1118206 |
Gecko Profiler uses nsThread before nsThreadManager is initialized |
NEW |
|
|
|
1210446 |
AutoProfilerTracing should also push label frames |
REOPENED |
[fxp] |
|
|
1234670 |
Gecko crashes when trying to profile talos locally |
NEW |
|
|
|
1307215 |
[meta] Stack sampling must be reliable |
NEW |
|
meta |
|
1321607 |
Expose the list of all threads which are available for profiling through nsIProfiler |
NEW |
|
|
|
1329181 |
[meta] Profiler crashes and hangs |
NEW |
|
meta |
|
1329212 |
[meta] Profiler performance issues / sources of skew due to profiler overhead |
NEW |
|
meta |
|
1329219 |
[meta] Profiler memory usage issues |
NEW |
|
meta |
|
1337558 |
[meta] Solid multi-thread profiling |
NEW |
|
meta |
|
1339982 |
Reduced profiling experience when running with many content processes (crashes, hangs, ...) |
NEW |
|
|
|
1369955 |
Off-the shelf control, enumeration, and documentation of performance sources |
ASSIGNED |
|
|
Jim Blandy :jimb |
1379286 |
Use a kernel driver for profiling on Windows |
NEW |
|
perf |
|
1382943 |
No stack frames during LdrLoadDll |
REOPENED |
inj+ |
|
|
1387528 |
implement non-interval markers for console.timeStamp console.time, and console.timeEnd |
NEW |
[fxp] |
|
|
1398137 |
Crash under memcpy | DoLULBacktrace |
REOPENED |
|
|
|
1417184 |
show separate profiler markers for js parse and compile |
NEW |
|
|
|
1434649 |
Push the worker script URL label frame even when the profiler is inactive |
NEW |
|
|
|
1446575 |
[meta] Remote profiling for GeckoView milestone 1.1 |
NEW |
[geckoview] |
meta |
|
1456606 |
Add support for capturing profiles locally on Android without having a computer connected to the phone |
NEW |
|
|
|
1458088 |
Indicate profiler initialization time properly (especially on Linux where LUL initialization takes a long time) |
NEW |
|
|
|
1463559 |
Hide native unsymbolicated JIT frames from Ion->C++ |
NEW |
|
|
|
1465869 |
Store the "stack type" (JS / label stack / native stack) of a call stack frame in the frameTable |
NEW |
|
|
|
1465926 |
Register every thread with the GeckoProfiler, and make it really hard to add new threads without registering |
NEW |
|
|
|
1466261 |
Marker for off thread parse/compile on content process |
NEW |
|
|
|
1466927 |
The page at https://androidify.com/en/#/create takes a very long time to load with the profiler enabled |
NEW |
|
|
|
1505818 |
Make the page registration lazy in profiler |
NEW |
|
|
Nazım Can Altınova [:canova][:canaltinova on phabricator] |
1507253 |
Android Gecko Profiler captures via MOZ_PROFILER_STARTUP/SHUTDOWN have to be symbolicated manually |
NEW |
|
|
|
1507566 |
Profiler Counters not showing up in profile outputs when forcebly injected into SpiderMonkey |
NEW |
|
|
|
1508838 |
Document the uses for the marker payloads |
NEW |
|
|
|
1510302 |
Make the Gecko profiler work on Android x86 |
NEW |
|
|
|
1522271 |
Capturing screenshots on Android uses too much CPU for swizzling image data |
NEW |
|
|
|
1524574 |
mainthreadio markers missing for closing a file on Windows (NtClose) |
NEW |
[fxp] |
|
|
1526317 |
Some content process profiles sometimes not discarded when stopping the profiler |
NEW |
|
|
|
1527023 |
[Meta] Surface Disk IO information in the profiler |
NEW |
|
meta |
|
1527110 |
Would be nice to have a way to figure out why we are painting |
NEW |
|
|
|
1527340 |
Surface additional payload information for file io |
NEW |
|
|
|
1527453 |
Examine IOInterposer thread safety |
NEW |
|
|
|
1527736 |
ProfilerIOInterposeObserver cannot be re-enabled when restarting the profiler |
NEW |
|
|
|
1527739 |
Write tests to verify that file io operations are working across various platforms |
NEW |
|
|
|
1527741 |
Turn the mainthreadio feature on by default |
NEW |
|
|
|
1528318 |
Style cause stack could use a profiler label that describes the restyled element |
NEW |
|
|
|
1529108 |
Unwinding from libc.so on Android 32 bit fails (bad stacks during syscalls) |
NEW |
|
|
|
1530610 |
System freeze when starting the profiler |
NEW |
|
|
|
1533018 |
Shutdown profiles don't have correct metadata information |
NEW |
|
|
|
1533779 |
NSPRInterposer should interpose open calls |
NEW |
|
|
|
1536109 |
NSPRIOInterposer doesn't interpose nsLocalFile.exists |
NEW |
|
|
|
1541291 |
Enabling the profiler can wreak havoc and causes failures of test_profile_worker*.html on Android |
NEW |
|
|
|
1541360 |
It seems like we miss network markers on the parent process |
NEW |
|
|
|
1541551 |
Profiles of local Windows builds get stuck "Waiting for symbol tables for library xul.pdb" |
NEW |
|
|
|
1542742 |
Missing File IO markers for GetDirectoryEntries |
NEW |
|
|
|
1542919 |
Symbolication of local arm64 builds on Windows does not work |
NEW |
|
|
|
1547284 |
Make JS allocation sampling probability configurable |
NEW |
|
|
|
1547989 |
Missing mainthread file IO markers for IsDirectory() call on Windows from MakeFileInputStream |
NEW |
|
|
|
1548003 |
WipeContainingBlock marker is invalid |
NEW |
|
|
|
1549729 |
Assertion failure: table.empty(), at /Users/florian/buildhg/mozilla/js/src/vm/JSScript.cpp:3353 when stopping the profiler |
NEW |
|
|
|
1549735 |
Profiler shouldn't call JS::GetProfiledFrames off-thread |
NEW |
|
|
|
1555702 |
Missing GPU and Parent samples in profiles when webrender screenshots are enabled |
NEW |
|
|
|
1555706 |
With webrender screenshots enabled, the awesomebar shows in a separate panel |
NEW |
|
|
|
1558116 |
The panel should be wider to accommodate for cross-plaform font differences |
NEW |
|
|
|
1559485 |
Use UniqueChars in RecordAllocationInfo |
NEW |
|
|
|
1560443 |
Profiler produces black screenshots |
NEW |
|
|
|
1560601 |
[network markers] we should have a way to expose the delay timing before we actually create a channel |
NEW |
|
|
|
1560602 |
[network markers] requests with a cache hit don't have timings reported. |
NEW |
|
|
|
1574816 |
Change Windows profiler default (and smallest?) interval to 2ms |
NEW |
|
|
|
1580159 |
[network markers] The "REDIRECT" marker should be in Redirect3Complete instead of Redirect1Begin |
NEW |
|
|
|
1586618 |
Profile serialization should not include entries added after the serialization starts |
NEW |
|
|
|
1589977 |
Review default profiler buffer sizes |
NEW |
|
regression |
|
1590708 |
[meta] Single tab profiling mode |
ASSIGNED |
|
meta |
Nazım Can Altınova [:canova][:canaltinova on phabricator] |
1594291 |
Write test verifying that the responsiveness is correctly captured in the profiler |
NEW |
|
|
|
1595687 |
Running a second profiler browser tests is much slower |
NEW |
|
|
|
1596235 |
Add innerWindowIDs for wasm frames |
NEW |
|
|
|
1609655 |
Add innerWindowID to profiler_add_js_marker API |
NEW |
|
|
|
1609667 |
Investigate and see if we can add innerWindowID for GC markers. |
NEW |
|
|
|
1611555 |
Perma PROCESS-CRASH | damp | application crashed [@ AslHashFree + 0x35028] |
NEW |
|
crash |
|
1616642 |
The profiler label in ThreadEventQueue::GetEvent should not use the category IDLE if it's spinning on a different event target |
NEW |
|
|
|
1616887 |
[meta] Profiling improvements for Firefox on Android |
NEW |
|
meta |
|
1623752 |
Markers with undefined intervals present in the profile |
NEW |
|
|
|
1623942 |
Update LUL and EHABIStackWalk tables when libraries get loaded and unloaded. Stack unwinding doesn't work on libraries that were loaded after profiler initialization (e.g. libfreebl3.so, libsoftokn3.so) |
NEW |
|
|
|
1624165 |
Intermittent runner.py | application crashed [@ libc.so.6 + 0x14d086] |
NEW |
|
crash, intermittent-failure |
|
1627871 |
Consider adding tests to monitor the health of most profile metadata |
NEW |
|
|
|
1635810 |
[meta] LUL initialization takes too long, slows down startup profiling and content process startup |
NEW |
[fxp] |
meta |
|
1635811 |
Consider caching LUL unwind information |
NEW |
|
|
|
1640817 |
Some IPC markers have a negative duration |
NEW |
[fxp] |
|
|
1640956 |
[meta] Markers 2.0 |
NEW |
|
meta |
|
1640970 |
Remove duplicate time information from marker payloads |
NEW |
|
|
|
1640987 |
Simplify stack serialization |
NEW |
[fxp] |
|
|
1640995 |
Audit PROFILER_MARKER_TEXT's, and convert them to structured data |
NEW |
|
|
|
1641119 |
Handle OutOfMemoryError exceptions when allocating Java samples buffer |
NEW |
|
|
|
1646714 |
Add innerWindowsID and stack to the marker struct |
NEW |
|
|
|
1648325 |
Investigate root cause of Talos intermittents when Base Profiler is running |
NEW |
|
|
|
1648770 |
Add markers for points of nsHttpChannel suspension and resume, track suspend time |
NEW |
|
|
|
1651047 |
Fenix crashes during first content url load when profiler is active on Android 10 |
NEW |
|
|
|
1658045 |
High amount of overhead from capturing and freeing cause callstacks for refresh driver ticks |
NEW |
|
|
|
1659880 |
Review functions that could be interrupted by the profiler sampler |
NEW |
|
|
|
1662246 |
Differentiate markers that are generated by cloned documents |
NEW |
|
|
|
1664418 |
Crash in [@ encoding_rs::utf_8::Utf8Decoder::decode_to_utf16_raw] |
NEW |
|
crash |
|
1666652 |
Add more unit tests for new marker API classes |
NEW |
|
|
|
1669273 |
Intermittent SUMMARY: ThreadSanitizer: SEGV /builds/worker/fetches/llvm-project/llvm/projects/compiler-rt/lib/tsan/rtl/tsan_rtl.cpp:848:55 in __tsan::MemoryAccess(__tsan::ThreadState*, unsigned long, unsigned long, int, bool, bool) |
NEW |
|
intermittent-failure |
|
1674930 |
MarkerCategory and related things should be renamed to clarify that they're "pairs" |
NEW |
|
|
|
1679823 |
Record the optimization characteristics of the profiled build |
NEW |
|
|
|
1686565 |
Stackwalking sometimes missing in startup profiles |
NEW |
|
|
|
1687510 |
Deadlock when stackwalking while a DLL is being loaded |
NEW |
|
|
|
1688135 |
On low core machines, the sampler thread seems to propagate bad performance from one thread to all sampled threads |
NEW |
|
|
|
1689204 |
Android sampler may rapidly "catch up" if sampling takes took long, seems undesirable |
NEW |
|
|
|
1690147 |
Final screenshot is not captured because there is no follow-up composite; should have a wait timeout for screenshot capturing |
NEW |
|
|
|
1691095 |
Make category colors an enum |
NEW |
|
|
|
1697953 |
Consider switching profiler buffer locking to atomics |
NEW |
[fxp] |
|
|
1699681 |
Enabling the profiler with the JS feature to profile tests leaks windows until shutdown |
ASSIGNED |
[fxp] |
regression |
Nazım Can Altınova [:canova][:canaltinova on phabricator] |
1710817 |
Add baseprofiler tests to catch sampling and marker issues |
NEW |
|
|
|
1714538 |
Review all rust thread spawns in the codebase and register them with the profiler using profiler rust API |
NEW |
|
|
|
1724255 |
Add a method in ChromeUtils to capture a profiler stack for use in a future marker |
NEW |
|
|
|
1725057 |
Removing too many leaf frames from allocation stacks on Windows? |
NEW |
|
|
|
1728243 |
Navigation::Start marker in content process on start up is missing URL in fenix |
NEW |
|
|
|
1728845 |
Add tests for the profiler Rust API |
NEW |
|
|
|
1730997 |
Consider making "cpu" feature non-optional |
NEW |
[fxp-profiler] |
|
|
1735074 |
The profiler memory track is relative, should be absolute. Need to know the initial memory at the beginning of profiling. (Counter should be initialized with current number of allocated bytes) |
NEW |
[fxp-profiler] [memory] |
|
|
1741717 |
Content process is missing in geckoview_example (fenix?) unless I load another page (allrecipes.com) |
NEW |
|
|
|
1749523 |
It looks like JIT "unsymbolicated" frames are showing up in profiles nowadays |
NEW |
|
|
|
1750375 |
Intermittent toolkit/components/extensions/test/xpcshell/test_ext_geckoProfiler_control.js | application crashed [@ ProfilingStack::~ProfilingStack] |
NEW |
|
crash, intermittent-failure |
|
1751036 |
Crash [@ mozilla::ProfilerChild::ProcessPendingUpdate()] when restarting for an update |
NEW |
|
|
|
1752859 |
[network-markers] Some network markers of the content process don't have an end and don't exist on the parent |
NEW |
|
|
|
1753704 |
[network markers] Canceled requests are not showed as such in the content process |
NEW |
|
|
|
1753912 |
Check that timestamps sent through ProfileBufferChunkManagerUpdate and DestroyReleasedChunksAtOrBefore are related to each other |
NEW |
|
|
|
1756519 |
Are the base profiler and gecko profiler using different units for CPU time? |
NEW |
|
|
|
1757350 |
Reduce overhead of profiler::AllocCallback and FreeCallback |
NEW |
|
|
|
1762526 |
Stop using a dedicated thread for profiler startup and control |
NEW |
|
memory-footprint |
|
1766970 |
Write test for multiple counters |
NEW |
|
|
|
1766971 |
Review json-streaming code that could skip adjacent entries of the same type |
NEW |
|
|
|
1769010 |
Profiles with re-used thread ids are confusing |
NEW |
|
|
|
1773286 |
Add an API to unregister a thread from another thread, by its thread id |
NEW |
|
|
|
1773313 |
In profiler shared library enumeration, obtain the ELF build ID from memory and not by reading the file again |
NEW |
|
|
|
1774326 |
Move marker type to top-level property, potentially using a unique type id |
NEW |
|
|
|
1774328 |
Gecko Profiler JSON marker schemas should also define a list of properties, and marker payloads should be stored as arrays |
NEW |
|
|
|
1774334 |
Don't output more than one sample for the duration of SLEEPing/idle periods |
NEW |
|
|
|
1782125 |
Startup profiling deadlocks on Windows |
NEW |
|
|
|
1792530 |
Long content process profile missing when capturing a profile soon after closing the tab |
NEW |
|
|
|
1796684 |
Lots of jank in the profiler on latest nightly. |
NEW |
|
|
|
1799109 |
Add an AUTO_PROFILER_MARKER_UNTYPED macro |
NEW |
|
|
|
1804062 |
Deadlock when profiling and recording power metrics |
NEW |
|
|
|
1805922 |
Inline functions don't show up in profiles of local builds. |
NEW |
|
|
Emilio Cobos Álvarez (:emilio) |
1806054 |
Memory profiling uses the expensive malloc_usable_size function. |
ASSIGNED |
[sp3] |
|
Paul Bone [:pbone] |
1811493 |
Profiles without the JS feature have null stack samples |
NEW |
|
|
|
1811807 |
Store the marker stacks inside the main profiler buffer rather than their own buffers |
NEW |
[fxp] |
|
|
1811865 |
The native allocations feature crashes on Linux |
NEW |
|
|
|
1812025 |
The java threads are sampled even when the nostacksampling feature is used to reduce overhead |
NEW |
|
|
|
1828574 |
Weird looking profile |
NEW |
|
|
|
1829433 |
GeckoProfilerEntryMarker takes time in js::RunScript even when Firefox profiler isn't running |
ASSIGNED |
[sp3][fxp] |
|
Adam Brouwers-Harries [:aabh] [he/him] ⌚GMT |
1833147 |
Enabling Firefox Profiler on large page crashes Firefox |
NEW |
|
|
|
1839299 |
Use a finer-grained strategy than stack walking suppressions to avoid stack walking deadlock on Windows x64 and aarch64 |
ASSIGNED |
|
|
Yannis Juglaret [:yannis] |
1855661 |
0.15% installer size (OSX) regression on Mon September 25 2023 |
ASSIGNED |
[fxp] |
perf-alert, regression |
Markus Stange [:mstange] |
1870022 |
Change the type of Reflow markers to "Reflow" and add a dedicated schema for them |
NEW |
[fxp] |
regression |
|
1871174 |
Crash in [@ stackoverflow | ExtractJsFrames] |
NEW |
|
crash |
|
1873170 |
Reentrancy deadlock in TimerThreadWrapper / nsTimerImpl on a profiler-only path |
NEW |
|
|
|
1878882 |
Mutex cycle causing deadlocks due to ProfilerBuffers and IPC |
ASSIGNED |
|
|
Nazım Can Altınova [:canova][:canaltinova on phabricator] |
1883903 |
Consider capturing profiles using signals for timed out tests in CI |
NEW |
|
|
Adam Brouwers-Harries [:aabh] [he/him] ⌚GMT |
1885000 |
Find non-failing alternatives if a download directory doesn't exist when stopping the profiler with a signal |
NEW |
[fxp] |
|
Adam Brouwers-Harries [:aabh] [he/him] ⌚GMT |
1886326 |
Crash when profiling AOT-compiled Blazor WebAssembly application |
NEW |
|
|
|
1887890 |
Add better frame labels for certificate stuff |
NEW |
[fxp] |
|
|
1891138 |
Crash in js::gc::HeaderWord::get when doing native allocation profiling of www.itemkey.co.uk/workerTests/workerTest.html |
NEW |
|
|
|
1892602 |
Profiling for startup on android misses a small chunk of startup. |
NEW |
[fxp] |
|
|
1897436 |
Capture the active tab id at the end of the profiling session instead of start |
NEW |
[fxp] |
|
|
1899722 |
Rust threads don't get their correct "stack top" during thread registration |
NEW |
|
|
|
1900589 |
Make the memory replace malloc machinery always available for dev edition |
NEW |
[fxp] |
|
|
1904639 |
Ensure that signal control of the profiler works on Android |
NEW |
|
|
|
1905079 |
Firefox Desktop and Fenix Mobile have separate preset lists |
NEW |
|
|
|
1906151 |
Sending SIGTERM to the profiler while writing a profile to disk results in a truncated profile file. |
NEW |
|
|
|