On 7/23/14 10:53 , Jeff Epler wrote:
> On Wed, Jul 23, 2014 at 09:29:57AM -0600, Sebastian Kuzminsky wrote:
>> bf7c06b hostmot2: Introduce queue_{read,write} methods
>>
>>       queue_read() and queue_write() seems almost but not quite identical
>>       to tram reads and writes.  The difference is the queued reads/writes
>>       are adjustable on a per-batch basis, whereas the tram reads/writes
>>       must be configured at driver load time.  But they're never used in
>>       that dynamic way, so maybe a stronger (generic) tram implementation
>>       would be better?
>
> Queue read/write can work on any addresses it wants.  In the ethernet
> product, this has an overhead of 4 bytes anytime it's not a contiguous
> region.  But the cost of sending two packets instead of one is much
> higher.  This is "good enough" to enable ethernet and I assume that's
> why micges did it this way rather than truly finishing use of tram.
> Do you feel strongly enough about this to keep hm2_eth out until it
> actually uses tram?

Ah, i didn't think of the inclusion of the address in the queued io vs 
the tram io, that makes sense.  Ignore that part of my feedback, don't 
make hm2_eth wait for tram.  We can change it later if/when we add tram 
support.


-- 
Sebastian Kuzminsky

------------------------------------------------------------------------------
Want fast and easy access to all the code in your enterprise? Index and
search up to 200,000 lines of code with a free copy of Black Duck
Code Sight - the same software that powers the world's largest code
search on Ohloh, the Black Duck Open Hub! Try it now.
http://p.sf.net/sfu/bds
_______________________________________________
Emc-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/emc-developers

Reply via email to