4 * Implements the architecture independant portion of the LWKT
12 #include <sys/param.h> /* MAXCOMLEN */
15 #include <sys/queue.h> /* TAILQ_* macros */
17 #ifndef _SYS_MSGPORT_H_
18 #include <sys/msgport.h> /* lwkt_port */
21 #include <sys/time.h> /* struct timeval */
26 #ifndef _SYS_SPINLOCK_H_
27 #include <sys/spinlock.h>
29 #ifndef _SYS_IOSCHED_H_
30 #include <sys/iosched.h>
32 #include <machine/thread.h> /* md_thread */
33 #include <machine/stdint.h>
51 typedef struct lwkt_queue *lwkt_queue_t;
52 typedef struct lwkt_token *lwkt_token_t;
53 typedef struct lwkt_tokref *lwkt_tokref_t;
55 typedef struct lwkt_cpu_msg *lwkt_cpu_msg_t;
56 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 _SYS_CPUMASK_H_
73 #include <sys/cpumask.h> /* cpumask_t */
76 #include <machine/frame.h>
83 * Tokens are used to serialize access to information. They are 'soft'
84 * serialization entities that only stay in effect while a thread is
85 * running. If the thread blocks, other threads can run holding the same
86 * token(s). The tokens are reacquired when the original thread resumes.
88 * Tokens guarantee that no deadlock can happen regardless of type or
89 * ordering. However, obtaining the same token first shared, then
90 * stacking exclusive, is not allowed and will panic.
92 * A thread can depend on its serialization remaining intact through a
93 * preemption. An interrupt which attempts to use the same token as the
94 * thread being preempted will reschedule itself for non-preemptive
95 * operation, so the new token code is capable of interlocking against
96 * interrupts as well as other cpus. This means that your token can only
97 * be (temporarily) lost if you *explicitly* block.
99 * Tokens are managed through a helper reference structure, lwkt_tokref. Each
100 * thread has a stack of tokref's to keep track of acquired tokens. Multiple
101 * tokref's may reference the same token.
104 * Acquiring an exclusive token requires acquiring the EXCLUSIVE bit
105 * with count == 0. If the exclusive bit cannot be acquired, EXCLREQ
106 * is set. Once acquired, EXCLREQ is cleared (but could get set by
107 * another thread also trying for an exclusive lock at any time).
110 * Acquiring a shared token requires waiting for the EXCLUSIVE bit
111 * to be cleared and then acquiring a count. A shared lock request
112 * can temporarily acquire a count and then back it out if it is
113 * unable to obtain the EXCLUSIVE bit, allowing fetchadd to be used.
115 * A thread attempting to get a single shared token will defer to
116 * pending exclusive requesters. However, a thread already holding
117 * one or more tokens and trying to get an additional shared token
118 * cannot defer to exclusive requesters because doing so can lead
121 * Multiple exclusive tokens are handled by treating the additional tokens
122 * as a special case of the shared token, incrementing the count value. This
123 * reduces the complexity of the token release code.
127 long t_count; /* Shared/exclreq/exclusive access */
128 struct lwkt_tokref *t_ref; /* Exclusive ref */
129 long t_collisions; /* Collision counter */
130 const char *t_desc; /* Descriptive name */
133 #define TOK_EXCLUSIVE 0x00000001 /* Exclusive lock held */
134 #define TOK_EXCLREQ 0x00000002 /* Exclusive request pending */
135 #define TOK_INCR 4 /* Shared count increment */
136 #define TOK_COUNTMASK (~(long)(TOK_EXCLUSIVE|TOK_EXCLREQ))
139 * Static initialization for a lwkt_token.
141 #define LWKT_TOKEN_INITIALIZER(name) \
150 * Assert that a particular token is held
152 #define LWKT_TOKEN_HELD_ANY(tok) _lwkt_token_held_any(tok, curthread)
153 #define LWKT_TOKEN_HELD_EXCL(tok) _lwkt_token_held_excl(tok, curthread)
155 #define ASSERT_LWKT_TOKEN_HELD(tok) \
156 KKASSERT(LWKT_TOKEN_HELD_ANY(tok))
158 #define ASSERT_LWKT_TOKEN_HELD_EXCL(tok) \
159 KKASSERT(LWKT_TOKEN_HELD_EXCL(tok))
161 #define ASSERT_NO_TOKENS_HELD(td) \
162 KKASSERT((td)->td_toks_stop == &td->td_toks_array[0])
165 lwkt_token_t tr_tok; /* token in question */
166 long tr_count; /* TOK_EXCLUSIVE|TOK_EXCLREQ or 0 */
167 struct thread *tr_owner; /* me */
170 #define MAXCPUFIFO 256 /* power of 2 */
171 #define MAXCPUFIFO_MASK (MAXCPUFIFO - 1)
172 #define LWKT_MAXTOKENS 32 /* max tokens beneficially held by thread */
174 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
176 * Always cast to ipifunc_t when registering an ipi. The actual ipi function
177 * is called with both the data and an interrupt frame, but the ipi function
178 * that is registered might only declare a data argument.
180 typedef void (*ipifunc1_t)(void *arg);
181 typedef void (*ipifunc2_t)(void *arg, int arg2);
182 typedef void (*ipifunc3_t)(void *arg, int arg2, struct intrframe *frame);
185 int ip_rindex; /* only written by target cpu */
186 int ip_xindex; /* written by target, indicates completion */
187 int ip_windex; /* only written by source cpu */
188 int ip_drain; /* drain source limit */
193 char filler[32 - sizeof(int) - sizeof(void *) * 2];
194 } ip_info[MAXCPUFIFO];
198 * CPU Synchronization structure. See lwkt_cpusync_init() and
199 * lwkt_cpusync_interlock() for more information.
201 typedef void (*cpusync_func_t)(void *arg);
203 struct lwkt_cpusync {
204 cpumask_t cs_mask; /* cpus running the sync */
205 cpumask_t cs_mack; /* mask acknowledge */
206 cpusync_func_t cs_func; /* function to execute */
207 void *cs_data; /* function data */
209 #endif /* _KERNEL || _KERNEL_STRUCTURES */
212 * The standard message and queue structure used for communications between
213 * cpus. Messages are typically queued via a machine-specific non-linked
214 * FIFO matrix allowing any cpu to send a message to any other cpu without
218 typedef struct lwkt_cpu_msg {
219 void (*cm_func)(lwkt_cpu_msg_t msg); /* primary dispatch function */
220 int cm_code; /* request code if applicable */
221 int cm_cpu; /* reply to cpu */
222 thread_t cm_originator; /* originating thread for wakeup */
227 * per-thread file descriptor cache
230 int fd; /* descriptor being cached */
232 struct file *fp; /* cached referenced fp */
237 #define NFDCACHE 4 /* max fd's cached by a thread */
240 * Thread structure. Note that ownership of a thread structure is special
241 * cased and there is no 'token'. A thread is always owned by the cpu
242 * represented by td_gd, any manipulation of the thread by some other cpu
243 * must be done through cpu_*msg() functions. e.g. you could request
244 * ownership of a thread that way, or hand a thread off to another cpu.
246 * NOTE: td_ucred is synchronized from the p_ucred on user->kernel syscall,
247 * trap, and AST/signal transitions to provide a stable ucred for
248 * (primarily) system calls. This field will be NULL for pure kernel
254 TAILQ_ENTRY(thread) td_threadq;
255 TAILQ_ENTRY(thread) td_allq;
256 TAILQ_ENTRY(thread) td_sleepq;
257 lwkt_port td_msgport; /* built-in message port for replies */
258 struct lwp *td_lwp; /* (optional) associated lwp */
259 struct proc *td_proc; /* (optional) associated process */
260 struct pcb *td_pcb; /* points to pcb and top of kstack */
261 struct globaldata *td_gd; /* associated with this cpu */
262 const char *td_wmesg; /* string name for blockage */
263 const volatile void *td_wchan; /* waiting on channel */
264 int td_pri; /* 0-31, 31=highest priority (note 1) */
265 int td_critcount; /* critical section priority */
266 u_int td_flags; /* TDF flags */
267 int td_wdomain; /* domain for wchan address (typ 0) */
268 void (*td_preemptable)(struct thread *td, int critcount);
269 void (*td_release)(struct thread *td);
270 char *td_kstack; /* kernel stack */
271 int td_kstack_size; /* size of kernel stack */
272 char *td_sp; /* kernel stack pointer for LWKT restore */
273 thread_t (*td_switch)(struct thread *ntd);
274 __uint64_t td_uticks; /* Statclock hits in user mode (uS) */
275 __uint64_t td_sticks; /* Statclock hits in system mode (uS) */
276 __uint64_t td_iticks; /* Statclock hits processing intr (uS) */
277 int td_locks; /* lockmgr lock debugging */
278 struct plimit *td_limit; /* synchronized from proc->p_limit */
279 int td_refs; /* hold position in gd_tdallq / hold free */
280 int td_nest_count; /* prevent splz nesting */
281 u_int td_contended; /* token contention count */
282 u_int td_mpflags; /* flags can be set by foreign cpus */
283 int td_cscount; /* cpu synchronization master */
284 int td_wakefromcpu; /* who woke me up? */
285 int td_upri; /* user priority (sub-priority under td_pri) */
286 int td_type; /* thread type, TD_TYPE_ */
287 int td_tracker; /* misc use (base value 0), recursion count */
289 int td_unused03[3]; /* for future fields */
290 struct iosched_data td_iosdata; /* Dynamic I/O scheduling data */
291 struct timeval td_start; /* start time for a thread/process */
292 char td_comm[MAXCOMLEN+1]; /* typ 16+1 bytes */
293 struct thread *td_preempted; /* we preempted this thread */
294 struct ucred *td_ucred; /* synchronized from proc->p_ucred */
295 void *td_vmm; /* vmm private data */
296 lwkt_tokref_t td_toks_have; /* tokens we own */
297 lwkt_tokref_t td_toks_stop; /* tokens we want */
298 struct lwkt_tokref td_toks_array[LWKT_MAXTOKENS];
299 int td_fairq_load; /* fairq */
300 int td_fairq_count; /* fairq */
301 struct globaldata *td_migrate_gd; /* target gd for thread migration */
302 struct fdcache td_fdcache[NFDCACHE];
303 void *td_linux_task; /* drm/linux support */
304 #ifdef DEBUG_CRIT_SECTIONS
305 #define CRIT_DEBUG_ARRAY_SIZE 32
306 #define CRIT_DEBUG_ARRAY_MASK (CRIT_DEBUG_ARRAY_SIZE - 1)
307 const char *td_crit_debug_array[CRIT_DEBUG_ARRAY_SIZE];
308 int td_crit_debug_index;
309 int td_in_crit_report;
311 struct md_thread td_mach;
313 #define SPINLOCK_DEBUG_ARRAY_SIZE 32
314 int td_spinlock_stack_id[SPINLOCK_DEBUG_ARRAY_SIZE];
315 struct spinlock *td_spinlock_stack[SPINLOCK_DEBUG_ARRAY_SIZE];
316 void *td_spinlock_caller_pc[SPINLOCK_DEBUG_ARRAY_SIZE];
319 * Track lockmgr locks held; lk->lk_filename:lk->lk_lineno is the holder
321 #define LOCKMGR_DEBUG_ARRAY_SIZE 8
322 int td_lockmgr_stack_id[LOCKMGR_DEBUG_ARRAY_SIZE];
323 struct lock *td_lockmgr_stack[LOCKMGR_DEBUG_ARRAY_SIZE];
327 #define td_toks_base td_toks_array[0]
328 #define td_toks_end td_toks_array[LWKT_MAXTOKENS]
330 #define TD_TOKS_HELD(td) ((td)->td_toks_stop != &(td)->td_toks_base)
331 #define TD_TOKS_NOT_HELD(td) ((td)->td_toks_stop == &(td)->td_toks_base)
334 * Thread flags. Note that TDF_RUNNING is cleared on the old thread after
335 * we switch to the new one, which is necessary because LWKTs don't need
336 * to hold the BGL. This flag is used by the exit code and the managed
337 * thread migration code. Note in addition that preemption will cause
338 * TDF_RUNNING to be cleared temporarily, so any code checking TDF_RUNNING
339 * must also check TDF_PREEMPT_LOCK.
341 * LWKT threads stay on their (per-cpu) run queue while running, not to
342 * be confused with user processes which are removed from the user scheduling
343 * run queue while actually running.
345 * td_threadq can represent the thread on one of three queues... the LWKT
346 * run queue, a tsleep queue, or an lwkt blocking queue. The LWKT subsystem
347 * does not allow a thread to be scheduled if it already resides on some
350 #define TDF_RUNNING 0x00000001 /* thread still active */
351 #define TDF_RUNQ 0x00000002 /* on an LWKT run queue */
352 #define TDF_PREEMPT_LOCK 0x00000004 /* I have been preempted */
353 #define TDF_PREEMPT_DONE 0x00000008 /* ac preemption complete */
354 #define TDF_NOSTART 0x00000010 /* do not schedule on create */
355 #define TDF_MIGRATING 0x00000020 /* thread is being migrated */
356 #define TDF_SINTR 0x00000040 /* interruptability for 'ps' */
357 #define TDF_TSLEEPQ 0x00000080 /* on a tsleep wait queue */
359 #define TDF_SYSTHREAD 0x00000100 /* reserve memory may be used */
360 #define TDF_ALLOCATED_THREAD 0x00000200 /* objcache allocated thread */
361 #define TDF_ALLOCATED_STACK 0x00000400 /* objcache allocated stack */
362 #define TDF_FPU_HEUR 0x00000800 /* active restore on switch */
363 #define TDF_DEADLKTREAT 0x00001000 /* special lockmgr treatment */
364 #define TDF_MARKER 0x00002000 /* tdallq list scan marker */
365 #define TDF_TIMEOUT_RUNNING 0x00004000 /* tsleep timeout race */
366 #define TDF_TIMEOUT 0x00008000 /* tsleep timeout */
367 #define TDF_INTTHREAD 0x00010000 /* interrupt thread */
368 #define TDF_TSLEEP_DESCHEDULED 0x00020000 /* tsleep core deschedule */
369 #define TDF_BLOCKED 0x00040000 /* Thread is blocked */
370 #define TDF_PANICWARN 0x00080000 /* panic warning in switch */
371 #define TDF_BLOCKQ 0x00100000 /* on block queue */
372 #define TDF_FORCE_SPINPORT 0x00200000
373 #define TDF_EXITING 0x00400000 /* thread exiting */
374 #define TDF_USINGFP 0x00800000 /* thread using fp coproc */
375 #define TDF_KERNELFP 0x01000000 /* kernel using fp coproc */
376 #define TDF_DELAYED_WAKEUP 0x02000000
377 #define TDF_FIXEDCPU 0x04000000 /* running cpu is fixed */
378 #define TDF_USERMODE 0x08000000 /* in or entering user mode */
379 #define TDF_NOFAULT 0x10000000 /* force onfault on fault */
380 #define TDF_CLKTHREAD 0x20000000 /* detect INTTHREAD clock */
382 #define TDF_MP_STOPREQ 0x00000001 /* suspend_kproc */
383 #define TDF_MP_WAKEREQ 0x00000002 /* resume_kproc */
384 #define TDF_MP_EXITWAIT 0x00000004 /* reaper, see lwp_wait() */
385 #define TDF_MP_EXITSIG 0x00000008 /* reaper, see lwp_wait() */
386 #define TDF_MP_BATCH_DEMARC 0x00000010 /* batch mode handling */
387 #define TDF_MP_DIDYIELD 0x00000020 /* effects scheduling */
389 #define TD_TYPE_GENERIC 0 /* generic thread */
390 #define TD_TYPE_CRYPTO 1 /* crypto thread */
391 #define TD_TYPE_NETISR 2 /* netisr thread */
394 * Thread priorities. Typically only one thread from any given
395 * user process scheduling queue is on the LWKT run queue at a time.
396 * Remember that there is one LWKT run queue per cpu.
398 * Critical sections are handled by bumping td_pri above TDPRI_MAX, which
399 * causes interrupts to be masked as they occur. When this occurs a
400 * rollup flag will be set in mycpu->gd_reqflags.
402 #define TDPRI_IDLE_THREAD 0 /* the idle thread */
403 #define TDPRI_IDLE_WORK 1 /* idle work (page zero, etc) */
404 #define TDPRI_USER_SCHEDULER 2 /* user scheduler helper */
405 #define TDPRI_USER_IDLE 4 /* user scheduler idle */
406 #define TDPRI_USER_NORM 6 /* user scheduler normal */
407 #define TDPRI_USER_REAL 8 /* user scheduler real time */
408 #define TDPRI_KERN_LPSCHED 9 /* (comparison point only) */
409 #define TDPRI_KERN_USER 10 /* kernel / block in syscall */
410 #define TDPRI_KERN_DAEMON 12 /* kernel daemon (pageout, etc) */
411 #define TDPRI_SOFT_NORM 14 /* kernel / normal */
412 #define TDPRI_SOFT_TIMER 16 /* kernel / timer */
413 #define TDPRI_UNUSED19 19
414 #define TDPRI_INT_SUPPORT 20 /* kernel / high priority support */
415 #define TDPRI_INT_LOW 27 /* low priority interrupt */
416 #define TDPRI_INT_MED 28 /* medium priority interrupt */
417 #define TDPRI_INT_HIGH 29 /* high priority interrupt */
420 #define LWKT_THREAD_STACK (UPAGES * PAGE_SIZE)
422 #define IN_CRITICAL_SECT(td) ((td)->td_critcount)
426 extern void (*linux_task_drop_callback)(struct thread *);
427 extern void (*linux_proc_drop_callback)(struct proc *);
432 extern struct lwkt_token mp_token;
433 extern struct lwkt_token pmap_token;
434 extern struct lwkt_token dev_token;
435 extern struct lwkt_token vm_token;
436 extern struct lwkt_token vmspace_token;
437 extern struct lwkt_token kvm_token;
438 extern struct lwkt_token sigio_token;
439 extern struct lwkt_token tty_token;
440 extern struct lwkt_token vnode_token;
441 extern struct lwkt_token revoke_token;
442 extern struct lwkt_token kbd_token;
443 extern struct lwkt_token vga_token;
448 struct thread *lwkt_alloc_thread(struct thread *, int, int, int);
449 void lwkt_init_thread(struct thread *, void *, int, int, struct globaldata *);
450 void lwkt_set_interrupt_support_thread(void);
451 void lwkt_set_comm(thread_t, const char *, ...) __printflike(2, 3);
452 void lwkt_free_thread(struct thread *);
453 void lwkt_gdinit(struct globaldata *);
454 void lwkt_switch(void);
455 void lwkt_switch_return(struct thread *);
456 void lwkt_preempt(thread_t, int);
457 void lwkt_schedule(thread_t);
458 void lwkt_schedule_noresched(thread_t);
459 void lwkt_schedule_self(thread_t);
460 void lwkt_deschedule(thread_t);
461 void lwkt_deschedule_self(thread_t);
462 void lwkt_yield(void);
463 void lwkt_yield_quick(void);
464 void lwkt_user_yield(void);
465 void lwkt_hold(thread_t);
466 void lwkt_rele(thread_t);
467 void lwkt_passive_release(thread_t);
468 void lwkt_maybe_splz(thread_t);
470 void lwkt_gettoken(lwkt_token_t);
471 void lwkt_gettoken_shared(lwkt_token_t);
472 int lwkt_trytoken(lwkt_token_t);
473 void lwkt_reltoken(lwkt_token_t);
474 int lwkt_cnttoken(lwkt_token_t, thread_t);
475 int lwkt_getalltokens(thread_t, int);
476 void lwkt_relalltokens(thread_t);
477 void lwkt_token_init(lwkt_token_t, const char *);
478 void lwkt_token_uninit(lwkt_token_t);
480 void lwkt_token_pool_init(void);
481 lwkt_token_t lwkt_token_pool_lookup(void *);
482 lwkt_token_t lwkt_getpooltoken(void *);
483 void lwkt_relpooltoken(void *);
485 void lwkt_token_swap(void);
487 void lwkt_setpri(thread_t, int);
488 void lwkt_setpri_initial(thread_t, int);
489 void lwkt_setpri_self(int);
490 void lwkt_schedulerclock(thread_t td);
491 void lwkt_setcpu_self(struct globaldata *);
492 void lwkt_migratecpu(int);
494 void lwkt_giveaway(struct thread *);
495 void lwkt_acquire(struct thread *);
496 int lwkt_send_ipiq3(struct globaldata *, ipifunc3_t, void *, int);
497 int lwkt_send_ipiq3_passive(struct globaldata *, ipifunc3_t, void *, int);
498 int lwkt_send_ipiq3_bycpu(int, ipifunc3_t, void *, int);
499 int lwkt_send_ipiq3_mask(cpumask_t, ipifunc3_t, void *, int);
500 void lwkt_wait_ipiq(struct globaldata *, int);
501 void lwkt_process_ipiq(void);
502 void lwkt_process_ipiq_frame(struct intrframe *);
503 void lwkt_smp_stopped(void);
504 void lwkt_synchronize_ipiqs(const char *);
506 /* lwkt_cpusync_init() - inline function in sys/thread2.h */
507 void lwkt_cpusync_simple(cpumask_t, cpusync_func_t, void *);
508 void lwkt_cpusync_interlock(lwkt_cpusync_t);
509 void lwkt_cpusync_deinterlock(lwkt_cpusync_t);
510 void lwkt_cpusync_quick(lwkt_cpusync_t);
512 void crit_panic(void) __dead2;
513 struct lwp *lwkt_preempted_proc(void);
515 int lwkt_create(void (*)(void *), void *, struct thread **, struct thread *,
516 int, int, const char *, ...) __printflike(7, 8);
517 void lwkt_exit(void) __dead2;
518 void lwkt_remove_tdallq(struct thread *);
522 #endif /* !_SYS_THREAD_H_ */