Dmitry Vyukov

I think we could try to mount debugfs during setup procedure in executor in setup_features. But we need to mount it before any features are checked, not just coverage, because...

While working on moving code from the target to the host, I see lots of utilities that duplicate code and pain to update. 4 more use cases that would benefit...

> For that we need some dynamically configurable layer that would return nil in Next() before we create the fuzzer object and explicitly plug it in. Potentially we could always...

> Repro and image testing are somewhat simpler of these 4 (if any VM crashed at any stage in the process, we don't retry, but just report the result), but...

> Syz-manager makes things a bit more complicated by dynamically resizing this VM pool to reproduce bugs. Potentially we could merge repro into the normal manager operation, repro will just...

> I'm not sure whether merging all these tools into syz-manager is a good idea, but taking this functionality to e.g. a separate package and using the package in all...

> I've pushed the changes that move `pkg/vminfo` to the new execution interface, but it has broken `tools/syz-runtest`. Looks like I'll need to port `pkg/runtest` as well. Now you feel...

Hi @r0xen, It's the first time I hear about KLEAF. So, no, there are no workarounds and I don't understand what exactly is the problem and how hard it is...

If you introduce randomness into tests, please use this thing: https://github.com/google/syzkaller/blob/master/pkg/testutil/testutil.go#L25-L35 It provides both determinism for CI coverage/failures, random seeds in local testing and ability to reproduce with particular seed.

Some test-only hook that will permute instructions looks better. Note that Go tests generally run in parallel, so permuting in each test won't work well. But we could permute in...