[PATCH net-next v2 0/3] net: atlantic: convert RX path to page_pool

[Yangyu Chen <cyy@cyyself.name>]

The first two patches are standalone fixes for long-standing ring
teardown leaks and carry Cc: stable tags with their affected ranges
(patch 1: v4.11+, patch 2: v5.2+). They apply and were build- and
runtime-tested independently of each other and of the conversion in
patch 3, which is net-next material only and not tagged for stable.
Backports of patch 1 to trees without XDP support (before v5.19) only
need the xdp_frame branch dropped.

On systems where the NIC sits behind an IOMMU, the atlantic RX path
may not reach line rate: every RX buffer is allocated with
dev_alloc_pages() and mapped with dma_map_page(), then unmapped and
freed once the stack has consumed the packet. Every map/unmap is an
IOTLB/pagetable operation, and depending on the IOMMU this can
dominate the RX path at 10G rates. On an AMD Strix Halo system with a
Thunderbolt-attached QNAP QNA-T310G1S (MTU 1500, TCP over IPv6,
iperf3 -R), RX tops out at about 2.2 Gbit/s.

An earlier patch [1] worked around this by making the RX page order
tunable via a module parameter, amortizing one map/unmap over eight
pages worth of frames. The review feedback was to convert the driver
to the page_pool API instead of adding a knob. This series does that
conversion.

Patches 1 and 2 fix two long-standing teardown leaks first: TX
buffers stranded beyond the budgeted single aq_ring_tx_clean() pass
or past the frozen hw_head, and RX pages parked in consumed but not
yet refilled slots that the bounded deinit walk never visited. Both
are silent leaks today; after the conversion each stranded buffer
would hold a page_pool fragment reference and turn every interface
down into a permanently stalled pool shutdown (reproduction logs are
in the notes of both patches), so they need fixing before the
conversion lands.

Patch 3 converts the RX path to page_pool with the fragment API.
Pages are DMA-mapped once when they enter the per-ring pool
(PP_FLAG_DMA_MAP) and stay mapped while they recycle between the
driver and the stack, so steady-state RX performs no IOMMU work.
page_pool_dev_alloc_frag() takes over the sub-page splitting the
driver's hand-rolled "page flip" scheme did based on page_ref_count(),
buffer ownership becomes transfer-based, and the XDP memory model
switches to MEM_TYPE_PAGE_POOL. The diff is generated with
--histogram so the removed per-page helpers show as whole-function
deletions and aq_get_rxpages() as an in-place rewrite.

Performance, QNA-T310G1S (AQC100) behind Thunderbolt/IOMMU, MTU 1500,
TCP over IPv6, iperf3 -R:

  before:  2.24 Gbit/s
  after:   9.14 Gbit/s

matching what previously required the rxpageorder=3 workaround, but
with order-0 pages and no tunable.

Tested on the same setup: line rate with plain RX and with XDP_PASS;
XDP_TX with a MAC-swap reflector under sustained traffic; repeated
ifdown/ifup and module unload cycles under both plain RX and XDP_TX
load complete without "stalled pool shutdown" warnings or other splats.

Changes in v2 (thanks to Mina Almasry for the review):
 - split the RX deinit gap leak out of the conversion into its own
   fix (patch 2) and generate the conversion diff with --histogram so
   it reads as whole-function removals plus an in-place
   aq_get_rxpages() rewrite
 - add Cc: stable tags with the affected ranges to patches 1 and 2
 - comment the ring-before-rxq-registration ordering and its error
   unwind in aq_vec_ring_alloc()
 - aq_ring_tx_deinit(): return early instead of goto, drop the
   likely()/unlikely() annotations
 - aq_ptp_ring_alloc(): use the err_exit_xdp_rxq label instead of
   open-coding the unregister in the memory model error path
 - rebase on current net-next

[1] https://lore.kernel.org/lkml/tencent_E71C2F71D9631843941A5DF87204D1B5B509@qq.com/

v1: https://lore.kernel.org/lkml/tencent_7DB01BE7F8FA056BB5F11D3570CF636C4309@qq.com/

Yangyu Chen (3):
  net: atlantic: free stranded TX buffers on ring deinit
  net: atlantic: free RX pages of consumed but not refilled buffers
  net: atlantic: convert RX path to page_pool

 drivers/net/ethernet/aquantia/Kconfig         |   1 +
 .../ethernet/aquantia/atlantic/aq_ethtool.c   |   3 -
 .../net/ethernet/aquantia/atlantic/aq_ptp.c   |  18 +-
 .../net/ethernet/aquantia/atlantic/aq_ring.c  | 261 +++++++++---------
 .../net/ethernet/aquantia/atlantic/aq_ring.h  |   7 +-
 .../net/ethernet/aquantia/atlantic/aq_vec.c   |  23 +-
 6 files changed, 169 insertions(+), 144 deletions(-)


base-commit: 89d8006259b81dd25c962f6cc8d7ab268d6ea426
-- 
2.47.3