lan966x: execute xdp_do_flush() before napi_complete_done()
authorMagnus Karlsson <magnus.karlsson@intel.com>
Wed, 25 Jan 2023 07:48:58 +0000 (08:48 +0100)
committerJakub Kicinski <kuba@kernel.org>
Sat, 28 Jan 2023 06:27:59 +0000 (22:27 -0800)
commit12b5717990c81fc2f6f2aa9d53d960b916478d4f
treeb59aabfb0507f8f797bdc146555661663fa145c6
parent2ccce20d51faa0178086163ccb6c84a099a87ab4
lan966x: execute xdp_do_flush() before napi_complete_done()

Make sure that xdp_do_flush() is always executed before
napi_complete_done(). This is important for two reasons. First, a
redirect to an XSKMAP assumes that a call to xdp_do_redirect() from
napi context X on CPU Y will be followed by a xdp_do_flush() from the
same napi context and CPU. This is not guaranteed if the
napi_complete_done() is executed before xdp_do_flush(), as it tells
the napi logic that it is fine to schedule napi context X on another
CPU. Details from a production system triggering this bug using the
veth driver can be found following the first link below.

The second reason is that the XDP_REDIRECT logic in itself relies on
being inside a single NAPI instance through to the xdp_do_flush() call
for RCU protection of all in-kernel data structures. Details can be
found in the second link below.

Fixes: a825b611c7c1 ("net: lan966x: Add support for XDP_REDIRECT")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Acked-by: Steen Hegelund <Steen.Hegelund@microchip.com>
Link: https://lore.kernel.org/r/20221220185903.1105011-1-sbohrer@cloudflare.com
Link: https://lore.kernel.org/all/20210624160609.292325-1-toke@redhat.com/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
drivers/net/ethernet/microchip/lan966x/lan966x_fdma.c