]> git.dujemihanovic.xyz Git - u-boot.git/commitdiff
mips: octeon: Add cvmx-helper-pko3.c
authorAaron Williams <awilliams@marvell.com>
Thu, 7 Apr 2022 07:11:21 +0000 (09:11 +0200)
committerStefan Roese <sr@denx.de>
Wed, 4 May 2022 01:38:21 +0000 (03:38 +0200)
Import cvmx-helper-pko3.c from 2013 U-Boot. It will be used by the later
added drivers to support networking on the MIPS Octeon II / III
platforms.

Signed-off-by: Aaron Williams <awilliams@marvell.com>
Signed-off-by: Stefan Roese <sr@denx.de>
arch/mips/mach-octeon/cvmx-helper-pko3.c [new file with mode: 0644]

diff --git a/arch/mips/mach-octeon/cvmx-helper-pko3.c b/arch/mips/mach-octeon/cvmx-helper-pko3.c
new file mode 100644 (file)
index 0000000..37fe45d
--- /dev/null
@@ -0,0 +1,1088 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2018-2022 Marvell International Ltd.
+ *
+ * PKOv3 helper file
+ */
+
+#include <errno.h>
+#include <log.h>
+#include <time.h>
+#include <linux/delay.h>
+
+#include <mach/cvmx-regs.h>
+#include <mach/cvmx-csr.h>
+#include <mach/cvmx-bootmem.h>
+#include <mach/octeon-model.h>
+#include <mach/cvmx-fuse.h>
+#include <mach/octeon-feature.h>
+#include <mach/cvmx-qlm.h>
+#include <mach/octeon_qlm.h>
+#include <mach/cvmx-pcie.h>
+#include <mach/cvmx-coremask.h>
+#include <mach/cvmx-range.h>
+#include <mach/cvmx-global-resources.h>
+
+#include <mach/cvmx-agl-defs.h>
+#include <mach/cvmx-bgxx-defs.h>
+#include <mach/cvmx-ciu-defs.h>
+#include <mach/cvmx-gmxx-defs.h>
+#include <mach/cvmx-gserx-defs.h>
+#include <mach/cvmx-ilk-defs.h>
+#include <mach/cvmx-ipd-defs.h>
+#include <mach/cvmx-pcsx-defs.h>
+#include <mach/cvmx-pcsxx-defs.h>
+#include <mach/cvmx-pki-defs.h>
+#include <mach/cvmx-pko-defs.h>
+#include <mach/cvmx-xcv-defs.h>
+
+#include <mach/cvmx-hwpko.h>
+#include <mach/cvmx-ilk.h>
+#include <mach/cvmx-ipd.h>
+#include <mach/cvmx-pki.h>
+#include <mach/cvmx-pko3.h>
+#include <mach/cvmx-pko3-queue.h>
+#include <mach/cvmx-pko3-resources.h>
+
+#include <mach/cvmx-helper.h>
+#include <mach/cvmx-helper-board.h>
+#include <mach/cvmx-helper-cfg.h>
+
+#include <mach/cvmx-helper-bgx.h>
+#include <mach/cvmx-helper-cfg.h>
+#include <mach/cvmx-helper-util.h>
+#include <mach/cvmx-helper-pki.h>
+
+/* channels are present at L2 queue level by default */
+static const enum cvmx_pko3_level_e cvmx_pko_default_channel_level =
+       CVMX_PKO_L2_QUEUES;
+
+static const int debug;
+
+static int __pko_pkt_budget, __pko_pkt_quota;
+
+/* These global variables are relevant for boot CPU only */
+static cvmx_fpa3_gaura_t __cvmx_pko3_aura[CVMX_MAX_NODES];
+
+/* This constant can not be modified, defined here for clarity only */
+#define CVMX_PKO3_POOL_BUFFER_SIZE 4096 /* 78XX PKO requires 4KB */
+
+/**
+ * @INTERNAL
+ *
+ * Build an owner tag based on interface/port
+ */
+static int __cvmx_helper_pko3_res_owner(int ipd_port)
+{
+       int res_owner;
+       const int res_owner_pfix = 0x19d0 << 14;
+
+       ipd_port &= 0x3fff; /* 12-bit for local CHAN_E value + node */
+
+       res_owner = res_owner_pfix | ipd_port;
+
+       return res_owner;
+}
+
+/**
+ * Configure an AURA/POOL designated for PKO internal use.
+ *
+ * This pool is used for (a) memory buffers that store PKO descriptor queues,
+ * (b) buffers for use with PKO_SEND_JUMP_S sub-header.
+ *
+ * The buffers of type (a) are never accessed by software, and their number
+ * should be at least equal to 4 times the number of descriptor queues
+ * in use.
+ *
+ * Type (b) buffers are consumed by PKO3 command-composition code,
+ * and are released by the hardware upon completion of transmission.
+ *
+ * @returns -1 if the pool could not be established or 12-bit AURA
+ * that includes the node number for use in PKO3 initialization call.
+ *
+ * NOTE: Linux kernel should pass its own aura to PKO3 initialization
+ * function so that the buffers can be mapped into kernel space
+ * for when software needs to adccess their contents.
+ *
+ */
+static int __cvmx_pko3_config_memory(unsigned int node)
+{
+       cvmx_fpa3_gaura_t aura;
+       int aura_num;
+       unsigned int buf_count;
+       bool small_mem;
+       int i, num_intf = 0;
+       const unsigned int pkt_per_buf =
+               (CVMX_PKO3_POOL_BUFFER_SIZE / sizeof(u64) / 16);
+       const unsigned int base_buf_count = 1024 * 4;
+
+       /* Simulator has limited memory, but uses one interface at a time */
+       //      small_mem = cvmx_sysinfo_get()->board_type == CVMX_BOARD_TYPE_SIM;
+       small_mem = false;
+
+       /* Count the number of live interfaces */
+       for (i = 0; i < cvmx_helper_get_number_of_interfaces(); i++) {
+               int xiface = cvmx_helper_node_interface_to_xiface(node, i);
+
+               if (CVMX_HELPER_INTERFACE_MODE_DISABLED !=
+                   cvmx_helper_interface_get_mode(xiface))
+                       num_intf++;
+       }
+
+       buf_count = 1024;
+       __pko_pkt_quota = buf_count * pkt_per_buf;
+       __pko_pkt_budget = __pko_pkt_quota * num_intf;
+       (void)small_mem;
+       (void)base_buf_count;
+
+       if (debug)
+               debug("%s: Creating AURA with %u buffers for up to %d total packets, %d packets per interface\n",
+                     __func__, buf_count, __pko_pkt_budget, __pko_pkt_quota);
+
+       aura = cvmx_fpa3_setup_aura_and_pool(node, -1, "PKO3 AURA", NULL,
+                                            CVMX_PKO3_POOL_BUFFER_SIZE,
+                                            buf_count);
+
+       if (!__cvmx_fpa3_aura_valid(aura)) {
+               printf("ERROR: %s AURA create failed\n", __func__);
+               return -1;
+       }
+
+       aura_num = aura.node << 10 | aura.laura;
+
+       /* Store handle for destruction */
+       __cvmx_pko3_aura[node] = aura;
+
+       return aura_num;
+}
+
+/** Initialize a channelized port
+ * This is intended for LOOP, ILK and NPI interfaces which have one MAC
+ * per interface and need a channel per subinterface (e.g. ring).
+ * Each channel then may have 'num_queues' descriptor queues
+ * attached to it, which can also be prioritized or fair.
+ */
+static int __cvmx_pko3_config_chan_interface(int xiface, unsigned int num_chans,
+                                            u8 num_queues, bool prioritized)
+{
+       int l1_q_num;
+       int l2_q_base;
+       enum cvmx_pko3_level_e level;
+       int res;
+       int parent_q, child_q;
+       unsigned int chan, dq;
+       int pko_mac_num;
+       u16 ipd_port;
+       int res_owner, prio;
+       unsigned int i;
+       struct cvmx_xiface xi = cvmx_helper_xiface_to_node_interface(xiface);
+       unsigned int node = xi.node;
+       char b1[12];
+
+       if (num_queues == 0)
+               num_queues = 1;
+       if ((cvmx_pko3_num_level_queues(CVMX_PKO_DESCR_QUEUES) / num_chans) < 3)
+               num_queues = 1;
+
+       if (prioritized && num_queues > 1)
+               prio = num_queues;
+       else
+               prio = -1;
+
+       if (debug)
+               debug("%s: configuring xiface %u:%u with %u chans %u queues each\n",
+                     __func__, xi.node, xi.interface, num_chans, num_queues);
+
+       /* all channels all go to the same mac */
+       pko_mac_num = __cvmx_pko3_get_mac_num(xiface, 0);
+       if (pko_mac_num < 0) {
+               printf("ERROR: %s: Invalid interface\n", __func__);
+               return -1;
+       }
+
+       /* Resources of all channels on this port have common owner */
+       ipd_port = cvmx_helper_get_ipd_port(xiface, 0);
+
+       /* Build an identifiable owner */
+       res_owner = __cvmx_helper_pko3_res_owner(ipd_port);
+
+       /* Start configuration at L1/PQ */
+       level = CVMX_PKO_PORT_QUEUES;
+
+       /* Reserve port queue to make sure the MAC is not already configured */
+       l1_q_num = cvmx_pko_alloc_queues(node, level, res_owner, -1, 1);
+
+       if (l1_q_num < 0) {
+               printf("ERROR: %s: Reserving L1 PQ\n", __func__);
+               return -1;
+       }
+
+       res = cvmx_pko3_pq_config(node, pko_mac_num, l1_q_num);
+       if (res < 0) {
+               printf("ERROR: %s: Configuring L1 PQ\n", __func__);
+               return -1;
+       }
+
+       /* next queue level = L2/SQ */
+       level = __cvmx_pko3_sq_lvl_next(level);
+
+       /* allocate level 2 queues, one per channel */
+       l2_q_base =
+               cvmx_pko_alloc_queues(node, level, res_owner, -1, num_chans);
+       if (l2_q_base < 0) {
+               printf("ERROR: %s: allocation L2 SQ\n", __func__);
+               return -1;
+       }
+
+       /* Configre <num_chans> L2 children for PQ, non-prioritized */
+       res = cvmx_pko3_sq_config_children(node, level, l1_q_num, l2_q_base,
+                                          num_chans, -1);
+
+       if (res < 0) {
+               printf("ERROR: %s: Failed channel queues\n", __func__);
+               return -1;
+       }
+
+       /* map channels to l2 queues */
+       for (chan = 0; chan < num_chans; chan++) {
+               ipd_port = cvmx_helper_get_ipd_port(xiface, chan);
+               cvmx_pko3_map_channel(node, l1_q_num, l2_q_base + chan,
+                                     ipd_port);
+       }
+
+       /* next queue level = L3/SQ */
+       level = __cvmx_pko3_sq_lvl_next(level);
+       parent_q = l2_q_base;
+
+       do {
+               child_q = cvmx_pko_alloc_queues(node, level, res_owner, -1,
+                                               num_chans);
+
+               if (child_q < 0) {
+                       printf("ERROR: %s: allocating %s\n", __func__,
+                              __cvmx_pko3_sq_str(b1, level, child_q));
+                       return -1;
+               }
+
+               for (i = 0; i < num_chans; i++) {
+                       res = cvmx_pko3_sq_config_children(
+                               node, level, parent_q + i, child_q + i, 1, 1);
+
+                       if (res < 0) {
+                               printf("ERROR: %s: configuring %s\n", __func__,
+                                      __cvmx_pko3_sq_str(b1, level, child_q));
+                               return -1;
+                       }
+
+               } /* for i */
+
+               parent_q = child_q;
+               level = __cvmx_pko3_sq_lvl_next(level);
+
+               /* Terminate loop on DQ level, it has special handling */
+       } while (level != CVMX_PKO_DESCR_QUEUES &&
+                level != CVMX_PKO_LEVEL_INVAL);
+
+       if (level != CVMX_PKO_DESCR_QUEUES) {
+               printf("ERROR: %s: level sequence error\n", __func__);
+               return -1;
+       }
+
+       /* Configure DQs, num_dqs per chan */
+       for (chan = 0; chan < num_chans; chan++) {
+               res = cvmx_pko_alloc_queues(node, level, res_owner, -1,
+                                           num_queues);
+
+               if (res < 0)
+                       goto _fail;
+               dq = res;
+
+               if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_0) && (dq & 7))
+                       debug("WARNING: %s: DQ# %u not integral of 8\n",
+                             __func__, dq);
+
+               res = cvmx_pko3_sq_config_children(node, level, parent_q + chan,
+                                                  dq, num_queues, prio);
+               if (res < 0)
+                       goto _fail;
+
+               /* register DQ range with the translation table */
+               res = __cvmx_pko3_ipd_dq_register(xiface, chan, dq, num_queues);
+               if (res < 0)
+                       goto _fail;
+       }
+
+       return 0;
+_fail:
+       debug("ERROR: %s: configuring queues for xiface %u:%u chan %u\n",
+             __func__, xi.node, xi.interface, i);
+       return -1;
+}
+
+/** Initialize a single Ethernet port with PFC-style channels
+ *
+ * One interface can contain multiple ports, this function is per-port
+ * Here, a physical port is allocated 8 logical channel, one per VLAN
+ * tag priority, one DQ is assigned to each channel, and all 8 DQs
+ * are registered for that IPD port.
+ * Note that the DQs are arrange such that the Ethernet QoS/PCP field
+ * can be used as an offset to the value returned by cvmx_pko_base_queue_get().
+ *
+ * For HighGig2 mode, 16 channels may be desired, instead of 8,
+ * but this function does not support that.
+ */
+static int __cvmx_pko3_config_pfc_interface(int xiface, unsigned int port)
+{
+       enum cvmx_pko3_level_e level;
+       int pko_mac_num;
+       int l1_q_num, l2_q_base;
+       int child_q, parent_q;
+       int dq_base;
+       int res;
+       const unsigned int num_chans = 8;
+       cvmx_xiface_t xi = cvmx_helper_xiface_to_node_interface(xiface);
+       unsigned int node = xi.node;
+       u16 ipd_port;
+       int res_owner;
+       char b1[12];
+       unsigned int i;
+
+       if (debug)
+               debug("%s: configuring xiface %u:%u port %u with %u PFC channels\n",
+                     __func__, node, xi.interface, port, num_chans);
+
+       /* Get MAC number for the iface/port */
+       pko_mac_num = __cvmx_pko3_get_mac_num(xiface, port);
+       if (pko_mac_num < 0) {
+               printf("ERROR: %s: Invalid interface\n", __func__);
+               return -1;
+       }
+
+       ipd_port = cvmx_helper_get_ipd_port(xiface, port);
+
+       /* Build an identifiable owner identifier */
+       res_owner = __cvmx_helper_pko3_res_owner(ipd_port);
+
+       level = CVMX_PKO_PORT_QUEUES;
+
+       /* Allocate port queue to make sure the MAC is not already configured */
+       l1_q_num = cvmx_pko_alloc_queues(node, level, res_owner, -1, 1);
+
+       if (l1_q_num < 0) {
+               printf("ERROR: %s: allocation L1 PQ\n", __func__);
+               return -1;
+       }
+
+       res = cvmx_pko3_pq_config(xi.node, pko_mac_num, l1_q_num);
+       if (res < 0) {
+               printf("ERROR: %s: Configuring %s\n", __func__,
+                      __cvmx_pko3_sq_str(b1, level, l1_q_num));
+               return -1;
+       }
+
+       /* Determine the next queue level */
+       level = __cvmx_pko3_sq_lvl_next(level);
+
+       /* Allocate 'num_chans' L2 queues, one per channel */
+       l2_q_base =
+               cvmx_pko_alloc_queues(node, level, res_owner, -1, num_chans);
+       if (l2_q_base < 0) {
+               printf("ERROR: %s: allocation L2 SQ\n", __func__);
+               return -1;
+       }
+
+       /* Configre <num_chans> L2 children for PQ, with static priority */
+       res = cvmx_pko3_sq_config_children(node, level, l1_q_num, l2_q_base,
+                                          num_chans, num_chans);
+
+       if (res < 0) {
+               printf("ERROR: %s: Configuring %s for PFC\n", __func__,
+                      __cvmx_pko3_sq_str(b1, level, l1_q_num));
+               return -1;
+       }
+
+       /* Map each of the allocated channels */
+       for (i = 0; i < num_chans; i++) {
+               u16 chan;
+
+               /* Get CHAN_E value for this PFC channel, PCP in low 3 bits */
+               chan = ipd_port | cvmx_helper_prio2qos(i);
+
+               cvmx_pko3_map_channel(node, l1_q_num, l2_q_base + i, chan);
+       }
+
+       /* Iterate through the levels until DQ and allocate 'num_chans'
+        * consecutive queues at each level and hook them up
+        * one-to-one with the parent level queues
+        */
+
+       parent_q = l2_q_base;
+       level = __cvmx_pko3_sq_lvl_next(level);
+
+       do {
+               child_q = cvmx_pko_alloc_queues(node, level, res_owner, -1,
+                                               num_chans);
+
+               if (child_q < 0) {
+                       printf("ERROR: %s: allocating %s\n", __func__,
+                              __cvmx_pko3_sq_str(b1, level, child_q));
+                       return -1;
+               }
+
+               for (i = 0; i < num_chans; i++) {
+                       res = cvmx_pko3_sq_config_children(
+                               node, level, parent_q + i, child_q + i, 1, 1);
+
+                       if (res < 0) {
+                               printf("ERROR: %s: configuring %s\n", __func__,
+                                      __cvmx_pko3_sq_str(b1, level, child_q));
+                               return -1;
+                       }
+
+               } /* for i */
+
+               parent_q = child_q;
+               level = __cvmx_pko3_sq_lvl_next(level);
+
+               /* Terminate loop on DQ level, it has special handling */
+       } while (level != CVMX_PKO_DESCR_QUEUES &&
+                level != CVMX_PKO_LEVEL_INVAL);
+
+       if (level != CVMX_PKO_DESCR_QUEUES) {
+               printf("ERROR: %s: level sequence error\n", __func__);
+               return -1;
+       }
+
+       dq_base = cvmx_pko_alloc_queues(node, level, res_owner, -1, num_chans);
+       if (dq_base < 0) {
+               printf("ERROR: %s: allocating %s\n", __func__,
+                      __cvmx_pko3_sq_str(b1, level, dq_base));
+               return -1;
+       }
+
+       /* Configure DQs in QoS order, so that QoS/PCP can be index */
+       for (i = 0; i < num_chans; i++) {
+               int dq_num = dq_base + cvmx_helper_prio2qos(i);
+
+               res = cvmx_pko3_sq_config_children(node, level, parent_q + i,
+                                                  dq_num, 1, 1);
+               if (res < 0) {
+                       printf("ERROR: %s: configuring %s\n", __func__,
+                              __cvmx_pko3_sq_str(b1, level, dq_num));
+                       return -1;
+               }
+       }
+
+       /* register entire DQ range with the IPD translation table */
+       __cvmx_pko3_ipd_dq_register(xiface, port, dq_base, num_chans);
+
+       return 0;
+}
+
+/**
+ * Initialize a simple interface with a a given number of
+ * fair or prioritized queues.
+ * This function will assign one channel per sub-interface.
+ */
+int __cvmx_pko3_config_gen_interface(int xiface, uint8_t subif, u8 num_queues,
+                                    bool prioritized)
+{
+       cvmx_xiface_t xi = cvmx_helper_xiface_to_node_interface(xiface);
+       u8 node = xi.node;
+       int l1_q_num;
+       int parent_q, child_q;
+       int dq;
+       int res, res_owner;
+       int pko_mac_num;
+       enum cvmx_pko3_level_e level;
+       u16 ipd_port;
+       int static_pri;
+       char b1[12];
+
+       num_queues = 1;
+
+       if (num_queues == 0) {
+               num_queues = 1;
+               printf("WARNING: %s: xiface %#x misconfigured\n", __func__,
+                      xiface);
+       }
+
+       /* Configure DQs relative priority (a.k.a. scheduling) */
+       if (prioritized) {
+               /* With 8 queues or fewer, use static priority, else WRR */
+               static_pri = (num_queues < 9) ? num_queues : 0;
+       } else {
+               /* Set equal-RR scheduling among queues */
+               static_pri = -1;
+       }
+
+       if (debug)
+               debug("%s: configuring xiface %u:%u/%u nq=%u %s\n", __func__,
+                     xi.node, xi.interface, subif, num_queues,
+                     (prioritized) ? "qos" : "fair");
+
+       /* Get MAC number for the iface/port */
+       pko_mac_num = __cvmx_pko3_get_mac_num(xiface, subif);
+       if (pko_mac_num < 0) {
+               printf("ERROR: %s: Invalid interface %u:%u\n", __func__,
+                      xi.node, xi.interface);
+               return -1;
+       }
+
+       ipd_port = cvmx_helper_get_ipd_port(xiface, subif);
+
+       if (debug)
+               debug("%s: xiface %u:%u/%u ipd_port=%#03x\n", __func__, xi.node,
+                     xi.interface, subif, ipd_port);
+
+       /* Build an identifiable owner identifier */
+       res_owner = __cvmx_helper_pko3_res_owner(ipd_port);
+
+       level = CVMX_PKO_PORT_QUEUES;
+
+       /* Reserve port queue to make sure the MAC is not already configured */
+       l1_q_num = cvmx_pko_alloc_queues(node, level, res_owner, -1, 1);
+
+       if (l1_q_num < 0) {
+               printf("ERROR %s: xiface %u:%u/%u failed allocation L1 PQ\n",
+                      __func__, xi.node, xi.interface, subif);
+               return -1;
+       }
+
+       res = cvmx_pko3_pq_config(node, pko_mac_num, l1_q_num);
+       if (res < 0) {
+               printf("ERROR %s: Configuring L1 PQ\n", __func__);
+               return -1;
+       }
+
+       parent_q = l1_q_num;
+
+       /* Determine the next queue level */
+       level = __cvmx_pko3_sq_lvl_next(level);
+
+       /* Simply chain queues 1-to-1 from L2 to one before DQ level */
+       do {
+               /* allocate next level queue */
+               child_q = cvmx_pko_alloc_queues(node, level, res_owner, -1, 1);
+
+               if (child_q < 0) {
+                       printf("ERROR: %s: allocating %s\n", __func__,
+                              __cvmx_pko3_sq_str(b1, level, child_q));
+                       return -1;
+               }
+
+               /* Configre newly allocated queue */
+               res = cvmx_pko3_sq_config_children(node, level, parent_q,
+                                                  child_q, 1, 1);
+
+               if (res < 0) {
+                       printf("ERROR: %s: configuring %s\n", __func__,
+                              __cvmx_pko3_sq_str(b1, level, child_q));
+                       return -1;
+               }
+
+               /* map IPD/channel to L2/L3 queues */
+               if (level == cvmx_pko_default_channel_level)
+                       cvmx_pko3_map_channel(node, l1_q_num, child_q,
+                                             ipd_port);
+
+               /* Prepare for next level */
+               level = __cvmx_pko3_sq_lvl_next(level);
+               parent_q = child_q;
+
+               /* Terminate loop on DQ level, it has special handling */
+       } while (level != CVMX_PKO_DESCR_QUEUES &&
+                level != CVMX_PKO_LEVEL_INVAL);
+
+       if (level != CVMX_PKO_DESCR_QUEUES) {
+               printf("ERROR: %s: level sequence error\n", __func__);
+               return -1;
+       }
+
+       /* Allocate descriptor queues for the port */
+       dq = cvmx_pko_alloc_queues(node, level, res_owner, -1, num_queues);
+       if (dq < 0) {
+               printf("ERROR: %s: could not reserve DQs\n", __func__);
+               return -1;
+       }
+
+       res = cvmx_pko3_sq_config_children(node, level, parent_q, dq,
+                                          num_queues, static_pri);
+       if (res < 0) {
+               printf("ERROR: %s: configuring %s\n", __func__,
+                      __cvmx_pko3_sq_str(b1, level, dq));
+               return -1;
+       }
+
+       /* register DQ/IPD translation */
+       __cvmx_pko3_ipd_dq_register(xiface, subif, dq, num_queues);
+
+       if (debug)
+               debug("%s: xiface %u:%u/%u qs %u-%u\n", __func__, xi.node,
+                     xi.interface, subif, dq, dq + num_queues - 1);
+       return 0;
+}
+
+/** Initialize the NULL interface
+ *
+ * A NULL interface is a special case in that it is not
+ * one of the enumerated interfaces in the system, and does
+ * not apply to input either. Still, it can be very handy
+ * for dealing with packets that should be discarded in
+ * a generic, streamlined way.
+ *
+ * The Descriptor Queue 0 will be reserved for the NULL interface
+ * and the normalized (i.e. IPD) port number has the all-ones value.
+ */
+static int __cvmx_pko3_config_null_interface(unsigned int node)
+{
+       int l1_q_num;
+       int parent_q, child_q;
+       enum cvmx_pko3_level_e level;
+       int i, res, res_owner;
+       int xiface, ipd_port;
+       int num_dq = 1;   /* # of DQs for NULL */
+       const int dq = 0; /* Reserve DQ#0 for NULL */
+       char pko_mac_num;
+       char b1[12];
+
+       if (OCTEON_IS_MODEL(OCTEON_CN78XX))
+               pko_mac_num = 0x1C; /* MAC# 28 virtual MAC for NULL */
+       else if (OCTEON_IS_MODEL(OCTEON_CN73XX))
+               pko_mac_num = 0x0F; /* MAC# 16 virtual MAC for NULL */
+       else if (OCTEON_IS_MODEL(OCTEON_CNF75XX))
+               pko_mac_num = 0x0A; /* MAC# 10 virtual MAC for NULL */
+       else
+               return -1;
+
+       if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_0))
+               num_dq = 8;
+
+       if (debug)
+               debug("%s: null iface dq=%u-%u\n", __func__, dq,
+                     dq + num_dq - 1);
+
+       ipd_port = cvmx_helper_node_to_ipd_port(node, CVMX_PKO3_IPD_PORT_NULL);
+
+       /* Build an identifiable owner identifier by MAC# for easy release */
+       res_owner = __cvmx_helper_pko3_res_owner(ipd_port);
+       if (res_owner < 0) {
+               debug("%s: ERROR Invalid interface\n", __func__);
+               return -1;
+       }
+
+       level = CVMX_PKO_PORT_QUEUES;
+
+       /* Allocate a port queue */
+       l1_q_num = cvmx_pko_alloc_queues(node, level, res_owner, -1, 1);
+
+       if (l1_q_num < 0) {
+               debug("%s: ERROR reserving L1 SQ\n", __func__);
+               return -1;
+       }
+
+       res = cvmx_pko3_pq_config(node, pko_mac_num, l1_q_num);
+       if (res < 0) {
+               printf("ERROR: %s: PQ/L1 queue configuration\n", __func__);
+               return -1;
+       }
+
+       parent_q = l1_q_num;
+
+       /* Determine the next queue level */
+       level = __cvmx_pko3_sq_lvl_next(level);
+
+       /* Simply chain queues 1-to-1 from L2 to one before DQ level */
+       do {
+               /* allocate next level queue */
+               child_q = cvmx_pko_alloc_queues(node, level, res_owner, -1, 1);
+
+               if (child_q < 0) {
+                       printf("ERROR: %s: allocating %s\n", __func__,
+                              __cvmx_pko3_sq_str(b1, level, child_q));
+                       return -1;
+               }
+
+               /* Configre newly allocated queue */
+               res = cvmx_pko3_sq_config_children(node, level, parent_q,
+                                                  child_q, 1, 1);
+
+               if (res < 0) {
+                       printf("ERROR: %s: configuring %s\n", __func__,
+                              __cvmx_pko3_sq_str(b1, level, child_q));
+                       return -1;
+               }
+
+               /* Prepare for next level */
+               level = __cvmx_pko3_sq_lvl_next(level);
+               parent_q = child_q;
+
+               /* Terminate loop on DQ level, it has special handling */
+       } while (level != CVMX_PKO_DESCR_QUEUES &&
+                level != CVMX_PKO_LEVEL_INVAL);
+
+       if (level != CVMX_PKO_DESCR_QUEUES) {
+               printf("ERROR: %s: level sequence error\n", __func__);
+               return -1;
+       }
+
+       /* Reserve 'num_dq' DQ's at 0 by convention */
+       res = cvmx_pko_alloc_queues(node, level, res_owner, dq, num_dq);
+       if (dq != res) {
+               debug("%s: ERROR: could not reserve DQs\n", __func__);
+               return -1;
+       }
+
+       res = cvmx_pko3_sq_config_children(node, level, parent_q, dq, num_dq,
+                                          num_dq);
+       if (res < 0) {
+               printf("ERROR: %s: configuring %s\n", __func__,
+                      __cvmx_pko3_sq_str(b1, level, dq));
+               return -1;
+       }
+
+       /* NULL interface does not need to map to a CHAN_E */
+
+       /* register DQ/IPD translation */
+       xiface = cvmx_helper_node_interface_to_xiface(node, __CVMX_XIFACE_NULL);
+       __cvmx_pko3_ipd_dq_register(xiface, 0, dq, num_dq);
+
+       /* open the null DQs here */
+       for (i = 0; i < num_dq; i++) {
+               unsigned int limit = 128; /* NULL never really uses much */
+
+               cvmx_pko_dq_open(node, dq + i);
+               cvmx_pko3_dq_set_limit(node, dq + i, limit);
+       }
+
+       return 0;
+}
+
+/** Open all descriptor queues belonging to an interface/port
+ * @INTERNAL
+ */
+int __cvmx_pko3_helper_dqs_activate(int xiface, int index, bool min_pad)
+{
+       int ipd_port, dq_base, dq_count, i;
+       struct cvmx_xiface xi = cvmx_helper_xiface_to_node_interface(xiface);
+       unsigned int limit;
+
+       /* Get local IPD port for the interface */
+       ipd_port = cvmx_helper_get_ipd_port(xiface, index);
+       if (ipd_port < 0) {
+               printf("ERROR: %s: No IPD port for interface %d port %d\n",
+                      __func__, xiface, index);
+               return -1;
+       }
+
+       /* Get DQ# range for the IPD port */
+       dq_base = cvmx_pko3_get_queue_base(ipd_port);
+       dq_count = cvmx_pko3_get_queue_num(ipd_port);
+       if (dq_base < 0 || dq_count <= 0) {
+               printf("ERROR: %s: No descriptor queues for interface %d port %d\n",
+                      __func__, xiface, index);
+               return -1;
+       }
+
+       /* Mask out node from global DQ# */
+       dq_base &= (1 << 10) - 1;
+
+       limit = __pko_pkt_quota / dq_count /
+               cvmx_helper_interface_enumerate(xiface);
+
+       for (i = 0; i < dq_count; i++) {
+               /* FIXME: 2ms at 1Gbps max packet rate, make speed dependent */
+               cvmx_pko_dq_open(xi.node, dq_base + i);
+               cvmx_pko3_dq_options(xi.node, dq_base + i, min_pad);
+
+               if (debug)
+                       debug("%s: DQ%u limit %d\n", __func__, dq_base + i,
+                             limit);
+
+               cvmx_pko3_dq_set_limit(xi.node, dq_base + i, limit);
+               __pko_pkt_budget -= limit;
+       }
+
+       if (__pko_pkt_budget < 0)
+               printf("WARNING: %s: PKO buffer deficit %d\n", __func__,
+                      __pko_pkt_budget);
+       else if (debug)
+               debug("%s: PKO remaining packet budget: %d\n", __func__,
+                     __pko_pkt_budget);
+
+       return i;
+}
+
+/** Configure and initialize PKO3 for an interface
+ *
+ * @param xiface is the interface number to configure
+ * @return 0 on success.
+ */
+int cvmx_helper_pko3_init_interface(int xiface)
+{
+       cvmx_helper_interface_mode_t mode;
+       int node, iface, subif, num_ports;
+       bool fcs_enable, pad_enable, pad_enable_pko;
+       u8 fcs_sof_off = 0;
+       u8 num_queues = 1;
+       bool qos = false, pfc = false;
+       int res = -1;
+       cvmx_xiface_t xi = cvmx_helper_xiface_to_node_interface(xiface);
+
+       node = xi.node;
+       iface = xi.interface;
+       mode = cvmx_helper_interface_get_mode(xiface);
+       num_ports = cvmx_helper_interface_enumerate(xiface);
+       subif = 0;
+
+       if ((unsigned int)iface <
+           NUM_ELEMENTS(__cvmx_pko_queue_static_config[node].pknd.pko_cfg_iface)) {
+               pfc = __cvmx_pko_queue_static_config[node]
+                             .pknd.pko_cfg_iface[iface]
+                             .pfc_enable;
+               num_queues = __cvmx_pko_queue_static_config[node]
+                                    .pknd.pko_cfg_iface[iface]
+                                    .queues_per_port;
+               qos = __cvmx_pko_queue_static_config[node]
+                             .pknd.pko_cfg_iface[iface]
+                             .qos_enable;
+       }
+
+       /* Force 8 DQs per port for pass 1.0 to circumvent limitations */
+       if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_0))
+               num_queues = 8;
+
+       /* For ILK there is one IPD port per channel */
+       if (mode == CVMX_HELPER_INTERFACE_MODE_ILK)
+               num_ports = __cvmx_helper_ilk_enumerate(xiface);
+
+       /* Skip non-existent interfaces */
+       if (num_ports < 1) {
+               debug("ERROR: %s: invalid iface %u:%u\n", __func__, node,
+                     iface);
+               return -1;
+       }
+
+       if (mode == CVMX_HELPER_INTERFACE_MODE_LOOP) {
+               num_queues = __cvmx_pko_queue_static_config[node]
+                                    .pknd.pko_cfg_loop.queues_per_port;
+               qos = __cvmx_pko_queue_static_config[node]
+                             .pknd.pko_cfg_loop.qos_enable;
+
+               if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_0))
+                       num_queues = 8;
+
+               res = __cvmx_pko3_config_chan_interface(xiface, num_ports,
+                                                       num_queues, qos);
+               if (res < 0)
+                       goto __cfg_error;
+       } else if (mode == CVMX_HELPER_INTERFACE_MODE_NPI) {
+               num_queues = __cvmx_pko_queue_static_config[node]
+                                    .pknd.pko_cfg_npi.queues_per_port;
+               qos = __cvmx_pko_queue_static_config[node]
+                             .pknd.pko_cfg_npi.qos_enable;
+
+               if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_0))
+                       num_queues = 8;
+
+               res = __cvmx_pko3_config_chan_interface(xiface, num_ports,
+                                                       num_queues, qos);
+               if (res < 0)
+                       goto __cfg_error;
+       }
+       /* ILK-specific queue configuration */
+       else if (mode == CVMX_HELPER_INTERFACE_MODE_ILK) {
+               unsigned int num_chans = __cvmx_helper_ilk_enumerate(xiface);
+
+               num_queues = 8;
+               qos = true;
+               pfc = false;
+
+               if (num_chans >= 128)
+                       num_queues = 1;
+               else if (num_chans >= 64)
+                       num_queues = 2;
+               else if (num_chans >= 32)
+                       num_queues = 4;
+               else
+                       num_queues = 8;
+
+               res = __cvmx_pko3_config_chan_interface(xiface, num_chans,
+                                                       num_queues, qos);
+       }
+       /* Setup all ethernet configured for PFC */
+       else if (pfc) {
+               /* PFC interfaces have 8 prioritized queues */
+               for (subif = 0; subif < num_ports; subif++) {
+                       res = __cvmx_pko3_config_pfc_interface(xiface, subif);
+                       if (res < 0)
+                               goto __cfg_error;
+
+                       /* Enable PFC/CBFC on BGX */
+                       __cvmx_helper_bgx_xaui_config_pfc(node, iface, subif,
+                                                         true);
+               }
+       } else {
+               /* All other interfaces follow static configuration */
+               for (subif = 0; subif < num_ports; subif++) {
+                       res = __cvmx_pko3_config_gen_interface(xiface, subif,
+                                                              num_queues, qos);
+                       if (res < 0)
+                               goto __cfg_error;
+               }
+       }
+
+       fcs_enable = __cvmx_helper_get_has_fcs(xiface);
+       pad_enable = __cvmx_helper_get_pko_padding(xiface);
+
+       /* Do not use PKO PAD/FCS generation on o78p1.x on BGX interfaces */
+       if (OCTEON_IS_MODEL(OCTEON_CN78XX_PASS1_X))
+               pad_enable_pko = false;
+       else
+               pad_enable_pko = pad_enable;
+
+       if (debug)
+               debug("%s: iface %u:%u FCS=%d pad=%d pko=%d\n", __func__, node,
+                     iface, fcs_enable, pad_enable, pad_enable_pko);
+
+       /* Setup interface options */
+       for (subif = 0; subif < num_ports; subif++) {
+               /* Open interface/port DQs to allow transmission to begin */
+               res = __cvmx_pko3_helper_dqs_activate(xiface, subif,
+                                                     pad_enable_pko);
+
+               if (res < 0)
+                       goto __cfg_error;
+
+               /* ILK has only one MAC, subif == logical-channel */
+               if (mode == CVMX_HELPER_INTERFACE_MODE_ILK && subif > 0)
+                       continue;
+
+               /* LOOP has only one MAC, subif == logical-channel */
+               if (mode == CVMX_HELPER_INTERFACE_MODE_LOOP && subif > 0)
+                       continue;
+
+               /* NPI has only one MAC, subif == 'ring' */
+               if (mode == CVMX_HELPER_INTERFACE_MODE_NPI && subif > 0)
+                       continue;
+
+               /* for sRIO there is 16 byte sRIO header, outside of FCS */
+               if (mode == CVMX_HELPER_INTERFACE_MODE_SRIO)
+                       fcs_sof_off = 16;
+
+               if (iface >= CVMX_HELPER_MAX_GMX) {
+                       /* Non-BGX interface, use PKO for FCS/PAD */
+                       res = cvmx_pko3_interface_options(xiface, subif,
+                                                         fcs_enable,
+                                                         pad_enable_pko,
+                                                         fcs_sof_off);
+               } else if (pad_enable == pad_enable_pko) {
+                       /* BGX interface: FCS/PAD done by PKO */
+                       res = cvmx_pko3_interface_options(xiface, subif,
+                                                         fcs_enable,
+                                                         pad_enable,
+                                                         fcs_sof_off);
+                       cvmx_helper_bgx_tx_options(node, iface, subif, false,
+                                                  false);
+               } else {
+                       /* BGX interface: FCS/PAD done by BGX */
+                       res = cvmx_pko3_interface_options(xiface, subif, false,
+                                                         false, fcs_sof_off);
+                       cvmx_helper_bgx_tx_options(node, iface, subif,
+                                                  fcs_enable, pad_enable);
+               }
+
+               if (res < 0)
+                       debug("WARNING: %s: option set failed on iface %u:%u/%u\n",
+                             __func__, node, iface, subif);
+               if (debug)
+                       debug("%s: face %u:%u/%u fifo size %d\n", __func__,
+                             node, iface, subif,
+                             cvmx_pko3_port_fifo_size(xiface, subif));
+       }
+       return 0;
+
+__cfg_error:
+       debug("ERROR: %s: failed on iface %u:%u/%u\n", __func__, node, iface,
+             subif);
+       return -1;
+}
+
+/**
+ * Global initialization for PKO3
+ *
+ * Should only be called once on each node
+ *
+ * TBD: Resolve the kernel case.
+ * When Linux eats up the entire memory, bootmem will be unable to
+ * satisfy our request, and the memory needs to come from Linux free pages.
+ */
+int __cvmx_helper_pko3_init_global(unsigned int node, uint16_t gaura)
+{
+       int res;
+
+       res = cvmx_pko3_hw_init_global(node, gaura);
+       if (res < 0) {
+               debug("ERROR: %s:failed block initialization\n", __func__);
+               return res;
+       }
+
+       /* configure channel level */
+       cvmx_pko3_channel_credit_level(node, cvmx_pko_default_channel_level);
+
+       /* add NULL MAC/DQ setup */
+       res = __cvmx_pko3_config_null_interface(node);
+       if (res < 0)
+               debug("ERROR: %s: creating NULL interface\n", __func__);
+
+       return res;
+}
+
+/**
+ * Global initialization for PKO3
+ *
+ * Should only be called once on each node
+ *
+ * When Linux eats up the entire memory, bootmem will be unable to
+ * satisfy our request, and the memory needs to come from Linux free pages.
+ */
+int cvmx_helper_pko3_init_global(unsigned int node)
+{
+       void *ptr;
+       int res = -1;
+       unsigned int aura_num = ~0;
+       cvmx_fpa3_gaura_t aura;
+
+       /* Allocate memory required by PKO3 */
+       res = __cvmx_pko3_config_memory(node);
+       if (res < 0) {
+               debug("ERROR: %s: PKO3 memory allocation error\n", __func__);
+               return res;
+       }
+
+       aura_num = res;
+       aura = __cvmx_pko3_aura[node];
+
+       /* Exercise the FPA to make sure the AURA is functional */
+       ptr = cvmx_fpa3_alloc(aura);
+
+       if (!ptr) {
+               res = -1;
+       } else {
+               cvmx_fpa3_free_nosync(ptr, aura, 0);
+               res = 0;
+       }
+
+       if (res < 0) {
+               debug("ERROR: %s: FPA failure AURA=%u:%d\n", __func__,
+                     aura.node, aura.laura);
+               return -1;
+       }
+
+       res = __cvmx_helper_pko3_init_global(node, aura_num);
+
+       if (res < 0)
+               debug("ERROR: %s: failed to start PPKO\n", __func__);
+
+       return res;
+}