[PATCH net-next v3 00/15] net: macb: implement context swapping

[Théo Lebrun]

MACB has a pretty primitive approach to buffer management. They are all
stored in `struct macb *bp`. On operations that require buffer realloc
(set_ringparam & change_mtu at the moment), the only option is to close
the interface, change our global state and re-open the interface.

Two issues:
- It doesn't fly on memory pressured systems; we free our precious
  buffers and don't manage to reallocate fully, meaning our machine
  just lost its network access.
- Anecdotally, it is pretty slow because it implies a full PHY reinit.

Instead, we shall:
 - allocate a new context (including buffers) first
 - if it fails, early return without any impact to the interface
 - stop interface
 - update global state (bp, netdev, etc)
 - pass newly allocated buffer pointers to the hardware
 - start interface
 - free old context

This is what we implement here. Both .set_ringparam() and
.ndo_change_mtu() are covered by this series. In the future,
at least .set_channels() [0], XDP [1] and XSK [2] would benefit.

The change is super intrusive so conflicts will be major. Sorry!

Thanks,
Have a nice day,
Théo

[0]: https://lore.kernel.org/netdev/20260317-macb-set-channels-v4-0-1bd4f4ffcfca@xxxxxxxxxxx/
[1]: https://lore.kernel.org/netdev/20260323221047.2749577-1-pvalerio@xxxxxxxxxx/
[2]: https://lore.kernel.org/netdev/20260304-macb-xsk-v1-0-ba2ebe2bdaa3@xxxxxxxxxxx/

Signed-off-by: Théo Lebrun <theo.lebrun@xxxxxxxxxxx>
---
Changes in v3:
- Use `const struct macb_info *info` instead of bare `u32 caps` as
  helper arguments, for type safety.
- macb_interrupt(): the pre-lock readl(ISR) to detect spurious
  interrupts is only done if CLEAR_ON_WRITE.
- Don't forget allocating context in at91ether_open().
- swap:
  - Refuse swap for EMAC HW; it would crash because codepaths are so
    different.
  - Grab new bp->mac_cfg_lock to serialise with phylink MAC callbacks.
    We cannot rely on phydev->lock because it isn't present in the SFP
    or fixed-link cases. We also want to avoid phylink_stop() which
    triggers a slow PHY retrain.
- swap start:
  - We used to do disable-irqs-and-hw then drain-all-bh-features, but
    then HW might be raced against. Instead we disable-irqs then
    drain-all-bh then disable-hw which means at disable-hw step no BH
    context can be active.
  - Use macb_halt_tx() helper to properly stop HW.
  - Disable BH features before netif_tx_disable() call to avoid queue
    wakeup races.
  - Use macb_queue_isr_clear() helper instead of manual if-then-writel.
- swap end:
  - Grab bp->lock for the hardware reinit sequence composed of DMACFG
    and NCR writes.
  - Drop now useless EMAC check (we refuse EMAC HW before swapping).
- nits:
  - New patch to rename macb_{alloc,free}_consistent() which don't only
    allocate consistent buffers since a long time ago.
  - Fix the start_xmit verbose netdev_vdbg() format string from %hu to %u
    because the queue index type changed.
  - Strong commit reword from "unify `struct macb *` naming convention"
    to "unify variable naming convention in at91ether functions" which
    was underselling the changes.
- Rebase upon latest net-next/main (1c664ec4b9ea).
- Link to v2: https://patch.msgid.link/20260410-macb-context-v2-0-af39f71d40b6@xxxxxxxxxxx

Changes in v2:
- Patch "add subset of `struct macb` to `struct macb_context`" was
  messed up. It contained much more than what the name implied. Split
  into three commits (I caused trouble by rebase reordering).
- Fix tieoff; V1 allocated it without initialisation.
- Fix NULL pointer dereference on context in mab_get_regs() and
  macb_get_ringparam() when interface is offline.
- Patch "unify device pointer naming convention":
  - Fix build issue when CONFIG_NETCONSOLE=y.
  - Rename `struct net_device *dev` to `netdev` in macb.h.
  - Rename `struct phy_device *phy` to `phydev` in macb_main.c.
- On swap, call netdev_tx_reset_queue() to reset all DQL counters.
- At end of swap, add missing kfree(old_ctx).
- During HW disabling in swap, grab bp->lock to protect against IRQ
  handler.
- On swap, cancel the three BH features MACB has:
  bp->hresp_err_bh_work, bp->tx_lpi_work and queue->tx_error_task.
- On swap, call macb_configure_dma() which writes buffer size to
  hardware registers. This is important because the change_mtu codepath
  changes the buffer size.
- Rebase onto latest net-next/main (58dd34dbd5b0) & resolve conflicts.
- Link to v1: https://patch.msgid.link/20260401-macb-context-v1-0-9590c5ab7272@xxxxxxxxxxx

---
Théo Lebrun (15):
      net: macb: drop "consistent" from alloc/free function names
      net: macb: unify device pointer naming convention
      net: macb: unify variable naming convention in at91ether functions
      net: macb: unify queue index variable naming convention and types
      net: macb: enforce reverse christmas tree (RCT) convention
      net: macb: allocate tieoff descriptor once across device lifetime
      net: macb: introduce macb_context struct for buffer management
      net: macb: avoid macb_init_rx_buffer_size() modifying state
      net: macb: make `struct macb` subset reachable from macb_context struct
      net: macb: change caps helpers signatures
      net: macb: change function signatures to take contexts
      net: macb: introduce macb_context_alloc() helper
      net: macb: re-read ISR inside IRQ handler locked section
      net: macb: use context swapping in .set_ringparam()
      net: macb: use context swapping in .ndo_change_mtu()

 drivers/net/ethernet/cadence/macb.h      |  127 +-
 drivers/net/ethernet/cadence/macb_main.c | 1849 +++++++++++++++++-------------
 drivers/net/ethernet/cadence/macb_pci.c  |   46 +-
 drivers/net/ethernet/cadence/macb_ptp.c  |   26 +-
 4 files changed, 1199 insertions(+), 849 deletions(-)
---
base-commit: 64e6f7613459a9b344fb09a129584435b281088b
change-id: 20260401-macb-context-bd0caf20414d

Best regards,
--  
Théo Lebrun <theo.lebrun@xxxxxxxxxxx>