[PATCH v2] net/intel: add IDPF PCI class ID support

[Bruce Richardson]

On Wed, Sep 24, 2025 at 06:27:49PM +0530, Vemula Venkatesh wrote:
> Current IDPF supports only the MEV device IDs. MMG has new set of
> device IDs and same might be the case for the future devices. Instead
> of adding new device IDs every time, make use of the IDPF PCI class
> ID(0x20001) to differentiate between PF and VF.
> 
> Write and read the VF_ARQBAL register to find if the current device is
> a PF or a VF.
> 
> v2: addressed reviewers comments.
> 
> Signed-off-by: Vemula Venkatesh <venkatesh.vemula at intel.com>
> ---
>  drivers/net/intel/cpfl/cpfl_ethdev.c          |  3 +-
>  drivers/net/intel/cpfl/cpfl_ethdev.h          |  3 +-
>  .../net/intel/idpf/base/idpf_controlq_api.h   |  1 +
>  drivers/net/intel/idpf/idpf_common_device.c   | 37 +++++++++++++++++--
>  drivers/net/intel/idpf/idpf_common_device.h   | 17 +++++++++
>  drivers/net/intel/idpf/idpf_ethdev.c          |  2 +
>  6 files changed, 57 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/net/intel/cpfl/cpfl_ethdev.c b/drivers/net/intel/cpfl/cpfl_ethdev.c
> index 6d7b23ad7a..8cec977ecf 100644
> --- a/drivers/net/intel/cpfl/cpfl_ethdev.c
> +++ b/drivers/net/intel/cpfl/cpfl_ethdev.c
> @@ -2606,7 +2606,8 @@ cpfl_dev_vport_init(struct rte_eth_dev *dev, void *init_params)
>  }
>  
>  static const struct rte_pci_id pci_id_cpfl_map[] = {
> -	{ RTE_PCI_DEVICE(IDPF_INTEL_VENDOR_ID, IDPF_DEV_ID_CPF) },
> +	{ RTE_PCI_DEVICE(IDPF_INTEL_VENDOR_ID, CPFL_DEV_ID_MEV) },
> +	{ RTE_PCI_DEVICE(IDPF_INTEL_VENDOR_ID, CPFL_DEV_ID_MMG) },
>  	{ .vendor_id = 0, /* sentinel */ },
>  };
>  
> diff --git a/drivers/net/intel/cpfl/cpfl_ethdev.h b/drivers/net/intel/cpfl/cpfl_ethdev.h
> index d4e1176ab1..566b395dae 100644
> --- a/drivers/net/intel/cpfl/cpfl_ethdev.h
> +++ b/drivers/net/intel/cpfl/cpfl_ethdev.h
> @@ -59,7 +59,8 @@
>  #define CPFL_ALARM_INTERVAL	50000 /* us */
>  
>  /* Device IDs */
> -#define IDPF_DEV_ID_CPF			0x1453
> +#define CPFL_DEV_ID_MMG			0x11E0
> +#define CPFL_DEV_ID_MEV			0x1453
>  #define VIRTCHNL2_QUEUE_GROUP_P2P	0x100
>  
>  #define CPFL_HOST_ID_NUM	2
> diff --git a/drivers/net/intel/idpf/base/idpf_controlq_api.h b/drivers/net/intel/idpf/base/idpf_controlq_api.h
> index 8a90258099..7c25aabb28 100644
> --- a/drivers/net/intel/idpf/base/idpf_controlq_api.h
> +++ b/drivers/net/intel/idpf/base/idpf_controlq_api.h
> @@ -184,6 +184,7 @@ struct idpf_hw {
>  	u16 subsystem_device_id;
>  	u16 subsystem_vendor_id;
>  	u8 revision_id;
> +	uint32_t cls_id;
>  	bool adapter_stopped;
>  

This is a poor location to add a new 32-bit field, right between two 8-bit
fields. Adding it here will likely cause more padding than necessary to be
added. Add the new field alongside other 32-bit fields to minimize padding.

>  	LIST_HEAD_TYPE(list_head, idpf_ctlq_info) cq_list_head;
> diff --git a/drivers/net/intel/idpf/idpf_common_device.c b/drivers/net/intel/idpf/idpf_common_device.c
> index ff1fbcd2b4..6f083770d8 100644
> --- a/drivers/net/intel/idpf/idpf_common_device.c
> +++ b/drivers/net/intel/idpf/idpf_common_device.c
> @@ -125,12 +125,12 @@ struct idpf_ctlq_create_info vf_ctlq_info[IDPF_CTLQ_NUM] = {
>  };
>  
>  static int
> -idpf_init_mbx(struct idpf_hw *hw)
> +idpf_init_mbx(struct idpf_hw *hw, bool is_vf)
>  {
>  	struct idpf_ctlq_info *ctlq;
>  	int ret = 0;
>  
> -	if (hw->device_id == IDPF_DEV_ID_SRIOV)
> +	if (is_vf == 1)
>  		ret = idpf_ctlq_init(hw, IDPF_CTLQ_NUM, vf_ctlq_info);
>  	else
>  		ret = idpf_ctlq_init(hw, IDPF_CTLQ_NUM, pf_ctlq_info);
> @@ -388,8 +388,21 @@ idpf_adapter_init(struct idpf_adapter *adapter)
>  {
>  	struct idpf_hw *hw = &adapter->hw;
>  	int ret;
> +	int err;
> +	bool is_vf = 0;
>  
> -	if (hw->device_id == IDPF_DEV_ID_SRIOV) {
> +	switch (hw->device_id) {
> +	case IDPF_DEV_ID_SRIOV:
> +			is_vf = 1;
> +			break;
> +	default:
> +			if (hw->cls_id == IDPF_CLASS_NETWORK_ETHERNET_PROGIF) {
> +				err = idpf_is_vf_device(hw, &is_vf);
> +				if (err)
> +					return err;
> +			}
> +	}

There is no need for a switch statement here, I think. Simplify the code by
just adding a "(hw->device_id == IDPF_DEV_ID_SRIOV)" check at the start of
idpf_is_vf_device function, and use that in all cases.

> +	if (is_vf) {
>  		ret = idpf_check_vf_reset_done(hw);
>  	} else {
>  		idpf_reset_pf(hw);
> @@ -400,7 +413,7 @@ idpf_adapter_init(struct idpf_adapter *adapter)
>  		goto err_check_reset;
>  	}
>  
> -	ret = idpf_init_mbx(hw);
> +	ret = idpf_init_mbx(hw, is_vf);
>  	if (ret != 0) {
>  		DRV_LOG(ERR, "Failed to init mailbox");
>  		goto err_check_reset;
> @@ -443,6 +456,22 @@ idpf_adapter_init(struct idpf_adapter *adapter)
>  	return ret;
>  }
>  
> +#define IDPF_VF_TEST_VAL		0xFEED0000
> +
> +/**
> + * idpf_is_vf_device - Helper to find if it is a VF device
> + * @pdev: PCI device information struct
> + * @is_vf: used to update VF device status
> + *
> + * Return: 0 on success, errno on failure.
> + */
> +int idpf_is_vf_device(struct idpf_hw *hw, bool *is_vf)
> +{
> +	IDPF_WRITE_REG(hw, VF_ARQBAL, IDPF_VF_TEST_VAL);
> +	*is_vf = (IDPF_READ_REG(hw, VF_ARQBAL) == IDPF_VF_TEST_VAL);
> +	return 0;
> +}
> +

Why pass in is_vf as parameter and return int here, given that there is no
possible error. I would rewrite as:

bool idpf_is_vf_device(struct idpf_hw *hw)
{
	if (hw->device_id == IDPF_DEV_ID_SRIOV)
		return 1;
	IDPF_WRITE_REG(hw, VF_ARQBAL, IDPF_VF_TEST_VAL);
	return IDPF_READ_REG(hw, VF_ARQBAL) == IDPF_VF_TEST_VAL;
}

While only one line longer, it removes the need for the switch statement
above and all the error handling for a non-zero result which can never
happen.


>  RTE_EXPORT_INTERNAL_SYMBOL(idpf_adapter_deinit)
>  int
>  idpf_adapter_deinit(struct idpf_adapter *adapter)
> diff --git a/drivers/net/intel/idpf/idpf_common_device.h b/drivers/net/intel/idpf/idpf_common_device.h
> index 5f3e4a4fcf..51da8214cb 100644
> --- a/drivers/net/intel/idpf/idpf_common_device.h
> +++ b/drivers/net/intel/idpf/idpf_common_device.h
> @@ -44,6 +44,23 @@
>  	(sizeof(struct virtchnl2_ptype) +				\
>  	 (((p)->proto_id_count ? ((p)->proto_id_count - 1) : 0) * sizeof((p)->proto_id[0])))
>  
> +/** Macro used to help building up tables of device IDs with PCI class */
> +#define IDPF_PCI_CLASS(cls)          \
> +	.class_id = (cls),      \
> +	.vendor_id = RTE_PCI_ANY_ID,  \
> +	.device_id = RTE_PCI_ANY_ID,  \
> +	.subsystem_vendor_id = RTE_PCI_ANY_ID, \
> +	.subsystem_device_id = RTE_PCI_ANY_ID
> +
> +
> +/* PCI Class network ethernet */
> +#define PCI_CLASS_NETWORK_ETHERNET  0x0200
> +#define IDPF_NETWORK_ETHERNET_PROGIF				0x01
> +#define IDPF_CLASS_NETWORK_ETHERNET_PROGIF			\
> +	(PCI_CLASS_NETWORK_ETHERNET << 8 | IDPF_NETWORK_ETHERNET_PROGIF)
> +

So the device does not identify itself as an ethernet device because it has
it's class id set as the ethernet-class shifted by 8?

> +int idpf_is_vf_device(struct idpf_hw *hw, bool *is_vf);
> +
>  struct idpf_adapter {
>  	struct idpf_hw hw;
>  	struct virtchnl2_version_info virtchnl_version;
> diff --git a/drivers/net/intel/idpf/idpf_ethdev.c b/drivers/net/intel/idpf/idpf_ethdev.c
> index 90720909bf..80bf53db0f 100644
> --- a/drivers/net/intel/idpf/idpf_ethdev.c
> +++ b/drivers/net/intel/idpf/idpf_ethdev.c
> @@ -1201,6 +1201,7 @@ idpf_adapter_ext_init(struct rte_pci_device *pci_dev, struct idpf_adapter_ext *a
>  	hw->vendor_id = pci_dev->id.vendor_id;
>  	hw->device_id = pci_dev->id.device_id;
>  	hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
> +	hw->cls_id = pci_dev->id.class_id;
>  
>  	strncpy(adapter->name, pci_dev->device.name, PCI_PRI_STR_SIZE);
>  
> @@ -1313,6 +1314,7 @@ idpf_dev_vport_init(struct rte_eth_dev *dev, void *init_params)
>  static const struct rte_pci_id pci_id_idpf_map[] = {
>  	{ RTE_PCI_DEVICE(IDPF_INTEL_VENDOR_ID, IDPF_DEV_ID_PF) },
>  	{ RTE_PCI_DEVICE(IDPF_INTEL_VENDOR_ID, IDPF_DEV_ID_SRIOV) },
> +	{ IDPF_PCI_CLASS(IDPF_CLASS_NETWORK_ETHERNET_PROGIF) },
>  	{ .vendor_id = 0, /* sentinel */ },
>  };
>  
> -- 
> 2.34.1
>