| 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 : Mon Apr  9 16:11:46 2018 | 
|---|
| 13 | // Update Count     : 24 | 
|---|
| 14 | // | 
|---|
| 15 |  | 
|---|
| 16 | //C Includes | 
|---|
| 17 | #include <stddef.h> | 
|---|
| 18 | #include <errno.h> | 
|---|
| 19 | #include <string.h> | 
|---|
| 20 | extern "C" { | 
|---|
| 21 | #include <stdio.h> | 
|---|
| 22 | #include <fenv.h> | 
|---|
| 23 | #include <sys/resource.h> | 
|---|
| 24 | #include <signal.h> | 
|---|
| 25 | #include <unistd.h> | 
|---|
| 26 | } | 
|---|
| 27 |  | 
|---|
| 28 | //CFA Includes | 
|---|
| 29 | #include "time.hfa" | 
|---|
| 30 | #include "kernel_private.hfa" | 
|---|
| 31 | #include "preemption.hfa" | 
|---|
| 32 | #include "startup.hfa" | 
|---|
| 33 |  | 
|---|
| 34 | //Private includes | 
|---|
| 35 | #define __CFA_INVOKE_PRIVATE__ | 
|---|
| 36 | #include "invoke.h" | 
|---|
| 37 |  | 
|---|
| 38 | //Start and stop routine for the kernel, declared first to make sure they run first | 
|---|
| 39 | static void kernel_startup(void)  __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) )); | 
|---|
| 40 | static void kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) )); | 
|---|
| 41 |  | 
|---|
| 42 | //----------------------------------------------------------------------------- | 
|---|
| 43 | // Kernel storage | 
|---|
| 44 | KERNEL_STORAGE(cluster,           mainCluster); | 
|---|
| 45 | KERNEL_STORAGE(processor,         mainProcessor); | 
|---|
| 46 | KERNEL_STORAGE(thread_desc,       mainThread); | 
|---|
| 47 | KERNEL_STORAGE(machine_context_t, mainThreadCtx); | 
|---|
| 48 |  | 
|---|
| 49 | cluster     * mainCluster; | 
|---|
| 50 | processor   * mainProcessor; | 
|---|
| 51 | thread_desc * mainThread; | 
|---|
| 52 |  | 
|---|
| 53 | extern "C" { | 
|---|
| 54 | struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters; | 
|---|
| 55 | } | 
|---|
| 56 |  | 
|---|
| 57 | //----------------------------------------------------------------------------- | 
|---|
| 58 | // Global state | 
|---|
| 59 | thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) = { | 
|---|
| 60 | NULL, | 
|---|
| 61 | NULL, | 
|---|
| 62 | NULL, | 
|---|
| 63 | { 1, false, false } | 
|---|
| 64 | }; | 
|---|
| 65 |  | 
|---|
| 66 | //----------------------------------------------------------------------------- | 
|---|
| 67 | // Struct to steal stack | 
|---|
| 68 | struct current_stack_info_t { | 
|---|
| 69 | machine_context_t ctx; | 
|---|
| 70 | unsigned int size;              // size of stack | 
|---|
| 71 | void *base;                             // base of stack | 
|---|
| 72 | void *storage;                  // pointer to stack | 
|---|
| 73 | void *limit;                    // stack grows towards stack limit | 
|---|
| 74 | void *context;                  // address of cfa_context_t | 
|---|
| 75 | void *top;                              // address of top of storage | 
|---|
| 76 | }; | 
|---|
| 77 |  | 
|---|
| 78 | void ?{}( current_stack_info_t & this ) { | 
|---|
| 79 | CtxGet( this.ctx ); | 
|---|
| 80 | this.base = this.ctx.FP; | 
|---|
| 81 | this.storage = this.ctx.SP; | 
|---|
| 82 |  | 
|---|
| 83 | rlimit r; | 
|---|
| 84 | getrlimit( RLIMIT_STACK, &r); | 
|---|
| 85 | this.size = r.rlim_cur; | 
|---|
| 86 |  | 
|---|
| 87 | this.limit = (void *)(((intptr_t)this.base) - this.size); | 
|---|
| 88 | this.context = &storage_mainThreadCtx; | 
|---|
| 89 | this.top = this.base; | 
|---|
| 90 | } | 
|---|
| 91 |  | 
|---|
| 92 | //----------------------------------------------------------------------------- | 
|---|
| 93 | // Main thread construction | 
|---|
| 94 | void ?{}( coStack_t & this, current_stack_info_t * info) with( this ) { | 
|---|
| 95 | size      = info->size; | 
|---|
| 96 | storage   = info->storage; | 
|---|
| 97 | limit     = info->limit; | 
|---|
| 98 | base      = info->base; | 
|---|
| 99 | context   = info->context; | 
|---|
| 100 | top       = info->top; | 
|---|
| 101 | userStack = true; | 
|---|
| 102 | } | 
|---|
| 103 |  | 
|---|
| 104 | void ?{}( coroutine_desc & this, current_stack_info_t * info) with( this ) { | 
|---|
| 105 | stack{ info }; | 
|---|
| 106 | name = "Main Thread"; | 
|---|
| 107 | errno_ = 0; | 
|---|
| 108 | state = Start; | 
|---|
| 109 | starter = NULL; | 
|---|
| 110 | } | 
|---|
| 111 |  | 
|---|
| 112 | void ?{}( thread_desc & this, current_stack_info_t * info) with( this ) { | 
|---|
| 113 | self_cor{ info }; | 
|---|
| 114 | curr_cor = &self_cor; | 
|---|
| 115 | curr_cluster = mainCluster; | 
|---|
| 116 | self_mon.owner = &this; | 
|---|
| 117 | self_mon.recursion = 1; | 
|---|
| 118 | self_mon_p = &self_mon; | 
|---|
| 119 | next = NULL; | 
|---|
| 120 |  | 
|---|
| 121 | node.next = NULL; | 
|---|
| 122 | node.prev = NULL; | 
|---|
| 123 | doregister(curr_cluster, this); | 
|---|
| 124 |  | 
|---|
| 125 | monitors{ &self_mon_p, 1, (fptr_t)0 }; | 
|---|
| 126 | } | 
|---|
| 127 |  | 
|---|
| 128 | //----------------------------------------------------------------------------- | 
|---|
| 129 | // Processor coroutine | 
|---|
| 130 | void ?{}(processorCtx_t & this) { | 
|---|
| 131 |  | 
|---|
| 132 | } | 
|---|
| 133 |  | 
|---|
| 134 | // Construct the processor context of non-main processors | 
|---|
| 135 | static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) { | 
|---|
| 136 | (this.__cor){ info }; | 
|---|
| 137 | this.proc = proc; | 
|---|
| 138 | } | 
|---|
| 139 |  | 
|---|
| 140 | static void start(processor * this); | 
|---|
| 141 | void ?{}(processor & this, const char * name, cluster & cltr) with( this ) { | 
|---|
| 142 | this.name = name; | 
|---|
| 143 | this.cltr = &cltr; | 
|---|
| 144 | terminated{ 0 }; | 
|---|
| 145 | do_terminate = false; | 
|---|
| 146 | preemption_alarm = NULL; | 
|---|
| 147 | pending_preemption = false; | 
|---|
| 148 | runner.proc = &this; | 
|---|
| 149 |  | 
|---|
| 150 | idleLock{}; | 
|---|
| 151 |  | 
|---|
| 152 | start( &this ); | 
|---|
| 153 | } | 
|---|
| 154 |  | 
|---|
| 155 | void ^?{}(processor & this) with( this ){ | 
|---|
| 156 | if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) { | 
|---|
| 157 | __cfaabi_dbg_print_safe("Kernel : core %p signaling termination\n", &this); | 
|---|
| 158 |  | 
|---|
| 159 | __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED); | 
|---|
| 160 | wake( &this ); | 
|---|
| 161 |  | 
|---|
| 162 | P( terminated ); | 
|---|
| 163 | verify( kernelTLS.this_processor != &this); | 
|---|
| 164 | } | 
|---|
| 165 |  | 
|---|
| 166 | pthread_join( kernel_thread, NULL ); | 
|---|
| 167 | } | 
|---|
| 168 |  | 
|---|
| 169 | void ?{}(cluster & this, const char * name, Duration preemption_rate) with( this ) { | 
|---|
| 170 | this.name = name; | 
|---|
| 171 | this.preemption_rate = preemption_rate; | 
|---|
| 172 | ready_queue{}; | 
|---|
| 173 | ready_queue_lock{}; | 
|---|
| 174 |  | 
|---|
| 175 | procs{ __get }; | 
|---|
| 176 | idles{ __get }; | 
|---|
| 177 | threads{ __get }; | 
|---|
| 178 |  | 
|---|
| 179 | doregister(this); | 
|---|
| 180 | } | 
|---|
| 181 |  | 
|---|
| 182 | void ^?{}(cluster & this) { | 
|---|
| 183 | unregister(this); | 
|---|
| 184 | } | 
|---|
| 185 |  | 
|---|
| 186 | //============================================================================================= | 
|---|
| 187 | // Kernel Scheduling logic | 
|---|
| 188 | //============================================================================================= | 
|---|
| 189 | static void runThread(processor * this, thread_desc * dst); | 
|---|
| 190 | static void finishRunning(processor * this); | 
|---|
| 191 | static void halt(processor * this); | 
|---|
| 192 |  | 
|---|
| 193 | //Main of the processor contexts | 
|---|
| 194 | void main(processorCtx_t & runner) { | 
|---|
| 195 | processor * this = runner.proc; | 
|---|
| 196 | verify(this); | 
|---|
| 197 |  | 
|---|
| 198 | __cfaabi_dbg_print_safe("Kernel : core %p starting\n", this); | 
|---|
| 199 |  | 
|---|
| 200 | doregister(this->cltr, this); | 
|---|
| 201 |  | 
|---|
| 202 | { | 
|---|
| 203 | // Setup preemption data | 
|---|
| 204 | preemption_scope scope = { this }; | 
|---|
| 205 |  | 
|---|
| 206 | __cfaabi_dbg_print_safe("Kernel : core %p started\n", this); | 
|---|
| 207 |  | 
|---|
| 208 | thread_desc * readyThread = NULL; | 
|---|
| 209 | for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) | 
|---|
| 210 | { | 
|---|
| 211 | readyThread = nextThread( this->cltr ); | 
|---|
| 212 |  | 
|---|
| 213 | if(readyThread) | 
|---|
| 214 | { | 
|---|
| 215 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 216 |  | 
|---|
| 217 | runThread(this, readyThread); | 
|---|
| 218 |  | 
|---|
| 219 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 220 |  | 
|---|
| 221 | //Some actions need to be taken from the kernel | 
|---|
| 222 | finishRunning(this); | 
|---|
| 223 |  | 
|---|
| 224 | spin_count = 0; | 
|---|
| 225 | } | 
|---|
| 226 | else | 
|---|
| 227 | { | 
|---|
| 228 | // spin(this, &spin_count); | 
|---|
| 229 | halt(this); | 
|---|
| 230 | } | 
|---|
| 231 | } | 
|---|
| 232 |  | 
|---|
| 233 | __cfaabi_dbg_print_safe("Kernel : core %p stopping\n", this); | 
|---|
| 234 | } | 
|---|
| 235 |  | 
|---|
| 236 | unregister(this->cltr, this); | 
|---|
| 237 |  | 
|---|
| 238 | V( this->terminated ); | 
|---|
| 239 |  | 
|---|
| 240 | __cfaabi_dbg_print_safe("Kernel : core %p terminated\n", this); | 
|---|
| 241 | } | 
|---|
| 242 |  | 
|---|
| 243 | // KERNEL ONLY | 
|---|
| 244 | // runThread runs a thread by context switching | 
|---|
| 245 | // from the processor coroutine to the target thread | 
|---|
| 246 | static void runThread(processor * this, thread_desc * dst) { | 
|---|
| 247 | assert(dst->curr_cor); | 
|---|
| 248 | coroutine_desc * proc_cor = get_coroutine(this->runner); | 
|---|
| 249 | coroutine_desc * thrd_cor = dst->curr_cor; | 
|---|
| 250 |  | 
|---|
| 251 | // Reset the terminating actions here | 
|---|
| 252 | this->finish.action_code = No_Action; | 
|---|
| 253 |  | 
|---|
| 254 | // Update global state | 
|---|
| 255 | kernelTLS.this_thread = dst; | 
|---|
| 256 |  | 
|---|
| 257 | // Context Switch to the thread | 
|---|
| 258 | ThreadCtxSwitch(proc_cor, thrd_cor); | 
|---|
| 259 | // when ThreadCtxSwitch returns we are back in the processor coroutine | 
|---|
| 260 | } | 
|---|
| 261 |  | 
|---|
| 262 | // KERNEL_ONLY | 
|---|
| 263 | static void returnToKernel() { | 
|---|
| 264 | coroutine_desc * proc_cor = get_coroutine(kernelTLS.this_processor->runner); | 
|---|
| 265 | coroutine_desc * thrd_cor = kernelTLS.this_thread->curr_cor = kernelTLS.this_coroutine; | 
|---|
| 266 | ThreadCtxSwitch(thrd_cor, proc_cor); | 
|---|
| 267 | } | 
|---|
| 268 |  | 
|---|
| 269 | // KERNEL_ONLY | 
|---|
| 270 | // Once a thread has finished running, some of | 
|---|
| 271 | // its final actions must be executed from the kernel | 
|---|
| 272 | static void finishRunning(processor * this) with( this->finish ) { | 
|---|
| 273 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 274 | choose( action_code ) { | 
|---|
| 275 | case No_Action: | 
|---|
| 276 | break; | 
|---|
| 277 | case Release: | 
|---|
| 278 | unlock( *lock ); | 
|---|
| 279 | case Schedule: | 
|---|
| 280 | ScheduleThread( thrd ); | 
|---|
| 281 | case Release_Schedule: | 
|---|
| 282 | unlock( *lock ); | 
|---|
| 283 | ScheduleThread( thrd ); | 
|---|
| 284 | case Release_Multi: | 
|---|
| 285 | for(int i = 0; i < lock_count; i++) { | 
|---|
| 286 | unlock( *locks[i] ); | 
|---|
| 287 | } | 
|---|
| 288 | case Release_Multi_Schedule: | 
|---|
| 289 | for(int i = 0; i < lock_count; i++) { | 
|---|
| 290 | unlock( *locks[i] ); | 
|---|
| 291 | } | 
|---|
| 292 | for(int i = 0; i < thrd_count; i++) { | 
|---|
| 293 | ScheduleThread( thrds[i] ); | 
|---|
| 294 | } | 
|---|
| 295 | case Callback: | 
|---|
| 296 | callback(); | 
|---|
| 297 | default: | 
|---|
| 298 | abort("KERNEL ERROR: Unexpected action to run after thread"); | 
|---|
| 299 | } | 
|---|
| 300 | } | 
|---|
| 301 |  | 
|---|
| 302 | // KERNEL_ONLY | 
|---|
| 303 | // Context invoker for processors | 
|---|
| 304 | // This is the entry point for processors (kernel threads) | 
|---|
| 305 | // It effectively constructs a coroutine by stealing the pthread stack | 
|---|
| 306 | static void * CtxInvokeProcessor(void * arg) { | 
|---|
| 307 | processor * proc = (processor *) arg; | 
|---|
| 308 | kernelTLS.this_processor = proc; | 
|---|
| 309 | kernelTLS.this_coroutine = NULL; | 
|---|
| 310 | kernelTLS.this_thread    = NULL; | 
|---|
| 311 | kernelTLS.preemption_state.[enabled, disable_count] = [false, 1]; | 
|---|
| 312 | // SKULLDUGGERY: We want to create a context for the processor coroutine | 
|---|
| 313 | // which is needed for the 2-step context switch. However, there is no reason | 
|---|
| 314 | // to waste the perfectly valid stack create by pthread. | 
|---|
| 315 | current_stack_info_t info; | 
|---|
| 316 | machine_context_t ctx; | 
|---|
| 317 | info.context = &ctx; | 
|---|
| 318 | (proc->runner){ proc, &info }; | 
|---|
| 319 |  | 
|---|
| 320 | __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.base); | 
|---|
| 321 |  | 
|---|
| 322 | //Set global state | 
|---|
| 323 | kernelTLS.this_coroutine = get_coroutine(proc->runner); | 
|---|
| 324 | kernelTLS.this_thread    = NULL; | 
|---|
| 325 |  | 
|---|
| 326 | //We now have a proper context from which to schedule threads | 
|---|
| 327 | __cfaabi_dbg_print_safe("Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx); | 
|---|
| 328 |  | 
|---|
| 329 | // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't | 
|---|
| 330 | // resume it to start it like it normally would, it will just context switch | 
|---|
| 331 | // back to here. Instead directly call the main since we already are on the | 
|---|
| 332 | // appropriate stack. | 
|---|
| 333 | get_coroutine(proc->runner)->state = Active; | 
|---|
| 334 | main( proc->runner ); | 
|---|
| 335 | get_coroutine(proc->runner)->state = Halted; | 
|---|
| 336 |  | 
|---|
| 337 | // Main routine of the core returned, the core is now fully terminated | 
|---|
| 338 | __cfaabi_dbg_print_safe("Kernel : core %p main ended (%p)\n", proc, &proc->runner); | 
|---|
| 339 |  | 
|---|
| 340 | return NULL; | 
|---|
| 341 | } | 
|---|
| 342 |  | 
|---|
| 343 | static void start(processor * this) { | 
|---|
| 344 | __cfaabi_dbg_print_safe("Kernel : Starting core %p\n", this); | 
|---|
| 345 |  | 
|---|
| 346 | pthread_create( &this->kernel_thread, NULL, CtxInvokeProcessor, (void*)this ); | 
|---|
| 347 |  | 
|---|
| 348 | __cfaabi_dbg_print_safe("Kernel : core %p started\n", this); | 
|---|
| 349 | } | 
|---|
| 350 |  | 
|---|
| 351 | // KERNEL_ONLY | 
|---|
| 352 | void kernel_first_resume(processor * this) { | 
|---|
| 353 | coroutine_desc * src = kernelTLS.this_coroutine; | 
|---|
| 354 | coroutine_desc * dst = get_coroutine(this->runner); | 
|---|
| 355 |  | 
|---|
| 356 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 357 |  | 
|---|
| 358 | create_stack(&dst->stack, dst->stack.size); | 
|---|
| 359 | CtxStart(&this->runner, CtxInvokeCoroutine); | 
|---|
| 360 |  | 
|---|
| 361 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 362 |  | 
|---|
| 363 | dst->last = src; | 
|---|
| 364 | dst->starter = dst->starter ? dst->starter : src; | 
|---|
| 365 |  | 
|---|
| 366 | // set state of current coroutine to inactive | 
|---|
| 367 | src->state = src->state == Halted ? Halted : Inactive; | 
|---|
| 368 |  | 
|---|
| 369 | // set new coroutine that task is executing | 
|---|
| 370 | kernelTLS.this_coroutine = dst; | 
|---|
| 371 |  | 
|---|
| 372 | // SKULLDUGGERY normally interrupts are enable before leaving a coroutine ctxswitch. | 
|---|
| 373 | // Therefore, when first creating a coroutine, interrupts are enable before calling the main. | 
|---|
| 374 | // This is consistent with thread creation. However, when creating the main processor coroutine, | 
|---|
| 375 | // we wan't interrupts to be disabled. Therefore, we double-disable interrupts here so they will | 
|---|
| 376 | // stay disabled. | 
|---|
| 377 | disable_interrupts(); | 
|---|
| 378 |  | 
|---|
| 379 | // context switch to specified coroutine | 
|---|
| 380 | assert( src->stack.context ); | 
|---|
| 381 | CtxSwitch( src->stack.context, dst->stack.context ); | 
|---|
| 382 | // when CtxSwitch returns we are back in the src coroutine | 
|---|
| 383 |  | 
|---|
| 384 | // set state of new coroutine to active | 
|---|
| 385 | src->state = Active; | 
|---|
| 386 |  | 
|---|
| 387 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 388 | } | 
|---|
| 389 |  | 
|---|
| 390 | //----------------------------------------------------------------------------- | 
|---|
| 391 | // Scheduler routines | 
|---|
| 392 |  | 
|---|
| 393 | // KERNEL ONLY | 
|---|
| 394 | void ScheduleThread( thread_desc * thrd ) { | 
|---|
| 395 | verify( thrd ); | 
|---|
| 396 | verify( thrd->self_cor.state != Halted ); | 
|---|
| 397 |  | 
|---|
| 398 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 399 |  | 
|---|
| 400 | verifyf( thrd->next == NULL, "Expected null got %p", thrd->next ); | 
|---|
| 401 |  | 
|---|
| 402 | with( *thrd->curr_cluster ) { | 
|---|
| 403 | lock  ( ready_queue_lock __cfaabi_dbg_ctx2 ); | 
|---|
| 404 | bool was_empty = !(ready_queue != 0); | 
|---|
| 405 | append( ready_queue, thrd ); | 
|---|
| 406 | unlock( ready_queue_lock ); | 
|---|
| 407 |  | 
|---|
| 408 | if(was_empty) { | 
|---|
| 409 | lock      (proc_list_lock __cfaabi_dbg_ctx2); | 
|---|
| 410 | if(idles) { | 
|---|
| 411 | wake_fast(idles.head); | 
|---|
| 412 | } | 
|---|
| 413 | unlock    (proc_list_lock); | 
|---|
| 414 | } | 
|---|
| 415 | else if( struct processor * idle = idles.head ) { | 
|---|
| 416 | wake_fast(idle); | 
|---|
| 417 | } | 
|---|
| 418 |  | 
|---|
| 419 | } | 
|---|
| 420 |  | 
|---|
| 421 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 422 | } | 
|---|
| 423 |  | 
|---|
| 424 | // KERNEL ONLY | 
|---|
| 425 | thread_desc * nextThread(cluster * this) with( *this ) { | 
|---|
| 426 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 427 | lock( ready_queue_lock __cfaabi_dbg_ctx2 ); | 
|---|
| 428 | thread_desc * head = pop_head( ready_queue ); | 
|---|
| 429 | unlock( ready_queue_lock ); | 
|---|
| 430 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 431 | return head; | 
|---|
| 432 | } | 
|---|
| 433 |  | 
|---|
| 434 | void BlockInternal() { | 
|---|
| 435 | disable_interrupts(); | 
|---|
| 436 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 437 | returnToKernel(); | 
|---|
| 438 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 439 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 440 | } | 
|---|
| 441 |  | 
|---|
| 442 | void BlockInternal( __spinlock_t * lock ) { | 
|---|
| 443 | disable_interrupts(); | 
|---|
| 444 | with( *kernelTLS.this_processor ) { | 
|---|
| 445 | finish.action_code = Release; | 
|---|
| 446 | finish.lock        = lock; | 
|---|
| 447 | } | 
|---|
| 448 |  | 
|---|
| 449 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 450 | returnToKernel(); | 
|---|
| 451 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 452 |  | 
|---|
| 453 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 454 | } | 
|---|
| 455 |  | 
|---|
| 456 | void BlockInternal( thread_desc * thrd ) { | 
|---|
| 457 | disable_interrupts(); | 
|---|
| 458 | with( * kernelTLS.this_processor ) { | 
|---|
| 459 | finish.action_code = Schedule; | 
|---|
| 460 | finish.thrd        = thrd; | 
|---|
| 461 | } | 
|---|
| 462 |  | 
|---|
| 463 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 464 | returnToKernel(); | 
|---|
| 465 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 466 |  | 
|---|
| 467 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 468 | } | 
|---|
| 469 |  | 
|---|
| 470 | void BlockInternal( __spinlock_t * lock, thread_desc * thrd ) { | 
|---|
| 471 | assert(thrd); | 
|---|
| 472 | disable_interrupts(); | 
|---|
| 473 | with( * kernelTLS.this_processor ) { | 
|---|
| 474 | finish.action_code = Release_Schedule; | 
|---|
| 475 | finish.lock        = lock; | 
|---|
| 476 | finish.thrd        = thrd; | 
|---|
| 477 | } | 
|---|
| 478 |  | 
|---|
| 479 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 480 | returnToKernel(); | 
|---|
| 481 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 482 |  | 
|---|
| 483 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 484 | } | 
|---|
| 485 |  | 
|---|
| 486 | void BlockInternal(__spinlock_t * locks [], unsigned short count) { | 
|---|
| 487 | disable_interrupts(); | 
|---|
| 488 | with( * kernelTLS.this_processor ) { | 
|---|
| 489 | finish.action_code = Release_Multi; | 
|---|
| 490 | finish.locks       = locks; | 
|---|
| 491 | finish.lock_count  = count; | 
|---|
| 492 | } | 
|---|
| 493 |  | 
|---|
| 494 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 495 | returnToKernel(); | 
|---|
| 496 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 497 |  | 
|---|
| 498 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 499 | } | 
|---|
| 500 |  | 
|---|
| 501 | void BlockInternal(__spinlock_t * locks [], unsigned short lock_count, thread_desc * thrds [], unsigned short thrd_count) { | 
|---|
| 502 | disable_interrupts(); | 
|---|
| 503 | with( *kernelTLS.this_processor ) { | 
|---|
| 504 | finish.action_code = Release_Multi_Schedule; | 
|---|
| 505 | finish.locks       = locks; | 
|---|
| 506 | finish.lock_count  = lock_count; | 
|---|
| 507 | finish.thrds       = thrds; | 
|---|
| 508 | finish.thrd_count  = thrd_count; | 
|---|
| 509 | } | 
|---|
| 510 |  | 
|---|
| 511 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 512 | returnToKernel(); | 
|---|
| 513 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 514 |  | 
|---|
| 515 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 516 | } | 
|---|
| 517 |  | 
|---|
| 518 | void BlockInternal(__finish_callback_fptr_t callback) { | 
|---|
| 519 | disable_interrupts(); | 
|---|
| 520 | with( *kernelTLS.this_processor ) { | 
|---|
| 521 | finish.action_code = Callback; | 
|---|
| 522 | finish.callback    = callback; | 
|---|
| 523 | } | 
|---|
| 524 |  | 
|---|
| 525 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 526 | returnToKernel(); | 
|---|
| 527 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 528 |  | 
|---|
| 529 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 530 | } | 
|---|
| 531 |  | 
|---|
| 532 | // KERNEL ONLY | 
|---|
| 533 | void LeaveThread(__spinlock_t * lock, thread_desc * thrd) { | 
|---|
| 534 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 535 | with( * kernelTLS.this_processor ) { | 
|---|
| 536 | finish.action_code = thrd ? Release_Schedule : Release; | 
|---|
| 537 | finish.lock        = lock; | 
|---|
| 538 | finish.thrd        = thrd; | 
|---|
| 539 | } | 
|---|
| 540 |  | 
|---|
| 541 | returnToKernel(); | 
|---|
| 542 | } | 
|---|
| 543 |  | 
|---|
| 544 | //============================================================================================= | 
|---|
| 545 | // Kernel Setup logic | 
|---|
| 546 | //============================================================================================= | 
|---|
| 547 | //----------------------------------------------------------------------------- | 
|---|
| 548 | // Kernel boot procedures | 
|---|
| 549 | static void kernel_startup(void) { | 
|---|
| 550 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 551 | __cfaabi_dbg_print_safe("Kernel : Starting\n"); | 
|---|
| 552 |  | 
|---|
| 553 | __cfa_dbg_global_clusters.list{ __get }; | 
|---|
| 554 | __cfa_dbg_global_clusters.lock{}; | 
|---|
| 555 |  | 
|---|
| 556 | // Initialize the main cluster | 
|---|
| 557 | mainCluster = (cluster *)&storage_mainCluster; | 
|---|
| 558 | (*mainCluster){"Main Cluster"}; | 
|---|
| 559 |  | 
|---|
| 560 | __cfaabi_dbg_print_safe("Kernel : Main cluster ready\n"); | 
|---|
| 561 |  | 
|---|
| 562 | // Start by initializing the main thread | 
|---|
| 563 | // SKULLDUGGERY: the mainThread steals the process main thread | 
|---|
| 564 | // which will then be scheduled by the mainProcessor normally | 
|---|
| 565 | mainThread = (thread_desc *)&storage_mainThread; | 
|---|
| 566 | current_stack_info_t info; | 
|---|
| 567 | (*mainThread){ &info }; | 
|---|
| 568 |  | 
|---|
| 569 | __cfaabi_dbg_print_safe("Kernel : Main thread ready\n"); | 
|---|
| 570 |  | 
|---|
| 571 |  | 
|---|
| 572 |  | 
|---|
| 573 | // Construct the processor context of the main processor | 
|---|
| 574 | void ?{}(processorCtx_t & this, processor * proc) { | 
|---|
| 575 | (this.__cor){ "Processor" }; | 
|---|
| 576 | this.__cor.starter = NULL; | 
|---|
| 577 | this.proc = proc; | 
|---|
| 578 | } | 
|---|
| 579 |  | 
|---|
| 580 | void ?{}(processor & this) with( this ) { | 
|---|
| 581 | name = "Main Processor"; | 
|---|
| 582 | cltr = mainCluster; | 
|---|
| 583 | terminated{ 0 }; | 
|---|
| 584 | do_terminate = false; | 
|---|
| 585 | preemption_alarm = NULL; | 
|---|
| 586 | pending_preemption = false; | 
|---|
| 587 | kernel_thread = pthread_self(); | 
|---|
| 588 |  | 
|---|
| 589 | runner{ &this }; | 
|---|
| 590 | __cfaabi_dbg_print_safe("Kernel : constructed main processor context %p\n", &runner); | 
|---|
| 591 | } | 
|---|
| 592 |  | 
|---|
| 593 | // Initialize the main processor and the main processor ctx | 
|---|
| 594 | // (the coroutine that contains the processing control flow) | 
|---|
| 595 | mainProcessor = (processor *)&storage_mainProcessor; | 
|---|
| 596 | (*mainProcessor){}; | 
|---|
| 597 |  | 
|---|
| 598 | //initialize the global state variables | 
|---|
| 599 | kernelTLS.this_processor = mainProcessor; | 
|---|
| 600 | kernelTLS.this_thread    = mainThread; | 
|---|
| 601 | kernelTLS.this_coroutine = &mainThread->self_cor; | 
|---|
| 602 |  | 
|---|
| 603 | // Enable preemption | 
|---|
| 604 | kernel_start_preemption(); | 
|---|
| 605 |  | 
|---|
| 606 | // Add the main thread to the ready queue | 
|---|
| 607 | // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread | 
|---|
| 608 | ScheduleThread(mainThread); | 
|---|
| 609 |  | 
|---|
| 610 | // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX | 
|---|
| 611 | // context. Hence, the main thread does not begin through CtxInvokeThread, like all other threads. The trick here is that | 
|---|
| 612 | // mainThread is on the ready queue when this call is made. | 
|---|
| 613 | kernel_first_resume( kernelTLS.this_processor ); | 
|---|
| 614 |  | 
|---|
| 615 |  | 
|---|
| 616 |  | 
|---|
| 617 | // THE SYSTEM IS NOW COMPLETELY RUNNING | 
|---|
| 618 | __cfaabi_dbg_print_safe("Kernel : Started\n--------------------------------------------------\n\n"); | 
|---|
| 619 |  | 
|---|
| 620 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 621 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 622 | verify( TL_GET( preemption_state.enabled ) ); | 
|---|
| 623 | } | 
|---|
| 624 |  | 
|---|
| 625 | static void kernel_shutdown(void) { | 
|---|
| 626 | __cfaabi_dbg_print_safe("\n--------------------------------------------------\nKernel : Shutting down\n"); | 
|---|
| 627 |  | 
|---|
| 628 | verify( TL_GET( preemption_state.enabled ) ); | 
|---|
| 629 | disable_interrupts(); | 
|---|
| 630 | verify( ! kernelTLS.preemption_state.enabled ); | 
|---|
| 631 |  | 
|---|
| 632 | // SKULLDUGGERY: Notify the mainProcessor it needs to terminates. | 
|---|
| 633 | // When its coroutine terminates, it return control to the mainThread | 
|---|
| 634 | // which is currently here | 
|---|
| 635 | __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE); | 
|---|
| 636 | returnToKernel(); | 
|---|
| 637 | mainThread->self_cor.state = Halted; | 
|---|
| 638 |  | 
|---|
| 639 | // THE SYSTEM IS NOW COMPLETELY STOPPED | 
|---|
| 640 |  | 
|---|
| 641 | // Disable preemption | 
|---|
| 642 | kernel_stop_preemption(); | 
|---|
| 643 |  | 
|---|
| 644 | // Destroy the main processor and its context in reverse order of construction | 
|---|
| 645 | // These were manually constructed so we need manually destroy them | 
|---|
| 646 | ^(mainProcessor->runner){}; | 
|---|
| 647 | ^(mainProcessor){}; | 
|---|
| 648 |  | 
|---|
| 649 | // Final step, destroy the main thread since it is no longer needed | 
|---|
| 650 | // Since we provided a stack to this taxk it will not destroy anything | 
|---|
| 651 | ^(mainThread){}; | 
|---|
| 652 |  | 
|---|
| 653 | ^(__cfa_dbg_global_clusters.list){}; | 
|---|
| 654 | ^(__cfa_dbg_global_clusters.lock){}; | 
|---|
| 655 |  | 
|---|
| 656 | __cfaabi_dbg_print_safe("Kernel : Shutdown complete\n"); | 
|---|
| 657 | } | 
|---|
| 658 |  | 
|---|
| 659 | //============================================================================================= | 
|---|
| 660 | // Kernel Quiescing | 
|---|
| 661 | //============================================================================================= | 
|---|
| 662 | static void halt(processor * this) with( *this ) { | 
|---|
| 663 | // verify( ! __atomic_load_n(&do_terminate, __ATOMIC_SEQ_CST) ); | 
|---|
| 664 |  | 
|---|
| 665 | with( *cltr ) { | 
|---|
| 666 | lock      (proc_list_lock __cfaabi_dbg_ctx2); | 
|---|
| 667 | remove    (procs, *this); | 
|---|
| 668 | push_front(idles, *this); | 
|---|
| 669 | unlock    (proc_list_lock); | 
|---|
| 670 | } | 
|---|
| 671 |  | 
|---|
| 672 | __cfaabi_dbg_print_safe("Kernel : Processor %p ready to sleep\n", this); | 
|---|
| 673 |  | 
|---|
| 674 | wait( idleLock ); | 
|---|
| 675 |  | 
|---|
| 676 | __cfaabi_dbg_print_safe("Kernel : Processor %p woke up and ready to run\n", this); | 
|---|
| 677 |  | 
|---|
| 678 | with( *cltr ) { | 
|---|
| 679 | lock      (proc_list_lock __cfaabi_dbg_ctx2); | 
|---|
| 680 | remove    (idles, *this); | 
|---|
| 681 | push_front(procs, *this); | 
|---|
| 682 | unlock    (proc_list_lock); | 
|---|
| 683 | } | 
|---|
| 684 | } | 
|---|
| 685 |  | 
|---|
| 686 | //============================================================================================= | 
|---|
| 687 | // Unexpected Terminating logic | 
|---|
| 688 | //============================================================================================= | 
|---|
| 689 | static __spinlock_t kernel_abort_lock; | 
|---|
| 690 | static bool kernel_abort_called = false; | 
|---|
| 691 |  | 
|---|
| 692 | void * kernel_abort(void) __attribute__ ((__nothrow__)) { | 
|---|
| 693 | // abort cannot be recursively entered by the same or different processors because all signal handlers return when | 
|---|
| 694 | // the globalAbort flag is true. | 
|---|
| 695 | lock( kernel_abort_lock __cfaabi_dbg_ctx2 ); | 
|---|
| 696 |  | 
|---|
| 697 | // first task to abort ? | 
|---|
| 698 | if ( kernel_abort_called ) {                    // not first task to abort ? | 
|---|
| 699 | unlock( kernel_abort_lock ); | 
|---|
| 700 |  | 
|---|
| 701 | sigset_t mask; | 
|---|
| 702 | sigemptyset( &mask ); | 
|---|
| 703 | sigaddset( &mask, SIGALRM );            // block SIGALRM signals | 
|---|
| 704 | sigsuspend( &mask );                    // block the processor to prevent further damage during abort | 
|---|
| 705 | _exit( EXIT_FAILURE );                  // if processor unblocks before it is killed, terminate it | 
|---|
| 706 | } | 
|---|
| 707 | else { | 
|---|
| 708 | kernel_abort_called = true; | 
|---|
| 709 | unlock( kernel_abort_lock ); | 
|---|
| 710 | } | 
|---|
| 711 |  | 
|---|
| 712 | return kernelTLS.this_thread; | 
|---|
| 713 | } | 
|---|
| 714 |  | 
|---|
| 715 | void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) { | 
|---|
| 716 | thread_desc * thrd = kernel_data; | 
|---|
| 717 |  | 
|---|
| 718 | if(thrd) { | 
|---|
| 719 | int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd ); | 
|---|
| 720 | __cfaabi_dbg_bits_write( abort_text, len ); | 
|---|
| 721 |  | 
|---|
| 722 | if ( get_coroutine(thrd) != kernelTLS.this_coroutine ) { | 
|---|
| 723 | len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", kernelTLS.this_coroutine->name, kernelTLS.this_coroutine ); | 
|---|
| 724 | __cfaabi_dbg_bits_write( abort_text, len ); | 
|---|
| 725 | } | 
|---|
| 726 | else { | 
|---|
| 727 | __cfaabi_dbg_bits_write( ".\n", 2 ); | 
|---|
| 728 | } | 
|---|
| 729 | } | 
|---|
| 730 | else { | 
|---|
| 731 | int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" ); | 
|---|
| 732 | __cfaabi_dbg_bits_write( abort_text, len ); | 
|---|
| 733 | } | 
|---|
| 734 | } | 
|---|
| 735 |  | 
|---|
| 736 | int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) { | 
|---|
| 737 | return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2; | 
|---|
| 738 | } | 
|---|
| 739 |  | 
|---|
| 740 | static __spinlock_t kernel_debug_lock; | 
|---|
| 741 |  | 
|---|
| 742 | extern "C" { | 
|---|
| 743 | void __cfaabi_dbg_bits_acquire() { | 
|---|
| 744 | lock( kernel_debug_lock __cfaabi_dbg_ctx2 ); | 
|---|
| 745 | } | 
|---|
| 746 |  | 
|---|
| 747 | void __cfaabi_dbg_bits_release() { | 
|---|
| 748 | unlock( kernel_debug_lock ); | 
|---|
| 749 | } | 
|---|
| 750 | } | 
|---|
| 751 |  | 
|---|
| 752 | //============================================================================================= | 
|---|
| 753 | // Kernel Utilities | 
|---|
| 754 | //============================================================================================= | 
|---|
| 755 | //----------------------------------------------------------------------------- | 
|---|
| 756 | // Locks | 
|---|
| 757 | void  ?{}( semaphore & this, int count = 1 ) { | 
|---|
| 758 | (this.lock){}; | 
|---|
| 759 | this.count = count; | 
|---|
| 760 | (this.waiting){}; | 
|---|
| 761 | } | 
|---|
| 762 | void ^?{}(semaphore & this) {} | 
|---|
| 763 |  | 
|---|
| 764 | void P(semaphore & this) with( this ){ | 
|---|
| 765 | lock( lock __cfaabi_dbg_ctx2 ); | 
|---|
| 766 | count -= 1; | 
|---|
| 767 | if ( count < 0 ) { | 
|---|
| 768 | // queue current task | 
|---|
| 769 | append( waiting, kernelTLS.this_thread ); | 
|---|
| 770 |  | 
|---|
| 771 | // atomically release spin lock and block | 
|---|
| 772 | BlockInternal( &lock ); | 
|---|
| 773 | } | 
|---|
| 774 | else { | 
|---|
| 775 | unlock( lock ); | 
|---|
| 776 | } | 
|---|
| 777 | } | 
|---|
| 778 |  | 
|---|
| 779 | void V(semaphore & this) with( this ) { | 
|---|
| 780 | thread_desc * thrd = NULL; | 
|---|
| 781 | lock( lock __cfaabi_dbg_ctx2 ); | 
|---|
| 782 | count += 1; | 
|---|
| 783 | if ( count <= 0 ) { | 
|---|
| 784 | // remove task at head of waiting list | 
|---|
| 785 | thrd = pop_head( waiting ); | 
|---|
| 786 | } | 
|---|
| 787 |  | 
|---|
| 788 | unlock( lock ); | 
|---|
| 789 |  | 
|---|
| 790 | // make new owner | 
|---|
| 791 | WakeThread( thrd ); | 
|---|
| 792 | } | 
|---|
| 793 |  | 
|---|
| 794 | //----------------------------------------------------------------------------- | 
|---|
| 795 | // Global Queues | 
|---|
| 796 | void doregister( cluster     & cltr ) { | 
|---|
| 797 | lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2); | 
|---|
| 798 | push_front( __cfa_dbg_global_clusters.list, cltr ); | 
|---|
| 799 | unlock    ( __cfa_dbg_global_clusters.lock ); | 
|---|
| 800 | } | 
|---|
| 801 |  | 
|---|
| 802 | void unregister( cluster     & cltr ) { | 
|---|
| 803 | lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2); | 
|---|
| 804 | remove( __cfa_dbg_global_clusters.list, cltr ); | 
|---|
| 805 | unlock( __cfa_dbg_global_clusters.lock ); | 
|---|
| 806 | } | 
|---|
| 807 |  | 
|---|
| 808 | void doregister( cluster * cltr, thread_desc & thrd ) { | 
|---|
| 809 | lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2); | 
|---|
| 810 | push_front(cltr->threads, thrd); | 
|---|
| 811 | unlock    (cltr->thread_list_lock); | 
|---|
| 812 | } | 
|---|
| 813 |  | 
|---|
| 814 | void unregister( cluster * cltr, thread_desc & thrd ) { | 
|---|
| 815 | lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2); | 
|---|
| 816 | remove(cltr->threads, thrd ); | 
|---|
| 817 | unlock(cltr->thread_list_lock); | 
|---|
| 818 | } | 
|---|
| 819 |  | 
|---|
| 820 | void doregister( cluster * cltr, processor * proc ) { | 
|---|
| 821 | lock      (cltr->proc_list_lock __cfaabi_dbg_ctx2); | 
|---|
| 822 | push_front(cltr->procs, *proc); | 
|---|
| 823 | unlock    (cltr->proc_list_lock); | 
|---|
| 824 | } | 
|---|
| 825 |  | 
|---|
| 826 | void unregister( cluster * cltr, processor * proc ) { | 
|---|
| 827 | lock  (cltr->proc_list_lock __cfaabi_dbg_ctx2); | 
|---|
| 828 | remove(cltr->procs, *proc ); | 
|---|
| 829 | unlock(cltr->proc_list_lock); | 
|---|
| 830 | } | 
|---|
| 831 |  | 
|---|
| 832 | //----------------------------------------------------------------------------- | 
|---|
| 833 | // Debug | 
|---|
| 834 | __cfaabi_dbg_debug_do( | 
|---|
| 835 | extern "C" { | 
|---|
| 836 | void __cfaabi_dbg_record(__spinlock_t & this, const char * prev_name) { | 
|---|
| 837 | this.prev_name = prev_name; | 
|---|
| 838 | this.prev_thrd = kernelTLS.this_thread; | 
|---|
| 839 | } | 
|---|
| 840 | } | 
|---|
| 841 | ) | 
|---|
| 842 | // Local Variables: // | 
|---|
| 843 | // mode: c // | 
|---|
| 844 | // tab-width: 4 // | 
|---|
| 845 | // End: // | 
|---|