static void
dr_start_user_threads(void)
{
kthread_id_t tp;
mutex_enter(&pidlock);
/* walk all threads and release them */
for (tp = curthread->t_next; tp != curthread; tp = tp->t_next) {
proc_t *p = ttoproc(tp);
/* skip kernel threads */
if (ttoproc(tp)->p_as == &kas)
continue;
mutex_enter(&p->p_lock);
tp->t_proc_flag &= ~TP_CHKPT;
mutex_exit(&p->p_lock);
thread_lock(tp);
if (CPR_ISTOPPED(tp)) {
/* back on the runq */
tp->t_schedflag |= TS_RESUME;
setrun_locked(tp);
}
thread_unlock(tp);
}
mutex_exit(&pidlock);
}
/*
* Checks and makes sure all user threads are stopped
*/
static int
cpr_check_user_threads()
{
kthread_id_t tp;
int rc = 0;
mutex_enter(&pidlock);
tp = curthread->t_next;
do {
if (ttoproc(tp)->p_as == &kas || ttoproc(tp)->p_stat == SZOMB)
continue;
thread_lock(tp);
/*
* make sure that we are off all the queues and in a stopped
* state.
*/
if (!CPR_ISTOPPED(tp)) {
thread_unlock(tp);
mutex_exit(&pidlock);
if (count == CPR_UTSTOP_RETRY) {
DEBUG1(errp("Suspend failed: cannt stop "
"uthread\n"));
cpr_err(CE_WARN, "Suspend cannot stop "
"process %s (%p:%x).",
ttoproc(tp)->p_user.u_psargs, (void *)tp,
tp->t_state);
cpr_err(CE_WARN, "Process may be waiting for"
" network request, please try again.");
}
DEBUG2(errp("cant stop t=%p state=%x pfg=%x sched=%x\n",
tp, tp->t_state, tp->t_proc_flag, tp->t_schedflag));
DEBUG2(errp("proc %p state=%x pid=%d\n",
ttoproc(tp), ttoproc(tp)->p_stat,
ttoproc(tp)->p_pidp->pid_id));
return (1);
}
thread_unlock(tp);
} while ((tp = tp->t_next) != curthread && rc == 0);
mutex_exit(&pidlock);
return (0);
}
/*
* start all threads that were stopped for checkpoint.
*/
void
cpr_start_user_threads()
{
kthread_id_t tp;
proc_t *p;
mutex_enter(&pidlock);
tp = curthread->t_next;
do {
p = ttoproc(tp);
/*
* kernel threads are callback'ed rather than setrun.
*/
if (ttoproc(tp)->p_as == &kas) continue;
/*
* t_proc_flag should have been cleared. Just to make sure here
*/
mutex_enter(&p->p_lock);
tp->t_proc_flag &= ~TP_CHKPT;
mutex_exit(&p->p_lock);
thread_lock(tp);
if (CPR_ISTOPPED(tp)) {
/*
* put it back on the runq
*/
tp->t_schedflag |= TS_RESUME;
setrun_locked(tp);
}
thread_unlock(tp);
/*
* DEBUG - Keep track of current and next thread pointer.
*/
} while ((tp = tp->t_next) != curthread);
mutex_exit(&pidlock);
}
/* ARGSUSED */
static int
dr_stop_user_threads(dr_sr_handle_t *srh)
{
int count;
int bailout;
dr_handle_t *handle = srh->sr_dr_handlep;
static fn_t f = "dr_stop_user_threads";
kthread_id_t tp;
extern void add_one_utstop();
extern void utstop_timedwait(clock_t);
extern void utstop_init(void);
#define DR_UTSTOP_RETRY 4
#define DR_UTSTOP_WAIT hz
if (dr_skip_user_threads)
return (DDI_SUCCESS);
utstop_init();
/* we need to try a few times to get past fork, etc. */
srh->sr_err_idx = 0;
for (count = 0; count < DR_UTSTOP_RETRY; count++) {
/* walk the entire threadlist */
mutex_enter(&pidlock);
for (tp = curthread->t_next; tp != curthread; tp = tp->t_next) {
proc_t *p = ttoproc(tp);
/* handle kernel threads separately */
if (p->p_as == &kas || p->p_stat == SZOMB)
continue;
mutex_enter(&p->p_lock);
thread_lock(tp);
if (tp->t_state == TS_STOPPED) {
/* add another reason to stop this thread */
tp->t_schedflag &= ~TS_RESUME;
} else {
tp->t_proc_flag |= TP_CHKPT;
thread_unlock(tp);
mutex_exit(&p->p_lock);
add_one_utstop();
mutex_enter(&p->p_lock);
thread_lock(tp);
aston(tp);
if (tp->t_state == TS_SLEEP &&
(tp->t_flag & T_WAKEABLE)) {
setrun_locked(tp);
}
}
/* grab thread if needed */
if (tp->t_state == TS_ONPROC && tp->t_cpu != CPU)
poke_cpu(tp->t_cpu->cpu_id);
thread_unlock(tp);
mutex_exit(&p->p_lock);
}
mutex_exit(&pidlock);
/* let everything catch up */
utstop_timedwait(count * count * DR_UTSTOP_WAIT);
/* now, walk the threadlist again to see if we are done */
mutex_enter(&pidlock);
for (tp = curthread->t_next, bailout = 0;
tp != curthread; tp = tp->t_next) {
proc_t *p = ttoproc(tp);
/* handle kernel threads separately */
if (p->p_as == &kas || p->p_stat == SZOMB)
continue;
/*
* If this thread didn't stop, and we don't allow
* unstopped blocked threads, bail.
*/
thread_lock(tp);
if (!CPR_ISTOPPED(tp) &&
!(dr_allow_blocked_threads &&
DR_VSTOPPED(tp))) {
bailout = 1;
if (count == DR_UTSTOP_RETRY - 1) {
/*
* save the pid for later reporting
*/
srh->sr_err_idx =
dr_add_int(srh->sr_err_ints,
srh->sr_err_idx, DR_MAX_ERR_INT,
(uint64_t)p->p_pid);
cmn_err(CE_WARN, "%s: "
"failed to stop thread: "
"process=%s, pid=%d",
f, p->p_user.u_psargs, p->p_pid);
PR_QR("%s: failed to stop thread: "
"process=%s, pid=%d, t_id=0x%p, "
"t_state=0x%x, t_proc_flag=0x%x, "
"t_schedflag=0x%x\n",
f, p->p_user.u_psargs, p->p_pid,
tp, tp->t_state, tp->t_proc_flag,
tp->t_schedflag);
}
}
thread_unlock(tp);
}
mutex_exit(&pidlock);
/* were all the threads stopped? */
if (!bailout)
break;
}
/* were we unable to stop all threads after a few tries? */
if (bailout) {
handle->h_err = drerr_int(ESBD_UTHREAD, srh->sr_err_ints,
srh->sr_err_idx, 0);
return (ESRCH);
}
return (DDI_SUCCESS);
}
static void
sbdp_resume_devices(dev_info_t *start, sbdp_sr_handle_t *srh)
{
int circ;
dev_info_t *dip, *next, *last = NULL;
char *bn;
sbd_error_t *sep;
sep = &srh->sep;
/* attach in reverse device tree order */
while (last != start) {
dip = start;
next = ddi_get_next_sibling(dip);
while (next != last && dip != SR_FAILED_DIP(srh)) {
dip = next;
next = ddi_get_next_sibling(dip);
}
if (dip == SR_FAILED_DIP(srh)) {
/* Release hold acquired in sbdp_suspend_devices() */
ndi_rele_devi(dip);
SR_FAILED_DIP(srh) = NULL;
} else if (sbdp_is_real_device(dip) &&
SR_FAILED_DIP(srh) == NULL) {
if (DEVI(dip)->devi_binding_name != NULL) {
bn = ddi_binding_name(dip);
}
#ifdef DEBUG
if (!sbdp_bypass_device(bn)) {
#else
{
#endif
char d_name[40], d_alias[40], *d_info;
d_name[0] = 0;
d_info = ddi_get_name_addr(dip);
if (d_info == NULL)
d_info = "<null>";
if (!sbdp_resolve_devname(dip, d_name,
d_alias)) {
if (d_alias[0] != 0) {
SBDP_DBG_QR("\tresuming "
"%s@%s (aka %s)\n",
d_name, d_info,
d_alias);
} else {
SBDP_DBG_QR("\tresuming "
"%s@%s\n",
d_name, d_info);
}
} else {
SBDP_DBG_QR("\tresuming %s@%s\n",
bn, d_info);
}
if (devi_attach(dip, DDI_RESUME) !=
DDI_SUCCESS) {
/*
* Print a console warning,
* set an errno of ESGT_RESUME,
* and save the driver major
* number in the e_str.
*/
(void) sprintf(sbdp_get_err_buf(sep),
"%s@%s",
d_name[0] ? d_name : bn, d_info);
SBDP_DBG_QR("\tFAILED to resume "
"%s\n", sbdp_get_err_buf(sep));
sbdp_set_err(sep,
ESGT_RESUME, NULL);
}
}
}
ndi_devi_enter(dip, &circ);
sbdp_resume_devices(ddi_get_child(dip), srh);
ndi_devi_exit(dip, circ);
last = dip;
}
}
/*
* True if thread is virtually stopped. Similar to CPR_VSTOPPED
* but from DR point of view. These user threads are waiting in
* the kernel. Once they return from kernel, they will process
* the stop signal and stop.
*/
#define SBDP_VSTOPPED(t) \
((t)->t_state == TS_SLEEP && \
(t)->t_wchan != NULL && \
(t)->t_astflag && \
((t)->t_proc_flag & TP_CHKPT))
static int
sbdp_stop_user_threads(sbdp_sr_handle_t *srh)
{
int count;
char cache_psargs[PSARGSZ];
kthread_id_t cache_tp;
uint_t cache_t_state;
int bailout;
sbd_error_t *sep;
kthread_id_t tp;
extern void add_one_utstop();
extern void utstop_timedwait(clock_t);
extern void utstop_init(void);
#define SBDP_UTSTOP_RETRY 4
#define SBDP_UTSTOP_WAIT hz
if (sbdp_skip_user_threads)
return (DDI_SUCCESS);
sep = &srh->sep;
ASSERT(sep);
utstop_init();
/* we need to try a few times to get past fork, etc. */
for (count = 0; count < SBDP_UTSTOP_RETRY; count++) {
/* walk the entire threadlist */
mutex_enter(&pidlock);
for (tp = curthread->t_next; tp != curthread; tp = tp->t_next) {
proc_t *p = ttoproc(tp);
/* handle kernel threads separately */
if (p->p_as == &kas || p->p_stat == SZOMB)
continue;
mutex_enter(&p->p_lock);
thread_lock(tp);
if (tp->t_state == TS_STOPPED) {
/* add another reason to stop this thread */
tp->t_schedflag &= ~TS_RESUME;
} else {
tp->t_proc_flag |= TP_CHKPT;
thread_unlock(tp);
mutex_exit(&p->p_lock);
add_one_utstop();
mutex_enter(&p->p_lock);
thread_lock(tp);
aston(tp);
if (ISWAKEABLE(tp) || ISWAITING(tp)) {
setrun_locked(tp);
}
}
/* grab thread if needed */
if (tp->t_state == TS_ONPROC && tp->t_cpu != CPU)
poke_cpu(tp->t_cpu->cpu_id);
thread_unlock(tp);
mutex_exit(&p->p_lock);
}
mutex_exit(&pidlock);
/* let everything catch up */
utstop_timedwait(count * count * SBDP_UTSTOP_WAIT);
/* now, walk the threadlist again to see if we are done */
mutex_enter(&pidlock);
for (tp = curthread->t_next, bailout = 0;
tp != curthread; tp = tp->t_next) {
proc_t *p = ttoproc(tp);
/* handle kernel threads separately */
if (p->p_as == &kas || p->p_stat == SZOMB)
continue;
/*
* If this thread didn't stop, and we don't allow
* unstopped blocked threads, bail.
*/
thread_lock(tp);
if (!CPR_ISTOPPED(tp) &&
!(sbdp_allow_blocked_threads &&
SBDP_VSTOPPED(tp))) {
/* nope, cache the details for later */
bcopy(p->p_user.u_psargs, cache_psargs,
sizeof (cache_psargs));
cache_tp = tp;
cache_t_state = tp->t_state;
bailout = 1;
}
thread_unlock(tp);
}
mutex_exit(&pidlock);
/* were all the threads stopped? */
if (!bailout)
break;
}
/* were we unable to stop all threads after a few tries? */
if (bailout) {
cmn_err(CE_NOTE, "process: %s id: %p state: %x\n",
cache_psargs, cache_tp, cache_t_state);
(void) sprintf(sbdp_get_err_buf(sep), "%s", cache_psargs);
sbdp_set_err(sep, ESGT_UTHREAD, NULL);
return (ESRCH);
}
return (DDI_SUCCESS);
}