Excellent wrap-up for this year's #Linux #perf_tools GSoC work: lore.kernel.org/all/CAH0uvohVu… Thanks a lot to @howardchu_ for the excellent work, having a BTF enabled 'perf trace' was something we really wanted for a long time, everything should land on Linux v6.12!

#Linux #perf_tools office hours tomorrow: lore.kernel.org/linux-perf-use…

Linux #perf_tools office hours tomorrow: lore.kernel.org/linux-perf-use…

Dear @Arm what is broken in #Linux #perf_tools: 1) BIG.little: lore.kernel.org/lkml/202405100… 2) gather perf events with offline CPUs: lore.kernel.org/lkml/202406030… 3) Apple M? events, was fixed but the fix was reverted due a broken name in a PMU on Ampere systems: git.kernel.org/pub/scm/linux/…

New perf tool python build, BPF filters on perf top, latest Intel events and TMA metrics 4.8. Thanks to @namhyung_foss for sending the first #perf_tools v6.11 PR: lore.kernel.org/lkml/202407170…

@acmel is the 8th highest tester, so a testament to the amount of work that goes into providing #perf_tools.

Going to plumbers, pet peeves/ideas for the #Linux #perf_tools? Consider this: lore.kernel.org/linux-perf-use…

#perf_tools office hours tomorrow: lore.kernel.org/linux-perf-use…

GSoC results are trickling in: @howardchu_ is doing a great job at using BTF information in 'perf trace', see how enumerations are being pretty printed now: sberm.cn/blog/perf-trac… Next goal: generic struct pointer syscall args! #Linux #perf_tools #BPF #BTF

#Linux #perf_tools office hours tomorrow: lore.kernel.org/linux-perf-use…

Cleaning up #perf_tools Python support. An issue had been in building flex/bison things in setuptools. A brain wave over switching the build into the Makefile/libraries actually worked while removing 110 loc: lore.kernel.org/lkml/202406121… I love it when a plan comes together.

@abhi9u Are Ubuntu still packaging the Debian wrapper script that ties #perf_tools to the kernel version? Something Debian hasn't done for years now and should be discouraged.

There appear to be game changing implications here, but also pitfalls. How best to implement a spin lock? I think we need a benchmark suite inside #perf_tools - note ARM already started with "lockhammer" github.com/ARM-software/s…

#perf_tools fixes and improvements for #Linux v6.10 just sent to Linus: lore.kernel.org/lkml/202405211…

I have to update a slide I wrote 9 days ago and presented this week: vger.kernel.org/~acme/prez/lsf… Now we have a 3rd (4th?) way of disassembling: lore.kernel.org/lkml/202405181… To recap: parse objdump -dS (llvm-objdump), capstone and llvm libs. #linux #perf_tools

Updated #perf_tools flame graph wiki to refer to inbuilt commands. I'd forgotten how good it and the gecko output option looked: perf.wiki.kernel.org/index.php/Tuto…

Slides for my presentation at #LSFMMBPF at 12:00 today: vger.kernel.org/~acme/prez/lsf… #Linux #bpf #pahole #perf_tools

Interesting article about using features @ABudankov implemented long ago to control perf from an app to measure app code, i.e. send start/stop: pramodkumbhar.com/2024/04/linux-… It would be cool to turn that into a perf man page, and have that app C++ code merged... #Linux #perf_tools

The most interest we've ever seen for GSoC on #Linux #perf_tools from contributors! The application period has now started: developers.google.com/open-source/gs… The energy from everyone in doing this is hugely appreciated, let me/us know how we can help.

Ian Rogers @iiirogers

2 years ago

The most interest we've ever seen for GSoC on #Linux #perf_tools from contributors! The application period has now started: developers.google.com/open-source/gs… The energy from everyone in doing this is hugely appreciated, let me/us know how we can help.

2 19 41 13K 8

@abhi9u @brendangregg @forked_franz Yes. Agner Fog's blog described it: agner.org/forum/viewtopi… On Intel look for perf events like MACHINE_CLEARS.MRN_NUKE. When memory renaming misbehaves it looks like bad speculation, and AMD have a bad_speculation metric in #Linux #perf_tools.