Context Navigation

kernel.cfa @ 70ac8d0

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationnew-astnew-ast-unique-exprpthread-emulationqualifiedEnum

Last change on this file since 70ac8d0 was 70ac8d0, checked in by Thierry Delisle <tdelisle@…>, 4 years ago
Merge branch 'master' of plg.uwaterloo.ca:software/cfa/cfa-cc
Property mode set to `100644`
File size: 35.2 KB

Line
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 : Tue Feb 4 13:03:15 2020
13	// Update Count : 58
14	//
15
16	#define __cforall_thread__
17	// #define __CFA_DEBUG_PRINT_RUNTIME_CORE__
18
19	//C Includes
20	#include <stddef.h>
21	#include <errno.h>
22	#include <string.h>
23	extern "C" {
24	#include <stdio.h>
25	#include <fenv.h>
26	#include <sys/resource.h>
27	#include <signal.h>
28	#include <unistd.h>
29	#include <limits.h> // PTHREAD_STACK_MIN
30	#include <sys/mman.h> // mprotect
31	}
32
33	//CFA Includes
34	#include "time.hfa"
35	#include "kernel_private.hfa"
36	#include "preemption.hfa"
37	#include "startup.hfa"
38
39	//Private includes
40	#define __CFA_INVOKE_PRIVATE__
41	#include "invoke.h"
42
43
44	//-----------------------------------------------------------------------------
45	// Some assembly required
46	#if defined( __i386 )
47	#define CtxGet( ctx ) \
48	__asm__ volatile ( \
49	"movl %%esp,%0\n"\
50	"movl %%ebp,%1\n"\
51	: "=rm" (ctx.SP),\
52	"=rm" (ctx.FP) \
53	)
54
55	// mxcr : SSE Status and Control bits (control bits are preserved across function calls)
56	// fcw : X87 FPU control word (preserved across function calls)
57	#define __x87_store \
58	uint32_t __mxcr; \
59	uint16_t __fcw; \
60	__asm__ volatile ( \
61	"stmxcsr %0\n" \
62	"fnstcw %1\n" \
63	: "=m" (__mxcr),\
64	"=m" (__fcw) \
65	)
66
67	#define __x87_load \
68	__asm__ volatile ( \
69	"fldcw %1\n" \
70	"ldmxcsr %0\n" \
71	::"m" (__mxcr),\
72	"m" (__fcw) \
73	)
74
75	#elif defined( __x86_64 )
76	#define CtxGet( ctx ) \
77	__asm__ volatile ( \
78	"movq %%rsp,%0\n"\
79	"movq %%rbp,%1\n"\
80	: "=rm" (ctx.SP),\
81	"=rm" (ctx.FP) \
82	)
83
84	#define __x87_store \
85	uint32_t __mxcr; \
86	uint16_t __fcw; \
87	__asm__ volatile ( \
88	"stmxcsr %0\n" \
89	"fnstcw %1\n" \
90	: "=m" (__mxcr),\
91	"=m" (__fcw) \
92	)
93
94	#define __x87_load \
95	__asm__ volatile ( \
96	"fldcw %1\n" \
97	"ldmxcsr %0\n" \
98	:: "m" (__mxcr),\
99	"m" (__fcw) \
100	)
101
102
103	#elif defined( __ARM_ARCH )
104	#define CtxGet( ctx ) __asm__ ( \
105	"mov %0,%%sp\n" \
106	"mov %1,%%r11\n" \
107	: "=rm" (ctx.SP), "=rm" (ctx.FP) )
108	#else
109	#error unknown hardware architecture
110	#endif
111
112	//-----------------------------------------------------------------------------
113	//Start and stop routine for the kernel, declared first to make sure they run first
114	static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
115	static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
116
117	//-----------------------------------------------------------------------------
118	// Kernel Scheduling logic
119	static $thread * __next_thread(cluster * this);
120	static void __run_thread(processor * this, $thread * dst);
121	static $thread * __halt(processor * this);
122	static bool __wake_one(cluster * cltr, bool was_empty);
123	static bool __wake_proc(processor *);
124
125	//-----------------------------------------------------------------------------
126	// Kernel storage
127	KERNEL_STORAGE(cluster, mainCluster);
128	KERNEL_STORAGE(processor, mainProcessor);
129	KERNEL_STORAGE($thread, mainThread);
130	KERNEL_STORAGE(__stack_t, mainThreadCtx);
131
132	cluster * mainCluster;
133	processor * mainProcessor;
134	$thread * mainThread;
135
136	extern "C" {
137	struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
138	}
139
140	size_t __page_size = 0;
141
142	//-----------------------------------------------------------------------------
143	// Global state
144	thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) = {
145	NULL, // cannot use 0p
146	NULL,
147	{ 1, false, false },
148	6u //this should be seeded better but due to a bug calling rdtsc doesn't work
149	};
150
151	//-----------------------------------------------------------------------------
152	// Struct to steal stack
153	struct current_stack_info_t {
154	__stack_t * storage; // pointer to stack object
155	void * base; // base of stack
156	void * limit; // stack grows towards stack limit
157	void * context; // address of cfa_context_t
158	};
159
160	void ?{}( current_stack_info_t & this ) {
161	__stack_context_t ctx;
162	CtxGet( ctx );
163	this.base = ctx.FP;
164
165	rlimit r;
166	getrlimit( RLIMIT_STACK, &r);
167	size_t size = r.rlim_cur;
168
169	this.limit = (void *)(((intptr_t)this.base) - size);
170	this.context = &storage_mainThreadCtx;
171	}
172
173	//-----------------------------------------------------------------------------
174	// Main thread construction
175
176	void ?{}( $coroutine & this, current_stack_info_t * info) with( this ) {
177	stack.storage = info->storage;
178	with(*stack.storage) {
179	limit = info->limit;
180	base = info->base;
181	}
182	__attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
183	*istorage \|= 0x1;
184	name = "Main Thread";
185	state = Start;
186	starter = 0p;
187	last = 0p;
188	cancellation = 0p;
189	}
190
191	void ?{}( $thread & this, current_stack_info_t * info) with( this ) {
192	state = Start;
193	self_cor{ info };
194	curr_cor = &self_cor;
195	curr_cluster = mainCluster;
196	self_mon.owner = &this;
197	self_mon.recursion = 1;
198	self_mon_p = &self_mon;
199	next = 0p;
200
201	node.next = 0p;
202	node.prev = 0p;
203	doregister(curr_cluster, this);
204
205	monitors{ &self_mon_p, 1, (fptr_t)0 };
206	}
207
208	//-----------------------------------------------------------------------------
209	// Processor coroutine
210	void ?{}(processorCtx_t & this) {
211
212	}
213
214	// Construct the processor context of non-main processors
215	static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
216	(this.__cor){ info };
217	this.proc = proc;
218	}
219
220	static void * __invoke_processor(void * arg);
221
222	void ?{}(processor & this, const char name[], cluster & cltr) with( this ) {
223	this.name = name;
224	this.cltr = &cltr;
225	terminated{ 0 };
226	destroyer = 0p;
227	do_terminate = false;
228	preemption_alarm = 0p;
229	pending_preemption = false;
230	runner.proc = &this;
231
232	idle{};
233
234	__cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
235
236	this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
237
238	__cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
239	}
240
241	void ^?{}(processor & this) with( this ){
242	if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
243	__cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
244
245	__atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
246	__wake_proc( &this );
247
248	P( terminated );
249	verify( kernelTLS.this_processor != &this);
250	}
251
252	int err = pthread_join( kernel_thread, 0p );
253	if( err != 0 ) abort("KERNEL ERROR: joining processor %p caused error %s\n", &this, strerror(err));
254
255	free( this.stack );
256	}
257
258	void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned io_flags) with( this ) {
259	this.name = name;
260	this.preemption_rate = preemption_rate;
261	ready_queue{};
262	ready_queue_lock{};
263
264	#if !defined(__CFA_NO_STATISTICS__)
265	print_stats = false;
266	#endif
267
268	procs{ __get };
269	idles{ __get };
270	threads{ __get };
271
272	__kernel_io_startup( this, io_flags, &this == mainCluster );
273
274	doregister(this);
275	}
276
277	void ^?{}(cluster & this) {
278	__kernel_io_shutdown( this, &this == mainCluster );
279
280	unregister(this);
281	}
282
283	//=============================================================================================
284	// Kernel Scheduling logic
285	//=============================================================================================
286	//Main of the processor contexts
287	void main(processorCtx_t & runner) {
288	// Because of a bug, we couldn't initialized the seed on construction
289	// Do it here
290	kernelTLS.rand_seed ^= rdtscl();
291
292	processor * this = runner.proc;
293	verify(this);
294
295	__cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
296
297	doregister(this->cltr, this);
298
299	{
300	// Setup preemption data
301	preemption_scope scope = { this };
302
303	__cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
304
305	$thread * readyThread = 0p;
306	for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) {
307	// Try to get the next thread
308	readyThread = __next_thread( this->cltr );
309
310	// If no ready thread
311	if( readyThread == 0p ) {
312	// Block until a thread is ready
313	readyThread = __halt(this);
314	}
315
316	// Check if we actually found a thread
317	if( readyThread ) {
318	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
319	/* paranoid */ verifyf( readyThread->state == Ready \|\| readyThread->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", readyThread->state, readyThread->preempted);
320	/* paranoid */ verifyf( readyThread->next == 0p, "Expected null got %p", readyThread->next );
321
322	// We found a thread run it
323	__run_thread(this, readyThread);
324
325	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
326	}
327	}
328
329	__cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
330	}
331
332	unregister(this->cltr, this);
333
334	V( this->terminated );
335
336	__cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
337
338	// HACK : the coroutine context switch expects this_thread to be set
339	// and it make sense for it to be set in all other cases except here
340	// fake it
341	if( this == mainProcessor ) kernelTLS.this_thread = mainThread;
342	}
343
344	static int * __volatile_errno() __attribute__((noinline));
345	static int * __volatile_errno() { asm(""); return &errno; }
346
347	// KERNEL ONLY
348	// runThread runs a thread by context switching
349	// from the processor coroutine to the target thread
350	static void __run_thread(processor * this, $thread * thrd_dst) {
351	$coroutine * proc_cor = get_coroutine(this->runner);
352
353	// Update global state
354	kernelTLS.this_thread = thrd_dst;
355
356	// set state of processor coroutine to inactive
357	verify(proc_cor->state == Active);
358	proc_cor->state = Blocked;
359
360	// Actually run the thread
361	RUNNING: while(true) {
362	if(unlikely(thrd_dst->preempted)) {
363	thrd_dst->preempted = __NO_PREEMPTION;
364	verify(thrd_dst->state == Active \|\| thrd_dst->state == Rerun);
365	} else {
366	verify(thrd_dst->state == Blocked \|\| thrd_dst->state == Ready); // Ready means scheduled normally, blocked means rerun
367	thrd_dst->state = Active;
368	}
369
370	__cfaabi_dbg_debug_do(
371	thrd_dst->park_stale = true;
372	thrd_dst->unpark_stale = true;
373	)
374
375	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
376	/* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
377	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ) \|\| thrd_dst->curr_cor == proc_cor, "ERROR : Destination $thread %p has been corrupted.\n StackPointer too small.\n", thrd_dst ); // add escape condition if we are setting up the processor
378	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit) \|\| thrd_dst->curr_cor == proc_cor, "ERROR : Destination $thread %p has been corrupted.\n StackPointer too large.\n", thrd_dst ); // add escape condition if we are setting up the processor
379
380	// set context switch to the thread that the processor is executing
381	verify( thrd_dst->context.SP );
382	__cfactx_switch( &proc_cor->context, &thrd_dst->context );
383	// when __cfactx_switch returns we are back in the processor coroutine
384
385	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit), "ERROR : Destination $thread %p has been corrupted.\n StackPointer too large.\n", thrd_dst );
386	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ), "ERROR : Destination $thread %p has been corrupted.\n StackPointer too small.\n", thrd_dst );
387	/* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
388	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
389
390
391	// We just finished running a thread, there are a few things that could have happened.
392	// 1 - Regular case : the thread has blocked and now one has scheduled it yet.
393	// 2 - Racy case : the thread has blocked but someone has already tried to schedule it.
394	// 4 - Preempted
395	// In case 1, we may have won a race so we can't write to the state again.
396	// In case 2, we lost the race so we now own the thread.
397
398	if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
399	// The thread was preempted, reschedule it and reset the flag
400	__schedule_thread( thrd_dst );
401	break RUNNING;
402	}
403
404	// set state of processor coroutine to active and the thread to inactive
405	static_assert(sizeof(thrd_dst->state) == sizeof(int));
406	enum coroutine_state old_state = __atomic_exchange_n(&thrd_dst->state, Blocked, __ATOMIC_SEQ_CST);
407	__cfaabi_dbg_debug_do( thrd_dst->park_result = old_state; )
408	switch(old_state) {
409	case Halted:
410	// The thread has halted, it should never be scheduled/run again, leave it back to Halted and move on
411	thrd_dst->state = Halted;
412
413	// We may need to wake someone up here since
414	unpark( this->destroyer __cfaabi_dbg_ctx2 );
415	this->destroyer = 0p;
416	break RUNNING;
417	case Active:
418	// This is case 1, the regular case, nothing more is needed
419	break RUNNING;
420	case Rerun:
421	// This is case 2, the racy case, someone tried to run this thread before it finished blocking
422	// In this case, just run it again.
423	continue RUNNING;
424	default:
425	// This makes no sense, something is wrong abort
426	abort("Finished running a thread that was Blocked/Start/Primed %d\n", old_state);
427	}
428	}
429
430	// Just before returning to the processor, set the processor coroutine to active
431	proc_cor->state = Active;
432	kernelTLS.this_thread = 0p;
433	}
434
435	// KERNEL_ONLY
436	void returnToKernel() {
437	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
438	$coroutine * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
439	$thread * thrd_src = kernelTLS.this_thread;
440
441	// Run the thread on this processor
442	{
443	int local_errno = *__volatile_errno();
444	#if defined( __i386 ) \|\| defined( __x86_64 )
445	__x87_store;
446	#endif
447	verify( proc_cor->context.SP );
448	__cfactx_switch( &thrd_src->context, &proc_cor->context );
449	#if defined( __i386 ) \|\| defined( __x86_64 )
450	__x87_load;
451	#endif
452	*__volatile_errno() = local_errno;
453	}
454
455	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
456	/* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) < ((uintptr_t)__get_stack(thrd_src->curr_cor)->base ), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too small.\n", thrd_src );
457	/* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) > ((uintptr_t)__get_stack(thrd_src->curr_cor)->limit), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too large.\n", thrd_src );
458	}
459
460	// KERNEL_ONLY
461	// Context invoker for processors
462	// This is the entry point for processors (kernel threads)
463	// It effectively constructs a coroutine by stealing the pthread stack
464	static void * __invoke_processor(void * arg) {
465	processor * proc = (processor *) arg;
466	kernelTLS.this_processor = proc;
467	kernelTLS.this_thread = 0p;
468	kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
469	// SKULLDUGGERY: We want to create a context for the processor coroutine
470	// which is needed for the 2-step context switch. However, there is no reason
471	// to waste the perfectly valid stack create by pthread.
472	current_stack_info_t info;
473	__stack_t ctx;
474	info.storage = &ctx;
475	(proc->runner){ proc, &info };
476
477	__cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
478
479	//Set global state
480	kernelTLS.this_thread = 0p;
481
482	//We now have a proper context from which to schedule threads
483	__cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
484
485	// SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
486	// resume it to start it like it normally would, it will just context switch
487	// back to here. Instead directly call the main since we already are on the
488	// appropriate stack.
489	get_coroutine(proc->runner)->state = Active;
490	main( proc->runner );
491	get_coroutine(proc->runner)->state = Halted;
492
493	// Main routine of the core returned, the core is now fully terminated
494	__cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
495
496	return 0p;
497	}
498
499	static void Abort( int ret, const char func[] ) {
500	if ( ret ) { // pthread routines return errno values
501	abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
502	} // if
503	} // Abort
504
505	void * __create_pthread( pthread_t * pthread, void * (start)(void ), void * arg ) {
506	pthread_attr_t attr;
507
508	Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
509
510	size_t stacksize;
511	// default stack size, normally defined by shell limit
512	Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
513	assert( stacksize >= PTHREAD_STACK_MIN );
514
515	void * stack;
516	__cfaabi_dbg_debug_do(
517	stack = memalign( __page_size, stacksize + __page_size );
518	// pthread has no mechanism to create the guard page in user supplied stack.
519	if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
520	abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
521	} // if
522	);
523	__cfaabi_dbg_no_debug_do(
524	stack = malloc( stacksize );
525	);
526
527	Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
528
529	Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
530	return stack;
531	}
532
533	// KERNEL_ONLY
534	static void __kernel_first_resume( processor * this ) {
535	$thread * src = mainThread;
536	$coroutine * dst = get_coroutine(this->runner);
537
538	verify( ! kernelTLS.preemption_state.enabled );
539
540	kernelTLS.this_thread->curr_cor = dst;
541	__stack_prepare( &dst->stack, 65000 );
542	__cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
543
544	verify( ! kernelTLS.preemption_state.enabled );
545
546	dst->last = &src->self_cor;
547	dst->starter = dst->starter ? dst->starter : &src->self_cor;
548
549	// make sure the current state is still correct
550	/* paranoid */ verify(src->state == Ready);
551
552	// context switch to specified coroutine
553	verify( dst->context.SP );
554	__cfactx_switch( &src->context, &dst->context );
555	// when __cfactx_switch returns we are back in the src coroutine
556
557	mainThread->curr_cor = &mainThread->self_cor;
558
559	// make sure the current state has been update
560	/* paranoid */ verify(src->state == Active);
561
562	verify( ! kernelTLS.preemption_state.enabled );
563	}
564
565	// KERNEL_ONLY
566	static void __kernel_last_resume( processor * this ) {
567	$coroutine * src = &mainThread->self_cor;
568	$coroutine * dst = get_coroutine(this->runner);
569
570	verify( ! kernelTLS.preemption_state.enabled );
571	verify( dst->starter == src );
572	verify( dst->context.SP );
573
574	// SKULLDUGGERY in debug the processors check that the
575	// stack is still within the limit of the stack limits after running a thread.
576	// that check doesn't make sense if we context switch to the processor using the
577	// coroutine semantics. Since this is a special case, use the current context
578	// info to populate these fields.
579	__cfaabi_dbg_debug_do(
580	__stack_context_t ctx;
581	CtxGet( ctx );
582	mainThread->context.SP = ctx.SP;
583	mainThread->context.FP = ctx.FP;
584	)
585
586	// context switch to the processor
587	__cfactx_switch( &src->context, &dst->context );
588	}
589
590	//-----------------------------------------------------------------------------
591	// Scheduler routines
592	// KERNEL ONLY
593	void __schedule_thread( $thread * thrd ) with( *thrd->curr_cluster ) {
594	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
595	/* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
596	/* paranoid */ if( thrd->state == Blocked \|\| thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
597	"Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
598	/* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active \|\| thrd->state == Rerun,
599	"Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
600	/* paranoid */ #endif
601	/* paranoid */ verifyf( thrd->next == 0p, "Expected null got %p", thrd->next );
602
603	if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
604
605	lock ( ready_queue_lock __cfaabi_dbg_ctx2 );
606	bool was_empty = !(ready_queue != 0);
607	append( ready_queue, thrd );
608	unlock( ready_queue_lock );
609
610	__wake_one(thrd->curr_cluster, was_empty);
611
612	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
613	}
614
615	// KERNEL ONLY
616	static $thread * __next_thread(cluster * this) with( *this ) {
617	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
618
619	lock( ready_queue_lock __cfaabi_dbg_ctx2 );
620	$thread * head = pop_head( ready_queue );
621	unlock( ready_queue_lock );
622
623	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
624	return head;
625	}
626
627	// KERNEL ONLY unpark with out disabling interrupts
628	void __unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
629	static_assert(sizeof(thrd->state) == sizeof(int));
630
631	// record activity
632	__cfaabi_dbg_record_thrd( *thrd, false, caller );
633
634	enum coroutine_state old_state = __atomic_exchange_n(&thrd->state, Rerun, __ATOMIC_SEQ_CST);
635	__cfaabi_dbg_debug_do( thrd->unpark_result = old_state; )
636	switch(old_state) {
637	case Active:
638	// Wake won the race, the thread will reschedule/rerun itself
639	break;
640	case Blocked:
641	/* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
642
643	// Wake lost the race,
644	thrd->state = Blocked;
645	__schedule_thread( thrd );
646	break;
647	case Rerun:
648	abort("More than one thread attempted to schedule thread %p\n", thrd);
649	break;
650	case Halted:
651	case Start:
652	case Primed:
653	default:
654	// This makes no sense, something is wrong abort
655	abort();
656	}
657	}
658
659	void unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
660	if( !thrd ) return;
661
662	disable_interrupts();
663	__unpark( thrd __cfaabi_dbg_ctx_fwd2 );
664	enable_interrupts( __cfaabi_dbg_ctx );
665	}
666
667	void park( __cfaabi_dbg_ctx_param ) {
668	/* paranoid */ verify( kernelTLS.preemption_state.enabled );
669	disable_interrupts();
670	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
671	/* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );
672
673	// record activity
674	__cfaabi_dbg_record_thrd( *kernelTLS.this_thread, true, caller );
675
676	returnToKernel();
677
678	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
679	enable_interrupts( __cfaabi_dbg_ctx );
680	/* paranoid */ verify( kernelTLS.preemption_state.enabled );
681
682	}
683
684	// KERNEL ONLY
685	void __leave_thread() {
686	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
687	returnToKernel();
688	abort();
689	}
690
691	// KERNEL ONLY
692	bool force_yield( __Preemption_Reason reason ) {
693	/* paranoid */ verify( kernelTLS.preemption_state.enabled );
694	disable_interrupts();
695	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
696
697	$thread * thrd = kernelTLS.this_thread;
698	/* paranoid */ verify(thrd->state == Active \|\| thrd->state == Rerun);
699
700	// SKULLDUGGERY: It is possible that we are preempting this thread just before
701	// it was going to park itself. If that is the case and it is already using the
702	// intrusive fields then we can't use them to preempt the thread
703	// If that is the case, abandon the preemption.
704	bool preempted = false;
705	if(thrd->next == 0p) {
706	preempted = true;
707	thrd->preempted = reason;
708	returnToKernel();
709	}
710
711	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
712	enable_interrupts_noPoll();
713	/* paranoid */ verify( kernelTLS.preemption_state.enabled );
714
715	return preempted;
716	}
717
718	//=============================================================================================
719	// Kernel Setup logic
720	//=============================================================================================
721	//-----------------------------------------------------------------------------
722	// Kernel boot procedures
723	static void __kernel_startup(void) {
724	verify( ! kernelTLS.preemption_state.enabled );
725	__cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
726
727	__page_size = sysconf( _SC_PAGESIZE );
728
729	__cfa_dbg_global_clusters.list{ __get };
730	__cfa_dbg_global_clusters.lock{};
731
732	// Initialize the main cluster
733	mainCluster = (cluster *)&storage_mainCluster;
734	(*mainCluster){"Main Cluster"};
735
736	__cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
737
738	// Start by initializing the main thread
739	// SKULLDUGGERY: the mainThread steals the process main thread
740	// which will then be scheduled by the mainProcessor normally
741	mainThread = ($thread *)&storage_mainThread;
742	current_stack_info_t info;
743	info.storage = (__stack_t*)&storage_mainThreadCtx;
744	(*mainThread){ &info };
745
746	__cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
747
748
749
750	// Construct the processor context of the main processor
751	void ?{}(processorCtx_t & this, processor * proc) {
752	(this.__cor){ "Processor" };
753	this.__cor.starter = 0p;
754	this.proc = proc;
755	}
756
757	void ?{}(processor & this) with( this ) {
758	name = "Main Processor";
759	cltr = mainCluster;
760	terminated{ 0 };
761	do_terminate = false;
762	preemption_alarm = 0p;
763	pending_preemption = false;
764	kernel_thread = pthread_self();
765
766	runner{ &this };
767	__cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
768	}
769
770	// Initialize the main processor and the main processor ctx
771	// (the coroutine that contains the processing control flow)
772	mainProcessor = (processor *)&storage_mainProcessor;
773	(*mainProcessor){};
774
775	//initialize the global state variables
776	kernelTLS.this_processor = mainProcessor;
777	kernelTLS.this_thread = mainThread;
778
779	// Enable preemption
780	kernel_start_preemption();
781
782	// Add the main thread to the ready queue
783	// once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
784	__schedule_thread(mainThread);
785
786	// SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
787	// context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
788	// mainThread is on the ready queue when this call is made.
789	__kernel_first_resume( kernelTLS.this_processor );
790
791
792	// THE SYSTEM IS NOW COMPLETELY RUNNING
793
794
795	// Now that the system is up, finish creating systems that need threading
796	__kernel_io_finish_start( *mainCluster );
797
798
799	__cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
800
801	verify( ! kernelTLS.preemption_state.enabled );
802	enable_interrupts( __cfaabi_dbg_ctx );
803	verify( TL_GET( preemption_state.enabled ) );
804	}
805
806	static void __kernel_shutdown(void) {
807	//Before we start shutting things down, wait for systems that need threading to shutdown
808	__kernel_io_prepare_stop( *mainCluster );
809
810	/* paranoid */ verify( TL_GET( preemption_state.enabled ) );
811	disable_interrupts();
812	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
813
814	__cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
815
816	// SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
817	// When its coroutine terminates, it return control to the mainThread
818	// which is currently here
819	__atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
820	__kernel_last_resume( kernelTLS.this_processor );
821	mainThread->self_cor.state = Halted;
822
823	// THE SYSTEM IS NOW COMPLETELY STOPPED
824
825	// Disable preemption
826	kernel_stop_preemption();
827
828	// Destroy the main processor and its context in reverse order of construction
829	// These were manually constructed so we need manually destroy them
830	void ^?{}(processor & this) with( this ){
831	/* paranoid */ verify( this.do_terminate == true );
832	}
833
834	^(*mainProcessor){};
835
836	// Final step, destroy the main thread since it is no longer needed
837	// Since we provided a stack to this taxk it will not destroy anything
838	/* paranoid / verify(mainThread->self_cor.stack.storage == (__stack_t)(((uintptr_t)&storage_mainThreadCtx)\| 0x1));
839	^(*mainThread){};
840
841	^(*mainCluster){};
842
843	^(__cfa_dbg_global_clusters.list){};
844	^(__cfa_dbg_global_clusters.lock){};
845
846	__cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
847	}
848
849	//=============================================================================================
850	// Kernel Idle Sleep
851	//=============================================================================================
852	static $thread * __halt(processor * this) with( *this ) {
853	if( do_terminate ) return 0p;
854
855	// First, lock the cluster idle
856	lock( cltr->idle_lock __cfaabi_dbg_ctx2 );
857
858	// Check if we can find a thread
859	if( $thread * found = __next_thread( cltr ) ) {
860	unlock( cltr->idle_lock );
861	return found;
862	}
863
864	// Move this processor from the active list to the idle list
865	move_to_front(cltr->procs, cltr->idles, *this);
866
867	// Unlock the idle lock so we don't go to sleep with a lock
868	unlock (cltr->idle_lock);
869
870	// We are ready to sleep
871	__cfadbg_print_safe(runtime_core, "Kernel : Processor %p ready to sleep\n", this);
872	wait( idle );
873
874	// We have woken up
875	__cfadbg_print_safe(runtime_core, "Kernel : Processor %p woke up and ready to run\n", this);
876
877	// Get ourself off the idle list
878	with( *cltr ) {
879	lock (idle_lock __cfaabi_dbg_ctx2);
880	move_to_front(idles, procs, *this);
881	unlock(idle_lock);
882	}
883
884	// Don't check the ready queue again, we may not be in a position to run a thread
885	return 0p;
886	}
887
888	// Wake a thread from the front if there are any
889	static bool __wake_one(cluster * this, __attribute__((unused)) bool force) {
890	// if we don't want to force check if we know it's false
891	// if( !this->idles.head && !force ) return false;
892
893	// First, lock the cluster idle
894	lock( this->idle_lock __cfaabi_dbg_ctx2 );
895
896	// Check if there is someone to wake up
897	if( !this->idles.head ) {
898	// Nope unlock and return false
899	unlock( this->idle_lock );
900	return false;
901	}
902
903	// Wake them up
904	__cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this->idles.head);
905	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
906	post( this->idles.head->idle );
907
908	// Unlock and return true
909	unlock( this->idle_lock );
910	return true;
911	}
912
913	// Unconditionnaly wake a thread
914	static bool __wake_proc(processor * this) {
915	__cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
916
917	disable_interrupts();
918	/* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
919	bool ret = post( this->idle );
920	enable_interrupts( __cfaabi_dbg_ctx );
921
922	return ret;
923	}
924
925	//=============================================================================================
926	// Unexpected Terminating logic
927	//=============================================================================================
928	static __spinlock_t kernel_abort_lock;
929	static bool kernel_abort_called = false;
930
931	void * kernel_abort(void) __attribute__ ((__nothrow__)) {
932	// abort cannot be recursively entered by the same or different processors because all signal handlers return when
933	// the globalAbort flag is true.
934	lock( kernel_abort_lock __cfaabi_dbg_ctx2 );
935
936	// first task to abort ?
937	if ( kernel_abort_called ) { // not first task to abort ?
938	unlock( kernel_abort_lock );
939
940	sigset_t mask;
941	sigemptyset( &mask );
942	sigaddset( &mask, SIGALRM ); // block SIGALRM signals
943	sigaddset( &mask, SIGUSR1 ); // block SIGALRM signals
944	sigsuspend( &mask ); // block the processor to prevent further damage during abort
945	_exit( EXIT_FAILURE ); // if processor unblocks before it is killed, terminate it
946	}
947	else {
948	kernel_abort_called = true;
949	unlock( kernel_abort_lock );
950	}
951
952	return kernelTLS.this_thread;
953	}
954
955	void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) {
956	$thread * thrd = kernel_data;
957
958	if(thrd) {
959	int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
960	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
961
962	if ( &thrd->self_cor != thrd->curr_cor ) {
963	len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
964	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
965	}
966	else {
967	__cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
968	}
969	}
970	else {
971	int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
972	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
973	}
974	}
975
976	int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
977	return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2;
978	}
979
980	static __spinlock_t kernel_debug_lock;
981
982	extern "C" {
983	void __cfaabi_bits_acquire() {
984	lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
985	}
986
987	void __cfaabi_bits_release() {
988	unlock( kernel_debug_lock );
989	}
990	}
991
992	//=============================================================================================
993	// Kernel Utilities
994	//=============================================================================================
995	//-----------------------------------------------------------------------------
996	// Locks
997	void ?{}( semaphore & this, int count = 1 ) {
998	(this.lock){};
999	this.count = count;
1000	(this.waiting){};
1001	}
1002	void ^?{}(semaphore & this) {}
1003
1004	bool P(semaphore & this) with( this ){
1005	lock( lock __cfaabi_dbg_ctx2 );
1006	count -= 1;
1007	if ( count < 0 ) {
1008	// queue current task
1009	append( waiting, kernelTLS.this_thread );
1010
1011	// atomically release spin lock and block
1012	unlock( lock );
1013	park( __cfaabi_dbg_ctx );
1014	return true;
1015	}
1016	else {
1017	unlock( lock );
1018	return false;
1019	}
1020	}
1021
1022	bool V(semaphore & this) with( this ) {
1023	$thread * thrd = 0p;
1024	lock( lock __cfaabi_dbg_ctx2 );
1025	count += 1;
1026	if ( count <= 0 ) {
1027	// remove task at head of waiting list
1028	thrd = pop_head( waiting );
1029	}
1030
1031	unlock( lock );
1032
1033	// make new owner
1034	unpark( thrd __cfaabi_dbg_ctx2 );
1035
1036	return thrd != 0p;
1037	}
1038
1039	bool V(semaphore & this, unsigned diff) with( this ) {
1040	$thread * thrd = 0p;
1041	lock( lock __cfaabi_dbg_ctx2 );
1042	int release = max(-count, (int)diff);
1043	count += diff;
1044	for(release) {
1045	unpark( pop_head( waiting ) __cfaabi_dbg_ctx2 );
1046	}
1047
1048	unlock( lock );
1049
1050	return thrd != 0p;
1051	}
1052
1053	//-----------------------------------------------------------------------------
1054	// Global Queues
1055	void doregister( cluster & cltr ) {
1056	lock ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
1057	push_front( __cfa_dbg_global_clusters.list, cltr );
1058	unlock ( __cfa_dbg_global_clusters.lock );
1059	}
1060
1061	void unregister( cluster & cltr ) {
1062	lock ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
1063	remove( __cfa_dbg_global_clusters.list, cltr );
1064	unlock( __cfa_dbg_global_clusters.lock );
1065	}
1066
1067	void doregister( cluster * cltr, $thread & thrd ) {
1068	lock (cltr->thread_list_lock __cfaabi_dbg_ctx2);
1069	cltr->nthreads += 1;
1070	push_front(cltr->threads, thrd);
1071	unlock (cltr->thread_list_lock);
1072	}
1073
1074	void unregister( cluster * cltr, $thread & thrd ) {
1075	lock (cltr->thread_list_lock __cfaabi_dbg_ctx2);
1076	remove(cltr->threads, thrd );
1077	cltr->nthreads -= 1;
1078	unlock(cltr->thread_list_lock);
1079	}
1080
1081	void doregister( cluster * cltr, processor * proc ) {
1082	lock (cltr->idle_lock __cfaabi_dbg_ctx2);
1083	cltr->nprocessors += 1;
1084	push_front(cltr->procs, *proc);
1085	unlock (cltr->idle_lock);
1086	}
1087
1088	void unregister( cluster * cltr, processor * proc ) {
1089	lock (cltr->idle_lock __cfaabi_dbg_ctx2);
1090	remove(cltr->procs, *proc );
1091	cltr->nprocessors -= 1;
1092	unlock(cltr->idle_lock);
1093	}
1094
1095	//-----------------------------------------------------------------------------
1096	// Debug
1097	__cfaabi_dbg_debug_do(
1098	extern "C" {
1099	void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
1100	this.prev_name = prev_name;
1101	this.prev_thrd = kernelTLS.this_thread;
1102	}
1103
1104	void __cfaabi_dbg_record_thrd($thread & this, bool park, const char prev_name[]) {
1105	if(park) {
1106	this.park_caller = prev_name;
1107	this.park_stale = false;
1108	}
1109	else {
1110	this.unpark_caller = prev_name;
1111	this.unpark_stale = false;
1112	}
1113	}
1114	}
1115	)
1116
1117	//-----------------------------------------------------------------------------
1118	// Debug
1119	bool threading_enabled(void) __attribute__((const)) {
1120	return true;
1121	}
1122	// Local Variables: //
1123	// mode: c //
1124	// tab-width: 4 //
1125	// End: //

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

Original Format