KVM: s390: only deliver the set service event bits
PATCH 1/1 KVM: s390: only deliver the set service event bits
The SCLP driver code masks off the last two bits of the parameter [1] to determine if a read is required, but doesn't care about the contents of those bits. Meanwhile, the KVM code that delivers event interrupts masks off those two bits but sends both to the guest, even if only one was specified by userspace [2].
This works for the driver code, but it means any nuances of those bits gets lost. Use the event pending mask as an actual mask, and only send the bit(s) that were specified in the pending interrupt.
[1] Linux: sclp_interrupt_handler() (drivers/s390/char/sclp.c:658)
[2] QEMU: service_interrupt() (hw/s390x/sclp.c:360..363)
--- arch/s390/kvm/interrupt.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/s390/kvm/interrupt.c +2 -2
@@ -1031,7 +1031,7 @@ static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)return 0 ; } ext = fi->srv_signal; -/* only clear the event bit */ +/* only clear the event bits */ fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING ;clear_bit (IRQ_PEND_EXT_SERVICE_EV , &fi->pending_irqs);spin_unlock (&fi->lock);
@@ -1041,7 +1041,7 @@ static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)trace_kvm_s390_deliver_interrupt (vcpu->vcpu_id,KVM_S390_INT_SERVICE , ext.ext_params,0 ); -return write_sclp (vcpu,SCCB_EVENT_PENDING ); +return write_sclp (vcpu, ext.ext_params &SCCB_EVENT_PENDING ); }static int __must_check __deliver_pfault_done (struct kvm_vcpu *vcpu)
Signature · 1 line
2.40.1
Am 05.02.24 um 22:43 schrieb Eric Farman:
Quoted 10 lines
The SCLP driver code masks off the last two bits of the parameter [1] to determine if a read is required, but doesn't care about the contents of those bits. Meanwhile, the KVM code that delivers event interrupts masks off those two bits but sends both to the guest, even if only one was specified by userspace [2].
This works for the driver code, but it means any nuances of those bits gets lost. Use the event pending mask as an actual mask, and only send the bit(s) that were specified in the pending interrupt.
Quoted 5 lines
[1] Linux:
sclp_interrupt_handler()(drivers/s390/char/sclp.c:658)
[2] QEMU:service_interrupt()(hw/s390x/sclp.c:360..363)Fixes 0890ddea1a90 ("KVM: s390: protvirt: Add SCLP interrupt handling")Signed-off-by Eric Farman <[email protected]>
makes sense
Quoted 26 lines
--- arch/s390/kvm/interrupt.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/s390/kvm/interrupt.c +2
-2@@ -1031,7 +1031,7 @@ static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)return 0 ; } ext = fi->srv_signal; -/* only clear the event bit */ +/* only clear the event bits */ fi->srv_signal.ext_params &= ~SCCB_EVENT_PENDING ;clear_bit (IRQ_PEND_EXT_SERVICE_EV , &fi->pending_irqs);spin_unlock (&fi->lock);@@ -1041,7 +1041,7 @@ static int __must_check __deliver_service_ev(struct kvm_vcpu *vcpu)trace_kvm_s390_deliver_interrupt (vcpu->vcpu_id,KVM_S390_INT_SERVICE , ext.ext_params,0 ); -return write_sclp (vcpu,SCCB_EVENT_PENDING ); +return write_sclp (vcpu, ext.ext_params &SCCB_EVENT_PENDING ); }static int __must_check __deliver_pfault_done (struct kvm_vcpu *vcpu)
Quoted 6 lines · Eric Farman
The SCLP driver code masks off the last two bits of the parameter [1] to determine if a read is required, but doesn't care about the contents of those bits. Meanwhile, the KVM code that delivers event interrupts masks off those two bits but sends both to the guest, even if only one was specified by userspace [2].
Quoted 6 lines
This works for the driver code, but it means any nuances of those bits gets lost. Use the event pending mask as an actual mask, and only send the bit(s) that were specified in the pending interrupt.
[1] Linux:
sclp_interrupt_handler()(drivers/s390/char/sclp.c:658)
[2] QEMU:service_interrupt()(hw/s390x/sclp.c:360..363)
Janosch Frank
maintainerQuoted 9 lines · Eric Farman
The SCLP driver code masks off the last two bits of the parameter [1] to determine if a read is required, but doesn't care about the contents of those bits. Meanwhile, the KVM code that delivers event interrupts masks off those two bits but sends both to the guest, even if only one was specified by userspace [2].
This works for the driver code, but it means any nuances of those bits gets lost. Use the event pending mask as an actual mask, and only send the bit(s) that were specified in the pending interrupt.
Thanks, picked