Re: [PATCH RESEND 1/3] OvmfPkg/XenSupport: remove usage of prefetchable PCI host bridge aperture

From: Laszlo Ersek <lersek@redhat.com>
To: Anthony PERARD <anthony.perard@citrix.com>,
	Igor Druzhinin <igor.druzhinin@citrix.com>
Cc: jordan.l.justen@intel.com, edk2-devel@lists.01.org,
	xen-devel@lists.xenproject.org, julien.grall@arm.com,
	ard.biesheuvel@linaro.org
Subject: Re: [PATCH RESEND 1/3] OvmfPkg/XenSupport: remove usage of prefetchable PCI host bridge aperture
Date: Wed, 20 Mar 2019 12:15:34 +0100	[thread overview]
Message-ID: <c97f13ad-b9c0-6716-7c34-9d2e226297a6__26326.5275558055$1553080591$gmane$org@redhat.com> (raw)
In-Reply-To: <20190319140319.GJ11621@perard.uk.xensource.com>

On 03/19/19 15:03, Anthony PERARD wrote:
> On Thu, Mar 14, 2019 at 07:45:56PM +0000, Igor Druzhinin wrote:
>> On 14/03/2019 17:41, Anthony PERARD wrote:
>>> Hi,
>>>
>>> On Wed, Mar 06, 2019 at 12:40:54PM +0000, Igor Druzhinin wrote:
>>>> This aperture doesn't exist in OVMF and trying to use it causes
>>>
>>> I'm trying to understand what you mean by writing "doesn't exist in
>>> OVMF". Are prefetchable BAR not handled by ScanForRootBridges() ?
>>> Or is it the emulation of the config space that isn't correct?
>>> Maybe QEMU should lies about a BAR been prefetchable?
>>
>> The problem here is: hvmloader places BARs initially disregarding
>> prefetchable bit in an arbitrary order because essentially there is only
>> 1 aperture for the host bridge in emulated system under Xen (and KVM as
>> well). In PcatPciRootBridgeParseBars() we construct apertures for high
>> level OVMF code by reading the BAR placement information after
>> hvmloader. It often appears that there are prefetchable and
>> non-prefetchable BARs coexist with each other and make prefetchable and
>> non-prefetchable apertures overlap. This eventually triggers an
>> assertion in high level OVMF code because that shouldn't happen.
> 
> Thanks for the explanation. Could you add it to the patch description?
> 
>> OVMF for KVM is not using prefetchable BAR at all - see
>> PciHostBridgeGetRootBridges() in which it passes mNonExistAperture dummy
>> object to high level code. I think it's wrong to construct a
>> prefetchable aperture for Xen and this code should be removed as it's
>> done for QEMU-KVM. Do you think this patch needs to do that?
> 
> It would be nice to remove the code that isn't useful so feel free to do
> it and/or keep the current patch with the description updated.

Right -- I'd like to add one keyword here, for background: EFI_PCI_HOST_BRIDGE_COMBINE_MEM_PMEM. (It's documented in both edk2 [MdePkg/Include/Protocol/PciHostBridgeResourceAllocation.h] and the PI spec [EFI_PCI_HOST_BRIDGE_RESOURCE_ALLOCATION_PROTOCOL.GetAllocAttributes()].)

Thanks
Laszlo

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel