mbox series

[iwl-next,v5,0/2] igc: Link IRQs and queues to NAPIs

Message ID 20241028195243.52488-1-jdamato@fastly.com
Headers show
Series igc: Link IRQs and queues to NAPIs | expand

Message

Joe Damato Oct. 28, 2024, 7:52 p.m. UTC 
Greetings:

Welcome to v5.

See changelog below and in each patch for changes from v4 [1].

This revision was created due to a report from Vitaly [2], that my v4
was re-introducing a potential deadlock in runtime_resume which was
fixed in commit: 6f31d6b: "igc: Refactor runtime power management flow."

As you'll see, I've modified patch 2 to include a small wrapper to
either hold rtnl (or not) depending on whether runtime_resume or resume
are being called.

Overall, this series adds support for netdev-genl to igc so that
userland apps can query IRQ, queue, and NAPI instance relationships.
This is useful because developers who have igc NICs (for example, in
their Intel NUCs) who are working on epoll-based busy polling apps and
using SO_INCOMING_NAPI_ID, need access to this API to map NAPI IDs back
to queues.

See the commit messages of each patch for example output I got on my igc
hardware.

Thanks to reviewers and maintainers for their comments/feedback!

Thanks,
Joe

[1]: https://lore.kernel.org/netdev/20241022215246.307821-1-jdamato@fastly.com/
[2]: https://lore.kernel.org/netdev/d7799132-7e4a-0ac2-cbda-c919ce434fe2@intel.com/

v5:
  - Add a small wrapper to patch 2 to only hold rtnl when resume is
    called, but avoid rtnl when runtime_resume is called which would
    trigger a deadlock.

v4: https://lore.kernel.org/netdev/20241022215246.307821-1-jdamato@fastly.com/
  - Fixed a typo in Patch 1's commit message for the "other" IRQ number
  - Based on a bug report for e1000, closer scrutiny of the code
    revealed two paths where rtnl_lock / rtnl_unlock should be added in
    Patch 2: igc_resume and igc_io_error_detected. The code added to
    igc_io_error_detected is inspired by ixgbe's
    ixgbe_io_error_detected

v3: https://lore.kernel.org/netdev/20241018171343.314835-1-jdamato@fastly.com/
  - No longer an RFC
  - Patch 1: no changes
  - Patch 2:
      - Replace igc_unset_queue_napi with igc_set_queue_napi(..., NULL),
        as suggested by Vinicius Costa Gomes
      - Simplify implementation of igc_set_queue_napi as suggested by Kurt
        Kanzenbach, with a minor change to use the ring->queue_index

rfcv2: https://lore.kernel.org/netdev/20241014213012.187976-1-jdamato@fastly.com/
  - Patch 1: update line wrapping to 80 chars
  - Patch 2:
    - Update commit message to include output for IGC_FLAG_QUEUE_PAIRS
      enabled and disabled
    - Significant refactor to move queue mapping code to helpers to be
      called from multiple locations
    - Adjusted code to handle IGC_FLAG_QUEUE_PAIRS disabled as suggested
      by Kurt Kanzenbach
    - Map / unmap queues in igc_xdp_disable_pool and
      igc_xdp_enable_pool, respectively, as suggested by Vinicius Costa
      Gomes to handle the XDP case

rfcv1: https://lore.kernel.org/lkml/20241003233850.199495-1-jdamato@fastly.com/

Joe Damato (2):
  igc: Link IRQs to NAPI instances
  igc: Link queues to NAPI instances

 drivers/net/ethernet/intel/igc/igc.h      |  2 +
 drivers/net/ethernet/intel/igc/igc_main.c | 55 ++++++++++++++++++++---
 drivers/net/ethernet/intel/igc/igc_xdp.c  |  2 +
 3 files changed, 52 insertions(+), 7 deletions(-)


base-commit: b8ee7a11c75436b85fa1641aa5f970de0f8a575c