[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH v3 8/8] xen: arm: allocate more than one bank for 1:1 domain 0 if needed

On 06/30/2014 05:27 PM, Ian Campbell wrote:
> On Mon, 2014-06-30 at 17:20 +0100, Julien Grall wrote:
>> On 06/30/2014 05:14 PM, Ian Campbell wrote:
>>> On Mon, 2014-06-30 at 17:10 +0100, Julien Grall wrote:
>>>> Hi Ian,
>>>> On 06/30/2014 04:58 PM, Ian Campbell wrote:
>>>>> On Fri, 2014-06-27 at 14:09 +0100, Ian Campbell wrote:
>>>>>> On Fri, 2014-06-27 at 14:04 +0100, Julien Grall wrote:
>>>>>>> On 27/06/14 14:01, Ian Campbell wrote:
>>>>>>>>> Futhermore, it looks like we always allocate memory from the top of 
>>>>>>>>> this
>>>>>>>>> zone. Is it normal?
>>>>>>>> Normal behaviour is to allocate from as high as possible so I'm not
>>>>>>>> surprised that applies with a zone as well.
>>>>>>> For DOM0 bank allocation, shouldn't we try to allocate as lower as 
>>>>>>> possible? If not, we may not be able to create a second bank for dom0.
>>>>>> The loop over bits is supposed to help there, but it's not as effective
>>>>>> with higher addresses I suppose.
>>>>>> We'd need a new flag fore the allocate. Which is doable I guess. I'll
>>>>>> give it a try.
>>>>> It's a bit tricky and involves messing with the guts of the page
>>>>> allocator, which I'd rather avoid...
>>>>> We do allow allocations to be merged onto the start of bank 0, which in
>>>>> the normal case is what I would expect to happen, although it isn't
>>>>> guaranteed. In that case we would try higher addresses and might find
>>>>> enough RAM there.
>>>>> I suppose I could relax the constraint a bit and allow new banks in
>>>>> front of bank zero if they were >= 128MB (say).
>>>> So, replacing the bank zero by the new bank, right? IIRC, some Linux
>>>> version are requiring the bank to be ordered.
>>> Inserting the new bank before, retaining the ordering.
>> Thanks, this plan sounds good!
> I've implemented it in the obvious way, but I can't for the life of me
> get the scenario to actually occur despite playing with module load
> location, fdt rsvmem, mem= and dom0_mem=.
> I suppose that gives us some confidence that this is not going to happen
> very often...

If you send me the patch, I can give a try on the versatile express to
see what happen.

Julien Grall

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.