Re: PAGE_SIZE (64KB), while block driver 'struct request' deals with < PAGE_SIZE (up to 44Kb). Was:Re: [RFC] Support of non-indirect grant backend on 64KB guest

[Julien Grall <julien.grall@citrix.com>]

Hi Konrad,

On 04/09/2015 18:32, Konrad Rzeszutek Wilk wrote:
> On Fri, Sep 04, 2015 at 05:15:13PM +0100, Julien Grall wrote:
>> On 04/09/15 16:41, Konrad Rzeszutek Wilk wrote:
>>>> Maybe we could fall back to the previous plan of modifying xen-blkfront
>>>> for the moment?
>>>
>>> Which afaic need to be reposted?
>>
>> Right. Although I didn't see any comment on the patch. All the comments
>> was about the problem. Does it mean that you and Roger are "happy" with
>> the way it's done?
>
> There were some #idef I think? And I recall seeing the segment limit being
> advertised as PAGE_SIZE / 2?

There is no ifdef but a PAGE_SIZE / 2. I know about the latter and plan 
to replace it. It was only to have a quick patch to expose my problem.

> I dug in the other block drivers to figure out what they
> do when the underlaying storage cannot handle < PAGE_SIZE data.
> And I couldn't find them. Which means this should be really dealt
> on the drivers side (xen-blkfront) as an quirk.
>
> Anyhow what I am going to do is - once it is reposted, force the
> driver under x86 to work under this condition. That is disable
> persistent support and only use 2048 .. and then drive some IO.
> If all goes well I will send it in a git pull to Jens.

It will also depends on 64KB series which I plan to repost next week. I 
will send as follow-up.

To test it properly on x86 it will be necessary to drop on BUG_ON in the 
code which ensure that the extra req is never use for 4KB
(see BUG_ON((XEN_PAGE_SIZE == PAGE_SIZE) && require_extra_req)).

Regards,

-- 
Julien Grall