Re: [RFC V2 net-next 1/3] ethtool: extend coalesce setting uAPI with CQE mode

On Mon, 31 May 2021 09:24:01 +0800 Huazhong Tan wrote:

quoted

quoted

@@ -975,6 +977,8 @@ Request contents:
    ``ETHTOOL_A_COALESCE_TX_USECS_HIGH``         u32     delay (us), high Tx
    ``ETHTOOL_A_COALESCE_TX_MAX_FRAMES_HIGH``    u32     max packets, high Tx
    ``ETHTOOL_A_COALESCE_RATE_SAMPLE_INTERVAL``  u32     rate sampling interval
+  ``ETHTOOL_A_COALESCE_USE_CQE_TX``	       bool    timer reset in CQE, Tx
+  ``ETHTOOL_A_COALESCE_USE_CQE_RX``	       bool    timer reset in CQE, Rx
    ===========================================  ======  =======================
  
  Request is rejected if it attributes declared as unsupported by driver (i.e.  

Did you provide the theory of operation for CQE vs EQE mode somewhere,
as I requested?  

the definition of enum dim_cq_period_mode in include/linux/dim.h has

below comment:

/**
  * enum dim_cq_period_mode - Modes for CQ period count
  *
  * @DIM_CQ_PERIOD_MODE_START_FROM_EQE: Start counting from EQE
  * @DIM_CQ_PERIOD_MODE_START_FROM_CQE: Start counting from CQE (implies 
timer reset)
  * @DIM_CQ_PERIOD_NUM_MODES: Number of modes
  */


is this comment suitable? and add reference in 
Documentation/networking/ethtool-netlink.rst to

the comment in dim.h.

DIM is kernel internals we need user facing, meaningful documentation.
I'm not 100% clea on what the exact difference is.

If the difference is whether timer is restarted on new packet arrival
or not - why mention CQE at all and not just call the configuration
knob "restart timer on new packet arrival"?