On 03/31/16 20:05, Leon Romanovsky wrote:
On Thu, Mar 31, 2016 at 05:01:39PM -0700, Bart Van Assche wrote:
That patch is wrong because it makes the ib_srpt driver use I/O
contexts allocated by transport_alloc_session_tags() but it does
not initialize these I/O contexts properly.
Did you have a chance to see which initializations are missing in this
case? What is needed to do if we decide to fix original patch?
Except these questions, the revert is fine :)
Reviewed-by: Leon Romanovsky <leonro@xxxxxxxxxxxx>
Hello Leon,
Thanks for the review. The initializations that are missing from that
patch are the 'buf' pointer in the srpt_ioctx structure and mapping that
buffer for DMA. Another bug introduced by that patch is that it doubles
the amount of memory that is allocated for I/O contexts. New I/O context
allocations were added by that patch but the existing I/O context
allocation code was not removed.
Regarding reconsidering the original patch: before we do that it has to
be shown with numbers that the percpu_ida conversion does not decrease
performance. This is something I had already asked two months ago. See
also
http://thread.gmane.org/gmane.linux.scsi.target.devel/11253/focus=110559.
Bart.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html