From: Maynard J. <may...@us...> - 2014-08-11 16:44:49
|
This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "oprofile". The annotated tag, PRE_RELEASE_1_0 has been created at 7c369f889b8f3df82cdd4fe693074d0ad2c1734f (tag) tagging d85b588879b432d8fb6bc2a52f5926d81e3b940e (commit) tagged by Maynard Johnson on Mon Aug 11 11:43:19 2014 -0500 - Log ----------------------------------------------------------------- Last commit point with opcontrol Aaro Koskinen (1): configure: fix test-for-synth check with GCC 4.9.0 Alan Modra (1): Tidy powerpc64 bfd target check Andi Kleen (16): Remove the latency_above_... event for Westmere which requires setting special MSRs that oprofile doesn't support. This patch adds the infrastructure to add extra flags for a unit mask This patch allows to specify unitmasks by name (first word in the unit Avoid extra spaces printed in individual small wordwraps. This fixes Implement the ANY (any thread) extra bit for Intel CPUs. Needed This adds the event list for Intel IvyBridge and the model number Add the Haswell client event lists and model numbers Add EXTRA_NONE flag for use with unit masks empty "extra:" field Add empty extra: lines to every Intel event with a unique first word Use names to make all non-unique Intel default unit masks unique Add some missing Haswell model numbers Add support for Intel Silvermont processor Add oprofile support for Broadwell microarchitecture Fix some problems in the Broadwell events Improve error message for non-unique unit mask Update the Haswell events to the latest version Andreas Arnez (1): Determine s390 cpu type from /proc/cpuinfo for operf/ocount Andreas Krebbel (4): S/390: Enhance the user space tools for System z hardware sampling Fix mis-spelling of S390_HW_SAMPLER_BUFSIZE variable Fix up s390 implementation to match what was accepted upstream in the kernel Add support for IBM zEnterprise EC12 (zEC12) Antonio Rosales (1): Fix compile error on Ubuntu 12.10 Breno Leitao (2): New man page for operf command. Add the new operf.1.in file for the new operf man page Carl Love (12): operf: print lost sample warning only if lost samples exceeds threshold Oprofile, opreport: fix to pass string of options from opannotate to objdump Oprofile, operf: Expanded error message when running multiple operf sessions Oprofile operf: Fix the code to strip the _GRP## from the event name operf, add throttling and multiplexing stats operf, remove support to report multiplexing. OProfile, fix the units for the reported CPU frequency Ocount, print the unit mask, kernel and user modes if specified for the event Duplicate event specs passed to ocount show up twice in output Add support for getting the Kernel symbols from /proc/kallsyms Fix kallsyms support for callgraph and debug-info opreport options opreport: header timestamps are different for kallsyms file Changbin Park (1): Ensure parsed_filename's jit_dumpfile_exists variable is initialized before use Daniel Hansel (7): JIT support (for profiling Java applications) added JIT support Fixed call of execvp() to execute opjitconv if it is installed in a custom directory (option "--prefix") Changed license of JVMPI agent from GPL v2 to LGPL v2.1 Fix memory leak according to opening opd_pipe Fix memory leak according to opening opd_pipe (added missing changes) Change location to store intermediate JIT dump files Dave Jones (39): * pp/oprofpp.c: use EXIT_FAILURE | EXIT_SUCCESS in exit() calls. Oops. Missed these ones. Film at 11: Programmer updates documentation. Update ignorance list. op_help was printing out... Hmm, oprofpp needs to support 4 counters where available. Update so that it doesn't say Intel only. remove op_start (Will re-add in a minute to see if it'll become +x) Executable bit on op_start set. oops, forgot yesterdays changelog Point Athlon users at the Athlon docs instead of Intel ones. This never happened. Mmkay ? Change MSR defines from MSR_IA32_xxx from MSR_P6_xxx, based on info on Add missing cvsignores Add recognition for alternative PIII string. Fix PPro recognition. * Makefile.in: Make documentation build again. Readd to CVS. * pp/op_time.cpp: fix for printing (NaN%) in the zero sample case. We can pretend an AMD Hammer in 32bit mode is an Athlon. Improved cpu detection. tiny cleanup libop/op_events.c: Add x86-64 specific events. libop/op_events.c: Add unit mask for Hypertransport events. s/HT cache/HT/ K8 HT events are utm_exclusive, not utm_bitmask Add additional K8 unit masks. * libop/op_cpu_type.c: Clue the user in on why the cpu type isn't * libop/op_events.c: Doh, unit masks use bit numbers, not values. * utils/opcontrol: Trying to use the options --pid-filter=pid and * module/x86/op_model_athlon.c: Don't poke reserved bits in the counter. * libop/op_events.c: segregload is utm_mask not utm_exclusive Reenable building for hammer. * module/x86/op_apic.[c|h]: move NMI gate setup to architecture remove dupe minimum tag * oprofile/module/x86/op_nmi.c: Fix another possible race condition. Remove stale debug code from yesterdays commit. * events/x86-64.hammer.events: Check for APIC on Athlon before enabling it. Gergely Kis (1): Detect MIPS CPUs based on the "cpu model" field of /proc/cpuinfo Gilles Allard (3): Fix for oprof_start when daemonrc file does not exist Don't show irrelevant bits in unit mask sub-window Fix size problem of oprof_start GUI Graydon Hoare (8): reviewed and okayed by Philippe Elie (with comments added) patch okayed by John Levon, Philippe Elie [ ChangeLog ] [ ChangeLog ] [ ChangeLog ] [ ChangeLog ] check bitmasks as well as exclusive values, for unit masks Handle printing multiple bitmask values Joakim Tjernlund (1): Print correct clock speed when using timer mode John L. Villalovos (1): Patch to add support for Westmere-EX processor. John Levon (1286): CVS Import Fixed max buffer size with kmalloc(). Deleted unused file Write out mapping information. Add sprofile.s target Removed duplicate target. Use -march=pentiumpro option. Use "asmlinkage" Add thread exit semaphore. Implement new mmap-based design in daemon. Removed cruft from repository mknod at the correct place. Hack for binary loader OP_DROP notification. Offset kernel/module image entries to allow for "negative" entries. Minor fixes. Minor documentation fixes. recursive make clean Removed duplicate. Allow environment $(CFLAGS) if set. Use "uint". Use uint/ulong. Use uint/ulong. Use uint/ulong. Removed cp_events.c duplicate. Removed hack for no sys_execve() capture. Added $Id$ Removed useless source. Temp. utility files go elsewhere. Remove useless sources. Fix srcdoc target to depend on filename. Fix broken usage of perror(). Simplified use of opd_handle_kernel_sample(). Use CLONE_* for kernel_thread(). Last bits of pp code. Better initial hash/buffer size values. Compare image strings starting from end. Should be faster as most Added srcdoc script snarfed from kernel. Use autoconf. Moved from dae/ Use autoconf include config.h Correct location of opd_util.h Fix LIBS typo. Added distclean target install target fixes Use version.h Remove duplicate flags. Fix reading in of footer. Only use preferred-stack-boundary if available add HTML source doc files to clean target. Add source docs. Merged in aclocal.m4 to configure.in. No need to inc/dec mm_users. Check for ld. Replaced un-necessary u8's with int's. Replace spinlock with mmap_sem. Comment fixes. Support --version/-v. Subst $(CC) Remove pointless kernel-as-zeroth-mapping. Only collect symbols in section with flag SEC_CODE. Intercept old_mmap(). Fix oprof_output_map(). Code cleanup. Code cleanup. Automatically set OP_EXPORTED_DO_NMI. Add tags target. Fix to arch check. Fix oprof_outstr(). Don't export any symbols. Code cleanups. Reintroduce necessary map_lock. Use map_lock in read method for (unlikely ?) clash with oprof_output_map(). Remove debugging code. Structure order fix. Get ASCII procs earlier. Update MODINSTALLDIR for new modutils kernel/ dir. Increase thread wake up frequency. Include hash map. oprof_opened changed to be x86-atomic to fix race possibility. Use hash map for generating maps. Just dumping Read from hash map. Typo fix. Typo fixes. Added .cvsignore files. Added config.* Atomic exclusive open. Atomic device open. Set up hash map device. Small error message fix. Fix hashmap type. Fix hash_map type. Small fixes. Disable interrupts for sensitive APIC setup. Various minor fixes. Should always output a number into the map buffer, even Should always output a number into the map buffer, even Added FIXME comment. Include watermark to set ready some time before reaching the One-structure header contains how many samples to read in eip member. Watermark for out8() as well. Remove debugging code. Hike watermark. Allocate hash_map on module load. Means daemon shouldn't struct op_sample shouldn't be aligned. Added FIXME Fixed silly unit mask typo. If at read() time, nextbuf hasn't overflowed, it should be Fix path name construction. Added FIXME. module->name not used, so it was removed. Require CAP_SYS_PTRACE. Moved variables into .c file ! Clarified opd_map_offset(). Make opd_is_in_map() inline. Small cleanup. Add feature to filter on pid or pgrp. Fix typo. Cleanup hash_access() macro. Small tidyup. Forgot unlock of map_lock. Stop profiling if module is loaded. Read mappings on OP_DROP. Clarify code. Cleanup printing of counter details. OP_DROP was a terrible name ! Debugging output fix. Add debugging output. Fix offset calculation. Update documentation. Fix offset. Small cleanups. Ignore crufty symbols. Should read from map device even on failure path. Remove debugging code. Check CAP_SYS_PTRACE capability on read(). Added debugging code. Use vmalloc()/vfree(). Increase watermark. Add a process even on error paths. Debugging output. Comments. Added FIXME for those who change PAGE_OFFSET. Comment fix. Temporary hack on module unload race. Use dprintf(). Use oprof_ready rather than global variable mapbuf_wakeup. Correctly offset kernel samples. Signal every 10 minutes not 20. flush stdout after printing stats. Increase default buffer + hash table sizes Implement process list hash. Only copy some of the buffer into userspace. Fiddle with parameters. Install in all-new modules location. Up pre-watermark. Fix reporting of map array depth. Only use bfd when it is opened properly. Add facility to specify binary image. Fix kernel image offset handling. To get working results on a multi-section file, we need to sort Add chmod's. Cleanups. Cleanup for --help option. Use ioctl(). Print sample percentages correctly. To do list. Make list_symbol output more like other things. Updated. Started some documentation. More updates. #define -> enum Expand um descriptions a little. Module is not unloaded anymore, there is an ioctl() interface. Added comments, added FIXMES to check. Slight fixes to srcdocs. Various cleanups. comments. make dist Update COPYING. Introduce sysctl, various other changes. oprofpp -l /bin/mv and friends. Added website address. update docs, change default source directory to use /lib/modules/`uname -r`/build Some small cleanup and allowing for UP NMI oopser. See ChangeLog New mapping rework, several important bug fixes Started results directory. Increase watermark Small fixes, results stuff update docs Use strcpy not strncpy in do_hash(). sync with the 24/1 - more updates to come. compile with -march, full clean sysctl, various fixes + improvements non-recursive do_hash(), minor fixes, more results gprof output working, add md5sum stuff, re-write docs and add manpage add URLs fix docbook work with 2.4.3, remove forced dumping when quiet bump version to 0.0.4 fix indentation. fix shell script when no parameters passed, unit mask fixes from Philippe remove libiberty header include add philippe's prototype tcl interface don't output logfile message on stderr fix CONFIG_MODVERSIONS op_start, tcl/tk fixes, new output options for oprofpp added latex run results unit mask fix for mmx from philippe, also better um description code. fix previous fix from philippe from philippe fixes for theoretical memory leaks on failure, s/sourceforge/sf/ make unloadable module possible. edge detect and footer expansion patches, small fix for unload option spacing and dead code fixes always have smp_apic_restore() update readme oprof_convert added. cvs ignore fix philippe bugfixes, remove rmmod in op_stop warnings fixes, malloc attribute, remove broken line bug fix and gui improvements from philippe more tcl updates, hide malloc compiler warnings add the source annotater ! commit silly fixes clarify vmlinux, remove generated files from CVS fix dist bug docs from philippe, and oprofile gui fixes small docs fixes from philippe several clean ups and bugfixes from philippe back up old sample files if the profiling details don't match update remove ability to have different values on different CPUs. I probably missed some docs on not setting silly counter reset values extra bit of warning docs stuff philippe: event count value minima, separate values for gui, small op_start fix filter patch from philippe fix to bug reported by Bob Montgomery, use mtime not md5 path canonicalisation patch enable unload and filter unconditionally. small fixes and v4 convert support. Philippe, please test/check my convert philippe's fix for me vomiting all over the tree, plus small beautifications tentative fix for no-symbol modules case include right header, small install fix whitespace changes whitespace convert fix fix introduced memleak bad module unload problems fixed, small fixes elsewhere small tidy compile fixes use cli/sti instead of explicit masking for installing the NMI. Hopefully small corner case fix when we miss a sample against a full count entry support 2.4.7 and above I got bored of typing echo 1 >/proc/sys/dev/oprofile/dump ... argh, versioning nightmare with complete_and_exit 2.4.8 still has up_and_exit masking again during set_gate fixes to the hash map ! last minute file leak patch from philippe make install fix bump version use op_dump qm.diff small fix uninstall patch from philippe symbolic changes from philippe. lots of stuff TODO ;) the athlon patch. more work needed, but it works. Thanks Dave !! recognise celerons in cpu type mptable parse philippe's huge patch philippe's patch remove bogus code small improvement to op_dump don't unload on SMP fix for lvtpc_apic_restore doc tidies s/CPU_TYPE/CPUTYPE in op_start a couple of fixes philippe's cleanup, and several fixes from me. extra sanity checks in op_start fix core dump pedantry cpu type detection fixes, irq stats move some code, remove some stale comments fix race leading to oops ! code tidies post-prof fixes for 4 counters, and some other changes. two small changes some pp cleanup add missing files autoconfiscation for Qt2 couple of bug fixes, fix mtime and changed binary issue (I doubt the over doc fixes, implement image hashing in daemon as far as I've got with the re-jig so far. Philippe, sorry, I've trampled over more gui changes - I moved the _impl file into oprof_start.cpp as I didn't nothing interesting ... .cvsignores nothing interesting ... d'oh ! gah ! likely/unlikely support. needs fixing and testing with pre13 Small fixes + big GUI update. some gui tidy. fork/exec for children in the GUI some more tidying boring little cleanup 2.4.10 support. small cleanup build fixes. robustness for script a little more remove tcl gui temp fix for compile error. philippe - iterator on std::vector<const char *> is not same as const char ** Fix silly bug. Dave, can you try again ? more robust whoops argh ! never mention this to me, or I will die of shame. fix is_profiler_started use pidof instead pidof not /sbin/pidof minor fixes flashy GUI stuff. or something. speed improvements rename to oprofpp.cpp per-counter config for GUI. Please test for corner cases I missed. fix GUI crash avoid temp files fixes for g++ 3.0 various fixes. add email address at request verbose support bump version for release two small fixes some little more fixes add AC_FIND_FILE back temp remove broken test update TODO fix config bug :P last fix, forever :) more build fixes. Are we ready for a release ? changes to filtering. Philippe, what do you think ? -> 0.0.6 remove extra debug line 0.0.7cvs (0.0.6 has been tagged as RELEASE_0_0_6) string pool no kernel thread use a separate map buffer. fix for likely()/unlikely() whoops, missed a bit cleanup - /bin/rm -rf /root/.oprofile ! mandrake workaround #1 - please test on kernels 2.4.8 and below if you have save/restore msr values on load/unload man fix, msr restore fix another mandrake work around - please check configure still works typo some fixes from H.J. Lu $DESTDIR support use $(LD) detect laptops. build re-org add notice about broken sysctl API move tests changelog over fix nr_interrupts to actually do something ;) allow_unload on UP fix error message test for and use hard_get_current(). fail if hard_get_current() exists change to 0.0.7 - no patches now please !!! fixes for release mark 0.0.8cvs bob's fix update to do a little some fixmes pedantic bastard ... whoops ! minor changes, Michel's patch move config.h.in to maintainerclean deps files in .deps - Philippe please check use enum op_cpu for cpu type little extra fix doc fix sigusr1 stuff remove config.h.in - run ./autogen.sh again whoops ! enable profiling again :P improve docs allow EAGAIN through read_device ... oops fix for note_pos remove the two config buttons, save on WM hide more updates/reorg etc. pin note pos to end on overflow. more wakeups are caused now in this case, new iteration of clean stop code. hopefully this one is OK. shutdown hang fix. print stats on exit minor crap remove extra dump little fixes, slight re-org change the partial_stop change. Phil, look please, if things are still wrong minor cleanups cleanups changelog dump stop fixes add security FIXME future fixme... loop waiting for start/stop ... add fixme update TODO gcc 3.0 and std:: - incomplete ... disable pednatic in util/ to allow compile. It will do ... fix small configure problem, bump to 0.0.8 - don't apply anything :) 0.0.9cvs ! fixes to compile on 2.4 add FIXME for 2.2 support little changes to allow build ... add fixme minor fixes clean up dcache. Some pretty major changes with regard to 2.2 and remove kernle header s from userland add doc for --with-linux absolute paths update todo Fix problem with op_dump shuttting down oprofiled. FIXMEs whitelist arugments rotate changelog for philippe add reference to older changelog fix defaults OK, step #1 to RTC interrupts. Each type of interrupt is hidden behind a RTC patches #2. This gets the RTC working with a fixed value. Set FORCE_RTC rtc round #3 fix remove unneede check fix op_start && op_stop nothing to see here, move on please ... some more RTC improvements. fix build fix RTC build for 2.2 modversions + 2.2 fix for RTC disable 2.2 SMP compile for now. split headers, little cleanup ->owner fixes x 2.2 build, rtc kernel/user, 2.2 NMI wake_up fix (maybe) fix ->owner problem, thanks phil split NMI asm code. phil, please look carefully. I've checked 2.4 still doesn't bump note size + watermark use bkl around execve fix deadlock on 2.2 SMP remove superfluous BKL lock sorry everyone ... remove this... see changelog, nothing interesting remove broken code for now.. MSR check and 2.2 SMP fixes. missed to commit thi missing changelogs. wake up fixes, minor clean missing cli thanks phil APIC / fixmap changes. Phil, this is slightly different from the one one more small fix depmod fix.. pte_page fix phile, I punted some more things out from this release - complain if you don't like something ... fix remap problem simply boring hacky output cleanup gcc 3.0 fixes - phil please look ! 2.2 remap fix oopsable bug spotted by Bob ! bump to 0.0.9 - no commits. bump version to 0.1cvs boring sysctl fix rtc patch from bob whhoops, wrong version committed... boring changes readability finish previous commit thoroughly untested map/note lock changes local symbols allowed update todo minor. OK Phil fix those bugs I asked you about and we're ready to roll right ? remove pedantic bump version, fix gcc3 open tree again, update TODO Fix build. a couple of the things we need to do for oprof_report some laptops work in proper mode ... comment fix to reflect situation with binutils fix build for some kernels move to docbook XML ... Phil, this is the sort of thing I'm talking about. a couple more things. I want bugzilla, badly. remove oprof_convert little whoopsie in configure.in Bob, Alex - prospect will now receive samples with EIP == 0x0, I suppose william's patch + a slight extension Get correct summaries for the test kernel module. Still to do : move pid/pgrp filter. Remove ignore-myself fix permissions problem. Most files touched, but actually nothing particularly exciting : remove results/ dir disable oprof_report by default for release couple more fixes. We're ready to go. Test test test ! prioritise TODO list a little some doc improvements bump to 0.2 open up to 0.3cvs nitial sweep at a 0.3 plan Mega-patch ! It builds now, but there are still some ... troubles remove util/ properly ! back out experiment I didn't mean to commit re-write module/Makefile.in some doxy changes Last tedious patch Sorry, I lied, I lied. missing backslash ! a couple of boring things various minor things some minor changes Qt3 patch from Cliff Woolley William's patch, and a little more gah missed this bit more boring splitting up and moving around. a small lock file change missing fi op_time crash fix, and follow on from it doc changes for libopt++, minor stuff Re-organise dae. Still needs review + test, but I'm tired restore lost commits ... Forgot to tell CVS about this dying. uncommit debug line very minor changes fix bug 563624 where shared library samples don't show up in oprofpp -kl fix DOCDIR path Some more refactoring... More refactoring (start libop++) move demangle_symbol Missed from previous commits Move sample file mangle/unmangle to libop++ remove not-so-useful check More re-org Factor some more code (derive_files) move op_bfd to libutil++, and add and use verbose_ostream Remove oprofpp_util, in favour of counter_util, etc. small cleanup 2002-06-05 John Levon <mo...@co...> small tweak fail if !root Some minor changes, turn on -Werror A couple more tidies, allow compilation with The Compiler That Time Forgot a couple of small bug fixes fix build for me and a couple of tweaks minor fixes forgot to finish the patch ... whitespace, use check_range in op_rtc move apic compat stuff around replace verbose_ostream with something simple small cleanup minor change add a scoped_ptr. scoped_array forthcoming fix blunder, add scoped_array Use Rules.make Some module fixes Some cleanups Fix a regression with check_mtime() Remove FIXME A number of small fixes a little refactor Use a structure for the module sample buffer. nothing really tiny cleanu A couple of bug fixes small cleanup update credits Drop our requirement for System.map : use vmlinux instead. Use a state variable { STOPPED, STOPPING, RUNNING } instead of three fix last samples on shutdown notification move eip 0 check into module again, at better place. Add some debug, fix a make indiciate fix for cpu hotplug race add two fixme's bump to 0.3, release time resize fix from Kevin Puetzk 0.4cvs, tree is open... Will's counters patch, add some another log stat Will's uname patch Will's printf patch Mention Qt 3 Will's 64 bit patch new Qt script fix moc/moc2 problem another bit Will's std:: patch tweak fix two comments give an example for unit mask usage. *I* got confused - this is bad ;) Fix a serious bug with multiple text sections. The model specific patch. match kernel MSR naming Merge part of the 2.5 patch userspace stuff. I'm going to merge little bits sigh. fix nr. 1 missing changelog accidental cvs clash commit removed Introduce new kernel/uspace interface and use where appropriate. Config split into old/new oprofile. GUI will only work in 2.4 for now. add missing header CPU RTC fix crash fix for rtc 2.5 patch start script 2.5 patch fix Graydon's P4 patch Some more 2.5 patch tweaks Will's TC_DELIVER fix 2.5 daemon fixes. missing commit cpu_type for 2.5 fix Fix. Still segfaulting practically immediately. GRRRR. Fail on P4 HT from Will nothing interesting .. update to 2.5 interface check mount from Will, and mkdir from me fix broken grep for the mounting Graydon's ABI patch. Doc patch from Graydon namespace from Will Upgrade to latest kernel API patch from Will fix for HT from Alex bump to 0.4 - tree frozen tweak to 0.4 re-open the tree. Welcome to 0.5cvs Initial IA/64 patch remove 2.5 support from old module. CONFIG_OPROFILE is the new way. Hold on to yours hats kids, this is Phil's move to automake remove another old makefile A couple more fixes more fixes. TODO: make --enable-abi work again use foreign for autogen.sh Make separate objdir work. module build fails but all is OK with --with-kernel-support small tweak More tweaks. Confirmed to build on automake 1.7.1 / autoconf 2.54 abi tidy Some steps to get --enable-abi to work again more .am fixes another tweak Will's Ia 64 doc patch Make include/sstream work again X libs fix fix --enable-abi don't dist dummy.cpp add FIXME a configure fix ABI fixes (de inline, gcc 3.2). Fix daemon Makefiles. Use popt:: namespace fix in docs anal crap proper fix for daemon c++ link remove kernel_only from 2.5 disable Pentium IV: it hangs Some small things, ignore s/db_tree_t/samples_db_t/ and related changes fix --deinit and another fix couple of minor cleanups remove annyoing AC_REVISION silly accidental commit fixed some --reset/--save fixes Mark op_dump/session as deprecated more deprecation. anything left ? --version for opcontrol docu updates for removed stuff little changes some cleanup, more on the way more format_output changes some more cleanups in format_output missing std:: small pedantries a small cleanup some more docs Lots of doc fixes and cleanups remove oprof_report - it breaks make dist Fix instlaling "op_start" as "op_start_25" Remove the ln -s arch hack. Instead we set path with -I and use AC_SUBST. Some more fixes for the build. make distcheck works now (when not module compiling, naturally) Rotate changelog Implement --start-daemon, some fixes ascii-string cpu types support. Not yet used ... 32/64 fixes w. /proc/kcore move complete_dump add opd_cookie.h Bail out if lookup of dcookie fails. We will fail later and more obscurely anyway fix ELF read from Randolph P4 HT cpu support. don't call op_help oprofiled that's on the path, prefer one in same dir as opcontrol first. Some TODO updates. What did I leave out ? Add CodingStyle and HACKING. Contributions please. remove -Werror for module code. Kernel headers too fucked. Falk's patch. add nr_blah vs. no_blah Two ppc fixes from Anton Format vma width right on 64 bit. Revert previous, bfd_vma is 64 bit on phil's x86 binutils disable x86_64 for 2.4 bump to 0.5. The tree is open again. Will, please commit the manpages patch Stale lock file detection cleanup and fix error message compile fix compile fix here too Some GUI fixes for recent opcontrol change, plus some 2.5 UI improvements. remove oprof_start item mention SIGTERM stuff couple of things throw() specifiers update update Module dropping for 2.5 stats fix paranoid tweak Fix that blasted separate-kernel bug ! Fix Params used vecho fix make dist fix X, non-Qt build say something about patches and cvs compile fix stuff Some projects run configure on autogen.sh. We don't and we shouldn't, so quickstart Fix up OP_DATA/BINDIR another rename, small fix Remove sanity checking from 2.4 module - read changelog oprofpp needs -P tweak bump to 0.5.1 Fix the dist tarball to include stl.pat Tree is open Fix IS_TIMER for cpu-as-string use % not %-age small fixes fix EV4 mis-detection Default event selection. Needs some more work (see TODO) update Scale the default event count update a little Events in separate text files instead of C. Fix blank line restriction remove /proc/ksyms reading for 2.5 todo update and some trivial TODO->FIXME. I can't be arsed with a changelog - so sue me .. some more re working nstall + little more doc tweak Turn kernel off when --separate=kernel some doc cleanups Allow disabling of --separate and --pid/pgrp-filter. Still the counters changelog Chris Mollers libdb patch + some changes on top More doc cleanups. op_merge --use-counter -> --counter oops add --ctrX-event=none fix gui for --separate=none, --ctrX-event=none opcontrol fixes. First steps towards making the GUI use daemonrc (this Pick up the event configs from daemonrc. Needs testing please ! tweak parser improvements --version --help for opcontrol now behave a bit more sanely --no-vmllinux for 2.5 kernels. It works for me. doc build tweak when no xsltproc etc. tweak the xsltproc test only sleep 2 after dump on 2.4 Fix ::is_prefix for older gccs Make --setup optional (see changelog) small opcontrol fix mention doc/www pedantry document autogen.sh in README Fix make distcheck Some op_help UI improvements Use XML catalog more docbook changes fix include more db macro fixes module handling fixes in daemon/ More module fixes for 2.5 small rename more 2.5 modules fixes mtime warning improvement opcontrol --dump fix doc some 2.5 module stuff Fix typedef.m4 for 2.13 move some macros to m4/ More configure.in moving about fix quoting problem Bryan's patch, doc patches remove stupid thinko Doc improvements fix dist bump to 0.5.2 - tree is frozen tree is open. Remove the old scripts at long last. fix make dist use libs explicitly rearrange code a bit small op_to_source cleanup some op_to_source work. Output info as footer not header when separate More small op_to_source cleanups. Phil, please look at the FIXME some renamings sed a renaming from the last sweep prepend annotations for op_to_source opcontrol fix line 0 fix another output tweak big cheat ... describe oprofile-www tiny tweaks Compile fix for IA64 from Alex validation errors small fix compiler option testing remove cpu speed check some comments on whitespace typo fixes remove spurious ws buffer rewrite, and fix a bug in a race window exposed tgid code handling update docs update back to 0.6cvs disallow --rtc-value when needed update update remove some unused abi code build fixes and remove some unused code update fix operator< unique_storage<I, V> more unique_storage cleanup hide some unique_storage stuff away. hide .id tiny tweaks ... make up an ocean whitespace fix (test) test(ish) another test yet another test std::pair not pair opp_symbol.h -> symbol.h and some header cleanups fix make dist(check) future-proof the code handlers for CODE_CTX_TGID athlon/hammer event fixes from jas...@am... mention timer bug in TODO mention opcontrol nastiness missing typename handle a truncated tgid entry properly. Fix the TGID fix removing an annoying verbose printf a vague release schedule The first parts of automatic counter initialisation in opcontrol docs for opcontrol --event= s/RTC/RTC_INTERRUPTS/ make unit mask, user, kernel optional in the event spec. Verify that credit jason yeh, fix changelog c89 fix update TODO c89 fix 64 bit fixes fix ChangeLog fix autogen.sh instructions fix typo remove some apparently bogus FIXMEs I added for the Hammer unit masks fix autogen.sh command example make "opcontrol" show the help message accept --foo blah as well as --foo=blah accept short forms of some opcontrol options Fix -e. echo has to be the worst designed utility in the world, which is generate Doxyfile froom autoconf don't accept -r - too easy to typo don't build opdiff hide opannotate's --base-dir option fix make distcheck update .cvsignore CPU_TIMER fixes, plus some trivial stuff w/o changelog move the default events into op_help. Needed step to fixing the GUI for fix default RTC event name fix op_help -c, move default event code into the library start fixing the gui for new events stuff more GUI fixes update todo CPU_TIMER_INT GUI fixes another timer int fix for opcontrol (damnit !) remove some done things from TODO, move some minor things to post-1.0 t's TODO update day ! testing something ... more testing Add some tentative insns on cvs branching. use default event in the GUI remove the uptime pretty printing from the GUI fix default event stuff when daemonrc exists select "No event" on a switch if needed UI fix for when vmlinux specification is wrong fix last cleanup (which was actually a fix) fix the fix to the cleanup (which was actually a fix) small fixes remove mention of oprof_start_event (it's gone) comment out doc entry for --base-dir small tweaks to the messages small tweaks select the default event replace folder icon update fix op_help call bump to 0.6 in preparation for release bump to 0.7cvs remove bitrotted oprof_report update manpage for --event fix a potential problem due to libbfd bogosity add sample-file: to TODO make flame_uops mandatory remove some apparently dead code fix _List_iterator matching avoid using ps in opcontrol, it's dog slow missing std:: remove --source-dir/--base-dir from opannotate (perhaps temporarily) --search-dirs for opannotate add --base-dirs update todo several API changes; --reset fixes update TODO missing s/void/int/ remove duplicate opd_printf.h remove unused mypid variable share some daemon code fix share opd_write_abi() more code sharing fix make dist whitespace change Some renamings to start getting rid of the notion of counters from pp code. more count_group changes, removal of pp_nr_counters. Now libpp/ itself allow event:BLEH,BLAH etc. count_array changes make sure opcontrol defaults to the default unit mask if one is not The Moderately-sized Renaming some docs updates The Son Of Renaming trivialities remove "zero samples" hint, always false small cleanup update HACKING s390 fix update comment about gcc2_compiled. being in the "wrong" place read /dev/oprofile/pointer_size. /proc/kcore is dying. ouch... report the kernel pointer size (see "report more details in the log missing std:: Check for duplicate tags in the event descriptions. fix err message abort() for "can't happen" errors propogate event error back up missed from last commit mention --reset in "Getting started" clean up change log small cleanup correct headers in op_lockfile.c correct some more headers configure cleanly on freebsd minimal docs for incorrect attribution warning 0.6.1 version bump open the tree again to 0.7cvs arrange_profiles, and make opreport use it opgprof - use arrange_profiles API op_annotate - use arrange_profiles Ding dong, the old code is dead (ish) rename split_sample_filename, sort profile classes Add code for naming classes Long names for each class, and use them column headers for image report more image report col header changes Turn off address show by default (for non-details), add --show-address remove dead FIXME remove dead TODO item invert_profiles - restore 0.6.1 speed of opreport, and some more Expand the commentary Use inverted profiles for opannotate and share the code fix -p for opgprof fix -p for opreport/opannotate detailed errors for too many unmerged profiles update TODO header fix more arrange_profiles commentary document --show-address -Wdeclaration-after-statement, update 2.5 refs to 2.6 small header file cleanup missed a 2.5 ref and another and some more fix link error daemon/ fix, some cleanups, rlimit thread profiling fix small logging improvements "separate" variables renaming updates from IRC locate_images cleanup, fix op_bfd for fakeness use inverted_profile API in opgprof push image flags down to the clients. Handle bfd format failures nicely update TODO small fix for last patch mage error flags cleanup remove unused popt additional help missing include from wcohen start an internals manual remove count group naming, add a glossary and overview to internals make -C doc/ clean fix back out make clean change add a minimal kernel todo disallow --global-percent when it's meaningless mask SIGTERM too in the critical section add comment Phil - ELOOP refers to path components not the link target itself fully resolve sym links passed on the command line move op_get_link to op_file, fix doxygen comment placements remove /proc/kcore support remove support for a missing CTX_TGID (i.e. older 2.5 kernels will break now) big commit: the oprofiled rewrite. See changelog for details. Woo, et cetera. remove the err_msg stuff in favour of errno-style returns per-cpu profiling is here bump LR_AMOUNT to 1000 rlimit 8192 -> 2048, RSS 38Mb -> 10Mb when --separate=all patch from marc herbert for stow support tiny fixes small cleanups in libabi I tested EMFILE libutil tests - note the rel_to_abs_path failure, phe what's the issue ? don't build the tests until "make check" time don't error out on "-x -m all" options workaround for recently introduced --shutdown race causing Fix Qt 2.3.1 compile some docs on why you might not get any results fix IA64 compile small cleanup remove --pid/pgrp-filter, --kernel-only update TODO for minimal stuff wanted for 0.7 random cookie failure works fine Perfmon support. mitigate signal race a bit - needs fixing forgot to cvs add small cleanup remove pfmlib support (well that lasted long !) in favour of hand-coding SIGTERM change for perfmon children fix xmemdump typo Use opd_parse_events API for 2.4 daemon. Fix multiple counters foor merge some more code, ~50 lines of opcontrol 2.5 references -> 2.6 itrivial cleanup of opcontrol style merge option handling into opd_util, 118 insertions(+), 253 deletions(-) perfmon fixes add some fflushes disable perfmon code use "--merge" consistently use a fixed value for the default event count allow non matching axes if all class's tid == tgid nicer error message on clashes profile class axes sample file header changes. version bumped. update todo doc fixes add thread separate gui + fixes re-enable perfmon, only enable if !timer fix make distcheck, bump release to 0.7 tree is open again more internals docs perfmon cleanups remove done TODO more internals docs remove outdated stuff from the manual add a diagram to the internals manual ifixes for 2.6 image filtering ipudate TODO small tidy update TOODO again Finally ! Merge the 2.4 daemon code into daemon/liblegacy doxygen fixes, reduce cookie hashstable size update TODO cleanup and improve TODO small cleanup small perfmon cleanup remove an unused include path from liblegacy opd_proc header cleanup share a little more daemon code document image filtering Make empty filter be --image=all specification. Clean up help message, Patch to oprofile(1) from Jason Lunz fix typo be rewriting error message missing std:: fix make check fix typo too string_filter unit tests remove emails from @author lines, no changelog improve chunk rules for docs Add unit tests for comma_list. make check now fails, as an issue some small makefile cleanups fix up confusion concerning generic_spec vs. comma_list make the comma_list matching understandable to a moron like me fix compilation on IA-64 dirname/basename/follow_link fixes small additions to CodingStyle --disable-werror, --disable-optimization remove unused -DKVERSION from gui minor Makefile.am cleanup bump to 0.7.1; tree is closed bump to 0.8cvs update TODO a bit use realpath(3) move bad_regexp ctor fix typos mention daemon auto-restart in TODO language cleanups, only allow --verbose at sensible times add HTML docs to SEE ALSO in man pages. check_style pedantry, no changelog. whoops some minor cleanups rename env var to OPROFILE_EVENTS_DIR check_style fixes, no changelog internals doc updates add some TODO improve error messages (from Anton Blanchard) Fix a BFD leak on format failure, plus a small race window when the improve the opstack docs Fix up syntax, been broken for a while it seems push some things back past 0.8 release s/dependant/dependent/ bump to 0.8 - tree is closed add some opstack weirdnesses some schedule updates etc. tree is re-open 64-bit fixes fixes for relative paths and opannotate small follow up fix hopefully fix dcookie aliasing bug, by Bin Ren clean up libdb API a little some trivial renamings add comment about bracing around if/while/for fix bug 964097 (event code of 0x00). Not tested. There's also a 2.6 add opstack(1) to oprofile(1)'s SEE ALSO remove sample-file:/binary: (praise the Lord). Allow either "lib" or include call graph files in oparchive quash timestamp warnings if displaying profiles from an archive document some more add "sample file parsed twice" problem minor mangling problem mention 32-bit amd64 bug objective C fix fix op_bfd bug spotted by Luca Rossato add doc tidbit to TODO 0.8.1 candidate... more doxygen fixes tree is open, it's time for 0.8.2cvs PPC64 support from the IBM guys Way to forget how CVS works ! disable x86-64 support on 2.4 fix compile warnings tiny fix Altix patch from Greg Banks default count for IA64 changed ppc64 event patch from Maynard Fix typo from Kristis fix AMD doc link update TODO MIPS support, some check_style.py fixes MIPS URLs from Ralf PPC e500 support from Andy Fleming Fix xscale1 unit mask further validation of kernel range ppc64 docs from Maynard, bump to 0.9cvs style fixes, whitespace only couple of small fixes update TODO remove x86-64 support on 2.4 kernels as it doesn't work Fix typo fix typo in docs trivial style fixes a couple more trivial style fixes fix the fix purious std:: header file cleanup more internals docs rename op_import and op_help ppc64 dotted symbols fix major re-work of call-graph output further synth sym patch merge opstack into opreport mplement callgraph --global-percent implement symbol filtering for opreport -c output improvement for opreport -c some re-jigging of debug stuff performance improovements in op_bfd opreport -c performance improvement some cleanups fix tests build trivial whitespace, spelling fix opannotate when it matches several binaries add -% and -D synonyms initial stab at diffing profiles update TODO trivial fix and again... mplement thresholding for diff, fix duplicate syms docs for profile diffing sanity check for matching profile classes when diffing compile fix for previous fabs() change bfd bug workaround from Maynard quoting fix from Nathan Tallent small cleanup of diff_container code clean ups remove TRACE_END, add a hack for 2.4 --no-vmlinux two fixes for diff profiles rejigging of BFD stuff. add --with-binutils, --with-gcc handle forthcoming NO_COOKIE (currently by dropping it on the floor) remove dead prototype fix 100% cpu on --no-vmlinux and --callgraph combination fix libabi for cg files add opcontrol --status merge anon mapping support small change in docs several important fixes in section processing, needs further testing fix for the 2.4 --no-vmlinux workaround fix manual typos image filtering fixes test commit fix a blooper in the diff output field, bump to 0.9 bump version to 0.9.1cvs fix gcc 2.95.4 and older glibc compilation hopefully fix ARM kernel problems fix ARM kernel MIPS 24K support from Ralf Baechle gcc 3.4 build fix for 2.4 module bump to 0.9.1, tree is closed fix make check 0.9.2cvs, tree is open update TODO document archive: in oprofile(1) e500v2 support, mips fix Maynard's patch use __MIPSEB__ not _MIPSEB cleanup from Sean Lee. man page tweak Maynard's patch for ppc64 CYCLES Xen support from HP. Fix typo (from Mike Carlson) mention 2.4 power management problem anon double free fix LRU improvement remove debug oops again... save kernel range in daemonrc (from Andreas Krebbel) intel core due support. New MIPS and PowerPC performance counter support from Ralf and Luca Barbato. core 2 support from Benjamin CR LaHaise make event names match BKDG for amd64. ppc updates from Mark Greer core 2 fixes from Dean Gaudet. static initialization fix. Remove bashism. Strip out boring ARM symbols. Both from Richard Purdie. Fix bug 1597054 (event names with a '/') session dir patch. XML support. style fixes add missing file. Complain about --xml and diff profiles; add some more doc on the latter. Further static initializer fix for ARM. report -X fix from Dave Nomura Anon region naming patch from Amitabha Roy Small fix from Rob Bradford IRC channel is on OFTC these days MIPS events fix from Manoj Ekbote PA6T support; ARM EABI fix Patches from Richard Purdie. Core 2 patch from Dean Gaudet XML callgraph patch. ARM events patches Various patches from Richard Purdie PPC64 demangling fix. amd64 events update from Dean Gaudet Patch from Dean Gaudet doc fix warnings fix from Melchior Franz Cell and IA64 Xen fixes fix SPU change style fixes from Daniel e300, avr32 support. oparchive fixes fix typo Split changelog, fix cvsignore GCC 4.3 fixes AMD update from Jason Yeh ARM big-endian fix Updates for Maynard. Joshua Emele (1): Fix incompatibility between opcontrol & busybox Leonid Moiseichuk (1): Add minimal (armv7-common) support for ARMv7 Krait Marcin Juszkiewicz (1): Add rmb() definition for AArch64 architecture Maynard Johnson (257): Fix some ppc64 event files that had non-UTF8 chars. Fix spelling errors in POWER7 events file. Signed-off-by: Maynard Johnson <may...@us...> Signed-off-by: Maynard Johnson <may...@us...> One more change for migration to git to ensure compile warnings result in errors. Fix symbol size problem that causes "start > end" erorr Fix --with-java configure option to follow autoconf standards Fix up wording about new named unit masks concept. Fix opcontrol --status to show accurate information for running daemon Change version number to 0.9.8git Fix opreport -X so total samples for the binary does not include any module counts. Doc changes reflecting removal of old oprofile kernel module code Fix oprofile build warnings when using gcc 4.6.1 or newer Update POWER7 events and groups. Fix regression caused by Oct 24 commit which broke opannotate --assembly Initial code drop for perf-events branch Add new files for perf-events branch Fixes/workarounds for early/buggy perf_events kernels Symptom: If runtime binaries have been stripped of symbol information, users Regarding the previous commit, I neglected to add a comment in the Fix debuginfo processing for ppc64 Fix debuginfo processing to handle no symbol info in debuginfo file Discard user context kernel samples for which we have no app_cookie Fix 'profile_t::samples_range(): start > end' error Update perf-events branch to match patchset posted to oprofile-list on Feb 23 Fix mmap'ing of perf data file when doing operf_read::convertPerfData. Merge branch 'master' into perf-events Fix up error handling and remove restriction that perf_event_paranoid be set to -1. Remove unused function. Fix opannotate --assembly to work with prelinked runtime lib and non-prelink debuginfo Add support for --pid option and fix timing bug with passing COMMAND option. Add support for kernel profiling; incidental namespace fixups Refactored some code into a library for use by both operf and (in future) oprofile daemon Add support for system-wide profiling to libperf_events and operf Fix problem when passed app name is of the form '<subdir>/<app_name>' Minor bug fixes and performance improvement for system-wide mode. Fix timing-related problems with app name Add support for recording kernel module samples. Fix seg fault when kernel sample received before kernel MMAP. Revert "Fix seg fault when kernel sample received before kernel MMAP." Fix regression causing seg fault when running with no vmlinux file. Add support for callgraph profiling Re-process samples dropped due to no mapping or no process info Miscellaneous minor fixes plus fix broken --separate-cpu option Fix to include unistd.h instead of sys/unistd.h for syscall Add support for JIT profiling to operf Fix regression to operf system-wide profiling caused by previous patch Make new --separate-thread option for operf Fix patch committed on Feb 29, 2012 so that samples from PID 0 are not discarded. Add same fix to master branch as already made to perf-events to not discard samples from process 0 Flesh out operf JIT support so old jitdump files are removed Make various minor cleanup fixes to operf Remove reset option from operf and replace with append option. Change operf to create oprofile sample files during profiling instead of afterwards Fix up anonymous mapping support in operf to include vdso, stack, and heap Minor fixes, changes, and code reorganizations to operf to improve error handling Various operf cleanups, mostly addressing issues reported by Valgrind Fix kernel profiling breakage caused by earlier commit Make operf print warning if non-root user is profiling with kptr_restrict set Add support to operf for recording hypervisor samples Flesh out operf support for kernel versus user domain sample recording. Fix for operf userland/kernelspace profiling and improvements to mmap of kernel data Switch operf from popt to getopt Build operf man page only when the operf program is also built Add reporting of profiling statistics to operf. Add comment in operf man page about using ophelp to list available events Add #include "config.h" before bfd.h Add #include "config.h" before bfd.h for perf-events branch Minor fixups for verbose output for operf Fix opjitconv to not end abnormally when no jit dump directory is found Set exclude_idle to 0 in operf call to perf_event_open Handle EINTR of operf read of sample data pipe Make number of mmap pages for perf_events data be based on pagesize Remove unused "--kernel-buffersize-multiplier" option from operf Fix operf handling of invalid options Make operf use OP_BINDIR when invoking ophelp Make opreport use <cur_dir>/oprofile data for default session-dir Document non-support for event-based profiling in guest environments Add --with-kernel configure option to specify location of kernel headers Fix up help text for --with-kernel option Merge remote branch 'origin/perf-events' Fixes to earlier commit that changed operf to use getopt vs popt Fix for bug 3309794: Change type for sample header mtime field to u64 Fix error when using 'operf --callgraph --events=blah' Fix configure to not alter user variables and remove non-working --with-gcc option Fix up oprofile user manual to add info about new operf program Add some helpful messages for various operf usage scenarios Fix how "Nr. non-backtrace samples" from oprofiled.log is calculated Change "Nr. event lost ..." field in the oprofiled.log to "Nr. samples lost ..." Fix oparchive to use absolute path for copying abi and oprofiled.log Add new operf.log file to what oparchive will archive Bug #3386923: Document that 'opannoate -a' needs symbol info Make git builds use -Werror and fix resulting warning message in operf.cpp Fix 'make check' warning treated as error when using newer gcc Allow libpfm4 to be used by ppc64 for obtaining event code Fix previous commit that broke non-ppc64 arch builds Fix build warning on Fedora 17 with gcc 4.7.0 Fix warning messages when building on Ubuntu 11.10 Fix errors found by coverity in new and changed code in perf_events port Fix operf to read unit mask in hex and document unit mask spec requirementes Remove extraneous debugging cerr from previous commit Change pp tools to abort if <cur_dir>/oprofile_data exists, but no samples found Fix operf profiling of forked processes Add a new option to operf to do conversion to oprofile format after profiling is done Cleanup verbose variable usage in operf for easier debugging Fix unit mask handling (including 'extra' bits) in operf Fix opimport to not try to import .jo ELF files Make opcontrol --status show session dir Rename ... [truncated message content] |