igb: add PF support
authorIntel <intel.com>
Mon, 3 Jun 2013 00:00:00 +0000 (00:00 +0000)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Tue, 17 Sep 2013 12:16:08 +0000 (14:16 +0200)
Signed-off-by: Intel
lib/librte_pmd_e1000/Makefile
lib/librte_pmd_e1000/e1000_ethdev.h
lib/librte_pmd_e1000/igb_ethdev.c
lib/librte_pmd_e1000/igb_pf.c [new file with mode: 0644]
lib/librte_pmd_e1000/igb_rxtx.c

index 6ac5424..eea5fba 100644 (file)
@@ -85,6 +85,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_IGB_PMD) += igb_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_IGB_PMD) += igb_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_EM_PMD) += em_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_EM_PMD) += em_rxtx.c
+SRCS-$(CONFIG_RTE_LIBRTE_EM_PMD) += igb_pf.c
 
 # this lib depends upon:
 DEPDIRS-$(CONFIG_RTE_LIBRTE_E1000_PMD) += lib/librte_eal lib/librte_ether
index af75290..9ebb5ba 100644 (file)
@@ -37,6 +37,7 @@
 
 /* need update link, bit flag */
 #define E1000_FLAG_NEED_LINK_UPDATE (uint32_t)(1 << 0)
+#define E1000_FLAG_MAILBOX          (uint32_t)(1 << 1)
 
 /*
  * Defines that were not part of e1000_hw.h as they are not used by the FreeBSD
@@ -63,6 +64,21 @@ struct e1000_vfta {
        uint32_t vfta[IGB_VFTA_SIZE];
 };
 
+/*
+ * VF data which used by PF host only
+ */
+#define E1000_MAX_VF_MC_ENTRIES         30
+struct e1000_vf_info {
+       uint8_t vf_mac_addresses[ETHER_ADDR_LEN];
+       uint16_t vf_mc_hashes[E1000_MAX_VF_MC_ENTRIES];
+       uint16_t num_vf_mc_hashes;
+       uint16_t default_vf_vlan_id;
+       uint16_t vlans_enabled;
+       uint16_t pf_qos;
+       uint16_t vlan_count;
+       uint16_t tx_rate;
+};
+
 /*
  * Structure to store private data for each driver instance (for each port).
  */
@@ -71,6 +87,7 @@ struct e1000_adapter {
        struct e1000_hw_stats   stats;
        struct e1000_interrupt  intr;
        struct e1000_vfta       shadow_vfta;
+       struct e1000_vf_info    *vfdata;
 };
 
 #define E1000_DEV_PRIVATE_TO_HW(adapter) \
@@ -85,6 +102,9 @@ struct e1000_adapter {
 #define E1000_DEV_PRIVATE_TO_VFTA(adapter) \
        (&((struct e1000_adapter *)adapter)->shadow_vfta)
 
+#define E1000_DEV_PRIVATE_TO_P_VFDATA(adapter) \
+        (&((struct e1000_adapter *)adapter)->vfdata)
+
 /*
  * RX/TX IGB function prototypes
  */
@@ -121,6 +141,15 @@ int eth_igbvf_rx_init(struct rte_eth_dev *dev);
 
 void eth_igbvf_tx_init(struct rte_eth_dev *dev);
 
+/*
+ * misc function prototypes
+ */
+void igb_pf_host_init(struct rte_eth_dev *eth_dev);
+void igb_pf_mbx_process(struct rte_eth_dev *eth_dev);
+int igb_pf_host_configure(struct rte_eth_dev *eth_dev);
+
 /*
  * RX/TX EM function prototypes
  */
index f9b03eb..4cea254 100644 (file)
@@ -139,6 +139,8 @@ static int eth_igb_rss_reta_query(struct rte_eth_dev *dev,
 #define IGB_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
 #define IGB_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
 
+#define IGBVF_PMD_NAME "rte_igbvf_pmd"     /* PMD name */
+
 static enum e1000_fc_mode igb_fc_setting = e1000_fc_full;
 
 /*
@@ -284,6 +286,23 @@ igb_intr_disable(struct e1000_hw *hw)
        E1000_WRITE_FLUSH(hw);
 }
 
+static inline int32_t
+igb_pf_reset_hw(struct e1000_hw *hw)
+{
+       uint32_t ctrl_ext;
+       int32_t status;
+       status = e1000_reset_hw(hw);
+       ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
+       /* Set PF Reset Done bit so PF/VF Mail Ops can work */
+       ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
+       E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
+       E1000_WRITE_FLUSH(hw);
+       return status;
+}
 static void
 igb_identify_hardware(struct rte_eth_dev *dev)
 {
@@ -310,6 +329,7 @@ eth_igb_dev_init(__attribute__((unused)) struct eth_driver *eth_drv,
                E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
        struct e1000_vfta * shadow_vfta =
                        E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
+       uint32_t ctrl_ext;
 
        pci_dev = eth_dev->pci_dev;
        eth_dev->dev_ops = &eth_igb_ops;
@@ -350,7 +370,7 @@ eth_igb_dev_init(__attribute__((unused)) struct eth_driver *eth_drv,
         * Start from a known state, this is important in reading the nvm
         * and mac from that.
         */
-       e1000_reset_hw(hw);
+       igb_pf_reset_hw(hw);
 
        /* Make sure we have a good EEPROM before we read from it */
        if (e1000_validate_nvm_checksum(hw) < 0) {
@@ -406,6 +426,15 @@ eth_igb_dev_init(__attribute__((unused)) struct eth_driver *eth_drv,
                                        "SOL/IDER session");
        }
 
+       /* initialize PF if max_vfs not zero */
+       igb_pf_host_init(eth_dev);
+       ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
+       /* Set PF Reset Done bit so PF/VF Mail Ops can work */
+       ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
+       E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
+       E1000_WRITE_FLUSH(hw);
+
        PMD_INIT_LOG(INFO, "port_id %d vendorID=0x%x deviceID=0x%x\n",
                     eth_dev->data->port_id, pci_dev->id.vendor_id,
                     pci_dev->id.device_id);
@@ -461,7 +490,7 @@ eth_igbvf_dev_init(__attribute__((unused)) struct eth_driver *eth_drv,
 
        /* Disable the interrupts for VF */
        igbvf_intr_disable(hw);
-
+       
        diag = hw->mac.ops.reset_hw(hw);
 
        /* Allocate memory for storing MAC addresses */
@@ -474,6 +503,7 @@ eth_igbvf_dev_init(__attribute__((unused)) struct eth_driver *eth_drv,
                        ETHER_ADDR_LEN * hw->mac.rar_entry_count);
                return -ENOMEM;
        }
+       
        /* Copy the permanent MAC address */
        ether_addr_copy((struct ether_addr *) hw->mac.perm_addr,
                        &eth_dev->data->mac_addrs[0]);
@@ -556,6 +586,7 @@ eth_igb_start(struct rte_eth_dev *dev)
        struct e1000_hw *hw =
                E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
        int ret, i, mask;
+       uint32_t ctrl_ext;
 
        PMD_INIT_LOG(DEBUG, ">>");
 
@@ -585,6 +616,15 @@ eth_igb_start(struct rte_eth_dev *dev)
 
        E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN);
 
+       ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
+       /* Set PF Reset Done bit so PF/VF Mail Ops can work */
+       ctrl_ext |= E1000_CTRL_EXT_PFRSTD;
+       E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext);
+       E1000_WRITE_FLUSH(hw);
+
+       /* configure PF module if SRIOV enabled */
+       igb_pf_host_configure(dev);
+
        /* Configure for OS presence */
        igb_init_manageability(hw);
 
@@ -714,7 +754,7 @@ eth_igb_stop(struct rte_eth_dev *dev)
        struct rte_eth_link link;
 
        igb_intr_disable(hw);
-       e1000_reset_hw(hw);
+       igb_pf_reset_hw(hw);
        E1000_WRITE_REG(hw, E1000_WUC, 0);
 
        /* Power down the phy. Needed to make the link go Down */
@@ -806,7 +846,7 @@ igb_hardware_init(struct e1000_hw *hw)
                hw->fc.requested_mode = e1000_fc_none;
 
        /* Issue a global reset */
-       e1000_reset_hw(hw);
+       igb_pf_reset_hw(hw);
        E1000_WRITE_REG(hw, E1000_WUC, 0);
 
        diag = e1000_init_hw(hw);
@@ -1493,6 +1533,9 @@ eth_igb_interrupt_get_status(struct rte_eth_dev *dev)
                intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
        }
 
+       if (icr & E1000_ICR_VMMB) 
+               intr->flags |= E1000_FLAG_MAILBOX;
+
        return 0;
 }
 
@@ -1517,6 +1560,10 @@ eth_igb_interrupt_action(struct rte_eth_dev *dev)
        struct rte_eth_link link;
        int ret;
 
+       if (intr->flags & E1000_FLAG_MAILBOX) {
+               igb_pf_mbx_process(dev);
+               intr->flags &= ~E1000_FLAG_MAILBOX;
+       }
 
        igb_intr_enable(dev);
        rte_intr_enable(&(dev->pci_dev->intr_handle));
@@ -1650,13 +1697,18 @@ eth_igb_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
        return (-EIO);
 }
 
+#define E1000_RAH_POOLSEL_SHIFT      (18)
 static void
 eth_igb_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
                uint32_t index, __rte_unused uint32_t pool)
 {
        struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t rah;
 
        e1000_rar_set(hw, mac_addr->addr_bytes, index);
+       rah = E1000_READ_REG(hw, E1000_RAH(index));
+       rah |= (0x1 << (E1000_RAH_POOLSEL_SHIFT + pool));
+       E1000_WRITE_REG(hw, E1000_RAH(index), rah);
 }
 
 static void
diff --git a/lib/librte_pmd_e1000/igb_pf.c b/lib/librte_pmd_e1000/igb_pf.c
new file mode 100644 (file)
index 0000000..e7c1de6
--- /dev/null
@@ -0,0 +1,494 @@
+/*-
+ *   BSD LICENSE
+ * 
+ *   Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ * 
+ *   Redistribution and use in source and binary forms, with or without 
+ *   modification, are permitted provided that the following conditions 
+ *   are met:
+ * 
+ *     * Redistributions of source code must retain the above copyright 
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright 
+ *       notice, this list of conditions and the following disclaimer in 
+ *       the documentation and/or other materials provided with the 
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its 
+ *       contributors may be used to endorse or promote products derived 
+ *       from this software without specific prior written permission.
+ * 
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ * 
+ */
+
+#include <stdio.h>
+#include <errno.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <stdarg.h>
+#include <inttypes.h>
+
+#include <rte_interrupts.h>
+#include <rte_log.h>
+#include <rte_debug.h>
+#include <rte_eal.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_memcpy.h>
+#include <rte_malloc.h>
+#include <rte_random.h>
+
+#include "e1000/e1000_defines.h"
+#include "e1000/e1000_regs.h"
+#include "e1000/e1000_hw.h"
+#include "e1000_ethdev.h"
+
+static inline 
+void eth_random_addr(uint8_t *addr)
+{
+       uint64_t rand = rte_rand();
+       uint8_t *p = (uint8_t*)&rand;
+
+       rte_memcpy(addr, p, ETHER_ADDR_LEN);
+       addr[0] &= 0xfe;        /* clear multicast bit */
+       addr[0] |= 0x02;        /* set local assignment bit (IEEE802) */
+}
+
+static inline uint16_t
+dev_num_vf(struct rte_eth_dev *eth_dev)
+{
+       return eth_dev->pci_dev->max_vfs;
+}
+
+static inline 
+int igb_vf_perm_addr_gen(struct rte_eth_dev *dev, uint16_t vf_num)
+{
+       unsigned char vf_mac_addr[ETHER_ADDR_LEN];
+       struct e1000_vf_info *vfinfo = 
+               *E1000_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
+       uint16_t vfn;
+
+       for (vfn = 0; vfn < vf_num; vfn++) {
+               eth_random_addr(vf_mac_addr);
+               /* keep the random address as default */
+               memcpy(vfinfo[vfn].vf_mac_addresses, vf_mac_addr, 
+                               ETHER_ADDR_LEN);
+       }
+
+       return 0;
+}
+
+static inline int
+igb_mb_intr_setup(struct rte_eth_dev *dev)
+{
+       struct e1000_interrupt *intr =
+               E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
+
+       intr->mask |= E1000_ICR_VMMB;
+
+       return 0;
+}
+
+void igb_pf_host_init(struct rte_eth_dev *eth_dev)
+{
+       struct e1000_vf_info **vfinfo = 
+               E1000_DEV_PRIVATE_TO_P_VFDATA(eth_dev->data->dev_private);
+       struct e1000_hw *hw = 
+               E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       uint16_t vf_num;
+       uint8_t nb_queue;
+
+       RTE_ETH_DEV_SRIOV(eth_dev).active = 0;
+       if (0 == (vf_num = dev_num_vf(eth_dev)))
+               return;
+
+       if (hw->mac.type == e1000_i350)
+               nb_queue = 1;
+       else if(hw->mac.type == e1000_82576)
+               /* per datasheet, it should be 2, but 1 seems correct */
+               nb_queue = 1;
+       else
+               return;
+
+       *vfinfo = rte_zmalloc("vf_info", sizeof(struct e1000_vf_info) * vf_num, 0);
+       if (*vfinfo == NULL)
+               rte_panic("Cannot allocate memory for private VF data\n");
+
+       RTE_ETH_DEV_SRIOV(eth_dev).active = ETH_8_POOLS;
+       RTE_ETH_DEV_SRIOV(eth_dev).nb_q_per_pool = nb_queue;
+       RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx = vf_num;
+       RTE_ETH_DEV_SRIOV(eth_dev).def_pool_q_idx = (uint16_t)(vf_num * nb_queue);
+
+       igb_vf_perm_addr_gen(eth_dev, vf_num);
+
+       /* set mb interrupt mask */
+       igb_mb_intr_setup(eth_dev);
+
+       return;
+}
+
+#define E1000_RAH_POOLSEL_SHIFT    (18)
+int igb_pf_host_configure(struct rte_eth_dev *eth_dev)
+{
+       uint32_t vtctl;
+       uint16_t vf_num;
+       struct e1000_hw *hw = 
+               E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       uint32_t vlanctrl;
+       int i;
+       uint32_t rah;
+
+       if (0 == (vf_num = dev_num_vf(eth_dev)))
+               return -1;
+
+       /* enable VMDq and set the default pool for PF */
+       vtctl = E1000_READ_REG(hw, E1000_VT_CTL);
+       vtctl &= ~E1000_VT_CTL_DEFAULT_POOL_MASK;
+       vtctl |= RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx 
+               << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
+       vtctl |= E1000_VT_CTL_VM_REPL_EN;
+       E1000_WRITE_REG(hw, E1000_VT_CTL, vtctl);
+
+       /* Enable pools reserved to PF only */
+       E1000_WRITE_REG(hw, E1000_VFRE, (~0) << vf_num);
+       E1000_WRITE_REG(hw, E1000_VFTE, (~0) << vf_num);
+
+       /* PFDMA Tx General Switch Control Enables VMDQ loopback */
+       if (hw->mac.type == e1000_i350)
+               E1000_WRITE_REG(hw, E1000_TXSWC, E1000_DTXSWC_VMDQ_LOOPBACK_EN);
+       else
+               E1000_WRITE_REG(hw, E1000_DTXSWC, E1000_DTXSWC_VMDQ_LOOPBACK_EN);
+
+       /* clear VMDq map to perment rar 0 */
+       rah = E1000_READ_REG(hw, E1000_RAH(0));
+       rah &= ~ (0xFF << E1000_RAH_POOLSEL_SHIFT);
+       E1000_WRITE_REG(hw, E1000_RAH(0), rah);
+
+       /* clear VMDq map to scan rar 32 */
+       rah = E1000_READ_REG(hw, E1000_RAH(hw->mac.rar_entry_count));
+       rah &= ~ (0xFF << E1000_RAH_POOLSEL_SHIFT);
+       E1000_WRITE_REG(hw, E1000_RAH(hw->mac.rar_entry_count), rah);
+
+       /* set VMDq map to default PF pool */
+       rah = E1000_READ_REG(hw, E1000_RAH(0));
+       rah |= (0x1 << (RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx +
+                       E1000_RAH_POOLSEL_SHIFT));
+       E1000_WRITE_REG(hw, E1000_RAH(0), rah);
+
+       /*
+        * enable vlan filtering and allow all vlan tags through 
+        */
+       vlanctrl = E1000_READ_REG(hw, E1000_RCTL);
+       vlanctrl |= E1000_RCTL_VFE ; /* enable vlan filters */
+       E1000_WRITE_REG(hw, E1000_RCTL, vlanctrl);
+
+       /* VFTA - enable all vlan filters */
+       for (i = 0; i < IGB_VFTA_SIZE; i++) {
+               E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, 0xFFFFFFFF);
+       }
+       
+       /* Enable/Disable MAC Anti-Spoofing */
+       e1000_vmdq_set_anti_spoofing_pf(hw, FALSE, vf_num);
+
+       return 0;
+}
+
+static void 
+set_rx_mode(struct rte_eth_dev *dev)
+{
+       struct rte_eth_dev_data *dev_data = 
+               (struct rte_eth_dev_data*)dev->data->dev_private;
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t fctrl, vmolr = E1000_VMOLR_BAM | E1000_VMOLR_AUPE;
+       uint16_t vfn = dev_num_vf(dev);
+
+       /* Check for Promiscuous and All Multicast modes */
+       fctrl = E1000_READ_REG(hw, E1000_RCTL);
+
+       /* set all bits that we expect to always be set */
+       fctrl &= ~E1000_RCTL_SBP; /* disable store-bad-packets */
+       fctrl |= E1000_RCTL_BAM;;
+
+       /* clear the bits we are changing the status of */
+       fctrl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
+
+       if (dev_data->promiscuous) {
+               fctrl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
+               vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
+       } else {
+               if (dev_data->all_multicast) {
+                       fctrl |= E1000_RCTL_MPE;
+                       vmolr |= E1000_VMOLR_MPME;
+               } else {
+                       vmolr |= E1000_VMOLR_ROMPE;
+               }
+       }
+
+       if ((hw->mac.type == e1000_82576) ||
+               (hw->mac.type == e1000_i350)) {
+               vmolr |= E1000_READ_REG(hw, E1000_VMOLR(vfn)) &
+                        ~(E1000_VMOLR_MPME | E1000_VMOLR_ROMPE |
+                          E1000_VMOLR_ROPE);
+               E1000_WRITE_REG(hw, E1000_VMOLR(vfn), vmolr);
+       }
+
+       E1000_WRITE_REG(hw, E1000_RCTL, fctrl);
+}
+
+static inline void 
+igb_vf_reset_event(struct rte_eth_dev *dev, uint16_t vf)
+{
+       struct e1000_hw *hw = 
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_vf_info *vfinfo = 
+               *(E1000_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
+       uint32_t vmolr = E1000_READ_REG(hw, E1000_VMOLR(vf));
+
+       vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | 
+                       E1000_VMOLR_BAM | E1000_VMOLR_AUPE);
+       E1000_WRITE_REG(hw, E1000_VMOLR(vf), vmolr);
+
+       E1000_WRITE_REG(hw, E1000_VMVIR(vf), 0);
+       
+       /* reset multicast table array for vf */
+       vfinfo[vf].num_vf_mc_hashes = 0;
+
+       /* reset rx mode */
+       set_rx_mode(dev);
+}
+
+static inline void 
+igb_vf_reset_msg(struct rte_eth_dev *dev, uint16_t vf)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t reg;
+
+       /* enable transmit and receive for vf */
+       reg = E1000_READ_REG(hw, E1000_VFTE);
+       reg |= (reg | (1 << vf));
+       E1000_WRITE_REG(hw, E1000_VFTE, reg);
+
+       reg = E1000_READ_REG(hw, E1000_VFRE);
+       reg |= (reg | (1 << vf));
+       E1000_WRITE_REG(hw, E1000_VFRE, reg);
+
+       igb_vf_reset_event(dev, vf);
+}
+
+static int
+igb_vf_reset(struct rte_eth_dev *dev, uint16_t vf, uint32_t *msgbuf)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_vf_info *vfinfo = 
+               *(E1000_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
+       unsigned char *vf_mac = vfinfo[vf].vf_mac_addresses;
+       int rar_entry = hw->mac.rar_entry_count - (vf + 1);
+       uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
+       uint32_t rah;
+
+       igb_vf_reset_msg(dev, vf);
+
+       hw->mac.ops.rar_set(hw, vf_mac, rar_entry);
+       rah = E1000_READ_REG(hw, E1000_RAH(rar_entry));
+       rah |= (0x1 << (vf + E1000_RAH_POOLSEL_SHIFT));
+       E1000_WRITE_REG(hw, E1000_RAH(rar_entry), rah);
+
+       /* reply to reset with ack and vf mac address */
+       msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
+       rte_memcpy(new_mac, vf_mac, ETHER_ADDR_LEN);
+       e1000_write_mbx(hw, msgbuf, 3, vf);
+
+       return 0;
+}
+
+static int
+igb_vf_set_mac_addr(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_vf_info *vfinfo = 
+               *(E1000_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
+       int rar_entry = hw->mac.rar_entry_count - (vf + 1);
+       uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
+
+       if (is_valid_assigned_ether_addr((struct ether_addr*)new_mac)) {
+               rte_memcpy(vfinfo[vf].vf_mac_addresses, new_mac, 6);
+               hw->mac.ops.rar_set(hw, new_mac, rar_entry);
+               return 0;
+       }
+       return -1;
+}
+
+static int
+igb_vf_set_multicast(struct rte_eth_dev *dev, __rte_unused uint32_t vf, uint32_t *msgbuf)
+{
+       int i;
+       uint32_t vector_bit;
+       uint32_t vector_reg;
+       uint32_t mta_reg;
+       int entries = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> 
+               E1000_VT_MSGINFO_SHIFT;
+       uint16_t *hash_list = (uint16_t *)&msgbuf[1];
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_vf_info *vfinfo = 
+               *(E1000_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
+
+       /* only so many hash values supported */
+       entries = RTE_MIN(entries, E1000_MAX_VF_MC_ENTRIES);
+
+       /*
+        * salt away the number of multi cast addresses assigned
+        * to this VF for later use to restore when the PF multi cast
+        * list changes
+        */
+       vfinfo->num_vf_mc_hashes = (uint16_t)entries;
+
+       /*
+        * VFs are limited to using the MTA hash table for their multicast
+        * addresses
+        */
+       for (i = 0; i < entries; i++) {
+               vfinfo->vf_mc_hashes[i] = hash_list[i];
+       }
+
+       for (i = 0; i < vfinfo->num_vf_mc_hashes; i++) {
+               vector_reg = (vfinfo->vf_mc_hashes[i] >> 5) & 0x7F;
+               vector_bit = vfinfo->vf_mc_hashes[i] & 0x1F;
+               mta_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, vector_reg);
+               mta_reg |= (1 << vector_bit);
+               E1000_WRITE_REG_ARRAY(hw, E1000_MTA, vector_reg, mta_reg);
+       }
+
+       return 0;
+}
+
+static int
+igb_vf_set_vlan(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
+{
+       int add, vid;
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_vf_info *vfinfo = 
+               *(E1000_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
+       uint32_t vid_idx, vid_bit, vfta;
+
+       add = (msgbuf[0] & E1000_VT_MSGINFO_MASK)
+               >> E1000_VT_MSGINFO_SHIFT;
+       vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
+
+       if (add)
+               vfinfo[vf].vlan_count++;
+       else if (vfinfo[vf].vlan_count)
+               vfinfo[vf].vlan_count--;
+
+       vid_idx = (uint32_t)((vid >> E1000_VFTA_ENTRY_SHIFT) & 
+                            E1000_VFTA_ENTRY_MASK);
+       vid_bit = (uint32_t)(1 << (vid & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
+       vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
+       if (add)
+               vfta |= vid_bit;
+       else
+               vfta &= ~vid_bit;
+       
+       E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
+       E1000_WRITE_FLUSH(hw);
+
+       return 0;
+}
+
+static int 
+igb_rcv_msg_from_vf(struct rte_eth_dev *dev, uint16_t vf)
+{
+       uint16_t mbx_size = E1000_VFMAILBOX_SIZE;
+       uint32_t msgbuf[E1000_VFMAILBOX_SIZE];
+       int32_t retval;
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       retval = e1000_read_mbx(hw, msgbuf, mbx_size, vf);
+       if (retval) {
+               RTE_LOG(ERR, PMD, "Error mbx recv msg from VF %d\n", vf);
+               return retval;
+       }
+
+       /* do nothing with the message already processed */
+       if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
+               return retval;
+
+       /* flush the ack before we write any messages back */
+       E1000_WRITE_FLUSH(hw);
+
+       /* perform VF reset */
+       if (msgbuf[0] == E1000_VF_RESET) {
+               return igb_vf_reset(dev, vf, msgbuf);
+       }
+
+       /* check & process VF to PF mailbox message */
+       switch ((msgbuf[0] & 0xFFFF)) {
+       case E1000_VF_SET_MAC_ADDR:
+               retval = igb_vf_set_mac_addr(dev, vf, msgbuf);
+               break;
+       case E1000_VF_SET_MULTICAST:
+               retval = igb_vf_set_multicast(dev, vf, msgbuf);
+               break;
+       case E1000_VF_SET_VLAN:
+               retval = igb_vf_set_vlan(dev, vf, msgbuf);
+               break;
+       default:
+               RTE_LOG(DEBUG, PMD, "Unhandled Msg %8.8x\n", (unsigned) msgbuf[0]);
+               retval = E1000_ERR_MBX;
+               break;
+       }
+
+       /* response the VF according to the message process result */
+       if (retval)
+               msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
+       else
+               msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
+
+       msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
+
+       e1000_write_mbx(hw, msgbuf, 1, vf);
+
+       return retval;
+}
+
+static inline void 
+igb_rcv_ack_from_vf(struct rte_eth_dev *dev, uint16_t vf)
+{
+       uint32_t msg = E1000_VT_MSGTYPE_NACK;
+       struct e1000_hw *hw = 
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       e1000_write_mbx(hw, &msg, 1, vf);
+}
+
+void igb_pf_mbx_process(struct rte_eth_dev *eth_dev)
+{
+       uint16_t vf;
+       struct e1000_hw *hw = 
+               E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+
+       for (vf = 0; vf < dev_num_vf(eth_dev); vf++) {
+               /* check & process vf function level reset */
+               if (!e1000_check_for_rst(hw, vf))
+                       igb_vf_reset_event(eth_dev, vf);
+
+               /* check & process vf mailbox messages */
+               if (!e1000_check_for_msg(hw, vf))
+                       igb_rcv_msg_from_vf(eth_dev, vf);
+
+               /* check & process acks from vf */
+               if (!e1000_check_for_ack(hw, vf))
+                       igb_rcv_ack_from_vf(eth_dev, vf);
+       }
+}
index dd24788..62c2985 100644 (file)
@@ -126,6 +126,7 @@ struct igb_rx_queue {
        uint16_t            nb_rx_hold; /**< number of held free RX desc. */
        uint16_t            rx_free_thresh; /**< max free RX desc to hold. */
        uint16_t            queue_id;   /**< RX queue index. */
+       uint16_t            reg_idx;    /**< RX queue register index. */
        uint8_t             port_id;    /**< Device port identifier. */
        uint8_t             pthresh;    /**< Prefetch threshold register. */
        uint8_t             hthresh;    /**< Host threshold register. */
@@ -166,6 +167,7 @@ struct igb_tx_queue {
        uint16_t               tx_head;
        /**< Index of first used TX descriptor. */
        uint16_t               queue_id; /**< TX queue index. */
+       uint16_t               reg_idx;  /**< TX queue register index. */
        uint8_t                port_id;  /**< Device port identifier. */
        uint8_t                pthresh;  /**< Prefetch threshold register. */
        uint8_t                hthresh;  /**< Host threshold register. */
@@ -1229,9 +1231,11 @@ eth_igb_tx_queue_setup(struct rte_eth_dev *dev,
        txq->hthresh = tx_conf->tx_thresh.hthresh;
        txq->wthresh = tx_conf->tx_thresh.wthresh;
        txq->queue_id = queue_idx;
+       txq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
+               queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
        txq->port_id = dev->data->port_id;
 
-       txq->tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(queue_idx));
+       txq->tdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_TDT(txq->reg_idx));
        txq->tx_ring_phys_addr = (uint64_t) tz->phys_addr;
        txq->tx_ring = (union e1000_adv_tx_desc *) tz->addr;
 
@@ -1345,6 +1349,8 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
        rxq->drop_en = rx_conf->rx_drop_en;
        rxq->rx_free_thresh = rx_conf->rx_free_thresh;
        rxq->queue_id = queue_idx;
+       rxq->reg_idx = (uint16_t)((RTE_ETH_DEV_SRIOV(dev).active == 0) ?
+               queue_idx : RTE_ETH_DEV_SRIOV(dev).def_pool_q_idx + queue_idx);
        rxq->port_id = dev->data->port_id;
        rxq->crc_len = (uint8_t) ((dev->data->dev_conf.rxmode.hw_strip_crc) ? 0 :
                                  ETHER_CRC_LEN);
@@ -1360,8 +1366,8 @@ eth_igb_rx_queue_setup(struct rte_eth_dev *dev,
                igb_rx_queue_release(rxq);
                return (-ENOMEM);
        }
-       rxq->rdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDT(queue_idx));
-       rxq->rdh_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDH(queue_idx));
+       rxq->rdt_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDT(rxq->reg_idx));
+       rxq->rdh_reg_addr = E1000_PCI_REG_ADDR(hw, E1000_RDH(rxq->reg_idx));
        rxq->rx_ring_phys_addr = (uint64_t) rz->phys_addr;
        rxq->rx_ring = (union e1000_adv_rx_desc *) rz->addr;
 
@@ -1641,12 +1647,12 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                                                        0 : ETHER_CRC_LEN);
 
                bus_addr = rxq->rx_ring_phys_addr;
-               E1000_WRITE_REG(hw, E1000_RDLEN(i),
+               E1000_WRITE_REG(hw, E1000_RDLEN(rxq->reg_idx),
                                rxq->nb_rx_desc *
                                sizeof(union e1000_adv_rx_desc));
-               E1000_WRITE_REG(hw, E1000_RDBAH(i),
+               E1000_WRITE_REG(hw, E1000_RDBAH(rxq->reg_idx),
                                (uint32_t)(bus_addr >> 32));
-               E1000_WRITE_REG(hw, E1000_RDBAL(i), (uint32_t)bus_addr);
+               E1000_WRITE_REG(hw, E1000_RDBAL(rxq->reg_idx), (uint32_t)bus_addr);
 
                srrctl = E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
 
@@ -1691,16 +1697,16 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                if (rxq->drop_en)
                        srrctl |= E1000_SRRCTL_DROP_EN;
 
-               E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
+               E1000_WRITE_REG(hw, E1000_SRRCTL(rxq->reg_idx), srrctl);
 
                /* Enable this RX queue. */
-               rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
+               rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(rxq->reg_idx));
                rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
                rxdctl &= 0xFFF00000;
                rxdctl |= (rxq->pthresh & 0x1F);
                rxdctl |= ((rxq->hthresh & 0x1F) << 8);
                rxdctl |= ((rxq->wthresh & 0x1F) << 16);
-               E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
+               E1000_WRITE_REG(hw, E1000_RXDCTL(rxq->reg_idx), rxdctl);
        }
 
        /*
@@ -1753,10 +1759,11 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                /* set STRCRC bit in all queues for Powerville/Springville */
                if (hw->mac.type == e1000_i350 || hw->mac.type == e1000_i210) {
                        for (i = 0; i < dev->data->nb_rx_queues; i++) {
+                               rxq = dev->data->rx_queues[i];
                                uint32_t dvmolr = E1000_READ_REG(hw,
-                                       E1000_DVMOLR(i));
+                                       E1000_DVMOLR(rxq->reg_idx));
                                dvmolr |= E1000_DVMOLR_STRCRC;
-                               E1000_WRITE_REG(hw, E1000_DVMOLR(i), dvmolr);
+                               E1000_WRITE_REG(hw, E1000_DVMOLR(rxq->reg_idx), dvmolr);
                        }
                }
        } else {
@@ -1765,10 +1772,11 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
                /* clear STRCRC bit in all queues for Powerville/Springville */
                if (hw->mac.type == e1000_i350 || hw->mac.type == e1000_i210) {
                        for (i = 0; i < dev->data->nb_rx_queues; i++) {
+                               rxq = dev->data->rx_queues[i];
                                uint32_t dvmolr = E1000_READ_REG(hw,
-                                       E1000_DVMOLR(i));
+                                       E1000_DVMOLR(rxq->reg_idx));
                                dvmolr &= ~E1000_DVMOLR_STRCRC;
-                               E1000_WRITE_REG(hw, E1000_DVMOLR(i), dvmolr);
+                               E1000_WRITE_REG(hw, E1000_DVMOLR(rxq->reg_idx), dvmolr);
                        }
                }
        }
@@ -1792,8 +1800,8 @@ eth_igb_rx_init(struct rte_eth_dev *dev)
         */
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                rxq = dev->data->rx_queues[i];
-               E1000_WRITE_REG(hw, E1000_RDH(i), 0);
-               E1000_WRITE_REG(hw, E1000_RDT(i), rxq->nb_rx_desc - 1);
+               E1000_WRITE_REG(hw, E1000_RDH(rxq->reg_idx), 0);
+               E1000_WRITE_REG(hw, E1000_RDT(rxq->reg_idx), rxq->nb_rx_desc - 1);
        }
 
        return 0;
@@ -1821,24 +1829,24 @@ eth_igb_tx_init(struct rte_eth_dev *dev)
                txq = dev->data->tx_queues[i];
                bus_addr = txq->tx_ring_phys_addr;
 
-               E1000_WRITE_REG(hw, E1000_TDLEN(i),
+               E1000_WRITE_REG(hw, E1000_TDLEN(txq->reg_idx),
                                txq->nb_tx_desc *
                                sizeof(union e1000_adv_tx_desc));
-               E1000_WRITE_REG(hw, E1000_TDBAH(i),
+               E1000_WRITE_REG(hw, E1000_TDBAH(txq->reg_idx),
                                (uint32_t)(bus_addr >> 32));
-               E1000_WRITE_REG(hw, E1000_TDBAL(i), (uint32_t)bus_addr);
+               E1000_WRITE_REG(hw, E1000_TDBAL(txq->reg_idx), (uint32_t)bus_addr);
 
                /* Setup the HW Tx Head and Tail descriptor pointers. */
-               E1000_WRITE_REG(hw, E1000_TDT(i), 0);
-               E1000_WRITE_REG(hw, E1000_TDH(i), 0);
+               E1000_WRITE_REG(hw, E1000_TDT(txq->reg_idx), 0);
+               E1000_WRITE_REG(hw, E1000_TDH(txq->reg_idx), 0);
 
                /* Setup Transmit threshold registers. */
-               txdctl = E1000_READ_REG(hw, E1000_TXDCTL(i));
+               txdctl = E1000_READ_REG(hw, E1000_TXDCTL(txq->reg_idx));
                txdctl |= txq->pthresh & 0x1F;
                txdctl |= ((txq->hthresh & 0x1F) << 8);
                txdctl |= ((txq->wthresh & 0x1F) << 16);
                txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
-               E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
+               E1000_WRITE_REG(hw, E1000_TXDCTL(txq->reg_idx), txdctl);
        }
 
        /* Program the Transmit Control Register. */