On Fri, Sep 23, 2011 at 7:24 AM, Jan Beulich <[email protected]> wrote:
>>>> On 22.09.11 at 17:44, "Kulkarni, Shanti" <[email protected]> wrote:
>> It works properly using the last OS 11.2 2.6.31 kernel on top of 11.4,
>> so I opened a bug
>> (https://bugzilla.novell.com/show_bug.cgi?id=719858). Thanks for your
>> help.
>
> Seems like I was wrong with the assumption that this would be a
> problem with the native kernel too. There was a resource handling
> change in 2.6.37 that isn't compatible with the Xen kernel's memory
> handling, which precludes resource re-assignment on any system
> with (roughly) memory extending past the 4G boundary.
>
> Jan
>

My system has 8G, so that would explain it. Thank you.

I assume I'll have to remove memory or keep my kernel < 2.6.37 for the
time being, but do you know if it's likely that this situation will
change with the 3.0-final kernel now that Xen's been merged into it?

>> On Thu, Sep 22, 2011 at 9:07 AM, Jan Beulich <[email protected]> wrote:
>>>>>> On 22.09.11 at 15:40, "Kulkarni, Shanti" <[email protected]> wrote:
>>>> Here is the kernel log for the new kernel (2.6.37).  You're correct
>>>> that I no longer have the old kernel installed. From my limited
>>>> research on the subject I think the entry "[    0.188870] pci
>>>> 0000:04:00.1: reg 10: [mem 0xfe6fec00-0xfe6fecff]" means that the
>>>> offending device is not being realigned, otherwise it would start on a
>>>> multiple of 0x1000.
>>>
>>> You didn't read on - the message stating that the code doing the
>>> alignment got triggered follows immediately.
>>>
>>>> [    0.180274] pci 0000:00:14.2: [1002:4383] type 0 class 0x000403
>>>> [    0.180296] pci 0000:00:14.2: reg 10: [mem 0xfe2f8000-0xfe2fbfff 64bit]
>>>> [    0.180341] pci 0000:00:14.2: Disabling memory decoding and
>>>> releasing memory resources.
>>>> [    0.180435] pci 0000:00:14.2: PME# supported from D0 D3hot D3cold
>>>> [    0.180439] pci 0000:00:14.2: PME# disabled
>>>>...
>>>> [    0.188602] pci 0000:04:00.0: [1033:0035] type 0 class 0x000c03
>>>> [    0.188627] pci 0000:04:00.0: reg 10: [mem 0xfe6ff000-0xfe6fffff]
>>>> [    0.188711] pci 0000:04:00.0: Disabling memory decoding and
>>>> releasing memory resources.
>>>> [    0.188814] pci 0000:04:00.0: supports D1 D2
>>>> [    0.188816] pci 0000:04:00.0: PME# supported from D0 D1 D2 D3hot
>>>> [    0.188821] pci 0000:04:00.0: PME# disabled
>>>> [    0.188844] pci 0000:04:00.1: [1033:00e0] type 0 class 0x000c03
>>>> [    0.188870] pci 0000:04:00.1: reg 10: [mem 0xfe6fec00-0xfe6fecff]
>>>> [    0.188953] pci 0000:04:00.1: Disabling memory decoding and
>>>> releasing memory resources.
>>>
>>> (namely here)
>>>
>>>> [    0.189038] pci 0000:04:00.1: Rounding up size of resource #0 to 0x1000.
>>>
>>> (and here)
>>>
>>>> [    0.189126] pci 0000:04:00.1: supports D1 D2
>>>> [    0.189127] pci 0000:04:00.1: PME# supported from D0 D1 D2 D3hot
>>>> [    0.189132] pci 0000:04:00.1: PME# disabled
>>>> [    0.189166] pci 0000:04:01.0: [1033:00e7] type 0 class 0x000c00
>>>> [    0.189192] pci 0000:04:01.0: reg 10: [mem 0xfe6fd000-0xfe6fdfff]
>>>> [    0.189275] pci 0000:04:01.0: Disabling memory decoding and
>>>> releasing memory resources.
>>>> [    0.189393] pci 0000:04:01.0: supports D1 D2
>>>> [    0.189394] pci 0000:04:01.0: PME# supported from D0 D1 D2 D3hot
>>>> [    0.189399] pci 0000:04:01.0: PME# disabled
>>>>...
>>>
>>> And then:
>>>
>>>> [    0.227832] pci 0000:00:14.2: BAR 0: can't assign mem (size 0x4000)
>>>>...
>>>
>>> and further
>>>
>>>> [    0.228453] pci 0000:04:00.0: BAR 0: can't assign mem (size 0x1000)
>>>> [    0.228516] pci 0000:04:00.1: BAR 0: can't assign mem (size 0x1000)
>>>> [    0.228579] pci 0000:04:01.0: BAR 0: can't assign mem (size 0x1000)
>>>
>>> That seems to be a problem not only with the Xen kernel (just try passing
>>> the same options to the native kernel and see whether you get the same
>>> errors), and is certainly dependent on how the BIOS does its original
>>> resource assignment.
>>>
>>> I'd suggest opening a bug (as it should minimally be able to re-use the
>>> BIOS-assigned memory ranges that already were 4k-aligned), but you
>>> should expect being asked to supply information documenting that this
>>> in fact worked on 2.6.31 (for this to be considered a regression).
>>>
>>> Jan
>>>
>>>
>> --
>> To unsubscribe, e-mail: [email protected]
>> For additional commands, e-mail: [email protected]
>
>
>
--
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to