[PATCH net-next v3 0/2] net: ethernet: mtk_eth_soc: 9K jumbo frame support

[Sandeep Haemoon <sandeeph@loljews.com>]

This series makes the MediaTek SoC ethernet driver support 9K jumbo
frames on the 10G XGMII MACs (MT7988).  The driver advertises a 2030-byte
max_mtu and programs the MAC max-rx field for at most 2048 bytes, so
MTU > 2K is rejected and 9K frames are not usable.

Patch 1 gates page_pool on the descriptor length of the ring: the pool
hands out order-0 pages and maps at most MTK_PP_MAX_BUF_SIZE, so a ring
whose descriptor length is larger has to use the
netdev_alloc_frag()/mtk_max_lro_buf_alloc() path.  It is pure
preparation, every ring fits today.

Patch 2 adds the MTK_NETSYS_RX_9K capability, advertises 9K max_mtu on
the XGMII MACs (from the DT phy_mode at probe, so DSA can raise the
conduit MTU before the netdev is opened), programs XMAC_RX_CFG2 and the
PDMA RX SDL, derives the shared RX buffer length from the largest MTU,
sizes the RX rings from it and keeps the MAC max-rx limits in sync with
it.  A runtime MTU change that needs larger buffers is rejected with
-EBUSY while DMA is running; shrinking is allowed.

This work is a forward-port of the MediaTek downstream jumbo work: Mason
Chang's "change default rx buffer length" and "add 9k jumbo frame
support" (MediaTek OpenWrt feed 999-eth-30/31, carried in
frank-w/BPI-Router-Linux), plus Frank Wunderlich's "try to fix crash the
right way".  The runtime reallocation of Mason's "dynamic rx buffer
adjustment" (feed 999-eth-33) is not included and will be sent
separately.

The equivalent change was tested on a BPI-R4 Pro 8X (MT7988, 10G lan6 /
MxL86252) on the 6.18.60 backport: with both ends at MTU 9000, 8K ICMP
passes and iperf3 with MSS 8900 reaches ~9.87 Gbit/s (vs ~2.57 Gbit/s at
MSS 1460); the DSA conduit is raised to 9004 at boot.  This v7.3 series
is compile-tested with CONFIG_NET_MEDIATEK_SOC=y and COMPILE_TEST=y; no
hardware run of this exact tree yet.

Changes in v3:
- gate page_pool on ring->buf_size <= MTK_PP_MAX_BUF_SIZE instead of
  ring->frag_size <= PAGE_SIZE: frag_size uses the netdev_alloc_frag()
  layout, so a ring could pass the old test while its descriptor length
  was already larger than the pool can map;
- keep 2K as the minimum RX buffer length, so MTUs up to 2K keep working
  at runtime on every SoC, DSA user ports included;
- restore MT7628's 1518 max_mtu, which v2 dropped by accident;
- in mtk_change_mtu() reject only a change that needs larger buffers, so
  a lowered MTU no longer locks the port out of further changes;
- program XMAC_RX_CFG2 with read-modify-write, keeping the bits outside
  the length field;
- replace the PDMA RX SDL field instead of OR-ing into it, which grew on
  every open with a different MTU;
- rewrite the MAC max-rx limits on every mac_config(), so a reopen with a
  smaller MTU narrows MACs that still carry the old limit;
- derive the XDP admission checks from the shared RX buffer length, so a
  program cannot attach and then never run;
- correct the commit messages: no runtime reallocation is included, and
  mtk_hw_init() now uses the shared length too;
- keep Simon Horman's Reviewed-by on 1/2 only, as 2/2 changed materially.

The order-2 allocation cost at 9K (one buffer per slot, high-order
GFP_ATOMIC refill) is unchanged from the downstream design and is called
out in the commit message.

v2: https://lore.kernel.org/all/patch-2793186c653124c4c0b3d463bde2ea3a@loljews.com/
v1: https://lore.kernel.org/all/patch-4f863a941889f0bceca1c082f2d79624@loljews.com/

Sandeep Haemoon (2):
  net: ethernet: mtk_eth_soc: skip page_pool for RX rings it cannot back
  net: ethernet: mtk_eth_soc: add 9K jumbo frame support

 drivers/net/ethernet/mediatek/mtk_eth_soc.c | 215 +++++++++++++++-----
 drivers/net/ethernet/mediatek/mtk_eth_soc.h |  10 +-
 2 files changed, 173 insertions(+), 52 deletions(-)

base-commit: c03c1a14507dfaf8b856c2608f9ac216021fc222
-- 
2.53.0