int
acquire_sys(struct snapshot *ss, kstat_ctl_t *kc)
{
size_t i;
kstat_named_t *knp;
kstat_t *ksp;
if ((ksp = kstat_lookup(kc, "unix", 0, "sysinfo")) == NULL)
return (errno);
if (kstat_read(kc, ksp, &ss->s_sys.ss_sysinfo) == -1)
return (errno);
if ((ksp = kstat_lookup(kc, "unix", 0, "vminfo")) == NULL)
return (errno);
if (kstat_read(kc, ksp, &ss->s_sys.ss_vminfo) == -1)
return (errno);
if ((ksp = kstat_lookup(kc, "unix", 0, "dnlcstats")) == NULL)
return (errno);
if (kstat_read(kc, ksp, &ss->s_sys.ss_nc) == -1)
return (errno);
if ((ksp = kstat_lookup(kc, "unix", 0, "system_misc")) == NULL)
return (errno);
if (kstat_read(kc, ksp, NULL) == -1)
return (errno);
knp = (kstat_named_t *)kstat_data_lookup(ksp, "clk_intr");
if (knp == NULL)
return (errno);
ss->s_sys.ss_ticks = knp->value.l;
knp = (kstat_named_t *)kstat_data_lookup(ksp, "deficit");
if (knp == NULL)
return (errno);
ss->s_sys.ss_deficit = knp->value.l;
for (i = 0; i < ss->s_nr_cpus; i++) {
if (!CPU_ACTIVE(&ss->s_cpus[i]))
continue;
if (kstat_add(&ss->s_cpus[i].cs_sys, &ss->s_sys.ss_agg_sys))
return (errno);
if (kstat_add(&ss->s_cpus[i].cs_vm, &ss->s_sys.ss_agg_vm))
return (errno);
ss->s_nr_active_cpus++;
}
return (0);
}
/*
* Jump to the fast reboot switcher. This function never returns.
*/
void
fast_reboot()
{
processorid_t bootcpuid = 0;
extern uintptr_t postbootkernelbase;
extern char fb_swtch_image[];
fastboot_file_t *fb;
int i;
postbootkernelbase = 0;
fb = &newkernel.fi_files[FASTBOOT_SWTCH];
/*
* Map the address into both the current proc's address
* space and the kernel's address space in case the panic
* is forced by kmdb.
*/
if (&kas != curproc->p_as) {
hat_devload(curproc->p_as->a_hat, (caddr_t)fb->fb_va,
MMU_PAGESIZE, mmu_btop(fb->fb_dest_pa),
PROT_READ | PROT_WRITE | PROT_EXEC,
HAT_LOAD_NOCONSIST | HAT_LOAD_LOCK);
}
bcopy((void *)fb_swtch_image, (void *)fb->fb_va, fb->fb_size);
/*
* Set fb_va to fake_va
*/
for (i = 0; i < FASTBOOT_MAX_FILES_MAP; i++) {
newkernel.fi_files[i].fb_va = fake_va;
}
if (panicstr && CPU->cpu_id != bootcpuid &&
CPU_ACTIVE(cpu_get(bootcpuid))) {
extern void panic_idle(void);
cpuset_t cpuset;
CPUSET_ZERO(cpuset);
CPUSET_ADD(cpuset, bootcpuid);
xc_priority((xc_arg_t)&newkernel, 0, 0, CPUSET2BV(cpuset),
(xc_func_t)fastboot_xc_func);
panic_idle();
} else
(void) fastboot_xc_func(&newkernel, 0, 0);
}
/*
* Class callback when a CPU is actually moving partitions
*/
static void
pg_cmt_cpupart_move(cpu_t *cp, cpupart_t *oldpp, cpupart_t *newpp)
{
cpu_t *cpp;
group_t *pgs;
pg_t *pg;
group_iter_t pg_iter;
pg_cpu_itr_t cpu_iter;
boolean_t found;
ASSERT(MUTEX_HELD(&cpu_lock));
if (cmt_sched_disabled)
return;
pgs = &cp->cpu_pg->pgs;
group_iter_init(&pg_iter);
/*
* Iterate over the CPUs CMT PGs
*/
while ((pg = group_iterate(pgs, &pg_iter)) != NULL) {
if (IS_CMT_PG(pg) == 0)
continue;
/*
* Add the PG to the bitset in the new partition.
*/
bitset_add(&newpp->cp_cmt_pgs, pg->pg_id);
/*
* Remove the PG from the bitset in the old partition
* if the last of the PG's CPUs have left.
*/
found = B_FALSE;
PG_CPU_ITR_INIT(pg, cpu_iter);
while ((cpp = pg_cpu_next(&cpu_iter)) != NULL) {
if (cpp == cp)
continue;
if (CPU_ACTIVE(cpp) &&
cpp->cpu_part->cp_id == oldpp->cp_id) {
found = B_TRUE;
break;
}
}
if (!found)
bitset_del(&cp->cpu_part->cp_cmt_pgs, pg->pg_id);
}
}
static void
cpr_save_mp_state(void)
{
cpu_t *cp;
ASSERT(MUTEX_HELD(&cpu_lock));
cp = cpu_list;
do {
cp->cpu_cpr_flags &= ~CPU_CPR_ONLINE;
if (CPU_ACTIVE(cp))
CPU_SET_CPR_FLAGS(cp, CPU_CPR_ONLINE);
} while ((cp = cp->cpu_next) != cpu_list);
}
static int
acquire_cpus(struct snapshot *ss, kstat_ctl_t *kc)
{
size_t i;
ss->s_nr_cpus = sysconf(_SC_CPUID_MAX) + 1;
ss->s_cpus = calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot));
if (ss->s_cpus == NULL)
goto out;
for (i = 0; i < ss->s_nr_cpus; i++) {
kstat_t *ksp;
ss->s_cpus[i].cs_id = ID_NO_CPU;
ss->s_cpus[i].cs_state = p_online(i, P_STATUS);
/* If no valid CPU is present, move on to the next one */
if (ss->s_cpus[i].cs_state == -1)
continue;
ss->s_cpus[i].cs_id = i;
if ((ksp = kstat_lookup_read(kc, "cpu_info", i, NULL)) == NULL)
goto out;
(void) pset_assign(PS_QUERY, i, &ss->s_cpus[i].cs_pset_id);
if (ss->s_cpus[i].cs_pset_id == PS_NONE)
ss->s_cpus[i].cs_pset_id = ID_NO_PSET;
if (!CPU_ACTIVE(&ss->s_cpus[i]))
continue;
if ((ksp = kstat_lookup_read(kc, "cpu", i, "vm")) == NULL)
goto out;
if (kstat_copy(ksp, &ss->s_cpus[i].cs_vm))
goto out;
if ((ksp = kstat_lookup_read(kc, "cpu", i, "sys")) == NULL)
goto out;
if (kstat_copy(ksp, &ss->s_cpus[i].cs_sys))
goto out;
}
errno = 0;
out:
return (errno);
}
int
cpr_mp_online(void)
{
cpu_t *cp, *bootcpu = CPU;
int rc = 0;
/*
* Do nothing for UP.
*/
if (ncpus == 1)
return (0);
/*
* cpr_save_mp_state() sets CPU_CPR_ONLINE in cpu_cpr_flags
* to indicate a cpu was online at the time of cpr_suspend();
* now restart those cpus that were marked as CPU_CPR_ONLINE
* and actually are offline.
*/
mutex_enter(&cpu_lock);
for (cp = bootcpu->cpu_next; cp != bootcpu; cp = cp->cpu_next) {
/*
* Clear the CPU_FROZEN flag in all cases.
*/
cp->cpu_flags &= ~CPU_FROZEN;
if (CPU_CPR_IS_OFFLINE(cp))
continue;
if (CPU_ACTIVE(cp))
continue;
if ((rc = cpr_p_online(cp, CPU_CPR_ONLINE))) {
mutex_exit(&cpu_lock);
return (rc);
}
}
/*
* turn off the boot cpu if it was offlined
*/
if (CPU_CPR_IS_OFFLINE(bootcpu)) {
if ((rc = cpr_p_online(bootcpu, CPU_CPR_OFFLINE))) {
mutex_exit(&cpu_lock);
return (rc);
}
}
mutex_exit(&cpu_lock);
return (0);
}
/*
* CPU ONLINE/OFFLINE CODE
*/
int
cpr_mp_offline(void)
{
cpu_t *cp, *bootcpu;
int rc = 0;
int brought_up_boot = 0;
/*
* Do nothing for UP.
*/
if (ncpus == 1)
return (0);
mutex_enter(&cpu_lock);
cpr_save_mp_state();
bootcpu = i_cpr_bootcpu();
if (!CPU_ACTIVE(bootcpu)) {
if ((rc = cpr_p_online(bootcpu, CPU_CPR_ONLINE))) {
mutex_exit(&cpu_lock);
return (rc);
}
brought_up_boot = 1;
}
cp = cpu_list;
do {
if (cp == bootcpu)
continue;
if (cp->cpu_flags & CPU_OFFLINE)
continue;
if ((rc = cpr_p_online(cp, CPU_CPR_OFFLINE))) {
mutex_exit(&cpu_lock);
return (rc);
}
} while ((cp = cp->cpu_next) != cpu_list);
if (brought_up_boot && (cpr_debug & (CPR_DEBUG1 | CPR_DEBUG6)))
prom_printf("changed cpu %p to state %d\n",
(void *)bootcpu, CPU_CPR_ONLINE);
mutex_exit(&cpu_lock);
return (rc);
}
/*ARGSUSED*/
void
mdboot(int cmd, int fcn, char *mdep, boolean_t invoke_cb)
{
processorid_t bootcpuid = 0;
static int is_first_quiesce = 1;
static int is_first_reset = 1;
int reset_status = 0;
static char fallback_str[] = "Falling back to regular reboot.\n";
if (fcn == AD_FASTREBOOT && !newkernel.fi_valid)
fcn = AD_BOOT;
if (!panicstr) {
kpreempt_disable();
if (fcn == AD_FASTREBOOT) {
mutex_enter(&cpu_lock);
if (CPU_ACTIVE(cpu_get(bootcpuid))) {
affinity_set(bootcpuid);
}
mutex_exit(&cpu_lock);
} else {
affinity_set(CPU_CURRENT);
}
}
if (force_shutdown_method != AD_UNKNOWN)
fcn = force_shutdown_method;
/*
* XXX - rconsvp is set to NULL to ensure that output messages
* are sent to the underlying "hardware" device using the
* monitor's printf routine since we are in the process of
* either rebooting or halting the machine.
*/
rconsvp = NULL;
/*
* Print the reboot message now, before pausing other cpus.
* There is a race condition in the printing support that
* can deadlock multiprocessor machines.
*/
if (!(fcn == AD_HALT || fcn == AD_POWEROFF))
prom_printf("rebooting...\n");
if (IN_XPV_PANIC())
reset();
/*
* We can't bring up the console from above lock level, so do it now
*/
pm_cfb_check_and_powerup();
/* make sure there are no more changes to the device tree */
devtree_freeze();
if (invoke_cb)
(void) callb_execute_class(CB_CL_MDBOOT, NULL);
/*
* Clear any unresolved UEs from memory.
*/
page_retire_mdboot();
#if defined(__xpv)
/*
* XXPV Should probably think some more about how we deal
* with panicing before it's really safe to panic.
* On hypervisors, we reboot very quickly.. Perhaps panic
* should only attempt to recover by rebooting if,
* say, we were able to mount the root filesystem,
* or if we successfully launched init(1m).
*/
if (panicstr && proc_init == NULL)
(void) HYPERVISOR_shutdown(SHUTDOWN_poweroff);
#endif
/*
* stop other cpus and raise our priority. since there is only
* one active cpu after this, and our priority will be too high
* for us to be preempted, we're essentially single threaded
* from here on out.
*/
(void) spl6();
if (!panicstr) {
mutex_enter(&cpu_lock);
pause_cpus(NULL, NULL);
mutex_exit(&cpu_lock);
}
/*
* If the system is panicking, the preloaded kernel is valid, and
* fastreboot_onpanic has been set, and the system has been up for
* longer than fastreboot_onpanic_uptime (default to 10 minutes),
* choose Fast Reboot.
*/
if (fcn == AD_BOOT && panicstr && newkernel.fi_valid &&
fastreboot_onpanic &&
(panic_lbolt - lbolt_at_boot) > fastreboot_onpanic_uptime) {
fcn = AD_FASTREBOOT;
}
/*
* Try to quiesce devices.
*/
if (is_first_quiesce) {
/*
* Clear is_first_quiesce before calling quiesce_devices()
* so that if quiesce_devices() causes panics, it will not
* be invoked again.
*/
is_first_quiesce = 0;
quiesce_active = 1;
quiesce_devices(ddi_root_node(), &reset_status);
if (reset_status == -1) {
if (fcn == AD_FASTREBOOT && !force_fastreboot) {
prom_printf("Driver(s) not capable of fast "
"reboot.\n");
prom_printf(fallback_str);
fastreboot_capable = 0;
fcn = AD_BOOT;
} else if (fcn != AD_FASTREBOOT)
fastreboot_capable = 0;
}
quiesce_active = 0;
}
/*
* Try to reset devices. reset_leaves() should only be called
* a) when there are no other threads that could be accessing devices,
* and
* b) on a system that's not capable of fast reboot (fastreboot_capable
* being 0), or on a system where quiesce_devices() failed to
* complete (quiesce_active being 1).
*/
if (is_first_reset && (!fastreboot_capable || quiesce_active)) {
/*
* Clear is_first_reset before calling reset_devices()
* so that if reset_devices() causes panics, it will not
* be invoked again.
*/
is_first_reset = 0;
reset_leaves();
}
/* Verify newkernel checksum */
if (fastreboot_capable && fcn == AD_FASTREBOOT &&
fastboot_cksum_verify(&newkernel) != 0) {
fastreboot_capable = 0;
prom_printf("Fast reboot: checksum failed for the new "
"kernel.\n");
prom_printf(fallback_str);
}
(void) spl8();
if (fastreboot_capable && fcn == AD_FASTREBOOT) {
/*
* psm_shutdown is called within fast_reboot()
*/
fast_reboot();
} else {
(*psm_shutdownf)(cmd, fcn);
if (fcn == AD_HALT || fcn == AD_POWEROFF)
halt((char *)NULL);
else
prom_reboot("");
}
/*NOTREACHED*/
}
/*
* Class callback when a CPU goes inactive (offline)
*
* This is called in a context where CPUs are paused
*/
static void
pg_cmt_cpu_inactive(cpu_t *cp)
{
int err;
group_t *pgs;
pg_cmt_t *pg;
cpu_t *cpp;
group_iter_t i;
pg_cpu_itr_t cpu_itr;
boolean_t found;
ASSERT(MUTEX_HELD(&cpu_lock));
if (cmt_sched_disabled)
return;
pgs = &cp->cpu_pg->pgs;
group_iter_init(&i);
while ((pg = group_iterate(pgs, &i)) != NULL) {
if (IS_CMT_PG(pg) == 0)
continue;
/*
* Remove the CPU from the CMT PGs active CPU group
* bitmap
*/
err = group_remove(&pg->cmt_cpus_actv, cp, GRP_NORESIZE);
ASSERT(err == 0);
bitset_del(&pg->cmt_cpus_actv_set, cp->cpu_seqid);
/*
* If there are no more active CPUs in this PG over which
* load was balanced, remove it as a balancing candidate.
*/
if (GROUP_SIZE(&pg->cmt_cpus_actv) == 0 &&
(pg->cmt_policy & (CMT_BALANCE | CMT_COALESCE))) {
err = group_remove(pg->cmt_siblings, pg, GRP_NORESIZE);
ASSERT(err == 0);
if (pg->cmt_parent == NULL &&
pg->cmt_siblings != &cmt_root->cl_pgs) {
err = group_remove(&cmt_root->cl_pgs, pg,
GRP_NORESIZE);
ASSERT(err == 0);
}
}
/*
* Assert the number of active CPUs does not exceed
* the total number of CPUs in the PG
*/
ASSERT(GROUP_SIZE(&pg->cmt_cpus_actv) <=
GROUP_SIZE(&((pg_t *)pg)->pg_cpus));
/*
* Update the PG bitset in the CPU's old partition
*/
found = B_FALSE;
PG_CPU_ITR_INIT(pg, cpu_itr);
while ((cpp = pg_cpu_next(&cpu_itr)) != NULL) {
if (cpp == cp)
continue;
if (CPU_ACTIVE(cpp) &&
cpp->cpu_part->cp_id == cp->cpu_part->cp_id) {
found = B_TRUE;
break;
}
}
if (!found) {
bitset_del(&cp->cpu_part->cp_cmt_pgs,
((pg_t *)pg)->pg_id);
}
}
}
static int
acquire_psets(struct snapshot *ss)
{
psetid_t *pids = NULL;
struct pset_snapshot *ps;
size_t pids_nr;
size_t i, j;
/*
* Careful in this code. We have to use pset_list
* twice, but inbetween pids_nr can change at will.
* We delay the setting of s_nr_psets until we have
* the "final" value of pids_nr.
*/
if (pset_list(NULL, &pids_nr) < 0)
return (errno);
if ((pids = calloc(pids_nr, sizeof (psetid_t))) == NULL)
goto out;
if (pset_list(pids, &pids_nr) < 0)
goto out;
ss->s_psets = calloc(pids_nr + 1, sizeof (struct pset_snapshot));
if (ss->s_psets == NULL)
goto out;
ss->s_nr_psets = pids_nr + 1;
/* CPUs not in any actual pset */
ps = &ss->s_psets[0];
ps->ps_id = 0;
ps->ps_cpus = calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot *));
if (ps->ps_cpus == NULL)
goto out;
/* CPUs in a a pset */
for (i = 1; i < ss->s_nr_psets; i++) {
ps = &ss->s_psets[i];
ps->ps_id = pids[i - 1];
ps->ps_cpus =
calloc(ss->s_nr_cpus, sizeof (struct cpu_snapshot *));
if (ps->ps_cpus == NULL)
goto out;
}
for (i = 0; i < ss->s_nr_psets; i++) {
ps = &ss->s_psets[i];
for (j = 0; j < ss->s_nr_cpus; j++) {
if (!CPU_ACTIVE(&ss->s_cpus[j]))
continue;
if (ss->s_cpus[j].cs_pset_id != ps->ps_id)
continue;
ps->ps_cpus[ps->ps_nr_cpus++] = &ss->s_cpus[j];
}
}
errno = 0;
out:
free(pids);
return (errno);
}