[e660761] | 1 | // |
---|
| 2 | // Cforall Version 1.0.0 Copyright (C) 2020 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/startup.cfa -- |
---|
| 8 | // |
---|
| 9 | // Author : Thierry Delisle |
---|
| 10 | // Created On : Thu Jul 30 15:12:54 2020 |
---|
| 11 | // Last Modified By : |
---|
| 12 | // Last Modified On : |
---|
| 13 | // Update Count : |
---|
| 14 | // |
---|
| 15 | |
---|
| 16 | #define __cforall_thread__ |
---|
| 17 | |
---|
| 18 | // C Includes |
---|
| 19 | #include <errno.h> // errno |
---|
| 20 | #include <string.h> // strerror |
---|
| 21 | #include <unistd.h> // sysconf |
---|
| 22 | extern "C" { |
---|
| 23 | #include <limits.h> // PTHREAD_STACK_MIN |
---|
| 24 | #include <sys/mman.h> // mprotect |
---|
| 25 | #include <sys/resource.h> // getrlimit |
---|
| 26 | } |
---|
| 27 | |
---|
| 28 | // CFA Includes |
---|
| 29 | #include "kernel_private.hfa" |
---|
| 30 | #include "startup.hfa" // STARTUP_PRIORITY_XXX |
---|
| 31 | |
---|
| 32 | //----------------------------------------------------------------------------- |
---|
| 33 | // Some assembly required |
---|
| 34 | #if defined( __i386 ) |
---|
| 35 | #define CtxGet( ctx ) \ |
---|
| 36 | __asm__ volatile ( \ |
---|
| 37 | "movl %%esp,%0\n"\ |
---|
| 38 | "movl %%ebp,%1\n"\ |
---|
| 39 | : "=rm" (ctx.SP),\ |
---|
| 40 | "=rm" (ctx.FP) \ |
---|
| 41 | ) |
---|
| 42 | #elif defined( __x86_64 ) |
---|
| 43 | #define CtxGet( ctx ) \ |
---|
| 44 | __asm__ volatile ( \ |
---|
| 45 | "movq %%rsp,%0\n"\ |
---|
| 46 | "movq %%rbp,%1\n"\ |
---|
| 47 | : "=rm" (ctx.SP),\ |
---|
| 48 | "=rm" (ctx.FP) \ |
---|
| 49 | ) |
---|
| 50 | #elif defined( __ARM_ARCH ) |
---|
| 51 | #define CtxGet( ctx ) __asm__ ( \ |
---|
| 52 | "mov %0,%%sp\n" \ |
---|
| 53 | "mov %1,%%r11\n" \ |
---|
| 54 | : "=rm" (ctx.SP), "=rm" (ctx.FP) ) |
---|
| 55 | #else |
---|
| 56 | #error unknown hardware architecture |
---|
| 57 | #endif |
---|
| 58 | |
---|
| 59 | //----------------------------------------------------------------------------- |
---|
| 60 | // Start and stop routine for the kernel, declared first to make sure they run first |
---|
| 61 | static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) )); |
---|
| 62 | static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) )); |
---|
| 63 | |
---|
| 64 | //----------------------------------------------------------------------------- |
---|
| 65 | // Static Forward Declarations |
---|
| 66 | struct current_stack_info_t; |
---|
| 67 | |
---|
| 68 | static void * __invoke_processor(void * arg); |
---|
| 69 | static void __kernel_first_resume( processor * this ); |
---|
| 70 | static void __kernel_last_resume ( processor * this ); |
---|
| 71 | static void init(processor & this, const char name[], cluster & _cltr); |
---|
| 72 | static void deinit(processor & this); |
---|
| 73 | static void doregister( struct cluster & cltr ); |
---|
| 74 | static void unregister( struct cluster & cltr ); |
---|
| 75 | static void ?{}( $coroutine & this, current_stack_info_t * info); |
---|
| 76 | static void ?{}( $thread & this, current_stack_info_t * info); |
---|
| 77 | static void ?{}(processorCtx_t & this) {} |
---|
| 78 | static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info); |
---|
| 79 | |
---|
| 80 | //----------------------------------------------------------------------------- |
---|
| 81 | // Forward Declarations for other modules |
---|
| 82 | extern void __kernel_alarm_startup(void); |
---|
| 83 | extern void __kernel_alarm_shutdown(void); |
---|
| 84 | extern void __kernel_io_startup (void); |
---|
| 85 | extern void __kernel_io_shutdown(void); |
---|
| 86 | |
---|
| 87 | //----------------------------------------------------------------------------- |
---|
| 88 | // Other Forward Declarations |
---|
[1eb239e4] | 89 | extern void __wake_proc(processor *); |
---|
[e660761] | 90 | |
---|
| 91 | //----------------------------------------------------------------------------- |
---|
| 92 | // Kernel storage |
---|
| 93 | KERNEL_STORAGE(cluster, mainCluster); |
---|
| 94 | KERNEL_STORAGE(processor, mainProcessor); |
---|
| 95 | KERNEL_STORAGE($thread, mainThread); |
---|
| 96 | KERNEL_STORAGE(__stack_t, mainThreadCtx); |
---|
| 97 | KERNEL_STORAGE(io_context, mainPollerThread); |
---|
| 98 | KERNEL_STORAGE(__scheduler_RWLock_t, __scheduler_lock); |
---|
| 99 | #if !defined(__CFA_NO_STATISTICS__) |
---|
| 100 | KERNEL_STORAGE(__stats_t, mainProcStats); |
---|
| 101 | #endif |
---|
| 102 | |
---|
| 103 | cluster * mainCluster; |
---|
| 104 | processor * mainProcessor; |
---|
| 105 | $thread * mainThread; |
---|
| 106 | __scheduler_RWLock_t * __scheduler_lock; |
---|
| 107 | |
---|
| 108 | extern "C" { |
---|
| 109 | struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters; |
---|
| 110 | } |
---|
| 111 | |
---|
| 112 | size_t __page_size = 0; |
---|
| 113 | |
---|
| 114 | //----------------------------------------------------------------------------- |
---|
| 115 | // Global state |
---|
| 116 | thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) @= { |
---|
| 117 | NULL, // cannot use 0p |
---|
| 118 | NULL, |
---|
| 119 | NULL, |
---|
| 120 | { 1, false, false }, |
---|
| 121 | }; |
---|
| 122 | |
---|
| 123 | //----------------------------------------------------------------------------- |
---|
| 124 | // Struct to steal stack |
---|
| 125 | struct current_stack_info_t { |
---|
| 126 | __stack_t * storage; // pointer to stack object |
---|
| 127 | void * base; // base of stack |
---|
| 128 | void * limit; // stack grows towards stack limit |
---|
| 129 | void * context; // address of cfa_context_t |
---|
| 130 | }; |
---|
| 131 | |
---|
| 132 | void ?{}( current_stack_info_t & this ) { |
---|
| 133 | __stack_context_t ctx; |
---|
| 134 | CtxGet( ctx ); |
---|
| 135 | this.base = ctx.FP; |
---|
| 136 | |
---|
| 137 | rlimit r; |
---|
| 138 | getrlimit( RLIMIT_STACK, &r); |
---|
| 139 | size_t size = r.rlim_cur; |
---|
| 140 | |
---|
| 141 | this.limit = (void *)(((intptr_t)this.base) - size); |
---|
| 142 | this.context = &storage_mainThreadCtx; |
---|
| 143 | } |
---|
| 144 | |
---|
| 145 | |
---|
| 146 | |
---|
| 147 | //============================================================================================= |
---|
| 148 | // Kernel Setup logic |
---|
| 149 | //============================================================================================= |
---|
| 150 | //----------------------------------------------------------------------------- |
---|
| 151 | // Kernel boot procedures |
---|
| 152 | static void __kernel_startup(void) { |
---|
| 153 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
| 154 | __cfadbg_print_safe(runtime_core, "Kernel : Starting\n"); |
---|
| 155 | |
---|
| 156 | __page_size = sysconf( _SC_PAGESIZE ); |
---|
| 157 | |
---|
| 158 | __cfa_dbg_global_clusters.list{ __get }; |
---|
| 159 | __cfa_dbg_global_clusters.lock{}; |
---|
| 160 | |
---|
| 161 | // Initialize the global scheduler lock |
---|
| 162 | __scheduler_lock = (__scheduler_RWLock_t*)&storage___scheduler_lock; |
---|
| 163 | (*__scheduler_lock){}; |
---|
| 164 | |
---|
| 165 | // Initialize the main cluster |
---|
| 166 | mainCluster = (cluster *)&storage_mainCluster; |
---|
| 167 | (*mainCluster){"Main Cluster", 0}; |
---|
| 168 | |
---|
| 169 | __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n"); |
---|
| 170 | |
---|
| 171 | // Start by initializing the main thread |
---|
| 172 | // SKULLDUGGERY: the mainThread steals the process main thread |
---|
| 173 | // which will then be scheduled by the mainProcessor normally |
---|
| 174 | mainThread = ($thread *)&storage_mainThread; |
---|
| 175 | current_stack_info_t info; |
---|
| 176 | info.storage = (__stack_t*)&storage_mainThreadCtx; |
---|
| 177 | (*mainThread){ &info }; |
---|
| 178 | |
---|
| 179 | __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n"); |
---|
| 180 | |
---|
| 181 | |
---|
| 182 | |
---|
| 183 | // Construct the processor context of the main processor |
---|
| 184 | void ?{}(processorCtx_t & this, processor * proc) { |
---|
| 185 | (this.__cor){ "Processor" }; |
---|
| 186 | this.__cor.starter = 0p; |
---|
| 187 | this.proc = proc; |
---|
| 188 | } |
---|
| 189 | |
---|
| 190 | void ?{}(processor & this) with( this ) { |
---|
| 191 | ( this.idle ){}; |
---|
| 192 | ( this.terminated ){ 0 }; |
---|
| 193 | ( this.runner ){}; |
---|
| 194 | init( this, "Main Processor", *mainCluster ); |
---|
| 195 | kernel_thread = pthread_self(); |
---|
| 196 | |
---|
| 197 | runner{ &this }; |
---|
| 198 | __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner); |
---|
| 199 | } |
---|
| 200 | |
---|
| 201 | // Initialize the main processor and the main processor ctx |
---|
| 202 | // (the coroutine that contains the processing control flow) |
---|
| 203 | mainProcessor = (processor *)&storage_mainProcessor; |
---|
| 204 | (*mainProcessor){}; |
---|
| 205 | |
---|
| 206 | //initialize the global state variables |
---|
| 207 | kernelTLS.this_processor = mainProcessor; |
---|
| 208 | kernelTLS.this_thread = mainThread; |
---|
| 209 | |
---|
| 210 | #if !defined( __CFA_NO_STATISTICS__ ) |
---|
| 211 | kernelTLS.this_stats = (__stats_t *)& storage_mainProcStats; |
---|
| 212 | __init_stats( kernelTLS.this_stats ); |
---|
| 213 | #endif |
---|
| 214 | |
---|
| 215 | // Enable preemption |
---|
| 216 | __kernel_alarm_startup(); |
---|
| 217 | |
---|
| 218 | // Start IO |
---|
| 219 | __kernel_io_startup(); |
---|
| 220 | |
---|
| 221 | // Add the main thread to the ready queue |
---|
| 222 | // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread |
---|
| 223 | __schedule_thread((__processor_id_t *)mainProcessor, mainThread); |
---|
| 224 | |
---|
| 225 | // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX |
---|
| 226 | // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that |
---|
| 227 | // mainThread is on the ready queue when this call is made. |
---|
| 228 | __kernel_first_resume( kernelTLS.this_processor ); |
---|
| 229 | |
---|
| 230 | |
---|
| 231 | // THE SYSTEM IS NOW COMPLETELY RUNNING |
---|
| 232 | |
---|
| 233 | |
---|
| 234 | // SKULLDUGGERY: The constructor for the mainCluster will call alloc with a dimension of 0 |
---|
| 235 | // malloc *can* return a non-null value, we should free it if that is the case |
---|
| 236 | free( mainCluster->io.ctxs ); |
---|
| 237 | |
---|
| 238 | // Now that the system is up, finish creating systems that need threading |
---|
| 239 | mainCluster->io.ctxs = (io_context *)&storage_mainPollerThread; |
---|
| 240 | mainCluster->io.cnt = 1; |
---|
| 241 | (*mainCluster->io.ctxs){ *mainCluster }; |
---|
| 242 | |
---|
| 243 | __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n"); |
---|
| 244 | |
---|
| 245 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
| 246 | enable_interrupts( __cfaabi_dbg_ctx ); |
---|
| 247 | verify( TL_GET( preemption_state.enabled ) ); |
---|
| 248 | } |
---|
| 249 | |
---|
| 250 | static void __kernel_shutdown(void) { |
---|
| 251 | //Before we start shutting things down, wait for systems that need threading to shutdown |
---|
| 252 | ^(*mainCluster->io.ctxs){}; |
---|
| 253 | mainCluster->io.cnt = 0; |
---|
| 254 | mainCluster->io.ctxs = 0p; |
---|
| 255 | |
---|
| 256 | /* paranoid */ verify( TL_GET( preemption_state.enabled ) ); |
---|
| 257 | disable_interrupts(); |
---|
| 258 | /* paranoid */ verify( ! kernelTLS.preemption_state.enabled ); |
---|
| 259 | |
---|
| 260 | __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n"); |
---|
| 261 | |
---|
| 262 | // SKULLDUGGERY: Notify the mainProcessor it needs to terminates. |
---|
| 263 | // When its coroutine terminates, it return control to the mainThread |
---|
| 264 | // which is currently here |
---|
| 265 | __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE); |
---|
| 266 | __kernel_last_resume( kernelTLS.this_processor ); |
---|
| 267 | mainThread->self_cor.state = Halted; |
---|
| 268 | |
---|
| 269 | // THE SYSTEM IS NOW COMPLETELY STOPPED |
---|
| 270 | |
---|
| 271 | // Disable preemption |
---|
| 272 | __kernel_alarm_shutdown(); |
---|
| 273 | |
---|
| 274 | // Stop IO |
---|
| 275 | __kernel_io_shutdown(); |
---|
| 276 | |
---|
| 277 | // Destroy the main processor and its context in reverse order of construction |
---|
| 278 | // These were manually constructed so we need manually destroy them |
---|
| 279 | void ^?{}(processor & this) with( this ){ |
---|
| 280 | deinit( this ); |
---|
| 281 | |
---|
| 282 | /* paranoid */ verify( this.do_terminate == true ); |
---|
| 283 | __cfaabi_dbg_print_safe("Kernel : destroyed main processor context %p\n", &runner); |
---|
| 284 | } |
---|
| 285 | |
---|
| 286 | ^(*mainProcessor){}; |
---|
| 287 | |
---|
| 288 | // Final step, destroy the main thread since it is no longer needed |
---|
| 289 | |
---|
| 290 | // Since we provided a stack to this taxk it will not destroy anything |
---|
| 291 | /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1)); |
---|
| 292 | ^(*mainThread){}; |
---|
| 293 | |
---|
| 294 | ^(*mainCluster){}; |
---|
| 295 | |
---|
| 296 | ^(*__scheduler_lock){}; |
---|
| 297 | |
---|
| 298 | ^(__cfa_dbg_global_clusters.list){}; |
---|
| 299 | ^(__cfa_dbg_global_clusters.lock){}; |
---|
| 300 | |
---|
| 301 | __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n"); |
---|
| 302 | } |
---|
| 303 | |
---|
| 304 | //============================================================================================= |
---|
| 305 | // Kernel Initial Scheduling logic |
---|
| 306 | //============================================================================================= |
---|
| 307 | |
---|
| 308 | // Context invoker for processors |
---|
| 309 | // This is the entry point for processors (kernel threads) *except* for the main processor |
---|
| 310 | // It effectively constructs a coroutine by stealing the pthread stack |
---|
| 311 | static void * __invoke_processor(void * arg) { |
---|
| 312 | #if !defined( __CFA_NO_STATISTICS__ ) |
---|
| 313 | __stats_t local_stats; |
---|
| 314 | __init_stats( &local_stats ); |
---|
| 315 | kernelTLS.this_stats = &local_stats; |
---|
| 316 | #endif |
---|
| 317 | |
---|
| 318 | processor * proc = (processor *) arg; |
---|
| 319 | kernelTLS.this_processor = proc; |
---|
| 320 | kernelTLS.this_thread = 0p; |
---|
| 321 | kernelTLS.preemption_state.[enabled, disable_count] = [false, 1]; |
---|
| 322 | // SKULLDUGGERY: We want to create a context for the processor coroutine |
---|
| 323 | // which is needed for the 2-step context switch. However, there is no reason |
---|
| 324 | // to waste the perfectly valid stack create by pthread. |
---|
| 325 | current_stack_info_t info; |
---|
| 326 | __stack_t ctx; |
---|
| 327 | info.storage = &ctx; |
---|
| 328 | (proc->runner){ proc, &info }; |
---|
| 329 | |
---|
| 330 | __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage); |
---|
| 331 | |
---|
| 332 | //Set global state |
---|
| 333 | kernelTLS.this_thread = 0p; |
---|
| 334 | |
---|
| 335 | //We now have a proper context from which to schedule threads |
---|
| 336 | __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx); |
---|
| 337 | |
---|
| 338 | // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't |
---|
| 339 | // resume it to start it like it normally would, it will just context switch |
---|
| 340 | // back to here. Instead directly call the main since we already are on the |
---|
| 341 | // appropriate stack. |
---|
| 342 | get_coroutine(proc->runner)->state = Active; |
---|
| 343 | main( proc->runner ); |
---|
| 344 | get_coroutine(proc->runner)->state = Halted; |
---|
| 345 | |
---|
| 346 | // Main routine of the core returned, the core is now fully terminated |
---|
| 347 | __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner); |
---|
| 348 | |
---|
| 349 | #if !defined(__CFA_NO_STATISTICS__) |
---|
| 350 | __tally_stats(proc->cltr->stats, &local_stats); |
---|
| 351 | if( 0 != proc->print_stats ) { |
---|
| 352 | __print_stats( &local_stats, proc->print_stats, true, proc->name, (void*)proc ); |
---|
| 353 | } |
---|
| 354 | #endif |
---|
| 355 | |
---|
| 356 | return 0p; |
---|
| 357 | } |
---|
| 358 | |
---|
| 359 | static void __kernel_first_resume( processor * this ) { |
---|
| 360 | $thread * src = mainThread; |
---|
| 361 | $coroutine * dst = get_coroutine(this->runner); |
---|
| 362 | |
---|
| 363 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
| 364 | |
---|
| 365 | kernelTLS.this_thread->curr_cor = dst; |
---|
| 366 | __stack_prepare( &dst->stack, 65000 ); |
---|
| 367 | __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine); |
---|
| 368 | |
---|
| 369 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
| 370 | |
---|
| 371 | dst->last = &src->self_cor; |
---|
| 372 | dst->starter = dst->starter ? dst->starter : &src->self_cor; |
---|
| 373 | |
---|
| 374 | // make sure the current state is still correct |
---|
| 375 | /* paranoid */ verify(src->state == Ready); |
---|
| 376 | |
---|
| 377 | // context switch to specified coroutine |
---|
| 378 | verify( dst->context.SP ); |
---|
| 379 | __cfactx_switch( &src->context, &dst->context ); |
---|
| 380 | // when __cfactx_switch returns we are back in the src coroutine |
---|
| 381 | |
---|
| 382 | mainThread->curr_cor = &mainThread->self_cor; |
---|
| 383 | |
---|
| 384 | // make sure the current state has been update |
---|
| 385 | /* paranoid */ verify(src->state == Active); |
---|
| 386 | |
---|
| 387 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
| 388 | } |
---|
| 389 | |
---|
| 390 | // KERNEL_ONLY |
---|
| 391 | static void __kernel_last_resume( processor * this ) { |
---|
| 392 | $coroutine * src = &mainThread->self_cor; |
---|
| 393 | $coroutine * dst = get_coroutine(this->runner); |
---|
| 394 | |
---|
| 395 | verify( ! kernelTLS.preemption_state.enabled ); |
---|
| 396 | verify( dst->starter == src ); |
---|
| 397 | verify( dst->context.SP ); |
---|
| 398 | |
---|
| 399 | // SKULLDUGGERY in debug the processors check that the |
---|
| 400 | // stack is still within the limit of the stack limits after running a thread. |
---|
| 401 | // that check doesn't make sense if we context switch to the processor using the |
---|
| 402 | // coroutine semantics. Since this is a special case, use the current context |
---|
| 403 | // info to populate these fields. |
---|
| 404 | __cfaabi_dbg_debug_do( |
---|
| 405 | __stack_context_t ctx; |
---|
| 406 | CtxGet( ctx ); |
---|
| 407 | mainThread->context.SP = ctx.SP; |
---|
| 408 | mainThread->context.FP = ctx.FP; |
---|
| 409 | ) |
---|
| 410 | |
---|
| 411 | // context switch to the processor |
---|
| 412 | __cfactx_switch( &src->context, &dst->context ); |
---|
| 413 | } |
---|
| 414 | |
---|
| 415 | |
---|
| 416 | //============================================================================================= |
---|
| 417 | // Kernel Object Constructors logic |
---|
| 418 | //============================================================================================= |
---|
| 419 | //----------------------------------------------------------------------------- |
---|
| 420 | // Main thread construction |
---|
| 421 | static void ?{}( $coroutine & this, current_stack_info_t * info) with( this ) { |
---|
| 422 | stack.storage = info->storage; |
---|
| 423 | with(*stack.storage) { |
---|
| 424 | limit = info->limit; |
---|
| 425 | base = info->base; |
---|
| 426 | } |
---|
| 427 | __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage; |
---|
| 428 | *istorage |= 0x1; |
---|
| 429 | name = "Main Thread"; |
---|
| 430 | state = Start; |
---|
| 431 | starter = 0p; |
---|
| 432 | last = 0p; |
---|
| 433 | cancellation = 0p; |
---|
| 434 | } |
---|
| 435 | |
---|
| 436 | static void ?{}( $thread & this, current_stack_info_t * info) with( this ) { |
---|
| 437 | ticket = 1; |
---|
| 438 | state = Start; |
---|
| 439 | self_cor{ info }; |
---|
| 440 | curr_cor = &self_cor; |
---|
| 441 | curr_cluster = mainCluster; |
---|
| 442 | self_mon.owner = &this; |
---|
| 443 | self_mon.recursion = 1; |
---|
| 444 | self_mon_p = &self_mon; |
---|
| 445 | link.next = 0p; |
---|
| 446 | link.prev = 0p; |
---|
| 447 | |
---|
| 448 | node.next = 0p; |
---|
| 449 | node.prev = 0p; |
---|
| 450 | doregister(curr_cluster, this); |
---|
| 451 | |
---|
| 452 | monitors{ &self_mon_p, 1, (fptr_t)0 }; |
---|
| 453 | } |
---|
| 454 | |
---|
| 455 | //----------------------------------------------------------------------------- |
---|
| 456 | // Processor |
---|
| 457 | // Construct the processor context of non-main processors |
---|
| 458 | static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) { |
---|
| 459 | (this.__cor){ info }; |
---|
| 460 | this.proc = proc; |
---|
| 461 | } |
---|
| 462 | |
---|
| 463 | static void init(processor & this, const char name[], cluster & _cltr) with( this ) { |
---|
| 464 | this.name = name; |
---|
| 465 | this.cltr = &_cltr; |
---|
| 466 | id = -1u; |
---|
| 467 | destroyer = 0p; |
---|
| 468 | do_terminate = false; |
---|
| 469 | preemption_alarm = 0p; |
---|
| 470 | pending_preemption = false; |
---|
| 471 | |
---|
| 472 | #if !defined(__CFA_NO_STATISTICS__) |
---|
| 473 | print_stats = 0; |
---|
| 474 | print_halts = false; |
---|
| 475 | #endif |
---|
| 476 | |
---|
[1eb239e4] | 477 | lock( this.cltr->idles ); |
---|
| 478 | int target = this.cltr->idles.total += 1u; |
---|
| 479 | unlock( this.cltr->idles ); |
---|
[e660761] | 480 | |
---|
| 481 | id = doregister((__processor_id_t*)&this); |
---|
| 482 | |
---|
| 483 | // Lock the RWlock so no-one pushes/pops while we are changing the queue |
---|
| 484 | uint_fast32_t last_size = ready_mutate_lock(); |
---|
| 485 | |
---|
| 486 | // Adjust the ready queue size |
---|
| 487 | ready_queue_grow( cltr, target ); |
---|
| 488 | |
---|
| 489 | // Unlock the RWlock |
---|
| 490 | ready_mutate_unlock( last_size ); |
---|
| 491 | |
---|
| 492 | __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this); |
---|
| 493 | } |
---|
| 494 | |
---|
| 495 | // Not a ctor, it just preps the destruction but should not destroy members |
---|
| 496 | static void deinit(processor & this) { |
---|
[1eb239e4] | 497 | lock( this.cltr->idles ); |
---|
| 498 | int target = this.cltr->idles.total -= 1u; |
---|
| 499 | unlock( this.cltr->idles ); |
---|
[e660761] | 500 | |
---|
| 501 | // Lock the RWlock so no-one pushes/pops while we are changing the queue |
---|
| 502 | uint_fast32_t last_size = ready_mutate_lock(); |
---|
| 503 | |
---|
| 504 | // Adjust the ready queue size |
---|
| 505 | ready_queue_shrink( this.cltr, target ); |
---|
| 506 | |
---|
| 507 | // Unlock the RWlock |
---|
| 508 | ready_mutate_unlock( last_size ); |
---|
| 509 | |
---|
| 510 | // Finally we don't need the read_lock any more |
---|
| 511 | unregister((__processor_id_t*)&this); |
---|
| 512 | } |
---|
| 513 | |
---|
| 514 | void ?{}(processor & this, const char name[], cluster & _cltr) { |
---|
| 515 | ( this.idle ){}; |
---|
| 516 | ( this.terminated ){ 0 }; |
---|
| 517 | ( this.runner ){}; |
---|
| 518 | init( this, name, _cltr ); |
---|
| 519 | |
---|
| 520 | __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this); |
---|
| 521 | |
---|
| 522 | this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this ); |
---|
| 523 | |
---|
| 524 | } |
---|
| 525 | |
---|
| 526 | void ^?{}(processor & this) with( this ){ |
---|
| 527 | if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) { |
---|
| 528 | __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this); |
---|
| 529 | |
---|
| 530 | __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED); |
---|
| 531 | __wake_proc( &this ); |
---|
| 532 | |
---|
| 533 | P( terminated ); |
---|
| 534 | verify( kernelTLS.this_processor != &this); |
---|
| 535 | } |
---|
| 536 | |
---|
| 537 | int err = pthread_join( kernel_thread, 0p ); |
---|
| 538 | if( err != 0 ) abort("KERNEL ERROR: joining processor %p caused error %s\n", &this, strerror(err)); |
---|
| 539 | |
---|
| 540 | free( this.stack ); |
---|
| 541 | |
---|
| 542 | deinit( this ); |
---|
| 543 | } |
---|
| 544 | |
---|
| 545 | //----------------------------------------------------------------------------- |
---|
| 546 | // Cluster |
---|
[1eb239e4] | 547 | static void ?{}(__cluster_idles & this) { |
---|
| 548 | this.lock = 0; |
---|
| 549 | this.idle = 0; |
---|
| 550 | this.total = 0; |
---|
| 551 | (this.list){}; |
---|
| 552 | } |
---|
| 553 | |
---|
[e660761] | 554 | void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params) with( this ) { |
---|
| 555 | this.name = name; |
---|
| 556 | this.preemption_rate = preemption_rate; |
---|
| 557 | ready_queue{}; |
---|
| 558 | |
---|
| 559 | #if !defined(__CFA_NO_STATISTICS__) |
---|
| 560 | print_stats = 0; |
---|
| 561 | stats = alloc(); |
---|
| 562 | __init_stats( stats ); |
---|
| 563 | #endif |
---|
| 564 | |
---|
| 565 | threads{ __get }; |
---|
| 566 | |
---|
| 567 | doregister(this); |
---|
| 568 | |
---|
| 569 | // Lock the RWlock so no-one pushes/pops while we are changing the queue |
---|
| 570 | uint_fast32_t last_size = ready_mutate_lock(); |
---|
| 571 | |
---|
| 572 | // Adjust the ready queue size |
---|
| 573 | ready_queue_grow( &this, 0 ); |
---|
| 574 | |
---|
| 575 | // Unlock the RWlock |
---|
| 576 | ready_mutate_unlock( last_size ); |
---|
| 577 | |
---|
| 578 | this.io.cnt = num_io; |
---|
| 579 | this.io.ctxs = aalloc(num_io); |
---|
| 580 | for(i; this.io.cnt) { |
---|
| 581 | (this.io.ctxs[i]){ this, io_params }; |
---|
| 582 | } |
---|
| 583 | } |
---|
| 584 | |
---|
| 585 | void ^?{}(cluster & this) { |
---|
| 586 | for(i; this.io.cnt) { |
---|
| 587 | ^(this.io.ctxs[i]){ true }; |
---|
| 588 | } |
---|
| 589 | free(this.io.ctxs); |
---|
| 590 | |
---|
| 591 | // Lock the RWlock so no-one pushes/pops while we are changing the queue |
---|
| 592 | uint_fast32_t last_size = ready_mutate_lock(); |
---|
| 593 | |
---|
| 594 | // Adjust the ready queue size |
---|
| 595 | ready_queue_shrink( &this, 0 ); |
---|
| 596 | |
---|
| 597 | // Unlock the RWlock |
---|
| 598 | ready_mutate_unlock( last_size ); |
---|
| 599 | |
---|
| 600 | #if !defined(__CFA_NO_STATISTICS__) |
---|
| 601 | if( 0 != this.print_stats ) { |
---|
| 602 | __print_stats( this.stats, this.print_stats, true, this.name, (void*)&this ); |
---|
| 603 | } |
---|
| 604 | free( this.stats ); |
---|
| 605 | #endif |
---|
| 606 | |
---|
| 607 | unregister(this); |
---|
| 608 | } |
---|
| 609 | |
---|
| 610 | //============================================================================================= |
---|
| 611 | // Miscellaneous Initialization |
---|
| 612 | //============================================================================================= |
---|
| 613 | //----------------------------------------------------------------------------- |
---|
| 614 | // Global Queues |
---|
| 615 | static void doregister( cluster & cltr ) { |
---|
| 616 | lock ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2); |
---|
| 617 | push_front( __cfa_dbg_global_clusters.list, cltr ); |
---|
| 618 | unlock ( __cfa_dbg_global_clusters.lock ); |
---|
| 619 | } |
---|
| 620 | |
---|
| 621 | static void unregister( cluster & cltr ) { |
---|
| 622 | lock ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2); |
---|
| 623 | remove( __cfa_dbg_global_clusters.list, cltr ); |
---|
| 624 | unlock( __cfa_dbg_global_clusters.lock ); |
---|
| 625 | } |
---|
| 626 | |
---|
| 627 | void doregister( cluster * cltr, $thread & thrd ) { |
---|
| 628 | lock (cltr->thread_list_lock __cfaabi_dbg_ctx2); |
---|
| 629 | cltr->nthreads += 1; |
---|
| 630 | push_front(cltr->threads, thrd); |
---|
| 631 | unlock (cltr->thread_list_lock); |
---|
| 632 | } |
---|
| 633 | |
---|
| 634 | void unregister( cluster * cltr, $thread & thrd ) { |
---|
| 635 | lock (cltr->thread_list_lock __cfaabi_dbg_ctx2); |
---|
| 636 | remove(cltr->threads, thrd ); |
---|
| 637 | cltr->nthreads -= 1; |
---|
| 638 | unlock(cltr->thread_list_lock); |
---|
| 639 | } |
---|
| 640 | |
---|
| 641 | static void check( int ret, const char func[] ) { |
---|
| 642 | if ( ret ) { // pthread routines return errno values |
---|
| 643 | abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) ); |
---|
| 644 | } // if |
---|
| 645 | } // Abort |
---|
| 646 | |
---|
| 647 | void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) { |
---|
| 648 | pthread_attr_t attr; |
---|
| 649 | |
---|
| 650 | check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute |
---|
| 651 | |
---|
| 652 | size_t stacksize; |
---|
| 653 | // default stack size, normally defined by shell limit |
---|
| 654 | check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" ); |
---|
| 655 | assert( stacksize >= PTHREAD_STACK_MIN ); |
---|
| 656 | |
---|
| 657 | void * stack; |
---|
| 658 | __cfaabi_dbg_debug_do( |
---|
| 659 | stack = memalign( __page_size, stacksize + __page_size ); |
---|
| 660 | // pthread has no mechanism to create the guard page in user supplied stack. |
---|
| 661 | if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) { |
---|
| 662 | abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) ); |
---|
| 663 | } // if |
---|
| 664 | ); |
---|
| 665 | __cfaabi_dbg_no_debug_do( |
---|
| 666 | stack = malloc( stacksize ); |
---|
| 667 | ); |
---|
| 668 | |
---|
| 669 | check( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" ); |
---|
| 670 | |
---|
| 671 | check( pthread_create( pthread, &attr, start, arg ), "pthread_create" ); |
---|
| 672 | return stack; |
---|
| 673 | } |
---|