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
10 * $DragonFly: src/sys/sys/thread.h,v 1.97 2008/09/20 04:31:02 sephe Exp $
13 #ifndef _SYS_THREAD_H_
14 #define _SYS_THREAD_H_
16 #ifndef _SYS_STDINT_H_
17 #include <sys/stdint.h> /* __int types */
20 #include <sys/param.h> /* MAXCOMLEN */
23 #include <sys/queue.h> /* TAILQ_* macros */
25 #ifndef _SYS_MSGPORT_H_
26 #include <sys/msgport.h> /* lwkt_port */
29 #include <sys/time.h> /* struct timeval */
31 #ifndef _SYS_SPINLOCK_H_
32 #include <sys/spinlock.h>
34 #ifndef _SYS_IOSCHED_H_
35 #include <sys/iosched.h>
37 #ifndef _MACHINE_THREAD_H_
38 #include <machine/thread.h>
56 typedef struct lwkt_queue *lwkt_queue_t;
57 typedef struct lwkt_token *lwkt_token_t;
58 typedef struct lwkt_tokref *lwkt_tokref_t;
59 typedef struct lwkt_cpu_msg *lwkt_cpu_msg_t;
60 typedef struct lwkt_cpu_port *lwkt_cpu_port_t;
61 typedef struct lwkt_ipiq *lwkt_ipiq_t;
62 typedef struct lwkt_cpusync *lwkt_cpusync_t;
63 typedef struct thread *thread_t;
65 typedef TAILQ_HEAD(lwkt_queue, thread) lwkt_queue;
68 * Differentiation between kernel threads and user threads. Userland
69 * programs which want to access to kernel structures have to define
70 * _KERNEL_STRUCTURES. This is a kinda safety valve to prevent badly
71 * written user programs from getting an LWKT thread that is neither the
72 * kernel nor the user version.
74 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
75 #ifndef _MACHINE_THREAD_H_
76 #include <machine/thread.h> /* md_thread */
78 #ifndef _MACHINE_FRAME_H_
79 #include <machine/frame.h>
86 * Tokens are used to serialize access to information. They are 'soft'
87 * serialization entities that only stay in effect while a thread is
88 * running. If the thread blocks, other threads can run holding the same
89 * token(s). The tokens are reacquired when the original thread resumes.
91 * A thread can depend on its serialization remaining intact through a
92 * preemption. An interrupt which attempts to use the same token as the
93 * thread being preempted will reschedule itself for non-preemptive
94 * operation, so the new token code is capable of interlocking against
95 * interrupts as well as other cpus. This means that your token can only
96 * be (temporarily) lost if you *explicitly* block.
98 * Tokens are managed through a helper reference structure, lwkt_tokref,
99 * which is typically declared on the caller's stack. Multiple tokref's
100 * may reference the same token.
102 * It is possible to detect that your token was temporarily lost via
103 * lwkt_token_is_stale(), which uses the t_lastowner field. This field
104 * does NOT necessarily represent the current owner and can become stale
105 * (not point to a valid structure). It is used solely to detect
106 * whether the token was temporarily lost to another thread. The lost
107 * state is cleared by the function.
110 typedef struct lwkt_token {
112 struct spinlock t_spinlock; /* Controls access */
114 struct spinlock t_unused01;
116 struct thread *t_owner; /* The current owner of the token */
117 int t_count; /* Per-thread count */
118 struct thread *t_lastowner; /* Last owner that acquired token */
122 #define LWKT_TOKEN_INITIALIZER(head) \
124 .t_spinlock = SPINLOCK_INITIALIZER(head.t_spinlock), \
126 .t_lastowner = NULL, \
130 #define LWKT_TOKEN_INITIALIZER(head) \
133 .t_lastowner = NULL, \
138 #define ASSERT_LWKT_TOKEN_HELD(token) \
139 KKASSERT((token)->t_owner == curthread)
141 typedef struct lwkt_tokref {
142 lwkt_token_t tr_tok; /* token in question */
143 lwkt_tokref_t tr_next; /* linked list */
144 int tr_state; /* 0 = don't have, 1 = have */
147 #define LWKT_TOKREF_INIT(tok) \
149 #define LWKT_TOKREF_DECLARE(name, tok) \
150 lwkt_tokref name = LWKT_TOKREF_INIT(tok)
152 #define MAXCPUFIFO 16 /* power of 2 */
153 #define MAXCPUFIFO_MASK (MAXCPUFIFO - 1)
154 #define LWKT_MAXTOKENS 16 /* max tokens beneficially held by thread */
157 * Always cast to ipifunc_t when registering an ipi. The actual ipi function
158 * is called with both the data and an interrupt frame, but the ipi function
159 * that is registered might only declare a data argument.
161 typedef void (*ipifunc1_t)(void *arg);
162 typedef void (*ipifunc2_t)(void *arg, int arg2);
163 typedef void (*ipifunc3_t)(void *arg, int arg2, struct intrframe *frame);
165 typedef struct lwkt_ipiq {
166 int ip_rindex; /* only written by target cpu */
167 int ip_xindex; /* written by target, indicates completion */
168 int ip_windex; /* only written by source cpu */
169 ipifunc3_t ip_func[MAXCPUFIFO];
170 void *ip_arg1[MAXCPUFIFO];
171 int ip_arg2[MAXCPUFIFO];
172 u_int ip_npoll; /* synchronization to avoid excess IPIs */
176 * CPU Synchronization structure. See lwkt_cpusync_start() and
177 * lwkt_cpusync_finish() for more information.
179 typedef void (*cpusync_func_t)(lwkt_cpusync_t poll);
180 typedef void (*cpusync_func2_t)(void *data);
182 struct lwkt_cpusync {
183 cpusync_func_t cs_run_func; /* run (tandem w/ acquire) */
184 cpusync_func_t cs_fin1_func; /* fin1 (synchronized) */
185 cpusync_func2_t cs_fin2_func; /* fin2 (tandem w/ release) */
188 volatile int cs_count;
193 * The standard message and queue structure used for communications between
194 * cpus. Messages are typically queued via a machine-specific non-linked
195 * FIFO matrix allowing any cpu to send a message to any other cpu without
198 typedef struct lwkt_cpu_msg {
199 void (*cm_func)(lwkt_cpu_msg_t msg); /* primary dispatch function */
200 int cm_code; /* request code if applicable */
201 int cm_cpu; /* reply to cpu */
202 thread_t cm_originator; /* originating thread for wakeup */
206 * Thread structure. Note that ownership of a thread structure is special
207 * cased and there is no 'token'. A thread is always owned by the cpu
208 * represented by td_gd, any manipulation of the thread by some other cpu
209 * must be done through cpu_*msg() functions. e.g. you could request
210 * ownership of a thread that way, or hand a thread off to another cpu.
212 * NOTE: td_pri is bumped by TDPRI_CRIT when entering a critical section,
213 * but this does not effect how the thread is scheduled by LWKT.
215 * NOTE: td_ucred is synchronized from the p_ucred on user->kernel syscall,
216 * trap, and AST/signal transitions to provide a stable ucred for
217 * (primarily) system calls. This field will be NULL for pure kernel
224 TAILQ_ENTRY(thread) td_threadq;
225 TAILQ_ENTRY(thread) td_allq;
226 TAILQ_ENTRY(thread) td_sleepq;
227 lwkt_port td_msgport; /* built-in message port for replies */
228 struct lwp *td_lwp; /* (optional) associated lwp */
229 struct proc *td_proc; /* (optional) associated process */
230 struct pcb *td_pcb; /* points to pcb and top of kstack */
231 struct globaldata *td_gd; /* associated with this cpu */
232 const char *td_wmesg; /* string name for blockage */
233 void *td_wchan; /* waiting on channel */
234 int td_pri; /* 0-31, 31=highest priority (note 1) */
235 int td_flags; /* TDF flags */
236 int td_wdomain; /* domain for wchan address (typ 0) */
237 void (*td_preemptable)(struct thread *td, int critpri);
238 void (*td_release)(struct thread *td);
239 char *td_kstack; /* kernel stack */
240 int td_kstack_size; /* size of kernel stack */
241 char *td_sp; /* kernel stack pointer for LWKT restore */
242 void (*td_switch)(struct thread *ntd);
243 __uint64_t td_uticks; /* Statclock hits in user mode (uS) */
244 __uint64_t td_sticks; /* Statclock hits in system mode (uS) */
245 __uint64_t td_iticks; /* Statclock hits processing intr (uS) */
246 int td_locks; /* lockmgr lock debugging */
248 int td_refs; /* hold position in gd_tdallq / hold free */
249 int td_nest_count; /* prevent splz nesting */
251 int td_mpcount; /* MP lock held (count) */
252 int td_cscount; /* cpu synchronization master */
254 int td_mpcount_unused; /* filler so size matches */
255 int td_cscount_unused;
257 struct iosched_data td_iosdata; /* Dynamic I/O scheduling data */
258 struct timeval td_start; /* start time for a thread/process */
259 char td_comm[MAXCOMLEN+1]; /* typ 16+1 bytes */
260 struct thread *td_preempted; /* we preempted this thread */
261 struct ucred *td_ucred; /* synchronized from p_ucred */
262 struct caps_kinfo *td_caps; /* list of client and server registrations */
263 lwkt_tokref_t td_toks; /* tokens beneficially held */
264 #ifdef DEBUG_CRIT_SECTIONS
265 #define CRIT_DEBUG_ARRAY_SIZE 32
266 #define CRIT_DEBUG_ARRAY_MASK (CRIT_DEBUG_ARRAY_SIZE - 1)
267 const char *td_crit_debug_array[CRIT_DEBUG_ARRAY_SIZE];
268 int td_crit_debug_index;
269 int td_in_crit_report;
271 struct md_thread td_mach;
275 * Thread flags. Note that TDF_RUNNING is cleared on the old thread after
276 * we switch to the new one, which is necessary because LWKTs don't need
277 * to hold the BGL. This flag is used by the exit code and the managed
278 * thread migration code. Note in addition that preemption will cause
279 * TDF_RUNNING to be cleared temporarily, so any code checking TDF_RUNNING
280 * must also check TDF_PREEMPT_LOCK.
282 * LWKT threads stay on their (per-cpu) run queue while running, not to
283 * be confused with user processes which are removed from the user scheduling
284 * run queue while actually running.
286 * td_threadq can represent the thread on one of three queues... the LWKT
287 * run queue, a tsleep queue, or an lwkt blocking queue. The LWKT subsystem
288 * does not allow a thread to be scheduled if it already resides on some
291 #define TDF_RUNNING 0x0001 /* thread still active */
292 #define TDF_RUNQ 0x0002 /* on an LWKT run queue */
293 #define TDF_PREEMPT_LOCK 0x0004 /* I have been preempted */
294 #define TDF_PREEMPT_DONE 0x0008 /* acknowledge preemption complete */
295 #define TDF_IDLE_NOHLT 0x0010 /* we need to spin */
296 #define TDF_MIGRATING 0x0020 /* thread is being migrated */
297 #define TDF_SINTR 0x0040 /* interruptability hint for 'ps' */
298 #define TDF_TSLEEPQ 0x0080 /* on a tsleep wait queue */
300 #define TDF_SYSTHREAD 0x0100 /* allocations may use reserve */
301 #define TDF_ALLOCATED_THREAD 0x0200 /* objcache allocated thread */
302 #define TDF_ALLOCATED_STACK 0x0400 /* objcache allocated stack */
303 #define TDF_VERBOSE 0x0800 /* verbose on exit */
304 #define TDF_DEADLKTREAT 0x1000 /* special lockmgr deadlock treatment */
305 #define TDF_STOPREQ 0x2000 /* suspend_kproc */
306 #define TDF_WAKEREQ 0x4000 /* resume_kproc */
307 #define TDF_TIMEOUT 0x8000 /* tsleep timeout */
308 #define TDF_INTTHREAD 0x00010000 /* interrupt thread */
309 #define TDF_TSLEEP_DESCHEDULED 0x00020000 /* tsleep core deschedule */
310 #define TDF_BLOCKED 0x00040000 /* Thread is blocked */
311 #define TDF_PANICWARN 0x00080000 /* panic warning in switch */
312 #define TDF_BLOCKQ 0x00100000 /* on block queue */
313 #define TDF_MPSAFE 0x00200000 /* (thread creation) */
314 #define TDF_EXITING 0x00400000 /* thread exiting */
315 #define TDF_USINGFP 0x00800000 /* thread using fp coproc */
316 #define TDF_KERNELFP 0x01000000 /* kernel using fp coproc */
317 #define TDF_NETWORK 0x02000000 /* network proto thread */
320 * Thread priorities. Typically only one thread from any given
321 * user process scheduling queue is on the LWKT run queue at a time.
322 * Remember that there is one LWKT run queue per cpu.
324 * Critical sections are handled by bumping td_pri above TDPRI_MAX, which
325 * causes interrupts to be masked as they occur. When this occurs a
326 * rollup flag will be set in mycpu->gd_reqflags.
328 #define TDPRI_IDLE_THREAD 0 /* the idle thread */
329 #define TDPRI_USER_SCHEDULER 2 /* user scheduler helper */
330 #define TDPRI_USER_IDLE 4 /* user scheduler idle */
331 #define TDPRI_USER_NORM 6 /* user scheduler normal */
332 #define TDPRI_USER_REAL 8 /* user scheduler real time */
333 #define TDPRI_KERN_LPSCHED 9 /* scheduler helper for userland sch */
334 #define TDPRI_KERN_USER 10 /* kernel / block in syscall */
335 #define TDPRI_KERN_DAEMON 12 /* kernel daemon (pageout, etc) */
336 #define TDPRI_SOFT_NORM 14 /* kernel / normal */
337 #define TDPRI_SOFT_TIMER 16 /* kernel / timer */
338 #define TDPRI_EXITING 19 /* exiting thread */
339 #define TDPRI_INT_SUPPORT 20 /* kernel / high priority support */
340 #define TDPRI_INT_LOW 27 /* low priority interrupt */
341 #define TDPRI_INT_MED 28 /* medium priority interrupt */
342 #define TDPRI_INT_HIGH 29 /* high priority interrupt */
345 #define TDPRI_MASK 31
346 #define TDPRI_CRIT 32 /* high bits of td_pri used for crit */
349 #define LWKT_THREAD_STACK (UPAGES * PAGE_SIZE)
352 #define CACHE_NTHREADS 6
354 #define IN_CRITICAL_SECT(td) ((td)->td_pri >= TDPRI_CRIT)
356 extern void lwkt_init(void);
357 extern struct thread *lwkt_alloc_thread(struct thread *, int, int, int);
358 extern void lwkt_init_thread(struct thread *, void *, int, int,
359 struct globaldata *);
360 extern void lwkt_set_comm(thread_t, const char *, ...);
361 extern void lwkt_wait_free(struct thread *);
362 extern void lwkt_free_thread(struct thread *);
363 extern void lwkt_gdinit(struct globaldata *);
364 extern void lwkt_switch(void);
365 extern void lwkt_preempt(thread_t, int);
366 extern void lwkt_schedule(thread_t);
367 extern void lwkt_schedule_noresched(thread_t);
368 extern void lwkt_schedule_self(thread_t);
369 extern void lwkt_deschedule(thread_t);
370 extern void lwkt_deschedule_self(thread_t);
371 extern void lwkt_yield(void);
372 extern void lwkt_user_yield(void);
373 extern void lwkt_token_wait(void);
374 extern void lwkt_hold(thread_t);
375 extern void lwkt_rele(thread_t);
376 extern void lwkt_passive_release(thread_t);
378 extern void lwkt_gettoken(lwkt_tokref_t, lwkt_token_t);
379 extern int lwkt_trytoken(lwkt_tokref_t, lwkt_token_t);
380 extern void lwkt_gettokref(lwkt_tokref_t);
381 extern int lwkt_trytokref(lwkt_tokref_t);
382 extern void lwkt_reltoken(lwkt_tokref_t);
383 extern int lwkt_getalltokens(thread_t);
384 extern void lwkt_relalltokens(thread_t);
385 extern void lwkt_drain_token_requests(void);
386 extern void lwkt_token_init(lwkt_token_t);
387 extern void lwkt_token_uninit(lwkt_token_t);
388 extern int lwkt_token_is_stale(lwkt_tokref_t);
390 extern void lwkt_token_pool_init(void);
391 extern lwkt_token_t lwkt_token_pool_get(void *);
393 extern void lwkt_setpri(thread_t, int);
394 extern void lwkt_setpri_initial(thread_t, int);
395 extern void lwkt_setpri_self(int);
396 extern int lwkt_check_resched(thread_t);
397 extern void lwkt_setcpu_self(struct globaldata *);
398 extern void lwkt_migratecpu(int);
402 extern void lwkt_giveaway(struct thread *);
403 extern void lwkt_acquire(struct thread *);
404 extern int lwkt_send_ipiq3(struct globaldata *, ipifunc3_t, void *, int);
405 extern int lwkt_send_ipiq3_passive(struct globaldata *, ipifunc3_t,
407 extern int lwkt_send_ipiq3_nowait(struct globaldata *, ipifunc3_t,
409 extern int lwkt_send_ipiq3_bycpu(int, ipifunc3_t, void *, int);
410 extern int lwkt_send_ipiq3_mask(cpumask_t, ipifunc3_t, void *, int);
411 extern void lwkt_wait_ipiq(struct globaldata *, int);
412 extern int lwkt_seq_ipiq(struct globaldata *);
413 extern void lwkt_process_ipiq(void);
415 extern void lwkt_process_ipiq_frame(struct intrframe *);
417 extern void lwkt_smp_stopped(void);
418 extern void lwkt_synchronize_ipiqs(const char *);
422 extern void lwkt_cpusync_simple(cpumask_t, cpusync_func_t, void *);
423 extern void lwkt_cpusync_fastdata(cpumask_t, cpusync_func2_t, void *);
424 extern void lwkt_cpusync_start(cpumask_t, lwkt_cpusync_t);
425 extern void lwkt_cpusync_add(cpumask_t, lwkt_cpusync_t);
426 extern void lwkt_cpusync_finish(lwkt_cpusync_t);
428 extern void crit_panic(void);
429 extern struct lwp *lwkt_preempted_proc(void);
431 extern int lwkt_create (void (*func)(void *), void *, struct thread **,
432 struct thread *, int, int, const char *, ...);
433 extern void lwkt_exit (void) __dead2;
434 extern void lwkt_remove_tdallq (struct thread *);
435 extern void lwkt_mp_lock_contested(void);
436 extern void lwkt_mp_lock_uncontested(void);