Re: [Xen-devel] [PATCH RFC 4/4] xen-block: introduce a new request type to unmap grants

[David Vrabel <david.vrabel@citrix.com>]

On 11/07/13 16:12, Roger Pau Monné wrote:
> On 11/07/13 15:48, David Vrabel wrote:
>> On 10/07/13 10:19, Roger Pau Monné wrote:
>>>
>>> >From 1ede72ba10a7ec13d57ba6d2af54e86a099d7125 Mon Sep 17 00:00:00 2001
>>> From: Roger Pau Monne <roger.pau@citrix.com>
>>> Date: Wed, 10 Jul 2013 10:22:19 +0200
>>> Subject: [PATCH RFC] xen-blkfront: revoke foreign access for grants not
>>>  mapped by the backend
>>> MIME-Version: 1.0
>>> Content-Type: text/plain; charset=UTF-8
>>> Content-Transfer-Encoding: 8bit
>>>
>>> There's no need to keep the foreign access in a grant if it is not
>>> persistently mapped by the backend. This allows us to free grants that
>>> are not mapped by the backend, thus preventing blkfront from hoarding
>>> all grants.
>>>
>>> The main effect of this is that blkfront will only persistently map
>>> the same grants as the backend, and it will always try to use grants
>>> that are already mapped by the backend. Also the number of persistent
>>> grants in blkfront is the same as in blkback (and is controlled by the
>>> value in blkback).
>>
>> Does this place in implicit requirement on the behaviour of the backend.
>>  i.e., the backend must persistently map the first N grants and always
>> unmap the remainder?  If so, this should be clearly documented.
> 
> No, the backend can persistently map whatever grants it wants, and in
> fact we have a LRU in the backend that keeps unmapping a certain amount
> of grants that have not been used over a period of time.

I don't see a mechanism by which the frontend can notice that the
backend has unmapped an unused grant.  It can only notice when a request
using that grant is completed.

>> It also seems odd to have the backend decide how much frontend resource
>> can be consumed at anyone time.  It's not clear how the backend is
>> supposed to know how many persistent grants it should hang on to.
> 
> blkfront has to at least persistently map the same grants as the
> backend. blkfront could persistently map all grants, but then we will
> have grant shortage, and I doubt there's much performance gain from
> persistently mapping grants in blkfront but not blkback (since the
> biggest performance penalty comes from the unmapping done in blkback and
> TLB flush involved).

I'm saying that the frontend needs to be able to set a cap on the number
of persistent grants kept by the backend.  If other demands on a
domain's grant table resource means it can only spare G grants for a
particular frontend it needs to be able to ensure this (with the
cooperation of the backend).

David