net/enic: support SR-IOV VF using admin channel

The Linux VIC PF driver now requires the use of the admin channel
between PF and VF drivers. Certain devcmds are disabled for VF. The VF
driver is supposed to send control messages through the admin channel
to the PF driver to perform those devcmds. This commit adds the admin
channel to the VF driver (net/enic).

The VF's admin channel consists of normal Tx/Rx queues. VIC firmware
hardwires those queues to PF. Control messages are specially crafted
but otherwise normal packets.

The Rx queue uses LSC interrupt (interrupt vector 0) to notify the
driver of new Rx control messages. The PF driver may send unsolicited
request messages (e.g. asking for VF stats) to VF. Such messages cause
LSC interrupts and are processed on the global interrupt thread.

For devcmds that must be sent through the admin channel, use wrapper
functions. They check if the device is a VF. If VF, use the admin
channel. Otherwise, perform devcmd directly.

Two complications:
- Soft Rx stats
VF on old VIC models does not have HW Rx counters. In this case, the
VF driver counts packets/bytes and reports them as device stats.

- Backward compatibility mode
Old VIC PF drivers on some operating systems may support only
VF_CAPABILITY_REQUEST message or not support the admin channel at
all. When the VF driver detects such PF driver, it reverts to the
compatibility mode and does not use the admin channel. In this mode,
trust mode (e.g. enabling promiscuous mode) does not work.

Signed-off-by: Hyong Youb Kim <[email protected]>
Reviewed-by: John Daley <[email protected]>

Author: Hyong Youb Kim

doc/guides/nics/enic.rst

@@ -37,7 +37,7 @@ Supported features
 - UDP RSS hashing (1400 series and later adapters)
 - Scattered Rx
 - MTU update
-- SR-IOV on UCS managed servers connected to Fabric Interconnects
+- SR-IOV virtual function
 - Flow API
 - Overlay offload
 
@@ -135,103 +135,87 @@ Configuration information
     TCP, IPv4, TCP-IPv4, IPv6, TCP-IPv6, IPv6 Extension, TCP-IPv6 Extension.
 
 
-SR-IOV mode utilization
+SR-IOV Virtual Function
 -----------------------
 
-UCS blade servers configured with dynamic vNIC connection policies in UCSM
-are capable of supporting SR-IOV. SR-IOV virtual functions (VFs) are
-specialized vNICs, distinct from regular Ethernet vNICs. These VFs can be
-directly assigned to virtual machines (VMs) as 'passthrough' devices.
+VIC 1400 and later series supports SR-IOV. It can be enabled via both
+UCSM and CIMC. Please refer to the following guides to enable SR-IOV
+virtual functions (VFs).
 
-In UCS, SR-IOV VFs require the use of the Cisco Virtual Machine Fabric Extender
-(VM-FEX), which gives the VM a dedicated
-interface on the Fabric Interconnect (FI). Layer 2 switching is done at
-the FI. This may eliminate the requirement for software switching on the
-host to route intra-host VM traffic.
+  - CIMC: `Managing vNICs <https://www.cisco.com/c/en/us/td/docs/unified_computing/ucs/c/sw/gui/config/guide/4_3/b_cisco_ucs_c-series_gui_configuration_guide_43/b_Cisco_UCS_C-series_GUI_Configuration_Guide_41_chapter_01011.html#d77871e5874a1635>`_
 
-Please refer to `Creating a Dynamic vNIC Connection Policy
-<http://www.cisco.com/c/en/us/td/docs/unified_computing/ucs/sw/vm_fex/vmware/gui/config_guide/b_GUI_VMware_VM-FEX_UCSM_Configuration_Guide/b_GUI_VMware_VM-FEX_UCSM_Configuration_Guide_chapter_010.html#task_433E01651F69464783A68E66DA8A47A5>`_
-for information on configuring SR-IOV adapter policies and port profiles
-using UCSM.
+  - UCSM: `Configuring SRIOV HPN Connection Policies <https://www.cisco.com/c/en/us/td/docs/unified_computing/ucs/ucs-manager/GUI-User-Guides/Network-Mgmt/4-3/b_UCSM_Network_Mgmt_Guide_4_3/b_UCSM_Network_Mgmt_Guide_chapter_01010.html#d21438e9555a1635>`_
 
-Once the policies are in place and the host OS is rebooted, VFs should be
-visible on the host, E.g.:
+Note that the previous SR-IOV implementation that is tied to VM-FEX
+(Cisco Virtual Machine Fabric Extender) has been discontinued, and
+ENIC PMD no longer supports it. The current SR-IOV implementation does
+not require the Fabric Interconnect (FI), as layer 2 switching is done
+within the VIC adapter.
+
+Once SR-IOV is enabled, reboot the host OS and follow OS specific
+steps to create VFs and assign them to virtual machines (VMs) or
+containers as necessary. The VIC physical function (PF) drivers for ESXi
+and Linux support SR-IOV. The following shows simplified steps for
+Linux.
 
 .. code-block:: console
 
+     # echo 4 > /sys/class/net/<pf-interface>/device/sriov_numvfs
+
      # lspci | grep Cisco | grep Ethernet
-     0d:00.0 Ethernet controller: Cisco Systems Inc VIC Ethernet NIC (rev a2)
-     0d:00.1 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
-     0d:00.2 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
-     0d:00.3 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
-     0d:00.4 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
-     0d:00.5 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
-     0d:00.6 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
-     0d:00.7 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
-
-Enable Intel IOMMU on the host and install KVM and libvirt, and reboot again as
-required. Then, using libvirt, create a VM instance with an assigned device.
-Below is an example ``interface`` block (part of the domain configuration XML)
-that adds the host VF 0d:00:01 to the VM. ``profileid='pp-vlan-25'`` indicates
-the port profile that has been configured in UCSM.
+     12:00.0 Ethernet controller: Cisco Systems Inc VIC Ethernet NIC (rev a2)
+     12:00.1 Ethernet controller: Cisco Systems Inc Device 02b7 (rev a2)
+     12:00.2 Ethernet controller: Cisco Systems Inc Device 02b7 (rev a2)
+     12:00.3 Ethernet controller: Cisco Systems Inc Device 02b7 (rev a2)
+     12:00.4 Ethernet controller: Cisco Systems Inc Device 02b7 (rev a2)
+
+Writing 4 to ``sriov_numvfs`` creates 4 VFs. ``lspci`` shows VFs and
+their PCI locations.  Interfaces with device ID ``02b7`` are the
+VFs. The following snippet for libvirt XML assigns VF at ``12:00.1``
+to VM.
 
 .. code-block:: console
 
-    <interface type='hostdev' managed='yes'>
-      <mac address='52:54:00:ac:ff:b6'/>
+    <interface type="hostdev" managed="yes">
+      <mac address="fa:16:3e:46:39:c5"/>
       <driver name='vfio'/>
       <source>
-        <address type='pci' domain='0x0000' bus='0x0d' slot='0x00' function='0x1'/>
+        <address type="pci" domain="0x0000" bus="0x12" slot="0x00" function="0x1"/>
       </source>
-      <virtualport type='802.1Qbh'>
-        <parameters profileid='pp-vlan-25'/>
-      </virtualport>
+      <vlan>
+        <tag id="1000"/>
+      </vlan>
     </interface>
 
-
-Alternatively, the configuration can be done in a separate file using the
-``network`` keyword. These methods are described in the libvirt documentation for
-`Network XML format <https://libvirt.org/formatnetwork.html>`_.
-
 When the VM instance is started, libvirt will bind the host VF to
-vfio, complete provisioning on the FI and bring up the link.
-
-.. note::
-
-    It is not possible to use a VF directly from the host because it is not
-    fully provisioned until libvirt brings up the VM that it is assigned
-    to.
-
-In the VM instance, the VF will now be visible. E.g., here the VF 00:04.0 is
-seen on the VM instance and should be available for binding to a DPDK.
+vfio-pci. In the VM instance, the VF will now be visible. In this
+example, VF at ``07:00.0`` is seen on the VM instance and is available
+for binding to DPDK.
 
 .. code-block:: console
 
-     # lspci | grep Ether
-     00:04.0 Ethernet controller: Cisco Systems Inc VIC SR-IOV VF (rev a2)
+     # lspci | grep Cisco
+     07:00.0 Ethernet controller: Cisco Systems Inc Device 02b7 (rev a2)
 
-Follow the normal DPDK install procedure, binding the VF to either ``igb_uio``
-or ``vfio`` in non-IOMMU mode.
+There are two known limitations of the current SR-IOV implementation.
 
-In the VM, the kernel enic driver may be automatically bound to the VF during
-boot. Unbinding it currently hangs due to a known issue with the driver. To
-work around the issue, block the enic module as follows.
-Please see :ref:`Limitations <enic_limitations>` for limitations in
-the use of SR-IOV.
+  - Software Rx statistics
 
-.. code-block:: console
+    VF on old VIC models does not have hardware Rx counters. In this case,
+    ENIC PMD counts packets/bytes and reports them as device statistics.
 
-     # cat /etc/modprobe.d/enic.conf
-     blacklist enic
+  - Backward compatibility mode
 
-     # dracut --force
+    Old PF drivers on ESXi may lack full admin channel support. ENIC PMD
+    detects such PF driver during initialization and reverts to the
+    compatibility mode. In this mode, ENIC PMD does not use the admin channel,
+    and trust mode (e.g. enabling promiscuous mode on VF) is not supported.
 
 .. note::
 
-    Passthrough does not require SR-IOV. If VM-FEX is not desired, the user
+    Passthrough does not require SR-IOV. If SR-IOV is not desired, the user
     may create as many regular vNICs as necessary and assign them to VMs as
-    passthrough devices. Since these vNICs are not SR-IOV VFs, using them as
-    passthrough devices do not require libvirt, port profiles, and VM-FEX.
+    passthrough devices.
 
 
 .. _enic-generic-flow-api:

doc/guides/rel_notes/release_24_11.rst

@@ -55,6 +55,10 @@ New Features
      Also, make sure to start the actual text at the margin.
      =======================================================
 
+* **Updated Cisco enic driver.**
+
+  * Added SR-IOV VF support.
+
 
 Removed Items
 -------------

drivers/net/enic/base/vnic_cq.c

@@ -24,6 +24,7 @@ int vnic_cq_alloc(struct vnic_dev *vdev, struct vnic_cq *cq, unsigned int index,
 
 	cq->index = index;
 	cq->vdev = vdev;
+	cq->admin_chan = false;
 
 	cq->ctrl = vnic_dev_get_res(vdev, RES_TYPE_CQ, index);
 	if (!cq->ctrl) {
@@ -40,6 +41,32 @@ int vnic_cq_alloc(struct vnic_dev *vdev, struct vnic_cq *cq, unsigned int index,
 	return 0;
 }
 
+int vnic_admin_cq_alloc(struct vnic_dev *vdev, struct vnic_cq *cq, unsigned int index,
+	unsigned int socket_id, unsigned int desc_count, unsigned int desc_size)
+{
+	int err;
+	char res_name[RTE_MEMZONE_NAMESIZE];
+	static int instance;
+
+	cq->index = index;
+	cq->vdev = vdev;
+	cq->admin_chan = true;
+
+	cq->ctrl = vnic_dev_get_res(vdev, RES_TYPE_ADMIN_CQ, index);
+	if (!cq->ctrl) {
+		pr_err("Failed to get admin CQ[%u] resource\n", index);
+		return -EINVAL;
+	}
+
+	snprintf(res_name, sizeof(res_name), "%d-admin-cq-%u", instance++, index);
+	err = vnic_dev_alloc_desc_ring(vdev, &cq->ring, desc_count, desc_size,
+		socket_id, res_name);
+	if (err)
+		return err;
+
+	return 0;
+}
+
 void vnic_cq_init(struct vnic_cq *cq, unsigned int flow_control_enable,
 	unsigned int color_enable, unsigned int cq_head, unsigned int cq_tail,
 	unsigned int cq_tail_color, unsigned int interrupt_enable,

drivers/net/enic/base/vnic_cq.h

@@ -59,12 +59,15 @@ struct vnic_cq {
 	unsigned int tobe_rx_coal_timeval;
 	ktime_t prev_ts;
 #endif
+	bool admin_chan;
 };
 
 void vnic_cq_free(struct vnic_cq *cq);
 int vnic_cq_alloc(struct vnic_dev *vdev, struct vnic_cq *cq, unsigned int index,
 	unsigned int socket_id,
 	unsigned int desc_count, unsigned int desc_size);
+int vnic_admin_cq_alloc(struct vnic_dev *vdev, struct vnic_cq *cq, unsigned int index,
+	unsigned int socket_id, unsigned int desc_count, unsigned int desc_size);
 void vnic_cq_init(struct vnic_cq *cq, unsigned int flow_control_enable,
 	unsigned int color_enable, unsigned int cq_head, unsigned int cq_tail,
 	unsigned int cq_tail_color, unsigned int interrupt_enable,

drivers/net/enic/base/vnic_dev.c

@@ -47,6 +47,8 @@ struct vnic_dev {
 	dma_addr_t linkstatus_pa;
 	struct vnic_stats *stats;
 	dma_addr_t stats_pa;
+	struct vnic_sriov_stats *sriov_stats;
+	dma_addr_t sriov_stats_pa;
 	struct vnic_devcmd_fw_info *fw_info;
 	dma_addr_t fw_info_pa;
 	struct fm_info *flowman_info;
@@ -164,6 +166,9 @@ static int vnic_dev_discover_res(struct vnic_dev *vdev,
 		case RES_TYPE_RQ:
 		case RES_TYPE_CQ:
 		case RES_TYPE_INTR_CTRL:
+		case RES_TYPE_ADMIN_WQ:
+		case RES_TYPE_ADMIN_RQ:
+		case RES_TYPE_ADMIN_CQ:
 			/* each count is stride bytes long */
 			len = count * VNIC_RES_STRIDE;
 			if (len + bar_offset > bar[bar_num].len) {
@@ -210,6 +215,9 @@ void __iomem *vnic_dev_get_res(struct vnic_dev *vdev, enum vnic_res_type type,
 	case RES_TYPE_RQ:
 	case RES_TYPE_CQ:
 	case RES_TYPE_INTR_CTRL:
+	case RES_TYPE_ADMIN_WQ:
+	case RES_TYPE_ADMIN_RQ:
+	case RES_TYPE_ADMIN_CQ:
 		return (char __iomem *)vdev->res[type].vaddr +
 			index * VNIC_RES_STRIDE;
 	default:
@@ -1143,6 +1151,18 @@ int vnic_dev_alloc_stats_mem(struct vnic_dev *vdev)
 	return vdev->stats == NULL ? -ENOMEM : 0;
 }
 
+int vnic_dev_alloc_sriov_stats_mem(struct vnic_dev *vdev)
+{
+	char name[RTE_MEMZONE_NAMESIZE];
+	static uint32_t instance;
+
+	snprintf((char *)name, sizeof(name), "vnic_sriov_stats-%u", instance++);
+	vdev->sriov_stats = vdev->alloc_consistent(vdev->priv,
+					     sizeof(struct vnic_sriov_stats),
+					     &vdev->sriov_stats_pa, (uint8_t *)name);
+	return vdev->sriov_stats == NULL ? -ENOMEM : 0;
+}
+
 void vnic_dev_unregister(struct vnic_dev *vdev)
 {
 	if (vdev) {
@@ -1155,6 +1175,10 @@ void vnic_dev_unregister(struct vnic_dev *vdev)
 			vdev->free_consistent(vdev->priv,
 				sizeof(struct vnic_stats),
 				vdev->stats, vdev->stats_pa);
+		if (vdev->sriov_stats)
+			vdev->free_consistent(vdev->priv,
+				sizeof(struct vnic_sriov_stats),
+				vdev->sriov_stats, vdev->sriov_stats_pa);
 		if (vdev->flowman_info)
 			vdev->free_consistent(vdev->priv,
 				sizeof(struct fm_info),
@@ -1355,3 +1379,27 @@ int vnic_dev_set_cq_entry_size(struct vnic_dev *vdev, uint32_t rq_idx,
 
 	return vnic_dev_cmd(vdev, CMD_CQ_ENTRY_SIZE_SET, &a0, &a1, wait);
 }
+
+int vnic_dev_enable_admin_qp(struct vnic_dev *vdev, uint32_t enable)
+{
+	uint64_t a0, a1;
+	int wait = 1000;
+
+	a0 = QP_TYPE_ADMIN;
+	a1 = enable;
+	return vnic_dev_cmd(vdev, CMD_QP_TYPE_SET, &a0, &a1, wait);
+}
+
+int vnic_dev_sriov_stats(struct vnic_dev *vdev, struct vnic_sriov_stats **stats)
+{
+	uint64_t a0, a1;
+	int wait = 1000;
+	int err;
+
+	a0 = vdev->sriov_stats_pa;
+	a1 = sizeof(struct vnic_sriov_stats);
+	err = vnic_dev_cmd(vdev, CMD_SRIOV_STATS_GET, &a0, &a1, wait);
+	if (!err)
+		*stats = vdev->sriov_stats;
+	return err;
+}

drivers/net/enic/base/vnic_dev.h

@@ -199,5 +199,8 @@ int vnic_dev_capable_geneve(struct vnic_dev *vdev);
 uint64_t vnic_dev_capable_cq_entry_size(struct vnic_dev *vdev);
 int vnic_dev_set_cq_entry_size(struct vnic_dev *vdev, uint32_t rq_idx,
 			       uint32_t size_flag);
+int vnic_dev_alloc_sriov_stats_mem(struct vnic_dev *vdev);
+int vnic_dev_sriov_stats(struct vnic_dev *vdev, struct vnic_sriov_stats **stats);
+int vnic_dev_enable_admin_qp(struct vnic_dev *vdev, uint32_t enable);
 
 #endif /* _VNIC_DEV_H_ */

drivers/net/enic/base/vnic_devcmd.h

@@ -646,6 +646,20 @@ enum vnic_devcmd_cmd {
 	 *	   bit 2: 64 bytes
 	 */
 	CMD_CQ_ENTRY_SIZE_SET = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 90),
+
+	/*
+	 * enable/disable wq/rq queue pair of qp_type on a PF/VF.
+	 * in: (u32) a0 = wq/rq qp_type
+	 * in: (u32) a0 = enable(1)/disable(0)
+	 */
+	CMD_QP_TYPE_SET = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 97),
+
+	/*
+	 * SRIOV vic stats get
+	 * in: (u64) a0 = host buffer addr for stats dump
+	 * in  (u32) a1 = length of the buffer
+	 */
+	CMD_SRIOV_STATS_GET = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 98),
 };
 
 /* Modes for exchanging advanced filter capabilities. The modes supported by
@@ -1194,4 +1208,39 @@ typedef enum {
 #define VNIC_RQ_CQ_ENTRY_SIZE_32_CAPABLE	(1 << VNIC_RQ_CQ_ENTRY_SIZE_32)
 #define VNIC_RQ_CQ_ENTRY_SIZE_64_CAPABLE	(1 << VNIC_RQ_CQ_ENTRY_SIZE_64)
 
+/* CMD_QP_TYPE_SET */
+#define QP_TYPE_ADMIN   0
+
+struct vnic_sriov_stats {
+	uint32_t ver;
+	uint8_t sriov_vlan_membership_cap;   /* sriov support vlan-membership */
+	uint8_t sriov_vlan_membership_enabled; /* Default is disabled (0) */
+	uint8_t sriov_rss_vf_full_cap;	     /* sriov VFs support full rss */
+	uint8_t sriov_host_rx_stats;	     /* host does rx stats */
+
+	/* IGx/EGx classifier TCAM
+	 */
+	uint32_t ig_classifier0_tcam_cfg;    /* IG0 TCAM config entries */
+	uint32_t ig_classifier0_tcam_free;   /* IG0 TCAM free count */
+	uint32_t eg_classifier0_tcam_cfg;    /* EG0 TCAM config entries */
+	uint32_t eg_classifier0_tcam_free;   /* EG0 TCAM free count */
+
+	uint32_t ig_classifier1_tcam_cfg;    /* IG1 TCAM config entries */
+	uint32_t ig_classifier1_tcam_free;   /* IG1 TCAM free count */
+	uint32_t eg_classifier1_tcam_cfg;    /* EG1 TCAM config entries */
+	uint32_t eg_classifier1_tcam_free;   /* EG1 TCAM free count */
+
+	/* IGx/EGx flow table entries
+	 */
+	uint32_t sriov_ig_flow_table_cfg;    /* sriov IG FTE config */
+	uint32_t sriov_ig_flow_table_free;   /* sriov IG FTE free */
+	uint32_t sriov_eg_flow_table_cfg;    /* sriov EG FTE config */
+	uint32_t sriov_eg_flow_table_free;   /* sriov EG FTE free */
+
+	uint8_t admin_qp_ready[32];	     /* admin_qp ready bits (256) */
+	uint16_t vf_index;		     /* VF index or SRIOV_PF_IDX */
+	uint16_t reserved1;
+	uint32_t reserved2[256 - 23];
+};
+
 #endif /* _VNIC_DEVCMD_H_ */