[lkml]   [2005]   [Feb]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [RFC][PATCH] add driver matching priorities
On Thu, 2005-02-10 at 00:41 -0800, Greg KH wrote:
> On Fri, Jan 28, 2005 at 05:30:04PM -0500, Adam Belay wrote:
> > Hi,
> >
> > This patch adds initial support for driver matching priorities to the
> > driver model. It is needed for my work on converting the pci bridge
> > driver to use "struct device_driver". It may also be helpful for driver
> > with more complex (or long id lists as I've seen in many cases) matching
> > criteria.
> >
> > "match" has been added to "struct device_driver". There are now two
> > steps in the matching process. The first step is a bus specific filter
> > that determines possible driver candidates. The second step is a driver
> > specific match function that verifies if the driver will work with the
> > hardware, and returns a priority code (how well it is able to handle the
> > device). The bus layer could override the driver's match function if
> > necessary (similar to how it passes *probe through it's layer and then
> > on to the actual driver).
> >
> > The current priorities are as follows:
> >
> > enum {
> > };
> >
> > let me know if any of this would need to be changed. For example, the
> > "struct bus_type" match function could return a priority code.
> >
> > Of course this patch is not going to be effective alone. We also need
> > to change the init order. If a driver is registered early but isn't the
> > best available, it will be bound to the device prematurely. This would
> > be a problem for carbus (yenta) bridges.
> >
> > I think we may have to load all in kernel drivers first, and then begin
> > matching them to hardware. Do you agree? If so, I'd be happy to make a
> > patch for that too.
> I think the issue that Al raises about drivers grabbing devices, and
> then trying to unbind them might be a real problem.

I agree. Do you think registering every in-kernel driver before probing
hardware would solve this problem?

> Also, why can't this just be done in the bus specific code, in the match
> function? I don't see how putting this into the driver core helps out
> any.

The match priority is a chararistic of the driver and how it's
implemented rather than the bus's matching mechanism. The type of match
doesn't necessarily reflect the driver's ability to control the hardware
(ex. a driver could match on a specific PCI id but only provide generic
support for the device).

Also, I think this is a feature that would be useful for all of the
buses. Therefore, it would seem implementing it in the driver core
might result in the least code duplication.

The second "*match" function in "struct device_driver" gives the driver
a chance to evaluate it's ability of controlling the device and solves a
few problems with the current implementation. (ex. it's not possible to
detect ISA Modems with only a list of PnP IDs, and some PCI devices
support a pool of IDs that is too large to put in an ID table).


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 14:10    [W:0.121 / U:4.236 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site