hw/xen: Implement EVTCHNOP_unmask

This finally comes with a mechanism for actually injecting events into
the guest vCPU, with all the atomic-test-and-set that's involved in
setting the bit in the shinfo, then the index in the vcpu_info, and
injecting either the lapic vector as MSI, or letting KVM inject the
bare vector.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Reviewed-by: Paul Durrant <paul@xen.org>

1 files changed, 175 insertions, 0 deletions

diff --git a/hw/i386/kvm/xen_evtchn.c b/hw/i386/kvm/xen_evtchn.c
index c57d36b492..54707f1f9f 100644
--- a/hw/i386/kvm/xen_evtchn.c
+++ b/hw/i386/kvm/xen_evtchn.c
@@ -224,6 +224,13 @@ int xen_evtchn_set_callback_param(uint64_t param)
     return ret;
 }
 
+static void inject_callback(XenEvtchnState *s, uint32_t vcpu)
+{
+    int type = s->callback_param >> CALLBACK_VIA_TYPE_SHIFT;
+
+    kvm_xen_inject_vcpu_callback_vector(vcpu, type);
+}
+
 static bool valid_port(evtchn_port_t port)
 {
     if (!port) {
@@ -294,6 +301,152 @@ int xen_evtchn_status_op(struct evtchn_status *status)
     return 0;
 }
 
+/*
+ * Never thought I'd hear myself say this, but C++ templates would be
+ * kind of nice here.
+ *
+ * template<class T> static int do_unmask_port(T *shinfo, ...);
+ */
+static int do_unmask_port_lm(XenEvtchnState *s, evtchn_port_t port,
+                             bool do_unmask, struct shared_info *shinfo,
+                             struct vcpu_info *vcpu_info)
+{
+    const int bits_per_word = BITS_PER_BYTE * sizeof(shinfo->evtchn_pending[0]);
+    typeof(shinfo->evtchn_pending[0]) mask;
+    int idx = port / bits_per_word;
+    int offset = port % bits_per_word;
+
+    mask = 1UL << offset;
+
+    if (idx >= bits_per_word) {
+        return -EINVAL;
+    }
+
+    if (do_unmask) {
+        /*
+         * If this is a true unmask operation, clear the mask bit. If
+         * it was already unmasked, we have nothing further to do.
+         */
+        if (!((qatomic_fetch_and(&shinfo->evtchn_mask[idx], ~mask) & mask))) {
+            return 0;
+        }
+    } else {
+        /*
+         * This is a pseudo-unmask for affinity changes. We don't
+         * change the mask bit, and if it's *masked* we have nothing
+         * else to do.
+         */
+        if (qatomic_fetch_or(&shinfo->evtchn_mask[idx], 0) & mask) {
+            return 0;
+        }
+    }
+
+    /* If the event was not pending, we're done. */
+    if (!(qatomic_fetch_or(&shinfo->evtchn_pending[idx], 0) & mask)) {
+        return 0;
+    }
+
+    /* Now on to the vcpu_info evtchn_pending_sel index... */
+    mask = 1UL << idx;
+
+    /* If a port in this word was already pending for this vCPU, all done. */
+    if (qatomic_fetch_or(&vcpu_info->evtchn_pending_sel, mask) & mask) {
+        return 0;
+    }
+
+    /* Set evtchn_upcall_pending for this vCPU */
+    if (qatomic_fetch_or(&vcpu_info->evtchn_upcall_pending, 1)) {
+        return 0;
+    }
+
+    inject_callback(s, s->port_table[port].vcpu);
+
+    return 0;
+}
+
+static int do_unmask_port_compat(XenEvtchnState *s, evtchn_port_t port,
+                                 bool do_unmask,
+                                 struct compat_shared_info *shinfo,
+                                 struct compat_vcpu_info *vcpu_info)
+{
+    const int bits_per_word = BITS_PER_BYTE * sizeof(shinfo->evtchn_pending[0]);
+    typeof(shinfo->evtchn_pending[0]) mask;
+    int idx = port / bits_per_word;
+    int offset = port % bits_per_word;
+
+    mask = 1UL << offset;
+
+    if (idx >= bits_per_word) {
+        return -EINVAL;
+    }
+
+    if (do_unmask) {
+        /*
+         * If this is a true unmask operation, clear the mask bit. If
+         * it was already unmasked, we have nothing further to do.
+         */
+        if (!((qatomic_fetch_and(&shinfo->evtchn_mask[idx], ~mask) & mask))) {
+            return 0;
+        }
+    } else {
+        /*
+         * This is a pseudo-unmask for affinity changes. We don't
+         * change the mask bit, and if it's *masked* we have nothing
+         * else to do.
+         */
+        if (qatomic_fetch_or(&shinfo->evtchn_mask[idx], 0) & mask) {
+            return 0;
+        }
+    }
+
+    /* If the event was not pending, we're done. */
+    if (!(qatomic_fetch_or(&shinfo->evtchn_pending[idx], 0) & mask)) {
+        return 0;
+    }
+
+    /* Now on to the vcpu_info evtchn_pending_sel index... */
+    mask = 1UL << idx;
+
+    /* If a port in this word was already pending for this vCPU, all done. */
+    if (qatomic_fetch_or(&vcpu_info->evtchn_pending_sel, mask) & mask) {
+        return 0;
+    }
+
+    /* Set evtchn_upcall_pending for this vCPU */
+    if (qatomic_fetch_or(&vcpu_info->evtchn_upcall_pending, 1)) {
+        return 0;
+    }
+
+    inject_callback(s, s->port_table[port].vcpu);
+
+    return 0;
+}
+
+static int unmask_port(XenEvtchnState *s, evtchn_port_t port, bool do_unmask)
+{
+    void *vcpu_info, *shinfo;
+
+    if (s->port_table[port].type == EVTCHNSTAT_closed) {
+        return -EINVAL;
+    }
+
+    shinfo = xen_overlay_get_shinfo_ptr();
+    if (!shinfo) {
+        return -ENOTSUP;
+    }
+
+    vcpu_info = kvm_xen_get_vcpu_info_hva(s->port_table[port].vcpu);
+    if (!vcpu_info) {
+        return -EINVAL;
+    }
+
+    if (xen_is_long_mode()) {
+        return do_unmask_port_lm(s, port, do_unmask, shinfo, vcpu_info);
+    } else {
+        return do_unmask_port_compat(s, port, do_unmask, shinfo, vcpu_info);
+    }
+}
+
 static int clear_port_pending(XenEvtchnState *s, evtchn_port_t port)
 {
     void *p = xen_overlay_get_shinfo_ptr();
@@ -380,3 +533,25 @@ int xen_evtchn_close_op(struct evtchn_close *close)
 
     return ret;
 }
+
+int xen_evtchn_unmask_op(struct evtchn_unmask *unmask)
+{
+    XenEvtchnState *s = xen_evtchn_singleton;
+    int ret;
+
+    if (!s) {
+        return -ENOTSUP;
+    }
+
+    if (!valid_port(unmask->port)) {
+        return -EINVAL;
+    }
+
+    qemu_mutex_lock(&s->port_lock);
+
+    ret = unmask_port(s, unmask->port, true);
+
+    qemu_mutex_unlock(&s->port_lock);
+
+    return ret;
+}