[discussions] rr everywhere

[lwintermelon]

Can rr work if

1. intel hybrid cores on KVM or hyper-v: It seems that PMU won't work if you are in KVM on intel hybrid cores, can anyone confirm that? @rocallahan I learned from your blog that you've built a new computer with an Intel hybrid-core processor. [added] I can confirm that PMU doesn't work on Intel hybrid cores in Hyper-V, nor does it work in VMware Workstation even for 11th-gen Intel CPUs that aren't hybrid cores.
2. AMD Zen on hypervisor: the workaround may work on bare mental, but not for hypervisor like hyper-v.
3. You can't get access to PMU: Don't worry, have you noticed rr.soft? I've tested it, and it passed 99% of the tests.

Summary: 1. and 2. , If you are on hypervisor with intel hybrid cores or AMD zen, or you are in VMware Workstation with newer intel cpus, you have trouble running rr.

[GitMensch]

For 2: ... and as part of Hyper-V, also not on WSL2 (at least I've found no solution, including custom kernel builds)

3 looks interesting (haven't tried, do I understand it correctly that you need a GCC/Clang plugin to make it usable, either in general or for acceptable performance)?

Also for 3, from its README about merging this upstream:

The ability to run without CPU HW performance counters is a large feature patch and would require additional time and effort to get merged upstream. It also depends on whether rr maintainers want dynamic/static instrumentation in their codebase or wish to continue keeping rr purely HW performance counters based.

... so: Do rr maintainers accept dynamic/static instrumentation in their codebase?

[sidkshatriya]

3 looks interesting (haven't tried, do I understand it correctly that you need a GCC/Clang plugin to make it usable, either in general or for acceptable performance)?

You don't require a GCC/Clang plugin to make it usable.

The plugin makes things a bit more robust as dynamic instrumentation is then not required. Try record/replaying without the use of a plugin first -- in most cases things should work fine. Reach for the plugin for enhanced robustness and if you are encountering issues with record/replay with dynamic instrumentation.

... so: Do rr maintainers accept dynamic/static instrumentation in their codebase?

I would very much love it if @rocallahan / @khuey accepted it in this repo. I don't think the code changes are very intrusive.

[GitMensch]

Thanks for the information on the plugin.

I would very much love it if [the maintainers] accepted it in this repo. I don't think the code changes are very intrusive.

Then please create a PR to show and discuss the changes.

[sidkshatriya]

Then please create a PR to show and discuss the changes.

The rr maintainers are aware of rr.soft and the broad strategy used to implement it.

It would take some time and effort to create a PR.

I would love for the maintainers to play around with rr.soft a bit (if they haven't already) and let me know they would like something like this in their code base. If so I can subsequently work on this.