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>
55 typedef struct lwkt_queue *lwkt_queue_t;
56 typedef struct lwkt_token *lwkt_token_t;
57 typedef struct lwkt_tokref *lwkt_tokref_t;
58 typedef struct lwkt_cpu_msg *lwkt_cpu_msg_t;
59 typedef struct lwkt_cpu_port *lwkt_cpu_port_t;
60 typedef struct lwkt_ipiq *lwkt_ipiq_t;
61 typedef struct lwkt_cpusync *lwkt_cpusync_t;
62 typedef struct thread *thread_t;
64 typedef TAILQ_HEAD(lwkt_queue, thread) lwkt_queue;
67 * Differentiation between kernel threads and user threads. Userland
68 * programs which want to access to kernel structures have to define
69 * _KERNEL_STRUCTURES. This is a kinda safety valve to prevent badly
70 * written user programs from getting an LWKT thread that is neither the
71 * kernel nor the user version.
73 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
74 #ifndef _MACHINE_THREAD_H_
75 #include <machine/thread.h> /* md_thread */
77 #ifndef _MACHINE_FRAME_H_
78 #include <machine/frame.h>
85 * Tokens are used to serialize access to information. They are 'soft'
86 * serialization entities that only stay in effect while a thread is
87 * running. If the thread blocks, other threads can run holding the same
88 * token(s). The tokens are reacquired when the original thread resumes.
90 * A thread can depend on its serialization remaining intact through a
91 * preemption. An interrupt which attempts to use the same token as the
92 * thread being preempted will reschedule itself for non-preemptive
93 * operation, so the new token code is capable of interlocking against
94 * interrupts as well as other cpus. This means that your token can only
95 * be (temporarily) lost if you *explicitly* block.
97 * Tokens are managed through a helper reference structure, lwkt_tokref. Each
98 * thread has a stack of tokref's to keep track of acquired tokens. Multiple
99 * tokref's may reference the same token.
101 * Tokens can be held shared or exclusive. An exclusive holder is able
102 * to set the TOK_EXCLUSIVE bit in t_count as long as no bit in the count
103 * mask is set. If unable to accomplish this TOK_EXCLREQ can be set instead
104 * which prevents any new shared acquisitions while the exclusive requestor
105 * spins in the scheduler. A shared holder can bump t_count by the increment
106 * value as long as neither TOK_EXCLUSIVE or TOK_EXCLREQ is set, else spin
109 * Multiple exclusive tokens are handled by treating the additional tokens
110 * as a special case of the shared token, incrementing the count value. This
111 * reduces the complexity of the token release code.
114 typedef struct lwkt_token {
115 long t_count; /* Shared/exclreq/exclusive access */
116 struct lwkt_tokref *t_ref; /* Exclusive ref */
117 long t_collisions; /* Collision counter */
118 const char *t_desc; /* Descriptive name */
121 #define TOK_EXCLUSIVE 0x00000001 /* Exclusive lock held */
122 #define TOK_EXCLREQ 0x00000002 /* Exclusive request pending */
123 #define TOK_INCR 4 /* Shared count increment */
124 #define TOK_COUNTMASK (~(long)(TOK_EXCLUSIVE|TOK_EXCLREQ))
127 * Static initialization for a lwkt_token.
129 #define LWKT_TOKEN_INITIALIZER(name) \
138 * Assert that a particular token is held
140 #define LWKT_TOKEN_HELD_ANY(tok) _lwkt_token_held_any(tok, curthread)
141 #define LWKT_TOKEN_HELD_EXCL(tok) _lwkt_token_held_excl(tok, curthread)
143 #define ASSERT_LWKT_TOKEN_HELD(tok) \
144 KKASSERT(LWKT_TOKEN_HELD_ANY(tok))
146 #define ASSERT_LWKT_TOKEN_HELD_EXCL(tok) \
147 KKASSERT(LWKT_TOKEN_HELD_EXCL(tok))
149 #define ASSERT_NO_TOKENS_HELD(td) \
150 KKASSERT((td)->td_toks_stop == &td->td_toks_array[0])
153 * Assert that a particular token is held and we are in a hard
154 * code execution section (interrupt, ipi, or hard code section).
155 * Hard code sections are not allowed to block or potentially block.
156 * e.g. lwkt_gettoken() would only be ok if the token were already
159 #define ASSERT_LWKT_TOKEN_HARD(tok) \
161 globaldata_t zgd __debugvar = mycpu; \
162 KKASSERT((tok)->t_ref && \
163 (tok)->t_ref->tr_owner == zgd->gd_curthread && \
164 zgd->gd_intr_nesting_level > 0); \
168 * Assert that a particular token is held and we are in a normal
169 * critical section. Critical sections will not be preempted but
170 * can explicitly block (tsleep, lwkt_gettoken, etc).
172 #define ASSERT_LWKT_TOKEN_CRIT(tok) \
174 globaldata_t zgd __debugvar = mycpu; \
175 KKASSERT((tok)->t_ref && \
176 (tok)->t_ref->tr_owner == zgd->gd_curthread && \
177 zgd->gd_curthread->td_critcount > 0); \
181 lwkt_token_t tr_tok; /* token in question */
182 long tr_count; /* TOK_EXCLUSIVE|TOK_EXCLREQ or 0 */
183 struct thread *tr_owner; /* me */
186 #define MAXCPUFIFO 32 /* power of 2 */
187 #define MAXCPUFIFO_MASK (MAXCPUFIFO - 1)
188 #define LWKT_MAXTOKENS 32 /* max tokens beneficially held by thread */
191 * Always cast to ipifunc_t when registering an ipi. The actual ipi function
192 * is called with both the data and an interrupt frame, but the ipi function
193 * that is registered might only declare a data argument.
195 typedef void (*ipifunc1_t)(void *arg);
196 typedef void (*ipifunc2_t)(void *arg, int arg2);
197 typedef void (*ipifunc3_t)(void *arg, int arg2, struct intrframe *frame);
199 typedef struct lwkt_ipiq {
200 int ip_rindex; /* only written by target cpu */
201 int ip_xindex; /* written by target, indicates completion */
202 int ip_windex; /* only written by source cpu */
207 char filler[32 - sizeof(int) - sizeof(void *) * 2];
208 } ip_info[MAXCPUFIFO];
212 * CPU Synchronization structure. See lwkt_cpusync_start() and
213 * lwkt_cpusync_finish() for more information.
215 typedef void (*cpusync_func_t)(void *arg);
217 struct lwkt_cpusync {
218 cpumask_t cs_mask; /* cpus running the sync */
219 cpumask_t cs_mack; /* mask acknowledge */
220 cpusync_func_t cs_func; /* function to execute */
221 void *cs_data; /* function data */
225 * The standard message and queue structure used for communications between
226 * cpus. Messages are typically queued via a machine-specific non-linked
227 * FIFO matrix allowing any cpu to send a message to any other cpu without
230 typedef struct lwkt_cpu_msg {
231 void (*cm_func)(lwkt_cpu_msg_t msg); /* primary dispatch function */
232 int cm_code; /* request code if applicable */
233 int cm_cpu; /* reply to cpu */
234 thread_t cm_originator; /* originating thread for wakeup */
238 * Thread structure. Note that ownership of a thread structure is special
239 * cased and there is no 'token'. A thread is always owned by the cpu
240 * represented by td_gd, any manipulation of the thread by some other cpu
241 * must be done through cpu_*msg() functions. e.g. you could request
242 * ownership of a thread that way, or hand a thread off to another cpu.
244 * NOTE: td_ucred is synchronized from the p_ucred on user->kernel syscall,
245 * trap, and AST/signal transitions to provide a stable ucred for
246 * (primarily) system calls. This field will be NULL for pure kernel
252 TAILQ_ENTRY(thread) td_threadq;
253 TAILQ_ENTRY(thread) td_allq;
254 TAILQ_ENTRY(thread) td_sleepq;
255 lwkt_port td_msgport; /* built-in message port for replies */
256 struct lwp *td_lwp; /* (optional) associated lwp */
257 struct proc *td_proc; /* (optional) associated process */
258 struct pcb *td_pcb; /* points to pcb and top of kstack */
259 struct globaldata *td_gd; /* associated with this cpu */
260 const char *td_wmesg; /* string name for blockage */
261 const volatile void *td_wchan; /* waiting on channel */
262 int td_pri; /* 0-31, 31=highest priority (note 1) */
263 int td_critcount; /* critical section priority */
264 u_int td_flags; /* TDF flags */
265 int td_wdomain; /* domain for wchan address (typ 0) */
266 void (*td_preemptable)(struct thread *td, int critcount);
267 void (*td_release)(struct thread *td);
268 char *td_kstack; /* kernel stack */
269 int td_kstack_size; /* size of kernel stack */
270 char *td_sp; /* kernel stack pointer for LWKT restore */
271 thread_t (*td_switch)(struct thread *ntd);
272 __uint64_t td_uticks; /* Statclock hits in user mode (uS) */
273 __uint64_t td_sticks; /* Statclock hits in system mode (uS) */
274 __uint64_t td_iticks; /* Statclock hits processing intr (uS) */
275 int td_locks; /* lockmgr lock debugging */
276 void *td_dsched_priv1; /* priv data for I/O schedulers */
277 int td_refs; /* hold position in gd_tdallq / hold free */
278 int td_nest_count; /* prevent splz nesting */
279 int td_contended; /* token contention count */
280 u_int td_mpflags; /* flags can be set by foreign cpus */
281 int td_cscount; /* cpu synchronization master */
282 int td_wakefromcpu; /* who woke me up? */
283 int td_upri; /* user priority (sub-priority under td_pri) */
284 int td_type; /* thread type, TD_TYPE_ */
285 int td_unused02[1]; /* for future fields */
286 int td_unused03[4]; /* for future fields */
287 struct iosched_data td_iosdata; /* Dynamic I/O scheduling data */
288 struct timeval td_start; /* start time for a thread/process */
289 char td_comm[MAXCOMLEN+1]; /* typ 16+1 bytes */
290 struct thread *td_preempted; /* we preempted this thread */
291 struct ucred *td_ucred; /* synchronized from p_ucred */
292 void *td_unused04; /* for future fields */
293 lwkt_tokref_t td_toks_have; /* tokens we own */
294 lwkt_tokref_t td_toks_stop; /* tokens we want */
295 struct lwkt_tokref td_toks_array[LWKT_MAXTOKENS];
296 int td_fairq_load; /* fairq */
297 int td_fairq_count; /* fairq */
298 struct globaldata *td_migrate_gd; /* target gd for thread migration */
299 #ifdef DEBUG_CRIT_SECTIONS
300 #define CRIT_DEBUG_ARRAY_SIZE 32
301 #define CRIT_DEBUG_ARRAY_MASK (CRIT_DEBUG_ARRAY_SIZE - 1)
302 const char *td_crit_debug_array[CRIT_DEBUG_ARRAY_SIZE];
303 int td_crit_debug_index;
304 int td_in_crit_report;
306 struct md_thread td_mach;
308 #define SPINLOCK_DEBUG_ARRAY_SIZE 32
309 int td_spinlock_stack_id[SPINLOCK_DEBUG_ARRAY_SIZE];
310 struct spinlock *td_spinlock_stack[SPINLOCK_DEBUG_ARRAY_SIZE];
311 void *td_spinlock_caller_pc[SPINLOCK_DEBUG_ARRAY_SIZE];
314 * Track lockmgr locks held; lk->lk_filename:lk->lk_lineno is the holder
316 #define LOCKMGR_DEBUG_ARRAY_SIZE 8
317 int td_lockmgr_stack_id[LOCKMGR_DEBUG_ARRAY_SIZE];
318 struct lock *td_lockmgr_stack[LOCKMGR_DEBUG_ARRAY_SIZE];
322 #define td_toks_base td_toks_array[0]
323 #define td_toks_end td_toks_array[LWKT_MAXTOKENS]
325 #define TD_TOKS_HELD(td) ((td)->td_toks_stop != &(td)->td_toks_base)
326 #define TD_TOKS_NOT_HELD(td) ((td)->td_toks_stop == &(td)->td_toks_base)
329 * Thread flags. Note that TDF_RUNNING is cleared on the old thread after
330 * we switch to the new one, which is necessary because LWKTs don't need
331 * to hold the BGL. This flag is used by the exit code and the managed
332 * thread migration code. Note in addition that preemption will cause
333 * TDF_RUNNING to be cleared temporarily, so any code checking TDF_RUNNING
334 * must also check TDF_PREEMPT_LOCK.
336 * LWKT threads stay on their (per-cpu) run queue while running, not to
337 * be confused with user processes which are removed from the user scheduling
338 * run queue while actually running.
340 * td_threadq can represent the thread on one of three queues... the LWKT
341 * run queue, a tsleep queue, or an lwkt blocking queue. The LWKT subsystem
342 * does not allow a thread to be scheduled if it already resides on some
345 #define TDF_RUNNING 0x00000001 /* thread still active */
346 #define TDF_RUNQ 0x00000002 /* on an LWKT run queue */
347 #define TDF_PREEMPT_LOCK 0x00000004 /* I have been preempted */
348 #define TDF_PREEMPT_DONE 0x00000008 /* ac preemption complete */
349 #define TDF_NOSTART 0x00000010 /* do not schedule on create */
350 #define TDF_MIGRATING 0x00000020 /* thread is being migrated */
351 #define TDF_SINTR 0x00000040 /* interruptability for 'ps' */
352 #define TDF_TSLEEPQ 0x00000080 /* on a tsleep wait queue */
354 #define TDF_SYSTHREAD 0x00000100 /* reserve memory may be used */
355 #define TDF_ALLOCATED_THREAD 0x00000200 /* objcache allocated thread */
356 #define TDF_ALLOCATED_STACK 0x00000400 /* objcache allocated stack */
357 #define TDF_VERBOSE 0x00000800 /* verbose on exit */
358 #define TDF_DEADLKTREAT 0x00001000 /* special lockmgr treatment */
359 #define TDF_MARKER 0x00002000 /* tdallq list scan marker */
360 #define TDF_TIMEOUT_RUNNING 0x00004000 /* tsleep timeout race */
361 #define TDF_TIMEOUT 0x00008000 /* tsleep timeout */
362 #define TDF_INTTHREAD 0x00010000 /* interrupt thread */
363 #define TDF_TSLEEP_DESCHEDULED 0x00020000 /* tsleep core deschedule */
364 #define TDF_BLOCKED 0x00040000 /* Thread is blocked */
365 #define TDF_PANICWARN 0x00080000 /* panic warning in switch */
366 #define TDF_BLOCKQ 0x00100000 /* on block queue */
367 #define TDF_FORCE_SPINPORT 0x00200000
368 #define TDF_EXITING 0x00400000 /* thread exiting */
369 #define TDF_USINGFP 0x00800000 /* thread using fp coproc */
370 #define TDF_KERNELFP 0x01000000 /* kernel using fp coproc */
371 #define TDF_DELAYED_WAKEUP 0x02000000
372 #define TDF_UNUSED1 0x04000000 /* unused */
373 #define TDF_USERMODE 0x08000000 /* in or entering user mode */
375 #define TDF_MP_STOPREQ 0x00000001 /* suspend_kproc */
376 #define TDF_MP_WAKEREQ 0x00000002 /* resume_kproc */
377 #define TDF_MP_EXITWAIT 0x00000004 /* reaper, see lwp_wait() */
378 #define TDF_MP_EXITSIG 0x00000008 /* reaper, see lwp_wait() */
379 #define TDF_MP_BATCH_DEMARC 0x00000010 /* batch mode handling */
381 #define TD_TYPE_GENERIC 0 /* generic thread */
382 #define TD_TYPE_CRYPTO 1 /* crypto thread */
383 #define TD_TYPE_NETISR 2 /* netisr thread */
386 * Thread priorities. Typically only one thread from any given
387 * user process scheduling queue is on the LWKT run queue at a time.
388 * Remember that there is one LWKT run queue per cpu.
390 * Critical sections are handled by bumping td_pri above TDPRI_MAX, which
391 * causes interrupts to be masked as they occur. When this occurs a
392 * rollup flag will be set in mycpu->gd_reqflags.
394 #define TDPRI_IDLE_THREAD 0 /* the idle thread */
395 #define TDPRI_IDLE_WORK 1 /* idle work (page zero, etc) */
396 #define TDPRI_USER_SCHEDULER 2 /* user scheduler helper */
397 #define TDPRI_USER_IDLE 4 /* user scheduler idle */
398 #define TDPRI_USER_NORM 6 /* user scheduler normal */
399 #define TDPRI_USER_REAL 8 /* user scheduler real time */
400 #define TDPRI_KERN_LPSCHED 9 /* scheduler helper for userland sch */
401 #define TDPRI_KERN_USER 10 /* kernel / block in syscall */
402 #define TDPRI_KERN_DAEMON 12 /* kernel daemon (pageout, etc) */
403 #define TDPRI_SOFT_NORM 14 /* kernel / normal */
404 #define TDPRI_SOFT_TIMER 16 /* kernel / timer */
405 #define TDPRI_EXITING 19 /* exiting thread */
406 #define TDPRI_INT_SUPPORT 20 /* kernel / high priority support */
407 #define TDPRI_INT_LOW 27 /* low priority interrupt */
408 #define TDPRI_INT_MED 28 /* medium priority interrupt */
409 #define TDPRI_INT_HIGH 29 /* high priority interrupt */
412 #define LWKT_THREAD_STACK (UPAGES * PAGE_SIZE)
414 #define IN_CRITICAL_SECT(td) ((td)->td_critcount)
421 extern struct lwkt_token mp_token;
422 extern struct lwkt_token pmap_token;
423 extern struct lwkt_token dev_token;
424 extern struct lwkt_token vm_token;
425 extern struct lwkt_token vmspace_token;
426 extern struct lwkt_token kvm_token;
427 extern struct lwkt_token proc_token;
428 extern struct lwkt_token tty_token;
429 extern struct lwkt_token vnode_token;
430 extern struct lwkt_token vmobj_token;
435 extern void lwkt_init(void);
436 extern struct thread *lwkt_alloc_thread(struct thread *, int, int, int);
437 extern void lwkt_init_thread(struct thread *, void *, int, int,
438 struct globaldata *);
439 extern void lwkt_set_interrupt_support_thread(void);
440 extern void lwkt_set_comm(thread_t, const char *, ...) __printflike(2, 3);
441 extern void lwkt_free_thread(struct thread *);
442 extern void lwkt_gdinit(struct globaldata *);
443 extern void lwkt_switch(void);
444 extern void lwkt_switch_return(struct thread *);
445 extern void lwkt_preempt(thread_t, int);
446 extern void lwkt_schedule(thread_t);
447 extern void lwkt_schedule_noresched(thread_t);
448 extern void lwkt_schedule_self(thread_t);
449 extern void lwkt_deschedule(thread_t);
450 extern void lwkt_deschedule_self(thread_t);
451 extern void lwkt_yield(void);
452 extern void lwkt_yield_quick(void);
453 extern void lwkt_user_yield(void);
454 extern void lwkt_token_wait(void);
455 extern void lwkt_hold(thread_t);
456 extern void lwkt_rele(thread_t);
457 extern void lwkt_passive_release(thread_t);
458 extern void lwkt_maybe_splz(thread_t);
460 extern void lwkt_gettoken(lwkt_token_t);
461 extern void lwkt_gettoken_shared(lwkt_token_t);
462 extern void lwkt_gettoken_hard(lwkt_token_t);
463 extern int lwkt_trytoken(lwkt_token_t);
464 extern void lwkt_reltoken(lwkt_token_t);
465 extern void lwkt_reltoken_hard(lwkt_token_t);
466 extern int lwkt_cnttoken(lwkt_token_t, thread_t);
467 extern int lwkt_getalltokens(thread_t, int);
468 extern void lwkt_relalltokens(thread_t);
469 extern void lwkt_drain_token_requests(void);
470 extern void lwkt_token_init(lwkt_token_t, const char *);
471 extern void lwkt_token_uninit(lwkt_token_t);
473 extern void lwkt_token_pool_init(void);
474 extern lwkt_token_t lwkt_token_pool_lookup(void *);
475 extern lwkt_token_t lwkt_getpooltoken(void *);
476 extern void lwkt_relpooltoken(void *);
478 extern void lwkt_token_swap(void);
480 extern void lwkt_setpri(thread_t, int);
481 extern void lwkt_setpri_initial(thread_t, int);
482 extern void lwkt_setpri_self(int);
483 extern void lwkt_schedulerclock(thread_t td);
484 extern void lwkt_setcpu_self(struct globaldata *);
485 extern void lwkt_migratecpu(int);
487 extern void lwkt_giveaway(struct thread *);
488 extern void lwkt_acquire(struct thread *);
489 extern int lwkt_send_ipiq3(struct globaldata *, ipifunc3_t, void *, int);
490 extern int lwkt_send_ipiq3_passive(struct globaldata *, ipifunc3_t,
492 extern int lwkt_send_ipiq3_nowait(struct globaldata *, ipifunc3_t,
494 extern int lwkt_send_ipiq3_bycpu(int, ipifunc3_t, void *, int);
495 extern int lwkt_send_ipiq3_mask(cpumask_t, ipifunc3_t, void *, int);
496 extern void lwkt_wait_ipiq(struct globaldata *, int);
497 extern int lwkt_seq_ipiq(struct globaldata *);
498 extern void lwkt_process_ipiq(void);
499 extern void lwkt_process_ipiq_frame(struct intrframe *);
500 extern void lwkt_smp_stopped(void);
501 extern void lwkt_synchronize_ipiqs(const char *);
503 /* lwkt_cpusync_init() - inline function in sys/thread2.h */
504 extern void lwkt_cpusync_simple(cpumask_t, cpusync_func_t, void *);
505 extern void lwkt_cpusync_interlock(lwkt_cpusync_t);
506 extern void lwkt_cpusync_deinterlock(lwkt_cpusync_t);
508 extern void crit_panic(void) __dead2;
509 extern struct lwp *lwkt_preempted_proc(void);
511 extern int lwkt_create (void (*func)(void *), void *, struct thread **,
512 struct thread *, int, int,
513 const char *, ...) __printflike(7, 8);
514 extern void lwkt_exit (void) __dead2;
515 extern void lwkt_remove_tdallq (struct thread *);