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

Re: [Xen-devel] [PATCH 1/4] Add XEN pvSCSI protocol description

>>> On 30.06.14 at 13:18, <JGross@xxxxxxxx> wrote:
> On 06/30/2014 01:02 PM, Jan Beulich wrote:
>>>>> On 27.06.14 at 19:11, <konrad.wilk@xxxxxxxxxx> wrote:
>>> On Fri, Jun 27, 2014 at 04:34:33PM +0200, jgross@xxxxxxxx wrote:
>>>> +/*
>>>> + * Maximum scatter/gather segments per request.
>>>> + *
>>>> + * Considering balance between allocating at least 16 "vscsiif_request"
>>>> + * structures on one page (4096 bytes) and the number of scatter/gather
>>>> + * elements needed, we decided to use 26 as a magic number.
>>>> + */
>>>> +#define VSCSIIF_SG_TABLESIZE              26
>>>> +
>>>> +/*
>>>> + * based on Linux kernel 2.6.18
>>> This being a bit more .. new, - do these sizes make sense anymore?
>>> Should they be extended a bit? Or have support for using the
>>> old ones (as default) and then negotiate new sizes with the kernel?
>>> (If of course there is a difference?)
>> As JÃrgen already said (and as you should have noticed yourself) -
>> this is an interface definition that we can't just change. Negotiation
>> of larger counts is an option, which is what VSCSIIF_ACT_SCSI_SG_PRESET
>> is intended for (the implementation of which isn't part of this patchset
>> afaics, but could be made available on top of it).
> I don't think this is the proper way to handle larger SG lists. The
> VSCSIIF_ACT_SCSI_SG_PRESET option would just bump the maximum length
> up to 31 (currently 26). Or was it thought to be incremental (multiple
> presets for one request)?

Yes, that's how it's to be used. See xen-vscsi-large-requests.patch
it the newer of our trees.

> I'd rather add a way to specify SG lists
> residing in an own (granted) page (or even multiple pages). This would
> allow 512*26 SG entries without having to change the request structure.
> The capability to handle this feature could be indicated via xenstore.

And yes, considering the similar changes in blkif I agree that this
would be the preferred method today. That patch, however,
predates those blkif enhancements.


Xen-devel mailing list



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