int
acpi_sleep_machdep(struct acpi_softc *sc, int state)
{
ACPI_STATUS status;
if (sc->acpi_wakeaddr == 0ul)
return (-1); /* couldn't alloc wake memory */
#ifdef SMP
suspcpus = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &suspcpus);
#endif
if (acpi_resume_beep != 0)
timer_spkr_acquire();
AcpiSetFirmwareWakingVector(WAKECODE_PADDR(sc));
intr_suspend();
if (savectx(susppcbs[0])) {
fpususpend(suspfpusave[0]);
#ifdef SMP
if (!CPU_EMPTY(&suspcpus) &&
suspend_cpus(suspcpus) == 0) {
device_printf(sc->acpi_dev, "Failed to suspend APs\n");
return (0); /* couldn't sleep */
}
#endif
WAKECODE_FIXUP(resume_beep, uint8_t, (acpi_resume_beep != 0));
WAKECODE_FIXUP(reset_video, uint8_t, (acpi_reset_video != 0));
WAKECODE_FIXUP(wakeup_pcb, struct pcb *, susppcbs[0]);
WAKECODE_FIXUP(wakeup_fpusave, void *, suspfpusave[0]);
WAKECODE_FIXUP(wakeup_gdt, uint16_t,
susppcbs[0]->pcb_gdt.rd_limit);
WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t,
susppcbs[0]->pcb_gdt.rd_base);
WAKECODE_FIXUP(wakeup_cpu, int, 0);
/* Call ACPICA to enter the desired sleep state */
if (state == ACPI_STATE_S4 && sc->acpi_s4bios)
status = AcpiEnterSleepStateS4bios();
else
status = AcpiEnterSleepState(state);
if (status != AE_OK) {
device_printf(sc->acpi_dev,
"AcpiEnterSleepState failed - %s\n",
AcpiFormatException(status));
return (0); /* couldn't sleep */
}
for (;;)
ia32_pause();
}
return (1); /* wakeup successfully */
}
/*
* Top level routine to direct suspend/resume of a domain.
*/
void
xen_suspend_domain(void)
{
extern void rtcsync(void);
extern hrtime_t hres_last_tick;
mfn_t start_info_mfn;
ulong_t flags;
pfn_t pfn;
int i;
/*
* Check that we are happy to suspend on this hypervisor.
*/
if (xen_hypervisor_supports_solaris(XEN_SUSPEND_CHECK) == 0) {
cpr_err(CE_WARN, "Cannot suspend on this hypervisor "
"version: v%lu.%lu%s, need at least version v3.0.4 or "
"-xvm based hypervisor", XENVER_CURRENT(xv_major),
XENVER_CURRENT(xv_minor), XENVER_CURRENT(xv_ver));
return;
}
/*
* XXPV - Are we definitely OK to suspend by the time we've connected
* the handler?
*/
cpr_err(CE_NOTE, "Domain suspending for save/migrate");
SUSPEND_DEBUG("xen_suspend_domain\n");
/*
* suspend interrupts and devices
* XXPV - we use suspend/resume for both save/restore domains (like sun
* cpr) and for migration. Would be nice to know the difference if
* possible. For save/restore where down time may be a long time, we
* may want to do more of the things that cpr does. (i.e. notify user
* processes, shrink memory footprint for faster restore, etc.)
*/
xen_suspend_devices();
SUSPEND_DEBUG("xenbus_suspend\n");
xenbus_suspend();
pfn = hat_getpfnum(kas.a_hat, (caddr_t)xen_info);
start_info_mfn = pfn_to_mfn(pfn);
/*
* XXPV: cpu hotplug can hold this under a xenbus watch. Are we safe
* wrt xenbus being suspended here?
*/
mutex_enter(&cpu_lock);
/*
* Suspend must be done on vcpu 0, as no context for other CPUs is
* saved.
*
* XXPV - add to taskq API ?
*/
thread_affinity_set(curthread, 0);
kpreempt_disable();
SUSPEND_DEBUG("xen_start_migrate\n");
xen_start_migrate();
if (ncpus > 1)
suspend_cpus();
/*
* We can grab the ec_lock as it's a spinlock with a high SPL. Hence
* any holder would have dropped it to get through suspend_cpus().
*/
mutex_enter(&ec_lock);
/*
* From here on in, we can't take locks.
*/
SUSPEND_DEBUG("ec_suspend\n");
ec_suspend();
SUSPEND_DEBUG("gnttab_suspend\n");
gnttab_suspend();
flags = intr_clear();
xpv_time_suspend();
/*
* Currently, the hypervisor incorrectly fails to bring back
* powered-down VCPUs. Thus we need to record any powered-down VCPUs
* to prevent any attempts to operate on them. But we have to do this
* *after* the very first time we do ec_suspend().
*/
for (i = 1; i < ncpus; i++) {
if (cpu[i] == NULL)
continue;
if (cpu_get_state(cpu[i]) == P_POWEROFF)
CPUSET_ATOMIC_ADD(cpu_suspend_lost_set, i);
}
/*
* The dom0 save/migrate code doesn't automatically translate
* these into PFNs, but expects them to be, so we do it here.
* We don't use mfn_to_pfn() because so many OS services have
* been disabled at this point.
*/
xen_info->store_mfn = mfn_to_pfn_mapping[xen_info->store_mfn];
xen_info->console.domU.mfn =
mfn_to_pfn_mapping[xen_info->console.domU.mfn];
if (CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask == 0) {
prom_printf("xen_suspend_domain(): "
"CPU->cpu_m.mcpu_vcpu_info->evtchn_upcall_mask not set\n");
(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
}
if (HYPERVISOR_update_va_mapping((uintptr_t)HYPERVISOR_shared_info,
0, UVMF_INVLPG)) {
prom_printf("xen_suspend_domain(): "
"HYPERVISOR_update_va_mapping() failed\n");
(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
}
SUSPEND_DEBUG("HYPERVISOR_suspend\n");
/*
* At this point we suspend and sometime later resume.
*/
if (HYPERVISOR_suspend(start_info_mfn)) {
prom_printf("xen_suspend_domain(): "
"HYPERVISOR_suspend() failed\n");
(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
}
/*
* Point HYPERVISOR_shared_info to its new value.
*/
if (HYPERVISOR_update_va_mapping((uintptr_t)HYPERVISOR_shared_info,
xen_info->shared_info | PT_NOCONSIST | PT_VALID | PT_WRITABLE,
UVMF_INVLPG))
(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
if (xen_info->nr_pages != mfn_count) {
prom_printf("xen_suspend_domain(): number of pages"
" changed, was 0x%lx, now 0x%lx\n", mfn_count,
xen_info->nr_pages);
(void) HYPERVISOR_shutdown(SHUTDOWN_crash);
}
xpv_time_resume();
cached_max_mfn = 0;
SUSPEND_DEBUG("gnttab_resume\n");
gnttab_resume();
/* XXPV: add a note that this must be lockless. */
SUSPEND_DEBUG("ec_resume\n");
ec_resume();
intr_restore(flags);
if (ncpus > 1)
resume_cpus();
mutex_exit(&ec_lock);
xen_end_migrate();
mutex_exit(&cpu_lock);
/*
* Now we can take locks again.
*/
/*
* Force the tick value used for tv_nsec in hres_tick() to be up to
* date. rtcsync() will reset the hrestime value appropriately.
*/
hres_last_tick = xpv_gethrtime();
/*
* XXPV: we need to have resumed the CPUs since this takes locks, but
* can remote CPUs see bad state? Presumably yes. Should probably nest
* taking of todlock inside of cpu_lock, or vice versa, then provide an
* unlocked version. Probably need to call clkinitf to reset cpu freq
* and re-calibrate if we migrated to a different speed cpu. Also need
* to make a (re)init_cpu_info call to update processor info structs
* and device tree info. That remains to be written at the moment.
*/
rtcsync();
rebuild_mfn_list();
SUSPEND_DEBUG("xenbus_resume\n");
xenbus_resume();
SUSPEND_DEBUG("xenbus_resume_devices\n");
xen_resume_devices();
thread_affinity_clear(curthread);
kpreempt_enable();
SUSPEND_DEBUG("finished xen_suspend_domain\n");
/*
* We have restarted our suspended domain, update the hypervisor
* details. NB: This must be done at the end of this function,
* since we need the domain to be completely resumed before
* these functions will work correctly.
*/
xen_set_version(XENVER_CURRENT_IDX);
/*
* We can check and report a warning, but we don't stop the
* process.
*/
if (xen_hypervisor_supports_solaris(XEN_SUSPEND_CHECK) == 0)
cmn_err(CE_WARN, "Found hypervisor version: v%lu.%lu%s "
"but need at least version v3.0.4",
XENVER_CURRENT(xv_major), XENVER_CURRENT(xv_minor),
XENVER_CURRENT(xv_ver));
cmn_err(CE_NOTE, "domain restore/migrate completed");
}
static void
xctrl_suspend()
{
#ifdef SMP
cpuset_t cpu_suspend_map;
#endif
int suspend_cancelled;
EVENTHANDLER_INVOKE(power_suspend);
if (smp_started) {
thread_lock(curthread);
sched_bind(curthread, 0);
thread_unlock(curthread);
}
KASSERT((PCPU_GET(cpuid) == 0), ("Not running on CPU#0"));
/*
* Clear our XenStore node so the toolstack knows we are
* responding to the suspend request.
*/
xs_write(XST_NIL, "control", "shutdown", "");
/*
* Be sure to hold Giant across DEVICE_SUSPEND/RESUME since non-MPSAFE
* drivers need this.
*/
mtx_lock(&Giant);
if (DEVICE_SUSPEND(root_bus) != 0) {
mtx_unlock(&Giant);
printf("%s: device_suspend failed\n", __func__);
return;
}
mtx_unlock(&Giant);
#ifdef SMP
CPU_ZERO(&cpu_suspend_map); /* silence gcc */
if (smp_started) {
/*
* Suspend other CPUs. This prevents IPIs while we
* are resuming, and will allow us to reset per-cpu
* vcpu_info on resume.
*/
cpu_suspend_map = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &cpu_suspend_map);
if (!CPU_EMPTY(&cpu_suspend_map))
suspend_cpus(cpu_suspend_map);
}
#endif
/*
* Prevent any races with evtchn_interrupt() handler.
*/
disable_intr();
intr_suspend();
xen_hvm_suspend();
suspend_cancelled = HYPERVISOR_suspend(0);
xen_hvm_resume(suspend_cancelled != 0);
intr_resume(suspend_cancelled != 0);
enable_intr();
/*
* Reset grant table info.
*/
gnttab_resume(NULL);
#ifdef SMP
/* Send an IPI_BITMAP in case there are pending bitmap IPIs. */
lapic_ipi_vectored(IPI_BITMAP_VECTOR, APIC_IPI_DEST_ALL);
if (smp_started && !CPU_EMPTY(&cpu_suspend_map)) {
/*
* Now that event channels have been initialized,
* resume CPUs.
*/
resume_cpus(cpu_suspend_map);
}
#endif
/*
* FreeBSD really needs to add DEVICE_SUSPEND_CANCEL or
* similar.
*/
mtx_lock(&Giant);
DEVICE_RESUME(root_bus);
mtx_unlock(&Giant);
if (smp_started) {
thread_lock(curthread);
sched_unbind(curthread);
thread_unlock(curthread);
}
EVENTHANDLER_INVOKE(power_resume);
if (bootverbose)
printf("System resumed after suspension\n");
}
int
acpi_sleep_machdep(struct acpi_softc *sc, int state)
{
ACPI_STATUS status;
struct pcb *pcb;
if (sc->acpi_wakeaddr == 0ul)
return (-1); /* couldn't alloc wake memory */
#ifdef SMP
suspcpus = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &suspcpus);
#endif
if (acpi_resume_beep != 0)
timer_spkr_acquire();
AcpiSetFirmwareWakingVector(sc->acpi_wakephys, 0);
intr_suspend();
pcb = &susppcbs[0]->sp_pcb;
if (savectx(pcb)) {
#ifdef __amd64__
fpususpend(susppcbs[0]->sp_fpususpend);
#elif defined(DEV_NPX)
npxsuspend(susppcbs[0]->sp_fpususpend);
#endif
#ifdef SMP
if (!CPU_EMPTY(&suspcpus) && suspend_cpus(suspcpus) == 0) {
device_printf(sc->acpi_dev, "Failed to suspend APs\n");
return (0); /* couldn't sleep */
}
#endif
WAKECODE_FIXUP(resume_beep, uint8_t, (acpi_resume_beep != 0));
WAKECODE_FIXUP(reset_video, uint8_t, (acpi_reset_video != 0));
#ifndef __amd64__
WAKECODE_FIXUP(wakeup_cr4, register_t, pcb->pcb_cr4);
#endif
WAKECODE_FIXUP(wakeup_pcb, struct pcb *, pcb);
WAKECODE_FIXUP(wakeup_gdt, uint16_t, pcb->pcb_gdt.rd_limit);
WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t, pcb->pcb_gdt.rd_base);
/* Call ACPICA to enter the desired sleep state */
if (state == ACPI_STATE_S4 && sc->acpi_s4bios)
status = AcpiEnterSleepStateS4bios();
else
status = AcpiEnterSleepState(state);
if (ACPI_FAILURE(status)) {
device_printf(sc->acpi_dev,
"AcpiEnterSleepState failed - %s\n",
AcpiFormatException(status));
return (0); /* couldn't sleep */
}
for (;;)
ia32_pause();
} else {
#ifdef __amd64__
fpuresume(susppcbs[0]->sp_fpususpend);
#elif defined(DEV_NPX)
npxresume(susppcbs[0]->sp_fpususpend);
#endif
}
return (1); /* wakeup successfully */
}
int
acpi_sleep_machdep(struct acpi_softc *sc, int state)
{
struct savefpu *stopfpu;
#ifdef SMP
cpumask_t wakeup_cpus;
#endif
register_t cr3, rf;
ACPI_STATUS status;
int ret;
ret = -1;
if (sc->acpi_wakeaddr == 0ul)
return (ret);
#ifdef SMP
wakeup_cpus = PCPU_GET(other_cpus);
#endif
AcpiSetFirmwareWakingVector(WAKECODE_PADDR(sc));
rf = intr_disable();
intr_suspend();
/*
* Temporarily switch to the kernel pmap because it provides
* an identity mapping (setup at boot) for the low physical
* memory region containing the wakeup code.
*/
cr3 = rcr3();
load_cr3(KPML4phys);
stopfpu = &stopxpcbs[0].xpcb_pcb.pcb_save;
if (acpi_savecpu(&stopxpcbs[0])) {
fpugetregs(curthread, stopfpu);
#ifdef SMP
if (wakeup_cpus != 0 && suspend_cpus(wakeup_cpus) == 0) {
device_printf(sc->acpi_dev,
"Failed to suspend APs: CPU mask = 0x%jx\n",
(uintmax_t)(wakeup_cpus & ~stopped_cpus));
goto out;
}
#endif
WAKECODE_FIXUP(resume_beep, uint8_t, (acpi_resume_beep != 0));
WAKECODE_FIXUP(reset_video, uint8_t, (acpi_reset_video != 0));
WAKECODE_FIXUP(wakeup_xpcb, struct xpcb *, &stopxpcbs[0]);
WAKECODE_FIXUP(wakeup_gdt, uint16_t,
stopxpcbs[0].xpcb_gdt.rd_limit);
WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t,
stopxpcbs[0].xpcb_gdt.rd_base);
WAKECODE_FIXUP(wakeup_cpu, int, 0);
/* Call ACPICA to enter the desired sleep state */
if (state == ACPI_STATE_S4 && sc->acpi_s4bios)
status = AcpiEnterSleepStateS4bios();
else
status = AcpiEnterSleepState(state);
if (status != AE_OK) {
device_printf(sc->acpi_dev,
"AcpiEnterSleepState failed - %s\n",
AcpiFormatException(status));
goto out;
}
for (;;)
ia32_pause();
} else {
int
acpi_sleep_machdep(struct acpi_softc *sc, int state)
{
ACPI_STATUS status;
struct pcb *pcb;
#ifdef __amd64__
struct pcpu *pc;
int i;
#endif
if (sc->acpi_wakeaddr == 0ul)
return (-1); /* couldn't alloc wake memory */
#ifdef SMP
suspcpus = all_cpus;
CPU_CLR(PCPU_GET(cpuid), &suspcpus);
#endif
if (acpi_resume_beep != 0)
timer_spkr_acquire();
AcpiSetFirmwareWakingVector(sc->acpi_wakephys, 0);
intr_suspend();
pcb = &susppcbs[0]->sp_pcb;
if (savectx(pcb)) {
#ifdef __amd64__
fpususpend(susppcbs[0]->sp_fpususpend);
#else
npxsuspend(susppcbs[0]->sp_fpususpend);
#endif
#ifdef SMP
if (!CPU_EMPTY(&suspcpus) && suspend_cpus(suspcpus) == 0) {
device_printf(sc->acpi_dev, "Failed to suspend APs\n");
return (0); /* couldn't sleep */
}
#endif
#ifdef __amd64__
hw_ibrs_active = 0;
hw_ssb_active = 0;
cpu_stdext_feature3 = 0;
CPU_FOREACH(i) {
pc = pcpu_find(i);
pc->pc_ibpb_set = 0;
}
#endif
WAKECODE_FIXUP(resume_beep, uint8_t, (acpi_resume_beep != 0));
WAKECODE_FIXUP(reset_video, uint8_t, (acpi_reset_video != 0));
#ifdef __amd64__
WAKECODE_FIXUP(wakeup_efer, uint64_t, rdmsr(MSR_EFER) &
~(EFER_LMA));
#else
if ((amd_feature & AMDID_NX) != 0)
WAKECODE_FIXUP(wakeup_efer, uint64_t, rdmsr(MSR_EFER));
WAKECODE_FIXUP(wakeup_cr4, register_t, pcb->pcb_cr4);
#endif
WAKECODE_FIXUP(wakeup_pcb, struct pcb *, pcb);
WAKECODE_FIXUP(wakeup_gdt, uint16_t, pcb->pcb_gdt.rd_limit);
WAKECODE_FIXUP(wakeup_gdt + 2, uint64_t, pcb->pcb_gdt.rd_base);
#ifdef __i386__
/*
* Map some low memory with virt == phys for ACPI wakecode
* to use to jump to high memory after enabling paging. This
* is the same as for similar jump in locore, except the
* jump is a single instruction, and we know its address
* more precisely so only need a single PTD, and we have to
* be careful to use the kernel map (PTD[0] is for curthread
* which may be a user thread in deprecated APIs).
*/
pmap_remap_lowptdi(true);
#endif
/* Call ACPICA to enter the desired sleep state */
if (state == ACPI_STATE_S4 && sc->acpi_s4bios)
status = AcpiEnterSleepStateS4bios();
else
status = AcpiEnterSleepState(state);
if (ACPI_FAILURE(status)) {
device_printf(sc->acpi_dev,
"AcpiEnterSleepState failed - %s\n",
AcpiFormatException(status));
return (0); /* couldn't sleep */
}
if (acpi_susp_bounce)
resumectx(pcb);
for (;;)
ia32_pause();
} else {
