kernel - lwkt_token revamp
[dragonfly.git] / sys / sys / thread.h
CommitLineData
2df9fa87
MD
1/*
2 * SYS/THREAD.H
3 *
4 * Implements the architecture independant portion of the LWKT
5 * subsystem.
05220613
MD
6 *
7 * Types which must already be defined when this header is included by
8 * userland: struct md_thread
2df9fa87 9 *
cbb1e737 10 * $DragonFly: src/sys/sys/thread.h,v 1.97 2008/09/20 04:31:02 sephe Exp $
2df9fa87
MD
11 */
12
13#ifndef _SYS_THREAD_H_
14#define _SYS_THREAD_H_
15
05220613
MD
16#ifndef _SYS_STDINT_H_
17#include <sys/stdint.h> /* __int types */
18#endif
19#ifndef _SYS_PARAM_H_
20#include <sys/param.h> /* MAXCOMLEN */
21#endif
8a8d5d85
MD
22#ifndef _SYS_QUEUE_H_
23#include <sys/queue.h> /* TAILQ_* macros */
24#endif
05220613
MD
25#ifndef _SYS_MSGPORT_H_
26#include <sys/msgport.h> /* lwkt_port */
27#endif
41f3429e
HP
28#ifndef _SYS_TIME_H_
29#include <sys/time.h> /* struct timeval */
30#endif
9d265729
MD
31#ifndef _SYS_SPINLOCK_H_
32#include <sys/spinlock.h>
33#endif
79eae878
MD
34#ifndef _SYS_IOSCHED_H_
35#include <sys/iosched.h>
36#endif
1bd40720
MD
37#ifndef _MACHINE_THREAD_H_
38#include <machine/thread.h>
39#endif
8a8d5d85 40
f1d1c3fa 41struct globaldata;
ef09c3ed 42struct lwp;
2df9fa87
MD
43struct proc;
44struct thread;
f1d1c3fa
MD
45struct lwkt_queue;
46struct lwkt_token;
41a01a4d 47struct lwkt_tokref;
96728c05 48struct lwkt_ipiq;
f1d1c3fa
MD
49struct lwkt_cpu_msg;
50struct lwkt_cpu_port;
ece04fd0
MD
51struct lwkt_msg;
52struct lwkt_port;
3b6b7bd1 53struct lwkt_cpusync;
4fd10eb6 54union sysunion;
f1d1c3fa
MD
55
56typedef struct lwkt_queue *lwkt_queue_t;
57typedef struct lwkt_token *lwkt_token_t;
41a01a4d 58typedef struct lwkt_tokref *lwkt_tokref_t;
f1d1c3fa
MD
59typedef struct lwkt_cpu_msg *lwkt_cpu_msg_t;
60typedef struct lwkt_cpu_port *lwkt_cpu_port_t;
96728c05 61typedef struct lwkt_ipiq *lwkt_ipiq_t;
3b6b7bd1 62typedef struct lwkt_cpusync *lwkt_cpusync_t;
f1d1c3fa
MD
63typedef struct thread *thread_t;
64
65typedef TAILQ_HEAD(lwkt_queue, thread) lwkt_queue;
f1d1c3fa 66
05220613
MD
67/*
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.
73 */
74#if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
85100692
MD
75#ifndef _MACHINE_THREAD_H_
76#include <machine/thread.h> /* md_thread */
77#endif
88c4d2f6
MD
78#ifndef _MACHINE_FRAME_H_
79#include <machine/frame.h>
80#endif
81#else
82struct intrframe;
ece04fd0 83#endif
f1d1c3fa
MD
84
85/*
41a01a4d 86 * Tokens are used to serialize access to information. They are 'soft'
dd55d707 87 * serialization entities that only stay in effect while a thread is
41a01a4d 88 * running. If the thread blocks, other threads can run holding the same
dd55d707 89 * token(s). The tokens are reacquired when the original thread resumes.
41a01a4d
MD
90 *
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
7eb611ef
MD
95 * interrupts as well as other cpus. This means that your token can only
96 * be (temporarily) lost if you *explicitly* block.
41a01a4d
MD
97 *
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.
f1d1c3fa 101 */
dd55d707 102
f1d1c3fa 103typedef struct lwkt_token {
c6fbe95a 104 struct lwkt_tokref *t_ref; /* Owning ref or NULL */
3b998fa9 105 intptr_t t_flags; /* MP lock required */
7eb611ef 106} lwkt_token;
dd55d707 107
3b998fa9
MD
108#define LWKT_TOKEN_MPSAFE 0x0001
109
110/*
111 * Static initialization for a lwkt_token.
112 * UP - Not MPSAFE (full MP lock will also be acquired)
113 * MP - Is MPSAFE (only the token will be acquired)
114 */
115#define LWKT_TOKEN_UP_INITIALIZER(head) \
c6fbe95a 116{ \
3b998fa9
MD
117 .t_ref = NULL, \
118 .t_flags = 0 \
119}
120
121#define LWKT_TOKEN_MP_INITIALIZER(head) \
122{ \
123 .t_ref = NULL, \
124 .t_flags = LWKT_TOKEN_MPSAFE \
0ca9b5e1 125}
0ca9b5e1 126
c6fbe95a
MD
127#define ASSERT_LWKT_TOKEN_HELD(tok) \
128 KKASSERT((tok)->t_ref->tr_owner == curthread)
8ed305a4 129
3b998fa9 130struct lwkt_tokref {
41a01a4d 131 lwkt_token_t tr_tok; /* token in question */
c6fbe95a 132 struct thread *tr_owner; /* me */
3b998fa9
MD
133 intptr_t tr_flags; /* copy of t_flags */
134};
41a01a4d 135
96728c05
MD
136#define MAXCPUFIFO 16 /* power of 2 */
137#define MAXCPUFIFO_MASK (MAXCPUFIFO - 1)
3b998fa9 138#define LWKT_MAXTOKENS 32 /* max tokens beneficially held by thread */
96728c05 139
88c4d2f6
MD
140/*
141 * Always cast to ipifunc_t when registering an ipi. The actual ipi function
142 * is called with both the data and an interrupt frame, but the ipi function
143 * that is registered might only declare a data argument.
144 */
b8a98473
MD
145typedef void (*ipifunc1_t)(void *arg);
146typedef void (*ipifunc2_t)(void *arg, int arg2);
147typedef void (*ipifunc3_t)(void *arg, int arg2, struct intrframe *frame);
96728c05
MD
148
149typedef struct lwkt_ipiq {
150 int ip_rindex; /* only written by target cpu */
166ec852 151 int ip_xindex; /* written by target, indicates completion */
96728c05 152 int ip_windex; /* only written by source cpu */
b8a98473
MD
153 ipifunc3_t ip_func[MAXCPUFIFO];
154 void *ip_arg1[MAXCPUFIFO];
155 int ip_arg2[MAXCPUFIFO];
4c9f5a7f 156 u_int ip_npoll; /* synchronization to avoid excess IPIs */
96728c05
MD
157} lwkt_ipiq;
158
f1d1c3fa 159/*
3b6b7bd1
MD
160 * CPU Synchronization structure. See lwkt_cpusync_start() and
161 * lwkt_cpusync_finish() for more information.
162 */
163typedef void (*cpusync_func_t)(lwkt_cpusync_t poll);
164typedef void (*cpusync_func2_t)(void *data);
165
166struct lwkt_cpusync {
167 cpusync_func_t cs_run_func; /* run (tandem w/ acquire) */
168 cpusync_func_t cs_fin1_func; /* fin1 (synchronized) */
169 cpusync_func2_t cs_fin2_func; /* fin2 (tandem w/ release) */
170 void *cs_data;
5c71a36a 171 int cs_maxcount;
3b6b7bd1 172 volatile int cs_count;
5c71a36a 173 cpumask_t cs_mask;
3b6b7bd1
MD
174};
175
176/*
f1d1c3fa
MD
177 * The standard message and queue structure used for communications between
178 * cpus. Messages are typically queued via a machine-specific non-linked
179 * FIFO matrix allowing any cpu to send a message to any other cpu without
180 * blocking.
181 */
182typedef struct lwkt_cpu_msg {
183 void (*cm_func)(lwkt_cpu_msg_t msg); /* primary dispatch function */
184 int cm_code; /* request code if applicable */
185 int cm_cpu; /* reply to cpu */
186 thread_t cm_originator; /* originating thread for wakeup */
187} lwkt_cpu_msg;
188
189/*
f1d1c3fa 190 * Thread structure. Note that ownership of a thread structure is special
a72187e9
MD
191 * cased and there is no 'token'. A thread is always owned by the cpu
192 * represented by td_gd, any manipulation of the thread by some other cpu
193 * must be done through cpu_*msg() functions. e.g. you could request
194 * ownership of a thread that way, or hand a thread off to another cpu.
4b5f931b
MD
195 *
196 * NOTE: td_pri is bumped by TDPRI_CRIT when entering a critical section,
197 * but this does not effect how the thread is scheduled by LWKT.
9910d07b
MD
198 *
199 * NOTE: td_ucred is synchronized from the p_ucred on user->kernel syscall,
200 * trap, and AST/signal transitions to provide a stable ucred for
201 * (primarily) system calls. This field will be NULL for pure kernel
202 * threads.
f1d1c3fa 203 */
96728c05 204struct md_intr_info;
f6bf3af1 205struct caps_kinfo;
96728c05 206
2df9fa87
MD
207struct thread {
208 TAILQ_ENTRY(thread) td_threadq;
73e4f7b9 209 TAILQ_ENTRY(thread) td_allq;
ae8e83e6 210 TAILQ_ENTRY(thread) td_sleepq;
ece04fd0 211 lwkt_port td_msgport; /* built-in message port for replies */
ef09c3ed 212 struct lwp *td_lwp; /* (optional) associated lwp */
2df9fa87
MD
213 struct proc *td_proc; /* (optional) associated process */
214 struct pcb *td_pcb; /* points to pcb and top of kstack */
26a0694b 215 struct globaldata *td_gd; /* associated with this cpu */
ae8050a4 216 const char *td_wmesg; /* string name for blockage */
5decebc7 217 const volatile void *td_wchan; /* waiting on channel */
4b5f931b 218 int td_pri; /* 0-31, 31=highest priority (note 1) */
b8337f35 219 int td_flags; /* TDF flags */
da5fb9ef 220 int td_wdomain; /* domain for wchan address (typ 0) */
96728c05 221 void (*td_preemptable)(struct thread *td, int critpri);
a2a5ad0d 222 void (*td_release)(struct thread *td);
7e1d4bf4 223 char *td_kstack; /* kernel stack */
f470d0c8 224 int td_kstack_size; /* size of kernel stack */
8ad65e08
MD
225 char *td_sp; /* kernel stack pointer for LWKT restore */
226 void (*td_switch)(struct thread *ntd);
05220613
MD
227 __uint64_t td_uticks; /* Statclock hits in user mode (uS) */
228 __uint64_t td_sticks; /* Statclock hits in system mode (uS) */
229 __uint64_t td_iticks; /* Statclock hits processing intr (uS) */
69d78e99 230 int td_locks; /* lockmgr lock debugging */
d666840a 231 int td_unused01;
8c72e3d5 232 void *td_dsched_priv1; /* priv data for I/O schedulers */
73e4f7b9 233 int td_refs; /* hold position in gd_tdallq / hold free */
46a3f46d 234 int td_nest_count; /* prevent splz nesting */
8a8d5d85
MD
235#ifdef SMP
236 int td_mpcount; /* MP lock held (count) */
0f7a3396 237 int td_cscount; /* cpu synchronization master */
8a8d5d85 238#else
386344f1
MD
239 int td_mpcount_unused; /* filler so size matches */
240 int td_cscount_unused;
8a8d5d85 241#endif
79eae878 242 struct iosched_data td_iosdata; /* Dynamic I/O scheduling data */
41f3429e 243 struct timeval td_start; /* start time for a thread/process */
0cfcada1 244 char td_comm[MAXCOMLEN+1]; /* typ 16+1 bytes */
99df837e 245 struct thread *td_preempted; /* we preempted this thread */
d86a23e0 246 struct ucred *td_ucred; /* synchronized from p_ucred */
f6bf3af1 247 struct caps_kinfo *td_caps; /* list of client and server registrations */
3b998fa9
MD
248 lwkt_tokref_t td_toks_stop;
249 struct lwkt_tokref td_toks_array[LWKT_MAXTOKENS];
02d8a449
MD
250#ifdef DEBUG_CRIT_SECTIONS
251#define CRIT_DEBUG_ARRAY_SIZE 32
252#define CRIT_DEBUG_ARRAY_MASK (CRIT_DEBUG_ARRAY_SIZE - 1)
253 const char *td_crit_debug_array[CRIT_DEBUG_ARRAY_SIZE];
254 int td_crit_debug_index;
255 int td_in_crit_report;
256#endif
85100692 257 struct md_thread td_mach;
2df9fa87
MD
258};
259
3b998fa9
MD
260#define td_toks_base td_toks_array[0]
261#define td_toks_end td_toks_array[LWKT_MAXTOKENS]
262
263#define TD_TOKS_HELD(td) ((td)->td_toks_stop != &(td)->td_toks_base)
264#define TD_TOKS_NOT_HELD(td) ((td)->td_toks_stop == &(td)->td_toks_base)
265
2df9fa87 266/*
d9eea1a5
MD
267 * Thread flags. Note that TDF_RUNNING is cleared on the old thread after
268 * we switch to the new one, which is necessary because LWKTs don't need
269 * to hold the BGL. This flag is used by the exit code and the managed
c1102e9f
MD
270 * thread migration code. Note in addition that preemption will cause
271 * TDF_RUNNING to be cleared temporarily, so any code checking TDF_RUNNING
272 * must also check TDF_PREEMPT_LOCK.
a2a5ad0d
MD
273 *
274 * LWKT threads stay on their (per-cpu) run queue while running, not to
275 * be confused with user processes which are removed from the user scheduling
276 * run queue while actually running.
344ad853
MD
277 *
278 * td_threadq can represent the thread on one of three queues... the LWKT
279 * run queue, a tsleep queue, or an lwkt blocking queue. The LWKT subsystem
280 * does not allow a thread to be scheduled if it already resides on some
281 * queue.
8ad65e08 282 */
d9eea1a5 283#define TDF_RUNNING 0x0001 /* thread still active */
a2a5ad0d 284#define TDF_RUNQ 0x0002 /* on an LWKT run queue */
26a0694b
MD
285#define TDF_PREEMPT_LOCK 0x0004 /* I have been preempted */
286#define TDF_PREEMPT_DONE 0x0008 /* acknowledge preemption complete */
a2a5ad0d 287#define TDF_IDLE_NOHLT 0x0010 /* we need to spin */
5d21b981 288#define TDF_MIGRATING 0x0020 /* thread is being migrated */
d9d6cb99 289#define TDF_SINTR 0x0040 /* interruptability hint for 'ps' */
344ad853 290#define TDF_TSLEEPQ 0x0080 /* on a tsleep wait queue */
26a0694b 291
4ecf7cc9 292#define TDF_SYSTHREAD 0x0100 /* allocations may use reserve */
40aaf5fc
NT
293#define TDF_ALLOCATED_THREAD 0x0200 /* objcache allocated thread */
294#define TDF_ALLOCATED_STACK 0x0400 /* objcache allocated stack */
99df837e 295#define TDF_VERBOSE 0x0800 /* verbose on exit */
dadab5e9 296#define TDF_DEADLKTREAT 0x1000 /* special lockmgr deadlock treatment */
0cfcada1
MD
297#define TDF_STOPREQ 0x2000 /* suspend_kproc */
298#define TDF_WAKEREQ 0x4000 /* resume_kproc */
299#define TDF_TIMEOUT 0x8000 /* tsleep timeout */
03aa8d99 300#define TDF_INTTHREAD 0x00010000 /* interrupt thread */
ae8e83e6 301#define TDF_TSLEEP_DESCHEDULED 0x00020000 /* tsleep core deschedule */
8ec60c3f 302#define TDF_BLOCKED 0x00040000 /* Thread is blocked */
a7422615 303#define TDF_PANICWARN 0x00080000 /* panic warning in switch */
344ad853 304#define TDF_BLOCKQ 0x00100000 /* on block queue */
c2fba90b 305#define TDF_MPSAFE 0x00200000 /* (thread creation) */
c1102e9f 306#define TDF_EXITING 0x00400000 /* thread exiting */
1b251f0a 307#define TDF_USINGFP 0x00800000 /* thread using fp coproc */
aad81e48 308#define TDF_KERNELFP 0x01000000 /* kernel using fp coproc */
cbb1e737 309#define TDF_NETWORK 0x02000000 /* network proto thread */
8ad65e08
MD
310
311/*
2df9fa87
MD
312 * Thread priorities. Typically only one thread from any given
313 * user process scheduling queue is on the LWKT run queue at a time.
314 * Remember that there is one LWKT run queue per cpu.
f1d1c3fa
MD
315 *
316 * Critical sections are handled by bumping td_pri above TDPRI_MAX, which
235957ed
MD
317 * causes interrupts to be masked as they occur. When this occurs a
318 * rollup flag will be set in mycpu->gd_reqflags.
2df9fa87 319 */
f1d1c3fa 320#define TDPRI_IDLE_THREAD 0 /* the idle thread */
bb6811be 321#define TDPRI_IDLE_WORK 1 /* idle work (page zero, etc) */
50017724 322#define TDPRI_USER_SCHEDULER 2 /* user scheduler helper */
f1d1c3fa
MD
323#define TDPRI_USER_IDLE 4 /* user scheduler idle */
324#define TDPRI_USER_NORM 6 /* user scheduler normal */
325#define TDPRI_USER_REAL 8 /* user scheduler real time */
9ae9ee8d 326#define TDPRI_KERN_LPSCHED 9 /* scheduler helper for userland sch */
f1d1c3fa 327#define TDPRI_KERN_USER 10 /* kernel / block in syscall */
26a0694b 328#define TDPRI_KERN_DAEMON 12 /* kernel daemon (pageout, etc) */
f1d1c3fa
MD
329#define TDPRI_SOFT_NORM 14 /* kernel / normal */
330#define TDPRI_SOFT_TIMER 16 /* kernel / timer */
ae8050a4 331#define TDPRI_EXITING 19 /* exiting thread */
f1d1c3fa
MD
332#define TDPRI_INT_SUPPORT 20 /* kernel / high priority support */
333#define TDPRI_INT_LOW 27 /* low priority interrupt */
334#define TDPRI_INT_MED 28 /* medium priority interrupt */
335#define TDPRI_INT_HIGH 29 /* high priority interrupt */
336#define TDPRI_MAX 31
337
338#define TDPRI_MASK 31
339#define TDPRI_CRIT 32 /* high bits of td_pri used for crit */
2df9fa87 340
d9d13bd5 341#ifdef _KERNEL
f470d0c8 342#define LWKT_THREAD_STACK (UPAGES * PAGE_SIZE)
d9d13bd5 343#endif
f470d0c8 344
ef0fdad1 345#define CACHE_NTHREADS 6
7e1d4bf4 346
f8c3996b
MD
347#define IN_CRITICAL_SECT(td) ((td)->td_pri >= TDPRI_CRIT)
348
40aaf5fc 349extern void lwkt_init(void);
b9665ad7
SW
350extern struct thread *lwkt_alloc_thread(struct thread *, int, int, int);
351extern void lwkt_init_thread(struct thread *, void *, int, int,
352 struct globaldata *);
fcefa6f2 353extern void lwkt_set_comm(thread_t, const char *, ...) __printflike(2, 3);
b9665ad7
SW
354extern void lwkt_wait_free(struct thread *);
355extern void lwkt_free_thread(struct thread *);
356extern void lwkt_gdinit(struct globaldata *);
8ad65e08 357extern void lwkt_switch(void);
b9665ad7
SW
358extern void lwkt_preempt(thread_t, int);
359extern void lwkt_schedule(thread_t);
361d01dd 360extern void lwkt_schedule_noresched(thread_t);
b9665ad7
SW
361extern void lwkt_schedule_self(thread_t);
362extern void lwkt_deschedule(thread_t);
363extern void lwkt_deschedule_self(thread_t);
f1d1c3fa 364extern void lwkt_yield(void);
3824f392 365extern void lwkt_user_yield(void);
41a01a4d 366extern void lwkt_token_wait(void);
b9665ad7
SW
367extern void lwkt_hold(thread_t);
368extern void lwkt_rele(thread_t);
3824f392 369extern void lwkt_passive_release(thread_t);
b9665ad7 370
3b998fa9
MD
371extern void lwkt_gettoken(lwkt_token_t);
372extern int lwkt_trytoken(lwkt_token_t);
373extern void lwkt_reltoken(lwkt_token_t);
b9665ad7
SW
374extern int lwkt_getalltokens(thread_t);
375extern void lwkt_relalltokens(thread_t);
41a01a4d 376extern void lwkt_drain_token_requests(void);
3b998fa9 377extern void lwkt_token_init(lwkt_token_t, int);
b9665ad7 378extern void lwkt_token_uninit(lwkt_token_t);
41a01a4d
MD
379
380extern void lwkt_token_pool_init(void);
c6fbe95a 381extern lwkt_token_t lwkt_token_pool_lookup(void *);
3b998fa9 382extern lwkt_token_t lwkt_getpooltoken(void *);
41a01a4d 383
b9665ad7 384extern void lwkt_setpri(thread_t, int);
03bd0a5e 385extern void lwkt_setpri_initial(thread_t, int);
b9665ad7 386extern void lwkt_setpri_self(int);
b9eb1c19 387extern int lwkt_check_resched(thread_t);
b9665ad7
SW
388extern void lwkt_setcpu_self(struct globaldata *);
389extern void lwkt_migratecpu(int);
b8a98473
MD
390
391#ifdef SMP
392
52eedfb5
MD
393extern void lwkt_giveaway(struct thread *);
394extern void lwkt_acquire(struct thread *);
b9665ad7
SW
395extern int lwkt_send_ipiq3(struct globaldata *, ipifunc3_t, void *, int);
396extern int lwkt_send_ipiq3_passive(struct globaldata *, ipifunc3_t,
397 void *, int);
398extern int lwkt_send_ipiq3_nowait(struct globaldata *, ipifunc3_t,
399 void *, int);
400extern int lwkt_send_ipiq3_bycpu(int, ipifunc3_t, void *, int);
401extern int lwkt_send_ipiq3_mask(cpumask_t, ipifunc3_t, void *, int);
402extern void lwkt_wait_ipiq(struct globaldata *, int);
403extern int lwkt_seq_ipiq(struct globaldata *);
96728c05 404extern void lwkt_process_ipiq(void);
88c4d2f6 405#ifdef _KERNEL
b9665ad7 406extern void lwkt_process_ipiq_frame(struct intrframe *);
88c4d2f6 407#endif
b8a98473 408extern void lwkt_smp_stopped(void);
6c92c1f2 409extern void lwkt_synchronize_ipiqs(const char *);
b8a98473
MD
410
411#endif /* SMP */
412
b9665ad7
SW
413extern void lwkt_cpusync_simple(cpumask_t, cpusync_func_t, void *);
414extern void lwkt_cpusync_fastdata(cpumask_t, cpusync_func2_t, void *);
415extern void lwkt_cpusync_start(cpumask_t, lwkt_cpusync_t);
416extern void lwkt_cpusync_add(cpumask_t, lwkt_cpusync_t);
417extern void lwkt_cpusync_finish(lwkt_cpusync_t);
b8a98473 418
26a0694b 419extern void crit_panic(void);
553ea3c8 420extern struct lwp *lwkt_preempted_proc(void);
4b5f931b 421
b9665ad7 422extern int lwkt_create (void (*func)(void *), void *, struct thread **,
fcefa6f2
SW
423 struct thread *, int, int,
424 const char *, ...) __printflike(7, 8);
b153f746 425extern void lwkt_exit (void) __dead2;
e56e4dea 426extern void lwkt_remove_tdallq (struct thread *);
99df837e 427
2df9fa87
MD
428#endif
429