xref: /illumos-gate/usr/src/uts/common/sys/cpuvar.h (revision b52a336e)
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 (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 #ifndef _SYS_CPUVAR_H
27 #define	_SYS_CPUVAR_H
28 
29 #include <sys/thread.h>
30 #include <sys/sysinfo.h>	/* has cpu_stat_t definition */
31 #include <sys/disp.h>
32 #include <sys/processor.h>
33 
34 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
35 #include <sys/machcpuvar.h>
36 #endif
37 
38 #include <sys/types.h>
39 #include <sys/file.h>
40 #include <sys/bitmap.h>
41 #include <sys/rwlock.h>
42 #include <sys/msacct.h>
43 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \
44 	(defined(__i386) || defined(__amd64))
45 #include <asm/cpuvar.h>
46 #endif
47 
48 #ifdef	__cplusplus
49 extern "C" {
50 #endif
51 
52 struct squeue_set_s;
53 
54 #define	CPU_CACHE_COHERENCE_SIZE	64
55 #define	S_LOADAVG_SZ	11
56 #define	S_MOVAVG_SZ	10
57 
58 struct loadavg_s {
59 	int lg_cur;		/* current loadavg entry */
60 	unsigned int lg_len;	/* number entries recorded */
61 	hrtime_t lg_total;	/* used to temporarily hold load totals */
62 	hrtime_t lg_loads[S_LOADAVG_SZ];	/* table of recorded entries */
63 };
64 
65 /*
66  * For fast event tracing.
67  */
68 struct ftrace_record;
69 typedef struct ftrace_data {
70 	int			ftd_state;	/* ftrace flags */
71 	kmutex_t		ftd_unused;	/* ftrace buffer lock, unused */
72 	struct ftrace_record	*ftd_cur;	/* current record */
73 	struct ftrace_record	*ftd_first;	/* first record */
74 	struct ftrace_record	*ftd_last;	/* last record */
75 } ftrace_data_t;
76 
77 struct cyc_cpu;
78 struct nvlist;
79 
80 /*
81  * Per-CPU data.
82  *
83  * Be careful adding new members: if they are not the same in all modules (e.g.
84  * change size depending on a #define), CTF uniquification can fail to work
85  * properly.  Furthermore, this is transitive in that it applies recursively to
86  * all types pointed to by cpu_t.
87  */
88 typedef struct cpu {
89 	processorid_t	cpu_id;			/* CPU number */
90 	processorid_t	cpu_seqid;	/* sequential CPU id (0..ncpus-1) */
91 	volatile cpu_flag_t cpu_flags;		/* flags indicating CPU state */
92 	struct cpu	*cpu_self;		/* pointer to itself */
93 	kthread_t	*cpu_thread;		/* current thread */
94 	kthread_t	*cpu_idle_thread;	/* idle thread for this CPU */
95 	kthread_t	*cpu_pause_thread;	/* pause thread for this CPU */
96 	klwp_id_t	cpu_lwp;		/* current lwp (if any) */
97 	klwp_id_t	cpu_fpowner;		/* currently loaded fpu owner */
98 	struct cpupart	*cpu_part;		/* partition with this CPU */
99 	struct lgrp_ld	*cpu_lpl;		/* pointer to this cpu's load */
100 	int		cpu_cache_offset;	/* see kmem.c for details */
101 
102 	/*
103 	 * Links to other CPUs.  It is safe to walk these lists if
104 	 * one of the following is true:
105 	 * 	- cpu_lock held
106 	 * 	- preemption disabled via kpreempt_disable
107 	 * 	- PIL >= DISP_LEVEL
108 	 * 	- acting thread is an interrupt thread
109 	 * 	- all other CPUs are paused
110 	 */
111 	struct cpu	*cpu_next;		/* next existing CPU */
112 	struct cpu	*cpu_prev;		/* prev existing CPU */
113 	struct cpu	*cpu_next_onln;		/* next online (enabled) CPU */
114 	struct cpu	*cpu_prev_onln;		/* prev online (enabled) CPU */
115 	struct cpu	*cpu_next_part;		/* next CPU in partition */
116 	struct cpu	*cpu_prev_part;		/* prev CPU in partition */
117 	struct cpu	*cpu_next_lgrp;		/* next CPU in latency group */
118 	struct cpu	*cpu_prev_lgrp;		/* prev CPU in latency group */
119 	struct cpu	*cpu_next_lpl;		/* next CPU in lgrp partition */
120 	struct cpu	*cpu_prev_lpl;
121 
122 	struct cpu_pg	*cpu_pg;		/* cpu's processor groups */
123 
124 	void		*cpu_reserved[4];	/* reserved for future use */
125 
126 	/*
127 	 * Scheduling variables.
128 	 */
129 	disp_t		*cpu_disp;		/* dispatch queue data */
130 	/*
131 	 * Note that cpu_disp is set before the CPU is added to the system
132 	 * and is never modified.  Hence, no additional locking is needed
133 	 * beyond what's necessary to access the cpu_t structure.
134 	 */
135 	char		cpu_runrun;	/* scheduling flag - set to preempt */
136 	char		cpu_kprunrun;		/* force kernel preemption */
137 	pri_t		cpu_chosen_level; 	/* priority at which cpu */
138 						/* was chosen for scheduling */
139 	kthread_t	*cpu_dispthread; /* thread selected for dispatch */
140 	disp_lock_t	cpu_thread_lock; /* dispatcher lock on current thread */
141 	uint8_t		cpu_disp_flags;	/* flags used by dispatcher */
142 	/*
143 	 * The following field is updated when ever the cpu_dispthread
144 	 * changes. Also in places, where the current thread(cpu_dispthread)
145 	 * priority changes. This is used in disp_lowpri_cpu()
146 	 */
147 	pri_t		cpu_dispatch_pri; /* priority of cpu_dispthread */
148 	clock_t		cpu_last_swtch;	/* last time switched to new thread */
149 
150 	/*
151 	 * Interrupt data.
152 	 */
153 	caddr_t		cpu_intr_stack;	/* interrupt stack */
154 	kthread_t	*cpu_intr_thread; /* interrupt thread list */
155 	uint_t		cpu_intr_actv;	/* interrupt levels active (bitmask) */
156 	int		cpu_base_spl;	/* priority for highest rupt active */
157 
158 	/*
159 	 * Statistics.
160 	 */
161 	cpu_stats_t	cpu_stats;		/* per-CPU statistics */
162 	struct kstat	*cpu_info_kstat;	/* kstat for cpu info */
163 
164 	uintptr_t	cpu_profile_pc;	/* kernel PC in profile interrupt */
165 	uintptr_t	cpu_profile_upc; /* user PC in profile interrupt */
166 	uintptr_t	cpu_profile_pil; /* PIL when profile interrupted */
167 
168 	ftrace_data_t	cpu_ftrace;		/* per cpu ftrace data */
169 
170 	clock_t		cpu_deadman_counter;	/* used by deadman() */
171 	uint_t		cpu_deadman_countdown;	/* used by deadman() */
172 
173 	kmutex_t	cpu_cpc_ctxlock; /* protects context for idle thread */
174 	kcpc_ctx_t	*cpu_cpc_ctx;	/* performance counter context */
175 
176 	/*
177 	 * Configuration information for the processor_info system call.
178 	 */
179 	processor_info_t cpu_type_info;	/* config info */
180 	time_t		cpu_state_begin; /* when CPU entered current state */
181 	char		cpu_cpr_flags;	/* CPR related info */
182 	struct cyc_cpu	*cpu_cyclic;	/* per cpu cyclic subsystem data */
183 	struct squeue_set_s *cpu_squeue_set;	/* per cpu squeue set */
184 	struct nvlist	*cpu_props;	/* pool-related properties */
185 
186 	krwlock_t	cpu_ft_lock;		/* DTrace: fasttrap lock */
187 	uintptr_t	cpu_dtrace_caller;	/* DTrace: caller, if any */
188 	hrtime_t	cpu_dtrace_chillmark;	/* DTrace: chill mark time */
189 	hrtime_t	cpu_dtrace_chilled;	/* DTrace: total chill time */
190 	volatile uint16_t cpu_mstate;		/* cpu microstate */
191 	volatile uint16_t cpu_mstate_gen;	/* generation counter */
192 	volatile hrtime_t cpu_mstate_start;	/* cpu microstate start time */
193 	volatile hrtime_t cpu_acct[NCMSTATES];	/* cpu microstate data */
194 	hrtime_t	cpu_intracct[NCMSTATES]; /* interrupt mstate data */
195 	hrtime_t	cpu_waitrq;		/* cpu run-queue wait time */
196 	struct loadavg_s cpu_loadavg;		/* loadavg info for this cpu */
197 
198 	char		*cpu_idstr;	/* for printing and debugging */
199 	char		*cpu_brandstr;	/* for printing */
200 
201 	/*
202 	 * Sum of all device interrupt weights that are currently directed at
203 	 * this cpu. Cleared at start of interrupt redistribution.
204 	 */
205 	int32_t		cpu_intr_weight;
206 	void		*cpu_vm_data;
207 
208 	struct cpu_physid *cpu_physid;	/* physical associations */
209 
210 	uint64_t	cpu_curr_clock;		/* current clock freq in Hz */
211 	char		*cpu_supp_freqs;	/* supported freqs in Hz */
212 
213 	uintptr_t	cpu_cpcprofile_pc;	/* kernel PC in cpc interrupt */
214 	uintptr_t	cpu_cpcprofile_upc;	/* user PC in cpc interrupt */
215 
216 	/*
217 	 * Interrupt load factor used by dispatcher & softcall
218 	 */
219 	hrtime_t	cpu_intrlast;   /* total interrupt time (nsec) */
220 	int		cpu_intrload;   /* interrupt load factor (0-99%) */
221 
222 	uint_t		cpu_rotor;	/* for cheap pseudo-random numbers */
223 
224 	struct cu_cpu_info	*cpu_cu_info;	/* capacity & util. info */
225 
226 	/*
227 	 * cpu_generation is updated whenever CPU goes on-line or off-line.
228 	 * Updates to cpu_generation are protected by cpu_lock.
229 	 *
230 	 * See CPU_NEW_GENERATION() macro below.
231 	 */
232 	volatile uint_t		cpu_generation;	/* tracking on/off-line */
233 
234 	/*
235 	 * New members must be added /before/ this member, as the CTF tools
236 	 * rely on this being the last field before cpu_m, so they can
237 	 * correctly calculate the offset when synthetically adding the cpu_m
238 	 * member in objects that do not have it.  This fixup is required for
239 	 * uniquification to work correctly.
240 	 */
241 	uintptr_t	cpu_m_pad;
242 
243 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
244 	struct machcpu	cpu_m;		/* per architecture info */
245 #endif
246 } cpu_t;
247 
248 /*
249  * The cpu_core structure consists of per-CPU state available in any context.
250  * On some architectures, this may mean that the page(s) containing the
251  * NCPU-sized array of cpu_core structures must be locked in the TLB -- it
252  * is up to the platform to assure that this is performed properly.  Note that
253  * the structure is sized to avoid false sharing.
254  */
255 #define	CPUC_SIZE		(sizeof (uint16_t) + sizeof (uint8_t) + \
256 				sizeof (uintptr_t) + sizeof (kmutex_t))
257 #define	CPUC_PADSIZE		CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE
258 
259 typedef struct cpu_core {
260 	uint16_t	cpuc_dtrace_flags;	/* DTrace flags */
261 	uint8_t		cpuc_dcpc_intr_state;	/* DCPC provider intr state */
262 	uint8_t		cpuc_pad[CPUC_PADSIZE];	/* padding */
263 	uintptr_t	cpuc_dtrace_illval;	/* DTrace illegal value */
264 	kmutex_t	cpuc_pid_lock;		/* DTrace pid provider lock */
265 } cpu_core_t;
266 
267 #ifdef _KERNEL
268 extern cpu_core_t cpu_core[];
269 #endif /* _KERNEL */
270 
271 /*
272  * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack.
273  * Note that this isn't a test for a high PIL.  For example, cpu_intr_actv
274  * does not get updated when we go through sys_trap from TL>0 at high PIL.
275  * getpil() should be used instead to check for PIL levels.
276  */
277 #define	CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1))
278 
279 /*
280  * Check to see if an interrupt thread might be active at a given ipl.
281  * If so return true.
282  * We must be conservative--it is ok to give a false yes, but a false no
283  * will cause disaster.  (But if the situation changes after we check it is
284  * ok--the caller is trying to ensure that an interrupt routine has been
285  * exited).
286  * This is used when trying to remove an interrupt handler from an autovector
287  * list in avintr.c.
288  */
289 #define	INTR_ACTIVE(cpup, level)	\
290 	((level) <= LOCK_LEVEL ? 	\
291 	((cpup)->cpu_intr_actv & (1 << (level))) : (CPU_ON_INTR(cpup)))
292 
293 /*
294  * CPU_PSEUDO_RANDOM() returns a per CPU value that changes each time one
295  * looks at it. It's meant as a cheap mechanism to be incorporated in routines
296  * wanting to avoid biasing, but where true randomness isn't needed (just
297  * something that changes).
298  */
299 #define	CPU_PSEUDO_RANDOM() (CPU->cpu_rotor++)
300 
301 #if defined(_KERNEL) || defined(_KMEMUSER)
302 
303 #define	INTR_STACK_SIZE	MAX(DEFAULTSTKSZ, PAGESIZE)
304 
305 /* MEMBERS PROTECTED BY "atomicity": cpu_flags */
306 
307 /*
308  * Flags in the CPU structure.
309  *
310  * These are protected by cpu_lock (except during creation).
311  *
312  * Offlined-CPUs have three stages of being offline:
313  *
314  * CPU_ENABLE indicates that the CPU is participating in I/O interrupts
315  * that can be directed at a number of different CPUs.  If CPU_ENABLE
316  * is off, the CPU will not be given interrupts that can be sent elsewhere,
317  * but will still get interrupts from devices associated with that CPU only,
318  * and from other CPUs.
319  *
320  * CPU_OFFLINE indicates that the dispatcher should not allow any threads
321  * other than interrupt threads to run on that CPU.  A CPU will not have
322  * CPU_OFFLINE set if there are any bound threads (besides interrupts).
323  *
324  * CPU_QUIESCED is set if p_offline was able to completely turn idle the
325  * CPU and it will not have to run interrupt threads.  In this case it'll
326  * stay in the idle loop until CPU_QUIESCED is turned off.
327  *
328  * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully
329  * suspended (in the suspend path), or have yet to be resumed (in the resume
330  * case).
331  *
332  * On some platforms CPUs can be individually powered off.
333  * The following flags are set for powered off CPUs: CPU_QUIESCED,
334  * CPU_OFFLINE, and CPU_POWEROFF.  The following flags are cleared:
335  * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE.
336  */
337 #define	CPU_RUNNING	0x001		/* CPU running */
338 #define	CPU_READY	0x002		/* CPU ready for cross-calls */
339 #define	CPU_QUIESCED	0x004		/* CPU will stay in idle */
340 #define	CPU_EXISTS	0x008		/* CPU is configured */
341 #define	CPU_ENABLE	0x010		/* CPU enabled for interrupts */
342 #define	CPU_OFFLINE	0x020		/* CPU offline via p_online */
343 #define	CPU_POWEROFF	0x040		/* CPU is powered off */
344 #define	CPU_FROZEN	0x080		/* CPU is frozen via CPR suspend */
345 #define	CPU_SPARE	0x100		/* CPU offline available for use */
346 #define	CPU_FAULTED	0x200		/* CPU offline diagnosed faulty */
347 
348 #define	FMT_CPU_FLAGS							\
349 	"\20\12fault\11spare\10frozen"					\
350 	"\7poweroff\6offline\5enable\4exist\3quiesced\2ready\1run"
351 
352 #define	CPU_ACTIVE(cpu)	(((cpu)->cpu_flags & CPU_OFFLINE) == 0)
353 
354 /*
355  * Flags for cpu_offline(), cpu_faulted(), and cpu_spare().
356  */
357 #define	CPU_FORCED	0x0001		/* Force CPU offline */
358 
359 /*
360  * DTrace flags.
361  */
362 #define	CPU_DTRACE_NOFAULT	0x0001	/* Don't fault */
363 #define	CPU_DTRACE_DROP		0x0002	/* Drop this ECB */
364 #define	CPU_DTRACE_BADADDR	0x0004	/* DTrace fault: bad address */
365 #define	CPU_DTRACE_BADALIGN	0x0008	/* DTrace fault: bad alignment */
366 #define	CPU_DTRACE_DIVZERO	0x0010	/* DTrace fault: divide by zero */
367 #define	CPU_DTRACE_ILLOP	0x0020	/* DTrace fault: illegal operation */
368 #define	CPU_DTRACE_NOSCRATCH	0x0040	/* DTrace fault: out of scratch */
369 #define	CPU_DTRACE_KPRIV	0x0080	/* DTrace fault: bad kernel access */
370 #define	CPU_DTRACE_UPRIV	0x0100	/* DTrace fault: bad user access */
371 #define	CPU_DTRACE_TUPOFLOW	0x0200	/* DTrace fault: tuple stack overflow */
372 #if defined(__sparc)
373 #define	CPU_DTRACE_FAKERESTORE	0x0400	/* pid provider hint to getreg */
374 #endif
375 #define	CPU_DTRACE_ENTRY	0x0800	/* pid provider hint to ustack() */
376 #define	CPU_DTRACE_BADSTACK	0x1000	/* DTrace fault: bad stack */
377 
378 #define	CPU_DTRACE_FAULT	(CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \
379 				CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \
380 				CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \
381 				CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \
382 				CPU_DTRACE_BADSTACK)
383 #define	CPU_DTRACE_ERROR	(CPU_DTRACE_FAULT | CPU_DTRACE_DROP)
384 
385 /*
386  * Dispatcher flags
387  * These flags must be changed only by the current CPU.
388  */
389 #define	CPU_DISP_DONTSTEAL	0x01	/* CPU undergoing context swtch */
390 #define	CPU_DISP_HALTED		0x02	/* CPU halted waiting for interrupt */
391 
392 #endif /* _KERNEL || _KMEMUSER */
393 
394 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
395 
396 /*
397  * Macros for manipulating sets of CPUs as a bitmap.  Note that this
398  * bitmap may vary in size depending on the maximum CPU id a specific
399  * platform supports.  This may be different than the number of CPUs
400  * the platform supports, since CPU ids can be sparse.  We define two
401  * sets of macros; one for platforms where the maximum CPU id is less
402  * than the number of bits in a single word (32 in a 32-bit kernel,
403  * 64 in a 64-bit kernel), and one for platforms that require bitmaps
404  * of more than one word.
405  */
406 
407 #define	CPUSET_WORDS	BT_BITOUL(NCPU)
408 #define	CPUSET_NOTINSET	((uint_t)-1)
409 
410 #if	CPUSET_WORDS > 1
411 
412 typedef struct cpuset {
413 	ulong_t	cpub[CPUSET_WORDS];
414 } cpuset_t;
415 
416 /*
417  * Private functions for manipulating cpusets that do not fit in a
418  * single word.  These should not be used directly; instead the
419  * CPUSET_* macros should be used so the code will be portable
420  * across different definitions of NCPU.
421  */
422 extern	void	cpuset_all(cpuset_t *);
423 extern	void	cpuset_all_but(cpuset_t *, uint_t);
424 extern	int	cpuset_isnull(cpuset_t *);
425 extern	int	cpuset_cmp(cpuset_t *, cpuset_t *);
426 extern	void	cpuset_only(cpuset_t *, uint_t);
427 extern	uint_t	cpuset_find(cpuset_t *);
428 extern	void	cpuset_bounds(cpuset_t *, uint_t *, uint_t *);
429 
430 #define	CPUSET_ALL(set)			cpuset_all(&(set))
431 #define	CPUSET_ALL_BUT(set, cpu)	cpuset_all_but(&(set), cpu)
432 #define	CPUSET_ONLY(set, cpu)		cpuset_only(&(set), cpu)
433 #define	CPU_IN_SET(set, cpu)		BT_TEST((set).cpub, cpu)
434 #define	CPUSET_ADD(set, cpu)		BT_SET((set).cpub, cpu)
435 #define	CPUSET_DEL(set, cpu)		BT_CLEAR((set).cpub, cpu)
436 #define	CPUSET_ISNULL(set)		cpuset_isnull(&(set))
437 #define	CPUSET_ISEQUAL(set1, set2)	cpuset_cmp(&(set1), &(set2))
438 
439 /*
440  * Find one CPU in the cpuset.
441  * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu
442  * could be found. (i.e. empty set)
443  */
444 #define	CPUSET_FIND(set, cpu)		{		\
445 	cpu = cpuset_find(&(set));			\
446 }
447 
448 /*
449  * Determine the smallest and largest CPU id in the set. Returns
450  * CPUSET_NOTINSET in smallest and largest when set is empty.
451  */
452 #define	CPUSET_BOUNDS(set, smallest, largest)	{		\
453 	cpuset_bounds(&(set), &(smallest), &(largest));		\
454 }
455 
456 /*
457  * Atomic cpuset operations
458  * These are safe to use for concurrent cpuset manipulations.
459  * "xdel" and "xadd" are exclusive operations, that set "result" to "0"
460  * if the add or del was successful, or "-1" if not successful.
461  * (e.g. attempting to add a cpu to a cpuset that's already there, or
462  * deleting a cpu that's not in the cpuset)
463  */
464 
465 #define	CPUSET_ATOMIC_DEL(set, cpu)	BT_ATOMIC_CLEAR((set).cpub, (cpu))
466 #define	CPUSET_ATOMIC_ADD(set, cpu)	BT_ATOMIC_SET((set).cpub, (cpu))
467 
468 #define	CPUSET_ATOMIC_XADD(set, cpu, result) \
469 	BT_ATOMIC_SET_EXCL((set).cpub, cpu, result)
470 
471 #define	CPUSET_ATOMIC_XDEL(set, cpu, result) \
472 	BT_ATOMIC_CLEAR_EXCL((set).cpub, cpu, result)
473 
474 
475 #define	CPUSET_OR(set1, set2)		{		\
476 	int _i;						\
477 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
478 		(set1).cpub[_i] |= (set2).cpub[_i];	\
479 }
480 
481 #define	CPUSET_XOR(set1, set2)		{		\
482 	int _i;						\
483 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
484 		(set1).cpub[_i] ^= (set2).cpub[_i];	\
485 }
486 
487 #define	CPUSET_AND(set1, set2)		{		\
488 	int _i;						\
489 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
490 		(set1).cpub[_i] &= (set2).cpub[_i];	\
491 }
492 
493 #define	CPUSET_ZERO(set)		{		\
494 	int _i;						\
495 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
496 		(set).cpub[_i] = 0;			\
497 }
498 
499 #elif	CPUSET_WORDS == 1
500 
501 typedef	ulong_t	cpuset_t;	/* a set of CPUs */
502 
503 #define	CPUSET(cpu)			(1UL << (cpu))
504 
505 #define	CPUSET_ALL(set)			((void)((set) = ~0UL))
506 #define	CPUSET_ALL_BUT(set, cpu)	((void)((set) = ~CPUSET(cpu)))
507 #define	CPUSET_ONLY(set, cpu)		((void)((set) = CPUSET(cpu)))
508 #define	CPU_IN_SET(set, cpu)		((set) & CPUSET(cpu))
509 #define	CPUSET_ADD(set, cpu)		((void)((set) |= CPUSET(cpu)))
510 #define	CPUSET_DEL(set, cpu)		((void)((set) &= ~CPUSET(cpu)))
511 #define	CPUSET_ISNULL(set)		((set) == 0)
512 #define	CPUSET_ISEQUAL(set1, set2)	((set1) == (set2))
513 #define	CPUSET_OR(set1, set2)		((void)((set1) |= (set2)))
514 #define	CPUSET_XOR(set1, set2)		((void)((set1) ^= (set2)))
515 #define	CPUSET_AND(set1, set2)		((void)((set1) &= (set2)))
516 #define	CPUSET_ZERO(set)		((void)((set) = 0))
517 
518 #define	CPUSET_FIND(set, cpu)		{		\
519 	cpu = (uint_t)(lowbit(set) - 1);				\
520 }
521 
522 #define	CPUSET_BOUNDS(set, smallest, largest)	{	\
523 	smallest = (uint_t)(lowbit(set) - 1);		\
524 	largest = (uint_t)(highbit(set) - 1);		\
525 }
526 
527 #define	CPUSET_ATOMIC_DEL(set, cpu)	atomic_and_long(&(set), ~CPUSET(cpu))
528 #define	CPUSET_ATOMIC_ADD(set, cpu)	atomic_or_long(&(set), CPUSET(cpu))
529 
530 #define	CPUSET_ATOMIC_XADD(set, cpu, result) \
531 	{ result = atomic_set_long_excl(&(set), (cpu)); }
532 
533 #define	CPUSET_ATOMIC_XDEL(set, cpu, result) \
534 	{ result = atomic_clear_long_excl(&(set), (cpu)); }
535 
536 #else	/* CPUSET_WORDS <= 0 */
537 
538 #error NCPU is undefined or invalid
539 
540 #endif	/* CPUSET_WORDS	*/
541 
542 extern cpuset_t cpu_seqid_inuse;
543 
544 #endif	/* (_KERNEL || _KMEMUSER) && _MACHDEP */
545 
546 #define	CPU_CPR_OFFLINE		0x0
547 #define	CPU_CPR_ONLINE		0x1
548 #define	CPU_CPR_IS_OFFLINE(cpu)	(((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0)
549 #define	CPU_CPR_IS_ONLINE(cpu)	((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE)
550 #define	CPU_SET_CPR_FLAGS(cpu, flag)	((cpu)->cpu_cpr_flags |= flag)
551 
552 #if defined(_KERNEL) || defined(_KMEMUSER)
553 
554 extern struct cpu	*cpu[];		/* indexed by CPU number */
555 extern struct cpu	**cpu_seq;	/* indexed by sequential CPU id */
556 extern cpu_t		*cpu_list;	/* list of CPUs */
557 extern cpu_t		*cpu_active;	/* list of active CPUs */
558 extern int		ncpus;		/* number of CPUs present */
559 extern int		ncpus_online;	/* number of CPUs not quiesced */
560 extern int		max_ncpus;	/* max present before ncpus is known */
561 extern int		boot_max_ncpus;	/* like max_ncpus but for real */
562 extern int		boot_ncpus;	/* # cpus present @ boot */
563 extern processorid_t	max_cpuid;	/* maximum CPU number */
564 extern struct cpu	*cpu_inmotion;	/* offline or partition move target */
565 extern cpu_t		*clock_cpu_list;
566 extern processorid_t	max_cpu_seqid_ever;	/* maximum seqid ever given */
567 
568 #if defined(__i386) || defined(__amd64)
569 extern struct cpu *curcpup(void);
570 #define	CPU		(curcpup())	/* Pointer to current CPU */
571 #else
572 #define	CPU		(curthread->t_cpu)	/* Pointer to current CPU */
573 #endif
574 
575 /*
576  * CPU_CURRENT indicates to thread_affinity_set to use CPU->cpu_id
577  * as the target and to grab cpu_lock instead of requiring the caller
578  * to grab it.
579  */
580 #define	CPU_CURRENT	-3
581 
582 /*
583  * Per-CPU statistics
584  *
585  * cpu_stats_t contains numerous system and VM-related statistics, in the form
586  * of gauges or monotonically-increasing event occurrence counts.
587  */
588 
589 #define	CPU_STATS_ENTER_K()	kpreempt_disable()
590 #define	CPU_STATS_EXIT_K()	kpreempt_enable()
591 
592 #define	CPU_STATS_ADD_K(class, stat, amount) \
593 	{	kpreempt_disable(); /* keep from switching CPUs */\
594 		CPU_STATS_ADDQ(CPU, class, stat, amount); \
595 		kpreempt_enable(); \
596 	}
597 
598 #define	CPU_STATS_ADDQ(cp, class, stat, amount)	{			\
599 	extern void __dtrace_probe___cpu_##class##info_##stat(uint_t,	\
600 	    uint64_t *, cpu_t *);					\
601 	uint64_t *stataddr = &((cp)->cpu_stats.class.stat);		\
602 	__dtrace_probe___cpu_##class##info_##stat((amount),		\
603 	    stataddr, cp);						\
604 	*(stataddr) += (amount);					\
605 }
606 
607 #define	CPU_STATS(cp, stat)                                       \
608 	((cp)->cpu_stats.stat)
609 
610 /*
611  * Increment CPU generation value.
612  * This macro should be called whenever CPU goes on-line or off-line.
613  * Updates to cpu_generation should be protected by cpu_lock.
614  */
615 #define	CPU_NEW_GENERATION(cp)	((cp)->cpu_generation++)
616 
617 #endif /* _KERNEL || _KMEMUSER */
618 
619 /*
620  * CPU support routines.
621  */
622 #if	defined(_KERNEL) && defined(__STDC__)	/* not for genassym.c */
623 
624 struct zone;
625 
626 void	cpu_list_init(cpu_t *);
627 void	cpu_add_unit(cpu_t *);
628 void	cpu_del_unit(int cpuid);
629 void	cpu_add_active(cpu_t *);
630 void	cpu_kstat_init(cpu_t *);
631 void	cpu_visibility_add(cpu_t *, struct zone *);
632 void	cpu_visibility_remove(cpu_t *, struct zone *);
633 void	cpu_visibility_configure(cpu_t *, struct zone *);
634 void	cpu_visibility_unconfigure(cpu_t *, struct zone *);
635 void	cpu_visibility_online(cpu_t *, struct zone *);
636 void	cpu_visibility_offline(cpu_t *, struct zone *);
637 void	cpu_create_intrstat(cpu_t *);
638 void	cpu_delete_intrstat(cpu_t *);
639 int	cpu_kstat_intrstat_update(kstat_t *, int);
640 void	cpu_intr_swtch_enter(kthread_t *);
641 void	cpu_intr_swtch_exit(kthread_t *);
642 
643 void	mbox_lock_init(void);	 /* initialize cross-call locks */
644 void	mbox_init(int cpun);	 /* initialize cross-calls */
645 void	poke_cpu(int cpun);	 /* interrupt another CPU (to preempt) */
646 
647 /*
648  * values for safe_list.  Pause state that CPUs are in.
649  */
650 #define	PAUSE_IDLE	0		/* normal state */
651 #define	PAUSE_READY	1		/* paused thread ready to spl */
652 #define	PAUSE_WAIT	2		/* paused thread is spl-ed high */
653 #define	PAUSE_DIE	3		/* tell pause thread to leave */
654 #define	PAUSE_DEAD	4		/* pause thread has left */
655 
656 void	mach_cpu_pause(volatile char *);
657 
658 void	pause_cpus(cpu_t *off_cp);
659 void	start_cpus(void);
660 int	cpus_paused(void);
661 
662 void	cpu_pause_init(void);
663 cpu_t	*cpu_get(processorid_t cpun);	/* get the CPU struct associated */
664 
665 int	cpu_online(cpu_t *cp);			/* take cpu online */
666 int	cpu_offline(cpu_t *cp, int flags);	/* take cpu offline */
667 int	cpu_spare(cpu_t *cp, int flags);	/* take cpu to spare */
668 int	cpu_faulted(cpu_t *cp, int flags);	/* take cpu to faulted */
669 int	cpu_poweron(cpu_t *cp);		/* take powered-off cpu to offline */
670 int	cpu_poweroff(cpu_t *cp);	/* take offline cpu to powered-off */
671 
672 cpu_t	*cpu_intr_next(cpu_t *cp);	/* get next online CPU taking intrs */
673 int	cpu_intr_count(cpu_t *cp);	/* count # of CPUs handling intrs */
674 int	cpu_intr_on(cpu_t *cp);		/* CPU taking I/O interrupts? */
675 void	cpu_intr_enable(cpu_t *cp);	/* enable I/O interrupts */
676 int	cpu_intr_disable(cpu_t *cp);	/* disable I/O interrupts */
677 void	cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */
678 
679 /*
680  * Routines for checking CPU states.
681  */
682 int	cpu_is_online(cpu_t *);		/* check if CPU is online */
683 int	cpu_is_nointr(cpu_t *);		/* check if CPU can service intrs */
684 int	cpu_is_active(cpu_t *);		/* check if CPU can run threads */
685 int	cpu_is_offline(cpu_t *);	/* check if CPU is offline */
686 int	cpu_is_poweredoff(cpu_t *);	/* check if CPU is powered off */
687 
688 int	cpu_flagged_online(cpu_flag_t);	/* flags show CPU is online */
689 int	cpu_flagged_nointr(cpu_flag_t);	/* flags show CPU not handling intrs */
690 int	cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */
691 int	cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */
692 int	cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */
693 
694 /*
695  * The processor_info(2) state of a CPU is a simplified representation suitable
696  * for use by an application program.  Kernel subsystems should utilize the
697  * internal per-CPU state as given by the cpu_flags member of the cpu structure,
698  * as this information may include platform- or architecture-specific state
699  * critical to a subsystem's disposition of a particular CPU.
700  */
701 void	cpu_set_state(cpu_t *);		/* record/timestamp current state */
702 int	cpu_get_state(cpu_t *);		/* get current cpu state */
703 const char *cpu_get_state_str(cpu_t *);	/* get current cpu state as string */
704 
705 
706 void	cpu_set_curr_clock(uint64_t);	/* indicate the current CPU's freq */
707 void	cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */
708 						/* frequencies */
709 
710 int	cpu_configure(int);
711 int	cpu_unconfigure(int);
712 void	cpu_destroy_bound_threads(cpu_t *cp);
713 
714 extern int cpu_bind_thread(kthread_t *tp, processorid_t bind,
715     processorid_t *obind, int *error);
716 extern int cpu_unbind(processorid_t cpu_id, boolean_t force);
717 extern void thread_affinity_set(kthread_t *t, int cpu_id);
718 extern void thread_affinity_clear(kthread_t *t);
719 extern void affinity_set(int cpu_id);
720 extern void affinity_clear(void);
721 extern void init_cpu_mstate(struct cpu *, int);
722 extern void term_cpu_mstate(struct cpu *);
723 extern void new_cpu_mstate(int, hrtime_t);
724 extern void get_cpu_mstate(struct cpu *, hrtime_t *);
725 extern void thread_nomigrate(void);
726 extern void thread_allowmigrate(void);
727 extern void weakbinding_stop(void);
728 extern void weakbinding_start(void);
729 
730 /*
731  * The following routines affect the CPUs participation in interrupt processing,
732  * if that is applicable on the architecture.  This only affects interrupts
733  * which aren't directed at the processor (not cross calls).
734  *
735  * cpu_disable_intr returns non-zero if interrupts were previously enabled.
736  */
737 int	cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */
738 void	cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */
739 
740 /*
741  * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus
742  * and ncpus_online counts.
743  */
744 extern kmutex_t	cpu_lock;	/* lock protecting CPU data */
745 
746 /*
747  * CPU state change events
748  *
749  * Various subsystems need to know when CPUs change their state. They get this
750  * information by registering  CPU state change callbacks using
751  * register_cpu_setup_func(). Whenever any CPU changes its state, the callback
752  * function is called. The callback function is passed three arguments:
753  *
754  *   Event, described by cpu_setup_t
755  *   CPU ID
756  *   Transparent pointer passed when registering the callback
757  *
758  * The callback function is called with cpu_lock held. The return value from the
759  * callback function is usually ignored, except for CPU_CONFIG and CPU_UNCONFIG
760  * events. For these two events, non-zero return value indicates a failure and
761  * prevents successful completion of the operation.
762  *
763  * New events may be added in the future. Callback functions should ignore any
764  * events that they do not understand.
765  *
766  * The following events provide notification callbacks:
767  *
768  *  CPU_INIT	A new CPU is started and added to the list of active CPUs
769  *		  This event is only used during boot
770  *
771  *  CPU_CONFIG	A newly inserted CPU is prepared for starting running code
772  *		  This event is called by DR code
773  *
774  *  CPU_UNCONFIG CPU has been powered off and needs cleanup
775  *		  This event is called by DR code
776  *
777  *  CPU_ON	CPU is enabled but does not run anything yet
778  *
779  *  CPU_INTR_ON	CPU is enabled and has interrupts enabled
780  *
781  *  CPU_OFF	CPU is going offline but can still run threads
782  *
783  *  CPU_CPUPART_OUT	CPU is going to move out of its partition
784  *
785  *  CPU_CPUPART_IN	CPU is going to move to a new partition
786  *
787  *  CPU_SETUP	CPU is set up during boot and can run threads
788  */
789 typedef enum {
790 	CPU_INIT,
791 	CPU_CONFIG,
792 	CPU_UNCONFIG,
793 	CPU_ON,
794 	CPU_OFF,
795 	CPU_CPUPART_IN,
796 	CPU_CPUPART_OUT,
797 	CPU_SETUP,
798 	CPU_INTR_ON
799 } cpu_setup_t;
800 
801 typedef int cpu_setup_func_t(cpu_setup_t, int, void *);
802 
803 /*
804  * Routines used to register interest in cpu's being added to or removed
805  * from the system.
806  */
807 extern void register_cpu_setup_func(cpu_setup_func_t *, void *);
808 extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *);
809 extern void cpu_state_change_notify(int, cpu_setup_t);
810 
811 /*
812  * Call specified function on the given CPU
813  */
814 typedef void (*cpu_call_func_t)(uintptr_t, uintptr_t);
815 extern void cpu_call(cpu_t *, cpu_call_func_t, uintptr_t, uintptr_t);
816 
817 
818 /*
819  * Create various strings that describe the given CPU for the
820  * processor_info system call and configuration-related kstats.
821  */
822 #define	CPU_IDSTRLEN	100
823 
824 extern void init_cpu_info(struct cpu *);
825 extern void populate_idstr(struct cpu *);
826 extern void cpu_vm_data_init(struct cpu *);
827 extern void cpu_vm_data_destroy(struct cpu *);
828 
829 #endif	/* _KERNEL */
830 
831 #ifdef	__cplusplus
832 }
833 #endif
834 
835 #endif /* _SYS_CPUVAR_H */
836