[src/trunk]: src/sys/arch/xen/x86 Use hypervisor provided interface to unmask...

[cherry <cherry%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/5e3b741c72db
branches:  trunk
changeset: 445907:5e3b741c72db
user:      cherry <cherry%NetBSD.org@localhost>
date:      Sat Nov 17 05:26:46 2018 +0000

description:
Use hypervisor provided interface to unmask specific ports.

Although at first glance this looks suboptimal, the unmask operation
fast path does not use hypervisor_unmask_event(). Instead, it directly
operates on the mask and pending bit arrays to provide what would
effectively be an "auto mask/eoi" semantic.

This change is thus not in the fast path, and has the advantage of
performance improvements since cross CPU state updates etc. is handled
within the hypervisor instead of domU IPIs.

diffstat:

 sys/arch/xen/x86/hypervisor_machdep.c |  47 ++++++++--------------------------
 1 files changed, 11 insertions(+), 36 deletions(-)

diffs (75 lines):

diff -r b35c351a38a4 -r 5e3b741c72db sys/arch/xen/x86/hypervisor_machdep.c
--- a/sys/arch/xen/x86/hypervisor_machdep.c     Sat Nov 17 01:45:25 2018 +0000
+++ b/sys/arch/xen/x86/hypervisor_machdep.c     Sat Nov 17 05:26:46 2018 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: hypervisor_machdep.c,v 1.29 2018/10/26 05:33:21 cherry Exp $   */
+/*     $NetBSD: hypervisor_machdep.c,v 1.30 2018/11/17 05:26:46 cherry Exp $   */
 
 /*
  *
@@ -54,7 +54,7 @@
 
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: hypervisor_machdep.c,v 1.29 2018/10/26 05:33:21 cherry Exp $");
+__KERNEL_RCSID(0, "$NetBSD: hypervisor_machdep.c,v 1.30 2018/11/17 05:26:46 cherry Exp $");
 
 #include <sys/param.h>
 #include <sys/systm.h>
@@ -322,47 +322,22 @@
 void
 hypervisor_unmask_event(unsigned int ev)
 {
-       volatile shared_info_t *s = HYPERVISOR_shared_info;
-       CPU_INFO_ITERATOR cii;
-       struct cpu_info *ci;
-       volatile struct vcpu_info *vci;
+
+       KASSERT(ev > 0 && ev < NR_EVENT_CHANNELS);
 
 #ifdef PORT_DEBUG
        if (ev == PORT_DEBUG)
                printf("hypervisor_unmask_event %d\n", ev);
 #endif
 
-       xen_atomic_clear_bit(&s->evtchn_mask[0], ev);
-       /*
-        * The following is basically the equivalent of
-        * 'hw_resend_irq'. Just like a real IO-APIC we 'lose the
-        * interrupt edge' if the channel is masked.
-        */
-       if (!xen_atomic_test_bit(&s->evtchn_pending[0], ev))
-               return;
+       /* Xen unmasks the evtchn_mask[0]:ev bit for us. */
+       evtchn_op_t op;
+       op.cmd = EVTCHNOP_unmask;
+       op.u.unmask.port = ev;
+       if (HYPERVISOR_event_channel_op(&op) != 0)
+               panic("Failed to unmask event %d\n", ev);
 
-       for (CPU_INFO_FOREACH(cii, ci)) {
-               if (!xen_atomic_test_bit(&ci->ci_evtmask[0], ev))
-                       continue;
-               vci = ci->ci_vcpu;
-               if (__predict_true(ci == curcpu())) {
-                       if (!xen_atomic_test_and_set_bit(&vci->evtchn_pending_sel,
-                               ev>>LONG_SHIFT))
-                               xen_atomic_set_bit(&vci->evtchn_upcall_pending, 0);
-               }
-               if (!vci->evtchn_upcall_mask) {
-                       if (__predict_true(ci == curcpu())) {
-                               hypervisor_force_callback();
-                       } else {
-                               if (__predict_false(
-                                   xen_send_ipi(ci, XEN_IPI_HVCB))) {
-                                       panic("xen_send_ipi(cpu%d, "
-                                           "XEN_IPI_HVCB) failed\n",
-                                           (int) ci->ci_cpuid);
-                               }
-                       }
-               }
-       }
+       return;
 }
 
 void