Re: [dpdk-dev] [PATCH v2] net/i40e: remove VF interrupt handler

[Stephen Hemminger <stephen@networkplumber.org>]

On Fri, 22 Jun 2018 08:44:14 +0800
Qi Zhang <qi.z.zhang@intel.com> wrote:

> For i40evf, internal rx interrupt and adminq interrupt share the same
> source, that cause a lot cpu cycles be wasted on interrupt handler
> on rx path. This is complained by customers which require low latency
> (when set I40E_ITR_INTERVAL to small value), but have to be sufferred by
> tremendous interrupts handling that eat significant CPU resources.
> 
> The patch disable pci interrupt and remove the interrupt handler,
> replace it with a low frequency (50ms) interrupt polling daemon
> which is implemented by registering a alarm callback periodly, this
> save CPU time significently: On a typical x86 server with 2.1GHz CPU,
> with low latency configure (32us) we saw CPU usage from top commmand
> reduced from 20% to 0% on management core in testpmd).
> 
> Also with the new method we can remove compile option: I40E_ITR_INTERVAL
> which is used to balance between low latency and low CPU usage previously.
> Now we don't need it since we can reach both at same time.
> 
> Suggested-by: Jingjing Wu <jingjing.wu@intel.com>
> Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
> ---
> 
> v2:
> - update doc
> 
>  config/common_base                |  2 --
>  doc/guides/nics/i40e.rst          |  5 -----
>  drivers/net/i40e/i40e_ethdev.c    |  3 +--
>  drivers/net/i40e/i40e_ethdev.h    | 22 +++++++++++-----------
>  drivers/net/i40e/i40e_ethdev_vf.c | 36 ++++++++++++++----------------------
>  5 files changed, 26 insertions(+), 42 deletions(-)
> 
> diff --git a/config/common_base b/config/common_base
> index 6b0d1cbbb..9e21c6865 100644
> --- a/config/common_base
> +++ b/config/common_base
> @@ -264,8 +264,6 @@ CONFIG_RTE_LIBRTE_I40E_INC_VECTOR=y
>  CONFIG_RTE_LIBRTE_I40E_16BYTE_RX_DESC=n
>  CONFIG_RTE_LIBRTE_I40E_QUEUE_NUM_PER_PF=64
>  CONFIG_RTE_LIBRTE_I40E_QUEUE_NUM_PER_VM=4
> -# interval up to 8160 us, aligned to 2 (or default value)
> -CONFIG_RTE_LIBRTE_I40E_ITR_INTERVAL=-1
>  
>  #
>  # Compile burst-oriented FM10K PMD
> diff --git a/doc/guides/nics/i40e.rst b/doc/guides/nics/i40e.rst
> index 18549bf5a..3fc4ceac7 100644
> --- a/doc/guides/nics/i40e.rst
> +++ b/doc/guides/nics/i40e.rst
> @@ -96,11 +96,6 @@ Please note that enabling debugging options may affect system performance.
>  
>    Number of queues reserved for each VMDQ Pool.
>  
> -- ``CONFIG_RTE_LIBRTE_I40E_ITR_INTERVAL`` (default ``-1``)
> -
> -  Interrupt Throttling interval.
> -
> -
>  Runtime Config Options
>  ~~~~~~~~~~~~~~~~~~~~~~
>  
> diff --git a/drivers/net/i40e/i40e_ethdev.c b/drivers/net/i40e/i40e_ethdev.c
> index 13c5d3296..c8f9566e0 100644
> --- a/drivers/net/i40e/i40e_ethdev.c
> +++ b/drivers/net/i40e/i40e_ethdev.c
> @@ -1829,8 +1829,7 @@ __vsi_queues_bind_intr(struct i40e_vsi *vsi, uint16_t msix_vect,
>  	/* Write first RX queue to Link list register as the head element */
>  	if (vsi->type != I40E_VSI_SRIOV) {
>  		uint16_t interval =
> -			i40e_calc_itr_interval(RTE_LIBRTE_I40E_ITR_INTERVAL, 1,
> -					       pf->support_multi_driver);
> +			i40e_calc_itr_interval(1, pf->support_multi_driver);
>  
>  		if (msix_vect == I40E_MISC_VEC_ID) {
>  			I40E_WRITE_REG(hw, I40E_PFINT_LNKLST0,
> diff --git a/drivers/net/i40e/i40e_ethdev.h b/drivers/net/i40e/i40e_ethdev.h
> index 11c4c76bd..599993dac 100644
> --- a/drivers/net/i40e/i40e_ethdev.h
> +++ b/drivers/net/i40e/i40e_ethdev.h
> @@ -178,7 +178,7 @@ enum i40e_flxpld_layer_idx {
>  #define I40E_ITR_INDEX_NONE             3
>  #define I40E_QUEUE_ITR_INTERVAL_DEFAULT 32 /* 32 us */
>  #define I40E_QUEUE_ITR_INTERVAL_MAX     8160 /* 8160 us */
> -#define I40E_VF_QUEUE_ITR_INTERVAL_DEFAULT 8160 /* 8160 us */
> +#define I40E_VF_QUEUE_ITR_INTERVAL_DEFAULT 32 /* 32 us */
>  /* Special FW support this floating VEB feature */
>  #define FLOATING_VEB_SUPPORTED_FW_MAJ 5
>  #define FLOATING_VEB_SUPPORTED_FW_MIN 0
> @@ -1328,17 +1328,17 @@ i40e_align_floor(int n)
>  }
>  
>  static inline uint16_t
> -i40e_calc_itr_interval(int16_t interval, bool is_pf, bool is_multi_drv)
> +i40e_calc_itr_interval(bool is_pf, bool is_multi_drv)
>  {
> -	if (interval < 0 || interval > I40E_QUEUE_ITR_INTERVAL_MAX) {
> -		if (is_multi_drv) {
> -			interval = I40E_QUEUE_ITR_INTERVAL_MAX;
> -		} else {
> -			if (is_pf)
> -				interval = I40E_QUEUE_ITR_INTERVAL_DEFAULT;
> -			else
> -				interval = I40E_VF_QUEUE_ITR_INTERVAL_DEFAULT;
> -		}
> +	uint16_t interval = 0;
> +
> +	if (is_multi_drv) {
> +		interval = I40E_QUEUE_ITR_INTERVAL_MAX;
> +	} else {
> +		if (is_pf)
> +			interval = I40E_QUEUE_ITR_INTERVAL_DEFAULT;
> +		else
> +			interval = I40E_VF_QUEUE_ITR_INTERVAL_DEFAULT;
>  	}
>  
>  	/* Convert to hardware count, as writing each 1 represents 2 us */
> diff --git a/drivers/net/i40e/i40e_ethdev_vf.c b/drivers/net/i40e/i40e_ethdev_vf.c
> index 804e44530..ad5c069e8 100644
> --- a/drivers/net/i40e/i40e_ethdev_vf.c
> +++ b/drivers/net/i40e/i40e_ethdev_vf.c
> @@ -44,6 +44,8 @@
>  #define I40EVF_BUSY_WAIT_COUNT 50
>  #define MAX_RESET_WAIT_CNT     20
>  
> +#define I40EVF_ALARM_INTERVAL 50000 /* us */
> +
>  struct i40evf_arq_msg_info {
>  	enum virtchnl_ops ops;
>  	enum i40e_status_code result;
> @@ -1133,7 +1135,7 @@ i40evf_init_vf(struct rte_eth_dev *dev)
>  	struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
>  	struct i40e_vf *vf = I40EVF_DEV_PRIVATE_TO_VF(dev->data->dev_private);
>  	uint16_t interval =
> -		i40e_calc_itr_interval(RTE_LIBRTE_I40E_ITR_INTERVAL, 0, 0);
> +		i40e_calc_itr_interval(0, 0);
>  
>  	vf->adapter = I40E_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
>  	vf->dev_data = dev->data;
> @@ -1370,7 +1372,7 @@ i40evf_handle_aq_msg(struct rte_eth_dev *dev)
>   *  void
>   */
>  static void
> -i40evf_dev_interrupt_handler(void *param)
> +i40evf_dev_alarm_handler(void *param)
>  {
>  	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
>  	struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
> @@ -1399,6 +1401,8 @@ i40evf_dev_interrupt_handler(void *param)
>  
>  done:
>  	i40evf_enable_irq0(hw);
> +	rte_eal_alarm_set(I40EVF_ALARM_INTERVAL,
> +			  i40evf_dev_alarm_handler, dev);
>  }
>  
>  static int
> @@ -1442,12 +1446,8 @@ i40evf_dev_init(struct rte_eth_dev *eth_dev)
>  		return -1;
>  	}
>  
> -	/* register callback func to eal lib */
> -	rte_intr_callback_register(&pci_dev->intr_handle,
> -		i40evf_dev_interrupt_handler, (void *)eth_dev);
> -
> -	/* enable uio intr after callback register */
> -	rte_intr_enable(&pci_dev->intr_handle);
> +	rte_eal_alarm_set(I40EVF_ALARM_INTERVAL,
> +			  i40evf_dev_alarm_handler, eth_dev);
>  
>  	/* configure and enable device interrupt */
>  	i40evf_enable_irq0(hw);
> @@ -1836,7 +1836,7 @@ i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
>  	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
>  	struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
>  	uint16_t interval =
> -		i40e_calc_itr_interval(RTE_LIBRTE_I40E_ITR_INTERVAL, 0, 0);
> +		i40e_calc_itr_interval(0, 0);
>  	uint16_t msix_intr;
>  
>  	msix_intr = intr_handle->intr_vec[queue_id];
> @@ -1859,8 +1859,6 @@ i40evf_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
>  
>  	I40EVF_WRITE_FLUSH(hw);
>  
> -	rte_intr_enable(&pci_dev->intr_handle);
> -
>  	return 0;
>  }
>  
> @@ -2023,10 +2021,8 @@ i40evf_dev_start(struct rte_eth_dev *dev)
>  	 * queue interrupt to other VFIO vectors.
>  	 * So clear uio/vfio intr/evevnfd first to avoid failure.
>  	 */
> -	if (dev->data->dev_conf.intr_conf.rxq != 0) {
> -		rte_intr_disable(intr_handle);
> +	if (dev->data->dev_conf.intr_conf.rxq != 0)
>  		rte_intr_enable(intr_handle);
> -	}
>  
>  	i40evf_enable_queues_intr(dev);
>  
> @@ -2050,6 +2046,9 @@ i40evf_dev_stop(struct rte_eth_dev *dev)
>  
>  	PMD_INIT_FUNC_TRACE();
>  
> +	if (dev->data->dev_conf.intr_conf.rxq != 0)
> +		rte_intr_disable(intr_handle);
> +
>  	if (hw->adapter_stopped == 1)
>  		return;
>  	i40evf_stop_queues(dev);
> @@ -2285,9 +2284,8 @@ static void
>  i40evf_dev_close(struct rte_eth_dev *dev)
>  {
>  	struct i40e_hw *hw = I40E_DEV_PRIVATE_TO_HW(dev->data->dev_private);
> -	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
> -	struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
>  
> +	rte_eal_alarm_cancel(i40evf_dev_alarm_handler, dev);
>  	i40evf_dev_stop(dev);
>  	i40e_dev_free_queues(dev);
>  	/*
> @@ -2300,12 +2298,6 @@ i40evf_dev_close(struct rte_eth_dev *dev)
>  
>  	i40evf_reset_vf(hw);
>  	i40e_shutdown_adminq(hw);
> -	/* disable uio intr before callback unregister */
> -	rte_intr_disable(intr_handle);
> -
> -	/* unregister callback func from eal lib */
> -	rte_intr_callback_unregister(intr_handle,
> -				     i40evf_dev_interrupt_handler, dev);
>  	i40evf_disable_irq0(hw);
>  }
>  

Rather than adding a polling routine internally, why not change the driver to not support
Link State or receive interrupts. Better yet, let the application decide.
Keep the interrupt logic but only enable interrupts if application has requested
LSC or recveive interrupt mode.