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

Re: [Xen-devel] [PATCH RFCv2 3/4] mm/memory_hotplug: Introduce and use more memory types



On 04.12.18 10:44, Michal Suchánek wrote:
> On Fri, 30 Nov 2018 18:59:21 +0100
> David Hildenbrand <david@xxxxxxxxxx> wrote:
> 
>> Let's introduce new types for different kinds of memory blocks and use
>> them in existing code. As I don't see an easy way to split this up,
>> do it in one hunk for now.
>>
>> acpi:
>>  Use DIMM or DIMM_UNREMOVABLE depending on hotremove support in the kernel.
>>  Properly change the type when trying to add memory that was already
>>  detected and used during boot (so this memory will correctly end up as
>>  "acpi" in user space).
>>
>> pseries:
>>  Use DIMM or DIMM_UNREMOVABLE depending on hotremove support in the kernel.
>>  As far as I see, handling like in the acpi case for existing blocks is
>>  not required.
>>
>> probed memory from user space:
>>  Use DIMM_UNREMOVABLE as there is no interface to get rid of this code
>>  again.
>>
>> hv_balloon,xen/balloon:
>>  Use BALLOON. As simple as that :)
>>
>> s390x/sclp:
>>  Use a dedicated type S390X_STANDBY as this type of memory and it's
>>  semantics are very s390x specific.
>>
>> powernv/memtrace:
>>  Only allow to use BOOT memory for memtrace. I consider this code in
>>  general dangerous, but we have to keep it working ... most probably just
>>  a debug feature.
> 
> I don't think it should be arbitrarily restricted like that.
> 

Well code that "randomly" offlines/onlines/removes/adds memory blocks
that it does not own (hint: nobody else in the kernel does that), should
be restricted to types we can guarantee to work.

> Thanks
> 
> Michal
> 


-- 

Thanks,

David / dhildenb

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

 


Rackspace

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