1 | /* $Id: tstTSC.cpp 4071 2007-08-07 17:07:59Z vboxsync $ */
|
---|
2 | /** @file
|
---|
3 | * innotek Portable Runtime Testcase - SMP TSC testcase.
|
---|
4 | */
|
---|
5 |
|
---|
6 | /*
|
---|
7 | * Copyright (C) 2006-2007 innotek GmbH
|
---|
8 | *
|
---|
9 | * This file is part of VirtualBox Open Source Edition (OSE), as
|
---|
10 | * available from http://www.virtualbox.org. This file is free software;
|
---|
11 | * you can redistribute it and/or modify it under the terms of the GNU
|
---|
12 | * General Public License as published by the Free Software Foundation,
|
---|
13 | * in version 2 as it comes in the "COPYING" file of the VirtualBox OSE
|
---|
14 | * distribution. VirtualBox OSE is distributed in the hope that it will
|
---|
15 | * be useful, but WITHOUT ANY WARRANTY of any kind.
|
---|
16 | */
|
---|
17 |
|
---|
18 | /*******************************************************************************
|
---|
19 | * Header Files *
|
---|
20 | *******************************************************************************/
|
---|
21 | #include <iprt/runtime.h>
|
---|
22 |
|
---|
23 | /*******************************************************************************
|
---|
24 | * Structures and Typedefs *
|
---|
25 | *******************************************************************************/
|
---|
26 | typedef struct TSCDATA
|
---|
27 | {
|
---|
28 | /** The TSC. */
|
---|
29 | uint64_t volatile TSC;
|
---|
30 | /** The APIC ID. */
|
---|
31 | uint8_t volatile u8ApicId;
|
---|
32 | /** Did it succeed? */
|
---|
33 | bool volatile fRead;
|
---|
34 | /** Did it fail? */
|
---|
35 | bool volatile fFailed;
|
---|
36 | /** The thread handle. */
|
---|
37 | RTTHREAD Thread;
|
---|
38 | } TSCDATA, *PTSCDATA;
|
---|
39 |
|
---|
40 | /*******************************************************************************
|
---|
41 | * Global Variables *
|
---|
42 | *******************************************************************************/
|
---|
43 | /** The number of CPUs waiting on their user event semaphore. */
|
---|
44 | static volatile uint32_t g_cWaiting;
|
---|
45 | /** The number of CPUs ready (in spin) to do the TSC read. */
|
---|
46 | static volatile uint32_t g_cReady;
|
---|
47 | /** The variable the CPUs are spinning on.
|
---|
48 | * 0: Spin.
|
---|
49 | * 1: Go ahead.
|
---|
50 | * 2: You're too late, back to square one. */
|
---|
51 | static volatile uint32_t g_u32Go;
|
---|
52 | /** The number of CPUs that managed to read the TSC. */
|
---|
53 | static volatile uint32_t g_cRead;
|
---|
54 | /** The number of CPUs that failed to read the TSC. */
|
---|
55 | static volatile uint32_t g_cFailed;
|
---|
56 |
|
---|
57 | /** Indicator forcing the threads to quit. */
|
---|
58 | static volatile bool g_fDone;
|
---|
59 |
|
---|
60 |
|
---|
61 | /*******************************************************************************
|
---|
62 | * Internal Functions *
|
---|
63 | *******************************************************************************/
|
---|
64 | static DECLCALLBACK(int) ThreadFunction(RTTHREAD Thread, void *pvUser);
|
---|
65 |
|
---|
66 |
|
---|
67 | /**
|
---|
68 | * Thread function for catching the other cpus.
|
---|
69 | *
|
---|
70 | * @returns VINF_SUCCESS (we don't care).
|
---|
71 | * @param Thread The thread handle.
|
---|
72 | * @param pvUser PTSCDATA.
|
---|
73 | */
|
---|
74 | static DECLCALLBACK(int) ThreadFunction(RTTHREAD Thread, void *pvUser)
|
---|
75 | {
|
---|
76 | PTSCDATA pTscData = (PTSCDATA)pvUser;
|
---|
77 |
|
---|
78 | while (!g_fDone)
|
---|
79 | {
|
---|
80 | /*
|
---|
81 | * Wait.
|
---|
82 | */
|
---|
83 | ASMAtomicIncU32(&g_cWaiting);
|
---|
84 | RTThreadUserWait(Thread, RT_INDEFINITE_WAIT);
|
---|
85 | RTThreadUserReset(Thread);
|
---|
86 | ASMAtomicDecU32(&g_cWaiting);
|
---|
87 | if (g_fDone)
|
---|
88 | break;
|
---|
89 |
|
---|
90 | /*
|
---|
91 | * Spin.
|
---|
92 | */
|
---|
93 | ASMAtomicIncU32(&g_cReady);
|
---|
94 | while (!g_fDone)
|
---|
95 | {
|
---|
96 | const uint8_t ApicId1 = ASMGetApicId();
|
---|
97 | const uint64_t TSC1 = ASMReadTSC();
|
---|
98 | const uint32_t u32Go = g_u32Go;
|
---|
99 | if (u32Go == 0)
|
---|
100 | continue;
|
---|
101 |
|
---|
102 | if (u32Go == 1)
|
---|
103 | {
|
---|
104 | /* do the reading. */
|
---|
105 | const uint8_t ApicId2 = ASMGetApicId();
|
---|
106 | const uint64_t TSC2 = ASMReadTSC();
|
---|
107 | const uint8_t ApicId3 = ASMGetApicId();
|
---|
108 | const uint64_t TSC3 = ASMReadTSC();
|
---|
109 | const uint8_t ApicId4 = ASMGetApicId();
|
---|
110 |
|
---|
111 | if ( ApicId1 == ApicId2
|
---|
112 | && ApicId1 == ApicId3
|
---|
113 | && ApicId1 == ApicId4
|
---|
114 | && TSC3 - TSC1 < 1750 /* WARNING: This is just a guess, increase if it doesn't work for you. */
|
---|
115 | && TSC2 - TSC1 < TSC3 - TSC1
|
---|
116 | )
|
---|
117 | {
|
---|
118 | /* succeeded. */
|
---|
119 | pTscData->TSC = TSC2;
|
---|
120 | pTscData->u8ApicId = ApicId1;
|
---|
121 | pTscData->fFailed = false;
|
---|
122 | pTscData->fRead = true;
|
---|
123 | ASMAtomicIncU32(&g_cRead);
|
---|
124 | break;
|
---|
125 | }
|
---|
126 | }
|
---|
127 |
|
---|
128 | /* failed */
|
---|
129 | pTscData->fFailed = true;
|
---|
130 | pTscData->fRead = false;
|
---|
131 | ASMAtomicIncU32(&g_cFailed);
|
---|
132 | break;
|
---|
133 | }
|
---|
134 | }
|
---|
135 |
|
---|
136 | return VINF_SUCCESS;
|
---|
137 | }
|
---|
138 |
|
---|
139 |
|
---|
140 | int main()
|
---|
141 | {
|
---|
142 | RTR3Init();
|
---|
143 |
|
---|
144 | /*
|
---|
145 | * This is only relevant to on SMP systems.
|
---|
146 | */
|
---|
147 | const unsigned cCpus = RTSystemProcessorGetCount();
|
---|
148 | if (cCpus <= 1)
|
---|
149 | {
|
---|
150 | RTPrintf("tstTSC: SKIPPED - Only relevant on SMP systems\n");
|
---|
151 | return 0;
|
---|
152 | }
|
---|
153 |
|
---|
154 | /*
|
---|
155 | * Create the threads.
|
---|
156 | */
|
---|
157 | static TSCDATA s_aData[254];
|
---|
158 | uint32_t i;
|
---|
159 | if (cCpus > RT_ELEMENTS(s_aData))
|
---|
160 | {
|
---|
161 | RTPrintf("tstTSC: FAILED - too many CPUs (%u)\n", cCpus);
|
---|
162 | return 1;
|
---|
163 | }
|
---|
164 |
|
---|
165 | /* ourselves. */
|
---|
166 | s_aData[0].Thread = RTThreadSelf();
|
---|
167 |
|
---|
168 | /* the others */
|
---|
169 | for (i = 1; i < cCpus; i++)
|
---|
170 | {
|
---|
171 | int rc = RTThreadCreate(&s_aData[i].Thread, ThreadFunction, &s_aData[i], 0, RTTHREADTYPE_TIMER, RTTHREADFLAGS_WAITABLE, "OTHERCPU");
|
---|
172 | if (RT_FAILURE(rc))
|
---|
173 | {
|
---|
174 | RTPrintf("tstTSC: FAILURE - RTThreatCreate failed when creating thread #%u, rc=%Rrc!\n", i, rc);
|
---|
175 | ASMAtomicXchgSize(&g_fDone, true);
|
---|
176 | while (i-- > 1)
|
---|
177 | {
|
---|
178 | RTThreadUserSignal(s_aData[i].Thread);
|
---|
179 | RTThreadWait(s_aData[i].Thread, 5000, NULL);
|
---|
180 | }
|
---|
181 | return 1;
|
---|
182 | }
|
---|
183 | }
|
---|
184 |
|
---|
185 | /*
|
---|
186 | * Retry untill we get lucky (or give up).
|
---|
187 | */
|
---|
188 | for (unsigned cTries = 0; ; cTries++)
|
---|
189 | {
|
---|
190 | if (cTries > 1024)
|
---|
191 | {
|
---|
192 | RTPrintf("tstTSC: FAILURE - %d attempts, giving.\n", cTries);
|
---|
193 | break;
|
---|
194 | }
|
---|
195 |
|
---|
196 | /*
|
---|
197 | * Wait for the other threads to get ready (brute force active wait, I'm lazy).
|
---|
198 | */
|
---|
199 | i = 0;
|
---|
200 | while (g_cWaiting < cCpus - 1)
|
---|
201 | {
|
---|
202 | if (i++ > _2G32)
|
---|
203 | break;
|
---|
204 | RTThreadSleep(i & 0xf);
|
---|
205 | }
|
---|
206 | if (g_cWaiting != cCpus - 1)
|
---|
207 | {
|
---|
208 | RTPrintf("tstTSC: FAILURE - threads failed to get waiting (%d != %d (i=%d))\n", g_cWaiting + 1, cCpus, i);
|
---|
209 | break;
|
---|
210 | }
|
---|
211 |
|
---|
212 | /*
|
---|
213 | * Send them spinning.
|
---|
214 | */
|
---|
215 | ASMAtomicXchgU32(&g_cReady, 0);
|
---|
216 | ASMAtomicXchgU32(&g_u32Go, 0);
|
---|
217 | ASMAtomicXchgU32(&g_cRead, 0);
|
---|
218 | ASMAtomicXchgU32(&g_cFailed, 0);
|
---|
219 | for (i = 1; i < cCpus; i++)
|
---|
220 | {
|
---|
221 | ASMAtomicXchgSize(&s_aData[i].fFailed, false);
|
---|
222 | ASMAtomicXchgSize(&s_aData[i].fRead, false);
|
---|
223 | ASMAtomicXchgU8(&s_aData[i].u8ApicId, 0xff);
|
---|
224 |
|
---|
225 | int rc = RTThreadUserSignal(s_aData[i].Thread);
|
---|
226 | if (RT_FAILURE(rc))
|
---|
227 | RTPrintf("tstTSC: WARNING - RTThreadUserSignal(%#u) -> rc=%Rrc!\n", i, rc);
|
---|
228 | }
|
---|
229 |
|
---|
230 | /* wait for them to get ready. */
|
---|
231 | i = 0;
|
---|
232 | while (g_cReady < cCpus - 1)
|
---|
233 | {
|
---|
234 | if (i++ > _2G32)
|
---|
235 | break;
|
---|
236 | }
|
---|
237 | if (g_cReady != cCpus - 1)
|
---|
238 | {
|
---|
239 | RTPrintf("tstTSC: FAILURE - threads failed to get ready (%d != %d, i=%d)\n", g_cWaiting + 1, cCpus, i);
|
---|
240 | break;
|
---|
241 | }
|
---|
242 |
|
---|
243 | /*
|
---|
244 | * Flip the "go" switch and do our readings.
|
---|
245 | * We give the other threads the slack it takes to two extra TSC and APIC ID reads.
|
---|
246 | */
|
---|
247 | const uint8_t ApicId1 = ASMGetApicId();
|
---|
248 | const uint64_t TSC1 = ASMReadTSC();
|
---|
249 | ASMAtomicXchgU32(&g_u32Go, 1);
|
---|
250 | const uint8_t ApicId2 = ASMGetApicId();
|
---|
251 | const uint64_t TSC2 = ASMReadTSC();
|
---|
252 | const uint8_t ApicId3 = ASMGetApicId();
|
---|
253 | const uint64_t TSC3 = ASMReadTSC();
|
---|
254 | const uint8_t ApicId4 = ASMGetApicId();
|
---|
255 | const uint64_t TSC4 = ASMReadTSC();
|
---|
256 | ASMAtomicXchgU32(&g_u32Go, 2);
|
---|
257 | const uint8_t ApicId5 = ASMGetApicId();
|
---|
258 | const uint64_t TSC5 = ASMReadTSC();
|
---|
259 | const uint8_t ApicId6 = ASMGetApicId();
|
---|
260 |
|
---|
261 | /* Compose our own result. */
|
---|
262 | if ( ApicId1 == ApicId2
|
---|
263 | && ApicId1 == ApicId3
|
---|
264 | && ApicId1 == ApicId4
|
---|
265 | && ApicId1 == ApicId5
|
---|
266 | && ApicId1 == ApicId6
|
---|
267 | && TSC5 - TSC1 < 2000 /* WARNING: This is just a guess, increase if it doesn't work for you. */
|
---|
268 | && TSC4 - TSC1 < TSC5 - TSC1
|
---|
269 | && TSC3 - TSC1 < TSC4 - TSC1
|
---|
270 | && TSC2 - TSC1 < TSC3 - TSC1
|
---|
271 | )
|
---|
272 | {
|
---|
273 | /* succeeded. */
|
---|
274 | s_aData[0].TSC = TSC2;
|
---|
275 | s_aData[0].u8ApicId = ApicId1;
|
---|
276 | s_aData[0].fFailed = false;
|
---|
277 | s_aData[0].fRead = true;
|
---|
278 | ASMAtomicIncU32(&g_cRead);
|
---|
279 | }
|
---|
280 | else
|
---|
281 | {
|
---|
282 | /* failed */
|
---|
283 | s_aData[0].fFailed = true;
|
---|
284 | s_aData[0].fRead = false;
|
---|
285 | ASMAtomicIncU32(&g_cFailed);
|
---|
286 | }
|
---|
287 |
|
---|
288 | /*
|
---|
289 | * Wait a little while to let the other ones to finish.
|
---|
290 | */
|
---|
291 | i = 0;
|
---|
292 | while (g_cRead + g_cFailed < cCpus)
|
---|
293 | {
|
---|
294 | if (i++ > _2G32)
|
---|
295 | break;
|
---|
296 | if (i > _1M)
|
---|
297 | RTThreadSleep(i & 0xf);
|
---|
298 | }
|
---|
299 | if (g_cRead + g_cFailed != cCpus)
|
---|
300 | {
|
---|
301 | RTPrintf("tstTSC: FAILURE - threads failed to complete reading (%d + %d != %d)\n", g_cRead, g_cFailed, cCpus);
|
---|
302 | break;
|
---|
303 | }
|
---|
304 |
|
---|
305 | /*
|
---|
306 | * If everone succeeded, print the results.
|
---|
307 | */
|
---|
308 | if (!g_cFailed)
|
---|
309 | {
|
---|
310 | RTPrintf(" # ID TSC\n"
|
---|
311 | "-----------------------\n");
|
---|
312 | for (i = 0; i < cCpus; i++)
|
---|
313 | RTPrintf("%2d %02x %RX64\n", i, s_aData[i].u8ApicId, s_aData[i].TSC);
|
---|
314 | RTPrintf("(Needed %u attempt%s.)\n", cTries + 1, cTries ? "s" : "");
|
---|
315 | break;
|
---|
316 | }
|
---|
317 | }
|
---|
318 |
|
---|
319 | /*
|
---|
320 | * Destroy the threads.
|
---|
321 | */
|
---|
322 | ASMAtomicXchgSize(&g_fDone, true);
|
---|
323 | for (i = 1; i < cCpus; i++)
|
---|
324 | {
|
---|
325 | int rc = RTThreadUserSignal(s_aData[i].Thread);
|
---|
326 | if (RT_FAILURE(rc))
|
---|
327 | RTPrintf("tstTSC: WARNING - RTThreadUserSignal(%#u) -> rc=%Rrc! (2)\n", i, rc);
|
---|
328 | }
|
---|
329 | for (i = 1; i < cCpus; i++)
|
---|
330 | {
|
---|
331 | int rc = RTThreadWait(s_aData[i].Thread, 5000, NULL);
|
---|
332 | if (RT_FAILURE(rc))
|
---|
333 | RTPrintf("tstTSC: WARNING - RTThreadWait(%#u) -> rc=%Rrc!\n", i, rc);
|
---|
334 | }
|
---|
335 |
|
---|
336 | return g_cFailed != 0 || g_cRead != cCpus;
|
---|
337 | }
|
---|