quoted 10 lines
virt/kvm/kvm_main.c +5
-0u64 halt_ns; +if (vcpu->halt_poll_ns > vcpu->kvm->max_halt_poll_ns) + vcpu->halt_poll_ns = vcpu->kvm->max_halt_poll_ns; + + do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns; + start = cur = poll_end =ktime_get ();if (do_halt_poll) {ktime_t stop =ktime_add_ns (start, vcpu->halt_poll_ns);
Thanks,
Yanan
RFC PATCH 2/3 KVM: Avoid re-reading kvm->max_halt_poll_ns during halt-polling
In reply to David MatlackAvoid re-reading kvm->max_halt_poll_ns multiple times during halt-polling except when it is explicitly useful, e.g. to check if the max time changed across a halt. kvm->max_halt_poll_ns can be changed at any time by userspace via KVM_CAP_HALT_POLL.
This bug is unlikely to cause any serious side-effects. In the worst case one halt polls for shorter or longer than it should, and then is fixed up on the next halt. Furthmore, this is still possible since kvm->max_halt_poll_ns are not synchronized with halts.
--- virt/kvm/kvm_main.c | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-)
virt/kvm/kvm_main.c +15 -6
@@ -3488,6 +3488,11 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, } } +static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) +{ +return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +} +/* * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt * polling is enabled, busy wait for a short time before blocking to avoid the
@@ -3496,14 +3501,15 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,*/ void kvm_vcpu_halt (struct kvm_vcpu *vcpu) { +unsigned int max_halt_poll_ns =kvm_vcpu_max_halt_poll_ns (vcpu);bool halt_poll_allowed = !kvm_arch_no_poll (vcpu);ktime_t start, cur, poll_end;bool waited = false;bool do_halt_poll;u64 halt_ns; -if (vcpu->halt_poll_ns > vcpu->kvm->max_halt_poll_ns) - vcpu->halt_poll_ns = vcpu->kvm->max_halt_poll_ns; +if (vcpu->halt_poll_ns > max_halt_poll_ns) + vcpu->halt_poll_ns = max_halt_poll_ns; do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
@@ -3545,18 +3551,21 @@ void kvm_vcpu_halt(struct kvm_vcpu *vcpu)update_halt_poll_stats (vcpu, start, poll_end, !waited);if (halt_poll_allowed) { +/* Recompute the max halt poll time in case it changed. */ + max_halt_poll_ns =kvm_vcpu_max_halt_poll_ns (vcpu); +if (!vcpu_valid_wakeup (vcpu)) {shrink_halt_poll_ns (vcpu); - }else if (vcpu->kvm->max_halt_poll_ns) { + }else if (max_halt_poll_ns) {if (halt_ns <= vcpu->halt_poll_ns) ;/* we had a long block, shrink polling */ else if (vcpu->halt_poll_ns && - halt_ns > vcpu->kvm->max_halt_poll_ns) + halt_ns > max_halt_poll_ns)shrink_halt_poll_ns (vcpu);/* we had a short halt and our poll time is too small */ -else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && - halt_ns < vcpu->kvm->max_halt_poll_ns) +else if (vcpu->halt_poll_ns < max_halt_poll_ns && + halt_ns < max_halt_poll_ns)grow_halt_poll_ns (vcpu); }else { vcpu->halt_poll_ns =0 ;
Signature · 1 line
2.38.1.431.g37b22c650d-goog
Re: [RFC PATCH 2/3] KVM: Avoid re-reading kvm->max_halt_poll_ns during halt-polling
Quoted 8 lines · David Matlack
Avoid re-reading
kvm->max_halt_poll_nsmultiple times during halt-polling except when it is explicitly useful, e.g. to check if the max time changed across a halt.kvm->max_halt_poll_nscan be changed at any time by userspace viaKVM_CAP_HALT_POLL.This bug is unlikely to cause any serious side-effects. In the worst case one halt polls for shorter or longer than it should, and then is fixed up on the next halt. Furthmore, this is still possible since
s/Furthmore/Furthermore
Quoted 7 lines
kvm->max_halt_poll_nsare not synchronized with halts.Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>--- virt/kvm/kvm_main.c | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-)
Looks good to me:
Thanks,
Yanan
Quoted 60 lines
virt/kvm/kvm_main.c +15
-6@@ -3488,6 +3488,11 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, } } +static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) +{ +return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +} +/* * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt * polling is enabled, busy wait for a short time before blocking to avoid the @@ -3496,14 +3501,15 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,*/ void kvm_vcpu_halt (struct kvm_vcpu *vcpu) { +unsigned int max_halt_poll_ns =kvm_vcpu_max_halt_poll_ns (vcpu);bool halt_poll_allowed = !kvm_arch_no_poll (vcpu);ktime_t start, cur, poll_end;bool waited = false;bool do_halt_poll;u64 halt_ns; -if (vcpu->halt_poll_ns > vcpu->kvm->max_halt_poll_ns) - vcpu->halt_poll_ns = vcpu->kvm->max_halt_poll_ns; +if (vcpu->halt_poll_ns > max_halt_poll_ns) + vcpu->halt_poll_ns = max_halt_poll_ns; do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;@@ -3545,18 +3551,21 @@ void kvm_vcpu_halt(struct kvm_vcpu *vcpu)update_halt_poll_stats (vcpu, start, poll_end, !waited);if (halt_poll_allowed) { +/* Recompute the max halt poll time in case it changed. */ + max_halt_poll_ns =kvm_vcpu_max_halt_poll_ns (vcpu); +if (!vcpu_valid_wakeup (vcpu)) {shrink_halt_poll_ns (vcpu); - }else if (vcpu->kvm->max_halt_poll_ns) { + }else if (max_halt_poll_ns) {if (halt_ns <= vcpu->halt_poll_ns) ;/* we had a long block, shrink polling */ else if (vcpu->halt_poll_ns && - halt_ns > vcpu->kvm->max_halt_poll_ns) + halt_ns > max_halt_poll_ns)shrink_halt_poll_ns (vcpu);/* we had a short halt and our poll time is too small */ -else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && - halt_ns < vcpu->kvm->max_halt_poll_ns) +else if (vcpu->halt_poll_ns < max_halt_poll_ns && + halt_ns < max_halt_poll_ns)grow_halt_poll_ns (vcpu); }else { vcpu->halt_poll_ns =0 ;
Re: [RFC PATCH 2/3] KVM: Avoid re-reading kvm->max_halt_poll_ns during halt-polling
In reply to David MatlackAm 17.11.22 um 01:16 schrieb David Matlack:
Quoted 5 lines
Avoid re-reading
kvm->max_halt_poll_nsmultiple times during halt-polling except when it is explicitly useful, e.g. to check if the max time changed across a halt.kvm->max_halt_poll_nscan be changed at any time by userspace viaKVM_CAP_HALT_POLL.
Quoted 7 lines
This bug is unlikely to cause any serious side-effects. In the worst case one halt polls for shorter or longer than it should, and then is fixed up on the next halt. Furthmore, this is still possible since
kvm->max_halt_poll_nsare not synchronized with halts.Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>
Quoted 64 lines
--- virt/kvm/kvm_main.c | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-)virt/kvm/kvm_main.c +15
-6@@ -3488,6 +3488,11 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start, } } +static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) +{ +return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +} +/* * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc... If halt * polling is enabled, busy wait for a short time before blocking to avoid the @@ -3496,14 +3501,15 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,*/ void kvm_vcpu_halt (struct kvm_vcpu *vcpu) { +unsigned int max_halt_poll_ns =kvm_vcpu_max_halt_poll_ns (vcpu);bool halt_poll_allowed = !kvm_arch_no_poll (vcpu);ktime_t start, cur, poll_end;bool waited = false;bool do_halt_poll;u64 halt_ns; -if (vcpu->halt_poll_ns > vcpu->kvm->max_halt_poll_ns) - vcpu->halt_poll_ns = vcpu->kvm->max_halt_poll_ns; +if (vcpu->halt_poll_ns > max_halt_poll_ns) + vcpu->halt_poll_ns = max_halt_poll_ns; do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;@@ -3545,18 +3551,21 @@ void kvm_vcpu_halt(struct kvm_vcpu *vcpu)update_halt_poll_stats (vcpu, start, poll_end, !waited);if (halt_poll_allowed) { +/* Recompute the max halt poll time in case it changed. */ + max_halt_poll_ns =kvm_vcpu_max_halt_poll_ns (vcpu); +if (!vcpu_valid_wakeup (vcpu)) {shrink_halt_poll_ns (vcpu); - }else if (vcpu->kvm->max_halt_poll_ns) { + }else if (max_halt_poll_ns) {if (halt_ns <= vcpu->halt_poll_ns) ;/* we had a long block, shrink polling */ else if (vcpu->halt_poll_ns && - halt_ns > vcpu->kvm->max_halt_poll_ns) + halt_ns > max_halt_poll_ns)shrink_halt_poll_ns (vcpu);/* we had a short halt and our poll time is too small */ -else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns && - halt_ns < vcpu->kvm->max_halt_poll_ns) +else if (vcpu->halt_poll_ns < max_halt_poll_ns && + halt_ns < max_halt_poll_ns)grow_halt_poll_ns (vcpu); }else { vcpu->halt_poll_ns =0 ;
RFC PATCH 3/3 KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
In reply to David MatlackObey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL on every halt, rather than just sampling the module parameter when the VM is first created. This restore the original behavior of kvm.halt_poll_ns for VMs that have not opted into KVM_CAP_HALT_POLL.
Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using KVM_CAP_HALT_POLL.
--- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 27 ++++++++++++++++++++++++--- 2 files changed, 25 insertions(+), 3 deletions(-)
include/linux/kvm_host.h +1 -0
@@ -788,6 +788,7 @@ struct kvm {struct srcu_struct srcu;struct srcu_struct irq_srcu;pid_t userspace_pid; +bool override_halt_poll_ns;unsigned int max_halt_poll_ns;u32 dirty_ring_size;bool vm_bugged;
virt/kvm/kvm_main.c +24 -3
@@ -1198,8 +1198,6 @@ static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) goto out_err_no_arch_destroy_vm; } - kvm->max_halt_poll_ns = halt_poll_ns; - r =kvm_arch_init_vm (kvm, type);if (r) goto out_err_no_arch_destroy_vm;
@@ -3490,7 +3488,20 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) { -return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +struct kvm *kvm = vcpu->kvm; + +if (kvm->override_halt_poll_ns) { +/* +* Ensure kvm->max_halt_poll_ns is not read before +* kvm->override_halt_poll_ns. +* +* Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. +*/ +smp_rmb (); +return READ_ONCE (kvm->max_halt_poll_ns); + } + +return READ_ONCE (halt_poll_ns); }/*
@@ -4600,6 +4611,16 @@ static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,return -EINVAL ; kvm->max_halt_poll_ns = cap->args[0 ]; + +/* +* Ensure kvm->override_halt_poll_ns does not become visible +* before kvm->max_halt_poll_ns. +* +* Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). +*/ +smp_wmb (); + kvm->override_halt_poll_ns = true; +return 0 ; }case KVM_CAP_DIRTY_LOG_RING :
Signature · 1 line
2.38.1.431.g37b22c650d-goog
Re: [RFC PATCH 3/3] KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
Hi David,
Quoted 2 lines · David Matlack
Obey kvm.halt_poll_ns in VMs not using
KVM_CAP_HALT_POLLon every halt, rather than just sampling the module parameter when the VM is first
s/first/firstly
Quoted 1 line
created. This restore the original behavior of kvm.halt_poll_ns for VMs
s/restore/restores
Quoted 5 lines
that have not opted into
KVM_CAP_HALT_POLL.Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using
KVM_CAP_HALT_POLL.
Should we add more detailed comments about relationship between KVM_CAP_HALT_POLL and kvm.halt_poll_ns in Documentation/virt/kvm/api.rst? Something like:
"once KVM_CAP_HALT_POLL is used for a target VM, it will completely ignores any future changes to kvm.halt_poll_ns..."
Quoted 62 lines
Reported-by Christian Borntraeger <[email protected]>Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>--- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 27 ++++++++++++++++++++++++--- 2 files changed, 25 insertions(+), 3 deletions(-)include/linux/kvm_host.h +1
-0@@ -788,6 +788,7 @@ struct kvm {struct srcu_struct srcu;struct srcu_struct irq_srcu;pid_t userspace_pid; +bool override_halt_poll_ns;unsigned int max_halt_poll_ns;u32 dirty_ring_size;bool vm_bugged;virt/kvm/kvm_main.c +17
-3@@ -1198,8 +1198,6 @@ static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) goto out_err_no_arch_destroy_vm; } - kvm->max_halt_poll_ns = halt_poll_ns; - r =kvm_arch_init_vm (kvm, type);if (r) goto out_err_no_arch_destroy_vm;@@ -3490,7 +3488,20 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) { -return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +struct kvm *kvm = vcpu->kvm; + +if (kvm->override_halt_poll_ns) { +/* +* Ensure kvm->max_halt_poll_ns is not read before +* kvm->override_halt_poll_ns. +* +* Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. +*/ +smp_rmb (); +return READ_ONCE (kvm->max_halt_poll_ns); + } + +return READ_ONCE (halt_poll_ns); }/* @@ -4600,6 +4611,16 @@ static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,return -EINVAL ; kvm->max_halt_poll_ns = cap->args[0 ]; + +/* +* Ensure kvm->override_halt_poll_ns does not become visible
virt/kvm/kvm_main.c static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, · line 4626
Quoted 10 lines
virt/kvm/kvm_main.c +7
-0+* before kvm->max_halt_poll_ns. +* +* Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). +*/ +smp_wmb (); + kvm->override_halt_poll_ns = true; +return 0 ; }case KVM_CAP_DIRTY_LOG_RING :
Looks good to me:
Thanks,
Yanan
Re: [RFC PATCH 3/3] KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
Quoted 11 lines · wangyanan (Y)
Hi David,
Quoted 2 lines · David Matlack
Obey kvm.halt_poll_ns in VMs not using
KVM_CAP_HALT_POLLon every halt, rather than just sampling the module parameter when the VM is firsts/first/firstly
Quoted 1 line
created. This restore the original behavior of kvm.halt_poll_ns for VMs
s/restore/restores
Quoted 2 lines
that have not opted into
KVM_CAP_HALT_POLL.
Quoted 8 lines
Quoted 3 lines
Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using
KVM_CAP_HALT_POLL.Should we add more detailed comments about relationship between
KVM_CAP_HALT_POLLand kvm.halt_poll_ns inDocumentation/virt/kvm/api.rst? Something like:
"onceKVM_CAP_HALT_POLLis used for a target VM, it will completely ignores any future changes to kvm.halt_poll_ns..."
Yes we should.
I will do some testing on this series today and then resend the series as a non-RFC with the Documentation changes.
Thanks for the reviews.
Quoted 77 lines
Quoted 62 lines
Reported-by Christian Borntraeger <[email protected]>Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>--- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 27 ++++++++++++++++++++++++--- 2 files changed, 25 insertions(+), 3 deletions(-)include/linux/kvm_host.h +1
-0@@ -788,6 +788,7 @@ struct kvm {struct srcu_struct srcu;struct srcu_struct irq_srcu;pid_t userspace_pid; +bool override_halt_poll_ns;unsigned int max_halt_poll_ns;u32 dirty_ring_size;bool vm_bugged;virt/kvm/kvm_main.c +17
-3@@ -1198,8 +1198,6 @@ static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) goto out_err_no_arch_destroy_vm; } - kvm->max_halt_poll_ns = halt_poll_ns; - r =kvm_arch_init_vm (kvm, type);if (r) goto out_err_no_arch_destroy_vm;@@ -3490,7 +3488,20 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) { -return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +struct kvm *kvm = vcpu->kvm; + +if (kvm->override_halt_poll_ns) { +/* +* Ensure kvm->max_halt_poll_ns is not read before +* kvm->override_halt_poll_ns. +* +* Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. +*/ +smp_rmb (); +return READ_ONCE (kvm->max_halt_poll_ns); + } + +return READ_ONCE (halt_poll_ns); }/* @@ -4600,6 +4611,16 @@ static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,return -EINVAL ; kvm->max_halt_poll_ns = cap->args[0 ]; + +/* +* Ensure kvm->override_halt_poll_ns does not become visible virt/kvm/kvm_main.c static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, · line 4626
Quoted 10 lines
virt/kvm/kvm_main.c +7
-0+* before kvm->max_halt_poll_ns. +* +* Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). +*/ +smp_wmb (); + kvm->override_halt_poll_ns = true; +return 0 ; }case KVM_CAP_DIRTY_LOG_RING :Looks good to me:
Reviewed-by Yanan Wang <[email protected]>Thanks,
Yanan
Re: [RFC PATCH 3/3] KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
Quoted 22 lines · David Matlack
Quoted 11 lines · wangyanan (Y)
Hi David,
Quoted 2 lines · David Matlack
Obey kvm.halt_poll_ns in VMs not using
KVM_CAP_HALT_POLLon every halt, rather than just sampling the module parameter when the VM is firsts/first/firstly
Quoted 1 line
created. This restore the original behavior of kvm.halt_poll_ns for VMs
s/restore/restores
Quoted 2 lines
that have not opted into
KVM_CAP_HALT_POLL.Quoted 8 lines
Quoted 3 lines
Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using
KVM_CAP_HALT_POLL.Should we add more detailed comments about relationship between
KVM_CAP_HALT_POLLand kvm.halt_poll_ns inDocumentation/virt/kvm/api.rst? Something like:
"onceKVM_CAP_HALT_POLLis used for a target VM, it will completely ignores any future changes to kvm.halt_poll_ns..."
Quoted 6 lines
Yes we should.
I will do some testing on this series today and then resend the series as a non-RFC with the Documentation changes.
Thanks for the reviews.
Initial testing looks good but I did not have time to finish due to a bug in how our VMM is currently using KVM_CAP_HALT_POLL. I will be out all next week so I'll pick this up the week after.
Quoted 78 lines
Quoted 78 lines
Quoted 63 lines
Reported-by Christian Borntraeger <[email protected]>Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>--- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 27 ++++++++++++++++++++++++--- 2 files changed, 25 insertions(+), 3 deletions(-)include/linux/kvm_host.h +1
-0@@ -788,6 +788,7 @@ struct kvm {struct srcu_struct srcu;struct srcu_struct irq_srcu;pid_t userspace_pid; +bool override_halt_poll_ns;unsigned int max_halt_poll_ns;u32 dirty_ring_size;bool vm_bugged;virt/kvm/kvm_main.c +17
-3@@ -1198,8 +1198,6 @@ static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) goto out_err_no_arch_destroy_vm; } - kvm->max_halt_poll_ns = halt_poll_ns; - r =kvm_arch_init_vm (kvm, type);if (r) goto out_err_no_arch_destroy_vm;@@ -3490,7 +3488,20 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) { -return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +struct kvm *kvm = vcpu->kvm; + +if (kvm->override_halt_poll_ns) { +/* +* Ensure kvm->max_halt_poll_ns is not read before +* kvm->override_halt_poll_ns. +* +* Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. +*/ +smp_rmb (); +return READ_ONCE (kvm->max_halt_poll_ns); + } + +return READ_ONCE (halt_poll_ns); }/* @@ -4600,6 +4611,16 @@ static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,return -EINVAL ; kvm->max_halt_poll_ns = cap->args[0 ]; + +/* +* Ensure kvm->override_halt_poll_ns does not become visible virt/kvm/kvm_main.c static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, · line 4626
Quoted 10 lines
virt/kvm/kvm_main.c +7
-0+* before kvm->max_halt_poll_ns. +* +* Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). +*/ +smp_wmb (); + kvm->override_halt_poll_ns = true; +return 0 ; }case KVM_CAP_DIRTY_LOG_RING :Looks good to me:
Reviewed-by Yanan Wang <[email protected]>Thanks,
Yanan
Re: [RFC PATCH 3/3] KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
Quoted 22 lines · David Matlack
Quoted 21 lines · David Matlack
Quoted 10 lines · wangyanan (Y)
Hi David,
Quoted 2 lines · David Matlack
Obey kvm.halt_poll_ns in VMs not using
KVM_CAP_HALT_POLLon every halt, rather than just sampling the module parameter when the VM is firsts/first/firstly
Quoted 1 line
created. This restore the original behavior of kvm.halt_poll_ns for VMs
s/restore/restores
Quoted 2 lines
that have not opted into
KVM_CAP_HALT_POLL.Quoted 8 lines
Quoted 3 lines
Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using
KVM_CAP_HALT_POLL.Should we add more detailed comments about relationship between
KVM_CAP_HALT_POLLand kvm.halt_poll_ns inDocumentation/virt/kvm/api.rst? Something like:
"onceKVM_CAP_HALT_POLLis used for a target VM, it will completely ignores any future changes to kvm.halt_poll_ns..."Yes we should.
Quoted 7 lines
Quoted 4 lines
I will do some testing on this series today and then resend the series as a non-RFC with the Documentation changes.
Thanks for the reviews.
Initial testing looks good but I did not have time to finish due to a bug in how our VMM is currently using
KVM_CAP_HALT_POLL. I will be out all next week so I'll pick this up the week after.
OK, thanks.
Yanan
Quoted 78 lines
Quoted 75 lines
Quoted 75 lines
Quoted 62 lines
Reported-by Christian Borntraeger <[email protected]>Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>--- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 27 ++++++++++++++++++++++++--- 2 files changed, 25 insertions(+), 3 deletions(-)include/linux/kvm_host.h +1
-0@@ -788,6 +788,7 @@ struct kvm {struct srcu_struct srcu;struct srcu_struct irq_srcu;pid_t userspace_pid; +bool override_halt_poll_ns;unsigned int max_halt_poll_ns;u32 dirty_ring_size;bool vm_bugged;virt/kvm/kvm_main.c +17
-3@@ -1198,8 +1198,6 @@ static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) goto out_err_no_arch_destroy_vm; } - kvm->max_halt_poll_ns = halt_poll_ns; - r =kvm_arch_init_vm (kvm, type);if (r) goto out_err_no_arch_destroy_vm;@@ -3490,7 +3488,20 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) { -return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +struct kvm *kvm = vcpu->kvm; + +if (kvm->override_halt_poll_ns) { +/* +* Ensure kvm->max_halt_poll_ns is not read before +* kvm->override_halt_poll_ns. +* +* Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. +*/ +smp_rmb (); +return READ_ONCE (kvm->max_halt_poll_ns); + } + +return READ_ONCE (halt_poll_ns); }/* @@ -4600,6 +4611,16 @@ static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,return -EINVAL ; kvm->max_halt_poll_ns = cap->args[0 ]; + +/* +* Ensure kvm->override_halt_poll_ns does not become visible virt/kvm/kvm_main.c static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, · line 4626
Quoted 10 lines
virt/kvm/kvm_main.c +7
-0+* before kvm->max_halt_poll_ns. +* +* Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). +*/ +smp_wmb (); + kvm->override_halt_poll_ns = true; +return 0 ; }case KVM_CAP_DIRTY_LOG_RING :Looks good to me:
Reviewed-by Yanan Wang <[email protected]>Quoted 2 lines
Quoted 2 lines
Thanks,
Yanan.
Re: [RFC PATCH 3/3] KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
Quoted 24 lines · wangyanan (Y)
Quoted 22 lines · David Matlack
Quoted 21 lines · David Matlack
Quoted 10 lines · wangyanan (Y)
Hi David,
Quoted 2 lines · David Matlack
Obey kvm.halt_poll_ns in VMs not using
KVM_CAP_HALT_POLLon every halt, rather than just sampling the module parameter when the VM is firsts/first/firstly
Quoted 1 line
created. This restore the original behavior of kvm.halt_poll_ns for VMs
s/restore/restores
Quoted 2 lines
that have not opted into
KVM_CAP_HALT_POLL.Quoted 8 lines
Quoted 3 lines
Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using
KVM_CAP_HALT_POLL.Should we add more detailed comments about relationship between
KVM_CAP_HALT_POLLand kvm.halt_poll_ns inDocumentation/virt/kvm/api.rst? Something like:
"onceKVM_CAP_HALT_POLLis used for a target VM, it will completely ignores any future changes to kvm.halt_poll_ns..."Yes we should.
Quoted 8 lines
Quoted 7 lines
Quoted 4 lines
I will do some testing on this series today and then resend the series as a non-RFC with the Documentation changes.
Thanks for the reviews.
Initial testing looks good but I did not have time to finish due to a bug in how our VMM is currently using
KVM_CAP_HALT_POLL. I will be out all next week so I'll pick this up the week after.OK, thanks.
I see that Linus already merged these patches into 6.1-rc7, so I've sent the Documentation changes as a separate series.
Re: [RFC PATCH 3/3] KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
In reply to David MatlackAm 17.11.22 um 01:16 schrieb David Matlack:
Quoted 5 lines
Obey kvm.halt_poll_ns in VMs not using
KVM_CAP_HALT_POLLon every halt, rather than just sampling the module parameter when the VM is first created. This restore the original behavior of kvm.halt_poll_ns for VMs that have not opted intoKVM_CAP_HALT_POLL.
Quoted 7 lines
Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using
KVM_CAP_HALT_POLL.Reported-by Christian Borntraeger <[email protected]>Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>
One thing. This does not apply without the other 2 patches. Not sure if we want to redo this somehow to allow for stable backports?
Quoted 69 lines
--- include/linux/kvm_host.h | 1 + virt/kvm/kvm_main.c | 27 ++++++++++++++++++++++++--- 2 files changed, 25 insertions(+), 3 deletions(-)include/linux/kvm_host.h +1
-0@@ -788,6 +788,7 @@ struct kvm {struct srcu_struct srcu;struct srcu_struct irq_srcu;pid_t userspace_pid; +bool override_halt_poll_ns;unsigned int max_halt_poll_ns;u32 dirty_ring_size;bool vm_bugged;virt/kvm/kvm_main.c +24
-3@@ -1198,8 +1198,6 @@ static struct kvm *kvm_create_vm(unsigned long type, const char *fdname) goto out_err_no_arch_destroy_vm; } - kvm->max_halt_poll_ns = halt_poll_ns; - r =kvm_arch_init_vm (kvm, type);if (r) goto out_err_no_arch_destroy_vm;@@ -3490,7 +3488,20 @@ static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,static unsigned int kvm_vcpu_max_halt_poll_ns (struct kvm_vcpu *vcpu) { -return READ_ONCE (vcpu->kvm->max_halt_poll_ns); +struct kvm *kvm = vcpu->kvm; + +if (kvm->override_halt_poll_ns) { +/* +* Ensure kvm->max_halt_poll_ns is not read before +* kvm->override_halt_poll_ns. +* +* Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL. +*/ +smp_rmb (); +return READ_ONCE (kvm->max_halt_poll_ns); + } + +return READ_ONCE (halt_poll_ns); }/* @@ -4600,6 +4611,16 @@ static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,return -EINVAL ; kvm->max_halt_poll_ns = cap->args[0 ]; + +/* +* Ensure kvm->override_halt_poll_ns does not become visible +* before kvm->max_halt_poll_ns. +* +* Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns(). +*/ +smp_wmb (); + kvm->override_halt_poll_ns = true; +return 0 ; }case KVM_CAP_DIRTY_LOG_RING :
Re: [RFC PATCH 3/3] KVM: Obey kvm.halt_poll_ns in VMs not using KVM_CAP_HALT_POLL
Am 21.11.22 um 10:22 schrieb Christian Borntraeger:
Quoted 14 lines
Am 17.11.22 um 01:16 schrieb David Matlack:
Quoted 5 lines
Obey kvm.halt_poll_ns in VMs not using
KVM_CAP_HALT_POLLon every halt, rather than just sampling the module parameter when the VM is first created. This restore the original behavior of kvm.halt_poll_ns for VMs that have not opted intoKVM_CAP_HALT_POLL.Quoted 7 lines
Notably, this change restores the ability for admins to disable or change the maximum halt-polling time system wide for VMs not using
KVM_CAP_HALT_POLL.Reported-by Christian Borntraeger <[email protected]>Fixes acd05785e48c ("kvm: add capability for halt polling")Signed-off-by David Matlack <[email protected]>
Quoted 5 lines
Reviewed-by Christian Borntraeger <[email protected]>One thing. This does not apply without the other 2 patches. Not sure if we want to redo this somehow to allow for stable backports?
One option would be that Paolo, when applying uses the stable syntax as outlined in Documentation/process/stable-kernel-rules.rst
for dependencies, e.g.
Cc: <[email protected]> # 3.3.x: a1f84a3: sched: Check for idle Cc: <[email protected]> # 3.3.x: 1b9508f: sched: Rate-limit newidle Cc: <[email protected]> # 3.3.x: fd21073: sched: Fix affinity logic