[illumos-gate] 15887 truss can fail to update data_model -- Keith M Wesolowski <wesolows⊙oc>

Anyone here know the email of any FreeBSD BHYVE folks (e.g. Michael Dexter)?

Please DM me if you know.

All set, thanks.

rmustacc: any drawbacks to keeping around a topo handle for the lifetime of a process (and just calling topo_snap_hold() / get_data / topo_snap_rele() as needed to get topo data?

i haven't found any guidance (since it's all undocumented), so not sure if I should do that, or also open/close anytime i need to refresh some topo data

the obvious thing to check would be what data associated with the handle stays alive

but I was pretty sure fmd kept it alive?

seems to keep it alive and refcount it

looking at topo_open(), for the most part, it seems ok to do that, but it does a topo_builtin_create() which does topo_mod_enumerate()

wait, no I got confused about something

it looks to re-open it in fmd_topo_update()

which is where I'm wondering if that might present issues

so it looks like fmd does a new topo_open() each fmd_topo_update() and then refcounts it as consumers (there) will outlive that function.

which is why I got confused

it looks hooked in to only update the topo on DR events though.

and I would assume, absent complaints from fishworkers, that fmd is the canonical long-lived consumer

so doing what it does would make sense to me