Пример #1
0
void
panicsys(const char *format, va_list alist, struct regs *rp, int on_panic_stack)
{
	int s = spl8();
	kthread_t *t = curthread;
	cpu_t *cp = CPU;

	caddr_t intr_stack = NULL;
	uint_t intr_actv;

	ushort_t schedflag = t->t_schedflag;
	cpu_t *bound_cpu = t->t_bound_cpu;
	char preempt = t->t_preempt;

	(void) setjmp(&t->t_pcb);
	t->t_flag |= T_PANIC;

	t->t_schedflag |= TS_DONT_SWAP;
	t->t_bound_cpu = cp;
	t->t_preempt++;

	/*
	 * Switch lbolt to event driven mode.
	 */
	lbolt_hybrid = lbolt_event_driven;

	panic_enter_hw(s);

	/*
	 * If we're on the interrupt stack and an interrupt thread is available
	 * in this CPU's pool, preserve the interrupt stack by detaching an
	 * interrupt thread and making its stack the intr_stack.
	 */
	if (CPU_ON_INTR(cp) && cp->cpu_intr_thread != NULL) {
		kthread_t *it = cp->cpu_intr_thread;

		intr_stack = cp->cpu_intr_stack;
		intr_actv = cp->cpu_intr_actv;

		cp->cpu_intr_stack = thread_stk_init(it->t_stk);
		cp->cpu_intr_thread = it->t_link;

		/*
		 * Clear only the high level bits of cpu_intr_actv.
		 * We want to indicate that high-level interrupts are
		 * not active without destroying the low-level interrupt
		 * information stored there.
		 */
		cp->cpu_intr_actv &= ((1 << (LOCK_LEVEL + 1)) - 1);
	}

	/*
	 * Record one-time panic information and quiesce the other CPUs.
	 * Then print out the panic message and stack trace.
	 */
	if (on_panic_stack) {
		panic_data_t *pdp = (panic_data_t *)panicbuf;

		pdp->pd_version = PANICBUFVERS;
		pdp->pd_msgoff = sizeof (panic_data_t) - sizeof (panic_nv_t);

		if (t->t_panic_trap != NULL)
			panic_savetrap(pdp, t->t_panic_trap);
		else
			panic_saveregs(pdp, rp);

		(void) vsnprintf(&panicbuf[pdp->pd_msgoff],
		    PANICBUFSIZE - pdp->pd_msgoff, format, alist);

		/*
		 * Call into the platform code to stop the other CPUs.
		 * We currently have all interrupts blocked, and expect that
		 * the platform code will lower ipl only as far as needed to
		 * perform cross-calls, and will acquire as *few* locks as is
		 * possible -- panicstr is not set so we can still deadlock.
		 */
		panic_stopcpus(cp, t, s);

		panicstr = (char *)format;
		va_copy(panicargs, alist);
		panic_lbolt = LBOLT_NO_ACCOUNT;
		panic_lbolt64 = LBOLT_NO_ACCOUNT64;
		panic_hrestime = hrestime;
		panic_hrtime = gethrtime_waitfree();
		panic_thread = t;
		panic_regs = t->t_pcb;
		panic_reg = rp;
		panic_cpu = *cp;
		panic_ipl = spltoipl(s);
		panic_schedflag = schedflag;
		panic_bound_cpu = bound_cpu;
		panic_preempt = preempt;

		if (intr_stack != NULL) {
			panic_cpu.cpu_intr_stack = intr_stack;
			panic_cpu.cpu_intr_actv = intr_actv;
		}

		/*
		 * Lower ipl to 10 to keep clock() from running, but allow
		 * keyboard interrupts to enter the debugger.  These callbacks
		 * are executed with panicstr set so they can bypass locks.
		 */
		splx(ipltospl(CLOCK_LEVEL));
		panic_quiesce_hw(pdp);
		(void) FTRACE_STOP();
		(void) callb_execute_class(CB_CL_PANIC, NULL);

		if (log_intrq != NULL)
			log_flushq(log_intrq);

		/*
		 * If log_consq has been initialized and syslogd has started,
		 * print any messages in log_consq that haven't been consumed.
		 */
		if (log_consq != NULL && log_consq != log_backlogq)
			log_printq(log_consq);

		fm_banner();

#if defined(__x86)
		/*
		 * A hypervisor panic originates outside of Solaris, so we
		 * don't want to prepend the panic message with misleading
		 * pointers from within Solaris.
		 */
		if (!IN_XPV_PANIC())
#endif
			printf("\n\rpanic[cpu%d]/thread=%p: ", cp->cpu_id,
			    (void *)t);
		vprintf(format, alist);
		printf("\n\n");

		if (t->t_panic_trap != NULL) {
			panic_showtrap(t->t_panic_trap);
			printf("\n");
		}

		traceregs(rp);
		printf("\n");

		if (((boothowto & RB_DEBUG) || obpdebug) &&
		    !nopanicdebug && !panic_forced) {
			if (dumpvp != NULL) {
				debug_enter("panic: entering debugger "
				    "(continue to save dump)");
			} else {
				debug_enter("panic: entering debugger "
				    "(no dump device, continue to reboot)");
			}
		}

	} else if (panic_dump != 0 || panic_sync != 0 || panicstr != NULL) {
		printf("\n\rpanic[cpu%d]/thread=%p: ", cp->cpu_id, (void *)t);
		vprintf(format, alist);
		printf("\n");
	} else
		goto spin;

	/*
	 * Prior to performing sync or dump, we make sure that do_polled_io is
	 * set, but we'll leave ipl at 10; deadman(), a CY_HIGH_LEVEL cyclic,
	 * will re-enter panic if we are not making progress with sync or dump.
	 */

	/*
	 * Sync the filesystems.  Reset t_cred if not set because much of
	 * the filesystem code depends on CRED() being valid.
	 */
	if (!in_sync && panic_trigger(&panic_sync)) {
		if (t->t_cred == NULL)
			t->t_cred = kcred;
		splx(ipltospl(CLOCK_LEVEL));
		do_polled_io = 1;
		vfs_syncall();
	}

	/*
	 * Take the crash dump.  If the dump trigger is already set, try to
	 * enter the debugger again before rebooting the system.
	 */
	if (panic_trigger(&panic_dump)) {
		panic_dump_hw(s);
		splx(ipltospl(CLOCK_LEVEL));
		errorq_panic();
		do_polled_io = 1;
		dumpsys();
	} else if (((boothowto & RB_DEBUG) || obpdebug) && !nopanicdebug) {
		debug_enter("panic: entering debugger (continue to reboot)");
	} else
		printf("dump aborted: please record the above information!\n");

	if (halt_on_panic)
		mdboot(A_REBOOT, AD_HALT, NULL, B_FALSE);
	else
		mdboot(A_REBOOT, panic_bootfcn, panic_bootstr, B_FALSE);
spin:
	/*
	 * Restore ipl to at most CLOCK_LEVEL so we don't end up spinning
	 * and unable to jump into the debugger.
	 */
	splx(MIN(s, ipltospl(CLOCK_LEVEL)));
	for (;;)
		;
}
Пример #2
0
/*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*/
}
Пример #3
0
/*ARGSUSED*/
int
suspend_start(char *error_reason, size_t max_reason_len)
{
	uint64_t	source_tick;
	uint64_t	source_stick;
	uint64_t	rv;
	timestruc_t	source_tod;
	int		spl;

	ASSERT(suspend_supported());
	DBG("suspend: %s", __func__);

	sfmmu_ctxdoms_lock();

	mutex_enter(&cpu_lock);

	/* Suspend the watchdog */
	watchdog_suspend();

	/* Record the TOD */
	mutex_enter(&tod_lock);
	source_tod = tod_get();
	mutex_exit(&tod_lock);

	/* Pause all other CPUs */
	pause_cpus(NULL);
	DBG_PROM("suspend: CPUs paused\n");

	/* Suspend cyclics */
	cyclic_suspend();
	DBG_PROM("suspend: cyclics suspended\n");

	/* Disable interrupts */
	spl = spl8();
	DBG_PROM("suspend: spl8()\n");

	source_tick = gettick_counter();
	source_stick = gettick();
	DBG_PROM("suspend: source_tick: 0x%lx\n", source_tick);
	DBG_PROM("suspend: source_stick: 0x%lx\n", source_stick);

	/*
	 * Call into the HV to initiate the suspend. hv_guest_suspend()
	 * returns after the guest has been resumed or if the suspend
	 * operation failed or was cancelled. After a successful suspend,
	 * the %tick and %stick registers may have changed by an amount
	 * that is not proportional to the amount of time that has passed.
	 * They may have jumped forwards or backwards. Some variation is
	 * allowed and accounted for using suspend_tick_stick_max_delta,
	 * but otherwise this jump must be uniform across all CPUs and we
	 * operate under the assumption that it is (maintaining two global
	 * offset variables--one for %tick and one for %stick.)
	 */
	DBG_PROM("suspend: suspending... \n");
	rv = hv_guest_suspend();
	if (rv != 0) {
		splx(spl);
		cyclic_resume();
		start_cpus();
		watchdog_resume();
		mutex_exit(&cpu_lock);
		sfmmu_ctxdoms_unlock();
		DBG("suspend: failed, rv: %ld\n", rv);
		return (rv);
	}

	suspend_count++;

	/* Update the global tick and stick offsets and the preserved TOD */
	set_tick_offsets(source_tick, source_stick, &source_tod);

	/* Ensure new offsets are globally visible before resuming CPUs */
	membar_sync();

	/* Enable interrupts */
	splx(spl);

	/* Set the {%tick,%stick}.NPT bits on all CPUs */
	if (enable_user_tick_stick_emulation) {
		xc_all((xcfunc_t *)enable_tick_stick_npt, NULL, NULL);
		xt_sync(cpu_ready_set);
		ASSERT(gettick_npt() != 0);
		ASSERT(getstick_npt() != 0);
	}

	/* If emulation is enabled, but not currently active, enable it */
	if (enable_user_tick_stick_emulation && !tick_stick_emulation_active) {
		tick_stick_emulation_active = B_TRUE;
	}

	sfmmu_ctxdoms_remove();

	/* Resume cyclics, unpause CPUs */
	cyclic_resume();
	start_cpus();

	/* Set the TOD */
	mutex_enter(&tod_lock);
	tod_set(source_tod);
	mutex_exit(&tod_lock);

	/* Re-enable the watchdog */
	watchdog_resume();

	mutex_exit(&cpu_lock);

	/* Download the latest MD */
	if ((rv = mach_descrip_update()) != 0)
		cmn_err(CE_PANIC, "suspend: mach_descrip_update failed: %ld",
		    rv);

	sfmmu_ctxdoms_update();
	sfmmu_ctxdoms_unlock();

	/* Get new MD, update CPU mappings/relationships */
	if (suspend_update_cpu_mappings)
		update_cpu_mappings();

	DBG("suspend: target tick: 0x%lx", gettick_counter());
	DBG("suspend: target stick: 0x%llx", gettick());
	DBG("suspend: user %%tick/%%stick emulation is %d",
	    tick_stick_emulation_active);
	DBG("suspend: finished");

	return (0);
}