xref: /illumos-gate/usr/src/uts/common/sys/cpuvar.h (revision 3c4c5929f4f5a266089758e0d018685b379767ad)
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  */
27 
28 #ifndef _SYS_CPUVAR_H
29 #define	_SYS_CPUVAR_H
30 
31 #include <sys/thread.h>
32 #include <sys/sysinfo.h>	/* has cpu_stat_t definition */
33 #include <sys/disp.h>
34 #include <sys/processor.h>
35 
36 #include <sys/loadavg.h>
37 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
38 #include <sys/machcpuvar.h>
39 #endif
40 
41 #include <sys/types.h>
42 #include <sys/file.h>
43 #include <sys/bitmap.h>
44 #include <sys/rwlock.h>
45 #include <sys/msacct.h>
46 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \
47 	(defined(__i386) || defined(__amd64))
48 #include <asm/cpuvar.h>
49 #endif
50 
51 #ifdef	__cplusplus
52 extern "C" {
53 #endif
54 
55 struct squeue_set_s;
56 
57 #define	CPU_CACHE_COHERENCE_SIZE	64
58 
59 /*
60  * For fast event tracing.
61  */
62 struct ftrace_record;
63 typedef struct ftrace_data {
64 	int			ftd_state;	/* ftrace flags */
65 	kmutex_t		ftd_unused;	/* ftrace buffer lock, unused */
66 	struct ftrace_record	*ftd_cur;	/* current record */
67 	struct ftrace_record	*ftd_first;	/* first record */
68 	struct ftrace_record	*ftd_last;	/* last record */
69 } ftrace_data_t;
70 
71 struct cyc_cpu;
72 struct nvlist;
73 
74 /*
75  * Per-CPU data.
76  *
77  * Be careful adding new members: if they are not the same in all modules (e.g.
78  * change size depending on a #define), CTF uniquification can fail to work
79  * properly.  Furthermore, this is transitive in that it applies recursively to
80  * all types pointed to by cpu_t.
81  */
82 typedef struct cpu {
83 	processorid_t	cpu_id;			/* CPU number */
84 	processorid_t	cpu_seqid;	/* sequential CPU id (0..ncpus-1) */
85 	volatile cpu_flag_t cpu_flags;		/* flags indicating CPU state */
86 	struct cpu	*cpu_self;		/* pointer to itself */
87 	kthread_t	*cpu_thread;		/* current thread */
88 	kthread_t	*cpu_idle_thread;	/* idle thread for this CPU */
89 	kthread_t	*cpu_pause_thread;	/* pause thread for this CPU */
90 	klwp_id_t	cpu_lwp;		/* current lwp (if any) */
91 	klwp_id_t	cpu_fpowner;		/* currently loaded fpu owner */
92 	struct cpupart	*cpu_part;		/* partition with this CPU */
93 	struct lgrp_ld	*cpu_lpl;		/* pointer to this cpu's load */
94 	int		cpu_cache_offset;	/* see kmem.c for details */
95 
96 	/*
97 	 * Links to other CPUs.  It is safe to walk these lists if
98 	 * one of the following is true:
99 	 * 	- cpu_lock held
100 	 * 	- preemption disabled via kpreempt_disable
101 	 * 	- PIL >= DISP_LEVEL
102 	 * 	- acting thread is an interrupt thread
103 	 * 	- all other CPUs are paused
104 	 */
105 	struct cpu	*cpu_next;		/* next existing CPU */
106 	struct cpu	*cpu_prev;		/* prev existing CPU */
107 	struct cpu	*cpu_next_onln;		/* next online (enabled) CPU */
108 	struct cpu	*cpu_prev_onln;		/* prev online (enabled) CPU */
109 	struct cpu	*cpu_next_part;		/* next CPU in partition */
110 	struct cpu	*cpu_prev_part;		/* prev CPU in partition */
111 	struct cpu	*cpu_next_lgrp;		/* next CPU in latency group */
112 	struct cpu	*cpu_prev_lgrp;		/* prev CPU in latency group */
113 	struct cpu	*cpu_next_lpl;		/* next CPU in lgrp partition */
114 	struct cpu	*cpu_prev_lpl;
115 
116 	struct cpu_pg	*cpu_pg;		/* cpu's processor groups */
117 
118 	void		*cpu_reserved[4];	/* reserved for future use */
119 
120 	/*
121 	 * Scheduling variables.
122 	 */
123 	disp_t		*cpu_disp;		/* dispatch queue data */
124 	/*
125 	 * Note that cpu_disp is set before the CPU is added to the system
126 	 * and is never modified.  Hence, no additional locking is needed
127 	 * beyond what's necessary to access the cpu_t structure.
128 	 */
129 	char		cpu_runrun;	/* scheduling flag - set to preempt */
130 	char		cpu_kprunrun;		/* force kernel preemption */
131 	pri_t		cpu_chosen_level; 	/* priority at which cpu */
132 						/* was chosen for scheduling */
133 	kthread_t	*cpu_dispthread; /* thread selected for dispatch */
134 	disp_lock_t	cpu_thread_lock; /* dispatcher lock on current thread */
135 	uint8_t		cpu_disp_flags;	/* flags used by dispatcher */
136 	/*
137 	 * The following field is updated when ever the cpu_dispthread
138 	 * changes. Also in places, where the current thread(cpu_dispthread)
139 	 * priority changes. This is used in disp_lowpri_cpu()
140 	 */
141 	pri_t		cpu_dispatch_pri; /* priority of cpu_dispthread */
142 	clock_t		cpu_last_swtch;	/* last time switched to new thread */
143 
144 	/*
145 	 * Interrupt data.
146 	 */
147 	caddr_t		cpu_intr_stack;	/* interrupt stack */
148 	kthread_t	*cpu_intr_thread; /* interrupt thread list */
149 	uint_t		cpu_intr_actv;	/* interrupt levels active (bitmask) */
150 	int		cpu_base_spl;	/* priority for highest rupt active */
151 
152 	/*
153 	 * Statistics.
154 	 */
155 	cpu_stats_t	cpu_stats;		/* per-CPU statistics */
156 	struct kstat	*cpu_info_kstat;	/* kstat for cpu info */
157 
158 	uintptr_t	cpu_profile_pc;	/* kernel PC in profile interrupt */
159 	uintptr_t	cpu_profile_upc; /* user PC in profile interrupt */
160 	uintptr_t	cpu_profile_pil; /* PIL when profile interrupted */
161 
162 	ftrace_data_t	cpu_ftrace;		/* per cpu ftrace data */
163 
164 	clock_t		cpu_deadman_counter;	/* used by deadman() */
165 	uint_t		cpu_deadman_countdown;	/* used by deadman() */
166 
167 	kmutex_t	cpu_cpc_ctxlock; /* protects context for idle thread */
168 	kcpc_ctx_t	*cpu_cpc_ctx;	/* performance counter context */
169 
170 	/*
171 	 * Configuration information for the processor_info system call.
172 	 */
173 	processor_info_t cpu_type_info;	/* config info */
174 	time_t		cpu_state_begin; /* when CPU entered current state */
175 	char		cpu_cpr_flags;	/* CPR related info */
176 	struct cyc_cpu	*cpu_cyclic;	/* per cpu cyclic subsystem data */
177 	struct squeue_set_s *cpu_squeue_set;	/* per cpu squeue set */
178 	struct nvlist	*cpu_props;	/* pool-related properties */
179 
180 	krwlock_t	cpu_ft_lock;		/* DTrace: fasttrap lock */
181 	uintptr_t	cpu_dtrace_caller;	/* DTrace: caller, if any */
182 	hrtime_t	cpu_dtrace_chillmark;	/* DTrace: chill mark time */
183 	hrtime_t	cpu_dtrace_chilled;	/* DTrace: total chill time */
184 	uint64_t	cpu_dtrace_probes;	/* DTrace: total probes fired */
185 	hrtime_t	cpu_dtrace_nsec;	/* DTrace: ns in dtrace_probe */
186 
187 	volatile uint16_t cpu_mstate;		/* cpu microstate */
188 	volatile uint16_t cpu_mstate_gen;	/* generation counter */
189 	volatile hrtime_t cpu_mstate_start;	/* cpu microstate start time */
190 	volatile hrtime_t cpu_acct[NCMSTATES];	/* cpu microstate data */
191 	hrtime_t	cpu_intracct[NCMSTATES]; /* interrupt mstate data */
192 	hrtime_t	cpu_waitrq;		/* cpu run-queue wait time */
193 	struct loadavg_s cpu_loadavg;		/* loadavg info for this cpu */
194 
195 	char		*cpu_idstr;	/* for printing and debugging */
196 	char		*cpu_brandstr;	/* for printing */
197 
198 	/*
199 	 * Sum of all device interrupt weights that are currently directed at
200 	 * this cpu. Cleared at start of interrupt redistribution.
201 	 */
202 	int32_t		cpu_intr_weight;
203 	void		*cpu_vm_data;
204 
205 	struct cpu_physid *cpu_physid;	/* physical associations */
206 
207 	uint64_t	cpu_curr_clock;		/* current clock freq in Hz */
208 	char		*cpu_supp_freqs;	/* supported freqs in Hz */
209 
210 	uintptr_t	cpu_cpcprofile_pc;	/* kernel PC in cpc interrupt */
211 	uintptr_t	cpu_cpcprofile_upc;	/* user PC in cpc interrupt */
212 
213 	/*
214 	 * Interrupt load factor used by dispatcher & softcall
215 	 */
216 	hrtime_t	cpu_intrlast;   /* total interrupt time (nsec) */
217 	int		cpu_intrload;   /* interrupt load factor (0-99%) */
218 
219 	uint_t		cpu_rotor;	/* for cheap pseudo-random numbers */
220 
221 	struct cu_cpu_info	*cpu_cu_info;	/* capacity & util. info */
222 
223 	/*
224 	 * cpu_generation is updated whenever CPU goes on-line or off-line.
225 	 * Updates to cpu_generation are protected by cpu_lock.
226 	 *
227 	 * See CPU_NEW_GENERATION() macro below.
228 	 */
229 	volatile uint_t		cpu_generation;	/* tracking on/off-line */
230 
231 	/*
232 	 * New members must be added /before/ this member, as the CTF tools
233 	 * rely on this being the last field before cpu_m, so they can
234 	 * correctly calculate the offset when synthetically adding the cpu_m
235 	 * member in objects that do not have it.  This fixup is required for
236 	 * uniquification to work correctly.
237 	 */
238 	uintptr_t	cpu_m_pad;
239 
240 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
241 	struct machcpu	cpu_m;		/* per architecture info */
242 #endif
243 } cpu_t;
244 
245 /*
246  * The cpu_core structure consists of per-CPU state available in any context.
247  * On some architectures, this may mean that the page(s) containing the
248  * NCPU-sized array of cpu_core structures must be locked in the TLB -- it
249  * is up to the platform to assure that this is performed properly.  Note that
250  * the structure is sized to avoid false sharing.
251  */
252 #define	CPUC_SIZE		(sizeof (uint16_t) + sizeof (uint8_t) + \
253 				sizeof (uintptr_t) + sizeof (kmutex_t))
254 #define	CPUC_PADSIZE		CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE
255 
256 typedef struct cpu_core {
257 	uint16_t	cpuc_dtrace_flags;	/* DTrace flags */
258 	uint8_t		cpuc_dcpc_intr_state;	/* DCPC provider intr state */
259 	uint8_t		cpuc_pad[CPUC_PADSIZE];	/* padding */
260 	uintptr_t	cpuc_dtrace_illval;	/* DTrace illegal value */
261 	kmutex_t	cpuc_pid_lock;		/* DTrace pid provider lock */
262 } cpu_core_t;
263 
264 #ifdef _KERNEL
265 extern cpu_core_t cpu_core[];
266 #endif /* _KERNEL */
267 
268 /*
269  * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack.
270  * Note that this isn't a test for a high PIL.  For example, cpu_intr_actv
271  * does not get updated when we go through sys_trap from TL>0 at high PIL.
272  * getpil() should be used instead to check for PIL levels.
273  */
274 #define	CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1))
275 
276 /*
277  * Check to see if an interrupt thread might be active at a given ipl.
278  * If so return true.
279  * We must be conservative--it is ok to give a false yes, but a false no
280  * will cause disaster.  (But if the situation changes after we check it is
281  * ok--the caller is trying to ensure that an interrupt routine has been
282  * exited).
283  * This is used when trying to remove an interrupt handler from an autovector
284  * list in avintr.c.
285  */
286 #define	INTR_ACTIVE(cpup, level)	\
287 	((level) <= LOCK_LEVEL ? 	\
288 	((cpup)->cpu_intr_actv & (1 << (level))) : (CPU_ON_INTR(cpup)))
289 
290 /*
291  * CPU_PSEUDO_RANDOM() returns a per CPU value that changes each time one
292  * looks at it. It's meant as a cheap mechanism to be incorporated in routines
293  * wanting to avoid biasing, but where true randomness isn't needed (just
294  * something that changes).
295  */
296 #define	CPU_PSEUDO_RANDOM() (CPU->cpu_rotor++)
297 
298 #if defined(_KERNEL) || defined(_KMEMUSER) || defined(_BOOT)
299 
300 #define	INTR_STACK_SIZE	MAX(DEFAULTSTKSZ, PAGESIZE)
301 
302 /* MEMBERS PROTECTED BY "atomicity": cpu_flags */
303 
304 /*
305  * Flags in the CPU structure.
306  *
307  * These are protected by cpu_lock (except during creation).
308  *
309  * Offlined-CPUs have three stages of being offline:
310  *
311  * CPU_ENABLE indicates that the CPU is participating in I/O interrupts
312  * that can be directed at a number of different CPUs.  If CPU_ENABLE
313  * is off, the CPU will not be given interrupts that can be sent elsewhere,
314  * but will still get interrupts from devices associated with that CPU only,
315  * and from other CPUs.
316  *
317  * CPU_OFFLINE indicates that the dispatcher should not allow any threads
318  * other than interrupt threads to run on that CPU.  A CPU will not have
319  * CPU_OFFLINE set if there are any bound threads (besides interrupts).
320  *
321  * CPU_QUIESCED is set if p_offline was able to completely turn idle the
322  * CPU and it will not have to run interrupt threads.  In this case it'll
323  * stay in the idle loop until CPU_QUIESCED is turned off.
324  *
325  * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully
326  * suspended (in the suspend path), or have yet to be resumed (in the resume
327  * case).
328  *
329  * On some platforms CPUs can be individually powered off.
330  * The following flags are set for powered off CPUs: CPU_QUIESCED,
331  * CPU_OFFLINE, and CPU_POWEROFF.  The following flags are cleared:
332  * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE.
333  */
334 #define	CPU_RUNNING	0x001		/* CPU running */
335 #define	CPU_READY	0x002		/* CPU ready for cross-calls */
336 #define	CPU_QUIESCED	0x004		/* CPU will stay in idle */
337 #define	CPU_EXISTS	0x008		/* CPU is configured */
338 #define	CPU_ENABLE	0x010		/* CPU enabled for interrupts */
339 #define	CPU_OFFLINE	0x020		/* CPU offline via p_online */
340 #define	CPU_POWEROFF	0x040		/* CPU is powered off */
341 #define	CPU_FROZEN	0x080		/* CPU is frozen via CPR suspend */
342 #define	CPU_SPARE	0x100		/* CPU offline available for use */
343 #define	CPU_FAULTED	0x200		/* CPU offline diagnosed faulty */
344 
345 #define	FMT_CPU_FLAGS							\
346 	"\20\12fault\11spare\10frozen"					\
347 	"\7poweroff\6offline\5enable\4exist\3quiesced\2ready\1run"
348 
349 #define	CPU_ACTIVE(cpu)	(((cpu)->cpu_flags & CPU_OFFLINE) == 0)
350 
351 /*
352  * Flags for cpu_offline(), cpu_faulted(), and cpu_spare().
353  */
354 #define	CPU_FORCED	0x0001		/* Force CPU offline */
355 
356 /*
357  * DTrace flags.
358  */
359 #define	CPU_DTRACE_NOFAULT	0x0001	/* Don't fault */
360 #define	CPU_DTRACE_DROP		0x0002	/* Drop this ECB */
361 #define	CPU_DTRACE_BADADDR	0x0004	/* DTrace fault: bad address */
362 #define	CPU_DTRACE_BADALIGN	0x0008	/* DTrace fault: bad alignment */
363 #define	CPU_DTRACE_DIVZERO	0x0010	/* DTrace fault: divide by zero */
364 #define	CPU_DTRACE_ILLOP	0x0020	/* DTrace fault: illegal operation */
365 #define	CPU_DTRACE_NOSCRATCH	0x0040	/* DTrace fault: out of scratch */
366 #define	CPU_DTRACE_KPRIV	0x0080	/* DTrace fault: bad kernel access */
367 #define	CPU_DTRACE_UPRIV	0x0100	/* DTrace fault: bad user access */
368 #define	CPU_DTRACE_TUPOFLOW	0x0200	/* DTrace fault: tuple stack overflow */
369 #if defined(__sparc)
370 #define	CPU_DTRACE_FAKERESTORE	0x0400	/* pid provider hint to getreg */
371 #endif
372 #define	CPU_DTRACE_ENTRY	0x0800	/* pid provider hint to ustack() */
373 #define	CPU_DTRACE_BADSTACK	0x1000	/* DTrace fault: bad stack */
374 
375 #define	CPU_DTRACE_FAULT	(CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \
376 				CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \
377 				CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \
378 				CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \
379 				CPU_DTRACE_BADSTACK)
380 #define	CPU_DTRACE_ERROR	(CPU_DTRACE_FAULT | CPU_DTRACE_DROP)
381 
382 /*
383  * Dispatcher flags
384  * These flags must be changed only by the current CPU.
385  */
386 #define	CPU_DISP_DONTSTEAL	0x01	/* CPU undergoing context swtch */
387 #define	CPU_DISP_HALTED		0x02	/* CPU halted waiting for interrupt */
388 
389 /* Note: inside ifdef: _KERNEL || _KMEMUSER || _BOOT */
390 #if defined(_MACHDEP)
391 
392 /*
393  * Macros for manipulating sets of CPUs as a bitmap.  Note that this
394  * bitmap may vary in size depending on the maximum CPU id a specific
395  * platform supports.  This may be different than the number of CPUs
396  * the platform supports, since CPU ids can be sparse.  We define two
397  * sets of macros; one for platforms where the maximum CPU id is less
398  * than the number of bits in a single word (32 in a 32-bit kernel,
399  * 64 in a 64-bit kernel), and one for platforms that require bitmaps
400  * of more than one word.
401  */
402 
403 #define	CPUSET_WORDS	BT_BITOUL(NCPU)
404 #define	CPUSET_NOTINSET	((uint_t)-1)
405 
406 #if	CPUSET_WORDS > 1
407 
408 typedef struct cpuset {
409 	ulong_t	cpub[CPUSET_WORDS];
410 } cpuset_t;
411 
412 /*
413  * Private functions for manipulating cpusets that do not fit in a
414  * single word.  These should not be used directly; instead the
415  * CPUSET_* macros should be used so the code will be portable
416  * across different definitions of NCPU.
417  */
418 extern	void	cpuset_all(cpuset_t *);
419 extern	void	cpuset_all_but(cpuset_t *, uint_t);
420 extern	int	cpuset_isnull(cpuset_t *);
421 extern	int	cpuset_cmp(cpuset_t *, cpuset_t *);
422 extern	void	cpuset_only(cpuset_t *, uint_t);
423 extern	uint_t	cpuset_find(cpuset_t *);
424 extern	void	cpuset_bounds(cpuset_t *, uint_t *, uint_t *);
425 
426 #define	CPUSET_ALL(set)			cpuset_all(&(set))
427 #define	CPUSET_ALL_BUT(set, cpu)	cpuset_all_but(&(set), cpu)
428 #define	CPUSET_ONLY(set, cpu)		cpuset_only(&(set), cpu)
429 #define	CPU_IN_SET(set, cpu)		BT_TEST((set).cpub, cpu)
430 #define	CPUSET_ADD(set, cpu)		BT_SET((set).cpub, cpu)
431 #define	CPUSET_DEL(set, cpu)		BT_CLEAR((set).cpub, cpu)
432 #define	CPUSET_ISNULL(set)		cpuset_isnull(&(set))
433 #define	CPUSET_ISEQUAL(set1, set2)	cpuset_cmp(&(set1), &(set2))
434 
435 /*
436  * Find one CPU in the cpuset.
437  * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu
438  * could be found. (i.e. empty set)
439  */
440 #define	CPUSET_FIND(set, cpu)		{		\
441 	cpu = cpuset_find(&(set));			\
442 }
443 
444 /*
445  * Determine the smallest and largest CPU id in the set. Returns
446  * CPUSET_NOTINSET in smallest and largest when set is empty.
447  */
448 #define	CPUSET_BOUNDS(set, smallest, largest)	{		\
449 	cpuset_bounds(&(set), &(smallest), &(largest));		\
450 }
451 
452 /*
453  * Atomic cpuset operations
454  * These are safe to use for concurrent cpuset manipulations.
455  * "xdel" and "xadd" are exclusive operations, that set "result" to "0"
456  * if the add or del was successful, or "-1" if not successful.
457  * (e.g. attempting to add a cpu to a cpuset that's already there, or
458  * deleting a cpu that's not in the cpuset)
459  */
460 
461 #define	CPUSET_ATOMIC_DEL(set, cpu)	BT_ATOMIC_CLEAR((set).cpub, (cpu))
462 #define	CPUSET_ATOMIC_ADD(set, cpu)	BT_ATOMIC_SET((set).cpub, (cpu))
463 
464 #define	CPUSET_ATOMIC_XADD(set, cpu, result) \
465 	BT_ATOMIC_SET_EXCL((set).cpub, cpu, result)
466 
467 #define	CPUSET_ATOMIC_XDEL(set, cpu, result) \
468 	BT_ATOMIC_CLEAR_EXCL((set).cpub, cpu, result)
469 
470 
471 #define	CPUSET_OR(set1, set2)		{		\
472 	int _i;						\
473 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
474 		(set1).cpub[_i] |= (set2).cpub[_i];	\
475 }
476 
477 #define	CPUSET_XOR(set1, set2)		{		\
478 	int _i;						\
479 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
480 		(set1).cpub[_i] ^= (set2).cpub[_i];	\
481 }
482 
483 #define	CPUSET_AND(set1, set2)		{		\
484 	int _i;						\
485 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
486 		(set1).cpub[_i] &= (set2).cpub[_i];	\
487 }
488 
489 #define	CPUSET_ZERO(set)		{		\
490 	int _i;						\
491 	for (_i = 0; _i < CPUSET_WORDS; _i++)		\
492 		(set).cpub[_i] = 0;			\
493 }
494 
495 #elif	CPUSET_WORDS == 1
496 
497 typedef	ulong_t	cpuset_t;	/* a set of CPUs */
498 
499 #define	CPUSET(cpu)			(1UL << (cpu))
500 
501 #define	CPUSET_ALL(set)			((void)((set) = ~0UL))
502 #define	CPUSET_ALL_BUT(set, cpu)	((void)((set) = ~CPUSET(cpu)))
503 #define	CPUSET_ONLY(set, cpu)		((void)((set) = CPUSET(cpu)))
504 #define	CPU_IN_SET(set, cpu)		((set) & CPUSET(cpu))
505 #define	CPUSET_ADD(set, cpu)		((void)((set) |= CPUSET(cpu)))
506 #define	CPUSET_DEL(set, cpu)		((void)((set) &= ~CPUSET(cpu)))
507 #define	CPUSET_ISNULL(set)		((set) == 0)
508 #define	CPUSET_ISEQUAL(set1, set2)	((set1) == (set2))
509 #define	CPUSET_OR(set1, set2)		((void)((set1) |= (set2)))
510 #define	CPUSET_XOR(set1, set2)		((void)((set1) ^= (set2)))
511 #define	CPUSET_AND(set1, set2)		((void)((set1) &= (set2)))
512 #define	CPUSET_ZERO(set)		((void)((set) = 0))
513 
514 #define	CPUSET_FIND(set, cpu)		{		\
515 	cpu = (uint_t)(lowbit(set) - 1);				\
516 }
517 
518 #define	CPUSET_BOUNDS(set, smallest, largest)	{	\
519 	smallest = (uint_t)(lowbit(set) - 1);		\
520 	largest = (uint_t)(highbit(set) - 1);		\
521 }
522 
523 #define	CPUSET_ATOMIC_DEL(set, cpu)	atomic_and_ulong(&(set), ~CPUSET(cpu))
524 #define	CPUSET_ATOMIC_ADD(set, cpu)	atomic_or_ulong(&(set), CPUSET(cpu))
525 
526 #define	CPUSET_ATOMIC_XADD(set, cpu, result) \
527 	{ result = atomic_set_long_excl(&(set), (cpu)); }
528 
529 #define	CPUSET_ATOMIC_XDEL(set, cpu, result) \
530 	{ result = atomic_clear_long_excl(&(set), (cpu)); }
531 
532 #else	/* CPUSET_WORDS <= 0 */
533 
534 #error NCPU is undefined or invalid
535 
536 #endif	/* CPUSET_WORDS	*/
537 
538 extern cpuset_t cpu_seqid_inuse;
539 
540 #endif	/* _MACHDEP */
541 #endif /* _KERNEL || _KMEMUSER || _BOOT */
542 
543 #define	CPU_CPR_OFFLINE		0x0
544 #define	CPU_CPR_ONLINE		0x1
545 #define	CPU_CPR_IS_OFFLINE(cpu)	(((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0)
546 #define	CPU_CPR_IS_ONLINE(cpu)	((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE)
547 #define	CPU_SET_CPR_FLAGS(cpu, flag)	((cpu)->cpu_cpr_flags |= flag)
548 
549 #if defined(_KERNEL) || defined(_KMEMUSER)
550 
551 extern struct cpu	*cpu[];		/* indexed by CPU number */
552 extern struct cpu	**cpu_seq;	/* indexed by sequential CPU id */
553 extern cpu_t		*cpu_list;	/* list of CPUs */
554 extern cpu_t		*cpu_active;	/* list of active CPUs */
555 extern int		ncpus;		/* number of CPUs present */
556 extern int		ncpus_online;	/* number of CPUs not quiesced */
557 extern int		max_ncpus;	/* max present before ncpus is known */
558 extern int		boot_max_ncpus;	/* like max_ncpus but for real */
559 extern int		boot_ncpus;	/* # cpus present @ boot */
560 extern processorid_t	max_cpuid;	/* maximum CPU number */
561 extern struct cpu	*cpu_inmotion;	/* offline or partition move target */
562 extern cpu_t		*clock_cpu_list;
563 extern processorid_t	max_cpu_seqid_ever;	/* maximum seqid ever given */
564 
565 #if defined(__i386) || defined(__amd64)
566 extern struct cpu *curcpup(void);
567 #define	CPU		(curcpup())	/* Pointer to current CPU */
568 #else
569 #define	CPU		(curthread->t_cpu)	/* Pointer to current CPU */
570 #endif
571 
572 /*
573  * CPU_CURRENT indicates to thread_affinity_set to use CPU->cpu_id
574  * as the target and to grab cpu_lock instead of requiring the caller
575  * to grab it.
576  */
577 #define	CPU_CURRENT	-3
578 
579 /*
580  * Per-CPU statistics
581  *
582  * cpu_stats_t contains numerous system and VM-related statistics, in the form
583  * of gauges or monotonically-increasing event occurrence counts.
584  */
585 
586 #define	CPU_STATS_ENTER_K()	kpreempt_disable()
587 #define	CPU_STATS_EXIT_K()	kpreempt_enable()
588 
589 #define	CPU_STATS_ADD_K(class, stat, amount) \
590 	{	kpreempt_disable(); /* keep from switching CPUs */\
591 		CPU_STATS_ADDQ(CPU, class, stat, amount); \
592 		kpreempt_enable(); \
593 	}
594 
595 #define	CPU_STATS_ADDQ(cp, class, stat, amount)	{			\
596 	extern void __dtrace_probe___cpu_##class##info_##stat(uint_t,	\
597 	    uint64_t *, cpu_t *);					\
598 	uint64_t *stataddr = &((cp)->cpu_stats.class.stat);		\
599 	__dtrace_probe___cpu_##class##info_##stat((amount),		\
600 	    stataddr, cp);						\
601 	*(stataddr) += (amount);					\
602 }
603 
604 #define	CPU_STATS(cp, stat)                                       \
605 	((cp)->cpu_stats.stat)
606 
607 /*
608  * Increment CPU generation value.
609  * This macro should be called whenever CPU goes on-line or off-line.
610  * Updates to cpu_generation should be protected by cpu_lock.
611  */
612 #define	CPU_NEW_GENERATION(cp)	((cp)->cpu_generation++)
613 
614 #endif /* _KERNEL || _KMEMUSER */
615 
616 /*
617  * CPU support routines.
618  */
619 #if	defined(_KERNEL) && defined(__STDC__)	/* not for genassym.c */
620 
621 struct zone;
622 
623 void	cpu_list_init(cpu_t *);
624 void	cpu_add_unit(cpu_t *);
625 void	cpu_del_unit(int cpuid);
626 void	cpu_add_active(cpu_t *);
627 void	cpu_kstat_init(cpu_t *);
628 void	cpu_visibility_add(cpu_t *, struct zone *);
629 void	cpu_visibility_remove(cpu_t *, struct zone *);
630 void	cpu_visibility_configure(cpu_t *, struct zone *);
631 void	cpu_visibility_unconfigure(cpu_t *, struct zone *);
632 void	cpu_visibility_online(cpu_t *, struct zone *);
633 void	cpu_visibility_offline(cpu_t *, struct zone *);
634 void	cpu_create_intrstat(cpu_t *);
635 void	cpu_delete_intrstat(cpu_t *);
636 int	cpu_kstat_intrstat_update(kstat_t *, int);
637 void	cpu_intr_swtch_enter(kthread_t *);
638 void	cpu_intr_swtch_exit(kthread_t *);
639 
640 void	mbox_lock_init(void);	 /* initialize cross-call locks */
641 void	mbox_init(int cpun);	 /* initialize cross-calls */
642 void	poke_cpu(int cpun);	 /* interrupt another CPU (to preempt) */
643 
644 /*
645  * values for safe_list.  Pause state that CPUs are in.
646  */
647 #define	PAUSE_IDLE	0		/* normal state */
648 #define	PAUSE_READY	1		/* paused thread ready to spl */
649 #define	PAUSE_WAIT	2		/* paused thread is spl-ed high */
650 #define	PAUSE_DIE	3		/* tell pause thread to leave */
651 #define	PAUSE_DEAD	4		/* pause thread has left */
652 
653 void	mach_cpu_pause(volatile char *);
654 
655 void	pause_cpus(cpu_t *off_cp, void *(*func)(void *));
656 void	start_cpus(void);
657 int	cpus_paused(void);
658 
659 void	cpu_pause_init(void);
660 cpu_t	*cpu_get(processorid_t cpun);	/* get the CPU struct associated */
661 
662 int	cpu_online(cpu_t *cp);			/* take cpu online */
663 int	cpu_offline(cpu_t *cp, int flags);	/* take cpu offline */
664 int	cpu_spare(cpu_t *cp, int flags);	/* take cpu to spare */
665 int	cpu_faulted(cpu_t *cp, int flags);	/* take cpu to faulted */
666 int	cpu_poweron(cpu_t *cp);		/* take powered-off cpu to offline */
667 int	cpu_poweroff(cpu_t *cp);	/* take offline cpu to powered-off */
668 
669 cpu_t	*cpu_intr_next(cpu_t *cp);	/* get next online CPU taking intrs */
670 int	cpu_intr_count(cpu_t *cp);	/* count # of CPUs handling intrs */
671 int	cpu_intr_on(cpu_t *cp);		/* CPU taking I/O interrupts? */
672 void	cpu_intr_enable(cpu_t *cp);	/* enable I/O interrupts */
673 int	cpu_intr_disable(cpu_t *cp);	/* disable I/O interrupts */
674 void	cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */
675 
676 /*
677  * Routines for checking CPU states.
678  */
679 int	cpu_is_online(cpu_t *);		/* check if CPU is online */
680 int	cpu_is_nointr(cpu_t *);		/* check if CPU can service intrs */
681 int	cpu_is_active(cpu_t *);		/* check if CPU can run threads */
682 int	cpu_is_offline(cpu_t *);	/* check if CPU is offline */
683 int	cpu_is_poweredoff(cpu_t *);	/* check if CPU is powered off */
684 
685 int	cpu_flagged_online(cpu_flag_t);	/* flags show CPU is online */
686 int	cpu_flagged_nointr(cpu_flag_t);	/* flags show CPU not handling intrs */
687 int	cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */
688 int	cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */
689 int	cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */
690 
691 /*
692  * The processor_info(2) state of a CPU is a simplified representation suitable
693  * for use by an application program.  Kernel subsystems should utilize the
694  * internal per-CPU state as given by the cpu_flags member of the cpu structure,
695  * as this information may include platform- or architecture-specific state
696  * critical to a subsystem's disposition of a particular CPU.
697  */
698 void	cpu_set_state(cpu_t *);		/* record/timestamp current state */
699 int	cpu_get_state(cpu_t *);		/* get current cpu state */
700 const char *cpu_get_state_str(cpu_t *);	/* get current cpu state as string */
701 
702 
703 void	cpu_set_curr_clock(uint64_t);	/* indicate the current CPU's freq */
704 void	cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */
705 						/* frequencies */
706 
707 int	cpu_configure(int);
708 int	cpu_unconfigure(int);
709 void	cpu_destroy_bound_threads(cpu_t *cp);
710 
711 extern int cpu_bind_thread(kthread_t *tp, processorid_t bind,
712     processorid_t *obind, int *error);
713 extern int cpu_unbind(processorid_t cpu_id, boolean_t force);
714 extern void thread_affinity_set(kthread_t *t, int cpu_id);
715 extern void thread_affinity_clear(kthread_t *t);
716 extern void affinity_set(int cpu_id);
717 extern void affinity_clear(void);
718 extern void init_cpu_mstate(struct cpu *, int);
719 extern void term_cpu_mstate(struct cpu *);
720 extern void new_cpu_mstate(int, hrtime_t);
721 extern void get_cpu_mstate(struct cpu *, hrtime_t *);
722 extern void thread_nomigrate(void);
723 extern void thread_allowmigrate(void);
724 extern void weakbinding_stop(void);
725 extern void weakbinding_start(void);
726 
727 /*
728  * The following routines affect the CPUs participation in interrupt processing,
729  * if that is applicable on the architecture.  This only affects interrupts
730  * which aren't directed at the processor (not cross calls).
731  *
732  * cpu_disable_intr returns non-zero if interrupts were previously enabled.
733  */
734 int	cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */
735 void	cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */
736 
737 /*
738  * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus
739  * and ncpus_online counts.
740  */
741 extern kmutex_t	cpu_lock;	/* lock protecting CPU data */
742 
743 /*
744  * CPU state change events
745  *
746  * Various subsystems need to know when CPUs change their state. They get this
747  * information by registering  CPU state change callbacks using
748  * register_cpu_setup_func(). Whenever any CPU changes its state, the callback
749  * function is called. The callback function is passed three arguments:
750  *
751  *   Event, described by cpu_setup_t
752  *   CPU ID
753  *   Transparent pointer passed when registering the callback
754  *
755  * The callback function is called with cpu_lock held. The return value from the
756  * callback function is usually ignored, except for CPU_CONFIG and CPU_UNCONFIG
757  * events. For these two events, non-zero return value indicates a failure and
758  * prevents successful completion of the operation.
759  *
760  * New events may be added in the future. Callback functions should ignore any
761  * events that they do not understand.
762  *
763  * The following events provide notification callbacks:
764  *
765  *  CPU_INIT	A new CPU is started and added to the list of active CPUs
766  *		  This event is only used during boot
767  *
768  *  CPU_CONFIG	A newly inserted CPU is prepared for starting running code
769  *		  This event is called by DR code
770  *
771  *  CPU_UNCONFIG CPU has been powered off and needs cleanup
772  *		  This event is called by DR code
773  *
774  *  CPU_ON	CPU is enabled but does not run anything yet
775  *
776  *  CPU_INTR_ON	CPU is enabled and has interrupts enabled
777  *
778  *  CPU_OFF	CPU is going offline but can still run threads
779  *
780  *  CPU_CPUPART_OUT	CPU is going to move out of its partition
781  *
782  *  CPU_CPUPART_IN	CPU is going to move to a new partition
783  *
784  *  CPU_SETUP	CPU is set up during boot and can run threads
785  */
786 typedef enum {
787 	CPU_INIT,
788 	CPU_CONFIG,
789 	CPU_UNCONFIG,
790 	CPU_ON,
791 	CPU_OFF,
792 	CPU_CPUPART_IN,
793 	CPU_CPUPART_OUT,
794 	CPU_SETUP,
795 	CPU_INTR_ON
796 } cpu_setup_t;
797 
798 typedef int cpu_setup_func_t(cpu_setup_t, int, void *);
799 
800 /*
801  * Routines used to register interest in cpu's being added to or removed
802  * from the system.
803  */
804 extern void register_cpu_setup_func(cpu_setup_func_t *, void *);
805 extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *);
806 extern void cpu_state_change_notify(int, cpu_setup_t);
807 
808 /*
809  * Call specified function on the given CPU
810  */
811 typedef void (*cpu_call_func_t)(uintptr_t, uintptr_t);
812 extern void cpu_call(cpu_t *, cpu_call_func_t, uintptr_t, uintptr_t);
813 
814 
815 /*
816  * Create various strings that describe the given CPU for the
817  * processor_info system call and configuration-related kstats.
818  */
819 #define	CPU_IDSTRLEN	100
820 
821 extern void init_cpu_info(struct cpu *);
822 extern void populate_idstr(struct cpu *);
823 extern void cpu_vm_data_init(struct cpu *);
824 extern void cpu_vm_data_destroy(struct cpu *);
825 
826 #endif	/* _KERNEL */
827 
828 #ifdef	__cplusplus
829 }
830 #endif
831 
832 #endif /* _SYS_CPUVAR_H */
833