PERFORCE change 38917 for review

[M. Warner Losh]

In message: <XFMail.20031003131759.jhb at FreeBSD.org>
            John Baldwin <jhb at freebsd.org> writes:
: 
: On 03-Oct-2003 M. Warner Losh wrote:
: > In message: <20031002022135.GA26666 at perrin.nxad.com>
: >             Hiten Pandya <hmp at nxad.com> writes:
: >:      Does this also include PCI Hotplugging or is this going to be
: >:      a separate effort?
: > 
: > Yes and no.
: > 
: > We already have removable devices in the tree.  CardBus, PC Card, USB
: > and Firewire.  Right now the drivers take liberties and chances and
: > there are a number of ugly races that rear their ugly heads.  This is
: > an effort to clean that up and make things totally safe in all the
: > cases that we can think of, and to document those areas where we
: > can't.
: > 
: > There's no specific plans on my part to implement the hot plug state
: > transition and deal with it.  People have raised their hands in the
: > past to work on it, but so far no one has made significant progress in
: > public view.
: 
: BTW, I thought of a locking concern in attach routines this morning.
: Basically, as soon as a driver does bus_setup_intr(), it needs to start
: using any locks shared with its interrupt handler and the hardware
: and driver state need to be able to handle having the handler invoked
: when that function is called.  The reason is that the device may be
: attached to an interrupt line shared with another device and thus the
: other device may generate an interrupt and call the new devices handler
: before bus_setup_intr() even returns.  Similarly, locking will be
: needed after calls to disk_create(), if_attach(), etc.

Yes.  That's true.  I'll keep it in mind.

Warner