1
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance 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 (c) 1992, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25 /*
26 * Copyright (c) 2009-2010, Intel Corporation.
27 * All rights reserved.
28 */
29
30 #define PSMI_1_7
31 #include <sys/smp_impldefs.h>
32 #include <sys/psm.h>
33 #include <sys/psm_modctl.h>
34 #include <sys/pit.h>
35 #include <sys/cmn_err.h>
36 #include <sys/strlog.h>
37 #include <sys/clock.h>
38 #include <sys/debug.h>
39 #include <sys/rtc.h>
40 #include <sys/x86_archext.h>
41 #include <sys/cpupart.h>
42 #include <sys/cpuvar.h>
43 #include <sys/cpu_event.h>
44 #include <sys/cmt.h>
45 #include <sys/cpu.h>
46 #include <sys/disp.h>
47 #include <sys/archsystm.h>
48 #include <sys/machsystm.h>
49 #include <sys/sysmacros.h>
50 #include <sys/memlist.h>
51 #include <sys/param.h>
52 #include <sys/promif.h>
53 #include <sys/cpu_pm.h>
54 #if defined(__xpv)
55 #include <sys/hypervisor.h>
56 #endif
57 #include <sys/mach_intr.h>
58 #include <vm/hat_i86.h>
59 #include <sys/kdi_machimpl.h>
60 #include <sys/sdt.h>
61 #include <sys/hpet.h>
62 #include <sys/sunddi.h>
63 #include <sys/sunndi.h>
64 #include <sys/cpc_pcbe.h>
65
66 #define OFFSETOF(s, m) (size_t)(&(((s *)0)->m))
67
68 /*
69 * Local function prototypes
70 */
71 static int mp_disable_intr(processorid_t cpun);
72 static void mp_enable_intr(processorid_t cpun);
73 static void mach_init();
74 static void mach_picinit();
75 static int machhztomhz(uint64_t cpu_freq_hz);
76 static uint64_t mach_getcpufreq(void);
77 static void mach_fixcpufreq(void);
78 static int mach_clkinit(int, int *);
79 static void mach_smpinit(void);
80 static int mach_softlvl_to_vect(int ipl);
81 static void mach_get_platform(int owner);
82 static void mach_construct_info();
83 static int mach_translate_irq(dev_info_t *dip, int irqno);
84 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
85 psm_intr_op_t, int *);
86 static void mach_notify_error(int level, char *errmsg);
87 static hrtime_t dummy_hrtime(void);
88 static void dummy_scalehrtime(hrtime_t *);
89 static uint64_t dummy_unscalehrtime(hrtime_t);
90 void cpu_idle(void);
91 static void cpu_wakeup(cpu_t *, int);
92 #ifndef __xpv
93 void cpu_idle_mwait(void);
94 static void cpu_wakeup_mwait(cpu_t *, int);
95 #endif
96 static int mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp);
97
98 /*
99 * External reference functions
100 */
101 extern void return_instr();
102 extern uint64_t freq_tsc(uint32_t *);
103 #if defined(__i386)
104 extern uint64_t freq_notsc(uint32_t *);
105 #endif
106 extern void pc_gethrestime(timestruc_t *);
107 extern int cpuid_get_coreid(cpu_t *);
108 extern int cpuid_get_chipid(cpu_t *);
109
110 /*
111 * PSM functions initialization
112 */
113 void (*psm_shutdownf)(int, int) = (void (*)(int, int))return_instr;
114 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
115 void (*psm_notifyf)(int) = (void (*)(int))return_instr;
116 void (*psm_set_idle_cpuf)(int) = (void (*)(int))return_instr;
117 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
118 void (*psminitf)() = mach_init;
119 void (*picinitf)() = return_instr;
120 int (*clkinitf)(int, int *) = (int (*)(int, int *))return_instr;
121 int (*ap_mlsetup)() = (int (*)(void))return_instr;
122 void (*send_dirintf)() = return_instr;
123 void (*setspl)(int) = (void (*)(int))return_instr;
124 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
125 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
126 int (*get_pending_spl)(void) = (int (*)(void))return_instr;
127 int (*addintr)(void *, int, avfunc, char *, int, caddr_t, caddr_t,
128 uint64_t *, dev_info_t *) = NULL;
129 void (*remintr)(void *, int, avfunc, int) = NULL;
130 void (*kdisetsoftint)(int, struct av_softinfo *)=
131 (void (*)(int, struct av_softinfo *))return_instr;
132 void (*setsoftint)(int, struct av_softinfo *)=
133 (void (*)(int, struct av_softinfo *))return_instr;
134 int (*slvltovect)(int) = (int (*)(int))return_instr;
135 int (*setlvl)(int, int *) = (int (*)(int, int *))return_instr;
136 void (*setlvlx)(int, int) = (void (*)(int, int))return_instr;
137 int (*psm_disable_intr)(int) = mp_disable_intr;
138 void (*psm_enable_intr)(int) = mp_enable_intr;
139 hrtime_t (*gethrtimef)(void) = dummy_hrtime;
140 hrtime_t (*gethrtimeunscaledf)(void) = dummy_hrtime;
141 void (*scalehrtimef)(hrtime_t *) = dummy_scalehrtime;
142 uint64_t (*unscalehrtimef)(hrtime_t) = dummy_unscalehrtime;
143 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
144 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
145 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
146 int (*psm_get_clockirq)(int) = NULL;
147 int (*psm_get_ipivect)(int, int) = NULL;
148 uchar_t (*psm_get_ioapicid)(uchar_t) = NULL;
149 uint32_t (*psm_get_localapicid)(uint32_t) = NULL;
150 uchar_t (*psm_xlate_vector_by_irq)(uchar_t) = NULL;
151
152 int (*psm_clkinit)(int) = NULL;
153 void (*psm_timer_reprogram)(hrtime_t) = NULL;
154 void (*psm_timer_enable)(void) = NULL;
155 void (*psm_timer_disable)(void) = NULL;
156 void (*psm_post_cyclic_setup)(void *arg) = NULL;
157 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
158 int *) = mach_intr_ops;
159 int (*psm_state)(psm_state_request_t *) = (int (*)(psm_state_request_t *))
160 return_instr;
161
162 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
163 void (*hrtime_tick)(void) = return_instr;
164
165 int (*psm_cpu_create_devinfo)(cpu_t *, dev_info_t **) = mach_cpu_create_devinfo;
166 int (*psm_cpu_get_devinfo)(cpu_t *, dev_info_t **) = NULL;
167
168 /* global IRM pool for APIX (PSM) module */
169 ddi_irm_pool_t *apix_irm_pool_p = NULL;
170
171 /*
172 * True if the generic TSC code is our source of hrtime, rather than whatever
173 * the PSM can provide.
174 */
175 #ifdef __xpv
176 int tsc_gethrtime_enable = 0;
177 #else
178 int tsc_gethrtime_enable = 1;
179 #endif
180 int tsc_gethrtime_initted = 0;
181
182 /*
183 * True if the hrtime implementation is "hires"; namely, better than microdata.
184 */
185 int gethrtime_hires = 0;
186
187 /*
188 * Local Static Data
189 */
190 static struct psm_ops mach_ops;
191 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
192 static ushort_t mach_ver[4] = {0, 0, 0, 0};
193
194 /*
195 * virtualization support for psm
196 */
197 void *psm_vt_ops = NULL;
198 /*
199 * If non-zero, idle cpus will become "halted" when there's
200 * no work to do.
201 */
202 int idle_cpu_use_hlt = 1;
203
204 #ifndef __xpv
205 /*
206 * If non-zero, idle cpus will use mwait if available to halt instead of hlt.
207 */
208 int idle_cpu_prefer_mwait = 1;
209 /*
210 * Set to 0 to avoid MONITOR+CLFLUSH assertion.
211 */
212 int idle_cpu_assert_cflush_monitor = 1;
213
214 /*
215 * If non-zero, idle cpus will not use power saving Deep C-States idle loop.
216 */
217 int idle_cpu_no_deep_c = 0;
218 /*
219 * Non-power saving idle loop and wakeup pointers.
220 * Allows user to toggle Deep Idle power saving feature on/off.
221 */
222 void (*non_deep_idle_cpu)() = cpu_idle;
223 void (*non_deep_idle_disp_enq_thread)(cpu_t *, int);
224
225 /*
226 * Object for the kernel to access the HPET.
227 */
228 hpet_t hpet;
229
230 #endif /* ifndef __xpv */
231
232 uint_t cp_haltset_fanout = 0;
233
234 /*ARGSUSED*/
235 int
pg_plat_hw_shared(cpu_t * cp,pghw_type_t hw)236 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
237 {
238 switch (hw) {
239 case PGHW_IPIPE:
240 if (is_x86_feature(x86_featureset, X86FSET_HTT)) {
241 /*
242 * Hyper-threading is SMT
243 */
244 return (1);
245 } else {
246 return (0);
247 }
248 case PGHW_PROCNODE:
249 if (cpuid_get_procnodes_per_pkg(cp) > 1)
250 return (1);
251 else
252 return (0);
253 case PGHW_CHIP:
254 if (is_x86_feature(x86_featureset, X86FSET_CMP) ||
255 is_x86_feature(x86_featureset, X86FSET_HTT))
256 return (1);
257 else
258 return (0);
259 case PGHW_CACHE:
260 if (cpuid_get_ncpu_sharing_last_cache(cp) > 1)
261 return (1);
262 else
263 return (0);
264 case PGHW_POW_ACTIVE:
265 if (cpupm_domain_id(cp, CPUPM_DTYPE_ACTIVE) != (id_t)-1)
266 return (1);
267 else
268 return (0);
269 case PGHW_POW_IDLE:
270 if (cpupm_domain_id(cp, CPUPM_DTYPE_IDLE) != (id_t)-1)
271 return (1);
272 else
273 return (0);
274 default:
275 return (0);
276 }
277 }
278
279 /*
280 * Compare two CPUs and see if they have a pghw_type_t sharing relationship
281 * If pghw_type_t is an unsupported hardware type, then return -1
282 */
283 int
pg_plat_cpus_share(cpu_t * cpu_a,cpu_t * cpu_b,pghw_type_t hw)284 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
285 {
286 id_t pgp_a, pgp_b;
287
288 pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
289 pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
290
291 if (pgp_a == -1 || pgp_b == -1)
292 return (-1);
293
294 return (pgp_a == pgp_b);
295 }
296
297 /*
298 * Return a physical instance identifier for known hardware sharing
299 * relationships
300 */
301 id_t
pg_plat_hw_instance_id(cpu_t * cpu,pghw_type_t hw)302 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
303 {
304 switch (hw) {
305 case PGHW_IPIPE:
306 return (cpuid_get_coreid(cpu));
307 case PGHW_CACHE:
308 return (cpuid_get_last_lvl_cacheid(cpu));
309 case PGHW_PROCNODE:
310 return (cpuid_get_procnodeid(cpu));
311 case PGHW_CHIP:
312 return (cpuid_get_chipid(cpu));
313 case PGHW_POW_ACTIVE:
314 return (cpupm_domain_id(cpu, CPUPM_DTYPE_ACTIVE));
315 case PGHW_POW_IDLE:
316 return (cpupm_domain_id(cpu, CPUPM_DTYPE_IDLE));
317 default:
318 return (-1);
319 }
320 }
321
322 /*
323 * Express preference for optimizing for sharing relationship
324 * hw1 vs hw2
325 */
326 pghw_type_t
pg_plat_hw_rank(pghw_type_t hw1,pghw_type_t hw2)327 pg_plat_hw_rank(pghw_type_t hw1, pghw_type_t hw2)
328 {
329 int i, rank1, rank2;
330
331 static pghw_type_t hw_hier[] = {
332 PGHW_IPIPE,
333 PGHW_CACHE,
334 PGHW_PROCNODE,
335 PGHW_CHIP,
336 PGHW_POW_IDLE,
337 PGHW_POW_ACTIVE,
338 PGHW_NUM_COMPONENTS
339 };
340
341 for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
342 if (hw_hier[i] == hw1)
343 rank1 = i;
344 if (hw_hier[i] == hw2)
345 rank2 = i;
346 }
347
348 if (rank1 > rank2)
349 return (hw1);
350 else
351 return (hw2);
352 }
353
354 /*
355 * Override the default CMT dispatcher policy for the specified
356 * hardware sharing relationship
357 */
358 pg_cmt_policy_t
pg_plat_cmt_policy(pghw_type_t hw)359 pg_plat_cmt_policy(pghw_type_t hw)
360 {
361 /*
362 * For shared caches, also load balance across them to
363 * maximize aggregate cache capacity
364 */
365 switch (hw) {
366 case PGHW_CACHE:
367 return (CMT_BALANCE|CMT_AFFINITY);
368 default:
369 return (CMT_NO_POLICY);
370 }
371 }
372
373 id_t
pg_plat_get_core_id(cpu_t * cpu)374 pg_plat_get_core_id(cpu_t *cpu)
375 {
376 return ((id_t)cpuid_get_coreid(cpu));
377 }
378
379 void
cmp_set_nosteal_interval(void)380 cmp_set_nosteal_interval(void)
381 {
382 /* Set the nosteal interval (used by disp_getbest()) to 100us */
383 nosteal_nsec = 100000UL;
384 }
385
386 /*
387 * Routine to ensure initial callers to hrtime gets 0 as return
388 */
389 static hrtime_t
dummy_hrtime(void)390 dummy_hrtime(void)
391 {
392 return (0);
393 }
394
395 /* ARGSUSED */
396 static void
dummy_scalehrtime(hrtime_t * ticks)397 dummy_scalehrtime(hrtime_t *ticks)
398 {}
399
400 static uint64_t
dummy_unscalehrtime(hrtime_t nsecs)401 dummy_unscalehrtime(hrtime_t nsecs)
402 {
403 return ((uint64_t)nsecs);
404 }
405
406 /*
407 * Supports Deep C-State power saving idle loop.
408 */
409 void
cpu_idle_adaptive(void)410 cpu_idle_adaptive(void)
411 {
412 (*CPU->cpu_m.mcpu_idle_cpu)();
413 }
414
415 /*
416 * Function called by CPU idle notification framework to check whether CPU
417 * has been awakened. It will be called with interrupt disabled.
418 * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
419 * notification framework.
420 */
421 /*ARGSUSED*/
422 static void
cpu_idle_check_wakeup(void * arg)423 cpu_idle_check_wakeup(void *arg)
424 {
425 /*
426 * Toggle interrupt flag to detect pending interrupts.
427 * If interrupt happened, do_interrupt() will notify CPU idle
428 * notification framework so no need to call cpu_idle_exit() here.
429 */
430 sti();
431 SMT_PAUSE();
432 cli();
433 }
434
435 /*
436 * Idle the present CPU until wakened via an interrupt
437 */
438 void
cpu_idle(void)439 cpu_idle(void)
440 {
441 cpu_t *cpup = CPU;
442 processorid_t cpu_sid = cpup->cpu_seqid;
443 cpupart_t *cp = cpup->cpu_part;
444 int hset_update = 1;
445
446 /*
447 * If this CPU is online, and there's multiple CPUs
448 * in the system, then we should notate our halting
449 * by adding ourselves to the partition's halted CPU
450 * bitmap. This allows other CPUs to find/awaken us when
451 * work becomes available.
452 */
453 if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
454 hset_update = 0;
455
456 /*
457 * Add ourselves to the partition's halted CPUs bitmap
458 * and set our HALTED flag, if necessary.
459 *
460 * When a thread becomes runnable, it is placed on the queue
461 * and then the halted CPU bitmap is checked to determine who
462 * (if anyone) should be awakened. We therefore need to first
463 * add ourselves to the bitmap, and and then check if there
464 * is any work available. The order is important to prevent a race
465 * that can lead to work languishing on a run queue somewhere while
466 * this CPU remains halted.
467 *
468 * Either the producing CPU will see we're halted and will awaken us,
469 * or this CPU will see the work available in disp_anywork().
470 *
471 * Note that memory barriers after updating the HALTED flag
472 * are not necessary since an atomic operation (updating the bitset)
473 * immediately follows. On x86 the atomic operation acts as a
474 * memory barrier for the update of cpu_disp_flags.
475 */
476 if (hset_update) {
477 cpup->cpu_disp_flags |= CPU_DISP_HALTED;
478 bitset_atomic_add(&cp->cp_haltset, cpu_sid);
479 }
480
481 /*
482 * Check to make sure there's really nothing to do.
483 * Work destined for this CPU may become available after
484 * this check. We'll be notified through the clearing of our
485 * bit in the halted CPU bitmap, and a poke.
486 */
487 if (disp_anywork()) {
488 if (hset_update) {
489 cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
490 bitset_atomic_del(&cp->cp_haltset, cpu_sid);
491 }
492 return;
493 }
494
495 /*
496 * We're on our way to being halted.
497 *
498 * Disable interrupts now, so that we'll awaken immediately
499 * after halting if someone tries to poke us between now and
500 * the time we actually halt.
501 *
502 * We check for the presence of our bit after disabling interrupts.
503 * If it's cleared, we'll return. If the bit is cleared after
504 * we check then the poke will pop us out of the halted state.
505 *
506 * This means that the ordering of the poke and the clearing
507 * of the bit by cpu_wakeup is important.
508 * cpu_wakeup() must clear, then poke.
509 * cpu_idle() must disable interrupts, then check for the bit.
510 */
511 cli();
512
513 if (hset_update && bitset_in_set(&cp->cp_haltset, cpu_sid) == 0) {
514 cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
515 sti();
516 return;
517 }
518
519 /*
520 * The check for anything locally runnable is here for performance
521 * and isn't needed for correctness. disp_nrunnable ought to be
522 * in our cache still, so it's inexpensive to check, and if there
523 * is anything runnable we won't have to wait for the poke.
524 */
525 if (cpup->cpu_disp->disp_nrunnable != 0) {
526 if (hset_update) {
527 cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
528 bitset_atomic_del(&cp->cp_haltset, cpu_sid);
529 }
530 sti();
531 return;
532 }
533
534 if (cpu_idle_enter(IDLE_STATE_C1, 0,
535 cpu_idle_check_wakeup, NULL) == 0) {
536 mach_cpu_idle();
537 cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
538 }
539
540 /*
541 * We're no longer halted
542 */
543 if (hset_update) {
544 cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
545 bitset_atomic_del(&cp->cp_haltset, cpu_sid);
546 }
547 }
548
549
550 /*
551 * If "cpu" is halted, then wake it up clearing its halted bit in advance.
552 * Otherwise, see if other CPUs in the cpu partition are halted and need to
553 * be woken up so that they can steal the thread we placed on this CPU.
554 * This function is only used on MP systems.
555 */
556 static void
cpu_wakeup(cpu_t * cpu,int bound)557 cpu_wakeup(cpu_t *cpu, int bound)
558 {
559 uint_t cpu_found;
560 processorid_t cpu_sid;
561 cpupart_t *cp;
562
563 cp = cpu->cpu_part;
564 cpu_sid = cpu->cpu_seqid;
565 if (bitset_in_set(&cp->cp_haltset, cpu_sid)) {
566 /*
567 * Clear the halted bit for that CPU since it will be
568 * poked in a moment.
569 */
570 bitset_atomic_del(&cp->cp_haltset, cpu_sid);
571 /*
572 * We may find the current CPU present in the halted cpuset
573 * if we're in the context of an interrupt that occurred
574 * before we had a chance to clear our bit in cpu_idle().
575 * Poking ourself is obviously unnecessary, since if
576 * we're here, we're not halted.
577 */
578 if (cpu != CPU)
579 poke_cpu(cpu->cpu_id);
580 return;
581 } else {
582 /*
583 * This cpu isn't halted, but it's idle or undergoing a
584 * context switch. No need to awaken anyone else.
585 */
586 if (cpu->cpu_thread == cpu->cpu_idle_thread ||
587 cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
588 return;
589 }
590
591 /*
592 * No need to wake up other CPUs if this is for a bound thread.
593 */
594 if (bound)
595 return;
596
597 /*
598 * The CPU specified for wakeup isn't currently halted, so check
599 * to see if there are any other halted CPUs in the partition,
600 * and if there are then awaken one.
601 */
602 do {
603 cpu_found = bitset_find(&cp->cp_haltset);
604 if (cpu_found == (uint_t)-1)
605 return;
606 } while (bitset_atomic_test_and_del(&cp->cp_haltset, cpu_found) < 0);
607
608 if (cpu_found != CPU->cpu_seqid) {
609 poke_cpu(cpu_seq[cpu_found]->cpu_id);
610 }
611 }
612
613 #ifndef __xpv
614 /*
615 * Function called by CPU idle notification framework to check whether CPU
616 * has been awakened. It will be called with interrupt disabled.
617 * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
618 * notification framework.
619 */
620 static void
cpu_idle_mwait_check_wakeup(void * arg)621 cpu_idle_mwait_check_wakeup(void *arg)
622 {
623 volatile uint32_t *mcpu_mwait = (volatile uint32_t *)arg;
624
625 ASSERT(arg != NULL);
626 if (*mcpu_mwait != MWAIT_HALTED) {
627 /*
628 * CPU has been awakened, notify CPU idle notification system.
629 */
630 cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
631 } else {
632 /*
633 * Toggle interrupt flag to detect pending interrupts.
634 * If interrupt happened, do_interrupt() will notify CPU idle
635 * notification framework so no need to call cpu_idle_exit()
636 * here.
637 */
638 sti();
639 SMT_PAUSE();
640 cli();
641 }
642 }
643
644 /*
645 * Idle the present CPU until awakened via touching its monitored line
646 */
647 void
cpu_idle_mwait(void)648 cpu_idle_mwait(void)
649 {
650 volatile uint32_t *mcpu_mwait = CPU->cpu_m.mcpu_mwait;
651 cpu_t *cpup = CPU;
652 processorid_t cpu_sid = cpup->cpu_seqid;
653 cpupart_t *cp = cpup->cpu_part;
654 int hset_update = 1;
655
656 /*
657 * Set our mcpu_mwait here, so we can tell if anyone tries to
658 * wake us between now and when we call mwait. No other cpu will
659 * attempt to set our mcpu_mwait until we add ourself to the halted
660 * CPU bitmap.
661 */
662 *mcpu_mwait = MWAIT_HALTED;
663
664 /*
665 * If this CPU is online, and there's multiple CPUs
666 * in the system, then we should note our halting
667 * by adding ourselves to the partition's halted CPU
668 * bitmap. This allows other CPUs to find/awaken us when
669 * work becomes available.
670 */
671 if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
672 hset_update = 0;
673
674 /*
675 * Add ourselves to the partition's halted CPUs bitmap
676 * and set our HALTED flag, if necessary.
677 *
678 * When a thread becomes runnable, it is placed on the queue
679 * and then the halted CPU bitmap is checked to determine who
680 * (if anyone) should be awakened. We therefore need to first
681 * add ourselves to the bitmap, and and then check if there
682 * is any work available.
683 *
684 * Note that memory barriers after updating the HALTED flag
685 * are not necessary since an atomic operation (updating the bitmap)
686 * immediately follows. On x86 the atomic operation acts as a
687 * memory barrier for the update of cpu_disp_flags.
688 */
689 if (hset_update) {
690 cpup->cpu_disp_flags |= CPU_DISP_HALTED;
691 bitset_atomic_add(&cp->cp_haltset, cpu_sid);
692 }
693
694 /*
695 * Check to make sure there's really nothing to do.
696 * Work destined for this CPU may become available after
697 * this check. We'll be notified through the clearing of our
698 * bit in the halted CPU bitmap, and a write to our mcpu_mwait.
699 *
700 * disp_anywork() checks disp_nrunnable, so we do not have to later.
701 */
702 if (disp_anywork()) {
703 if (hset_update) {
704 cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
705 bitset_atomic_del(&cp->cp_haltset, cpu_sid);
706 }
707 return;
708 }
709
710 /*
711 * We're on our way to being halted.
712 * To avoid a lost wakeup, arm the monitor before checking if another
713 * cpu wrote to mcpu_mwait to wake us up.
714 */
715 i86_monitor(mcpu_mwait, 0, 0);
716 if (*mcpu_mwait == MWAIT_HALTED) {
717 if (cpu_idle_enter(IDLE_STATE_C1, 0,
718 cpu_idle_mwait_check_wakeup, (void *)mcpu_mwait) == 0) {
719 if (*mcpu_mwait == MWAIT_HALTED) {
720 i86_mwait(0, 0);
721 }
722 cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
723 }
724 }
725
726 /*
727 * We're no longer halted
728 */
729 if (hset_update) {
730 cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
731 bitset_atomic_del(&cp->cp_haltset, cpu_sid);
732 }
733 }
734
735 /*
736 * If "cpu" is halted in mwait, then wake it up clearing its halted bit in
737 * advance. Otherwise, see if other CPUs in the cpu partition are halted and
738 * need to be woken up so that they can steal the thread we placed on this CPU.
739 * This function is only used on MP systems.
740 */
741 static void
cpu_wakeup_mwait(cpu_t * cp,int bound)742 cpu_wakeup_mwait(cpu_t *cp, int bound)
743 {
744 cpupart_t *cpu_part;
745 uint_t cpu_found;
746 processorid_t cpu_sid;
747
748 cpu_part = cp->cpu_part;
749 cpu_sid = cp->cpu_seqid;
750
751 /*
752 * Clear the halted bit for that CPU since it will be woken up
753 * in a moment.
754 */
755 if (bitset_in_set(&cpu_part->cp_haltset, cpu_sid)) {
756 /*
757 * Clear the halted bit for that CPU since it will be
758 * poked in a moment.
759 */
760 bitset_atomic_del(&cpu_part->cp_haltset, cpu_sid);
761 /*
762 * We may find the current CPU present in the halted cpuset
763 * if we're in the context of an interrupt that occurred
764 * before we had a chance to clear our bit in cpu_idle().
765 * Waking ourself is obviously unnecessary, since if
766 * we're here, we're not halted.
767 *
768 * monitor/mwait wakeup via writing to our cache line is
769 * harmless and less expensive than always checking if we
770 * are waking ourself which is an uncommon case.
771 */
772 MWAIT_WAKEUP(cp); /* write to monitored line */
773 return;
774 } else {
775 /*
776 * This cpu isn't halted, but it's idle or undergoing a
777 * context switch. No need to awaken anyone else.
778 */
779 if (cp->cpu_thread == cp->cpu_idle_thread ||
780 cp->cpu_disp_flags & CPU_DISP_DONTSTEAL)
781 return;
782 }
783
784 /*
785 * No need to wake up other CPUs if the thread we just enqueued
786 * is bound.
787 */
788 if (bound || ncpus == 1)
789 return;
790
791 /*
792 * See if there's any other halted CPUs. If there are, then
793 * select one, and awaken it.
794 * It's possible that after we find a CPU, somebody else
795 * will awaken it before we get the chance.
796 * In that case, look again.
797 */
798 do {
799 cpu_found = bitset_find(&cpu_part->cp_haltset);
800 if (cpu_found == (uint_t)-1)
801 return;
802 } while (bitset_atomic_test_and_del(&cpu_part->cp_haltset,
803 cpu_found) < 0);
804
805 /*
806 * Do not check if cpu_found is ourself as monitor/mwait
807 * wakeup is cheap.
808 */
809 MWAIT_WAKEUP(cpu_seq[cpu_found]); /* write to monitored line */
810 }
811
812 #endif
813
814 void (*cpu_pause_handler)(volatile char *) = NULL;
815
816 static int
mp_disable_intr(int cpun)817 mp_disable_intr(int cpun)
818 {
819 /*
820 * switch to the offline cpu
821 */
822 affinity_set(cpun);
823 /*
824 * raise ipl to just below cross call
825 */
826 splx(XC_SYS_PIL - 1);
827 /*
828 * set base spl to prevent the next swtch to idle from
829 * lowering back to ipl 0
830 */
831 CPU->cpu_intr_actv |= (1 << (XC_SYS_PIL - 1));
832 set_base_spl();
833 affinity_clear();
834 return (DDI_SUCCESS);
835 }
836
837 static void
mp_enable_intr(int cpun)838 mp_enable_intr(int cpun)
839 {
840 /*
841 * switch to the online cpu
842 */
843 affinity_set(cpun);
844 /*
845 * clear the interrupt active mask
846 */
847 CPU->cpu_intr_actv &= ~(1 << (XC_SYS_PIL - 1));
848 set_base_spl();
849 (void) spl0();
850 affinity_clear();
851 }
852
853 static void
mach_get_platform(int owner)854 mach_get_platform(int owner)
855 {
856 void **srv_opsp;
857 void **clt_opsp;
858 int i;
859 int total_ops;
860
861 /* fix up psm ops */
862 srv_opsp = (void **)mach_set[0];
863 clt_opsp = (void **)mach_set[owner];
864 if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
865 total_ops = sizeof (struct psm_ops_ver01) /
866 sizeof (void (*)(void));
867 else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
868 /* no psm_notify_func */
869 total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
870 sizeof (void (*)(void));
871 else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
872 /* no psm_timer funcs */
873 total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
874 sizeof (void (*)(void));
875 else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
876 /* no psm_preshutdown function */
877 total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
878 sizeof (void (*)(void));
879 else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
880 /* no psm_intr_ops function */
881 total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
882 sizeof (void (*)(void));
883 else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_5)
884 /* no psm_state function */
885 total_ops = OFFSETOF(struct psm_ops, psm_state) /
886 sizeof (void (*)(void));
887 else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_6)
888 /* no psm_cpu_ops function */
889 total_ops = OFFSETOF(struct psm_ops, psm_cpu_ops) /
890 sizeof (void (*)(void));
891 else
892 total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
893
894 /*
895 * Save the version of the PSM module, in case we need to
896 * behave differently based on version.
897 */
898 mach_ver[0] = mach_ver[owner];
899
900 for (i = 0; i < total_ops; i++)
901 if (clt_opsp[i] != NULL)
902 srv_opsp[i] = clt_opsp[i];
903 }
904
905 static void
mach_construct_info()906 mach_construct_info()
907 {
908 struct psm_sw *swp;
909 int mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
910 int conflict_owner = 0;
911
912 if (psmsw->psw_forw == psmsw)
913 panic("No valid PSM modules found");
914 mutex_enter(&psmsw_lock);
915 for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
916 if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
917 continue;
918 mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
919 mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
920 mach_cnt[swp->psw_infop->p_owner]++;
921 }
922 mutex_exit(&psmsw_lock);
923
924 mach_get_platform(PSM_OWN_SYS_DEFAULT);
925
926 /* check to see are there any conflicts */
927 if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
928 conflict_owner = PSM_OWN_EXCLUSIVE;
929 if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
930 conflict_owner = PSM_OWN_OVERRIDE;
931 if (conflict_owner) {
932 /* remove all psm modules except uppc */
933 cmn_err(CE_WARN,
934 "Conflicts detected on the following PSM modules:");
935 mutex_enter(&psmsw_lock);
936 for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
937 if (swp->psw_infop->p_owner == conflict_owner)
938 cmn_err(CE_WARN, "%s ",
939 swp->psw_infop->p_mach_idstring);
940 }
941 mutex_exit(&psmsw_lock);
942 cmn_err(CE_WARN,
943 "Setting the system back to SINGLE processor mode!");
944 cmn_err(CE_WARN,
945 "Please edit /etc/mach to remove the invalid PSM module.");
946 return;
947 }
948
949 if (mach_set[PSM_OWN_EXCLUSIVE])
950 mach_get_platform(PSM_OWN_EXCLUSIVE);
951
952 if (mach_set[PSM_OWN_OVERRIDE])
953 mach_get_platform(PSM_OWN_OVERRIDE);
954 }
955
956 static void
mach_init()957 mach_init()
958 {
959 struct psm_ops *pops;
960
961 mach_construct_info();
962
963 pops = mach_set[0];
964
965 /* register the interrupt and clock initialization rotuines */
966 picinitf = mach_picinit;
967 clkinitf = mach_clkinit;
968 psm_get_clockirq = pops->psm_get_clockirq;
969
970 /* register the interrupt setup code */
971 slvltovect = mach_softlvl_to_vect;
972 addspl = pops->psm_addspl;
973 delspl = pops->psm_delspl;
974
975 if (pops->psm_translate_irq)
976 psm_translate_irq = pops->psm_translate_irq;
977 if (pops->psm_intr_ops)
978 psm_intr_ops = pops->psm_intr_ops;
979
980 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
981 /*
982 * Time-of-day functionality now handled in TOD modules.
983 * (Warn about PSM modules that think that we're going to use
984 * their ops vectors.)
985 */
986 if (pops->psm_tod_get)
987 cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
988 (void *)pops->psm_tod_get);
989
990 if (pops->psm_tod_set)
991 cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
992 (void *)pops->psm_tod_set);
993 #endif
994
995 if (pops->psm_notify_error) {
996 psm_notify_error = mach_notify_error;
997 notify_error = pops->psm_notify_error;
998 }
999
1000 (*pops->psm_softinit)();
1001
1002 /*
1003 * Initialize the dispatcher's function hooks to enable CPU halting
1004 * when idle. Set both the deep-idle and non-deep-idle hooks.
1005 *
1006 * Assume we can use power saving deep-idle loop cpu_idle_adaptive.
1007 * Platform deep-idle driver will reset our idle loop to
1008 * non_deep_idle_cpu if power saving deep-idle feature is not available.
1009 *
1010 * Do not use monitor/mwait if idle_cpu_use_hlt is not set(spin idle)
1011 * or idle_cpu_prefer_mwait is not set.
1012 * Allocate monitor/mwait buffer for cpu0.
1013 */
1014 #ifndef __xpv
1015 non_deep_idle_disp_enq_thread = disp_enq_thread;
1016 #endif
1017 if (idle_cpu_use_hlt) {
1018 idle_cpu = cpu_idle_adaptive;
1019 CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1020 #ifndef __xpv
1021 if (is_x86_feature(x86_featureset, X86FSET_MWAIT) &&
1022 idle_cpu_prefer_mwait) {
1023 CPU->cpu_m.mcpu_mwait = cpuid_mwait_alloc(CPU);
1024 /*
1025 * Protect ourself from insane mwait size.
1026 */
1027 if (CPU->cpu_m.mcpu_mwait == NULL) {
1028 #ifdef DEBUG
1029 cmn_err(CE_NOTE, "Using hlt idle. Cannot "
1030 "handle cpu 0 mwait size.");
1031 #endif
1032 idle_cpu_prefer_mwait = 0;
1033 CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1034 } else {
1035 CPU->cpu_m.mcpu_idle_cpu = cpu_idle_mwait;
1036 }
1037 } else {
1038 CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1039 }
1040 non_deep_idle_cpu = CPU->cpu_m.mcpu_idle_cpu;
1041
1042 /*
1043 * Disable power saving deep idle loop?
1044 */
1045 if (idle_cpu_no_deep_c) {
1046 idle_cpu = non_deep_idle_cpu;
1047 }
1048 #endif
1049 }
1050
1051 mach_smpinit();
1052 }
1053
1054 static void
mach_smpinit(void)1055 mach_smpinit(void)
1056 {
1057 struct psm_ops *pops;
1058 processorid_t cpu_id;
1059 int cnt;
1060 cpuset_t cpumask;
1061
1062 pops = mach_set[0];
1063 CPUSET_ZERO(cpumask);
1064
1065 cpu_id = -1;
1066 cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1067 /*
1068 * Only add boot_ncpus CPUs to mp_cpus. Other CPUs will be handled
1069 * by CPU DR driver at runtime.
1070 */
1071 for (cnt = 0; cpu_id != -1 && cnt < boot_ncpus; cnt++) {
1072 CPUSET_ADD(cpumask, cpu_id);
1073 cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1074 }
1075
1076 mp_cpus = cpumask;
1077
1078 /* MP related routines */
1079 ap_mlsetup = pops->psm_post_cpu_start;
1080 send_dirintf = pops->psm_send_ipi;
1081
1082 /* optional MP related routines */
1083 if (pops->psm_shutdown)
1084 psm_shutdownf = pops->psm_shutdown;
1085 if (pops->psm_preshutdown)
1086 psm_preshutdownf = pops->psm_preshutdown;
1087 if (pops->psm_notify_func)
1088 psm_notifyf = pops->psm_notify_func;
1089 if (pops->psm_set_idlecpu)
1090 psm_set_idle_cpuf = pops->psm_set_idlecpu;
1091 if (pops->psm_unset_idlecpu)
1092 psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
1093
1094 psm_clkinit = pops->psm_clkinit;
1095
1096 if (pops->psm_timer_reprogram)
1097 psm_timer_reprogram = pops->psm_timer_reprogram;
1098
1099 if (pops->psm_timer_enable)
1100 psm_timer_enable = pops->psm_timer_enable;
1101
1102 if (pops->psm_timer_disable)
1103 psm_timer_disable = pops->psm_timer_disable;
1104
1105 if (pops->psm_post_cyclic_setup)
1106 psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
1107
1108 if (pops->psm_state)
1109 psm_state = pops->psm_state;
1110
1111 /*
1112 * Set these vectors here so they can be used by Suspend/Resume
1113 * on UP machines.
1114 */
1115 if (pops->psm_disable_intr)
1116 psm_disable_intr = pops->psm_disable_intr;
1117 if (pops->psm_enable_intr)
1118 psm_enable_intr = pops->psm_enable_intr;
1119
1120 /* check for multiple CPUs */
1121 if (cnt < 2 && plat_dr_support_cpu() == B_FALSE)
1122 return;
1123
1124 /* check for MP platforms */
1125 if (pops->psm_cpu_start == NULL)
1126 return;
1127
1128 /*
1129 * Set the dispatcher hook to enable cpu "wake up"
1130 * when a thread becomes runnable.
1131 */
1132 if (idle_cpu_use_hlt) {
1133 disp_enq_thread = cpu_wakeup;
1134 #ifndef __xpv
1135 if (is_x86_feature(x86_featureset, X86FSET_MWAIT) &&
1136 idle_cpu_prefer_mwait)
1137 disp_enq_thread = cpu_wakeup_mwait;
1138 non_deep_idle_disp_enq_thread = disp_enq_thread;
1139 #endif
1140 }
1141
1142 psm_get_ipivect = pops->psm_get_ipivect;
1143
1144 (void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_intr",
1145 (*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
1146 NULL, NULL, NULL, NULL);
1147
1148 (void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
1149 }
1150
1151 static void
mach_picinit()1152 mach_picinit()
1153 {
1154 struct psm_ops *pops;
1155
1156 pops = mach_set[0];
1157
1158 /* register the interrupt handlers */
1159 setlvl = pops->psm_intr_enter;
1160 setlvlx = pops->psm_intr_exit;
1161
1162 /* initialize the interrupt hardware */
1163 (*pops->psm_picinit)();
1164
1165 /* set interrupt mask for current ipl */
1166 setspl = pops->psm_setspl;
1167 cli();
1168 setspl(CPU->cpu_pri);
1169 }
1170
1171 uint_t cpu_freq; /* MHz */
1172 uint64_t cpu_freq_hz; /* measured (in hertz) */
1173
1174 #define MEGA_HZ 1000000
1175
1176 #ifdef __xpv
1177
1178 int xpv_cpufreq_workaround = 1;
1179 int xpv_cpufreq_verbose = 0;
1180
1181 #else /* __xpv */
1182
1183 static uint64_t
mach_calchz(uint32_t pit_counter,uint64_t * processor_clks)1184 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
1185 {
1186 uint64_t cpu_hz;
1187
1188 if ((pit_counter == 0) || (*processor_clks == 0) ||
1189 (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
1190 return (0);
1191
1192 cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
1193
1194 return (cpu_hz);
1195 }
1196
1197 #endif /* __xpv */
1198
1199 static uint64_t
mach_getcpufreq(void)1200 mach_getcpufreq(void)
1201 {
1202 #if defined(__xpv)
1203 vcpu_time_info_t *vti = &CPU->cpu_m.mcpu_vcpu_info->time;
1204 uint64_t cpu_hz;
1205
1206 /*
1207 * During dom0 bringup, it was noted that on at least one older
1208 * Intel HT machine, the hypervisor initially gives a tsc_to_system_mul
1209 * value that is quite wrong (the 3.06GHz clock was reported
1210 * as 4.77GHz)
1211 *
1212 * The curious thing is, that if you stop the kernel at entry,
1213 * breakpoint here and inspect the value with kmdb, the value
1214 * is correct - but if you don't stop and simply enable the
1215 * printf statement (below), you can see the bad value printed
1216 * here. Almost as if something kmdb did caused the hypervisor to
1217 * figure it out correctly. And, note that the hypervisor
1218 * eventually -does- figure it out correctly ... if you look at
1219 * the field later in the life of dom0, it is correct.
1220 *
1221 * For now, on dom0, we employ a slightly cheesy workaround of
1222 * using the DOM0_PHYSINFO hypercall.
1223 */
1224 if (DOMAIN_IS_INITDOMAIN(xen_info) && xpv_cpufreq_workaround) {
1225 cpu_hz = 1000 * xpv_cpu_khz();
1226 } else {
1227 cpu_hz = (UINT64_C(1000000000) << 32) / vti->tsc_to_system_mul;
1228
1229 if (vti->tsc_shift < 0)
1230 cpu_hz <<= -vti->tsc_shift;
1231 else
1232 cpu_hz >>= vti->tsc_shift;
1233 }
1234
1235 if (xpv_cpufreq_verbose)
1236 printf("mach_getcpufreq: system_mul 0x%x, shift %d, "
1237 "cpu_hz %" PRId64 "Hz\n",
1238 vti->tsc_to_system_mul, vti->tsc_shift, cpu_hz);
1239
1240 return (cpu_hz);
1241 #else /* __xpv */
1242 uint32_t pit_counter;
1243 uint64_t processor_clks;
1244
1245 if (is_x86_feature(x86_featureset, X86FSET_TSC)) {
1246 /*
1247 * We have a TSC. freq_tsc() knows how to measure the number
1248 * of clock cycles sampled against the PIT.
1249 */
1250 ulong_t flags = clear_int_flag();
1251 processor_clks = freq_tsc(&pit_counter);
1252 restore_int_flag(flags);
1253 return (mach_calchz(pit_counter, &processor_clks));
1254 } else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
1255 #if defined(__amd64)
1256 panic("mach_getcpufreq: no TSC!");
1257 #elif defined(__i386)
1258 /*
1259 * We are a Cyrix based on a 6x86 core or an Intel Pentium
1260 * for which freq_notsc() knows how to measure the number of
1261 * elapsed clock cycles sampled against the PIT
1262 */
1263 ulong_t flags = clear_int_flag();
1264 processor_clks = freq_notsc(&pit_counter);
1265 restore_int_flag(flags);
1266 return (mach_calchz(pit_counter, &processor_clks));
1267 #endif /* __i386 */
1268 }
1269
1270 /* We do not know how to calculate cpu frequency for this cpu. */
1271 return (0);
1272 #endif /* __xpv */
1273 }
1274
1275 /*
1276 * If the clock speed of a cpu is found to be reported incorrectly, do not add
1277 * to this array, instead improve the accuracy of the algorithm that determines
1278 * the clock speed of the processor or extend the implementation to support the
1279 * vendor as appropriate. This is here only to support adjusting the speed on
1280 * older slower processors that mach_fixcpufreq() would not be able to account
1281 * for otherwise.
1282 */
1283 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
1284
1285 /*
1286 * On fast processors the clock frequency that is measured may be off by
1287 * a few MHz from the value printed on the part. This is a combination of
1288 * the factors that for such fast parts being off by this much is within
1289 * the tolerances for manufacture and because of the difficulties in the
1290 * measurement that can lead to small error. This function uses some
1291 * heuristics in order to tweak the value that was measured to match what
1292 * is most likely printed on the part.
1293 *
1294 * Some examples:
1295 * AMD Athlon 1000 mhz measured as 998 mhz
1296 * Intel Pentium III Xeon 733 mhz measured as 731 mhz
1297 * Intel Pentium IV 1500 mhz measured as 1495mhz
1298 *
1299 * If in the future this function is no longer sufficient to correct
1300 * for the error in the measurement, then the algorithm used to perform
1301 * the measurement will have to be improved in order to increase accuracy
1302 * rather than adding horrible and questionable kludges here.
1303 *
1304 * This is called after the cyclics subsystem because of the potential
1305 * that the heuristics within may give a worse estimate of the clock
1306 * frequency than the value that was measured.
1307 */
1308 static void
mach_fixcpufreq(void)1309 mach_fixcpufreq(void)
1310 {
1311 uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
1312
1313 freq = (uint32_t)cpu_freq;
1314
1315 /*
1316 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
1317 * measured speed taking into account that the 667 MHz parts were
1318 * the first to round-up.
1319 */
1320 mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
1321 near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
1322 delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
1323
1324 /* Find the nearest integer multiple of 50 MHz to the measured speed */
1325 mul = (freq + 25) / 50;
1326 near50 = mul * 50;
1327 delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
1328
1329 /* Find the closer of the two */
1330 if (delta66 < delta50) {
1331 fixed = near66;
1332 delta = delta66;
1333 } else {
1334 fixed = near50;
1335 delta = delta50;
1336 }
1337
1338 if (fixed > INT_MAX)
1339 return;
1340
1341 /*
1342 * Some older parts have a core clock frequency that is not an
1343 * integral multiple of 50 or 66 MHz. Check if one of the old
1344 * clock frequencies is closer to the measured value than any
1345 * of the integral multiples of 50 an 66, and if so set fixed
1346 * and delta appropriately to represent the closest value.
1347 */
1348 i = sizeof (x86_cpu_freq) / sizeof (int);
1349 while (i > 0) {
1350 i--;
1351
1352 if (x86_cpu_freq[i] <= freq) {
1353 mul = freq - x86_cpu_freq[i];
1354
1355 if (mul < delta) {
1356 fixed = x86_cpu_freq[i];
1357 delta = mul;
1358 }
1359
1360 break;
1361 }
1362
1363 mul = x86_cpu_freq[i] - freq;
1364
1365 if (mul < delta) {
1366 fixed = x86_cpu_freq[i];
1367 delta = mul;
1368 }
1369 }
1370
1371 /*
1372 * Set a reasonable maximum for how much to correct the measured
1373 * result by. This check is here to prevent the adjustment made
1374 * by this function from being more harm than good. It is entirely
1375 * possible that in the future parts will be made that are not
1376 * integral multiples of 66 or 50 in clock frequency or that
1377 * someone may overclock a part to some odd frequency. If the
1378 * measured value is farther from the corrected value than
1379 * allowed, then assume the corrected value is in error and use
1380 * the measured value.
1381 */
1382 if (6 < delta)
1383 return;
1384
1385 cpu_freq = (int)fixed;
1386 }
1387
1388
1389 static int
machhztomhz(uint64_t cpu_freq_hz)1390 machhztomhz(uint64_t cpu_freq_hz)
1391 {
1392 uint64_t cpu_mhz;
1393
1394 /* Round to nearest MHZ */
1395 cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
1396
1397 if (cpu_mhz > INT_MAX)
1398 return (0);
1399
1400 return ((int)cpu_mhz);
1401
1402 }
1403
1404
1405 static int
mach_clkinit(int preferred_mode,int * set_mode)1406 mach_clkinit(int preferred_mode, int *set_mode)
1407 {
1408 struct psm_ops *pops;
1409 int resolution;
1410
1411 pops = mach_set[0];
1412
1413 cpu_freq_hz = mach_getcpufreq();
1414
1415 cpu_freq = machhztomhz(cpu_freq_hz);
1416
1417 if (!is_x86_feature(x86_featureset, X86FSET_TSC) || (cpu_freq == 0))
1418 tsc_gethrtime_enable = 0;
1419
1420 #ifndef __xpv
1421 if (tsc_gethrtime_enable) {
1422 tsc_hrtimeinit(cpu_freq_hz);
1423 } else
1424 #endif
1425 {
1426 if (pops->psm_hrtimeinit)
1427 (*pops->psm_hrtimeinit)();
1428 gethrtimef = pops->psm_gethrtime;
1429 gethrtimeunscaledf = gethrtimef;
1430 /* scalehrtimef will remain dummy */
1431 }
1432
1433 mach_fixcpufreq();
1434
1435 if (mach_ver[0] >= PSM_INFO_VER01_3) {
1436 if (preferred_mode == TIMER_ONESHOT) {
1437
1438 resolution = (*pops->psm_clkinit)(0);
1439 if (resolution != 0) {
1440 *set_mode = TIMER_ONESHOT;
1441 return (resolution);
1442 }
1443 }
1444
1445 /*
1446 * either periodic mode was requested or could not set to
1447 * one-shot mode
1448 */
1449 resolution = (*pops->psm_clkinit)(hz);
1450 /*
1451 * psm should be able to do periodic, so we do not check
1452 * for return value of psm_clkinit here.
1453 */
1454 *set_mode = TIMER_PERIODIC;
1455 return (resolution);
1456 } else {
1457 /*
1458 * PSMI interface prior to PSMI_3 does not define a return
1459 * value for psm_clkinit, so the return value is ignored.
1460 */
1461 (void) (*pops->psm_clkinit)(hz);
1462 *set_mode = TIMER_PERIODIC;
1463 return (nsec_per_tick);
1464 }
1465 }
1466
1467
1468 /*ARGSUSED*/
1469 static int
mach_softlvl_to_vect(int ipl)1470 mach_softlvl_to_vect(int ipl)
1471 {
1472 setsoftint = av_set_softint_pending;
1473 kdisetsoftint = kdi_av_set_softint_pending;
1474
1475 return (PSM_SV_SOFTWARE);
1476 }
1477
1478 #ifdef DEBUG
1479 /*
1480 * This is here to allow us to simulate cpus that refuse to start.
1481 */
1482 cpuset_t cpufailset;
1483 #endif
1484
1485 int
mach_cpu_start(struct cpu * cp,void * ctx)1486 mach_cpu_start(struct cpu *cp, void *ctx)
1487 {
1488 struct psm_ops *pops = mach_set[0];
1489 processorid_t id = cp->cpu_id;
1490
1491 #ifdef DEBUG
1492 if (CPU_IN_SET(cpufailset, id))
1493 return (0);
1494 #endif
1495 return ((*pops->psm_cpu_start)(id, ctx));
1496 }
1497
1498 int
mach_cpuid_start(processorid_t id,void * ctx)1499 mach_cpuid_start(processorid_t id, void *ctx)
1500 {
1501 struct psm_ops *pops = mach_set[0];
1502
1503 #ifdef DEBUG
1504 if (CPU_IN_SET(cpufailset, id))
1505 return (0);
1506 #endif
1507 return ((*pops->psm_cpu_start)(id, ctx));
1508 }
1509
1510 int
mach_cpu_stop(cpu_t * cp,void * ctx)1511 mach_cpu_stop(cpu_t *cp, void *ctx)
1512 {
1513 struct psm_ops *pops = mach_set[0];
1514 psm_cpu_request_t request;
1515
1516 if (pops->psm_cpu_ops == NULL) {
1517 return (ENOTSUP);
1518 }
1519
1520 ASSERT(cp->cpu_id != -1);
1521 request.pcr_cmd = PSM_CPU_STOP;
1522 request.req.cpu_stop.cpuid = cp->cpu_id;
1523 request.req.cpu_stop.ctx = ctx;
1524
1525 return ((*pops->psm_cpu_ops)(&request));
1526 }
1527
1528 int
mach_cpu_add(mach_cpu_add_arg_t * argp,processorid_t * cpuidp)1529 mach_cpu_add(mach_cpu_add_arg_t *argp, processorid_t *cpuidp)
1530 {
1531 int rc;
1532 struct psm_ops *pops = mach_set[0];
1533 psm_cpu_request_t request;
1534
1535 if (pops->psm_cpu_ops == NULL) {
1536 return (ENOTSUP);
1537 }
1538
1539 request.pcr_cmd = PSM_CPU_ADD;
1540 request.req.cpu_add.argp = argp;
1541 request.req.cpu_add.cpuid = -1;
1542 rc = (*pops->psm_cpu_ops)(&request);
1543 if (rc == 0) {
1544 ASSERT(request.req.cpu_add.cpuid != -1);
1545 *cpuidp = request.req.cpu_add.cpuid;
1546 }
1547
1548 return (rc);
1549 }
1550
1551 int
mach_cpu_remove(processorid_t cpuid)1552 mach_cpu_remove(processorid_t cpuid)
1553 {
1554 struct psm_ops *pops = mach_set[0];
1555 psm_cpu_request_t request;
1556
1557 if (pops->psm_cpu_ops == NULL) {
1558 return (ENOTSUP);
1559 }
1560
1561 request.pcr_cmd = PSM_CPU_REMOVE;
1562 request.req.cpu_remove.cpuid = cpuid;
1563
1564 return ((*pops->psm_cpu_ops)(&request));
1565 }
1566
1567 /*
1568 * Default handler to create device node for CPU.
1569 * One reference count will be held on created device node.
1570 */
1571 static int
mach_cpu_create_devinfo(cpu_t * cp,dev_info_t ** dipp)1572 mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp)
1573 {
1574 int rv, circ;
1575 dev_info_t *dip;
1576 static kmutex_t cpu_node_lock;
1577 static dev_info_t *cpu_nex_devi = NULL;
1578
1579 ASSERT(cp != NULL);
1580 ASSERT(dipp != NULL);
1581 *dipp = NULL;
1582
1583 if (cpu_nex_devi == NULL) {
1584 mutex_enter(&cpu_node_lock);
1585 /* First check whether cpus exists. */
1586 cpu_nex_devi = ddi_find_devinfo("cpus", -1, 0);
1587 /* Create cpus if it doesn't exist. */
1588 if (cpu_nex_devi == NULL) {
1589 ndi_devi_enter(ddi_root_node(), &circ);
1590 rv = ndi_devi_alloc(ddi_root_node(), "cpus",
1591 (pnode_t)DEVI_SID_NODEID, &dip);
1592 if (rv != NDI_SUCCESS) {
1593 mutex_exit(&cpu_node_lock);
1594 cmn_err(CE_CONT,
1595 "?failed to create cpu nexus device.\n");
1596 return (PSM_FAILURE);
1597 }
1598 ASSERT(dip != NULL);
1599 (void) ndi_devi_online(dip, 0);
1600 ndi_devi_exit(ddi_root_node(), circ);
1601 cpu_nex_devi = dip;
1602 }
1603 mutex_exit(&cpu_node_lock);
1604 }
1605
1606 /*
1607 * create a child node for cpu identified as 'cpu_id'
1608 */
1609 ndi_devi_enter(cpu_nex_devi, &circ);
1610 dip = ddi_add_child(cpu_nex_devi, "cpu", DEVI_SID_NODEID, -1);
1611 if (dip == NULL) {
1612 cmn_err(CE_CONT,
1613 "?failed to create device node for cpu%d.\n", cp->cpu_id);
1614 rv = PSM_FAILURE;
1615 } else {
1616 *dipp = dip;
1617 (void) ndi_hold_devi(dip);
1618 rv = PSM_SUCCESS;
1619 }
1620 ndi_devi_exit(cpu_nex_devi, circ);
1621
1622 return (rv);
1623 }
1624
1625 /*
1626 * Create cpu device node in device tree and online it.
1627 * Return created dip with reference count held if requested.
1628 */
1629 int
mach_cpu_create_device_node(struct cpu * cp,dev_info_t ** dipp)1630 mach_cpu_create_device_node(struct cpu *cp, dev_info_t **dipp)
1631 {
1632 int rv;
1633 dev_info_t *dip = NULL;
1634
1635 ASSERT(psm_cpu_create_devinfo != NULL);
1636 rv = psm_cpu_create_devinfo(cp, &dip);
1637 if (rv == PSM_SUCCESS) {
1638 cpuid_set_cpu_properties(dip, cp->cpu_id, cp->cpu_m.mcpu_cpi);
1639 /* Recursively attach driver for parent nexus device. */
1640 if (i_ddi_attach_node_hierarchy(ddi_get_parent(dip)) ==
1641 DDI_SUCCESS) {
1642 /* Configure cpu itself and descendants. */
1643 (void) ndi_devi_online(dip,
1644 NDI_ONLINE_ATTACH | NDI_CONFIG);
1645 }
1646 if (dipp != NULL) {
1647 *dipp = dip;
1648 } else {
1649 (void) ndi_rele_devi(dip);
1650 }
1651 }
1652
1653 return (rv);
1654 }
1655
1656 /*
1657 * The dipp contains one of following values on return:
1658 * - NULL if no device node found
1659 * - pointer to device node if found
1660 */
1661 int
mach_cpu_get_device_node(struct cpu * cp,dev_info_t ** dipp)1662 mach_cpu_get_device_node(struct cpu *cp, dev_info_t **dipp)
1663 {
1664 *dipp = NULL;
1665 if (psm_cpu_get_devinfo != NULL) {
1666 if (psm_cpu_get_devinfo(cp, dipp) == PSM_SUCCESS) {
1667 return (PSM_SUCCESS);
1668 }
1669 }
1670
1671 return (PSM_FAILURE);
1672 }
1673
1674 /*ARGSUSED*/
1675 static int
mach_translate_irq(dev_info_t * dip,int irqno)1676 mach_translate_irq(dev_info_t *dip, int irqno)
1677 {
1678 return (irqno); /* default to NO translation */
1679 }
1680
1681 static void
mach_notify_error(int level,char * errmsg)1682 mach_notify_error(int level, char *errmsg)
1683 {
1684 /*
1685 * SL_FATAL is pass in once panicstr is set, deliver it
1686 * as CE_PANIC. Also, translate SL_ codes back to CE_
1687 * codes for the psmi handler
1688 */
1689 if (level & SL_FATAL)
1690 (*notify_error)(CE_PANIC, errmsg);
1691 else if (level & SL_WARN)
1692 (*notify_error)(CE_WARN, errmsg);
1693 else if (level & SL_NOTE)
1694 (*notify_error)(CE_NOTE, errmsg);
1695 else if (level & SL_CONSOLE)
1696 (*notify_error)(CE_CONT, errmsg);
1697 }
1698
1699 /*
1700 * It provides the default basic intr_ops interface for the new DDI
1701 * interrupt framework if the PSM doesn't have one.
1702 *
1703 * Input:
1704 * dip - pointer to the dev_info structure of the requested device
1705 * hdlp - pointer to the internal interrupt handle structure for the
1706 * requested interrupt
1707 * intr_op - opcode for this call
1708 * result - pointer to the integer that will hold the result to be
1709 * passed back if return value is PSM_SUCCESS
1710 *
1711 * Output:
1712 * return value is either PSM_SUCCESS or PSM_FAILURE
1713 */
1714 static int
mach_intr_ops(dev_info_t * dip,ddi_intr_handle_impl_t * hdlp,psm_intr_op_t intr_op,int * result)1715 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1716 psm_intr_op_t intr_op, int *result)
1717 {
1718 struct intrspec *ispec;
1719
1720 switch (intr_op) {
1721 case PSM_INTR_OP_CHECK_MSI:
1722 *result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1723 DDI_INTR_TYPE_MSIX);
1724 break;
1725 case PSM_INTR_OP_ALLOC_VECTORS:
1726 if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1727 *result = 1;
1728 else
1729 *result = 0;
1730 break;
1731 case PSM_INTR_OP_FREE_VECTORS:
1732 break;
1733 case PSM_INTR_OP_NAVAIL_VECTORS:
1734 if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1735 *result = 1;
1736 else
1737 *result = 0;
1738 break;
1739 case PSM_INTR_OP_XLATE_VECTOR:
1740 ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1741 *result = psm_translate_irq(dip, ispec->intrspec_vec);
1742 break;
1743 case PSM_INTR_OP_GET_CAP:
1744 *result = 0;
1745 break;
1746 case PSM_INTR_OP_GET_PENDING:
1747 case PSM_INTR_OP_CLEAR_MASK:
1748 case PSM_INTR_OP_SET_MASK:
1749 case PSM_INTR_OP_GET_SHARED:
1750 case PSM_INTR_OP_SET_PRI:
1751 case PSM_INTR_OP_SET_CAP:
1752 case PSM_INTR_OP_SET_CPU:
1753 case PSM_INTR_OP_GET_INTR:
1754 default:
1755 return (PSM_FAILURE);
1756 }
1757 return (PSM_SUCCESS);
1758 }
1759 /*
1760 * Return 1 if CMT load balancing policies should be
1761 * implemented across instances of the specified hardware
1762 * sharing relationship.
1763 */
1764 int
pg_cmt_load_bal_hw(pghw_type_t hw)1765 pg_cmt_load_bal_hw(pghw_type_t hw)
1766 {
1767 if (hw == PGHW_IPIPE ||
1768 hw == PGHW_FPU ||
1769 hw == PGHW_PROCNODE ||
1770 hw == PGHW_CHIP)
1771 return (1);
1772 else
1773 return (0);
1774 }
1775 /*
1776 * Return 1 if thread affinity polices should be implemented
1777 * for instances of the specifed hardware sharing relationship.
1778 */
1779 int
pg_cmt_affinity_hw(pghw_type_t hw)1780 pg_cmt_affinity_hw(pghw_type_t hw)
1781 {
1782 if (hw == PGHW_CACHE)
1783 return (1);
1784 else
1785 return (0);
1786 }
1787
1788 /*
1789 * Return number of counter events requested to measure hardware capacity and
1790 * utilization and setup CPC requests for specified CPU as needed
1791 *
1792 * May return 0 when platform or processor specific code knows that no CPC
1793 * events should be programmed on this CPU or -1 when platform or processor
1794 * specific code doesn't know which counter events are best to use and common
1795 * code should decide for itself
1796 */
1797 int
1798 /* LINTED E_FUNC_ARG_UNUSED */
cu_plat_cpc_init(cpu_t * cp,kcpc_request_list_t * reqs,int nreqs)1799 cu_plat_cpc_init(cpu_t *cp, kcpc_request_list_t *reqs, int nreqs)
1800 {
1801 const char *impl_name;
1802
1803 /*
1804 * Return error if pcbe_ops not set
1805 */
1806 if (pcbe_ops == NULL)
1807 return (-1);
1808
1809 /*
1810 * Return that no CPC events should be programmed on hyperthreaded
1811 * Pentium 4 and return error for all other x86 processors to tell
1812 * common code to decide what counter events to program on those CPUs
1813 * for measuring hardware capacity and utilization
1814 */
1815 impl_name = pcbe_ops->pcbe_impl_name();
1816 if (impl_name != NULL && strcmp(impl_name, PCBE_IMPL_NAME_P4HT) == 0)
1817 return (0);
1818 else
1819 return (-1);
1820 }
1821