13275 bhyve needs richer INIT/SIPI support
Reviewed by: Robert Mustacchi <rm@fingolfin.org>
Approved by: Gordon Ross <gordon.w.ross@gmail.com>

*** 516,532 ****
  #ifdef __FreeBSD__
  void
  fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
  #else
  void
! fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip,
!     bool suspend)
  #endif
  {
          int error;
  
          assert(fromcpu == BSP);
  
          /*
           * The 'newcpu' must be activated in the context of 'fromcpu'. If
           * vm_activate_cpu() is delayed until newcpu's pthread starts running
           * then vmm.ko is out-of-sync with bhyve and this can create a race
--- 516,533 ----
  #ifdef __FreeBSD__
  void
  fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
  #else
  void
! fbsdrun_addcpu(struct vmctx *ctx, int newcpu, uint64_t rip, bool suspend)
  #endif
  {
          int error;
  
+ #ifdef __FreeBSD__
          assert(fromcpu == BSP);
+ #endif
  
          /*
           * The 'newcpu' must be activated in the context of 'fromcpu'. If
           * vm_activate_cpu() is delayed until newcpu's pthread starts running
           * then vmm.ko is out-of-sync with bhyve and this can create a race
*** 575,585 ****
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_mmio *mmio = &entry->u.mmio;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_COMPLETE_MMIO;
          mmio->bytes = bytes;
          mmio->read = 1;
          mmio->gpa = gpa;
          mmio->data = data;
  }
--- 576,586 ----
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_mmio *mmio = &entry->u.mmio;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_FULFILL_MMIO;
          mmio->bytes = bytes;
          mmio->read = 1;
          mmio->gpa = gpa;
          mmio->data = data;
  }
*** 590,600 ****
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_mmio *mmio = &entry->u.mmio;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_COMPLETE_MMIO;
          mmio->bytes = bytes;
          mmio->read = 0;
          mmio->gpa = gpa;
          mmio->data = 0;
  }
--- 591,601 ----
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_mmio *mmio = &entry->u.mmio;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_FULFILL_MMIO;
          mmio->bytes = bytes;
          mmio->read = 0;
          mmio->gpa = gpa;
          mmio->data = 0;
  }
*** 605,615 ****
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_inout *inout = &entry->u.inout;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_COMPLETE_INOUT;
          inout->bytes = bytes;
          inout->flags = INOUT_IN;
          inout->port = port;
          inout->eax = data;
  }
--- 606,616 ----
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_inout *inout = &entry->u.inout;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_FULFILL_INOUT;
          inout->bytes = bytes;
          inout->flags = INOUT_IN;
          inout->port = port;
          inout->eax = data;
  }
*** 620,630 ****
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_inout *inout = &entry->u.inout;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_COMPLETE_INOUT;
          inout->bytes = bytes;
          inout->flags = 0;
          inout->port = port;
          inout->eax = 0;
  }
--- 621,631 ----
          struct vm_entry *entry = &vmentry[vcpu];
          struct vm_inout *inout = &entry->u.inout;
  
          assert(entry->cmd == VEC_DEFAULT);
  
!         entry->cmd = VEC_FULFILL_INOUT;
          inout->bytes = bytes;
          inout->flags = 0;
          inout->port = port;
          inout->eax = 0;
  }
*** 729,747 ****
--- 730,761 ----
                  }
          }
          return (VMEXIT_CONTINUE);
  }
  
+ #ifdef __FreeBSD__
  static int
  vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
  {
  
          (void)spinup_ap(ctx, *pvcpu,
                      vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip);
  
          return (VMEXIT_CONTINUE);
  }
+ #else
+ static int
+ vmexit_run_state(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
+ {
+         /*
+          * Run-state transitions (INIT, SIPI, etc) are handled in-kernel, so an
+          * exit to userspace with that code is not expected.
+          */
+         fprintf(stderr, "unexpected run-state VM exit");
+         return (VMEXIT_ABORT);
+ }
+ #endif /* __FreeBSD__ */
  
  #ifdef __FreeBSD__
  #define DEBUG_EPT_MISCONFIG
  #else
  /* EPT misconfig debugging not possible now that raw VMCS access is gone */
*** 1015,1025 ****
--- 1029,1043 ----
          [VM_EXITCODE_REQIDLE] = vmexit_reqidle,
          [VM_EXITCODE_RDMSR]  = vmexit_rdmsr,
          [VM_EXITCODE_WRMSR]  = vmexit_wrmsr,
          [VM_EXITCODE_MTRAP]  = vmexit_mtrap,
          [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
+ #ifdef __FreeBSD__
          [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap,
+ #else
+         [VM_EXITCODE_RUN_STATE] = vmexit_run_state,
+ #endif
          [VM_EXITCODE_SUSPENDED] = vmexit_suspend,
          [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
          [VM_EXITCODE_DEBUG] = vmexit_debug,
          [VM_EXITCODE_BPT] = vmexit_breakpoint,
  };
*** 1545,1562 ****
  
          if (caph_enter() == -1)
                  errx(EX_OSERR, "cap_enter() failed");
  #endif
  
          /*
           * Add CPU 0
           */
- #ifdef __FreeBSD__
          fbsdrun_addcpu(ctx, BSP, BSP, rip);
  #else
!         fbsdrun_addcpu(ctx, BSP, BSP, rip, suspend);
  
          mark_provisioned();
  #endif
  
          /*
           * Head off to the main event dispatch loop
--- 1563,1587 ----
  
          if (caph_enter() == -1)
                  errx(EX_OSERR, "cap_enter() failed");
  #endif
  
+ #ifdef __FreeBSD__
          /*
           * Add CPU 0
           */
          fbsdrun_addcpu(ctx, BSP, BSP, rip);
  #else
!         /* Set BSP to run (unlike the APs which wait for INIT) */
!         error = vm_set_run_state(ctx, BSP, VRS_RUN, 0);
!         assert(error == 0);
!         fbsdrun_addcpu(ctx, BSP, rip, suspend);
  
+         /* Add subsequent CPUs, which will wait until INIT/SIPI-ed */
+         for (uint_t i = 1; i < guest_ncpus; i++) {
+                 spinup_halted_ap(ctx, i);
+         }
          mark_provisioned();
  #endif
  
          /*
           * Head off to the main event dispatch loop