RE: linux-next: build failure after merge of the xen-two tree

Rafael J. Wysocki wrote:

On Saturday, February 16, 2013 01:52:00 AM Stephen Rothwell wrote:

quoted

Hi Rafael,

Hi,

quoted

On Fri, 15 Feb 2013 15:53:34 +0100 "Rafael J. Wysocki" [off-list ref]
wrote: 

quoted

On Saturday, February 16, 2013 12:50:14 AM Stephen Rothwell wrote:

quoted

On Fri, 15 Feb 2013 08:26:24 -0500 Konrad Rzeszutek Wilk
[off-list ref] wrote: 

quoted

Thank you. I keep on forgetting - but would it be OK for me to
take this patch in my tree? Or should I not since this is a new
functionality that Rafael is going to introduce in v3.9?

It is an API change in the pm tree that is not yet in Linus' tree.

quoted

And if so, perhaps I should tack it on in my tree, once Rafael
does a git pull to Linus? Or just point Linus to this git commit?

You should point Linus at this patch if the pm tree is merged
first, or 
Rafael should do the same if the reverse happens.

Alternatively, Konrad can pull the acpi-scan branch containing the
changes in question from my tree into his tree and rebase the new
material on top of that.

Or pull the acpi-scan branch into his tree and use my conflict
resolution in the resulting merge thus requiring no rebasing. 
However, Linus likes to see such interactions, so it can be left up
to when the latter of the two tress is merged by Linus.

Well, I'm afraid this won't be sufficient this time, because of this
commit in my tree (which is not on the acpi-scan branch):

commit 3757b94802fb65d8f696597a74053cf21738da0b
Author: Rafael J. Wysocki [off-list ref]
Date:   Wed Feb 13 14:36:47 2013 +0100

    ACPI / hotplug: Fix concurrency issues and memory leaks

after which acpi_bus_scan() and acpi_bus_trim() have to be run under
acpi_scan_lock (new in my tree as well).

Yes, we noticed that and only need minor updates at xen side, will send out 2 xen patches later accordingly, for cleanup and adding lock.

Thanks,
Jinsong

Moreover, I think that the introduction of ACPI-based CPU hotplug
into Xen and this point would be premature, because we need to rework
the original ACPI-based CPU hotplug and quite frankly it shouldn't
call acpi_bus_scan() directly at all.

Konrad?

Thanks,
Rafael