]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/commitdiff
qed: Add Light L2 support
authorYuval Mintz <Yuval.Mintz@caviumnetworks.com>
Sat, 1 Oct 2016 18:59:55 +0000 (21:59 +0300)
committerDavid S. Miller <davem@davemloft.net>
Tue, 4 Oct 2016 03:22:46 +0000 (23:22 -0400)
Other protocols beside the networking driver need the ability
of passing some L2 traffic, usually [although not limited] for the
purpose of some management traffic.

Signed-off-by: Yuval Mintz <Yuval.Mintz@caviumnetworks.com>
Signed-off-by: Ram Amrani <Ram.Amrani@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
13 files changed:
drivers/net/ethernet/qlogic/Kconfig
drivers/net/ethernet/qlogic/qed/Makefile
drivers/net/ethernet/qlogic/qed/qed.h
drivers/net/ethernet/qlogic/qed/qed_cxt.c
drivers/net/ethernet/qlogic/qed/qed_dev.c
drivers/net/ethernet/qlogic/qed/qed_dev_api.h
drivers/net/ethernet/qlogic/qed/qed_ll2.c [new file with mode: 0644]
drivers/net/ethernet/qlogic/qed/qed_ll2.h [new file with mode: 0644]
drivers/net/ethernet/qlogic/qed/qed_main.c
drivers/net/ethernet/qlogic/qed/qed_reg_addr.h
drivers/net/ethernet/qlogic/qed/qed_sp.h
include/linux/qed/qed_if.h
include/linux/qed/qed_ll2_if.h [new file with mode: 0644]

index 6ba48406899e75286db243a31985e78c62b4bdc1..9eb3b1914cf5211693717d8a5b1566656525eb24 100644 (file)
@@ -88,6 +88,14 @@ config QED
        ---help---
          This enables the support for ...
 
+config QED_LL2
+       bool "Qlogic QED Light L2 interface"
+       default n
+       depends on QED
+       ---help---
+       This enables support for Light L2 interface which is required
+       by all qed protocol drivers other than qede.
+
 config QED_SRIOV
        bool "QLogic QED 25/40/100Gb SR-IOV support"
        depends on QED && PCI_IOV
index 86a5b4f5f870e14f3759c411223029a51d75894d..e067098f10a9bedec8021cfeb1abf864d604b10b 100644 (file)
@@ -4,3 +4,4 @@ qed-y := qed_cxt.o qed_dev.o qed_hw.o qed_init_fw_funcs.o qed_init_ops.o \
         qed_int.o qed_main.o qed_mcp.o qed_sp_commands.o qed_spq.o qed_l2.o \
         qed_selftest.o qed_dcbx.o qed_debug.o
 qed-$(CONFIG_QED_SRIOV) += qed_sriov.o qed_vf.o
+qed-$(CONFIG_QED_LL2) += qed_ll2.o
index 0929582fc82b747fd8609fa29e85691921434836..91b571a3670b4673dfe08b968e3e2607866f743e 100644 (file)
@@ -72,6 +72,7 @@ struct qed_sb_info;
 struct qed_sb_attn_info;
 struct qed_cxt_mngr;
 struct qed_sb_sp_info;
+struct qed_ll2_info;
 struct qed_mcp_info;
 
 struct qed_rt_data {
@@ -152,6 +153,7 @@ enum QED_RESOURCES {
        QED_MAC,
        QED_VLAN,
        QED_ILT,
+       QED_LL2_QUEUE,
        QED_MAX_RESC,
 };
 
@@ -360,6 +362,8 @@ struct qed_hwfn {
        struct qed_sb_attn_info         *p_sb_attn;
 
        /* Protocol related */
+       bool                            using_ll2;
+       struct qed_ll2_info             *p_ll2_info;
        struct qed_pf_params            pf_params;
 
        bool b_rdma_enabled_in_prs;
@@ -564,6 +568,11 @@ struct qed_dev {
 
        struct qed_dbg_params           dbg_params;
 
+#ifdef CONFIG_QED_LL2
+       struct qed_cb_ll2_info          *ll2;
+       u8                              ll2_mac_address[ETH_ALEN];
+#endif
+
        const struct firmware           *firmware;
 };
 
index dd579b2ef224bb1bbad0db79a0ad8fcc957e86b0..d9bea2a9c9f7249c9f15211852db49653596d0e8 100644 (file)
@@ -1839,6 +1839,8 @@ int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn)
        /* Set the number of required CORE connections */
        u32 core_cids = 1; /* SPQ */
 
+       if (p_hwfn->using_ll2)
+               core_cids += 4;
        qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
 
        switch (p_hwfn->hw_info.personality) {
index 13d8b4075b01253b9caf3bedc0895666582f9581..9a8e153df841beef200540f148d04a6628ab8416 100644 (file)
@@ -29,6 +29,7 @@
 #include "qed_hw.h"
 #include "qed_init_ops.h"
 #include "qed_int.h"
+#include "qed_ll2.h"
 #include "qed_mcp.h"
 #include "qed_reg_addr.h"
 #include "qed_sp.h"
@@ -147,6 +148,9 @@ void qed_resc_free(struct qed_dev *cdev)
                qed_eq_free(p_hwfn, p_hwfn->p_eq);
                qed_consq_free(p_hwfn, p_hwfn->p_consq);
                qed_int_free(p_hwfn);
+#ifdef CONFIG_QED_LL2
+               qed_ll2_free(p_hwfn, p_hwfn->p_ll2_info);
+#endif
                qed_iov_free(p_hwfn);
                qed_dmae_info_free(p_hwfn);
                qed_dcbx_info_free(p_hwfn, p_hwfn->p_dcbx_info);
@@ -403,6 +407,9 @@ int qed_qm_reconf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
 
 int qed_resc_alloc(struct qed_dev *cdev)
 {
+#ifdef CONFIG_QED_LL2
+       struct qed_ll2_info *p_ll2_info;
+#endif
        struct qed_consq *p_consq;
        struct qed_eq *p_eq;
        int i, rc = 0;
@@ -513,6 +520,15 @@ int qed_resc_alloc(struct qed_dev *cdev)
                        goto alloc_no_mem;
                p_hwfn->p_consq = p_consq;
 
+#ifdef CONFIG_QED_LL2
+               if (p_hwfn->using_ll2) {
+                       p_ll2_info = qed_ll2_alloc(p_hwfn);
+                       if (!p_ll2_info)
+                               goto alloc_no_mem;
+                       p_hwfn->p_ll2_info = p_ll2_info;
+               }
+#endif
+
                /* DMA info initialization */
                rc = qed_dmae_info_alloc(p_hwfn);
                if (rc)
@@ -561,6 +577,10 @@ void qed_resc_setup(struct qed_dev *cdev)
                qed_int_setup(p_hwfn, p_hwfn->p_main_ptt);
 
                qed_iov_setup(p_hwfn, p_hwfn->p_main_ptt);
+#ifdef CONFIG_QED_LL2
+               if (p_hwfn->using_ll2)
+                       qed_ll2_setup(p_hwfn, p_hwfn->p_ll2_info);
+#endif
        }
 }
 
@@ -1304,6 +1324,7 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
        resc_num[QED_VLAN] = (ETH_NUM_VLAN_FILTERS - 1 /*For vlan0*/) /
                             num_funcs;
        resc_num[QED_ILT] = PXP_NUM_ILT_RECORDS_BB / num_funcs;
+       resc_num[QED_LL2_QUEUE] = MAX_NUM_LL2_RX_QUEUES / num_funcs;
 
        for (i = 0; i < QED_MAX_RESC; i++)
                resc_start[i] = resc_num[i] * enabled_func_idx;
@@ -1327,7 +1348,8 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
                   "RL = %d start = %d\n"
                   "MAC = %d start = %d\n"
                   "VLAN = %d start = %d\n"
-                  "ILT = %d start = %d\n",
+                  "ILT = %d start = %d\n"
+                  "LL2_QUEUE = %d start = %d\n",
                   p_hwfn->hw_info.resc_num[QED_SB],
                   p_hwfn->hw_info.resc_start[QED_SB],
                   p_hwfn->hw_info.resc_num[QED_L2_QUEUE],
@@ -1343,7 +1365,9 @@ static int qed_hw_get_resc(struct qed_hwfn *p_hwfn)
                   p_hwfn->hw_info.resc_num[QED_VLAN],
                   p_hwfn->hw_info.resc_start[QED_VLAN],
                   p_hwfn->hw_info.resc_num[QED_ILT],
-                  p_hwfn->hw_info.resc_start[QED_ILT]);
+                  p_hwfn->hw_info.resc_start[QED_ILT],
+                  RESC_NUM(p_hwfn, QED_LL2_QUEUE),
+                  RESC_START(p_hwfn, QED_LL2_QUEUE));
 
        return 0;
 }
@@ -2133,6 +2157,98 @@ int qed_fw_rss_eng(struct qed_hwfn *p_hwfn, u8 src_id, u8 *dst_id)
        return 0;
 }
 
+static void qed_llh_mac_to_filter(u32 *p_high, u32 *p_low,
+                                 u8 *p_filter)
+{
+       *p_high = p_filter[1] | (p_filter[0] << 8);
+       *p_low = p_filter[5] | (p_filter[4] << 8) |
+                (p_filter[3] << 16) | (p_filter[2] << 24);
+}
+
+int qed_llh_add_mac_filter(struct qed_hwfn *p_hwfn,
+                          struct qed_ptt *p_ptt, u8 *p_filter)
+{
+       u32 high = 0, low = 0, en;
+       int i;
+
+       if (!(IS_MF_SI(p_hwfn) || IS_MF_DEFAULT(p_hwfn)))
+               return 0;
+
+       qed_llh_mac_to_filter(&high, &low, p_filter);
+
+       /* Find a free entry and utilize it */
+       for (i = 0; i < NIG_REG_LLH_FUNC_FILTER_EN_SIZE; i++) {
+               en = qed_rd(p_hwfn, p_ptt,
+                           NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32));
+               if (en)
+                       continue;
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_VALUE +
+                      2 * i * sizeof(u32), low);
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_VALUE +
+                      (2 * i + 1) * sizeof(u32), high);
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_MODE + i * sizeof(u32), 0);
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE +
+                      i * sizeof(u32), 0);
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32), 1);
+               break;
+       }
+       if (i >= NIG_REG_LLH_FUNC_FILTER_EN_SIZE) {
+               DP_NOTICE(p_hwfn,
+                         "Failed to find an empty LLH filter to utilize\n");
+               return -EINVAL;
+       }
+
+       DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
+                  "mac: %pM is added at %d\n",
+                  p_filter, i);
+
+       return 0;
+}
+
+void qed_llh_remove_mac_filter(struct qed_hwfn *p_hwfn,
+                              struct qed_ptt *p_ptt, u8 *p_filter)
+{
+       u32 high = 0, low = 0;
+       int i;
+
+       if (!(IS_MF_SI(p_hwfn) || IS_MF_DEFAULT(p_hwfn)))
+               return;
+
+       qed_llh_mac_to_filter(&high, &low, p_filter);
+
+       /* Find the entry and clean it */
+       for (i = 0; i < NIG_REG_LLH_FUNC_FILTER_EN_SIZE; i++) {
+               if (qed_rd(p_hwfn, p_ptt,
+                          NIG_REG_LLH_FUNC_FILTER_VALUE +
+                          2 * i * sizeof(u32)) != low)
+                       continue;
+               if (qed_rd(p_hwfn, p_ptt,
+                          NIG_REG_LLH_FUNC_FILTER_VALUE +
+                          (2 * i + 1) * sizeof(u32)) != high)
+                       continue;
+
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_EN + i * sizeof(u32), 0);
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_VALUE + 2 * i * sizeof(u32), 0);
+               qed_wr(p_hwfn, p_ptt,
+                      NIG_REG_LLH_FUNC_FILTER_VALUE +
+                      (2 * i + 1) * sizeof(u32), 0);
+
+               DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
+                          "mac: %pM is removed from %d\n",
+                          p_filter, i);
+               break;
+       }
+       if (i >= NIG_REG_LLH_FUNC_FILTER_EN_SIZE)
+               DP_NOTICE(p_hwfn, "Tried to remove a non-configured filter\n");
+}
+
 static int qed_set_coalesce(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
                            u32 hw_addr, void *p_eth_qzone,
                            size_t eth_qzone_size, u8 timeset)
index 343bb0344f623c53ff8013473ac0470f1a2e81e2..b6711c106597bc53654457c8c45c68e7085f0016 100644 (file)
@@ -309,6 +309,26 @@ int qed_fw_rss_eng(struct qed_hwfn *p_hwfn,
                   u8 src_id,
                   u8 *dst_id);
 
+/**
+ * @brief qed_llh_add_mac_filter - configures a MAC filter in llh
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param p_filter - MAC to add
+ */
+int qed_llh_add_mac_filter(struct qed_hwfn *p_hwfn,
+                          struct qed_ptt *p_ptt, u8 *p_filter);
+
+/**
+ * @brief qed_llh_remove_mac_filter - removes a MAC filter from llh
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param p_filter - MAC to remove
+ */
+void qed_llh_remove_mac_filter(struct qed_hwfn *p_hwfn,
+                              struct qed_ptt *p_ptt, u8 *p_filter);
+
 /**
  * *@brief Cleanup of previous driver remains prior to load
  *
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.c b/drivers/net/ethernet/qlogic/qed/qed_ll2.c
new file mode 100644 (file)
index 0000000..e0ec8ed
--- /dev/null
@@ -0,0 +1,1699 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#include <linux/types.h>
+#include <asm/byteorder.h>
+#include <linux/dma-mapping.h>
+#include <linux/if_vlan.h>
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/stddef.h>
+#include <linux/version.h>
+#include <linux/workqueue.h>
+#include <net/ipv6.h>
+#include <linux/bitops.h>
+#include <linux/delay.h>
+#include <linux/errno.h>
+#include <linux/etherdevice.h>
+#include <linux/io.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/qed/qed_ll2_if.h>
+#include "qed.h"
+#include "qed_cxt.h"
+#include "qed_dev_api.h"
+#include "qed_hsi.h"
+#include "qed_hw.h"
+#include "qed_int.h"
+#include "qed_ll2.h"
+#include "qed_mcp.h"
+#include "qed_reg_addr.h"
+#include "qed_sp.h"
+
+#define QED_LL2_RX_REGISTERED(ll2)     ((ll2)->rx_queue.b_cb_registred)
+#define QED_LL2_TX_REGISTERED(ll2)     ((ll2)->tx_queue.b_cb_registred)
+
+#define QED_LL2_TX_SIZE (256)
+#define QED_LL2_RX_SIZE (4096)
+
+struct qed_cb_ll2_info {
+       int rx_cnt;
+       u32 rx_size;
+       u8 handle;
+       bool frags_mapped;
+
+       /* Lock protecting LL2 buffer lists in sleepless context */
+       spinlock_t lock;
+       struct list_head list;
+
+       const struct qed_ll2_cb_ops *cbs;
+       void *cb_cookie;
+};
+
+struct qed_ll2_buffer {
+       struct list_head list;
+       void *data;
+       dma_addr_t phys_addr;
+};
+
+static void qed_ll2b_complete_tx_packet(struct qed_hwfn *p_hwfn,
+                                       u8 connection_handle,
+                                       void *cookie,
+                                       dma_addr_t first_frag_addr,
+                                       bool b_last_fragment,
+                                       bool b_last_packet)
+{
+       struct qed_dev *cdev = p_hwfn->cdev;
+       struct sk_buff *skb = cookie;
+
+       /* All we need to do is release the mapping */
+       dma_unmap_single(&p_hwfn->cdev->pdev->dev, first_frag_addr,
+                        skb_headlen(skb), DMA_TO_DEVICE);
+
+       if (cdev->ll2->cbs && cdev->ll2->cbs->tx_cb)
+               cdev->ll2->cbs->tx_cb(cdev->ll2->cb_cookie, skb,
+                                     b_last_fragment);
+
+       if (cdev->ll2->frags_mapped)
+               /* Case where mapped frags were received, need to
+                * free skb with nr_frags marked as 0
+                */
+               skb_shinfo(skb)->nr_frags = 0;
+
+       dev_kfree_skb_any(skb);
+}
+
+static int qed_ll2_alloc_buffer(struct qed_dev *cdev,
+                               u8 **data, dma_addr_t *phys_addr)
+{
+       *data = kmalloc(cdev->ll2->rx_size, GFP_ATOMIC);
+       if (!(*data)) {
+               DP_INFO(cdev, "Failed to allocate LL2 buffer data\n");
+               return -ENOMEM;
+       }
+
+       *phys_addr = dma_map_single(&cdev->pdev->dev,
+                                   ((*data) + NET_SKB_PAD),
+                                   cdev->ll2->rx_size, DMA_FROM_DEVICE);
+       if (dma_mapping_error(&cdev->pdev->dev, *phys_addr)) {
+               DP_INFO(cdev, "Failed to map LL2 buffer data\n");
+               kfree((*data));
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+static int qed_ll2_dealloc_buffer(struct qed_dev *cdev,
+                                struct qed_ll2_buffer *buffer)
+{
+       spin_lock_bh(&cdev->ll2->lock);
+
+       dma_unmap_single(&cdev->pdev->dev, buffer->phys_addr,
+                        cdev->ll2->rx_size, DMA_FROM_DEVICE);
+       kfree(buffer->data);
+       list_del(&buffer->list);
+
+       cdev->ll2->rx_cnt--;
+       if (!cdev->ll2->rx_cnt)
+               DP_INFO(cdev, "All LL2 entries were removed\n");
+
+       spin_unlock_bh(&cdev->ll2->lock);
+
+       return 0;
+}
+
+static void qed_ll2_kill_buffers(struct qed_dev *cdev)
+{
+       struct qed_ll2_buffer *buffer, *tmp_buffer;
+
+       list_for_each_entry_safe(buffer, tmp_buffer, &cdev->ll2->list, list)
+               qed_ll2_dealloc_buffer(cdev, buffer);
+}
+
+void qed_ll2b_complete_rx_packet(struct qed_hwfn *p_hwfn,
+                                u8 connection_handle,
+                                struct qed_ll2_rx_packet *p_pkt,
+                                struct core_rx_fast_path_cqe *p_cqe,
+                                bool b_last_packet)
+{
+       u16 packet_length = le16_to_cpu(p_cqe->packet_length);
+       struct qed_ll2_buffer *buffer = p_pkt->cookie;
+       struct qed_dev *cdev = p_hwfn->cdev;
+       u16 vlan = le16_to_cpu(p_cqe->vlan);
+       u32 opaque_data_0, opaque_data_1;
+       u8 pad = p_cqe->placement_offset;
+       dma_addr_t new_phys_addr;
+       struct sk_buff *skb;
+       bool reuse = false;
+       int rc = -EINVAL;
+       u8 *new_data;
+
+       opaque_data_0 = le32_to_cpu(p_cqe->opaque_data.data[0]);
+       opaque_data_1 = le32_to_cpu(p_cqe->opaque_data.data[1]);
+
+       DP_VERBOSE(p_hwfn,
+                  (NETIF_MSG_RX_STATUS | QED_MSG_STORAGE | NETIF_MSG_PKTDATA),
+                  "Got an LL2 Rx completion: [Buffer at phys 0x%llx, offset 0x%02x] Length 0x%04x Parse_flags 0x%04x vlan 0x%04x Opaque data [0x%08x:0x%08x]\n",
+                  (u64)p_pkt->rx_buf_addr, pad, packet_length,
+                  le16_to_cpu(p_cqe->parse_flags.flags), vlan,
+                  opaque_data_0, opaque_data_1);
+
+       if ((cdev->dp_module & NETIF_MSG_PKTDATA) && buffer->data) {
+               print_hex_dump(KERN_INFO, "",
+                              DUMP_PREFIX_OFFSET, 16, 1,
+                              buffer->data, packet_length, false);
+       }
+
+       /* Determine if data is valid */
+       if (packet_length < ETH_HLEN)
+               reuse = true;
+
+       /* Allocate a replacement for buffer; Reuse upon failure */
+       if (!reuse)
+               rc = qed_ll2_alloc_buffer(p_hwfn->cdev, &new_data,
+                                         &new_phys_addr);
+
+       /* If need to reuse or there's no replacement buffer, repost this */
+       if (rc)
+               goto out_post;
+
+       skb = build_skb(buffer->data, 0);
+       if (!skb) {
+               rc = -ENOMEM;
+               goto out_post;
+       }
+
+       pad += NET_SKB_PAD;
+       skb_reserve(skb, pad);
+       skb_put(skb, packet_length);
+       skb_checksum_none_assert(skb);
+
+       /* Get parital ethernet information instead of eth_type_trans(),
+        * Since we don't have an associated net_device.
+        */
+       skb_reset_mac_header(skb);
+       skb->protocol = eth_hdr(skb)->h_proto;
+
+       /* Pass SKB onward */
+       if (cdev->ll2->cbs && cdev->ll2->cbs->rx_cb) {
+               if (vlan)
+                       __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan);
+               cdev->ll2->cbs->rx_cb(cdev->ll2->cb_cookie, skb,
+                                     opaque_data_0, opaque_data_1);
+       }
+
+       /* Update Buffer information and update FW producer */
+       buffer->data = new_data;
+       buffer->phys_addr = new_phys_addr;
+
+out_post:
+       rc = qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev), cdev->ll2->handle,
+                                   buffer->phys_addr, 0,  buffer, 1);
+
+       if (rc)
+               qed_ll2_dealloc_buffer(cdev, buffer);
+}
+
+static struct qed_ll2_info *__qed_ll2_handle_sanity(struct qed_hwfn *p_hwfn,
+                                                   u8 connection_handle,
+                                                   bool b_lock,
+                                                   bool b_only_active)
+{
+       struct qed_ll2_info *p_ll2_conn, *p_ret = NULL;
+
+       if (connection_handle >= QED_MAX_NUM_OF_LL2_CONNECTIONS)
+               return NULL;
+
+       if (!p_hwfn->p_ll2_info)
+               return NULL;
+
+       p_ll2_conn = &p_hwfn->p_ll2_info[connection_handle];
+
+       if (b_only_active) {
+               if (b_lock)
+                       mutex_lock(&p_ll2_conn->mutex);
+               if (p_ll2_conn->b_active)
+                       p_ret = p_ll2_conn;
+               if (b_lock)
+                       mutex_unlock(&p_ll2_conn->mutex);
+       } else {
+               p_ret = p_ll2_conn;
+       }
+
+       return p_ret;
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity(struct qed_hwfn *p_hwfn,
+                                                 u8 connection_handle)
+{
+       return __qed_ll2_handle_sanity(p_hwfn, connection_handle, false, true);
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity_lock(struct qed_hwfn *p_hwfn,
+                                                      u8 connection_handle)
+{
+       return __qed_ll2_handle_sanity(p_hwfn, connection_handle, true, true);
+}
+
+static struct qed_ll2_info *qed_ll2_handle_sanity_inactive(struct qed_hwfn
+                                                          *p_hwfn,
+                                                          u8 connection_handle)
+{
+       return __qed_ll2_handle_sanity(p_hwfn, connection_handle, false, false);
+}
+
+static void qed_ll2_txq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+       bool b_last_packet = false, b_last_frag = false;
+       struct qed_ll2_tx_packet *p_pkt = NULL;
+       struct qed_ll2_info *p_ll2_conn;
+       struct qed_ll2_tx_queue *p_tx;
+
+       p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return;
+
+       p_tx = &p_ll2_conn->tx_queue;
+
+       while (!list_empty(&p_tx->active_descq)) {
+               p_pkt = list_first_entry(&p_tx->active_descq,
+                                        struct qed_ll2_tx_packet, list_entry);
+               if (!p_pkt)
+                       break;
+
+               list_del(&p_pkt->list_entry);
+               b_last_packet = list_empty(&p_tx->active_descq);
+               list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+               p_tx->cur_completing_packet = *p_pkt;
+               p_tx->cur_completing_bd_idx = 1;
+               b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
+
+               qed_ll2b_complete_tx_packet(p_hwfn, p_ll2_conn->my_id,
+                                           p_pkt->cookie,
+                                           p_pkt->bds_set[0].tx_frag,
+                                           b_last_frag, b_last_packet);
+       }
+}
+
+static int qed_ll2_txq_completion(struct qed_hwfn *p_hwfn, void *p_cookie)
+{
+       struct qed_ll2_info *p_ll2_conn = p_cookie;
+       struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+       u16 new_idx = 0, num_bds = 0, num_bds_in_packet = 0;
+       struct qed_ll2_tx_packet *p_pkt;
+       bool b_last_frag = false;
+       unsigned long flags;
+       int rc = -EINVAL;
+
+       spin_lock_irqsave(&p_tx->lock, flags);
+       if (p_tx->b_completing_packet) {
+               rc = -EBUSY;
+               goto out;
+       }
+
+       new_idx = le16_to_cpu(*p_tx->p_fw_cons);
+       num_bds = ((s16)new_idx - (s16)p_tx->bds_idx);
+       while (num_bds) {
+               if (list_empty(&p_tx->active_descq))
+                       goto out;
+
+               p_pkt = list_first_entry(&p_tx->active_descq,
+                                        struct qed_ll2_tx_packet, list_entry);
+               if (!p_pkt)
+                       goto out;
+
+               p_tx->b_completing_packet = true;
+               p_tx->cur_completing_packet = *p_pkt;
+               num_bds_in_packet = p_pkt->bd_used;
+               list_del(&p_pkt->list_entry);
+
+               if (num_bds < num_bds_in_packet) {
+                       DP_NOTICE(p_hwfn,
+                                 "Rest of BDs does not cover whole packet\n");
+                       goto out;
+               }
+
+               num_bds -= num_bds_in_packet;
+               p_tx->bds_idx += num_bds_in_packet;
+               while (num_bds_in_packet--)
+                       qed_chain_consume(&p_tx->txq_chain);
+
+               p_tx->cur_completing_bd_idx = 1;
+               b_last_frag = p_tx->cur_completing_bd_idx == p_pkt->bd_used;
+               list_add_tail(&p_pkt->list_entry, &p_tx->free_descq);
+
+               spin_unlock_irqrestore(&p_tx->lock, flags);
+               qed_ll2b_complete_tx_packet(p_hwfn,
+                                           p_ll2_conn->my_id,
+                                           p_pkt->cookie,
+                                           p_pkt->bds_set[0].tx_frag,
+                                           b_last_frag, !num_bds);
+               spin_lock_irqsave(&p_tx->lock, flags);
+       }
+
+       p_tx->b_completing_packet = false;
+       rc = 0;
+out:
+       spin_unlock_irqrestore(&p_tx->lock, flags);
+       return rc;
+}
+
+static int qed_ll2_rxq_completion_reg(struct qed_hwfn *p_hwfn,
+                                     struct qed_ll2_info *p_ll2_conn,
+                                     union core_rx_cqe_union *p_cqe,
+                                     unsigned long lock_flags,
+                                     bool b_last_cqe)
+{
+       struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+       struct qed_ll2_rx_packet *p_pkt = NULL;
+
+       if (!list_empty(&p_rx->active_descq))
+               p_pkt = list_first_entry(&p_rx->active_descq,
+                                        struct qed_ll2_rx_packet, list_entry);
+       if (!p_pkt) {
+               DP_NOTICE(p_hwfn,
+                         "LL2 Rx completion but active_descq is empty\n");
+               return -EIO;
+       }
+       list_del(&p_pkt->list_entry);
+
+       if (qed_chain_consume(&p_rx->rxq_chain) != p_pkt->rxq_bd)
+               DP_NOTICE(p_hwfn,
+                         "Mismatch between active_descq and the LL2 Rx chain\n");
+       list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+       spin_unlock_irqrestore(&p_rx->lock, lock_flags);
+       qed_ll2b_complete_rx_packet(p_hwfn, p_ll2_conn->my_id,
+                                   p_pkt, &p_cqe->rx_cqe_fp, b_last_cqe);
+       spin_lock_irqsave(&p_rx->lock, lock_flags);
+
+       return 0;
+}
+
+static int qed_ll2_rxq_completion(struct qed_hwfn *p_hwfn, void *cookie)
+{
+       struct qed_ll2_info *p_ll2_conn = cookie;
+       struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+       union core_rx_cqe_union *cqe = NULL;
+       u16 cq_new_idx = 0, cq_old_idx = 0;
+       unsigned long flags = 0;
+       int rc = 0;
+
+       spin_lock_irqsave(&p_rx->lock, flags);
+       cq_new_idx = le16_to_cpu(*p_rx->p_fw_cons);
+       cq_old_idx = qed_chain_get_cons_idx(&p_rx->rcq_chain);
+
+       while (cq_new_idx != cq_old_idx) {
+               bool b_last_cqe = (cq_new_idx == cq_old_idx);
+
+               cqe = qed_chain_consume(&p_rx->rcq_chain);
+               cq_old_idx = qed_chain_get_cons_idx(&p_rx->rcq_chain);
+
+               DP_VERBOSE(p_hwfn,
+                          QED_MSG_LL2,
+                          "LL2 [sw. cons %04x, fw. at %04x] - Got Packet of type %02x\n",
+                          cq_old_idx, cq_new_idx, cqe->rx_cqe_sp.type);
+
+               switch (cqe->rx_cqe_sp.type) {
+               case CORE_RX_CQE_TYPE_SLOW_PATH:
+                       DP_NOTICE(p_hwfn, "LL2 - unexpected Rx CQE slowpath\n");
+                       rc = -EINVAL;
+                       break;
+               case CORE_RX_CQE_TYPE_REGULAR:
+                       rc = qed_ll2_rxq_completion_reg(p_hwfn, p_ll2_conn,
+                                                       cqe, flags, b_last_cqe);
+                       break;
+               default:
+                       rc = -EIO;
+               }
+       }
+
+       spin_unlock_irqrestore(&p_rx->lock, flags);
+       return rc;
+}
+
+void qed_ll2_rxq_flush(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+       struct qed_ll2_info *p_ll2_conn = NULL;
+       struct qed_ll2_rx_packet *p_pkt = NULL;
+       struct qed_ll2_rx_queue *p_rx;
+
+       p_ll2_conn = qed_ll2_handle_sanity_inactive(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return;
+
+       p_rx = &p_ll2_conn->rx_queue;
+
+       while (!list_empty(&p_rx->active_descq)) {
+               dma_addr_t rx_buf_addr;
+               void *cookie;
+               bool b_last;
+
+               p_pkt = list_first_entry(&p_rx->active_descq,
+                                        struct qed_ll2_rx_packet, list_entry);
+               if (!p_pkt)
+                       break;
+
+               list_del(&p_pkt->list_entry);
+               list_add_tail(&p_pkt->list_entry, &p_rx->free_descq);
+
+               rx_buf_addr = p_pkt->rx_buf_addr;
+               cookie = p_pkt->cookie;
+
+               b_last = list_empty(&p_rx->active_descq);
+       }
+}
+
+static int qed_sp_ll2_rx_queue_start(struct qed_hwfn *p_hwfn,
+                                    struct qed_ll2_info *p_ll2_conn,
+                                    u8 action_on_error)
+{
+       enum qed_ll2_conn_type conn_type = p_ll2_conn->conn_type;
+       struct qed_ll2_rx_queue *p_rx = &p_ll2_conn->rx_queue;
+       struct core_rx_start_ramrod_data *p_ramrod = NULL;
+       struct qed_spq_entry *p_ent = NULL;
+       struct qed_sp_init_data init_data;
+       u16 cqe_pbl_size;
+       int rc = 0;
+
+       /* Get SPQ entry */
+       memset(&init_data, 0, sizeof(init_data));
+       init_data.cid = p_ll2_conn->cid;
+       init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+       init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+       rc = qed_sp_init_request(p_hwfn, &p_ent,
+                                CORE_RAMROD_RX_QUEUE_START,
+                                PROTOCOLID_CORE, &init_data);
+       if (rc)
+               return rc;
+
+       p_ramrod = &p_ent->ramrod.core_rx_queue_start;
+
+       p_ramrod->sb_id = cpu_to_le16(qed_int_get_sp_sb_id(p_hwfn));
+       p_ramrod->sb_index = p_rx->rx_sb_index;
+       p_ramrod->complete_event_flg = 1;
+
+       p_ramrod->mtu = cpu_to_le16(p_ll2_conn->mtu);
+       DMA_REGPAIR_LE(p_ramrod->bd_base,
+                      p_rx->rxq_chain.p_phys_addr);
+       cqe_pbl_size = (u16)qed_chain_get_page_cnt(&p_rx->rcq_chain);
+       p_ramrod->num_of_pbl_pages = cpu_to_le16(cqe_pbl_size);
+       DMA_REGPAIR_LE(p_ramrod->cqe_pbl_addr,
+                      qed_chain_get_pbl_phys(&p_rx->rcq_chain));
+
+       p_ramrod->drop_ttl0_flg = p_ll2_conn->rx_drop_ttl0_flg;
+       p_ramrod->inner_vlan_removal_en = p_ll2_conn->rx_vlan_removal_en;
+       p_ramrod->queue_id = p_ll2_conn->queue_id;
+       p_ramrod->main_func_queue = 1;
+
+       if ((IS_MF_DEFAULT(p_hwfn) || IS_MF_SI(p_hwfn)) &&
+           p_ramrod->main_func_queue && (conn_type != QED_LL2_TYPE_ROCE)) {
+               p_ramrod->mf_si_bcast_accept_all = 1;
+               p_ramrod->mf_si_mcast_accept_all = 1;
+       } else {
+               p_ramrod->mf_si_bcast_accept_all = 0;
+               p_ramrod->mf_si_mcast_accept_all = 0;
+       }
+
+       p_ramrod->action_on_error.error_type = action_on_error;
+       return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_tx_queue_start(struct qed_hwfn *p_hwfn,
+                                    struct qed_ll2_info *p_ll2_conn)
+{
+       enum qed_ll2_conn_type conn_type = p_ll2_conn->conn_type;
+       struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+       struct core_tx_start_ramrod_data *p_ramrod = NULL;
+       struct qed_spq_entry *p_ent = NULL;
+       struct qed_sp_init_data init_data;
+       union qed_qm_pq_params pq_params;
+       u16 pq_id = 0, pbl_size;
+       int rc = -EINVAL;
+
+       if (!QED_LL2_TX_REGISTERED(p_ll2_conn))
+               return 0;
+
+       /* Get SPQ entry */
+       memset(&init_data, 0, sizeof(init_data));
+       init_data.cid = p_ll2_conn->cid;
+       init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+       init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+       rc = qed_sp_init_request(p_hwfn, &p_ent,
+                                CORE_RAMROD_TX_QUEUE_START,
+                                PROTOCOLID_CORE, &init_data);
+       if (rc)
+               return rc;
+
+       p_ramrod = &p_ent->ramrod.core_tx_queue_start;
+
+       p_ramrod->sb_id = cpu_to_le16(qed_int_get_sp_sb_id(p_hwfn));
+       p_ramrod->sb_index = p_tx->tx_sb_index;
+       p_ramrod->mtu = cpu_to_le16(p_ll2_conn->mtu);
+       p_ll2_conn->tx_stats_en = 1;
+       p_ramrod->stats_en = p_ll2_conn->tx_stats_en;
+       p_ramrod->stats_id = p_ll2_conn->tx_stats_id;
+
+       DMA_REGPAIR_LE(p_ramrod->pbl_base_addr,
+                      qed_chain_get_pbl_phys(&p_tx->txq_chain));
+       pbl_size = qed_chain_get_page_cnt(&p_tx->txq_chain);
+       p_ramrod->pbl_size = cpu_to_le16(pbl_size);
+
+       memset(&pq_params, 0, sizeof(pq_params));
+       pq_params.core.tc = p_ll2_conn->tx_tc;
+       pq_id = qed_get_qm_pq(p_hwfn, PROTOCOLID_CORE, &pq_params);
+       p_ramrod->qm_pq_id = cpu_to_le16(pq_id);
+
+       switch (conn_type) {
+       case QED_LL2_TYPE_ISCSI:
+       case QED_LL2_TYPE_ISCSI_OOO:
+               p_ramrod->conn_type = PROTOCOLID_ISCSI;
+               break;
+       case QED_LL2_TYPE_ROCE:
+               p_ramrod->conn_type = PROTOCOLID_ROCE;
+               break;
+       default:
+               p_ramrod->conn_type = PROTOCOLID_ETH;
+               DP_NOTICE(p_hwfn, "Unknown connection type: %d\n", conn_type);
+       }
+
+       return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_rx_queue_stop(struct qed_hwfn *p_hwfn,
+                                   struct qed_ll2_info *p_ll2_conn)
+{
+       struct core_rx_stop_ramrod_data *p_ramrod = NULL;
+       struct qed_spq_entry *p_ent = NULL;
+       struct qed_sp_init_data init_data;
+       int rc = -EINVAL;
+
+       /* Get SPQ entry */
+       memset(&init_data, 0, sizeof(init_data));
+       init_data.cid = p_ll2_conn->cid;
+       init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+       init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+       rc = qed_sp_init_request(p_hwfn, &p_ent,
+                                CORE_RAMROD_RX_QUEUE_STOP,
+                                PROTOCOLID_CORE, &init_data);
+       if (rc)
+               return rc;
+
+       p_ramrod = &p_ent->ramrod.core_rx_queue_stop;
+
+       p_ramrod->complete_event_flg = 1;
+       p_ramrod->queue_id = p_ll2_conn->queue_id;
+
+       return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int qed_sp_ll2_tx_queue_stop(struct qed_hwfn *p_hwfn,
+                                   struct qed_ll2_info *p_ll2_conn)
+{
+       struct qed_spq_entry *p_ent = NULL;
+       struct qed_sp_init_data init_data;
+       int rc = -EINVAL;
+
+       /* Get SPQ entry */
+       memset(&init_data, 0, sizeof(init_data));
+       init_data.cid = p_ll2_conn->cid;
+       init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+       init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+       rc = qed_sp_init_request(p_hwfn, &p_ent,
+                                CORE_RAMROD_TX_QUEUE_STOP,
+                                PROTOCOLID_CORE, &init_data);
+       if (rc)
+               return rc;
+
+       return qed_spq_post(p_hwfn, p_ent, NULL);
+}
+
+static int
+qed_ll2_acquire_connection_rx(struct qed_hwfn *p_hwfn,
+                             struct qed_ll2_info *p_ll2_info, u16 rx_num_desc)
+{
+       struct qed_ll2_rx_packet *p_descq;
+       u32 capacity;
+       int rc = 0;
+
+       if (!rx_num_desc)
+               goto out;
+
+       rc = qed_chain_alloc(p_hwfn->cdev,
+                            QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+                            QED_CHAIN_MODE_NEXT_PTR,
+                            QED_CHAIN_CNT_TYPE_U16,
+                            rx_num_desc,
+                            sizeof(struct core_rx_bd),
+                            &p_ll2_info->rx_queue.rxq_chain);
+       if (rc) {
+               DP_NOTICE(p_hwfn, "Failed to allocate ll2 rxq chain\n");
+               goto out;
+       }
+
+       capacity = qed_chain_get_capacity(&p_ll2_info->rx_queue.rxq_chain);
+       p_descq = kcalloc(capacity, sizeof(struct qed_ll2_rx_packet),
+                         GFP_KERNEL);
+       if (!p_descq) {
+               rc = -ENOMEM;
+               DP_NOTICE(p_hwfn, "Failed to allocate ll2 Rx desc\n");
+               goto out;
+       }
+       p_ll2_info->rx_queue.descq_array = p_descq;
+
+       rc = qed_chain_alloc(p_hwfn->cdev,
+                            QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+                            QED_CHAIN_MODE_PBL,
+                            QED_CHAIN_CNT_TYPE_U16,
+                            rx_num_desc,
+                            sizeof(struct core_rx_fast_path_cqe),
+                            &p_ll2_info->rx_queue.rcq_chain);
+       if (rc) {
+               DP_NOTICE(p_hwfn, "Failed to allocate ll2 rcq chain\n");
+               goto out;
+       }
+
+       DP_VERBOSE(p_hwfn, QED_MSG_LL2,
+                  "Allocated LL2 Rxq [Type %08x] with 0x%08x buffers\n",
+                  p_ll2_info->conn_type, rx_num_desc);
+
+out:
+       return rc;
+}
+
+static int qed_ll2_acquire_connection_tx(struct qed_hwfn *p_hwfn,
+                                        struct qed_ll2_info *p_ll2_info,
+                                        u16 tx_num_desc)
+{
+       struct qed_ll2_tx_packet *p_descq;
+       u32 capacity;
+       int rc = 0;
+
+       if (!tx_num_desc)
+               goto out;
+
+       rc = qed_chain_alloc(p_hwfn->cdev,
+                            QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+                            QED_CHAIN_MODE_PBL,
+                            QED_CHAIN_CNT_TYPE_U16,
+                            tx_num_desc,
+                            sizeof(struct core_tx_bd),
+                            &p_ll2_info->tx_queue.txq_chain);
+       if (rc)
+               goto out;
+
+       capacity = qed_chain_get_capacity(&p_ll2_info->tx_queue.txq_chain);
+       p_descq = kcalloc(capacity, sizeof(struct qed_ll2_tx_packet),
+                         GFP_KERNEL);
+       if (!p_descq) {
+               rc = -ENOMEM;
+               goto out;
+       }
+       p_ll2_info->tx_queue.descq_array = p_descq;
+
+       DP_VERBOSE(p_hwfn, QED_MSG_LL2,
+                  "Allocated LL2 Txq [Type %08x] with 0x%08x buffers\n",
+                  p_ll2_info->conn_type, tx_num_desc);
+
+out:
+       if (rc)
+               DP_NOTICE(p_hwfn,
+                         "Can't allocate memory for Tx LL2 with 0x%08x buffers\n",
+                         tx_num_desc);
+       return rc;
+}
+
+int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
+                              struct qed_ll2_info *p_params,
+                              u16 rx_num_desc,
+                              u16 tx_num_desc,
+                              u8 *p_connection_handle)
+{
+       qed_int_comp_cb_t comp_rx_cb, comp_tx_cb;
+       struct qed_ll2_info *p_ll2_info = NULL;
+       int rc;
+       u8 i;
+
+       if (!p_connection_handle || !p_hwfn->p_ll2_info)
+               return -EINVAL;
+
+       /* Find a free connection to be used */
+       for (i = 0; (i < QED_MAX_NUM_OF_LL2_CONNECTIONS); i++) {
+               mutex_lock(&p_hwfn->p_ll2_info[i].mutex);
+               if (p_hwfn->p_ll2_info[i].b_active) {
+                       mutex_unlock(&p_hwfn->p_ll2_info[i].mutex);
+                       continue;
+               }
+
+               p_hwfn->p_ll2_info[i].b_active = true;
+               p_ll2_info = &p_hwfn->p_ll2_info[i];
+               mutex_unlock(&p_hwfn->p_ll2_info[i].mutex);
+               break;
+       }
+       if (!p_ll2_info)
+               return -EBUSY;
+
+       p_ll2_info->conn_type = p_params->conn_type;
+       p_ll2_info->mtu = p_params->mtu;
+       p_ll2_info->rx_drop_ttl0_flg = p_params->rx_drop_ttl0_flg;
+       p_ll2_info->rx_vlan_removal_en = p_params->rx_vlan_removal_en;
+       p_ll2_info->tx_tc = p_params->tx_tc;
+       p_ll2_info->tx_dest = p_params->tx_dest;
+       p_ll2_info->ai_err_packet_too_big = p_params->ai_err_packet_too_big;
+       p_ll2_info->ai_err_no_buf = p_params->ai_err_no_buf;
+
+       rc = qed_ll2_acquire_connection_rx(p_hwfn, p_ll2_info, rx_num_desc);
+       if (rc)
+               goto q_allocate_fail;
+
+       rc = qed_ll2_acquire_connection_tx(p_hwfn, p_ll2_info, tx_num_desc);
+       if (rc)
+               goto q_allocate_fail;
+
+       /* Register callbacks for the Rx/Tx queues */
+       comp_rx_cb = qed_ll2_rxq_completion;
+       comp_tx_cb = qed_ll2_txq_completion;
+
+       if (rx_num_desc) {
+               qed_int_register_cb(p_hwfn, comp_rx_cb,
+                                   &p_hwfn->p_ll2_info[i],
+                                   &p_ll2_info->rx_queue.rx_sb_index,
+                                   &p_ll2_info->rx_queue.p_fw_cons);
+               p_ll2_info->rx_queue.b_cb_registred = true;
+       }
+
+       if (tx_num_desc) {
+               qed_int_register_cb(p_hwfn,
+                                   comp_tx_cb,
+                                   &p_hwfn->p_ll2_info[i],
+                                   &p_ll2_info->tx_queue.tx_sb_index,
+                                   &p_ll2_info->tx_queue.p_fw_cons);
+               p_ll2_info->tx_queue.b_cb_registred = true;
+       }
+
+       *p_connection_handle = i;
+       return rc;
+
+q_allocate_fail:
+       qed_ll2_release_connection(p_hwfn, i);
+       return -ENOMEM;
+}
+
+static int qed_ll2_establish_connection_rx(struct qed_hwfn *p_hwfn,
+                                          struct qed_ll2_info *p_ll2_conn)
+{
+       u8 action_on_error = 0;
+
+       if (!QED_LL2_RX_REGISTERED(p_ll2_conn))
+               return 0;
+
+       DIRECT_REG_WR(p_ll2_conn->rx_queue.set_prod_addr, 0x0);
+
+       SET_FIELD(action_on_error,
+                 CORE_RX_ACTION_ON_ERROR_PACKET_TOO_BIG,
+                 p_ll2_conn->ai_err_packet_too_big);
+       SET_FIELD(action_on_error,
+                 CORE_RX_ACTION_ON_ERROR_NO_BUFF, p_ll2_conn->ai_err_no_buf);
+
+       return qed_sp_ll2_rx_queue_start(p_hwfn, p_ll2_conn, action_on_error);
+}
+
+int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+       struct qed_ll2_info *p_ll2_conn;
+       struct qed_ll2_rx_queue *p_rx;
+       struct qed_ll2_tx_queue *p_tx;
+       int rc = -EINVAL;
+       u32 i, capacity;
+       u8 qid;
+
+       p_ll2_conn = qed_ll2_handle_sanity_lock(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return -EINVAL;
+       p_rx = &p_ll2_conn->rx_queue;
+       p_tx = &p_ll2_conn->tx_queue;
+
+       qed_chain_reset(&p_rx->rxq_chain);
+       qed_chain_reset(&p_rx->rcq_chain);
+       INIT_LIST_HEAD(&p_rx->active_descq);
+       INIT_LIST_HEAD(&p_rx->free_descq);
+       INIT_LIST_HEAD(&p_rx->posting_descq);
+       spin_lock_init(&p_rx->lock);
+       capacity = qed_chain_get_capacity(&p_rx->rxq_chain);
+       for (i = 0; i < capacity; i++)
+               list_add_tail(&p_rx->descq_array[i].list_entry,
+                             &p_rx->free_descq);
+       *p_rx->p_fw_cons = 0;
+
+       qed_chain_reset(&p_tx->txq_chain);
+       INIT_LIST_HEAD(&p_tx->active_descq);
+       INIT_LIST_HEAD(&p_tx->free_descq);
+       INIT_LIST_HEAD(&p_tx->sending_descq);
+       spin_lock_init(&p_tx->lock);
+       capacity = qed_chain_get_capacity(&p_tx->txq_chain);
+       for (i = 0; i < capacity; i++)
+               list_add_tail(&p_tx->descq_array[i].list_entry,
+                             &p_tx->free_descq);
+       p_tx->cur_completing_bd_idx = 0;
+       p_tx->bds_idx = 0;
+       p_tx->b_completing_packet = false;
+       p_tx->cur_send_packet = NULL;
+       p_tx->cur_send_frag_num = 0;
+       p_tx->cur_completing_frag_num = 0;
+       *p_tx->p_fw_cons = 0;
+
+       qed_cxt_acquire_cid(p_hwfn, PROTOCOLID_CORE, &p_ll2_conn->cid);
+
+       qid = p_hwfn->hw_info.resc_start[QED_LL2_QUEUE] + connection_handle;
+       p_ll2_conn->queue_id = qid;
+       p_ll2_conn->tx_stats_id = qid;
+       p_rx->set_prod_addr = (u8 __iomem *)p_hwfn->regview +
+                                           GTT_BAR0_MAP_REG_TSDM_RAM +
+                                           TSTORM_LL2_RX_PRODS_OFFSET(qid);
+       p_tx->doorbell_addr = (u8 __iomem *)p_hwfn->doorbells +
+                                           qed_db_addr(p_ll2_conn->cid,
+                                                       DQ_DEMS_LEGACY);
+
+       rc = qed_ll2_establish_connection_rx(p_hwfn, p_ll2_conn);
+       if (rc)
+               return rc;
+
+       rc = qed_sp_ll2_tx_queue_start(p_hwfn, p_ll2_conn);
+       if (rc)
+               return rc;
+
+       if (p_hwfn->hw_info.personality != QED_PCI_ETH_ROCE)
+               qed_wr(p_hwfn, p_hwfn->p_main_ptt, PRS_REG_USE_LIGHT_L2, 1);
+
+       return rc;
+}
+
+static void qed_ll2_post_rx_buffer_notify_fw(struct qed_hwfn *p_hwfn,
+                                            struct qed_ll2_rx_queue *p_rx,
+                                            struct qed_ll2_rx_packet *p_curp)
+{
+       struct qed_ll2_rx_packet *p_posting_packet = NULL;
+       struct core_ll2_rx_prod rx_prod = { 0, 0, 0 };
+       bool b_notify_fw = false;
+       u16 bd_prod, cq_prod;
+
+       /* This handles the flushing of already posted buffers */
+       while (!list_empty(&p_rx->posting_descq)) {
+               p_posting_packet = list_first_entry(&p_rx->posting_descq,
+                                                   struct qed_ll2_rx_packet,
+                                                   list_entry);
+               list_del(&p_posting_packet->list_entry);
+               list_add_tail(&p_posting_packet->list_entry,
+                             &p_rx->active_descq);
+               b_notify_fw = true;
+       }
+
+       /* This handles the supplied packet [if there is one] */
+       if (p_curp) {
+               list_add_tail(&p_curp->list_entry, &p_rx->active_descq);
+               b_notify_fw = true;
+       }
+
+       if (!b_notify_fw)
+               return;
+
+       bd_prod = qed_chain_get_prod_idx(&p_rx->rxq_chain);
+       cq_prod = qed_chain_get_prod_idx(&p_rx->rcq_chain);
+       rx_prod.bd_prod = cpu_to_le16(bd_prod);
+       rx_prod.cqe_prod = cpu_to_le16(cq_prod);
+       DIRECT_REG_WR(p_rx->set_prod_addr, *((u32 *)&rx_prod));
+}
+
+int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+                          u8 connection_handle,
+                          dma_addr_t addr,
+                          u16 buf_len, void *cookie, u8 notify_fw)
+{
+       struct core_rx_bd_with_buff_len *p_curb = NULL;
+       struct qed_ll2_rx_packet *p_curp = NULL;
+       struct qed_ll2_info *p_ll2_conn;
+       struct qed_ll2_rx_queue *p_rx;
+       unsigned long flags;
+       void *p_data;
+       int rc = 0;
+
+       p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return -EINVAL;
+       p_rx = &p_ll2_conn->rx_queue;
+
+       spin_lock_irqsave(&p_rx->lock, flags);
+       if (!list_empty(&p_rx->free_descq))
+               p_curp = list_first_entry(&p_rx->free_descq,
+                                         struct qed_ll2_rx_packet, list_entry);
+       if (p_curp) {
+               if (qed_chain_get_elem_left(&p_rx->rxq_chain) &&
+                   qed_chain_get_elem_left(&p_rx->rcq_chain)) {
+                       p_data = qed_chain_produce(&p_rx->rxq_chain);
+                       p_curb = (struct core_rx_bd_with_buff_len *)p_data;
+                       qed_chain_produce(&p_rx->rcq_chain);
+               }
+       }
+
+       /* If we're lacking entires, let's try to flush buffers to FW */
+       if (!p_curp || !p_curb) {
+               rc = -EBUSY;
+               p_curp = NULL;
+               goto out_notify;
+       }
+
+       /* We have an Rx packet we can fill */
+       DMA_REGPAIR_LE(p_curb->addr, addr);
+       p_curb->buff_length = cpu_to_le16(buf_len);
+       p_curp->rx_buf_addr = addr;
+       p_curp->cookie = cookie;
+       p_curp->rxq_bd = p_curb;
+       p_curp->buf_length = buf_len;
+       list_del(&p_curp->list_entry);
+
+       /* Check if we only want to enqueue this packet without informing FW */
+       if (!notify_fw) {
+               list_add_tail(&p_curp->list_entry, &p_rx->posting_descq);
+               goto out;
+       }
+
+out_notify:
+       qed_ll2_post_rx_buffer_notify_fw(p_hwfn, p_rx, p_curp);
+out:
+       spin_unlock_irqrestore(&p_rx->lock, flags);
+       return rc;
+}
+
+static void qed_ll2_prepare_tx_packet_set(struct qed_hwfn *p_hwfn,
+                                         struct qed_ll2_tx_queue *p_tx,
+                                         struct qed_ll2_tx_packet *p_curp,
+                                         u8 num_of_bds,
+                                         dma_addr_t first_frag,
+                                         u16 first_frag_len, void *p_cookie,
+                                         u8 notify_fw)
+{
+       list_del(&p_curp->list_entry);
+       p_curp->cookie = p_cookie;
+       p_curp->bd_used = num_of_bds;
+       p_curp->notify_fw = notify_fw;
+       p_tx->cur_send_packet = p_curp;
+       p_tx->cur_send_frag_num = 0;
+
+       p_curp->bds_set[p_tx->cur_send_frag_num].tx_frag = first_frag;
+       p_curp->bds_set[p_tx->cur_send_frag_num].frag_len = first_frag_len;
+       p_tx->cur_send_frag_num++;
+}
+
+static void qed_ll2_prepare_tx_packet_set_bd(struct qed_hwfn *p_hwfn,
+                                            struct qed_ll2_info *p_ll2,
+                                            struct qed_ll2_tx_packet *p_curp,
+                                            u8 num_of_bds,
+                                            enum core_tx_dest tx_dest,
+                                            u16 vlan,
+                                            u8 bd_flags,
+                                            u16 l4_hdr_offset_w,
+                                            dma_addr_t first_frag,
+                                            u16 first_frag_len)
+{
+       struct qed_chain *p_tx_chain = &p_ll2->tx_queue.txq_chain;
+       u16 prod_idx = qed_chain_get_prod_idx(p_tx_chain);
+       struct core_tx_bd *start_bd = NULL;
+       u16 frag_idx;
+
+       start_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
+       start_bd->nw_vlan_or_lb_echo = cpu_to_le16(vlan);
+       SET_FIELD(start_bd->bitfield1, CORE_TX_BD_L4_HDR_OFFSET_W,
+                 cpu_to_le16(l4_hdr_offset_w));
+       SET_FIELD(start_bd->bitfield1, CORE_TX_BD_TX_DST, tx_dest);
+       start_bd->bd_flags.as_bitfield = bd_flags;
+       start_bd->bd_flags.as_bitfield |= CORE_TX_BD_FLAGS_START_BD_MASK <<
+           CORE_TX_BD_FLAGS_START_BD_SHIFT;
+       SET_FIELD(start_bd->bitfield0, CORE_TX_BD_NBDS, num_of_bds);
+       DMA_REGPAIR_LE(start_bd->addr, first_frag);
+       start_bd->nbytes = cpu_to_le16(first_frag_len);
+
+       DP_VERBOSE(p_hwfn,
+                  (NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
+                  "LL2 [q 0x%02x cid 0x%08x type 0x%08x] Tx Producer at [0x%04x] - set with a %04x bytes %02x BDs buffer at %08x:%08x\n",
+                  p_ll2->queue_id,
+                  p_ll2->cid,
+                  p_ll2->conn_type,
+                  prod_idx,
+                  first_frag_len,
+                  num_of_bds,
+                  le32_to_cpu(start_bd->addr.hi),
+                  le32_to_cpu(start_bd->addr.lo));
+
+       if (p_ll2->tx_queue.cur_send_frag_num == num_of_bds)
+               return;
+
+       /* Need to provide the packet with additional BDs for frags */
+       for (frag_idx = p_ll2->tx_queue.cur_send_frag_num;
+            frag_idx < num_of_bds; frag_idx++) {
+               struct core_tx_bd **p_bd = &p_curp->bds_set[frag_idx].txq_bd;
+
+               *p_bd = (struct core_tx_bd *)qed_chain_produce(p_tx_chain);
+               (*p_bd)->bd_flags.as_bitfield = 0;
+               (*p_bd)->bitfield1 = 0;
+               (*p_bd)->bitfield0 = 0;
+               p_curp->bds_set[frag_idx].tx_frag = 0;
+               p_curp->bds_set[frag_idx].frag_len = 0;
+       }
+}
+
+/* This should be called while the Txq spinlock is being held */
+static void qed_ll2_tx_packet_notify(struct qed_hwfn *p_hwfn,
+                                    struct qed_ll2_info *p_ll2_conn)
+{
+       bool b_notify = p_ll2_conn->tx_queue.cur_send_packet->notify_fw;
+       struct qed_ll2_tx_queue *p_tx = &p_ll2_conn->tx_queue;
+       struct qed_ll2_tx_packet *p_pkt = NULL;
+       struct core_db_data db_msg = { 0, 0, 0 };
+       u16 bd_prod;
+
+       /* If there are missing BDs, don't do anything now */
+       if (p_ll2_conn->tx_queue.cur_send_frag_num !=
+           p_ll2_conn->tx_queue.cur_send_packet->bd_used)
+               return;
+
+       /* Push the current packet to the list and clean after it */
+       list_add_tail(&p_ll2_conn->tx_queue.cur_send_packet->list_entry,
+                     &p_ll2_conn->tx_queue.sending_descq);
+       p_ll2_conn->tx_queue.cur_send_packet = NULL;
+       p_ll2_conn->tx_queue.cur_send_frag_num = 0;
+
+       /* Notify FW of packet only if requested to */
+       if (!b_notify)
+               return;
+
+       bd_prod = qed_chain_get_prod_idx(&p_ll2_conn->tx_queue.txq_chain);
+
+       while (!list_empty(&p_tx->sending_descq)) {
+               p_pkt = list_first_entry(&p_tx->sending_descq,
+                                        struct qed_ll2_tx_packet, list_entry);
+               if (!p_pkt)
+                       break;
+
+               list_del(&p_pkt->list_entry);
+               list_add_tail(&p_pkt->list_entry, &p_tx->active_descq);
+       }
+
+       SET_FIELD(db_msg.params, CORE_DB_DATA_DEST, DB_DEST_XCM);
+       SET_FIELD(db_msg.params, CORE_DB_DATA_AGG_CMD, DB_AGG_CMD_SET);
+       SET_FIELD(db_msg.params, CORE_DB_DATA_AGG_VAL_SEL,
+                 DQ_XCM_CORE_TX_BD_PROD_CMD);
+       db_msg.agg_flags = DQ_XCM_CORE_DQ_CF_CMD;
+       db_msg.spq_prod = cpu_to_le16(bd_prod);
+
+       /* Make sure the BDs data is updated before ringing the doorbell */
+       wmb();
+
+       DIRECT_REG_WR(p_tx->doorbell_addr, *((u32 *)&db_msg));
+
+       DP_VERBOSE(p_hwfn,
+                  (NETIF_MSG_TX_QUEUED | QED_MSG_LL2),
+                  "LL2 [q 0x%02x cid 0x%08x type 0x%08x] Doorbelled [producer 0x%04x]\n",
+                  p_ll2_conn->queue_id,
+                  p_ll2_conn->cid, p_ll2_conn->conn_type, db_msg.spq_prod);
+}
+
+int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+                             u8 connection_handle,
+                             u8 num_of_bds,
+                             u16 vlan,
+                             u8 bd_flags,
+                             u16 l4_hdr_offset_w,
+                             dma_addr_t first_frag,
+                             u16 first_frag_len, void *cookie, u8 notify_fw)
+{
+       struct qed_ll2_tx_packet *p_curp = NULL;
+       struct qed_ll2_info *p_ll2_conn = NULL;
+       struct qed_ll2_tx_queue *p_tx;
+       struct qed_chain *p_tx_chain;
+       unsigned long flags;
+       int rc = 0;
+
+       p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return -EINVAL;
+       p_tx = &p_ll2_conn->tx_queue;
+       p_tx_chain = &p_tx->txq_chain;
+
+       if (num_of_bds > CORE_LL2_TX_MAX_BDS_PER_PACKET)
+               return -EIO;
+
+       spin_lock_irqsave(&p_tx->lock, flags);
+       if (p_tx->cur_send_packet) {
+               rc = -EEXIST;
+               goto out;
+       }
+
+       /* Get entry, but only if we have tx elements for it */
+       if (!list_empty(&p_tx->free_descq))
+               p_curp = list_first_entry(&p_tx->free_descq,
+                                         struct qed_ll2_tx_packet, list_entry);
+       if (p_curp && qed_chain_get_elem_left(p_tx_chain) < num_of_bds)
+               p_curp = NULL;
+
+       if (!p_curp) {
+               rc = -EBUSY;
+               goto out;
+       }
+
+       /* Prepare packet and BD, and perhaps send a doorbell to FW */
+       qed_ll2_prepare_tx_packet_set(p_hwfn, p_tx, p_curp,
+                                     num_of_bds, first_frag,
+                                     first_frag_len, cookie, notify_fw);
+       qed_ll2_prepare_tx_packet_set_bd(p_hwfn, p_ll2_conn, p_curp,
+                                        num_of_bds, CORE_TX_DEST_NW,
+                                        vlan, bd_flags, l4_hdr_offset_w,
+                                        first_frag, first_frag_len);
+
+       qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
+
+out:
+       spin_unlock_irqrestore(&p_tx->lock, flags);
+       return rc;
+}
+
+int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+                                     u8 connection_handle,
+                                     dma_addr_t addr, u16 nbytes)
+{
+       struct qed_ll2_tx_packet *p_cur_send_packet = NULL;
+       struct qed_ll2_info *p_ll2_conn = NULL;
+       u16 cur_send_frag_num = 0;
+       struct core_tx_bd *p_bd;
+       unsigned long flags;
+
+       p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return -EINVAL;
+
+       if (!p_ll2_conn->tx_queue.cur_send_packet)
+               return -EINVAL;
+
+       p_cur_send_packet = p_ll2_conn->tx_queue.cur_send_packet;
+       cur_send_frag_num = p_ll2_conn->tx_queue.cur_send_frag_num;
+
+       if (cur_send_frag_num >= p_cur_send_packet->bd_used)
+               return -EINVAL;
+
+       /* Fill the BD information, and possibly notify FW */
+       p_bd = p_cur_send_packet->bds_set[cur_send_frag_num].txq_bd;
+       DMA_REGPAIR_LE(p_bd->addr, addr);
+       p_bd->nbytes = cpu_to_le16(nbytes);
+       p_cur_send_packet->bds_set[cur_send_frag_num].tx_frag = addr;
+       p_cur_send_packet->bds_set[cur_send_frag_num].frag_len = nbytes;
+
+       p_ll2_conn->tx_queue.cur_send_frag_num++;
+
+       spin_lock_irqsave(&p_ll2_conn->tx_queue.lock, flags);
+       qed_ll2_tx_packet_notify(p_hwfn, p_ll2_conn);
+       spin_unlock_irqrestore(&p_ll2_conn->tx_queue.lock, flags);
+
+       return 0;
+}
+
+int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+       struct qed_ll2_info *p_ll2_conn = NULL;
+       int rc = -EINVAL;
+
+       p_ll2_conn = qed_ll2_handle_sanity_lock(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return -EINVAL;
+
+       /* Stop Tx & Rx of connection, if needed */
+       if (QED_LL2_TX_REGISTERED(p_ll2_conn)) {
+               rc = qed_sp_ll2_tx_queue_stop(p_hwfn, p_ll2_conn);
+               if (rc)
+                       return rc;
+               qed_ll2_txq_flush(p_hwfn, connection_handle);
+       }
+
+       if (QED_LL2_RX_REGISTERED(p_ll2_conn)) {
+               rc = qed_sp_ll2_rx_queue_stop(p_hwfn, p_ll2_conn);
+               if (rc)
+                       return rc;
+               qed_ll2_rxq_flush(p_hwfn, connection_handle);
+       }
+
+       return rc;
+}
+
+void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle)
+{
+       struct qed_ll2_info *p_ll2_conn = NULL;
+
+       p_ll2_conn = qed_ll2_handle_sanity(p_hwfn, connection_handle);
+       if (!p_ll2_conn)
+               return;
+
+       if (QED_LL2_RX_REGISTERED(p_ll2_conn)) {
+               p_ll2_conn->rx_queue.b_cb_registred = false;
+               qed_int_unregister_cb(p_hwfn, p_ll2_conn->rx_queue.rx_sb_index);
+       }
+
+       if (QED_LL2_TX_REGISTERED(p_ll2_conn)) {
+               p_ll2_conn->tx_queue.b_cb_registred = false;
+               qed_int_unregister_cb(p_hwfn, p_ll2_conn->tx_queue.tx_sb_index);
+       }
+
+       kfree(p_ll2_conn->tx_queue.descq_array);
+       qed_chain_free(p_hwfn->cdev, &p_ll2_conn->tx_queue.txq_chain);
+
+       kfree(p_ll2_conn->rx_queue.descq_array);
+       qed_chain_free(p_hwfn->cdev, &p_ll2_conn->rx_queue.rxq_chain);
+       qed_chain_free(p_hwfn->cdev, &p_ll2_conn->rx_queue.rcq_chain);
+
+       qed_cxt_release_cid(p_hwfn, p_ll2_conn->cid);
+
+       mutex_lock(&p_ll2_conn->mutex);
+       p_ll2_conn->b_active = false;
+       mutex_unlock(&p_ll2_conn->mutex);
+}
+
+struct qed_ll2_info *qed_ll2_alloc(struct qed_hwfn *p_hwfn)
+{
+       struct qed_ll2_info *p_ll2_connections;
+       u8 i;
+
+       /* Allocate LL2's set struct */
+       p_ll2_connections = kcalloc(QED_MAX_NUM_OF_LL2_CONNECTIONS,
+                                   sizeof(struct qed_ll2_info), GFP_KERNEL);
+       if (!p_ll2_connections) {
+               DP_NOTICE(p_hwfn, "Failed to allocate `struct qed_ll2'\n");
+               return NULL;
+       }
+
+       for (i = 0; i < QED_MAX_NUM_OF_LL2_CONNECTIONS; i++)
+               p_ll2_connections[i].my_id = i;
+
+       return p_ll2_connections;
+}
+
+void qed_ll2_setup(struct qed_hwfn *p_hwfn,
+                  struct qed_ll2_info *p_ll2_connections)
+{
+       int i;
+
+       for (i = 0; i < QED_MAX_NUM_OF_LL2_CONNECTIONS; i++)
+               mutex_init(&p_ll2_connections[i].mutex);
+}
+
+void qed_ll2_free(struct qed_hwfn *p_hwfn,
+                 struct qed_ll2_info *p_ll2_connections)
+{
+       kfree(p_ll2_connections);
+}
+
+static void _qed_ll2_get_tstats(struct qed_hwfn *p_hwfn,
+                               struct qed_ptt *p_ptt,
+                               struct qed_ll2_info *p_ll2_conn,
+                               struct qed_ll2_stats *p_stats)
+{
+       struct core_ll2_tstorm_per_queue_stat tstats;
+       u8 qid = p_ll2_conn->queue_id;
+       u32 tstats_addr;
+
+       memset(&tstats, 0, sizeof(tstats));
+       tstats_addr = BAR0_MAP_REG_TSDM_RAM +
+                     CORE_LL2_TSTORM_PER_QUEUE_STAT_OFFSET(qid);
+       qed_memcpy_from(p_hwfn, p_ptt, &tstats, tstats_addr, sizeof(tstats));
+
+       p_stats->packet_too_big_discard =
+                       HILO_64_REGPAIR(tstats.packet_too_big_discard);
+       p_stats->no_buff_discard = HILO_64_REGPAIR(tstats.no_buff_discard);
+}
+
+static void _qed_ll2_get_ustats(struct qed_hwfn *p_hwfn,
+                               struct qed_ptt *p_ptt,
+                               struct qed_ll2_info *p_ll2_conn,
+                               struct qed_ll2_stats *p_stats)
+{
+       struct core_ll2_ustorm_per_queue_stat ustats;
+       u8 qid = p_ll2_conn->queue_id;
+       u32 ustats_addr;
+
+       memset(&ustats, 0, sizeof(ustats));
+       ustats_addr = BAR0_MAP_REG_USDM_RAM +
+                     CORE_LL2_USTORM_PER_QUEUE_STAT_OFFSET(qid);
+       qed_memcpy_from(p_hwfn, p_ptt, &ustats, ustats_addr, sizeof(ustats));
+
+       p_stats->rcv_ucast_bytes = HILO_64_REGPAIR(ustats.rcv_ucast_bytes);
+       p_stats->rcv_mcast_bytes = HILO_64_REGPAIR(ustats.rcv_mcast_bytes);
+       p_stats->rcv_bcast_bytes = HILO_64_REGPAIR(ustats.rcv_bcast_bytes);
+       p_stats->rcv_ucast_pkts = HILO_64_REGPAIR(ustats.rcv_ucast_pkts);
+       p_stats->rcv_mcast_pkts = HILO_64_REGPAIR(ustats.rcv_mcast_pkts);
+       p_stats->rcv_bcast_pkts = HILO_64_REGPAIR(ustats.rcv_bcast_pkts);
+}
+
+static void _qed_ll2_get_pstats(struct qed_hwfn *p_hwfn,
+                               struct qed_ptt *p_ptt,
+                               struct qed_ll2_info *p_ll2_conn,
+                               struct qed_ll2_stats *p_stats)
+{
+       struct core_ll2_pstorm_per_queue_stat pstats;
+       u8 stats_id = p_ll2_conn->tx_stats_id;
+       u32 pstats_addr;
+
+       memset(&pstats, 0, sizeof(pstats));
+       pstats_addr = BAR0_MAP_REG_PSDM_RAM +
+                     CORE_LL2_PSTORM_PER_QUEUE_STAT_OFFSET(stats_id);
+       qed_memcpy_from(p_hwfn, p_ptt, &pstats, pstats_addr, sizeof(pstats));
+
+       p_stats->sent_ucast_bytes = HILO_64_REGPAIR(pstats.sent_ucast_bytes);
+       p_stats->sent_mcast_bytes = HILO_64_REGPAIR(pstats.sent_mcast_bytes);
+       p_stats->sent_bcast_bytes = HILO_64_REGPAIR(pstats.sent_bcast_bytes);
+       p_stats->sent_ucast_pkts = HILO_64_REGPAIR(pstats.sent_ucast_pkts);
+       p_stats->sent_mcast_pkts = HILO_64_REGPAIR(pstats.sent_mcast_pkts);
+       p_stats->sent_bcast_pkts = HILO_64_REGPAIR(pstats.sent_bcast_pkts);
+}
+
+int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+                     u8 connection_handle, struct qed_ll2_stats *p_stats)
+{
+       struct qed_ll2_info *p_ll2_conn = NULL;
+       struct qed_ptt *p_ptt;
+
+       memset(p_stats, 0, sizeof(*p_stats));
+
+       if ((connection_handle >= QED_MAX_NUM_OF_LL2_CONNECTIONS) ||
+           !p_hwfn->p_ll2_info)
+               return -EINVAL;
+
+       p_ll2_conn = &p_hwfn->p_ll2_info[connection_handle];
+
+       p_ptt = qed_ptt_acquire(p_hwfn);
+       if (!p_ptt) {
+               DP_ERR(p_hwfn, "Failed to acquire ptt\n");
+               return -EINVAL;
+       }
+
+       _qed_ll2_get_tstats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+       _qed_ll2_get_ustats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+       if (p_ll2_conn->tx_stats_en)
+               _qed_ll2_get_pstats(p_hwfn, p_ptt, p_ll2_conn, p_stats);
+
+       qed_ptt_release(p_hwfn, p_ptt);
+       return 0;
+}
+
+static void qed_ll2_register_cb_ops(struct qed_dev *cdev,
+                                   const struct qed_ll2_cb_ops *ops,
+                                   void *cookie)
+{
+       cdev->ll2->cbs = ops;
+       cdev->ll2->cb_cookie = cookie;
+}
+
+static int qed_ll2_start(struct qed_dev *cdev, struct qed_ll2_params *params)
+{
+       struct qed_ll2_info ll2_info;
+       struct qed_ll2_buffer *buffer;
+       enum qed_ll2_conn_type conn_type;
+       struct qed_ptt *p_ptt;
+       int rc, i;
+
+       /* Initialize LL2 locks & lists */
+       INIT_LIST_HEAD(&cdev->ll2->list);
+       spin_lock_init(&cdev->ll2->lock);
+       cdev->ll2->rx_size = NET_SKB_PAD + ETH_HLEN +
+                            L1_CACHE_BYTES + params->mtu;
+       cdev->ll2->frags_mapped = params->frags_mapped;
+
+       /*Allocate memory for LL2 */
+       DP_INFO(cdev, "Allocating LL2 buffers of size %08x bytes\n",
+               cdev->ll2->rx_size);
+       for (i = 0; i < QED_LL2_RX_SIZE; i++) {
+               buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
+               if (!buffer) {
+                       DP_INFO(cdev, "Failed to allocate LL2 buffers\n");
+                       goto fail;
+               }
+
+               rc = qed_ll2_alloc_buffer(cdev, (u8 **)&buffer->data,
+                                         &buffer->phys_addr);
+               if (rc) {
+                       kfree(buffer);
+                       goto fail;
+               }
+
+               list_add_tail(&buffer->list, &cdev->ll2->list);
+       }
+
+       switch (QED_LEADING_HWFN(cdev)->hw_info.personality) {
+       case QED_PCI_ISCSI:
+               conn_type = QED_LL2_TYPE_ISCSI;
+               break;
+       case QED_PCI_ETH_ROCE:
+               conn_type = QED_LL2_TYPE_ROCE;
+               break;
+       default:
+               conn_type = QED_LL2_TYPE_TEST;
+       }
+
+       /* Prepare the temporary ll2 information */
+       memset(&ll2_info, 0, sizeof(ll2_info));
+       ll2_info.conn_type = conn_type;
+       ll2_info.mtu = params->mtu;
+       ll2_info.rx_drop_ttl0_flg = params->drop_ttl0_packets;
+       ll2_info.rx_vlan_removal_en = params->rx_vlan_stripping;
+       ll2_info.tx_tc = 0;
+       ll2_info.tx_dest = CORE_TX_DEST_NW;
+
+       rc = qed_ll2_acquire_connection(QED_LEADING_HWFN(cdev), &ll2_info,
+                                       QED_LL2_RX_SIZE, QED_LL2_TX_SIZE,
+                                       &cdev->ll2->handle);
+       if (rc) {
+               DP_INFO(cdev, "Failed to acquire LL2 connection\n");
+               goto fail;
+       }
+
+       rc = qed_ll2_establish_connection(QED_LEADING_HWFN(cdev),
+                                         cdev->ll2->handle);
+       if (rc) {
+               DP_INFO(cdev, "Failed to establish LL2 connection\n");
+               goto release_fail;
+       }
+
+       /* Post all Rx buffers to FW */
+       spin_lock_bh(&cdev->ll2->lock);
+       list_for_each_entry(buffer, &cdev->ll2->list, list) {
+               rc = qed_ll2_post_rx_buffer(QED_LEADING_HWFN(cdev),
+                                           cdev->ll2->handle,
+                                           buffer->phys_addr, 0, buffer, 1);
+               if (rc) {
+                       DP_INFO(cdev,
+                               "Failed to post an Rx buffer; Deleting it\n");
+                       dma_unmap_single(&cdev->pdev->dev, buffer->phys_addr,
+                                        cdev->ll2->rx_size, DMA_FROM_DEVICE);
+                       kfree(buffer->data);
+                       list_del(&buffer->list);
+                       kfree(buffer);
+               } else {
+                       cdev->ll2->rx_cnt++;
+               }
+       }
+       spin_unlock_bh(&cdev->ll2->lock);
+
+       if (!cdev->ll2->rx_cnt) {
+               DP_INFO(cdev, "Failed passing even a single Rx buffer\n");
+               goto release_terminate;
+       }
+
+       if (!is_valid_ether_addr(params->ll2_mac_address)) {
+               DP_INFO(cdev, "Invalid Ethernet address\n");
+               goto release_terminate;
+       }
+
+       p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+       if (!p_ptt) {
+               DP_INFO(cdev, "Failed to acquire PTT\n");
+               goto release_terminate;
+       }
+
+       rc = qed_llh_add_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+                                   params->ll2_mac_address);
+       qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+       if (rc) {
+               DP_ERR(cdev, "Failed to allocate LLH filter\n");
+               goto release_terminate_all;
+       }
+
+       ether_addr_copy(cdev->ll2_mac_address, params->ll2_mac_address);
+
+       return 0;
+
+release_terminate_all:
+
+release_terminate:
+       qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+release_fail:
+       qed_ll2_release_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+fail:
+       qed_ll2_kill_buffers(cdev);
+       cdev->ll2->handle = QED_LL2_UNUSED_HANDLE;
+       return -EINVAL;
+}
+
+static int qed_ll2_stop(struct qed_dev *cdev)
+{
+       struct qed_ptt *p_ptt;
+       int rc;
+
+       if (cdev->ll2->handle == QED_LL2_UNUSED_HANDLE)
+               return 0;
+
+       p_ptt = qed_ptt_acquire(QED_LEADING_HWFN(cdev));
+       if (!p_ptt) {
+               DP_INFO(cdev, "Failed to acquire PTT\n");
+               goto fail;
+       }
+
+       qed_llh_remove_mac_filter(QED_LEADING_HWFN(cdev), p_ptt,
+                                 cdev->ll2_mac_address);
+       qed_ptt_release(QED_LEADING_HWFN(cdev), p_ptt);
+       eth_zero_addr(cdev->ll2_mac_address);
+
+       rc = qed_ll2_terminate_connection(QED_LEADING_HWFN(cdev),
+                                         cdev->ll2->handle);
+       if (rc)
+               DP_INFO(cdev, "Failed to terminate LL2 connection\n");
+
+       qed_ll2_kill_buffers(cdev);
+
+       qed_ll2_release_connection(QED_LEADING_HWFN(cdev), cdev->ll2->handle);
+       cdev->ll2->handle = QED_LL2_UNUSED_HANDLE;
+
+       return rc;
+fail:
+       return -EINVAL;
+}
+
+static int qed_ll2_start_xmit(struct qed_dev *cdev, struct sk_buff *skb)
+{
+       const skb_frag_t *frag;
+       int rc = -EINVAL, i;
+       dma_addr_t mapping;
+       u16 vlan = 0;
+       u8 flags = 0;
+
+       if (unlikely(skb->ip_summed != CHECKSUM_NONE)) {
+               DP_INFO(cdev, "Cannot transmit a checksumed packet\n");
+               return -EINVAL;
+       }
+
+       if (1 + skb_shinfo(skb)->nr_frags > CORE_LL2_TX_MAX_BDS_PER_PACKET) {
+               DP_ERR(cdev, "Cannot transmit a packet with %d fragments\n",
+                      1 + skb_shinfo(skb)->nr_frags);
+               return -EINVAL;
+       }
+
+       mapping = dma_map_single(&cdev->pdev->dev, skb->data,
+                                skb->len, DMA_TO_DEVICE);
+       if (unlikely(dma_mapping_error(&cdev->pdev->dev, mapping))) {
+               DP_NOTICE(cdev, "SKB mapping failed\n");
+               return -EINVAL;
+       }
+
+       /* Request HW to calculate IP csum */
+       if (!((vlan_get_protocol(skb) == htons(ETH_P_IPV6)) &&
+             ipv6_hdr(skb)->nexthdr == NEXTHDR_IPV6))
+               flags |= BIT(CORE_TX_BD_FLAGS_IP_CSUM_SHIFT);
+
+       if (skb_vlan_tag_present(skb)) {
+               vlan = skb_vlan_tag_get(skb);
+               flags |= BIT(CORE_TX_BD_FLAGS_VLAN_INSERTION_SHIFT);
+       }
+
+       rc = qed_ll2_prepare_tx_packet(QED_LEADING_HWFN(cdev),
+                                      cdev->ll2->handle,
+                                      1 + skb_shinfo(skb)->nr_frags,
+                                      vlan, flags, 0, mapping,
+                                      skb->len, skb, 1);
+       if (rc)
+               goto err;
+
+       for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+               frag = &skb_shinfo(skb)->frags[i];
+               if (!cdev->ll2->frags_mapped) {
+                       mapping = skb_frag_dma_map(&cdev->pdev->dev, frag, 0,
+                                                  skb_frag_size(frag),
+                                                  DMA_TO_DEVICE);
+
+                       if (unlikely(dma_mapping_error(&cdev->pdev->dev,
+                                                      mapping))) {
+                               DP_NOTICE(cdev,
+                                         "Unable to map frag - dropping packet\n");
+                               goto err;
+                       }
+               } else {
+                       mapping = page_to_phys(skb_frag_page(frag)) |
+                           frag->page_offset;
+               }
+
+               rc = qed_ll2_set_fragment_of_tx_packet(QED_LEADING_HWFN(cdev),
+                                                      cdev->ll2->handle,
+                                                      mapping,
+                                                      skb_frag_size(frag));
+
+               /* if failed not much to do here, partial packet has been posted
+                * we can't free memory, will need to wait for completion.
+                */
+               if (rc)
+                       goto err2;
+       }
+
+       return 0;
+
+err:
+       dma_unmap_single(&cdev->pdev->dev, mapping, skb->len, DMA_TO_DEVICE);
+
+err2:
+       return rc;
+}
+
+static int qed_ll2_stats(struct qed_dev *cdev, struct qed_ll2_stats *stats)
+{
+       if (!cdev->ll2)
+               return -EINVAL;
+
+       return qed_ll2_get_stats(QED_LEADING_HWFN(cdev),
+                                cdev->ll2->handle, stats);
+}
+
+const struct qed_ll2_ops qed_ll2_ops_pass = {
+       .start = &qed_ll2_start,
+       .stop = &qed_ll2_stop,
+       .start_xmit = &qed_ll2_start_xmit,
+       .register_cb_ops = &qed_ll2_register_cb_ops,
+       .get_stats = &qed_ll2_stats,
+};
+
+int qed_ll2_alloc_if(struct qed_dev *cdev)
+{
+       cdev->ll2 = kzalloc(sizeof(*cdev->ll2), GFP_KERNEL);
+       return cdev->ll2 ? 0 : -ENOMEM;
+}
+
+void qed_ll2_dealloc_if(struct qed_dev *cdev)
+{
+       kfree(cdev->ll2);
+       cdev->ll2 = NULL;
+}
diff --git a/drivers/net/ethernet/qlogic/qed/qed_ll2.h b/drivers/net/ethernet/qlogic/qed/qed_ll2.h
new file mode 100644 (file)
index 0000000..a037c48
--- /dev/null
@@ -0,0 +1,289 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#ifndef _QED_LL2_H
+#define _QED_LL2_H
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/qed/qed_chain.h>
+#include <linux/qed/qed_ll2_if.h>
+#include "qed.h"
+#include "qed_hsi.h"
+#include "qed_sp.h"
+
+#define QED_MAX_NUM_OF_LL2_CONNECTIONS                    (4)
+
+enum qed_ll2_conn_type {
+       QED_LL2_TYPE_RESERVED,
+       QED_LL2_TYPE_ISCSI,
+       QED_LL2_TYPE_TEST,
+       QED_LL2_TYPE_ISCSI_OOO,
+       QED_LL2_TYPE_RESERVED2,
+       QED_LL2_TYPE_ROCE,
+       QED_LL2_TYPE_RESERVED3,
+       MAX_QED_LL2_RX_CONN_TYPE
+};
+
+struct qed_ll2_rx_packet {
+       struct list_head list_entry;
+       struct core_rx_bd_with_buff_len *rxq_bd;
+       dma_addr_t rx_buf_addr;
+       u16 buf_length;
+       void *cookie;
+       u8 placement_offset;
+       u16 parse_flags;
+       u16 packet_length;
+       u16 vlan;
+       u32 opaque_data[2];
+};
+
+struct qed_ll2_tx_packet {
+       struct list_head list_entry;
+       u16 bd_used;
+       u16 vlan;
+       u16 l4_hdr_offset_w;
+       u8 bd_flags;
+       bool notify_fw;
+       void *cookie;
+
+       struct {
+               struct core_tx_bd *txq_bd;
+               dma_addr_t tx_frag;
+               u16 frag_len;
+       } bds_set[ETH_TX_MAX_BDS_PER_NON_LSO_PACKET];
+};
+
+struct qed_ll2_rx_queue {
+       /* Lock protecting the Rx queue manipulation */
+       spinlock_t lock;
+       struct qed_chain rxq_chain;
+       struct qed_chain rcq_chain;
+       u8 rx_sb_index;
+       bool b_cb_registred;
+       __le16 *p_fw_cons;
+       struct list_head active_descq;
+       struct list_head free_descq;
+       struct list_head posting_descq;
+       struct qed_ll2_rx_packet *descq_array;
+       void __iomem *set_prod_addr;
+};
+
+struct qed_ll2_tx_queue {
+       /* Lock protecting the Tx queue manipulation */
+       spinlock_t lock;
+       struct qed_chain txq_chain;
+       u8 tx_sb_index;
+       bool b_cb_registred;
+       __le16 *p_fw_cons;
+       struct list_head active_descq;
+       struct list_head free_descq;
+       struct list_head sending_descq;
+       struct qed_ll2_tx_packet *descq_array;
+       struct qed_ll2_tx_packet *cur_send_packet;
+       struct qed_ll2_tx_packet cur_completing_packet;
+       u16 cur_completing_bd_idx;
+       void __iomem *doorbell_addr;
+       u16 bds_idx;
+       u16 cur_send_frag_num;
+       u16 cur_completing_frag_num;
+       bool b_completing_packet;
+};
+
+struct qed_ll2_info {
+       /* Lock protecting the state of LL2 */
+       struct mutex mutex;
+       enum qed_ll2_conn_type conn_type;
+       u32 cid;
+       u8 my_id;
+       u8 queue_id;
+       u8 tx_stats_id;
+       bool b_active;
+       u16 mtu;
+       u8 rx_drop_ttl0_flg;
+       u8 rx_vlan_removal_en;
+       u8 tx_tc;
+       enum core_tx_dest tx_dest;
+       enum core_error_handle ai_err_packet_too_big;
+       enum core_error_handle ai_err_no_buf;
+       u8 tx_stats_en;
+       struct qed_ll2_rx_queue rx_queue;
+       struct qed_ll2_tx_queue tx_queue;
+};
+
+/**
+ * @brief qed_ll2_acquire_connection - allocate resources,
+ *        starts rx & tx (if relevant) queues pair. Provides
+ *        connecion handler as output parameter.
+ *
+ * @param p_hwfn
+ * @param p_params             Contain various configuration properties
+ * @param rx_num_desc
+ * @param tx_num_desc
+ *
+ * @param p_connection_handle  Output container for LL2 connection's handle
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_acquire_connection(struct qed_hwfn *p_hwfn,
+                              struct qed_ll2_info *p_params,
+                              u16 rx_num_desc,
+                              u16 tx_num_desc,
+                              u8 *p_connection_handle);
+
+/**
+ * @brief qed_ll2_establish_connection - start previously
+ *        allocated LL2 queues pair
+ *
+ * @param p_hwfn
+ * @param p_ptt
+ * @param connection_handle    LL2 connection's handle obtained from
+ *                              qed_ll2_require_connection
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_establish_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_post_rx_buffers - submit buffers to LL2 Rx queue.
+ *
+ * @param p_hwfn
+ * @param connection_handle    LL2 connection's handle obtained from
+ *                             qed_ll2_require_connection
+ * @param addr                 rx (physical address) buffers to submit
+ * @param cookie
+ * @param notify_fw            produce corresponding Rx BD immediately
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_post_rx_buffer(struct qed_hwfn *p_hwfn,
+                          u8 connection_handle,
+                          dma_addr_t addr,
+                          u16 buf_len, void *cookie, u8 notify_fw);
+
+/**
+ * @brief qed_ll2_prepare_tx_packet - request for start Tx BD
+ *                                   to prepare Tx packet submission to FW.
+ *
+ * @param p_hwfn
+ * @param connection_handle    LL2 connection's handle obtained from
+ *                             qed_ll2_require_connection
+ * @param num_of_bds           a number of requested BD equals a number of
+ *                             fragments in Tx packet
+ * @param vlan                 VLAN to insert to packet (if insertion set)
+ * @param bd_flags
+ * @param l4_hdr_offset_w      L4 Header Offset from start of packet
+ *                             (in words). This is needed if both l4_csum
+ *                             and ipv6_ext are set
+ * @param first_frag
+ * @param first_frag_len
+ * @param cookie
+ *
+ * @param notify_fw
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_prepare_tx_packet(struct qed_hwfn *p_hwfn,
+                             u8 connection_handle,
+                             u8 num_of_bds,
+                             u16 vlan,
+                             u8 bd_flags,
+                             u16 l4_hdr_offset_w,
+                             dma_addr_t first_frag,
+                             u16 first_frag_len, void *cookie, u8 notify_fw);
+
+/**
+ * @brief qed_ll2_release_connection - releases resources
+ *                                     allocated for LL2 connection
+ *
+ * @param p_hwfn
+ * @param connection_handle            LL2 connection's handle obtained from
+ *                                     qed_ll2_require_connection
+ */
+void qed_ll2_release_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_set_fragment_of_tx_packet -  provides fragments to fill
+ *                                             Tx BD of BDs requested by
+ *                                             qed_ll2_prepare_tx_packet
+ *
+ * @param p_hwfn
+ * @param connection_handle                    LL2 connection's handle
+ *                                             obtained from
+ *                                             qed_ll2_require_connection
+ * @param addr
+ * @param nbytes
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_set_fragment_of_tx_packet(struct qed_hwfn *p_hwfn,
+                                     u8 connection_handle,
+                                     dma_addr_t addr, u16 nbytes);
+
+/**
+ * @brief qed_ll2_terminate_connection -       stops Tx/Rx queues
+ *
+ *
+ * @param p_hwfn
+ * @param connection_handle                    LL2 connection's handle
+ *                                             obtained from
+ *                                             qed_ll2_require_connection
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_terminate_connection(struct qed_hwfn *p_hwfn, u8 connection_handle);
+
+/**
+ * @brief qed_ll2_get_stats -  get LL2 queue's statistics
+ *
+ *
+ * @param p_hwfn
+ * @param connection_handle    LL2 connection's handle obtained from
+ *                             qed_ll2_require_connection
+ * @param p_stats
+ *
+ * @return 0 on success, failure otherwise
+ */
+int qed_ll2_get_stats(struct qed_hwfn *p_hwfn,
+                     u8 connection_handle, struct qed_ll2_stats *p_stats);
+
+/**
+ * @brief qed_ll2_alloc - Allocates LL2 connections set
+ *
+ * @param p_hwfn
+ *
+ * @return pointer to alocated qed_ll2_info or NULL
+ */
+struct qed_ll2_info *qed_ll2_alloc(struct qed_hwfn *p_hwfn);
+
+/**
+ * @brief qed_ll2_setup - Inits LL2 connections set
+ *
+ * @param p_hwfn
+ * @param p_ll2_connections
+ *
+ */
+void qed_ll2_setup(struct qed_hwfn *p_hwfn,
+                  struct qed_ll2_info *p_ll2_connections);
+
+/**
+ * @brief qed_ll2_free - Releases LL2 connections set
+ *
+ * @param p_hwfn
+ * @param p_ll2_connections
+ *
+ */
+void qed_ll2_free(struct qed_hwfn *p_hwfn,
+                 struct qed_ll2_info *p_ll2_connections);
+
+#endif
index b730a632c3830910efb71316a6a6fbd5dd5b289e..48cdf62c025b57715074e769556cd9bfcfc059f7 100644 (file)
 #include <linux/etherdevice.h>
 #include <linux/vmalloc.h>
 #include <linux/qed/qed_if.h>
+#include <linux/qed/qed_ll2_if.h>
 
 #include "qed.h"
 #include "qed_sriov.h"
 #include "qed_sp.h"
 #include "qed_dev_api.h"
+#include "qed_ll2.h"
 #include "qed_mcp.h"
 #include "qed_hw.h"
 #include "qed_selftest.h"
@@ -608,7 +610,16 @@ static int qed_nic_reset(struct qed_dev *cdev)
 
 static int qed_nic_setup(struct qed_dev *cdev)
 {
-       int rc;
+       int rc, i;
+
+       /* Determine if interface is going to require LL2 */
+       if (QED_LEADING_HWFN(cdev)->hw_info.personality != QED_PCI_ETH) {
+               for (i = 0; i < cdev->num_hwfns; i++) {
+                       struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
+
+                       p_hwfn->using_ll2 = true;
+               }
+       }
 
        rc = qed_resc_alloc(cdev);
        if (rc)
@@ -873,6 +884,12 @@ static int qed_slowpath_start(struct qed_dev *cdev,
        DP_INFO(cdev,
                "HW initialization and function start completed successfully\n");
 
+       /* Allocate LL2 interface if needed */
+       if (QED_LEADING_HWFN(cdev)->using_ll2) {
+               rc = qed_ll2_alloc_if(cdev);
+               if (rc)
+                       goto err3;
+       }
        if (IS_PF(cdev)) {
                hwfn = QED_LEADING_HWFN(cdev);
                drv_version.version = (params->drv_major << 24) |
@@ -893,6 +910,8 @@ static int qed_slowpath_start(struct qed_dev *cdev,
 
        return 0;
 
+err3:
+       qed_hw_stop(cdev);
 err2:
        qed_hw_timers_stop_all(cdev);
        if (IS_PF(cdev))
@@ -915,6 +934,8 @@ static int qed_slowpath_stop(struct qed_dev *cdev)
        if (!cdev)
                return -ENODEV;
 
+       qed_ll2_dealloc_if(cdev);
+
        if (IS_PF(cdev)) {
                qed_free_stream_mem(cdev);
                if (IS_QED_ETH_IF(cdev))
index 759cb04e02b09bbffc2dbce03fe2fc4e86d0f1e9..e75738d2178306cf7147ce5dbe69513495dd396a 100644 (file)
        0x50196cUL
 #define NIG_REG_LLH_CLS_TYPE_DUALMODE \
        0x501964UL
+#define NIG_REG_LLH_FUNC_FILTER_VALUE \
+       0x501a00UL
+#define NIG_REG_LLH_FUNC_FILTER_VALUE_SIZE \
+       32
+#define NIG_REG_LLH_FUNC_FILTER_EN \
+       0x501a80UL
+#define NIG_REG_LLH_FUNC_FILTER_EN_SIZE        \
+       16
+#define NIG_REG_LLH_FUNC_FILTER_MODE \
+       0x501ac0UL
+#define NIG_REG_LLH_FUNC_FILTER_MODE_SIZE \
+       16
+#define NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE \
+       0x501b00UL
+#define NIG_REG_LLH_FUNC_FILTER_PROTOCOL_TYPE_SIZE \
+       16
+#define NIG_REG_LLH_FUNC_FILTER_HDR_SEL        \
+       0x501b40UL
+#define NIG_REG_LLH_FUNC_FILTER_HDR_SEL_SIZE \
+       16
 #define  NCSI_REG_CONFIG       \
        0x040200UL
 #define  PBF_REG_INIT \
        0x1f0a1cUL
 #define PRS_REG_ROCE_DEST_QP_MAX_PF \
        0x1f0430UL
+#define PRS_REG_USE_LIGHT_L2 \
+       0x1f096cUL
 #define  PSDM_REG_ENABLE_IN1 \
        0xfa0004UL
 #define  PSEM_REG_ENABLE_IN \
index a548504c3420d86bcc2d06945b6b9f3f90d6578b..a3c539f1c2ac1890bc115d99166644ac311a0720 100644 (file)
@@ -61,6 +61,10 @@ union ramrod_data {
        struct vport_start_ramrod_data vport_start;
        struct vport_stop_ramrod_data vport_stop;
        struct vport_update_ramrod_data vport_update;
+       struct core_rx_start_ramrod_data core_rx_queue_start;
+       struct core_rx_stop_ramrod_data core_rx_queue_stop;
+       struct core_tx_start_ramrod_data core_tx_queue_start;
+       struct core_tx_stop_ramrod_data core_tx_queue_stop;
        struct vport_filter_update_ramrod_data vport_filter_update;
 
        struct rdma_init_func_ramrod_data rdma_init_func;
index e4546abcea08f3bb6cd7581cf2853ab602f11e29..c2d74e8785cf7a96a1f57ade2fbfffa64212ae47 100644 (file)
@@ -627,6 +627,7 @@ enum DP_MODULE {
        QED_MSG_SP      = 0x100000,
        QED_MSG_STORAGE = 0x200000,
        QED_MSG_CXT     = 0x800000,
+       QED_MSG_LL2     = 0x1000000,
        QED_MSG_ILT     = 0x2000000,
        QED_MSG_ROCE    = 0x4000000,
        QED_MSG_DEBUG   = 0x8000000,
diff --git a/include/linux/qed/qed_ll2_if.h b/include/linux/qed/qed_ll2_if.h
new file mode 100644 (file)
index 0000000..fd75c26
--- /dev/null
@@ -0,0 +1,139 @@
+/* QLogic qed NIC Driver
+ *
+ * Copyright (c) 2015 QLogic Corporation
+ *
+ * This software is available under the terms of the GNU General Public License
+ * (GPL) Version 2, available from the file COPYING in the main directory of
+ * this source tree.
+ */
+
+#ifndef _QED_LL2_IF_H
+#define _QED_LL2_IF_H
+
+#include <linux/types.h>
+#include <linux/interrupt.h>
+#include <linux/netdevice.h>
+#include <linux/pci.h>
+#include <linux/skbuff.h>
+#include <linux/version.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/qed/qed_if.h>
+
+struct qed_ll2_stats {
+       u64 gsi_invalid_hdr;
+       u64 gsi_invalid_pkt_length;
+       u64 gsi_unsupported_pkt_typ;
+       u64 gsi_crcchksm_error;
+
+       u64 packet_too_big_discard;
+       u64 no_buff_discard;
+
+       u64 rcv_ucast_bytes;
+       u64 rcv_mcast_bytes;
+       u64 rcv_bcast_bytes;
+       u64 rcv_ucast_pkts;
+       u64 rcv_mcast_pkts;
+       u64 rcv_bcast_pkts;
+
+       u64 sent_ucast_bytes;
+       u64 sent_mcast_bytes;
+       u64 sent_bcast_bytes;
+       u64 sent_ucast_pkts;
+       u64 sent_mcast_pkts;
+       u64 sent_bcast_pkts;
+};
+
+#define QED_LL2_UNUSED_HANDLE   (0xff)
+
+struct qed_ll2_cb_ops {
+       int (*rx_cb)(void *, struct sk_buff *, u32, u32);
+       int (*tx_cb)(void *, struct sk_buff *, bool);
+};
+
+struct qed_ll2_params {
+       u16 mtu;
+       bool drop_ttl0_packets;
+       bool rx_vlan_stripping;
+       u8 tx_tc;
+       bool frags_mapped;
+       u8 ll2_mac_address[ETH_ALEN];
+};
+
+struct qed_ll2_ops {
+/**
+ * @brief start - initializes ll2
+ *
+ * @param cdev
+ * @param params - protocol driver configuration for the ll2.
+ *
+ * @return 0 on success, otherwise error value.
+ */
+       int (*start)(struct qed_dev *cdev, struct qed_ll2_params *params);
+
+/**
+ * @brief stop - stops the ll2
+ *
+ * @param cdev
+ *
+ * @return 0 on success, otherwise error value.
+ */
+       int (*stop)(struct qed_dev *cdev);
+
+/**
+ * @brief start_xmit - transmits an skb over the ll2 interface
+ *
+ * @param cdev
+ * @param skb
+ *
+ * @return 0 on success, otherwise error value.
+ */
+       int (*start_xmit)(struct qed_dev *cdev, struct sk_buff *skb);
+
+/**
+ * @brief register_cb_ops - protocol driver register the callback for Rx/Tx
+ * packets. Should be called before `start'.
+ *
+ * @param cdev
+ * @param cookie - to be passed to the callback functions.
+ * @param ops - the callback functions to register for Rx / Tx.
+ *
+ * @return 0 on success, otherwise error value.
+ */
+       void (*register_cb_ops)(struct qed_dev *cdev,
+                               const struct qed_ll2_cb_ops *ops,
+                               void *cookie);
+
+/**
+ * @brief get LL2 related statistics
+ *
+ * @param cdev
+ * @param stats - pointer to struct that would be filled with stats
+ *
+ * @return 0 on success, error otherwise.
+ */
+       int (*get_stats)(struct qed_dev *cdev, struct qed_ll2_stats *stats);
+};
+
+#ifdef CONFIG_QED_LL2
+int qed_ll2_alloc_if(struct qed_dev *);
+void qed_ll2_dealloc_if(struct qed_dev *);
+#else
+static const struct qed_ll2_ops qed_ll2_ops_pass = {
+       .start = NULL,
+       .stop = NULL,
+       .start_xmit = NULL,
+       .register_cb_ops = NULL,
+       .get_stats = NULL,
+};
+
+static inline int qed_ll2_alloc_if(struct qed_dev *cdev)
+{
+       return 0;
+}
+
+static inline void qed_ll2_dealloc_if(struct qed_dev *cdev)
+{
+}
+#endif
+#endif