Commit 23be7075 authored by Jesse Brandeburg's avatar Jesse Brandeburg Committed by Tony Nguyen
Browse files

ice: fix software generating extra interrupts 



The driver tried to work around missing completion events that occurred
while interrupts are disabled, by triggering a software interrupt
whenever we exit polling (but we had to have polled at least once).

This was causing a *lot* of extra interrupts for some workloads like
NVMe over TCP, which resulted in regressions in performance. It was also
visible when polling didn't prevent interrupts when busy_poll was
enabled.

Fix the extra interrupts by utilizing our previously unused 3rd ITR
(interrupt throttle) index and set it to 20K interrupts per second, and
then trigger a software interrupt within that rate limit.

While here, slightly refactor the code to avoid an overwrite of a local
variable in the case of wb_en = true.

Fixes: b7306b42 ("ice: manage interrupts during poll exit")
Signed-off-by: default avatarJesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: default avatarGurucharan G <gurucharanx.g@intel.com>
Signed-off-by: default avatarTony Nguyen <anthony.l.nguyen@intel.com>
parent d16a4f45

drivers/net/ethernet/intel/ice/ice_hw_autogen.h

+1 −0
Original line number Diff line number Diff line
@@ -182,6 +182,7 @@ 
#define GLINT_DYN_CTL_INTERVAL_S		5

#define GLINT_DYN_CTL_INTERVAL_M		ICE_M(0xFFF, 5)

#define GLINT_DYN_CTL_SW_ITR_INDX_ENA_M		BIT(24)

#define GLINT_DYN_CTL_SW_ITR_INDX_S		25

#define GLINT_DYN_CTL_SW_ITR_INDX_M		ICE_M(0x3, 25)

#define GLINT_DYN_CTL_WB_ON_ITR_M		BIT(30)

#define GLINT_DYN_CTL_INTENA_MSK_M		BIT(31)

drivers/net/ethernet/intel/ice/ice_txrx.c

+15 −11
Original line number Diff line number Diff line
@@ -1361,21 +1361,25 @@ static void ice_enable_interrupt(struct ice_q_vector *q_vector) 
	if (test_bit(ICE_DOWN, vsi->state))

		return;



	/* When exiting WB_ON_ITR, let ITR resume its normal

	 * interrupts-enabled path.

	/* trigger an ITR delayed software interrupt when exiting busy poll, to

	 * make sure to catch any pending cleanups that might have been missed

	 * due to interrupt state transition. If busy poll or poll isn't

	 * enabled, then don't update ITR, and just enable the interrupt.

	 */

	if (wb_en)

	if (!wb_en) {

		itr_val = ice_buildreg_itr(ICE_ITR_NONE, 0);

	} else {

		q_vector->wb_on_itr = false;



	itr_val = ice_buildreg_itr(ICE_ITR_NONE, 0);

	/* trigger an immediate software interrupt when exiting

	 * busy poll, to make sure to catch any pending cleanups

	 * that might have been missed due to interrupt state

	 * transition.

		/* do two things here with a single write. Set up the third ITR

		 * index to be used for software interrupt moderation, and then

		 * trigger a software interrupt with a rate limit of 20K on

		 * software interrupts, this will help avoid high interrupt

		 * loads due to frequently polling and exiting polling.

		 */

	if (wb_en) {

		itr_val = ice_buildreg_itr(ICE_IDX_ITR2, ICE_ITR_20K);

		itr_val |= GLINT_DYN_CTL_SWINT_TRIG_M |

			   GLINT_DYN_CTL_SW_ITR_INDX_M |

			   ICE_IDX_ITR2 << GLINT_DYN_CTL_SW_ITR_INDX_S |

			   GLINT_DYN_CTL_SW_ITR_INDX_ENA_M;

	}

	wr32(&vsi->back->hw, GLINT_DYN_CTL(q_vector->reg_idx), itr_val);