[SLOF] [PATCH] pci: Reserve free space at the end of bridge windows instead of at the beginning

[David Gibson]

On Fri, May 19, 2017 at 09:27:44AM +1000, Alexey Kardashevskiy wrote:
> On 16/05/17 17:05, Thomas Huth wrote:
> > On 16.05.2017 08:47, Nikunj A Dadhania wrote:
> >> Thomas Huth <thuth at redhat.com> writes:
> >>
> >>> This reverts commit e53c2541784fba7951c8aa6ccdbe4412fb03fca6
> >>> ("pci-scan: reserve memory for pci-bridge without devices").
> >>
> >> The above patch was added for an issue that was reported when an empty
> >> pci-bridge was added along with the network adapter, rhel6.5/sles11sp3
> >> was not able to get the dhcp address for the network.
> >>
> >> The actual problem was while enumerating an empty pci-bridge, stale
> >> range value programmed before the bridge probing stayed. Which then
> >> caused overlapping pci address allocation.
> >>
> >> This does not seem to be an issue any more.
> > 
> > Ok, good to know what the original problem was. Looking at the git
> > history, it seems like the main fix for the problem was the commit
> > 7cc0b6c5ce87549ba7bc396bfca1018c076a7351 already, and the "reserve
> > memory" patch was just the additional sugar (since hot-plugging on pci
> > bridges is not possible yet, as far as I know)?
> > 
> >>> It's better to reserve the free space at the end of the memory windows
> >>> instead (in the pci-bridge-set-[mem|mmio|io]-limit functions), and
> >>> with regards to the scarce I/O space, we should also reserve less
> >>> I/O memory on each bridge, so we use a limit of 2k (plus alignment)
> >>> here now.
> >>>
> >>> Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1443433
> >>> Signed-off-by: Thomas Huth <thuth at redhat.com>
> >>
> >> Reviewed-by: Nikunj A Dadhania <nikunj at linux.vnet.ibm.com>
> > 
> > Thanks!
> > 
> > Any chance that you could commit this while Alexey is away?
> 
> What is the hurry (just curious)?

It'd be nice to fix for RHEL7.4 which is very close now.

That said, I just realized there's another wrinkle here - we're going
to need to introduce a compatibility mode for SLOF (presumably keyed
off a device tree variable) so that we don't introduce a guest-visible
change in older machine types.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.ozlabs.org/pipermail/slof/attachments/20170519/5122c767/attachment.sig>