On 8/23/19 4:19 AM, Stephen Hemminger wrote:
On Thu, 22 Aug 2019 02:17:50 +0530
<pbhagavat...@marvell.com> wrote:
From: Pavan Nikhilesh <pbhagavat...@marvell.com>
Add new Rx offload flag `DEV_RX_OFFLOAD_RSS_HASH` which can be used to
enable/disable PMDs write to `rte_mbuf::hash::rss`.
PMDs notify the validity of `rte_mbuf::hash:rss` to the applcation
by enabling `PKT_RX_RSS_HASH ` flag in `rte_mbuf::ol_flags`.
Signed-off-by: Pavan Nikhilesh <pbhagavat...@marvell.com>
Reviewed-by: Andrew Rybchenko <arybche...@solarflare.com>
Is this really a good idea?
Every bit of hardware that works on Windows with RSS is required to
supply the hash (for software steering). So if adding an additional
capability, just adds another bit of complexity, code coverage, and one
more thing that won't be tested by everyone.
If hardware has it why not set it? Adding a branch is more expensive than
an unused assignment.
If a PMD sees no gain in disabling the offload and addition branches,
nothing obligates PMD to do. Just keep as is - no problem.
However, it opens the door to skip delivery of the RSS hash
from NIC HW to host CPU if the hash is not required.
Why should it be delivered to CPU and eat PCIe bandwidth if
the information is not required?
