Hi!
Some NICs (e.g. bnxt) change their RSS indirection table size based on
the queue count, because the hardware table is a shared resource. The
ethtool core locks ctx->indir_size at context creation, so drivers
have to reject channel changes when RSS contexts exist.
This series adds resize helpers and wires them up in bnxt and
netdevsim.
Patch 1 adds three core helpers:
ethtool_rxfh_indir_can_resize() - read-only validation
ethtool_rxfh_indir_resize() - fold/unfold a raw table in place
ethtool_rxfh_contexts_resize_all() - resize all non-default
contexts (all-or-none), with locking and RSS_NTF notifications
Patch 2 uses them in bnxt_set_channels(), replacing the -EINVAL
rejection. Context 0 is validated first, then non-default contexts are
resized, then context 0 is mutated.
Patch 3 adds RSS context support to netdevsim with dynamic table
sizing (roundup_pow_of_two(channels) * 16), exercising the full resize
path.
Patches 4-5 add tests:
ethtool-rss.sh - netdevsim-specific bash test
rss_drv.py - Python test that runs on any device with dynamic
table sizing (skips otherwise)
Running the tests:
# netdevsim bash test (requires root, netdevsim module)
cd tools/testing/selftests/drivers/net/netdevsim
sudo ./ethtool-rss.sh
# Python test on netdevsim
cd tools/testing/selftests/drivers/net/hw
sudo ./rss_drv.py
# Python test on real hardware
sudo NETIF=eth0 ./rss_drv.py
Open items:
- Pre-existing bug in bnxt: the IFF_RXFH_CONFIGURED guard in
bnxt_set_channels() only protects context 0. If context 0 uses
defaults but non-default contexts exist, channel changes that
change the table size succeed without updating ctx->indir_size for
those contexts. The core's ethtool_check_max_channel() prevents
invalid queue references but not the size mismatch itself. This
series fixes it by resizing all contexts. Should a separate fix be
sent to net for the existing bug?
- No user-controlled minimum table size yet. The plan is to record
the user-provided indirection table length in the context (e.g.
ctx->user_indir_size) and use it as a floor when folding: reject if
new_size < user_indir_size. This way the user's original table size
serves as an implicit minimum, preventing the driver from shrinking
below what the user intended. Left for a follow-up.
- rss_drv.py lives in drivers/net/hw/ but runs on netdevsim too now.
Maybe move up a level?
Björn Töpel (5):
ethtool: Add RSS indirection table resize helpers
bnxt_en: Resize RSS contexts on channel count change
netdevsim: Add RSS context support with dynamic table sizing
selftests: netdevsim: Add RSS indirection table resize test
selftests: rss_drv: Add RSS indirection table resize tests
.../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 23 +++-
drivers/net/netdevsim/ethtool.c | 119 ++++++++++++++++-
drivers/net/netdevsim/netdevsim.h | 4 +
include/linux/ethtool.h | 3 +
net/ethtool/common.c | 126 ++++++++++++++++++
.../selftests/drivers/net/hw/rss_drv.py | 101 +++++++++++++-
.../selftests/drivers/net/netdevsim/Makefile | 1 +
.../drivers/net/netdevsim/ethtool-rss.sh | 123 +++++++++++++++++
8 files changed, 488 insertions(+), 12 deletions(-)
create mode 100755 tools/testing/selftests/drivers/net/netdevsim/ethtool-rss.sh
base-commit: ed0abfe93fd135dac223e87a3c945017b1fa8bfc
--
2.53.0
