xref: /illumos-gate/usr/src/uts/common/sys/cpuvar.h (revision b91bbe3818919523e866b0d622e2dd56874530a7)
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  * Copyright (c) 2012 by Delphix. All rights reserved.
25  * Copyright 2014 Igor Kozhukhov <ikozhukhov@gmail.com>.
26  * Copyright 2017 RackTop Systems.
27  * Copyright 2019 Joyent, Inc.
28  * Copyright 2021 Oxide Computer Company
29  */
30 
31 #ifndef _SYS_CPUVAR_H
32 #define	_SYS_CPUVAR_H
33 
34 #include <sys/thread.h>
35 #include <sys/sysinfo.h>	/* has cpu_stat_t definition */
36 #include <sys/disp.h>
37 #include <sys/processor.h>
38 #include <sys/kcpc.h>		/* has kcpc_ctx_t definition */
39 
40 #include <sys/loadavg.h>
41 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
42 #include <sys/machcpuvar.h>
43 #endif
44 
45 #include <sys/types.h>
46 #include <sys/file.h>
47 #include <sys/bitmap.h>
48 #include <sys/rwlock.h>
49 #include <sys/msacct.h>
50 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \
51 	(defined(__i386) || defined(__amd64))
52 #include <asm/cpuvar.h>
53 #endif
54 
55 #ifdef	__cplusplus
56 extern "C" {
57 #endif
58 
59 struct squeue_set_s;
60 
61 #define	CPU_CACHE_COHERENCE_SIZE	64
62 
63 /*
64  * For fast event tracing.
65  */
66 struct ftrace_record;
67 typedef struct ftrace_data {
68 	int			ftd_state;	/* ftrace flags */
69 	kmutex_t		ftd_unused;	/* ftrace buffer lock, unused */
70 	struct ftrace_record	*ftd_cur;	/* current record */
71 	struct ftrace_record	*ftd_first;	/* first record */
72 	struct ftrace_record	*ftd_last;	/* last record */
73 } ftrace_data_t;
74 
75 struct cyc_cpu;
76 struct nvlist;
77 
78 /*
79  * Per-CPU data.
80  *
81  * Be careful adding new members: if they are not the same in all modules (e.g.
82  * change size depending on a #define), CTF uniquification can fail to work
83  * properly.  Furthermore, this is transitive in that it applies recursively to
84  * all types pointed to by cpu_t.
85  */
86 typedef struct cpu {
87 	processorid_t	cpu_id;			/* CPU number */
88 	processorid_t	cpu_seqid;	/* sequential CPU id (0..ncpus-1) */
89 	volatile cpu_flag_t cpu_flags;		/* flags indicating CPU state */
90 	struct cpu	*cpu_self;		/* pointer to itself */
91 	kthread_t	*cpu_thread;		/* current thread */
92 	kthread_t	*cpu_idle_thread;	/* idle thread for this CPU */
93 	kthread_t	*cpu_pause_thread;	/* pause thread for this CPU */
94 	klwp_id_t	cpu_lwp;		/* current lwp (if any) */
95 	klwp_id_t	cpu_fpowner;		/* currently loaded fpu owner */
96 	struct cpupart	*cpu_part;		/* partition with this CPU */
97 	struct lgrp_ld	*cpu_lpl;		/* pointer to this cpu's load */
98 	int		cpu_cache_offset;	/* see kmem.c for details */
99 
100 	/*
101 	 * Links to other CPUs.  It is safe to walk these lists if
102 	 * one of the following is true:
103 	 *	- cpu_lock held
104 	 *	- preemption disabled via kpreempt_disable
105 	 *	- PIL >= DISP_LEVEL
106 	 *	- acting thread is an interrupt thread
107 	 *	- all other CPUs are paused
108 	 */
109 	struct cpu	*cpu_next;		/* next existing CPU */
110 	struct cpu	*cpu_prev;		/* prev existing CPU */
111 	struct cpu	*cpu_next_onln;		/* next online (enabled) CPU */
112 	struct cpu	*cpu_prev_onln;		/* prev online (enabled) CPU */
113 	struct cpu	*cpu_next_part;		/* next CPU in partition */
114 	struct cpu	*cpu_prev_part;		/* prev CPU in partition */
115 	struct cpu	*cpu_next_lgrp;		/* next CPU in latency group */
116 	struct cpu	*cpu_prev_lgrp;		/* prev CPU in latency group */
117 	struct cpu	*cpu_next_lpl;		/* next CPU in lgrp partition */
118 	struct cpu	*cpu_prev_lpl;
119 
120 	struct cpu_pg	*cpu_pg;		/* cpu's processor groups */
121 
122 	void		*cpu_reserved[4];	/* reserved for future use */
123 
124 	/*
125 	 * Scheduling variables.
126 	 */
127 	disp_t		*cpu_disp;		/* dispatch queue data */
128 	/*
129 	 * Note that cpu_disp is set before the CPU is added to the system
130 	 * and is never modified.  Hence, no additional locking is needed
131 	 * beyond what's necessary to access the cpu_t structure.
132 	 */
133 	char		cpu_runrun;	/* scheduling flag - set to preempt */
134 	char		cpu_kprunrun;		/* force kernel preemption */
135 	pri_t		cpu_chosen_level;	/* priority at which cpu */
136 						/* was chosen for scheduling */
137 	kthread_t	*cpu_dispthread; /* thread selected for dispatch */
138 	disp_lock_t	cpu_thread_lock; /* dispatcher lock on current thread */
139 	uint8_t		cpu_disp_flags;	/* flags used by dispatcher */
140 	/*
141 	 * The following field is updated when ever the cpu_dispthread
142 	 * changes. Also in places, where the current thread(cpu_dispthread)
143 	 * priority changes. This is used in disp_lowpri_cpu()
144 	 */
145 	pri_t		cpu_dispatch_pri; /* priority of cpu_dispthread */
146 	clock_t		cpu_last_swtch;	/* last time switched to new thread */
147 
148 	/*
149 	 * Interrupt data.
150 	 */
151 	caddr_t		cpu_intr_stack;	/* interrupt stack */
152 	kthread_t	*cpu_intr_thread; /* interrupt thread list */
153 	uint_t		cpu_intr_actv;	/* interrupt levels active (bitmask) */
154 	int		cpu_base_spl;	/* priority for highest rupt active */
155 
156 	/*
157 	 * Statistics.
158 	 */
159 	cpu_stats_t	cpu_stats;		/* per-CPU statistics */
160 	struct kstat	*cpu_info_kstat;	/* kstat for cpu info */
161 
162 	uintptr_t	cpu_profile_pc;	/* kernel PC in profile interrupt */
163 	uintptr_t	cpu_profile_upc; /* user PC in profile interrupt */
164 	uintptr_t	cpu_profile_pil; /* PIL when profile interrupted */
165 
166 	ftrace_data_t	cpu_ftrace;		/* per cpu ftrace data */
167 
168 	clock_t		cpu_deadman_counter;	/* used by deadman() */
169 	uint_t		cpu_deadman_countdown;	/* used by deadman() */
170 
171 	kmutex_t	cpu_cpc_ctxlock; /* protects context for idle thread */
172 	kcpc_ctx_t	*cpu_cpc_ctx;	/* performance counter context */
173 
174 	/*
175 	 * Configuration information for the processor_info system call.
176 	 */
177 	processor_info_t cpu_type_info;	/* config info */
178 	time_t		cpu_state_begin; /* when CPU entered current state */
179 	char		cpu_cpr_flags;	/* CPR related info */
180 	struct cyc_cpu	*cpu_cyclic;	/* per cpu cyclic subsystem data */
181 	struct squeue_set_s *cpu_squeue_set;	/* per cpu squeue set */
182 	struct nvlist	*cpu_props;	/* pool-related properties */
183 
184 	krwlock_t	cpu_ft_lock;		/* DTrace: fasttrap lock */
185 	uintptr_t	cpu_dtrace_caller;	/* DTrace: caller, if any */
186 	hrtime_t	cpu_dtrace_chillmark;	/* DTrace: chill mark time */
187 	hrtime_t	cpu_dtrace_chilled;	/* DTrace: total chill time */
188 	uint64_t	cpu_dtrace_probes;	/* DTrace: total probes fired */
189 	hrtime_t	cpu_dtrace_nsec;	/* DTrace: ns in dtrace_probe */
190 
191 	volatile uint16_t cpu_mstate;		/* cpu microstate */
192 	volatile uint16_t cpu_mstate_gen;	/* generation counter */
193 	volatile hrtime_t cpu_mstate_start;	/* cpu microstate start time */
194 	volatile hrtime_t cpu_acct[NCMSTATES];	/* cpu microstate data */
195 	hrtime_t	cpu_intracct[NCMSTATES]; /* interrupt mstate data */
196 	hrtime_t	cpu_waitrq;		/* cpu run-queue wait time */
197 	struct loadavg_s cpu_loadavg;		/* loadavg info for this cpu */
198 
199 	char		*cpu_idstr;	/* for printing and debugging */
200 	char		*cpu_brandstr;	/* for printing */
201 
202 	/*
203 	 * Sum of all device interrupt weights that are currently directed at
204 	 * this cpu. Cleared at start of interrupt redistribution.
205 	 */
206 	int32_t		cpu_intr_weight;
207 	void		*cpu_vm_data;
208 
209 	struct cpu_physid *cpu_physid;	/* physical associations */
210 
211 	uint64_t	cpu_curr_clock;		/* current clock freq in Hz */
212 	char		*cpu_supp_freqs;	/* supported freqs in Hz */
213 
214 	uintptr_t	cpu_cpcprofile_pc;	/* kernel PC in cpc interrupt */
215 	uintptr_t	cpu_cpcprofile_upc;	/* user PC in cpc interrupt */
216 
217 	/*
218 	 * Interrupt load factor used by dispatcher & softcall
219 	 */
220 	hrtime_t	cpu_intrlast;   /* total interrupt time (nsec) */
221 	int		cpu_intrload;   /* interrupt load factor (0-99%) */
222 
223 	uint_t		cpu_rotor;	/* for cheap pseudo-random numbers */
224 
225 	struct cu_cpu_info	*cpu_cu_info;	/* capacity & util. info */
226 
227 	/*
228 	 * cpu_generation is updated whenever CPU goes on-line or off-line.
229 	 * Updates to cpu_generation are protected by cpu_lock.
230 	 *
231 	 * See CPU_NEW_GENERATION() macro below.
232 	 */
233 	volatile uint_t		cpu_generation;	/* tracking on/off-line */
234 
235 	/*
236 	 * New members must be added /before/ this member, as the CTF tools
237 	 * rely on this being the last field before cpu_m, so they can
238 	 * correctly calculate the offset when synthetically adding the cpu_m
239 	 * member in objects that do not have it.  This fixup is required for
240 	 * uniquification to work correctly.
241 	 */
242 	uintptr_t	cpu_m_pad;
243 
244 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
245 	struct machcpu	cpu_m;		/* per architecture info */
246 #endif
247 } cpu_t;
248 
249 /*
250  * The cpu_core structure consists of per-CPU state available in any context.
251  * On some architectures, this may mean that the page(s) containing the
252  * NCPU-sized array of cpu_core structures must be locked in the TLB -- it
253  * is up to the platform to assure that this is performed properly.  Note that
254  * the structure is sized to avoid false sharing.
255  */
256 #define	CPUC_SIZE		(sizeof (uint16_t) + sizeof (uint8_t) + \
257 				sizeof (uintptr_t) + sizeof (kmutex_t))
258 #define	CPUC_PADSIZE		CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE
259 
260 typedef struct cpu_core {
261 	uint16_t	cpuc_dtrace_flags;	/* DTrace flags */
262 	uint8_t		cpuc_dcpc_intr_state;	/* DCPC provider intr state */
263 	uint8_t		cpuc_pad[CPUC_PADSIZE];	/* padding */
264 	uintptr_t	cpuc_dtrace_illval;	/* DTrace illegal value */
265 	kmutex_t	cpuc_pid_lock;		/* DTrace pid provider lock */
266 } cpu_core_t;
267 
268 #ifdef _KERNEL
269 extern cpu_core_t cpu_core[];
270 #endif /* _KERNEL */
271 
272 /*
273  * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack.
274  * Note that this isn't a test for a high PIL.  For example, cpu_intr_actv
275  * does not get updated when we go through sys_trap from TL>0 at high PIL.
276  * getpil() should be used instead to check for PIL levels.
277  */
278 #define	CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1))
279 
280 /*
281  * Check to see if an interrupt thread might be active at a given ipl.
282  * If so return true.
283  * We must be conservative--it is ok to give a false yes, but a false no
284  * will cause disaster.  (But if the situation changes after we check it is
285  * ok--the caller is trying to ensure that an interrupt routine has been
286  * exited).
287  * This is used when trying to remove an interrupt handler from an autovector
288  * list in avintr.c.
289  */
290 #define	INTR_ACTIVE(cpup, level)	\
291 	((level) <= LOCK_LEVEL ?	\
292 	((cpup)->cpu_intr_actv & (1 << (level))) : (CPU_ON_INTR(cpup)))
293 
294 /*
295  * CPU_PSEUDO_RANDOM() returns a per CPU value that changes each time one
296  * looks at it. It's meant as a cheap mechanism to be incorporated in routines
297  * wanting to avoid biasing, but where true randomness isn't needed (just
298  * something that changes).
299  */
300 #define	CPU_PSEUDO_RANDOM() (CPU->cpu_rotor++)
301 
302 #if defined(_KERNEL) || defined(_KMEMUSER) || defined(_BOOT)
303 
304 #define	INTR_STACK_SIZE	MAX(DEFAULTSTKSZ, PAGESIZE)
305 
306 /* MEMBERS PROTECTED BY "atomicity": cpu_flags */
307 
308 /*
309  * Flags in the CPU structure.
310  *
311  * These are protected by cpu_lock (except during creation).
312  *
313  * Offlined-CPUs have three stages of being offline:
314  *
315  * CPU_ENABLE indicates that the CPU is participating in I/O interrupts
316  * that can be directed at a number of different CPUs.  If CPU_ENABLE
317  * is off, the CPU will not be given interrupts that can be sent elsewhere,
318  * but will still get interrupts from devices associated with that CPU only,
319  * and from other CPUs.
320  *
321  * CPU_OFFLINE indicates that the dispatcher should not allow any threads
322  * other than interrupt threads to run on that CPU.  A CPU will not have
323  * CPU_OFFLINE set if there are any bound threads (besides interrupts).
324  *
325  * CPU_QUIESCED is set if p_offline was able to completely turn idle the
326  * CPU and it will not have to run interrupt threads.  In this case it'll
327  * stay in the idle loop until CPU_QUIESCED is turned off.
328  *
329  * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully
330  * suspended (in the suspend path), or have yet to be resumed (in the resume
331  * case).
332  *
333  * CPU_DISABLED is used for disabling SMT. It is similar to CPU_OFFLINE, but
334  * cannot be onlined without being forced.
335  *
336  * On some platforms CPUs can be individually powered off.
337  * The following flags are set for powered off CPUs: CPU_QUIESCED,
338  * CPU_OFFLINE, and CPU_POWEROFF.  The following flags are cleared:
339  * CPU_RUNNING, CPU_READY, CPU_EXISTS, CPU_DISABLED and CPU_ENABLE.
340  */
341 #define	CPU_RUNNING	0x001		/* CPU running */
342 #define	CPU_READY	0x002		/* CPU ready for cross-calls */
343 #define	CPU_QUIESCED	0x004		/* CPU will stay in idle */
344 #define	CPU_EXISTS	0x008		/* CPU is configured */
345 #define	CPU_ENABLE	0x010		/* CPU enabled for interrupts */
346 #define	CPU_OFFLINE	0x020		/* CPU offline via p_online */
347 #define	CPU_POWEROFF	0x040		/* CPU is powered off */
348 #define	CPU_FROZEN	0x080		/* CPU is frozen via CPR suspend */
349 #define	CPU_SPARE	0x100		/* CPU offline available for use */
350 #define	CPU_FAULTED	0x200		/* CPU offline diagnosed faulty */
351 #define	CPU_DISABLED	0x400		/* CPU explicitly disabled (HT) */
352 
353 #define	CPU_ACTIVE(cpu)	(((cpu)->cpu_flags & CPU_OFFLINE) == 0)
354 
355 /*
356  * Flags for cpu_offline(), cpu_faulted(), and cpu_spare().
357  */
358 #define	CPU_FORCED	0x0001		/* Force CPU offline */
359 
360 /*
361  * DTrace flags.
362  */
363 #define	CPU_DTRACE_NOFAULT	0x0001	/* Don't fault */
364 #define	CPU_DTRACE_DROP		0x0002	/* Drop this ECB */
365 #define	CPU_DTRACE_BADADDR	0x0004	/* DTrace fault: bad address */
366 #define	CPU_DTRACE_BADALIGN	0x0008	/* DTrace fault: bad alignment */
367 #define	CPU_DTRACE_DIVZERO	0x0010	/* DTrace fault: divide by zero */
368 #define	CPU_DTRACE_ILLOP	0x0020	/* DTrace fault: illegal operation */
369 #define	CPU_DTRACE_NOSCRATCH	0x0040	/* DTrace fault: out of scratch */
370 #define	CPU_DTRACE_KPRIV	0x0080	/* DTrace fault: bad kernel access */
371 #define	CPU_DTRACE_UPRIV	0x0100	/* DTrace fault: bad user access */
372 #define	CPU_DTRACE_TUPOFLOW	0x0200	/* DTrace fault: tuple stack overflow */
373 #if defined(__sparc)
374 #define	CPU_DTRACE_FAKERESTORE	0x0400	/* pid provider hint to getreg */
375 #endif
376 #define	CPU_DTRACE_ENTRY	0x0800	/* pid provider hint to ustack() */
377 #define	CPU_DTRACE_BADSTACK	0x1000	/* DTrace fault: bad stack */
378 
379 #define	CPU_DTRACE_FAULT	(CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \
380 				CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \
381 				CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \
382 				CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \
383 				CPU_DTRACE_BADSTACK)
384 #define	CPU_DTRACE_ERROR	(CPU_DTRACE_FAULT | CPU_DTRACE_DROP)
385 
386 /*
387  * Dispatcher flags
388  * These flags must be changed only by the current CPU.
389  */
390 #define	CPU_DISP_DONTSTEAL	0x01	/* CPU undergoing context swtch */
391 #define	CPU_DISP_HALTED		0x02	/* CPU halted waiting for interrupt */
392 
393 /*
394  * Macros for manipulating sets of CPUs as a bitmap.  Note that this
395  * bitmap may vary in size depending on the maximum CPU id a specific
396  * platform supports.  This may be different than the number of CPUs
397  * the platform supports, since CPU ids can be sparse.  We define two
398  * sets of macros; one for platforms where the maximum CPU id is less
399  * than the number of bits in a single word (32 in a 32-bit kernel,
400  * 64 in a 64-bit kernel), and one for platforms that require bitmaps
401  * of more than one word.
402  */
403 
404 #define	CPUSET_WORDS	BT_BITOUL(NCPU)
405 #define	CPUSET_NOTINSET	((uint_t)-1)
406 
407 #if defined(_MACHDEP)
408 struct cpuset {
409 	ulong_t	cpub[CPUSET_WORDS];
410 };
411 #else
412 struct cpuset;
413 #endif
414 
415 typedef struct cpuset cpuset_t;
416 
417 extern cpuset_t	*cpuset_alloc(int);
418 extern void	cpuset_free(cpuset_t *);
419 
420 /*
421  * Functions for manipulating cpusets.  These were previously considered
422  * private when some cpuset_t handling was performed in the CPUSET_* macros.
423  * They are now acceptable to use in non-_MACHDEP code.
424  */
425 extern void	cpuset_all(cpuset_t *);
426 extern void	cpuset_all_but(cpuset_t *, const uint_t);
427 extern int	cpuset_isnull(const cpuset_t *);
428 extern int	cpuset_isequal(const cpuset_t *, const cpuset_t *);
429 extern void	cpuset_only(cpuset_t *, const uint_t);
430 extern long	cpu_in_set(const cpuset_t *, const uint_t);
431 extern void	cpuset_add(cpuset_t *, const uint_t);
432 extern void	cpuset_del(cpuset_t *, const uint_t);
433 extern uint_t	cpuset_find(const cpuset_t *);
434 extern void	cpuset_bounds(const cpuset_t *, uint_t *, uint_t *);
435 extern void	cpuset_atomic_del(cpuset_t *, const uint_t);
436 extern void	cpuset_atomic_add(cpuset_t *, const uint_t);
437 extern long	cpuset_atomic_xadd(cpuset_t *, const uint_t);
438 extern long	cpuset_atomic_xdel(cpuset_t *, const uint_t);
439 extern void	cpuset_or(cpuset_t *, const cpuset_t *);
440 extern void	cpuset_xor(cpuset_t *, const cpuset_t *);
441 extern void	cpuset_and(cpuset_t *, const cpuset_t *);
442 extern void	cpuset_zero(cpuset_t *);
443 
444 
445 #if defined(_MACHDEP)
446 
447 /*
448  * Prior to the cpuset_t restructuring, the CPUSET_* macros contained
449  * significant logic, rather than directly invoking the backend functions.
450  * They are maintained here so that existing _MACHDEP code can use them.
451  */
452 
453 #define	CPUSET_ALL(set)			cpuset_all(&(set))
454 #define	CPUSET_ALL_BUT(set, cpu)	cpuset_all_but(&(set), cpu)
455 #define	CPUSET_ONLY(set, cpu)		cpuset_only(&(set), cpu)
456 #define	CPU_IN_SET(set, cpu)		cpu_in_set(&(set), cpu)
457 #define	CPUSET_ADD(set, cpu)		cpuset_add(&(set), cpu)
458 #define	CPUSET_DEL(set, cpu)		cpuset_del(&(set), cpu)
459 #define	CPUSET_ISNULL(set)		cpuset_isnull(&(set))
460 #define	CPUSET_ISEQUAL(set1, set2)	cpuset_isequal(&(set1), &(set2))
461 
462 /*
463  * Find one CPU in the cpuset.
464  * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu
465  * could be found. (i.e. empty set)
466  */
467 #define	CPUSET_FIND(set, cpu)		{		\
468 	cpu = cpuset_find(&(set));			\
469 }
470 
471 /*
472  * Determine the smallest and largest CPU id in the set. Returns
473  * CPUSET_NOTINSET in smallest and largest when set is empty.
474  */
475 #define	CPUSET_BOUNDS(set, smallest, largest)	{		\
476 	cpuset_bounds(&(set), &(smallest), &(largest));		\
477 }
478 
479 /*
480  * Atomic cpuset operations
481  * These are safe to use for concurrent cpuset manipulations.
482  * "xdel" and "xadd" are exclusive operations, that set "result" to "0"
483  * if the add or del was successful, or "-1" if not successful.
484  * (e.g. attempting to add a cpu to a cpuset that's already there, or
485  * deleting a cpu that's not in the cpuset)
486  */
487 
488 #define	CPUSET_ATOMIC_DEL(set, cpu)	cpuset_atomic_del(&(set), cpu)
489 #define	CPUSET_ATOMIC_ADD(set, cpu)	cpuset_atomic_add(&(set), cpu)
490 
491 #define	CPUSET_ATOMIC_XADD(set, cpu, result)	\
492 	(result) = cpuset_atomic_xadd(&(set), cpu)
493 
494 #define	CPUSET_ATOMIC_XDEL(set, cpu, result)	\
495 	(result) = cpuset_atomic_xdel(&(set), cpu)
496 
497 #define	CPUSET_OR(set1, set2)	cpuset_or(&(set1), &(set2))
498 
499 #define	CPUSET_XOR(set1, set2)	cpuset_xor(&(set1), &(set2))
500 
501 #define	CPUSET_AND(set1, set2)	cpuset_and(&(set1), &(set2))
502 
503 #define	CPUSET_ZERO(set)	cpuset_zero(&(set))
504 
505 #endif /* defined(_MACHDEP) */
506 
507 
508 extern cpuset_t cpu_seqid_inuse;
509 
510 extern struct cpu	*cpu[];		/* indexed by CPU number */
511 extern struct cpu	**cpu_seq;	/* indexed by sequential CPU id */
512 extern cpu_t		*cpu_list;	/* list of CPUs */
513 extern cpu_t		*cpu_active;	/* list of active CPUs */
514 extern cpuset_t		cpu_active_set;	/* cached set of active CPUs */
515 extern int		ncpus;		/* number of CPUs present */
516 extern int		ncpus_online;	/* number of CPUs not quiesced */
517 extern int		ncpus_intr_enabled; /* nr of CPUs taking I/O intrs */
518 extern int		max_ncpus;	/* max present before ncpus is known */
519 extern int		boot_max_ncpus;	/* like max_ncpus but for real */
520 extern int		boot_ncpus;	/* # cpus present @ boot */
521 extern processorid_t	max_cpuid;	/* maximum CPU number */
522 extern struct cpu	*cpu_inmotion;	/* offline or partition move target */
523 extern cpu_t		*clock_cpu_list;
524 extern processorid_t	max_cpu_seqid_ever;	/* maximum seqid ever given */
525 
526 #if defined(__i386) || defined(__amd64)
527 extern struct cpu *curcpup(void);
528 #define	CPU		(curcpup())	/* Pointer to current CPU */
529 #else
530 #define	CPU		(curthread->t_cpu)	/* Pointer to current CPU */
531 #endif
532 
533 /*
534  * CPU_CURRENT indicates to thread_affinity_set() to use whatever curthread's
535  * current CPU is; holding cpu_lock is not required.
536  */
537 #define	CPU_CURRENT	-3
538 
539 /*
540  * CPU_BEST can be used by thread_affinity_set() callers to set affinity to a
541  * good CPU (in particular, an ht_acquire()-friendly choice); holding cpu_lock
542  * is not required.
543  */
544 #define	CPU_BEST	-4
545 
546 /*
547  * Per-CPU statistics
548  *
549  * cpu_stats_t contains numerous system and VM-related statistics, in the form
550  * of gauges or monotonically-increasing event occurrence counts.
551  */
552 
553 #define	CPU_STATS_ENTER_K()	kpreempt_disable()
554 #define	CPU_STATS_EXIT_K()	kpreempt_enable()
555 
556 #define	CPU_STATS_ADD_K(class, stat, amount) \
557 	{	kpreempt_disable(); /* keep from switching CPUs */\
558 		CPU_STATS_ADDQ(CPU, class, stat, amount); \
559 		kpreempt_enable(); \
560 	}
561 
562 #define	CPU_STATS_ADDQ(cp, class, stat, amount)	{			\
563 	extern void __dtrace_probe___cpu_##class##info_##stat(uint_t,	\
564 	    uint64_t *, cpu_t *);					\
565 	uint64_t *stataddr = &((cp)->cpu_stats.class.stat);		\
566 	__dtrace_probe___cpu_##class##info_##stat((amount),		\
567 	    stataddr, cp);						\
568 	*(stataddr) += (amount);					\
569 }
570 
571 #define	CPU_STATS(cp, stat)                                       \
572 	((cp)->cpu_stats.stat)
573 
574 /*
575  * Increment CPU generation value.
576  * This macro should be called whenever CPU goes on-line or off-line.
577  * Updates to cpu_generation should be protected by cpu_lock.
578  */
579 #define	CPU_NEW_GENERATION(cp)	((cp)->cpu_generation++)
580 
581 #endif /* defined(_KERNEL) || defined(_KMEMUSER) */
582 
583 #define	CPU_CPR_OFFLINE		0x0
584 #define	CPU_CPR_ONLINE		0x1
585 #define	CPU_CPR_IS_OFFLINE(cpu)	(((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0)
586 #define	CPU_CPR_IS_ONLINE(cpu)	((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE)
587 #define	CPU_SET_CPR_FLAGS(cpu, flag)	((cpu)->cpu_cpr_flags |= flag)
588 
589 /*
590  * CPU support routines (not for genassym.c)
591  */
592 #if	(defined(_KERNEL) || defined(_FAKE_KERNEL)) && defined(__STDC__)
593 
594 struct zone;
595 
596 void	cpu_list_init(cpu_t *);
597 void	cpu_add_unit(cpu_t *);
598 void	cpu_del_unit(int cpuid);
599 void	cpu_add_active(cpu_t *);
600 void	cpu_kstat_init(cpu_t *);
601 void	cpu_visibility_add(cpu_t *, struct zone *);
602 void	cpu_visibility_remove(cpu_t *, struct zone *);
603 void	cpu_visibility_configure(cpu_t *, struct zone *);
604 void	cpu_visibility_unconfigure(cpu_t *, struct zone *);
605 void	cpu_visibility_online(cpu_t *, struct zone *);
606 void	cpu_visibility_offline(cpu_t *, struct zone *);
607 void	cpu_create_intrstat(cpu_t *);
608 void	cpu_delete_intrstat(cpu_t *);
609 int	cpu_kstat_intrstat_update(kstat_t *, int);
610 void	cpu_intr_swtch_enter(kthread_t *);
611 void	cpu_intr_swtch_exit(kthread_t *);
612 
613 void	mbox_lock_init(void);	 /* initialize cross-call locks */
614 void	mbox_init(int cpun);	 /* initialize cross-calls */
615 void	poke_cpu(int cpun);	 /* interrupt another CPU (to preempt) */
616 
617 /*
618  * values for safe_list.  Pause state that CPUs are in.
619  */
620 #define	PAUSE_IDLE	0		/* normal state */
621 #define	PAUSE_READY	1		/* paused thread ready to spl */
622 #define	PAUSE_WAIT	2		/* paused thread is spl-ed high */
623 #define	PAUSE_DIE	3		/* tell pause thread to leave */
624 #define	PAUSE_DEAD	4		/* pause thread has left */
625 
626 void	mach_cpu_pause(volatile char *);
627 
628 void	pause_cpus(cpu_t *off_cp, void *(*func)(void *));
629 void	start_cpus(void);
630 int	cpus_paused(void);
631 
632 void	cpu_pause_init(void);
633 cpu_t	*cpu_get(processorid_t cpun);	/* get the CPU struct associated */
634 
635 int	cpu_online(cpu_t *, int);	/* take cpu online */
636 int	cpu_offline(cpu_t *, int);	/* take cpu offline */
637 int	cpu_spare(cpu_t *, int);	/* take cpu to spare */
638 int	cpu_faulted(cpu_t *, int);	/* take cpu to faulted */
639 int	cpu_poweron(cpu_t *);		/* take powered-off cpu to offline */
640 int	cpu_poweroff(cpu_t *);		/* take offline cpu to powered-off */
641 
642 cpu_t	*cpu_intr_next(cpu_t *cp);	/* get next online CPU taking intrs */
643 int	cpu_intr_count(cpu_t *cp);	/* count # of CPUs handling intrs */
644 int	cpu_intr_on(cpu_t *cp);		/* CPU taking I/O interrupts? */
645 void	cpu_intr_enable(cpu_t *cp);	/* enable I/O interrupts */
646 int	cpu_intr_disable(cpu_t *cp);	/* disable I/O interrupts */
647 void	cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */
648 
649 /*
650  * Routines for checking CPU states.
651  */
652 int	cpu_is_online(cpu_t *);		/* check if CPU is online */
653 int	cpu_is_nointr(cpu_t *);		/* check if CPU can service intrs */
654 int	cpu_is_active(cpu_t *);		/* check if CPU can run threads */
655 int	cpu_is_offline(cpu_t *);	/* check if CPU is offline */
656 int	cpu_is_poweredoff(cpu_t *);	/* check if CPU is powered off */
657 
658 int	cpu_flagged_online(cpu_flag_t);	/* flags show CPU is online */
659 int	cpu_flagged_nointr(cpu_flag_t);	/* flags show CPU not handling intrs */
660 int	cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */
661 int	cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */
662 int	cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */
663 
664 /*
665  * The processor_info(2) state of a CPU is a simplified representation suitable
666  * for use by an application program.  Kernel subsystems should utilize the
667  * internal per-CPU state as given by the cpu_flags member of the cpu structure,
668  * as this information may include platform- or architecture-specific state
669  * critical to a subsystem's disposition of a particular CPU.
670  */
671 void	cpu_set_state(cpu_t *);		/* record/timestamp current state */
672 int	cpu_get_state(cpu_t *);		/* get current cpu state */
673 const char *cpu_get_state_str(cpu_flag_t);
674 
675 
676 void	cpu_set_curr_clock(uint64_t);	/* indicate the current CPU's freq */
677 void	cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */
678 						/* frequencies */
679 
680 int	cpu_configure(int);
681 int	cpu_unconfigure(int);
682 void	cpu_destroy_bound_threads(cpu_t *cp);
683 
684 extern int cpu_bind_thread(kthread_t *tp, processorid_t bind,
685     processorid_t *obind, int *error);
686 extern int cpu_unbind(processorid_t cpu_id, boolean_t force);
687 extern void thread_affinity_set(kthread_t *t, int cpu_id);
688 extern void thread_affinity_clear(kthread_t *t);
689 extern void affinity_set(int cpu_id);
690 extern void affinity_clear(void);
691 extern void init_cpu_mstate(struct cpu *, int);
692 extern void term_cpu_mstate(struct cpu *);
693 extern void new_cpu_mstate(int, hrtime_t);
694 extern void get_cpu_mstate(struct cpu *, hrtime_t *);
695 extern void thread_nomigrate(void);
696 extern void thread_allowmigrate(void);
697 extern void weakbinding_stop(void);
698 extern void weakbinding_start(void);
699 
700 /*
701  * The following routines affect the CPUs participation in interrupt processing,
702  * if that is applicable on the architecture.  This only affects interrupts
703  * which aren't directed at the processor (not cross calls).
704  *
705  * cpu_disable_intr returns non-zero if interrupts were previously enabled.
706  */
707 int	cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */
708 void	cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */
709 
710 /*
711  * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus
712  * and ncpus_online counts.
713  */
714 extern kmutex_t	cpu_lock;	/* lock protecting CPU data */
715 
716 /*
717  * CPU state change events
718  *
719  * Various subsystems need to know when CPUs change their state. They get this
720  * information by registering  CPU state change callbacks using
721  * register_cpu_setup_func(). Whenever any CPU changes its state, the callback
722  * function is called. The callback function is passed three arguments:
723  *
724  *   Event, described by cpu_setup_t
725  *   CPU ID
726  *   Transparent pointer passed when registering the callback
727  *
728  * The callback function is called with cpu_lock held. The return value from the
729  * callback function is usually ignored, except for CPU_CONFIG and CPU_UNCONFIG
730  * events. For these two events, non-zero return value indicates a failure and
731  * prevents successful completion of the operation.
732  *
733  * New events may be added in the future. Callback functions should ignore any
734  * events that they do not understand.
735  *
736  * The following events provide notification callbacks:
737  *
738  *  CPU_INIT	A new CPU is started and added to the list of active CPUs
739  *		  This event is only used during boot
740  *
741  *  CPU_CONFIG	A newly inserted CPU is prepared for starting running code
742  *		  This event is called by DR code
743  *
744  *  CPU_UNCONFIG CPU has been powered off and needs cleanup
745  *		  This event is called by DR code
746  *
747  *  CPU_ON	CPU is enabled but does not run anything yet
748  *
749  *  CPU_INTR_ON	CPU is enabled and has interrupts enabled
750  *
751  *  CPU_OFF	CPU is going offline but can still run threads
752  *
753  *  CPU_CPUPART_OUT	CPU is going to move out of its partition
754  *
755  *  CPU_CPUPART_IN	CPU is going to move to a new partition
756  *
757  *  CPU_SETUP	CPU is set up during boot and can run threads
758  */
759 typedef enum {
760 	CPU_INIT,
761 	CPU_CONFIG,
762 	CPU_UNCONFIG,
763 	CPU_ON,
764 	CPU_OFF,
765 	CPU_CPUPART_IN,
766 	CPU_CPUPART_OUT,
767 	CPU_SETUP,
768 	CPU_INTR_ON
769 } cpu_setup_t;
770 
771 typedef int cpu_setup_func_t(cpu_setup_t, int, void *);
772 
773 /*
774  * Routines used to register interest in cpu's being added to or removed
775  * from the system.
776  */
777 extern void register_cpu_setup_func(cpu_setup_func_t *, void *);
778 extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *);
779 extern void cpu_state_change_notify(int, cpu_setup_t);
780 
781 /*
782  * Call specified function on the given CPU
783  */
784 typedef void (*cpu_call_func_t)(uintptr_t, uintptr_t);
785 extern void cpu_call(cpu_t *, cpu_call_func_t, uintptr_t, uintptr_t);
786 
787 
788 /*
789  * Create various strings that describe the given CPU for the
790  * processor_info system call and configuration-related kstats.
791  */
792 #define	CPU_IDSTRLEN	100
793 
794 extern void init_cpu_info(struct cpu *);
795 extern void populate_idstr(struct cpu *);
796 extern void cpu_vm_data_init(struct cpu *);
797 extern void cpu_vm_data_destroy(struct cpu *);
798 
799 #endif	/* _KERNEL || _FAKE_KERNEL */
800 
801 #ifdef	__cplusplus
802 }
803 #endif
804 
805 #endif /* _SYS_CPUVAR_H */
806