iSCSI and scatterlists

iSCSI and scatterlists

[Matthew Wilcox]

On Wed, Mar 16, 2005 at 10:44:50AM -0800, Mike Christie wrote:
> I got lost here. If you are talking about the need to kmap a sglist then 
> software iscsi has it. iscsi-sfnet used to do
> 
> while (...)
> 	kmap()
> 
> but I fixed that (I think I need to use kmap_atomic though, is that 
> correct or is it just a performance improvement - I am calling kmap from 
> a thread too so). I just added kmap_atomic to open-iscsi and I believe 
> pyx does something similar to the loop above.

Sounds like networking should grow an interface to accept a sglist as
input.  I'm really not familiar with Linux's networking stack to know
how to do it ... cc'ing netdev to get their thoughts.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: iSCSI and scatterlists

[Dmitry Yusupov]

On Wed, 2005-03-16 at 18:53 +0000, Matthew Wilcox wrote:
> On Wed, Mar 16, 2005 at 10:44:50AM -0800, Mike Christie wrote:
> > I got lost here. If you are talking about the need to kmap a sglist then 
> > software iscsi has it. iscsi-sfnet used to do
> > 
> > while (...)
> > 	kmap()
> > 
> > but I fixed that (I think I need to use kmap_atomic though, is that 
> > correct or is it just a performance improvement - I am calling kmap from 
> > a thread too so). I just added kmap_atomic to open-iscsi and I believe 
> > pyx does something similar to the loop above.
> 
> Sounds like networking should grow an interface to accept a sglist as
> input.  I'm really not familiar with Linux's networking stack to know
> how to do it ... cc'ing netdev to get their thoughts.

This is a nice idea, but will not always work with iSCSI protocol simply
because iSCSI PDU's data sizes might be negotiated to be lesser/bigger
than original WRITE's sglist size. It might help to optimize some data
path when PDU's data segment size >= sglist size. i.e. entire sglist
needs to be passed down to the stack.

i'm cross-posting to open-iscsi mailing list, so open-iscsi folks might
participate in the discussion.

Dmitry