/* -*- Mode: C; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- * * The contents of this file are subject to the Mozilla Public * License Version 1.1 (the "License"); you may not use this file * except in compliance with the License. You may obtain a copy of * the License at http://www.mozilla.org/MPL/ * * Software distributed under the License is distributed on an "AS * IS" basis, WITHOUT WARRANTY OF ANY KIND, either express or * implied. See the License for the specific language governing * rights and limitations under the License. * * The Original Code is spacetrace.h/spacetrace.c code, released * Nov 6, 2001. * * The Initial Developer of the Original Code is Netscape * Communications Corporation. Portions created by Netscape are * Copyright (C) 2001 Netscape Communications Corporation. All * Rights Reserved. * * Contributor(s): * Garrett Arch Blythe, 31-October-2001 * * Alternatively, the contents of this file may be used under the * terms of the GNU Public License (the "GPL"), in which case the * provisions of the GPL are applicable instead of those above. * If you wish to allow use of your version of this file only * under the terms of the GPL and not to allow others to use your * version of this file under the MPL, indicate your decision by * deleting the provisions above and replace them with the notice * and other provisions required by the GPL. If you do not delete * the provisions above, a recipient may use your version of this * file under either the MPL or the GPL. */ /* ** spacetrace.c ** ** SpaceTrace is meant to take the output of trace-malloc and present ** a picture of allocations over the run of the application. */ /* ** Required include files. */ #include "spacetrace.h" #include #include #include #include #if defined(XP_WIN32) #include /* _heapMin */ #endif #if defined(HAVE_BOUTELL_GD) /* ** See http://www.boutell.com/gd for the GD graphics library. ** Ports for many platorms exist. ** Your box may already have the lib (mine did, redhat 7.1 workstation). */ #include #include #include #include #endif /* HAVE_BOUTELL_GD */ /* ** Ugh, MSVC6's qsort is too slow... */ #include "nsQuickSort.h" /* ** Turn on to attempt adding support for graphs on your platform. */ #if defined(HAVE_BOUTELL_GD) #define WANT_GRAPHS 1 #endif /* HAVE_BOUTELL_GD */ #if !defined(WANT_GRAPHS) #define WANT_GRAPHS 0 #endif /* ** Turn on to add the ability to quit the server from the client. ** A dubious feature at best. */ #define WANT_QUIT 0 /* ** the globals variables. happy joy. */ static STGlobals globals; /* ** have the heap cleanup at opportune times, if possible. */ void heapCompact(void) { #if defined(XP_WIN32) _heapmin(); #endif } /* ** showHelp ** ** Give simple command line help. ** Returns !0 if the help was showed. */ int showHelp(void) { int retval = 0; if(0 != globals.mOptions.mShowHelp) { PR_fprintf(PR_STDOUT, "Usage:\t%s [OPTION]... [-|filename]\n\n", globals.mOptions.mProgramName); PR_fprintf(PR_STDOUT, "%s", "OPTIONS:\n" " -h Show this help.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -p Listen for http requests on the specified .\n" " Default port is '1969'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -d Place -b output in .\n" " The directory must exist.\n" " The default directory is '.'.\n" " Very important to not have a trailing slash!\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -b Execute in batch mode, multiple -b's allowed.\n" " Save into -d, then exit.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -l Set the maximum number of items to display in a list.\n" " The default is '500'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -o Sets the order in which lists are sorted when displayed.\n" " '0' is by weight (lifespan * byte size).\n" " '1' is by byte size.\n" " '2' is by time (lifetime).\n" ""); PR_fprintf(PR_STDOUT, "%s", " '3' is by allocation object count.\n" " '4' is by heap operation runtime cost.\n" " By default, is '0'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -smin Set the minimum byte size to exclude smaller allocations.\n" " The default is '0'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -smax Set the maximum byte size to exclude larger allocations.\n" " By default, there is no maximum.\n" "\n"); PR_fprintf(PR_STDOUT, " -tmin Set the minimum allocation lifetime in seconds.\n" " Excludes allocations which do not live at least said seconds.\n" " The default is '%u' seconds.\n" "\n", ST_DEFAULT_LIFETIME_MIN); PR_fprintf(PR_STDOUT, "%s", " -tmax Set the maximum allocation lifetime in seconds.\n" " Excludes allocations which live longer than the said seconds.\n" " By default, there is no maximum.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -wmin Set the minimum allocation weight.\n" " Weight is lifespan * byte size.\n" " Excludes allocations which do not have the weight.\n" " The default is '0'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -wmax Set the maximum allocation weight.\n" " Weight is lifespan * byte size.\n" " Excludes allocations which are over weight.\n" " By default, there is no maximum.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -imin Set the minimum in seconds.\n" " Excludes allocations existing solely before said second.\n" " The default is '0'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -imax Set the maximum in seconds.\n" " Excludes allocations existing solely after said second.\n" " By default, there is no maximum.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -amin Set the allocation minimum in seconds.\n" " Excludes allocations created before said second.\n" " The default is '0'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -amax Set the allocation maximum in seconds.\n" " Excludes allocations created after said second.\n" " By default, there is no maximum.\n" "\n"); #if WANT_GRAPHS PR_fprintf(PR_STDOUT, "%s", " -gmin Set the graph minimum in seconds.\n" " Excludes representing graph intervals before said second.\n" " The default is '0'.\n" "\n"); PR_fprintf(PR_STDOUT, "%s", " -gmax Set the graph maximum in seconds.\n" " Excludes representing graph intervals after said second.\n" " By default, there is no maximum.\n" "\n"); #endif /* WANT_GRAPHS */ PR_fprintf(PR_STDOUT, " -a Set an allocation alignment boundry.\n" " All allocations are a factor of .\n" " Meaning, an allocation of 1 byte would actually count as bytes.\n" ); PR_fprintf(PR_STDOUT, " Set to '1' in order to see the actual allocation sizes.\n" " Alignment is taken into account prior to allocation overhead.\n" " By default, is %u.\n" "\n", ST_DEFAULT_ALIGNMENT_SIZE); PR_fprintf(PR_STDOUT, " -h Set allocation overhead.\n" " All allocations cost an additional bytes.\n" " Overhead is taken into account after allocation alignment.\n" " Set to '0' in order to see the actual allocation sizes.\n" " By default, is %u.\n" "\n", ST_DEFAULT_OVERHEAD_SIZE); PR_fprintf(PR_STDOUT, " -c Restrict callsite backtraces to only those containing .\n" " Allows targeting of specific object creation methods.\n" " A maximum of %d multiple restrictions can be specified.\n" " By default, there is no restriction.\n" "\n", ST_SUBSTRING_MATCH_MAX); /* ** Showed something. */ retval = __LINE__; } return retval; } /* ** ticks2xsec ** ** Convert platform specific ticks to second units ** Returns 0 on success. */ PRUint32 ticks2xsec(tmreader* aReader, PRUint32 aTicks, PRUint32 aResolution) { PRUint32 retval = 0; PRUint64 bigone; PRUint64 tmp64; LL_UI2L(bigone, aResolution); LL_UI2L(tmp64, aTicks); LL_MUL(bigone, bigone, tmp64); LL_UI2L(tmp64, aReader->ticksPerSec); LL_DIV(bigone, bigone, tmp64); LL_L2UI(retval, bigone); return retval; } #define ticks2msec(reader, ticks) ticks2xsec((reader), (ticks), 1000) #define ticks2usec(reader, ticks) ticks2xsec((reader), (ticks), 1000000) /* ** initOptions ** ** Determine global settings for the application. ** Returns 0 on success. */ int initOptions(int aArgCount, char** aArgArray) { int retval = 0; int traverse = 0; const char* stdinDash = "-"; const char* outputDir = "."; const PRUint32 httpdPort = 1969; const PRUint32 listItemMax = 500; PRStatus prStatus = PR_SUCCESS; /* ** Set the program name. */ globals.mOptions.mProgramName = aArgArray[0]; /* ** As a default, stdin is the input. */ globals.mOptions.mFileName = stdinDash; /* ** As a default, this directory is the output. */ globals.mOptions.mOutputDir = outputDir; /* ** As a default, this is the port to listen for http requests. */ globals.mOptions.mHttpdPort = httpdPort; /* ** As a default, the number of list items to limit display to. */ globals.mOptions.mListItemMax = listItemMax; /* ** As a default, there is no maximum timeval on the dataset. */ globals.mOptions.mTimevalMax = ST_TIMEVAL_MAX; /* ** As a default, there is no maximum allocation size on the dataset. */ globals.mOptions.mSizeMax = (PRUint32)-1; /* ** As a default, want to look at allocations which live at least... */ globals.mOptions.mLifetimeMin = ST_DEFAULT_LIFETIME_MIN * ST_TIMEVAL_RESOLUTION; /* ** As a default, there is no maximum allocation lifetime timeval. */ globals.mOptions.mLifetimeMax = ST_TIMEVAL_MAX; /* ** As a default, there is no maximum weight allowed. */ globals.mOptions.mWeightMax64 = LL_INIT(0xFFFFFFFF, 0xFFFFFFFF); /* ** As a default, there is no maximum allocation timeval. */ globals.mOptions.mAllocationTimevalMax = ST_TIMEVAL_MAX; /* ** As a default, there is no maximum graph timeval. */ globals.mOptions.mGraphTimevalMax = ST_TIMEVAL_MAX; /* ** As a default, we align byte sizes to a particular size. */ globals.mOptions.mAlignBy = ST_DEFAULT_ALIGNMENT_SIZE; globals.mOptions.mOverhead = ST_DEFAULT_OVERHEAD_SIZE; /* ** category file **/ globals.mOptions.mCategoryFile = "rules.txt"; /* ** Default category name - name of root category */ globals.mOptions.mCategoryName = ST_ROOT_CATEGORY_NAME; /* ** Go through all arguments. ** If argument does not being with a dash it is a file name. ** If argument does begin with a dash but is only a dash ** it means input comes from stdin. ** If argument begins with a dash and does not end, then it ** maybe an option. */ for(traverse = 1; traverse < aArgCount; traverse++) { if('-' == aArgArray[traverse][0]) { /* ** Regular dash options. ** Detect what to do. */ switch(tolower(aArgArray[traverse][1])) { case '\0': { /* ** If the entire option is a dash, ** then input is stdin. */ globals.mOptions.mFileName = stdinDash; } break; case 'h': { if('\0' != aArgArray[traverse][2]) { PRInt32 scanRes = 0; /* ** Allocation overhead. */ scanRes = PR_sscanf(&aArgArray[traverse][2], "%u", &globals.mOptions.mOverhead); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } else { /* ** Help. */ globals.mOptions.mShowHelp = __LINE__; } } break; case 'p': { PRInt32 scanRes = 0; /* ** Port. */ scanRes = PR_sscanf(&aArgArray[traverse][2], "%u", &globals.mOptions.mHttpdPort); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'o': { PRInt32 scanRes = 0; /* ** Sort Order. */ scanRes = PR_sscanf(&aArgArray[traverse][2], "%u", &globals.mOptions.mOrderBy); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'd': { /* ** Where to stick the '-b' output. */ if('\0' != aArgArray[traverse][2]) { globals.mOptions.mOutputDir = &aArgArray[traverse][2]; } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'b': { /* ** Batch mode request. */ if('\0' != aArgArray[traverse][2]) { const char** expand = NULL; /* ** Increase size of batch buffer. */ expand = (const char**)realloc((void*)globals.mOptions.mBatchRequests, sizeof(const char*) * (globals.mOptions.mBatchRequestCount + 1)); if(NULL != expand) { /* ** Reassign in case of pointer move. */ globals.mOptions.mBatchRequests = expand; /* ** Add new entry, increase the count. */ globals.mOptions.mBatchRequests[globals.mOptions.mBatchRequestCount] = &aArgArray[traverse][2]; globals.mOptions.mBatchRequestCount++; } else { retval = __LINE__; REPORT_ERROR(__LINE__, realloc); } } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'l': { PRInt32 scanRes = 0; /* ** List item max. */ scanRes = PR_sscanf(&aArgArray[traverse][2], "%u", &globals.mOptions.mListItemMax); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'i': { if(0 == strncmp(&aArgArray[traverse][2], "min", 3)) { PRInt32 scanRes = 0; /* ** Timeval min. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mTimevalMin); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mTimevalMin *= ST_TIMEVAL_RESOLUTION; } } else if(0 == strncmp(&aArgArray[traverse][2], "max", 3)) { PRInt32 scanRes = 0; /* ** Timeval max */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mTimevalMax); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mTimevalMax *= ST_TIMEVAL_RESOLUTION; } } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'a': { if(0 == strncmp(&aArgArray[traverse][2], "min", 3)) { PRInt32 scanRes = 0; /* ** Allocation Timeval min. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mAllocationTimevalMin); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mAllocationTimevalMin *= ST_TIMEVAL_RESOLUTION; } } else if(0 == strncmp(&aArgArray[traverse][2], "max", 3)) { PRInt32 scanRes = 0; /* ** Allocation timeval max */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mAllocationTimevalMax); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mAllocationTimevalMax *= ST_TIMEVAL_RESOLUTION; } } else { PRInt32 scanRes = 0; /* ** Align by. */ scanRes = PR_sscanf(&aArgArray[traverse][2], "%u", &globals.mOptions.mAlignBy); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } } break; #if WANT_GRAPHS case 'g': { if(0 == strncmp(&aArgArray[traverse][2], "min", 3)) { PRInt32 scanRes = 0; /* ** Graph Timeval min. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mGraphTimevalMin); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mGraphTimevalMin *= ST_TIMEVAL_RESOLUTION; } } else if(0 == strncmp(&aArgArray[traverse][2], "max", 3)) { PRInt32 scanRes = 0; /* ** Graph Timeval max */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mGraphTimevalMax); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mGraphTimevalMax *= ST_TIMEVAL_RESOLUTION; } } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; #endif /* WANT_GRAPHS */ case 's': { if(0 == strncmp(&aArgArray[traverse][2], "min", 3)) { PRInt32 scanRes = 0; /* ** Size min. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mSizeMin); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } else if(0 == strncmp(&aArgArray[traverse][2], "max", 3)) { PRInt32 scanRes = 0; /* ** Size max. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mSizeMax); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 't': { if(0 == strncmp(&aArgArray[traverse][2], "min", 3)) { PRInt32 scanRes = 0; /* ** Lifetime min. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mLifetimeMin); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mLifetimeMin *= ST_TIMEVAL_RESOLUTION; } } else if(0 == strncmp(&aArgArray[traverse][2], "max", 3)) { PRInt32 scanRes = 0; /* ** Lifetime max. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%u", &globals.mOptions.mLifetimeMax); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } else { globals.mOptions.mLifetimeMax *= ST_TIMEVAL_RESOLUTION; } } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'w': { if(0 == strncmp(&aArgArray[traverse][2], "min", 3)) { PRInt32 scanRes = 0; /* ** Weight min. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%llu", &globals.mOptions.mWeightMin64); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } else if(0 == strncmp(&aArgArray[traverse][2], "max", 3)) { PRInt32 scanRes = 0; /* ** Weight max. */ scanRes = PR_sscanf(&aArgArray[traverse][5], "%llu", &globals.mOptions.mWeightMax64); if(1 != scanRes) { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; case 'c': { /* ** Restrict callsite text. */ if('\0' != aArgArray[traverse][2]) { int looper = 0; /* ** Figure out if we have any free space. ** If so, copy it. */ for(looper = 0; ST_SUBSTRING_MATCH_MAX > looper; looper++) { if(NULL != globals.mOptions.mRestrictText[looper]) { continue; } globals.mOptions.mRestrictText[looper] = strdup(&aArgArray[traverse][2]); if(NULL == globals.mOptions.mRestrictText[looper]) { retval = __LINE__; } break; } /* ** Error on no more space left. */ if(ST_SUBSTRING_MATCH_MAX == looper) { retval = __LINE__; } } else { retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } } break; default: { /* ** Unknown option. ** Error and show help. */ retval = __LINE__; globals.mOptions.mShowHelp = __LINE__; } break; } /* ** Check for some type of error, so we know to break the ** loop if need be. */ if(0 != retval) { break; } } else { /* ** File to process. */ globals.mOptions.mFileName = aArgArray[traverse]; } } /* ** initalize the categories */ initCategories(&globals); return retval; } #if WANT_GRAPHS /* ** createGraph ** ** Create a GD image with the common properties of a graph. ** Upon return, you normally allocate legend colors, ** draw your graph inside the region ** STGD_MARGIN,STGD_MARGIN,STGD_WIDTH-STGD_MARGIN,STGD_HEIGH-STGD_MARGIN, ** and then call drawGraph to format the surrounding information. ** ** You should use the normal GD image release function, gdImageDestroy ** when done with it. ** ** Image attributes: ** STGD_WIDTHxSTGD_HEIGHT ** trasparent (white) background ** incremental display */ gdImagePtr createGraph(int* aTransparencyColor) { gdImagePtr retval = NULL; if(NULL != aTransparencyColor) { *aTransparencyColor = -1; retval = gdImageCreate(STGD_WIDTH, STGD_HEIGHT); if(NULL != retval) { /* ** Background color (first one). */ *aTransparencyColor = gdImageColorAllocate(retval, 255, 255, 255); if(-1 != *aTransparencyColor) { /* ** As transparency. */ gdImageColorTransparent(retval, *aTransparencyColor); } /* ** And to set interlacing. */ gdImageInterlace(retval, 1); } else { REPORT_ERROR(__LINE__, gdImageCreate); } } else { REPORT_ERROR(__LINE__, createGraph); } return retval; } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS /* ** drawGraph ** ** This function mainly exists to simplify putitng all the pretty lace ** around a home made graph. */ void drawGraph(gdImagePtr aImage, int aColor, const char* aGraphTitle, const char* aXAxisTitle, const char* aYAxisTitle, PRUint32 aXMarkCount, PRUint32* aXMarkPercents, const char** aXMarkTexts, PRUint32 aYMarkCount, PRUint32* aYMarkPercents, const char** aYMarkTexts, PRUint32 aLegendCount, int* aLegendColors, const char** aLegendTexts) { if(NULL != aImage && NULL != aGraphTitle && NULL != aXAxisTitle && NULL != aYAxisTitle && (0 == aXMarkCount || (NULL != aXMarkPercents && NULL != aXMarkTexts)) && (0 == aYMarkCount || (NULL != aYMarkPercents && NULL != aYMarkTexts)) && (0 == aLegendCount || (NULL != aLegendColors && NULL != aLegendTexts))) { int margin = 1; PRUint32 traverse = 0; PRUint32 target = 0; const int markSize = 2; int x1 = 0; int y1 = 0; int x2 = 0; int y2 = 0; time_t theTimeT = time(NULL); char* theTime = ctime(&theTimeT); const char* logo = "SpaceTrace"; gdFontPtr titleFont = gdFontMediumBold; gdFontPtr markFont = gdFontTiny; gdFontPtr dateFont = gdFontTiny; gdFontPtr axisFont = gdFontSmall; gdFontPtr legendFont = gdFontTiny; gdFontPtr logoFont = gdFontTiny; /* ** Fixup the color. ** Black by default. */ if(-1 == aColor) { aColor = gdImageColorAllocate(aImage, 0, 0, 0); } if(-1 == aColor) { aColor = gdImageColorClosest(aImage, 0, 0, 0); } /* ** Output the box. */ x1 = STGD_MARGIN - margin; y1 = STGD_MARGIN - margin; x2 = STGD_WIDTH - x1; y2 = STGD_HEIGHT - y1; gdImageRectangle(aImage, x1, y1, x2, y2, aColor); margin++; /* ** Need to make small markings on the graph to indicate where the ** labels line up exactly. ** While we're at it, draw the label text. */ for(traverse = 0; traverse < aXMarkCount; traverse++) { target = ((STGD_WIDTH - (STGD_MARGIN * 2)) * aXMarkPercents[traverse]) / 100; x1 = STGD_MARGIN + target; y1 = STGD_MARGIN - margin; x2 = x1; y2 = y1 - markSize; gdImageLine(aImage, x1, y1, x2, y2, aColor); y1 = STGD_HEIGHT - y1; y2 = STGD_HEIGHT - y2; gdImageLine(aImage, x1, y1, x2, y2, aColor); if(NULL != aXMarkTexts[traverse]) { x1 = STGD_MARGIN + target - (markFont->h / 2); y1 = STGD_HEIGHT - STGD_MARGIN + margin + markSize + (strlen(aXMarkTexts[traverse]) * markFont->w); gdImageStringUp(aImage, markFont, x1, y1, (unsigned char*)aXMarkTexts[traverse], aColor); } } for(traverse = 0; traverse < aYMarkCount; traverse++) { target = ((STGD_HEIGHT - (STGD_MARGIN * 2)) * (100 - aYMarkPercents[traverse])) / 100; x1 = STGD_MARGIN - margin; y1 = STGD_MARGIN + target; x2 = x1 - markSize; y2 = y1; gdImageLine(aImage, x1, y1, x2, y2, aColor); x1 = STGD_WIDTH - x1; x2 = STGD_WIDTH - x2; gdImageLine(aImage, x1, y1, x2, y2, aColor); if(NULL != aYMarkTexts[traverse]) { x1 = STGD_MARGIN - margin - markSize - (strlen(aYMarkTexts[traverse]) * markFont->w); y1 = STGD_MARGIN + target - (markFont->h / 2); gdImageString(aImage, markFont, x1, y1, (unsigned char*)aYMarkTexts[traverse], aColor); } } margin += markSize; /* ** Title will be centered above the image. */ x1 = (STGD_WIDTH / 2) - ((strlen(aGraphTitle) * titleFont->w) / 2); y1 = ((STGD_MARGIN - margin) / 2) - (titleFont->h / 2); gdImageString(aImage, titleFont, x1, y1, (unsigned char*)aGraphTitle, aColor); /* ** Upper left will be the date. */ x1 = 0; y1 = 0; traverse = strlen(theTime) - 1; if(isspace(theTime[traverse])) { theTime[traverse] = '\0'; } gdImageString(aImage, dateFont, x1, y1, (unsigned char*)theTime, aColor); /* ** Lower right will be the logo. */ x1 = STGD_WIDTH - (strlen(logo) * logoFont->w); y1 = STGD_HEIGHT - logoFont->h; gdImageString(aImage, logoFont, x1, y1, (unsigned char*)logo, aColor); /* ** X and Y axis titles */ x1 = (STGD_WIDTH / 2) - ((strlen(aXAxisTitle) * axisFont->w) / 2); y1 = STGD_HEIGHT - axisFont->h; gdImageString(aImage, axisFont, x1, y1, (unsigned char*)aXAxisTitle, aColor); x1 = 0; y1 = (STGD_HEIGHT / 2) + ((strlen(aYAxisTitle) * axisFont->w) / 2); gdImageStringUp(aImage, axisFont, x1, y1, (unsigned char*)aYAxisTitle, aColor); /* ** The legend. ** Centered on the right hand side, going up. */ x1 = STGD_WIDTH - STGD_MARGIN + margin + (aLegendCount * legendFont->h) / 2; x2 = STGD_WIDTH - (aLegendCount * legendFont->h); if(x1 > x2) { x1 = x2; } y1 = 0; for(traverse = 0; traverse < aLegendCount; traverse++) { y2 = (STGD_HEIGHT / 2) + ((strlen(aLegendTexts[traverse]) * legendFont->w) / 2); if(y2 > y1) { y1 = y2; } } for(traverse = 0; traverse < aLegendCount; traverse++) { gdImageStringUp(aImage, legendFont, x1, y1, (unsigned char*)aLegendTexts[traverse], aLegendColors[traverse]); x1 += legendFont->h; } } } #endif /* WANT_GRAPHS */ #if defined(HAVE_BOUTELL_GD) /* ** pngSink ** ** GD callback, used to write out the png. */ int pngSink(void* aContext, const char* aBuffer, int aLen) { return PR_Write((PRFileDesc*)aContext, aBuffer, aLen); } #endif /* HAVE_BOUTELL_GD */ /* ** FormatNumber ** ** Formats a number with thousands separator. Dont free the result. Returns ** static data. */ char *FormatNumber(PRInt32 num) { static char buf[64]; char tmpbuf[64]; int len = 0; int bufindex = sizeof(buf) - 1; int mod3; PR_snprintf(tmpbuf, sizeof(tmpbuf), "%d", num); /* now insert the thousands separator */ mod3 = 0; len = strlen(tmpbuf); while (len >= 0) { if (tmpbuf[len] >= '0' && tmpbuf[len] <= '9') { if (mod3 == 3) { buf[bufindex--] = ','; mod3 = 0; } mod3++; } buf[bufindex--] = tmpbuf[len--]; } return buf+bufindex+1; } /* ** actualByteSize ** ** Apply alignment and overhead to size to figure out actual byte size */ PRUint32 actualByteSize(PRUint32 retval) { /* ** Need to bump the result by our alignment and overhead. ** The idea here is that an allocation actually costs you more than you ** thought. ** ** The msvcrt malloc has an alignment of 16 with an overhead of 8. ** The win32 HeapAlloc has an alignment of 8 with an overhead of 8. */ if(0 != retval) { PRUint32 eval = 0; PRUint32 over = 0; eval = retval - 1; if(0 != globals.mOptions.mAlignBy) { over = eval % globals.mOptions.mAlignBy; } retval = eval + globals.mOptions.mOverhead + globals.mOptions.mAlignBy - over; } return retval; } /* ** byteSize ** ** Figuring the byte size of an allocation. ** Might expand in the future to report size at a given time. ** For now, just use last relevant event. */ PRUint32 byteSize(STAllocation* aAlloc) { PRUint32 retval = 0; if(NULL != aAlloc && 0 != aAlloc->mEventCount) { PRUint32 index = aAlloc->mEventCount; /* ** Generally, the size is the last event's size. */ do { index--; retval = aAlloc->mEvents[index].mHeapSize; } while(0 == retval && 0 != index); } return actualByteSize(retval); } /* ** recalculateAllocationCost ** ** Given an allocation, does a recalculation of Cost - weight, heapcount etc. ** and does the right thing to propogate the cost upwards. */ int recalculateAllocationCost(STRun* aRun, STAllocation* aAllocation) { /* ** Now, see if they desire a callsite update. ** As mentioned previously, we decide if the run desires us to ** manipulate the callsite data only if it's stamp is set. ** We change all callsites and parent callsites to have that ** stamp as well, so as to mark them as being relevant to ** the current run in question. */ if(0 != aRun->mStats.mStamp) { PRUint32 timeval = aAllocation->mMaxTimeval - aAllocation->mMinTimeval; PRUint32 size = byteSize(aAllocation); PRUint64 weight64 = LL_INIT(0, 0); PRUint32 heapCost = aAllocation->mHeapRuntimeCost; PRUint64 timeval64 = LL_INIT(0, 0); PRUint64 size64 = LL_INIT(0, 0); LL_UI2L(timeval64, timeval); LL_UI2L(size64, size); LL_MUL(weight64, timeval64, size64); /* ** First, update this run. */ aRun->mStats.mCompositeCount++; aRun->mStats.mHeapRuntimeCost += heapCost; aRun->mStats.mSize += size; LL_ADD(aRun->mStats.mTimeval64, aRun->mStats.mTimeval64, timeval64); LL_ADD(aRun->mStats.mWeight64, aRun->mStats.mWeight64, weight64); /* ** Use the first event of the allocation to update the parent ** callsites. ** This has positive effect of not updating realloc callsites ** with the same data over and over again. */ if(0 < aAllocation->mEventCount) { tmcallsite* callsite = aAllocation->mEvents[0].mCallsite; STRun* callsiteRun = NULL; /* ** Go up parents till we drop. */ while(NULL != callsite && NULL != callsite->method) { callsiteRun = CALLSITE_RUN(callsite); if(NULL != callsiteRun) { /* ** Do we init it? */ if(callsiteRun->mStats.mStamp != aRun->mStats.mStamp) { memset(&callsiteRun->mStats, 0, sizeof(STCallsiteStats)); callsiteRun->mStats.mStamp = aRun->mStats.mStamp; } /* ** Add the values. ** Note that if the allocation was ever realloced, ** we are actually recording the final size. ** Also, the composite count does not include ** calls to realloc (or free for that matter), ** but rather is simply a count of actual heap ** allocation objects, from which someone will ** draw conclusions regarding number of malloc ** and free calls. ** It is possible to generate the exact number ** of calls to free/malloc/realloc should the ** absolute need arise to count them individually, ** but I fear it will take mucho memory and this ** is perhaps good enough for now. */ callsiteRun->mStats.mCompositeCount++; callsiteRun->mStats.mHeapRuntimeCost += heapCost; callsiteRun->mStats.mSize += size; LL_ADD(callsiteRun->mStats.mTimeval64, callsiteRun->mStats.mTimeval64, timeval64); LL_ADD(callsiteRun->mStats.mWeight64, callsiteRun->mStats.mWeight64, weight64); } callsite = callsite->parent; } } } return 0; } /* ** appendAllocation ** ** Given a run, append the allocation to it. ** No DUP checks are done. ** Also, we might want to update the parent callsites with stats. ** We decide to do this heavy duty work only if the run we are appending ** to has a non ZERO mStats.mStamp, meaning that it is asking to track ** such information when it was created. ** Returns !0 on success. */ int appendAllocation(STRun* aRun, STAllocation* aAllocation) { int retval = 0; if(NULL != aRun && NULL != aAllocation) { STAllocation** expand = NULL; /* ** Expand the size of the array if needed. */ expand = (STAllocation**)realloc(aRun->mAllocations, sizeof(STAllocation*) * (aRun->mAllocationCount + 1)); if(NULL != expand) { /* ** Reassign in case of pointer move. */ aRun->mAllocations = expand; /* ** Stick the allocation in. */ aRun->mAllocations[aRun->mAllocationCount] = aAllocation; /* ** If this is the global run, we need to let the allocation ** track the index back to us. */ if(&globals.mRun == aRun) { aAllocation->mRunIndex = aRun->mAllocationCount; } /* ** Increase the count. */ aRun->mAllocationCount++; /* ** We're good. */ retval = __LINE__; /* ** update allocation cost */ recalculateAllocationCost(aRun, aAllocation); } else { REPORT_ERROR(__LINE__, appendAllocation); } } else { REPORT_ERROR(__LINE__, appendAllocation); } return retval; } /* ** hasCallsiteMatch ** ** Determine if the callsite or the other callsites has the matching text. ** ** Returns 0 if there is no match. */ int hasCallsiteMatch(tmcallsite* aCallsite, const char* aMatch, int aDirection) { int retval = 0; if(NULL != aCallsite && NULL != aCallsite->method && NULL != aMatch && '\0' != *aMatch) { const char* methodName = NULL; do { methodName = tmmethodnode_name(aCallsite->method); if(NULL != methodName && NULL != strstr(methodName, aMatch)) { /* ** Contains the text. */ retval = __LINE__; break; } else { switch(aDirection) { case ST_FOLLOW_SIBLINGS: aCallsite = aCallsite->siblings; break; case ST_FOLLOW_PARENTS: aCallsite = aCallsite->parent; break; default: aCallsite = NULL; REPORT_ERROR(__LINE__, hasCallsiteMatch); break; } } } while(NULL != aCallsite && NULL != aCallsite->method); } else { REPORT_ERROR(__LINE__, hasCallsiteMatch); } return retval; } /* ** harvestRun ** ** Provide a simply way to go over a run, and yield the relevant allocations. ** The restrictions are easily set via the options page or the command ** line switches. ** ** On any match, add the allocation to the provided run. ** ** This makes it much easier for all the code to respect the options in ** force. ** ** You can override the global options by passing in your own options ** pointer if you need a custom harvest. ** ** Returns !0 on error, though aOutRun may contain a partial data set. */ int harvestRun(const STRun* aInRun, STRun* aOutRun, STOptions* aOptions) { int retval = 0; #if defined(DEBUG_dp) PRIntervalTime start = PR_IntervalNow(); fprintf(stderr, "DEBUG: harvesting run...\n"); #endif if(NULL != aInRun && NULL != aOutRun && aInRun != aOutRun) { PRUint32 traverse = 0; STAllocation* current = NULL; /* ** Fixup options if not provided. */ if(NULL == aOptions) { aOptions = &globals.mOptions; } for(traverse = 0; 0 == retval && traverse < aInRun->mAllocationCount; traverse++) { current = aInRun->mAllocations[traverse]; if(NULL != current) { PRUint32 lifetime = 0; PRUint32 bytesize = 0; PRUint64 weight64 = LL_INIT(0, 0); PRUint64 bytesize64 = LL_INIT(0, 0); PRUint64 lifetime64 = LL_INIT(0, 0); int appendRes = 0; int looper = 0; PRBool matched = PR_FALSE; /* ** Use this as an opportune time to fixup a memory ** leaked timeval, so as to not completely skew ** the weights. */ if(ST_TIMEVAL_MAX == current->mMaxTimeval) { current->mMaxTimeval = globals.mMaxTimeval; } /* ** Check allocation timeval restrictions. ** We have to slide the recorded timevals to be zero ** based, so that the comparisons make sense. */ if(aOptions->mAllocationTimevalMin > (current->mMinTimeval - globals.mMinTimeval)) { continue; } else if(aOptions->mAllocationTimevalMax < (current->mMinTimeval - globals.mMinTimeval)) { continue; } /* ** Check timeval restrictions. ** We have to slide the recorded timevals to be zero ** based, so that the comparisons make sense. */ if(aOptions->mTimevalMin > (current->mMaxTimeval - globals.mMinTimeval)) { continue; } else if(aOptions->mTimevalMax < (current->mMinTimeval - globals.mMinTimeval)) { continue; } /* ** Check lifetime restrictions. */ lifetime = current->mMaxTimeval - current->mMinTimeval; if(lifetime < aOptions->mLifetimeMin) { continue; } else if(lifetime > aOptions->mLifetimeMax) { continue; } /* ** Check byte size restrictions. */ bytesize = byteSize(current); if(bytesize < aOptions->mSizeMin) { continue; } else if(bytesize > aOptions->mSizeMax) { continue; } /* ** Check weight restrictions. */ LL_UI2L(bytesize64, bytesize); LL_UI2L(lifetime64, lifetime); LL_MUL(weight64, bytesize64, lifetime64); if(LL_UCMP(weight64, <, aOptions->mWeightMin64)) { continue; } else if(LL_UCMP(weight64, >, aOptions->mWeightMax64)) { continue; } /* ** Possibly restrict the callsite by text. ** Do this last, as it is a heavier check. ** ** One day, we may need to expand the logic to check for ** events beyond the initial allocation event. */ for(looper = 0; ST_SUBSTRING_MATCH_MAX > looper; looper++) { if(NULL != globals.mOptions.mRestrictText[looper] && '\0' != globals.mOptions.mRestrictText[looper][0]) { if(0 == hasCallsiteMatch(current->mEvents[0].mCallsite, globals.mOptions.mRestrictText[looper], ST_FOLLOW_PARENTS)) { break; } } else { matched = PR_TRUE; break; } } if(ST_SUBSTRING_MATCH_MAX == looper) { matched = PR_TRUE; } if(PR_FALSE == matched) { continue; } /* ** You get here, we add to the run. */ appendRes = appendAllocation(aOutRun, current); if(0 == appendRes) { retval = __LINE__; REPORT_ERROR(__LINE__, appendAllocation); } } } } #if defined(DEBUG_dp) fprintf(stderr, "DEBUG: harvesting ends: %dms [%d allocations]\n", PR_IntervalToMilliseconds(PR_IntervalNow() - start), aInRun->mAllocationCount); #endif return retval; } /* ** recalculateRunCost ** ** Goes over all allocations of a run and recalculates and propogates ** the allocation costs - weight, heapcount, size */ int recalculateRunCost(STRun* aRun) { PRUint32 traverse = 0; STAllocation* current = NULL; #if defined(DEBUG_dp) PRIntervalTime start = PR_IntervalNow(); fprintf(stderr, "DEBUG: recalculateRunCost...\n"); #endif if (NULL == aRun) return -1; /* reset stats of this run to 0 to begin recalculation */ memset(&aRun->mStats, 0, sizeof(STCallsiteStats)); /* reset timestamp to force propogation of cost */ aRun->mStats.mStamp = PR_IntervalNow(); for(traverse = 0; traverse < aRun->mAllocationCount; traverse++) { current = aRun->mAllocations[traverse]; if(NULL != current) { recalculateAllocationCost(aRun, current); } } #if defined(DEBUG_dp) fprintf(stderr, "DEBUG: recalculateRunCost ends: %dms [%d allocations]\n", PR_IntervalToMilliseconds(PR_IntervalNow() - start), aRun->mAllocationCount); #endif return 0; } /* ** compareAllocations ** ** qsort callback. ** Compare the allocations as specified by the options. */ int compareAllocations(const void* aAlloc1, const void* aAlloc2, void* aContext) { int retval = 0; if(NULL != aAlloc1 && NULL != aAlloc2) { STAllocation* alloc1 = *((STAllocation**)aAlloc1); STAllocation* alloc2 = *((STAllocation**)aAlloc2); if(NULL != alloc1 && NULL != alloc2) { /* ** Logic determined by pref/option. */ switch(globals.mOptions.mOrderBy) { case ST_COUNT: /* ** "By count" on a single allocation means nothing, ** fall through to weight. */ case ST_WEIGHT: { PRUint64 weight164 = LL_INIT(0, 0); PRUint64 weight264 = LL_INIT(0, 0); PRUint64 bytesize164 = LL_INIT(0, 0); PRUint64 bytesize264 = LL_INIT(0, 0); PRUint64 timeval164 = LL_INIT(0, 0); PRUint64 timeval264 = LL_INIT(0, 0); LL_UI2L(bytesize164, byteSize(alloc1)); LL_UI2L(timeval164, (alloc1->mMaxTimeval - alloc1->mMinTimeval)); LL_MUL(weight164, bytesize164, timeval164); LL_UI2L(bytesize264, byteSize(alloc2)); LL_UI2L(timeval264, (alloc2->mMaxTimeval - alloc2->mMinTimeval)); LL_MUL(weight264, bytesize264, timeval264); if(LL_UCMP(weight164, <, weight264)) { retval = __LINE__; } else if(LL_UCMP(weight164, >, weight264)) { retval = - __LINE__; } } break; case ST_SIZE: { PRUint32 size1 = byteSize(alloc1); PRUint32 size2 = byteSize(alloc2); if(size1 < size2) { retval = __LINE__; } else if(size1 > size2) { retval = - __LINE__; } } break; case ST_TIMEVAL: { PRUint32 timeval1 = (alloc1->mMaxTimeval - alloc1->mMinTimeval); PRUint32 timeval2 = (alloc2->mMaxTimeval - alloc2->mMinTimeval); if(timeval1 < timeval2) { retval = __LINE__; } else if(timeval1 > timeval2) { retval = - __LINE__; } } break; case ST_HEAPCOST: { PRUint32 cost1 = alloc1->mHeapRuntimeCost; PRUint32 cost2 = alloc2->mHeapRuntimeCost; if(cost1 < cost2) { retval = __LINE__; } else if(cost1 > cost2) { retval = - __LINE__; } } break; default: { REPORT_ERROR(__LINE__, compareAllocations); } break; } } } return retval; } /* ** sortRun ** ** Given a run, sort it in the manner specified by the options. ** Returns !0 on failure. */ int sortRun(STRun* aRun) { int retval = 0; if(NULL != aRun) { if(NULL != aRun->mAllocations && 0 < aRun->mAllocationCount) { NS_QuickSort(aRun->mAllocations, aRun->mAllocationCount, sizeof(STAllocation*), compareAllocations, NULL); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, sortRun); } return retval; } /* ** createRun ** ** Returns a newly allocated run, properly initialized. ** Must call freeRun() with the new STRun. ** ** ONLY PASS IN A NON_ZERO STAMP IF YOU KNOW WHAT YOU ARE DOING!!! ** A non zero stamp in a run has side effects all over the ** callsites of the allocations added to the run and their ** parents. ** ** Returns NULL on failure. */ STRun* createRun(PRUint32 aStamp) { STRun* retval = NULL; retval = (STRun*)calloc(1, sizeof(STRun)); if(NULL != retval) { retval->mStats.mStamp = aStamp; } return retval; } /* ** freeRun ** ** Free off the run and the associated data. */ void freeRun(STRun* aRun) { if(NULL != aRun) { if(NULL != aRun->mAllocations) { /* ** We do not free the allocations themselves. ** They are likely pointed to by at least 2 other existing ** runs. */ free(aRun->mAllocations); aRun->mAllocations = NULL; } free(aRun); aRun = NULL; } } /* ** createRunFromGlobal ** ** Harvest the global run, then sort it. ** Returns NULL on failure. ** Must call freeRun() with the new STRun. */ STRun* createRunFromGlobal(void) { STRun* retval = NULL; /* ** We stamp the run. ** As things are appended to it, it realizes that it should stamp the ** callsite backtrace with the information as well. ** In this manner, we can provide meaningful callsite data. */ retval = createRun(PR_IntervalNow()); if(NULL != retval) { int failure = 0; int harvestRes = harvestRun(&globals.mRun, retval, NULL); if(0 == harvestRes) { int sortRes = sortRun(retval); if(0 != sortRes) { failure = __LINE__; } } else { failure = __LINE__; } if(0 != failure) { freeRun(retval); retval = NULL; REPORT_ERROR(failure, createRunFromGlobal); } /* ** Categorize the run */ failure = categorizeRun(retval, &globals); if (0 != failure) { REPORT_ERROR(__LINE__, appendAllocation); } /* ** if we are focussing on a category, return that run instead of ** the harvested run. Make sure to recalculate cost. */ if (globals.mOptions.mCategoryName) { STCategoryNode* node = findCategoryNode(globals.mOptions.mCategoryName, &globals); if (node && node->run) { /* Recalculate cost of run */ recalculateRunCost(node->run); retval = node->run; } } } return retval; } /* ** getLiveAllocationByHeapID ** ** Go through a run and find the right heap ID. ** At the time of the call to this function, the allocation must be LIVE, ** meaning that it can not be freed. ** Go through the run backwards, in hopes of finding it near the end. ** ** Returns the allocation on success, otherwise NULL. */ STAllocation* getLiveAllocationByHeapID(STRun* aRun, PRUint32 aHeapID) { STAllocation* retval = NULL; if(NULL != aRun && 0 != aHeapID) { PRUint32 traverse = aRun->mAllocationCount; STAllocation* eval = NULL; /* ** Go through in reverse order. ** Stop when we have a return value. */ while(0 < traverse && NULL == retval) { /* ** Back up one to align with zero based index. */ traverse--; /* ** Take the pointer math out of further operations. */ eval = aRun->mAllocations[traverse]; /* ** Take a look at the events in reverse order. ** Basically the last event must NOT be a free. ** The last event must NOT be a realloc of size zero (free). ** Otherwise, try to match up the heapID of the event. */ if(0 != eval->mEventCount) { STAllocEvent* event = eval->mEvents + (eval->mEventCount - 1); switch(event->mEventType) { case TM_EVENT_FREE: { /* ** No freed allocation can match. */ } break; case TM_EVENT_REALLOC: case TM_EVENT_CALLOC: case TM_EVENT_MALLOC: { /* ** Heap IDs must match. */ if(aHeapID == event->mHeapID) { retval = eval; } } break; default: { REPORT_ERROR(__LINE__, getAllocationByHeapID); } break; } } } } else { REPORT_ERROR(__LINE__, getAllocationByHeapID); } return retval; } /* ** appendEvent ** ** Given an allocation, append a new event to it's lifetime. ** Returns the new event on success, otherwise NULL. */ STAllocEvent* appendEvent(STAllocation* aAllocation, PRUint32 aTimeval, char aEventType, PRUint32 aHeapID, PRUint32 aHeapSize, tmcallsite* aCallsite) { STAllocEvent* retval = NULL; if(NULL != aAllocation && NULL != aCallsite) { STAllocEvent* expand = NULL; /* ** Expand the allocation's event array. */ expand = (STAllocEvent*)realloc(aAllocation->mEvents, sizeof(STAllocEvent) * (aAllocation->mEventCount + 1)); if(NULL != expand) { /* ** Reassign in case of pointer move. */ aAllocation->mEvents = expand; /* ** Remove the pointer math from rest of code. */ retval = aAllocation->mEvents + aAllocation->mEventCount; /* ** Increase event array count. */ aAllocation->mEventCount++; /* ** Fill in the event. */ retval->mTimeval = aTimeval; retval->mEventType = aEventType; retval->mHeapID = aHeapID; retval->mHeapSize = aHeapSize; retval->mCallsite = aCallsite; /* ** Allocation may need to update idea of lifetime. ** See allocationTracker to see mMinTimeval inited to ST_TIMEVAL_MAX. */ if(aAllocation->mMinTimeval > aTimeval) { aAllocation->mMinTimeval = aTimeval; } /* ** This a free event? ** Can only set max timeval on a free. ** Otherwise, mMaxTimeval remains ST_TIMEVAL_MAX. ** Set in allocationTracker. */ if(TM_EVENT_FREE == aEventType) { aAllocation->mMaxTimeval = aTimeval; } } else { REPORT_ERROR(__LINE__, appendEvent); } } else { REPORT_ERROR(__LINE__, appendEvent); } return retval; } /* ** hasAllocation ** ** Determine if a given run has an allocation. ** This is really nothing more than a pointer comparison loop. ** Returns !0 if the run has the allocation. */ int hasAllocation(STRun* aRun, STAllocation* aTestFor) { int retval = 0; if(NULL != aRun && NULL != aTestFor) { PRUint32 traverse = aRun->mAllocationCount; /* ** Go through reverse, in the hopes it exists nearer the end. */ while(0 < traverse) { /* ** Back up. */ traverse--; if(aTestFor == aRun->mAllocations[traverse]) { retval = __LINE__; break; } } } else { REPORT_ERROR(__LINE__, hasAllocation); } return retval; } /* ** allocationTracker ** ** Important to keep track of all allocations unique so as to determine ** their lifetimes. ** ** Returns a pointer to the allocation on success. ** Return NULL on failure. */ STAllocation* allocationTracker(PRUint32 aTimeval, char aType, PRUint32 aHeapRuntimeCost, tmcallsite* aCallsite, PRUint32 aHeapID, PRUint32 aSize, tmcallsite* aOldCallsite, PRUint32 aOldHeapID, PRUint32 aOldSize) { STAllocation* retval = NULL; static int compactor = 1; const int frequency = 10000; PRUint32 actualSize, actualOldSize = 0; actualSize = actualByteSize(aSize); if (aOldSize) actualOldSize = actualByteSize(aOldSize); if(NULL != aCallsite) { int newAllocation = 0; tmcallsite* searchCallsite = NULL; PRUint32 searchHeapID = 0; STAllocation* allocation = NULL; /* ** Global operation ID increases. */ globals.mOperationCount++; /* ** Fix up the timevals if needed. */ if(aTimeval < globals.mMinTimeval) { globals.mMinTimeval = aTimeval; } if(aTimeval > globals.mMaxTimeval) { globals.mMaxTimeval = aTimeval; } switch(aType) { case TM_EVENT_FREE: { /* ** Update the global counter. */ globals.mFreeCount++; /* ** Update our peak memory used counter */ globals.mMemoryUsed -= actualSize; /* ** Not a new allocation, will need to search passed in site ** for the original allocation. */ searchCallsite = aCallsite; searchHeapID = aHeapID; } break; case TM_EVENT_MALLOC: { /* ** Update the global counter. */ globals.mMallocCount++; /* ** Update our peak memory used counter */ globals.mMemoryUsed += actualSize; if (globals.mMemoryUsed > globals.mPeakMemoryUsed) { globals.mPeakMemoryUsed = globals.mMemoryUsed; } /* ** This will be a new allocation. */ newAllocation = __LINE__; } break; case TM_EVENT_CALLOC: { /* ** Update the global counter. */ globals.mCallocCount++; /* ** Update our peak memory used counter */ globals.mMemoryUsed += actualSize; if (globals.mMemoryUsed > globals.mPeakMemoryUsed) { globals.mPeakMemoryUsed = globals.mMemoryUsed; } /* ** This will be a new allocation. */ newAllocation = __LINE__; } break; case TM_EVENT_REALLOC: { /* ** Update the global counter. */ globals.mReallocCount++; /* ** Update our peak memory used counter */ globals.mMemoryUsed += actualSize - actualOldSize; if (globals.mMemoryUsed > globals.mPeakMemoryUsed) { globals.mPeakMemoryUsed = globals.mMemoryUsed; } /* ** This might be a new allocation. */ if(NULL == aOldCallsite) { newAllocation = __LINE__; } else { /* ** Need to search for the original callsite for the ** index to the allocation. */ searchCallsite = aOldCallsite; searchHeapID = aOldHeapID; } } break; default: { REPORT_ERROR(__LINE__, allocationTracker); } break; } /* ** We are either modifying an existing allocation or we are creating ** a new one. */ if(0 != newAllocation) { allocation = (STAllocation*)calloc(1, sizeof(STAllocation)); if(NULL != allocation) { /* ** Fixup the min timeval so if logic later will just work. */ allocation->mMinTimeval = ST_TIMEVAL_MAX; allocation->mMaxTimeval = ST_TIMEVAL_MAX; } } else if(NULL != searchCallsite && NULL != CALLSITE_RUN(searchCallsite) && 0 != searchHeapID) { /* ** We know what to search for, and we reduce what we search ** by only looking for those allocations at a known callsite. */ allocation = getLiveAllocationByHeapID(CALLSITE_RUN(searchCallsite), searchHeapID); } else { REPORT_ERROR(__LINE__, allocationTracker); } if(NULL != allocation) { STAllocEvent* appendResult = NULL; /* ** Record the amount of time this allocation event took. */ allocation->mHeapRuntimeCost += aHeapRuntimeCost; /* ** Now that we have an allocation, we need to make sure it has ** the proper event. */ appendResult = appendEvent(allocation, aTimeval, aType, aHeapID, aSize, aCallsite); if(NULL != appendResult) { if(0 != newAllocation) { int runAppendResult = 0; int callsiteAppendResult = 0; /* ** A new allocation needs to be added to the global run. ** A new allocation needs to be added to the callsite. */ runAppendResult = appendAllocation(&(globals.mRun), allocation); callsiteAppendResult = appendAllocation(CALLSITE_RUN(aCallsite), allocation); if(0 != runAppendResult && 0 != callsiteAppendResult) { /* ** Success. */ retval = allocation; } else { REPORT_ERROR(__LINE__, appendAllocation); } } else { /* ** An existing allocation, if a realloc situation, ** may need to be added to the new callsite. ** This can only occur if the new and old callsites ** differ. ** Even then, a brute force check will need to be made ** to ensure the allocation was not added twice; ** consider a realloc scenario where two different ** call stacks bump the allocation back and forth. */ if(aCallsite != searchCallsite) { int found = 0; found = hasAllocation(CALLSITE_RUN(aCallsite), allocation); if(0 == found) { int appendResult = 0; appendResult = appendAllocation(CALLSITE_RUN(aCallsite), allocation); if(0 != appendResult) { /* ** Success. */ retval = allocation; } else { REPORT_ERROR(__LINE__, appendAllocation); } } else { /* ** Already there. */ retval = allocation; } } else { /* ** Success. */ retval = allocation; } } } else { REPORT_ERROR(__LINE__, appendEvent); } } else { REPORT_ERROR(__LINE__, allocationTracker); } } else { REPORT_ERROR(__LINE__, allocationTracker); } /* ** Compact the heap a bit if you can. */ compactor++; if(0 == (compactor % frequency)) { heapCompact(); } return retval; } /* ** trackEvent ** ** An allocation event has dropped in on us. ** We need to do the right thing and track it. */ void trackEvent(PRUint32 aTimeval, char aType, PRUint32 aHeapRuntimeCost, tmcallsite* aCallsite, PRUint32 aHeapID, PRUint32 aSize, tmcallsite* aOldCallsite, PRUint32 aOldHeapID, PRUint32 aOldSize) { if(NULL != aCallsite) { /* ** Verify the old callsite just in case. */ if(NULL != CALLSITE_RUN(aCallsite) && (NULL == aOldCallsite || NULL != CALLSITE_RUN(aOldCallsite))) { STAllocation* allocation = NULL; /* ** Add to the allocation tracking code. */ allocation = allocationTracker(aTimeval, aType, aHeapRuntimeCost, aCallsite, aHeapID, aSize, aOldCallsite, aOldHeapID, aOldSize); if(NULL == allocation) { REPORT_ERROR(__LINE__, allocationTracker); } } else { REPORT_ERROR(__LINE__, trackEvent); } } else { REPORT_ERROR(__LINE__, trackEvent); } } /* ** tmEventHandler ** ** Callback from the tmreader_eventloop function. ** Simply tries to sort out what we desire to know. */ void tmEventHandler(tmreader* aReader, tmevent* aEvent) { if(NULL != aReader && NULL != aEvent) { switch(aEvent->type) { /* ** Events we ignore. */ case TM_EVENT_LIBRARY: case TM_EVENT_METHOD: case TM_EVENT_STATS: case TM_EVENT_TIMESTAMP: case TM_EVENT_FILENAME: break; /* ** Allocation events need to be tracked. */ case TM_EVENT_MALLOC: case TM_EVENT_CALLOC: case TM_EVENT_REALLOC: case TM_EVENT_FREE: { PRUint32 oldptr = 0; PRUint32 oldsize = 0; tmcallsite* callsite = NULL; tmcallsite* oldcallsite = NULL; if(TM_EVENT_REALLOC == aEvent->type) { /* ** Only care about old arguments if there were any. */ if(0 != aEvent->u.alloc.oldserial) { oldptr = aEvent->u.alloc.oldptr; oldsize = aEvent->u.alloc.oldsize; oldcallsite = tmreader_callsite(aReader, aEvent->u.alloc.oldserial); if(NULL == oldcallsite) { REPORT_ERROR(__LINE__, tmreader_callsite); } } } callsite = tmreader_callsite(aReader, aEvent->serial); if(NULL != callsite) { /* ** Verify a callsite run is there. ** If not, we are ignoring this callsite. */ if(NULL != CALLSITE_RUN(callsite)) { char eventType = aEvent->type; /* ** Play a nasty trick on reallocs of size zero. ** They are to become free events. ** This allows me to avoid all types of special case code. */ if(0 == aEvent->u.alloc.size && TM_EVENT_REALLOC == aEvent->type) { eventType = TM_EVENT_FREE; } trackEvent(ticks2msec(aReader, aEvent->u.alloc.interval), eventType, ticks2usec(aReader, aEvent->u.alloc.cost), callsite, aEvent->u.alloc.ptr, aEvent->u.alloc.size, oldcallsite, oldptr, oldsize); } } else { REPORT_ERROR(__LINE__, tmreader_callsite); } } break; /* ** Callsite, set up the callsite run if it does not exist. */ case TM_EVENT_CALLSITE: { tmcallsite* callsite = tmreader_callsite(aReader, aEvent->serial); if(NULL != callsite) { if(NULL == CALLSITE_RUN(callsite)) { int createrun = __LINE__; #if defined(MOZILLA_CLIENT) /* ** For a mozilla spacetrace, ignore this particular ** callsite as it is just noise, and causes us to ** use a lot of memory. ** ** This callsite is present on the linux build, ** not sure if the other platforms have it. */ if(0 != hasCallsiteMatch(callsite, "g_main_is_running", ST_FOLLOW_PARENTS)) { createrun = 0; } #endif /* MOZILLA_CLIENT */ if(0 != createrun) { callsite->data = createRun(0); } } } else { REPORT_ERROR(__LINE__, tmreader_callsite); } } break; /* ** Unhandled events should not be allowed. */ default: { REPORT_ERROR(__LINE__, tmEventHandler); } break; } } } /* ** htmlAnchor ** ** Output an HTML anchor, or just the text depending on the mode. */ void htmlAnchor(const char* aHref, const char* aText, const char* aTarget) { if(NULL == aTarget || '\0' == aTarget[0]) { aTarget = "_st_content"; } if(NULL != aHref && '\0' != *aHref && NULL != aText && '\0' != *aText) { int anchorLive = 1; /* ** In batch mode, we need to verify the anchor is live. */ if(0 != globals.mOptions.mBatchRequestCount) { PRUint32 loop = 0; int comparison = 1; for(loop = 0; loop < globals.mOptions.mBatchRequestCount; loop++) { comparison = strcmp(aHref, globals.mOptions.mBatchRequests[loop]); if(0 == comparison) { break; } } /* ** Did we find it? */ if(0 == comparison) { anchorLive = 0; } } /* ** In any mode, don't make an href to the current page. */ if(0 != anchorLive && NULL != globals.mRequest.mFileName) { if(0 == strcmp(aHref, globals.mRequest.mFileName)) { anchorLive = 0; } } /* ** Do the right thing. */ if(0 != anchorLive) { PR_fprintf(globals.mRequest.mFD, "%s\n", aTarget, aHref, aText); } else { PR_fprintf(globals.mRequest.mFD, "%s\n", aText); } } else { REPORT_ERROR(__LINE__, htmlAnchor); } } /* ** htmlAllocationAnchor ** ** Output an html achor that will resolve to the allocation in question. */ void htmlAllocationAnchor(STAllocation* aAllocation, const char* aText) { if(NULL != aAllocation && NULL != aText && '\0' != *aText) { char buffer[128]; /* ** This is a total hack. ** The filename contains the index of the allocation in globals.mRun. ** Safer than using the raw pointer value.... */ PR_snprintf(buffer, sizeof(buffer), "allocation_%u.html", aAllocation->mRunIndex); htmlAnchor(buffer, aText, NULL); } else { REPORT_ERROR(__LINE__, htmlAllocationAnchor); } } /* ** resolveSourceFile ** ** Easy way to get a readable/short name. ** NULL if not present, not resolvable. */ const char* resolveSourceFile(tmmethodnode* aMethod) { const char* retval = NULL; if(NULL != aMethod) { const char* methodSays = NULL; methodSays = aMethod->sourcefile; if(NULL != methodSays && '\0' != methodSays[0] && 0 != strcmp("noname", methodSays)) { retval = strrchr(methodSays, '/'); if(NULL != retval) { retval++; } else { retval = methodSays; } } } return retval; } /* ** htmlCallsiteAnchor ** ** Output an html anchor that will resolve to the callsite in question. ** If no text is provided, we provide our own. ** ** RealName determines wether or not we crawl our parents until the point ** we no longer match stats. */ void htmlCallsiteAnchor(tmcallsite* aCallsite, const char* aText, int aRealName) { if(NULL != aCallsite) { char textBuf[512]; char hrefBuf[128]; tmcallsite* namesite = aCallsite; /* ** Should we use a different name? */ if(0 == aRealName && NULL != namesite->parent && NULL != namesite->parent->method) { STRun* myRun = NULL; STRun* upRun = NULL; do { myRun = CALLSITE_RUN(namesite); upRun = CALLSITE_RUN(namesite->parent); if(0 != memcmp(&myRun->mStats, &upRun->mStats, sizeof(STCallsiteStats))) { /* ** Doesn't match, stop. */ break; } else { /* ** Matches, keep going up. */ namesite = namesite->parent; } } while(NULL != namesite->parent && NULL != namesite->parent->method); } /* ** If no text, provide our own. */ if(NULL == aText || '\0' == *aText) { const char* methodName = NULL; const char* sourceFile = NULL; if(NULL != namesite->method) { methodName = tmmethodnode_name(namesite->method); } else { methodName = "==NONAME=="; } /* ** Decide which format to use to identify the callsite. ** If we can detect availability, hook up the filename with lxr information. */ sourceFile = resolveSourceFile(namesite->method); if(NULL != sourceFile && 0 == strncmp("mozilla/", namesite->method->sourcefile, 8)) { char lxrHREFBuf[512]; PR_snprintf(lxrHREFBuf, sizeof(lxrHREFBuf), "(%s:%u)", namesite->method->sourcefile + 8, namesite->method->linenumber, sourceFile, namesite->method->linenumber); PR_snprintf(textBuf, sizeof(textBuf), "%s%s", methodName, lxrHREFBuf); } else if(NULL != sourceFile) { PR_snprintf(textBuf, sizeof(textBuf), "%s(%s:%u)", methodName, sourceFile, namesite->method->linenumber); } else { PR_snprintf(textBuf, sizeof(textBuf), "%s+%u(%u)", methodName, namesite->offset, (PRUint32)namesite->entry.key); } aText = textBuf; } PR_snprintf(hrefBuf, sizeof(hrefBuf), "callsite_%u.html", (PRUint32)aCallsite->entry.key); htmlAnchor(hrefBuf, aText, NULL); } else { REPORT_ERROR(__LINE__, htmlCallsiteAnchor); } } /* ** htmlHeader ** ** Output a standard header in the report files. */ void htmlHeader(const char* aTitle) { PR_fprintf(globals.mRequest.mFD, "\n" "\n" "%s\n" "\n" "\n" "
\n" "\n" , aTitle, globals.mOptions.mCategoryName); PR_fprintf(globals.mRequest.mFD,"\n"); PR_fprintf(globals.mRequest.mFD,"\n"); PR_fprintf(globals.mRequest.mFD,"
Category: %s"); htmlAnchor("index.html", "[Index]", NULL); PR_fprintf(globals.mRequest.mFD,""); htmlAnchor("options.html", "[Options]", NULL); PR_fprintf(globals.mRequest.mFD,"
\n"); /* ** This is a dubious feature at best. */ #if WANT_QUIT htmlAnchor("quit.html", "[Quit]", NULL); #endif PR_fprintf(globals.mRequest.mFD, "
\n
\n"); } /* ** htmlFooter ** ** Output a standard footer in the report file. */ void htmlFooter(void) { PR_fprintf(globals.mRequest.mFD, "
\n" "
\n" "SpaceTrace\n" "
\n" "\n" "\n" ); } /* ** htmlNotFound ** ** Not found message. */ void htmlNotFound(void) { htmlHeader("File Not Found"); PR_fprintf(globals.mRequest.mFD, "File Not Found\n"); htmlFooter(); } /* ** callsiteArrayFromCallsite ** ** Simply return an array of the callsites divulged from the site passed in, ** including the site passed in. ** Do not worry about dups, or the order of the items. ** ** Returns the number of items in the array. ** If the same as aExistingCount, then nothing happened. */ PRUint32 callsiteArrayFromCallsite(tmcallsite*** aArray, PRUint32 aExistingCount, tmcallsite* aSite, int aFollow) { PRUint32 retval = 0; if(NULL != aArray && NULL != aSite) { tmcallsite** expand = NULL; /* ** If we have an existing count, we just keep expanding this. */ retval = aExistingCount; /* ** Go through every allocation. */ do { /* ** expand the array. */ expand = (tmcallsite**)realloc(*aArray, sizeof(tmcallsite*) * (retval + 1)); if(NULL != expand) { /* ** Set the callsite in case of pointer move. */ *aArray = expand; /* ** Assign the value. */ (*aArray)[retval] = aSite; retval++; } else { REPORT_ERROR(__LINE__, realloc); break; } /* ** What do we follow? */ switch(aFollow) { case ST_FOLLOW_SIBLINGS: aSite = aSite->siblings; break; case ST_FOLLOW_PARENTS: aSite = aSite->parent; break; default: aSite = NULL; REPORT_ERROR(__LINE__, callsiteArrayFromCallsite); break; } } while(NULL != aSite && NULL != aSite->method); } return retval; } /* ** callsiteArrayFromRun ** ** Simply return an array of the callsites from the run allocations. ** We only pay attention to callsites that were not free callsites. ** Do not worry about dups, or the order of the items. ** ** Returns the number of items in the array. ** If the same as aExistingCount, then nothing happened. */ PRUint32 callsiteArrayFromRun(tmcallsite*** aArray, PRUint32 aExistingCount, STRun* aRun) { PRUint32 retval = 0; if(NULL != aArray && NULL != aRun && 0 < aRun->mAllocationCount) { PRUint32 allocLoop = 0; PRUint32 eventLoop = 0; int stopLoops = 0; /* ** If we have an existing count, we just keep expanding this. */ retval = aExistingCount; /* ** Go through every allocation. */ for(allocLoop = 0; 0 == stopLoops && allocLoop < aRun->mAllocationCount; allocLoop++) { /* ** Go through every event. */ for(eventLoop = 0; 0 == stopLoops && eventLoop < aRun->mAllocations[allocLoop]->mEventCount; eventLoop++) { /* ** Skip the free events. */ if(TM_EVENT_FREE != aRun->mAllocations[allocLoop]->mEvents[eventLoop].mEventType) { tmcallsite** expand = NULL; /* ** expand the array. */ expand = (tmcallsite**)realloc(*aArray, sizeof(tmcallsite*) * (retval + 1)); if(NULL != expand) { /* ** Set the callsite in case of pointer move. */ *aArray = expand; /* ** Assign the value. */ (*aArray)[retval] = aRun->mAllocations[allocLoop]->mEvents[eventLoop].mCallsite; retval++; } else { REPORT_ERROR(__LINE__, realloc); stopLoops = __LINE__; } } } } } return retval; } /* ** getDataPRUint* ** ** Helper to avoid cut and paste code. ** Failure to find aCheckFor does not mean failure. ** Returns !0 on failure. */ int getDataPRUint32Base(const char* aGetData, const char* aCheckFor, void* aStoreResult, PRUint32 aBits) { int retval = 0; if(NULL != aGetData && NULL != aCheckFor && NULL != aStoreResult) { const char* found = NULL; PRInt32 scanRes = 0; /* ** Looking for the presence. */ found = strstr(aGetData, aCheckFor); if(NULL != found) { int length = 0; const char* dataPoint = NULL; /* ** Skip the varname. ** Skip the '=' sign. */ length = strlen(aCheckFor); dataPoint = found + length + 1; /* ** Just attempt to scan from here. */ if(64 == aBits) { scanRes = PR_sscanf(dataPoint, "%llu", aStoreResult); } else { scanRes = PR_sscanf(dataPoint, "%u", aStoreResult); } if(1 != scanRes) { retval = __LINE__; REPORT_ERROR(__LINE__, PR_sscanf); } } } else { retval = __LINE__; REPORT_ERROR(__LINE__, getDataPRUint32); } return retval; } int getDataPRUint32(const char* aGetData, const char* aCheckFor, PRUint32* aStoreResult, int* aChanged, PRUint32 aConversion) { int retval = 0; PRUint32 value = *aStoreResult; retval = getDataPRUint32Base(aGetData, aCheckFor, aStoreResult, 32); *aStoreResult *= aConversion; if(NULL != aChanged && value != *aStoreResult) { (*aChanged) = (*aChanged) + 1; } return retval; } int getDataPRUint64(const char* aGetData, const char* aCheckFor, PRUint64* aStoreResult64, int* aChanged) { int retval = 0; PRUint64 value64 = *aStoreResult64; retval = getDataPRUint32Base(aGetData, aCheckFor, aStoreResult64, 64); if(NULL != aChanged && LL_NE(value64, *aStoreResult64)) { (*aChanged) = (*aChanged) + 1; } return retval; } /* ** getDataString ** ** Pull out the string data, if specified. ** Return !0 on failure. */ int getDataString(const char* aGetData, const char* aCheckFor, char** aStoreResult, int* aChanged) { int retval = 0; if(NULL != aGetData && NULL != aCheckFor && NULL != aStoreResult) { const char* found = NULL; PRInt32 scanRes = 0; /* ** Check for presence. */ found = strstr(aGetData, aCheckFor); if(NULL != found) { int length = 0; const char* dataPoint = NULL; const char* endPoint = NULL; char* oldResult = NULL; int theLen = 0; /* ** Skip the varname. ** Skip the '=' sign. */ length = strlen(aCheckFor); dataPoint = found + length + 1; /* ** The length is up to a '&' or until end of string. */ endPoint = strchr(dataPoint, '&'); if(NULL == endPoint) { endPoint = dataPoint + strlen(dataPoint); } /* ** Store original value if present. */ if(NULL != *aStoreResult) { oldResult = *aStoreResult; } /* ** Allocate space for new string. */ theLen = (int)(endPoint - dataPoint); *aStoreResult = (char*)malloc((size_t)(theLen + 1)); if(NULL != *aStoreResult) { int index1 = 0; int index2 = 0; strncpy(*aStoreResult, dataPoint, theLen); (*aStoreResult)[theLen] = '\0'; /* ** Here's a totally suboptimal and bug prone unhexcaper. */ for(; index1 <= theLen; index1++) { if('%' == (*aStoreResult)[index1] && '\0' != (*aStoreResult)[index1 + 1] && '\0' != (*aStoreResult)[index1 + 2]) { int unhex = 0; if('9' >= (*aStoreResult)[index1 + 1]) { unhex |= (((*aStoreResult)[index1 + 1] - '0') << 4); } else { unhex |= ((toupper((*aStoreResult)[index1 + 1]) - 'A' + 10) << 4); } if('9' >= (*aStoreResult)[index1 + 2]) { unhex |= ((*aStoreResult)[index1 + 2] - '0'); } else { unhex |= (toupper((*aStoreResult)[index1 + 2]) - 'A' + 10); } index1 += 2; (*aStoreResult)[index1] = unhex; } (*aStoreResult)[index2++] = (*aStoreResult)[index1]; } /* ** See if the value actually changed. */ if(NULL != aChanged) { if(NULL != oldResult) { if(0 != strcmp(oldResult, *aStoreResult)) { (*aChanged) = (*aChanged) + 1; } } else if('\0' != (*aStoreResult)[0]) { (*aChanged) = (*aChanged) + 1; } } /* ** Free off the prior value if relevant. */ if(NULL != oldResult) { free(oldResult); oldResult = NULL; } } else { retval = __LINE__; REPORT_ERROR(__LINE__, malloc); } } } else { retval = __LINE__; REPORT_ERROR(__LINE__, getDataPRUint32); } return retval; } /* ** displayTopAllocations ** ** Present the top allocations. ** The run must be passed in, and it must be pre-sorted. ** ** Returns !0 on failure. */ int displayTopAllocations(STRun* aRun, int aWantCallsite) { int retval = 0; if(NULL != aRun) { if(0 < aRun->mAllocationCount) { PRUint32 loop = 0; STAllocation* current = NULL; PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); if(0 != aWantCallsite) { PR_fprintf(globals.mRequest.mFD, "\n"); } PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Loop over the items, up to some limit or until the end. */ for(loop = 0; loop < globals.mOptions.mListItemMax && loop < aRun->mAllocationCount; loop++) { current = aRun->mAllocations[loop]; if(NULL != current) { PRUint32 lifespan = current->mMaxTimeval - current->mMinTimeval; PRUint32 size = byteSize(current); PRUint32 heapCost = current->mHeapRuntimeCost; PRUint64 weight64 = LL_INIT(0, 0); PRUint64 size64 = LL_INIT(0, 0); PRUint64 lifespan64 = LL_INIT(0, 0); char buffer[32]; LL_UI2L(size64, size); LL_UI2L(lifespan64, lifespan); LL_MUL(weight64, size64, lifespan64); PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Rank. */ PR_fprintf(globals.mRequest.mFD, "\n", loop + 1); /* ** Index. */ PR_snprintf(buffer, sizeof(buffer), "%u", current->mRunIndex); PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Byte Size. */ PR_fprintf(globals.mRequest.mFD, "\n", size); /* ** Lifespan. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_TIMEVAL_PRINTABLE(lifespan)); /* ** Weight. */ PR_fprintf(globals.mRequest.mFD, "\n", weight64); /* ** Heap operation cost. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_MICROVAL_PRINTABLE(heapCost)); if(0 != aWantCallsite) { /* ** Callsite. */ PR_fprintf(globals.mRequest.mFD, "\n"); } PR_fprintf(globals.mRequest.mFD, "\n"); } } PR_fprintf(globals.mRequest.mFD, "
RankIndexByte SizeLifespan SecondsWeightHeap Operation SecondsOrigin Callsite
%u\n"); htmlAllocationAnchor(current, buffer); PR_fprintf(globals.mRequest.mFD, "%u" ST_TIMEVAL_FORMAT "%llu" ST_MICROVAL_FORMAT ""); htmlCallsiteAnchor(current->mEvents[0].mCallsite, NULL, 0); PR_fprintf(globals.mRequest.mFD, "
\n"); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, displayTopAllocations); } return retval; } /* ** displayMemoryLeaks ** ** Present the top memory leaks. ** The run must be passed in, and it must be pre-sorted. ** ** Returns !0 on failure. */ int displayMemoryLeaks(STRun* aRun) { int retval = 0; if(NULL != aRun) { PRUint32 loop = 0; PRUint32 displayed = 0; STAllocation* current = NULL; PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Loop over all of the items, or until we've displayed enough. */ for(loop = 0; displayed < globals.mOptions.mListItemMax && loop < aRun->mAllocationCount; loop++) { current = aRun->mAllocations[loop]; if(NULL != current && 0 != current->mEventCount) { /* ** In order to be a leak, the last event of it's life must ** NOT be a free operation. ** ** A free operation is just that, a free. */ if(TM_EVENT_FREE != current->mEvents[current->mEventCount - 1].mEventType) { PRUint32 lifespan = current->mMaxTimeval - current->mMinTimeval; PRUint32 size = byteSize(current); PRUint32 heapCost = current->mHeapRuntimeCost; PRUint64 weight64 = LL_INIT(0, 0); PRUint64 size64 = LL_INIT(0, 0); PRUint64 lifespan64 = LL_INIT(0, 0); char buffer[32]; LL_UI2L(size64, size); LL_UI2L(lifespan64, lifespan); LL_MUL(weight64, size64, lifespan64); /* ** One more shown. */ displayed++; PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Rank. */ PR_fprintf(globals.mRequest.mFD, "\n", displayed); /* ** Index. */ PR_snprintf(buffer, sizeof(buffer), "%u", current->mRunIndex); PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Byte Size. */ PR_fprintf(globals.mRequest.mFD, "\n", size); /* ** Lifespan. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_TIMEVAL_PRINTABLE(lifespan)); /* ** Weight. */ PR_fprintf(globals.mRequest.mFD, "\n", weight64); /* ** Heap Operation Seconds. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_MICROVAL_PRINTABLE(heapCost)); /* ** Callsite. */ PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); } } } PR_fprintf(globals.mRequest.mFD, "
RankIndexByte SizeLifespan SecondsWeightHeap Operation SecondsOrigin Callsite
%u\n"); htmlAllocationAnchor(current, buffer); PR_fprintf(globals.mRequest.mFD, "%u" ST_TIMEVAL_FORMAT "%llu" ST_MICROVAL_FORMAT ""); htmlCallsiteAnchor(current->mEvents[0].mCallsite, NULL, 0); PR_fprintf(globals.mRequest.mFD, "
\n"); } else { retval = __LINE__; REPORT_ERROR(__LINE__, displayMemoryLeaks); } return retval; } /* ** displayCallsites ** ** Display a table of callsites. ** If the stamp is non zero, then must match that stamp. ** If the stamp is zero, then must match the global sorted run stamp. ** Return !0 on error. */ int displayCallsites(tmcallsite* aCallsite, int aFollow, PRUint32 aStamp, int aRealNames) { int retval = 0; if(NULL != aCallsite && NULL != aCallsite->method) { int headerDisplayed = 0; STRun* run = NULL; /* ** Corrent the stamp if need be. */ if(0 == aStamp && NULL != globals.mCache.mSortedRun) { aStamp = globals.mCache.mSortedRun->mStats.mStamp; } /* ** Loop over the callsites looking for a stamp match. ** A stamp guarantees there is something interesting to look at too. ** If found, output it. */ while(NULL != aCallsite && NULL != aCallsite->method) { run = CALLSITE_RUN(aCallsite); if(NULL != run) { if(aStamp == run->mStats.mStamp) { /* ** We got a header? */ if(0 == headerDisplayed) { headerDisplayed = __LINE__; PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); } /* ** Output the information. */ PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Method name. */ PR_fprintf(globals.mRequest.mFD, ""); /* ** Byte Size. */ PR_fprintf(globals.mRequest.mFD, "\n", run->mStats.mSize); /* ** Seconds. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_TIMEVAL_PRINTABLE64(run->mStats.mTimeval64)); /* ** Weight. */ PR_fprintf(globals.mRequest.mFD, "\n", run->mStats.mWeight64); /* ** Allocation object count. */ PR_fprintf(globals.mRequest.mFD, "\n", run->mStats.mCompositeCount); /* ** Heap Operation Seconds. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_MICROVAL_PRINTABLE(run->mStats.mHeapRuntimeCost)); PR_fprintf(globals.mRequest.mFD, "\n"); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, displayCallsites); break; } /* ** What do we follow? */ switch(aFollow) { case ST_FOLLOW_SIBLINGS: aCallsite = aCallsite->siblings; break; case ST_FOLLOW_PARENTS: aCallsite = aCallsite->parent; break; default: aCallsite = NULL; retval = __LINE__; REPORT_ERROR(__LINE__, displayCallsites); break; } } /* ** Terminate the table if we should. */ if(0 != headerDisplayed) { PR_fprintf(globals.mRequest.mFD, "
CallsiteComposite Byte SizeComposite SecondsComposite WeightHeap Object CountComposite Heap Operation Seconds
"); htmlCallsiteAnchor(aCallsite, NULL, aRealNames); PR_fprintf(globals.mRequest.mFD, "%u" ST_TIMEVAL_FORMAT "%llu%u" ST_MICROVAL_FORMAT "
\n"); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, displayCallsites); } return retval; } /* ** displayAllocationDetails ** ** Report what we know about the allocation. ** ** Returns !0 on error. */ int displayAllocationDetails(STAllocation* aAllocation) { int retval = 0; if(NULL != aAllocation) { PRUint32 traverse = 0; PRUint32 bytesize = byteSize(aAllocation); PRUint32 timeval = aAllocation->mMaxTimeval - aAllocation->mMinTimeval; PRUint32 heapCost = aAllocation->mHeapRuntimeCost; PRUint64 weight64 = LL_INIT(0, 0); PRUint64 bytesize64 = LL_INIT(0, 0); PRUint64 timeval64 = LL_INIT(0, 0); PRUint32 cacheval = 0; int displayRes = 0; LL_UI2L(bytesize64, bytesize); LL_UI2L(timeval64, timeval); LL_MUL(weight64, bytesize64, timeval64); PR_fprintf(globals.mRequest.mFD, "Allocation %u Details:

\n", aAllocation->mRunIndex); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n", bytesize); PR_fprintf(globals.mRequest.mFD, "\n", ST_TIMEVAL_PRINTABLE(timeval)); PR_fprintf(globals.mRequest.mFD, "\n", weight64); PR_fprintf(globals.mRequest.mFD, "\n", ST_MICROVAL_PRINTABLE(heapCost)); PR_fprintf(globals.mRequest.mFD, "
Final Size:%u
Lifespan Seconds:" ST_TIMEVAL_FORMAT "
Weight:%llu
Heap Operation Seconds:" ST_MICROVAL_FORMAT "

\n"); /* ** The events. */ PR_fprintf(globals.mRequest.mFD, "%u Life Event(s):
\n", aAllocation->mEventCount); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); for(traverse = 0; traverse < aAllocation->mEventCount && traverse < globals.mOptions.mListItemMax; traverse++) { PR_fprintf(globals.mRequest.mFD, "\n"); /* ** count. */ PR_fprintf(globals.mRequest.mFD, "\n", traverse + 1); /* ** Operation. */ PR_fprintf(globals.mRequest.mFD, ""); /* ** Size. */ PR_fprintf(globals.mRequest.mFD, "\n", aAllocation->mEvents[traverse].mHeapSize); /* ** Timeval. */ cacheval = aAllocation->mEvents[traverse].mTimeval - globals.mMinTimeval; PR_fprintf(globals.mRequest.mFD, "\n", ST_TIMEVAL_PRINTABLE(cacheval)); /* ** Callsite backtrace. ** Only relevant backtrace is for event 0 for now until ** trace-malloc outputs proper callsites for all others. */ PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); } PR_fprintf(globals.mRequest.mFD, "
OperationSizeSeconds
%u."); switch(aAllocation->mEvents[traverse].mEventType) { case TM_EVENT_CALLOC: PR_fprintf(globals.mRequest.mFD, "calloc"); break; case TM_EVENT_FREE: PR_fprintf(globals.mRequest.mFD, "free"); break; case TM_EVENT_MALLOC: PR_fprintf(globals.mRequest.mFD, "malloc"); break; case TM_EVENT_REALLOC: PR_fprintf(globals.mRequest.mFD, "realloc"); break; default: retval = __LINE__; REPORT_ERROR(__LINE__, displayAllocationDetails); break; } PR_fprintf(globals.mRequest.mFD, "%u" ST_TIMEVAL_FORMAT "\n"); if(0 == traverse) { displayRes = displayCallsites(aAllocation->mEvents[traverse].mCallsite, ST_FOLLOW_PARENTS, 0, __LINE__); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayCallsite); } } PR_fprintf(globals.mRequest.mFD, "

\n"); } else { retval = __LINE__; REPORT_ERROR(__LINE__, displayAllocationDetails); } return retval; } /* ** compareCallsites ** ** qsort callback. ** Compare the callsites as specified by the options. ** There must be NO equal callsites, unless they really are duplicates, ** this is so that a duplicate detector loop can ** simply skip sorted items until the callsite is different. */ int compareCallsites(const void* aSite1, const void* aSite2, void* aContext) { int retval = 0; if(NULL != aSite1 && NULL != aSite2) { tmcallsite* site1 = *((tmcallsite**)aSite1); tmcallsite* site2 = *((tmcallsite**)aSite2); if(NULL != site1 && NULL != site2) { STRun* run1 = CALLSITE_RUN(site1); STRun* run2 = CALLSITE_RUN(site2); if(NULL != run1 && NULL != run2) { STCallsiteStats* stats1 = &(run1->mStats); STCallsiteStats* stats2 = &(run2->mStats); /* ** Logic determined by pref/option. */ switch(globals.mOptions.mOrderBy) { case ST_WEIGHT: { PRUint64 weight164 = stats1->mWeight64; PRUint64 weight264 = stats2->mWeight64; if(LL_UCMP(weight164, <, weight264)) { retval = __LINE__; } else if(LL_UCMP(weight164, >, weight264)) { retval = - __LINE__; } } break; case ST_SIZE: { PRUint32 size1 = stats1->mSize; PRUint32 size2 = stats2->mSize; if(size1 < size2) { retval = __LINE__; } else if(size1 > size2) { retval = - __LINE__; } } break; case ST_TIMEVAL: { PRUint64 timeval164 = stats1->mTimeval64; PRUint64 timeval264 = stats2->mTimeval64; if(LL_UCMP(timeval164, <, timeval264)) { retval = __LINE__; } else if(LL_UCMP(timeval164, >, timeval264)) { retval = - __LINE__; } } break; case ST_COUNT: { PRUint32 count1 = stats1->mCompositeCount; PRUint32 count2 = stats2->mCompositeCount; if(count1 < count2) { retval = __LINE__; } else if(count1 > count2) { retval = - __LINE__; } } break; case ST_HEAPCOST: { PRUint32 cost1 = stats1->mHeapRuntimeCost; PRUint32 cost2 = stats2->mHeapRuntimeCost; if(cost1 < cost2) { retval = __LINE__; } else if(cost1 > cost2) { retval = - __LINE__; } } break; default: { REPORT_ERROR(__LINE__, compareAllocations); } break; } /* ** If the return value is still zero, do a pointer compare. ** This makes sure we return zero, only iff the same object. */ if(0 == retval) { if(stats1 < stats2) { retval = __LINE__; } else if(stats1 > stats2) { retval = - __LINE__; } } } } } return retval; } /* ** displayTopCallsites ** ** Given a list of callsites, sort it, and output skipping dups. ** The passed in callsite array is side effected, as in that it will come ** back sorted. This function will not release the array. ** ** Note: If the stamp passed in is non zero, then all callsites must match. ** If the stamp is zero, all callsites must match global sorted run stamp. ** ** Returns !0 on error. */ int displayTopCallsites(tmcallsite** aCallsites, PRUint32 aCallsiteCount, PRUint32 aStamp, int aRealName) { int retval = 0; if(NULL != aCallsites && 0 < aCallsiteCount) { PRUint32 traverse = 0; STRun* run = NULL; tmcallsite* site = NULL; int headerDisplayed = 0; PRUint32 displayed = 0; /* ** Fixup the stamp. */ if(0 == aStamp && NULL != globals.mCache.mSortedRun) { aStamp = globals.mCache.mSortedRun->mStats.mStamp; } /* ** Sort the things. */ NS_QuickSort(aCallsites, aCallsiteCount, sizeof(tmcallsite*), compareCallsites, NULL); /* ** Time for output. */ for(traverse = 0; traverse < aCallsiteCount && globals.mOptions.mListItemMax > displayed; traverse++) { site = aCallsites[traverse]; run = CALLSITE_RUN(site); /* ** Only if the same stamp.... */ if(aStamp == run->mStats.mStamp) { /* ** We got a header yet? */ if(0 == headerDisplayed) { headerDisplayed = __LINE__; PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n"); } displayed++; PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Rank. */ PR_fprintf(globals.mRequest.mFD, "\n", displayed); /* ** Method. */ PR_fprintf(globals.mRequest.mFD, "\n"); /* ** Size. */ PR_fprintf(globals.mRequest.mFD, "\n", run->mStats.mSize); /* ** Timeval. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_TIMEVAL_PRINTABLE64(run->mStats.mTimeval64)); /* ** Weight. */ PR_fprintf(globals.mRequest.mFD, "\n", run->mStats.mWeight64); /* ** Allocation object count. */ PR_fprintf(globals.mRequest.mFD, "\n", run->mStats.mCompositeCount); /* ** Heap operation seconds. */ PR_fprintf(globals.mRequest.mFD, "\n", ST_MICROVAL_PRINTABLE(run->mStats.mHeapRuntimeCost)); PR_fprintf(globals.mRequest.mFD, "\n"); if(globals.mOptions.mListItemMax > displayed) { /* ** Skip any dups. */ while(((traverse + 1) < aCallsiteCount) && (site == aCallsites[traverse + 1])) { traverse++; } } } } /* ** We need to terminate anything? */ if(0 != headerDisplayed) { PR_fprintf(globals.mRequest.mFD, "
RankCallsiteComposite SizeComposite SecondsComposite WeightHeap Object CountHeap Operation Seconds
%u"); htmlCallsiteAnchor(site, NULL, aRealName); PR_fprintf(globals.mRequest.mFD, "%u" ST_TIMEVAL_FORMAT "%llu%u" ST_MICROVAL_FORMAT "
\n"); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, displayTopCallsites); } return retval; } /* ** displayCallsiteDetails ** ** The callsite specific report. ** Try to report what we know. ** This one hits a little harder than the rest. ** ** Returns !0 on error. */ int displayCallsiteDetails(tmcallsite* aCallsite) { int retval = 0; if(NULL != aCallsite && NULL != aCallsite->method) { STRun* sortedRun = NULL; STRun* thisRun = CALLSITE_RUN(aCallsite); const char* sourceFile = NULL; sourceFile = resolveSourceFile(aCallsite->method); if(NULL != sourceFile) { PR_fprintf(globals.mRequest.mFD, "%s", tmmethodnode_name(aCallsite->method)); PR_fprintf(globals.mRequest.mFD, "(%s:%u)", aCallsite->method->sourcefile, aCallsite->method->linenumber, sourceFile, aCallsite->method->linenumber); PR_fprintf(globals.mRequest.mFD, " Callsite Details:

\n"); } else { PR_fprintf(globals.mRequest.mFD, "%s+%u(%u) Callsite Details:

\n", tmmethodnode_name(aCallsite->method), aCallsite->offset, (PRUint32)aCallsite->entry.key); } PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "\n", thisRun->mStats.mSize); PR_fprintf(globals.mRequest.mFD, "\n", ST_TIMEVAL_PRINTABLE64(thisRun->mStats.mTimeval64)); PR_fprintf(globals.mRequest.mFD, "\n", thisRun->mStats.mWeight64); PR_fprintf(globals.mRequest.mFD, "\n", thisRun->mStats.mCompositeCount); PR_fprintf(globals.mRequest.mFD, "\n", ST_MICROVAL_PRINTABLE(thisRun->mStats.mHeapRuntimeCost)); PR_fprintf(globals.mRequest.mFD, "
Composite Byte Size:%u
Composite Seconds:" ST_TIMEVAL_FORMAT "
Composite Weight:%llu
Heap Object Count:%u
Heap Operation Seconds:" ST_MICROVAL_FORMAT "
\n

\n"); /* ** Kids (callsites we call): */ if(NULL != aCallsite->kids && NULL != aCallsite->kids->method) { int displayRes = 0; PRUint32 siteCount = 0; tmcallsite** sites = NULL; /* ** Collect the kid sibling callsites. ** Doing it this way sorts them for relevance. */ siteCount = callsiteArrayFromCallsite(&sites, 0, aCallsite->kids, ST_FOLLOW_SIBLINGS); if(0 != siteCount && NULL != sites) { /* ** Got something to show. */ PR_fprintf(globals.mRequest.mFD, "Children Callsites:
\n"); displayRes = displayTopCallsites(sites, siteCount, 0, __LINE__); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayTopCallsites); } PR_fprintf(globals.mRequest.mFD, "

\n"); /* ** Done with array. */ free(sites); sites = NULL; } } /* ** Parents (those who call us): */ if(NULL != aCallsite->parent && NULL != aCallsite->parent->method) { int displayRes = 0; PR_fprintf(globals.mRequest.mFD, "Parent Callsites:
\n"); displayRes = displayCallsites(aCallsite->parent, ST_FOLLOW_PARENTS, 0, __LINE__); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayCallsites); } PR_fprintf(globals.mRequest.mFD, "

\n"); } /* ** Allocations we did. ** Simply harvest our own run. */ sortedRun = createRun(0); if(NULL != sortedRun) { int harvestRes = 0; harvestRes = harvestRun(CALLSITE_RUN(aCallsite), sortedRun, NULL); if(0 == harvestRes) { if(0 != sortedRun->mAllocationCount) { int sortRes = 0; sortRes = sortRun(sortedRun); if(0 == sortRes) { int displayRes = 0; PR_fprintf(globals.mRequest.mFD, "Allocations:
\n"); displayRes = displayTopAllocations(sortedRun, 0); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayTopAllocations); } PR_fprintf(globals.mRequest.mFD, "

\n"); } else { retval = __LINE__; REPORT_ERROR(__LINE__, sortRun); } } } else { retval = __LINE__; REPORT_ERROR(__LINE__, harvestRun); } /* ** Done with the run. */ freeRun(sortedRun); sortedRun = NULL; } else { retval = __LINE__; REPORT_ERROR(__LINE__, createRun); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, displayCallsiteDetails); } return retval; } #if WANT_GRAPHS /* ** graphFootprint ** ** Output a PNG graph of the memory usage of the run. ** ** Draw the graph within these boundaries. ** STGD_MARGIN,STGD_MARGIN,STGD_WIDTH-STGD_MARGIN,STGD_HEIGHT-STGD_MARGIN ** ** Returns !0 on failure. */ int graphFootprint(STRun* aRun) { int retval = 0; if(NULL != aRun) { PRUint32 *YData = NULL; PRUint32 YDataArray[STGD_SPACE_X]; PRUint32 traverse = 0; PRUint32 timeval = globals.mOptions.mGraphTimevalMin; PRUint32 loop = 0; /* ** Decide if this is custom or we should use the global cache. */ if(aRun == globals.mCache.mSortedRun) { YData = globals.mCache.mFootprintYData; } else { YData = YDataArray; } /* ** Only do the computations if we aren't cached already. */ if(YData != globals.mCache.mFootprintYData || 0 == globals.mCache.mFootprintCached) { memset(YData, 0, sizeof(PRUint32) * STGD_SPACE_X); /* ** Initialize our Y data. ** Pretty brutal loop here.... */ for(traverse = 0; 0 == retval && traverse < STGD_SPACE_X; traverse++) { /* ** Compute what timeval this Y data lands in. */ timeval = ((traverse * (globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin)) / STGD_SPACE_X) + globals.mMinTimeval + globals.mOptions.mGraphTimevalMin; /* ** Loop over the run. ** Should an allocation contain said Timeval, we're good. */ for(loop = 0; loop < aRun->mAllocationCount; loop++) { if(timeval >= aRun->mAllocations[loop]->mMinTimeval && timeval <= aRun->mAllocations[loop]->mMaxTimeval) { YData[traverse] += byteSize(aRun->mAllocations[loop]); } } } /* ** Did we cache this? */ if(YData == globals.mCache.mFootprintYData) { globals.mCache.mFootprintCached = __LINE__; } } if(0 == retval) { PRUint32 minMemory = (PRUint32)-1; PRUint32 maxMemory = 0; int transparent = 0; gdImagePtr graph = NULL; /* ** Go through and find the minimum and maximum sizes. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { if(YData[traverse] < minMemory) { minMemory = YData[traverse]; } if(YData[traverse] > maxMemory) { maxMemory = YData[traverse]; } } /* ** We can now draw the graph. */ graph = createGraph(&transparent); if(NULL != graph) { gdSink theSink; int red = 0; int x1 = 0; int y1 = 0; int x2 = 0; int y2 = 0; PRUint32 percents[11] = { 0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100 }; char* timevals[11]; char* bytes[11]; char timevalSpace[11][32]; char byteSpace[11][32]; int legendColors[1]; const char* legends[1] = { "Memory in Use" }; PRUint32 cached = 0; /* ** Figure out what the labels will say. */ for(traverse = 0; traverse < 11; traverse++) { timevals[traverse] = timevalSpace[traverse]; bytes[traverse] = byteSpace[traverse]; cached = (((globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin) * percents[traverse]) / 100) + globals.mOptions.mGraphTimevalMin; PR_snprintf(timevals[traverse], 32, ST_TIMEVAL_FORMAT, ST_TIMEVAL_PRINTABLE(cached)); PR_snprintf(bytes[traverse], 32, "%u", ((maxMemory - minMemory) * percents[traverse]) / 100); } red = gdImageColorAllocate(graph, 255, 0, 0); legendColors[0] = red; drawGraph(graph, -1, "Memory Footprint Over Time", "Seconds", "Bytes", 11, percents, (const char**)timevals, 11, percents, (const char**)bytes, 1, legendColors, legends); if(maxMemory != minMemory) { PRInt64 in64 = LL_INIT(0, 0); PRInt64 ydata64 = LL_INIT(0, 0); PRInt64 spacey64 = LL_INIT(0, 0); PRInt64 mem64 = LL_INIT(0, 0); PRInt32 in32 = 0; /* ** Go through our Y data and mark it up. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { x1 = traverse + STGD_MARGIN; y1 = STGD_HEIGHT - STGD_MARGIN; /* ** Need to do this math in 64 bits. */ LL_I2L(ydata64, YData[traverse]); LL_I2L(spacey64, STGD_SPACE_Y); LL_I2L(mem64, (maxMemory - minMemory)); LL_MUL(in64, ydata64, spacey64); LL_DIV(in64, in64, mem64); LL_L2I(in32, in64); x2 = x1; y2 = y1 - in32; gdImageLine(graph, x1, y1, x2, y2, red); } } theSink.context = globals.mRequest.mFD; theSink.sink = pngSink; gdImagePngToSink(graph, &theSink); gdImageDestroy(graph); } else { retval = __LINE__; REPORT_ERROR(__LINE__, createGraph); } } } else { retval = __LINE__; REPORT_ERROR(__LINE__, graphFootprint); } return retval; } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS /* ** graphTimeval ** ** Output a PNG graph of when the memory is allocated. ** ** Draw the graph within these boundaries. ** STGD_MARGIN,STGD_MARGIN,STGD_WIDTH-STGD_MARGIN,STGD_HEIGHT-STGD_MARGIN ** ** Returns !0 on failure. */ int graphTimeval(STRun* aRun) { int retval = 0; if(NULL != aRun) { PRUint32 *YData = NULL; PRUint32 YDataArray[STGD_SPACE_X]; PRUint32 traverse = 0; PRUint32 timeval = globals.mOptions.mGraphTimevalMin + globals.mMinTimeval; PRUint32 loop = 0; /* ** Decide if this is custom or we should use the global cache. */ if(aRun == globals.mCache.mSortedRun) { YData = globals.mCache.mTimevalYData; } else { YData = YDataArray; } /* ** Only do the computations if we aren't cached already. */ if(YData != globals.mCache.mTimevalYData || 0 == globals.mCache.mTimevalCached) { PRUint32 prevTimeval = 0; memset(YData, 0, sizeof(PRUint32) * STGD_SPACE_X); /* ** Initialize our Y data. ** Pretty brutal loop here.... */ for(traverse = 0; 0 == retval && traverse < STGD_SPACE_X; traverse++) { /* ** Compute what timeval this Y data lands in. */ prevTimeval = timeval; timeval = ((traverse * (globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin)) / STGD_SPACE_X) + globals.mMinTimeval + globals.mOptions.mGraphTimevalMin; /* ** Loop over the run. ** Should an allocation have been allocated between ** prevTimeval and timeval.... */ for(loop = 0; loop < aRun->mAllocationCount; loop++) { if(prevTimeval < aRun->mAllocations[loop]->mMinTimeval && timeval >= aRun->mAllocations[loop]->mMinTimeval) { YData[traverse] += byteSize(aRun->mAllocations[loop]); } } } /* ** Did we cache this? */ if(YData == globals.mCache.mTimevalYData) { globals.mCache.mTimevalCached = __LINE__; } } if(0 == retval) { PRUint32 minMemory = (PRUint32)-1; PRUint32 maxMemory = 0; int transparent = 0; gdImagePtr graph = NULL; /* ** Go through and find the minimum and maximum sizes. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { if(YData[traverse] < minMemory) { minMemory = YData[traverse]; } if(YData[traverse] > maxMemory) { maxMemory = YData[traverse]; } } /* ** We can now draw the graph. */ graph = createGraph(&transparent); if(NULL != graph) { gdSink theSink; int red = 0; int x1 = 0; int y1 = 0; int x2 = 0; int y2 = 0; PRUint32 percents[11] = { 0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100 }; char* timevals[11]; char* bytes[11]; char timevalSpace[11][32]; char byteSpace[11][32]; int legendColors[1]; const char* legends[1] = { "Memory Allocated" }; PRUint32 cached = 0; /* ** Figure out what the labels will say. */ for(traverse = 0; traverse < 11; traverse++) { timevals[traverse] = timevalSpace[traverse]; bytes[traverse] = byteSpace[traverse]; cached = (((globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin) * percents[traverse]) / 100) + globals.mOptions.mGraphTimevalMin; PR_snprintf(timevals[traverse], 32, ST_TIMEVAL_FORMAT, ST_TIMEVAL_PRINTABLE(cached)); PR_snprintf(bytes[traverse], 32, "%u", ((maxMemory - minMemory) * percents[traverse]) / 100); } red = gdImageColorAllocate(graph, 255, 0, 0); legendColors[0] = red; drawGraph(graph, -1, "Allocation Times", "Seconds", "Bytes", 11, percents, (const char**)timevals, 11, percents, (const char**)bytes, 1, legendColors, legends); if(maxMemory != minMemory) { PRInt64 in64 = LL_INIT(0, 0); PRInt64 ydata64 = LL_INIT(0, 0); PRInt64 spacey64 = LL_INIT(0, 0); PRInt64 mem64 = LL_INIT(0, 0); PRInt32 in32 = 0; /* ** Go through our Y data and mark it up. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { x1 = traverse + STGD_MARGIN; y1 = STGD_HEIGHT - STGD_MARGIN; /* ** Need to do this math in 64 bits. */ LL_I2L(ydata64, YData[traverse]); LL_I2L(spacey64, STGD_SPACE_Y); LL_I2L(mem64, (maxMemory - minMemory)); LL_MUL(in64, ydata64, spacey64); LL_DIV(in64, in64, mem64); LL_L2I(in32, in64); x2 = x1; y2 = y1 - in32; gdImageLine(graph, x1, y1, x2, y2, red); } } theSink.context = globals.mRequest.mFD; theSink.sink = pngSink; gdImagePngToSink(graph, &theSink); gdImageDestroy(graph); } else { retval = __LINE__; REPORT_ERROR(__LINE__, createGraph); } } } else { retval = __LINE__; REPORT_ERROR(__LINE__, graphTimeval); } return retval; } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS /* ** graphLifespan ** ** Output a PNG graph of how long memory lived. ** ** Draw the graph within these boundaries. ** STGD_MARGIN,STGD_MARGIN,STGD_WIDTH-STGD_MARGIN,STGD_HEIGHT-STGD_MARGIN ** ** Returns !0 on failure. */ int graphLifespan(STRun* aRun) { int retval = 0; if(NULL != aRun) { PRUint32 *YData = NULL; PRUint32 YDataArray[STGD_SPACE_X]; PRUint32 traverse = 0; PRUint32 timeval = globals.mOptions.mGraphTimevalMin; PRUint32 loop = 0; /* ** Decide if this is custom or we should use the global cache. */ if(aRun == globals.mCache.mSortedRun) { YData = globals.mCache.mLifespanYData; } else { YData = YDataArray; } /* ** Only do the computations if we aren't cached already. */ if(YData != globals.mCache.mLifespanYData || 0 == globals.mCache.mLifespanCached) { PRUint32 prevTimeval = 0; PRUint32 lifespan = 0; memset(YData, 0, sizeof(PRUint32) * STGD_SPACE_X); /* ** Initialize our Y data. ** Pretty brutal loop here.... */ for(traverse = 0; 0 == retval && traverse < STGD_SPACE_X; traverse++) { /* ** Compute what timeval this Y data lands in. */ prevTimeval = timeval; timeval = ((traverse * (globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin)) / STGD_SPACE_X) + globals.mOptions.mGraphTimevalMin; /* ** Loop over the run. ** Should an allocation have lived between ** prevTimeval and timeval.... */ for(loop = 0; loop < aRun->mAllocationCount; loop++) { lifespan = aRun->mAllocations[loop]->mMaxTimeval - aRun->mAllocations[loop]->mMinTimeval; if(prevTimeval < lifespan && timeval >= lifespan) { YData[traverse] += byteSize(aRun->mAllocations[loop]); } } } /* ** Did we cache this? */ if(YData == globals.mCache.mLifespanYData) { globals.mCache.mLifespanCached = __LINE__; } } if(0 == retval) { PRUint32 minMemory = (PRUint32)-1; PRUint32 maxMemory = 0; int transparent = 0; gdImagePtr graph = NULL; /* ** Go through and find the minimum and maximum sizes. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { if(YData[traverse] < minMemory) { minMemory = YData[traverse]; } if(YData[traverse] > maxMemory) { maxMemory = YData[traverse]; } } /* ** We can now draw the graph. */ graph = createGraph(&transparent); if(NULL != graph) { gdSink theSink; int red = 0; int x1 = 0; int y1 = 0; int x2 = 0; int y2 = 0; PRUint32 percents[11] = { 0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100 }; char* timevals[11]; char* bytes[11]; char timevalSpace[11][32]; char byteSpace[11][32]; int legendColors[1]; const char* legends[1] = { "Live Memory" }; PRUint32 cached = 0; /* ** Figure out what the labels will say. */ for(traverse = 0; traverse < 11; traverse++) { timevals[traverse] = timevalSpace[traverse]; bytes[traverse] = byteSpace[traverse]; cached = (((globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin) * percents[traverse]) / 100) + globals.mOptions.mGraphTimevalMin; PR_snprintf(timevals[traverse], 32, ST_TIMEVAL_FORMAT, ST_TIMEVAL_PRINTABLE(cached)); PR_snprintf(bytes[traverse], 32, "%u", ((maxMemory - minMemory) * percents[traverse]) / 100); } red = gdImageColorAllocate(graph, 255, 0, 0); legendColors[0] = red; drawGraph(graph, -1, "Allocation Lifespans", "Lifespan", "Bytes", 11, percents, (const char**)timevals, 11, percents, (const char**)bytes, 1, legendColors, legends); if(maxMemory != minMemory) { PRInt64 in64 = LL_INIT(0, 0); PRInt64 ydata64 = LL_INIT(0, 0); PRInt64 spacey64 = LL_INIT(0, 0); PRInt64 mem64 = LL_INIT(0, 0); PRInt32 in32 = 0; /* ** Go through our Y data and mark it up. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { x1 = traverse + STGD_MARGIN; y1 = STGD_HEIGHT - STGD_MARGIN; /* ** Need to do this math in 64 bits. */ LL_I2L(ydata64, YData[traverse]); LL_I2L(spacey64, STGD_SPACE_Y); LL_I2L(mem64, (maxMemory - minMemory)); LL_MUL(in64, ydata64, spacey64); LL_DIV(in64, in64, mem64); LL_L2I(in32, in64); x2 = x1; y2 = y1 - in32; gdImageLine(graph, x1, y1, x2, y2, red); } } theSink.context = globals.mRequest.mFD; theSink.sink = pngSink; gdImagePngToSink(graph, &theSink); gdImageDestroy(graph); } else { retval = __LINE__; REPORT_ERROR(__LINE__, createGraph); } } } else { retval = __LINE__; REPORT_ERROR(__LINE__, graphLifespan); } return retval; } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS /* ** graphWeight ** ** Output a PNG graph of Allocations by Weight ** ** Draw the graph within these boundaries. ** STGD_MARGIN,STGD_MARGIN,STGD_WIDTH-STGD_MARGIN,STGD_HEIGHT-STGD_MARGIN ** ** Returns !0 on failure. */ int graphWeight(STRun* aRun) { int retval = 0; if(NULL != aRun) { PRUint64 *YData64 = NULL; PRUint64 YDataArray64[STGD_SPACE_X]; PRUint32 traverse = 0; PRUint32 timeval = globals.mOptions.mGraphTimevalMin + globals.mMinTimeval; PRUint32 loop = 0; /* ** Decide if this is custom or we should use the global cache. */ if(aRun == globals.mCache.mSortedRun) { YData64 = globals.mCache.mWeightYData64; } else { YData64 = YDataArray64; } /* ** Only do the computations if we aren't cached already. */ if(YData64 != globals.mCache.mWeightYData64 || 0 == globals.mCache.mWeightCached) { PRUint32 prevTimeval = 0; memset(YData64, 0, sizeof(PRUint64) * STGD_SPACE_X); /* ** Initialize our Y data. ** Pretty brutal loop here.... */ for(traverse = 0; 0 == retval && traverse < STGD_SPACE_X; traverse++) { /* ** Compute what timeval this Y data lands in. */ prevTimeval = timeval; timeval = ((traverse * (globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin)) / STGD_SPACE_X) + globals.mMinTimeval + globals.mOptions.mGraphTimevalMin; /* ** Loop over the run. ** Should an allocation have been allocated between ** prevTimeval and timeval.... */ for(loop = 0; loop < aRun->mAllocationCount; loop++) { if(prevTimeval < aRun->mAllocations[loop]->mMinTimeval && timeval >= aRun->mAllocations[loop]->mMinTimeval) { PRUint64 size64 = LL_INIT(0, 0); PRUint64 lifespan64 = LL_INIT(0, 0); PRUint64 weight64 = LL_INIT(0, 0); LL_UI2L(size64, byteSize(aRun->mAllocations[loop])); LL_UI2L(lifespan64, (aRun->mAllocations[loop]->mMaxTimeval - aRun->mAllocations[loop]->mMinTimeval)); LL_MUL(weight64, size64, lifespan64); LL_ADD(YData64[traverse], YData64[traverse], weight64); } } } /* ** Did we cache this? */ if(YData64 == globals.mCache.mWeightYData64) { globals.mCache.mWeightCached = __LINE__; } } if(0 == retval) { PRUint64 minWeight64 = LL_INIT(0xFFFFFFFF, 0xFFFFFFFF); PRUint64 maxWeight64 = LL_INIT(0, 0); int transparent = 0; gdImagePtr graph = NULL; /* ** Go through and find the minimum and maximum weights. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { if(LL_UCMP(YData64[traverse], <, minWeight64)) { minWeight64 = YData64[traverse]; } if(LL_UCMP(YData64[traverse], >, maxWeight64)) { maxWeight64 = YData64[traverse]; } } /* ** We can now draw the graph. */ graph = createGraph(&transparent); if(NULL != graph) { gdSink theSink; int red = 0; int x1 = 0; int y1 = 0; int x2 = 0; int y2 = 0; PRUint32 percents[11] = { 0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100 }; char* timevals[11]; char* bytes[11]; char timevalSpace[11][32]; char byteSpace[11][32]; int legendColors[1]; const char* legends[1] = { "Memory Weight" }; PRUint64 percent64 = LL_INIT(0, 0); PRUint64 result64 = LL_INIT(0, 0); PRUint64 hundred64 = LL_INIT(0, 0); PRUint32 cached = 0; LL_UI2L(hundred64, 100); /* ** Figure out what the labels will say. */ for(traverse = 0; traverse < 11; traverse++) { timevals[traverse] = timevalSpace[traverse]; bytes[traverse] = byteSpace[traverse]; cached = (((globals.mOptions.mGraphTimevalMax - globals.mOptions.mGraphTimevalMin) * percents[traverse]) / 100) + globals.mOptions.mGraphTimevalMin; PR_snprintf(timevals[traverse], 32, ST_TIMEVAL_FORMAT, ST_TIMEVAL_PRINTABLE(cached)); LL_UI2L(percent64, percents[traverse]); LL_SUB(result64, maxWeight64, minWeight64); LL_MUL(result64, result64, percent64); LL_DIV(result64, result64, hundred64); PR_snprintf(bytes[traverse], 32, "%llu", result64); } red = gdImageColorAllocate(graph, 255, 0, 0); legendColors[0] = red; drawGraph(graph, -1, "Allocation Weights", "Seconds", "Weight", 11, percents, (const char**)timevals, 11, percents, (const char**)bytes, 1, legendColors, legends); if(LL_NE(maxWeight64, minWeight64)) { PRInt64 in64 = LL_INIT(0, 0); PRInt64 spacey64 = LL_INIT(0, 0); PRInt64 weight64 = LL_INIT(0, 0); PRInt32 in32 = 0; /* ** Go through our Y data and mark it up. */ for(traverse = 0; traverse < STGD_SPACE_X; traverse++) { x1 = traverse + STGD_MARGIN; y1 = STGD_HEIGHT - STGD_MARGIN; /* ** Need to do this math in 64 bits. */ LL_I2L(spacey64, STGD_SPACE_Y); LL_SUB(weight64, maxWeight64, minWeight64); LL_MUL(in64, YData64[traverse], spacey64); LL_DIV(in64, in64, weight64); LL_L2I(in32, in64); x2 = x1; y2 = y1 - in32; gdImageLine(graph, x1, y1, x2, y2, red); } } theSink.context = globals.mRequest.mFD; theSink.sink = pngSink; gdImagePngToSink(graph, &theSink); gdImageDestroy(graph); } else { retval = __LINE__; REPORT_ERROR(__LINE__, createGraph); } } } else { retval = __LINE__; REPORT_ERROR(__LINE__, graphWeight); } return retval; } #endif /* WANT_GRAPHS */ /* ** applySettings ** ** Apply settings and update global options. ** Returns 0 on success. Nonzero on failure. */ int applySettings(void) { int getRes = 0; int changedSet = 0; int changedOrder = 0; int changedGraph = 0; int changedDontCare = 0; char looper_buf[32]; PRIntn looper = 0; /* ** If we've got get data, we need to attempt to enact the changes. ** That way, when we show the page, it will have the new changes. */ if(NULL == globals.mRequest.mGetData || '\0' == *globals.mRequest.mGetData) return 0; getRes += getDataPRUint32(globals.mRequest.mGetData, "mListItemMax", &globals.mOptions.mListItemMax, &changedDontCare, 1); getRes += getDataPRUint32(globals.mRequest.mGetData, "mTimevalMin", &globals.mOptions.mTimevalMin, &changedSet, ST_TIMEVAL_RESOLUTION); getRes += getDataPRUint32(globals.mRequest.mGetData, "mTimevalMax", &globals.mOptions.mTimevalMax, &changedSet, ST_TIMEVAL_RESOLUTION); getRes += getDataPRUint32(globals.mRequest.mGetData, "mAllocationTimevalMin", &globals.mOptions.mAllocationTimevalMin, &changedSet, ST_TIMEVAL_RESOLUTION); getRes += getDataPRUint32(globals.mRequest.mGetData, "mAllocationTimevalMax", &globals.mOptions.mAllocationTimevalMax, &changedSet, ST_TIMEVAL_RESOLUTION); #if WANT_GRAPHS getRes += getDataPRUint32(globals.mRequest.mGetData, "mGraphTimevalMin", &globals.mOptions.mGraphTimevalMin, &changedGraph, ST_TIMEVAL_RESOLUTION); getRes += getDataPRUint32(globals.mRequest.mGetData, "mGraphTimevalMax", &globals.mOptions.mGraphTimevalMax, &changedGraph, ST_TIMEVAL_RESOLUTION); #endif /* WANT_GRAPHS */ getRes += getDataPRUint32(globals.mRequest.mGetData, "mSizeMin", &globals.mOptions.mSizeMin, &changedSet, 1); getRes += getDataPRUint32(globals.mRequest.mGetData, "mSizeMax", &globals.mOptions.mSizeMax, &changedSet, 1); getRes += getDataPRUint32(globals.mRequest.mGetData, "mAlignBy", &globals.mOptions.mAlignBy, &changedSet, 1); getRes += getDataPRUint32(globals.mRequest.mGetData, "mOverhead", &globals.mOptions.mOverhead, &changedSet, 1); getRes += getDataPRUint32(globals.mRequest.mGetData, "mOrderBy", &globals.mOptions.mOrderBy, &changedOrder, 1); getRes += getDataPRUint32(globals.mRequest.mGetData, "mLifetimeMin", &globals.mOptions.mLifetimeMin, &changedSet, ST_TIMEVAL_RESOLUTION); getRes += getDataPRUint32(globals.mRequest.mGetData, "mLifetimeMax", &globals.mOptions.mLifetimeMax, &changedSet, ST_TIMEVAL_RESOLUTION); getRes += getDataPRUint64(globals.mRequest.mGetData, "mWeightMin", &globals.mOptions.mWeightMin64, &changedSet); getRes += getDataPRUint64(globals.mRequest.mGetData, "mWeightMax", &globals.mOptions.mWeightMax64, &changedSet); for(looper = 0; ST_SUBSTRING_MATCH_MAX > looper; looper++) { PR_snprintf(looper_buf, sizeof(looper_buf), "mRestrictText%d", looper); getRes += getDataString(globals.mRequest.mGetData, looper_buf, &globals.mOptions.mRestrictText[looper], &changedSet); } getRes += getDataString(globals.mRequest.mGetData, "mCategoryName", &globals.mOptions.mCategoryName, &changedSet); /* ** Sanity check options */ if (!globals.mOptions.mCategoryName || !*globals.mOptions.mCategoryName) { globals.mOptions.mCategoryName = ST_ROOT_CATEGORY_NAME; } else if (!findCategoryNode(globals.mOptions.mCategoryName, &globals)) { /* Category node is invalid. Reset to root. */ globals.mOptions.mCategoryName = ST_ROOT_CATEGORY_NAME; } /* ** Resort the global based on new prefs if needed. */ if(0 != changedSet || 0 != changedOrder) { /* ** Dont free globals.mCache.mSortedRun anymore. It is held in the root category node. ** It will get freed automatically when categorization happens. */ globals.mCache.mSortedRun = createRunFromGlobal(); if(NULL == globals.mCache.mSortedRun) { getRes = __LINE__; REPORT_ERROR(__LINE__, createRunFromGlobal); } } #if WANT_GRAPHS /* ** If any of the set was changed, we need to throw away all our ** cached graphs. */ if(0 != changedSet || 0 != changedGraph) { /* ** Automove the graph timeval if required. */ if((globals.mMaxTimeval - globals.mMinTimeval) < globals.mOptions.mGraphTimevalMax) { globals.mOptions.mGraphTimevalMax = (globals.mMaxTimeval - globals.mMinTimeval); } globals.mCache.mFootprintCached = 0; globals.mCache.mTimevalCached = 0; globals.mCache.mLifespanCached = 0; globals.mCache.mWeightCached = 0; } #endif /* WANT_GRAPHS */ return getRes; } /* ** displaySettings ** ** Present the settings for change during execution. ** Returns !0 on error. ** ** Changes are effected via the get data. */ int displaySettings(void) { int retval = 0; PRUint32 cached = 0; PRIntn looper = 0; int getRes; getRes = applySettings(); /* ** Display header for settings page. Do this after applying settings. ** */ htmlHeader("SpaceTrace Settings"); /* ** Report on the operation. */ if(0 != getRes) { REPORT_ERROR(getRes, applySettings); PR_fprintf(globals.mRequest.mFD, "%u: There was a problem. Some changes may have been applied.


\n", PR_IntervalNow()); } else { PR_fprintf(globals.mRequest.mFD, "%u: Your changes have been applied.

\n", PR_IntervalNow()); } /* ** A small blurb regarding the options. */ PR_fprintf(globals.mRequest.mFD, "NOTES:

\n"); cached = globals.mMaxTimeval - globals.mMinTimeval; PR_fprintf(globals.mRequest.mFD, "The total seconds in this run is: 0 to " ST_TIMEVAL_FORMAT "
\n", ST_TIMEVAL_PRINTABLE(cached)); PR_fprintf(globals.mRequest.mFD, "All options should have command line equivalents to support batch mode.
\n"); PR_fprintf(globals.mRequest.mFD, "Changes to the options take effect immediately.
\n"); PR_fprintf(globals.mRequest.mFD, "


\n"); /* ** We've got a form to create. */ PR_fprintf(globals.mRequest.mFD, "
\n"); PR_fprintf(globals.mRequest.mFD, "Maximum number of items to display in a list?
\n"); PR_fprintf(globals.mRequest.mFD, "This option exists to control how much information you are willing to accept.

\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mListItemMax); PR_fprintf(globals.mRequest.mFD, "


\n"); PR_fprintf(globals.mRequest.mFD, "This option controls the sort order of the lists presented.
\n"); PR_fprintf(globals.mRequest.mFD, "There are several choices:
\n"); PR_fprintf(globals.mRequest.mFD, "
  • 0 is by weight (byte size * seconds).
  • 1 is by byte size.
  • 2 is by seconds (lifetime).
  • 3 is by allocation object count.
  • 4 is by heap operation runtime cost.

\n"); PR_fprintf(globals.mRequest.mFD, "Desired sort order?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mOrderBy); PR_fprintf(globals.mRequest.mFD, "


\n"); #if WANT_GRAPHS PR_fprintf(globals.mRequest.mFD, "Modify the seconds for which the graphs cover;
\n"); PR_fprintf(globals.mRequest.mFD, "meaning that a narrower range will produce a more detailed graph for that timespan.

\n"); PR_fprintf(globals.mRequest.mFD, "Minimum graph second?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mGraphTimevalMin / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "Maximum graph second?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mGraphTimevalMax / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "


\n"); #endif /* WANT_GRAPHS */ PR_fprintf(globals.mRequest.mFD, "Modify the secondss to target allocations created during a particular timespan;
\n"); PR_fprintf(globals.mRequest.mFD, "meaning that the allocations created only within the timespan are of interest.

\n"); PR_fprintf(globals.mRequest.mFD, "Minimum allocation second?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mAllocationTimevalMin / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "Maximum allocation second?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mAllocationTimevalMax / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "


\n"); PR_fprintf(globals.mRequest.mFD, "Modify the byte sizes to target allocations of a particular byte size.

\n"); PR_fprintf(globals.mRequest.mFD, "Minimum byte size?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mSizeMin); PR_fprintf(globals.mRequest.mFD, "Maximum byte size?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mSizeMax); PR_fprintf(globals.mRequest.mFD, "


\n"); PR_fprintf(globals.mRequest.mFD, "Modify the alignment boundry and heap overhead of allocations to see the actual impact an allocation has on a heap;
\n"); PR_fprintf(globals.mRequest.mFD, "meaning that normally an allocation of 1 bytes actually costs more bytes depending on your heap implementation.
\n"); PR_fprintf(globals.mRequest.mFD, "Overhead is taken into account after allocation alignment.
\n"); PR_fprintf(globals.mRequest.mFD, "i.e. the msvcrt malloc has an alignment of 16 with an overhead of 8 (1 byte allocation costs 24 heap bytes), the win32 HeapAlloc has an alignment of 8 with an overhead of 8 (1 byte allocation costs 16 heap bytes).

\n"); PR_fprintf(globals.mRequest.mFD, "Align by?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mAlignBy); PR_fprintf(globals.mRequest.mFD, "Overhead?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mOverhead); PR_fprintf(globals.mRequest.mFD, "


\n"); PR_fprintf(globals.mRequest.mFD, "Modify the seconds to target allocations of a particular lifespan/duration;
\n"); PR_fprintf(globals.mRequest.mFD, "meaning that the allocations existed at least or at most the specified span of time.

\n"); PR_fprintf(globals.mRequest.mFD, "Minimum lifetime?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mLifetimeMin / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "Maximum lifetime?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mLifetimeMax / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "


\n"); PR_fprintf(globals.mRequest.mFD, "Modify the numbers to target allocations of particular weights;
\n"); PR_fprintf(globals.mRequest.mFD, "the weight of an allocation is the byte size multiplied by the lifespan.

\n"); PR_fprintf(globals.mRequest.mFD, "Minimum weight?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mWeightMin64); PR_fprintf(globals.mRequest.mFD, "Maximum weight?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mWeightMax64); PR_fprintf(globals.mRequest.mFD, "


\n"); PR_fprintf(globals.mRequest.mFD, "By manipulating the time range, you narrow or widen the set of live allocations evaluated. Allocations existing solely before the minimum or solely after the maximum will not be considered.

\n"); PR_fprintf(globals.mRequest.mFD, "Minimum second?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mTimevalMin / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "Maximum timeval?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mTimevalMax / ST_TIMEVAL_RESOLUTION); PR_fprintf(globals.mRequest.mFD, "


\n"); PR_fprintf(globals.mRequest.mFD, "Restrict callsite backtraces to thost only containing the specified text.\n"); PR_fprintf(globals.mRequest.mFD, "This allows targeting of specific creation functions.\n"); PR_fprintf(globals.mRequest.mFD, "Keep all the strings together near the top of this list, as the code will stop trying to match on the first empty string.
\n"); for(looper = 0; ST_SUBSTRING_MATCH_MAX > looper; looper++) { PR_fprintf(globals.mRequest.mFD, "
\n", looper, NULL == globals.mOptions.mRestrictText[looper] ? "" : globals.mOptions.mRestrictText[looper]); } PR_fprintf(globals.mRequest.mFD, "
\n"); /* ** category selection */ PR_fprintf(globals.mRequest.mFD, "By giving a category name, you can focus on allocations belonging to that particular category only like css, js, xpcom. Until we have a flashy new ui to browse categories, use a category name in rules.txt

\n"); PR_fprintf(globals.mRequest.mFD, "Category to Focus on?
\n"); PR_fprintf(globals.mRequest.mFD, "
\n", globals.mOptions.mCategoryName); PR_fprintf(globals.mRequest.mFD, "


\n"); /* ** And last but not least, the submission button. */ PR_fprintf(globals.mRequest.mFD, "
\n"); PR_fprintf(globals.mRequest.mFD, "
\n"); return retval; } /* ** displayIndex ** ** Present a list of the reports you can drill down into. ** Returns !0 on failure. */ int displayIndex(void) { int retval = 0; /* ** Present reports in a list format. */ PR_fprintf(globals.mRequest.mFD, "
    "); PR_fprintf(globals.mRequest.mFD, "\n
  • "); htmlAnchor("root_callsites.html", "Root Callsites", NULL); PR_fprintf(globals.mRequest.mFD, "\n
  • "); htmlAnchor("top_callsites.html", "Top Callsites Report", NULL); PR_fprintf(globals.mRequest.mFD, "\n
  • "); htmlAnchor("top_allocations.html", "Top Allocations Report", NULL); PR_fprintf(globals.mRequest.mFD, "\n
  • "); htmlAnchor("memory_leaks.html", "Memory Leak Report", NULL); #if WANT_GRAPHS PR_fprintf(globals.mRequest.mFD, "\n
  • Graphs"); PR_fprintf(globals.mRequest.mFD, "
      "); PR_fprintf(globals.mRequest.mFD, "\n
    • "); htmlAnchor("footprint_graph.html", "Footprint", NULL); PR_fprintf(globals.mRequest.mFD, "\n
    • "); htmlAnchor("lifespan_graph.html", "Allocation Lifespans", NULL); PR_fprintf(globals.mRequest.mFD, "\n
    • "); htmlAnchor("times_graph.html", "Allocation Times", NULL); PR_fprintf(globals.mRequest.mFD, "\n
    • "); htmlAnchor("weight_graph.html", "Allocation Weights", NULL); PR_fprintf(globals.mRequest.mFD, "\n
    \n"); #endif /* WANT_GRAPHS */ PR_fprintf(globals.mRequest.mFD, "\n
\n"); return retval; } /* ** handleRequest ** ** Based on what file they are asking for, perform some processing. ** Output the results to aFD. ** ** Returns !0 on error. */ int handleRequest(tmreader* aTMR, PRFileDesc* aFD, const char* aFileName, const char* aGetData) { int retval = 0; if(NULL != aTMR && NULL != aFD && NULL != aFileName && '\0' != *aFileName) { /* ** Init the global request. */ globals.mRequest.mFD = aFD; globals.mRequest.mTMR = aTMR; globals.mRequest.mFileName = aFileName; globals.mRequest.mGetData = aGetData; /* ** Attempt to find the file of interest. */ if(0 == strcmp("index.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Index"); displayRes = displayIndex(); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayIndex); } htmlFooter(); } else if(0 == strcmp("settings.html", aFileName) || 0 == strcmp("options.html", aFileName)) { int settingsRes = 0; settingsRes = displaySettings(); if(0 != settingsRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displaySettings); } htmlFooter(); } else if(0 == strcmp("top_allocations.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Top Allocations Report"); displayRes = displayTopAllocations(globals.mCache.mSortedRun, 1); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayTopAllocations); } htmlFooter(); } else if(0 == strcmp("top_callsites.html", aFileName)) { int displayRes = 0; tmcallsite** array = NULL; PRUint32 arrayCount = 0; htmlHeader("SpaceTrace Top Callsites Report"); if(0 < globals.mCache.mSortedRun->mAllocationCount) { arrayCount = callsiteArrayFromRun(&array, 0, globals.mCache.mSortedRun); if(0 != arrayCount && NULL != array) { displayRes = displayTopCallsites(array, arrayCount, 0, 0); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayTopCallsites); } /* ** Done with the array. */ free(array); array = NULL; } } else { retval = __LINE__; REPORT_ERROR(__LINE__, handleRequest); } htmlFooter(); } else if(0 == strcmp("memory_leaks.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Memory Leaks Report"); displayRes = displayMemoryLeaks(globals.mCache.mSortedRun); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayMemoryLeaks); } htmlFooter(); } else if(0 == strncmp("allocation_", aFileName, 11)) { int scanRes = 0; PRUint32 allocationIndex = 0; /* ** Oh, what a hack.... ** The index to the allocation structure in the global run ** is in the filename. Better than the pointer value.... */ scanRes = PR_sscanf(aFileName + 11, "%u", &allocationIndex); if(1 == scanRes && globals.mRun.mAllocationCount > allocationIndex && NULL != globals.mRun.mAllocations[allocationIndex]) { STAllocation* allocation = globals.mRun.mAllocations[allocationIndex]; char buffer[128]; int displayRes = 0; PR_snprintf(buffer, sizeof(buffer), "SpaceTrace Allocation %u Details Report", allocationIndex); htmlHeader(buffer); displayRes = displayAllocationDetails(allocation); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayAllocationDetails); } htmlFooter(); } else { htmlNotFound(); } } else if(0 == strncmp("callsite_", aFileName, 9)) { int scanRes = 0; PRUint32 callsiteSerial = 0; tmcallsite* resolved = NULL; /* ** Oh, what a hack.... ** The serial(key) to the callsite structure in the hash table ** is in the filename. Better than the pointer value.... */ scanRes = PR_sscanf(aFileName + 9, "%u", &callsiteSerial); if(1 == scanRes && 0 != callsiteSerial && NULL != (resolved = tmreader_callsite(aTMR, callsiteSerial))) { char buffer[128]; int displayRes = 0; PR_snprintf(buffer, sizeof(buffer), "SpaceTrace Callsite %u Details Report", callsiteSerial); htmlHeader(buffer); displayRes = displayCallsiteDetails(resolved); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayAllocationDetails); } htmlFooter(); } else { htmlNotFound(); } } else if(0 == strcmp("root_callsites.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Root Callsites"); displayRes = displayCallsites(aTMR->calltree_root.kids, ST_FOLLOW_SIBLINGS, 0, __LINE__); if(0 != displayRes) { retval = __LINE__; REPORT_ERROR(__LINE__, displayCallsites); } htmlFooter(); } #if WANT_GRAPHS else if(0 == strcmp("footprint_graph.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Memory Footprint Report"); PR_fprintf(globals.mRequest.mFD, "
\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "
\n"); htmlFooter(); } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS else if(0 == strcmp("times_graph.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Allocation Times Report"); PR_fprintf(globals.mRequest.mFD, "
\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "
\n"); htmlFooter(); } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS else if(0 == strcmp("lifespan_graph.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Allocation Lifespans Report"); PR_fprintf(globals.mRequest.mFD, "
\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "
\n"); htmlFooter(); } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS else if(0 == strcmp("weight_graph.html", aFileName)) { int displayRes = 0; htmlHeader("SpaceTrace Allocation Weights Report"); PR_fprintf(globals.mRequest.mFD, "
\n"); PR_fprintf(globals.mRequest.mFD, "\n"); PR_fprintf(globals.mRequest.mFD, "
\n"); htmlFooter(); } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS else if(0 == strcmp("footprint.png", aFileName)) { int graphRes = 0; graphRes = graphFootprint(globals.mCache.mSortedRun); if(0 != graphRes) { retval = __LINE__; REPORT_ERROR(__LINE__, graphFootprint); } } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS else if(0 == strcmp("times.png", aFileName)) { int graphRes = 0; graphRes = graphTimeval(globals.mCache.mSortedRun); if(0 != graphRes) { retval = __LINE__; REPORT_ERROR(__LINE__, graphTimeval); } } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS else if(0 == strcmp("lifespan.png", aFileName)) { int graphRes = 0; graphRes = graphLifespan(globals.mCache.mSortedRun); if(0 != graphRes) { retval = __LINE__; REPORT_ERROR(__LINE__, graphLifespan); } } #endif /* WANT_GRAPHS */ #if WANT_GRAPHS else if(0 == strcmp("weight.png", aFileName)) { int graphRes = 0; graphRes = graphWeight(globals.mCache.mSortedRun); if(0 != graphRes) { retval = __LINE__; REPORT_ERROR(__LINE__, graphWeight); } } #endif /* WANT_GRAPHS */ #if WANT_QUIT else if(0 == strcmp("quit.html", aFileName) || 0 == strcmp("exit.html", aFileName)) { /* ** Request to quit the server. */ globals.mStopHttpd = __LINE__; htmlHeader("SpaceTrace Goodbye"); PR_fprintf(globals.mRequest.mFD, "The server is exiting.\n"); htmlFooter(); } #endif /* WANT_QUIT */ else { htmlNotFound(); } /* ** Clear out global request. */ memset(&globals.mRequest, 0, sizeof(globals.mRequest)); } else { retval = __LINE__; REPORT_ERROR(__LINE__, handleRequest); } /* ** Compact a little if you can after each request. */ heapCompact(); return retval; } /* ** handleClient ** ** Read the fd for the request. ** Output the results. ** Returns !0 on error. */ int handleClient(tmreader* aTMR, PRFileDesc* aFD) { int retval = 0; if(NULL != aTMR && NULL != aFD) { char aBuffer[2048]; PRInt32 readRes = 0; readRes = PR_Read(aFD, aBuffer, sizeof(aBuffer)); if(0 <= readRes) { const char* sanityCheck = "GET /"; if(0 == strncmp(sanityCheck, aBuffer, 5)) { char* eourl = NULL; char* start = &aBuffer[5]; char* getData = NULL; int realFun = 0; const char* crlf = "\015\012"; char* eoline = NULL; /* ** Truncate the line if possible. ** Only want first one. */ eoline = strstr(aBuffer, crlf); if(NULL != eoline) { *eoline = '\0'; } /* ** Find the whitespace. ** That is either end of line or the " HTTP/1.x" suffix. ** We do not care. */ for(eourl = start; 0 == isspace(*eourl) && '\0' != *eourl; eourl++) { /* ** No body. */ } /* ** Cap it off. ** Convert empty '/' to index.html. */ *eourl = '\0'; if('\0' == *start) { strcpy(start, "index.html"); } /* ** Have we got any GET form data? */ getData = strchr(start, '?'); if(NULL != getData) { /* ** Whack it off. */ *getData = '\0'; getData++; } /* ** This is totally a hack, but oh well.... ** ** Send that the request was OK, regardless. ** ** Other code will tell the user they were wrong. ** If the filename contains a ".png", then send the image ** mime type, otherwise, say it is text/html. */ PR_fprintf(aFD, "HTTP/1.0 200%s", crlf); PR_fprintf(aFD, "Server: SpaceTrace/0.1%s", crlf); PR_fprintf(aFD, "Content-type: "); if(NULL != strstr(start, ".png")) { PR_fprintf(aFD, "image/png"); } else if(NULL != strstr(start, ".jpg")) { PR_fprintf(aFD, "image/jpeg"); } else if(NULL != strstr(start, ".txt")) { PR_fprintf(aFD, "text/plain"); } else { PR_fprintf(aFD, "text/html"); } PR_fprintf(aFD, crlf); /* ** One more to seperate headers from content. */ PR_fprintf(aFD, crlf); /* ** Ready for the real fun. */ realFun = handleRequest(aTMR, aFD, start, getData); if(0 != realFun) { retval = __LINE__; REPORT_ERROR(__LINE__, handleRequest); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, handleClient); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, lineReader); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, handleClient); } return retval; } /* ** serverMode ** ** List on a port as a httpd. ** Output results interactively on demand. ** ** Returns !0 on error. */ int serverMode(tmreader* aTMR) { int retval = 0; PRFileDesc* socket = NULL; /* ** Create a socket. */ socket = PR_NewTCPSocket(); if(NULL != socket) { PRStatus closeRes = PR_SUCCESS; PRNetAddr bindAddr; PRStatus bindRes = PR_SUCCESS; /* ** Bind it to an interface/port. ** Any interface. */ bindAddr.inet.family = PR_AF_INET; bindAddr.inet.port = PR_htons((PRUint16)globals.mOptions.mHttpdPort); bindAddr.inet.ip = PR_htonl(PR_INADDR_ANY); bindRes = PR_Bind(socket, &bindAddr); if(PR_SUCCESS == bindRes) { PRStatus listenRes = PR_SUCCESS; const int backlog = 10; /* ** Start listening for clients. ** Give a decent backlog, some of our processing will take ** a bit. */ listenRes = PR_Listen(socket, backlog); if(PR_SUCCESS == listenRes) { PRFileDesc* connection = NULL; int handleRes = 0; int failureSum = 0; char message[80]; /* ** Output a little message saying we are receiving. */ PR_snprintf(message, sizeof(message), "server accepting connections on port %u....", globals.mOptions.mHttpdPort); REPORT_INFO(message); PR_fprintf(PR_STDOUT, "Peak memory used: %s bytes\n", FormatNumber(globals.mPeakMemoryUsed)); PR_fprintf(PR_STDOUT, "Total calls : %s", FormatNumber(globals.mMallocCount + globals.mCallocCount + globals.mReallocCount + globals.mFreeCount)); PR_fprintf(PR_STDOUT, " [%s", FormatNumber(globals.mMallocCount)); PR_fprintf(PR_STDOUT, " + %s", FormatNumber(globals.mCallocCount)); PR_fprintf(PR_STDOUT, " + %s", FormatNumber(globals.mReallocCount)); PR_fprintf(PR_STDOUT, " + %s]\n", FormatNumber(globals.mFreeCount)); /* ** Keep accepting until we know otherwise. ** We stay single threaded, as a result page may output ** a URL to an image which requires the same processing, ** and we serialize the caching of that common data by ** only accepting one connection at a time. ** Plus, I can defend I'm still sane by not trying to ** write a full on HTTPD if I stay single threaded. */ while(0 == globals.mStopHttpd && 0 == retval) { connection = PR_Accept(socket, NULL, PR_INTERVAL_NO_TIMEOUT); if(NULL != connection) { /* ** Hand off the connection. ** However, no error for each connection will cause us ** to stop. ** The load time is too painful to bail out here. */ handleRes = handleClient(aTMR, connection); if(0 != handleRes) { failureSum += __LINE__; REPORT_ERROR(__LINE__, handleClient); } /* ** Done with the connection. */ closeRes = PR_Close(connection); if(PR_SUCCESS != closeRes) { failureSum += __LINE__; REPORT_ERROR(__LINE__, PR_Close); } } else { failureSum += __LINE__; REPORT_ERROR(__LINE__, PR_Accept); } } if(0 != failureSum) { retval = __LINE__; } /* ** Output a little message saying it is all over. */ REPORT_INFO("server no longer accepting connections...."); } else { retval = __LINE__; REPORT_ERROR(__LINE__, PR_Listen); } } else { retval = __LINE__; REPORT_ERROR(__LINE__, PR_Bind); } /* ** Done with socket. */ closeRes = PR_Close(socket); if(PR_SUCCESS != closeRes) { retval = __LINE__; REPORT_ERROR(__LINE__, PR_Close); } socket = NULL; } else { retval = __LINE__; REPORT_ERROR(__LINE__, PR_NewTCPSocket); } return retval; } /* ** batchMode ** ** Perform whatever batch requests we were asked to do. */ int batchMode(tmreader* aTMR) { int retval = 0; if(NULL != aTMR && 0 != globals.mOptions.mBatchRequestCount) { PRUint32 loop = 0; int failureSum = 0; int handleRes = 0; char aFileName[1024]; PRUint32 sprintfRes = 0; /* ** Go through and process the various files requested. ** We do not stop on failure, as it is too costly to rerun the ** batch job. */ for(loop = 0; loop < globals.mOptions.mBatchRequestCount; loop++) { sprintfRes = PR_snprintf(aFileName, sizeof(aFileName), "%s%c%s", globals.mOptions.mOutputDir, PR_GetDirectorySeparator(), globals.mOptions.mBatchRequests[loop]); if((PRUint32)-1 != sprintfRes) { PRFileDesc* outFile = NULL; outFile = PR_Open(aFileName, ST_FLAGS, ST_PERMS); if(NULL != outFile) { PRStatus closeRes = PR_SUCCESS; handleRes = handleRequest(aTMR, outFile, globals.mOptions.mBatchRequests[loop], NULL); if(0 != handleRes) { failureSum += __LINE__; REPORT_ERROR(__LINE__, handleRequest); } closeRes = PR_Close(outFile); if(PR_SUCCESS != closeRes) { failureSum += __LINE__; REPORT_ERROR(__LINE__, PR_Close); } } else { failureSum += __LINE__; REPORT_ERROR(__LINE__, PR_Open); } } else { failureSum += __LINE__; REPORT_ERROR(__LINE__, PR_snprintf); } } if(0 != failureSum) { retval = __LINE__; } } else { retval = __LINE__; REPORT_ERROR(__LINE__, outputReports); } return retval; } /* ** doRun ** ** Perform the actual processing this program requires. ** Returns !0 on failure. */ int doRun(void) { int retval = 0; tmreader* tmr = NULL; /* ** Create the new trace-malloc reader. */ tmr = tmreader_new(globals.mOptions.mProgramName, NULL); if(NULL != tmr) { int tmResult = 0; int outputResult = 0; #if defined(DEBUG_dp) PRIntervalTime start = PR_IntervalNow(); fprintf(stderr, "DEBUG: reading tracemalloc data...\n"); #endif tmResult = tmreader_eventloop(tmr, globals.mOptions.mFileName, tmEventHandler); #if defined(DEBUG_dp) fprintf(stderr, "DEBUG: reading tracemalloc data ends: %dms [%d allocations]\n", PR_IntervalToMilliseconds(PR_IntervalNow() - start), globals.mRun.mAllocationCount); #endif if(0 == tmResult) { REPORT_ERROR(__LINE__, tmreader_eventloop); retval = __LINE__; } if(0 == retval) { #if WANT_GRAPHS /* ** May want to set the max graph timeval, now that we have it. */ if(ST_TIMEVAL_MAX == globals.mOptions.mGraphTimevalMax) { globals.mOptions.mGraphTimevalMax = (globals.mMaxTimeval - globals.mMinTimeval); } #endif /* WANT_GRAPHS */ /* ** Create the default sorted run. */ /* NS_ASSERTION(globals.mCache.mSortedRun == NULL, "mSortedRun not null"); */ globals.mCache.mSortedRun = createRunFromGlobal(); if(NULL != globals.mCache.mSortedRun) { /* ** Decide if we're going into batch mode or server mode. */ if(0 != globals.mOptions.mBatchRequestCount) { /* ** Output in one big step while everything still exists. */ outputResult = batchMode(tmr); if(0 != outputResult) { REPORT_ERROR(__LINE__, batchMode); retval = __LINE__; } } else { int serverRes = 0; /* ** httpd time. */ serverRes = serverMode(tmr); if(0 != serverRes) { REPORT_ERROR(__LINE__, serverMode); retval = __LINE__; } } /* ** Done with global sorted run. ** Check for NULL again, may have been realloced at some ** point with failure. */ globals.mCache.mSortedRun = NULL; /* ** Clear our categorization tree */ freeCategories(&globals); } else { retval = __LINE__; REPORT_ERROR(__LINE__, createRunFromGlobal); } } /* ** All done. */ tmreader_destroy(tmr); tmr = NULL; } else { REPORT_ERROR(__LINE__, tmreader_new); retval = __LINE__; } return retval; } /* ** main ** ** Process entry and exit. */ int main(int aArgCount, char** aArgArray) { int retval = 0; int optionsResult = 0; PRStatus prResult = PR_SUCCESS; int showedHelp = 0; int looper = 0; /* ** Set the minimum timeval really high so other code ** that checks the timeval will get it right. */ globals.mMinTimeval = ST_TIMEVAL_MAX; /* ** Initialize globals */ globals.mPeakMemoryUsed = 0; globals.mMemoryUsed = 0; /* ** NSPR init. */ PR_Init(PR_USER_THREAD, PR_PRIORITY_NORMAL, 0); /* ** Handle initializing options. */ optionsResult = initOptions(aArgCount, aArgArray); if(0 != optionsResult) { REPORT_ERROR(optionsResult, initOptions); retval = __LINE__; } /* ** Show help on usage if need be. */ showedHelp = showHelp(); /* ** Only perform the run if everything is checking out. */ if(0 == showedHelp && 0 == retval) { int runResult = 0; runResult = doRun(); if(0 != runResult) { REPORT_ERROR(runResult, doRun); retval = __LINE__; } } /* ** Options cleanup. */ for(looper = 0; ST_SUBSTRING_MATCH_MAX > looper; looper++) { if(NULL != globals.mOptions.mRestrictText[looper]) { free(globals.mOptions.mRestrictText[looper]); globals.mOptions.mRestrictText[looper] = NULL; } } /* ** All done. */ if(0 != retval) { REPORT_ERROR(retval, main); } prResult = PR_Cleanup(); if(PR_SUCCESS != prResult) { REPORT_ERROR(retval, PR_Cleanup); retval = __LINE__; } return retval; }