Linus Torvalds <torvalds@transmeta.com> writes:
> On Sun, 5 Jan 2003, Ivan Kokshaysky wrote:
> >
> > Hopefully this patch should solve most problems with probing the BARs.
> > The changes are quite minimal as everything still is done in one pass.
>
> Can you do the same with a multi-pass thing?
>
> I really think the single-pass approach is broken, because it means that
> we _cannot_ have a fixup for device that runs _before_ the fixup for the
> bridge that bridges to the device.
>
> As such, the "PCI_FIXUP_EARLY" is not _nearly_ early enough, since it's
> way too late for the actual problem that started this whole thread (ie in
> order to turn off a bridge, we have to make sure that everything behind
> the bridge is turned off _first_).
>
> In other words, we really should be able to do all the bus number setup
> _first_. That isn't dependent ont eh BAR's or anything else. The actual
> _sizing_ of the bus is clearly somethign we cannot do early, but we can
> (and should) enumerate the devices first in phase #1.
>
> Alternatively, we could even have a very limited phase #1 that only
> enumerates _reachable_ devices (ie it doesn't even try to create bus
> numbers, it only enumerates devices and buses that have already been set
> up by the firmware, and ignores bridges that aren't set up yet). A pure
> discovery phase, without any configuration at all.
>
> Hmm?
I have done something similar to this in LinuxBIOS, perhaps a
description of that will spark ideas.
In the enumeration phase I look at the vendor+device id and then hdr
type to assign methods. The methods I have are:
scan_bus,
read_bases,
write_bases.
Then for all children of a bus that have a scan_bus function I
recurse. This fits in very naturally with devices not necessarily
being pci devices.
When it comes time to assign pci resources I call the read_bases
method to get the size and possibly fixed location of the resource. I
play with my data structures to get a non conflicting set of resources
and then I call write_bases on each device to write the resource
assignments to the actual device.
The setup works well enough that I don't have to differentiation later
in the code between pci busses and normal pci devices.
I have not had to push beyond the standard pci devices yet, but I
don't think I have missed anything that would cause me problems.
Linus is this the kind of thing you are thinking of?
Eric
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Jan 07 2003 - 22:00:31 EST