kern: Add phashinit_flags(), which allows malloc(M_NOWAIT)

[sepherosa_gmail.com]

Summary

It will be used for the upcoming LRO hash table initialization. And probably will be useful in other cases, when M_WAITOK can't be used.

[jhb]

sys/sys/systm.h

193

I think you can drop the comment for the variable in the prototype as that isn't typical style in FreeBSD. There is a manpage for hashinit/phashinit that should be updated for this new function as well. That is where the meaning of the flags argument should be documented instead.

[kib]

sys/kern/subr_hash.c

111

To my taste, the new name is too long. We often call such enhanced functions e.g. phashinit1(). I understand that you just copied what was done to hashinit.

130

You wrote the assert expression in very compact way. Why spend 4 lines to translate flags ?

m_flags = (flags & HASH_NOWAIT) != 0 ? M_NOWAIT : M_WAITOK;

[sepherosa_gmail.com]

sys/kern/subr_hash.c

111

Just want to make it consistent w/ hashinit :)

130

KASSERT was copied too :P

And, well, when copying the malloc, I just noticed that part duplicates too much code (two malloc(...) lines), so we saw what I posted here.

Yeah, compat form is fine w/ me.

[sepherosa_gmail.com]

Address jhb and kib's concerns

[jhb]

Maybe as a followup fix hashinit_flags() to not duplicate the malloc and instead use the logic you used in phashinit_flags(). That's optional though (and I wouldn't include it as part of this change).

[sepherosa_gmail.com]

In D6138#130946, @jhb wrote:

Maybe as a followup fix hashinit_flags() to not duplicate the malloc and instead use the logic you used in phashinit_flags(). That's optional though (and I wouldn't include it as part of this change).

Yeah, sure.