xref: /illumos-gate/usr/src/uts/i86pc/os/cmi.c (revision 257873cfc1dd3337766407f80397db60a56f2f5a)
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 /*
23  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * Public interface to routines implemented by CPU modules
29  */
30 
31 #include <sys/types.h>
32 #include <sys/atomic.h>
33 #include <sys/x86_archext.h>
34 #include <sys/cpu_module_impl.h>
35 #include <sys/cpu_module_ms.h>
36 #include <sys/fm/util.h>
37 #include <sys/reboot.h>
38 #include <sys/modctl.h>
39 #include <sys/param.h>
40 #include <sys/cmn_err.h>
41 #include <sys/systm.h>
42 #include <sys/fm/protocol.h>
43 #include <sys/pcb.h>
44 #include <sys/ontrap.h>
45 #include <sys/psw.h>
46 #include <sys/privregs.h>
47 #include <sys/machsystm.h>
48 
49 /*
50  * Set to force cmi_init to fail.
51  */
52 int cmi_no_init = 0;
53 
54 /*
55  * Set to avoid MCA initialization.
56  */
57 int cmi_no_mca_init = 0;
58 
59 /*
60  * If cleared for debugging we will not attempt to load a model-specific
61  * cpu module but will load the generic cpu module instead.
62  */
63 int cmi_force_generic = 0;
64 
65 /*
66  * If cleared for debugging, we will suppress panicking on fatal hardware
67  * errors.  This should *only* be used for debugging; it use can and will
68  * cause data corruption if actual hardware errors are detected by the system.
69  */
70 int cmi_panic_on_uncorrectable_error = 1;
71 
72 #ifndef __xpv
73 /*
74  * Set to indicate whether we are able to enable cmci interrupt.
75  */
76 int cmi_enable_cmci = 0;
77 #endif
78 
79 /*
80  * Subdirectory (relative to the module search path) in which we will
81  * look for cpu modules.
82  */
83 #define	CPUMOD_SUBDIR	"cpu"
84 
85 /*
86  * CPU modules have a filenames such as "cpu.AuthenticAMD.15" and
87  * "cpu.generic" - the "cpu" prefix is specified by the following.
88  */
89 #define	CPUMOD_PREFIX	"cpu"
90 
91 /*
92  * Structure used to keep track of cpu modules we have loaded and their ops
93  */
94 typedef struct cmi {
95 	struct cmi *cmi_next;
96 	struct cmi *cmi_prev;
97 	const cmi_ops_t *cmi_ops;
98 	struct modctl *cmi_modp;
99 	uint_t cmi_refcnt;
100 } cmi_t;
101 
102 static cmi_t *cmi_list;
103 static kmutex_t cmi_load_lock;
104 
105 /*
106  * Functions we need from cmi_hw.c that are not part of the cpu_module.h
107  * interface.
108  */
109 extern cmi_hdl_t cmi_hdl_create(enum cmi_hdl_class, uint_t, uint_t, uint_t);
110 extern void cmi_hdl_setcmi(cmi_hdl_t, void *, void *);
111 extern void *cmi_hdl_getcmi(cmi_hdl_t);
112 extern void cmi_hdl_setmc(cmi_hdl_t, const struct cmi_mc_ops *, void *);
113 extern void cmi_hdl_inj_begin(cmi_hdl_t);
114 extern void cmi_hdl_inj_end(cmi_hdl_t);
115 
116 #define	HDL2CMI(hdl)		cmi_hdl_getcmi(hdl)
117 
118 #define	CMI_OPS(cmi)		(cmi)->cmi_ops
119 #define	CMI_OP_PRESENT(cmi, op)	((cmi) && CMI_OPS(cmi)->op != NULL)
120 
121 #define	CMI_MATCH_VENDOR	0	/* Just match on vendor */
122 #define	CMI_MATCH_FAMILY	1	/* Match down to family */
123 #define	CMI_MATCH_MODEL		2	/* Match down to model */
124 #define	CMI_MATCH_STEPPING	3	/* Match down to stepping */
125 
126 static void
127 cmi_link(cmi_t *cmi)
128 {
129 	ASSERT(MUTEX_HELD(&cmi_load_lock));
130 
131 	cmi->cmi_prev = NULL;
132 	cmi->cmi_next = cmi_list;
133 	if (cmi_list != NULL)
134 		cmi_list->cmi_prev = cmi;
135 	cmi_list = cmi;
136 }
137 
138 static void
139 cmi_unlink(cmi_t *cmi)
140 {
141 	ASSERT(MUTEX_HELD(&cmi_load_lock));
142 	ASSERT(cmi->cmi_refcnt == 0);
143 
144 	if (cmi->cmi_prev != NULL)
145 		cmi->cmi_prev = cmi->cmi_next;
146 
147 	if (cmi->cmi_next != NULL)
148 		cmi->cmi_next->cmi_prev = cmi->cmi_prev;
149 
150 	if (cmi_list == cmi)
151 		cmi_list = cmi->cmi_next;
152 }
153 
154 /*
155  * Hold the module in memory.  We call to CPU modules without using the
156  * stubs mechanism, so these modules must be manually held in memory.
157  * The mod_ref acts as if another loaded module has a dependency on us.
158  */
159 static void
160 cmi_hold(cmi_t *cmi)
161 {
162 	ASSERT(MUTEX_HELD(&cmi_load_lock));
163 
164 	mutex_enter(&mod_lock);
165 	cmi->cmi_modp->mod_ref++;
166 	mutex_exit(&mod_lock);
167 	cmi->cmi_refcnt++;
168 }
169 
170 static void
171 cmi_rele(cmi_t *cmi)
172 {
173 	ASSERT(MUTEX_HELD(&cmi_load_lock));
174 
175 	mutex_enter(&mod_lock);
176 	cmi->cmi_modp->mod_ref--;
177 	mutex_exit(&mod_lock);
178 
179 	if (--cmi->cmi_refcnt == 0) {
180 		cmi_unlink(cmi);
181 		kmem_free(cmi, sizeof (cmi_t));
182 	}
183 }
184 
185 static cmi_ops_t *
186 cmi_getops(modctl_t *modp)
187 {
188 	cmi_ops_t *ops;
189 
190 	if ((ops = (cmi_ops_t *)modlookup_by_modctl(modp, "_cmi_ops")) ==
191 	    NULL) {
192 		cmn_err(CE_WARN, "cpu module '%s' is invalid: no _cmi_ops "
193 		    "found", modp->mod_modname);
194 		return (NULL);
195 	}
196 
197 	if (ops->cmi_init == NULL) {
198 		cmn_err(CE_WARN, "cpu module '%s' is invalid: no cmi_init "
199 		    "entry point", modp->mod_modname);
200 		return (NULL);
201 	}
202 
203 	return (ops);
204 }
205 
206 static cmi_t *
207 cmi_load_modctl(modctl_t *modp)
208 {
209 	cmi_ops_t *ops;
210 	uintptr_t ver;
211 	cmi_t *cmi;
212 	cmi_api_ver_t apiver;
213 
214 	ASSERT(MUTEX_HELD(&cmi_load_lock));
215 
216 	for (cmi = cmi_list; cmi != NULL; cmi = cmi->cmi_next) {
217 		if (cmi->cmi_modp == modp)
218 			return (cmi);
219 	}
220 
221 	if ((ver = modlookup_by_modctl(modp, "_cmi_api_version")) == NULL) {
222 		/*
223 		 * Apparently a cpu module before versioning was introduced -
224 		 * we call this version 0.
225 		 */
226 		apiver = CMI_API_VERSION_0;
227 	} else {
228 		apiver = *((cmi_api_ver_t *)ver);
229 		if (!CMI_API_VERSION_CHKMAGIC(apiver)) {
230 			cmn_err(CE_WARN, "cpu module '%s' is invalid: "
231 			    "_cmi_api_version 0x%x has bad magic",
232 			    modp->mod_modname, apiver);
233 			return (NULL);
234 		}
235 	}
236 
237 	if (apiver != CMI_API_VERSION) {
238 		cmn_err(CE_WARN, "cpu module '%s' has API version %d, "
239 		    "kernel requires API version %d", modp->mod_modname,
240 		    CMI_API_VERSION_TOPRINT(apiver),
241 		    CMI_API_VERSION_TOPRINT(CMI_API_VERSION));
242 		return (NULL);
243 	}
244 
245 	if ((ops = cmi_getops(modp)) == NULL)
246 		return (NULL);
247 
248 	cmi = kmem_zalloc(sizeof (*cmi), KM_SLEEP);
249 	cmi->cmi_ops = ops;
250 	cmi->cmi_modp = modp;
251 
252 	cmi_link(cmi);
253 
254 	return (cmi);
255 }
256 
257 static int
258 cmi_cpu_match(cmi_hdl_t hdl1, cmi_hdl_t hdl2, int match)
259 {
260 	if (match >= CMI_MATCH_VENDOR &&
261 	    cmi_hdl_vendor(hdl1) != cmi_hdl_vendor(hdl2))
262 		return (0);
263 
264 	if (match >= CMI_MATCH_FAMILY &&
265 	    cmi_hdl_family(hdl1) != cmi_hdl_family(hdl2))
266 		return (0);
267 
268 	if (match >= CMI_MATCH_MODEL &&
269 	    cmi_hdl_model(hdl1) != cmi_hdl_model(hdl2))
270 		return (0);
271 
272 	if (match >= CMI_MATCH_STEPPING &&
273 	    cmi_hdl_stepping(hdl1) != cmi_hdl_stepping(hdl2))
274 		return (0);
275 
276 	return (1);
277 }
278 
279 static int
280 cmi_search_list_cb(cmi_hdl_t whdl, void *arg1, void *arg2, void *arg3)
281 {
282 	cmi_hdl_t thdl = (cmi_hdl_t)arg1;
283 	int match = *((int *)arg2);
284 	cmi_hdl_t *rsltp = (cmi_hdl_t *)arg3;
285 
286 	if (cmi_cpu_match(thdl, whdl, match)) {
287 		cmi_hdl_hold(whdl);	/* short-term hold */
288 		*rsltp = whdl;
289 		return (CMI_HDL_WALK_DONE);
290 	} else {
291 		return (CMI_HDL_WALK_NEXT);
292 	}
293 }
294 
295 static cmi_t *
296 cmi_search_list(cmi_hdl_t hdl, int match)
297 {
298 	cmi_hdl_t dhdl = NULL;
299 	cmi_t *cmi = NULL;
300 
301 	ASSERT(MUTEX_HELD(&cmi_load_lock));
302 
303 	cmi_hdl_walk(cmi_search_list_cb, (void *)hdl, (void *)&match, &dhdl);
304 	if (dhdl) {
305 		cmi = HDL2CMI(dhdl);
306 		cmi_hdl_rele(dhdl);	/* held in cmi_search_list_cb */
307 	}
308 
309 	return (cmi);
310 }
311 
312 static cmi_t *
313 cmi_load_module(cmi_hdl_t hdl, int match, int *chosenp)
314 {
315 	modctl_t *modp;
316 	cmi_t *cmi;
317 	int modid;
318 	uint_t s[3];
319 
320 	ASSERT(MUTEX_HELD(&cmi_load_lock));
321 	ASSERT(match == CMI_MATCH_STEPPING || match == CMI_MATCH_MODEL ||
322 	    match == CMI_MATCH_FAMILY || match == CMI_MATCH_VENDOR);
323 
324 	/*
325 	 * Have we already loaded a module for a cpu with the same
326 	 * vendor/family/model/stepping?
327 	 */
328 	if ((cmi = cmi_search_list(hdl, match)) != NULL) {
329 		cmi_hold(cmi);
330 		return (cmi);
331 	}
332 
333 	s[0] = cmi_hdl_family(hdl);
334 	s[1] = cmi_hdl_model(hdl);
335 	s[2] = cmi_hdl_stepping(hdl);
336 	modid = modload_qualified(CPUMOD_SUBDIR, CPUMOD_PREFIX,
337 	    cmi_hdl_vendorstr(hdl), ".", s, match, chosenp);
338 
339 	if (modid == -1)
340 		return (NULL);
341 
342 	modp = mod_hold_by_id(modid);
343 	cmi = cmi_load_modctl(modp);
344 	if (cmi)
345 		cmi_hold(cmi);
346 	mod_release_mod(modp);
347 
348 	return (cmi);
349 }
350 
351 /*
352  * Try to load a cpu module with specific support for this chip type.
353  */
354 static cmi_t *
355 cmi_load_specific(cmi_hdl_t hdl, void **datap)
356 {
357 	cmi_t *cmi;
358 	int err;
359 	int i;
360 
361 	ASSERT(MUTEX_HELD(&cmi_load_lock));
362 
363 	for (i = CMI_MATCH_STEPPING; i >= CMI_MATCH_VENDOR; i--) {
364 		int suffixlevel;
365 
366 		if ((cmi = cmi_load_module(hdl, i, &suffixlevel)) == NULL)
367 			return (NULL);
368 
369 		/*
370 		 * A module has loaded and has a _cmi_ops structure, and the
371 		 * module has been held for this instance.  Call its cmi_init
372 		 * entry point - we expect success (0) or ENOTSUP.
373 		 */
374 		if ((err = cmi->cmi_ops->cmi_init(hdl, datap)) == 0) {
375 			if (boothowto & RB_VERBOSE) {
376 				printf("initialized cpu module '%s' on "
377 				    "chip %d core %d strand %d\n",
378 				    cmi->cmi_modp->mod_modname,
379 				    cmi_hdl_chipid(hdl), cmi_hdl_coreid(hdl),
380 				    cmi_hdl_strandid(hdl));
381 			}
382 			return (cmi);
383 		} else if (err != ENOTSUP) {
384 			cmn_err(CE_WARN, "failed to init cpu module '%s' on "
385 			    "chip %d core %d strand %d: err=%d\n",
386 			    cmi->cmi_modp->mod_modname,
387 			    cmi_hdl_chipid(hdl), cmi_hdl_coreid(hdl),
388 			    cmi_hdl_strandid(hdl), err);
389 		}
390 
391 		/*
392 		 * The module failed or declined to init, so release
393 		 * it and update i to be equal to the number
394 		 * of suffices actually used in the last module path.
395 		 */
396 		cmi_rele(cmi);
397 		i = suffixlevel;
398 	}
399 
400 	return (NULL);
401 }
402 
403 /*
404  * Load the generic IA32 MCA cpu module, which may still supplement
405  * itself with model-specific support through cpu model-specific modules.
406  */
407 static cmi_t *
408 cmi_load_generic(cmi_hdl_t hdl, void **datap)
409 {
410 	modctl_t *modp;
411 	cmi_t *cmi;
412 	int modid;
413 	int err;
414 
415 	ASSERT(MUTEX_HELD(&cmi_load_lock));
416 
417 	if ((modid = modload(CPUMOD_SUBDIR, CPUMOD_PREFIX ".generic")) == -1)
418 		return (NULL);
419 
420 	modp = mod_hold_by_id(modid);
421 	cmi = cmi_load_modctl(modp);
422 	if (cmi)
423 		cmi_hold(cmi);
424 	mod_release_mod(modp);
425 
426 	if (cmi == NULL)
427 		return (NULL);
428 
429 	if ((err = cmi->cmi_ops->cmi_init(hdl, datap)) != 0) {
430 		if (err != ENOTSUP)
431 			cmn_err(CE_WARN, CPUMOD_PREFIX ".generic failed to "
432 			    "init: err=%d", err);
433 		cmi_rele(cmi);
434 		return (NULL);
435 	}
436 
437 	return (cmi);
438 }
439 
440 cmi_hdl_t
441 cmi_init(enum cmi_hdl_class class, uint_t chipid, uint_t coreid,
442     uint_t strandid)
443 {
444 	cmi_t *cmi = NULL;
445 	cmi_hdl_t hdl;
446 	void *data;
447 
448 	if (cmi_no_init) {
449 		cmi_no_mca_init = 1;
450 		return (NULL);
451 	}
452 
453 	mutex_enter(&cmi_load_lock);
454 
455 	if ((hdl = cmi_hdl_create(class, chipid, coreid, strandid)) == NULL) {
456 		mutex_exit(&cmi_load_lock);
457 		cmn_err(CE_WARN, "There will be no MCA support on chip %d "
458 		    "core %d strand %d (cmi_hdl_create returned NULL)\n",
459 		    chipid, coreid, strandid);
460 		return (NULL);
461 	}
462 
463 	if (!cmi_force_generic)
464 		cmi = cmi_load_specific(hdl, &data);
465 
466 	if (cmi == NULL && (cmi = cmi_load_generic(hdl, &data)) == NULL) {
467 		cmn_err(CE_WARN, "There will be no MCA support on chip %d "
468 		    "core %d strand %d\n", chipid, coreid, strandid);
469 		cmi_hdl_rele(hdl);
470 		mutex_exit(&cmi_load_lock);
471 		return (NULL);
472 	}
473 
474 	cmi_hdl_setcmi(hdl, cmi, data);
475 
476 	cms_init(hdl);
477 
478 	mutex_exit(&cmi_load_lock);
479 
480 	return (hdl);
481 }
482 
483 /*
484  * cmi_fini is not called at the moment.  It is intended to be called
485  * on DR deconfigure of a cpu resource.  It should not be called at
486  * simple offline of a cpu.
487  */
488 void
489 cmi_fini(cmi_hdl_t hdl)
490 {
491 	cmi_t *cmi = HDL2CMI(hdl);
492 
493 	if (cms_present(hdl))
494 		cms_fini(hdl);
495 
496 	if (CMI_OP_PRESENT(cmi, cmi_fini))
497 		CMI_OPS(cmi)->cmi_fini(hdl);
498 
499 	cmi_hdl_rele(hdl);	/* release hold obtained in cmi_hdl_create */
500 }
501 
502 /*
503  * cmi_post_startup is called from post_startup for the boot cpu only (no
504  * other cpus are started yet).
505  */
506 void
507 cmi_post_startup(void)
508 {
509 	cmi_hdl_t hdl;
510 	cmi_t *cmi;
511 
512 	if (cmi_no_mca_init != 0 ||
513 	    (hdl = cmi_hdl_any()) == NULL)	/* short-term hold */
514 		return;
515 
516 	cmi = HDL2CMI(hdl);
517 
518 	if (CMI_OP_PRESENT(cmi, cmi_post_startup))
519 		CMI_OPS(cmi)->cmi_post_startup(hdl);
520 
521 	cmi_hdl_rele(hdl);
522 }
523 
524 /*
525  * Called just once from start_other_cpus when all processors are started.
526  * This will not be called for each cpu, so the registered op must not
527  * assume it is called as such.  We are not necessarily executing on
528  * the boot cpu.
529  */
530 void
531 cmi_post_mpstartup(void)
532 {
533 	cmi_hdl_t hdl;
534 	cmi_t *cmi;
535 
536 	if (cmi_no_mca_init != 0 ||
537 	    (hdl = cmi_hdl_any()) == NULL)	/* short-term hold */
538 		return;
539 
540 	cmi = HDL2CMI(hdl);
541 
542 	if (CMI_OP_PRESENT(cmi, cmi_post_mpstartup))
543 		CMI_OPS(cmi)->cmi_post_mpstartup(hdl);
544 
545 	cmi_hdl_rele(hdl);
546 }
547 
548 void
549 cmi_faulted_enter(cmi_hdl_t hdl)
550 {
551 	cmi_t *cmi = HDL2CMI(hdl);
552 
553 	if (cmi_no_mca_init != 0)
554 		return;
555 
556 	if (CMI_OP_PRESENT(cmi, cmi_faulted_enter))
557 		CMI_OPS(cmi)->cmi_faulted_enter(hdl);
558 }
559 
560 void
561 cmi_faulted_exit(cmi_hdl_t hdl)
562 {
563 	cmi_t *cmi = HDL2CMI(hdl);
564 
565 	if (cmi_no_mca_init != 0)
566 		return;
567 
568 	if (CMI_OP_PRESENT(cmi, cmi_faulted_exit))
569 		CMI_OPS(cmi)->cmi_faulted_exit(hdl);
570 }
571 
572 void
573 cmi_mca_init(cmi_hdl_t hdl)
574 {
575 	cmi_t *cmi;
576 
577 	if (cmi_no_mca_init != 0)
578 		return;
579 
580 	cmi = HDL2CMI(hdl);
581 
582 	if (CMI_OP_PRESENT(cmi, cmi_mca_init))
583 		CMI_OPS(cmi)->cmi_mca_init(hdl);
584 }
585 
586 #define	CMI_RESPONSE_PANIC		0x0	/* panic must have value 0 */
587 #define	CMI_RESPONSE_NONE		0x1
588 #define	CMI_RESPONSE_CKILL		0x2
589 #define	CMI_RESPONSE_REBOOT		0x3	/* not implemented */
590 #define	CMI_RESPONSE_ONTRAP_PROT	0x4
591 #define	CMI_RESPONSE_LOFAULT_PROT	0x5
592 
593 /*
594  * Return 0 if we will panic in response to this machine check, otherwise
595  * non-zero.  If the caller is cmi_mca_trap in this file then the nonzero
596  * return values are to be interpreted from CMI_RESPONSE_* above.
597  *
598  * This function must just return what will be done without actually
599  * doing anything; this includes not changing the regs.
600  */
601 int
602 cmi_mce_response(struct regs *rp, uint64_t disp)
603 {
604 	int panicrsp = cmi_panic_on_uncorrectable_error ? CMI_RESPONSE_PANIC :
605 	    CMI_RESPONSE_NONE;
606 	on_trap_data_t *otp;
607 
608 	ASSERT(rp != NULL);	/* don't call for polling, only on #MC */
609 
610 	/*
611 	 * If no bits are set in the disposition then there is nothing to
612 	 * worry about and we do not need to trampoline to ontrap or
613 	 * lofault handlers.
614 	 */
615 	if (disp == 0)
616 		return (CMI_RESPONSE_NONE);
617 
618 	/*
619 	 * Unconstrained errors cannot be forgiven, even by ontrap or
620 	 * lofault protection.  The data is not poisoned and may not
621 	 * even belong to the trapped context - eg a writeback of
622 	 * data that is found to be bad.
623 	 */
624 	if (disp & CMI_ERRDISP_UC_UNCONSTRAINED)
625 		return (panicrsp);
626 
627 	/*
628 	 * ontrap OT_DATA_EC and lofault protection forgive any disposition
629 	 * other than unconstrained, even those normally forced fatal.
630 	 */
631 	if ((otp = curthread->t_ontrap) != NULL && otp->ot_prot & OT_DATA_EC)
632 		return (CMI_RESPONSE_ONTRAP_PROT);
633 	else if (curthread->t_lofault)
634 		return (CMI_RESPONSE_LOFAULT_PROT);
635 
636 	/*
637 	 * Forced-fatal errors are terminal even in user mode.
638 	 */
639 	if (disp & CMI_ERRDISP_FORCEFATAL)
640 		return (panicrsp);
641 
642 	/*
643 	 * If the trapped context is corrupt or we have no instruction pointer
644 	 * to resume at (and aren't trampolining to a fault handler)
645 	 * then in the kernel case we must panic and in usermode we
646 	 * kill the affected contract.
647 	 */
648 	if (disp & (CMI_ERRDISP_CURCTXBAD | CMI_ERRDISP_RIPV_INVALID))
649 		return (USERMODE(rp->r_cs) ?  CMI_RESPONSE_CKILL : panicrsp);
650 
651 	/*
652 	 * Anything else is harmless
653 	 */
654 	return (CMI_RESPONSE_NONE);
655 }
656 
657 int cma_mca_trap_panic_suppressed = 0;
658 
659 static void
660 cmi_mca_panic(void)
661 {
662 	if (cmi_panic_on_uncorrectable_error) {
663 		fm_panic("Unrecoverable Machine-Check Exception");
664 	} else {
665 		cmn_err(CE_WARN, "suppressing panic from fatal #mc");
666 		cma_mca_trap_panic_suppressed++;
667 	}
668 }
669 
670 
671 int cma_mca_trap_contract_kills = 0;
672 int cma_mca_trap_ontrap_forgiven = 0;
673 int cma_mca_trap_lofault_forgiven = 0;
674 
675 /*
676  * Native #MC handler - we branch to here from mcetrap
677  */
678 /*ARGSUSED*/
679 void
680 cmi_mca_trap(struct regs *rp)
681 {
682 #ifndef	__xpv
683 	cmi_hdl_t hdl = NULL;
684 	uint64_t disp;
685 	cmi_t *cmi;
686 	int s;
687 
688 	if (cmi_no_mca_init != 0)
689 		return;
690 
691 	/*
692 	 * This function can call cmn_err, and the cpu module cmi_mca_trap
693 	 * entry point may also elect to call cmn_err (e.g., if it can't
694 	 * log the error onto an errorq, say very early in boot).
695 	 * We need to let cprintf know that we must not block.
696 	 */
697 	s = spl8();
698 
699 	if ((hdl = cmi_hdl_lookup(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
700 	    cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) == NULL ||
701 	    (cmi = HDL2CMI(hdl)) == NULL ||
702 	    !CMI_OP_PRESENT(cmi, cmi_mca_trap)) {
703 
704 		cmn_err(CE_WARN, "#MC exception on cpuid %d: %s",
705 		    CPU->cpu_id,
706 		    hdl ? "handle lookup ok but no #MC handler found" :
707 		    "handle lookup failed");
708 
709 		if (hdl != NULL)
710 			cmi_hdl_rele(hdl);
711 
712 		splx(s);
713 		return;
714 	}
715 
716 	disp = CMI_OPS(cmi)->cmi_mca_trap(hdl, rp);
717 
718 	switch (cmi_mce_response(rp, disp)) {
719 	default:
720 		cmn_err(CE_WARN, "Invalid response from cmi_mce_response");
721 		/*FALLTHRU*/
722 
723 	case CMI_RESPONSE_PANIC:
724 		cmi_mca_panic();
725 		break;
726 
727 	case CMI_RESPONSE_NONE:
728 		break;
729 
730 	case CMI_RESPONSE_CKILL:
731 		ttolwp(curthread)->lwp_pcb.pcb_flags |= ASYNC_HWERR;
732 		aston(curthread);
733 		cma_mca_trap_contract_kills++;
734 		break;
735 
736 	case CMI_RESPONSE_ONTRAP_PROT: {
737 		on_trap_data_t *otp = curthread->t_ontrap;
738 		otp->ot_trap = OT_DATA_EC;
739 		rp->r_pc = otp->ot_trampoline;
740 		cma_mca_trap_ontrap_forgiven++;
741 		break;
742 	}
743 
744 	case CMI_RESPONSE_LOFAULT_PROT:
745 		rp->r_r0 = EFAULT;
746 		rp->r_pc = curthread->t_lofault;
747 		cma_mca_trap_lofault_forgiven++;
748 		break;
749 	}
750 
751 	cmi_hdl_rele(hdl);
752 	splx(s);
753 #endif	/* __xpv */
754 }
755 
756 void
757 cmi_hdl_poke(cmi_hdl_t hdl)
758 {
759 	cmi_t *cmi = HDL2CMI(hdl);
760 
761 	if (!CMI_OP_PRESENT(cmi, cmi_hdl_poke))
762 		return;
763 
764 	CMI_OPS(cmi)->cmi_hdl_poke(hdl);
765 }
766 
767 #ifndef	__xpv
768 void
769 cmi_cmci_trap()
770 {
771 	cmi_hdl_t hdl = NULL;
772 	cmi_t *cmi;
773 
774 	if (cmi_no_mca_init != 0)
775 		return;
776 
777 	if ((hdl = cmi_hdl_lookup(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
778 	    cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) == NULL ||
779 	    (cmi = HDL2CMI(hdl)) == NULL ||
780 	    !CMI_OP_PRESENT(cmi, cmi_cmci_trap)) {
781 
782 		cmn_err(CE_WARN, "CMCI interrupt on cpuid %d: %s",
783 		    CPU->cpu_id,
784 		    hdl ? "handle lookup ok but no CMCI handler found" :
785 		    "handle lookup failed");
786 
787 		if (hdl != NULL)
788 			cmi_hdl_rele(hdl);
789 
790 		return;
791 	}
792 
793 	CMI_OPS(cmi)->cmi_cmci_trap(hdl);
794 
795 	cmi_hdl_rele(hdl);
796 }
797 #endif	/* __xpv */
798 
799 void
800 cmi_mc_register(cmi_hdl_t hdl, const cmi_mc_ops_t *mcops, void *mcdata)
801 {
802 	if (!cmi_no_mca_init)
803 		cmi_hdl_setmc(hdl, mcops, mcdata);
804 }
805 
806 void
807 cmi_mc_sw_memscrub_disable(void)
808 {
809 	memscrub_disable();
810 }
811 
812 cmi_errno_t
813 cmi_mc_patounum(uint64_t pa, uint8_t valid_hi, uint8_t valid_lo, uint32_t synd,
814     int syndtype, mc_unum_t *up)
815 {
816 	const struct cmi_mc_ops *mcops;
817 	cmi_hdl_t hdl;
818 	cmi_errno_t rv;
819 
820 	if (cmi_no_mca_init ||
821 	    (hdl = cmi_hdl_any()) == NULL)	/* short-term hold */
822 		return (CMIERR_MC_ABSENT);
823 
824 	if ((mcops = cmi_hdl_getmcops(hdl)) == NULL ||
825 	    mcops->cmi_mc_patounum == NULL) {
826 		cmi_hdl_rele(hdl);
827 		return (CMIERR_MC_NOTSUP);
828 	}
829 
830 	rv = mcops->cmi_mc_patounum(cmi_hdl_getmcdata(hdl), pa, valid_hi,
831 	    valid_lo, synd, syndtype, up);
832 
833 	cmi_hdl_rele(hdl);
834 
835 	return (rv);
836 }
837 
838 cmi_errno_t
839 cmi_mc_unumtopa(mc_unum_t *up, nvlist_t *nvl, uint64_t *pap)
840 {
841 	const struct cmi_mc_ops *mcops;
842 	cmi_hdl_t hdl;
843 	cmi_errno_t rv;
844 	nvlist_t *hcsp;
845 
846 	if (up != NULL && nvl != NULL)
847 		return (CMIERR_API);	/* convert from just one form */
848 
849 	if (cmi_no_mca_init ||
850 	    (hdl = cmi_hdl_any()) == NULL)	/* short-term hold */
851 		return (CMIERR_MC_ABSENT);
852 
853 	if ((mcops = cmi_hdl_getmcops(hdl)) == NULL ||
854 	    mcops->cmi_mc_unumtopa == NULL) {
855 		cmi_hdl_rele(hdl);
856 
857 		if (nvl != NULL && nvlist_lookup_nvlist(nvl,
858 		    FM_FMRI_HC_SPECIFIC, &hcsp) == 0 &&
859 		    (nvlist_lookup_uint64(hcsp,
860 		    "asru-" FM_FMRI_HC_SPECIFIC_PHYSADDR, pap) == 0 ||
861 		    nvlist_lookup_uint64(hcsp, FM_FMRI_HC_SPECIFIC_PHYSADDR,
862 		    pap) == 0)) {
863 			return (CMIERR_MC_PARTIALUNUMTOPA);
864 		} else {
865 			return (mcops && mcops->cmi_mc_unumtopa ?
866 			    CMIERR_MC_NOTSUP : CMIERR_MC_ABSENT);
867 		}
868 	}
869 
870 	rv = mcops->cmi_mc_unumtopa(cmi_hdl_getmcdata(hdl), up, nvl, pap);
871 
872 	cmi_hdl_rele(hdl);
873 
874 	return (rv);
875 }
876 
877 void
878 cmi_mc_logout(cmi_hdl_t hdl, boolean_t ismc, boolean_t sync)
879 {
880 	const struct cmi_mc_ops *mcops;
881 
882 	if (cmi_no_mca_init || (mcops = cmi_hdl_getmcops(hdl)) == NULL)
883 		return;
884 
885 	if (mcops->cmi_mc_logout != NULL)
886 		mcops->cmi_mc_logout(hdl, ismc, sync);
887 }
888 
889 cmi_errno_t
890 cmi_hdl_msrinject(cmi_hdl_t hdl, cmi_mca_regs_t *regs, uint_t nregs,
891     int force)
892 {
893 	cmi_t *cmi = cmi_hdl_getcmi(hdl);
894 	cmi_errno_t rc;
895 
896 	if (!CMI_OP_PRESENT(cmi, cmi_msrinject))
897 		return (CMIERR_NOTSUP);
898 
899 	cmi_hdl_inj_begin(hdl);
900 	rc = CMI_OPS(cmi)->cmi_msrinject(hdl, regs, nregs, force);
901 	cmi_hdl_inj_end(hdl);
902 
903 	return (rc);
904 }
905 
906 boolean_t
907 cmi_panic_on_ue(void)
908 {
909 	return (cmi_panic_on_uncorrectable_error ? B_TRUE : B_FALSE);
910 }
911 
912 void
913 cmi_panic_callback(void)
914 {
915 	cmi_hdl_t hdl;
916 	cmi_t *cmi;
917 
918 	if (cmi_no_mca_init || (hdl = cmi_hdl_any()) == NULL)
919 		return;
920 
921 	cmi = cmi_hdl_getcmi(hdl);
922 	if (CMI_OP_PRESENT(cmi, cmi_panic_callback))
923 		CMI_OPS(cmi)->cmi_panic_callback();
924 
925 	cmi_hdl_rele(hdl);
926 }
927