RTDECL(uint64_t) RTTimeSystemNanoTS(void)
{
return RTTimeNanoTS();
}
RTDECL(int) RTFileAioCtxWait(RTFILEAIOCTX hAioCtx, size_t cMinReqs, RTMSINTERVAL cMillies,
PRTFILEAIOREQ pahReqs, size_t cReqs, uint32_t *pcReqs)
{
/*
* Validate the parameters, making sure to always set pcReqs.
*/
AssertPtrReturn(pcReqs, VERR_INVALID_POINTER);
*pcReqs = 0; /* always set */
PRTFILEAIOCTXINTERNAL pCtxInt = hAioCtx;
RTFILEAIOCTX_VALID_RETURN(pCtxInt);
AssertPtrReturn(pahReqs, VERR_INVALID_POINTER);
AssertReturn(cReqs != 0, VERR_INVALID_PARAMETER);
AssertReturn(cReqs >= cMinReqs, VERR_OUT_OF_RANGE);
/*
* Can't wait if there are not requests around.
*/
if (RT_UNLIKELY(ASMAtomicUoReadS32(&pCtxInt->cRequests) == 0))
return VERR_FILE_AIO_NO_REQUEST;
/*
* Convert the timeout if specified.
*/
struct timespec *pTimeout = NULL;
struct timespec Timeout = {0,0};
uint64_t StartNanoTS = 0;
if (cMillies != RT_INDEFINITE_WAIT)
{
Timeout.tv_sec = cMillies / 1000;
Timeout.tv_nsec = cMillies % 1000 * 1000000;
pTimeout = &Timeout;
StartNanoTS = RTTimeNanoTS();
}
/* Wait for at least one. */
if (!cMinReqs)
cMinReqs = 1;
/* For the wakeup call. */
Assert(pCtxInt->hThreadWait == NIL_RTTHREAD);
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, RTThreadSelf());
/*
* Loop until we're woken up, hit an error (incl timeout), or
* have collected the desired number of requests.
*/
int rc = VINF_SUCCESS;
int cRequestsCompleted = 0;
while (!pCtxInt->fWokenUp)
{
LNXKAIOIOEVENT aPortEvents[AIO_MAXIMUM_REQUESTS_PER_CONTEXT];
int cRequestsToWait = RT_MIN(cReqs, AIO_MAXIMUM_REQUESTS_PER_CONTEXT);
ASMAtomicXchgBool(&pCtxInt->fWaiting, true);
rc = rtFileAsyncIoLinuxGetEvents(pCtxInt->AioContext, cMinReqs, cRequestsToWait, &aPortEvents[0], pTimeout);
ASMAtomicXchgBool(&pCtxInt->fWaiting, false);
if (RT_FAILURE(rc))
break;
uint32_t const cDone = rc;
rc = VINF_SUCCESS;
/*
* Process received events / requests.
*/
for (uint32_t i = 0; i < cDone; i++)
{
/*
* The iocb is the first element in our request structure.
* So we can safely cast it directly to the handle (see above)
*/
PRTFILEAIOREQINTERNAL pReqInt = (PRTFILEAIOREQINTERNAL)aPortEvents[i].pIoCB;
AssertPtr(pReqInt);
Assert(pReqInt->u32Magic == RTFILEAIOREQ_MAGIC);
/** @todo aeichner: The rc field contains the result code
* like you can find in errno for the normal read/write ops.
* But there is a second field called rc2. I don't know the
* purpose for it yet.
*/
if (RT_UNLIKELY(aPortEvents[i].rc < 0))
pReqInt->Rc = RTErrConvertFromErrno(-aPortEvents[i].rc); /* Convert to positive value. */
else
{
pReqInt->Rc = VINF_SUCCESS;
pReqInt->cbTransfered = aPortEvents[i].rc;
}
/* Mark the request as finished. */
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
pahReqs[cRequestsCompleted++] = (RTFILEAIOREQ)pReqInt;
}
/*
* Done Yet? If not advance and try again.
*/
if (cDone >= cMinReqs)
break;
cMinReqs -= cDone;
cReqs -= cDone;
if (cMillies != RT_INDEFINITE_WAIT)
{
/* The API doesn't return ETIMEDOUT, so we have to fix that ourselves. */
uint64_t NanoTS = RTTimeNanoTS();
uint64_t cMilliesElapsed = (NanoTS - StartNanoTS) / 1000000;
if (cMilliesElapsed >= cMillies)
{
rc = VERR_TIMEOUT;
break;
}
/* The syscall supposedly updates it, but we're paranoid. :-) */
Timeout.tv_sec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) / 1000;
Timeout.tv_nsec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) % 1000 * 1000000;
}
}
/*
* Update the context state and set the return value.
*/
*pcReqs = cRequestsCompleted;
ASMAtomicSubS32(&pCtxInt->cRequests, cRequestsCompleted);
Assert(pCtxInt->hThreadWait == RTThreadSelf());
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, NIL_RTTHREAD);
/*
* Clear the wakeup flag and set rc.
*/
if ( pCtxInt->fWokenUp
&& RT_SUCCESS(rc))
{
ASMAtomicXchgBool(&pCtxInt->fWokenUp, false);
rc = VERR_INTERRUPTED;
}
return rc;
}
RTDECL(uint64_t) RTTimeMilliTS(void)
{
return RTTimeNanoTS() / 1000000;
}
int main()
{
/*
* Init runtime
*/
unsigned cErrors = 0;
int rc = RTR3Init();
if (RT_FAILURE(rc))
{
RTPrintf("tstTimer: RTR3Init() -> %d\n", rc);
return 1;
}
/*
* Check that the clock is reliable.
*/
RTPrintf("tstTimer: TESTING - RTTimeNanoTS() for 2sec\n");
uint64_t uTSMillies = RTTimeMilliTS();
uint64_t uTSBegin = RTTimeNanoTS();
uint64_t uTSLast = uTSBegin;
uint64_t uTSDiff;
uint64_t cIterations = 0;
do
{
uint64_t uTS = RTTimeNanoTS();
if (uTS < uTSLast)
{
RTPrintf("tstTimer: FAILURE - RTTimeNanoTS() is unreliable. uTS=%RU64 uTSLast=%RU64\n", uTS, uTSLast);
cErrors++;
}
if (++cIterations > (2*1000*1000*1000))
{
RTPrintf("tstTimer: FAILURE - RTTimeNanoTS() is unreliable. cIterations=%RU64 uTS=%RU64 uTSBegin=%RU64\n", cIterations, uTS, uTSBegin);
return 1;
}
uTSLast = uTS;
uTSDiff = uTSLast - uTSBegin;
} while (uTSDiff < (2*1000*1000*1000));
uTSMillies = RTTimeMilliTS() - uTSMillies;
if (uTSMillies >= 2500 || uTSMillies <= 1500)
{
RTPrintf("tstTimer: FAILURE - uTSMillies=%RI64 uTSBegin=%RU64 uTSLast=%RU64 uTSDiff=%RU64\n",
uTSMillies, uTSBegin, uTSLast, uTSDiff);
cErrors++;
}
if (!cErrors)
RTPrintf("tstTimer: OK - RTTimeNanoTS()\n");
/*
* Tests.
*/
static struct
{
unsigned uMicroInterval;
unsigned uMilliesWait;
unsigned cLower;
unsigned cUpper;
} aTests[] =
{
{ 32000, 2000, 0, 0 },
{ 20000, 2000, 0, 0 },
{ 10000, 2000, 0, 0 },
{ 8000, 2000, 0, 0 },
{ 2000, 2000, 0, 0 },
{ 1000, 2000, 0, 0 },
{ 500, 5000, 0, 0 },
{ 200, 5000, 0, 0 },
{ 100, 5000, 0, 0 }
};
unsigned i = 0;
for (i = 0; i < RT_ELEMENTS(aTests); i++)
{
aTests[i].cLower = (aTests[i].uMilliesWait*1000 - aTests[i].uMilliesWait*100) / aTests[i].uMicroInterval;
aTests[i].cUpper = (aTests[i].uMilliesWait*1000 + aTests[i].uMilliesWait*100) / aTests[i].uMicroInterval;
gu64Norm = aTests[i].uMicroInterval*1000;
RTPrintf("\n"
"tstTimer: TESTING - %d us interval, %d ms wait, expects %d-%d ticks.\n",
aTests[i].uMicroInterval, aTests[i].uMilliesWait, aTests[i].cLower, aTests[i].cUpper);
/*
* Start timer which ticks every 10ms.
*/
gcTicks = 0;
PRTTIMER pTimer;
gu64Max = 0;
gu64Min = UINT64_MAX;
gu64Prev = 0;
RT_ZERO(cFrequency);
#ifdef RT_OS_WINDOWS
if (aTests[i].uMicroInterval < 1000)
continue;
rc = RTTimerCreate(&pTimer, aTests[i].uMicroInterval / 1000, TimerCallback, NULL);
#else
rc = RTTimerCreateEx(&pTimer, aTests[i].uMicroInterval * (uint64_t)1000, 0, TimerCallback, NULL);
#endif
if (RT_FAILURE(rc))
{
RTPrintf("tstTimer: FAILURE - RTTimerCreateEx(,%u*1M,,,) -> %Rrc\n", aTests[i].uMicroInterval, rc);
cErrors++;
continue;
}
/*
* Start the timer and active waiting for the requested test period.
*/
uTSBegin = RTTimeNanoTS();
#ifndef RT_OS_WINDOWS
rc = RTTimerStart(pTimer, 0);
if (RT_FAILURE(rc))
{
RTPrintf("tstTimer: FAILURE - RTTimerStart(,0) -> %Rrc\n", aTests[i].uMicroInterval, rc);
cErrors++;
}
#endif
while (RTTimeNanoTS() - uTSBegin < (uint64_t)aTests[i].uMilliesWait * 1000000)
/* nothing */;
/* destroy the timer */
uint64_t uTSEnd = RTTimeNanoTS();
uTSDiff = uTSEnd - uTSBegin;
rc = RTTimerDestroy(pTimer);
if (RT_FAILURE(rc))
{
RTPrintf("tstTimer: FAILURE - RTTimerDestroy() -> %d gcTicks=%d\n", rc, gcTicks);
cErrors++;
}
RTPrintf("tstTimer: uTS=%RI64 (%RU64 - %RU64)\n", uTSDiff, uTSBegin, uTSEnd);
unsigned cTicks = gcTicks;
RTThreadSleep(aTests[i].uMicroInterval/1000 * 3);
if (gcTicks != cTicks)
{
RTPrintf("tstTimer: FAILURE - RTTimerDestroy() didn't really stop the timer! gcTicks=%d cTicks=%d\n", gcTicks, cTicks);
cErrors++;
continue;
}
/*
* Check the number of ticks.
*/
if (gcTicks < aTests[i].cLower)
{
RTPrintf("tstTimer: FAILURE - Too few ticks gcTicks=%d (expected %d-%d)", gcTicks, aTests[i].cUpper, aTests[i].cLower);
cErrors++;
}
else if (gcTicks > aTests[i].cUpper)
{
RTPrintf("tstTimer: FAILURE - Too many ticks gcTicks=%d (expected %d-%d)", gcTicks, aTests[i].cUpper, aTests[i].cLower);
cErrors++;
}
else
RTPrintf("tstTimer: OK - gcTicks=%d", gcTicks);
RTPrintf(" min=%RU64 max=%RU64\n", gu64Min, gu64Max);
for (int j = 0; j < (int)RT_ELEMENTS(cFrequency); j++)
{
uint32_t len = cFrequency[j] * 70 / gcTicks;
uint32_t deviation = j - RT_ELEMENTS(cFrequency) / 2;
uint64_t u64FreqPercent = (uint64_t)cFrequency[j] * 10000 / gcTicks;
uint64_t u64FreqPercentFrac = u64FreqPercent % 100;
u64FreqPercent = u64FreqPercent / 100;
RTPrintf("%+4d%c %6u %3llu.%02llu%% ",
deviation, deviation == 0 ? ' ' : '%', cFrequency[j],
u64FreqPercent, u64FreqPercentFrac);
for (unsigned k = 0; k < len; k++)
RTPrintf("*");
RTPrintf("\n");
}
}
/*
* Summary.
*/
if (!cErrors)
RTPrintf("tstTimer: SUCCESS\n");
else
RTPrintf("tstTimer: FAILURE %d errors\n", cErrors);
return !!cErrors;
}
RTDECL(uint64_t) RTTimeSystemMilliTS(void)
{
return RTTimeNanoTS() / RT_NS_1MS;
}
int main()
{
/*
* Init.
*/
RTTEST hTest;
RTEXITCODE rcExit = RTTestInitExAndCreate(0, NULL, RTR3INIT_FLAGS_SUPLIB, "tstRTTime", &hTest);
if (rcExit != RTEXITCODE_SUCCESS)
return rcExit;
RTTestBanner(hTest);
/*
* RTNanoTimeTS() shall never return something which
* is less or equal to the return of the previous call.
*/
RTTimeSystemNanoTS(); RTTimeNanoTS(); RTThreadYield();
uint64_t u64RTStartTS = RTTimeNanoTS();
uint64_t u64OSStartTS = RTTimeSystemNanoTS();
uint32_t i;
uint64_t u64Prev = RTTimeNanoTS();
for (i = 0; i < 100*_1M; i++)
{
uint64_t u64 = RTTimeNanoTS();
if (u64 <= u64Prev)
{
/** @todo wrapping detection. */
RTTestFailed(hTest, "i=%#010x u64=%#llx u64Prev=%#llx (1)\n", i, u64, u64Prev);
if (RTTestErrorCount(hTest) >= 256)
break;
RTThreadYield();
u64 = RTTimeNanoTS();
}
else if (u64 - u64Prev > 1000000000 /* 1sec */)
{
RTTestFailed(hTest, "i=%#010x u64=%#llx u64Prev=%#llx delta=%lld\n", i, u64, u64Prev, u64 - u64Prev);
if (RTTestErrorCount(hTest) >= 256)
break;
RTThreadYield();
u64 = RTTimeNanoTS();
}
if (!(i & (_1M*2 - 1)))
{
RTTestPrintf(hTest, RTTESTLVL_INFO, "i=%#010x u64=%#llx u64Prev=%#llx delta=%lld\n", i, u64, u64Prev, u64 - u64Prev);
RTThreadYield();
u64 = RTTimeNanoTS();
}
u64Prev = u64;
}
RTTimeSystemNanoTS(); RTTimeNanoTS(); RTThreadYield();
uint64_t u64RTElapsedTS = RTTimeNanoTS();
uint64_t u64OSElapsedTS = RTTimeSystemNanoTS();
u64RTElapsedTS -= u64RTStartTS;
u64OSElapsedTS -= u64OSStartTS;
int64_t i64Diff = u64OSElapsedTS >= u64RTElapsedTS ? u64OSElapsedTS - u64RTElapsedTS : u64RTElapsedTS - u64OSElapsedTS;
if (i64Diff > (int64_t)(u64OSElapsedTS / 1000))
RTTestFailed(hTest, "total time differs too much! u64OSElapsedTS=%#llx u64RTElapsedTS=%#llx delta=%lld\n",
u64OSElapsedTS, u64RTElapsedTS, u64OSElapsedTS - u64RTElapsedTS);
else
{
if (u64OSElapsedTS >= u64RTElapsedTS)
RTTestValue(hTest, "Total time delta", u64OSElapsedTS - u64RTElapsedTS, RTTESTUNIT_NS);
else
RTTestValue(hTest, "Total time delta", u64RTElapsedTS - u64OSElapsedTS, RTTESTUNIT_NS);
RTTestPrintf(hTest, RTTESTLVL_INFO, "total time difference: u64OSElapsedTS=%#llx u64RTElapsedTS=%#llx delta=%lld\n",
u64OSElapsedTS, u64RTElapsedTS, u64OSElapsedTS - u64RTElapsedTS);
}
#if defined(RT_ARCH_AMD64) || defined(RT_ARCH_X86) /** @todo This isn't really x86 or AMD64 specific... */
RTTestValue(hTest, "RTTimeDbgSteps", RTTimeDbgSteps(), RTTESTUNIT_OCCURRENCES);
RTTestValue(hTest, "RTTimeDbgSteps pp", ((uint64_t)RTTimeDbgSteps() * 1000) / i, RTTESTUNIT_PP1K);
RTTestValue(hTest, "RTTimeDbgExpired", RTTimeDbgExpired(), RTTESTUNIT_OCCURRENCES);
RTTestValue(hTest, "RTTimeDbgExpired pp", ((uint64_t)RTTimeDbgExpired() * 1000) / i, RTTESTUNIT_PP1K);
RTTestValue(hTest, "RTTimeDbgBad", RTTimeDbgBad(), RTTESTUNIT_OCCURRENCES);
RTTestValue(hTest, "RTTimeDbgBad pp", ((uint64_t)RTTimeDbgBad() * 1000) / i, RTTESTUNIT_PP1K);
RTTestValue(hTest, "RTTimeDbgRaces", RTTimeDbgRaces(), RTTESTUNIT_OCCURRENCES);
RTTestValue(hTest, "RTTimeDbgRaces pp", ((uint64_t)RTTimeDbgRaces() * 1000) / i, RTTESTUNIT_PP1K);
#endif
return RTTestSummaryAndDestroy(hTest);
}
/**
* Receive thread.
* This is reading stuff from the network.
*/
DECLCALLBACK(int) ReceiveThread(RTTHREAD hThreadSelf, void *pvArg)
{
uint32_t cbReceived = 0;
uint32_t cLostFrames = 0;
uint32_t iFrame = UINT32_MAX;
PMYARGS pArgs = (PMYARGS)pvArg;
NOREF(hThreadSelf);
for (;;)
{
/*
* Read data.
*/
while (IntNetRingHasMoreToRead(&pArgs->pBuf->Recv))
{
uint8_t abBuf[16384 + 1024];
MYFRAMEHDR *pHdr = (MYFRAMEHDR *)&abBuf[0];
uint32_t cb = IntNetRingReadAndSkipFrame(&pArgs->pBuf->Recv, abBuf);
/* check for termination frame. */
if ( pHdr->iFrame == 0xffffdead
&& pHdr->auEos[0] == 0xffffdead
&& pHdr->auEos[1] == 0xffffdead
&& pHdr->auEos[2] == 0xffffdead)
{
pArgs->u64End = RTTimeNanoTS();
RTThreadSleep(10);
RTTestPrintf(g_hTest, RTTESTLVL_ALWAYS,
"receiver thread %.6Rhxs terminating.\n"
" iFrame=%u cb=%'u c=%'u %'uKB/s %'ufps cLost=%'u \n",
&pArgs->Mac, iFrame, cbReceived, iFrame - cLostFrames,
(unsigned)(cbReceived * 1000000000.0 / 1024 / (pArgs->u64End - pArgs->u64Start)),
(unsigned)((iFrame - cLostFrames) * 1000000000.0 / (pArgs->u64End - pArgs->u64Start)),
cLostFrames);
return VINF_SUCCESS;
}
/* validate frame header */
if ( pHdr->DstMac.au16[0] != pArgs->Mac.au16[0]
|| pHdr->DstMac.au16[1] != pArgs->Mac.au16[1]
|| pHdr->DstMac.au16[2] != pArgs->Mac.au16[2]
|| pHdr->SrcMac.au16[0] != pArgs->Mac.au16[0]
|| pHdr->SrcMac.au16[1] != pArgs->Mac.au16[1]
|| pHdr->SrcMac.au16[2] != (pArgs->Mac.au16[2] + 1) % 2)
{
RTTestFailed(g_hTest, "receiver thread %.6Rhxs received frame header: %.16Rhxs\n", &pArgs->Mac, abBuf);
}
/* frame stuff and stats. */
int32_t off = pHdr->iFrame - (iFrame + 1);
if (off)
{
if (off > 0)
{
#ifndef IGNORE_LOST_FRAMES
RTTestFailed(g_hTest, "receiver thread %.6Rhxs: iFrame=%#x *puFrame=%#x off=%d\n",
&pArgs->Mac, iFrame, pHdr->iFrame, off);
#endif
cLostFrames += off;
}
else
{
cLostFrames++;
RTTestFailed(g_hTest, "receiver thread %.6Rhxs: iFrame=%#x *puFrame=%#x off=%d\n",
&pArgs->Mac, iFrame, pHdr->iFrame, off);
}
}
iFrame = pHdr->iFrame;
cbReceived += cb;
}
/*
* Wait for data.
*/
int rc = IntNetR0IfWait(pArgs->hIf, g_pSession, RT_INDEFINITE_WAIT);
switch (rc)
{
case VERR_INTERRUPTED:
case VINF_SUCCESS:
break;
case VERR_SEM_DESTROYED:
RTTestPrintf(g_hTest, RTTESTLVL_ALWAYS,
"receiver thread %.6Rhxs terminating. iFrame=%u cb=%'u c=%'u cLost=%'u\n",
&pArgs->Mac, iFrame, cbReceived, iFrame - cLostFrames, cLostFrames);
return VINF_SUCCESS;
default:
RTTestFailed(g_hTest, "receiver thread %.6Rhxs got odd return value %Rrc! iFrame=%u cb=%'u c=%'u cLost=%'u\n",
&pArgs->Mac, rc, iFrame, cbReceived, iFrame - cLostFrames, cLostFrames);
return rc;
}
}
}
static void Test4(unsigned cThreads, unsigned cSeconds, unsigned uWritePercent, bool fYield, bool fQuiet)
{
unsigned i;
uint64_t acIterations[32];
RTTHREAD aThreads[RT_ELEMENTS(acIterations)];
AssertRelease(cThreads <= RT_ELEMENTS(acIterations));
RTTestSubF(g_hTest, "Test4 - %u threads, %u sec, %u%% writes, %syielding",
cThreads, cSeconds, uWritePercent, fYield ? "" : "non-");
/*
* Init globals.
*/
g_fYield = fYield;
g_fQuiet = fQuiet;
g_fTerminate = false;
g_uWritePercent = uWritePercent;
g_cConcurrentWriters = 0;
g_cConcurrentReaders = 0;
RTTEST_CHECK_RC_RETV(g_hTest, RTSemRWCreate(&g_hSemRW), VINF_SUCCESS);
/*
* Create the threads and let them block on the semrw.
*/
RTTEST_CHECK_RC_RETV(g_hTest, RTSemRWRequestWrite(g_hSemRW, RT_INDEFINITE_WAIT), VINF_SUCCESS);
for (i = 0; i < cThreads; i++)
{
acIterations[i] = 0;
RTTEST_CHECK_RC_RETV(g_hTest, RTThreadCreateF(&aThreads[i], Test4Thread, &acIterations[i], 0,
RTTHREADTYPE_DEFAULT, RTTHREADFLAGS_WAITABLE,
"test-%u", i), VINF_SUCCESS);
}
/*
* Do the test run.
*/
uint32_t cErrorsBefore = RTTestErrorCount(g_hTest);
uint64_t u64StartTS = RTTimeNanoTS();
RTTEST_CHECK_RC(g_hTest, RTSemRWReleaseWrite(g_hSemRW), VINF_SUCCESS);
RTThreadSleep(cSeconds * 1000);
ASMAtomicWriteBool(&g_fTerminate, true);
uint64_t ElapsedNS = RTTimeNanoTS() - u64StartTS;
/*
* Clean up the threads and semaphore.
*/
for (i = 0; i < cThreads; i++)
RTTEST_CHECK_RC(g_hTest, RTThreadWait(aThreads[i], 5000, NULL), VINF_SUCCESS);
RTTEST_CHECK_MSG(g_hTest, g_cConcurrentWriters == 0, (g_hTest, "g_cConcurrentWriters=%u at end of test\n", g_cConcurrentWriters));
RTTEST_CHECK_MSG(g_hTest, g_cConcurrentReaders == 0, (g_hTest, "g_cConcurrentReaders=%u at end of test\n", g_cConcurrentReaders));
RTTEST_CHECK_RC(g_hTest, RTSemRWDestroy(g_hSemRW), VINF_SUCCESS);
g_hSemRW = NIL_RTSEMRW;
if (RTTestErrorCount(g_hTest) != cErrorsBefore)
RTThreadSleep(100);
/*
* Collect and display the results.
*/
uint64_t cItrTotal = acIterations[0];
for (i = 1; i < cThreads; i++)
cItrTotal += acIterations[i];
uint64_t cItrNormal = cItrTotal / cThreads;
uint64_t cItrMinOK = cItrNormal / 20; /* 5% */
uint64_t cItrMaxDeviation = 0;
for (i = 0; i < cThreads; i++)
{
uint64_t cItrDelta = RT_ABS((int64_t)(acIterations[i] - cItrNormal));
if (acIterations[i] < cItrMinOK)
RTTestFailed(g_hTest, "Thread %u did less than 5%% of the iterations - %llu (it) vs. %llu (5%%) - %llu%%\n",
i, acIterations[i], cItrMinOK, cItrDelta * 100 / cItrNormal);
else if (cItrDelta > cItrNormal / 2)
RTTestPrintf(g_hTest, RTTESTLVL_ALWAYS,
"Warning! Thread %u deviates by more than 50%% - %llu (it) vs. %llu (avg) - %llu%%\n",
i, acIterations[i], cItrNormal, cItrDelta * 100 / cItrNormal);
if (cItrDelta > cItrMaxDeviation)
cItrMaxDeviation = cItrDelta;
}
//RTTestPrintf(g_hTest, RTTESTLVL_ALWAYS,
// "Threads: %u Total: %llu Per Sec: %llu Avg: %llu ns Max dev: %llu%%\n",
// cThreads,
// cItrTotal,
// cItrTotal / cSeconds,
// ElapsedNS / cItrTotal,
// cItrMaxDeviation * 100 / cItrNormal
// );
//
RTTestValue(g_hTest, "Thruput", cItrTotal * UINT32_C(1000000000) / ElapsedNS, RTTESTUNIT_CALLS_PER_SEC);
RTTestValue(g_hTest, "Max diviation", cItrMaxDeviation * 100 / cItrNormal, RTTESTUNIT_PCT);
}
void tstFileAioTestReadWriteBasic(RTFILE File, bool fWrite, void *pvTestBuf,
size_t cbTestBuf, size_t cbTestFile, uint32_t cMaxReqsInFlight)
{
/* Allocate request array. */
RTFILEAIOREQ *paReqs;
paReqs = (PRTFILEAIOREQ)RTTestGuardedAllocHead(g_hTest, cMaxReqsInFlight * sizeof(RTFILEAIOREQ));
RTTESTI_CHECK_RETV(paReqs);
RT_BZERO(paReqs, sizeof(cMaxReqsInFlight * sizeof(RTFILEAIOREQ)));
/* Allocate array holding pointer to data buffers. */
void **papvBuf = (void **)RTTestGuardedAllocHead(g_hTest, cMaxReqsInFlight * sizeof(void *));
RTTESTI_CHECK_RETV(papvBuf);
/* Allocate the buffers*/
for (unsigned i = 0; i < cMaxReqsInFlight; i++)
{
RTTESTI_CHECK_RC_OK_RETV(RTTestGuardedAlloc(g_hTest, cbTestBuf, PAGE_SIZE, true /*fHead*/, &papvBuf[i]));
if (fWrite)
memcpy(papvBuf[i], pvTestBuf, cbTestBuf);
if (fWrite)
memcpy(papvBuf[i], pvTestBuf, cbTestBuf);
else
RT_BZERO(papvBuf[i], cbTestBuf);
}
/* Allocate array holding completed requests. */
RTFILEAIOREQ *paReqsCompleted;
paReqsCompleted = (PRTFILEAIOREQ)RTTestGuardedAllocHead(g_hTest, cMaxReqsInFlight * sizeof(RTFILEAIOREQ));
RTTESTI_CHECK_RETV(paReqsCompleted);
RT_BZERO(paReqsCompleted, cMaxReqsInFlight * sizeof(RTFILEAIOREQ));
/* Create a context and associate the file handle with it. */
RTFILEAIOCTX hAioContext;
RTTESTI_CHECK_RC_RETV(RTFileAioCtxCreate(&hAioContext, cMaxReqsInFlight, 0 /* fFlags */), VINF_SUCCESS);
RTTESTI_CHECK_RC_RETV(RTFileAioCtxAssociateWithFile(hAioContext, File), VINF_SUCCESS);
/* Initialize requests. */
for (unsigned i = 0; i < cMaxReqsInFlight; i++)
RTFileAioReqCreate(&paReqs[i]);
RTFOFF off = 0;
int cRuns = 0;
uint64_t NanoTS = RTTimeNanoTS();
size_t cbLeft = cbTestFile;
while (cbLeft)
{
int rc;
int cReqs = 0;
for (unsigned i = 0; i < cMaxReqsInFlight; i++)
{
size_t cbTransfer = cbLeft < cbTestBuf ? cbLeft : cbTestBuf;
if (!cbTransfer)
break;
if (fWrite)
rc = RTFileAioReqPrepareWrite(paReqs[i], File, off, papvBuf[i],
cbTransfer, papvBuf[i]);
else
rc = RTFileAioReqPrepareRead(paReqs[i], File, off, papvBuf[i],
cbTransfer, papvBuf[i]);
RTTESTI_CHECK_RC(rc, VINF_SUCCESS);
cbLeft -= cbTransfer;
off += cbTransfer;
cReqs++;
}
rc = RTFileAioCtxSubmit(hAioContext, paReqs, cReqs);
RTTESTI_CHECK_MSG(rc == VINF_SUCCESS, ("Failed to submit tasks after %d runs. rc=%Rrc\n", cRuns, rc));
if (rc != VINF_SUCCESS)
break;
/* Wait */
uint32_t cCompleted = 0;
RTTESTI_CHECK_RC(rc = RTFileAioCtxWait(hAioContext, cReqs, RT_INDEFINITE_WAIT,
paReqsCompleted, cMaxReqsInFlight, &cCompleted),
VINF_SUCCESS);
if (rc != VINF_SUCCESS)
break;
if (!fWrite)
{
for (uint32_t i = 0; i < cCompleted; i++)
{
/* Compare that we read the right stuff. */
void *pvBuf = RTFileAioReqGetUser(paReqsCompleted[i]);
RTTESTI_CHECK(pvBuf);
size_t cbTransfered;
RTTESTI_CHECK_RC(rc = RTFileAioReqGetRC(paReqsCompleted[i], &cbTransfered), VINF_SUCCESS);
if (rc != VINF_SUCCESS)
break;
RTTESTI_CHECK_MSG(cbTransfered == cbTestBuf, ("cbTransfered=%zd\n", cbTransfered));
RTTESTI_CHECK_RC_OK(rc = (memcmp(pvBuf, pvTestBuf, cbTestBuf) == 0 ? VINF_SUCCESS : VERR_BAD_EXE_FORMAT));
if (rc != VINF_SUCCESS)
break;
memset(pvBuf, 0, cbTestBuf);
}
}
cRuns++;
if (RT_FAILURE(rc))
break;
}
NanoTS = RTTimeNanoTS() - NanoTS;
uint64_t SpeedKBs = (uint64_t)(cbTestFile / (NanoTS / 1000000000.0) / 1024);
RTTestValue(g_hTest, "Throughput", SpeedKBs, RTTESTUNIT_KILOBYTES_PER_SEC);
/* cleanup */
for (unsigned i = 0; i < cMaxReqsInFlight; i++)
RTTestGuardedFree(g_hTest, papvBuf[i]);
RTTestGuardedFree(g_hTest, papvBuf);
for (unsigned i = 0; i < cMaxReqsInFlight; i++)
RTTESTI_CHECK_RC(RTFileAioReqDestroy(paReqs[i]), VINF_SUCCESS);
RTTESTI_CHECK_RC(RTFileAioCtxDestroy(hAioContext), VINF_SUCCESS);
RTTestGuardedFree(g_hTest, paReqs);
}
/**
* Initializes the guest object.
*/
HRESULT Guest::init(Console *aParent)
{
LogFlowThisFunc(("aParent=%p\n", aParent));
ComAssertRet(aParent, E_INVALIDARG);
/* Enclose the state transition NotReady->InInit->Ready */
AutoInitSpan autoInitSpan(this);
AssertReturn(autoInitSpan.isOk(), E_FAIL);
unconst(mParent) = aParent;
/* Confirm a successful initialization when it's the case */
autoInitSpan.setSucceeded();
ULONG aMemoryBalloonSize;
HRESULT hr = mParent->machine()->COMGETTER(MemoryBalloonSize)(&aMemoryBalloonSize);
if (hr == S_OK) /** @todo r=andy SUCCEEDED? */
mMemoryBalloonSize = aMemoryBalloonSize;
else
mMemoryBalloonSize = 0; /* Default is no ballooning */
BOOL fPageFusionEnabled;
hr = mParent->machine()->COMGETTER(PageFusionEnabled)(&fPageFusionEnabled);
if (hr == S_OK) /** @todo r=andy SUCCEEDED? */
mfPageFusionEnabled = fPageFusionEnabled;
else
mfPageFusionEnabled = false; /* Default is no page fusion*/
mStatUpdateInterval = 0; /* Default is not to report guest statistics at all */
mCollectVMMStats = false;
/* Clear statistics. */
mNetStatRx = mNetStatTx = 0;
mNetStatLastTs = RTTimeNanoTS();
for (unsigned i = 0 ; i < GUESTSTATTYPE_MAX; i++)
mCurrentGuestStat[i] = 0;
mVmValidStats = pm::VMSTATMASK_NONE;
mMagic = GUEST_MAGIC;
int vrc = RTTimerLRCreate(&mStatTimer, 1000 /* ms */,
&Guest::staticUpdateStats, this);
AssertMsgRC(vrc, ("Failed to create guest statistics update timer (%Rrc)\n", vrc));
#ifdef VBOX_WITH_GUEST_CONTROL
unconst(mEventSource).createObject();
Assert(!mEventSource.isNull());
hr = mEventSource->init(static_cast<IGuest*>(this));
#else
hr = S_OK;
#endif
try
{
#ifdef VBOX_WITH_DRAG_AND_DROP
m_pGuestDnD = new GuestDnD(this);
AssertPtr(m_pGuestDnD);
#endif
}
catch(std::bad_alloc &)
{
hr = E_OUTOFMEMORY;
}
return hr;
}
static int Test1(unsigned cThreads, unsigned cSeconds, bool fYield, bool fQuiet)
{
int rc;
unsigned i;
uint64_t g_au64[32];
RTTHREAD aThreads[RT_ELEMENTS(g_au64)];
AssertRelease(cThreads <= RT_ELEMENTS(g_au64));
/*
* Init globals.
*/
g_fYield = fYield;
g_fQuiet = fQuiet;
g_fTerminate = false;
rc = RTSemMutexCreate(&g_hMutex);
if (RT_FAILURE(rc))
return PrintError("RTSemMutexCreate failed (rc=%Rrc)\n", rc);
/*
* Create the threads and let them block on the mutex.
*/
rc = RTSemMutexRequest(g_hMutex, RT_INDEFINITE_WAIT);
if (RT_FAILURE(rc))
return PrintError("RTSemMutexRequest failed (rc=%Rrc)\n", rc);
for (i = 0; i < cThreads; i++)
{
g_au64[i] = 0;
rc = RTThreadCreate(&aThreads[i], ThreadTest1, &g_au64[i], 0, RTTHREADTYPE_DEFAULT, RTTHREADFLAGS_WAITABLE, "test");
if (RT_FAILURE(rc))
return PrintError("RTThreadCreate failed for thread %u (rc=%Rrc)\n", i, rc);
}
if (!fQuiet)
RTPrintf("tstSemMutex: %zu Threads created. Racing them for %u seconds (%s) ...\n",
cThreads, cSeconds, g_fYield ? "yielding" : "no yielding");
uint64_t u64StartTS = RTTimeNanoTS();
rc = RTSemMutexRelease(g_hMutex);
if (RT_FAILURE(rc))
PrintError("RTSemMutexRelease failed (rc=%Rrc)\n", rc);
RTThreadSleep(cSeconds * 1000);
ASMAtomicXchgBool(&g_fTerminate, true);
uint64_t ElapsedNS = RTTimeNanoTS() - u64StartTS;
for (i = 0; i < cThreads; i++)
{
rc = RTThreadWait(aThreads[i], 5000, NULL);
if (RT_FAILURE(rc))
PrintError("RTThreadWait failed for thread %u (rc=%Rrc)\n", i, rc);
}
rc = RTSemMutexDestroy(g_hMutex);
if (RT_FAILURE(rc))
PrintError("RTSemMutexDestroy failed - %Rrc\n", rc);
g_hMutex = NIL_RTSEMMUTEX;
if (g_cErrors)
RTThreadSleep(100);
/*
* Collect and display the results.
*/
uint64_t Total = g_au64[0];
for (i = 1; i < cThreads; i++)
Total += g_au64[i];
uint64_t Normal = Total / cThreads;
uint64_t MaxDeviation = 0;
for (i = 0; i < cThreads; i++)
{
uint64_t Delta = RT_ABS((int64_t)(g_au64[i] - Normal));
if (Delta > Normal / 2)
RTPrintf("tstSemMutex: Warning! Thread %d deviates by more than 50%% - %llu (it) vs. %llu (avg)\n",
i, g_au64[i], Normal);
if (Delta > MaxDeviation)
MaxDeviation = Delta;
}
RTPrintf("tstSemMutex: Threads: %u Total: %llu Per Sec: %llu Avg: %llu ns Max dev: %llu%%\n",
cThreads,
Total,
Total / cSeconds,
ElapsedNS / Total,
MaxDeviation * 100 / Normal
);
return 0;
}
void Guest::updateStats(uint64_t iTick)
{
uint64_t cbFreeTotal = 0;
uint64_t cbAllocTotal = 0;
uint64_t cbBalloonedTotal = 0;
uint64_t cbSharedTotal = 0;
uint64_t cbSharedMem = 0;
ULONG uNetStatRx = 0;
ULONG uNetStatTx = 0;
ULONG aGuestStats[GUESTSTATTYPE_MAX];
RT_ZERO(aGuestStats);
AutoWriteLock alock(this COMMA_LOCKVAL_SRC_POS);
ULONG validStats = mVmValidStats;
/* Check if we have anything to report */
if (validStats)
{
mVmValidStats = pm::VMSTATMASK_NONE;
memcpy(aGuestStats, mCurrentGuestStat, sizeof(aGuestStats));
}
alock.release();
/*
* Calling SessionMachine may take time as the object resides in VBoxSVC
* process. This is why we took a snapshot of currently collected stats
* and released the lock.
*/
Console::SafeVMPtrQuiet ptrVM(mParent);
if (ptrVM.isOk())
{
int rc;
/*
* There is no point in collecting VM shared memory if other memory
* statistics are not available yet. Or is there?
*/
if (validStats)
{
/* Query the missing per-VM memory statistics. */
uint64_t cbTotalMemIgn, cbPrivateMemIgn, cbZeroMemIgn;
rc = PGMR3QueryMemoryStats(ptrVM.rawUVM(), &cbTotalMemIgn, &cbPrivateMemIgn, &cbSharedMem, &cbZeroMemIgn);
if (rc == VINF_SUCCESS)
validStats |= pm::VMSTATMASK_GUEST_MEMSHARED;
}
if (mCollectVMMStats)
{
rc = PGMR3QueryGlobalMemoryStats(ptrVM.rawUVM(), &cbAllocTotal, &cbFreeTotal, &cbBalloonedTotal, &cbSharedTotal);
AssertRC(rc);
if (rc == VINF_SUCCESS)
validStats |= pm::VMSTATMASK_VMM_ALLOC | pm::VMSTATMASK_VMM_FREE
| pm::VMSTATMASK_VMM_BALOON | pm::VMSTATMASK_VMM_SHARED;
}
uint64_t uRxPrev = mNetStatRx;
uint64_t uTxPrev = mNetStatTx;
mNetStatRx = mNetStatTx = 0;
rc = STAMR3Enum(ptrVM.rawUVM(), "*/ReceiveBytes|*/TransmitBytes", staticEnumStatsCallback, this);
AssertRC(rc);
uint64_t uTsNow = RTTimeNanoTS();
uint64_t cNsPassed = uTsNow - mNetStatLastTs;
if (cNsPassed >= 1000)
{
mNetStatLastTs = uTsNow;
uNetStatRx = (ULONG)((mNetStatRx - uRxPrev) * 1000000 / (cNsPassed / 1000)); /* in bytes per second */
uNetStatTx = (ULONG)((mNetStatTx - uTxPrev) * 1000000 / (cNsPassed / 1000)); /* in bytes per second */
validStats |= pm::VMSTATMASK_NET_RX | pm::VMSTATMASK_NET_TX;
LogFlowThisFunc(("Net Rx=%llu Tx=%llu Ts=%llu Delta=%llu\n", mNetStatRx, mNetStatTx, uTsNow, cNsPassed));
}
else
{
/* Can happen on resume or if we're using a non-monotonic clock
source for the timer and the time is adjusted. */
mNetStatRx = uRxPrev;
mNetStatTx = uTxPrev;
LogThisFunc(("Net Ts=%llu cNsPassed=%llu - too small interval\n", uTsNow, cNsPassed));
}
}
mParent->reportVmStatistics(validStats,
aGuestStats[GUESTSTATTYPE_CPUUSER],
aGuestStats[GUESTSTATTYPE_CPUKERNEL],
aGuestStats[GUESTSTATTYPE_CPUIDLE],
/* Convert the units for RAM usage stats: page (4K) -> 1KB units */
mCurrentGuestStat[GUESTSTATTYPE_MEMTOTAL] * (_4K/_1K),
mCurrentGuestStat[GUESTSTATTYPE_MEMFREE] * (_4K/_1K),
mCurrentGuestStat[GUESTSTATTYPE_MEMBALLOON] * (_4K/_1K),
(ULONG)(cbSharedMem / _1K), /* bytes -> KB */
mCurrentGuestStat[GUESTSTATTYPE_MEMCACHE] * (_4K/_1K),
mCurrentGuestStat[GUESTSTATTYPE_PAGETOTAL] * (_4K/_1K),
(ULONG)(cbAllocTotal / _1K), /* bytes -> KB */
(ULONG)(cbFreeTotal / _1K),
(ULONG)(cbBalloonedTotal / _1K),
(ULONG)(cbSharedTotal / _1K),
uNetStatRx,
uNetStatTx);
}
DECL_FORCE_INLINE(int) rtSemRWRequestWrite(RTSEMRW hRWSem, RTMSINTERVAL cMillies, bool fInterruptible, PCRTLOCKVALSRCPOS pSrcPos)
{
/*
* Validate handle.
*/
struct RTSEMRWINTERNAL *pThis = hRWSem;
AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
AssertReturn(pThis->u32Magic == RTSEMRW_MAGIC, VERR_INVALID_HANDLE);
RTMSINTERVAL cMilliesInitial = cMillies;
uint64_t tsStart = 0;
if (cMillies != RT_INDEFINITE_WAIT && cMillies != 0)
tsStart = RTTimeNanoTS();
#ifdef RTSEMRW_STRICT
RTTHREAD hThreadSelf = NIL_RTTHREAD;
if (cMillies)
{
hThreadSelf = RTThreadSelfAutoAdopt();
int rc9 = RTLockValidatorRecExclCheckOrder(&pThis->ValidatorWrite, hThreadSelf, pSrcPos, cMillies);
if (RT_FAILURE(rc9))
return rc9;
}
#endif
/*
* Take critsect.
*/
int rc = RTCritSectEnter(&pThis->CritSect);
if (RT_FAILURE(rc))
{
AssertMsgFailed(("RTCritSectEnter failed on rwsem %p, rc=%Rrc\n", hRWSem, rc));
return rc;
}
/*
* Check if the state of affairs allows write access.
*/
RTNATIVETHREAD hNativeSelf = pThis->CritSect.NativeThreadOwner;
if ( !pThis->cReads
&& ( ( !pThis->cWrites
&& ( !pThis->cWritesWaiting /* play fair if we can wait */
|| !cMillies)
)
|| pThis->hWriter == hNativeSelf
)
)
{
/*
* Reset the reader event semaphore if necessary.
*/
if (pThis->fNeedResetReadEvent)
{
pThis->fNeedResetReadEvent = false;
rc = RTSemEventMultiReset(pThis->ReadEvent);
AssertMsgRC(rc, ("Failed to reset readers, rwsem %p, rc=%Rrc.\n", hRWSem, rc));
}
pThis->cWrites++;
pThis->hWriter = hNativeSelf;
#ifdef RTSEMRW_STRICT
RTLockValidatorRecExclSetOwner(&pThis->ValidatorWrite, hThreadSelf, pSrcPos, pThis->cWrites == 1);
#endif
RTCritSectLeave(&pThis->CritSect);
return VINF_SUCCESS;
}
/*
* Signal writer presence.
*/
if (cMillies != 0)
pThis->cWritesWaiting++;
RTCritSectLeave(&pThis->CritSect);
/*
* Wait till it's ready for writing.
*/
if (cMillies == 0)
return VERR_TIMEOUT;
#ifndef RTSEMRW_STRICT
RTTHREAD hThreadSelf = RTThreadSelf();
#endif
for (;;)
{
if (cMillies != RT_INDEFINITE_WAIT)
{
int64_t tsDelta = RTTimeNanoTS() - tsStart;
if (tsDelta >= 1000000)
{
tsDelta /= 1000000;
if ((uint64_t)tsDelta < cMilliesInitial)
cMilliesInitial = (RTMSINTERVAL)tsDelta;
else
cMilliesInitial = 1;
}
}
#ifdef RTSEMRW_STRICT
rc = RTLockValidatorRecExclCheckBlocking(&pThis->ValidatorWrite, hThreadSelf, pSrcPos, true,
cMillies, RTTHREADSTATE_RW_WRITE, false);
if (RT_FAILURE(rc))
break;
#else
RTThreadBlocking(hThreadSelf, RTTHREADSTATE_RW_WRITE, false);
#endif
int rcWait;
if (fInterruptible)
rcWait = rc = RTSemEventWaitNoResume(pThis->WriteEvent, cMillies);
else
rcWait = rc = RTSemEventWait(pThis->WriteEvent, cMillies);
RTThreadUnblocked(hThreadSelf, RTTHREADSTATE_RW_WRITE);
if (RT_UNLIKELY(RT_FAILURE_NP(rc) && rc != VERR_TIMEOUT)) /* timeouts are handled below */
{
AssertMsgRC(rc, ("RTSemEventWait failed on rwsem %p, rc=%Rrc\n", hRWSem, rc));
break;
}
if (RT_UNLIKELY(pThis->u32Magic != RTSEMRW_MAGIC))
{
rc = VERR_SEM_DESTROYED;
break;
}
/*
* Re-take critsect and repeat the check we did prior to this loop.
*/
rc = RTCritSectEnter(&pThis->CritSect);
if (RT_FAILURE(rc))
{
AssertMsgFailed(("RTCritSectEnter failed on rwsem %p, rc=%Rrc\n", hRWSem, rc));
break;
}
if (!pThis->cReads && (!pThis->cWrites || pThis->hWriter == hNativeSelf))
{
/*
* Reset the reader event semaphore if necessary.
*/
if (pThis->fNeedResetReadEvent)
{
pThis->fNeedResetReadEvent = false;
rc = RTSemEventMultiReset(pThis->ReadEvent);
AssertMsgRC(rc, ("Failed to reset readers, rwsem %p, rc=%Rrc.\n", hRWSem, rc));
}
pThis->cWrites++;
pThis->hWriter = hNativeSelf;
pThis->cWritesWaiting--;
#ifdef RTSEMRW_STRICT
RTLockValidatorRecExclSetOwner(&pThis->ValidatorWrite, hThreadSelf, pSrcPos, true);
#endif
RTCritSectLeave(&pThis->CritSect);
return VINF_SUCCESS;
}
RTCritSectLeave(&pThis->CritSect);
/*
* Quit if the wait already timed out.
*/
if (rcWait == VERR_TIMEOUT)
{
rc = VERR_TIMEOUT;
break;
}
}
/*
* Timeout/error case, clean up.
*/
if (pThis->u32Magic == RTSEMRW_MAGIC)
{
RTCritSectEnter(&pThis->CritSect);
/* Adjust this counter, whether we got the critsect or not. */
pThis->cWritesWaiting--;
RTCritSectLeave(&pThis->CritSect);
}
return rc;
}
DECL_FORCE_INLINE(int) rtSemRWRequestRead(RTSEMRW hRWSem, RTMSINTERVAL cMillies, bool fInterruptible, PCRTLOCKVALSRCPOS pSrcPos)
{
/*
* Validate handle.
*/
struct RTSEMRWINTERNAL *pThis = hRWSem;
AssertPtrReturn(pThis, VERR_INVALID_HANDLE);
AssertReturn(pThis->u32Magic == RTSEMRW_MAGIC, VERR_INVALID_HANDLE);
RTMSINTERVAL cMilliesInitial = cMillies;
uint64_t tsStart = 0;
if (cMillies != RT_INDEFINITE_WAIT && cMillies != 0)
tsStart = RTTimeNanoTS();
#ifdef RTSEMRW_STRICT
RTTHREAD hThreadSelf = RTThreadSelfAutoAdopt();
if (cMillies > 0)
{
int rc9;
if (pThis->hWriter != NIL_RTTHREAD && pThis->hWriter == RTThreadNativeSelf())
rc9 = RTLockValidatorRecExclCheckOrder(&pThis->ValidatorWrite, hThreadSelf, pSrcPos, cMillies);
else
rc9 = RTLockValidatorRecSharedCheckOrder(&pThis->ValidatorRead, hThreadSelf, pSrcPos, cMillies);
if (RT_FAILURE(rc9))
return rc9;
}
#endif
/*
* Take critsect.
*/
int rc = RTCritSectEnter(&pThis->CritSect);
if (RT_FAILURE(rc))
{
AssertMsgFailed(("RTCritSectEnter failed on rwsem %p, rc=%Rrc\n", hRWSem, rc));
return rc;
}
/*
* Check if the state of affairs allows read access.
* Do not block further readers if there is a writer waiting, as
* that will break/deadlock reader recursion.
*/
if ( pThis->hWriter == NIL_RTNATIVETHREAD
#if 0
&& ( !pThis->cWritesWaiting
|| pThis->cReads)
#endif
)
{
pThis->cReads++;
Assert(pThis->cReads > 0);
#ifdef RTSEMRW_STRICT
RTLockValidatorRecSharedAddOwner(&pThis->ValidatorRead, hThreadSelf, pSrcPos);
#endif
RTCritSectLeave(&pThis->CritSect);
return VINF_SUCCESS;
}
RTNATIVETHREAD hNativeSelf = pThis->CritSect.NativeThreadOwner;
if (pThis->hWriter == hNativeSelf)
{
#ifdef RTSEMRW_STRICT
int rc9 = RTLockValidatorRecExclRecursionMixed(&pThis->ValidatorWrite, &pThis->ValidatorRead.Core, pSrcPos);
if (RT_FAILURE(rc9))
{
RTCritSectLeave(&pThis->CritSect);
return rc9;
}
#endif
pThis->cWriterReads++;
Assert(pThis->cWriterReads > 0);
RTCritSectLeave(&pThis->CritSect);
return VINF_SUCCESS;
}
RTCritSectLeave(&pThis->CritSect);
/*
* Wait till it's ready for reading.
*/
if (cMillies == 0)
return VERR_TIMEOUT;
#ifndef RTSEMRW_STRICT
RTTHREAD hThreadSelf = RTThreadSelf();
#endif
for (;;)
{
if (cMillies != RT_INDEFINITE_WAIT)
{
int64_t tsDelta = RTTimeNanoTS() - tsStart;
if (tsDelta >= 1000000)
{
tsDelta /= 1000000;
if ((uint64_t)tsDelta < cMilliesInitial)
cMilliesInitial = (RTMSINTERVAL)tsDelta;
else
cMilliesInitial = 1;
}
}
#ifdef RTSEMRW_STRICT
rc = RTLockValidatorRecSharedCheckBlocking(&pThis->ValidatorRead, hThreadSelf, pSrcPos, true,
cMillies, RTTHREADSTATE_RW_READ, false);
if (RT_FAILURE(rc))
break;
#else
RTThreadBlocking(hThreadSelf, RTTHREADSTATE_RW_READ, false);
#endif
int rcWait;
if (fInterruptible)
rcWait = rc = RTSemEventMultiWaitNoResume(pThis->ReadEvent, cMillies);
else
rcWait = rc = RTSemEventMultiWait(pThis->ReadEvent, cMillies);
RTThreadUnblocked(hThreadSelf, RTTHREADSTATE_RW_READ);
if (RT_FAILURE(rc) && rc != VERR_TIMEOUT) /* handle timeout below */
{
AssertMsgRC(rc, ("RTSemEventMultiWait failed on rwsem %p, rc=%Rrc\n", hRWSem, rc));
break;
}
if (pThis->u32Magic != RTSEMRW_MAGIC)
{
rc = VERR_SEM_DESTROYED;
break;
}
/*
* Re-take critsect and repeat the check we did before the loop.
*/
rc = RTCritSectEnter(&pThis->CritSect);
if (RT_FAILURE(rc))
{
AssertMsgFailed(("RTCritSectEnter failed on rwsem %p, rc=%Rrc\n", hRWSem, rc));
break;
}
if ( pThis->hWriter == NIL_RTNATIVETHREAD
#if 0
&& ( !pThis->cWritesWaiting
|| pThis->cReads)
#endif
)
{
pThis->cReads++;
Assert(pThis->cReads > 0);
#ifdef RTSEMRW_STRICT
RTLockValidatorRecSharedAddOwner(&pThis->ValidatorRead, hThreadSelf, pSrcPos);
#endif
RTCritSectLeave(&pThis->CritSect);
return VINF_SUCCESS;
}
RTCritSectLeave(&pThis->CritSect);
/*
* Quit if the wait already timed out.
*/
if (rcWait == VERR_TIMEOUT)
{
rc = VERR_TIMEOUT;
break;
}
}
/* failed */
return rc;
}
RTDECL(int) RTFileAioCtxWait(RTFILEAIOCTX hAioCtx, size_t cMinReqs, RTMSINTERVAL cMillies,
PRTFILEAIOREQ pahReqs, size_t cReqs, uint32_t *pcReqs)
{
int rc = VINF_SUCCESS;
int cRequestsCompleted = 0;
/*
* Validate the parameters, making sure to always set pcReqs.
*/
AssertPtrReturn(pcReqs, VERR_INVALID_POINTER);
*pcReqs = 0; /* always set */
PRTFILEAIOCTXINTERNAL pCtxInt = hAioCtx;
RTFILEAIOCTX_VALID_RETURN(pCtxInt);
AssertPtrReturn(pahReqs, VERR_INVALID_POINTER);
AssertReturn(cReqs != 0, VERR_INVALID_PARAMETER);
AssertReturn(cReqs >= cMinReqs, VERR_OUT_OF_RANGE);
if (RT_UNLIKELY(ASMAtomicReadS32(&pCtxInt->cRequests) == 0))
return VERR_FILE_AIO_NO_REQUEST;
/*
* Convert the timeout if specified.
*/
struct timespec *pTimeout = NULL;
struct timespec Timeout = {0,0};
uint64_t StartNanoTS = 0;
if (cMillies != RT_INDEFINITE_WAIT)
{
Timeout.tv_sec = cMillies / 1000;
Timeout.tv_nsec = cMillies % 1000 * 1000000;
pTimeout = &Timeout;
StartNanoTS = RTTimeNanoTS();
}
/* Wait for at least one. */
if (!cMinReqs)
cMinReqs = 1;
/* For the wakeup call. */
Assert(pCtxInt->hThreadWait == NIL_RTTHREAD);
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, RTThreadSelf());
while ( cMinReqs
&& RT_SUCCESS_NP(rc))
{
struct kevent aKEvents[AIO_MAXIMUM_REQUESTS_PER_CONTEXT];
int cRequestsToWait = cMinReqs < AIO_MAXIMUM_REQUESTS_PER_CONTEXT ? cReqs : AIO_MAXIMUM_REQUESTS_PER_CONTEXT;
int rcBSD;
uint64_t StartTime;
ASMAtomicXchgBool(&pCtxInt->fWaiting, true);
rcBSD = kevent(pCtxInt->iKQueue, NULL, 0, aKEvents, cRequestsToWait, pTimeout);
ASMAtomicXchgBool(&pCtxInt->fWaiting, false);
if (RT_UNLIKELY(rcBSD < 0))
{
rc = RTErrConvertFromErrno(errno);
break;
}
uint32_t const cDone = rcBSD;
/* Process received events. */
for (uint32_t i = 0; i < cDone; i++)
{
PRTFILEAIOREQINTERNAL pReqInt = (PRTFILEAIOREQINTERNAL)aKEvents[i].udata;
AssertPtr(pReqInt);
Assert(pReqInt->u32Magic == RTFILEAIOREQ_MAGIC);
/*
* Retrieve the status code here already because the
* user may omit the RTFileAioReqGetRC() call and
* we will leak kernel resources then.
* This will result in errors during submission
* of other requests as soon as the max_aio_queue_per_proc
* limit is reached.
*/
int cbTransfered = aio_return(&pReqInt->AioCB);
if (cbTransfered < 0)
{
pReqInt->Rc = RTErrConvertFromErrno(cbTransfered);
pReqInt->cbTransfered = 0;
}
else
{
pReqInt->Rc = VINF_SUCCESS;
pReqInt->cbTransfered = cbTransfered;
}
RTFILEAIOREQ_SET_STATE(pReqInt, COMPLETED);
pahReqs[cRequestsCompleted++] = (RTFILEAIOREQ)pReqInt;
}
/*
* Done Yet? If not advance and try again.
*/
if (cDone >= cMinReqs)
break;
cMinReqs -= cDone;
cReqs -= cDone;
if (cMillies != RT_INDEFINITE_WAIT)
{
/* The API doesn't return ETIMEDOUT, so we have to fix that ourselves. */
uint64_t NanoTS = RTTimeNanoTS();
uint64_t cMilliesElapsed = (NanoTS - StartNanoTS) / 1000000;
if (cMilliesElapsed >= cMillies)
{
rc = VERR_TIMEOUT;
break;
}
/* The syscall supposedly updates it, but we're paranoid. :-) */
Timeout.tv_sec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) / 1000;
Timeout.tv_nsec = (cMillies - (RTMSINTERVAL)cMilliesElapsed) % 1000 * 1000000;
}
}
/*
* Update the context state and set the return value.
*/
*pcReqs = cRequestsCompleted;
ASMAtomicSubS32(&pCtxInt->cRequests, cRequestsCompleted);
Assert(pCtxInt->hThreadWait == RTThreadSelf());
ASMAtomicWriteHandle(&pCtxInt->hThreadWait, NIL_RTTHREAD);
/*
* Clear the wakeup flag and set rc.
*/
if ( pCtxInt->fWokenUp
&& RT_SUCCESS(rc))
{
ASMAtomicXchgBool(&pCtxInt->fWokenUp, false);
rc = VERR_INTERRUPTED;
}
return rc;
}
RTDECL(int) RTMpOnSpecific(RTCPUID idCpu, PFNRTMPWORKER pfnWorker, void *pvUser1, void *pvUser2)
{
/*
* Don't try mess with an offline CPU.
*/
if (!RTMpIsCpuOnline(idCpu))
return !RTMpIsCpuPossible(idCpu)
? VERR_CPU_NOT_FOUND
: VERR_CPU_OFFLINE;
/*
* Use the broadcast IPI routine if there are no more than two CPUs online,
* or if the current IRQL is unsuitable for KeWaitForSingleObject.
*/
if ( g_pfnrtKeIpiGenericCall
&& ( RTMpGetOnlineCount() <= 2
|| KeGetCurrentIrql() > APC_LEVEL) )
return rtMpCallUsingBroadcastIpi(pfnWorker, pvUser1, pvUser2, rtmpNtOnSpecificBroadcastIpiWrapper, 0);
#if 0 /** @todo untested code. needs some tuning. */
/*
* Initialize the argument package and the objects within it.
* The package is referenced counted to avoid unnecessary spinning to
* synchronize cleanup and prevent stack corruption.
*/
PRTMPNTONSPECIFICARGS pArgs = (PRTMPNTONSPECIFICARGS)ExAllocatePoolWithTag(NonPagedPool, sizeof(*pArgs), (ULONG)'RTMp');
if (!pArgs)
return VERR_NO_MEMORY;
pArgs->cRefs = 2;
pArgs->fExecuting = false;
pArgs->fDone = false;
pArgs->CallbackArgs.pfnWorker = pfnWorker;
pArgs->CallbackArgs.pvUser1 = pvUser1;
pArgs->CallbackArgs.pvUser2 = pvUser2;
pArgs->CallbackArgs.idCpu = idCpu;
pArgs->CallbackArgs.cHits = 0;
pArgs->CallbackArgs.cRefs = 2;
KeInitializeEvent(&pArgs->DoneEvt, SynchronizationEvent, FALSE /* not signalled */);
KeInitializeDpc(&pArgs->Dpc, rtMpNtOnSpecificDpcWrapper, pArgs);
KeSetImportanceDpc(&pArgs->Dpc, HighImportance);
KeSetTargetProcessorDpc(&pArgs->Dpc, (int)idCpu);
/*
* Disable preemption while we check the current processor and inserts the DPC.
*/
KIRQL bOldIrql;
KeRaiseIrql(DISPATCH_LEVEL, &bOldIrql);
ASMCompilerBarrier(); /* paranoia */
if (RTMpCpuId() == idCpu)
{
/* Just execute the callback on the current CPU. */
pfnWorker(idCpu, pvUser1, pvUser2);
KeLowerIrql(bOldIrql);
ExFreePool(pArgs);
return VINF_SUCCESS;
}
/* Different CPU, so queue it if the CPU is still online. */
int rc;
if (RTMpIsCpuOnline(idCpu))
{
BOOLEAN fRc = KeInsertQueueDpc(&pArgs->Dpc, 0, 0);
Assert(fRc);
KeLowerIrql(bOldIrql);
uint64_t const nsRealWaitTS = RTTimeNanoTS();
/*
* Wait actively for a while in case the CPU/thread responds quickly.
*/
uint32_t cLoopsLeft = 0x20000;
while (cLoopsLeft-- > 0)
{
if (pArgs->fDone)
{
rtMpNtOnSpecificRelease(pArgs);
return VINF_SUCCESS;
}
ASMNopPause();
}
/*
* It didn't respond, so wait on the event object, poking the CPU if it's slow.
*/
LARGE_INTEGER Timeout;
Timeout.QuadPart = -10000; /* 1ms */
NTSTATUS rcNt = KeWaitForSingleObject(&pArgs->DoneEvt, Executive, KernelMode, FALSE /* Alertable */, &Timeout);
if (rcNt == STATUS_SUCCESS)
{
rtMpNtOnSpecificRelease(pArgs);
return VINF_SUCCESS;
}
/* If it hasn't respondend yet, maybe poke it and wait some more. */
if (rcNt == STATUS_TIMEOUT)
{
if ( !pArgs->fExecuting
&& ( g_pfnrtMpPokeCpuWorker == rtMpPokeCpuUsingHalSendSoftwareInterrupt
|| g_pfnrtMpPokeCpuWorker == rtMpPokeCpuUsingHalReqestIpiW7Plus
|| g_pfnrtMpPokeCpuWorker == rtMpPokeCpuUsingHalReqestIpiPreW7))
RTMpPokeCpu(idCpu);
Timeout.QuadPart = -1280000; /* 128ms */
rcNt = KeWaitForSingleObject(&pArgs->DoneEvt, Executive, KernelMode, FALSE /* Alertable */, &Timeout);
if (rcNt == STATUS_SUCCESS)
{
rtMpNtOnSpecificRelease(pArgs);
return VINF_SUCCESS;
}
}
/*
* Something weird is happening, try bail out.
*/
if (KeRemoveQueueDpc(&pArgs->Dpc))
{
ExFreePool(pArgs); /* DPC was still queued, so we can return without further ado. */
LogRel(("RTMpOnSpecific(%#x): Not processed after %llu ns: rcNt=%#x\n", idCpu, RTTimeNanoTS() - nsRealWaitTS, rcNt));
}
else
{
/* DPC is running, wait a good while for it to complete. */
LogRel(("RTMpOnSpecific(%#x): Still running after %llu ns: rcNt=%#x\n", idCpu, RTTimeNanoTS() - nsRealWaitTS, rcNt));
Timeout.QuadPart = -30*1000*1000*10; /* 30 seconds */
rcNt = KeWaitForSingleObject(&pArgs->DoneEvt, Executive, KernelMode, FALSE /* Alertable */, &Timeout);
if (rcNt != STATUS_SUCCESS)
LogRel(("RTMpOnSpecific(%#x): Giving up on running worker after %llu ns: rcNt=%#x\n", idCpu, RTTimeNanoTS() - nsRealWaitTS, rcNt));
}
rc = RTErrConvertFromNtStatus(rcNt);
}
else
{
/* CPU is offline.*/
KeLowerIrql(bOldIrql);
rc = !RTMpIsCpuPossible(idCpu) ? VERR_CPU_NOT_FOUND : VERR_CPU_OFFLINE;
}
rtMpNtOnSpecificRelease(pArgs);
return rc;
#else
return rtMpCallUsingDpcs(pfnWorker, pvUser1, pvUser2, RT_NT_CPUID_SPECIFIC, idCpu);
#endif
}
