On 02.06.2016 16:07, Bharata B Rao wrote:
> On Thu, Jun 02, 2016 at 09:03:15AM +0200, Thomas Huth wrote:
>> On 02.06.2016 06:39, Bharata B Rao wrote:
>> ...
>>> Agreed. Here is the updated patch:
>>>
>>> spapr: Increase hotpluggable memory slots to 256
>>>
>>> From: Bharata B Rao <bhar...@linux.vnet.ibm.com>
>>>
>>> KVM now supports 512 memslots on PowerPC (earlier it was 32). Allow half
>>> of it (256) to be used as hotpluggable memory slots.
>>>
>>> Instead of hard coding the max value, use the KVM supplied value if KVM
>>> is enabled. Otherwise resort to the default value of 32.
>>>
>>> Signed-off-by: Bharata B Rao <bhar...@linux.vnet.ibm.com>
>>> ---
>>> hw/ppc/spapr.c | 15 +++++++++++++--
>>> 1 file changed, 13 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
>>> index 44e401a..c82adef 100644
>>> --- a/hw/ppc/spapr.c
>>> +++ b/hw/ppc/spapr.c
>>> @@ -1816,11 +1816,22 @@ static void ppc_spapr_init(MachineState *machine)
>>> /* initialize hotplug memory address space */
>>> if (machine->ram_size < machine->maxram_size) {
>>> ram_addr_t hotplug_mem_size = machine->maxram_size -
>>> machine->ram_size;
>>> + /*
>>> + * Number of memslots supported by KVM on PowerPC was increased
>>> + * from 32 to 512. Let us limit the number of hotpluggable slots
>>> + * to half of that (256). However ensure that number of slots
>>> + * doesn't drop below 32 on older hosts.
>>> + */
>>
>> Using "hard-coded" information like "increased to 512" in comments is
>> true for the current state, but this has a risk of being out of date
>> soon. Once we change the memslots in the kernel, this comment is not
>> true anymore and might cause confusion. Better talk about leaving half
>> of the kernel memslots for PCI and other devices, or so.
>
> Just want to note that even though we are limiting hotpluggable memory
> slots to half of max, it is always possible for other devices to eat
> into the memory hotplug slots, right ?
Right. But 256 slots for those other devices is already plenty, I think.
Let's hope that it is enough for a while - if it's not enough, we've got
to rework the kernel code again (and maybe switch to dynamic memslots
allocation instead).
Thomas
