source: libcfa/src/concurrency/kernel.hfa @ a2a4566

ADTast-experimentalenumforall-pointer-decaypthread-emulationqualifiedEnum
Last change on this file since a2a4566 was a2a4566, checked in by Thierry Delisle <tdelisle@…>, 2 years ago

Added new ready-queue that uses per-thread queues but with some cpu awarness

  • Property mode set to 100644
File size: 9.2 KB
Line 
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 -- Header containing the core of the kernel API
8//
9// Author           : Thierry Delisle
10// Created On       : Tue Jan 17 12:27:26 2017
11// Last Modified By : Peter A. Buhr
12// Last Modified On : Tue Feb  4 12:29:26 2020
13// Update Count     : 22
14//
15
16#pragma once
17
18#include "invoke.h"
19#include "time_t.hfa"
20#include "coroutine.hfa"
21
22#include "containers/list.hfa"
23
24extern "C" {
25        #include <bits/pthreadtypes.h>
26        #include <pthread.h>
27        #include <linux/types.h>
28}
29
30#ifdef __CFA_WITH_VERIFY__
31        extern bool __cfaabi_dbg_in_kernel();
32#endif
33
34//-----------------------------------------------------------------------------
35// I/O
36struct cluster;
37struct $io_context;
38struct $io_arbiter;
39
40struct io_context_params {
41        int num_entries;
42};
43
44void  ?{}(io_context_params & this);
45
46//-----------------------------------------------------------------------------
47// Processor
48extern struct cluster * mainCluster;
49
50// Processor id, required for scheduling threads
51
52
53coroutine processorCtx_t {
54        struct processor * proc;
55};
56
57// Wrapper around kernel threads
58struct __attribute__((aligned(128))) processor {
59        // Cluster from which to get threads
60        struct cluster * cltr;
61
62        // Ready Queue state per processor
63        struct {
64                unsigned short its;
65                unsigned short itr;
66                unsigned id;
67                unsigned target;
68                unsigned last;
69                signed   cpu;
70                // unsigned long long int cutoff;
71        } rdq;
72
73        // Set to true to notify the processor should terminate
74        volatile bool do_terminate;
75
76        // Coroutine ctx who does keeps the state of the processor
77        struct processorCtx_t runner;
78
79        // Name of the processor
80        const char * name;
81
82        // Handle to pthreads
83        pthread_t kernel_thread;
84
85        // Unique id for the processor (not per cluster)
86        unsigned unique_id;
87
88        struct {
89                $io_context * ctx;
90                bool pending;
91                bool dirty;
92        } io;
93
94        // Preemption data
95        // Node which is added in the discrete event simulaiton
96        struct alarm_node_t * preemption_alarm;
97
98        // If true, a preemption was triggered in an unsafe region, the processor must preempt as soon as possible
99        bool pending_preemption;
100
101        // Idle lock (kernel semaphore)
102        int idle_fd;
103
104        // Termination synchronisation (user semaphore)
105        oneshot terminated;
106
107        // pthread Stack
108        void * stack;
109
110        // Link lists fields
111        inline dlink(processor);
112
113        // special init fields
114        // This is needed for memcached integration
115        // once memcached experiments are done this should probably be removed
116        // it is not a particularly safe scheme as it can make processors less homogeneous
117        struct {
118                thread$ * thrd;
119        } init;
120
121        struct KernelThreadData * local_data;
122
123        #if !defined(__CFA_NO_STATISTICS__)
124                int print_stats;
125                bool print_halts;
126        #endif
127
128#ifdef __CFA_DEBUG__
129        // Last function to enable preemption on this processor
130        const char * last_enable;
131#endif
132};
133P9_EMBEDDED( processor, dlink(processor) )
134
135void  ?{}(processor & this, const char name[], struct cluster & cltr);
136void ^?{}(processor & this);
137
138static inline void  ?{}(processor & this)                        { this{ "Anonymous Processor", *mainCluster}; }
139static inline void  ?{}(processor & this, struct cluster & cltr) { this{ "Anonymous Processor", cltr}; }
140static inline void  ?{}(processor & this, const char name[])     { this{name, *mainCluster}; }
141
142//-----------------------------------------------------------------------------
143// Cluster Tools
144
145// Intrusives lanes which are used by the ready queue
146struct __attribute__((aligned(128))) __intrusive_lane_t;
147void  ?{}(__intrusive_lane_t & this);
148void ^?{}(__intrusive_lane_t & this);
149
150// Aligned timestamps which are used by the relaxed ready queue
151struct __attribute__((aligned(128))) __timestamp_t {
152        volatile unsigned long long tv;
153        volatile unsigned long long ma;
154};
155
156struct __attribute__((aligned(128))) __cache_id_t {
157        volatile unsigned id;
158};
159
160// Aligned timestamps which are used by the relaxed ready queue
161struct __attribute__((aligned(128))) __help_cnts_t {
162        volatile unsigned long long src;
163        volatile unsigned long long dst;
164        volatile unsigned long long tri;
165};
166
167static inline void  ?{}(__timestamp_t & this) { this.tv = 0; this.ma = 0; }
168static inline void ^?{}(__timestamp_t & this) {}
169
170struct __attribute__((aligned(128))) __ready_queue_caches_t;
171void  ?{}(__ready_queue_caches_t & this);
172void ^?{}(__ready_queue_caches_t & this);
173
174//TODO adjust cache size to ARCHITECTURE
175// Structure holding the ready queue
176struct __ready_queue_t {
177        // Data tracking the actual lanes
178        // On a seperate cacheline from the used struct since
179        // used can change on each push/pop but this data
180        // only changes on shrink/grow
181        struct {
182                // Arary of lanes
183                __intrusive_lane_t * volatile data;
184
185                // Array of times
186                __timestamp_t * volatile tscs;
187
188                __cache_id_t * volatile caches;
189
190                // Array of stats
191                __help_cnts_t * volatile help;
192
193                // Number of lanes (empty or not)
194                volatile size_t count;
195        } lanes;
196};
197
198void  ?{}(__ready_queue_t & this);
199void ^?{}(__ready_queue_t & this);
200#if !defined(__CFA_NO_STATISTICS__)
201        unsigned cnt(const __ready_queue_t & this, unsigned idx);
202#endif
203
204// Idle Sleep
205struct __cluster_proc_list {
206        // Spin lock protecting the queue
207        __spinlock_t lock;
208
209        // FD to use to wake a processor
210        volatile int fd;
211
212        // Total number of processors
213        unsigned total;
214
215        // Total number of idle processors
216        unsigned idle;
217
218        // List of idle processors
219        dlist(processor) idles;
220
221        // List of active processors
222        dlist(processor) actives;
223};
224
225//-----------------------------------------------------------------------------
226// Cluster
227struct __attribute__((aligned(128))) cluster {
228        // Ready queue for threads
229        __ready_queue_t ready_queue;
230
231        // Name of the cluster
232        const char * name;
233
234        // Preemption rate on this cluster
235        Duration preemption_rate;
236
237        // List of idle processors
238        __cluster_proc_list procs;
239
240        // List of threads
241        __spinlock_t thread_list_lock;
242        __dllist_t(struct thread$) threads;
243        unsigned int nthreads;
244
245        // Link lists fields
246        struct __dbg_node_cltr {
247                cluster * next;
248                cluster * prev;
249        } node;
250
251        struct {
252                $io_arbiter * arbiter;
253                io_context_params params;
254        } io;
255
256        #if !defined(__CFA_NO_STATISTICS__)
257                struct __stats_t * stats;
258                int print_stats;
259        #endif
260};
261extern Duration default_preemption();
262
263void ?{} (cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params);
264void ^?{}(cluster & this);
265
266static inline void ?{} (cluster & this)                                            { io_context_params default_params;    this{"Anonymous Cluster", default_preemption(), 1, default_params}; }
267static inline void ?{} (cluster & this, Duration preemption_rate)                  { io_context_params default_params;    this{"Anonymous Cluster", preemption_rate, 1, default_params}; }
268static inline void ?{} (cluster & this, const char name[])                         { io_context_params default_params;    this{name, default_preemption(), 1, default_params}; }
269static inline void ?{} (cluster & this, unsigned num_io)                           { io_context_params default_params;    this{"Anonymous Cluster", default_preemption(), num_io, default_params}; }
270static inline void ?{} (cluster & this, Duration preemption_rate, unsigned num_io) { io_context_params default_params;    this{"Anonymous Cluster", preemption_rate, num_io, default_params}; }
271static inline void ?{} (cluster & this, const char name[], unsigned num_io)        { io_context_params default_params;    this{name, default_preemption(), num_io, default_params}; }
272static inline void ?{} (cluster & this, const io_context_params & io_params)                                            { this{"Anonymous Cluster", default_preemption(), 1, io_params}; }
273static inline void ?{} (cluster & this, Duration preemption_rate, const io_context_params & io_params)                  { this{"Anonymous Cluster", preemption_rate, 1, io_params}; }
274static inline void ?{} (cluster & this, const char name[], const io_context_params & io_params)                         { this{name, default_preemption(), 1, io_params}; }
275static inline void ?{} (cluster & this, unsigned num_io, const io_context_params & io_params)                           { this{"Anonymous Cluster", default_preemption(), num_io, io_params}; }
276static inline void ?{} (cluster & this, Duration preemption_rate, unsigned num_io, const io_context_params & io_params) { this{"Anonymous Cluster", preemption_rate, num_io, io_params}; }
277static inline void ?{} (cluster & this, const char name[], unsigned num_io, const io_context_params & io_params)        { this{name, default_preemption(), num_io, io_params}; }
278
279static inline [cluster *&, cluster *& ] __get( cluster & this ) __attribute__((const)) { return this.node.[next, prev]; }
280
281static inline struct processor * active_processor() { return publicTLS_get( this_processor ); } // UNSAFE
282static inline struct cluster   * active_cluster  () { return publicTLS_get( this_processor )->cltr; }
283
284#if !defined(__CFA_NO_STATISTICS__)
285        void print_stats_now( cluster & this, int flags );
286
287        static inline void print_stats_at_exit( cluster & this, int flags ) {
288                this.print_stats |= flags;
289        }
290
291        static inline void print_stats_at_exit( processor & this, int flags ) {
292                this.print_stats |= flags;
293        }
294
295        void print_halts( processor & this );
296#endif
297
298// Local Variables: //
299// mode: c //
300// tab-width: 4 //
301// End: //
Note: See TracBrowser for help on using the repository browser.