1 | // |
---|
2 | // Cforall Version 1.0.0 Copyright (C) 2016 University of Waterloo |
---|
3 | // |
---|
4 | // The contents of this file are covered under the licence agreement in the |
---|
5 | // file "LICENCE" distributed with Cforall. |
---|
6 | // |
---|
7 | // kernel.c -- |
---|
8 | // |
---|
9 | // Author : Thierry Delisle |
---|
10 | // Created On : Tue Jan 17 12:27:26 2017 |
---|
11 | // Last Modified By : Peter A. Buhr |
---|
12 | // Last Modified On : Thu Jan 30 22:55:50 2020 |
---|
13 | // Update Count : 56 |
---|
14 | // |
---|
15 | |
---|
16 | #define __cforall_thread__ |
---|
17 | |
---|
18 | //C Includes |
---|
19 | #include <stddef.h> |
---|
20 | #include <errno.h> |
---|
21 | #include <string.h> |
---|
22 | extern "C" { |
---|
23 | #include <stdio.h> |
---|
24 | #include <fenv.h> |
---|
25 | #include <sys/resource.h> |
---|
26 | #include <signal.h> |
---|
27 | #include <unistd.h> |
---|
28 | #include <limits.h> // PTHREAD_STACK_MIN |
---|
29 | #include <sys/mman.h> // mprotect |
---|
30 | } |
---|
31 | |
---|
32 | //CFA Includes |
---|
33 | #include "time.hfa" |
---|
34 | #include "kernel_private.hfa" |
---|
35 | #include "preemption.hfa" |
---|
36 | #include "startup.hfa" |
---|
37 | |
---|
38 | //Private includes |
---|
39 | #define __CFA_INVOKE_PRIVATE__ |
---|
40 | #include "invoke.h" |
---|
41 | |
---|
42 | //----------------------------------------------------------------------------- |
---|
43 | // Some assembly required |
---|
44 | #if defined( __i386 ) |
---|
45 | #define CtxGet( ctx ) \ |
---|
46 | __asm__ volatile ( \ |
---|
47 | "movl %%esp,%0\n"\ |
---|
48 | "movl %%ebp,%1\n"\ |
---|
49 | : "=rm" (ctx.SP),\ |
---|
50 | "=rm" (ctx.FP) \ |
---|
51 | ) |
---|
52 | |
---|
53 | // mxcr : SSE Status and Control bits (control bits are preserved across function calls) |
---|
54 | // fcw : X87 FPU control word (preserved across function calls) |
---|
55 | #define __x87_store \ |
---|
56 | uint32_t __mxcr; \ |
---|
57 | uint16_t __fcw; \ |
---|
58 | __asm__ volatile ( \ |
---|
59 | "stmxcsr %0\n" \ |
---|
60 | "fnstcw %1\n" \ |
---|
61 | : "=m" (__mxcr),\ |
---|
62 | "=m" (__fcw) \ |
---|
63 | ) |
---|
64 | |
---|
65 | #define __x87_load \ |
---|
66 | __asm__ volatile ( \ |
---|
67 | "fldcw %1\n" \ |
---|
68 | "ldmxcsr %0\n" \ |
---|
69 | ::"m" (__mxcr),\ |
---|
70 | "m" (__fcw) \ |
---|
71 | ) |
---|
72 | |
---|
73 | #elif defined( __x86_64 ) |
---|
74 | #define CtxGet( ctx ) \ |
---|
75 | __asm__ volatile ( \ |
---|
76 | "movq %%rsp,%0\n"\ |
---|
77 | "movq %%rbp,%1\n"\ |
---|
78 | : "=rm" (ctx.SP),\ |
---|
79 | "=rm" (ctx.FP) \ |
---|
80 | ) |
---|
81 | |
---|
82 | #define __x87_store \ |
---|
83 | uint32_t __mxcr; \ |
---|
84 | uint16_t __fcw; \ |
---|
85 | __asm__ volatile ( \ |
---|
86 | "stmxcsr %0\n" \ |
---|
87 | "fnstcw %1\n" \ |
---|
88 | : "=m" (__mxcr),\ |
---|
89 | "=m" (__fcw) \ |
---|
90 | ) |
---|
91 | |
---|
92 | #define __x87_load \ |
---|
93 | __asm__ volatile ( \ |
---|
94 | "fldcw %1\n" \ |
---|
95 | "ldmxcsr %0\n" \ |
---|
96 | :: "m" (__mxcr),\ |
---|
97 | "m" (__fcw) \ |
---|
98 | ) |
---|
99 | |
---|
100 | |
---|
101 | #elif defined( __ARM_ARCH ) |
---|
102 | #define CtxGet( ctx ) __asm__ ( \ |
---|
103 | "mov %0,%%sp\n" \ |
---|
104 | "mov %1,%%r11\n" \ |
---|
105 | : "=rm" (ctx.SP), "=rm" (ctx.FP) ) |
---|
106 | #else |
---|
107 | #error unknown hardware architecture |
---|
108 | #endif |
---|
109 | |
---|
110 | //----------------------------------------------------------------------------- |
---|
111 | //Start and stop routine for the kernel, declared first to make sure they run first |
---|
112 | static void kernel_startup(void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) )); |
---|
113 | static void kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) )); |
---|
114 | |
---|
115 | //----------------------------------------------------------------------------- |
---|
116 | // Kernel storage |
---|
117 | KERNEL_STORAGE(cluster, mainCluster); |
---|
118 | KERNEL_STORAGE(processor, mainProcessor); |
---|
119 | KERNEL_STORAGE(thread_desc, mainThread); |
---|
120 | KERNEL_STORAGE(__stack_t, mainThreadCtx); |
---|
121 | |
---|
122 | cluster * mainCluster; |
---|
123 | processor * mainProcessor; |
---|
124 | thread_desc * mainThread; |
---|
125 | |
---|
126 | extern "C" { |
---|
127 | struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters; |
---|
128 | } |
---|
129 | |
---|
130 | size_t __page_size = 0; |
---|
131 | |
---|
132 | //----------------------------------------------------------------------------- |
---|
133 | // Global state |
---|
134 | thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) = { |
---|
135 | NULL, // cannot use 0p |
---|
136 | NULL, |
---|
137 | { 1, false, false }, |
---|
138 | 6u //this should be seeded better but due to a bug calling rdtsc doesn't work |
---|
139 | }; |
---|
140 | |
---|
141 | //----------------------------------------------------------------------------- |
---|
142 | // Struct to steal stack |
---|
143 | struct current_stack_info_t { |
---|
144 | __stack_t * storage; // pointer to stack object |
---|
145 | void * base; // base of stack |
---|
146 | void * limit; // stack grows towards stack limit |
---|
147 | void * context; // address of cfa_context_t |
---|
148 | }; |
---|
149 | |
---|
150 | void ?{}( current_stack_info_t & this ) { |
---|
151 | __stack_context_t ctx; |
---|
152 | CtxGet( ctx ); |
---|
153 | this.base = ctx.FP; |
---|
154 | |
---|
155 | rlimit r; |
---|
156 | getrlimit( RLIMIT_STACK, &r); |
---|
157 | size_t size = r.rlim_cur; |
---|
158 | |
---|
159 | this.limit = (void *)(((intptr_t)this.base) - size); |
---|
160 | this.context = &storage_mainThreadCtx; |
---|
161 | } |
---|
162 | |
---|
163 | //----------------------------------------------------------------------------- |
---|
164 | // Main thread construction |
---|
165 | |
---|
166 | void ?{}( coroutine_desc & this, current_stack_info_t * info) with( this ) { |
---|
167 | stack.storage = info->storage; |
---|
168 | with(*stack.storage) { |
---|
169 | limit = info->limit; |
---|
170 | base = info->base; |
---|
171 | } |
---|
172 | __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage; |
---|
173 | *istorage |= 0x1; |
---|
174 | name = "Main Thread"; |
---|
175 | state = Start; |
---|
176 | starter = 0p; |
---|
177 | last = 0p; |
---|
178 | cancellation = 0p; |
---|
179 | } |
---|
180 | |
---|
181 | void ?{}( thread_desc & this, current_stack_info_t * info) with( this ) { |
---|
182 | state = Start; |
---|
183 | self_cor{ info }; |
---|
184 | curr_cor = &self_cor; |
---|
185 | curr_cluster = mainCluster; |
---|
186 | self_mon.owner = &this; |
---|
187 | self_mon.recursion = 1; |
---|
188 | self_mon_p = &self_mon; |
---|
189 | next = 0p; |
---|
190 | |
---|
191 | node.next = 0p; |
---|
192 | node.prev = 0p; |
---|
193 | doregister(curr_cluster, this); |
---|
194 | |
---|
195 | monitors{ &self_mon_p, 1, (fptr_t)0 }; |
---|
196 | } |
---|
197 | |
---|
198 | //----------------------------------------------------------------------------- |
---|
199 | // Processor coroutine |
---|
200 | void ?{}(processorCtx_t & this) { |
---|
201 | |
---|
202 | } |
---|
203 | |
---|
204 | // Construct the processor context of non-main processors |
---|
205 | static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) { |
---|
206 | (this.__cor){ info }; |
---|
207 | this.proc = proc; |
---|
208 | } |
---|
209 | |
---|
210 | static void start(processor * this); |
---|
211 | void ?{}(processor & this, const char * name, cluster & cltr) with( this ) { |
---|
212 | this.name = name; |
---|
213 | this.cltr = &cltr; |
---|
214 | terminated{ 0 }; |
---|
215 | do_terminate = false; |
---|
216 | preemption_alarm = 0p; |
---|
217 | pending_preemption = false; |
---|
218 | runner.proc = &this; |
---|
219 | |
---|
220 | idleLock{}; |
---|
221 | |
---|
222 | start( &this ); |
---|
223 | } |
---|
224 | |
---|
225 | void ^?{}(processor & this) with( this ){ |
---|
226 | if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) { |
---|
227 | __cfaabi_dbg_print_safe("Kernel : core %p signaling termination\n", &this); |
---|
228 | |
---|
229 | __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED); |
---|
230 | wake( &this ); |
---|
231 | |
---|
232 | P( terminated ); |
---|
233 | verify( kernelTLS.this_processor != &this); |
---|
234 | } |
---|
235 | |
---|
236 | pthread_join( kernel_thread, 0p ); |
---|
237 | free( this.stack ); |
---|
238 | } |
---|
239 | |
---|
240 | void ?{}(cluster & this, const char * name, Duration preemption_rate) with( this ) { |
---|
241 | this.name = name; |
---|
242 | this.preemption_rate = preemption_rate; |
---|
243 | ready_queue{}; |
---|
244 | ready_queue_lock{}; |
---|
245 | |
---|
246 | procs{ __get }; |
---|
247 | idles{ __get }; |
---|
248 | threads{ __get }; |
---|
249 | |
---|
250 | doregister(this); |
---|
251 | } |
---|
252 | |
---|
253 | void ^?{}(cluster & this) { |
---|
254 | unregister(this); |
---|
255 | } |
---|
256 | |
---|
257 | //============================================================================================= |
---|
258 | // Kernel Scheduling logic |
---|
259 | //============================================================================================= |
---|
260 | static void runThread(processor * this, thread_desc * dst); |
---|
261 | static void finishRunning(processor * this); |
---|
262 | static void halt(processor * this); |
---|
263 | |
---|
264 | //Main of the processor contexts |
---|
265 | void main(processorCtx_t & runner) { |
---|
266 | // Because of a bug, we couldn't initialized the seed on construction |
---|
267 | // Do it here |
---|
268 | kernelTLS.rand_seed ^= rdtscl(); |
---|
269 | |
---|
270 | processor * this = runner.proc; |
---|
271 | verify(this); |
---|
272 | |
---|
273 | __cfaabi_dbg_print_safe("Kernel : core %p starting\n", this); |
---|
274 | |
---|
275 | doregister(this->cltr, this); |
---|
276 | |
---|
277 | { |
---|
278 | // Setup preemption data |
---|
279 | preemption_scope scope = { this }; |
---|
280 | |
---|
281 | __cfaabi_dbg_print_safe("Kernel : core %p started\n", this); |
---|
282 | |
---|
283 | thread_desc * readyThread = 0p; |
---|
284 | for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) { |
---|
285 | readyThread = nextThread( this->cltr ); |
---|
286 | |
---|
287 | if(readyThread) { |
---|
288 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
289 | |
---|
290 | runThread(this, readyThread); |
---|
291 | |
---|
292 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
293 | |
---|
294 | //Some actions need to be taken from the kernel |
---|
295 | finishRunning(this); |
---|
296 | |
---|
297 | spin_count = 0; |
---|
298 | } else { |
---|
299 | // spin(this, &spin_count); |
---|
300 | halt(this); |
---|
301 | } |
---|
302 | } |
---|
303 | |
---|
304 | __cfaabi_dbg_print_safe("Kernel : core %p stopping\n", this); |
---|
305 | } |
---|
306 | |
---|
307 | unregister(this->cltr, this); |
---|
308 | |
---|
309 | V( this->terminated ); |
---|
310 | |
---|
311 | __cfaabi_dbg_print_safe("Kernel : core %p terminated\n", this); |
---|
312 | } |
---|
313 | |
---|
314 | static int * __volatile_errno() __attribute__((noinline)); |
---|
315 | static int * __volatile_errno() { asm(""); return &errno; } |
---|
316 | |
---|
317 | // KERNEL ONLY |
---|
318 | // runThread runs a thread by context switching |
---|
319 | // from the processor coroutine to the target thread |
---|
320 | static void runThread(processor * this, thread_desc * thrd_dst) { |
---|
321 | coroutine_desc * proc_cor = get_coroutine(this->runner); |
---|
322 | |
---|
323 | // Reset the terminating actions here |
---|
324 | this->finish.action_code = No_Action; |
---|
325 | |
---|
326 | // Update global state |
---|
327 | kernelTLS.this_thread = thrd_dst; |
---|
328 | |
---|
329 | // set state of processor coroutine to inactive |
---|
330 | verify(proc_cor->state == Active); |
---|
331 | proc_cor->state = Inactive; |
---|
332 | |
---|
333 | // Actually run the thread |
---|
334 | RUN: |
---|
335 | { |
---|
336 | if(unlikely(thrd_dst->preempted)) { |
---|
337 | thrd_dst->preempted = false; |
---|
338 | } else { |
---|
339 | thrd_dst->state = Active; |
---|
340 | } |
---|
341 | |
---|
342 | // set context switch to the thread that the processor is executing |
---|
343 | verify( thrd_dst->context.SP ); |
---|
344 | CtxSwitch( &proc_cor->context, &thrd_dst->context ); |
---|
345 | // when CtxSwitch returns we are back in the processor coroutine |
---|
346 | } |
---|
347 | |
---|
348 | // We just finished running a thread, there are a few things that could have happened. |
---|
349 | // 1 - Regular case : the thread has blocked and now one has scheduled it yet. |
---|
350 | // 2 - Racy case : the thread has blocked but someone has already tried to schedule it. |
---|
351 | // 3 - Polite Racy case : the thread has blocked, someone has already tried to schedule it, but the thread is nice and wants to go through the ready-queue any way |
---|
352 | // 4 - Preempted |
---|
353 | // In case 1, we may have won a race so we can't write to the state again. |
---|
354 | // In case 2, we lost the race so we now own the thread. |
---|
355 | // In case 3, we lost the race but can just reschedule the thread. |
---|
356 | |
---|
357 | if(unlikely(thrd_dst->preempted)) { |
---|
358 | // The thread was preempted, reschedule it and reset the flag |
---|
359 | ScheduleThread( thrd_dst ); |
---|
360 | |
---|
361 | // Just before returning to the processor, set the processor coroutine to active |
---|
362 | proc_cor->state = Active; |
---|
363 | return; |
---|
364 | } |
---|
365 | |
---|
366 | // set state of processor coroutine to active and the thread to inactive |
---|
367 | enum coroutine_state old_state = __atomic_exchange_n(&thrd_dst->state, Inactive, __ATOMIC_SEQ_CST); |
---|
368 | switch(old_state) { |
---|
369 | case Halted: |
---|
370 | // The thread has halted, it should never be scheduled/run again, leave it back to Halted and move on |
---|
371 | thrd_dst->state = Halted; |
---|
372 | break; |
---|
373 | case Active: |
---|
374 | // This is case 1, the regular case, nothing more is needed |
---|
375 | break; |
---|
376 | case Rerun: |
---|
377 | // This is case 2, the racy case, someone tried to run this thread before it finished blocking |
---|
378 | // In this case, just run it again. |
---|
379 | goto RUN; |
---|
380 | case Reschedule: |
---|
381 | // This is case 3, someone tried to run this before it finished blocking |
---|
382 | // but it must go through the ready-queue |
---|
383 | thrd_dst->state = Inactive; /*restore invariant */ |
---|
384 | ScheduleThread( thrd_dst ); |
---|
385 | break; |
---|
386 | case Inactive: |
---|
387 | case Start: |
---|
388 | case Primed: |
---|
389 | default: |
---|
390 | // This makes no sense, something is wrong abort |
---|
391 | abort("Finished running a thread that was Inactive/Start/Primed %d\n", old_state); |
---|
392 | } |
---|
393 | |
---|
394 | // Just before returning to the processor, set the processor coroutine to active |
---|
395 | proc_cor->state = Active; |
---|
396 | } |
---|
397 | |
---|
398 | // KERNEL_ONLY |
---|
399 | static void returnToKernel() { |
---|
400 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
401 | coroutine_desc * proc_cor = get_coroutine(kernelTLS.this_processor->runner); |
---|
402 | thread_desc * thrd_src = kernelTLS.this_thread; |
---|
403 | |
---|
404 | // Run the thread on this processor |
---|
405 | { |
---|
406 | int local_errno = *__volatile_errno(); |
---|
407 | #if defined( __i386 ) || defined( __x86_64 ) |
---|
408 | __x87_store; |
---|
409 | #endif |
---|
410 | verify( proc_cor->context.SP ); |
---|
411 | CtxSwitch( &thrd_src->context, &proc_cor->context ); |
---|
412 | #if defined( __i386 ) || defined( __x86_64 ) |
---|
413 | __x87_load; |
---|
414 | #endif |
---|
415 | *__volatile_errno() = local_errno; |
---|
416 | } |
---|
417 | |
---|
418 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
419 | } |
---|
420 | |
---|
421 | // KERNEL_ONLY |
---|
422 | // Once a thread has finished running, some of |
---|
423 | // its final actions must be executed from the kernel |
---|
424 | static void finishRunning(processor * this) with( this->finish ) { |
---|
425 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
426 | verify( action_code == No_Action ); |
---|
427 | choose( action_code ) { |
---|
428 | case No_Action: |
---|
429 | break; |
---|
430 | case Release: |
---|
431 | unlock( *lock ); |
---|
432 | case Schedule: |
---|
433 | ScheduleThread( thrd ); |
---|
434 | case Release_Schedule: |
---|
435 | unlock( *lock ); |
---|
436 | ScheduleThread( thrd ); |
---|
437 | case Release_Multi: |
---|
438 | for(int i = 0; i < lock_count; i++) { |
---|
439 | unlock( *locks[i] ); |
---|
440 | } |
---|
441 | case Release_Multi_Schedule: |
---|
442 | for(int i = 0; i < lock_count; i++) { |
---|
443 | unlock( *locks[i] ); |
---|
444 | } |
---|
445 | for(int i = 0; i < thrd_count; i++) { |
---|
446 | ScheduleThread( thrds[i] ); |
---|
447 | } |
---|
448 | case Callback: |
---|
449 | callback(); |
---|
450 | default: |
---|
451 | abort("KERNEL ERROR: Unexpected action to run after thread"); |
---|
452 | } |
---|
453 | } |
---|
454 | |
---|
455 | // KERNEL_ONLY |
---|
456 | // Context invoker for processors |
---|
457 | // This is the entry point for processors (kernel threads) |
---|
458 | // It effectively constructs a coroutine by stealing the pthread stack |
---|
459 | static void * CtxInvokeProcessor(void * arg) { |
---|
460 | processor * proc = (processor *) arg; |
---|
461 | kernelTLS.this_processor = proc; |
---|
462 | kernelTLS.this_thread = 0p; |
---|
463 | kernelTLS.preemption_state.[enabled, disable_count] = [false, 1]; |
---|
464 | // SKULLDUGGERY: We want to create a context for the processor coroutine |
---|
465 | // which is needed for the 2-step context switch. However, there is no reason |
---|
466 | // to waste the perfectly valid stack create by pthread. |
---|
467 | current_stack_info_t info; |
---|
468 | __stack_t ctx; |
---|
469 | info.storage = &ctx; |
---|
470 | (proc->runner){ proc, &info }; |
---|
471 | |
---|
472 | __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage); |
---|
473 | |
---|
474 | //Set global state |
---|
475 | kernelTLS.this_thread = 0p; |
---|
476 | |
---|
477 | //We now have a proper context from which to schedule threads |
---|
478 | __cfaabi_dbg_print_safe("Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx); |
---|
479 | |
---|
480 | // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't |
---|
481 | // resume it to start it like it normally would, it will just context switch |
---|
482 | // back to here. Instead directly call the main since we already are on the |
---|
483 | // appropriate stack. |
---|
484 | get_coroutine(proc->runner)->state = Active; |
---|
485 | main( proc->runner ); |
---|
486 | get_coroutine(proc->runner)->state = Halted; |
---|
487 | |
---|
488 | // Main routine of the core returned, the core is now fully terminated |
---|
489 | __cfaabi_dbg_print_safe("Kernel : core %p main ended (%p)\n", proc, &proc->runner); |
---|
490 | |
---|
491 | return 0p; |
---|
492 | } |
---|
493 | |
---|
494 | static void Abort( int ret, const char * func ) { |
---|
495 | if ( ret ) { // pthread routines return errno values |
---|
496 | abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) ); |
---|
497 | } // if |
---|
498 | } // Abort |
---|
499 | |
---|
500 | void * create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) { |
---|
501 | pthread_attr_t attr; |
---|
502 | |
---|
503 | Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute |
---|
504 | |
---|
505 | size_t stacksize; |
---|
506 | // default stack size, normally defined by shell limit |
---|
507 | Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" ); |
---|
508 | assert( stacksize >= PTHREAD_STACK_MIN ); |
---|
509 | |
---|
510 | void * stack; |
---|
511 | __cfaabi_dbg_debug_do( |
---|
512 | stack = memalign( __page_size, stacksize + __page_size ); |
---|
513 | // pthread has no mechanism to create the guard page in user supplied stack. |
---|
514 | if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) { |
---|
515 | abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) ); |
---|
516 | } // if |
---|
517 | ); |
---|
518 | __cfaabi_dbg_no_debug_do( |
---|
519 | stack = malloc( stacksize ); |
---|
520 | ); |
---|
521 | |
---|
522 | Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" ); |
---|
523 | |
---|
524 | Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" ); |
---|
525 | return stack; |
---|
526 | } |
---|
527 | |
---|
528 | static void start(processor * this) { |
---|
529 | __cfaabi_dbg_print_safe("Kernel : Starting core %p\n", this); |
---|
530 | |
---|
531 | this->stack = create_pthread( &this->kernel_thread, CtxInvokeProcessor, (void *)this ); |
---|
532 | |
---|
533 | __cfaabi_dbg_print_safe("Kernel : core %p started\n", this); |
---|
534 | } |
---|
535 | |
---|
536 | // KERNEL_ONLY |
---|
537 | void kernel_first_resume( processor * this ) { |
---|
538 | thread_desc * src = mainThread; |
---|
539 | coroutine_desc * dst = get_coroutine(this->runner); |
---|
540 | |
---|
541 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
542 | |
---|
543 | kernelTLS.this_thread->curr_cor = dst; |
---|
544 | __stack_prepare( &dst->stack, 65000 ); |
---|
545 | CtxStart(main, dst, this->runner, CtxInvokeCoroutine); |
---|
546 | |
---|
547 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
548 | |
---|
549 | dst->last = &src->self_cor; |
---|
550 | dst->starter = dst->starter ? dst->starter : &src->self_cor; |
---|
551 | |
---|
552 | // set state of current coroutine to inactive |
---|
553 | src->state = src->state == Halted ? Halted : Inactive; |
---|
554 | |
---|
555 | // context switch to specified coroutine |
---|
556 | verify( dst->context.SP ); |
---|
557 | CtxSwitch( &src->context, &dst->context ); |
---|
558 | // when CtxSwitch returns we are back in the src coroutine |
---|
559 | |
---|
560 | mainThread->curr_cor = &mainThread->self_cor; |
---|
561 | |
---|
562 | // set state of new coroutine to active |
---|
563 | src->state = Active; |
---|
564 | |
---|
565 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
566 | } |
---|
567 | |
---|
568 | // KERNEL_ONLY |
---|
569 | void kernel_last_resume( processor * this ) { |
---|
570 | coroutine_desc * src = &mainThread->self_cor; |
---|
571 | coroutine_desc * dst = get_coroutine(this->runner); |
---|
572 | |
---|
573 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
574 | verify( dst->starter == src ); |
---|
575 | verify( dst->context.SP ); |
---|
576 | |
---|
577 | // context switch to the processor |
---|
578 | CtxSwitch( &src->context, &dst->context ); |
---|
579 | } |
---|
580 | |
---|
581 | //----------------------------------------------------------------------------- |
---|
582 | // Scheduler routines |
---|
583 | |
---|
584 | // KERNEL ONLY |
---|
585 | void ScheduleThread( thread_desc * thrd ) with( *thrd->curr_cluster ) { |
---|
586 | /* paranoid */ verify( ! kernelTLS.preemption_state.enabled ); |
---|
587 | /* paranoid */ verifyf( thrd->state == Inactive || thrd->state == Start || thrd->preempted, "state : %d, preempted %d\n", thrd->state, thrd->preempted); |
---|
588 | /* paranoid */ verifyf( thrd->next == 0p, "Expected null got %p", thrd->next ); |
---|
589 | |
---|
590 | lock ( ready_queue_lock __cfaabi_dbg_ctx2 ); |
---|
591 | bool was_empty = !(ready_queue != 0); |
---|
592 | append( ready_queue, thrd ); |
---|
593 | unlock( ready_queue_lock ); |
---|
594 | |
---|
595 | if(was_empty) { |
---|
596 | lock (proc_list_lock __cfaabi_dbg_ctx2); |
---|
597 | if(idles) { |
---|
598 | wake_fast(idles.head); |
---|
599 | } |
---|
600 | unlock (proc_list_lock); |
---|
601 | } |
---|
602 | else if( struct processor * idle = idles.head ) { |
---|
603 | wake_fast(idle); |
---|
604 | } |
---|
605 | |
---|
606 | /* paranoid */ verify( ! kernelTLS.preemption_state.enabled ); |
---|
607 | } |
---|
608 | |
---|
609 | // KERNEL ONLY |
---|
610 | thread_desc * nextThread(cluster * this) with( *this ) { |
---|
611 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
612 | lock( ready_queue_lock __cfaabi_dbg_ctx2 ); |
---|
613 | thread_desc * head = pop_head( ready_queue ); |
---|
614 | unlock( ready_queue_lock ); |
---|
615 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
616 | return head; |
---|
617 | } |
---|
618 | |
---|
619 | void BlockInternal() { |
---|
620 | disable_interrupts(); |
---|
621 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
622 | returnToKernel(); |
---|
623 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
624 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
625 | } |
---|
626 | |
---|
627 | void BlockInternal( __spinlock_t * lock ) { |
---|
628 | disable_interrupts(); |
---|
629 | unlock( *lock ); |
---|
630 | |
---|
631 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
632 | returnToKernel(); |
---|
633 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
634 | |
---|
635 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
636 | } |
---|
637 | |
---|
638 | void BlockInternal( thread_desc * thrd ) { |
---|
639 | disable_interrupts(); |
---|
640 | WakeThread( thrd, false ); |
---|
641 | |
---|
642 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
643 | returnToKernel(); |
---|
644 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
645 | |
---|
646 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
647 | } |
---|
648 | |
---|
649 | void BlockInternal( __spinlock_t * lock, thread_desc * thrd ) { |
---|
650 | disable_interrupts(); |
---|
651 | unlock( *lock ); |
---|
652 | WakeThread( thrd, false ); |
---|
653 | |
---|
654 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
655 | returnToKernel(); |
---|
656 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
657 | |
---|
658 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
659 | } |
---|
660 | |
---|
661 | void BlockInternal(__spinlock_t * locks [], unsigned short count) { |
---|
662 | disable_interrupts(); |
---|
663 | for(int i = 0; i < count; i++) { |
---|
664 | unlock( *locks[i] ); |
---|
665 | } |
---|
666 | |
---|
667 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
668 | returnToKernel(); |
---|
669 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
670 | |
---|
671 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
672 | } |
---|
673 | |
---|
674 | void BlockInternal(__spinlock_t * locks [], unsigned short lock_count, thread_desc * thrds [], unsigned short thrd_count) { |
---|
675 | disable_interrupts(); |
---|
676 | for(int i = 0; i < lock_count; i++) { |
---|
677 | unlock( *locks[i] ); |
---|
678 | } |
---|
679 | for(int i = 0; i < thrd_count; i++) { |
---|
680 | WakeThread( thrds[i], false ); |
---|
681 | } |
---|
682 | |
---|
683 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
684 | returnToKernel(); |
---|
685 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
686 | |
---|
687 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
688 | } |
---|
689 | |
---|
690 | void BlockInternal(__finish_callback_fptr_t callback) { |
---|
691 | disable_interrupts(); |
---|
692 | callback(); |
---|
693 | |
---|
694 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
695 | returnToKernel(); |
---|
696 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
697 | |
---|
698 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
699 | } |
---|
700 | |
---|
701 | // KERNEL ONLY |
---|
702 | void LeaveThread(__spinlock_t * lock, thread_desc * thrd) { |
---|
703 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
704 | unlock( *lock ); |
---|
705 | WakeThread( thrd, false ); |
---|
706 | |
---|
707 | returnToKernel(); |
---|
708 | } |
---|
709 | |
---|
710 | //============================================================================================= |
---|
711 | // Kernel Setup logic |
---|
712 | //============================================================================================= |
---|
713 | //----------------------------------------------------------------------------- |
---|
714 | // Kernel boot procedures |
---|
715 | static void kernel_startup(void) { |
---|
716 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
717 | __cfaabi_dbg_print_safe("Kernel : Starting\n"); |
---|
718 | |
---|
719 | __page_size = sysconf( _SC_PAGESIZE ); |
---|
720 | |
---|
721 | __cfa_dbg_global_clusters.list{ __get }; |
---|
722 | __cfa_dbg_global_clusters.lock{}; |
---|
723 | |
---|
724 | // Initialize the main cluster |
---|
725 | mainCluster = (cluster *)&storage_mainCluster; |
---|
726 | (*mainCluster){"Main Cluster"}; |
---|
727 | |
---|
728 | __cfaabi_dbg_print_safe("Kernel : Main cluster ready\n"); |
---|
729 | |
---|
730 | // Start by initializing the main thread |
---|
731 | // SKULLDUGGERY: the mainThread steals the process main thread |
---|
732 | // which will then be scheduled by the mainProcessor normally |
---|
733 | mainThread = (thread_desc *)&storage_mainThread; |
---|
734 | current_stack_info_t info; |
---|
735 | info.storage = (__stack_t*)&storage_mainThreadCtx; |
---|
736 | (*mainThread){ &info }; |
---|
737 | |
---|
738 | __cfaabi_dbg_print_safe("Kernel : Main thread ready\n"); |
---|
739 | |
---|
740 | |
---|
741 | |
---|
742 | // Construct the processor context of the main processor |
---|
743 | void ?{}(processorCtx_t & this, processor * proc) { |
---|
744 | (this.__cor){ "Processor" }; |
---|
745 | this.__cor.starter = 0p; |
---|
746 | this.proc = proc; |
---|
747 | } |
---|
748 | |
---|
749 | void ?{}(processor & this) with( this ) { |
---|
750 | name = "Main Processor"; |
---|
751 | cltr = mainCluster; |
---|
752 | terminated{ 0 }; |
---|
753 | do_terminate = false; |
---|
754 | preemption_alarm = 0p; |
---|
755 | pending_preemption = false; |
---|
756 | kernel_thread = pthread_self(); |
---|
757 | |
---|
758 | runner{ &this }; |
---|
759 | __cfaabi_dbg_print_safe("Kernel : constructed main processor context %p\n", &runner); |
---|
760 | } |
---|
761 | |
---|
762 | // Initialize the main processor and the main processor ctx |
---|
763 | // (the coroutine that contains the processing control flow) |
---|
764 | mainProcessor = (processor *)&storage_mainProcessor; |
---|
765 | (*mainProcessor){}; |
---|
766 | |
---|
767 | //initialize the global state variables |
---|
768 | kernelTLS.this_processor = mainProcessor; |
---|
769 | kernelTLS.this_thread = mainThread; |
---|
770 | |
---|
771 | // Enable preemption |
---|
772 | kernel_start_preemption(); |
---|
773 | |
---|
774 | // Add the main thread to the ready queue |
---|
775 | // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread |
---|
776 | ScheduleThread(mainThread); |
---|
777 | |
---|
778 | // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX |
---|
779 | // context. Hence, the main thread does not begin through CtxInvokeThread, like all other threads. The trick here is that |
---|
780 | // mainThread is on the ready queue when this call is made. |
---|
781 | kernel_first_resume( kernelTLS.this_processor ); |
---|
782 | |
---|
783 | |
---|
784 | |
---|
785 | // THE SYSTEM IS NOW COMPLETELY RUNNING |
---|
786 | __cfaabi_dbg_print_safe("Kernel : Started\n--------------------------------------------------\n\n"); |
---|
787 | |
---|
788 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
789 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
790 | verify( TL_GET( preemption_state.enabled ) ); |
---|
791 | } |
---|
792 | |
---|
793 | static void kernel_shutdown(void) { |
---|
794 | __cfaabi_dbg_print_safe("\n--------------------------------------------------\nKernel : Shutting down\n"); |
---|
795 | |
---|
796 | verify( TL_GET( preemption_state.enabled ) ); |
---|
797 | disable_interrupts(); |
---|
798 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
799 | |
---|
800 | // SKULLDUGGERY: Notify the mainProcessor it needs to terminates. |
---|
801 | // When its coroutine terminates, it return control to the mainThread |
---|
802 | // which is currently here |
---|
803 | __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE); |
---|
804 | kernel_last_resume( kernelTLS.this_processor ); |
---|
805 | mainThread->self_cor.state = Halted; |
---|
806 | |
---|
807 | // THE SYSTEM IS NOW COMPLETELY STOPPED |
---|
808 | |
---|
809 | // Disable preemption |
---|
810 | kernel_stop_preemption(); |
---|
811 | |
---|
812 | // Destroy the main processor and its context in reverse order of construction |
---|
813 | // These were manually constructed so we need manually destroy them |
---|
814 | ^(mainProcessor->runner){}; |
---|
815 | ^(mainProcessor){}; |
---|
816 | |
---|
817 | // Final step, destroy the main thread since it is no longer needed |
---|
818 | // Since we provided a stack to this taxk it will not destroy anything |
---|
819 | ^(mainThread){}; |
---|
820 | |
---|
821 | ^(__cfa_dbg_global_clusters.list){}; |
---|
822 | ^(__cfa_dbg_global_clusters.lock){}; |
---|
823 | |
---|
824 | __cfaabi_dbg_print_safe("Kernel : Shutdown complete\n"); |
---|
825 | } |
---|
826 | |
---|
827 | //============================================================================================= |
---|
828 | // Kernel Quiescing |
---|
829 | //============================================================================================= |
---|
830 | static void halt(processor * this) with( *this ) { |
---|
831 | // verify( ! __atomic_load_n(&do_terminate, __ATOMIC_SEQ_CST) ); |
---|
832 | |
---|
833 | with( *cltr ) { |
---|
834 | lock (proc_list_lock __cfaabi_dbg_ctx2); |
---|
835 | remove (procs, *this); |
---|
836 | push_front(idles, *this); |
---|
837 | unlock (proc_list_lock); |
---|
838 | } |
---|
839 | |
---|
840 | __cfaabi_dbg_print_safe("Kernel : Processor %p ready to sleep\n", this); |
---|
841 | |
---|
842 | wait( idleLock ); |
---|
843 | |
---|
844 | __cfaabi_dbg_print_safe("Kernel : Processor %p woke up and ready to run\n", this); |
---|
845 | |
---|
846 | with( *cltr ) { |
---|
847 | lock (proc_list_lock __cfaabi_dbg_ctx2); |
---|
848 | remove (idles, *this); |
---|
849 | push_front(procs, *this); |
---|
850 | unlock (proc_list_lock); |
---|
851 | } |
---|
852 | } |
---|
853 | |
---|
854 | //============================================================================================= |
---|
855 | // Unexpected Terminating logic |
---|
856 | //============================================================================================= |
---|
857 | static __spinlock_t kernel_abort_lock; |
---|
858 | static bool kernel_abort_called = false; |
---|
859 | |
---|
860 | void * kernel_abort(void) __attribute__ ((__nothrow__)) { |
---|
861 | // abort cannot be recursively entered by the same or different processors because all signal handlers return when |
---|
862 | // the globalAbort flag is true. |
---|
863 | lock( kernel_abort_lock __cfaabi_dbg_ctx2 ); |
---|
864 | |
---|
865 | // first task to abort ? |
---|
866 | if ( kernel_abort_called ) { // not first task to abort ? |
---|
867 | unlock( kernel_abort_lock ); |
---|
868 | |
---|
869 | sigset_t mask; |
---|
870 | sigemptyset( &mask ); |
---|
871 | sigaddset( &mask, SIGALRM ); // block SIGALRM signals |
---|
872 | sigaddset( &mask, SIGUSR1 ); // block SIGALRM signals |
---|
873 | sigsuspend( &mask ); // block the processor to prevent further damage during abort |
---|
874 | _exit( EXIT_FAILURE ); // if processor unblocks before it is killed, terminate it |
---|
875 | } |
---|
876 | else { |
---|
877 | kernel_abort_called = true; |
---|
878 | unlock( kernel_abort_lock ); |
---|
879 | } |
---|
880 | |
---|
881 | return kernelTLS.this_thread; |
---|
882 | } |
---|
883 | |
---|
884 | void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) { |
---|
885 | thread_desc * thrd = kernel_data; |
---|
886 | |
---|
887 | if(thrd) { |
---|
888 | int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd ); |
---|
889 | __cfaabi_bits_write( STDERR_FILENO, abort_text, len ); |
---|
890 | |
---|
891 | if ( &thrd->self_cor != thrd->curr_cor ) { |
---|
892 | len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor ); |
---|
893 | __cfaabi_bits_write( STDERR_FILENO, abort_text, len ); |
---|
894 | } |
---|
895 | else { |
---|
896 | __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 ); |
---|
897 | } |
---|
898 | } |
---|
899 | else { |
---|
900 | int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" ); |
---|
901 | __cfaabi_bits_write( STDERR_FILENO, abort_text, len ); |
---|
902 | } |
---|
903 | } |
---|
904 | |
---|
905 | int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) { |
---|
906 | return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2; |
---|
907 | } |
---|
908 | |
---|
909 | static __spinlock_t kernel_debug_lock; |
---|
910 | |
---|
911 | extern "C" { |
---|
912 | void __cfaabi_bits_acquire() { |
---|
913 | lock( kernel_debug_lock __cfaabi_dbg_ctx2 ); |
---|
914 | } |
---|
915 | |
---|
916 | void __cfaabi_bits_release() { |
---|
917 | unlock( kernel_debug_lock ); |
---|
918 | } |
---|
919 | } |
---|
920 | |
---|
921 | //============================================================================================= |
---|
922 | // Kernel Utilities |
---|
923 | //============================================================================================= |
---|
924 | //----------------------------------------------------------------------------- |
---|
925 | // Locks |
---|
926 | void ?{}( semaphore & this, int count = 1 ) { |
---|
927 | (this.lock){}; |
---|
928 | this.count = count; |
---|
929 | (this.waiting){}; |
---|
930 | } |
---|
931 | void ^?{}(semaphore & this) {} |
---|
932 | |
---|
933 | void P(semaphore & this) with( this ){ |
---|
934 | lock( lock __cfaabi_dbg_ctx2 ); |
---|
935 | count -= 1; |
---|
936 | if ( count < 0 ) { |
---|
937 | // queue current task |
---|
938 | append( waiting, kernelTLS.this_thread ); |
---|
939 | |
---|
940 | // atomically release spin lock and block |
---|
941 | BlockInternal( &lock ); |
---|
942 | } |
---|
943 | else { |
---|
944 | unlock( lock ); |
---|
945 | } |
---|
946 | } |
---|
947 | |
---|
948 | void V(semaphore & this) with( this ) { |
---|
949 | thread_desc * thrd = 0p; |
---|
950 | lock( lock __cfaabi_dbg_ctx2 ); |
---|
951 | count += 1; |
---|
952 | if ( count <= 0 ) { |
---|
953 | // remove task at head of waiting list |
---|
954 | thrd = pop_head( waiting ); |
---|
955 | } |
---|
956 | |
---|
957 | unlock( lock ); |
---|
958 | |
---|
959 | // make new owner |
---|
960 | WakeThread( thrd, false ); |
---|
961 | } |
---|
962 | |
---|
963 | //----------------------------------------------------------------------------- |
---|
964 | // Global Queues |
---|
965 | void doregister( cluster & cltr ) { |
---|
966 | lock ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2); |
---|
967 | push_front( __cfa_dbg_global_clusters.list, cltr ); |
---|
968 | unlock ( __cfa_dbg_global_clusters.lock ); |
---|
969 | } |
---|
970 | |
---|
971 | void unregister( cluster & cltr ) { |
---|
972 | lock ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2); |
---|
973 | remove( __cfa_dbg_global_clusters.list, cltr ); |
---|
974 | unlock( __cfa_dbg_global_clusters.lock ); |
---|
975 | } |
---|
976 | |
---|
977 | void doregister( cluster * cltr, thread_desc & thrd ) { |
---|
978 | lock (cltr->thread_list_lock __cfaabi_dbg_ctx2); |
---|
979 | cltr->nthreads += 1; |
---|
980 | push_front(cltr->threads, thrd); |
---|
981 | unlock (cltr->thread_list_lock); |
---|
982 | } |
---|
983 | |
---|
984 | void unregister( cluster * cltr, thread_desc & thrd ) { |
---|
985 | lock (cltr->thread_list_lock __cfaabi_dbg_ctx2); |
---|
986 | remove(cltr->threads, thrd ); |
---|
987 | cltr->nthreads -= 1; |
---|
988 | unlock(cltr->thread_list_lock); |
---|
989 | } |
---|
990 | |
---|
991 | void doregister( cluster * cltr, processor * proc ) { |
---|
992 | lock (cltr->proc_list_lock __cfaabi_dbg_ctx2); |
---|
993 | cltr->nprocessors += 1; |
---|
994 | push_front(cltr->procs, *proc); |
---|
995 | unlock (cltr->proc_list_lock); |
---|
996 | } |
---|
997 | |
---|
998 | void unregister( cluster * cltr, processor * proc ) { |
---|
999 | lock (cltr->proc_list_lock __cfaabi_dbg_ctx2); |
---|
1000 | remove(cltr->procs, *proc ); |
---|
1001 | cltr->nprocessors -= 1; |
---|
1002 | unlock(cltr->proc_list_lock); |
---|
1003 | } |
---|
1004 | |
---|
1005 | //----------------------------------------------------------------------------- |
---|
1006 | // Debug |
---|
1007 | __cfaabi_dbg_debug_do( |
---|
1008 | extern "C" { |
---|
1009 | void __cfaabi_dbg_record(__spinlock_t & this, const char * prev_name) { |
---|
1010 | this.prev_name = prev_name; |
---|
1011 | this.prev_thrd = kernelTLS.this_thread; |
---|
1012 | } |
---|
1013 | } |
---|
1014 | ) |
---|
1015 | |
---|
1016 | //----------------------------------------------------------------------------- |
---|
1017 | // Debug |
---|
1018 | bool threading_enabled(void) { |
---|
1019 | return true; |
---|
1020 | } |
---|
1021 | // Local Variables: // |
---|
1022 | // mode: c // |
---|
1023 | // tab-width: 4 // |
---|
1024 | // End: // |
---|