| 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 Aug 31 07:08:20 2020 | 
|---|
| 13 | // Update Count     : 71 | 
|---|
| 14 | // | 
|---|
| 15 |  | 
|---|
| 16 | #define __cforall_thread__ | 
|---|
| 17 | // #define __CFA_DEBUG_PRINT_RUNTIME_CORE__ | 
|---|
| 18 |  | 
|---|
| 19 | //C Includes | 
|---|
| 20 | #include <errno.h> | 
|---|
| 21 | #include <stdio.h> | 
|---|
| 22 | #include <signal.h> | 
|---|
| 23 | #include <unistd.h> | 
|---|
| 24 |  | 
|---|
| 25 | //CFA Includes | 
|---|
| 26 | #include "kernel_private.hfa" | 
|---|
| 27 | #include "preemption.hfa" | 
|---|
| 28 |  | 
|---|
| 29 | //Private includes | 
|---|
| 30 | #define __CFA_INVOKE_PRIVATE__ | 
|---|
| 31 | #include "invoke.h" | 
|---|
| 32 |  | 
|---|
| 33 |  | 
|---|
| 34 | //----------------------------------------------------------------------------- | 
|---|
| 35 | // Some assembly required | 
|---|
| 36 | #if defined( __i386 ) | 
|---|
| 37 | // mxcr : SSE Status and Control bits (control bits are preserved across function calls) | 
|---|
| 38 | // fcw  : X87 FPU control word (preserved across function calls) | 
|---|
| 39 | #define __x87_store         \ | 
|---|
| 40 | uint32_t __mxcr;      \ | 
|---|
| 41 | uint16_t __fcw;       \ | 
|---|
| 42 | __asm__ volatile (    \ | 
|---|
| 43 | "stmxcsr %0\n"  \ | 
|---|
| 44 | "fnstcw  %1\n"  \ | 
|---|
| 45 | : "=m" (__mxcr),\ | 
|---|
| 46 | "=m" (__fcw)  \ | 
|---|
| 47 | ) | 
|---|
| 48 |  | 
|---|
| 49 | #define __x87_load         \ | 
|---|
| 50 | __asm__ volatile (   \ | 
|---|
| 51 | "fldcw  %1\n"  \ | 
|---|
| 52 | "ldmxcsr %0\n" \ | 
|---|
| 53 | ::"m" (__mxcr),\ | 
|---|
| 54 | "m" (__fcw)  \ | 
|---|
| 55 | ) | 
|---|
| 56 |  | 
|---|
| 57 | #elif defined( __x86_64 ) | 
|---|
| 58 | #define __x87_store         \ | 
|---|
| 59 | uint32_t __mxcr;      \ | 
|---|
| 60 | uint16_t __fcw;       \ | 
|---|
| 61 | __asm__ volatile (    \ | 
|---|
| 62 | "stmxcsr %0\n"  \ | 
|---|
| 63 | "fnstcw  %1\n"  \ | 
|---|
| 64 | : "=m" (__mxcr),\ | 
|---|
| 65 | "=m" (__fcw)  \ | 
|---|
| 66 | ) | 
|---|
| 67 |  | 
|---|
| 68 | #define __x87_load          \ | 
|---|
| 69 | __asm__ volatile (    \ | 
|---|
| 70 | "fldcw  %1\n"   \ | 
|---|
| 71 | "ldmxcsr %0\n"  \ | 
|---|
| 72 | :: "m" (__mxcr),\ | 
|---|
| 73 | "m" (__fcw)  \ | 
|---|
| 74 | ) | 
|---|
| 75 |  | 
|---|
| 76 | #elif defined( __arm__ ) | 
|---|
| 77 | #define __x87_store | 
|---|
| 78 | #define __x87_load | 
|---|
| 79 |  | 
|---|
| 80 | #elif defined( __aarch64__ ) | 
|---|
| 81 | #define __x87_store              \ | 
|---|
| 82 | uint32_t __fpcntl[2];    \ | 
|---|
| 83 | __asm__ volatile (    \ | 
|---|
| 84 | "mrs x9, FPCR\n" \ | 
|---|
| 85 | "mrs x10, FPSR\n"  \ | 
|---|
| 86 | "stp x9, x10, %0\n"  \ | 
|---|
| 87 | : "=m" (__fpcntl) : : "x9", "x10" \ | 
|---|
| 88 | ) | 
|---|
| 89 |  | 
|---|
| 90 | #define __x87_load         \ | 
|---|
| 91 | __asm__ volatile (    \ | 
|---|
| 92 | "ldp x9, x10, %0\n"  \ | 
|---|
| 93 | "msr FPSR, x10\n"  \ | 
|---|
| 94 | "msr FPCR, x9\n" \ | 
|---|
| 95 | : "=m" (__fpcntl) : : "x9", "x10" \ | 
|---|
| 96 | ) | 
|---|
| 97 |  | 
|---|
| 98 | #else | 
|---|
| 99 | #error unsupported hardware architecture | 
|---|
| 100 | #endif | 
|---|
| 101 |  | 
|---|
| 102 | extern $thread * mainThread; | 
|---|
| 103 | extern processor * mainProcessor; | 
|---|
| 104 |  | 
|---|
| 105 | //----------------------------------------------------------------------------- | 
|---|
| 106 | // Kernel Scheduling logic | 
|---|
| 107 | static $thread * __next_thread(cluster * this); | 
|---|
| 108 | static $thread * __next_thread_slow(cluster * this); | 
|---|
| 109 | static void __run_thread(processor * this, $thread * dst); | 
|---|
| 110 | static void __wake_one(cluster * cltr); | 
|---|
| 111 |  | 
|---|
| 112 | static void push  (__cluster_idles & idles, processor & proc); | 
|---|
| 113 | static void remove(__cluster_idles & idles, processor & proc); | 
|---|
| 114 | static [unsigned idle, unsigned total, * processor] query( & __cluster_idles idles ); | 
|---|
| 115 |  | 
|---|
| 116 |  | 
|---|
| 117 | //============================================================================================= | 
|---|
| 118 | // Kernel Scheduling logic | 
|---|
| 119 | //============================================================================================= | 
|---|
| 120 | //Main of the processor contexts | 
|---|
| 121 | void main(processorCtx_t & runner) { | 
|---|
| 122 | // Because of a bug, we couldn't initialized the seed on construction | 
|---|
| 123 | // Do it here | 
|---|
| 124 | __cfaabi_tls.rand_seed ^= rdtscl(); | 
|---|
| 125 | __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner); | 
|---|
| 126 | __tls_rand_advance_bck(); | 
|---|
| 127 |  | 
|---|
| 128 | processor * this = runner.proc; | 
|---|
| 129 | verify(this); | 
|---|
| 130 |  | 
|---|
| 131 | __cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this); | 
|---|
| 132 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 133 | if( this->print_halts ) { | 
|---|
| 134 | __cfaabi_bits_print_safe( STDOUT_FILENO, "Processor : %d - %s (%p)\n", this->id, this->name, (void*)this); | 
|---|
| 135 | } | 
|---|
| 136 | #endif | 
|---|
| 137 |  | 
|---|
| 138 | { | 
|---|
| 139 | // Setup preemption data | 
|---|
| 140 | preemption_scope scope = { this }; | 
|---|
| 141 |  | 
|---|
| 142 | __cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this); | 
|---|
| 143 |  | 
|---|
| 144 | $thread * readyThread = 0p; | 
|---|
| 145 | MAIN_LOOP: | 
|---|
| 146 | for() { | 
|---|
| 147 | // Try to get the next thread | 
|---|
| 148 | readyThread = __next_thread( this->cltr ); | 
|---|
| 149 |  | 
|---|
| 150 | if( !readyThread ) { | 
|---|
| 151 | readyThread = __next_thread_slow( this->cltr ); | 
|---|
| 152 | } | 
|---|
| 153 |  | 
|---|
| 154 | HALT: | 
|---|
| 155 | if( !readyThread ) { | 
|---|
| 156 | // Don't block if we are done | 
|---|
| 157 | if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP; | 
|---|
| 158 |  | 
|---|
| 159 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 160 | __tls_stats()->ready.sleep.halts++; | 
|---|
| 161 | #endif | 
|---|
| 162 |  | 
|---|
| 163 | // Push self to idle stack | 
|---|
| 164 | push(this->cltr->idles, * this); | 
|---|
| 165 |  | 
|---|
| 166 | // Confirm the ready-queue is empty | 
|---|
| 167 | readyThread = __next_thread_slow( this->cltr ); | 
|---|
| 168 | if( readyThread ) { | 
|---|
| 169 | // A thread was found, cancel the halt | 
|---|
| 170 | remove(this->cltr->idles, * this); | 
|---|
| 171 |  | 
|---|
| 172 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 173 | __tls_stats()->ready.sleep.cancels++; | 
|---|
| 174 | #endif | 
|---|
| 175 |  | 
|---|
| 176 | // continue the mai loop | 
|---|
| 177 | break HALT; | 
|---|
| 178 | } | 
|---|
| 179 |  | 
|---|
| 180 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 181 | if(this->print_halts) { | 
|---|
| 182 | __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 0\n", this->id, rdtscl()); | 
|---|
| 183 | } | 
|---|
| 184 | #endif | 
|---|
| 185 |  | 
|---|
| 186 | wait( this->idle ); | 
|---|
| 187 |  | 
|---|
| 188 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 189 | if(this->print_halts) { | 
|---|
| 190 | __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 1\n", this->id, rdtscl()); | 
|---|
| 191 | } | 
|---|
| 192 | #endif | 
|---|
| 193 |  | 
|---|
| 194 | // We were woken up, remove self from idle | 
|---|
| 195 | remove(this->cltr->idles, * this); | 
|---|
| 196 |  | 
|---|
| 197 | // DON'T just proceed, start looking again | 
|---|
| 198 | continue MAIN_LOOP; | 
|---|
| 199 | } | 
|---|
| 200 |  | 
|---|
| 201 | /* paranoid */ verify( readyThread ); | 
|---|
| 202 |  | 
|---|
| 203 | // We found a thread run it | 
|---|
| 204 | __run_thread(this, readyThread); | 
|---|
| 205 |  | 
|---|
| 206 | // Are we done? | 
|---|
| 207 | if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP; | 
|---|
| 208 | } | 
|---|
| 209 |  | 
|---|
| 210 | __cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this); | 
|---|
| 211 | } | 
|---|
| 212 |  | 
|---|
| 213 | V( this->terminated ); | 
|---|
| 214 |  | 
|---|
| 215 | if(this == mainProcessor) { | 
|---|
| 216 | // HACK : the coroutine context switch expects this_thread to be set | 
|---|
| 217 | // and it make sense for it to be set in all other cases except here | 
|---|
| 218 | // fake it | 
|---|
| 219 | __cfaabi_tls.this_thread = mainThread; | 
|---|
| 220 | } | 
|---|
| 221 |  | 
|---|
| 222 | __cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this); | 
|---|
| 223 | } | 
|---|
| 224 |  | 
|---|
| 225 | static int * __volatile_errno() __attribute__((noinline)); | 
|---|
| 226 | static int * __volatile_errno() { asm(""); return &errno; } | 
|---|
| 227 |  | 
|---|
| 228 | // KERNEL ONLY | 
|---|
| 229 | // runThread runs a thread by context switching | 
|---|
| 230 | // from the processor coroutine to the target thread | 
|---|
| 231 | static void __run_thread(processor * this, $thread * thrd_dst) { | 
|---|
| 232 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 233 | /* paranoid */ verifyf( thrd_dst->state == Ready || thrd_dst->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", thrd_dst->state, thrd_dst->preempted); | 
|---|
| 234 | /* paranoid */ verifyf( thrd_dst->link.next == 0p, "Expected null got %p", thrd_dst->link.next ); | 
|---|
| 235 | __builtin_prefetch( thrd_dst->context.SP ); | 
|---|
| 236 |  | 
|---|
| 237 | $coroutine * proc_cor = get_coroutine(this->runner); | 
|---|
| 238 |  | 
|---|
| 239 | // set state of processor coroutine to inactive | 
|---|
| 240 | verify(proc_cor->state == Active); | 
|---|
| 241 | proc_cor->state = Blocked; | 
|---|
| 242 |  | 
|---|
| 243 | // Actually run the thread | 
|---|
| 244 | RUNNING:  while(true) { | 
|---|
| 245 | thrd_dst->preempted = __NO_PREEMPTION; | 
|---|
| 246 | thrd_dst->state = Active; | 
|---|
| 247 |  | 
|---|
| 248 | // Update global state | 
|---|
| 249 | kernelTLS().this_thread = thrd_dst; | 
|---|
| 250 |  | 
|---|
| 251 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 252 | /* paranoid */ verify( kernelTLS().this_thread == thrd_dst ); | 
|---|
| 253 | /* paranoid */ verify( thrd_dst->curr_cluster == this->cltr ); | 
|---|
| 254 | /* paranoid */ verify( thrd_dst->context.SP ); | 
|---|
| 255 | /* paranoid */ verify( thrd_dst->state != Halted ); | 
|---|
| 256 | /* 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 | 
|---|
| 257 | /* 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 | 
|---|
| 258 | /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary ); | 
|---|
| 259 |  | 
|---|
| 260 |  | 
|---|
| 261 |  | 
|---|
| 262 | // set context switch to the thread that the processor is executing | 
|---|
| 263 | __cfactx_switch( &proc_cor->context, &thrd_dst->context ); | 
|---|
| 264 | // when __cfactx_switch returns we are back in the processor coroutine | 
|---|
| 265 |  | 
|---|
| 266 | /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary ); | 
|---|
| 267 | /* 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 ); | 
|---|
| 268 | /* 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 ); | 
|---|
| 269 | /* paranoid */ verify( thrd_dst->context.SP ); | 
|---|
| 270 | /* paranoid */ verify( thrd_dst->curr_cluster == this->cltr ); | 
|---|
| 271 | /* paranoid */ verify( kernelTLS().this_thread == thrd_dst ); | 
|---|
| 272 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 273 |  | 
|---|
| 274 | // Reset global state | 
|---|
| 275 | kernelTLS().this_thread = 0p; | 
|---|
| 276 |  | 
|---|
| 277 | // We just finished running a thread, there are a few things that could have happened. | 
|---|
| 278 | // 1 - Regular case : the thread has blocked and now one has scheduled it yet. | 
|---|
| 279 | // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it. | 
|---|
| 280 | // 4 - Preempted | 
|---|
| 281 | // In case 1, we may have won a race so we can't write to the state again. | 
|---|
| 282 | // In case 2, we lost the race so we now own the thread. | 
|---|
| 283 |  | 
|---|
| 284 | if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) { | 
|---|
| 285 | // The thread was preempted, reschedule it and reset the flag | 
|---|
| 286 | __schedule_thread( thrd_dst ); | 
|---|
| 287 | break RUNNING; | 
|---|
| 288 | } | 
|---|
| 289 |  | 
|---|
| 290 | if(unlikely(thrd_dst->state == Halting)) { | 
|---|
| 291 | // The thread has halted, it should never be scheduled/run again | 
|---|
| 292 | // finish the thread | 
|---|
| 293 | __thread_finish( thrd_dst ); | 
|---|
| 294 | break RUNNING; | 
|---|
| 295 | } | 
|---|
| 296 |  | 
|---|
| 297 | /* paranoid */ verify( thrd_dst->state == Active ); | 
|---|
| 298 | thrd_dst->state = Blocked; | 
|---|
| 299 |  | 
|---|
| 300 | // set state of processor coroutine to active and the thread to inactive | 
|---|
| 301 | int old_ticket = __atomic_fetch_sub(&thrd_dst->ticket, 1, __ATOMIC_SEQ_CST); | 
|---|
| 302 | switch(old_ticket) { | 
|---|
| 303 | case TICKET_RUNNING: | 
|---|
| 304 | // This is case 1, the regular case, nothing more is needed | 
|---|
| 305 | break RUNNING; | 
|---|
| 306 | case TICKET_UNBLOCK: | 
|---|
| 307 | // This is case 2, the racy case, someone tried to run this thread before it finished blocking | 
|---|
| 308 | // In this case, just run it again. | 
|---|
| 309 | continue RUNNING; | 
|---|
| 310 | default: | 
|---|
| 311 | // This makes no sense, something is wrong abort | 
|---|
| 312 | abort(); | 
|---|
| 313 | } | 
|---|
| 314 | } | 
|---|
| 315 |  | 
|---|
| 316 | // Just before returning to the processor, set the processor coroutine to active | 
|---|
| 317 | proc_cor->state = Active; | 
|---|
| 318 |  | 
|---|
| 319 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 320 | } | 
|---|
| 321 |  | 
|---|
| 322 | // KERNEL_ONLY | 
|---|
| 323 | void returnToKernel() { | 
|---|
| 324 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 325 | $coroutine * proc_cor = get_coroutine(kernelTLS().this_processor->runner); | 
|---|
| 326 | $thread * thrd_src = kernelTLS().this_thread; | 
|---|
| 327 |  | 
|---|
| 328 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 329 | struct processor * last_proc = kernelTLS().this_processor; | 
|---|
| 330 | #endif | 
|---|
| 331 |  | 
|---|
| 332 | // Run the thread on this processor | 
|---|
| 333 | { | 
|---|
| 334 | int local_errno = *__volatile_errno(); | 
|---|
| 335 | #if defined( __i386 ) || defined( __x86_64 ) | 
|---|
| 336 | __x87_store; | 
|---|
| 337 | #endif | 
|---|
| 338 | /* paranoid */ verify( proc_cor->context.SP ); | 
|---|
| 339 | /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary ); | 
|---|
| 340 | __cfactx_switch( &thrd_src->context, &proc_cor->context ); | 
|---|
| 341 | /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary ); | 
|---|
| 342 | #if defined( __i386 ) || defined( __x86_64 ) | 
|---|
| 343 | __x87_load; | 
|---|
| 344 | #endif | 
|---|
| 345 | *__volatile_errno() = local_errno; | 
|---|
| 346 | } | 
|---|
| 347 |  | 
|---|
| 348 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 349 | if(last_proc != kernelTLS().this_processor) { | 
|---|
| 350 | __tls_stats()->ready.threads.migration++; | 
|---|
| 351 | } | 
|---|
| 352 | #endif | 
|---|
| 353 |  | 
|---|
| 354 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 355 | /* 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 ); | 
|---|
| 356 | /* 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 ); | 
|---|
| 357 | } | 
|---|
| 358 |  | 
|---|
| 359 | //----------------------------------------------------------------------------- | 
|---|
| 360 | // Scheduler routines | 
|---|
| 361 | // KERNEL ONLY | 
|---|
| 362 | void __schedule_thread( $thread * thrd ) { | 
|---|
| 363 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 364 | /* paranoid */ verify( kernelTLS().this_proc_id ); | 
|---|
| 365 | /* paranoid */ verify( thrd ); | 
|---|
| 366 | /* paranoid */ verify( thrd->state != Halted ); | 
|---|
| 367 | /* paranoid */ verify( thrd->curr_cluster ); | 
|---|
| 368 | /* paranoid */ #if defined( __CFA_WITH_VERIFY__ ) | 
|---|
| 369 | /* paranoid */  if( thrd->state == Blocked || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION, | 
|---|
| 370 | "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted ); | 
|---|
| 371 | /* paranoid */  if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active, | 
|---|
| 372 | "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted ); | 
|---|
| 373 | /* paranoid */ #endif | 
|---|
| 374 | /* paranoid */ verifyf( thrd->link.next == 0p, "Expected null got %p", thrd->link.next ); | 
|---|
| 375 | /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary ); | 
|---|
| 376 |  | 
|---|
| 377 |  | 
|---|
| 378 | if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready; | 
|---|
| 379 |  | 
|---|
| 380 | ready_schedule_lock(); | 
|---|
| 381 | // Dereference the thread now because once we push it, there is not guaranteed it's still valid. | 
|---|
| 382 | struct cluster * cl = thrd->curr_cluster; | 
|---|
| 383 |  | 
|---|
| 384 | // push the thread to the cluster ready-queue | 
|---|
| 385 | push( cl, thrd ); | 
|---|
| 386 |  | 
|---|
| 387 | // variable thrd is no longer safe to use | 
|---|
| 388 |  | 
|---|
| 389 | // wake the cluster using the save variable. | 
|---|
| 390 | __wake_one( cl ); | 
|---|
| 391 | ready_schedule_unlock(); | 
|---|
| 392 |  | 
|---|
| 393 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 394 | } | 
|---|
| 395 |  | 
|---|
| 396 | // KERNEL ONLY | 
|---|
| 397 | static inline $thread * __next_thread(cluster * this) with( *this ) { | 
|---|
| 398 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 399 | /* paranoid */ verify( kernelTLS().this_proc_id ); | 
|---|
| 400 |  | 
|---|
| 401 | ready_schedule_lock(); | 
|---|
| 402 | $thread * thrd = pop( this ); | 
|---|
| 403 | ready_schedule_unlock(); | 
|---|
| 404 |  | 
|---|
| 405 | /* paranoid */ verify( kernelTLS().this_proc_id ); | 
|---|
| 406 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 407 | return thrd; | 
|---|
| 408 | } | 
|---|
| 409 |  | 
|---|
| 410 | // KERNEL ONLY | 
|---|
| 411 | static inline $thread * __next_thread_slow(cluster * this) with( *this ) { | 
|---|
| 412 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 413 | /* paranoid */ verify( kernelTLS().this_proc_id ); | 
|---|
| 414 |  | 
|---|
| 415 | ready_schedule_lock(); | 
|---|
| 416 | $thread * thrd = pop_slow( this ); | 
|---|
| 417 | ready_schedule_unlock(); | 
|---|
| 418 |  | 
|---|
| 419 | /* paranoid */ verify( kernelTLS().this_proc_id ); | 
|---|
| 420 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 421 | return thrd; | 
|---|
| 422 | } | 
|---|
| 423 |  | 
|---|
| 424 | void unpark( $thread * thrd ) { | 
|---|
| 425 | if( !thrd ) return; | 
|---|
| 426 |  | 
|---|
| 427 | int old_ticket = __atomic_fetch_add(&thrd->ticket, 1, __ATOMIC_SEQ_CST); | 
|---|
| 428 | switch(old_ticket) { | 
|---|
| 429 | case TICKET_RUNNING: | 
|---|
| 430 | // Wake won the race, the thread will reschedule/rerun itself | 
|---|
| 431 | break; | 
|---|
| 432 | case TICKET_BLOCKED: | 
|---|
| 433 | /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION ); | 
|---|
| 434 | /* paranoid */ verify( thrd->state == Blocked ); | 
|---|
| 435 |  | 
|---|
| 436 | { | 
|---|
| 437 | /* paranoid */ verify( publicTLS_get(this_proc_id) ); | 
|---|
| 438 | bool full = publicTLS_get(this_proc_id)->full_proc; | 
|---|
| 439 | if(full) disable_interrupts(); | 
|---|
| 440 |  | 
|---|
| 441 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 442 |  | 
|---|
| 443 | // Wake lost the race, | 
|---|
| 444 | __schedule_thread( thrd ); | 
|---|
| 445 |  | 
|---|
| 446 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 447 |  | 
|---|
| 448 | if(full) enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 449 | /* paranoid */ verify( publicTLS_get(this_proc_id) ); | 
|---|
| 450 | } | 
|---|
| 451 |  | 
|---|
| 452 | break; | 
|---|
| 453 | default: | 
|---|
| 454 | // This makes no sense, something is wrong abort | 
|---|
| 455 | abort("Thread %p (%s) has mismatch park/unpark\n", thrd, thrd->self_cor.name); | 
|---|
| 456 | } | 
|---|
| 457 | } | 
|---|
| 458 |  | 
|---|
| 459 | void park( void ) { | 
|---|
| 460 | /* paranoid */ verify( __preemption_enabled() ); | 
|---|
| 461 | disable_interrupts(); | 
|---|
| 462 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 463 | /* paranoid */ verify( kernelTLS().this_thread->preempted == __NO_PREEMPTION ); | 
|---|
| 464 |  | 
|---|
| 465 | returnToKernel(); | 
|---|
| 466 |  | 
|---|
| 467 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 468 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 469 | /* paranoid */ verify( __preemption_enabled() ); | 
|---|
| 470 |  | 
|---|
| 471 | } | 
|---|
| 472 |  | 
|---|
| 473 | extern "C" { | 
|---|
| 474 | // Leave the thread monitor | 
|---|
| 475 | // last routine called by a thread. | 
|---|
| 476 | // Should never return | 
|---|
| 477 | void __cfactx_thrd_leave() { | 
|---|
| 478 | $thread * thrd = active_thread(); | 
|---|
| 479 | $monitor * this = &thrd->self_mon; | 
|---|
| 480 |  | 
|---|
| 481 | // Lock the monitor now | 
|---|
| 482 | lock( this->lock __cfaabi_dbg_ctx2 ); | 
|---|
| 483 |  | 
|---|
| 484 | disable_interrupts(); | 
|---|
| 485 |  | 
|---|
| 486 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 487 | /* paranoid */ verify( thrd->state == Active ); | 
|---|
| 488 | /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary ); | 
|---|
| 489 | /* paranoid */ verify( kernelTLS().this_thread == thrd ); | 
|---|
| 490 | /* paranoid */ verify( thrd->context.SP ); | 
|---|
| 491 | /* paranoid */ verifyf( ((uintptr_t)thrd->context.SP) > ((uintptr_t)__get_stack(thrd->curr_cor)->limit), "ERROR : $thread %p has been corrupted.\n StackPointer too large.\n", thrd ); | 
|---|
| 492 | /* paranoid */ verifyf( ((uintptr_t)thrd->context.SP) < ((uintptr_t)__get_stack(thrd->curr_cor)->base ), "ERROR : $thread %p has been corrupted.\n StackPointer too small.\n", thrd ); | 
|---|
| 493 |  | 
|---|
| 494 | thrd->state = Halting; | 
|---|
| 495 | if( TICKET_RUNNING != thrd->ticket ) { abort( "Thread terminated with pending unpark" ); } | 
|---|
| 496 | if( thrd != this->owner ) { abort( "Thread internal monitor has incorrect owner" ); } | 
|---|
| 497 | if( this->recursion != 1) { abort( "Thread internal monitor has unbalanced recursion" ); } | 
|---|
| 498 |  | 
|---|
| 499 | // Leave the thread | 
|---|
| 500 | returnToKernel(); | 
|---|
| 501 |  | 
|---|
| 502 | // Control flow should never reach here! | 
|---|
| 503 | abort(); | 
|---|
| 504 | } | 
|---|
| 505 | } | 
|---|
| 506 |  | 
|---|
| 507 | // KERNEL ONLY | 
|---|
| 508 | bool force_yield( __Preemption_Reason reason ) { | 
|---|
| 509 | /* paranoid */ verify( __preemption_enabled() ); | 
|---|
| 510 | disable_interrupts(); | 
|---|
| 511 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 512 |  | 
|---|
| 513 | $thread * thrd = kernelTLS().this_thread; | 
|---|
| 514 | /* paranoid */ verify(thrd->state == Active); | 
|---|
| 515 |  | 
|---|
| 516 | // SKULLDUGGERY: It is possible that we are preempting this thread just before | 
|---|
| 517 | // it was going to park itself. If that is the case and it is already using the | 
|---|
| 518 | // intrusive fields then we can't use them to preempt the thread | 
|---|
| 519 | // If that is the case, abandon the preemption. | 
|---|
| 520 | bool preempted = false; | 
|---|
| 521 | if(thrd->link.next == 0p) { | 
|---|
| 522 | preempted = true; | 
|---|
| 523 | thrd->preempted = reason; | 
|---|
| 524 | returnToKernel(); | 
|---|
| 525 | } | 
|---|
| 526 |  | 
|---|
| 527 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 528 | enable_interrupts_noPoll(); | 
|---|
| 529 | /* paranoid */ verify( __preemption_enabled() ); | 
|---|
| 530 |  | 
|---|
| 531 | return preempted; | 
|---|
| 532 | } | 
|---|
| 533 |  | 
|---|
| 534 | //============================================================================================= | 
|---|
| 535 | // Kernel Idle Sleep | 
|---|
| 536 | //============================================================================================= | 
|---|
| 537 | // Wake a thread from the front if there are any | 
|---|
| 538 | static void __wake_one(cluster * this) { | 
|---|
| 539 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 540 | /* paranoid */ verify( ready_schedule_islocked() ); | 
|---|
| 541 |  | 
|---|
| 542 | // Check if there is a sleeping processor | 
|---|
| 543 | processor * p; | 
|---|
| 544 | unsigned idle; | 
|---|
| 545 | unsigned total; | 
|---|
| 546 | [idle, total, p] = query(this->idles); | 
|---|
| 547 |  | 
|---|
| 548 | // If no one is sleeping, we are done | 
|---|
| 549 | if( idle == 0 ) return; | 
|---|
| 550 |  | 
|---|
| 551 | // We found a processor, wake it up | 
|---|
| 552 | post( p->idle ); | 
|---|
| 553 |  | 
|---|
| 554 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 555 | __tls_stats()->ready.sleep.wakes++; | 
|---|
| 556 | #endif | 
|---|
| 557 |  | 
|---|
| 558 | /* paranoid */ verify( ready_schedule_islocked() ); | 
|---|
| 559 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 560 |  | 
|---|
| 561 | return; | 
|---|
| 562 | } | 
|---|
| 563 |  | 
|---|
| 564 | // Unconditionnaly wake a thread | 
|---|
| 565 | void __wake_proc(processor * this) { | 
|---|
| 566 | __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this); | 
|---|
| 567 |  | 
|---|
| 568 | disable_interrupts(); | 
|---|
| 569 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 570 | post( this->idle ); | 
|---|
| 571 | enable_interrupts( __cfaabi_dbg_ctx ); | 
|---|
| 572 | } | 
|---|
| 573 |  | 
|---|
| 574 | static void push  (__cluster_idles & this, processor & proc) { | 
|---|
| 575 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 576 | lock( this ); | 
|---|
| 577 | this.idle++; | 
|---|
| 578 | /* paranoid */ verify( this.idle <= this.total ); | 
|---|
| 579 |  | 
|---|
| 580 | insert_first(this.list, proc); | 
|---|
| 581 | unlock( this ); | 
|---|
| 582 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 583 | } | 
|---|
| 584 |  | 
|---|
| 585 | static void remove(__cluster_idles & this, processor & proc) { | 
|---|
| 586 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 587 | lock( this ); | 
|---|
| 588 | this.idle--; | 
|---|
| 589 | /* paranoid */ verify( this.idle >= 0 ); | 
|---|
| 590 |  | 
|---|
| 591 | remove(proc); | 
|---|
| 592 | unlock( this ); | 
|---|
| 593 | /* paranoid */ verify( ! __preemption_enabled() ); | 
|---|
| 594 | } | 
|---|
| 595 |  | 
|---|
| 596 | static [unsigned idle, unsigned total, * processor] query( & __cluster_idles this ) { | 
|---|
| 597 | for() { | 
|---|
| 598 | uint64_t l = __atomic_load_n(&this.lock, __ATOMIC_SEQ_CST); | 
|---|
| 599 | if( 1 == (l % 2) ) { Pause(); continue; } | 
|---|
| 600 | unsigned idle    = this.idle; | 
|---|
| 601 | unsigned total   = this.total; | 
|---|
| 602 | processor * proc = &this.list`first; | 
|---|
| 603 | // Compiler fence is unnecessary, but gcc-8 and older incorrectly reorder code without it | 
|---|
| 604 | asm volatile("": : :"memory"); | 
|---|
| 605 | if(l != __atomic_load_n(&this.lock, __ATOMIC_SEQ_CST)) { Pause(); continue; } | 
|---|
| 606 | return [idle, total, proc]; | 
|---|
| 607 | } | 
|---|
| 608 | } | 
|---|
| 609 |  | 
|---|
| 610 | //============================================================================================= | 
|---|
| 611 | // Unexpected Terminating logic | 
|---|
| 612 | //============================================================================================= | 
|---|
| 613 | static __spinlock_t kernel_abort_lock; | 
|---|
| 614 | static bool kernel_abort_called = false; | 
|---|
| 615 |  | 
|---|
| 616 | void * kernel_abort(void) __attribute__ ((__nothrow__)) { | 
|---|
| 617 | // abort cannot be recursively entered by the same or different processors because all signal handlers return when | 
|---|
| 618 | // the globalAbort flag is true. | 
|---|
| 619 | lock( kernel_abort_lock __cfaabi_dbg_ctx2 ); | 
|---|
| 620 |  | 
|---|
| 621 | // disable interrupts, it no longer makes sense to try to interrupt this processor | 
|---|
| 622 | disable_interrupts(); | 
|---|
| 623 |  | 
|---|
| 624 | // first task to abort ? | 
|---|
| 625 | if ( kernel_abort_called ) {                    // not first task to abort ? | 
|---|
| 626 | unlock( kernel_abort_lock ); | 
|---|
| 627 |  | 
|---|
| 628 | sigset_t mask; | 
|---|
| 629 | sigemptyset( &mask ); | 
|---|
| 630 | sigaddset( &mask, SIGALRM );            // block SIGALRM signals | 
|---|
| 631 | sigaddset( &mask, SIGUSR1 );            // block SIGALRM signals | 
|---|
| 632 | sigsuspend( &mask );                            // block the processor to prevent further damage during abort | 
|---|
| 633 | _exit( EXIT_FAILURE );                          // if processor unblocks before it is killed, terminate it | 
|---|
| 634 | } | 
|---|
| 635 | else { | 
|---|
| 636 | kernel_abort_called = true; | 
|---|
| 637 | unlock( kernel_abort_lock ); | 
|---|
| 638 | } | 
|---|
| 639 |  | 
|---|
| 640 | return __cfaabi_tls.this_thread; | 
|---|
| 641 | } | 
|---|
| 642 |  | 
|---|
| 643 | void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) { | 
|---|
| 644 | $thread * thrd = ( $thread * ) kernel_data; | 
|---|
| 645 |  | 
|---|
| 646 | if(thrd) { | 
|---|
| 647 | int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd ); | 
|---|
| 648 | __cfaabi_bits_write( STDERR_FILENO, abort_text, len ); | 
|---|
| 649 |  | 
|---|
| 650 | if ( &thrd->self_cor != thrd->curr_cor ) { | 
|---|
| 651 | len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor ); | 
|---|
| 652 | __cfaabi_bits_write( STDERR_FILENO, abort_text, len ); | 
|---|
| 653 | } | 
|---|
| 654 | else { | 
|---|
| 655 | __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 ); | 
|---|
| 656 | } | 
|---|
| 657 | } | 
|---|
| 658 | else { | 
|---|
| 659 | int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" ); | 
|---|
| 660 | __cfaabi_bits_write( STDERR_FILENO, abort_text, len ); | 
|---|
| 661 | } | 
|---|
| 662 | } | 
|---|
| 663 |  | 
|---|
| 664 | int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) { | 
|---|
| 665 | return get_coroutine(kernelTLS().this_thread) == get_coroutine(mainThread) ? 4 : 2; | 
|---|
| 666 | } | 
|---|
| 667 |  | 
|---|
| 668 | static __spinlock_t kernel_debug_lock; | 
|---|
| 669 |  | 
|---|
| 670 | extern "C" { | 
|---|
| 671 | void __cfaabi_bits_acquire() { | 
|---|
| 672 | lock( kernel_debug_lock __cfaabi_dbg_ctx2 ); | 
|---|
| 673 | } | 
|---|
| 674 |  | 
|---|
| 675 | void __cfaabi_bits_release() { | 
|---|
| 676 | unlock( kernel_debug_lock ); | 
|---|
| 677 | } | 
|---|
| 678 | } | 
|---|
| 679 |  | 
|---|
| 680 | //============================================================================================= | 
|---|
| 681 | // Kernel Utilities | 
|---|
| 682 | //============================================================================================= | 
|---|
| 683 | //----------------------------------------------------------------------------- | 
|---|
| 684 | // Locks | 
|---|
| 685 | void  ?{}( semaphore & this, int count = 1 ) { | 
|---|
| 686 | (this.lock){}; | 
|---|
| 687 | this.count = count; | 
|---|
| 688 | (this.waiting){}; | 
|---|
| 689 | } | 
|---|
| 690 | void ^?{}(semaphore & this) {} | 
|---|
| 691 |  | 
|---|
| 692 | bool P(semaphore & this) with( this ){ | 
|---|
| 693 | lock( lock __cfaabi_dbg_ctx2 ); | 
|---|
| 694 | count -= 1; | 
|---|
| 695 | if ( count < 0 ) { | 
|---|
| 696 | // queue current task | 
|---|
| 697 | append( waiting, active_thread() ); | 
|---|
| 698 |  | 
|---|
| 699 | // atomically release spin lock and block | 
|---|
| 700 | unlock( lock ); | 
|---|
| 701 | park(); | 
|---|
| 702 | return true; | 
|---|
| 703 | } | 
|---|
| 704 | else { | 
|---|
| 705 | unlock( lock ); | 
|---|
| 706 | return false; | 
|---|
| 707 | } | 
|---|
| 708 | } | 
|---|
| 709 |  | 
|---|
| 710 | bool V(semaphore & this) with( this ) { | 
|---|
| 711 | $thread * thrd = 0p; | 
|---|
| 712 | lock( lock __cfaabi_dbg_ctx2 ); | 
|---|
| 713 | count += 1; | 
|---|
| 714 | if ( count <= 0 ) { | 
|---|
| 715 | // remove task at head of waiting list | 
|---|
| 716 | thrd = pop_head( waiting ); | 
|---|
| 717 | } | 
|---|
| 718 |  | 
|---|
| 719 | unlock( lock ); | 
|---|
| 720 |  | 
|---|
| 721 | // make new owner | 
|---|
| 722 | unpark( thrd ); | 
|---|
| 723 |  | 
|---|
| 724 | return thrd != 0p; | 
|---|
| 725 | } | 
|---|
| 726 |  | 
|---|
| 727 | bool V(semaphore & this, unsigned diff) with( this ) { | 
|---|
| 728 | $thread * thrd = 0p; | 
|---|
| 729 | lock( lock __cfaabi_dbg_ctx2 ); | 
|---|
| 730 | int release = max(-count, (int)diff); | 
|---|
| 731 | count += diff; | 
|---|
| 732 | for(release) { | 
|---|
| 733 | unpark( pop_head( waiting ) ); | 
|---|
| 734 | } | 
|---|
| 735 |  | 
|---|
| 736 | unlock( lock ); | 
|---|
| 737 |  | 
|---|
| 738 | return thrd != 0p; | 
|---|
| 739 | } | 
|---|
| 740 |  | 
|---|
| 741 | //----------------------------------------------------------------------------- | 
|---|
| 742 | // Debug | 
|---|
| 743 | __cfaabi_dbg_debug_do( | 
|---|
| 744 | extern "C" { | 
|---|
| 745 | void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) { | 
|---|
| 746 | this.prev_name = prev_name; | 
|---|
| 747 | this.prev_thrd = kernelTLS().this_thread; | 
|---|
| 748 | } | 
|---|
| 749 | } | 
|---|
| 750 | ) | 
|---|
| 751 |  | 
|---|
| 752 | //----------------------------------------------------------------------------- | 
|---|
| 753 | // Debug | 
|---|
| 754 | bool threading_enabled(void) __attribute__((const)) { | 
|---|
| 755 | return true; | 
|---|
| 756 | } | 
|---|
| 757 |  | 
|---|
| 758 | //----------------------------------------------------------------------------- | 
|---|
| 759 | // Statistics | 
|---|
| 760 | #if !defined(__CFA_NO_STATISTICS__) | 
|---|
| 761 | void print_halts( processor & this ) { | 
|---|
| 762 | this.print_halts = true; | 
|---|
| 763 | } | 
|---|
| 764 |  | 
|---|
| 765 | void print_stats_now( cluster & this, int flags ) { | 
|---|
| 766 | __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this ); | 
|---|
| 767 | } | 
|---|
| 768 |  | 
|---|
| 769 | extern int __print_alarm_stats; | 
|---|
| 770 | void print_alarm_stats() { | 
|---|
| 771 | __print_alarm_stats = -1; | 
|---|
| 772 | } | 
|---|
| 773 | #endif | 
|---|
| 774 | // Local Variables: // | 
|---|
| 775 | // mode: c // | 
|---|
| 776 | // tab-width: 4 // | 
|---|
| 777 | // End: // | 
|---|