Context Navigation

kernel.cfa@ f5bace8

Visit:

ADT ast-experimental enum pthread-emulation qualifiedEnum

Last change on this file since f5bace8 was 202c80b, checked in by Thierry Delisle <tdelisle@…>, 4 years ago
Added verify and changed print_now to not ignore flags.
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
[708ae38]	37	#include "kernel/private.hfa"
[73abe95]	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
[4479890]	138	extern bool __cfa_io_drain( processor * proc ) __attribute__((nonnull (1)));
[c7b2215]	139	extern bool __cfa_io_flush( processor *, int min_comp );
[c1c95b1]	140	static inline bool __maybe_io_drain( processor * );
[dddb3dd0]	141
[d3605f8]	142	#if defined(CFA_WITH_IO_URING_IDLE)
	143	extern bool __kernel_read(processor * proc, io_future_t & future, iovec &, int fd);
[6ddef36]	144	#endif
[7ef162b2]	145
[dddb3dd0]	146	extern void __disable_interrupts_hard();
	147	extern void __enable_interrupts_hard();
[c84e80a]	148
[7ef162b2]	149
[75f3522]	150	//=============================================================================================
	151	// Kernel Scheduling logic
	152	//=============================================================================================
[8fcbb4c]	153	//Main of the processor contexts
[83a071f9]	154	void main(processorCtx_t & runner) {
[21184e3]	155	// Because of a bug, we couldn't initialized the seed on construction
	156	// Do it here
[8fc652e0]	157	__cfaabi_tls.rand_seed ^= rdtscl();
	158	__cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner);
[f2384c9a]	159	__tls_rand_advance_bck();
[21184e3]	160
[83a071f9]	161	processor * this = runner.proc;
[094476d]	162	verify(this);
[c81ebf9]	163
[22226e4]	164	/* paranoid */ verify( this->idle_wctx.ftr != 0p );
	165	/* paranoid */ verify( this->idle_wctx.rdbuf != 0p );
[7ef162b2]	166
[22226e4]	167	// used for idle sleep when io_uring is present
	168	// mark it as already fulfilled so we know if there is a pending request or not
	169	this->idle_wctx.ftr->self.ptr = 1p;
	170	iovec idle_iovec = { this->idle_wctx.rdbuf, sizeof(eventfd_t) };
[dddb3dd0]	171
[4069faad]	172	__cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
[28d73c1]	173	#if !defined(__CFA_NO_STATISTICS__)
	174	if( this->print_halts ) {
[c993b15]	175	__cfaabi_bits_print_safe( STDOUT_FILENO, "Processor : %d - %s (%p)\n", this->unique_id, this->name, (void*)this);
[28d73c1]	176	}
	177	#endif
[b798713]	178
[75f3522]	179	{
[c81ebf9]	180	// Setup preemption data
	181	preemption_scope scope = { this };
	182
[a5e7233]	183	// if we need to run some special setup, now is the time to do it.
	184	if(this->init.thrd) {
	185	this->init.thrd->curr_cluster = this->cltr;
	186	__run_thread(this, this->init.thrd);
	187	}
[325e6ea]	188
[4069faad]	189	__cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
[8118303]	190
[e84ab3d]	191	thread$ * readyThread = 0p;
[1eb239e4]	192	MAIN_LOOP:
	193	for() {
[dddb3dd0]	194	// Check if there is pending io
	195	__maybe_io_drain( this );
	196
[92e7631]	197	// Try to get the next thread
[8c50aed]	198	readyThread = __next_thread( this->cltr );
[75f3522]	199
[1eb239e4]	200	if( !readyThread ) {
[c9c1c1cb]	201	__IO_STATS__(true, io.flush.idle++; )
[c7b2215]	202	__cfa_io_flush( this, 0 );
[c33c2af]	203
[c9c1c1cb]	204	readyThread = __next_thread( this->cltr );
	205	}
	206
	207	if( !readyThread ) for(5) {
	208	__IO_STATS__(true, io.flush.idle++; )
	209
[1eb239e4]	210	readyThread = __next_thread_slow( this->cltr );
[c9c1c1cb]	211
	212	if( readyThread ) break;
	213
	214	__cfa_io_flush( this, 0 );
[1eb239e4]	215	}
[4e6fb8e]	216
[1eb239e4]	217	HALT:
	218	if( !readyThread ) {
	219	// Don't block if we are done
	220	if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[c81ebf9]	221
[1eb239e4]	222	// Push self to idle stack
[5f5a729]	223	if(!mark_idle(this->cltr->procs, * this)) continue MAIN_LOOP;
[398e8e9]	224
[1eb239e4]	225	// Confirm the ready-queue is empty
[c9c1c1cb]	226	readyThread = __next_thread_search( this->cltr );
[1eb239e4]	227	if( readyThread ) {
	228	// A thread was found, cancel the halt
[6a9b12b]	229	mark_awake(this->cltr->procs, * this);
[1eb239e4]	230
[c9c1c1cb]	231	__STATS__(true, ready.sleep.cancels++; )
[1eb239e4]	232
	233	// continue the mai loop
	234	break HALT;
	235	}
	236
[22226e4]	237	idle_sleep( this, *this->idle_wctx.ftr, idle_iovec );
[1eb239e4]	238
	239	// We were woken up, remove self from idle
[6a9b12b]	240	mark_awake(this->cltr->procs, * this);
[1eb239e4]	241
	242	// DON'T just proceed, start looking again
	243	continue MAIN_LOOP;
[64a7146]	244	}
[1eb239e4]	245
	246	/* paranoid */ verify( readyThread );
	247
[dddb3dd0]	248	// Reset io dirty bit
	249	this->io.dirty = false;
	250
[1eb239e4]	251	// We found a thread run it
	252	__run_thread(this, readyThread);
	253
	254	// Are we done?
	255	if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[325e6ea]	256
[d529ad0]	257	if(__atomic_load_n(&this->io.pending, __ATOMIC_RELAXED) && !__atomic_load_n(&this->io.dirty, __ATOMIC_RELAXED)) {
[c9c1c1cb]	258	__IO_STATS__(true, io.flush.dirty++; )
[c7b2215]	259	__cfa_io_flush( this, 0 );
[dddb3dd0]	260	}
[c81ebf9]	261	}
	262
[4069faad]	263	__cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
[c84e80a]	264	}
[8118303]	265
[454f478]	266	post( this->terminated );
[bdeba0b]	267
[28d73c1]	268	if(this == mainProcessor) {
[6a490b2]	269	// HACK : the coroutine context switch expects this_thread to be set
	270	// and it make sense for it to be set in all other cases except here
	271	// fake it
[8fc652e0]	272	__cfaabi_tls.this_thread = mainThread;
[6a490b2]	273	}
[7768b8d]	274
[4069faad]	275	__cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
[c84e80a]	276	}
	277
[5c1a531]	278	static int * __volatile_errno() __attribute__((noinline));
	279	static int * __volatile_errno() { asm(""); return &errno; }
	280
[14a61b5]	281	// KERNEL ONLY
[1c273d0]	282	// runThread runs a thread by context switching
	283	// from the processor coroutine to the target thread
[e84ab3d]	284	static void __run_thread(processor * this, thread$ * thrd_dst) {
[8fc652e0]	285	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	286	/* paranoid */ verifyf( thrd_dst->state == Ready \|\| thrd_dst->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", thrd_dst->state, thrd_dst->preempted);
	287	/* paranoid */ verifyf( thrd_dst->link.next == 0p, "Expected null got %p", thrd_dst->link.next );
	288	__builtin_prefetch( thrd_dst->context.SP );
	289
[dddb3dd0]	290	__cfadbg_print_safe(runtime_core, "Kernel : core %p running thread %p (%s)\n", this, thrd_dst, thrd_dst->self_cor.name);
	291
[e84ab3d]	292	coroutine$ * proc_cor = get_coroutine(this->runner);
[8fcbb4c]	293
[9f575ea]	294	// set state of processor coroutine to inactive
	295	verify(proc_cor->state == Active);
[ae7be7a]	296	proc_cor->state = Blocked;
[e8e457e]	297
[9f575ea]	298	// Actually run the thread
[3381ed7]	299	RUNNING: while(true) {
[ff79d5e]	300	thrd_dst->preempted = __NO_PREEMPTION;
	301	thrd_dst->state = Active;
[e8e457e]	302
[58d64a4]	303	// Update global state
[8fc652e0]	304	kernelTLS().this_thread = thrd_dst;
[75f3522]	305
[8fc652e0]	306	/* paranoid */ verify( ! __preemption_enabled() );
	307	/* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
[9d6e1b8a]	308	/* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[b4b63e8]	309	/* paranoid */ verify( thrd_dst->context.SP );
[5afb49a]	310	/* paranoid */ verify( thrd_dst->state != Halted );
[e84ab3d]	311	/* 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
	312	/* 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]	313	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[b4b63e8]	314
[3381ed7]	315
[58d64a4]	316
[9f575ea]	317	// set context switch to the thread that the processor is executing
[c7a900a]	318	__cfactx_switch( &proc_cor->context, &thrd_dst->context );
	319	// when __cfactx_switch returns we are back in the processor coroutine
[9f575ea]	320
[50871b4]	321
	322
[ac12f1f]	323	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[e84ab3d]	324	/* 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 );
	325	/* 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]	326	/* paranoid */ verify( thrd_dst->context.SP );
[9d6e1b8a]	327	/* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[8fc652e0]	328	/* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
	329	/* paranoid */ verify( ! __preemption_enabled() );
[75f3522]	330
[58d64a4]	331	// Reset global state
[8fc652e0]	332	kernelTLS().this_thread = 0p;
[3381ed7]	333
	334	// We just finished running a thread, there are a few things that could have happened.
	335	// 1 - Regular case : the thread has blocked and now one has scheduled it yet.
	336	// 2 - Racy case : the thread has blocked but someone has already tried to schedule it.
	337	// 4 - Preempted
	338	// In case 1, we may have won a race so we can't write to the state again.
	339	// In case 2, we lost the race so we now own the thread.
	340
	341	if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
	342	// The thread was preempted, reschedule it and reset the flag
[24e321c]	343	schedule_thread$( thrd_dst, UNPARK_LOCAL );
[3381ed7]	344	break RUNNING;
	345	}
[75f3522]	346
[3ea8ad1]	347	if(unlikely(thrd_dst->state == Halting)) {
[ff79d5e]	348	// The thread has halted, it should never be scheduled/run again
[5afb49a]	349	// finish the thread
	350	__thread_finish( thrd_dst );
[ff79d5e]	351	break RUNNING;
	352	}
	353
	354	/* paranoid */ verify( thrd_dst->state == Active );
	355	thrd_dst->state = Blocked;
	356
[3381ed7]	357	// set state of processor coroutine to active and the thread to inactive
[ff79d5e]	358	int old_ticket = __atomic_fetch_sub(&thrd_dst->ticket, 1, __ATOMIC_SEQ_CST);
	359	switch(old_ticket) {
[6a77224]	360	case TICKET_RUNNING:
[3381ed7]	361	// This is case 1, the regular case, nothing more is needed
	362	break RUNNING;
[6a77224]	363	case TICKET_UNBLOCK:
[c9c1c1cb]	364	__STATS__(true, ready.threads.threads++; )
[3381ed7]	365	// This is case 2, the racy case, someone tried to run this thread before it finished blocking
	366	// In this case, just run it again.
	367	continue RUNNING;
	368	default:
	369	// This makes no sense, something is wrong abort
[ff79d5e]	370	abort();
[3381ed7]	371	}
[9f575ea]	372	}
[e8e457e]	373
[9f575ea]	374	// Just before returning to the processor, set the processor coroutine to active
[e8e457e]	375	proc_cor->state = Active;
[1eb239e4]	376
[dddb3dd0]	377	__cfadbg_print_safe(runtime_core, "Kernel : core %p finished running thread %p\n", this, thrd_dst);
	378
[c9c1c1cb]	379	__STATS__(true, ready.threads.threads--; )
[ec43cf9]	380
[8fc652e0]	381	/* paranoid */ verify( ! __preemption_enabled() );
[82c948c]	382	}
	383
[14a61b5]	384	// KERNEL_ONLY
[b0c7419]	385	void returnToKernel() {
[8fc652e0]	386	/* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]	387	coroutine$ * proc_cor = get_coroutine(kernelTLS().this_processor->runner);
	388	thread$ * thrd_src = kernelTLS().this_thread;
[e8e457e]	389
[c9c1c1cb]	390	__STATS_DEF( thrd_src->last_proc = kernelTLS().this_processor; )
[29cb302]	391
[9f575ea]	392	// Run the thread on this processor
	393	{
	394	int local_errno = *__volatile_errno();
	395	#if defined( __i386 ) \|\| defined( __x86_64 )
	396	__x87_store;
	397	#endif
[b4b63e8]	398	/* paranoid */ verify( proc_cor->context.SP );
[ac12f1f]	399	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[c7a900a]	400	__cfactx_switch( &thrd_src->context, &proc_cor->context );
[ac12f1f]	401	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[9f575ea]	402	#if defined( __i386 ) \|\| defined( __x86_64 )
	403	__x87_load;
	404	#endif
	405	*__volatile_errno() = local_errno;
[8fcbb4c]	406	}
[deca0f5]	407
[29cb302]	408	#if !defined(__CFA_NO_STATISTICS__)
[89eff25]	409	/* paranoid */ verify( thrd_src->last_proc != 0p );
	410	if(thrd_src->last_proc != kernelTLS().this_processor) {
[29cb302]	411	__tls_stats()->ready.threads.migration++;
	412	}
	413	#endif
	414
[8fc652e0]	415	/* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]	416	/* 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 );
	417	/* 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]	418	}
	419
[8def349]	420	//-----------------------------------------------------------------------------
	421	// Scheduler routines
[14a61b5]	422	// KERNEL ONLY
[24e321c]	423	static void __schedule_thread( thread$ * thrd, unpark_hint hint ) {
[8fc652e0]	424	/* paranoid */ verify( ! __preemption_enabled() );
[254ad1b]	425	/* paranoid */ verify( ready_schedule_islocked());
[6a490b2]	426	/* paranoid */ verify( thrd );
	427	/* paranoid */ verify( thrd->state != Halted );
[9d6e1b8a]	428	/* paranoid */ verify( thrd->curr_cluster );
[3381ed7]	429	/* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
[504a7dc]	430	/* paranoid */ if( thrd->state == Blocked \|\| thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
	431	"Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
[ff79d5e]	432	/* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active,
[504a7dc]	433	"Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
[3381ed7]	434	/* paranoid */ #endif
[6a490b2]	435	/* paranoid */ verifyf( thrd->link.next == 0p, "Expected null got %p", thrd->link.next );
[ac12f1f]	436	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
[b4b63e8]	437
[ae7be7a]	438	if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
[6b4cdd3]	439
[ec43cf9]	440	// Dereference the thread now because once we push it, there is not guaranteed it's still valid.
	441	struct cluster * cl = thrd->curr_cluster;
[c9c1c1cb]	442	__STATS_DEF(bool outside = hint == UNPARK_LOCAL && thrd->last_proc && thrd->last_proc != kernelTLS().this_processor; )
[32a8b61]	443
[254ad1b]	444	// push the thread to the cluster ready-queue
[24e321c]	445	push( cl, thrd, hint );
[32a8b61]	446
[254ad1b]	447	// variable thrd is no longer safe to use
[734908c]	448	thrd = 0xdeaddeaddeaddeadp;
[32a8b61]	449
[254ad1b]	450	// wake the cluster using the save variable.
	451	__wake_one( cl );
[1c273d0]	452
[ec43cf9]	453	#if !defined(__CFA_NO_STATISTICS__)
	454	if( kernelTLS().this_stats ) {
	455	__tls_stats()->ready.threads.threads++;
[89eff25]	456	if(outside) {
	457	__tls_stats()->ready.threads.extunpark++;
	458	}
[ec43cf9]	459	}
	460	else {
	461	__atomic_fetch_add(&cl->stats->ready.threads.threads, 1, __ATOMIC_RELAXED);
[89eff25]	462	__atomic_fetch_add(&cl->stats->ready.threads.extunpark, 1, __ATOMIC_RELAXED);
[ec43cf9]	463	}
	464	#endif
	465
[254ad1b]	466	/* paranoid */ verify( ready_schedule_islocked());
[8fc652e0]	467	/* paranoid */ verify( ! __preemption_enabled() );
[db6f06a]	468	}
	469
[24e321c]	470	void schedule_thread$( thread$ * thrd, unpark_hint hint ) {
[254ad1b]	471	ready_schedule_lock();
[24e321c]	472	__schedule_thread( thrd, hint );
[254ad1b]	473	ready_schedule_unlock();
	474	}
	475
[14a61b5]	476	// KERNEL ONLY
[e84ab3d]	477	static inline thread$ * __next_thread(cluster * this) with( *this ) {
[8fc652e0]	478	/* paranoid */ verify( ! __preemption_enabled() );
[7768b8d]	479
[e873838]	480	ready_schedule_lock();
[e84ab3d]	481	thread$ * thrd = pop_fast( this );
[e873838]	482	ready_schedule_unlock();
[7768b8d]	483
[8fc652e0]	484	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	485	return thrd;
[eb2e723]	486	}
	487
[64a7146]	488	// KERNEL ONLY
[e84ab3d]	489	static inline thread$ * __next_thread_slow(cluster * this) with( *this ) {
[8fc652e0]	490	/* paranoid */ verify( ! __preemption_enabled() );
[64a7146]	491
[e873838]	492	ready_schedule_lock();
[c9c1c1cb]	493	thread$ * thrd = pop_slow( this );
	494	ready_schedule_unlock();
	495
	496	/* paranoid */ verify( ! __preemption_enabled() );
	497	return thrd;
	498	}
[fc59df78]	499
[c9c1c1cb]	500	// KERNEL ONLY
	501	static inline thread$ * __next_thread_search(cluster * this) with( *this ) {
	502	/* paranoid */ verify( ! __preemption_enabled() );
	503
	504	ready_schedule_lock();
	505	thread$ * thrd = pop_search( this );
[e873838]	506	ready_schedule_unlock();
[64a7146]	507
[8fc652e0]	508	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	509	return thrd;
[64a7146]	510	}
	511
[e84ab3d]	512	static inline bool __must_unpark( thread$ * thrd ) {
[ff79d5e]	513	int old_ticket = __atomic_fetch_add(&thrd->ticket, 1, __ATOMIC_SEQ_CST);
	514	switch(old_ticket) {
[6a77224]	515	case TICKET_RUNNING:
[3381ed7]	516	// Wake won the race, the thread will reschedule/rerun itself
[c6c7e6c]	517	return false;
[6a77224]	518	case TICKET_BLOCKED:
[3381ed7]	519	/* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
[ff79d5e]	520	/* paranoid */ verify( thrd->state == Blocked );
[c6c7e6c]	521	return true;
[3381ed7]	522	default:
	523	// This makes no sense, something is wrong abort
[7ee8153]	524	abort("Thread %p (%s) has mismatch park/unpark\n", thrd, thrd->self_cor.name);
[de6319f]	525	}
[eb2e723]	526	}
	527
[24e321c]	528	void __kernel_unpark( thread$ * thrd, unpark_hint hint ) {
[e9c0b4c]	529	/* paranoid */ verify( ! __preemption_enabled() );
	530	/* paranoid */ verify( ready_schedule_islocked());
	531
	532	if( !thrd ) return;
	533
	534	if(__must_unpark(thrd)) {
	535	// Wake lost the race,
[24e321c]	536	__schedule_thread( thrd, hint );
[e9c0b4c]	537	}
	538
	539	/* paranoid */ verify( ready_schedule_islocked());
	540	/* paranoid */ verify( ! __preemption_enabled() );
	541	}
	542
[24e321c]	543	void unpark( thread$ * thrd, unpark_hint hint ) {
[c6c7e6c]	544	if( !thrd ) return;
[0b33412]	545
[c6c7e6c]	546	if(__must_unpark(thrd)) {
	547	disable_interrupts();
[a3821fa]	548	// Wake lost the race,
[24e321c]	549	schedule_thread$( thrd, hint );
[a3821fa]	550	enable_interrupts(false);
[de6319f]	551	}
[eb2e723]	552	}
[0b33412]	553
[e235429]	554	void park( void ) {
[a3821fa]	555	__disable_interrupts_checked();
	556	/* paranoid */ verify( kernelTLS().this_thread->preempted == __NO_PREEMPTION );
	557	returnToKernel();
	558	__enable_interrupts_checked();
[0c78741]	559
	560	}
[09800e9]	561
[5afb49a]	562	extern "C" {
	563	// Leave the thread monitor
	564	// last routine called by a thread.
	565	// Should never return
	566	void __cfactx_thrd_leave() {
[e84ab3d]	567	thread$ * thrd = active_thread();
	568	monitor$ * this = &thrd->self_mon;
[5afb49a]	569
	570	// Lock the monitor now
	571	lock( this->lock __cfaabi_dbg_ctx2 );
	572
	573	disable_interrupts();
	574
[9d6e1b8a]	575	/* paranoid */ verify( ! __preemption_enabled() );
	576	/* paranoid */ verify( thrd->state == Active );
	577	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
	578	/* paranoid */ verify( kernelTLS().this_thread == thrd );
	579	/* paranoid */ verify( thrd->context.SP );
[e84ab3d]	580	/* 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 );
	581	/* 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]	582
[3ea8ad1]	583	thrd->state = Halting;
[58688bf]	584	if( TICKET_RUNNING != thrd->ticket ) { abort( "Thread terminated with pending unpark" ); }
[9d6e1b8a]	585	if( thrd != this->owner ) { abort( "Thread internal monitor has incorrect owner" ); }
	586	if( this->recursion != 1) { abort( "Thread internal monitor has unbalanced recursion" ); }
[5afb49a]	587
	588	// Leave the thread
	589	returnToKernel();
	590
	591	// Control flow should never reach here!
[9d6e1b8a]	592	abort();
[5afb49a]	593	}
[09800e9]	594	}
	595
[14a61b5]	596	// KERNEL ONLY
[3381ed7]	597	bool force_yield( __Preemption_Reason reason ) {
[a3821fa]	598	__disable_interrupts_checked();
[e84ab3d]	599	thread$ * thrd = kernelTLS().this_thread;
[a3821fa]	600	/* paranoid */ verify(thrd->state == Active);
	601
	602	// SKULLDUGGERY: It is possible that we are preempting this thread just before
	603	// it was going to park itself. If that is the case and it is already using the
	604	// intrusive fields then we can't use them to preempt the thread
	605	// If that is the case, abandon the preemption.
	606	bool preempted = false;
	607	if(thrd->link.next == 0p) {
	608	preempted = true;
	609	thrd->preempted = reason;
	610	returnToKernel();
	611	}
	612	__enable_interrupts_checked( false );
[3381ed7]	613	return preempted;
[f2b12406]	614	}
	615
[14a61b5]	616	//=============================================================================================
[92e7631]	617	// Kernel Idle Sleep
[14a61b5]	618	//=============================================================================================
[64a7146]	619	// Wake a thread from the front if there are any
[e873838]	620	static void __wake_one(cluster * this) {
[7cf3b1d]	621	eventfd_t val;
	622
[8fc652e0]	623	/* paranoid */ verify( ! __preemption_enabled() );
[e873838]	624	/* paranoid */ verify( ready_schedule_islocked() );
[14a61b5]	625
[64a7146]	626	// Check if there is a sleeping processor
[7cf3b1d]	627	struct __fd_waitctx * fdp = __atomic_load_n(&this->procs.fdw, __ATOMIC_SEQ_CST);
[14a61b5]	628
[7cf3b1d]	629	// If no one is sleeping: we are done
	630	if( fdp == 0p ) return;
[14a61b5]	631
[7cf3b1d]	632	int fd = 1;
[22226e4]	633	if( __atomic_load_n(&fdp->sem, __ATOMIC_SEQ_CST) != 1 ) {
	634	fd = __atomic_exchange_n(&fdp->sem, 1, __ATOMIC_RELAXED);
[7cf3b1d]	635	}
[14a61b5]	636
[7cf3b1d]	637	switch(fd) {
	638	case 0:
	639	// If the processor isn't ready to sleep then the exchange will already wake it up
	640	#if !defined(__CFA_NO_STATISTICS__)
	641	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.early++;
	642	} else { __atomic_fetch_add(&this->stats->ready.sleep.early, 1, __ATOMIC_RELAXED); }
	643	#endif
	644	break;
	645	case 1:
	646	// If someone else already said they will wake them: we are done
	647	#if !defined(__CFA_NO_STATISTICS__)
	648	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.seen++;
	649	} else { __atomic_fetch_add(&this->stats->ready.sleep.seen, 1, __ATOMIC_RELAXED); }
	650	#endif
	651	break;
	652	default:
	653	// If the processor was ready to sleep, we need to wake it up with an actual write
	654	val = 1;
	655	eventfd_write( fd, val );
	656
	657	#if !defined(__CFA_NO_STATISTICS__)
	658	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.wakes++;
	659	} else { __atomic_fetch_add(&this->stats->ready.sleep.wakes, 1, __ATOMIC_RELAXED); }
	660	#endif
	661	break;
	662	}
[1eb239e4]	663
[e873838]	664	/* paranoid */ verify( ready_schedule_islocked() );
[8fc652e0]	665	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	666
	667	return;
[64a7146]	668	}
[14a61b5]	669
[64a7146]	670	// Unconditionnaly wake a thread
[1eb239e4]	671	void __wake_proc(processor * this) {
[7d0ebd0]	672	/* paranoid */ verify( ! __preemption_enabled() );
	673
[64a7146]	674	__cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
[14a61b5]	675
[22226e4]	676	this->idle_wctx.sem = 1;
[7cf3b1d]	677
[7d0ebd0]	678	eventfd_t val;
	679	val = 1;
[22226e4]	680	eventfd_write( this->idle_wctx.evfd, val );
[7d0ebd0]	681
	682	/* paranoid */ verify( ! __preemption_enabled() );
[92e7631]	683	}
	684
[d3605f8]	685	static void idle_sleep(processor * this, io_future_t & future, iovec & iov) {
[202c80b]	686	/* paranoid */ verify( this->idle_wctx.evfd != 1 );
	687	/* paranoid */ verify( this->idle_wctx.evfd != 2 );
	688
[7cf3b1d]	689	// Tell everyone we are ready to go do sleep
	690	for() {
[22226e4]	691	int expected = this->idle_wctx.sem;
[7cf3b1d]	692
	693	// Someone already told us to wake-up! No time for a nap.
	694	if(expected == 1) { return; }
	695
	696	// Try to mark that we are going to sleep
[22226e4]	697	if(__atomic_compare_exchange_n(&this->idle_wctx.sem, &expected, this->idle_wctx.evfd, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST) ) {
[7cf3b1d]	698	// Every one agreed, taking a nap
	699	break;
	700	}
	701	}
	702
	703
[d3605f8]	704	#if !defined(CFA_WITH_IO_URING_IDLE)
[7ef162b2]	705	#if !defined(__CFA_NO_STATISTICS__)
	706	if(this->print_halts) {
	707	__cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 0\n", this->unique_id, rdtscl());
	708	}
	709	#endif
[1757f98]	710
[7ef162b2]	711	__cfadbg_print_safe(runtime_core, "Kernel : core %p waiting on eventfd %d\n", this, this->idle_fd);
[1757f98]	712
[7ef162b2]	713	{
	714	eventfd_t val;
[22226e4]	715	ssize_t ret = read( this->idle_wctx.evfd, &val, sizeof(val) );
[7ef162b2]	716	if(ret < 0) {
	717	switch((int)errno) {
	718	case EAGAIN:
	719	#if EAGAIN != EWOULDBLOCK
	720	case EWOULDBLOCK:
	721	#endif
	722	case EINTR:
	723	// No need to do anything special here, just assume it's a legitimate wake-up
	724	break;
	725	default:
	726	abort( "KERNEL : internal error, read failure on idle eventfd, error(%d) %s.", (int)errno, strerror( (int)errno ) );
	727	}
[1757f98]	728	}
	729	}
	730
[7ef162b2]	731	#if !defined(__CFA_NO_STATISTICS__)
	732	if(this->print_halts) {
	733	__cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 1\n", this->unique_id, rdtscl());
	734	}
	735	#endif
	736	#else
	737	// Do we already have a pending read
	738	if(available(future)) {
	739	// There is no pending read, we need to add one
	740	reset(future);
	741
[22226e4]	742	__kernel_read(this, future, iov, this->idle_wctx.evfd );
[1757f98]	743	}
[7ef162b2]	744
[c7b2215]	745	__cfa_io_flush( this, 1 );
[1757f98]	746	#endif
	747	}
	748
[5f5a729]	749	static bool mark_idle(__cluster_proc_list & this, processor & proc) {
[c9c1c1cb]	750	__STATS__(true, ready.sleep.halts++; )
[7cf3b1d]	751
[22226e4]	752	proc.idle_wctx.sem = 0;
[7cf3b1d]	753
[8fc652e0]	754	/* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]	755	if(!try_lock( this )) return false;
[1eb239e4]	756	this.idle++;
	757	/* paranoid */ verify( this.idle <= this.total );
[fc59b580]	758	remove(proc);
[6a9b12b]	759	insert_first(this.idles, proc);
[34b8cb7]	760
[7cf3b1d]	761	__atomic_store_n(&this.fdw, &proc.idle_wctx, __ATOMIC_SEQ_CST);
[1eb239e4]	762	unlock( this );
[8fc652e0]	763	/* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]	764
	765	return true;
[1eb239e4]	766	}
[8e16177]	767
[6a9b12b]	768	static void mark_awake(__cluster_proc_list & this, processor & proc) {
[8fc652e0]	769	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	770	lock( this );
	771	this.idle--;
	772	/* paranoid */ verify( this.idle >= 0 );
	773	remove(proc);
[fc59b580]	774	insert_last(this.actives, proc);
[6a9b12b]	775
[a633f6f]	776	{
[7cf3b1d]	777	struct __fd_waitctx * wctx = 0;
	778	if(!this.idles`isEmpty) wctx = &this.idles`first.idle_wctx;
	779	__atomic_store_n(&this.fdw, wctx, __ATOMIC_SEQ_CST);
[a633f6f]	780	}
	781
[34b8cb7]	782	unlock( this );
[6a9b12b]	783	/* paranoid */ verify( ! __preemption_enabled() );
[6b4cdd3]	784	}
	785
[dbe9b08]	786	//=============================================================================================
	787	// Unexpected Terminating logic
	788	//=============================================================================================
[92bfda0]	789	void __kernel_abort_msg( char * abort_text, int abort_text_size ) {
[e84ab3d]	790	thread$ * thrd = __cfaabi_tls.this_thread;
[9d944b2]	791
[de94a60]	792	if(thrd) {
	793	int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
[1c40091]	794	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]	795
[212c2187]	796	if ( &thrd->self_cor != thrd->curr_cor ) {
	797	len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
[1c40091]	798	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]	799	}
	800	else {
[1c40091]	801	__cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
[de94a60]	802	}
[1c273d0]	803	}
[9d944b2]	804	else {
[de94a60]	805	int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
[1c40091]	806	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[9d944b2]	807	}
	808	}
	809
[92bfda0]	810	int __kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
	811	return get_coroutine(__cfaabi_tls.this_thread) == get_coroutine(mainThread) ? 4 : 2;
[2b8bc41]	812	}
	813
[de94a60]	814	static __spinlock_t kernel_debug_lock;
	815
[9d944b2]	816	extern "C" {
[1c40091]	817	void __cfaabi_bits_acquire() {
[36982fc]	818	lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
[9d944b2]	819	}
	820
[1c40091]	821	void __cfaabi_bits_release() {
[ea7d2b0]	822	unlock( kernel_debug_lock );
[9d944b2]	823	}
[8118303]	824	}
	825
[fa21ac9]	826	//=============================================================================================
	827	// Kernel Utilities
	828	//=============================================================================================
[dddb3dd0]	829	#if defined(CFA_HAVE_LINUX_IO_URING_H)
	830	#include "io/types.hfa"
	831	#endif
	832
[c1c95b1]	833	static inline bool __maybe_io_drain( processor * proc ) {
[4479890]	834	/* paranoid */ verify( proc );
[e9c0b4c]	835	bool ret = false;
[dddb3dd0]	836	#if defined(CFA_HAVE_LINUX_IO_URING_H)
	837	__cfadbg_print_safe(runtime_core, "Kernel : core %p checking io for ring %d\n", proc, proc->io.ctx->fd);
	838
	839	// Check if we should drain the queue
	840	$io_context * ctx = proc->io.ctx;
	841	unsigned head = *ctx->cq.head;
	842	unsigned tail = *ctx->cq.tail;
[c1c95b1]	843	if(head == tail) return false;
[040334e]	844	ready_schedule_lock();
[4479890]	845	ret = __cfa_io_drain( proc );
[040334e]	846	ready_schedule_unlock();
[dddb3dd0]	847	#endif
[e9c0b4c]	848	return ret;
[dddb3dd0]	849	}
	850
[de94a60]	851	//-----------------------------------------------------------------------------
	852	// Debug
	853	__cfaabi_dbg_debug_do(
[1997b4e]	854	extern "C" {
[ae66348]	855	void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
[1997b4e]	856	this.prev_name = prev_name;
[8fc652e0]	857	this.prev_thrd = kernelTLS().this_thread;
[1997b4e]	858	}
[9181f1d]	859	}
[f7d6bb0]	860	)
[2026bb6]	861
	862	//-----------------------------------------------------------------------------
	863	// Debug
[8c50aed]	864	bool threading_enabled(void) __attribute__((const)) {
[2026bb6]	865	return true;
	866	}
[c34ebf2]	867
	868	//-----------------------------------------------------------------------------
	869	// Statistics
	870	#if !defined(__CFA_NO_STATISTICS__)
	871	void print_halts( processor & this ) {
	872	this.print_halts = true;
	873	}
[58688bf]	874
[69914cbc]	875	static void crawl_list( cluster * cltr, dlist(processor) & list, unsigned count ) {
[8464edf]	876	/* paranoid */ verify( cltr->stats );
	877
	878	processor * it = &list`first;
	879	for(unsigned i = 0; i < count; i++) {
	880	/* paranoid */ verifyf( it, "Unexpected null iterator, at index %u of %u\n", i, count);
	881	/* paranoid */ verify( it->local_data->this_stats );
[c33c2af]	882	// __print_stats( it->local_data->this_stats, cltr->print_stats, "Processor", it->name, (void*)it );
[8464edf]	883	__tally_stats( cltr->stats, it->local_data->this_stats );
	884	it = &(*it)`next;
	885	}
	886	}
	887
	888	void crawl_cluster_stats( cluster & this ) {
	889	// Stop the world, otherwise stats could get really messed-up
	890	// this doesn't solve all problems but does solve many
	891	// so it's probably good enough
[c33c2af]	892	disable_interrupts();
[8464edf]	893	uint_fast32_t last_size = ready_mutate_lock();
	894
	895	crawl_list(&this, this.procs.actives, this.procs.total - this.procs.idle);
	896	crawl_list(&this, this.procs.idles , this.procs.idle );
	897
	898	// Unlock the RWlock
	899	ready_mutate_unlock( last_size );
[c33c2af]	900	enable_interrupts();
[8464edf]	901	}
	902
	903
[58688bf]	904	void print_stats_now( cluster & this, int flags ) {
[8464edf]	905	crawl_cluster_stats( this );
[202c80b]	906	__print_stats( this.stats, flags, "Cluster", this.name, (void*)&this );
[1b033b8]	907	}
[c34ebf2]	908	#endif
[8118303]	909	// Local Variables: //
	910	// mode: c //
	911	// tab-width: 4 //
	912	// End: //

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

Download in other formats:

Original Format