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 2010 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 * Copyright (c) 2018, Joyent, Inc.
26 */
27
28 /*
29 * Public interface to routines implemented by CPU modules
30 */
31
32 #include <sys/types.h>
33 #include <sys/atomic.h>
34 #include <sys/x86_archext.h>
35 #include <sys/cpu_module_impl.h>
36 #include <sys/cpu_module_ms.h>
37 #include <sys/fm/util.h>
38 #include <sys/reboot.h>
39 #include <sys/modctl.h>
40 #include <sys/param.h>
41 #include <sys/cmn_err.h>
42 #include <sys/systm.h>
43 #include <sys/fm/protocol.h>
44 #include <sys/pcb.h>
45 #include <sys/ontrap.h>
46 #include <sys/psw.h>
47 #include <sys/privregs.h>
48 #include <sys/machsystm.h>
49
50 /*
51 * Set to force cmi_init to fail.
52 */
53 int cmi_no_init = 0;
54
55 /*
56 * Set to avoid MCA initialization.
57 */
58 int cmi_no_mca_init = 0;
59
60 /*
61 * If cleared for debugging we will not attempt to load a model-specific
62 * cpu module but will load the generic cpu module instead.
63 */
64 int cmi_force_generic = 0;
65
66 /*
67 * If cleared for debugging, we will suppress panicking on fatal hardware
68 * errors. This should *only* be used for debugging; it use can and will
69 * cause data corruption if actual hardware errors are detected by the system.
70 */
71 int cmi_panic_on_uncorrectable_error = 1;
72
73 /*
74 * Subdirectory (relative to the module search path) in which we will
75 * look for cpu modules.
76 */
77 #define CPUMOD_SUBDIR "cpu"
78
79 /*
80 * CPU modules have a filenames such as "cpu.AuthenticAMD.15" and
81 * "cpu.generic" - the "cpu" prefix is specified by the following.
82 */
83 #define CPUMOD_PREFIX "cpu"
84
85 /*
86 * Structure used to keep track of cpu modules we have loaded and their ops
87 */
88 typedef struct cmi {
89 struct cmi *cmi_next;
90 struct cmi *cmi_prev;
91 const cmi_ops_t *cmi_ops;
92 struct modctl *cmi_modp;
93 uint_t cmi_refcnt;
94 } cmi_t;
95
96 static cmi_t *cmi_list;
97 static const cmi_mc_ops_t *cmi_mc_global_ops;
98 static void *cmi_mc_global_data;
99 static kmutex_t cmi_load_lock;
100
101 /*
102 * Functions we need from cmi_hw.c that are not part of the cpu_module.h
103 * interface.
104 */
105 extern cmi_hdl_t cmi_hdl_create(enum cmi_hdl_class, uint_t, uint_t, uint_t);
106 extern void cmi_hdl_destroy(cmi_hdl_t ophdl);
107 extern void cmi_hdl_setcmi(cmi_hdl_t, void *, void *);
108 extern void *cmi_hdl_getcmi(cmi_hdl_t);
109 extern void cmi_hdl_setmc(cmi_hdl_t, const struct cmi_mc_ops *, void *);
110 extern void cmi_hdl_inj_begin(cmi_hdl_t);
111 extern void cmi_hdl_inj_end(cmi_hdl_t);
112 extern void cmi_read_smbios(cmi_hdl_t);
113
114 #define HDL2CMI(hdl) cmi_hdl_getcmi(hdl)
115
116 #define CMI_OPS(cmi) (cmi)->cmi_ops
117 #define CMI_OP_PRESENT(cmi, op) ((cmi) && CMI_OPS(cmi)->op != NULL)
118
119 #define CMI_MATCH_VENDOR 0 /* Just match on vendor */
120 #define CMI_MATCH_FAMILY 1 /* Match down to family */
121 #define CMI_MATCH_MODEL 2 /* Match down to model */
122 #define CMI_MATCH_STEPPING 3 /* Match down to stepping */
123
124 static void
cmi_link(cmi_t * cmi)125 cmi_link(cmi_t *cmi)
126 {
127 ASSERT(MUTEX_HELD(&cmi_load_lock));
128
129 cmi->cmi_prev = NULL;
130 cmi->cmi_next = cmi_list;
131 if (cmi_list != NULL)
132 cmi_list->cmi_prev = cmi;
133 cmi_list = cmi;
134 }
135
136 static void
cmi_unlink(cmi_t * cmi)137 cmi_unlink(cmi_t *cmi)
138 {
139 ASSERT(MUTEX_HELD(&cmi_load_lock));
140 ASSERT(cmi->cmi_refcnt == 0);
141
142 if (cmi->cmi_prev != NULL)
143 cmi->cmi_prev = cmi->cmi_next;
144
145 if (cmi->cmi_next != NULL)
146 cmi->cmi_next->cmi_prev = cmi->cmi_prev;
147
148 if (cmi_list == cmi)
149 cmi_list = cmi->cmi_next;
150 }
151
152 /*
153 * Hold the module in memory. We call to CPU modules without using the
154 * stubs mechanism, so these modules must be manually held in memory.
155 * The mod_ref acts as if another loaded module has a dependency on us.
156 */
157 static void
cmi_hold(cmi_t * cmi)158 cmi_hold(cmi_t *cmi)
159 {
160 ASSERT(MUTEX_HELD(&cmi_load_lock));
161
162 mutex_enter(&mod_lock);
163 cmi->cmi_modp->mod_ref++;
164 mutex_exit(&mod_lock);
165 cmi->cmi_refcnt++;
166 }
167
168 static void
cmi_rele(cmi_t * cmi)169 cmi_rele(cmi_t *cmi)
170 {
171 ASSERT(MUTEX_HELD(&cmi_load_lock));
172
173 mutex_enter(&mod_lock);
174 cmi->cmi_modp->mod_ref--;
175 mutex_exit(&mod_lock);
176
177 if (--cmi->cmi_refcnt == 0) {
178 cmi_unlink(cmi);
179 kmem_free(cmi, sizeof (cmi_t));
180 }
181 }
182
183 static cmi_ops_t *
cmi_getops(modctl_t * modp)184 cmi_getops(modctl_t *modp)
185 {
186 cmi_ops_t *ops;
187
188 if ((ops = (cmi_ops_t *)modlookup_by_modctl(modp, "_cmi_ops")) ==
189 NULL) {
190 cmn_err(CE_WARN, "cpu module '%s' is invalid: no _cmi_ops "
191 "found", modp->mod_modname);
192 return (NULL);
193 }
194
195 if (ops->cmi_init == NULL) {
196 cmn_err(CE_WARN, "cpu module '%s' is invalid: no cmi_init "
197 "entry point", modp->mod_modname);
198 return (NULL);
199 }
200
201 return (ops);
202 }
203
204 static cmi_t *
cmi_load_modctl(modctl_t * modp)205 cmi_load_modctl(modctl_t *modp)
206 {
207 cmi_ops_t *ops;
208 uintptr_t ver;
209 cmi_t *cmi;
210 cmi_api_ver_t apiver;
211
212 ASSERT(MUTEX_HELD(&cmi_load_lock));
213
214 for (cmi = cmi_list; cmi != NULL; cmi = cmi->cmi_next) {
215 if (cmi->cmi_modp == modp)
216 return (cmi);
217 }
218
219 if ((ver = modlookup_by_modctl(modp, "_cmi_api_version")) == 0) {
220 /*
221 * Apparently a cpu module before versioning was introduced -
222 * we call this version 0.
223 */
224 apiver = CMI_API_VERSION_0;
225 } else {
226 apiver = *((cmi_api_ver_t *)ver);
227 if (!CMI_API_VERSION_CHKMAGIC(apiver)) {
228 cmn_err(CE_WARN, "cpu module '%s' is invalid: "
229 "_cmi_api_version 0x%x has bad magic",
230 modp->mod_modname, apiver);
231 return (NULL);
232 }
233 }
234
235 if (apiver != CMI_API_VERSION) {
236 cmn_err(CE_WARN, "cpu module '%s' has API version %d, "
237 "kernel requires API version %d", modp->mod_modname,
238 CMI_API_VERSION_TOPRINT(apiver),
239 CMI_API_VERSION_TOPRINT(CMI_API_VERSION));
240 return (NULL);
241 }
242
243 if ((ops = cmi_getops(modp)) == NULL)
244 return (NULL);
245
246 cmi = kmem_zalloc(sizeof (*cmi), KM_SLEEP);
247 cmi->cmi_ops = ops;
248 cmi->cmi_modp = modp;
249
250 cmi_link(cmi);
251
252 return (cmi);
253 }
254
255 static int
cmi_cpu_match(cmi_hdl_t hdl1,cmi_hdl_t hdl2,int match)256 cmi_cpu_match(cmi_hdl_t hdl1, cmi_hdl_t hdl2, int match)
257 {
258 if (match >= CMI_MATCH_VENDOR &&
259 cmi_hdl_vendor(hdl1) != cmi_hdl_vendor(hdl2))
260 return (0);
261
262 if (match >= CMI_MATCH_FAMILY &&
263 cmi_hdl_family(hdl1) != cmi_hdl_family(hdl2))
264 return (0);
265
266 if (match >= CMI_MATCH_MODEL &&
267 cmi_hdl_model(hdl1) != cmi_hdl_model(hdl2))
268 return (0);
269
270 if (match >= CMI_MATCH_STEPPING &&
271 cmi_hdl_stepping(hdl1) != cmi_hdl_stepping(hdl2))
272 return (0);
273
274 return (1);
275 }
276
277 static int
cmi_search_list_cb(cmi_hdl_t whdl,void * arg1,void * arg2,void * arg3)278 cmi_search_list_cb(cmi_hdl_t whdl, void *arg1, void *arg2, void *arg3)
279 {
280 cmi_hdl_t thdl = (cmi_hdl_t)arg1;
281 int match = *((int *)arg2);
282 cmi_hdl_t *rsltp = (cmi_hdl_t *)arg3;
283
284 if (cmi_cpu_match(thdl, whdl, match)) {
285 cmi_hdl_hold(whdl); /* short-term hold */
286 *rsltp = whdl;
287 return (CMI_HDL_WALK_DONE);
288 } else {
289 return (CMI_HDL_WALK_NEXT);
290 }
291 }
292
293 static cmi_t *
cmi_search_list(cmi_hdl_t hdl,int match)294 cmi_search_list(cmi_hdl_t hdl, int match)
295 {
296 cmi_hdl_t dhdl = NULL;
297 cmi_t *cmi = NULL;
298
299 ASSERT(MUTEX_HELD(&cmi_load_lock));
300
301 cmi_hdl_walk(cmi_search_list_cb, (void *)hdl, (void *)&match, &dhdl);
302 if (dhdl) {
303 cmi = HDL2CMI(dhdl);
304 cmi_hdl_rele(dhdl); /* held in cmi_search_list_cb */
305 }
306
307 return (cmi);
308 }
309
310 static cmi_t *
cmi_load_module(cmi_hdl_t hdl,int match,int * chosenp)311 cmi_load_module(cmi_hdl_t hdl, int match, int *chosenp)
312 {
313 modctl_t *modp;
314 cmi_t *cmi;
315 int modid;
316 uint_t s[3];
317
318 ASSERT(MUTEX_HELD(&cmi_load_lock));
319 ASSERT(match == CMI_MATCH_STEPPING || match == CMI_MATCH_MODEL ||
320 match == CMI_MATCH_FAMILY || match == CMI_MATCH_VENDOR);
321
322 /*
323 * Have we already loaded a module for a cpu with the same
324 * vendor/family/model/stepping?
325 */
326 if ((cmi = cmi_search_list(hdl, match)) != NULL) {
327 cmi_hold(cmi);
328 return (cmi);
329 }
330
331 s[0] = cmi_hdl_family(hdl);
332 s[1] = cmi_hdl_model(hdl);
333 s[2] = cmi_hdl_stepping(hdl);
334 modid = modload_qualified(CPUMOD_SUBDIR, CPUMOD_PREFIX,
335 cmi_hdl_vendorstr(hdl), ".", s, match, chosenp);
336
337 if (modid == -1)
338 return (NULL);
339
340 modp = mod_hold_by_id(modid);
341 cmi = cmi_load_modctl(modp);
342 if (cmi)
343 cmi_hold(cmi);
344 mod_release_mod(modp);
345
346 return (cmi);
347 }
348
349 /*
350 * Try to load a cpu module with specific support for this chip type.
351 */
352 static cmi_t *
cmi_load_specific(cmi_hdl_t hdl,void ** datap)353 cmi_load_specific(cmi_hdl_t hdl, void **datap)
354 {
355 cmi_t *cmi;
356 int err;
357 int i;
358
359 ASSERT(MUTEX_HELD(&cmi_load_lock));
360
361 for (i = CMI_MATCH_STEPPING; i >= CMI_MATCH_VENDOR; i--) {
362 int suffixlevel;
363
364 if ((cmi = cmi_load_module(hdl, i, &suffixlevel)) == NULL)
365 return (NULL);
366
367 /*
368 * A module has loaded and has a _cmi_ops structure, and the
369 * module has been held for this instance. Call its cmi_init
370 * entry point - we expect success (0) or ENOTSUP.
371 */
372 if ((err = cmi->cmi_ops->cmi_init(hdl, datap)) == 0) {
373 if (boothowto & RB_VERBOSE) {
374 printf("initialized cpu module '%s' on "
375 "chip %d core %d strand %d\n",
376 cmi->cmi_modp->mod_modname,
377 cmi_hdl_chipid(hdl), cmi_hdl_coreid(hdl),
378 cmi_hdl_strandid(hdl));
379 }
380 return (cmi);
381 } else if (err != ENOTSUP) {
382 cmn_err(CE_WARN, "failed to init cpu module '%s' on "
383 "chip %d core %d strand %d: err=%d\n",
384 cmi->cmi_modp->mod_modname,
385 cmi_hdl_chipid(hdl), cmi_hdl_coreid(hdl),
386 cmi_hdl_strandid(hdl), err);
387 }
388
389 /*
390 * The module failed or declined to init, so release
391 * it and update i to be equal to the number
392 * of suffices actually used in the last module path.
393 */
394 cmi_rele(cmi);
395 i = suffixlevel;
396 }
397
398 return (NULL);
399 }
400
401 /*
402 * Load the generic IA32 MCA cpu module, which may still supplement
403 * itself with model-specific support through cpu model-specific modules.
404 */
405 static cmi_t *
cmi_load_generic(cmi_hdl_t hdl,void ** datap)406 cmi_load_generic(cmi_hdl_t hdl, void **datap)
407 {
408 modctl_t *modp;
409 cmi_t *cmi;
410 int modid;
411 int err;
412
413 ASSERT(MUTEX_HELD(&cmi_load_lock));
414
415 if ((modid = modload(CPUMOD_SUBDIR, CPUMOD_PREFIX ".generic")) == -1)
416 return (NULL);
417
418 modp = mod_hold_by_id(modid);
419 cmi = cmi_load_modctl(modp);
420 if (cmi)
421 cmi_hold(cmi);
422 mod_release_mod(modp);
423
424 if (cmi == NULL)
425 return (NULL);
426
427 if ((err = cmi->cmi_ops->cmi_init(hdl, datap)) != 0) {
428 if (err != ENOTSUP)
429 cmn_err(CE_WARN, CPUMOD_PREFIX ".generic failed to "
430 "init: err=%d", err);
431 cmi_rele(cmi);
432 return (NULL);
433 }
434
435 return (cmi);
436 }
437
438 cmi_hdl_t
cmi_init(enum cmi_hdl_class class,uint_t chipid,uint_t coreid,uint_t strandid)439 cmi_init(enum cmi_hdl_class class, uint_t chipid, uint_t coreid,
440 uint_t strandid)
441 {
442 cmi_t *cmi = NULL;
443 cmi_hdl_t hdl;
444 void *data;
445
446 if (cmi_no_init) {
447 cmi_no_mca_init = 1;
448 return (NULL);
449 }
450
451 mutex_enter(&cmi_load_lock);
452
453 if ((hdl = cmi_hdl_create(class, chipid, coreid, strandid)) == NULL) {
454 mutex_exit(&cmi_load_lock);
455 cmn_err(CE_WARN, "There will be no MCA support on chip %d "
456 "core %d strand %d (cmi_hdl_create returned NULL)\n",
457 chipid, coreid, strandid);
458 return (NULL);
459 }
460
461 if (!cmi_force_generic)
462 cmi = cmi_load_specific(hdl, &data);
463
464 if (cmi == NULL && (cmi = cmi_load_generic(hdl, &data)) == NULL) {
465 cmn_err(CE_WARN, "There will be no MCA support on chip %d "
466 "core %d strand %d\n", chipid, coreid, strandid);
467 cmi_hdl_rele(hdl);
468 mutex_exit(&cmi_load_lock);
469 return (NULL);
470 }
471
472 cmi_hdl_setcmi(hdl, cmi, data);
473
474 cms_init(hdl);
475
476 cmi_read_smbios(hdl);
477
478 mutex_exit(&cmi_load_lock);
479
480 return (hdl);
481 }
482
483 /*
484 * cmi_fini is called on DR deconfigure of a cpu resource.
485 * It should not be called at simple offline of a cpu.
486 */
487 void
cmi_fini(cmi_hdl_t hdl)488 cmi_fini(cmi_hdl_t hdl)
489 {
490 cmi_t *cmi = HDL2CMI(hdl);
491
492 if (cms_present(hdl))
493 cms_fini(hdl);
494
495 if (CMI_OP_PRESENT(cmi, cmi_fini))
496 CMI_OPS(cmi)->cmi_fini(hdl);
497
498 cmi_hdl_destroy(hdl);
499 }
500
501 /*
502 * cmi_post_startup is called from post_startup for the boot cpu only (no
503 * other cpus are started yet).
504 */
505 void
cmi_post_startup(void)506 cmi_post_startup(void)
507 {
508 cmi_hdl_t hdl;
509 cmi_t *cmi;
510
511 if (cmi_no_mca_init != 0 ||
512 (hdl = cmi_hdl_any()) == NULL) /* short-term hold */
513 return;
514
515 cmi = HDL2CMI(hdl);
516
517 if (CMI_OP_PRESENT(cmi, cmi_post_startup))
518 CMI_OPS(cmi)->cmi_post_startup(hdl);
519
520 cmi_hdl_rele(hdl);
521 }
522
523 /*
524 * Called just once from start_other_cpus when all processors are started.
525 * This will not be called for each cpu, so the registered op must not
526 * assume it is called as such. We are not necessarily executing on
527 * the boot cpu.
528 */
529 void
cmi_post_mpstartup(void)530 cmi_post_mpstartup(void)
531 {
532 cmi_hdl_t hdl;
533 cmi_t *cmi;
534
535 if (cmi_no_mca_init != 0 ||
536 (hdl = cmi_hdl_any()) == NULL) /* short-term hold */
537 return;
538
539 cmi = HDL2CMI(hdl);
540
541 if (CMI_OP_PRESENT(cmi, cmi_post_mpstartup))
542 CMI_OPS(cmi)->cmi_post_mpstartup(hdl);
543
544 cmi_hdl_rele(hdl);
545 }
546
547 void
cmi_faulted_enter(cmi_hdl_t hdl)548 cmi_faulted_enter(cmi_hdl_t hdl)
549 {
550 cmi_t *cmi = HDL2CMI(hdl);
551
552 if (cmi_no_mca_init != 0)
553 return;
554
555 if (CMI_OP_PRESENT(cmi, cmi_faulted_enter))
556 CMI_OPS(cmi)->cmi_faulted_enter(hdl);
557 }
558
559 void
cmi_faulted_exit(cmi_hdl_t hdl)560 cmi_faulted_exit(cmi_hdl_t hdl)
561 {
562 cmi_t *cmi = HDL2CMI(hdl);
563
564 if (cmi_no_mca_init != 0)
565 return;
566
567 if (CMI_OP_PRESENT(cmi, cmi_faulted_exit))
568 CMI_OPS(cmi)->cmi_faulted_exit(hdl);
569 }
570
571 void
cmi_mca_init(cmi_hdl_t hdl)572 cmi_mca_init(cmi_hdl_t hdl)
573 {
574 cmi_t *cmi;
575
576 if (cmi_no_mca_init != 0)
577 return;
578
579 cmi = HDL2CMI(hdl);
580
581 if (CMI_OP_PRESENT(cmi, cmi_mca_init))
582 CMI_OPS(cmi)->cmi_mca_init(hdl);
583 }
584
585 #define CMI_RESPONSE_PANIC 0x0 /* panic must have value 0 */
586 #define CMI_RESPONSE_NONE 0x1
587 #define CMI_RESPONSE_CKILL 0x2
588 #define CMI_RESPONSE_REBOOT 0x3 /* not implemented */
589 #define CMI_RESPONSE_ONTRAP_PROT 0x4
590 #define CMI_RESPONSE_LOFAULT_PROT 0x5
591
592 /*
593 * Return 0 if we will panic in response to this machine check, otherwise
594 * non-zero. If the caller is cmi_mca_trap in this file then the nonzero
595 * return values are to be interpreted from CMI_RESPONSE_* above.
596 *
597 * This function must just return what will be done without actually
598 * doing anything; this includes not changing the regs.
599 */
600 int
cmi_mce_response(struct regs * rp,uint64_t disp)601 cmi_mce_response(struct regs *rp, uint64_t disp)
602 {
603 int panicrsp = cmi_panic_on_uncorrectable_error ? CMI_RESPONSE_PANIC :
604 CMI_RESPONSE_NONE;
605 on_trap_data_t *otp;
606
607 ASSERT(rp != NULL); /* don't call for polling, only on #MC */
608
609 /*
610 * If no bits are set in the disposition then there is nothing to
611 * worry about and we do not need to trampoline to ontrap or
612 * lofault handlers.
613 */
614 if (disp == 0)
615 return (CMI_RESPONSE_NONE);
616
617 /*
618 * Unconstrained errors cannot be forgiven, even by ontrap or
619 * lofault protection. The data is not poisoned and may not
620 * even belong to the trapped context - eg a writeback of
621 * data that is found to be bad.
622 */
623 if (disp & CMI_ERRDISP_UC_UNCONSTRAINED)
624 return (panicrsp);
625
626 /*
627 * ontrap OT_DATA_EC and lofault protection forgive any disposition
628 * other than unconstrained, even those normally forced fatal.
629 */
630 if ((otp = curthread->t_ontrap) != NULL && otp->ot_prot & OT_DATA_EC)
631 return (CMI_RESPONSE_ONTRAP_PROT);
632 else if (curthread->t_lofault)
633 return (CMI_RESPONSE_LOFAULT_PROT);
634
635 /*
636 * Forced-fatal errors are terminal even in user mode.
637 */
638 if (disp & CMI_ERRDISP_FORCEFATAL)
639 return (panicrsp);
640
641 /*
642 * If the trapped context is corrupt or we have no instruction pointer
643 * to resume at (and aren't trampolining to a fault handler)
644 * then in the kernel case we must panic and in usermode we
645 * kill the affected contract.
646 */
647 if (disp & (CMI_ERRDISP_CURCTXBAD | CMI_ERRDISP_RIPV_INVALID))
648 return (USERMODE(rp->r_cs) ? CMI_RESPONSE_CKILL : panicrsp);
649
650 /*
651 * Anything else is harmless
652 */
653 return (CMI_RESPONSE_NONE);
654 }
655
656 int cma_mca_trap_panic_suppressed = 0;
657
658 static void
cmi_mca_panic(void)659 cmi_mca_panic(void)
660 {
661 if (cmi_panic_on_uncorrectable_error) {
662 fm_panic("Unrecoverable Machine-Check Exception");
663 } else {
664 cmn_err(CE_WARN, "suppressing panic from fatal #mc");
665 cma_mca_trap_panic_suppressed++;
666 }
667 }
668
669
670 int cma_mca_trap_contract_kills = 0;
671 int cma_mca_trap_ontrap_forgiven = 0;
672 int cma_mca_trap_lofault_forgiven = 0;
673
674 /*
675 * Native #MC handler - we branch to here from mcetrap
676 */
677 /*ARGSUSED*/
678 void
cmi_mca_trap(struct regs * rp)679 cmi_mca_trap(struct regs *rp)
680 {
681 #ifndef __xpv
682 cmi_hdl_t hdl = NULL;
683 uint64_t disp;
684 cmi_t *cmi;
685 int s;
686
687 if (cmi_no_mca_init != 0)
688 return;
689
690 /*
691 * This function can call cmn_err, and the cpu module cmi_mca_trap
692 * entry point may also elect to call cmn_err (e.g., if it can't
693 * log the error onto an errorq, say very early in boot).
694 * We need to let cprintf know that we must not block.
695 */
696 s = spl8();
697
698 if ((hdl = cmi_hdl_lookup(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
699 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) == NULL ||
700 (cmi = HDL2CMI(hdl)) == NULL ||
701 !CMI_OP_PRESENT(cmi, cmi_mca_trap)) {
702
703 cmn_err(CE_WARN, "#MC exception on cpuid %d: %s",
704 CPU->cpu_id,
705 hdl ? "handle lookup ok but no #MC handler found" :
706 "handle lookup failed");
707
708 if (hdl != NULL)
709 cmi_hdl_rele(hdl);
710
711 splx(s);
712 return;
713 }
714
715 disp = CMI_OPS(cmi)->cmi_mca_trap(hdl, rp);
716
717 switch (cmi_mce_response(rp, disp)) {
718 default:
719 cmn_err(CE_WARN, "Invalid response from cmi_mce_response");
720 /*FALLTHRU*/
721
722 case CMI_RESPONSE_PANIC:
723 cmi_mca_panic();
724 break;
725
726 case CMI_RESPONSE_NONE:
727 break;
728
729 case CMI_RESPONSE_CKILL:
730 ttolwp(curthread)->lwp_pcb.pcb_flags |= ASYNC_HWERR;
731 aston(curthread);
732 cma_mca_trap_contract_kills++;
733 break;
734
735 case CMI_RESPONSE_ONTRAP_PROT: {
736 on_trap_data_t *otp = curthread->t_ontrap;
737 otp->ot_trap = OT_DATA_EC;
738 rp->r_pc = otp->ot_trampoline;
739 cma_mca_trap_ontrap_forgiven++;
740 break;
741 }
742
743 case CMI_RESPONSE_LOFAULT_PROT:
744 rp->r_r0 = EFAULT;
745 rp->r_pc = curthread->t_lofault;
746 cma_mca_trap_lofault_forgiven++;
747 break;
748 }
749
750 cmi_hdl_rele(hdl);
751 splx(s);
752 #endif /* __xpv */
753 }
754
755 void
cmi_hdl_poke(cmi_hdl_t hdl)756 cmi_hdl_poke(cmi_hdl_t hdl)
757 {
758 cmi_t *cmi = HDL2CMI(hdl);
759
760 if (!CMI_OP_PRESENT(cmi, cmi_hdl_poke))
761 return;
762
763 CMI_OPS(cmi)->cmi_hdl_poke(hdl);
764 }
765
766 #ifndef __xpv
767 void
cmi_cmci_trap()768 cmi_cmci_trap()
769 {
770 cmi_hdl_t hdl = NULL;
771 cmi_t *cmi;
772
773 if (cmi_no_mca_init != 0)
774 return;
775
776 if ((hdl = cmi_hdl_lookup(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
777 cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) == NULL ||
778 (cmi = HDL2CMI(hdl)) == NULL ||
779 !CMI_OP_PRESENT(cmi, cmi_cmci_trap)) {
780
781 cmn_err(CE_WARN, "CMCI interrupt on cpuid %d: %s",
782 CPU->cpu_id,
783 hdl ? "handle lookup ok but no CMCI handler found" :
784 "handle lookup failed");
785
786 if (hdl != NULL)
787 cmi_hdl_rele(hdl);
788
789 return;
790 }
791
792 CMI_OPS(cmi)->cmi_cmci_trap(hdl);
793
794 cmi_hdl_rele(hdl);
795 }
796 #endif /* __xpv */
797
798 void
cmi_mc_register(cmi_hdl_t hdl,const cmi_mc_ops_t * mcops,void * mcdata)799 cmi_mc_register(cmi_hdl_t hdl, const cmi_mc_ops_t *mcops, void *mcdata)
800 {
801 if (!cmi_no_mca_init)
802 cmi_hdl_setmc(hdl, mcops, mcdata);
803 }
804
805 cmi_errno_t
cmi_mc_register_global(const cmi_mc_ops_t * mcops,void * mcdata)806 cmi_mc_register_global(const cmi_mc_ops_t *mcops, void *mcdata)
807 {
808 if (!cmi_no_mca_init) {
809 if (cmi_mc_global_ops != NULL || cmi_mc_global_data != NULL ||
810 mcops == NULL || mcops->cmi_mc_patounum == NULL ||
811 mcops->cmi_mc_unumtopa == NULL) {
812 return (CMIERR_UNKNOWN);
813 }
814 cmi_mc_global_data = mcdata;
815 cmi_mc_global_ops = mcops;
816 }
817 return (CMI_SUCCESS);
818 }
819
820 void
cmi_mc_sw_memscrub_disable(void)821 cmi_mc_sw_memscrub_disable(void)
822 {
823 memscrub_disable();
824 }
825
826 cmi_errno_t
cmi_mc_patounum(uint64_t pa,uint8_t valid_hi,uint8_t valid_lo,uint32_t synd,int syndtype,mc_unum_t * up)827 cmi_mc_patounum(uint64_t pa, uint8_t valid_hi, uint8_t valid_lo, uint32_t synd,
828 int syndtype, mc_unum_t *up)
829 {
830 const struct cmi_mc_ops *mcops;
831 cmi_hdl_t hdl;
832 cmi_errno_t rv;
833
834 if (cmi_no_mca_init)
835 return (CMIERR_MC_ABSENT);
836
837 if (cmi_mc_global_ops != NULL) {
838 if (cmi_mc_global_ops->cmi_mc_patounum == NULL)
839 return (CMIERR_MC_NOTSUP);
840 return (cmi_mc_global_ops->cmi_mc_patounum(cmi_mc_global_data,
841 pa, valid_hi, valid_lo, synd, syndtype, up));
842 }
843
844 if ((hdl = cmi_hdl_any()) == NULL) /* short-term hold */
845 return (CMIERR_MC_ABSENT);
846
847 if ((mcops = cmi_hdl_getmcops(hdl)) == NULL ||
848 mcops->cmi_mc_patounum == NULL) {
849 cmi_hdl_rele(hdl);
850 return (CMIERR_MC_NOTSUP);
851 }
852
853 rv = mcops->cmi_mc_patounum(cmi_hdl_getmcdata(hdl), pa, valid_hi,
854 valid_lo, synd, syndtype, up);
855
856 cmi_hdl_rele(hdl);
857
858 return (rv);
859 }
860
861 cmi_errno_t
cmi_mc_unumtopa(mc_unum_t * up,nvlist_t * nvl,uint64_t * pap)862 cmi_mc_unumtopa(mc_unum_t *up, nvlist_t *nvl, uint64_t *pap)
863 {
864 const struct cmi_mc_ops *mcops;
865 cmi_hdl_t hdl;
866 cmi_errno_t rv;
867 nvlist_t *hcsp;
868
869 if (up != NULL && nvl != NULL)
870 return (CMIERR_API); /* convert from just one form */
871
872 if (cmi_no_mca_init)
873 return (CMIERR_MC_ABSENT);
874
875 if (cmi_mc_global_ops != NULL) {
876 if (cmi_mc_global_ops->cmi_mc_unumtopa == NULL)
877 return (CMIERR_MC_NOTSUP);
878 return (cmi_mc_global_ops->cmi_mc_unumtopa(cmi_mc_global_data,
879 up, nvl, pap));
880 }
881
882 if ((hdl = cmi_hdl_any()) == NULL) /* short-term hold */
883 return (CMIERR_MC_ABSENT);
884
885 if ((mcops = cmi_hdl_getmcops(hdl)) == NULL ||
886 mcops->cmi_mc_unumtopa == NULL) {
887 cmi_hdl_rele(hdl);
888
889 if (nvl != NULL && nvlist_lookup_nvlist(nvl,
890 FM_FMRI_HC_SPECIFIC, &hcsp) == 0 &&
891 (nvlist_lookup_uint64(hcsp,
892 "asru-" FM_FMRI_HC_SPECIFIC_PHYSADDR, pap) == 0 ||
893 nvlist_lookup_uint64(hcsp, FM_FMRI_HC_SPECIFIC_PHYSADDR,
894 pap) == 0)) {
895 return (CMIERR_MC_PARTIALUNUMTOPA);
896 } else {
897 return (mcops && mcops->cmi_mc_unumtopa == NULL ?
898 CMIERR_MC_NOTSUP : CMIERR_MC_ABSENT);
899 }
900 }
901
902 rv = mcops->cmi_mc_unumtopa(cmi_hdl_getmcdata(hdl), up, nvl, pap);
903
904 cmi_hdl_rele(hdl);
905
906 return (rv);
907 }
908
909 void
cmi_mc_logout(cmi_hdl_t hdl,boolean_t ismc,boolean_t sync)910 cmi_mc_logout(cmi_hdl_t hdl, boolean_t ismc, boolean_t sync)
911 {
912 const struct cmi_mc_ops *mcops;
913
914 if (cmi_no_mca_init)
915 return;
916
917 if (cmi_mc_global_ops != NULL)
918 mcops = cmi_mc_global_ops;
919 else
920 mcops = cmi_hdl_getmcops(hdl);
921
922 if (mcops != NULL && mcops->cmi_mc_logout != NULL)
923 mcops->cmi_mc_logout(hdl, ismc, sync);
924 }
925
926 cmi_errno_t
cmi_hdl_msrinject(cmi_hdl_t hdl,cmi_mca_regs_t * regs,uint_t nregs,int force)927 cmi_hdl_msrinject(cmi_hdl_t hdl, cmi_mca_regs_t *regs, uint_t nregs,
928 int force)
929 {
930 cmi_t *cmi = cmi_hdl_getcmi(hdl);
931 cmi_errno_t rc;
932
933 if (!CMI_OP_PRESENT(cmi, cmi_msrinject))
934 return (CMIERR_NOTSUP);
935
936 cmi_hdl_inj_begin(hdl);
937 rc = CMI_OPS(cmi)->cmi_msrinject(hdl, regs, nregs, force);
938 cmi_hdl_inj_end(hdl);
939
940 return (rc);
941 }
942
943 boolean_t
cmi_panic_on_ue(void)944 cmi_panic_on_ue(void)
945 {
946 return (cmi_panic_on_uncorrectable_error ? B_TRUE : B_FALSE);
947 }
948
949 void
cmi_panic_callback(void)950 cmi_panic_callback(void)
951 {
952 cmi_hdl_t hdl;
953 cmi_t *cmi;
954
955 if (cmi_no_mca_init || (hdl = cmi_hdl_any()) == NULL)
956 return;
957
958 cmi = cmi_hdl_getcmi(hdl);
959 if (CMI_OP_PRESENT(cmi, cmi_panic_callback))
960 CMI_OPS(cmi)->cmi_panic_callback();
961
962 cmi_hdl_rele(hdl);
963 }
964
965
966 const char *
cmi_hdl_chipident(cmi_hdl_t hdl)967 cmi_hdl_chipident(cmi_hdl_t hdl)
968 {
969 cmi_t *cmi = cmi_hdl_getcmi(hdl);
970
971 if (!CMI_OP_PRESENT(cmi, cmi_ident))
972 return (NULL);
973
974 return (CMI_OPS(cmi)->cmi_ident(hdl));
975 }
976