OS-8007 want PORT_SOURCE_DEVICE for device-specific port events

[joyent-automation]

OS-8007 want PORT_SOURCE_DEVICE for device-specific port events

This PR was migrated-from-gerrit, https://cr.joyent.us/#/c/6948/.
The raw archive of this CR is here.
See MANTA-4594 for info on Joyent Eng's migration from Gerrit.

CR discussion

@johnlevon commented at 2019-10-09T17:35:48

Patch Set 1:

(12 comments)

Patch Set 1 code comments

usr/src/man/man3c/port_associate.3c#159 @johnlevon

nit, s/in in/in /

usr/src/man/man3c/port_create.3c#109 @johnlevon

I'm unclear on the fork semantics of device ports, given that we cache the pid. For example, a child can't dissociate a port.

usr/src/uts/common/fs/portfs/port.c#85 @johnlevon

this is missing PORT_SOURCE_DEV

usr/src/uts/common/fs/portfs/port.c#194 @johnlevon

this is missing PORT_SOURCE_DEV

usr/src/uts/common/fs/portfs/port.c#239 @johnlevon

this is missing PORT_SOURCE_DEV

usr/src/uts/common/fs/portfs/port_dev.c#19 @johnlevon

We could do with a rationale here: I'm guessing something just pointing out that poll+ioctl means a lot of hand-written infrastructure better served by a port source

usr/src/uts/common/fs/portfs/port_dev.c#99 @johnlevon

nit, missing space after 'the'

usr/src/uts/common/fs/portfs/port_dev.c#157 @johnlevon

the naming here is pretty confusing:

port_dev_hash_get() - get the hash table
port_dev_hash_destroy() - destroy the hash table
port_dev_hash_delete() - internal helper to remove an item from the hash
port_dev_hash_find*() - find an item in the hash
port_dev_hash_insert() - insert an item into the hash

Could we maybe rename the first two? port_dev_hashtable_ maybe? Or remove hash from the others?

usr/src/uts/common/fs/portfs/port_dev.c#210 @johnlevon

it's kind of nasty that we have to manage our own lists here.
Why aren't we using mod_hash_create_extended() with a key type of object+pid ?

usr/src/uts/common/fs/portfs/port_dev.c#380 @johnlevon

I'm not following the "sizeof (port_dev_t *)" here, is that from a previous version?

usr/src/uts/common/fs/portfs/port_dev.c#644 @johnlevon

We don't hold the device in any way while we have active
associations. What is stopping the device from detaching here while such ports are active? Can't ->p_ops become stale then?

Basically, I don't understand the relationship between a particular driver's minor device lifetime, and the lifetime of a particular port associated with it, and the user objects associated on top.

usr/src/uts/common/os/port_subr.c#522 @johnlevon

This was changed, but :536 and co wasn't. Can we do one or the either, and get rid of :517-:518