Skip to contentSkip to conversation

Progress on BTF linking and deduplication using the toolchain

LSF/MM/BPF TOPIC Nick Alcock 2 participants 2 messages Feb 19 – Feb 20 BPF

lore
Share & tools
lore
#0 lore raw

This talk will provide an update on the work we have done in binutils and pahole to support BTF.

We have been working on pahole modifications, intended for use with GCC-generated BTF information, providing a more direct handling of BTF without the need for DWARF to BTF conversion.

This is done using libctf from binutils, which as of our latest version (not yet on binutils master), can read and write BTF in a way that is fully compatible with the existing mechanism. We also have added support for merging and deduplicating BTF in the linker. In addition we can now dump BTF with objdump.

We will show comparisons between the old and new approach within pahole, which will exemplify differences in speed, as well as the lack of differences in output.

This talk will give a brief overview of how this works as well as a short demonstration.

(Like several others, I'm hoping I can do this remotely: transport from the UK across Bank Holiday Monday is something between awful and abominable.

I've added a couple of other possible attendees -- Cc:ed -- to the Google form side of this: if this was overstepping some sort of bound, since I'm not a subsystem maintainer of any sort, my apologies.)

#1 lore raw
Quoted 21 lines · Nick Alcock

This talk will provide an update on the work we have done in binutils and pahole to support BTF.

We have been working on pahole modifications, intended for use with GCC-generated BTF information, providing a more direct handling of BTF without the need for DWARF to BTF conversion.

This is done using libctf from binutils, which as of our latest version (not yet on binutils master), can read and write BTF in a way that is fully compatible with the existing mechanism. We also have added support for merging and deduplicating BTF in the linker. In addition we can now dump BTF with objdump.

We will show comparisons between the old and new approach within pahole, which will exemplify differences in speed, as well as the lack of differences in output.

This talk will give a brief overview of how this works as well as a short demonstration.

Quoted 7 lines

(Like several others, I'm hoping I can do this remotely: transport from the UK across Bank Holiday Monday is something between awful and abominable.

I've added a couple of other possible attendees -- Cc:ed -- to the Google form side of this: if this was overstepping some sort of bound, since I'm not a subsystem maintainer of any sort, my apologies.)

To clarify on the attendance requests: (sorry cannot edit the form) Nick Alcock -> preferred remote,
Elena Zannoni -> in person,
Bruce McCulloch -> either

thanks
elena