link: support bpf perf event link, add bpf_cookies

[mmat11]

This PR adds bpf_cookie and bpf_perf_link. bpf_perf_link is used automatically, if supported, in place of perfEvent. bpf_cookie can be specified only if the link is implemented via bpf_perf_link; if this is not the case, an error will be returned

[lmb]

I did another pass, thanks for clarifying my questions re bpf.h pointers. Regarding Link and how to implement that: I think we should split the code that does perf event set up from the code that does perf event attachment. Rough outline:

• Add type perfEventIoctl struct and type perfEventLink struct. The former has stub functions to implement Link, the latter just wraps a RawLink similar to rawTracepoint (maybe we can just use RawLink directly?)
• perfEvent.attach becomes a free function attachPerfEvent or similar which returns a Link, either perfEventIoctl or perfEventLink.

[mmat11]

@lmb I did some changes, not sure if this is what you asked (I needed to reimplement certain RawLink methods). Also, do you know where I could check which operations for which link type are supported? I am asking because I kept getting EINVAL in all testLink()/update

[lmb]

Please take another look at the changes I outlined in #565 (review). My idea for the final state is as follows:

• perfEvent: encapsulates handling tracefs / pmu, creating the perf event really.
• perfEventLink: attaches a bpf program to a perfEvent using BPF_LINK_CREATE.
• perfEventIoctl: attaches a bpf program to a perfEvent using ioctl().

perfEventLink and perfEventIoctl will wrap perfEvent so that they can do clean up. perfEventLink will additionally embed RawLink.

perfEvent should not implement Link, only perfEventLink and perfEventIoctl do. This is because we want a Link to always represent the result of attaching a program.

If we follow this schema we don't have to change kprobePmu, etc. and can keep attachPerfEvent more straighforward.

[mmat11]

@lmb I set this as draft as I am still figuring out what to do with pinning

[lmb]

Looks good, I left a few nits. Thanks for working in the feedback.

I'm currently traveling, so I'll probably merge this Friday if you're not in a hurry.

[lmb]

Thanks a lot!