[dpdk-dev] [PATCH v6 10/19] net/ngbe: support link update

[Andrew Rybchenko]

On 6/17/21 1:59 PM, Jiawen Wu wrote:
> Register to handle device interrupt.
> 
> Signed-off-by: Jiawen Wu <jiawenwu at trustnetic.com>

[snip]

> diff --git a/doc/guides/nics/ngbe.rst b/doc/guides/nics/ngbe.rst
> index 54d0665db9..0918cc2918 100644
> --- a/doc/guides/nics/ngbe.rst
> +++ b/doc/guides/nics/ngbe.rst
> @@ -8,6 +8,11 @@ The NGBE PMD (librte_pmd_ngbe) provides poll mode driver support
>  for Wangxun 1 Gigabit Ethernet NICs.
>  
>  
> +Features
> +--------
> +
> +- Link state information
> +

Two empty lines before the section.

>  Prerequisites
>  -------------
>  

[snip]

> diff --git a/drivers/net/ngbe/ngbe_ethdev.c b/drivers/net/ngbe/ngbe_ethdev.c
> index deca64137d..c952023e8b 100644
> --- a/drivers/net/ngbe/ngbe_ethdev.c
> +++ b/drivers/net/ngbe/ngbe_ethdev.c
> @@ -141,6 +175,23 @@ eth_ngbe_dev_init(struct rte_eth_dev *eth_dev, void *init_params __rte_unused)
>  		return -ENOMEM;
>  	}
>  
> +	ctrl_ext = rd32(hw, NGBE_PORTCTL);
> +	/* let hardware know driver is loaded */
> +	ctrl_ext |= NGBE_PORTCTL_DRVLOAD;
> +	/* Set PF Reset Done bit so PF/VF Mail Ops can work */
> +	ctrl_ext |= NGBE_PORTCTL_RSTDONE;
> +	wr32(hw, NGBE_PORTCTL, ctrl_ext);
> +	ngbe_flush(hw);
> +
> +	rte_intr_callback_register(intr_handle,
> +				   ngbe_dev_interrupt_handler, eth_dev);
> +
> +	/* enable uio/vfio intr/eventfd mapping */
> +	rte_intr_enable(intr_handle);
> +
> +	/* enable support intr */
> +	ngbe_enable_intr(eth_dev);
> +

I don't understand why it is done unconditionally regardless
of the corresponding bit in dev_conf.

>  	return 0;
>  }
>  
> @@ -180,11 +231,25 @@ static int eth_ngbe_pci_remove(struct rte_pci_device *pci_dev)
>  
>  static struct rte_pci_driver rte_ngbe_pmd = {
>  	.id_table = pci_id_ngbe_map,
> -	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
> +	.drv_flags = RTE_PCI_DRV_NEED_MAPPING |
> +		     RTE_PCI_DRV_INTR_LSC,
>  	.probe = eth_ngbe_pci_probe,
>  	.remove = eth_ngbe_pci_remove,
>  };
>  
> +static int
> +ngbe_dev_configure(struct rte_eth_dev *dev)
> +{
> +	struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
> +
> +	PMD_INIT_FUNC_TRACE();
> +
> +	/* set flag to update link status after init */
> +	intr->flags |= NGBE_FLAG_NEED_LINK_UPDATE;
> +
> +	return 0;
> +}
> +
>  /*
>   * Reset and stop device.
>   */
> @@ -198,6 +263,315 @@ ngbe_dev_close(struct rte_eth_dev *dev)
>  	return -EINVAL;
>  }
>  
> +static int
> +ngbe_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
> +{
> +	RTE_SET_USED(dev);
> +
> +	dev_info->speed_capa = ETH_LINK_SPEED_1G | ETH_LINK_SPEED_100M |
> +				ETH_LINK_SPEED_10M;
> +
> +	return 0;
> +}
> +
> +/* return 0 means link status changed, -1 means not changed */
> +int
> +ngbe_dev_link_update_share(struct rte_eth_dev *dev,
> +			    int wait_to_complete)
> +{
> +	struct ngbe_hw *hw = ngbe_dev_hw(dev);
> +	struct rte_eth_link link;
> +	u32 link_speed = NGBE_LINK_SPEED_UNKNOWN;
> +	u32 lan_speed = 0;
> +	struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
> +	bool link_up;
> +	int err;
> +	int wait = 1;
> +
> +	memset(&link, 0, sizeof(link));
> +	link.link_status = ETH_LINK_DOWN;
> +	link.link_speed = ETH_SPEED_NUM_NONE;
> +	link.link_duplex = ETH_LINK_HALF_DUPLEX;
> +	link.link_autoneg = !(dev->data->dev_conf.link_speeds &
> +			~ETH_LINK_SPEED_AUTONEG);
> +
> +	hw->mac.get_link_status = true;
> +
> +	if (intr->flags & NGBE_FLAG_NEED_LINK_CONFIG)
> +		return rte_eth_linkstatus_set(dev, &link);
> +
> +	/* check if it needs to wait to complete, if lsc interrupt is enabled */
> +	if (wait_to_complete == 0 || dev->data->dev_conf.intr_conf.lsc != 0)
> +		wait = 0;
> +
> +	err = hw->mac.check_link(hw, &link_speed, &link_up, wait);
> +

Please, remove empty line which adds unnecessary logical
separation of the operation and its result check.

> +	if (err != 0) {
> +		link.link_speed = ETH_SPEED_NUM_100M;
> +		link.link_duplex = ETH_LINK_FULL_DUPLEX;
> +		return rte_eth_linkstatus_set(dev, &link);
> +	}
> +
> +	if (!link_up)
> +		return rte_eth_linkstatus_set(dev, &link);
> +
> +	intr->flags &= ~NGBE_FLAG_NEED_LINK_CONFIG;
> +	link.link_status = ETH_LINK_UP;
> +	link.link_duplex = ETH_LINK_FULL_DUPLEX;
> +
> +	switch (link_speed) {
> +	default:
> +	case NGBE_LINK_SPEED_UNKNOWN:
> +		link.link_duplex = ETH_LINK_FULL_DUPLEX;
> +		link.link_speed = ETH_SPEED_NUM_100M;

May be ETH_SPEED_NUM_NONE?

> +		break;
> +
> +	case NGBE_LINK_SPEED_10M_FULL:
> +		link.link_speed = ETH_SPEED_NUM_10M;
> +		lan_speed = 0;
> +		break;
> +
> +	case NGBE_LINK_SPEED_100M_FULL:
> +		link.link_speed = ETH_SPEED_NUM_100M;
> +		lan_speed = 1;
> +		break;
> +
> +	case NGBE_LINK_SPEED_1GB_FULL:
> +		link.link_speed = ETH_SPEED_NUM_1G;
> +		lan_speed = 2;
> +		break;
> +	}
> +
> +	if (hw->is_pf) {
> +		wr32m(hw, NGBE_LAN_SPEED, NGBE_LAN_SPEED_MASK, lan_speed);
> +		if (link_speed & (NGBE_LINK_SPEED_1GB_FULL |
> +			NGBE_LINK_SPEED_100M_FULL | NGBE_LINK_SPEED_10M_FULL)) {

Such indent is very confusing since it is the same as on the
next line. Please, add extra TAB to indent above line further.

> +			wr32m(hw, NGBE_MACTXCFG, NGBE_MACTXCFG_SPEED_MASK,
> +				NGBE_MACTXCFG_SPEED_1G | NGBE_MACTXCFG_TE);
> +		}
> +	}
> +
> +	return rte_eth_linkstatus_set(dev, &link);
> +}
> +
> +static int
> +ngbe_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
> +{
> +	return ngbe_dev_link_update_share(dev, wait_to_complete);
> +}
> +
> +/*
> + * It reads ICR and sets flag for the link_update.
> + *
> + * @param dev
> + *  Pointer to struct rte_eth_dev.
> + *
> + * @return
> + *  - On success, zero.
> + *  - On failure, a negative value.
> + */
> +static int
> +ngbe_dev_interrupt_get_status(struct rte_eth_dev *dev)
> +{
> +	uint32_t eicr;
> +	struct ngbe_hw *hw = ngbe_dev_hw(dev);
> +	struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
> +
> +	/* clear all cause mask */
> +	ngbe_disable_intr(hw);
> +
> +	/* read-on-clear nic registers here */
> +	eicr = ((u32 *)hw->isb_mem)[NGBE_ISB_MISC];
> +	PMD_DRV_LOG(DEBUG, "eicr %x", eicr);
> +
> +	intr->flags = 0;
> +
> +	/* set flag for async link update */
> +	if (eicr & NGBE_ICRMISC_PHY)
> +		intr->flags |= NGBE_FLAG_NEED_LINK_UPDATE;
> +
> +	if (eicr & NGBE_ICRMISC_VFMBX)
> +		intr->flags |= NGBE_FLAG_MAILBOX;
> +
> +	if (eicr & NGBE_ICRMISC_LNKSEC)
> +		intr->flags |= NGBE_FLAG_MACSEC;
> +
> +	if (eicr & NGBE_ICRMISC_GPIO)
> +		intr->flags |= NGBE_FLAG_NEED_LINK_UPDATE;
> +
> +	return 0;
> +}
> +
> +/**
> + * It gets and then prints the link status.
> + *
> + * @param dev
> + *  Pointer to struct rte_eth_dev.
> + *
> + * @return
> + *  - On success, zero.
> + *  - On failure, a negative value.
> + */
> +static void
> +ngbe_dev_link_status_print(struct rte_eth_dev *dev)
> +{
> +	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
> +	struct rte_eth_link link;
> +
> +	rte_eth_linkstatus_get(dev, &link);
> +
> +	if (link.link_status == ETH_LINK_UP) {
> +		PMD_INIT_LOG(INFO, "Port %d: Link Up - speed %u Mbps - %s",
> +					(int)(dev->data->port_id),
> +					(unsigned int)link.link_speed,
> +			link.link_duplex == ETH_LINK_FULL_DUPLEX ?
> +					"full-duplex" : "half-duplex");
> +	} else {
> +		PMD_INIT_LOG(INFO, " Port %d: Link Down",
> +				(int)(dev->data->port_id));
> +	}
> +	PMD_INIT_LOG(DEBUG, "PCI Address: " PCI_PRI_FMT,
> +				pci_dev->addr.domain,
> +				pci_dev->addr.bus,
> +				pci_dev->addr.devid,
> +				pci_dev->addr.function);
> +}
> +
> +/*
> + * It executes link_update after knowing an interrupt occurred.
> + *
> + * @param dev
> + *  Pointer to struct rte_eth_dev.
> + *
> + * @return
> + *  - On success, zero.
> + *  - On failure, a negative value.
> + */
> +static int
> +ngbe_dev_interrupt_action(struct rte_eth_dev *dev)
> +{
> +	struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
> +	int64_t timeout;
> +
> +	PMD_DRV_LOG(DEBUG, "intr action type %d", intr->flags);
> +
> +	if (intr->flags & NGBE_FLAG_NEED_LINK_UPDATE) {
> +		struct rte_eth_link link;
> +
> +		/*get the link status before link update, for predicting later*/
> +		rte_eth_linkstatus_get(dev, &link);
> +
> +		ngbe_dev_link_update(dev, 0);
> +
> +		/* likely to up */
> +		if (link.link_status != ETH_LINK_UP)
> +			/* handle it 1 sec later, wait it being stable */
> +			timeout = NGBE_LINK_UP_CHECK_TIMEOUT;
> +		/* likely to down */
> +		else
> +			/* handle it 4 sec later, wait it being stable */
> +			timeout = NGBE_LINK_DOWN_CHECK_TIMEOUT;
> +
> +		ngbe_dev_link_status_print(dev);
> +		if (rte_eal_alarm_set(timeout * 1000,
> +				      ngbe_dev_interrupt_delayed_handler,
> +				      (void *)dev) < 0) {
> +			PMD_DRV_LOG(ERR, "Error setting alarm");
> +		} else {
> +			/* remember original mask */
> +			intr->mask_misc_orig = intr->mask_misc;
> +			/* only disable lsc interrupt */
> +			intr->mask_misc &= ~NGBE_ICRMISC_PHY;
> +
> +			intr->mask_orig = intr->mask;
> +			/* only disable all misc interrupts */
> +			intr->mask &= ~(1ULL << NGBE_MISC_VEC_ID);
> +		}
> +	}
> +
> +	PMD_DRV_LOG(DEBUG, "enable intr immediately");
> +	ngbe_enable_intr(dev);
> +
> +	return 0;
> +}
> +
> +/**
> + * Interrupt handler which shall be registered for alarm callback for delayed
> + * handling specific interrupt to wait for the stable nic state. As the
> + * NIC interrupt state is not stable for ngbe after link is just down,
> + * it needs to wait 4 seconds to get the stable status.
> + *
> + * @param handle
> + *  Pointer to interrupt handle.
> + * @param param
> + *  The address of parameter (struct rte_eth_dev *) registered before.
> + *
> + * @return
> + *  void

Such documentation of the void functions is useless.
It may be simply omited.

> + */
> +static void
> +ngbe_dev_interrupt_delayed_handler(void *param)
> +{
> +	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
> +	struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
> +	struct ngbe_hw *hw = ngbe_dev_hw(dev);
> +	uint32_t eicr;
> +
> +	ngbe_disable_intr(hw);
> +
> +	eicr = ((u32 *)hw->isb_mem)[NGBE_ISB_MISC];
> +
> +	if (intr->flags & NGBE_FLAG_NEED_LINK_UPDATE) {
> +		ngbe_dev_link_update(dev, 0);
> +		intr->flags &= ~NGBE_FLAG_NEED_LINK_UPDATE;
> +		ngbe_dev_link_status_print(dev);
> +		rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC,
> +					      NULL);
> +	}
> +
> +	if (intr->flags & NGBE_FLAG_MACSEC) {
> +		rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_MACSEC,
> +					      NULL);
> +		intr->flags &= ~NGBE_FLAG_MACSEC;
> +	}
> +
> +	/* restore original mask */
> +	intr->mask_misc = intr->mask_misc_orig;
> +	intr->mask_misc_orig = 0;
> +	intr->mask = intr->mask_orig;
> +	intr->mask_orig = 0;
> +
> +	PMD_DRV_LOG(DEBUG, "enable intr in delayed handler S[%08x]", eicr);
> +	ngbe_enable_intr(dev);
> +}
> +
> +/**
> + * Interrupt handler triggered by NIC  for handling
> + * specific interrupt.
> + *
> + * @param handle
> + *  Pointer to interrupt handle.
> + * @param param
> + *  The address of parameter (struct rte_eth_dev *) registered before.
> + *
> + * @return
> + *  void

Such documentation of the void functions is useless.
It may be simply omited.

> + */
> +static void
> +ngbe_dev_interrupt_handler(void *param)
> +{
> +	struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
> +
> +	ngbe_dev_interrupt_get_status(dev);
> +	ngbe_dev_interrupt_action(dev);
> +}
> +
> +static const struct eth_dev_ops ngbe_eth_dev_ops = {
> +	.dev_configure              = ngbe_dev_configure,
> +	.dev_infos_get              = ngbe_dev_info_get,
> +	.link_update                = ngbe_dev_link_update,
> +};
> +
>  RTE_PMD_REGISTER_PCI(net_ngbe, rte_ngbe_pmd);
>  RTE_PMD_REGISTER_PCI_TABLE(net_ngbe, pci_id_ngbe_map);
>  RTE_PMD_REGISTER_KMOD_DEP(net_ngbe, "* igb_uio | uio_pci_generic | vfio-pci");
> diff --git a/drivers/net/ngbe/ngbe_ethdev.h b/drivers/net/ngbe/ngbe_ethdev.h
> index 87cc1cff6b..b67508a3de 100644
> --- a/drivers/net/ngbe/ngbe_ethdev.h
> +++ b/drivers/net/ngbe/ngbe_ethdev.h
> @@ -6,11 +6,30 @@
>  #ifndef _NGBE_ETHDEV_H_
>  #define _NGBE_ETHDEV_H_
>  
> +/* need update link, bit flag */
> +#define NGBE_FLAG_NEED_LINK_UPDATE (uint32_t)(1 << 0)
> +#define NGBE_FLAG_MAILBOX          (uint32_t)(1 << 1)
> +#define NGBE_FLAG_PHY_INTERRUPT    (uint32_t)(1 << 2)
> +#define NGBE_FLAG_MACSEC           (uint32_t)(1 << 3)
> +#define NGBE_FLAG_NEED_LINK_CONFIG (uint32_t)(1 << 4)
> +
> +#define NGBE_MISC_VEC_ID               RTE_INTR_VEC_ZERO_OFFSET
> +
> +/* structure for interrupt relative data */
> +struct ngbe_interrupt {
> +	uint32_t flags;
> +	uint32_t mask_misc;
> +	uint32_t mask_misc_orig; /* save mask during delayed handler */
> +	uint64_t mask;
> +	uint64_t mask_orig; /* save mask during delayed handler */
> +};
> +
>  /*
>   * Structure to store private data for each driver instance (for each port).
>   */
>  struct ngbe_adapter {
>  	struct ngbe_hw             hw;
> +	struct ngbe_interrupt      intr;
>  };
>  
>  static inline struct ngbe_adapter *
> @@ -30,6 +49,21 @@ ngbe_dev_hw(struct rte_eth_dev *dev)
>  	return hw;
>  }
>  
> +static inline struct ngbe_interrupt *
> +ngbe_dev_intr(struct rte_eth_dev *dev)
> +{
> +	struct ngbe_adapter *ad = ngbe_dev_adapter(dev);
> +	struct ngbe_interrupt *intr = &ad->intr;
> +
> +	return intr;
> +}
> +
> +int
> +ngbe_dev_link_update_share(struct rte_eth_dev *dev,
> +		int wait_to_complete);
> +
> +#define NGBE_LINK_DOWN_CHECK_TIMEOUT 4000 /* ms */
> +#define NGBE_LINK_UP_CHECK_TIMEOUT   1000 /* ms */
>  #define NGBE_VMDQ_NUM_UC_MAC         4096 /* Maximum nb. of UC MAC addr. */
>  
>  #endif /* _NGBE_ETHDEV_H_ */
>