Context Navigation

kernel.cfa@ 040334e

Visit:

ADT ast-experimental enum forall-pointer-decay pthread-emulation qualifiedEnum stuck-waitfor-destruct

Last change on this file since 040334e was 040334e, checked in by Thierry Delisle <tdelisle@…>, 4 years ago
Removed so-called 'new proc main' which wasn't useful.
Property mode set to `100644`
File size: 28.4 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
[8118303]	21	//C Includes
[214e8da]	22	#include <errno.h>
[9d944b2]	23	#include <stdio.h>
[445f984]	24	#include <string.h>
[58b6d1b]	25	#include <signal.h>
[9d944b2]	26	#include <unistd.h>
[dddb3dd0]	27	extern "C" {
	28	#include <sys/eventfd.h>
[d3605f8]	29	#include <sys/uio.h>
[dddb3dd0]	30	}
[8118303]	31
	32	//CFA Includes
[73abe95]	33	#include "kernel_private.hfa"
	34	#include "preemption.hfa"
[445f984]	35	#include "strstream.hfa"
	36	#include "device/cpu.hfa"
[7ef162b2]	37	#include "io/types.hfa"
[8118303]	38
	39	//Private includes
	40	#define __CFA_INVOKE_PRIVATE__
	41	#include "invoke.h"
	42
[89eff25]	43	#if !defined(__CFA_NO_STATISTICS__)
	44	#define __STATS( ...) __VA_ARGS__
	45	#else
	46	#define __STATS( ...)
	47	#endif
[4069faad]	48
[deca0f5]	49	//-----------------------------------------------------------------------------
	50	// Some assembly required
[1805b1b]	51	#if defined( __i386 )
[deca0f5]	52	// mxcr : SSE Status and Control bits (control bits are preserved across function calls)
	53	// fcw : X87 FPU control word (preserved across function calls)
	54	#define __x87_store \
	55	uint32_t __mxcr; \
	56	uint16_t __fcw; \
	57	__asm__ volatile ( \
	58	"stmxcsr %0\n" \
	59	"fnstcw %1\n" \
	60	: "=m" (__mxcr),\
	61	"=m" (__fcw) \
	62	)
	63
	64	#define __x87_load \
	65	__asm__ volatile ( \
	66	"fldcw %1\n" \
	67	"ldmxcsr %0\n" \
	68	::"m" (__mxcr),\
	69	"m" (__fcw) \
	70	)
	71
	72	#elif defined( __x86_64 )
	73	#define __x87_store \
	74	uint32_t __mxcr; \
	75	uint16_t __fcw; \
	76	__asm__ volatile ( \
	77	"stmxcsr %0\n" \
	78	"fnstcw %1\n" \
	79	: "=m" (__mxcr),\
	80	"=m" (__fcw) \
	81	)
	82
	83	#define __x87_load \
	84	__asm__ volatile ( \
	85	"fldcw %1\n" \
	86	"ldmxcsr %0\n" \
	87	:: "m" (__mxcr),\
	88	"m" (__fcw) \
	89	)
	90
[0190480]	91	#elif defined( __arm__ )
	92	#define __x87_store
	93	#define __x87_load
	94
	95	#elif defined( __aarch64__ )
[74f5c83]	96	#define __x87_store \
	97	uint32_t __fpcntl[2]; \
	98	__asm__ volatile ( \
	99	"mrs x9, FPCR\n" \
	100	"mrs x10, FPSR\n" \
	101	"stp x9, x10, %0\n" \
	102	: "=m" (__fpcntl) : : "x9", "x10" \
	103	)
	104
	105	#define __x87_load \
	106	__asm__ volatile ( \
	107	"ldp x9, x10, %0\n" \
	108	"msr FPSR, x10\n" \
	109	"msr FPCR, x9\n" \
	110	: "=m" (__fpcntl) : : "x9", "x10" \
	111	)
	112
[deca0f5]	113	#else
[0190480]	114	#error unsupported hardware architecture
[deca0f5]	115	#endif
	116
[e84ab3d]	117	extern thread$ * mainThread;
[e660761]	118	extern processor * mainProcessor;
[2ac095d]	119
[92e7631]	120	//-----------------------------------------------------------------------------
	121	// Kernel Scheduling logic
[e84ab3d]	122	static thread$ * __next_thread(cluster * this);
	123	static thread$ * __next_thread_slow(cluster * this);
	124	static inline bool __must_unpark( thread$ * thrd ) __attribute((nonnull(1)));
	125	static void __run_thread(processor * this, thread$ * dst);
[e873838]	126	static void __wake_one(cluster * cltr);
[1eb239e4]	127
[d3605f8]	128	static void idle_sleep(processor * proc, io_future_t & future, iovec & iov);
[5f5a729]	129	static bool mark_idle (__cluster_proc_list & idles, processor & proc);
[6a9b12b]	130	static void mark_awake(__cluster_proc_list & idles, processor & proc);
[1eb239e4]	131
[dddb3dd0]	132	extern void __cfa_io_start( processor * );
[c1c95b1]	133	extern bool __cfa_io_drain( processor * );
[c7b2215]	134	extern bool __cfa_io_flush( processor *, int min_comp );
[dddb3dd0]	135	extern void __cfa_io_stop ( processor * );
[c1c95b1]	136	static inline bool __maybe_io_drain( processor * );
[dddb3dd0]	137
[d3605f8]	138	#if defined(CFA_WITH_IO_URING_IDLE)
	139	extern bool __kernel_read(processor * proc, io_future_t & future, iovec &, int fd);
[6ddef36]	140	#endif
[7ef162b2]	141
[dddb3dd0]	142	extern void __disable_interrupts_hard();
	143	extern void __enable_interrupts_hard();
[c84e80a]	144
[7ef162b2]	145
[75f3522]	146	//=============================================================================================
	147	// Kernel Scheduling logic
	148	//=============================================================================================
[8fcbb4c]	149	//Main of the processor contexts
[83a071f9]	150	void main(processorCtx_t & runner) {
[21184e3]	151	// Because of a bug, we couldn't initialized the seed on construction
	152	// Do it here
[8fc652e0]	153	__cfaabi_tls.rand_seed ^= rdtscl();
	154	__cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner);
[f2384c9a]	155	__tls_rand_advance_bck();
[21184e3]	156
[83a071f9]	157	processor * this = runner.proc;
[094476d]	158	verify(this);
[c81ebf9]	159
[7ef162b2]	160	io_future_t future; // used for idle sleep when io_uring is present
	161	future.self.ptr = 1p; // mark it as already fulfilled so we know if there is a pending request or not
[c7b2215]	162	eventfd_t idle_val;
[d3605f8]	163	iovec idle_iovec = { &idle_val, sizeof(idle_val) };
[7ef162b2]	164
[dddb3dd0]	165	__cfa_io_start( this );
	166
[4069faad]	167	__cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
[28d73c1]	168	#if !defined(__CFA_NO_STATISTICS__)
	169	if( this->print_halts ) {
[c993b15]	170	__cfaabi_bits_print_safe( STDOUT_FILENO, "Processor : %d - %s (%p)\n", this->unique_id, this->name, (void*)this);
[28d73c1]	171	}
	172	#endif
[b798713]	173
[75f3522]	174	{
[c81ebf9]	175	// Setup preemption data
	176	preemption_scope scope = { this };
	177
[a5e7233]	178	// if we need to run some special setup, now is the time to do it.
	179	if(this->init.thrd) {
	180	this->init.thrd->curr_cluster = this->cltr;
	181	__run_thread(this, this->init.thrd);
	182	}
[325e6ea]	183
[4069faad]	184	__cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
[8118303]	185
[e84ab3d]	186	thread$ * readyThread = 0p;
[1eb239e4]	187	MAIN_LOOP:
	188	for() {
[dddb3dd0]	189	// Check if there is pending io
	190	__maybe_io_drain( this );
	191
[92e7631]	192	// Try to get the next thread
[8c50aed]	193	readyThread = __next_thread( this->cltr );
[75f3522]	194
[1eb239e4]	195	if( !readyThread ) {
[70b4aeb9]	196	__tls_stats()->io.flush.idle++;
[c7b2215]	197	__cfa_io_flush( this, 0 );
[c33c2af]	198
[1eb239e4]	199	readyThread = __next_thread_slow( this->cltr );
	200	}
[4e6fb8e]	201
[1eb239e4]	202	HALT:
	203	if( !readyThread ) {
	204	// Don't block if we are done
	205	if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[c81ebf9]	206
[1eb239e4]	207	// Push self to idle stack
[5f5a729]	208	if(!mark_idle(this->cltr->procs, * this)) continue MAIN_LOOP;
[398e8e9]	209
[1eb239e4]	210	// Confirm the ready-queue is empty
	211	readyThread = __next_thread_slow( this->cltr );
	212	if( readyThread ) {
	213	// A thread was found, cancel the halt
[6a9b12b]	214	mark_awake(this->cltr->procs, * this);
[1eb239e4]	215
	216	#if !defined(__CFA_NO_STATISTICS__)
	217	__tls_stats()->ready.sleep.cancels++;
	218	#endif
	219
	220	// continue the mai loop
	221	break HALT;
	222	}
	223
[d3605f8]	224	idle_sleep( this, future, idle_iovec );
[1eb239e4]	225
	226	// We were woken up, remove self from idle
[6a9b12b]	227	mark_awake(this->cltr->procs, * this);
[1eb239e4]	228
	229	// DON'T just proceed, start looking again
	230	continue MAIN_LOOP;
[64a7146]	231	}
[1eb239e4]	232
	233	/* paranoid */ verify( readyThread );
	234
[dddb3dd0]	235	// Reset io dirty bit
	236	this->io.dirty = false;
	237
[1eb239e4]	238	// We found a thread run it
	239	__run_thread(this, readyThread);
	240
	241	// Are we done?
	242	if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[325e6ea]	243
[dddb3dd0]	244	if(this->io.pending && !this->io.dirty) {
[70b4aeb9]	245	__tls_stats()->io.flush.dirty++;
[c7b2215]	246	__cfa_io_flush( this, 0 );
[dddb3dd0]	247	}
[c81ebf9]	248	}
	249
[4069faad]	250	__cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
[c84e80a]	251	}
[8118303]	252
[3bb4f85]	253	for(int i = 0; !available(future); i++) {
	254	if(i > 1000) __cfaabi_dbg_write( "ERROR: kernel has bin spinning on a flush after exit loop.\n", 60);
	255	__cfa_io_flush( this, 1 );
	256	}
	257
[dddb3dd0]	258	__cfa_io_stop( this );
	259
[454f478]	260	post( this->terminated );
[bdeba0b]	261
[28d73c1]	262	if(this == mainProcessor) {
[6a490b2]	263	// HACK : the coroutine context switch expects this_thread to be set
	264	// and it make sense for it to be set in all other cases except here
	265	// fake it
[8fc652e0]	266	__cfaabi_tls.this_thread = mainThread;
[6a490b2]	267	}
[7768b8d]	268
[4069faad]	269	__cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
[c84e80a]	270	}
	271
[5c1a531]	272	static int * __volatile_errno() __attribute__((noinline));
	273	static int * __volatile_errno() { asm(""); return &errno; }
	274
[14a61b5]	275	// KERNEL ONLY
[1c273d0]	276	// runThread runs a thread by context switching
	277	// from the processor coroutine to the target thread
[e84ab3d]	278	static void __run_thread(processor * this, thread$ * thrd_dst) {
[8fc652e0]	279	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	280	/* paranoid */ verifyf( thrd_dst->state == Ready \|\| thrd_dst->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", thrd_dst->state, thrd_dst->preempted);
	281	/* paranoid */ verifyf( thrd_dst->link.next == 0p, "Expected null got %p", thrd_dst->link.next );
	282	__builtin_prefetch( thrd_dst->context.SP );
	283
[dddb3dd0]	284	__cfadbg_print_safe(runtime_core, "Kernel : core %p running thread %p (%s)\n", this, thrd_dst, thrd_dst->self_cor.name);
	285
[e84ab3d]	286	coroutine$ * proc_cor = get_coroutine(this->runner);
[8fcbb4c]	287
[9f575ea]	288	// set state of processor coroutine to inactive
	289	verify(proc_cor->state == Active);
[ae7be7a]	290	proc_cor->state = Blocked;
[e8e457e]	291
[9f575ea]	292	// Actually run the thread
[3381ed7]	293	RUNNING: while(true) {
[ff79d5e]	294	thrd_dst->preempted = __NO_PREEMPTION;
	295	thrd_dst->state = Active;
[e8e457e]	296
[58d64a4]	297	// Update global state
[8fc652e0]	298	kernelTLS().this_thread = thrd_dst;
[75f3522]	299
[8fc652e0]	300	/* paranoid */ verify( ! __preemption_enabled() );
	301	/* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
[9d6e1b8a]	302	/* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[b4b63e8]	303	/* paranoid */ verify( thrd_dst->context.SP );
[5afb49a]	304	/* paranoid */ verify( thrd_dst->state != Halted );
[e84ab3d]	305	/* 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
	306	/* 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]	307	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[b4b63e8]	308
[3381ed7]	309
[58d64a4]	310
[9f575ea]	311	// set context switch to the thread that the processor is executing
[c7a900a]	312	__cfactx_switch( &proc_cor->context, &thrd_dst->context );
	313	// when __cfactx_switch returns we are back in the processor coroutine
[9f575ea]	314
[50871b4]	315
	316
[ac12f1f]	317	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[e84ab3d]	318	/* 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 );
	319	/* 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]	320	/* paranoid */ verify( thrd_dst->context.SP );
[9d6e1b8a]	321	/* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[8fc652e0]	322	/* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
	323	/* paranoid */ verify( ! __preemption_enabled() );
[75f3522]	324
[58d64a4]	325	// Reset global state
[8fc652e0]	326	kernelTLS().this_thread = 0p;
[3381ed7]	327
	328	// We just finished running a thread, there are a few things that could have happened.
	329	// 1 - Regular case : the thread has blocked and now one has scheduled it yet.
	330	// 2 - Racy case : the thread has blocked but someone has already tried to schedule it.
	331	// 4 - Preempted
	332	// In case 1, we may have won a race so we can't write to the state again.
	333	// In case 2, we lost the race so we now own the thread.
	334
	335	if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
	336	// The thread was preempted, reschedule it and reset the flag
[24e321c]	337	schedule_thread$( thrd_dst, UNPARK_LOCAL );
[3381ed7]	338	break RUNNING;
	339	}
[75f3522]	340
[3ea8ad1]	341	if(unlikely(thrd_dst->state == Halting)) {
[ff79d5e]	342	// The thread has halted, it should never be scheduled/run again
[5afb49a]	343	// finish the thread
	344	__thread_finish( thrd_dst );
[ff79d5e]	345	break RUNNING;
	346	}
	347
	348	/* paranoid */ verify( thrd_dst->state == Active );
	349	thrd_dst->state = Blocked;
	350
[3381ed7]	351	// set state of processor coroutine to active and the thread to inactive
[ff79d5e]	352	int old_ticket = __atomic_fetch_sub(&thrd_dst->ticket, 1, __ATOMIC_SEQ_CST);
	353	switch(old_ticket) {
[6a77224]	354	case TICKET_RUNNING:
[3381ed7]	355	// This is case 1, the regular case, nothing more is needed
	356	break RUNNING;
[6a77224]	357	case TICKET_UNBLOCK:
[ec43cf9]	358	#if !defined(__CFA_NO_STATISTICS__)
	359	__tls_stats()->ready.threads.threads++;
	360	#endif
[3381ed7]	361	// This is case 2, the racy case, someone tried to run this thread before it finished blocking
	362	// In this case, just run it again.
	363	continue RUNNING;
	364	default:
	365	// This makes no sense, something is wrong abort
[ff79d5e]	366	abort();
[3381ed7]	367	}
[9f575ea]	368	}
[e8e457e]	369
[9f575ea]	370	// Just before returning to the processor, set the processor coroutine to active
[e8e457e]	371	proc_cor->state = Active;
[1eb239e4]	372
[dddb3dd0]	373	__cfadbg_print_safe(runtime_core, "Kernel : core %p finished running thread %p\n", this, thrd_dst);
	374
[ec43cf9]	375	#if !defined(__CFA_NO_STATISTICS__)
	376	__tls_stats()->ready.threads.threads--;
	377	#endif
	378
[8fc652e0]	379	/* paranoid */ verify( ! __preemption_enabled() );
[82c948c]	380	}
	381
[14a61b5]	382	// KERNEL_ONLY
[b0c7419]	383	void returnToKernel() {
[8fc652e0]	384	/* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]	385	coroutine$ * proc_cor = get_coroutine(kernelTLS().this_processor->runner);
	386	thread$ * thrd_src = kernelTLS().this_thread;
[e8e457e]	387
[89eff25]	388	__STATS( thrd_src->last_proc = kernelTLS().this_processor; )
[29cb302]	389
[9f575ea]	390	// Run the thread on this processor
	391	{
	392	int local_errno = *__volatile_errno();
	393	#if defined( __i386 ) \|\| defined( __x86_64 )
	394	__x87_store;
	395	#endif
[b4b63e8]	396	/* paranoid */ verify( proc_cor->context.SP );
[ac12f1f]	397	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[c7a900a]	398	__cfactx_switch( &thrd_src->context, &proc_cor->context );
[ac12f1f]	399	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[9f575ea]	400	#if defined( __i386 ) \|\| defined( __x86_64 )
	401	__x87_load;
	402	#endif
	403	*__volatile_errno() = local_errno;
[8fcbb4c]	404	}
[deca0f5]	405
[29cb302]	406	#if !defined(__CFA_NO_STATISTICS__)
[89eff25]	407	/* paranoid */ verify( thrd_src->last_proc != 0p );
	408	if(thrd_src->last_proc != kernelTLS().this_processor) {
[29cb302]	409	__tls_stats()->ready.threads.migration++;
	410	}
	411	#endif
	412
[8fc652e0]	413	/* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]	414	/* 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 );
	415	/* 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]	416	}
	417
[8def349]	418	//-----------------------------------------------------------------------------
	419	// Scheduler routines
[14a61b5]	420	// KERNEL ONLY
[24e321c]	421	static void __schedule_thread( thread$ * thrd, unpark_hint hint ) {
[8fc652e0]	422	/* paranoid */ verify( ! __preemption_enabled() );
[254ad1b]	423	/* paranoid */ verify( ready_schedule_islocked());
[6a490b2]	424	/* paranoid */ verify( thrd );
	425	/* paranoid */ verify( thrd->state != Halted );
[9d6e1b8a]	426	/* paranoid */ verify( thrd->curr_cluster );
[3381ed7]	427	/* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
[504a7dc]	428	/* paranoid */ if( thrd->state == Blocked \|\| thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
	429	"Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
[ff79d5e]	430	/* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active,
[504a7dc]	431	"Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
[3381ed7]	432	/* paranoid */ #endif
[6a490b2]	433	/* paranoid */ verifyf( thrd->link.next == 0p, "Expected null got %p", thrd->link.next );
[ac12f1f]	434	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
[b4b63e8]	435
[ae7be7a]	436	if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
[6b4cdd3]	437
[ec43cf9]	438	// Dereference the thread now because once we push it, there is not guaranteed it's still valid.
	439	struct cluster * cl = thrd->curr_cluster;
[24e321c]	440	__STATS(bool outside = hint == UNPARK_LOCAL && thrd->last_proc && thrd->last_proc != kernelTLS().this_processor; )
[32a8b61]	441
[254ad1b]	442	// push the thread to the cluster ready-queue
[24e321c]	443	push( cl, thrd, hint );
[32a8b61]	444
[254ad1b]	445	// variable thrd is no longer safe to use
[734908c]	446	thrd = 0xdeaddeaddeaddeadp;
[32a8b61]	447
[254ad1b]	448	// wake the cluster using the save variable.
	449	__wake_one( cl );
[1c273d0]	450
[ec43cf9]	451	#if !defined(__CFA_NO_STATISTICS__)
	452	if( kernelTLS().this_stats ) {
	453	__tls_stats()->ready.threads.threads++;
[89eff25]	454	if(outside) {
	455	__tls_stats()->ready.threads.extunpark++;
	456	}
[ec43cf9]	457	}
	458	else {
	459	__atomic_fetch_add(&cl->stats->ready.threads.threads, 1, __ATOMIC_RELAXED);
[89eff25]	460	__atomic_fetch_add(&cl->stats->ready.threads.extunpark, 1, __ATOMIC_RELAXED);
[ec43cf9]	461	}
	462	#endif
	463
[254ad1b]	464	/* paranoid */ verify( ready_schedule_islocked());
[8fc652e0]	465	/* paranoid */ verify( ! __preemption_enabled() );
[db6f06a]	466	}
	467
[24e321c]	468	void schedule_thread$( thread$ * thrd, unpark_hint hint ) {
[254ad1b]	469	ready_schedule_lock();
[24e321c]	470	__schedule_thread( thrd, hint );
[254ad1b]	471	ready_schedule_unlock();
	472	}
	473
[14a61b5]	474	// KERNEL ONLY
[e84ab3d]	475	static inline thread$ * __next_thread(cluster * this) with( *this ) {
[8fc652e0]	476	/* paranoid */ verify( ! __preemption_enabled() );
[7768b8d]	477
[e873838]	478	ready_schedule_lock();
[e84ab3d]	479	thread$ * thrd = pop_fast( this );
[e873838]	480	ready_schedule_unlock();
[7768b8d]	481
[8fc652e0]	482	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	483	return thrd;
[eb2e723]	484	}
	485
[64a7146]	486	// KERNEL ONLY
[e84ab3d]	487	static inline thread$ * __next_thread_slow(cluster * this) with( *this ) {
[8fc652e0]	488	/* paranoid */ verify( ! __preemption_enabled() );
[64a7146]	489
[e873838]	490	ready_schedule_lock();
[e84ab3d]	491	thread$ * thrd;
[fc59df78]	492	for(25) {
	493	thrd = pop_slow( this );
	494	if(thrd) goto RET;
	495	}
	496	thrd = pop_search( this );
	497
	498	RET:
[e873838]	499	ready_schedule_unlock();
[64a7146]	500
[8fc652e0]	501	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	502	return thrd;
[64a7146]	503	}
	504
[e84ab3d]	505	static inline bool __must_unpark( thread$ * thrd ) {
[ff79d5e]	506	int old_ticket = __atomic_fetch_add(&thrd->ticket, 1, __ATOMIC_SEQ_CST);
	507	switch(old_ticket) {
[6a77224]	508	case TICKET_RUNNING:
[3381ed7]	509	// Wake won the race, the thread will reschedule/rerun itself
[c6c7e6c]	510	return false;
[6a77224]	511	case TICKET_BLOCKED:
[3381ed7]	512	/* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
[ff79d5e]	513	/* paranoid */ verify( thrd->state == Blocked );
[c6c7e6c]	514	return true;
[3381ed7]	515	default:
	516	// This makes no sense, something is wrong abort
[7ee8153]	517	abort("Thread %p (%s) has mismatch park/unpark\n", thrd, thrd->self_cor.name);
[de6319f]	518	}
[eb2e723]	519	}
	520
[24e321c]	521	void __kernel_unpark( thread$ * thrd, unpark_hint hint ) {
[e9c0b4c]	522	/* paranoid */ verify( ! __preemption_enabled() );
	523	/* paranoid */ verify( ready_schedule_islocked());
	524
	525	if( !thrd ) return;
	526
	527	if(__must_unpark(thrd)) {
	528	// Wake lost the race,
[24e321c]	529	__schedule_thread( thrd, hint );
[e9c0b4c]	530	}
	531
	532	/* paranoid */ verify( ready_schedule_islocked());
	533	/* paranoid */ verify( ! __preemption_enabled() );
	534	}
	535
[24e321c]	536	void unpark( thread$ * thrd, unpark_hint hint ) {
[c6c7e6c]	537	if( !thrd ) return;
[0b33412]	538
[c6c7e6c]	539	if(__must_unpark(thrd)) {
	540	disable_interrupts();
[a3821fa]	541	// Wake lost the race,
[24e321c]	542	schedule_thread$( thrd, hint );
[a3821fa]	543	enable_interrupts(false);
[de6319f]	544	}
[eb2e723]	545	}
[0b33412]	546
[e235429]	547	void park( void ) {
[a3821fa]	548	__disable_interrupts_checked();
	549	/* paranoid */ verify( kernelTLS().this_thread->preempted == __NO_PREEMPTION );
	550	returnToKernel();
	551	__enable_interrupts_checked();
[0c78741]	552
	553	}
[09800e9]	554
[5afb49a]	555	extern "C" {
	556	// Leave the thread monitor
	557	// last routine called by a thread.
	558	// Should never return
	559	void __cfactx_thrd_leave() {
[e84ab3d]	560	thread$ * thrd = active_thread();
	561	monitor$ * this = &thrd->self_mon;
[5afb49a]	562
	563	// Lock the monitor now
	564	lock( this->lock __cfaabi_dbg_ctx2 );
	565
	566	disable_interrupts();
	567
[9d6e1b8a]	568	/* paranoid */ verify( ! __preemption_enabled() );
	569	/* paranoid */ verify( thrd->state == Active );
	570	/* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
	571	/* paranoid */ verify( kernelTLS().this_thread == thrd );
	572	/* paranoid */ verify( thrd->context.SP );
[e84ab3d]	573	/* 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 );
	574	/* 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]	575
[3ea8ad1]	576	thrd->state = Halting;
[58688bf]	577	if( TICKET_RUNNING != thrd->ticket ) { abort( "Thread terminated with pending unpark" ); }
[9d6e1b8a]	578	if( thrd != this->owner ) { abort( "Thread internal monitor has incorrect owner" ); }
	579	if( this->recursion != 1) { abort( "Thread internal monitor has unbalanced recursion" ); }
[5afb49a]	580
	581	// Leave the thread
	582	returnToKernel();
	583
	584	// Control flow should never reach here!
[9d6e1b8a]	585	abort();
[5afb49a]	586	}
[09800e9]	587	}
	588
[14a61b5]	589	// KERNEL ONLY
[3381ed7]	590	bool force_yield( __Preemption_Reason reason ) {
[a3821fa]	591	__disable_interrupts_checked();
[e84ab3d]	592	thread$ * thrd = kernelTLS().this_thread;
[a3821fa]	593	/* paranoid */ verify(thrd->state == Active);
	594
	595	// SKULLDUGGERY: It is possible that we are preempting this thread just before
	596	// it was going to park itself. If that is the case and it is already using the
	597	// intrusive fields then we can't use them to preempt the thread
	598	// If that is the case, abandon the preemption.
	599	bool preempted = false;
	600	if(thrd->link.next == 0p) {
	601	preempted = true;
	602	thrd->preempted = reason;
	603	returnToKernel();
	604	}
	605	__enable_interrupts_checked( false );
[3381ed7]	606	return preempted;
[f2b12406]	607	}
	608
[14a61b5]	609	//=============================================================================================
[92e7631]	610	// Kernel Idle Sleep
[14a61b5]	611	//=============================================================================================
[64a7146]	612	// Wake a thread from the front if there are any
[e873838]	613	static void __wake_one(cluster * this) {
[7cf3b1d]	614	eventfd_t val;
	615
[8fc652e0]	616	/* paranoid */ verify( ! __preemption_enabled() );
[e873838]	617	/* paranoid */ verify( ready_schedule_islocked() );
[14a61b5]	618
[64a7146]	619	// Check if there is a sleeping processor
[7cf3b1d]	620	struct __fd_waitctx * fdp = __atomic_load_n(&this->procs.fdw, __ATOMIC_SEQ_CST);
[14a61b5]	621
[7cf3b1d]	622	// If no one is sleeping: we are done
	623	if( fdp == 0p ) return;
[14a61b5]	624
[7cf3b1d]	625	int fd = 1;
	626	if( __atomic_load_n(&fdp->fd, __ATOMIC_SEQ_CST) != 1 ) {
	627	fd = __atomic_exchange_n(&fdp->fd, 1, __ATOMIC_RELAXED);
	628	}
[14a61b5]	629
[7cf3b1d]	630	switch(fd) {
	631	case 0:
	632	// If the processor isn't ready to sleep then the exchange will already wake it up
	633	#if !defined(__CFA_NO_STATISTICS__)
	634	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.early++;
	635	} else { __atomic_fetch_add(&this->stats->ready.sleep.early, 1, __ATOMIC_RELAXED); }
	636	#endif
	637	break;
	638	case 1:
	639	// If someone else already said they will wake them: we are done
	640	#if !defined(__CFA_NO_STATISTICS__)
	641	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.seen++;
	642	} else { __atomic_fetch_add(&this->stats->ready.sleep.seen, 1, __ATOMIC_RELAXED); }
	643	#endif
	644	break;
	645	default:
	646	// If the processor was ready to sleep, we need to wake it up with an actual write
	647	val = 1;
	648	eventfd_write( fd, val );
	649
	650	#if !defined(__CFA_NO_STATISTICS__)
	651	if( kernelTLS().this_stats ) { __tls_stats()->ready.sleep.wakes++;
	652	} else { __atomic_fetch_add(&this->stats->ready.sleep.wakes, 1, __ATOMIC_RELAXED); }
	653	#endif
	654	break;
	655	}
[1eb239e4]	656
[e873838]	657	/* paranoid */ verify( ready_schedule_islocked() );
[8fc652e0]	658	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	659
	660	return;
[64a7146]	661	}
[14a61b5]	662
[64a7146]	663	// Unconditionnaly wake a thread
[1eb239e4]	664	void __wake_proc(processor * this) {
[7d0ebd0]	665	/* paranoid */ verify( ! __preemption_enabled() );
	666
[64a7146]	667	__cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
[14a61b5]	668
[7cf3b1d]	669	this->idle_wctx.fd = 1;
	670
[7d0ebd0]	671	eventfd_t val;
	672	val = 1;
	673	eventfd_write( this->idle_fd, val );
	674
	675	/* paranoid */ verify( ! __preemption_enabled() );
[92e7631]	676	}
	677
[d3605f8]	678	static void idle_sleep(processor * this, io_future_t & future, iovec & iov) {
[7cf3b1d]	679	// Tell everyone we are ready to go do sleep
	680	for() {
	681	int expected = this->idle_wctx.fd;
	682
	683	// Someone already told us to wake-up! No time for a nap.
	684	if(expected == 1) { return; }
	685
	686	// Try to mark that we are going to sleep
	687	if(__atomic_compare_exchange_n(&this->idle_wctx.fd, &expected, this->idle_fd, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST) ) {
	688	// Every one agreed, taking a nap
	689	break;
	690	}
	691	}
	692
	693
[d3605f8]	694	#if !defined(CFA_WITH_IO_URING_IDLE)
[7ef162b2]	695	#if !defined(__CFA_NO_STATISTICS__)
	696	if(this->print_halts) {
	697	__cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 0\n", this->unique_id, rdtscl());
	698	}
	699	#endif
[1757f98]	700
[7ef162b2]	701	__cfadbg_print_safe(runtime_core, "Kernel : core %p waiting on eventfd %d\n", this, this->idle_fd);
[1757f98]	702
[7ef162b2]	703	{
	704	eventfd_t val;
	705	ssize_t ret = read( this->idle_fd, &val, sizeof(val) );
	706	if(ret < 0) {
	707	switch((int)errno) {
	708	case EAGAIN:
	709	#if EAGAIN != EWOULDBLOCK
	710	case EWOULDBLOCK:
	711	#endif
	712	case EINTR:
	713	// No need to do anything special here, just assume it's a legitimate wake-up
	714	break;
	715	default:
	716	abort( "KERNEL : internal error, read failure on idle eventfd, error(%d) %s.", (int)errno, strerror( (int)errno ) );
	717	}
[1757f98]	718	}
	719	}
	720
[7ef162b2]	721	#if !defined(__CFA_NO_STATISTICS__)
	722	if(this->print_halts) {
	723	__cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 1\n", this->unique_id, rdtscl());
	724	}
	725	#endif
	726	#else
	727	// Do we already have a pending read
	728	if(available(future)) {
	729	// There is no pending read, we need to add one
	730	reset(future);
	731
[d3605f8]	732	__kernel_read(this, future, iov, this->idle_fd );
[1757f98]	733	}
[7ef162b2]	734
[c7b2215]	735	__cfa_io_flush( this, 1 );
[1757f98]	736	#endif
	737	}
	738
[5f5a729]	739	static bool mark_idle(__cluster_proc_list & this, processor & proc) {
[7cf3b1d]	740	#if !defined(__CFA_NO_STATISTICS__)
	741	__tls_stats()->ready.sleep.halts++;
	742	#endif
	743
	744	proc.idle_wctx.fd = 0;
	745
[8fc652e0]	746	/* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]	747	if(!try_lock( this )) return false;
[1eb239e4]	748	this.idle++;
	749	/* paranoid */ verify( this.idle <= this.total );
[fc59b580]	750	remove(proc);
[6a9b12b]	751	insert_first(this.idles, proc);
[34b8cb7]	752
[7cf3b1d]	753	__atomic_store_n(&this.fdw, &proc.idle_wctx, __ATOMIC_SEQ_CST);
[1eb239e4]	754	unlock( this );
[8fc652e0]	755	/* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]	756
	757	return true;
[1eb239e4]	758	}
[8e16177]	759
[6a9b12b]	760	static void mark_awake(__cluster_proc_list & this, processor & proc) {
[8fc652e0]	761	/* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]	762	lock( this );
	763	this.idle--;
	764	/* paranoid */ verify( this.idle >= 0 );
	765	remove(proc);
[fc59b580]	766	insert_last(this.actives, proc);
[6a9b12b]	767
[a633f6f]	768	{
[7cf3b1d]	769	struct __fd_waitctx * wctx = 0;
	770	if(!this.idles`isEmpty) wctx = &this.idles`first.idle_wctx;
	771	__atomic_store_n(&this.fdw, wctx, __ATOMIC_SEQ_CST);
[a633f6f]	772	}
	773
[34b8cb7]	774	unlock( this );
[6a9b12b]	775	/* paranoid */ verify( ! __preemption_enabled() );
[6b4cdd3]	776	}
	777
[dbe9b08]	778	//=============================================================================================
	779	// Unexpected Terminating logic
	780	//=============================================================================================
[92bfda0]	781	void __kernel_abort_msg( char * abort_text, int abort_text_size ) {
[e84ab3d]	782	thread$ * thrd = __cfaabi_tls.this_thread;
[9d944b2]	783
[de94a60]	784	if(thrd) {
	785	int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
[1c40091]	786	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]	787
[212c2187]	788	if ( &thrd->self_cor != thrd->curr_cor ) {
	789	len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
[1c40091]	790	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]	791	}
	792	else {
[1c40091]	793	__cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
[de94a60]	794	}
[1c273d0]	795	}
[9d944b2]	796	else {
[de94a60]	797	int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
[1c40091]	798	__cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[9d944b2]	799	}
	800	}
	801
[92bfda0]	802	int __kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
	803	return get_coroutine(__cfaabi_tls.this_thread) == get_coroutine(mainThread) ? 4 : 2;
[2b8bc41]	804	}
	805
[de94a60]	806	static __spinlock_t kernel_debug_lock;
	807
[9d944b2]	808	extern "C" {
[1c40091]	809	void __cfaabi_bits_acquire() {
[36982fc]	810	lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
[9d944b2]	811	}
	812
[1c40091]	813	void __cfaabi_bits_release() {
[ea7d2b0]	814	unlock( kernel_debug_lock );
[9d944b2]	815	}
[8118303]	816	}
	817
[fa21ac9]	818	//=============================================================================================
	819	// Kernel Utilities
	820	//=============================================================================================
[dddb3dd0]	821	#if defined(CFA_HAVE_LINUX_IO_URING_H)
	822	#include "io/types.hfa"
	823	#endif
	824
[c1c95b1]	825	static inline bool __maybe_io_drain( processor * proc ) {
[e9c0b4c]	826	bool ret = false;
[dddb3dd0]	827	#if defined(CFA_HAVE_LINUX_IO_URING_H)
	828	__cfadbg_print_safe(runtime_core, "Kernel : core %p checking io for ring %d\n", proc, proc->io.ctx->fd);
	829
	830	// Check if we should drain the queue
	831	$io_context * ctx = proc->io.ctx;
	832	unsigned head = *ctx->cq.head;
	833	unsigned tail = *ctx->cq.tail;
[c1c95b1]	834	if(head == tail) return false;
[040334e]	835	ready_schedule_lock();
	836	ret = __cfa_io_drain( proc );
	837	ready_schedule_unlock();
[dddb3dd0]	838	#endif
[e9c0b4c]	839	return ret;
[dddb3dd0]	840	}
	841
[de94a60]	842	//-----------------------------------------------------------------------------
	843	// Debug
	844	__cfaabi_dbg_debug_do(
[1997b4e]	845	extern "C" {
[ae66348]	846	void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
[1997b4e]	847	this.prev_name = prev_name;
[8fc652e0]	848	this.prev_thrd = kernelTLS().this_thread;
[1997b4e]	849	}
[9181f1d]	850	}
[f7d6bb0]	851	)
[2026bb6]	852
	853	//-----------------------------------------------------------------------------
	854	// Debug
[8c50aed]	855	bool threading_enabled(void) __attribute__((const)) {
[2026bb6]	856	return true;
	857	}
[c34ebf2]	858
	859	//-----------------------------------------------------------------------------
	860	// Statistics
	861	#if !defined(__CFA_NO_STATISTICS__)
	862	void print_halts( processor & this ) {
	863	this.print_halts = true;
	864	}
[58688bf]	865
[69914cbc]	866	static void crawl_list( cluster * cltr, dlist(processor) & list, unsigned count ) {
[8464edf]	867	/* paranoid */ verify( cltr->stats );
	868
	869	processor * it = &list`first;
	870	for(unsigned i = 0; i < count; i++) {
	871	/* paranoid */ verifyf( it, "Unexpected null iterator, at index %u of %u\n", i, count);
	872	/* paranoid */ verify( it->local_data->this_stats );
[c33c2af]	873	// __print_stats( it->local_data->this_stats, cltr->print_stats, "Processor", it->name, (void*)it );
[8464edf]	874	__tally_stats( cltr->stats, it->local_data->this_stats );
	875	it = &(*it)`next;
	876	}
	877	}
	878
	879	void crawl_cluster_stats( cluster & this ) {
	880	// Stop the world, otherwise stats could get really messed-up
	881	// this doesn't solve all problems but does solve many
	882	// so it's probably good enough
[c33c2af]	883	disable_interrupts();
[8464edf]	884	uint_fast32_t last_size = ready_mutate_lock();
	885
	886	crawl_list(&this, this.procs.actives, this.procs.total - this.procs.idle);
	887	crawl_list(&this, this.procs.idles , this.procs.idle );
	888
	889	// Unlock the RWlock
	890	ready_mutate_unlock( last_size );
[c33c2af]	891	enable_interrupts();
[8464edf]	892	}
	893
	894
[58688bf]	895	void print_stats_now( cluster & this, int flags ) {
[8464edf]	896	crawl_cluster_stats( this );
[1b033b8]	897	__print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
	898	}
[c34ebf2]	899	#endif
[8118303]	900	// Local Variables: //
	901	// mode: c //
	902	// tab-width: 4 //
	903	// End: //

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

Download in other formats:

Original Format