// SPDX-License-Identifier: GPL-2.0+ /* * Copyright 2014-2016 Freescale Semiconductor, Inc. * Copyright 2017, 2023 NXP */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ldpaa_eth.h" #ifdef CONFIG_PHYLIB static void init_phy(struct udevice *dev) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); priv->phy = dm_eth_phy_connect(dev); if (!priv->phy) return; phy_config(priv->phy); } #endif static void ldpaa_eth_collect_dpni_stats(struct udevice *dev, u64 *data) { union dpni_statistics dpni_stats; int dpni_stats_page_size[DPNI_STATISTICS_CNT] = { sizeof(dpni_stats.page_0), sizeof(dpni_stats.page_1), sizeof(dpni_stats.page_2), sizeof(dpni_stats.page_3), sizeof(dpni_stats.page_4), sizeof(dpni_stats.page_5), sizeof(dpni_stats.page_6), }; int j, k, num_cnt, err, i = 0; for (j = 0; j <= 6; j++) { /* We're not interested in pages 4 & 5 for now */ if (j == 4 || j == 5) continue; err = dpni_get_statistics(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, j, 0, &dpni_stats); if (err) { memset(&dpni_stats, 0, sizeof(dpni_stats)); printf("dpni_get_stats(%d) failed\n", j); } num_cnt = dpni_stats_page_size[j] / sizeof(u64); for (k = 0; k < num_cnt; k++) *(data + i++) = dpni_stats.raw.counter[k]; } } static void ldpaa_eth_add_dpni_stats(struct udevice *dev, u64 *data) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); int i; for (i = 0; i < LDPAA_ETH_DPNI_NUM_STATS; i++) priv->dpni_stats[i] += data[i]; } static void ldpaa_eth_collect_dpmac_stats(struct udevice *dev, u64 *data) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); int err, i; u64 value; for (i = 0; i < LDPAA_ETH_DPMAC_NUM_STATS; i++) { err = dpmac_get_counter(dflt_mc_io, MC_CMD_NO_FLAGS, priv->dpmac_handle, i, &value); if (err) printf("dpmac_get_counter(%d) failed\n", i); *(data + i) = value; } } static void ldpaa_eth_add_dpmac_stats(struct udevice *dev, u64 *data) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); int i; for (i = 0; i < LDPAA_ETH_DPMAC_NUM_STATS; i++) priv->dpmac_stats[i] += data[i]; } #ifdef DEBUG static void ldpaa_eth_dump_dpni_stats(struct udevice *dev, u64 *data) { int i; printf("DPNI counters:\n"); for (i = 0; i < LDPAA_ETH_DPNI_NUM_STATS; i++) printf(" %s: %llu\n", ldpaa_eth_dpni_stat_strings[i], data[i]); } static void ldpaa_eth_dump_dpmac_stats(struct udevice *dev, u64 *data) { int i; printf("DPMAC counters:\n"); for (i = 0; i < LDPAA_ETH_DPMAC_NUM_STATS; i++) printf(" %s: %llu\n", ldpaa_eth_dpmac_stat_strings[i], data[i]); } #endif static void ldpaa_eth_rx(struct ldpaa_eth_priv *priv, const struct dpaa_fd *fd) { u64 fd_addr; uint16_t fd_offset; uint32_t fd_length; struct ldpaa_fas *fas; uint32_t status, err; u32 timeo = (CONFIG_SYS_HZ * 2) / 1000; u32 time_start; struct qbman_release_desc releasedesc; struct qbman_swp *swp = dflt_dpio->sw_portal; fd_addr = ldpaa_fd_get_addr(fd); fd_offset = ldpaa_fd_get_offset(fd); fd_length = ldpaa_fd_get_len(fd); debug("Rx frame:data addr=0x%p size=0x%x\n", (u64 *)fd_addr, fd_length); if (fd->simple.frc & LDPAA_FD_FRC_FASV) { /* Read the frame annotation status word and check for errors */ fas = (struct ldpaa_fas *) ((uint8_t *)(fd_addr) + dflt_dpni->buf_layout.private_data_size); status = le32_to_cpu(fas->status); if (status & LDPAA_ETH_RX_ERR_MASK) { printf("Rx frame error(s): 0x%08x\n", status & LDPAA_ETH_RX_ERR_MASK); goto error; } else if (status & LDPAA_ETH_RX_UNSUPP_MASK) { printf("Unsupported feature in bitmask: 0x%08x\n", status & LDPAA_ETH_RX_UNSUPP_MASK); goto error; } } debug("Rx frame: To Upper layer\n"); net_process_received_packet((uint8_t *)(fd_addr) + fd_offset, fd_length); error: flush_dcache_range(fd_addr, fd_addr + LDPAA_ETH_RX_BUFFER_SIZE); qbman_release_desc_clear(&releasedesc); qbman_release_desc_set_bpid(&releasedesc, dflt_dpbp->dpbp_attr.bpid); time_start = get_timer(0); do { /* Release buffer into the QBMAN */ err = qbman_swp_release(swp, &releasedesc, &fd_addr, 1); } while (get_timer(time_start) < timeo && err == -EBUSY); if (err == -EBUSY) printf("Rx frame: QBMAN buffer release fails\n"); return; } static int ldpaa_eth_pull_dequeue_rx(struct udevice *dev, int flags, uchar **packetp) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); const struct ldpaa_dq *dq; const struct dpaa_fd *fd; int i = 5, err = 0, status; u32 timeo = (CONFIG_SYS_HZ * 2) / 1000; u32 time_start; static struct qbman_pull_desc pulldesc; struct qbman_swp *swp = dflt_dpio->sw_portal; while (--i) { qbman_pull_desc_clear(&pulldesc); qbman_pull_desc_set_numframes(&pulldesc, 1); qbman_pull_desc_set_fq(&pulldesc, priv->rx_dflt_fqid); err = qbman_swp_pull(swp, &pulldesc); if (err < 0) { printf("Dequeue frames error:0x%08x\n", err); continue; } time_start = get_timer(0); do { dq = qbman_swp_dqrr_next(swp); } while (get_timer(time_start) < timeo && !dq); if (dq) { /* Check for valid frame. If not sent a consume * confirmation to QBMAN otherwise give it to NADK * application and then send consume confirmation to * QBMAN. */ status = (uint8_t)ldpaa_dq_flags(dq); if ((status & LDPAA_DQ_STAT_VALIDFRAME) == 0) { debug("Dequeue RX frames:"); debug("No frame delivered\n"); qbman_swp_dqrr_consume(swp, dq); continue; } fd = ldpaa_dq_fd(dq); /* Obtain FD and process it */ ldpaa_eth_rx(priv, fd); qbman_swp_dqrr_consume(swp, dq); break; } else { err = -ENODATA; debug("No DQRR entries\n"); break; } } return err; } static int ldpaa_eth_tx(struct udevice *dev, void *buf, int len) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); struct dpaa_fd fd; u64 buffer_start; int data_offset, err; u32 timeo = (CONFIG_SYS_HZ * 10) / 1000; u32 time_start; struct qbman_swp *swp = dflt_dpio->sw_portal; struct qbman_eq_desc ed; struct qbman_release_desc releasedesc; /* Setup the FD fields */ memset(&fd, 0, sizeof(fd)); data_offset = priv->tx_data_offset; do { err = qbman_swp_acquire(dflt_dpio->sw_portal, dflt_dpbp->dpbp_attr.bpid, &buffer_start, 1); } while (err == -EBUSY); if (err <= 0) { printf("qbman_swp_acquire() failed\n"); return -ENOMEM; } debug("TX data: malloc buffer start=0x%p\n", (u64 *)buffer_start); memcpy(((uint8_t *)(buffer_start) + data_offset), buf, len); flush_dcache_range(buffer_start, buffer_start + LDPAA_ETH_RX_BUFFER_SIZE); ldpaa_fd_set_addr(&fd, (u64)buffer_start); ldpaa_fd_set_offset(&fd, (uint16_t)(data_offset)); ldpaa_fd_set_bpid(&fd, dflt_dpbp->dpbp_attr.bpid); ldpaa_fd_set_len(&fd, len); fd.simple.ctrl = LDPAA_FD_CTRL_ASAL | LDPAA_FD_CTRL_PTA | LDPAA_FD_CTRL_PTV1; qbman_eq_desc_clear(&ed); qbman_eq_desc_set_no_orp(&ed, 0); qbman_eq_desc_set_qd(&ed, priv->tx_qdid, priv->tx_flow_id, 0); time_start = get_timer(0); while (get_timer(time_start) < timeo) { err = qbman_swp_enqueue(swp, &ed, (const struct qbman_fd *)(&fd)); if (err != -EBUSY) break; } if (err < 0) { printf("error enqueueing Tx frame\n"); goto error; } return err; error: qbman_release_desc_clear(&releasedesc); qbman_release_desc_set_bpid(&releasedesc, dflt_dpbp->dpbp_attr.bpid); time_start = get_timer(0); do { /* Release buffer into the QBMAN */ err = qbman_swp_release(swp, &releasedesc, &buffer_start, 1); } while (get_timer(time_start) < timeo && err == -EBUSY); if (err == -EBUSY) printf("TX data: QBMAN buffer release fails\n"); return err; } static int ldpaa_get_dpmac_state(struct ldpaa_eth_priv *priv, struct dpmac_link_state *state) { phy_interface_t enet_if; struct phy_device *phydev = NULL; int err; /* let's start off with maximum capabilities */ enet_if = wriop_get_enet_if(priv->dpmac_id); switch (enet_if) { case PHY_INTERFACE_MODE_XGMII: state->rate = SPEED_10000; break; default: state->rate = SPEED_1000; break; } state->up = 1; state->options |= DPMAC_LINK_OPT_AUTONEG; phydev = priv->phy; if (phydev) { err = phy_startup(phydev); if (err) { printf("%s: Could not initialize\n", phydev->dev->name); state->up = 0; } else if (phydev->link) { state->rate = min(state->rate, (uint32_t)phydev->speed); if (!phydev->duplex) state->options |= DPMAC_LINK_OPT_HALF_DUPLEX; if (!phydev->autoneg) state->options &= ~DPMAC_LINK_OPT_AUTONEG; } else { state->up = 0; } } if (!phydev) state->options &= ~DPMAC_LINK_OPT_AUTONEG; if (!state->up) { state->rate = 0; state->options = 0; return -ENOLINK; } return 0; } static int ldpaa_eth_open(struct udevice *dev) { struct eth_pdata *plat = dev_get_plat(dev); struct ldpaa_eth_priv *priv = dev_get_priv(dev); struct dpmac_link_state dpmac_link_state = { 0 }; #ifdef DEBUG struct dpni_link_state link_state; #endif int err = 0; struct dpni_queue d_queue_cfg = { 0 }; struct dpni_queue_id d_queue; if (eth_is_active(dev)) return 0; if (get_mc_boot_status() != 0) { printf("ERROR (MC is not booted)\n"); return -ENODEV; } if (get_dpl_apply_status() == 0) { printf("ERROR (DPL is deployed. No device available)\n"); return -ENODEV; } /* DPMAC initialization */ err = ldpaa_dpmac_setup(priv); if (err < 0) goto err_dpmac_setup; err = ldpaa_get_dpmac_state(priv, &dpmac_link_state); if (err < 0) goto err_dpmac_bind; /* DPMAC binding DPNI */ err = ldpaa_dpmac_bind(priv); if (err) goto err_dpmac_bind; /* DPNI initialization */ err = ldpaa_dpni_setup(priv); if (err < 0) goto err_dpni_setup; err = ldpaa_dpbp_setup(); if (err < 0) goto err_dpbp_setup; /* DPNI binding DPBP */ err = ldpaa_dpni_bind(priv); if (err) goto err_dpni_bind; err = dpni_add_mac_addr(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, plat->enetaddr, 0, 0, 0); if (err) { printf("dpni_add_mac_addr() failed\n"); return err; } err = dpni_enable(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle); if (err < 0) { printf("dpni_enable() failed\n"); return err; } err = dpmac_set_link_state(dflt_mc_io, MC_CMD_NO_FLAGS, priv->dpmac_handle, &dpmac_link_state); if (err < 0) { printf("dpmac_set_link_state() failed\n"); return err; } #ifdef DEBUG printf("DPMAC link status: %d - ", dpmac_link_state.up); dpmac_link_state.up == 0 ? printf("down\n") : dpmac_link_state.up == 1 ? printf("up\n") : printf("error state\n"); err = dpni_get_link_state(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, &link_state); if (err < 0) { printf("dpni_get_link_state() failed\n"); return err; } printf("DPNI link status: %d - ", link_state.up); link_state.up == 0 ? printf("down\n") : link_state.up == 1 ? printf("up\n") : printf("error state\n"); #endif memset(&d_queue, 0, sizeof(struct dpni_queue)); err = dpni_get_queue(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, DPNI_QUEUE_RX, 0, 0, &d_queue_cfg, &d_queue); if (err) { printf("dpni_get_queue failed\n"); goto err_get_queue; } priv->rx_dflt_fqid = d_queue.fqid; err = dpni_get_qdid(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, DPNI_QUEUE_TX, &priv->tx_qdid); if (err) { printf("dpni_get_qdid() failed\n"); goto err_qdid; } return dpmac_link_state.up; err_qdid: err_get_queue: dpni_disable(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle); err_dpni_bind: ldpaa_dpbp_free(); err_dpbp_setup: dpni_close(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle); err_dpni_setup: err_dpmac_bind: dpmac_close(dflt_mc_io, MC_CMD_NO_FLAGS, priv->dpmac_handle); dpmac_destroy(dflt_mc_io, dflt_dprc_handle, MC_CMD_NO_FLAGS, priv->dpmac_id); err_dpmac_setup: return err; } static void ldpaa_eth_stop(struct udevice *dev) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); struct phy_device *phydev = NULL; int err = 0; u64 *data; if (!eth_is_active(dev)) return; data = kzalloc(sizeof(u64) * LDPAA_ETH_DPNI_NUM_STATS, GFP_KERNEL); if (data) { ldpaa_eth_collect_dpni_stats(dev, data); ldpaa_eth_add_dpni_stats(dev, data); #ifdef DEBUG ldpaa_eth_dump_dpni_stats(dev, data); #endif } kfree(data); data = kzalloc(sizeof(u64) * LDPAA_ETH_DPMAC_NUM_STATS, GFP_KERNEL); if (data) { ldpaa_eth_collect_dpmac_stats(dev, data); ldpaa_eth_add_dpmac_stats(dev, data); #ifdef DEBUG ldpaa_eth_dump_dpmac_stats(dev, data); #endif } kfree(data); err = dprc_disconnect(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dprc_handle, &dpmac_endpoint); if (err < 0) printf("dprc_disconnect() failed dpmac_endpoint\n"); err = dpmac_close(dflt_mc_io, MC_CMD_NO_FLAGS, priv->dpmac_handle); if (err < 0) printf("dpmac_close() failed\n"); err = dpmac_destroy(dflt_mc_io, dflt_dprc_handle, MC_CMD_NO_FLAGS, priv->dpmac_id); if (err < 0) printf("dpmac_destroy() failed\n"); /* Stop Tx and Rx traffic */ err = dpni_disable(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle); if (err < 0) printf("dpni_disable() failed\n"); phydev = priv->phy; if (phydev) phy_shutdown(phydev); /* Free DPBP handle and reset. */ ldpaa_dpbp_free(); dpni_reset(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle); if (err < 0) printf("dpni_reset() failed\n"); dpni_close(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle); if (err < 0) printf("dpni_close() failed\n"); } static void ldpaa_dpbp_drain_cnt(int count) { uint64_t buf_array[7]; void *addr; int ret, i; BUG_ON(count > 7); do { ret = qbman_swp_acquire(dflt_dpio->sw_portal, dflt_dpbp->dpbp_attr.bpid, buf_array, count); if (ret < 0) { printf("qbman_swp_acquire() failed\n"); return; } for (i = 0; i < ret; i++) { addr = (void *)buf_array[i]; debug("Free: buffer addr =0x%p\n", addr); free(addr); } } while (ret); } static void ldpaa_dpbp_drain(void) { int i; for (i = 0; i < LDPAA_ETH_NUM_BUFS; i += 7) ldpaa_dpbp_drain_cnt(7); } static int ldpaa_bp_add_7(uint16_t bpid) { uint64_t buf_array[7]; u8 *addr; int i; struct qbman_release_desc rd; for (i = 0; i < 7; i++) { addr = memalign(LDPAA_ETH_BUF_ALIGN, LDPAA_ETH_RX_BUFFER_SIZE); if (!addr) { printf("addr allocation failed\n"); goto err_alloc; } memset(addr, 0x00, LDPAA_ETH_RX_BUFFER_SIZE); flush_dcache_range((u64)addr, (u64)(addr + LDPAA_ETH_RX_BUFFER_SIZE)); buf_array[i] = (uint64_t)addr; debug("Release: buffer addr =0x%p\n", addr); } release_bufs: /* In case the portal is busy, retry until successful. * This function is guaranteed to succeed in a reasonable amount * of time. */ do { mdelay(1); qbman_release_desc_clear(&rd); qbman_release_desc_set_bpid(&rd, bpid); } while (qbman_swp_release(dflt_dpio->sw_portal, &rd, buf_array, i)); return i; err_alloc: if (i) goto release_bufs; return 0; } static int ldpaa_dpbp_seed(uint16_t bpid) { int i; int count; for (i = 0; i < LDPAA_ETH_NUM_BUFS; i += 7) { count = ldpaa_bp_add_7(bpid); if (count < 7) printf("Buffer Seed= %d\n", count); } return 0; } static int ldpaa_dpbp_setup(void) { int err; err = dpbp_open(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_attr.id, &dflt_dpbp->dpbp_handle); if (err) { printf("dpbp_open() failed\n"); goto err_open; } err = dpbp_enable(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_handle); if (err) { printf("dpbp_enable() failed\n"); goto err_enable; } err = dpbp_get_attributes(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_handle, &dflt_dpbp->dpbp_attr); if (err) { printf("dpbp_get_attributes() failed\n"); goto err_get_attr; } err = ldpaa_dpbp_seed(dflt_dpbp->dpbp_attr.bpid); if (err) { printf("Buffer seeding failed for DPBP %d (bpid=%d)\n", dflt_dpbp->dpbp_attr.id, dflt_dpbp->dpbp_attr.bpid); goto err_seed; } return 0; err_seed: err_get_attr: dpbp_disable(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_handle); err_enable: dpbp_close(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_handle); err_open: return err; } static void ldpaa_dpbp_free(void) { ldpaa_dpbp_drain(); dpbp_disable(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_handle); dpbp_reset(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_handle); dpbp_close(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpbp->dpbp_handle); } static int ldpaa_dpmac_version_check(struct fsl_mc_io *mc_io, struct ldpaa_eth_priv *priv) { int error; uint16_t major_ver, minor_ver; error = dpmac_get_api_version(dflt_mc_io, 0, &major_ver, &minor_ver); if ((major_ver < DPMAC_VER_MAJOR) || (major_ver == DPMAC_VER_MAJOR && minor_ver < DPMAC_VER_MINOR)) { printf("DPMAC version mismatch found %u.%u,", major_ver, minor_ver); printf("supported version is %u.%u\n", DPMAC_VER_MAJOR, DPMAC_VER_MINOR); return error; } return error; } static int ldpaa_dpmac_setup(struct ldpaa_eth_priv *priv) { int err = 0; struct dpmac_cfg dpmac_cfg; dpmac_cfg.mac_id = priv->dpmac_id; err = dpmac_create(dflt_mc_io, dflt_dprc_handle, MC_CMD_NO_FLAGS, &dpmac_cfg, &priv->dpmac_id); if (err) printf("dpmac_create() failed\n"); err = ldpaa_dpmac_version_check(dflt_mc_io, priv); if (err < 0) { printf("ldpaa_dpmac_version_check() failed: %d\n", err); goto err_version_check; } err = dpmac_open(dflt_mc_io, MC_CMD_NO_FLAGS, priv->dpmac_id, &priv->dpmac_handle); if (err < 0) { printf("dpmac_open() failed: %d\n", err); goto err_open; } return err; err_open: err_version_check: dpmac_destroy(dflt_mc_io, dflt_dprc_handle, MC_CMD_NO_FLAGS, priv->dpmac_id); return err; } static int ldpaa_dpmac_bind(struct ldpaa_eth_priv *priv) { int err = 0; struct dprc_connection_cfg dprc_connection_cfg = { /* If both rates are zero the connection */ /* will be configured in "best effort" mode. */ .committed_rate = 0, .max_rate = 0 }; #ifdef DEBUG struct dprc_endpoint dbg_endpoint; int state = 0; #endif memset(&dpmac_endpoint, 0, sizeof(struct dprc_endpoint)); strcpy(dpmac_endpoint.type, "dpmac"); dpmac_endpoint.id = priv->dpmac_id; memset(&dpni_endpoint, 0, sizeof(struct dprc_endpoint)); strcpy(dpni_endpoint.type, "dpni"); dpni_endpoint.id = dflt_dpni->dpni_id; err = dprc_connect(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dprc_handle, &dpmac_endpoint, &dpni_endpoint, &dprc_connection_cfg); if (err) printf("dprc_connect() failed\n"); #ifdef DEBUG err = dprc_get_connection(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dprc_handle, &dpni_endpoint, &dbg_endpoint, &state); printf("%s, DPMAC Type= %s\n", __func__, dbg_endpoint.type); printf("%s, DPMAC ID= %d\n", __func__, dbg_endpoint.id); printf("%s, DPMAC State= %d\n", __func__, state); memset(&dbg_endpoint, 0, sizeof(struct dprc_endpoint)); err = dprc_get_connection(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dprc_handle, &dpmac_endpoint, &dbg_endpoint, &state); printf("%s, DPNI Type= %s\n", __func__, dbg_endpoint.type); printf("%s, DPNI ID= %d\n", __func__, dbg_endpoint.id); printf("%s, DPNI State= %d\n", __func__, state); #endif return err; } static int ldpaa_dpni_setup(struct ldpaa_eth_priv *priv) { int err; /* and get a handle for the DPNI this interface is associate with */ err = dpni_open(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_id, &dflt_dpni->dpni_handle); if (err) { printf("dpni_open() failed\n"); goto err_open; } err = dpni_get_attributes(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, &dflt_dpni->dpni_attrs); if (err) { printf("dpni_get_attributes() failed (err=%d)\n", err); goto err_get_attr; } /* Configure our buffers' layout */ dflt_dpni->buf_layout.options = DPNI_BUF_LAYOUT_OPT_PARSER_RESULT | DPNI_BUF_LAYOUT_OPT_FRAME_STATUS | DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE | DPNI_BUF_LAYOUT_OPT_DATA_ALIGN; dflt_dpni->buf_layout.pass_parser_result = true; dflt_dpni->buf_layout.pass_frame_status = true; dflt_dpni->buf_layout.private_data_size = LDPAA_ETH_SWA_SIZE; /* HW erratum mandates data alignment in multiples of 256 */ dflt_dpni->buf_layout.data_align = LDPAA_ETH_BUF_ALIGN; /* ...rx, ... */ err = dpni_set_buffer_layout(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, DPNI_QUEUE_RX, &dflt_dpni->buf_layout); if (err) { printf("dpni_set_buffer_layout() failed"); goto err_buf_layout; } /* ... tx, ... */ /* remove Rx-only options */ dflt_dpni->buf_layout.options &= ~(DPNI_BUF_LAYOUT_OPT_DATA_ALIGN | DPNI_BUF_LAYOUT_OPT_PARSER_RESULT); err = dpni_set_buffer_layout(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, DPNI_QUEUE_TX, &dflt_dpni->buf_layout); if (err) { printf("dpni_set_buffer_layout() failed"); goto err_buf_layout; } /* ... tx-confirm. */ dflt_dpni->buf_layout.options &= ~DPNI_BUF_LAYOUT_OPT_PRIVATE_DATA_SIZE; err = dpni_set_buffer_layout(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, DPNI_QUEUE_TX_CONFIRM, &dflt_dpni->buf_layout); if (err) { printf("dpni_set_buffer_layout() failed"); goto err_buf_layout; } /* Now that we've set our tx buffer layout, retrieve the minimum * required tx data offset. */ err = dpni_get_tx_data_offset(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, &priv->tx_data_offset); if (err) { printf("dpni_get_tx_data_offset() failed\n"); goto err_data_offset; } /* Warn in case TX data offset is not multiple of 64 bytes. */ WARN_ON(priv->tx_data_offset % 64); /* Accomodate SWA space. */ priv->tx_data_offset += LDPAA_ETH_SWA_SIZE; debug("priv->tx_data_offset=%d\n", priv->tx_data_offset); return 0; err_data_offset: err_buf_layout: err_get_attr: dpni_close(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle); err_open: return err; } static int ldpaa_dpni_bind(struct ldpaa_eth_priv *priv) { struct dpni_pools_cfg pools_params; struct dpni_queue tx_queue; int err = 0; memset(&pools_params, 0, sizeof(pools_params)); pools_params.num_dpbp = 1; pools_params.pools[0].dpbp_id = (uint16_t)dflt_dpbp->dpbp_attr.id; pools_params.pools[0].buffer_size = LDPAA_ETH_RX_BUFFER_SIZE; err = dpni_set_pools(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, &pools_params); if (err) { printf("dpni_set_pools() failed\n"); return err; } memset(&tx_queue, 0, sizeof(struct dpni_queue)); err = dpni_set_queue(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, DPNI_QUEUE_TX, 0, 0, 0, &tx_queue); if (err) { printf("dpni_set_queue() failed\n"); return err; } err = dpni_set_tx_confirmation_mode(dflt_mc_io, MC_CMD_NO_FLAGS, dflt_dpni->dpni_handle, 0, DPNI_CONF_DISABLE); if (err) { printf("dpni_set_tx_confirmation_mode() failed\n"); return err; } return 0; } static int ldpaa_eth_probe(struct udevice *dev) { struct ofnode_phandle_args phandle; /* Nothing to do if there is no "phy-handle" in the DTS node */ if (dev_read_phandle_with_args(dev, "phy-handle", NULL, 0, 0, &phandle)) { return 0; } init_phy(dev); return 0; } uint32_t ldpaa_eth_get_dpmac_id(struct udevice *dev) { int port_node = dev_of_offset(dev); return fdtdec_get_uint(gd->fdt_blob, port_node, "reg", -1); } static int ldpaa_eth_bind(struct udevice *dev) { uint32_t dpmac_id; char eth_name[16]; int phy_mode = -1; phy_mode = dev_read_phy_mode(dev); if (phy_mode == PHY_INTERFACE_MODE_NA) { dev_err(dev, "incorrect phy mode\n"); return -EINVAL; } dpmac_id = ldpaa_eth_get_dpmac_id(dev); if (dpmac_id == -1) { dev_err(dev, "missing reg field from the dpmac node\n"); return -EINVAL; } sprintf(eth_name, "DPMAC%d@%s", dpmac_id, phy_string_for_interface(phy_mode)); device_set_name(dev, eth_name); return 0; } static int ldpaa_eth_of_to_plat(struct udevice *dev) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); priv->dpmac_id = ldpaa_eth_get_dpmac_id(dev); priv->phy_mode = dev_read_phy_mode(dev); return 0; } static int ldpaa_eth_get_sset_count(struct udevice *dev) { return LDPAA_ETH_DPNI_NUM_STATS + LDPAA_ETH_DPMAC_NUM_STATS; } static void ldpaa_eth_get_strings(struct udevice *dev, u8 *data) { u8 *p = data; int i; for (i = 0; i < LDPAA_ETH_DPNI_NUM_STATS; i++) { strlcpy(p, ldpaa_eth_dpni_stat_strings[i], ETH_GSTRING_LEN); p += ETH_GSTRING_LEN; } for (i = 0; i < LDPAA_ETH_DPMAC_NUM_STATS; i++) { strlcpy(p, ldpaa_eth_dpmac_stat_strings[i], ETH_GSTRING_LEN); p += ETH_GSTRING_LEN; } } static void ldpaa_eth_get_stats(struct udevice *dev, u64 *data) { struct ldpaa_eth_priv *priv = dev_get_priv(dev); int i, j = 0; for (i = 0; i < LDPAA_ETH_DPNI_NUM_STATS; i++) *(data + j++) = priv->dpni_stats[i]; for (i = 0; i < LDPAA_ETH_DPMAC_NUM_STATS; i++) *(data + j++) = priv->dpmac_stats[i]; } static const struct eth_ops ldpaa_eth_ops = { .start = ldpaa_eth_open, .send = ldpaa_eth_tx, .recv = ldpaa_eth_pull_dequeue_rx, .stop = ldpaa_eth_stop, .get_sset_count = ldpaa_eth_get_sset_count, .get_strings = ldpaa_eth_get_strings, .get_stats = ldpaa_eth_get_stats, }; static const struct udevice_id ldpaa_eth_of_ids[] = { { .compatible = "fsl,qoriq-mc-dpmac" }, }; U_BOOT_DRIVER(ldpaa_eth) = { .name = LDPAA_ETH_DRIVER_NAME, .id = UCLASS_ETH, .of_match = ldpaa_eth_of_ids, .of_to_plat = ldpaa_eth_of_to_plat, .bind = ldpaa_eth_bind, .probe = ldpaa_eth_probe, .ops = &ldpaa_eth_ops, .priv_auto = sizeof(struct ldpaa_eth_priv), .plat_auto = sizeof(struct eth_pdata), };