4 * Implements the architecture independant portion of the LWKT
7 * Types which must already be defined when this header is included by
8 * userland: struct md_thread
11 #ifndef _SYS_THREAD_H_
12 #define _SYS_THREAD_H_
14 #ifndef _SYS_STDINT_H_
15 #include <sys/stdint.h> /* __int types */
18 #include <sys/param.h> /* MAXCOMLEN */
21 #include <sys/queue.h> /* TAILQ_* macros */
23 #ifndef _SYS_MSGPORT_H_
24 #include <sys/msgport.h> /* lwkt_port */
27 #include <sys/time.h> /* struct timeval */
32 #ifndef _SYS_SPINLOCK_H_
33 #include <sys/spinlock.h>
35 #ifndef _SYS_IOSCHED_H_
36 #include <sys/iosched.h>
38 #include <machine/thread.h>
53 typedef struct lwkt_queue *lwkt_queue_t;
54 typedef struct lwkt_token *lwkt_token_t;
55 typedef struct lwkt_tokref *lwkt_tokref_t;
56 typedef struct lwkt_cpu_msg *lwkt_cpu_msg_t;
57 typedef struct lwkt_cpu_port *lwkt_cpu_port_t;
58 typedef struct lwkt_ipiq *lwkt_ipiq_t;
59 typedef struct lwkt_cpusync *lwkt_cpusync_t;
60 typedef struct thread *thread_t;
62 typedef TAILQ_HEAD(lwkt_queue, thread) lwkt_queue;
65 * Differentiation between kernel threads and user threads. Userland
66 * programs which want to access to kernel structures have to define
67 * _KERNEL_STRUCTURES. This is a kinda safety valve to prevent badly
68 * written user programs from getting an LWKT thread that is neither the
69 * kernel nor the user version.
71 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
72 #ifndef _MACHINE_FRAME_H_
73 #include <machine/frame.h>
80 * Tokens are used to serialize access to information. They are 'soft'
81 * serialization entities that only stay in effect while a thread is
82 * running. If the thread blocks, other threads can run holding the same
83 * token(s). The tokens are reacquired when the original thread resumes.
85 * A thread can depend on its serialization remaining intact through a
86 * preemption. An interrupt which attempts to use the same token as the
87 * thread being preempted will reschedule itself for non-preemptive
88 * operation, so the new token code is capable of interlocking against
89 * interrupts as well as other cpus. This means that your token can only
90 * be (temporarily) lost if you *explicitly* block.
92 * Tokens are managed through a helper reference structure, lwkt_tokref. Each
93 * thread has a stack of tokref's to keep track of acquired tokens. Multiple
94 * tokref's may reference the same token.
96 * Tokens can be held shared or exclusive. An exclusive holder is able
97 * to set the TOK_EXCLUSIVE bit in t_count as long as no bit in the count
98 * mask is set. If unable to accomplish this TOK_EXCLREQ can be set instead
99 * which prevents any new shared acquisitions while the exclusive requestor
100 * spins in the scheduler. A shared holder can bump t_count by the increment
101 * value as long as neither TOK_EXCLUSIVE or TOK_EXCLREQ is set, else spin
104 * Multiple exclusive tokens are handled by treating the additional tokens
105 * as a special case of the shared token, incrementing the count value. This
106 * reduces the complexity of the token release code.
109 typedef struct lwkt_token {
110 long t_count; /* Shared/exclreq/exclusive access */
111 struct lwkt_tokref *t_ref; /* Exclusive ref */
112 long t_collisions; /* Collision counter */
113 const char *t_desc; /* Descriptive name */
116 #define TOK_EXCLUSIVE 0x00000001 /* Exclusive lock held */
117 #define TOK_EXCLREQ 0x00000002 /* Exclusive request pending */
118 #define TOK_INCR 4 /* Shared count increment */
119 #define TOK_COUNTMASK (~(long)(TOK_EXCLUSIVE|TOK_EXCLREQ))
122 * Static initialization for a lwkt_token.
124 #define LWKT_TOKEN_INITIALIZER(name) \
133 * Assert that a particular token is held
135 #define LWKT_TOKEN_HELD_ANY(tok) _lwkt_token_held_any(tok, curthread)
136 #define LWKT_TOKEN_HELD_EXCL(tok) _lwkt_token_held_excl(tok, curthread)
138 #define ASSERT_LWKT_TOKEN_HELD(tok) \
139 KKASSERT(LWKT_TOKEN_HELD_ANY(tok))
141 #define ASSERT_LWKT_TOKEN_HELD_EXCL(tok) \
142 KKASSERT(LWKT_TOKEN_HELD_EXCL(tok))
144 #define ASSERT_NO_TOKENS_HELD(td) \
145 KKASSERT((td)->td_toks_stop == &td->td_toks_array[0])
148 * Assert that a particular token is held and we are in a hard
149 * code execution section (interrupt, ipi, or hard code section).
150 * Hard code sections are not allowed to block or potentially block.
151 * e.g. lwkt_gettoken() would only be ok if the token were already
154 #define ASSERT_LWKT_TOKEN_HARD(tok) \
156 globaldata_t zgd __debugvar = mycpu; \
157 KKASSERT((tok)->t_ref && \
158 (tok)->t_ref->tr_owner == zgd->gd_curthread && \
159 zgd->gd_intr_nesting_level > 0); \
163 * Assert that a particular token is held and we are in a normal
164 * critical section. Critical sections will not be preempted but
165 * can explicitly block (tsleep, lwkt_gettoken, etc).
167 #define ASSERT_LWKT_TOKEN_CRIT(tok) \
169 globaldata_t zgd __debugvar = mycpu; \
170 KKASSERT((tok)->t_ref && \
171 (tok)->t_ref->tr_owner == zgd->gd_curthread && \
172 zgd->gd_curthread->td_critcount > 0); \
176 lwkt_token_t tr_tok; /* token in question */
177 long tr_count; /* TOK_EXCLUSIVE|TOK_EXCLREQ or 0 */
178 struct thread *tr_owner; /* me */
181 #define MAXCPUFIFO 32 /* power of 2 */
182 #define MAXCPUFIFO_MASK (MAXCPUFIFO - 1)
183 #define LWKT_MAXTOKENS 32 /* max tokens beneficially held by thread */
186 * Always cast to ipifunc_t when registering an ipi. The actual ipi function
187 * is called with both the data and an interrupt frame, but the ipi function
188 * that is registered might only declare a data argument.
190 typedef void (*ipifunc1_t)(void *arg);
191 typedef void (*ipifunc2_t)(void *arg, int arg2);
192 typedef void (*ipifunc3_t)(void *arg, int arg2, struct intrframe *frame);
195 int ip_rindex; /* only written by target cpu */
196 int ip_xindex; /* written by target, indicates completion */
197 int ip_windex; /* only written by source cpu */
202 char filler[32 - sizeof(int) - sizeof(void *) * 2];
203 } ip_info[MAXCPUFIFO];
207 * CPU Synchronization structure. See lwkt_cpusync_start() and
208 * lwkt_cpusync_finish() for more information.
210 typedef void (*cpusync_func_t)(void *arg);
212 struct lwkt_cpusync {
213 cpumask_t cs_mask; /* cpus running the sync */
214 cpumask_t cs_mack; /* mask acknowledge */
215 cpusync_func_t cs_func; /* function to execute */
216 void *cs_data; /* function data */
220 * The standard message and queue structure used for communications between
221 * cpus. Messages are typically queued via a machine-specific non-linked
222 * FIFO matrix allowing any cpu to send a message to any other cpu without
225 typedef struct lwkt_cpu_msg {
226 void (*cm_func)(lwkt_cpu_msg_t msg); /* primary dispatch function */
227 int cm_code; /* request code if applicable */
228 int cm_cpu; /* reply to cpu */
229 thread_t cm_originator; /* originating thread for wakeup */
233 * Thread structure. Note that ownership of a thread structure is special
234 * cased and there is no 'token'. A thread is always owned by the cpu
235 * represented by td_gd, any manipulation of the thread by some other cpu
236 * must be done through cpu_*msg() functions. e.g. you could request
237 * ownership of a thread that way, or hand a thread off to another cpu.
239 * NOTE: td_ucred is synchronized from the p_ucred on user->kernel syscall,
240 * trap, and AST/signal transitions to provide a stable ucred for
241 * (primarily) system calls. This field will be NULL for pure kernel
247 TAILQ_ENTRY(thread) td_threadq;
248 TAILQ_ENTRY(thread) td_allq;
249 TAILQ_ENTRY(thread) td_sleepq;
250 lwkt_port td_msgport; /* built-in message port for replies */
251 struct lwp *td_lwp; /* (optional) associated lwp */
252 struct proc *td_proc; /* (optional) associated process */
253 struct pcb *td_pcb; /* points to pcb and top of kstack */
254 struct globaldata *td_gd; /* associated with this cpu */
255 const char *td_wmesg; /* string name for blockage */
256 const volatile void *td_wchan; /* waiting on channel */
257 int td_pri; /* 0-31, 31=highest priority (note 1) */
258 int td_critcount; /* critical section priority */
259 u_int td_flags; /* TDF flags */
260 int td_wdomain; /* domain for wchan address (typ 0) */
261 void (*td_preemptable)(struct thread *td, int critcount);
262 void (*td_release)(struct thread *td);
263 char *td_kstack; /* kernel stack */
264 int td_kstack_size; /* size of kernel stack */
265 char *td_sp; /* kernel stack pointer for LWKT restore */
266 thread_t (*td_switch)(struct thread *ntd);
267 __uint64_t td_uticks; /* Statclock hits in user mode (uS) */
268 __uint64_t td_sticks; /* Statclock hits in system mode (uS) */
269 __uint64_t td_iticks; /* Statclock hits processing intr (uS) */
270 int td_locks; /* lockmgr lock debugging */
271 void *td_unused01; /* (future I/O scheduler heuristic) */
272 int td_refs; /* hold position in gd_tdallq / hold free */
273 int td_nest_count; /* prevent splz nesting */
274 int td_contended; /* token contention count */
275 u_int td_mpflags; /* flags can be set by foreign cpus */
276 int td_cscount; /* cpu synchronization master */
277 int td_wakefromcpu; /* who woke me up? */
278 int td_upri; /* user priority (sub-priority under td_pri) */
279 int td_type; /* thread type, TD_TYPE_ */
280 int td_tracker; /* for callers to debug lock counts */
281 int td_unused03[4]; /* for future fields */
282 struct iosched_data td_iosdata; /* Dynamic I/O scheduling data */
283 struct timeval td_start; /* start time for a thread/process */
284 char td_comm[MAXCOMLEN+1]; /* typ 16+1 bytes */
285 struct thread *td_preempted; /* we preempted this thread */
286 struct ucred *td_ucred; /* synchronized from p_ucred */
287 void *td_vmm; /* vmm private data */
288 lwkt_tokref_t td_toks_have; /* tokens we own */
289 lwkt_tokref_t td_toks_stop; /* tokens we want */
290 struct lwkt_tokref td_toks_array[LWKT_MAXTOKENS];
291 int td_fairq_load; /* fairq */
292 int td_fairq_count; /* fairq */
293 struct globaldata *td_migrate_gd; /* target gd for thread migration */
294 #ifdef DEBUG_CRIT_SECTIONS
295 #define CRIT_DEBUG_ARRAY_SIZE 32
296 #define CRIT_DEBUG_ARRAY_MASK (CRIT_DEBUG_ARRAY_SIZE - 1)
297 const char *td_crit_debug_array[CRIT_DEBUG_ARRAY_SIZE];
298 int td_crit_debug_index;
299 int td_in_crit_report;
301 struct md_thread td_mach;
303 #define SPINLOCK_DEBUG_ARRAY_SIZE 32
304 int td_spinlock_stack_id[SPINLOCK_DEBUG_ARRAY_SIZE];
305 struct spinlock *td_spinlock_stack[SPINLOCK_DEBUG_ARRAY_SIZE];
306 void *td_spinlock_caller_pc[SPINLOCK_DEBUG_ARRAY_SIZE];
309 * Track lockmgr locks held; lk->lk_filename:lk->lk_lineno is the holder
311 #define LOCKMGR_DEBUG_ARRAY_SIZE 8
312 int td_lockmgr_stack_id[LOCKMGR_DEBUG_ARRAY_SIZE];
313 struct lock *td_lockmgr_stack[LOCKMGR_DEBUG_ARRAY_SIZE];
317 #define td_toks_base td_toks_array[0]
318 #define td_toks_end td_toks_array[LWKT_MAXTOKENS]
320 #define TD_TOKS_HELD(td) ((td)->td_toks_stop != &(td)->td_toks_base)
321 #define TD_TOKS_NOT_HELD(td) ((td)->td_toks_stop == &(td)->td_toks_base)
324 * Thread flags. Note that TDF_RUNNING is cleared on the old thread after
325 * we switch to the new one, which is necessary because LWKTs don't need
326 * to hold the BGL. This flag is used by the exit code and the managed
327 * thread migration code. Note in addition that preemption will cause
328 * TDF_RUNNING to be cleared temporarily, so any code checking TDF_RUNNING
329 * must also check TDF_PREEMPT_LOCK.
331 * LWKT threads stay on their (per-cpu) run queue while running, not to
332 * be confused with user processes which are removed from the user scheduling
333 * run queue while actually running.
335 * td_threadq can represent the thread on one of three queues... the LWKT
336 * run queue, a tsleep queue, or an lwkt blocking queue. The LWKT subsystem
337 * does not allow a thread to be scheduled if it already resides on some
340 #define TDF_RUNNING 0x00000001 /* thread still active */
341 #define TDF_RUNQ 0x00000002 /* on an LWKT run queue */
342 #define TDF_PREEMPT_LOCK 0x00000004 /* I have been preempted */
343 #define TDF_PREEMPT_DONE 0x00000008 /* ac preemption complete */
344 #define TDF_NOSTART 0x00000010 /* do not schedule on create */
345 #define TDF_MIGRATING 0x00000020 /* thread is being migrated */
346 #define TDF_SINTR 0x00000040 /* interruptability for 'ps' */
347 #define TDF_TSLEEPQ 0x00000080 /* on a tsleep wait queue */
349 #define TDF_SYSTHREAD 0x00000100 /* reserve memory may be used */
350 #define TDF_ALLOCATED_THREAD 0x00000200 /* objcache allocated thread */
351 #define TDF_ALLOCATED_STACK 0x00000400 /* objcache allocated stack */
352 #define TDF_VERBOSE 0x00000800 /* verbose on exit */
353 #define TDF_DEADLKTREAT 0x00001000 /* special lockmgr treatment */
354 #define TDF_MARKER 0x00002000 /* tdallq list scan marker */
355 #define TDF_TIMEOUT_RUNNING 0x00004000 /* tsleep timeout race */
356 #define TDF_TIMEOUT 0x00008000 /* tsleep timeout */
357 #define TDF_INTTHREAD 0x00010000 /* interrupt thread */
358 #define TDF_TSLEEP_DESCHEDULED 0x00020000 /* tsleep core deschedule */
359 #define TDF_BLOCKED 0x00040000 /* Thread is blocked */
360 #define TDF_PANICWARN 0x00080000 /* panic warning in switch */
361 #define TDF_BLOCKQ 0x00100000 /* on block queue */
362 #define TDF_FORCE_SPINPORT 0x00200000
363 #define TDF_EXITING 0x00400000 /* thread exiting */
364 #define TDF_USINGFP 0x00800000 /* thread using fp coproc */
365 #define TDF_KERNELFP 0x01000000 /* kernel using fp coproc */
366 #define TDF_DELAYED_WAKEUP 0x02000000
367 #define TDF_FIXEDCPU 0x04000000 /* running cpu is fixed */
368 #define TDF_USERMODE 0x08000000 /* in or entering user mode */
369 #define TDF_NOFAULT 0x10000000 /* force onfault on fault */
371 #define TDF_MP_STOPREQ 0x00000001 /* suspend_kproc */
372 #define TDF_MP_WAKEREQ 0x00000002 /* resume_kproc */
373 #define TDF_MP_EXITWAIT 0x00000004 /* reaper, see lwp_wait() */
374 #define TDF_MP_EXITSIG 0x00000008 /* reaper, see lwp_wait() */
375 #define TDF_MP_BATCH_DEMARC 0x00000010 /* batch mode handling */
376 #define TDF_MP_DIDYIELD 0x00000020 /* effects scheduling */
378 #define TD_TYPE_GENERIC 0 /* generic thread */
379 #define TD_TYPE_CRYPTO 1 /* crypto thread */
380 #define TD_TYPE_NETISR 2 /* netisr thread */
383 * Thread priorities. Typically only one thread from any given
384 * user process scheduling queue is on the LWKT run queue at a time.
385 * Remember that there is one LWKT run queue per cpu.
387 * Critical sections are handled by bumping td_pri above TDPRI_MAX, which
388 * causes interrupts to be masked as they occur. When this occurs a
389 * rollup flag will be set in mycpu->gd_reqflags.
391 #define TDPRI_IDLE_THREAD 0 /* the idle thread */
392 #define TDPRI_IDLE_WORK 1 /* idle work (page zero, etc) */
393 #define TDPRI_USER_SCHEDULER 2 /* user scheduler helper */
394 #define TDPRI_USER_IDLE 4 /* user scheduler idle */
395 #define TDPRI_USER_NORM 6 /* user scheduler normal */
396 #define TDPRI_USER_REAL 8 /* user scheduler real time */
397 #define TDPRI_KERN_LPSCHED 9 /* scheduler helper for userland sch */
398 #define TDPRI_KERN_USER 10 /* kernel / block in syscall */
399 #define TDPRI_KERN_DAEMON 12 /* kernel daemon (pageout, etc) */
400 #define TDPRI_SOFT_NORM 14 /* kernel / normal */
401 #define TDPRI_SOFT_TIMER 16 /* kernel / timer */
402 #define TDPRI_EXITING 19 /* exiting thread */
403 #define TDPRI_INT_SUPPORT 20 /* kernel / high priority support */
404 #define TDPRI_INT_LOW 27 /* low priority interrupt */
405 #define TDPRI_INT_MED 28 /* medium priority interrupt */
406 #define TDPRI_INT_HIGH 29 /* high priority interrupt */
409 #define LWKT_THREAD_STACK (UPAGES * PAGE_SIZE)
411 #define IN_CRITICAL_SECT(td) ((td)->td_critcount)
418 extern struct lwkt_token mp_token;
419 extern struct lwkt_token pmap_token;
420 extern struct lwkt_token dev_token;
421 extern struct lwkt_token vm_token;
422 extern struct lwkt_token vmspace_token;
423 extern struct lwkt_token kvm_token;
424 extern struct lwkt_token sigio_token;
425 extern struct lwkt_token tty_token;
426 extern struct lwkt_token vnode_token;
427 extern struct lwkt_token revoke_token;
432 extern struct thread *lwkt_alloc_thread(struct thread *, int, int, int);
433 extern void lwkt_init_thread(struct thread *, void *, int, int,
434 struct globaldata *);
435 extern void lwkt_set_interrupt_support_thread(void);
436 extern void lwkt_set_comm(thread_t, const char *, ...) __printflike(2, 3);
437 extern void lwkt_free_thread(struct thread *);
438 extern void lwkt_gdinit(struct globaldata *);
439 extern void lwkt_switch(void);
440 extern void lwkt_switch_return(struct thread *);
441 extern void lwkt_preempt(thread_t, int);
442 extern void lwkt_schedule(thread_t);
443 extern void lwkt_schedule_noresched(thread_t);
444 extern void lwkt_schedule_self(thread_t);
445 extern void lwkt_deschedule(thread_t);
446 extern void lwkt_deschedule_self(thread_t);
447 extern void lwkt_yield(void);
448 extern void lwkt_yield_quick(void);
449 extern void lwkt_user_yield(void);
450 extern void lwkt_hold(thread_t);
451 extern void lwkt_rele(thread_t);
452 extern void lwkt_passive_release(thread_t);
453 extern void lwkt_maybe_splz(thread_t);
455 extern void lwkt_gettoken(lwkt_token_t);
456 extern void lwkt_gettoken_shared(lwkt_token_t);
457 extern void lwkt_gettoken_hard(lwkt_token_t);
458 extern int lwkt_trytoken(lwkt_token_t);
459 extern void lwkt_reltoken(lwkt_token_t);
460 extern void lwkt_reltoken_hard(lwkt_token_t);
461 extern int lwkt_cnttoken(lwkt_token_t, thread_t);
462 extern int lwkt_getalltokens(thread_t, int);
463 extern void lwkt_relalltokens(thread_t);
464 extern void lwkt_token_init(lwkt_token_t, const char *);
465 extern void lwkt_token_uninit(lwkt_token_t);
467 extern void lwkt_token_pool_init(void);
468 extern lwkt_token_t lwkt_token_pool_lookup(void *);
469 extern lwkt_token_t lwkt_getpooltoken(void *);
470 extern void lwkt_relpooltoken(void *);
472 extern void lwkt_token_swap(void);
474 extern void lwkt_setpri(thread_t, int);
475 extern void lwkt_setpri_initial(thread_t, int);
476 extern void lwkt_setpri_self(int);
477 extern void lwkt_schedulerclock(thread_t td);
478 extern void lwkt_setcpu_self(struct globaldata *);
479 extern void lwkt_migratecpu(int);
481 extern void lwkt_giveaway(struct thread *);
482 extern void lwkt_acquire(struct thread *);
483 extern int lwkt_send_ipiq3(struct globaldata *, ipifunc3_t, void *, int);
484 extern int lwkt_send_ipiq3_passive(struct globaldata *, ipifunc3_t,
486 extern int lwkt_send_ipiq3_nowait(struct globaldata *, ipifunc3_t,
488 extern int lwkt_send_ipiq3_bycpu(int, ipifunc3_t, void *, int);
489 extern int lwkt_send_ipiq3_mask(cpumask_t, ipifunc3_t, void *, int);
490 extern void lwkt_wait_ipiq(struct globaldata *, int);
491 extern int lwkt_seq_ipiq(struct globaldata *);
492 extern void lwkt_process_ipiq(void);
493 extern void lwkt_process_ipiq_frame(struct intrframe *);
494 extern void lwkt_smp_stopped(void);
495 extern void lwkt_synchronize_ipiqs(const char *);
497 /* lwkt_cpusync_init() - inline function in sys/thread2.h */
498 extern void lwkt_cpusync_simple(cpumask_t, cpusync_func_t, void *);
499 extern void lwkt_cpusync_interlock(lwkt_cpusync_t);
500 extern void lwkt_cpusync_deinterlock(lwkt_cpusync_t);
501 extern void lwkt_cpusync_quick(lwkt_cpusync_t);
503 extern void crit_panic(void) __dead2;
504 extern struct lwp *lwkt_preempted_proc(void);
506 extern int lwkt_create (void (*func)(void *), void *, struct thread **,
507 struct thread *, int, int,
508 const char *, ...) __printflike(7, 8);
509 extern void lwkt_exit (void) __dead2;
510 extern void lwkt_remove_tdallq (struct thread *);