mbox series

[v2,net-next,0/6] sfc: TXQ refactor

Message ID 02b2bb5a-f360-68cb-3c13-b72ced1ecd7b@solarflare.com
Headers show
Series sfc: TXQ refactor | expand

Message

Edward Cree Sept. 3, 2020, 9:30 p.m. UTC 
Refactor and unify partner-TXQ handling in the EF100 and legacy drivers.

The main thrust of this series is to remove from the legacy (Siena/EF10)
 driver the assumption that a netdev TX queue has precisely two hardware
 TXQs (checksummed and unchecksummed) associated with it, so that in
 future we can have more (e.g. for handling inner-header checksums) or
 fewer (e.g. to free up hardware queues for XDP usage).

Changes from v1:
 * better explain patch #1 in the commit message, and rename
   xmit_more_available to xmit_pending
 * add new patch #2 applying the same approach to ef100, for consistency

Edward Cree (6):
  sfc: add and use efx_tx_send_pending in tx.c
  sfc: make ef100 xmit_more handling look more like ef10's
  sfc: use tx_queue->old_read_count in EF100 TX path
  sfc: use efx_channel_tx_[old_]fill_level() in Siena/EF10 TX datapath
  sfc: rewrite efx_tx_may_pio
  sfc: remove efx_tx_queue_partner

 drivers/net/ethernet/sfc/ef10.c       |  2 +-
 drivers/net/ethernet/sfc/ef100_tx.c   | 34 +++++----
 drivers/net/ethernet/sfc/ef100_tx.h   |  1 -
 drivers/net/ethernet/sfc/farch.c      |  2 +-
 drivers/net/ethernet/sfc/net_driver.h | 22 ++++--
 drivers/net/ethernet/sfc/nic_common.h | 40 +----------
 drivers/net/ethernet/sfc/tx.c         | 99 +++++++++++++++++----------
 drivers/net/ethernet/sfc/tx_common.c  |  9 +--
 8 files changed, 104 insertions(+), 105 deletions(-)

Comments

Jakub Kicinski Sept. 5, 2020, 7:24 p.m. UTC | #1 
On Thu, 3 Sep 2020 22:30:58 +0100 Edward Cree wrote:
> Refactor and unify partner-TXQ handling in the EF100 and legacy drivers.
> 
> The main thrust of this series is to remove from the legacy (Siena/EF10)
>  driver the assumption that a netdev TX queue has precisely two hardware
>  TXQs (checksummed and unchecksummed) associated with it, so that in
>  future we can have more (e.g. for handling inner-header checksums) or
>  fewer (e.g. to free up hardware queues for XDP usage).
> 
> Changes from v1:
>  * better explain patch #1 in the commit message, and rename
>    xmit_more_available to xmit_pending
>  * add new patch #2 applying the same approach to ef100, for consistency

Applied, thanks!