libsys, libc: provide rfork_thread() and pdrfork_thread() on all arches

[kib]

Summary

libsys, libc: provide rfork_thread() and pdrfork_thread() on all arches

posix_spawn: use rfork_thread on all arches

[kib]

After this lands, RFSPAWN tests can be enabled on x86 by using *fork_thread() instead of raw *rfork().

[markj]

lib/libc/gen/posix_spawn.c

346 ↗

(On Diff #170487)

348 ↗

(On Diff #170487)

349 ↗

(On Diff #170487)

I wonder why the stack cannot be allocated in the arch-specific (pd)rfork_thread().

[kib]

lib/libc/gen/posix_spawn.c

349 ↗

(On Diff #170487)

I do not follow, sorry. rfork_thread() is the public API, it takes explicit pointer to the stack. I implemented it for arches where it was missed. So the implementation is arch-specific but the interface is not.

Are you proposing to add yet another function, now with the arch-specific interface? I would prefer not.

[kib]

Grammar in comment.

[markj]

lib/libc/gen/posix_spawn.c

349 ↗

(On Diff #170487)

Sorry, I wasn't thinking clearly. Really, I am wondering why not keep the first ifdef in this function, and just pass a NULL pointer for the stack on !x86. It seems a bit silly to do the allocation and argv traversal there.

[kib]

lib/libc/gen/posix_spawn.c

349 ↗

(On Diff #170487)

Hm, so you mean that the _gen implementation for !x86 is flawed, because it does not switch to the provided stack. I agree.

But still, having it consistently provided on all arches is a right step IMO.

[kib]

Only allocate the stack on x86 for now

[markj]

lib/libc/gen/posix_spawn.c

349 ↗

(On Diff #170487)

Perhaps the generic implementation should return an error if a non-NULL stack is provided then?

[kib]

Require NULL stack on _gen

[kib]

Add __used

[markj]

lib/libc/gen/posix_spawn.c

289 ↗

(On Diff #170612)

Just declare cnt in the if (use_env_path) { ... block below?

lib/libsys/rfork_thread_gen.c

28

[kib]

Hopefully the final touches