On 25/10/2018 14:37, Michael Mueller wrote:
The GIB (Guest Information Block) links the GISA of all guests
that have adapter interrupts pending. These interrupts cannot be
deliverd because no vcpu of these guests is currently running in
SIE context. Instead, a GIB alert is issued on the host to schedule
these guests to run.
This mechanism allows to process adapter interrupts for currently
not running guests.
The GIB is created during host initialization and associated with
the Adapter Interruption Facility in case an Adapter Interruption
Virtualization Facility is available.
Signed-off-by: Michael Mueller <mimu@xxxxxxxxxxxxx>
---
arch/s390/include/asm/kvm_host.h | 10 ++++++++
arch/s390/kvm/interrupt.c | 43 ++++++++++++++++++++++++++++++++
arch/s390/kvm/kvm-s390.c | 1 +
arch/s390/kvm/kvm-s390.h | 2 ++
4 files changed, 56 insertions(+)
diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
index 29c940bf8506..01be3b666b11 100644
--- a/arch/s390/include/asm/kvm_host.h
+++ b/arch/s390/include/asm/kvm_host.h
@@ -776,6 +776,15 @@ struct kvm_s390_gisa {
};
};
+struct kvm_s390_gib {
+ atomic_t alert_list_origin;
+ u32 reserved01;
+ u8:5;
+ u8 nisc:3;
+ u8 reserved03[3];
+ u32 reserved04[5];
+};
+
/*
* sie_page2 has to be allocated as DMA because fac_list, crycb and
* gisa need 31bit addresses in the sie control block.
@@ -829,6 +838,7 @@ struct kvm_arch{
/* subset of available cpu features enabled by user space */
DECLARE_BITMAP(cpu_feat, KVM_S390_VM_CPU_FEAT_NR_BITS);
struct kvm_s390_gisa *gisa;
+ int gib_in_use;
};
#define KVM_HVA_ERR_BAD (-1UL)
diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c
index fcb55b02990e..3789a9ad8dee 100644
--- a/arch/s390/kvm/interrupt.c
+++ b/arch/s390/kvm/interrupt.c
@@ -31,6 +31,8 @@
#define PFAULT_DONE 0x0680
#define VIRTIO_PARAM 0x0d00
+static struct kvm_s390_gib *gib;
+
/* handle external calls via sigp interpretation facility */
static int sca_ext_call_pending(struct kvm_vcpu *vcpu, int *src_id)
{
@@ -2899,6 +2901,7 @@ void kvm_s390_gisa_init(struct kvm *kvm)
kvm->arch.gisa = &kvm->arch.sie_page2->gisa;
VM_EVENT(kvm, 3, "gisa 0x%pK initialized", kvm->arch.gisa);
kvm_s390_gisa_clear(kvm);
+ kvm->arch.gib_in_use = !!gib;
Following out answer in another thread about my questions on gib_in_use.
I do not understand why we should survive if gib is not initialized in
gib_init(), see here under, the gib_init() implenetation.
Then, if gib is never NULL, gib_in_use here only represent the kvm-guest
interrupt handling using GISA
}
}
@@ -2908,3 +2911,43 @@ void kvm_s390_gisa_destroy(struct kvm *kvm)
return;
kvm->arch.gisa = NULL;
}
+
+void kvm_s390_gib_destroy(void)
+{
+ if (!gib)
+ return;
+ chsc_sgib(0);
+ free_page((unsigned long)gib);
+ gib = NULL;
+}
+
+void kvm_s390_gib_init(u8 nisc)
+{
+ int rc;
+
+ if (gib)
+ return;
+
+ if (!css_general_characteristics.aiv) {
+ KVM_EVENT(3, "%s", "gib not initialized, no AIV facility");
+ return;
+ }
+
+ gib = (struct kvm_s390_gib *)get_zeroed_page(GFP_KERNEL | GFP_DMA);
+ if (!gib) {
+ KVM_EVENT(3, "%s", "gib memory allocation failed");
+ return;
+ }
Why should we survive this error?
+
+ gib->nisc = nisc;
+ rc = chsc_sgib((u32)(u64)gib);
+ if (rc) {
+ KVM_EVENT(3, "gib 0x%pK AIV association failed rc: %d",
+ gib, rc);
+ free_page((unsigned long)gib);
+ gib = NULL;
+ return;
+ }
or this failure ?
shouldn't we better return an error and abort loading KVM?
is there a use case for which we can continue without GIB when AIV is
available?
For my understanding, there is no case for which we can safely use GISA
without GIB.
Regards,
Pierre
--
Pierre Morel
Linux/KVM/QEMU in Böblingen - Germany
