| 1 | /* $Id: prfcore.cpp.h 29 2009-07-01 20:30:29Z bird $ */
|
|---|
| 2 | /** @file
|
|---|
| 3 | * kProfiler Mark 2 - Core Code Template.
|
|---|
| 4 | */
|
|---|
| 5 |
|
|---|
| 6 | /*
|
|---|
| 7 | * Copyright (c) 2006-2007 Knut St. Osmundsen <bird-kStuff-spamix@anduin.net>
|
|---|
| 8 | *
|
|---|
| 9 | * Permission is hereby granted, free of charge, to any person
|
|---|
| 10 | * obtaining a copy of this software and associated documentation
|
|---|
| 11 | * files (the "Software"), to deal in the Software without
|
|---|
| 12 | * restriction, including without limitation the rights to use,
|
|---|
| 13 | * copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|---|
| 14 | * copies of the Software, and to permit persons to whom the
|
|---|
| 15 | * Software is furnished to do so, subject to the following
|
|---|
| 16 | * conditions:
|
|---|
| 17 | *
|
|---|
| 18 | * The above copyright notice and this permission notice shall be
|
|---|
| 19 | * included in all copies or substantial portions of the Software.
|
|---|
| 20 | *
|
|---|
| 21 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
|---|
| 22 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
|
|---|
| 23 | * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|---|
| 24 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
|
|---|
| 25 | * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
|
|---|
| 26 | * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
|
|---|
| 27 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
|
|---|
| 28 | * OTHER DEALINGS IN THE SOFTWARE.
|
|---|
| 29 | */
|
|---|
| 30 |
|
|---|
| 31 |
|
|---|
| 32 | /**
|
|---|
| 33 | * Gets a function, create a new one if necessary.
|
|---|
| 34 | */
|
|---|
| 35 | static KPRF_TYPE(P,FUNC) KPRF_NAME(GetFunction)(KPRF_TYPE(P,HDR) pHdr, KPRF_TYPE(,UPTR) uPC)
|
|---|
| 36 | {
|
|---|
| 37 | /*
|
|---|
| 38 | * Perform a binary search of the function lookup table.
|
|---|
| 39 | */
|
|---|
| 40 | KPRF_TYPE(P,FUNC) paFunctions = KPRF_OFF2PTR(P,FUNC, pHdr->offFunctions, pHdr);
|
|---|
| 41 |
|
|---|
| 42 | KPRF_FUNCS_READ_LOCK();
|
|---|
| 43 | KI32 iStart = 0;
|
|---|
| 44 | KI32 iLast = pHdr->cFunctions - 1;
|
|---|
| 45 | KI32 i = iLast / 2;
|
|---|
| 46 | for (;;)
|
|---|
| 47 | {
|
|---|
| 48 | KU32 iFunction = pHdr->aiFunctions[i];
|
|---|
| 49 | KPRF_TYPE(,IPTR) iDiff = uPC - paFunctions[iFunction].uEntryPtr;
|
|---|
| 50 | if (!iDiff)
|
|---|
| 51 | {
|
|---|
| 52 | KPRF_FUNCS_READ_UNLOCK();
|
|---|
| 53 | return &paFunctions[iFunction];
|
|---|
| 54 | }
|
|---|
| 55 | if (iLast == iStart)
|
|---|
| 56 | break;
|
|---|
| 57 | if (iDiff < 0)
|
|---|
| 58 | iLast = i - 1;
|
|---|
| 59 | else
|
|---|
| 60 | iStart = i + 1;
|
|---|
| 61 | if (iLast < iStart)
|
|---|
| 62 | break;
|
|---|
| 63 | i = iStart + (iLast - iStart) / 2;
|
|---|
| 64 | }
|
|---|
| 65 | KPRF_FUNCS_READ_UNLOCK();
|
|---|
| 66 |
|
|---|
| 67 | /*
|
|---|
| 68 | * It wasn't found, try add it.
|
|---|
| 69 | */
|
|---|
| 70 | if (pHdr->cFunctions < pHdr->cMaxFunctions)
|
|---|
| 71 | return KPRF_NAME(NewFunction)(pHdr, uPC);
|
|---|
| 72 | return NULL;
|
|---|
| 73 | }
|
|---|
| 74 |
|
|---|
| 75 |
|
|---|
| 76 | /**
|
|---|
| 77 | * Unwind one frame.
|
|---|
| 78 | */
|
|---|
| 79 | static KU64* KPRF_NAME(UnwindOne)(KPRF_TYPE(P,HDR) pHdr, KPRF_TYPE(P,STACK) pStack, KPRF_TYPE(,UPTR) uPC, KU64 TS)
|
|---|
| 80 | {
|
|---|
| 81 | /*
|
|---|
| 82 | * Pop off the frame and update the frame below / thread.
|
|---|
| 83 | */
|
|---|
| 84 | KPRF_TYPE(P,FRAME) pFrame = &pStack->aFrames[--pStack->cFrames];
|
|---|
| 85 | KU64 *pCurOverheadTicks;
|
|---|
| 86 | if (pStack->cFrames)
|
|---|
| 87 | {
|
|---|
| 88 | KPRF_TYPE(P,FRAME) pTopFrame = pFrame - 1;
|
|---|
| 89 | pTopFrame->OverheadTicks += pFrame->OverheadTicks + pFrame->CurOverheadTicks;
|
|---|
| 90 | pTopFrame->SleepTicks += pFrame->SleepTicks;
|
|---|
| 91 | pTopFrame->OnTopOfStackStart = TS;
|
|---|
| 92 | pTopFrame->CurOverheadTicks = 0;
|
|---|
| 93 |
|
|---|
| 94 | pCurOverheadTicks = &pTopFrame->CurOverheadTicks;
|
|---|
| 95 | }
|
|---|
| 96 | else
|
|---|
| 97 | {
|
|---|
| 98 | KPRF_TYPE(P,THREAD) pThread = KPRF_OFF2PTR(P,THREAD, pStack->offThread, pHdr);
|
|---|
| 99 | pThread->ProfiledTicks += TS - pFrame->OnStackStart - pFrame->CurOverheadTicks - pFrame->OverheadTicks - pFrame->SleepTicks;
|
|---|
| 100 | pThread->OverheadTicks += pFrame->OverheadTicks + pFrame->CurOverheadTicks;
|
|---|
| 101 | pThread->SleepTicks += pFrame->SleepTicks;
|
|---|
| 102 |
|
|---|
| 103 | pCurOverheadTicks = &pThread->OverheadTicks;
|
|---|
| 104 | }
|
|---|
| 105 |
|
|---|
| 106 | /*
|
|---|
| 107 | * Update the function (if any).
|
|---|
| 108 | */
|
|---|
| 109 | if (pFrame->offFunction)
|
|---|
| 110 | {
|
|---|
| 111 | KPRF_TYPE(P,FUNC) pFunc = KPRF_OFF2PTR(P,FUNC, pFrame->offFunction, pHdr);
|
|---|
| 112 |
|
|---|
| 113 | /* Time on stack */
|
|---|
| 114 | KU64 Ticks = TS - pFrame->OnStackStart;
|
|---|
| 115 | Ticks -= pFrame->OverheadTicks + pFrame->CurOverheadTicks + pFrame->SleepTicks;
|
|---|
| 116 | /** @todo adjust overhead */
|
|---|
| 117 | KPRF_ASSERT(!(Ticks >> 63));
|
|---|
| 118 | if (pFunc->OnStack.MinTicks > Ticks)
|
|---|
| 119 | KPRF_ATOMIC_SET64(&pFunc->OnStack.MinTicks, Ticks);
|
|---|
| 120 | if (pFunc->OnStack.MaxTicks < Ticks)
|
|---|
| 121 | KPRF_ATOMIC_SET64(&pFunc->OnStack.MaxTicks, Ticks);
|
|---|
| 122 | KPRF_ATOMIC_ADD64(&pFunc->OnStack.SumTicks, Ticks);
|
|---|
| 123 |
|
|---|
| 124 | /* Time on top of stack */
|
|---|
| 125 | Ticks = TS - pFrame->OnTopOfStackStart;
|
|---|
| 126 | Ticks -= pFrame->CurOverheadTicks;
|
|---|
| 127 | Ticks += pFrame->OnTopOfStackTicks;
|
|---|
| 128 | /** @todo adjust overhead */
|
|---|
| 129 | KPRF_ASSERT(!(Ticks >> 63));
|
|---|
| 130 | if (pFunc->OnTopOfStack.MinTicks > Ticks)
|
|---|
| 131 | KPRF_ATOMIC_SET64(&pFunc->OnTopOfStack.MinTicks, Ticks);
|
|---|
| 132 | if (pFunc->OnTopOfStack.MaxTicks < Ticks)
|
|---|
| 133 | KPRF_ATOMIC_SET64(&pFunc->OnTopOfStack.MaxTicks, Ticks);
|
|---|
| 134 | KPRF_ATOMIC_ADD64(&pFunc->OnTopOfStack.SumTicks, Ticks);
|
|---|
| 135 |
|
|---|
| 136 | /* calls */
|
|---|
| 137 | if (pFrame->cCalls)
|
|---|
| 138 | KPRF_ATOMIC_ADD64(&pFunc->cCalls, pFrame->cCalls);
|
|---|
| 139 | }
|
|---|
| 140 |
|
|---|
| 141 | return pCurOverheadTicks;
|
|---|
| 142 | }
|
|---|
| 143 |
|
|---|
| 144 |
|
|---|
| 145 | /**
|
|---|
| 146 | * Unwinds the stack.
|
|---|
| 147 | *
|
|---|
| 148 | * On MSC+AMD64 we have to be very very careful here, because the uFramePtr cannot be trusted.
|
|---|
| 149 | */
|
|---|
| 150 | static KU64* KPRF_NAME(UnwindInt)(KPRF_TYPE(P,HDR) pHdr, KPRF_TYPE(P,STACK) pStack, KPRF_TYPE(,UPTR) uPC, KPRF_TYPE(,UPTR) uFramePtr, KU64 TS)
|
|---|
| 151 | {
|
|---|
| 152 | /** @todo need to deal with alternative stacks! */
|
|---|
| 153 |
|
|---|
| 154 | /*
|
|---|
| 155 | * Pop the stack until we're down below the current frame (uFramePtr).
|
|---|
| 156 | */
|
|---|
| 157 | KI32 iFrame = pStack->cFrames - 1;
|
|---|
| 158 | KPRF_TYPE(P,FRAME) pFrame = &pStack->aFrames[iFrame];
|
|---|
| 159 |
|
|---|
| 160 | /* the most frequent case first. */
|
|---|
| 161 | #if K_OS == K_OS_WINDOWS && K_ARCH == K_ARCH_AMD64
|
|---|
| 162 | if ( uFramePtr == pFrame->uFramePtr
|
|---|
| 163 | || ( pFrame->uFramePtr < uFramePtr
|
|---|
| 164 | && iFrame > 0
|
|---|
| 165 | && pFrame[-1].uFramePtr > uFramePtr))
|
|---|
| 166 | return KPRF_NAME(UnwindOne)(pHdr, pStack, uPC, TS);
|
|---|
| 167 | #else
|
|---|
| 168 | if (uFramePtr == pFrame->uFramePtr)
|
|---|
| 169 | return KPRF_NAME(UnwindOne)(pHdr, pStack, uPC, TS);
|
|---|
| 170 | #endif
|
|---|
| 171 |
|
|---|
| 172 | /* none? */
|
|---|
| 173 | if (pFrame->uFramePtr > uFramePtr)
|
|---|
| 174 | return &pFrame->CurOverheadTicks;
|
|---|
| 175 |
|
|---|
| 176 | /* one or more, possibly all */
|
|---|
| 177 | KU64 *pCurOverheadTicks = KPRF_NAME(UnwindOne)(pHdr, pStack, uPC, TS);
|
|---|
| 178 | pFrame--;
|
|---|
| 179 | if ( iFrame > 0
|
|---|
| 180 | #if K_OS == K_OS_WINDOWS && K_ARCH == K_ARCH_AMD64
|
|---|
| 181 | && pFrame->uFramePtr <= uFramePtr
|
|---|
| 182 | && pFrame[-1].uFramePtr > uFramePtr)
|
|---|
| 183 | #else
|
|---|
| 184 | && pFrame->uFramePtr <= uFramePtr)
|
|---|
| 185 | #endif
|
|---|
| 186 | {
|
|---|
| 187 | KPRF_TYPE(P,THREAD) pThread = KPRF_OFF2PTR(P,THREAD, pStack->offThread, pHdr);
|
|---|
| 188 | pThread->cUnwinds++; /* (This is the reason for what looks like a bad loop unrolling.) */
|
|---|
| 189 |
|
|---|
| 190 | pCurOverheadTicks = KPRF_NAME(UnwindOne)(pHdr, pStack, uPC, TS);
|
|---|
| 191 | iFrame -= 2;
|
|---|
| 192 | pFrame--;
|
|---|
| 193 | #if K_OS == K_OS_WINDOWS && K_ARCH == K_ARCH_AMD64
|
|---|
| 194 | while ( iFrame > 0
|
|---|
| 195 | && pFrame->uFramePtr <= uFramePtr
|
|---|
| 196 | && pFrame[-1].uFramePtr > uFramePtr)
|
|---|
| 197 | #else
|
|---|
| 198 | while ( iFrame >= 0
|
|---|
| 199 | && pFrame->uFramePtr <= uFramePtr)
|
|---|
| 200 | #endif
|
|---|
| 201 | {
|
|---|
| 202 | pCurOverheadTicks = KPRF_NAME(UnwindOne)(pHdr, pStack, uPC, TS);
|
|---|
| 203 | iFrame--;
|
|---|
| 204 | pFrame--;
|
|---|
| 205 | }
|
|---|
| 206 | }
|
|---|
| 207 |
|
|---|
| 208 | return pCurOverheadTicks;
|
|---|
| 209 | }
|
|---|
| 210 |
|
|---|
| 211 |
|
|---|
| 212 |
|
|---|
| 213 | /**
|
|---|
| 214 | * Enter function.
|
|---|
| 215 | *
|
|---|
| 216 | * @returns Where to account overhead.
|
|---|
| 217 | * @returns NULL if profiling is inactive.
|
|---|
| 218 | *
|
|---|
| 219 | * @param uPC The program counter register. (not relative)
|
|---|
| 220 | * @param uFramePtr The stack frame address. This must match the one passed to kPrfLeave. (not relative)
|
|---|
| 221 | * @param TS The timestamp when we entered into the profiler.
|
|---|
| 222 | * This must not be modified touched!
|
|---|
| 223 | *
|
|---|
| 224 | * @internal ?
|
|---|
| 225 | */
|
|---|
| 226 | KPRF_DECL_FUNC(KU64 *, Enter)(KPRF_TYPE(,UPTR) uPC, KPRF_TYPE(,UPTR) uFramePtr, const KU64 TS)
|
|---|
| 227 | {
|
|---|
| 228 | /*
|
|---|
| 229 | * Is profiling active ?
|
|---|
| 230 | */
|
|---|
| 231 | if (!KPRF_IS_ACTIVE())
|
|---|
| 232 | return NULL;
|
|---|
| 233 |
|
|---|
| 234 | /*
|
|---|
| 235 | * Get the header and adjust input addresses.
|
|---|
| 236 | */
|
|---|
| 237 | KPRF_TYPE(P,HDR) pHdr = KPRF_GET_HDR();
|
|---|
| 238 | if (!pHdr)
|
|---|
| 239 | return NULL;
|
|---|
| 240 | const KPRF_TYPE(,UPTR) uBasePtr = pHdr->uBasePtr;
|
|---|
| 241 | if (uBasePtr)
|
|---|
| 242 | {
|
|---|
| 243 | uFramePtr -= uBasePtr;
|
|---|
| 244 | uPC -= uBasePtr;
|
|---|
| 245 | }
|
|---|
| 246 |
|
|---|
| 247 | /*
|
|---|
| 248 | * Get the current thread. Reject unknown, inactive (in whatever way),
|
|---|
| 249 | * and thread which has performed a stack switch.
|
|---|
| 250 | */
|
|---|
| 251 | KPRF_TYPE(P,THREAD) pThread = KPRF_GET_THREAD();
|
|---|
| 252 | if (!pThread)
|
|---|
| 253 | return NULL;
|
|---|
| 254 | KPRF_TYPE(,THREADSTATE) enmThreadState = pThread->enmState;
|
|---|
| 255 | if ( enmThreadState != KPRF_TYPE(,THREADSTATE_ACTIVE)
|
|---|
| 256 | && enmThreadState != KPRF_TYPE(,THREADSTATE_OVERFLOWED)
|
|---|
| 257 | )
|
|---|
| 258 | return NULL;
|
|---|
| 259 | if (pThread->uStackBasePtr < uFramePtr) /* ASSUMES stack direction */
|
|---|
| 260 | {
|
|---|
| 261 | pThread->cStackSwitchRejects++;
|
|---|
| 262 | return NULL;
|
|---|
| 263 | }
|
|---|
| 264 | pThread->enmState = KPRF_TYPE(,THREADSTATE_SUSPENDED);
|
|---|
| 265 |
|
|---|
| 266 |
|
|---|
| 267 | /*
|
|---|
| 268 | * Update the thread statistics.
|
|---|
| 269 | */
|
|---|
| 270 | pThread->cCalls++;
|
|---|
| 271 | KPRF_TYPE(,UPTR) cbStack = pThread->uStackBasePtr - uFramePtr; /* ASSUMES stack direction */
|
|---|
| 272 | if (pThread->cbMaxStack < cbStack)
|
|---|
| 273 | pThread->cbMaxStack = cbStack;
|
|---|
| 274 |
|
|---|
| 275 | /*
|
|---|
| 276 | * Check if an longjmp or throw has taken place.
|
|---|
| 277 | * This check will not work if a stack switch has taken place (can fix that later).
|
|---|
| 278 | */
|
|---|
| 279 | KPRF_TYPE(P,STACK) pStack = KPRF_OFF2PTR(P,STACK, pThread->offStack, pHdr);
|
|---|
| 280 | KU32 iFrame = pStack->cFrames;
|
|---|
| 281 | KPRF_TYPE(P,FRAME) pFrame = &pStack->aFrames[iFrame];
|
|---|
| 282 | if ( iFrame
|
|---|
| 283 | #if K_OS == K_OS_WINDOWS && K_ARCH == K_ARCH_AMD64
|
|---|
| 284 | && 0) /* don't bother her yet because of _penter/_pexit frame problems. */
|
|---|
| 285 | #else
|
|---|
| 286 | && pThread->uStackBasePtr >= uFramePtr /* ASSUMES stack direction */
|
|---|
| 287 | && pFrame[-1].uFramePtr + (KPRF_BITS - 8) / 8 < uFramePtr) /* ASSUMES stack direction */
|
|---|
| 288 | #endif
|
|---|
| 289 | {
|
|---|
| 290 | KPRF_NAME(UnwindInt)(pHdr, pStack, uPC, uFramePtr, TS);
|
|---|
| 291 | iFrame = pStack->cFrames;
|
|---|
| 292 | }
|
|---|
| 293 |
|
|---|
| 294 | /*
|
|---|
| 295 | * Allocate a new stack frame.
|
|---|
| 296 | */
|
|---|
| 297 | if (iFrame >= pHdr->cMaxStackFrames)
|
|---|
| 298 | {
|
|---|
| 299 | /* overflow */
|
|---|
| 300 | pThread->enmState = KPRF_TYPE(,THREADSTATE_OVERFLOWED);
|
|---|
| 301 | pThread->cOverflows += enmThreadState != KPRF_TYPE(,THREADSTATE_OVERFLOWED);
|
|---|
| 302 | return &pStack->aFrames[iFrame - 1].CurOverheadTicks;
|
|---|
| 303 | }
|
|---|
| 304 | pStack->cFrames++;
|
|---|
| 305 |
|
|---|
| 306 | /*
|
|---|
| 307 | * Update the old top frame if any.
|
|---|
| 308 | */
|
|---|
| 309 | if (iFrame)
|
|---|
| 310 | {
|
|---|
| 311 | KPRF_TYPE(P,FRAME) pOldFrame = pFrame - 1;
|
|---|
| 312 | pOldFrame->OnTopOfStackTicks += TS - pOldFrame->OnTopOfStackStart;
|
|---|
| 313 | pOldFrame->cCalls++;
|
|---|
| 314 | }
|
|---|
| 315 |
|
|---|
| 316 | /*
|
|---|
| 317 | * Fill in the new frame.
|
|---|
| 318 | */
|
|---|
| 319 | pFrame->CurOverheadTicks = 0;
|
|---|
| 320 | pFrame->OverheadTicks = 0;
|
|---|
| 321 | pFrame->SleepTicks = 0;
|
|---|
| 322 | pFrame->OnStackStart = TS;
|
|---|
| 323 | pFrame->OnTopOfStackStart = TS;
|
|---|
| 324 | pFrame->OnTopOfStackTicks = 0;
|
|---|
| 325 | pFrame->cCalls = 0;
|
|---|
| 326 | pFrame->uFramePtr = uFramePtr;
|
|---|
| 327 |
|
|---|
| 328 | /*
|
|---|
| 329 | * Find the relevant function.
|
|---|
| 330 | */
|
|---|
| 331 | KPRF_TYPE(P,FUNC) pFunc = KPRF_NAME(GetFunction)(pHdr, uPC);
|
|---|
| 332 | if (pFunc)
|
|---|
| 333 | {
|
|---|
| 334 | pFrame->offFunction = KPRF_PTR2OFF(pFunc, pHdr);
|
|---|
| 335 | pFunc->cOnStack++;
|
|---|
| 336 | }
|
|---|
| 337 | else
|
|---|
| 338 | pFrame->offFunction = 0;
|
|---|
| 339 |
|
|---|
| 340 | /*
|
|---|
| 341 | * Nearly done, We only have to reactivate the thread and account overhead.
|
|---|
| 342 | * The latter is delegated to the caller.
|
|---|
| 343 | */
|
|---|
| 344 | pThread->enmState = KPRF_TYPE(,THREADSTATE_ACTIVE);
|
|---|
| 345 | return &pFrame->CurOverheadTicks;
|
|---|
| 346 | }
|
|---|
| 347 |
|
|---|
| 348 |
|
|---|
| 349 | /**
|
|---|
| 350 | * Leave function.
|
|---|
| 351 | *
|
|---|
| 352 | * @returns Where to account overhead.
|
|---|
| 353 | * @returns NULL if profiling is inactive.
|
|---|
| 354 | *
|
|---|
| 355 | * @param uPC The program counter register.
|
|---|
| 356 | * @param uFramePtr The stack frame address. This must match the one passed to kPrfEnter.
|
|---|
| 357 | * @param TS The timestamp when we entered into the profiler.
|
|---|
| 358 | * This must not be modified because the caller could be using it!
|
|---|
| 359 | * @internal
|
|---|
| 360 | */
|
|---|
| 361 | KPRF_DECL_FUNC(KU64 *, Leave)(KPRF_TYPE(,UPTR) uPC, KPRF_TYPE(,UPTR) uFramePtr, const KU64 TS)
|
|---|
| 362 | {
|
|---|
| 363 | /*
|
|---|
| 364 | * Is profiling active ?
|
|---|
| 365 | */
|
|---|
| 366 | if (!KPRF_IS_ACTIVE())
|
|---|
| 367 | return NULL;
|
|---|
| 368 |
|
|---|
| 369 | /*
|
|---|
| 370 | * Get the header and adjust input addresses.
|
|---|
| 371 | */
|
|---|
| 372 | KPRF_TYPE(P,HDR) pHdr = KPRF_GET_HDR();
|
|---|
| 373 | if (!pHdr)
|
|---|
| 374 | return NULL;
|
|---|
| 375 | const KPRF_TYPE(,UPTR) uBasePtr = pHdr->uBasePtr;
|
|---|
| 376 | if (uBasePtr)
|
|---|
| 377 | {
|
|---|
| 378 | uFramePtr -= uBasePtr;
|
|---|
| 379 | uPC -= uBasePtr;
|
|---|
| 380 | }
|
|---|
| 381 |
|
|---|
| 382 | /*
|
|---|
| 383 | * Get the current thread and suspend profiling of the thread until we leave this function.
|
|---|
| 384 | * Also reject threads which aren't active in some way.
|
|---|
| 385 | */
|
|---|
| 386 | KPRF_TYPE(P,THREAD) pThread = KPRF_GET_THREAD();
|
|---|
| 387 | if (!pThread)
|
|---|
| 388 | return NULL;
|
|---|
| 389 | KPRF_TYPE(,THREADSTATE) enmThreadState = pThread->enmState;
|
|---|
| 390 | if ( enmThreadState != KPRF_TYPE(,THREADSTATE_ACTIVE)
|
|---|
| 391 | && enmThreadState != KPRF_TYPE(,THREADSTATE_OVERFLOWED)
|
|---|
| 392 | )
|
|---|
| 393 | return NULL;
|
|---|
| 394 | KPRF_TYPE(P,STACK) pStack = KPRF_OFF2PTR(P,STACK, pThread->offStack, pHdr);
|
|---|
| 395 | if (!pStack->cFrames)
|
|---|
| 396 | return NULL;
|
|---|
| 397 | pThread->enmState = KPRF_TYPE(,THREADSTATE_SUSPENDED);
|
|---|
| 398 |
|
|---|
| 399 | /*
|
|---|
| 400 | * Unwind the stack down to and including the entry indicated by uFramePtr.
|
|---|
| 401 | * Leave it to the caller to update the overhead.
|
|---|
| 402 | */
|
|---|
| 403 | KU64 *pCurOverheadTicks = KPRF_NAME(UnwindInt)(pHdr, pStack, uPC, uFramePtr, TS);
|
|---|
| 404 |
|
|---|
| 405 | pThread->enmState = enmThreadState;
|
|---|
| 406 | return pCurOverheadTicks;
|
|---|
| 407 | }
|
|---|
| 408 |
|
|---|
| 409 |
|
|---|
| 410 | /**
|
|---|
| 411 | * Register the current thread.
|
|---|
| 412 | *
|
|---|
| 413 | * A thread can only be profiled if it has been registered by a call to this function.
|
|---|
| 414 | *
|
|---|
| 415 | * @param uPC The program counter register.
|
|---|
| 416 | * @param uStackBasePtr The base of the stack.
|
|---|
| 417 | */
|
|---|
| 418 | KPRF_DECL_FUNC(KPRF_TYPE(P,THREAD), RegisterThread)(KPRF_TYPE(,UPTR) uStackBasePtr, const char *pszName)
|
|---|
| 419 | {
|
|---|
| 420 | /*
|
|---|
| 421 | * Get the header and adjust input address.
|
|---|
| 422 | * (It doesn't matter whether we're active or not.)
|
|---|
| 423 | */
|
|---|
| 424 | KPRF_TYPE(P,HDR) pHdr = KPRF_GET_HDR();
|
|---|
| 425 | if (!pHdr)
|
|---|
| 426 | return NULL;
|
|---|
| 427 | const KPRF_TYPE(,UPTR) uBasePtr = pHdr->uBasePtr;
|
|---|
| 428 | if (uBasePtr)
|
|---|
| 429 | uStackBasePtr -= uBasePtr;
|
|---|
| 430 |
|
|---|
| 431 |
|
|---|
| 432 | /*
|
|---|
| 433 | * Allocate a thread and a stack.
|
|---|
| 434 | */
|
|---|
| 435 | KPRF_THREADS_LOCK();
|
|---|
| 436 | if (pHdr->cThreads < pHdr->cMaxThreads)
|
|---|
| 437 | {
|
|---|
| 438 | KPRF_TYPE(P,STACK) pStack = KPRF_OFF2PTR(P,STACK, pHdr->offStacks, pHdr);
|
|---|
| 439 | KU32 cLeft = pHdr->cMaxStacks;
|
|---|
| 440 | do
|
|---|
| 441 | {
|
|---|
| 442 | if (!pStack->offThread)
|
|---|
| 443 | {
|
|---|
| 444 | /* init the stack. */
|
|---|
| 445 | pStack->cFrames = 0;
|
|---|
| 446 | pStack->offThread = pHdr->offThreads + pHdr->cbThread * pHdr->cThreads++;
|
|---|
| 447 | pHdr->cStacks++;
|
|---|
| 448 |
|
|---|
| 449 | /* init the thread */
|
|---|
| 450 | KPRF_TYPE(P,THREAD) pThread = KPRF_OFF2PTR(P,THREAD, pStack->offThread, pHdr);
|
|---|
| 451 | pThread->ThreadId = KPRF_GET_THREADID();
|
|---|
| 452 | unsigned i = 0;
|
|---|
| 453 | if (pszName)
|
|---|
| 454 | while (i < sizeof(pThread->szName) - 1 && *pszName)
|
|---|
| 455 | pThread->szName[i++] = *pszName++;
|
|---|
| 456 | while (i < sizeof(pThread->szName))
|
|---|
| 457 | pThread->szName[i++] = '\0';
|
|---|
| 458 | pThread->enmState = KPRF_TYPE(,THREADSTATE_SUSPENDED);
|
|---|
| 459 | pThread->Reserved0 = KPRF_TYPE(,THREADSTATE_TERMINATED);
|
|---|
| 460 | pThread->uStackBasePtr = uStackBasePtr;
|
|---|
| 461 | pThread->cbMaxStack = 0;
|
|---|
| 462 | pThread->cCalls = 0;
|
|---|
| 463 | pThread->cOverflows = 0;
|
|---|
| 464 | pThread->cStackSwitchRejects = 0;
|
|---|
| 465 | pThread->cUnwinds = 0;
|
|---|
| 466 | pThread->ProfiledTicks = 0;
|
|---|
| 467 | pThread->OverheadTicks = 0;
|
|---|
| 468 | pThread->SleepTicks = 0;
|
|---|
| 469 | pThread->offStack = KPRF_PTR2OFF(pStack, pHdr);
|
|---|
| 470 |
|
|---|
| 471 |
|
|---|
| 472 | /* set the thread and make it active. */
|
|---|
| 473 | KPRF_THREADS_UNLOCK();
|
|---|
| 474 | KPRF_SET_THREAD(pThread);
|
|---|
| 475 | pThread->enmState = KPRF_TYPE(,THREADSTATE_ACTIVE);
|
|---|
| 476 | return pThread;
|
|---|
| 477 | }
|
|---|
| 478 |
|
|---|
| 479 | /* next */
|
|---|
| 480 | pStack = KPRF_TYPE(P,STACK)(((KPRF_TYPE(,UPTR))pStack + pHdr->cbStack));
|
|---|
| 481 | } while (--cLeft > 0);
|
|---|
| 482 | }
|
|---|
| 483 |
|
|---|
| 484 | KPRF_THREADS_UNLOCK();
|
|---|
| 485 | return NULL;
|
|---|
| 486 | }
|
|---|
| 487 |
|
|---|
| 488 |
|
|---|
| 489 | /**
|
|---|
| 490 | * Terminates a thread.
|
|---|
| 491 | *
|
|---|
| 492 | * To terminate the current thread use DeregisterThread(), because that
|
|---|
| 493 | * cleans up the TLS entry too.
|
|---|
| 494 | *
|
|---|
| 495 | * @param pHdr The profiler data set header.
|
|---|
| 496 | * @param pThread The thread to terminate.
|
|---|
| 497 | * @param TS The timestamp to use when terminating the thread.
|
|---|
| 498 | */
|
|---|
| 499 | KPRF_DECL_FUNC(void, TerminateThread)(KPRF_TYPE(P,HDR) pHdr, KPRF_TYPE(P,THREAD) pThread, KU64 TS)
|
|---|
| 500 | {
|
|---|
| 501 | if (pThread->enmState == KPRF_TYPE(,THREADSTATE_TERMINATED))
|
|---|
| 502 | return;
|
|---|
| 503 | pThread->enmState = KPRF_TYPE(,THREADSTATE_TERMINATED);
|
|---|
| 504 |
|
|---|
| 505 | /*
|
|---|
| 506 | * Unwind the entire stack.
|
|---|
| 507 | */
|
|---|
| 508 | if (pThread->offStack)
|
|---|
| 509 | {
|
|---|
| 510 | KPRF_TYPE(P,STACK) pStack = KPRF_OFF2PTR(P,STACK, pThread->offStack, pHdr);
|
|---|
| 511 | for (KU32 cFrames = pStack->cFrames; cFrames > 0; cFrames--)
|
|---|
| 512 | KPRF_NAME(UnwindOne)(pHdr, pStack, 0, TS);
|
|---|
| 513 |
|
|---|
| 514 | /*
|
|---|
| 515 | * Free the stack.
|
|---|
| 516 | */
|
|---|
| 517 | pThread->offStack = 0;
|
|---|
| 518 | KPRF_THREADS_LOCK();
|
|---|
| 519 | pStack->offThread = 0;
|
|---|
| 520 | pHdr->cStacks--;
|
|---|
| 521 | KPRF_THREADS_UNLOCK();
|
|---|
| 522 | }
|
|---|
| 523 | }
|
|---|
| 524 |
|
|---|
| 525 |
|
|---|
| 526 | /**
|
|---|
| 527 | * Deregister (terminate) the current thread.
|
|---|
| 528 | */
|
|---|
| 529 | KPRF_DECL_FUNC(void, DeregisterThread)(void)
|
|---|
| 530 | {
|
|---|
| 531 | KU64 TS = KPRF_NOW();
|
|---|
| 532 |
|
|---|
| 533 | /*
|
|---|
| 534 | * Get the header, then get the thread and mark it terminated.
|
|---|
| 535 | * (It doesn't matter whether we're active or not.)
|
|---|
| 536 | */
|
|---|
| 537 | KPRF_TYPE(P,HDR) pHdr = KPRF_GET_HDR();
|
|---|
| 538 | if (!pHdr)
|
|---|
| 539 | return;
|
|---|
| 540 |
|
|---|
| 541 | KPRF_TYPE(P,THREAD) pThread = KPRF_GET_THREAD();
|
|---|
| 542 | KPRF_SET_THREAD(NULL);
|
|---|
| 543 | if (!pThread)
|
|---|
| 544 | return;
|
|---|
| 545 | KPRF_NAME(TerminateThread)(pHdr, pThread, TS);
|
|---|
| 546 | }
|
|---|
| 547 |
|
|---|
| 548 |
|
|---|
| 549 | /**
|
|---|
| 550 | * Resumes / restarts a thread.
|
|---|
| 551 | *
|
|---|
| 552 | * @param fReset If set the stack is reset.
|
|---|
| 553 | */
|
|---|
| 554 | KPRF_DECL_FUNC(void, ResumeThread)(int fReset)
|
|---|
| 555 | {
|
|---|
| 556 | KU64 TS = KPRF_NOW();
|
|---|
| 557 |
|
|---|
| 558 | /*
|
|---|
| 559 | * Get the header, then get the thread and mark it terminated.
|
|---|
| 560 | * (It doesn't matter whether we're active or not.)
|
|---|
| 561 | */
|
|---|
| 562 | KPRF_TYPE(P,HDR) pHdr = KPRF_GET_HDR();
|
|---|
| 563 | if (!pHdr)
|
|---|
| 564 | return;
|
|---|
| 565 |
|
|---|
| 566 | KPRF_TYPE(P,THREAD) pThread = KPRF_GET_THREAD();
|
|---|
| 567 | if (!pThread)
|
|---|
| 568 | return;
|
|---|
| 569 | if (pThread->enmState != KPRF_TYPE(,THREADSTATE_SUSPENDED))
|
|---|
| 570 | return;
|
|---|
| 571 |
|
|---|
| 572 | /*
|
|---|
| 573 | * Reset (unwind) the stack?
|
|---|
| 574 | */
|
|---|
| 575 | KPRF_TYPE(P,STACK) pStack = KPRF_OFF2PTR(P,STACK, pThread->offStack, pHdr);
|
|---|
| 576 | if (fReset)
|
|---|
| 577 | {
|
|---|
| 578 | KU32 cFrames = pStack->cFrames;
|
|---|
| 579 | while (cFrames-- > 0)
|
|---|
| 580 | KPRF_NAME(UnwindOne)(pHdr, pStack, 0, TS);
|
|---|
| 581 | }
|
|---|
| 582 | /*
|
|---|
| 583 | * If we've got any thing on the stack, we'll have to stop the sleeping period.
|
|---|
| 584 | */
|
|---|
| 585 | else if (pStack->cFrames > 0)
|
|---|
| 586 | {
|
|---|
| 587 | KPRF_TYPE(P,FRAME) pFrame = &pStack->aFrames[pStack->cFrames - 1];
|
|---|
| 588 |
|
|---|
| 589 | /* update the sleeping time and set the start of the new top-of-stack period. */
|
|---|
| 590 | pFrame->SleepTicks += TS - pFrame->OnTopOfStackStart;
|
|---|
| 591 | pFrame->OnTopOfStackStart = TS;
|
|---|
| 592 | }
|
|---|
| 593 | /** @todo we're not accounting overhead here! */
|
|---|
| 594 |
|
|---|
| 595 | /*
|
|---|
| 596 | * We're done, switch the thread to active state.
|
|---|
| 597 | */
|
|---|
| 598 | pThread->enmState = KPRF_TYPE(,THREADSTATE_ACTIVE);
|
|---|
| 599 | }
|
|---|
| 600 |
|
|---|
| 601 |
|
|---|
| 602 | /**
|
|---|
| 603 | * Suspend / completes a thread.
|
|---|
| 604 | *
|
|---|
| 605 | * The thread will be in a suspend state where the time will be accounted for as sleeping.
|
|---|
| 606 | *
|
|---|
| 607 | * @param fUnwind If set the stack is unwound and the thread statistics updated.
|
|---|
| 608 | */
|
|---|
| 609 | KPRF_DECL_FUNC(void, SuspendThread)(int fUnwind)
|
|---|
| 610 | {
|
|---|
| 611 | KU64 TS = KPRF_NOW();
|
|---|
| 612 |
|
|---|
| 613 | /*
|
|---|
| 614 | * Get the header, then get the thread and mark it terminated.
|
|---|
| 615 | * (It doesn't matter whether we're active or not.)
|
|---|
| 616 | */
|
|---|
| 617 | KPRF_TYPE(P,HDR) pHdr = KPRF_GET_HDR();
|
|---|
| 618 | if (!pHdr)
|
|---|
| 619 | return;
|
|---|
| 620 |
|
|---|
| 621 | KPRF_TYPE(P,THREAD) pThread = KPRF_GET_THREAD();
|
|---|
| 622 | if (!pThread)
|
|---|
| 623 | return;
|
|---|
| 624 | if ( pThread->enmState != KPRF_TYPE(,THREADSTATE_ACTIVE)
|
|---|
| 625 | && pThread->enmState != KPRF_TYPE(,THREADSTATE_OVERFLOWED)
|
|---|
| 626 | && (pThread->enmState != KPRF_TYPE(,THREADSTATE_SUSPENDED) || fUnwind))
|
|---|
| 627 | return;
|
|---|
| 628 |
|
|---|
| 629 | pThread->enmState = KPRF_TYPE(,THREADSTATE_SUSPENDED);
|
|---|
| 630 |
|
|---|
| 631 | /*
|
|---|
| 632 | * Unwind the stack?
|
|---|
| 633 | */
|
|---|
| 634 | KPRF_TYPE(P,STACK) pStack = KPRF_OFF2PTR(P,STACK, pThread->offStack, pHdr);
|
|---|
| 635 | if (fUnwind)
|
|---|
| 636 | {
|
|---|
| 637 | KU32 cFrames = pStack->cFrames;
|
|---|
| 638 | while (cFrames-- > 0)
|
|---|
| 639 | KPRF_NAME(UnwindOne)(pHdr, pStack, 0, TS);
|
|---|
| 640 | }
|
|---|
| 641 | /*
|
|---|
| 642 | * If we've got any thing on the stack, we'll have to record the sleeping period
|
|---|
| 643 | * of the thread. If not we'll ignore it (for now at least).
|
|---|
| 644 | */
|
|---|
| 645 | else if (pStack->cFrames > 0)
|
|---|
| 646 | {
|
|---|
| 647 | KPRF_TYPE(P,FRAME) pFrame = &pStack->aFrames[pStack->cFrames - 1];
|
|---|
| 648 |
|
|---|
| 649 | /* update the top of stack time and set the start of the sleep period. */
|
|---|
| 650 | pFrame->OnTopOfStackTicks += TS - pFrame->OnTopOfStackStart;
|
|---|
| 651 | pFrame->OnTopOfStackStart = TS;
|
|---|
| 652 | }
|
|---|
| 653 |
|
|---|
| 654 | /** @todo we're not accounting overhead here! */
|
|---|
| 655 | }
|
|---|
| 656 |
|
|---|
| 657 |
|
|---|