]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/commitdiff
i40e: Add infrastructure for queue channel support
authorAmritha Nambiar <amritha.nambiar@intel.com>
Thu, 7 Sep 2017 11:00:17 +0000 (04:00 -0700)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Fri, 13 Oct 2017 20:38:17 +0000 (13:38 -0700)
This patch sets up the infrastructure for offloading TCs and
queue configurations to the hardware by creating HW channels(VSI).
A new channel is created for each of the traffic class
configuration offloaded via mqprio framework except for the first TC
(TC0). TC0 for the main VSI is also reconfigured as per user provided
queue parameters. Queue counts that are not power-of-2 are handled by
reconfiguring RSS by reprogramming LUTs using the queue count value.
This patch also handles configuring the TX rings for the channels,
setting up the RX queue map for channel.

Also, the channels so created are removed and all the queue
configuration is set to default when the qdisc is detached from the
root of the device.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/i40e/i40e.h
drivers/net/ethernet/intel/i40e/i40e_main.c
drivers/net/ethernet/intel/i40e/i40e_txrx.h

index e7c7a853cf7febf0aba068525b692ff2e22b7efc..bde982541772f1f905d2823faea826f1999b3ab2 100644 (file)
@@ -87,6 +87,7 @@
 #define I40E_AQ_LEN                    256
 #define I40E_AQ_WORK_LIMIT             66 /* max number of VFs + a little */
 #define I40E_MAX_USER_PRIORITY         8
+#define I40E_MAX_QUEUES_PER_CH         64
 #define I40E_DEFAULT_TRAFFIC_CLASS     BIT(0)
 #define I40E_DEFAULT_MSG_ENABLE                4
 #define I40E_QUEUE_WAIT_RETRY_LIMIT    10
@@ -340,6 +341,23 @@ struct i40e_flex_pit {
        u8 pit_index;
 };
 
+struct i40e_channel {
+       struct list_head list;
+       bool initialized;
+       u8 type;
+       u16 vsi_number; /* Assigned VSI number from AQ 'Add VSI' response */
+       u16 stat_counter_idx;
+       u16 base_queue;
+       u16 num_queue_pairs; /* Requested by user */
+       u16 seid;
+
+       u8 enabled_tc;
+       struct i40e_aqc_vsi_properties_data info;
+
+       /* track this channel belongs to which VSI */
+       struct i40e_vsi *parent_vsi;
+};
+
 /* struct that defines the Ethernet device */
 struct i40e_pf {
        struct pci_dev *pdev;
@@ -456,6 +474,7 @@ struct i40e_pf {
 #define I40E_FLAG_CLIENT_RESET                 BIT(26)
 #define I40E_FLAG_LINK_DOWN_ON_CLOSE_ENABLED   BIT(27)
 #define I40E_FLAG_SOURCE_PRUNING_DISABLED      BIT(28)
+#define I40E_FLAG_TC_MQPRIO                    BIT(29)
 
        struct i40e_client_instance *cinst;
        bool stat_offsets_loaded;
@@ -536,6 +555,8 @@ struct i40e_pf {
        u32 ioremap_len;
        u32 fd_inv;
        u16 phy_led_val;
+
+       u16 override_q_count;
 };
 
 /**
@@ -700,6 +721,15 @@ struct i40e_vsi {
        bool current_isup;      /* Sync 'link up' logging */
        enum i40e_aq_link_speed current_speed;  /* Sync link speed logging */
 
+       /* channel specific fields */
+       u16 cnt_q_avail;        /* num of queues available for channel usage */
+       u16 orig_rss_size;
+       u16 current_rss_size;
+
+       u16 next_base_queue;    /* next queue to be used for channel setup */
+
+       struct list_head ch_list;
+
        void *priv;     /* client driver data reference. */
 
        /* VSI specific handlers */
@@ -1004,4 +1034,6 @@ static inline bool i40e_enabled_xdp_vsi(struct i40e_vsi *vsi)
 {
        return !!vsi->xdp_prog;
 }
+
+int i40e_create_queue_channel(struct i40e_vsi *vsi, struct i40e_channel *ch);
 #endif /* _I40E_H_ */
index 6190257eecfec8915b6572d0a0b9188ea0efe6ac..e23105bee6d10ee7db4d0dbb7dae5ee196268c7f 100644 (file)
@@ -2881,7 +2881,7 @@ static void i40e_config_xps_tx_ring(struct i40e_ring *ring)
 {
        int cpu;
 
-       if (!ring->q_vector || !ring->netdev)
+       if (!ring->q_vector || !ring->netdev || ring->ch)
                return;
 
        /* We only initialize XPS once, so as not to overwrite user settings */
@@ -2944,7 +2944,14 @@ static int i40e_configure_tx_ring(struct i40e_ring *ring)
         * initialization. This has to be done regardless of
         * DCB as by default everything is mapped to TC0.
         */
-       tx_ctx.rdylist = le16_to_cpu(vsi->info.qs_handle[ring->dcb_tc]);
+
+       if (ring->ch)
+               tx_ctx.rdylist =
+                       le16_to_cpu(ring->ch->info.qs_handle[ring->dcb_tc]);
+
+       else
+               tx_ctx.rdylist = le16_to_cpu(vsi->info.qs_handle[ring->dcb_tc]);
+
        tx_ctx.rdylist_act = 0;
 
        /* clear the context in the HMC */
@@ -2966,12 +2973,23 @@ static int i40e_configure_tx_ring(struct i40e_ring *ring)
        }
 
        /* Now associate this queue with this PCI function */
-       if (vsi->type == I40E_VSI_VMDQ2) {
-               qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
-               qtx_ctl |= ((vsi->id) << I40E_QTX_CTL_VFVM_INDX_SHIFT) &
-                          I40E_QTX_CTL_VFVM_INDX_MASK;
+       if (ring->ch) {
+               if (ring->ch->type == I40E_VSI_VMDQ2)
+                       qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
+               else
+                       return -EINVAL;
+
+               qtx_ctl |= (ring->ch->vsi_number <<
+                           I40E_QTX_CTL_VFVM_INDX_SHIFT) &
+                           I40E_QTX_CTL_VFVM_INDX_MASK;
        } else {
-               qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
+               if (vsi->type == I40E_VSI_VMDQ2) {
+                       qtx_ctl = I40E_QTX_CTL_VM_QUEUE;
+                       qtx_ctl |= ((vsi->id) << I40E_QTX_CTL_VFVM_INDX_SHIFT) &
+                                   I40E_QTX_CTL_VFVM_INDX_MASK;
+               } else {
+                       qtx_ctl = I40E_QTX_CTL_PF_QUEUE;
+               }
        }
 
        qtx_ctl |= ((hw->pf_id << I40E_QTX_CTL_PF_INDX_SHIFT) &
@@ -5161,6 +5179,643 @@ out:
        return ret;
 }
 
+/**
+ * i40e_remove_queue_channels - Remove queue channels for the TCs
+ * @vsi: VSI to be configured
+ *
+ * Remove queue channels for the TCs
+ **/
+static void i40e_remove_queue_channels(struct i40e_vsi *vsi)
+{
+       struct i40e_channel *ch, *ch_tmp;
+       int ret, i;
+
+       /* Reset rss size that was stored when reconfiguring rss for
+        * channel VSIs with non-power-of-2 queue count.
+        */
+       vsi->current_rss_size = 0;
+
+       /* perform cleanup for channels if they exist */
+       if (list_empty(&vsi->ch_list))
+               return;
+
+       list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+               struct i40e_vsi *p_vsi;
+
+               list_del(&ch->list);
+               p_vsi = ch->parent_vsi;
+               if (!p_vsi || !ch->initialized) {
+                       kfree(ch);
+                       continue;
+               }
+               /* Reset queue contexts */
+               for (i = 0; i < ch->num_queue_pairs; i++) {
+                       struct i40e_ring *tx_ring, *rx_ring;
+                       u16 pf_q;
+
+                       pf_q = ch->base_queue + i;
+                       tx_ring = vsi->tx_rings[pf_q];
+                       tx_ring->ch = NULL;
+
+                       rx_ring = vsi->rx_rings[pf_q];
+                       rx_ring->ch = NULL;
+               }
+
+               /* delete VSI from FW */
+               ret = i40e_aq_delete_element(&vsi->back->hw, ch->seid,
+                                            NULL);
+               if (ret)
+                       dev_err(&vsi->back->pdev->dev,
+                               "unable to remove channel (%d) for parent VSI(%d)\n",
+                               ch->seid, p_vsi->seid);
+               kfree(ch);
+       }
+       INIT_LIST_HEAD(&vsi->ch_list);
+}
+
+/**
+ * i40e_is_any_channel - channel exist or not
+ * @vsi: ptr to VSI to which channels are associated with
+ *
+ * Returns true or false if channel(s) exist for associated VSI or not
+ **/
+static bool i40e_is_any_channel(struct i40e_vsi *vsi)
+{
+       struct i40e_channel *ch, *ch_tmp;
+
+       list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+               if (ch->initialized)
+                       return true;
+       }
+
+       return false;
+}
+
+/**
+ * i40e_get_max_queues_for_channel
+ * @vsi: ptr to VSI to which channels are associated with
+ *
+ * Helper function which returns max value among the queue counts set on the
+ * channels/TCs created.
+ **/
+static int i40e_get_max_queues_for_channel(struct i40e_vsi *vsi)
+{
+       struct i40e_channel *ch, *ch_tmp;
+       int max = 0;
+
+       list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+               if (!ch->initialized)
+                       continue;
+               if (ch->num_queue_pairs > max)
+                       max = ch->num_queue_pairs;
+       }
+
+       return max;
+}
+
+/**
+ * i40e_validate_num_queues - validate num_queues w.r.t channel
+ * @pf: ptr to PF device
+ * @num_queues: number of queues
+ * @vsi: the parent VSI
+ * @reconfig_rss: indicates should the RSS be reconfigured or not
+ *
+ * This function validates number of queues in the context of new channel
+ * which is being established and determines if RSS should be reconfigured
+ * or not for parent VSI.
+ **/
+static int i40e_validate_num_queues(struct i40e_pf *pf, int num_queues,
+                                   struct i40e_vsi *vsi, bool *reconfig_rss)
+{
+       int max_ch_queues;
+
+       if (!reconfig_rss)
+               return -EINVAL;
+
+       *reconfig_rss = false;
+
+       if (num_queues > I40E_MAX_QUEUES_PER_CH) {
+               dev_err(&pf->pdev->dev,
+                       "Failed to create VMDq VSI. User requested num_queues (%d) > I40E_MAX_QUEUES_PER_VSI (%u)\n",
+                       num_queues, I40E_MAX_QUEUES_PER_CH);
+               return -EINVAL;
+       }
+
+       if (vsi->current_rss_size) {
+               if (num_queues > vsi->current_rss_size) {
+                       dev_dbg(&pf->pdev->dev,
+                               "Error: num_queues (%d) > vsi's current_size(%d)\n",
+                               num_queues, vsi->current_rss_size);
+                       return -EINVAL;
+               } else if ((num_queues < vsi->current_rss_size) &&
+                          (!is_power_of_2(num_queues))) {
+                       dev_dbg(&pf->pdev->dev,
+                               "Error: num_queues (%d) < vsi's current_size(%d), but not power of 2\n",
+                               num_queues, vsi->current_rss_size);
+                       return -EINVAL;
+               }
+       }
+
+       if (!is_power_of_2(num_queues)) {
+               /* Find the max num_queues configured for channel if channel
+                * exist.
+                * if channel exist, then enforce 'num_queues' to be more than
+                * max ever queues configured for channel.
+                */
+               max_ch_queues = i40e_get_max_queues_for_channel(vsi);
+               if (num_queues < max_ch_queues) {
+                       dev_dbg(&pf->pdev->dev,
+                               "Error: num_queues (%d) < max queues configured for channel(%d)\n",
+                               num_queues, max_ch_queues);
+                       return -EINVAL;
+               }
+               *reconfig_rss = true;
+       }
+
+       return 0;
+}
+
+/**
+ * i40e_vsi_reconfig_rss - reconfig RSS based on specified rss_size
+ * @vsi: the VSI being setup
+ * @rss_size: size of RSS, accordingly LUT gets reprogrammed
+ *
+ * This function reconfigures RSS by reprogramming LUTs using 'rss_size'
+ **/
+static int i40e_vsi_reconfig_rss(struct i40e_vsi *vsi, u16 rss_size)
+{
+       struct i40e_pf *pf = vsi->back;
+       u8 seed[I40E_HKEY_ARRAY_SIZE];
+       struct i40e_hw *hw = &pf->hw;
+       int local_rss_size;
+       u8 *lut;
+       int ret;
+
+       if (!vsi->rss_size)
+               return -EINVAL;
+
+       if (rss_size > vsi->rss_size)
+               return -EINVAL;
+
+       local_rss_size = min_t(int, vsi->rss_size, rss_size);
+       lut = kzalloc(vsi->rss_table_size, GFP_KERNEL);
+       if (!lut)
+               return -ENOMEM;
+
+       /* Ignoring user configured lut if there is one */
+       i40e_fill_rss_lut(pf, lut, vsi->rss_table_size, local_rss_size);
+
+       /* Use user configured hash key if there is one, otherwise
+        * use default.
+        */
+       if (vsi->rss_hkey_user)
+               memcpy(seed, vsi->rss_hkey_user, I40E_HKEY_ARRAY_SIZE);
+       else
+               netdev_rss_key_fill((void *)seed, I40E_HKEY_ARRAY_SIZE);
+
+       ret = i40e_config_rss(vsi, seed, lut, vsi->rss_table_size);
+       if (ret) {
+               dev_info(&pf->pdev->dev,
+                        "Cannot set RSS lut, err %s aq_err %s\n",
+                        i40e_stat_str(hw, ret),
+                        i40e_aq_str(hw, hw->aq.asq_last_status));
+               kfree(lut);
+               return ret;
+       }
+       kfree(lut);
+
+       /* Do the update w.r.t. storing rss_size */
+       if (!vsi->orig_rss_size)
+               vsi->orig_rss_size = vsi->rss_size;
+       vsi->current_rss_size = local_rss_size;
+
+       return ret;
+}
+
+/**
+ * i40e_channel_setup_queue_map - Setup a channel queue map
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ctxt: VSI context structure
+ * @ch: ptr to channel structure
+ *
+ * Setup queue map for a specific channel
+ **/
+static void i40e_channel_setup_queue_map(struct i40e_pf *pf,
+                                        struct i40e_vsi_context *ctxt,
+                                        struct i40e_channel *ch)
+{
+       u16 qcount, qmap, sections = 0;
+       u8 offset = 0;
+       int pow;
+
+       sections = I40E_AQ_VSI_PROP_QUEUE_MAP_VALID;
+       sections |= I40E_AQ_VSI_PROP_SCHED_VALID;
+
+       qcount = min_t(int, ch->num_queue_pairs, pf->num_lan_msix);
+       ch->num_queue_pairs = qcount;
+
+       /* find the next higher power-of-2 of num queue pairs */
+       pow = ilog2(qcount);
+       if (!is_power_of_2(qcount))
+               pow++;
+
+       qmap = (offset << I40E_AQ_VSI_TC_QUE_OFFSET_SHIFT) |
+               (pow << I40E_AQ_VSI_TC_QUE_NUMBER_SHIFT);
+
+       /* Setup queue TC[0].qmap for given VSI context */
+       ctxt->info.tc_mapping[0] = cpu_to_le16(qmap);
+
+       ctxt->info.up_enable_bits = 0x1; /* TC0 enabled */
+       ctxt->info.mapping_flags |= cpu_to_le16(I40E_AQ_VSI_QUE_MAP_CONTIG);
+       ctxt->info.queue_mapping[0] = cpu_to_le16(ch->base_queue);
+       ctxt->info.valid_sections |= cpu_to_le16(sections);
+}
+
+/**
+ * i40e_add_channel - add a channel by adding VSI
+ * @pf: ptr to PF device
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @ch: ptr to channel structure
+ *
+ * Add a channel (VSI) using add_vsi and queue_map
+ **/
+static int i40e_add_channel(struct i40e_pf *pf, u16 uplink_seid,
+                           struct i40e_channel *ch)
+{
+       struct i40e_hw *hw = &pf->hw;
+       struct i40e_vsi_context ctxt;
+       u8 enabled_tc = 0x1; /* TC0 enabled */
+       int ret;
+
+       if (ch->type != I40E_VSI_VMDQ2) {
+               dev_info(&pf->pdev->dev,
+                        "add new vsi failed, ch->type %d\n", ch->type);
+               return -EINVAL;
+       }
+
+       memset(&ctxt, 0, sizeof(ctxt));
+       ctxt.pf_num = hw->pf_id;
+       ctxt.vf_num = 0;
+       ctxt.uplink_seid = uplink_seid;
+       ctxt.connection_type = I40E_AQ_VSI_CONN_TYPE_NORMAL;
+       if (ch->type == I40E_VSI_VMDQ2)
+               ctxt.flags = I40E_AQ_VSI_TYPE_VMDQ2;
+
+       if (pf->flags & I40E_FLAG_VEB_MODE_ENABLED) {
+               ctxt.info.valid_sections |=
+                    cpu_to_le16(I40E_AQ_VSI_PROP_SWITCH_VALID);
+               ctxt.info.switch_id =
+                  cpu_to_le16(I40E_AQ_VSI_SW_ID_FLAG_ALLOW_LB);
+       }
+
+       /* Set queue map for a given VSI context */
+       i40e_channel_setup_queue_map(pf, &ctxt, ch);
+
+       /* Now time to create VSI */
+       ret = i40e_aq_add_vsi(hw, &ctxt, NULL);
+       if (ret) {
+               dev_info(&pf->pdev->dev,
+                        "add new vsi failed, err %s aq_err %s\n",
+                        i40e_stat_str(&pf->hw, ret),
+                        i40e_aq_str(&pf->hw,
+                                    pf->hw.aq.asq_last_status));
+               return -ENOENT;
+       }
+
+       /* Success, update channel */
+       ch->enabled_tc = enabled_tc;
+       ch->seid = ctxt.seid;
+       ch->vsi_number = ctxt.vsi_number;
+       ch->stat_counter_idx = cpu_to_le16(ctxt.info.stat_counter_idx);
+
+       /* copy just the sections touched not the entire info
+        * since not all sections are valid as returned by
+        * update vsi params
+        */
+       ch->info.mapping_flags = ctxt.info.mapping_flags;
+       memcpy(&ch->info.queue_mapping,
+              &ctxt.info.queue_mapping, sizeof(ctxt.info.queue_mapping));
+       memcpy(&ch->info.tc_mapping, ctxt.info.tc_mapping,
+              sizeof(ctxt.info.tc_mapping));
+
+       return 0;
+}
+
+static int i40e_channel_config_bw(struct i40e_vsi *vsi, struct i40e_channel *ch,
+                                 u8 *bw_share)
+{
+       struct i40e_aqc_configure_vsi_tc_bw_data bw_data;
+       i40e_status ret;
+       int i;
+
+       bw_data.tc_valid_bits = ch->enabled_tc;
+       for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
+               bw_data.tc_bw_credits[i] = bw_share[i];
+
+       ret = i40e_aq_config_vsi_tc_bw(&vsi->back->hw, ch->seid,
+                                      &bw_data, NULL);
+       if (ret) {
+               dev_info(&vsi->back->pdev->dev,
+                        "Config VSI BW allocation per TC failed, aq_err: %d for new_vsi->seid %u\n",
+                        vsi->back->hw.aq.asq_last_status, ch->seid);
+               return -EINVAL;
+       }
+
+       for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++)
+               ch->info.qs_handle[i] = bw_data.qs_handles[i];
+
+       return 0;
+}
+
+/**
+ * i40e_channel_config_tx_ring - config TX ring associated with new channel
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ch: ptr to channel structure
+ *
+ * Configure TX rings associated with channel (VSI) since queues are being
+ * from parent VSI.
+ **/
+static int i40e_channel_config_tx_ring(struct i40e_pf *pf,
+                                      struct i40e_vsi *vsi,
+                                      struct i40e_channel *ch)
+{
+       i40e_status ret;
+       int i;
+       u8 bw_share[I40E_MAX_TRAFFIC_CLASS] = {0};
+
+       /* Enable ETS TCs with equal BW Share for now across all VSIs */
+       for (i = 0; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+               if (ch->enabled_tc & BIT(i))
+                       bw_share[i] = 1;
+       }
+
+       /* configure BW for new VSI */
+       ret = i40e_channel_config_bw(vsi, ch, bw_share);
+       if (ret) {
+               dev_info(&vsi->back->pdev->dev,
+                        "Failed configuring TC map %d for channel (seid %u)\n",
+                        ch->enabled_tc, ch->seid);
+               return ret;
+       }
+
+       for (i = 0; i < ch->num_queue_pairs; i++) {
+               struct i40e_ring *tx_ring, *rx_ring;
+               u16 pf_q;
+
+               pf_q = ch->base_queue + i;
+
+               /* Get to TX ring ptr of main VSI, for re-setup TX queue
+                * context
+                */
+               tx_ring = vsi->tx_rings[pf_q];
+               tx_ring->ch = ch;
+
+               /* Get the RX ring ptr */
+               rx_ring = vsi->rx_rings[pf_q];
+               rx_ring->ch = ch;
+       }
+
+       return 0;
+}
+
+/**
+ * i40e_setup_hw_channel - setup new channel
+ * @pf: ptr to PF device
+ * @vsi: the VSI being setup
+ * @ch: ptr to channel structure
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @type: type of channel to be created (VMDq2/VF)
+ *
+ * Setup new channel (VSI) based on specified type (VMDq2/VF)
+ * and configures TX rings accordingly
+ **/
+static inline int i40e_setup_hw_channel(struct i40e_pf *pf,
+                                       struct i40e_vsi *vsi,
+                                       struct i40e_channel *ch,
+                                       u16 uplink_seid, u8 type)
+{
+       int ret;
+
+       ch->initialized = false;
+       ch->base_queue = vsi->next_base_queue;
+       ch->type = type;
+
+       /* Proceed with creation of channel (VMDq2) VSI */
+       ret = i40e_add_channel(pf, uplink_seid, ch);
+       if (ret) {
+               dev_info(&pf->pdev->dev,
+                        "failed to add_channel using uplink_seid %u\n",
+                        uplink_seid);
+               return ret;
+       }
+
+       /* Mark the successful creation of channel */
+       ch->initialized = true;
+
+       /* Reconfigure TX queues using QTX_CTL register */
+       ret = i40e_channel_config_tx_ring(pf, vsi, ch);
+       if (ret) {
+               dev_info(&pf->pdev->dev,
+                        "failed to configure TX rings for channel %u\n",
+                        ch->seid);
+               return ret;
+       }
+
+       /* update 'next_base_queue' */
+       vsi->next_base_queue = vsi->next_base_queue + ch->num_queue_pairs;
+       dev_dbg(&pf->pdev->dev,
+               "Added channel: vsi_seid %u, vsi_number %u, stat_counter_idx %u, num_queue_pairs %u, pf->next_base_queue %d\n",
+               ch->seid, ch->vsi_number, ch->stat_counter_idx,
+               ch->num_queue_pairs,
+               vsi->next_base_queue);
+       return ret;
+}
+
+/**
+ * i40e_setup_channel - setup new channel using uplink element
+ * @pf: ptr to PF device
+ * @type: type of channel to be created (VMDq2/VF)
+ * @uplink_seid: underlying HW switching element (VEB) ID
+ * @ch: ptr to channel structure
+ *
+ * Setup new channel (VSI) based on specified type (VMDq2/VF)
+ * and uplink switching element (uplink_seid)
+ **/
+static bool i40e_setup_channel(struct i40e_pf *pf, struct i40e_vsi *vsi,
+                              struct i40e_channel *ch)
+{
+       u8 vsi_type;
+       u16 seid;
+       int ret;
+
+       if (vsi->type == I40E_VSI_MAIN) {
+               vsi_type = I40E_VSI_VMDQ2;
+       } else {
+               dev_err(&pf->pdev->dev, "unsupported parent vsi type(%d)\n",
+                       vsi->type);
+               return false;
+       }
+
+       /* underlying switching element */
+       seid = pf->vsi[pf->lan_vsi]->uplink_seid;
+
+       /* create channel (VSI), configure TX rings */
+       ret = i40e_setup_hw_channel(pf, vsi, ch, seid, vsi_type);
+       if (ret) {
+               dev_err(&pf->pdev->dev, "failed to setup hw_channel\n");
+               return false;
+       }
+
+       return ch->initialized ? true : false;
+}
+
+/**
+ * i40e_create_queue_channel - function to create channel
+ * @vsi: VSI to be configured
+ * @ch: ptr to channel (it contains channel specific params)
+ *
+ * This function creates channel (VSI) using num_queues specified by user,
+ * reconfigs RSS if needed.
+ **/
+int i40e_create_queue_channel(struct i40e_vsi *vsi,
+                             struct i40e_channel *ch)
+{
+       struct i40e_pf *pf = vsi->back;
+       bool reconfig_rss;
+       int err;
+
+       if (!ch)
+               return -EINVAL;
+
+       if (!ch->num_queue_pairs) {
+               dev_err(&pf->pdev->dev, "Invalid num_queues requested: %d\n",
+                       ch->num_queue_pairs);
+               return -EINVAL;
+       }
+
+       /* validate user requested num_queues for channel */
+       err = i40e_validate_num_queues(pf, ch->num_queue_pairs, vsi,
+                                      &reconfig_rss);
+       if (err) {
+               dev_info(&pf->pdev->dev, "Failed to validate num_queues (%d)\n",
+                        ch->num_queue_pairs);
+               return -EINVAL;
+       }
+
+       /* By default we are in VEPA mode, if this is the first VF/VMDq
+        * VSI to be added switch to VEB mode.
+        */
+       if ((!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) ||
+           (!i40e_is_any_channel(vsi))) {
+               if (!is_power_of_2(vsi->tc_config.tc_info[0].qcount)) {
+                       dev_dbg(&pf->pdev->dev,
+                               "Failed to create channel. Override queues (%u) not power of 2\n",
+                               vsi->tc_config.tc_info[0].qcount);
+                       return -EINVAL;
+               }
+
+               if (!(pf->flags & I40E_FLAG_VEB_MODE_ENABLED)) {
+                       pf->flags |= I40E_FLAG_VEB_MODE_ENABLED;
+
+                       if (vsi->type == I40E_VSI_MAIN) {
+                               if (pf->flags & I40E_FLAG_TC_MQPRIO)
+                                       i40e_do_reset(pf, I40E_PF_RESET_FLAG,
+                                                     true);
+                               else
+                                       i40e_do_reset_safe(pf,
+                                                          I40E_PF_RESET_FLAG);
+                       }
+               }
+               /* now onwards for main VSI, number of queues will be value
+                * of TC0's queue count
+                */
+       }
+
+       /* By this time, vsi->cnt_q_avail shall be set to non-zero and
+        * it should be more than num_queues
+        */
+       if (!vsi->cnt_q_avail || vsi->cnt_q_avail < ch->num_queue_pairs) {
+               dev_dbg(&pf->pdev->dev,
+                       "Error: cnt_q_avail (%u) less than num_queues %d\n",
+                       vsi->cnt_q_avail, ch->num_queue_pairs);
+               return -EINVAL;
+       }
+
+       /* reconfig_rss only if vsi type is MAIN_VSI */
+       if (reconfig_rss && (vsi->type == I40E_VSI_MAIN)) {
+               err = i40e_vsi_reconfig_rss(vsi, ch->num_queue_pairs);
+               if (err) {
+                       dev_info(&pf->pdev->dev,
+                                "Error: unable to reconfig rss for num_queues (%u)\n",
+                                ch->num_queue_pairs);
+                       return -EINVAL;
+               }
+       }
+
+       if (!i40e_setup_channel(pf, vsi, ch)) {
+               dev_info(&pf->pdev->dev, "Failed to setup channel\n");
+               return -EINVAL;
+       }
+
+       dev_info(&pf->pdev->dev,
+                "Setup channel (id:%u) utilizing num_queues %d\n",
+                ch->seid, ch->num_queue_pairs);
+
+       /* in case of VF, this will be main SRIOV VSI */
+       ch->parent_vsi = vsi;
+
+       /* and update main_vsi's count for queue_available to use */
+       vsi->cnt_q_avail -= ch->num_queue_pairs;
+
+       return 0;
+}
+
+/**
+ * i40e_configure_queue_channels - Add queue channel for the given TCs
+ * @vsi: VSI to be configured
+ *
+ * Configures queue channel mapping to the given TCs
+ **/
+static int i40e_configure_queue_channels(struct i40e_vsi *vsi)
+{
+       struct i40e_channel *ch;
+       int ret = 0, i;
+
+       /* Create app vsi with the TCs. Main VSI with TC0 is already set up */
+       for (i = 1; i < I40E_MAX_TRAFFIC_CLASS; i++) {
+               if (vsi->tc_config.enabled_tc & BIT(i)) {
+                       ch = kzalloc(sizeof(*ch), GFP_KERNEL);
+                       if (!ch) {
+                               ret = -ENOMEM;
+                               goto err_free;
+                       }
+
+                       INIT_LIST_HEAD(&ch->list);
+                       ch->num_queue_pairs =
+                               vsi->tc_config.tc_info[i].qcount;
+                       ch->base_queue =
+                               vsi->tc_config.tc_info[i].qoffset;
+
+                       list_add_tail(&ch->list, &vsi->ch_list);
+
+                       ret = i40e_create_queue_channel(vsi, ch);
+                       if (ret) {
+                               dev_err(&vsi->back->pdev->dev,
+                                       "Failed creating queue channel with TC%d: queues %d\n",
+                                       i, ch->num_queue_pairs);
+                               goto err_free;
+                       }
+               }
+       }
+       return ret;
+
+err_free:
+       i40e_remove_queue_channels(vsi);
+       return ret;
+}
+
 /**
  * i40e_veb_config_tc - Configure TCs for given VEB
  * @veb: given VEB
@@ -5612,10 +6267,18 @@ static int i40e_setup_tc(struct net_device *netdev, u8 tc)
                goto exit;
        }
 
-       /* Unquiesce VSI */
-       i40e_unquiesce_vsi(vsi);
+       if (pf->flags & I40E_FLAG_TC_MQPRIO) {
+               ret = i40e_configure_queue_channels(vsi);
+               if (ret) {
+                       netdev_info(netdev,
+                                   "Failed configuring queue channels\n");
+                       goto exit;
+               }
+       }
 
 exit:
+       /* Unquiesce VSI */
+       i40e_unquiesce_vsi(vsi);
        return ret;
 }
 
@@ -7030,6 +7693,35 @@ static void i40e_fdir_teardown(struct i40e_pf *pf)
                i40e_vsi_release(vsi);
 }
 
+/**
+ * i40e_rebuild_channels - Rebuilds channel VSIs if they existed before reset
+ * @vsi: PF main vsi
+ *
+ * Rebuilds channel VSIs if they existed before reset
+ **/
+static int i40e_rebuild_channels(struct i40e_vsi *vsi)
+{
+       struct i40e_channel *ch, *ch_tmp;
+       i40e_status ret;
+
+       if (list_empty(&vsi->ch_list))
+               return 0;
+
+       list_for_each_entry_safe(ch, ch_tmp, &vsi->ch_list, list) {
+               if (!ch->initialized)
+                       break;
+               /* Proceed with creation of channel (VMDq2) VSI */
+               ret = i40e_add_channel(vsi->back, vsi->uplink_seid, ch);
+               if (ret) {
+                       dev_info(&vsi->back->pdev->dev,
+                                "failed to rebuild channels using uplink_seid %u\n",
+                                vsi->uplink_seid);
+                       return ret;
+               }
+       }
+       return 0;
+}
+
 /**
  * i40e_prep_for_reset - prep for the core to reset
  * @pf: board private structure
@@ -7295,6 +7987,13 @@ static void i40e_rebuild(struct i40e_pf *pf, bool reinit, bool lock_acquired)
                }
        }
 
+       /* PF Main VSI is rebuild by now, go ahead and rebuild channel VSIs
+        * for this main VSI if they exist
+        */
+       ret = i40e_rebuild_channels(pf->vsi[pf->lan_vsi]);
+       if (ret)
+               goto end_unlock;
+
        /* Reconfigure hardware for allowing smaller MSS in the case
         * of TSO, so that we avoid the MDD being fired and causing
         * a reset in the case of small MSS+TSO.
@@ -11608,6 +12307,7 @@ static int i40e_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
                dev_info(&pdev->dev, "setup_pf_switch failed: %d\n", err);
                goto err_vsis;
        }
+       INIT_LIST_HEAD(&pf->vsi[pf->lan_vsi]->ch_list);
 
        /* Make sure flow control is set according to current settings */
        err = i40e_set_fc(hw, &set_fc_aq_fail, true);
index ff57ae451524f2994b40155a834acd90f7685fb6..fbae1182e2eabc9d278969d382df06c3fb51762b 100644 (file)
@@ -426,6 +426,8 @@ struct i40e_ring {
                                         * i40e_clean_rx_ring_irq() is called
                                         * for this ring.
                                         */
+
+       struct i40e_channel *ch;
 } ____cacheline_internodealigned_in_smp;
 
 static inline bool ring_uses_build_skb(struct i40e_ring *ring)