Context Navigation

kernel.cfa@ d824715

Visit:

ADT ast-experimental enum pthread-emulation qualifiedEnum

Last change on this file since d824715 was bfb9bf5, checked in by Thierry Delisle <tdelisle@…>, 4 years ago
Fixed some warnings
Property mode set to `100644`
File size: 28.9 KB

Rev	Line
[8118303]	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
[75f3522]	10	// Created On : Tue Jan 17 12:27:26 2017
[6b0b624]	11	// Last Modified By : Peter A. Buhr
[0190480]	12	// Last Modified On : Mon Aug 31 07:08:20 2020
	13	// Update Count : 71
[8118303]	14	//
	15
[2026bb6]	16	#define __cforall_thread__
[43784ac]	17	#define _GNU_SOURCE
	18
[4069faad]	19	// #define __CFA_DEBUG_PRINT_RUNTIME_CORE__
[2026bb6]	20
[bfb9bf5]	21	#pragma GCC diagnostic push
	22	#pragma GCC diagnostic ignored "-Waddress-of-packed-member"
	23
[8118303]	24	//C Includes
[214e8da]	25	#include <errno.h>
[9d944b2]	26	#include <stdio.h>
[445f984]	27	#include <string.h>
[58b6d1b]	28	#include <signal.h>
[9d944b2]	29	#include <unistd.h>
[bfb9bf5]	30
[dddb3dd0]	31	extern "C" {
	32	#include <sys/eventfd.h>
[d3605f8]	33	#include <sys/uio.h>
[dddb3dd0]	34	}
[8118303]	35
	36	//CFA Includes
[73abe95]	37	#include "kernel_private.hfa"
	38	#include "preemption.hfa"
[445f984]	39	#include "strstream.hfa"
	40	#include "device/cpu.hfa"
[7ef162b2]	41	#include "io/types.hfa"
[8118303]	42
	43	//Private includes
	44	#define __CFA_INVOKE_PRIVATE__
	45	#include "invoke.h"
[bfb9bf5]	46	#pragma GCC diagnostic pop
[8118303]	47
[89eff25]	48	#if !defined(__CFA_NO_STATISTICS__)
[c9c1c1cb]	49	#define __STATS_DEF( ...) __VA_ARGS__
[89eff25]	50	#else
[c9c1c1cb]	51	#define __STATS_DEF( ...)
[89eff25]	52	#endif
[4069faad]	53
[deca0f5]	54	//-----------------------------------------------------------------------------
	55	// Some assembly required
[1805b1b]	56	#if defined( __i386 )
[deca0f5]	57	// mxcr : SSE Status and Control bits (control bits are preserved across function calls)
	58	// fcw : X87 FPU control word (preserved across function calls)
	59	#define __x87_store \
	60	uint32_t __mxcr; \
	61	uint16_t __fcw; \
	62	__asm__ volatile ( \
	63	"stmxcsr %0\n" \
	64	"fnstcw %1\n" \
	65	: "=m" (__mxcr),\
	66	"=m" (__fcw) \
	67	)
	68
	69	#define __x87_load \
	70	__asm__ volatile ( \
	71	"fldcw %1\n" \
	72	"ldmxcsr %0\n" \
	73	::"m" (__mxcr),\
	74	"m" (__fcw) \
	75	)
	76
	77	#elif defined( __x86_64 )
	78	#define __x87_store \
	79	uint32_t __mxcr; \
	80	uint16_t __fcw; \
	81	__asm__ volatile ( \
	82	"stmxcsr %0\n" \
	83	"fnstcw %1\n" \
	84	: "=m" (__mxcr),\
	85	"=m" (__fcw) \
	86	)
	87
	88	#define __x87_load \
	89	__asm__ volatile ( \
	90	"fldcw %1\n" \
	91	"ldmxcsr %0\n" \
	92	:: "m" (__mxcr),\
	93	"m" (__fcw) \
	94	)
	95
[0190480]	96	#elif defined( __arm__ )
	97	#define __x87_store
	98	#define __x87_load
	99
	100	#elif defined( __aarch64__ )
[74f5c83]	101	#define __x87_store \
	102	uint32_t __fpcntl[2]; \
	103	__asm__ volatile ( \
	104	"mrs x9, FPCR\n" \
	105	"mrs x10, FPSR\n" \
	106	"stp x9, x10, %0\n" \
	107	: "=m" (__fpcntl) : : "x9", "x10" \
	108	)
	109
	110	#define __x87_load \
	111	__asm__ volatile ( \
	112	"ldp x9, x10, %0\n" \
	113	"msr FPSR, x10\n" \
	114	"msr FPCR, x9\n" \
	115	: "=m" (__fpcntl) : : "x9", "x10" \
	116	)
	117
[deca0f5]	118	#else
[0190480]	119	#error unsupported hardware architecture
[deca0f5]	120	#endif
	121
[e84ab3d]	122	extern thread$ * mainThread;
[e660761]	123	extern processor * mainProcessor;
[2ac095d]	124
[92e7631]	125	//-----------------------------------------------------------------------------
	126	// Kernel Scheduling logic
[e84ab3d]	127	static thread$ * __next_thread(cluster * this);
	128	static thread$ * __next_thread_slow(cluster * this);
[c9c1c1cb]	129	static thread$ * __next_thread_search(cluster * this);
[e84ab3d]	130	static inline bool __must_unpark( thread$ * thrd ) __attribute((nonnull(1)));
	131	static void __run_thread(processor * this, thread$ * dst);
[e873838]	132	static void __wake_one(cluster * cltr);
[1eb239e4]	133
[d3605f8]	134	static void idle_sleep(processor * proc, io_future_t & future, iovec & iov);
[5f5a729]	135	static bool mark_idle (__cluster_proc_list & idles, processor & proc);
[6a9b12b]	136	static void mark_awake(__cluster_proc_list & idles, processor & proc);
[1eb239e4]	137
[dddb3dd0]	138	extern void __cfa_io_start( processor * );
[c1c95b1]	139	extern bool __cfa_io_drain( processor * );
[c7b2215]	140	extern bool __cfa_io_flush( processor *, int min_comp );
[dddb3dd0]	141	extern void __cfa_io_stop ( processor * );
[c1c95b1]	142	static inline bool __maybe_io_drain( processor * );
[dddb3dd0]	143
[d3605f8]	144	#if defined(CFA_WITH_IO_URING_IDLE)
	145	extern bool __kernel_read(processor * proc, io_future_t & future, iovec &, int fd);
[6ddef36]	146	#endif
[7ef162b2]	147
[dddb3dd0]	148	extern void __disable_interrupts_hard();
	149	extern void __enable_interrupts_hard();
[c84e80a]	150
[7ef162b2]	151
[75f3522]	152	//=============================================================================================
	153	// Kernel Scheduling logic
	154	//=============================================================================================
[8fcbb4c]	155	//Main of the processor contexts
[83a071f9]	156	void main(processorCtx_t & runner) {
[21184e3]	157	// Because of a bug, we couldn't initialized the seed on construction
	158	// Do it here
[8fc652e0]	159	__cfaabi_tls.rand_seed ^= rdtscl();
	160	__cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner);
[f2384c9a]	161	__tls_rand_advance_bck();
[21184e3]	162
[83a071f9]	163	processor * this = runner.proc;
[094476d]	164	verify(this);
[c81ebf9]	165
[7ef162b2]	166	io_future_t future; // used for idle sleep when io_uring is present
	167	future.self.ptr = 1p; // mark it as already fulfilled so we know if there is a pending request or not
[c7b2215]	168	eventfd_t idle_val;
[d3605f8]	169	iovec idle_iovec = { &idle_val, sizeof(idle_val) };
[7ef162b2]	170
[dddb3dd0]	171	__cfa_io_start( this );
	172
[4069faad]	173	__cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
[28d73c1]	174	#if !defined(__CFA_NO_STATISTICS__)
	175	if( this->print_halts ) {
[c993b15]	176	__cfaabi_bits_print_safe( STDOUT_FILENO, "Processor : %d - %s (%p)\n", this->unique_id, this->name, (void*)this);
[28d73c1]	177	}
	178	#endif
[b798713]	179
[75f3522]	180	{
[c81ebf9]	181	// Setup preemption data
	182	preemption_scope scope = { this };
	183
[a5e7233]	184	// if we need to run some special setup, now is the time to do it.
	185	if(this->init.thrd) {
	186	this->init.thrd->curr_cluster = this->cltr;
	187	__run_thread(this, this->init.thrd);
	188	}
[325e6ea]	189
[4069faad]	190	__cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
[8118303]	191
[e84ab3d]	192	thread$ * readyThread = 0p;
[1eb239e4]	193	MAIN_LOOP:
	194	for() {
[dddb3dd0]	195	// Check if there is pending io
	196	__maybe_io_drain( this );
	197
[92e7631]	198	// Try to get the next thread
[8c50aed]	199	readyThread = __next_thread( this->cltr );
[75f3522]	200
[1eb239e4]	201	if( !readyThread ) {
[c9c1c1cb]	202	__IO_STATS__(true, io.flush.idle++; )
[c7b2215]	203	__cfa_io_flush( this, 0 );
[c33c2af]	204
[c9c1c1cb]	205	readyThread = __next_thread( this->cltr );
	206	}
	207
	208	if( !readyThread ) for(5) {
	209	__IO_STATS__(true, io.flush.idle++; )
	210
[1eb239e4]	211	readyThread = __next_thread_slow( this->cltr );
[c9c1c1cb]	212
	213	if( readyThread ) break;
	214
	215	__cfa_io_flush( this, 0 );
[1eb239e4]	216	}
[4e6fb8e]	217
[1eb239e4]	218	HALT:
	219	if( !readyThread ) {
	220	// Don't block if we are done
	221	if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[c81ebf9]	222
[1eb239e4]	223	// Push self to idle stack
[5f5a729]	224	if(!mark_idle(this->cltr->procs, * this)) continue MAIN_LOOP;
[398e8e9]	225
[1eb239e4]	226	// Confirm the ready-queue is empty
[c9c1c1cb]	227	readyThread = __next_thread_search( this->cltr );
[1eb239e4]	228	if( readyThread ) {
	229	// A thread was found, cancel the halt
[6a9b12b]	230	mark_awake(this->cltr->procs, * this);
[1eb239e4]	231
[c9c1c1cb]	232	__STATS__(true, ready.sleep.cancels++; )
[1eb239e4]	233
	234	// continue the mai loop
	235	break HALT;
	236	}
	237
[d3605f8]	238	idle_sleep( this, future, idle_iovec );
[1eb239e4]	239
	240	// We were woken up, remove self from idle
[6a9b12b]	241	mark_awake(this->cltr->procs, * this);
[1eb239e4]	242
	243	// DON'T just proceed, start looking again
	244	continue MAIN_LOOP;
[64a7146]	245	}
[1eb239e4]	246
	247	/* paranoid */ verify( readyThread );
	248
[dddb3dd0]	249	// Reset io dirty bit
	250	this->io.dirty = false;
	251
[1eb239e4]	252	// We found a thread run it
	253	__run_thread(this, readyThread);
	254
	255	// Are we done?
	256	if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[325e6ea]	257
[d529ad0]	258	if(__atomic_load_n(&this->io.pending, __ATOMIC_RELAXED) && !__atomic_load_n(&this->io.dirty, __ATOMIC_RELAXED)) {
[c9c1c1cb]	259	__IO_STATS__(true, io.flush.dirty++; )
[c7b2215]	260	__cfa_io_flush( this, 0 );
[dddb3dd0]	261	}
[c81ebf9]	262	}
	263
[4069faad]	264	__cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
[c84e80a]	265	}
[8118303]	266
[3bb4f85]	267	for(int i = 0; !available(future); i++) {
	268	if(i > 1000) __cfaabi_dbg_write( "ERROR: kernel has bin spinning on a flush after exit loop.\n", 60);
	269	__cfa_io_flush( this, 1 );
	270	}
	271
[dddb3dd0]	272	__cfa_io_stop( this );
	273
[454f478]	274	post( this->terminated );
[bdeba0b]	275
[28d73c1]	276	if(this == mainProcessor) {
[6a490b2]	277	// HACK : the coroutine context switch expects this_thread to be set
	278	// and it make sense for it to be set in all other cases except here
	279	// fake it
[8fc652e0]	280	__cfaabi_tls.this_thread = mainThread;
[6a490b2]	281	}
[7768b8d]	282
[4069faad]	283	__cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
[c84e80a]	284	}
	285
[5c1a531]	286	static int * __volatile_errno() __attribute__((noinline));
	287	static int * __volatile_errno() { asm(""); return &errno; }
	288
[14a61b5]	289	// KERNEL ONLY
[1c273d0]	290	// runThread runs a thread by context switching
	291	// from the processor coroutine to the target thread
[e84ab3d]	292	static void __run_thread(processor * this, thread$ * thrd_dst) {
[8fc652e0]	293	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	294	/* paranoid */ verifyf( thrd_dst->state == Ready \|\| thrd_dst->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", thrd_dst->state, thrd_dst->preempted);
	295	/* paranoid */ verifyf( thrd_dst->link.next == 0p, "Expected null got %p", thrd_dst->link.next );
	296	__builtin_prefetch( thrd_dst->context.SP );
	297
[dddb3dd0]	298	__cfadbg_print_safe(runtime_core, "Kernel : core %p running thread %p (%s)\n", this, thrd_dst, thrd_dst->self_cor.name);
	299
[e84ab3d]	300	coroutine$ * proc_cor = get_coroutine(this->runner);
[8fcbb4c]	301
[9f575ea]	302	// set state of processor coroutine to inactive
	303	verify(proc_cor->state == Active);
[ae7be7a]	304	proc_cor->state = Blocked;
[e8e457e]	305
[9f575ea]	306	// Actually run the thread
[3381ed7]	307	RUNNING: while(true) {
[ff79d5e]	308	thrd_dst->preempted = __NO_PREEMPTION;
	309	thrd_dst->state = Active;
[e8e457e]	310
[58d64a4]	311	// Update global state
[8fc652e0]	312	kernelTLS().this_thread = thrd_dst;
[75f3522]	313
[8fc652e0]	314	/* paranoid */ verify( ! __preemption_enabled() );
	315	/* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
[9d6e1b8a]	316	/* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[b4b63e8]	317	/* paranoid */ verify( thrd_dst->context.SP );
[5afb49a]	318	/* paranoid */ verify( thrd_dst->state != Halted );
[e84ab3d]	319	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ) \|\| thrd_dst->curr_cor == proc_cor \|\| thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too small.\n", thrd_dst ); // add escape condition if we are setting up the processor
	320	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit) \|\| thrd_dst->curr_cor == proc_cor \|\| thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too large.\n", thrd_dst ); // add escape condition if we are setting up the processor
[ac12f1f]	321	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[b4b63e8]	322
[3381ed7]	323
[58d64a4]	324
[9f575ea]	325	// set context switch to the thread that the processor is executing
[c7a900a]	326	__cfactx_switch( &proc_cor->context, &thrd_dst->context );
	327	// when __cfactx_switch returns we are back in the processor coroutine
[9f575ea]	328
[50871b4]	329
	330
[ac12f1f]	331	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[e84ab3d]	332	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit) \|\| thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too large.\n", thrd_dst );
	333	/* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ) \|\| thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too small.\n", thrd_dst );
[b4b63e8]	334	/* paranoid */ verify( thrd_dst->context.SP );
[9d6e1b8a]	335	/* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[8fc652e0]	336	/* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
	337	/* paranoid */ verify( ! __preemption_enabled() );
[75f3522]	338
[58d64a4]	339	// Reset global state
[8fc652e0]	340	kernelTLS().this_thread = 0p;
[3381ed7]	341
	342	// We just finished running a thread, there are a few things that could have happened.
	343	// 1 - Regular case : the thread has blocked and now one has scheduled it yet.
	344	// 2 - Racy case : the thread has blocked but someone has already tried to schedule it.
	345	// 4 - Preempted
	346	// In case 1, we may have won a race so we can't write to the state again.
	347	// In case 2, we lost the race so we now own the thread.
	348
	349	if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
	350	// The thread was preempted, reschedule it and reset the flag
[24e321c]	351	schedule_thread$( thrd_dst, UNPARK_LOCAL );
[3381ed7]	352	break RUNNING;
	353	}
[75f3522]	354
[3ea8ad1]	355	if(unlikely(thrd_dst->state == Halting)) {
[ff79d5e]	356	// The thread has halted, it should never be scheduled/run again
[5afb49a]	357	// finish the thread
	358	__thread_finish( thrd_dst );
[ff79d5e]	359	break RUNNING;
	360	}
	361
	362	/* paranoid */ verify( thrd_dst->state == Active );
	363	thrd_dst->state = Blocked;
	364
[3381ed7]	365	// set state of processor coroutine to active and the thread to inactive
[ff79d5e]	366	int old_ticket = __atomic_fetch_sub(&thrd_dst->ticket, 1, __ATOMIC_SEQ_CST);
	367	switch(old_ticket) {
[6a77224]	368	case TICKET_RUNNING:
[3381ed7]	369	// This is case 1, the regular case, nothing more is needed
	370	break RUNNING;
[6a77224]	371	case TICKET_UNBLOCK:
[c9c1c1cb]	372	__STATS__(true, ready.threads.threads++; )
[3381ed7]	373	// This is case 2, the racy case, someone tried to run this thread before it finished blocking
	374	// In this case, just run it again.
	375	continue RUNNING;
	376	default:
	377	// This makes no sense, something is wrong abort
[ff79d5e]	378	abort();
[3381ed7]	379	}
[9f575ea]	380	}
[e8e457e]	381
[9f575ea]	382	// Just before returning to the processor, set the processor coroutine to active
[e8e457e]	383	proc_cor->state = Active;
[1eb239e4]	384
[dddb3dd0]	385	__cfadbg_print_safe(runtime_core, "Kernel : core %p finished running thread %p\n", this, thrd_dst);
	386
[c9c1c1cb]	387	__STATS__(true, ready.threads.threads--; )
[ec43cf9]	388
[8fc652e0]	389	/* paranoid */ verify( ! __preemption_enabled() );
[82c948c]	390	}
	391
[14a61b5]	392	// KERNEL_ONLY
[b0c7419]	393	void returnToKernel() {
[8fc652e0]	394	/* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]	395	coroutine$ * proc_cor = get_coroutine(kernelTLS().this_processor->runner);
	396	thread$ * thrd_src = kernelTLS().this_thread;
[e8e457e]	397
[c9c1c1cb]	398	__STATS_DEF( thrd_src->last_proc = kernelTLS().this_processor; )
[29cb302]	399
[9f575ea]	400	// Run the thread on this processor
	401	{
	402	int local_errno = *__volatile_errno();
	403	#if defined( __i386 ) \|\| defined( __x86_64 )
	404	__x87_store;
	405	#endif
[b4b63e8]	406	/* paranoid */ verify( proc_cor->context.SP );
[ac12f1f]	407	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[c7a900a]	408	__cfactx_switch( &thrd_src->context, &proc_cor->context );
[ac12f1f]	409	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[9f575ea]	410	#if defined( __i386 ) \|\| defined( __x86_64 )
	411	__x87_load;
	412	#endif
	413	*__volatile_errno() = local_errno;
[8fcbb4c]	414	}
[deca0f5]	415
[29cb302]	416	#if !defined(__CFA_NO_STATISTICS__)
[89eff25]	417	/* paranoid */ verify( thrd_src->last_proc != 0p );
	418	if(thrd_src->last_proc != kernelTLS().this_processor) {
[29cb302]	419	__tls_stats()->ready.threads.migration++;
	420	}
	421	#endif
	422
[8fc652e0]	423	/* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]	424	/* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) < ((uintptr_t)__get_stack(thrd_src->curr_cor)->base ) \|\| thrd_src->corctx_flag, "ERROR : Returning thread$ %p has been corrupted.\n StackPointer too small.\n", thrd_src );
	425	/* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) > ((uintptr_t)__get_stack(thrd_src->curr_cor)->limit) \|\| thrd_src->corctx_flag, "ERROR : Returning thread$ %p has been corrupted.\n StackPointer too large.\n", thrd_src );
[c84e80a]	426	}
	427
[8def349]	428	//-----------------------------------------------------------------------------
	429	// Scheduler routines
[14a61b5]	430	// KERNEL ONLY
[24e321c]	431	static void __schedule_thread( thread$ * thrd, unpark_hint hint ) {
[8fc652e0]	432	/* paranoid */ verify( ! __preemption_enabled() );
[254ad1b]	433	/* paranoid */ verify( ready_schedule_islocked());
[6a490b2]	434	/* paranoid */ verify( thrd );
	435	/* paranoid */ verify( thrd->state != Halted );
[9d6e1b8a]	436	/* paranoid */ verify( thrd->curr_cluster );
[3381ed7]	437	/* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
[504a7dc]	438	/* paranoid */ if( thrd->state == Blocked \|\| thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
	439	"Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
[ff79d5e]	440	/* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active,
[504a7dc]	441	"Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
[3381ed7]	442	/* paranoid */ #endif
[6a490b2]	443	/* paranoid */ verifyf( thrd->link.next == 0p, "Expected null got %p", thrd->link.next );
[ac12f1f]	444	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
[b4b63e8]	445
[ae7be7a]	446	if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
[6b4cdd3]	447
[ec43cf9]	448	// Dereference the thread now because once we push it, there is not guaranteed it's still valid.
	449	struct cluster * cl = thrd->curr_cluster;
[c9c1c1cb]	450	__STATS_DEF(bool outside = hint == UNPARK_LOCAL && thrd->last_proc && thrd->last_proc != kernelTLS().this_processor; )
[32a8b61]	451
[254ad1b]	452	// push the thread to the cluster ready-queue
[24e321c]	453	push( cl, thrd, hint );
[32a8b61]	454
[254ad1b]	455	// variable thrd is no longer safe to use
[734908c]	456	thrd = 0xdeaddeaddeaddeadp;
[32a8b61]	457
[254ad1b]	458	// wake the cluster using the save variable.
	459	__wake_one( cl );
[1c273d0]	460
[ec43cf9]	461	#if !defined(__CFA_NO_STATISTICS__)
	462	if( kernelTLS().this_stats ) {
	463	__tls_stats()->ready.threads.threads++;
[89eff25]	464	if(outside) {
	465	__tls_stats()->ready.threads.extunpark++;
	466	}
[ec43cf9]	467	}
	468	else {
	469	__atomic_fetch_add(&cl->stats->ready.threads.threads, 1, __ATOMIC_RELAXED);
[89eff25]	470	__atomic_fetch_add(&cl->stats->ready.threads.extunpark, 1, __ATOMIC_RELAXED);
[ec43cf9]	471	}
	472	#endif
	473
[254ad1b]	474	/* paranoid */ verify( ready_schedule_islocked());
[8fc652e0]	475	/* paranoid */ verify( ! __preemption_enabled() );
[db6f06a]	476	}
	477
[24e321c]	478	void schedule_thread$( thread$ * thrd, unpark_hint hint ) {
[254ad1b]	479	ready_schedule_lock();
[24e321c]	480	__schedule_thread( thrd, hint );
[254ad1b]	481	ready_schedule_unlock();
	482	}
	483
[14a61b5]	484	// KERNEL ONLY
[e84ab3d]	485	static inline thread$ * __next_thread(cluster * this) with( *this ) {
[8fc652e0]	486	/* paranoid */ verify( ! __preemption_enabled() );
[7768b8d]	487
[e873838]	488	ready_schedule_lock();
[e84ab3d]	489	thread$ * thrd = pop_fast( this );
[e873838]	490	ready_schedule_unlock();
[7768b8d]	491
[8fc652e0]	492	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	493	return thrd;
[eb2e723]	494	}
	495
[64a7146]	496	// KERNEL ONLY
[e84ab3d]	497	static inline thread$ * __next_thread_slow(cluster * this) with( *this ) {
[8fc652e0]	498	/* paranoid */ verify( ! __preemption_enabled() );
[64a7146]	499
[e873838]	500	ready_schedule_lock();
[c9c1c1cb]	501	thread$ * thrd = pop_slow( this );
	502	ready_schedule_unlock();
	503
	504	/* paranoid */ verify( ! __preemption_enabled() );
	505	return thrd;
	506	}
[fc59df78]	507
[c9c1c1cb]	508	// KERNEL ONLY
	509	static inline thread$ * __next_thread_search(cluster * this) with( *this ) {
	510	/* paranoid */ verify( ! __preemption_enabled() );
	511
	512	ready_schedule_lock();
	513	thread$ * thrd = pop_search( this );
[e873838]	514	ready_schedule_unlock();
[64a7146]	515
[8fc652e0]	516	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	517	return thrd;
[64a7146]	518	}
	519
[e84ab3d]	520	static inline bool __must_unpark( thread$ * thrd ) {
[ff79d5e]	521	int old_ticket = __atomic_fetch_add(&thrd->ticket, 1, __ATOMIC_SEQ_CST);
	522	switch(old_ticket) {
[6a77224]	523	case TICKET_RUNNING:
[3381ed7]	524	// Wake won the race, the thread will reschedule/rerun itself
[c6c7e6c]	525	return false;
[6a77224]	526	case TICKET_BLOCKED:
[3381ed7]	527	/* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
[ff79d5e]	528	/* paranoid */ verify( thrd->state == Blocked );
[c6c7e6c]	529	return true;
[3381ed7]	530	default:
	531	// This makes no sense, something is wrong abort
[7ee8153]	532	abort("Thread %p (%s) has mismatch park/unpark\n", thrd, thrd->self_cor.name);
[de6319f]	533	}
[eb2e723]	534	}
	535
[24e321c]	536	void __kernel_unpark( thread$ * thrd, unpark_hint hint ) {
[e9c0b4c]	537	/* paranoid */ verify( ! __preemption_enabled() );
	538	/* paranoid */ verify( ready_schedule_islocked());
	539
	540	if( !thrd ) return;
	541
	542	if(__must_unpark(thrd)) {
	543	// Wake lost the race,
[24e321c]	544	__schedule_thread( thrd, hint );
[e9c0b4c]	545	}
	546
	547	/* paranoid */ verify( ready_schedule_islocked());
	548	/* paranoid */ verify( ! __preemption_enabled() );
	549	}
	550
[24e321c]	551	void unpark( thread$ * thrd, unpark_hint hint ) {
[c6c7e6c]	552	if( !thrd ) return;
[0b33412]	553
[c6c7e6c]	554	if(__must_unpark(thrd)) {
	555	disable_interrupts();
[a3821fa]	556	// Wake lost the race,
[24e321c]	557	schedule_thread$( thrd, hint );
[a3821fa]	558	enable_interrupts(false);
[de6319f]	559	}
[eb2e723]	560	}
[0b33412]	561
[e235429]	562	void park( void ) {
[a3821fa]	563	__disable_interrupts_checked();
	564	/* paranoid */ verify( kernelTLS().this_thread->preempted == __NO_PREEMPTION );
	565	returnToKernel();
	566	__enable_interrupts_checked();
[0c78741]	567
	568	}
[09800e9]	569
[5afb49a]	570	extern "C" {
	571	// Leave the thread monitor
	572	// last routine called by a thread.
	573	// Should never return
	574	void __cfactx_thrd_leave() {
[e84ab3d]	575	thread$ * thrd = active_thread();
	576	monitor$ * this = &thrd->self_mon;
[5afb49a]	577
	578	// Lock the monitor now
	579	lock( this->lock __cfaabi_dbg_ctx2 );
	580
	581	disable_interrupts();
	582
[9d6e1b8a]	583	/* paranoid */ verify( ! __preemption_enabled() );
	584	/* paranoid */ verify( thrd->state == Active );
	585	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
	586	/* paranoid */ verify( kernelTLS().this_thread == thrd );
	587	/* paranoid */ verify( thrd->context.SP );
[e84ab3d]	588	/* 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 );
	589	/* 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 );
[9d6e1b8a]	590
[3ea8ad1]	591	thrd->state = Halting;
[58688bf]	592	if( TICKET_RUNNING != thrd->ticket ) { abort( "Thread terminated with pending unpark" ); }
[9d6e1b8a]	593	if( thrd != this->owner ) { abort( "Thread internal monitor has incorrect owner" ); }
	594	if( this->recursion != 1) { abort( "Thread internal monitor has unbalanced recursion" ); }
[5afb49a]	595
	596	// Leave the thread
	597	returnToKernel();
	598
	599	// Control flow should never reach here!
[9d6e1b8a]	600	abort();
[5afb49a]	601	}
[09800e9]	602	}
	603
[14a61b5]	604	// KERNEL ONLY
[3381ed7]	605	bool force_yield( __Preemption_Reason reason ) {
[a3821fa]	606	__disable_interrupts_checked();
[e84ab3d]	607	thread$ * thrd = kernelTLS().this_thread;
[a3821fa]	608	/* paranoid */ verify(thrd->state == Active);
	609
	610	// SKULLDUGGERY: It is possible that we are preempting this thread just before
	611	// it was going to park itself. If that is the case and it is already using the
	612	// intrusive fields then we can't use them to preempt the thread
	613	// If that is the case, abandon the preemption.
	614	bool preempted = false;
	615	if(thrd->link.next == 0p) {
	616	preempted = true;
	617	thrd->preempted = reason;
	618	returnToKernel();
	619	}
	620	__enable_interrupts_checked( false );
[3381ed7]	621	return preempted;
[f2b12406]	622	}
	623
[14a61b5]	624	//=============================================================================================
[92e7631]	625	// Kernel Idle Sleep
[14a61b5]	626	//=============================================================================================
[64a7146]	627	// Wake a thread from the front if there are any
[e873838]	628	static void __wake_one(cluster * this) {
[7cf3b1d]	629	eventfd_t val;
	630
[8fc652e0]	631	/* paranoid */ verify( ! __preemption_enabled() );
[e873838]	632	/* paranoid */ verify( ready_schedule_islocked() );
[14a61b5]	633
[64a7146]	634	// Check if there is a sleeping processor
[7cf3b1d]	635	struct __fd_waitctx * fdp = __atomic_load_n(&this->procs.fdw, __ATOMIC_SEQ_CST);
[14a61b5]	636
[7cf3b1d]	637	// If no one is sleeping: we are done
	638	if( fdp == 0p ) return;
[14a61b5]	639
[7cf3b1d]	640	int fd = 1;
	641	if( __atomic_load_n(&fdp->fd, __ATOMIC_SEQ_CST) != 1 ) {
	642	fd = __atomic_exchange_n(&fdp->fd, 1, __ATOMIC_RELAXED);
	643	}
[14a61b5]	644
[7cf3b1d]	645	switch(fd) {
	646	case 0:
	647	// If the processor isn't ready to sleep then the exchange will already wake it up
	648	#if !defined(__CFA_NO_STATISTICS__)
	649	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.early++;
	650	} else { __atomic_fetch_add(&this->stats->ready.sleep.early, 1, __ATOMIC_RELAXED); }
	651	#endif
	652	break;
	653	case 1:
	654	// If someone else already said they will wake them: we are done
	655	#if !defined(__CFA_NO_STATISTICS__)
	656	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.seen++;
	657	} else { __atomic_fetch_add(&this->stats->ready.sleep.seen, 1, __ATOMIC_RELAXED); }
	658	#endif
	659	break;
	660	default:
	661	// If the processor was ready to sleep, we need to wake it up with an actual write
	662	val = 1;
	663	eventfd_write( fd, val );
	664
	665	#if !defined(__CFA_NO_STATISTICS__)
	666	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.wakes++;
	667	} else { __atomic_fetch_add(&this->stats->ready.sleep.wakes, 1, __ATOMIC_RELAXED); }
	668	#endif
	669	break;
	670	}
[1eb239e4]	671
[e873838]	672	/* paranoid */ verify( ready_schedule_islocked() );
[8fc652e0]	673	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	674
	675	return;
[64a7146]	676	}
[14a61b5]	677
[64a7146]	678	// Unconditionnaly wake a thread
[1eb239e4]	679	void __wake_proc(processor * this) {
[7d0ebd0]	680	/* paranoid */ verify( ! __preemption_enabled() );
	681
[64a7146]	682	__cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
[14a61b5]	683
[7cf3b1d]	684	this->idle_wctx.fd = 1;
	685
[7d0ebd0]	686	eventfd_t val;
	687	val = 1;
	688	eventfd_write( this->idle_fd, val );
	689
	690	/* paranoid */ verify( ! __preemption_enabled() );
[92e7631]	691	}
	692
[d3605f8]	693	static void idle_sleep(processor * this, io_future_t & future, iovec & iov) {
[7cf3b1d]	694	// Tell everyone we are ready to go do sleep
	695	for() {
	696	int expected = this->idle_wctx.fd;
	697
	698	// Someone already told us to wake-up! No time for a nap.
	699	if(expected == 1) { return; }
	700
	701	// Try to mark that we are going to sleep
	702	if(__atomic_compare_exchange_n(&this->idle_wctx.fd, &expected, this->idle_fd, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST) ) {
	703	// Every one agreed, taking a nap
	704	break;
	705	}
	706	}
	707
	708
[d3605f8]	709	#if !defined(CFA_WITH_IO_URING_IDLE)
[7ef162b2]	710	#if !defined(__CFA_NO_STATISTICS__)
	711	if(this->print_halts) {
	712	__cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 0\n", this->unique_id, rdtscl());
	713	}
	714	#endif
[1757f98]	715
[7ef162b2]	716	__cfadbg_print_safe(runtime_core, "Kernel : core %p waiting on eventfd %d\n", this, this->idle_fd);
[1757f98]	717
[7ef162b2]	718	{
	719	eventfd_t val;
	720	ssize_t ret = read( this->idle_fd, &val, sizeof(val) );
	721	if(ret < 0) {
	722	switch((int)errno) {
	723	case EAGAIN:
	724	#if EAGAIN != EWOULDBLOCK
	725	case EWOULDBLOCK:
	726	#endif
	727	case EINTR:
	728	// No need to do anything special here, just assume it's a legitimate wake-up
	729	break;
	730	default:
	731	abort( "KERNEL : internal error, read failure on idle eventfd, error(%d) %s.", (int)errno, strerror( (int)errno ) );
	732	}
[1757f98]	733	}
	734	}
	735
[7ef162b2]	736	#if !defined(__CFA_NO_STATISTICS__)
	737	if(this->print_halts) {
	738	__cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 1\n", this->unique_id, rdtscl());
	739	}
	740	#endif
	741	#else
	742	// Do we already have a pending read
	743	if(available(future)) {
	744	// There is no pending read, we need to add one
	745	reset(future);
	746
[d3605f8]	747	__kernel_read(this, future, iov, this->idle_fd );
[1757f98]	748	}
[7ef162b2]	749
[c7b2215]	750	__cfa_io_flush( this, 1 );
[1757f98]	751	#endif
	752	}
	753
[5f5a729]	754	static bool mark_idle(__cluster_proc_list & this, processor & proc) {
[c9c1c1cb]	755	__STATS__(true, ready.sleep.halts++; )
[7cf3b1d]	756
	757	proc.idle_wctx.fd = 0;
	758
[8fc652e0]	759	/* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]	760	if(!try_lock( this )) return false;
[1eb239e4]	761	this.idle++;
	762	/* paranoid */ verify( this.idle <= this.total );
[fc59b580]	763	remove(proc);
[6a9b12b]	764	insert_first(this.idles, proc);
[34b8cb7]	765
[7cf3b1d]	766	__atomic_store_n(&this.fdw, &proc.idle_wctx, __ATOMIC_SEQ_CST);
[1eb239e4]	767	unlock( this );
[8fc652e0]	768	/* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]	769
	770	return true;
[1eb239e4]	771	}
[8e16177]	772
[6a9b12b]	773	static void mark_awake(__cluster_proc_list & this, processor & proc) {
[8fc652e0]	774	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	775	lock( this );
	776	this.idle--;
	777	/* paranoid */ verify( this.idle >= 0 );
	778	remove(proc);
[fc59b580]	779	insert_last(this.actives, proc);
[6a9b12b]	780
[a633f6f]	781	{
[7cf3b1d]	782	struct __fd_waitctx * wctx = 0;
	783	if(!this.idles`isEmpty) wctx = &this.idles`first.idle_wctx;
	784	__atomic_store_n(&this.fdw, wctx, __ATOMIC_SEQ_CST);
[a633f6f]	785	}
	786
[34b8cb7]	787	unlock( this );
[6a9b12b]	788	/* paranoid */ verify( ! __preemption_enabled() );
[6b4cdd3]	789	}
	790
[dbe9b08]	791	//=============================================================================================
	792	// Unexpected Terminating logic
	793	//=============================================================================================
[92bfda0]	794	void __kernel_abort_msg( char * abort_text, int abort_text_size ) {
[e84ab3d]	795	thread$ * thrd = __cfaabi_tls.this_thread;
[9d944b2]	796
[de94a60]	797	if(thrd) {
	798	int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
[1c40091]	799	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]	800
[212c2187]	801	if ( &thrd->self_cor != thrd->curr_cor ) {
	802	len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
[1c40091]	803	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]	804	}
	805	else {
[1c40091]	806	__cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
[de94a60]	807	}
[1c273d0]	808	}
[9d944b2]	809	else {
[de94a60]	810	int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
[1c40091]	811	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[9d944b2]	812	}
	813	}
	814
[92bfda0]	815	int __kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
	816	return get_coroutine(__cfaabi_tls.this_thread) == get_coroutine(mainThread) ? 4 : 2;
[2b8bc41]	817	}
	818
[de94a60]	819	static __spinlock_t kernel_debug_lock;
	820
[9d944b2]	821	extern "C" {
[1c40091]	822	void __cfaabi_bits_acquire() {
[36982fc]	823	lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
[9d944b2]	824	}
	825
[1c40091]	826	void __cfaabi_bits_release() {
[ea7d2b0]	827	unlock( kernel_debug_lock );
[9d944b2]	828	}
[8118303]	829	}
	830
[fa21ac9]	831	//=============================================================================================
	832	// Kernel Utilities
	833	//=============================================================================================
[dddb3dd0]	834	#if defined(CFA_HAVE_LINUX_IO_URING_H)
	835	#include "io/types.hfa"
	836	#endif
	837
[c1c95b1]	838	static inline bool __maybe_io_drain( processor * proc ) {
[e9c0b4c]	839	bool ret = false;
[dddb3dd0]	840	#if defined(CFA_HAVE_LINUX_IO_URING_H)
	841	__cfadbg_print_safe(runtime_core, "Kernel : core %p checking io for ring %d\n", proc, proc->io.ctx->fd);
	842
	843	// Check if we should drain the queue
	844	$io_context * ctx = proc->io.ctx;
	845	unsigned head = *ctx->cq.head;
	846	unsigned tail = *ctx->cq.tail;
[c1c95b1]	847	if(head == tail) return false;
[040334e]	848	ready_schedule_lock();
	849	ret = __cfa_io_drain( proc );
	850	ready_schedule_unlock();
[dddb3dd0]	851	#endif
[e9c0b4c]	852	return ret;
[dddb3dd0]	853	}
	854
[de94a60]	855	//-----------------------------------------------------------------------------
	856	// Debug
	857	__cfaabi_dbg_debug_do(
[1997b4e]	858	extern "C" {
[ae66348]	859	void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
[1997b4e]	860	this.prev_name = prev_name;
[8fc652e0]	861	this.prev_thrd = kernelTLS().this_thread;
[1997b4e]	862	}
[9181f1d]	863	}
[f7d6bb0]	864	)
[2026bb6]	865
	866	//-----------------------------------------------------------------------------
	867	// Debug
[8c50aed]	868	bool threading_enabled(void) __attribute__((const)) {
[2026bb6]	869	return true;
	870	}
[c34ebf2]	871
	872	//-----------------------------------------------------------------------------
	873	// Statistics
	874	#if !defined(__CFA_NO_STATISTICS__)
	875	void print_halts( processor & this ) {
	876	this.print_halts = true;
	877	}
[58688bf]	878
[69914cbc]	879	static void crawl_list( cluster * cltr, dlist(processor) & list, unsigned count ) {
[8464edf]	880	/* paranoid */ verify( cltr->stats );
	881
	882	processor * it = &list`first;
	883	for(unsigned i = 0; i < count; i++) {
	884	/* paranoid */ verifyf( it, "Unexpected null iterator, at index %u of %u\n", i, count);
	885	/* paranoid */ verify( it->local_data->this_stats );
[c33c2af]	886	// __print_stats( it->local_data->this_stats, cltr->print_stats, "Processor", it->name, (void*)it );
[8464edf]	887	__tally_stats( cltr->stats, it->local_data->this_stats );
	888	it = &(*it)`next;
	889	}
	890	}
	891
	892	void crawl_cluster_stats( cluster & this ) {
	893	// Stop the world, otherwise stats could get really messed-up
	894	// this doesn't solve all problems but does solve many
	895	// so it's probably good enough
[c33c2af]	896	disable_interrupts();
[8464edf]	897	uint_fast32_t last_size = ready_mutate_lock();
	898
	899	crawl_list(&this, this.procs.actives, this.procs.total - this.procs.idle);
	900	crawl_list(&this, this.procs.idles , this.procs.idle );
	901
	902	// Unlock the RWlock
	903	ready_mutate_unlock( last_size );
[c33c2af]	904	enable_interrupts();
[8464edf]	905	}
	906
	907
[58688bf]	908	void print_stats_now( cluster & this, int flags ) {
[8464edf]	909	crawl_cluster_stats( this );
[1b033b8]	910	__print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
	911	}
[c34ebf2]	912	#endif
[8118303]	913	// Local Variables: //
	914	// mode: c //
	915	// tab-width: 4 //
	916	// End: //

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

Download in other formats:

Original Format