2001-12-04 00:21:04

by Daniel Stodden

[permalink] [raw]
Subject: pci: dev->driver "runtime" acquisition?

hi.

supposing i've got a "hotplug" style pci_driver,
but who's going to gain access to a device _not_ at initialization time,
i.e. not within pci_register_driver(), but rather somehere in a bottom
half. no pci_device_id* at pci_register_driver( ), since i don't
know the signature yet.

how do i request access here?
could anyone comment on the following please:

1. simple, but intrusive somehow:
dev_probe_lock()
if ( dev->driver == NULL )
dev->driver = my_driver;
dev_probe_unlock()
is the locking correct?

2. rather use pci_announce_device()?
my_driver->id_table =
somthing_i_just_built_on_the_fly_and_i_know_it_matches;
pci_announce_device( my_driver, dev );
/* probe() does the rest */

3. patch drivers/pci/pci.c?
pci_acquire_device( struct pci *dev, struct pci_driver *drv )
{
/* fill in 1. */
}

4. simply rely on the "compat", i.e. just taking over some resource?
i think that's not why it's been called "compat"

any better ideas? something i'm missing?
besides: policy? -- do i _have_ to pci_register_driver() at during
initialization before considering to actually drive something?

any help appreciated.

regards,
dns


--
__________________________________________________________________
mailto: [email protected]


2001-12-04 03:26:31

by Greg KH

[permalink] [raw]
Subject: Re: pci: dev->driver "runtime" acquisition?

On Mon, Dec 03, 2001 at 06:29:57PM +0100, Daniel Stodden wrote:
> hi.
>
> supposing i've got a "hotplug" style pci_driver,

Any hints on what kind of driver this is? :)

> but who's going to gain access to a device _not_ at initialization time,
> i.e. not within pci_register_driver(), but rather somehere in a bottom
> half. no pci_device_id* at pci_register_driver( ), since i don't
> know the signature yet.

Take a look at the pci_read_config_*_nodev() and
pci_write_config_*_nodev() functions in the
drivers/hotplug/pci_hotplug.h file. Do these functions help you out?

thanks,

greg k-h