xref: /illumos-gate/usr/src/uts/sun4u/os/mach_cpu_states.c (revision 747cfc73193da6c99aeb06f23e70552319393cc4)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/t_lock.h>
30 #include <sys/uadmin.h>
31 #include <sys/panic.h>
32 #include <sys/reboot.h>
33 #include <sys/autoconf.h>
34 #include <sys/machsystm.h>
35 #include <sys/promif.h>
36 #include <sys/membar.h>
37 #include <vm/hat_sfmmu.h>
38 #include <sys/cpu_module.h>
39 #include <sys/cpu_sgnblk_defs.h>
40 #include <sys/intreg.h>
41 #include <sys/consdev.h>
42 #include <sys/kdi_impl.h>
43 #include <sys/callb.h>
44 
45 #ifdef	TRAPTRACE
46 #include <sys/traptrace.h>
47 u_longlong_t panic_tick;
48 #endif /* TRAPTRACE */
49 
50 extern u_longlong_t	gettick();
51 static void reboot_machine(char *);
52 int disable_watchdog_on_exit = 0;
53 
54 extern void consconfig_teardown();
55 
56 /*
57  * Machine dependent code to reboot.
58  * "mdep" is interpreted as a character pointer; if non-null, it is a pointer
59  * to a string to be used as the argument string when rebooting.
60  *
61  * "invoke_cb" is a boolean. It is set to true when mdboot() can safely
62  * invoke CB_CL_MDBOOT callbacks before shutting the system down, i.e. when
63  * we are in a normal shutdown sequence (interrupts are not blocked, the
64  * system is not panic'ing or being suspended).
65  */
66 /*ARGSUSED*/
67 void
68 mdboot(int cmd, int fcn, char *bootstr, boolean_t invoke_cb)
69 {
70 	extern void pm_cfb_check_and_powerup(void);
71 
72 	/*
73 	 * Disable the hw watchdog timer.
74 	 */
75 	if (disable_watchdog_on_exit && watchdog_activated) {
76 		mutex_enter(&tod_lock);
77 		(void) tod_ops.tod_clear_watchdog_timer();
78 		mutex_exit(&tod_lock);
79 	}
80 
81 	/*
82 	 * At a high interrupt level we can't:
83 	 *	1) bring up the console
84 	 * or
85 	 *	2) wait for pending interrupts prior to redistribution
86 	 *	   to the current CPU
87 	 *
88 	 * so we do them now.
89 	 */
90 	pm_cfb_check_and_powerup();
91 
92 	/* make sure there are no more changes to the device tree */
93 	devtree_freeze();
94 
95 	if (invoke_cb)
96 		(void) callb_execute_class(CB_CL_MDBOOT, NULL);
97 
98 	/*
99 	 * Clear any unresolved UEs from memory.
100 	 */
101 	page_retire_mdboot();
102 
103 	/*
104 	 * stop other cpus which also raise our priority. since there is only
105 	 * one active cpu after this, and our priority will be too high
106 	 * for us to be preempted, we're essentially single threaded
107 	 * from here on out.
108 	 */
109 	stop_other_cpus();
110 
111 	consconfig_teardown();
112 
113 	/*
114 	 * try and reset leaf devices.  reset_leaves() should only
115 	 * be called when there are no other threads that could be
116 	 * accessing devices
117 	 */
118 	reset_leaves();
119 
120 	if (fcn == AD_HALT) {
121 		halt((char *)NULL);
122 	} else if (fcn == AD_POWEROFF) {
123 		power_down(NULL);
124 	} else {
125 		if (bootstr == NULL) {
126 			switch (fcn) {
127 
128 			case AD_BOOT:
129 				bootstr = "";
130 				break;
131 
132 			case AD_IBOOT:
133 				bootstr = "-a";
134 				break;
135 
136 			case AD_SBOOT:
137 				bootstr = "-s";
138 				break;
139 
140 			case AD_SIBOOT:
141 				bootstr = "-sa";
142 				break;
143 			default:
144 				cmn_err(CE_WARN,
145 				    "mdboot: invalid function %d", fcn);
146 				bootstr = "";
147 				break;
148 			}
149 		}
150 		reboot_machine(bootstr);
151 	}
152 	/* MAYBE REACHED */
153 }
154 
155 /* mdpreboot - may be called prior to mdboot while root fs still mounted */
156 /*ARGSUSED*/
157 void
158 mdpreboot(int cmd, int fcn, char *bootstr)
159 {
160 }
161 
162 /*
163  * Halt the machine and then reboot with the device
164  * and arguments specified in bootstr.
165  */
166 static void
167 reboot_machine(char *bootstr)
168 {
169 	flush_windows();
170 	stop_other_cpus();		/* send stop signal to other CPUs */
171 	prom_printf("rebooting...\n");
172 	/*
173 	 * For platforms that use CPU signatures, we
174 	 * need to set the signature block to OS and
175 	 * the state to exiting for all the processors.
176 	 */
177 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_REBOOT, -1);
178 	prom_reboot(bootstr);
179 	/*NOTREACHED*/
180 }
181 
182 /*
183  * We use the x-trap mechanism and idle_stop_xcall() to stop the other CPUs.
184  * Once in panic_idle() they raise spl, record their location, and spin.
185  */
186 static void
187 panic_idle(void)
188 {
189 	cpu_async_panic_callb(); /* check for async errors */
190 
191 	(void) spl7();
192 
193 	debug_flush_windows();
194 	(void) setjmp(&curthread->t_pcb);
195 
196 	CPU->cpu_m.in_prom = 1;
197 	membar_stld();
198 
199 	for (;;)
200 		continue;
201 }
202 
203 /*
204  * Force the other CPUs to trap into panic_idle(), and then remove them
205  * from the cpu_ready_set so they will no longer receive cross-calls.
206  */
207 /*ARGSUSED*/
208 void
209 panic_stopcpus(cpu_t *cp, kthread_t *t, int spl)
210 {
211 	cpuset_t cps;
212 	int i;
213 
214 	(void) splzs();
215 	CPUSET_ALL_BUT(cps, cp->cpu_id);
216 	xt_some(cps, (xcfunc_t *)idle_stop_xcall, (uint64_t)&panic_idle, NULL);
217 
218 	for (i = 0; i < NCPU; i++) {
219 		if (i != cp->cpu_id && CPU_XCALL_READY(i)) {
220 			int ntries = 0x10000;
221 
222 			while (!cpu[i]->cpu_m.in_prom && ntries) {
223 				DELAY(50);
224 				ntries--;
225 			}
226 
227 			if (!cpu[i]->cpu_m.in_prom)
228 				printf("panic: failed to stop cpu%d\n", i);
229 
230 			cpu[i]->cpu_flags &= ~CPU_READY;
231 			cpu[i]->cpu_flags |= CPU_QUIESCED;
232 			CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
233 		}
234 	}
235 }
236 
237 /*
238  * Platform callback following each entry to panicsys().  If we've panicked at
239  * level 14, we examine t_panic_trap to see if a fatal trap occurred.  If so,
240  * we disable further %tick_cmpr interrupts.  If not, an explicit call to panic
241  * was made and so we re-enqueue an interrupt request structure to allow
242  * further level 14 interrupts to be processed once we lower PIL.  This allows
243  * us to handle panics from the deadman() CY_HIGH_LEVEL cyclic.
244  */
245 void
246 panic_enter_hw(int spl)
247 {
248 	if (spl == ipltospl(PIL_14)) {
249 		uint_t opstate = disable_vec_intr();
250 
251 		if (curthread->t_panic_trap != NULL) {
252 			tickcmpr_disable();
253 			intr_dequeue_req(PIL_14, cbe_level14_inum);
254 		} else {
255 			if (!tickcmpr_disabled())
256 				intr_enqueue_req(PIL_14, cbe_level14_inum);
257 			/*
258 			 * Clear SOFTINT<14>, SOFTINT<0> (TICK_INT)
259 			 * and SOFTINT<16> (STICK_INT) to indicate
260 			 * that the current level 14 has been serviced.
261 			 */
262 			wr_clr_softint((1 << PIL_14) |
263 			    TICK_INT_MASK | STICK_INT_MASK);
264 		}
265 
266 		enable_vec_intr(opstate);
267 	}
268 }
269 
270 /*
271  * Miscellaneous hardware-specific code to execute after panicstr is set
272  * by the panic code: we also print and record PTL1 panic information here.
273  */
274 /*ARGSUSED*/
275 void
276 panic_quiesce_hw(panic_data_t *pdp)
277 {
278 	extern uint_t getpstate(void);
279 	extern void setpstate(uint_t);
280 
281 #ifdef TRAPTRACE
282 	/*
283 	 * Turn off TRAPTRACE and save the current %tick value in panic_tick.
284 	 */
285 	if (!panic_tick)
286 		panic_tick = gettick();
287 	TRAPTRACE_FREEZE;
288 #endif
289 	/*
290 	 * For Platforms that use CPU signatures, we
291 	 * need to set the signature block to OS, the state to
292 	 * exiting, and the substate to panic for all the processors.
293 	 */
294 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_PANIC, -1);
295 
296 	/*
297 	 * De-activate ECC functions and disable the watchdog timer now that
298 	 * we've made it through the critical part of the panic code.
299 	 */
300 	if (watchdog_enable)
301 		(void) tod_ops.tod_clear_watchdog_timer();
302 
303 	/*
304 	 * Disable further ECC errors from the CPU module and the bus nexus.
305 	 */
306 	cpu_disable_errors();
307 	(void) bus_func_invoke(BF_TYPE_ERRDIS);
308 
309 	/*
310 	 * Redirect all interrupts to the current CPU.
311 	 */
312 	intr_redist_all_cpus_shutdown();
313 
314 	/*
315 	 * This call exists solely to support dumps to network
316 	 * devices after sync from OBP.
317 	 *
318 	 * If we came here via the sync callback, then on some
319 	 * platforms, interrupts may have arrived while we were
320 	 * stopped in OBP.  OBP will arrange for those interrupts to
321 	 * be redelivered if you say "go", but not if you invoke a
322 	 * client callback like 'sync'.	 For some dump devices
323 	 * (network swap devices), we need interrupts to be
324 	 * delivered in order to dump, so we have to call the bus
325 	 * nexus driver to reset the interrupt state machines.
326 	 */
327 	(void) bus_func_invoke(BF_TYPE_RESINTR);
328 
329 	setpstate(getpstate() | PSTATE_IE);
330 }
331 
332 /*
333  * Platforms that use CPU signatures need to set the signature block to OS and
334  * the state to exiting for all CPUs. PANIC_CONT indicates that we're about to
335  * write the crash dump, which tells the SSP/SMS to begin a timeout routine to
336  * reboot the machine if the dump never completes.
337  */
338 /*ARGSUSED*/
339 void
340 panic_dump_hw(int spl)
341 {
342 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
343 }
344 
345 /*
346  * for ptl1_panic
347  */
348 void
349 ptl1_init_cpu(struct cpu *cpu)
350 {
351 	ptl1_state_t *pstate = &cpu->cpu_m.ptl1_state;
352 
353 	/*CONSTCOND*/
354 	if (sizeof (struct cpu) + PTL1_SSIZE > CPU_ALLOC_SIZE) {
355 		panic("ptl1_init_cpu: not enough space left for ptl1_panic "
356 		    "stack, sizeof (struct cpu) = %lu", sizeof (struct cpu));
357 	}
358 
359 	pstate->ptl1_stktop = (uintptr_t)cpu + CPU_ALLOC_SIZE;
360 	cpu_pa[cpu->cpu_id] = va_to_pa(cpu);
361 }
362 
363 void
364 ptl1_panic_handler(ptl1_state_t *pstate)
365 {
366 	static const char *ptl1_reasons[] = {
367 #ifdef	PTL1_PANIC_DEBUG
368 		"trap for debug purpose",	/* PTL1_BAD_DEBUG */
369 #else
370 		"unknown trap",			/* PTL1_BAD_DEBUG */
371 #endif
372 		"register window trap",		/* PTL1_BAD_WTRAP */
373 		"kernel MMU miss",		/* PTL1_BAD_KMISS */
374 		"kernel protection fault",	/* PTL1_BAD_KPROT_FAULT */
375 		"ISM MMU miss",			/* PTL1_BAD_ISM */
376 		"kernel MMU trap",		/* PTL1_BAD_MMUTRAP */
377 		"kernel trap handler state",	/* PTL1_BAD_TRAP */
378 		"floating point trap",		/* PTL1_BAD_FPTRAP */
379 #ifdef	DEBUG
380 		"pointer to intr_vec",		/* PTL1_BAD_INTR_VEC */
381 #else
382 		"unknown trap",			/* PTL1_BAD_INTR_VEC */
383 #endif
384 #ifdef	TRAPTRACE
385 		"TRACE_PTR state",		/* PTL1_BAD_TRACE_PTR */
386 #else
387 		"unknown trap",			/* PTL1_BAD_TRACE_PTR */
388 #endif
389 		"stack overflow",		/* PTL1_BAD_STACK */
390 		"DTrace flags",			/* PTL1_BAD_DTRACE_FLAGS */
391 		"attempt to steal locked ctx",  /* PTL1_BAD_CTX_STEAL */
392 		"CPU ECC error loop",		/* PTL1_BAD_ECC */
393 		"non-kernel context in sys/priv_trap() below or",
394 						/* PTL1_BAD_CTX */
395 	};
396 
397 	uint_t reason = pstate->ptl1_regs.ptl1_g1;
398 	uint_t tl = pstate->ptl1_regs.ptl1_trap_regs[0].ptl1_tl;
399 	struct panic_trap_info ti = { 0 };
400 
401 	/*
402 	 * Use trap_info for a place holder to call panic_savetrap() and
403 	 * panic_showtrap() to save and print out ptl1_panic information.
404 	 */
405 	if (curthread->t_panic_trap == NULL)
406 		curthread->t_panic_trap = &ti;
407 
408 	if (reason < sizeof (ptl1_reasons) / sizeof (ptl1_reasons[0]))
409 		panic("bad %s at TL %u", ptl1_reasons[reason], tl);
410 	else
411 		panic("ptl1_panic reason 0x%x at TL %u", reason, tl);
412 }
413 
414 void
415 clear_watchdog_on_exit()
416 {
417 	/*
418 	 * Only shut down an active hardware watchdog timer if the platform
419 	 * has expressed an interest to.
420 	 */
421 	if (disable_watchdog_on_exit && watchdog_activated) {
422 		prom_printf("Debugging requested; hardware watchdog "
423 		    "disabled; reboot to re-enable.\n");
424 		cmn_err(CE_WARN, "!Debugging requested; hardware watchdog "
425 		    "disabled; reboot to re-enable.");
426 		mutex_enter(&tod_lock);
427 		(void) tod_ops.tod_clear_watchdog_timer();
428 		mutex_exit(&tod_lock);
429 	}
430 }
431 
432 /*
433  * This null routine is only used by sun4v watchdog timer support.
434  */
435 void
436 restore_watchdog_on_entry(void)
437 {
438 }
439 
440 int
441 kdi_watchdog_disable(void)
442 {
443 	if (watchdog_activated) {
444 		mutex_enter(&tod_lock);
445 		(void) tod_ops.tod_clear_watchdog_timer();
446 		mutex_exit(&tod_lock);
447 	}
448 
449 	return (watchdog_activated);
450 }
451 
452 void
453 kdi_watchdog_restore(void)
454 {
455 	if (watchdog_enable) {
456 		mutex_enter(&tod_lock);
457 		(void) tod_ops.tod_set_watchdog_timer(watchdog_timeout_seconds);
458 		mutex_exit(&tod_lock);
459 	}
460 }
461 
462 /*ARGSUSED*/
463 void
464 mach_dump_buffer_init(void)
465 {
466 	/*
467 	 * setup dump buffer to store extra crash information
468 	 * not applicable to sun4u
469 	 */
470 }
471 
472 /*
473  * xt_sync - wait for previous x-traps to finish
474  */
475 void
476 xt_sync(cpuset_t cpuset)
477 {
478 	kpreempt_disable();
479 	CPUSET_DEL(cpuset, CPU->cpu_id);
480 	CPUSET_AND(cpuset, cpu_ready_set);
481 	xt_some(cpuset, (xcfunc_t *)xt_sync_tl1, 0, 0);
482 	kpreempt_enable();
483 }
484 
485 /*
486  * mach_soft_state_init() - dummy routine for sun4v soft state
487  */
488 void
489 mach_soft_state_init(void)
490 {}
491