uma: delay bucket_init() until we might actually enable buckets

[rlibby]

Summary

This helps with a bootstrapping probleming in upcoming work.

We don't first enable buckets until uma_startup2(), so we can delay
bucket creation until then. The other two paths to bucket_enable() are
both later, one in the pageout daemon (SI_SUB_KTHREAD_PAGE vs SI_SUB_VM)
and one in uma_timeout() (first activated in uma_startup3()). Note that
although some bucket functions are accessible before uma_startup2()
(e.g. bucket_select() in zone_ctor()), none of them inspect ubz_zone.

Test Plan

boot

[markj]

I think this is probably fine, I can't see any reason bucket zones need to be enabled so early.

sys/vm/uma_core.c

2450 ↗

(On Diff #65502)

Is it useful to have separate bucket_init() and _enable() routines at this point?

[rlibby]

sys/vm/uma_core.c

2450 ↗

(On Diff #65502)

I think so? You mean, "are there other callers", or "do there need to be other callers"? It looks like bucket_enable() may toggle bucketdisable on a running system according to vm conditions, by either uma_timeout() or the pagedaemon via uma_reclaim().

[markj]

sys/vm/uma_core.c

2450 ↗

(On Diff #65502)

Sorry for the noise, I didn't think that question through. I was really just wondering if bucket_init() should call bucket_enable() directly; I can't see why uma_reclaim_lock needs to be initialized first. There are some paths where we lock uma_reclaim_lock only if booted >= BOOT_BUCKETS, but that has nothing to do with buckets per se. It is a minor detail though.

[rlibby]

sys/vm/uma_core.c

2450 ↗

(On Diff #65502)

I see. Yeah, it's not about the uma_reclaim_lock, it's about booted = BOOT_BUCKETS. I don't have a strong feeling on the organization there. To move the bucket_enable into bucket_init, I'd either need to revert the new KASSERT or move booted = BOOT_BUCKETS before bucket_init, which may be a little surprising (but I think we are still single-threaded here so I think this is just aesthetics).

Would it maybe read better arranged like so:

sx_init(&uma_reclaim_lock, "umareclaim");
bucket_init();
booted = BOOT_BUCKETS;
bucket_enable();

[markj]

sys/vm/uma_core.c

2450 ↗

(On Diff #65502)

I do find that clearer.

[rlibby]

markj feedback

[markj]

Thanks.