xref: /titanic_44/usr/src/uts/sun4u/os/mach_cpu_states.c (revision 32529ec11ac8e93a62985721612a18ee6bb8659a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/types.h>
30 #include <sys/t_lock.h>
31 #include <sys/uadmin.h>
32 #include <sys/panic.h>
33 #include <sys/reboot.h>
34 #include <sys/autoconf.h>
35 #include <sys/machsystm.h>
36 #include <sys/promif.h>
37 #include <sys/membar.h>
38 #include <vm/hat_sfmmu.h>
39 #include <sys/cpu_module.h>
40 #include <sys/cpu_sgnblk_defs.h>
41 #include <sys/intreg.h>
42 #include <sys/consdev.h>
43 #include <sys/kdi_impl.h>
44 #include <sys/callb.h>
45 
46 #ifdef	TRAPTRACE
47 #include <sys/traptrace.h>
48 u_longlong_t panic_tick;
49 #endif /* TRAPTRACE */
50 
51 extern u_longlong_t	gettick();
52 static void reboot_machine(char *);
53 extern int disable_watchdog_on_exit;
54 
55 /*
56  * Machine dependent code to reboot.
57  * "mdep" is interpreted as a character pointer; if non-null, it is a pointer
58  * to a string to be used as the argument string when rebooting.
59  *
60  * "invoke_cb" is a boolean. It is set to true when mdboot() can safely
61  * invoke CB_CL_MDBOOT callbacks before shutting the system down, i.e. when
62  * we are in a normal shutdown sequence (interrupts are not blocked, the
63  * system is not panic'ing or being suspended).
64  */
65 /*ARGSUSED*/
66 void
67 mdboot(int cmd, int fcn, char *bootstr, boolean_t invoke_cb)
68 {
69 	extern void pm_cfb_check_and_powerup(void);
70 
71 	/*
72 	 * Disable the hw watchdog timer.
73 	 */
74 	if (disable_watchdog_on_exit && watchdog_activated) {
75 		mutex_enter(&tod_lock);
76 		(void) tod_ops.tod_clear_watchdog_timer();
77 		mutex_exit(&tod_lock);
78 	}
79 
80 	/*
81 	 * XXX - rconsvp is set to NULL to ensure that output messages
82 	 * are sent to the underlying "hardware" device using the
83 	 * monitor's printf routine since we are in the process of
84 	 * either rebooting or halting the machine.
85 	 */
86 	rconsvp = NULL;
87 
88 	/*
89 	 * At a high interrupt level we can't:
90 	 *	1) bring up the console
91 	 * or
92 	 *	2) wait for pending interrupts prior to redistribution
93 	 *	   to the current CPU
94 	 *
95 	 * so we do them now.
96 	 */
97 	pm_cfb_check_and_powerup();
98 
99 	/* make sure there are no more changes to the device tree */
100 	devtree_freeze();
101 
102 	if (invoke_cb)
103 		(void) callb_execute_class(CB_CL_MDBOOT, NULL);
104 
105 	/*
106 	 * Clear any unresolved UEs from memory.
107 	 */
108 	if (memsegs != NULL)
109 		page_retire_hunt(page_retire_mdboot_cb);
110 
111 	/*
112 	 * stop other cpus which also raise our priority. since there is only
113 	 * one active cpu after this, and our priority will be too high
114 	 * for us to be preempted, we're essentially single threaded
115 	 * from here on out.
116 	 */
117 	stop_other_cpus();
118 
119 	/*
120 	 * try and reset leaf devices.  reset_leaves() should only
121 	 * be called when there are no other threads that could be
122 	 * accessing devices
123 	 */
124 	reset_leaves();
125 
126 	if (fcn == AD_HALT) {
127 		halt((char *)NULL);
128 	} else if (fcn == AD_POWEROFF) {
129 		power_down(NULL);
130 	} else {
131 		if (bootstr == NULL) {
132 			switch (fcn) {
133 
134 			case AD_BOOT:
135 				bootstr = "";
136 				break;
137 
138 			case AD_IBOOT:
139 				bootstr = "-a";
140 				break;
141 
142 			case AD_SBOOT:
143 				bootstr = "-s";
144 				break;
145 
146 			case AD_SIBOOT:
147 				bootstr = "-sa";
148 				break;
149 			default:
150 				cmn_err(CE_WARN,
151 				    "mdboot: invalid function %d", fcn);
152 				bootstr = "";
153 				break;
154 			}
155 		}
156 		reboot_machine(bootstr);
157 	}
158 	/* MAYBE REACHED */
159 }
160 
161 /* mdpreboot - may be called prior to mdboot while root fs still mounted */
162 /*ARGSUSED*/
163 void
164 mdpreboot(int cmd, int fcn, char *bootstr)
165 {
166 }
167 
168 /*
169  * Halt the machine and then reboot with the device
170  * and arguments specified in bootstr.
171  */
172 static void
173 reboot_machine(char *bootstr)
174 {
175 	flush_windows();
176 	stop_other_cpus();		/* send stop signal to other CPUs */
177 	prom_printf("rebooting...\n");
178 	/*
179 	 * For platforms that use CPU signatures, we
180 	 * need to set the signature block to OS and
181 	 * the state to exiting for all the processors.
182 	 */
183 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_REBOOT, -1);
184 	prom_reboot(bootstr);
185 	/*NOTREACHED*/
186 }
187 
188 /*
189  * We use the x-trap mechanism and idle_stop_xcall() to stop the other CPUs.
190  * Once in panic_idle() they raise spl, record their location, and spin.
191  */
192 static void
193 panic_idle(void)
194 {
195 	cpu_async_panic_callb(); /* check for async errors */
196 
197 	(void) spl7();
198 
199 	debug_flush_windows();
200 	(void) setjmp(&curthread->t_pcb);
201 
202 	CPU->cpu_m.in_prom = 1;
203 	membar_stld();
204 
205 	for (;;);
206 }
207 
208 /*
209  * Force the other CPUs to trap into panic_idle(), and then remove them
210  * from the cpu_ready_set so they will no longer receive cross-calls.
211  */
212 /*ARGSUSED*/
213 void
214 panic_stopcpus(cpu_t *cp, kthread_t *t, int spl)
215 {
216 	cpuset_t cps;
217 	int i;
218 
219 	(void) splzs();
220 	CPUSET_ALL_BUT(cps, cp->cpu_id);
221 	xt_some(cps, (xcfunc_t *)idle_stop_xcall, (uint64_t)&panic_idle, NULL);
222 
223 	for (i = 0; i < NCPU; i++) {
224 		if (i != cp->cpu_id && CPU_XCALL_READY(i)) {
225 			int ntries = 0x10000;
226 
227 			while (!cpu[i]->cpu_m.in_prom && ntries) {
228 				DELAY(50);
229 				ntries--;
230 			}
231 
232 			if (!cpu[i]->cpu_m.in_prom)
233 				printf("panic: failed to stop cpu%d\n", i);
234 
235 			cpu[i]->cpu_flags &= ~CPU_READY;
236 			cpu[i]->cpu_flags |= CPU_QUIESCED;
237 			CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
238 		}
239 	}
240 }
241 
242 /*
243  * Platform callback following each entry to panicsys().  If we've panicked at
244  * level 14, we examine t_panic_trap to see if a fatal trap occurred.  If so,
245  * we disable further %tick_cmpr interrupts.  If not, an explicit call to panic
246  * was made and so we re-enqueue an interrupt request structure to allow
247  * further level 14 interrupts to be processed once we lower PIL.  This allows
248  * us to handle panics from the deadman() CY_HIGH_LEVEL cyclic.
249  */
250 void
251 panic_enter_hw(int spl)
252 {
253 	if (spl == ipltospl(PIL_14)) {
254 		uint_t opstate = disable_vec_intr();
255 
256 		if (curthread->t_panic_trap != NULL) {
257 			tickcmpr_disable();
258 			intr_dequeue_req(PIL_14, cbe_level14_inum);
259 		} else {
260 			if (!tickcmpr_disabled())
261 				intr_enqueue_req(PIL_14, cbe_level14_inum);
262 			/*
263 			 * Clear SOFTINT<14>, SOFTINT<0> (TICK_INT)
264 			 * and SOFTINT<16> (STICK_INT) to indicate
265 			 * that the current level 14 has been serviced.
266 			 */
267 			wr_clr_softint((1 << PIL_14) |
268 				TICK_INT_MASK | STICK_INT_MASK);
269 		}
270 
271 		enable_vec_intr(opstate);
272 	}
273 }
274 
275 /*
276  * Miscellaneous hardware-specific code to execute after panicstr is set
277  * by the panic code: we also print and record PTL1 panic information here.
278  */
279 /*ARGSUSED*/
280 void
281 panic_quiesce_hw(panic_data_t *pdp)
282 {
283 	extern uint_t getpstate(void);
284 	extern void setpstate(uint_t);
285 
286 #ifdef TRAPTRACE
287 	/*
288 	 * Turn off TRAPTRACE and save the current %tick value in panic_tick.
289 	 */
290 	if (!panic_tick)
291 		panic_tick = gettick();
292 	TRAPTRACE_FREEZE;
293 #endif
294 	/*
295 	 * For Platforms that use CPU signatures, we
296 	 * need to set the signature block to OS, the state to
297 	 * exiting, and the substate to panic for all the processors.
298 	 */
299 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_PANIC, -1);
300 
301 	/*
302 	 * De-activate ECC functions and disable the watchdog timer now that
303 	 * we've made it through the critical part of the panic code.
304 	 */
305 	if (watchdog_enable)
306 		(void) tod_ops.tod_clear_watchdog_timer();
307 
308 	/*
309 	 * Disable further ECC errors from the CPU module and the bus nexus.
310 	 */
311 	cpu_disable_errors();
312 	(void) bus_func_invoke(BF_TYPE_ERRDIS);
313 
314 	/*
315 	 * Redirect all interrupts to the current CPU.
316 	 */
317 	intr_redist_all_cpus_shutdown();
318 
319 	/*
320 	 * This call exists solely to support dumps to network
321 	 * devices after sync from OBP.
322 	 *
323 	 * If we came here via the sync callback, then on some
324 	 * platforms, interrupts may have arrived while we were
325 	 * stopped in OBP.  OBP will arrange for those interrupts to
326 	 * be redelivered if you say "go", but not if you invoke a
327 	 * client callback like 'sync'.	 For some dump devices
328 	 * (network swap devices), we need interrupts to be
329 	 * delivered in order to dump, so we have to call the bus
330 	 * nexus driver to reset the interrupt state machines.
331 	 */
332 	(void) bus_func_invoke(BF_TYPE_RESINTR);
333 
334 	setpstate(getpstate() | PSTATE_IE);
335 }
336 
337 /*
338  * Platforms that use CPU signatures need to set the signature block to OS and
339  * the state to exiting for all CPUs. PANIC_CONT indicates that we're about to
340  * write the crash dump, which tells the SSP/SMS to begin a timeout routine to
341  * reboot the machine if the dump never completes.
342  */
343 /*ARGSUSED*/
344 void
345 panic_dump_hw(int spl)
346 {
347 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
348 }
349 
350 /*
351  * for ptl1_panic
352  */
353 void
354 ptl1_init_cpu(struct cpu *cpu)
355 {
356 	ptl1_state_t *pstate = &cpu->cpu_m.ptl1_state;
357 
358 	/*CONSTCOND*/
359 	if (sizeof (struct cpu) + PTL1_SSIZE > CPU_ALLOC_SIZE) {
360 		panic("ptl1_init_cpu: not enough space left for ptl1_panic "
361 		    "stack, sizeof (struct cpu) = %d", sizeof (struct cpu));
362 	}
363 
364 	pstate->ptl1_stktop = (uintptr_t)cpu + CPU_ALLOC_SIZE;
365 	cpu_pa[cpu->cpu_id] = va_to_pa(cpu);
366 }
367 
368 void
369 ptl1_panic_handler(ptl1_state_t *pstate)
370 {
371 	static const char *ptl1_reasons[] = {
372 #ifdef	PTL1_PANIC_DEBUG
373 		"trap for debug purpose",	/* PTL1_BAD_DEBUG */
374 #else
375 		"unknown trap",			/* PTL1_BAD_DEBUG */
376 #endif
377 		"register window trap",		/* PTL1_BAD_WTRAP */
378 		"kernel MMU miss",		/* PTL1_BAD_KMISS */
379 		"kernel protection fault",	/* PTL1_BAD_KPROT_FAULT */
380 		"ISM MMU miss",			/* PTL1_BAD_ISM */
381 		"kernel MMU trap",		/* PTL1_BAD_MMUTRAP */
382 		"kernel trap handler state",	/* PTL1_BAD_TRAP */
383 		"floating point trap",		/* PTL1_BAD_FPTRAP */
384 #ifdef	DEBUG
385 		"pointer to intr_req",		/* PTL1_BAD_INTR_REQ */
386 #else
387 		"unknown trap",			/* PTL1_BAD_INTR_REQ */
388 #endif
389 #ifdef	TRAPTRACE
390 		"TRACE_PTR state",		/* PTL1_BAD_TRACE_PTR */
391 #else
392 		"unknown trap",			/* PTL1_BAD_TRACE_PTR */
393 #endif
394 		"stack overflow",		/* PTL1_BAD_STACK */
395 		"DTrace flags",			/* PTL1_BAD_DTRACE_FLAGS */
396 		"attempt to steal locked ctx",  /* PTL1_BAD_CTX_STEAL */
397 		"CPU ECC error loop",		/* PTL1_BAD_ECC */
398 		"unknown trap",			/* PTL1_BAD_HCALL */
399 	};
400 
401 	uint_t reason = pstate->ptl1_regs.ptl1_g1;
402 	uint_t tl = pstate->ptl1_regs.ptl1_trap_regs[0].ptl1_tl;
403 	struct trap_info ti = { 0 };
404 
405 	/*
406 	 * Use trap_info for a place holder to call panic_savetrap() and
407 	 * panic_showtrap() to save and print out ptl1_panic information.
408 	 */
409 	if (curthread->t_panic_trap == NULL)
410 		curthread->t_panic_trap = &ti;
411 
412 	if (reason < sizeof (ptl1_reasons) / sizeof (ptl1_reasons[0]))
413 		panic("bad %s at TL %u", ptl1_reasons[reason], tl);
414 	else
415 		panic("ptl1_panic reason 0x%x at TL %u", reason, tl);
416 }
417 
418 void
419 clear_watchdog_on_exit()
420 {
421 	/*
422 	 * Only shut down an active hardware watchdog timer if the platform
423 	 * has expressed an interest to.
424 	 */
425 	if (disable_watchdog_on_exit && watchdog_activated) {
426 		prom_printf("Debugging requested; hardware watchdog "
427 		    "disabled; reboot to re-enable.\n");
428 		cmn_err(CE_WARN, "!Debugging requested; hardware watchdog "
429 		    "disabled; reboot to re-enable.");
430 		mutex_enter(&tod_lock);
431 		(void) tod_ops.tod_clear_watchdog_timer();
432 		mutex_exit(&tod_lock);
433 	}
434 }
435 
436 int
437 kdi_watchdog_disable(void)
438 {
439 	if (watchdog_activated) {
440 		mutex_enter(&tod_lock);
441 		(void) tod_ops.tod_clear_watchdog_timer();
442 		mutex_exit(&tod_lock);
443 	}
444 
445 	return (watchdog_activated);
446 }
447 
448 void
449 kdi_watchdog_restore(void)
450 {
451 	if (watchdog_enable) {
452 		mutex_enter(&tod_lock);
453 		(void) tod_ops.tod_set_watchdog_timer(watchdog_timeout_seconds);
454 		mutex_exit(&tod_lock);
455 	}
456 }
457 
458 /*ARGSUSED*/
459 void
460 mach_dump_buffer_init(void)
461 {
462 	/*
463 	 * setup dump buffer to store extra crash information
464 	 * not applicable to sun4u
465 	 */
466 }
467 
468 /*
469  * xt_sync - wait for previous x-traps to finish
470  */
471 void
472 xt_sync(cpuset_t cpuset)
473 {
474 	kpreempt_disable();
475 	CPUSET_DEL(cpuset, CPU->cpu_id);
476 	CPUSET_AND(cpuset, cpu_ready_set);
477 	xt_some(cpuset, (xcfunc_t *)xt_sync_tl1, 0, 0);
478 	kpreempt_enable();
479 }
480