Re: [Xen-devel] [PATCH] xen/mcfg: Call PHYSDEVOP_pci_mmcfg_reserved before PCI enumeration

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 22.09.15 at 15:26, <eswierk@xxxxxxxxxxxxxxxxxx> wrote:
> On Tue, Sep 22, 2015 at 5:35 AM, Ed Swierk <eswierk@xxxxxxxxxxxxxxxxxx> wrote:
>> So if the contract is that Dom0 tells Xen about mmcfgs before the
>> devices they cover, then Linux ought to call pci_mmcfg_reserved from
>> (or immediately after) both pci_mmcfg_early_init() and
>> pci_mmcfg_late_init().
> 
> Brainstorming possible approaches:
> 
> I don't see an obvious way to hook into those functions (or their
> callers) without injecting Xen-specific code. Is there a precedent to
> follow?
> 
> Alternatively, we could just call the equivalent of xen_mcfg_late()
> from the existing xen_{add,remove}_device() notifiers. This would
> generate a lot of useless pci_mmcfg_reserved hypercalls (number of
> devices times number of mmcfg areas), but pci_mmcfg_arch_enable() in
> Xen should happily ignore the redundant ones. The advantage of this
> approach other than simplicity is that it makes the mmcfg -> device
> setup ordering very explicit.

Well, I can see this as a reasonable approach to avoid adding Xen-
specific code to generic x86 code, but I don't like such redundancy.
Note however that the redundancy could be brought down: For one
you'd only need to report the MMCFG area covering the device in
question. And then the kernel could track which ones it reported
already.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel