source: libcfa/src/concurrency/kernel.hfa @ 1b64344

ADTast-experimentalpthread-emulationqualifiedEnum
Last change on this file since 1b64344 was efa28d5, checked in by Thierry Delisle <tdelisle@…>, 3 years ago

Change wake_time to be as long as other timestamps to ease debugging.

  • Property mode set to 100644
File size: 9.4 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// Coroutine used py processors for the 2-step context switch
51coroutine processorCtx_t {
52        struct processor * proc;
53};
54
55struct io_future_t;
56
57// Information needed for idle sleep
58struct __fd_waitctx {
59        // semaphore/future like object
60        // values can be 0, 1 or some file descriptor.
61        // 0 - is the default state
62        // 1 - means the proc should wake-up immediately
63        // FD - means the proc is going asleep and should be woken by writing to the FD.
64        volatile int sem;
65
66        // The event FD that corresponds to this processor
67        int evfd;
68
69        // buffer into which the proc will read from evfd
70        // unused if not using io_uring for idle sleep
71        void * rdbuf;
72
73        // future use to track the read of the eventfd
74        // unused if not using io_uring for idle sleep
75        io_future_t * ftr;
76
77        volatile unsigned long long wake__time;
78        volatile unsigned long long sleep_time;
79        volatile unsigned long long drain_time;
80};
81
82// Wrapper around kernel threads
83struct __attribute__((aligned(128))) processor {
84        // Cluster from which to get threads
85        struct cluster * cltr;
86
87        // Ready Queue state per processor
88        struct {
89                unsigned short its;
90                unsigned short itr;
91                unsigned id;
92                unsigned target;
93                unsigned last;
94                signed   cpu;
95        } rdq;
96
97        // Set to true to notify the processor should terminate
98        volatile bool do_terminate;
99
100        // Coroutine ctx who does keeps the state of the processor
101        struct processorCtx_t runner;
102
103        // Name of the processor
104        const char * name;
105
106        // Handle to pthreads
107        pthread_t kernel_thread;
108
109        // Unique id for the processor (not per cluster)
110        unsigned unique_id;
111
112        struct {
113                $io_context * ctx;
114                unsigned target;
115                volatile bool pending;
116                volatile bool dirty;
117        } io;
118
119        // Preemption data
120        // Node which is added in the discrete event simulaiton
121        struct alarm_node_t * preemption_alarm;
122
123        // If true, a preemption was triggered in an unsafe region, the processor must preempt as soon as possible
124        bool pending_preemption;
125
126        // context for idle sleep
127        struct __fd_waitctx idle_wctx;
128
129        // Termination synchronisation (user semaphore)
130        oneshot terminated;
131
132        // pthread Stack
133        void * stack;
134
135        // Link lists fields
136        inline dlink(processor);
137
138        // special init fields
139        // This is needed for memcached integration
140        // once memcached experiments are done this should probably be removed
141        // it is not a particularly safe scheme as it can make processors less homogeneous
142        struct {
143                thread$ * thrd;
144        } init;
145
146        struct KernelThreadData * local_data;
147
148        #if !defined(__CFA_NO_STATISTICS__)
149                int print_stats;
150                bool print_halts;
151        #endif
152
153#ifdef __CFA_DEBUG__
154        // Last function to enable preemption on this processor
155        const char * last_enable;
156#endif
157};
158P9_EMBEDDED( processor, dlink(processor) )
159
160void  ?{}(processor & this, const char name[], struct cluster & cltr);
161void ^?{}(processor & this);
162
163static inline void  ?{}(processor & this)                        { this{ "Anonymous Processor", *mainCluster}; }
164static inline void  ?{}(processor & this, struct cluster & cltr) { this{ "Anonymous Processor", cltr}; }
165static inline void  ?{}(processor & this, const char name[])     { this{name, *mainCluster}; }
166
167//-----------------------------------------------------------------------------
168// Cluster Tools
169
170// Intrusives lanes which are used by the ready queue
171struct __attribute__((aligned(128))) __intrusive_lane_t;
172void  ?{}(__intrusive_lane_t & this);
173void ^?{}(__intrusive_lane_t & this);
174
175// Aligned timestamps which are used by the ready queue and io subsystem
176struct __attribute__((aligned(128))) __timestamp_t {
177        volatile unsigned long long tv;
178        volatile unsigned long long ma;
179};
180
181static inline void  ?{}(__timestamp_t & this) { this.tv = 0; this.ma = 0; }
182static inline void ^?{}(__timestamp_t &) {}
183
184
185struct __attribute__((aligned(16))) __cache_id_t {
186        volatile unsigned id;
187};
188
189// Idle Sleep
190struct __cluster_proc_list {
191        // Spin lock protecting the queue
192        __spinlock_t lock;
193
194        // FD to use to wake a processor
195        struct __fd_waitctx * volatile fdw;
196
197        // Total number of processors
198        unsigned total;
199
200        // Total number of idle processors
201        unsigned idle;
202
203        // List of idle processors
204        dlist(processor) idles;
205
206        // List of active processors
207        dlist(processor) actives;
208};
209
210//-----------------------------------------------------------------------------
211// Cluster
212struct __attribute__((aligned(128))) cluster {
213        struct {
214                struct {
215                        // Arary of subqueues
216                        __intrusive_lane_t * data;
217
218                        // Time since subqueues were processed
219                        __timestamp_t * tscs;
220
221                        // Number of subqueue / timestamps
222                        size_t count;
223                } readyQ;
224
225                struct {
226                        // Array of $io_
227                        $io_context ** data;
228
229                        // Time since subqueues were processed
230                        __timestamp_t * tscs;
231
232                        // Number of I/O subqueues
233                        size_t count;
234                } io;
235
236                // Cache each kernel thread belongs to
237                __cache_id_t * caches;
238        } sched;
239
240        // // Ready queue for threads
241        // __ready_queue_t ready_queue;
242
243        // Name of the cluster
244        const char * name;
245
246        // Preemption rate on this cluster
247        Duration preemption_rate;
248
249        // List of idle processors
250        __cluster_proc_list procs;
251
252        // List of threads
253        __spinlock_t thread_list_lock;
254        __dllist_t(struct thread$) threads;
255        unsigned int nthreads;
256
257        // Link lists fields
258        struct __dbg_node_cltr {
259                cluster * next;
260                cluster * prev;
261        } node;
262
263        struct {
264                $io_arbiter * arbiter;
265                io_context_params params;
266        } io;
267
268        #if !defined(__CFA_NO_STATISTICS__)
269                struct __stats_t * stats;
270                int print_stats;
271        #endif
272};
273extern Duration default_preemption();
274
275void ?{} (cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params);
276void ^?{}(cluster & this);
277
278static inline void ?{} (cluster & this)                                            { io_context_params default_params;    this{"Anonymous Cluster", default_preemption(), 1, default_params}; }
279static inline void ?{} (cluster & this, Duration preemption_rate)                  { io_context_params default_params;    this{"Anonymous Cluster", preemption_rate, 1, default_params}; }
280static inline void ?{} (cluster & this, const char name[])                         { io_context_params default_params;    this{name, default_preemption(), 1, default_params}; }
281static inline void ?{} (cluster & this, unsigned num_io)                           { io_context_params default_params;    this{"Anonymous Cluster", default_preemption(), num_io, default_params}; }
282static inline void ?{} (cluster & this, Duration preemption_rate, unsigned num_io) { io_context_params default_params;    this{"Anonymous Cluster", preemption_rate, num_io, default_params}; }
283static inline void ?{} (cluster & this, const char name[], unsigned num_io)        { io_context_params default_params;    this{name, default_preemption(), num_io, default_params}; }
284static inline void ?{} (cluster & this, const io_context_params & io_params)                                            { this{"Anonymous Cluster", default_preemption(), 1, io_params}; }
285static inline void ?{} (cluster & this, Duration preemption_rate, const io_context_params & io_params)                  { this{"Anonymous Cluster", preemption_rate, 1, io_params}; }
286static inline void ?{} (cluster & this, const char name[], const io_context_params & io_params)                         { this{name, default_preemption(), 1, io_params}; }
287static inline void ?{} (cluster & this, unsigned num_io, const io_context_params & io_params)                           { this{"Anonymous Cluster", default_preemption(), num_io, io_params}; }
288static 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}; }
289static inline void ?{} (cluster & this, const char name[], unsigned num_io, const io_context_params & io_params)        { this{name, default_preemption(), num_io, io_params}; }
290
291static inline [cluster *&, cluster *& ] __get( cluster & this ) __attribute__((const)) { return this.node.[next, prev]; }
292
293static inline struct processor * active_processor() { return publicTLS_get( this_processor ); } // UNSAFE
294static inline struct cluster   * active_cluster  () { return publicTLS_get( this_processor )->cltr; }
295
296#if !defined(__CFA_NO_STATISTICS__)
297        void print_stats_now( cluster & this, int flags );
298
299        static inline void print_stats_at_exit( cluster & this, int flags ) {
300                this.print_stats |= flags;
301        }
302
303        static inline void print_stats_at_exit( processor & this, int flags ) {
304                this.print_stats |= flags;
305        }
306
307        void print_halts( processor & this );
308#endif
309
310// Local Variables: //
311// mode: c //
312// tab-width: 4 //
313// End: //
Note: See TracBrowser for help on using the repository browser.