On Fri, 27 May 2022 12:04:14 +0200 Ingo Molnar <mingo@xxxxxxxxxx> wrote: > For those which implement objtool, it certainly should: as we parse through > each object file during the build, generating kallsyms data structures is > relatively straightforward. > > Objtool availability is a big gating condition though. :-/ > > [ ... and still Acked-by on -v4 too. ] I just sent out a v5 and removed your Acked-by because the changes to v5 are non-trivial like the previous changes in the other versions were. The big difference was that I needed place holders for the invalid functions in the available_filter_functions file, as I forgot that libtracefs uses the line number of these functions as a way to enable them in the set_ftrace_filter and set_ftrace_notrace files. Removing them made the indexing not in sync, and broke trace-cmd. I also added a work queue at boot up to run through all the records and mark any of the ones that fail the kallsyms check as DISABLED. If you want, feel free to review and ack that change too. https://lore.kernel.org/all/20220527163205.421c7828@xxxxxxxxxxxxxxxxxx/ I need to add a selftest to test the indexing code as well. The only reason I found it was that I was writing my presentation for Embedded Recipes and was using it as an example. And when the filtering wasn't working, I had to figure out why. -- Steve
