On 2/23/23 17:22, Nico Boehr wrote:
We sometimes put a virtual address in next_alert, which should always be
a physical address, since it is shared with hardware.
This currently works, because virtual and physical addresses are
the same.
I'd replace that with something like:
The gisa next alert address is defined as a host absolute address so
let's use virt_to_phys() to make sure we always write an absolute
address to this hardware structure.
This is not a bug since we're currently still running as a virtual ==
physical kernel but plan to move away from that.
Add phys_to_virt() to resolve the virtual-physical confusion.
Reviewed-by: Janosch Frank <frankja@xxxxxxxxxxxxx>
Signed-off-by: Nico Boehr <nrb@xxxxxxxxxxxxx>
---
arch/s390/kvm/interrupt.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c
index ab26aa53ee37..20743c5b000a 100644
--- a/arch/s390/kvm/interrupt.c
+++ b/arch/s390/kvm/interrupt.c
@@ -305,7 +305,7 @@ static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
{
- return READ_ONCE(gisa->next_alert) != (u32)(u64)gisa;
+ return READ_ONCE(gisa->next_alert) != (u32)virt_to_phys(gisa);
}
static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
@@ -3167,7 +3167,7 @@ void kvm_s390_gisa_init(struct kvm *kvm)
hrtimer_init(&gi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
gi->timer.function = gisa_vcpu_kicker;
memset(gi->origin, 0, sizeof(struct kvm_s390_gisa));
- gi->origin->next_alert = (u32)(u64)gi->origin;
+ gi->origin->next_alert = (u32)virt_to_phys(gi->origin);
VM_EVENT(kvm, 3, "gisa 0x%pK initialized", gi->origin);
}
