Re: [Qemu-devel] [PATCH 2/2] hw/pci-host/x86: extend the 64-bit PCI hole relative to the fw-assigned base

[Laszlo Ersek]

On 09/27/18 09:03, Igor Mammedov wrote:
> On Thu, 27 Sep 2018 00:31:07 +0200
> Laszlo Ersek <address@hidden> wrote:

>> - I guess I could ascertain the mis-alignment by using small guest RAM
>> (128MB), a single DIMM hotplug slot so that reserved-memory-end is
>> rounded up to 5GB (through the 1GB alignment), and a single ivshmem
>> device with a 2GB BAR (mis-aligned with the reserved-memory-end at 5GB).
>> Is this the best compromise?
>>
>> (The last option is basically an optimization of my previous reproducer,
>> namely 5GB guest RAM and 4GB BAR. The 5GB guest RAM made sure that RAM
>> would end at an *odd* GB number -- we can optimize that by using an
>> empty DIMM hotplug area rather than guest RAM. And the 4GB BAR can be
>> optimized to the smallest *even* (positive) number of GBs, namely 2GB.
> that probably would do the job

There's some black magic going on in the sizing of the reserved memory
area, due to at least commit 085f8e88ba73 ("pc: count in 1Gb hugepage
alignment when sizing hotplug-memory container", 2014-11-24). At this
point I'm basically trying to find whatever numbers that (a) don't take
up many resources, (b) reproduce the issue. :)

Thanks
Laszlo