Control: tag -1 moreinfo
On Thu, 2013-02-14 at 08:34 +0100, Bernhard Schmidt wrote:
> Package: linux-2.6
> Version: 2.6.32-46
> Severity: normal
> Tags: squeeze
>
> This is happening with a locally patched kernel with the patch from
> Bug #699913, otherwise there is another problem that also affects
> memory hotplug on both Squeeze and Wheezy.
>
> This particular problem has only been observed with Squeeze, but pretty
> reliable there. When you hotplug system memory in VMware ESXi 5.0 one
> memory bank fails to come online with the following error:
[...]
> This machine had 1GB before (memory0-7) and has been upgraded to 2GB. Note
> that memory8 is missing, consequently the system is short of 128MB.
>
> I have found a VMware forum article that describes this problem quite well:
>
> http://communities.vmware.com/message/1299666#1299666
>
> There is a patch linked there that apparently has been in the upstream
> kernel once, but has been reverted (?).
Yes, it was very quickly reverted, presumably because it broke some
systems.
[...]
> 40000000-40007fff : 0000:00:0f.0 seems to be allocating memory where it wants
> to plug the new memory bank.
[...]
This is not memory, it's MMIO (memory-mapped input/output) for the
emulated VGA card. The kernel controls the addresses used for MMIO for
PCI devices, but does not control the addresses used for RAM.
Apparently the system firmware (or in this case the hypervisor) can
provide hints through ACPI about what addresses it should use, to avoid
conflicting with hot-added RAM. But the kernel version in squeeze
ignores those hints.
Does the kernel parameter 'pci=use_crs' avoid this?
Ben.
--
Ben Hutchings
We get into the habit of living before acquiring the habit of thinking.
- Albert Camus
Attachment:
signature.asc
Description: This is a digitally signed message part