Re: [Xen-devel] grant table size

[Jan Beulich <jbeulich@xxxxxxxx>]

On 20.11.2019 12:18,  Durrant, Paul  wrote:
>> -----Original Message-----
>> From: Jan Beulich <jbeulich@xxxxxxxx>
>> Sent: 20 November 2019 12:09
>> To: Durrant, Paul <pdurrant@xxxxxxxxxx>
>> Cc: Roger Pau Monné <roger.pau@xxxxxxxxxx>; xen-devel@xxxxxxxxxxxxxxxxxxxx
>> Subject: Re: [Xen-devel] grant table size
>>
>> On 20.11.2019 11:49,  Durrant, Paul  wrote:
>>>> From: Roger Pau Monné <roger.pau@xxxxxxxxxx>
>>>> Sent: 20 November 2019 11:06
>>>>
>>>> Do you have in mind to signal this somehow to guests, or the
>>>> expectation is that the guest will have to poll GNTTABOP_query_size
>>>> and at some point the size will increase?
>>>
>>> I don't think the guest need care until its grant table grows to the
>>> max. At that point, rather than giving up, the guest would re-query
>>> the max value to see if there is now more headroom and then re-size
>>> its internal data structures accordingly.
>>
>> If we consider dynamic adjustments, what about shrinking of the
>> table? This would of course require some form of guest consent,
>> but it would be nice if the option would at least be accounted
>> for when working out how all of this should behave, even if the
>> case may not get handled right now.
>>
> 
> Well, perhaps we could have a set_size gnttab op where a guest would
> be allowed to call it with a value less than (or equal to) its current
> max, so that it can voluntarily yield its headroom, but only a
> privileged guest would be allowed to call it with an increased max
> value?

Ah yes, this sounds good.

> I'm not sure what mechanism would be best for requesting a guest
> reduction though, I guess probably xenstore... something akin to
> balloon target pages?

Perhaps.

> A guest reduction of max is of pretty limited value though AFAICT as
> only in-use frames really use any memory. The (active/shared/status)
> arrays could, of course, be reduced in size but that only gets you a
> few bytes back.

Well, if this really was about just "a few bytes", why wouldn't we
allow arbitrary size grant tables to begin with?

Jan

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