static void
vm_handle_rendezvous(struct vm *vm, int vcpuid)
{
KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
mtx_lock(&vm->rendezvous_mtx);
while (vm->rendezvous_func != NULL) {
/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
if (vcpuid != -1 &&
CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
!CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
}
if (CPU_CMP(&vm->rendezvous_req_cpus,
&vm->rendezvous_done_cpus) == 0) {
VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
vm_set_rendezvous_func(vm, NULL);
wakeup(&vm->rendezvous_func);
break;
}
RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
"vmrndv", 0);
}
mtx_unlock(&vm->rendezvous_mtx);
}
struct cpu_group *
smp_topo(void)
{
char cpusetbuf[CPUSETBUFSIZ], cpusetbuf2[CPUSETBUFSIZ];
struct cpu_group *top;
/*
* Check for a fake topology request for debugging purposes.
*/
switch (smp_topology) {
case 1:
/* Dual core with no sharing. */
top = smp_topo_1level(CG_SHARE_NONE, 2, 0);
break;
case 2:
/* No topology, all cpus are equal. */
top = smp_topo_none();
break;
case 3:
/* Dual core with shared L2. */
top = smp_topo_1level(CG_SHARE_L2, 2, 0);
break;
case 4:
/* quad core, shared l3 among each package, private l2. */
top = smp_topo_1level(CG_SHARE_L3, 4, 0);
break;
case 5:
/* quad core, 2 dualcore parts on each package share l2. */
top = smp_topo_2level(CG_SHARE_NONE, 2, CG_SHARE_L2, 2, 0);
break;
case 6:
/* Single-core 2xHTT */
top = smp_topo_1level(CG_SHARE_L1, 2, CG_FLAG_HTT);
break;
case 7:
/* quad core with a shared l3, 8 threads sharing L2. */
top = smp_topo_2level(CG_SHARE_L3, 4, CG_SHARE_L2, 8,
CG_FLAG_SMT);
break;
default:
/* Default, ask the system what it wants. */
top = cpu_topo();
break;
}
/*
* Verify the returned topology.
*/
if (top->cg_count != mp_ncpus)
panic("Built bad topology at %p. CPU count %d != %d",
top, top->cg_count, mp_ncpus);
if (CPU_CMP(&top->cg_mask, &all_cpus))
panic("Built bad topology at %p. CPU mask (%s) != (%s)",
top, cpusetobj_strprint(cpusetbuf, &top->cg_mask),
cpusetobj_strprint(cpusetbuf2, &all_cpus));
return (top);
}
void
_rm_wlock(struct rmlock *rm)
{
struct rm_priotracker *prio;
struct turnstile *ts;
cpuset_t readcpus;
if (SCHEDULER_STOPPED())
return;
if (rm->lock_object.lo_flags & LO_SLEEPABLE)
sx_xlock(&rm->rm_lock_sx);
else
mtx_lock(&rm->rm_lock_mtx);
if (CPU_CMP(&rm->rm_writecpus, &all_cpus)) {
/* Get all read tokens back */
readcpus = all_cpus;
CPU_NAND(&readcpus, &rm->rm_writecpus);
rm->rm_writecpus = all_cpus;
/*
* Assumes rm->rm_writecpus update is visible on other CPUs
* before rm_cleanIPI is called.
*/
#ifdef SMP
smp_rendezvous_cpus(readcpus,
smp_no_rendevous_barrier,
rm_cleanIPI,
smp_no_rendevous_barrier,
rm);
#else
rm_cleanIPI(rm);
#endif
mtx_lock_spin(&rm_spinlock);
while ((prio = LIST_FIRST(&rm->rm_activeReaders)) != NULL) {
ts = turnstile_trywait(&rm->lock_object);
prio->rmp_flags = RMPF_ONQUEUE | RMPF_SIGNAL;
mtx_unlock_spin(&rm_spinlock);
turnstile_wait(ts, prio->rmp_thread,
TS_EXCLUSIVE_QUEUE);
mtx_lock_spin(&rm_spinlock);
}
mtx_unlock_spin(&rm_spinlock);
}
}
int
vm_reinit(struct vm *vm)
{
int error;
/*
* A virtual machine can be reset only if all vcpus are suspended.
*/
if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
vm_cleanup(vm, false);
vm_init(vm, false);
error = 0;
} else {
error = EBUSY;
}
return (error);
}
static void test_cpu_equal_case_1(void)
{
/*
* CPU_EQUAL
*/
puts( "Exercise CPU_ZERO, CPU_EQUAL, CPU_CMP, and CPU_EMPTY" );
CPU_ZERO(&set1);
CPU_ZERO(&set2);
/* test that all bits are equal */
rtems_test_assert( CPU_EQUAL(&set1, &set2) );
/* compare all bits */
rtems_test_assert( CPU_CMP(&set1, &set2) );
/* compare all bits */
rtems_test_assert( CPU_EMPTY(&set1) );
}
/*
* Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
*/
static int
vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled)
{
struct vcpu *vcpu;
const char *wmesg;
int vcpu_halted, vm_halted;
const struct timespec ts = {.tv_sec = 1, .tv_nsec = 0}; /* 1 second */
KASSERT(!CPU_ISSET(((unsigned) vcpuid), &vm->halted_cpus),
("vcpu already halted"));
vcpu = &vm->vcpu[vcpuid];
vcpu_halted = 0;
vm_halted = 0;
vcpu_lock(vcpu);
while (1) {
/*
* Do a final check for pending NMI or interrupts before
* really putting this thread to sleep. Also check for
* software events that would cause this vcpu to wakeup.
*
* These interrupts/events could have happened after the
* vcpu returned from VMRUN() and before it acquired the
* vcpu lock above.
*/
if (vm->rendezvous_func != NULL || vm->suspend)
break;
if (vm_nmi_pending(vm, vcpuid))
break;
if (!intr_disabled) {
if (vm_extint_pending(vm, vcpuid) ||
vlapic_pending_intr(vcpu->vlapic, NULL)) {
break;
}
}
/*
* Some Linux guests implement "halt" by having all vcpus
* execute HLT with interrupts disabled. 'halted_cpus' keeps
* track of the vcpus that have entered this state. When all
* vcpus enter the halted state the virtual machine is halted.
*/
if (intr_disabled) {
wmesg = "vmhalt";
VCPU_CTR0(vm, vcpuid, "Halted");
if (!vcpu_halted && halt_detection_enabled) {
vcpu_halted = 1;
CPU_SET_ATOMIC(((unsigned) vcpuid), &vm->halted_cpus);
}
if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
vm_halted = 1;
break;
}
} else {
wmesg = "vmidle";
}
//t = ticks;
vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
/*
* XXX msleep_spin() cannot be interrupted by signals so
* wake up periodically to check pending signals.
*/
pthread_mutex_lock(&vcpu->vcpu_sleep_mtx);
vcpu_unlock(vcpu);
pthread_cond_timedwait_relative_np(&vcpu->vcpu_sleep_cnd,
&vcpu->vcpu_sleep_mtx, &ts);
vcpu_lock(vcpu);
pthread_mutex_unlock(&vcpu->vcpu_sleep_mtx);
//msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
vcpu_require_state_locked(vcpu, VCPU_FROZEN);
//vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
}
if (vcpu_halted)
CPU_CLR_ATOMIC(((unsigned) vcpuid), &vm->halted_cpus);
vcpu_unlock(vcpu);
if (vm_halted)
vm_suspend(vm, VM_SUSPEND_HALT);
return (0);
}
static int
vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
{
struct vie *vie;
struct vcpu *vcpu;
struct vm_exit *vme;
uint64_t gla, gpa, cs_base;
struct vm_guest_paging *paging;
mem_region_read_t mread;
mem_region_write_t mwrite;
enum vm_cpu_mode cpu_mode;
int cs_d, error, fault, length;
vcpu = &vm->vcpu[vcpuid];
vme = &vcpu->exitinfo;
gla = vme->u.inst_emul.gla;
gpa = vme->u.inst_emul.gpa;
cs_base = vme->u.inst_emul.cs_base;
cs_d = vme->u.inst_emul.cs_d;
vie = &vme->u.inst_emul.vie;
paging = &vme->u.inst_emul.paging;
cpu_mode = paging->cpu_mode;
VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#llx", gpa);
/* Fetch, decode and emulate the faulting instruction */
if (vie->num_valid == 0) {
/*
* If the instruction length is not known then assume a
* maximum size instruction.
*/
length = vme->inst_length ? vme->inst_length : VIE_INST_SIZE;
error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
cs_base, length, vie, &fault);
} else {
/*
* The instruction bytes have already been copied into 'vie'
*/
error = fault = 0;
}
if (error || fault)
return (error);
if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#llx",
vme->rip + cs_base);
*retu = true; /* dump instruction bytes in userspace */
return (0);
}
/*
* If the instruction length was not specified then update it now
* along with 'nextrip'.
*/
if (vme->inst_length == 0) {
vme->inst_length = vie->num_processed;
vcpu->nextrip += vie->num_processed;
}
/* return to userland unless this is an in-kernel emulated device */
if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + XHYVE_PAGE_SIZE) {
mread = lapic_mmio_read;
mwrite = lapic_mmio_write;
} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
mread = vioapic_mmio_read;
mwrite = vioapic_mmio_write;
} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
mread = vhpet_mmio_read;
mwrite = vhpet_mmio_write;
} else {
*retu = true;
return (0);
}
error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
mread, mwrite, retu);
return (error);
}
static int
vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
{
int i, done;
struct vcpu *vcpu;
const struct timespec ts = {.tv_sec = 1, .tv_nsec = 0}; /* 1 second */
done = 0;
vcpu = &vm->vcpu[vcpuid];
CPU_SET_ATOMIC(((unsigned) vcpuid), &vm->suspended_cpus);
/*
* Wait until all 'active_cpus' have suspended themselves.
*
* Since a VM may be suspended at any time including when one or
* more vcpus are doing a rendezvous we need to call the rendezvous
* handler while we are waiting to prevent a deadlock.
*/
vcpu_lock(vcpu);
while (1) {
if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
break;
}
if (vm->rendezvous_func == NULL) {
VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
pthread_mutex_lock(&vcpu->vcpu_sleep_mtx);
vcpu_unlock(vcpu);
pthread_cond_timedwait_relative_np(&vcpu->vcpu_sleep_cnd,
&vcpu->vcpu_sleep_mtx, &ts);
vcpu_lock(vcpu);
pthread_mutex_unlock(&vcpu->vcpu_sleep_mtx);
//msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
vcpu_require_state_locked(vcpu, VCPU_FROZEN);
} else {
VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
vcpu_unlock(vcpu);
vm_handle_rendezvous(vm, vcpuid);
vcpu_lock(vcpu);
}
}
vcpu_unlock(vcpu);
/*
* Wakeup the other sleeping vcpus and return to userspace.
*/
for (i = 0; i < VM_MAXCPU; i++) {
if (CPU_ISSET(((unsigned) i), &vm->suspended_cpus)) {
vcpu_notify_event(vm, i, false);
}
}
*retu = true;
return (0);
}
int
vm_suspend(struct vm *vm, enum vm_suspend_how how)
{
int i;
if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
return (EINVAL);
if (atomic_cmpset_int(((volatile u_int *) &vm->suspend), 0, how) == 0) {
VM_CTR2(vm, "virtual machine already suspended %d/%d",
vm->suspend, how);
return (EALREADY);
}
VM_CTR1(vm, "virtual machine successfully suspended %d", how);
/*
* Notify all active vcpus that they are now suspended.
*/
for (i = 0; i < VM_MAXCPU; i++) {
if (CPU_ISSET(((unsigned) i), &vm->active_cpus))
vcpu_notify_event(vm, i, false);
}
return (0);
}
