KVM: s390: add function process_gib_alert_list()
Quoted 28 lines · Michael Mueller
This function processes a gib alert list. It is required to run when either a gib alert interruption has been received or a gisa that might be in the alert list is cleared or dropped.
Signed-off-by Michael Mueller <[email protected]>--- arch/s390/kvm/interrupt.c | 38 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 38 insertions(+)arch/s390/kvm/interrupt.c +11
-0@@ -2900,6 +2900,44 @@ static void nullify_gisa(struct kvm_s390_gisa *gisa) gisa->next_alert = (u32 )(u64 )gisa; } +/* +* Before processing, the gib alert list needs to be cut-off from +* the gib by means of function unlink_gib_alert_list(). If non NULL, +* the list is processed from its latest to oldest entry. +* +* Processing an gisa entry needs to wake-up a vcpu of the kvm this gisa +* belongs to. Thus, the pending guest interruption will be processed +* in SIE context. +* +* Whenever a gisa is cleared (e.g. on vm reset) or a gisa is dropped +* (e.g. on vm termination) it might be part of the gib alert list.
arch/s390/kvm/interrupt.c static void nullify_gisa(struct kvm_s390_gisa *gisa) · line 2923
Quoted 10 lines
arch/s390/kvm/interrupt.c +10
-0+* Thus, these operations need to process the alert list as well. +*/ +static void __maybe_unused process_gib_alert_list ( +struct kvm_s390_gisa *gisa, +struct kvm_s390_gisa *gisa_to_nullify, +struct kvm_s390_gisa *gisa_to_drop) +{ +struct kvm_s390_gisa *next_alert; +struct kvm *kvm; +
Isn't this function a little bit complex?
I think we can make it clearer.
- The first argument is always AFAIU unlink_gib_alert_list(), then why not just call this function inside of process_gib_alert_list() ?
- The action to do is related to the KVM/GISA, may be some state inside KVM or GISA to differentiate the processing?
All the GISA in this list need to be handled anyway if we are coming from an IRQ, a CLEAR_IRQ or a RESET but...
arch/s390/kvm/interrupt.c static void nullify_gisa(struct kvm_s390_gisa *gisa) · line 2931
Quoted 8 lines
arch/s390/kvm/interrupt.c +8
-0+for (; gisa; gisa = next_alert) { + next_alert = (struct kvm_s390_gisa *)(u64 )gisa->next_alert; +/* unlink from alert list */ + gisa->next_alert = (u32 )(u64 )gisa; +/* skip if to clear or drop */ +if (gisa == gisa_to_nullify || + gisa == gisa_to_drop) +continue ;
... here:
I am not sure what happens if two guest are being reset at a time, it seems to me that one is reset and one get kicked.
arch/s390/kvm/interrupt.c static void nullify_gisa(struct kvm_s390_gisa *gisa) · line 2934
Quoted 3 lines
arch/s390/kvm/interrupt.c +3
-0+/* wake-up a vcpu of the kvm this gisa belongs to */ + kvm = container_of(gisa, struct sie_page2, gisa)->kvm; +__floating_irq_kick (kvm,KVM_S390_INT_IO (1 ,0 ,0 ,0 ));
Do we really need to insert a floating interrupt?
Wouldn't a simple kvm_s390_vcpu_wakeup() after choosing the vcpu be enough?
arch/s390/kvm/interrupt.c static void nullify_gisa(struct kvm_s390_gisa *gisa) · line 2938
Quoted 4 lines
arch/s390/kvm/interrupt.c +4
-0+ } + +if (gisa_to_nullify) +nullify_gisa (gisa_to_nullify);
we already reset the next_alert pointer
if we want to clear the interrupt we just need to clear IPM
Quoted 6 lines
arch/s390/kvm/interrupt.c +2
-0+} +void kvm_s390_gisa_clear (struct kvm *kvm) {if (kvm->arch.gisa) {
Signature · 2 lines
Pierre Morel Linux/KVM/QEMU in B�blingen - Germany