> Should the device attributes (for instance max_send_wr) be the max > supported by the HW or the max supported by the OS or something else?
I think they should be what the consumer can expect to work -- ie the upper limits on the current system, whether those limits are imposed by the HW, driver, or whatever. > For instance: Chelsio's HW can handle very large work queues, but > since Linux limits the size of contiguous dma coherent memory > allocations, the actual limits are much smaller. Which should I be > using for the device attributes? The smaller limit I think. > Also, the chelsio device uses a single work queue to implement the SQ > and RQ abstractions. So the max SQ depth depends on the RQ depth and > vice versa. This leads to device max attributes that aren't that > useful. Not nice... I guess you could return half of the work queue depth for each max since that is guaranteed to work? > I'm wondering what application writes should glean from these attributes... I think the limits are pointless unless they give something that an application can actually request. Some of the limits are pretty useless, eg max # of CQs, since another app may have already used up all the CQs. - R. _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
