diff options
Diffstat (limited to 'driver/aim-network')
-rw-r--r-- | driver/aim-network/networking.c | 575 | ||||
-rw-r--r-- | driver/aim-network/networking.h | 21 |
2 files changed, 596 insertions, 0 deletions
diff --git a/driver/aim-network/networking.c b/driver/aim-network/networking.c new file mode 100644 index 0000000..2f42de4 --- /dev/null +++ b/driver/aim-network/networking.c @@ -0,0 +1,575 @@ +/* + * Networking AIM - Networking Application Interface Module for MostCore + * + * Copyright (C) 2015, Microchip Technology Germany II GmbH & Co. KG + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * This file is licensed under GPLv2. + */ + +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include <linux/module.h> +#include <linux/netdevice.h> +#include <linux/etherdevice.h> +#include <linux/slab.h> +#include <linux/init.h> +#include <linux/list.h> +#include <linux/wait.h> +#include <linux/kobject.h> +#include "mostcore.h" +#include "networking.h" + +#define MEP_HDR_LEN 8 +#define MDP_HDR_LEN 16 +#define MAMAC_DATA_LEN (1024 - MDP_HDR_LEN) + +#define PMHL 5 + +#define PMS_TELID_UNSEGM_MAMAC 0x0A +#define PMS_FIFONO_MDP 0x01 +#define PMS_FIFONO_MEP 0x04 +#define PMS_MSGTYPE_DATA 0x04 +#define PMS_DEF_PRIO 0 +#define MEP_DEF_RETRY 15 + +#define PMS_FIFONO_MASK 0x07 +#define PMS_FIFONO_SHIFT 3 +#define PMS_RETRY_SHIFT 4 +#define PMS_TELID_MASK 0x0F +#define PMS_TELID_SHIFT 4 + +#define HB(value) ((u8)((u16)(value) >> 8)) +#define LB(value) ((u8)(value)) + +#define EXTRACT_BIT_SET(bitset_name, value) \ + (((value) >> bitset_name##_SHIFT) & bitset_name##_MASK) + +#define PMS_IS_MEP(buf, len) \ + ((len) > MEP_HDR_LEN && \ + EXTRACT_BIT_SET(PMS_FIFONO, (buf)[3]) == PMS_FIFONO_MEP) + +#define PMS_IS_MAMAC(buf, len) \ + ((len) > MDP_HDR_LEN && \ + EXTRACT_BIT_SET(PMS_FIFONO, (buf)[3]) == PMS_FIFONO_MDP && \ + EXTRACT_BIT_SET(PMS_TELID, (buf)[14]) == PMS_TELID_UNSEGM_MAMAC) + +struct net_dev_channel { + bool linked; + int ch_id; +}; + +struct net_dev_context { + struct most_interface *iface; + bool channels_opened; + bool is_mamac; + unsigned char link_stat; + struct net_device *dev; + struct net_dev_channel rx; + struct net_dev_channel tx; + struct list_head list; +}; + +static struct list_head net_devices = LIST_HEAD_INIT(net_devices); +static struct spinlock list_lock; +static struct most_aim aim; + +static int skb_to_mamac(const struct sk_buff *skb, struct mbo *mbo) +{ + u8 *buff = mbo->virt_address; + const u8 broadcast[] = { 0x03, 0xFF }; + const u8 *dest_addr = skb->data + 4; + const u8 *eth_type = skb->data + 12; + unsigned int payload_len = skb->len - ETH_HLEN; + unsigned int mdp_len = payload_len + MDP_HDR_LEN; + + if (mbo->buffer_length < mdp_len) { + pr_err("drop: too small buffer! (%d for %d)\n", + mbo->buffer_length, mdp_len); + return -EINVAL; + } + + if (skb->len < ETH_HLEN) { + pr_err("drop: too small packet! (%d)\n", skb->len); + return -EINVAL; + } + + if (dest_addr[0] == 0xFF && dest_addr[1] == 0xFF) + dest_addr = broadcast; + + *buff++ = HB(mdp_len - 2); + *buff++ = LB(mdp_len - 2); + + *buff++ = PMHL; + *buff++ = (PMS_FIFONO_MDP << PMS_FIFONO_SHIFT) | PMS_MSGTYPE_DATA; + *buff++ = PMS_DEF_PRIO; + *buff++ = dest_addr[0]; + *buff++ = dest_addr[1]; + *buff++ = 0x00; + + *buff++ = HB(payload_len + 6); + *buff++ = LB(payload_len + 6); + + /* end of FPH here */ + + *buff++ = eth_type[0]; + *buff++ = eth_type[1]; + *buff++ = 0; + *buff++ = 0; + + *buff++ = PMS_TELID_UNSEGM_MAMAC << 4 | HB(payload_len); + *buff++ = LB(payload_len); + + memcpy(buff, skb->data + ETH_HLEN, payload_len); + mbo->buffer_length = mdp_len; + return 0; +} + +static int skb_to_mep(const struct sk_buff *skb, struct mbo *mbo) +{ + u8 *buff = mbo->virt_address; + unsigned int mep_len = skb->len + MEP_HDR_LEN; + + if (mbo->buffer_length < mep_len) { + pr_err("drop: too small buffer! (%d for %d)\n", + mbo->buffer_length, mep_len); + return -EINVAL; + } + + *buff++ = HB(mep_len - 2); + *buff++ = LB(mep_len - 2); + + *buff++ = PMHL; + *buff++ = (PMS_FIFONO_MEP << PMS_FIFONO_SHIFT) | PMS_MSGTYPE_DATA; + *buff++ = (MEP_DEF_RETRY << PMS_RETRY_SHIFT) | PMS_DEF_PRIO; + *buff++ = 0; + *buff++ = 0; + *buff++ = 0; + + memcpy(buff, skb->data, skb->len); + mbo->buffer_length = mep_len; + return 0; +} + +static int most_nd_set_mac_address(struct net_device *dev, void *p) +{ + struct net_dev_context *nd = dev->ml_priv; + int err = eth_mac_addr(dev, p); + + if (err) + return err; + + BUG_ON(nd->dev != dev); + + nd->is_mamac = + (dev->dev_addr[0] == 0 && dev->dev_addr[1] == 0 && + dev->dev_addr[2] == 0 && dev->dev_addr[3] == 0); + + /* + * Set default MTU for the given packet type. + * It is still possible to change MTU using ip tools afterwards. + */ + dev->mtu = nd->is_mamac ? MAMAC_DATA_LEN : ETH_DATA_LEN; + + return 0; +} + +static int most_nd_open(struct net_device *dev) +{ + struct net_dev_context *nd = dev->ml_priv; + + netdev_info(dev, "open net device\n"); + + BUG_ON(nd->dev != dev); + + if (nd->channels_opened) + return -EFAULT; + + BUG_ON(!nd->tx.linked || !nd->rx.linked); + + if (most_start_channel(nd->iface, nd->rx.ch_id, &aim)) { + netdev_err(dev, "most_start_channel() failed\n"); + return -EBUSY; + } + + if (most_start_channel(nd->iface, nd->tx.ch_id, &aim)) { + netdev_err(dev, "most_start_channel() failed\n"); + most_stop_channel(nd->iface, nd->rx.ch_id, &aim); + return -EBUSY; + } + + nd->channels_opened = true; + + if (nd->is_mamac) { + nd->link_stat = 1; + netif_wake_queue(dev); + } else { + nd->iface->request_netinfo(nd->iface, nd->tx.ch_id); + } + + return 0; +} + +static int most_nd_stop(struct net_device *dev) +{ + struct net_dev_context *nd = dev->ml_priv; + + netdev_info(dev, "stop net device\n"); + + BUG_ON(nd->dev != dev); + netif_stop_queue(dev); + + if (nd->channels_opened) { + most_stop_channel(nd->iface, nd->rx.ch_id, &aim); + most_stop_channel(nd->iface, nd->tx.ch_id, &aim); + nd->channels_opened = false; + } + + return 0; +} + +static netdev_tx_t most_nd_start_xmit(struct sk_buff *skb, + struct net_device *dev) +{ + struct net_dev_context *nd = dev->ml_priv; + struct mbo *mbo; + int ret; + + BUG_ON(nd->dev != dev); + + mbo = most_get_mbo(nd->iface, nd->tx.ch_id, &aim); + + if (!mbo) { + netif_stop_queue(dev); + dev->stats.tx_fifo_errors++; + return NETDEV_TX_BUSY; + } + + if (nd->is_mamac) + ret = skb_to_mamac(skb, mbo); + else + ret = skb_to_mep(skb, mbo); + + if (ret) { + most_put_mbo(mbo); + dev->stats.tx_dropped++; + kfree_skb(skb); + return NETDEV_TX_OK; + } + + most_submit_mbo(mbo); + dev->stats.tx_packets++; + dev->stats.tx_bytes += skb->len; + kfree_skb(skb); + return NETDEV_TX_OK; +} + +static const struct net_device_ops most_nd_ops = { + .ndo_open = most_nd_open, + .ndo_stop = most_nd_stop, + .ndo_start_xmit = most_nd_start_xmit, + .ndo_set_mac_address = most_nd_set_mac_address, +}; + +static void most_nd_setup(struct net_device *dev) +{ + netdev_info(dev, "setup net device\n"); + ether_setup(dev); + dev->netdev_ops = &most_nd_ops; +} + +static void most_net_rm_netdev_safe(struct net_dev_context *nd) +{ + if (!nd->dev) + return; + + pr_info("remove net device %p\n", nd->dev); + + unregister_netdev(nd->dev); + free_netdev(nd->dev); + nd->dev = NULL; +} + +static struct net_dev_context *get_net_dev_context( + struct most_interface *iface) +{ + struct net_dev_context *nd, *tmp; + + spin_lock(&list_lock); + list_for_each_entry_safe(nd, tmp, &net_devices, list) { + if (nd->iface == iface) { + spin_unlock(&list_lock); + return nd; + } + } + spin_unlock(&list_lock); + return NULL; +} + +static int aim_probe_channel(struct most_interface *iface, int channel_idx, + struct most_channel_config *ccfg, + struct kobject *parent, char *name) +{ + struct net_dev_context *nd; + struct net_dev_channel *ch; + + if (!iface) + return -EINVAL; + + if (ccfg->data_type != MOST_CH_ASYNC) + return -EINVAL; + + nd = get_net_dev_context(iface); + + if (!nd) { + nd = kzalloc(sizeof(*nd), GFP_KERNEL); + if (!nd) + return -ENOMEM; + + nd->iface = iface; + + spin_lock(&list_lock); + list_add(&nd->list, &net_devices); + spin_unlock(&list_lock); + } + + ch = ccfg->direction == MOST_CH_TX ? &nd->tx : &nd->rx; + if (ch->linked) { + pr_err("only one channel per instance & direction allowed\n"); + return -EINVAL; + } + + if (nd->tx.linked || nd->rx.linked) { + struct net_device *dev = + alloc_netdev(0, "meth%d", NET_NAME_UNKNOWN, + most_nd_setup); + + if (!dev) { + pr_err("no memory for net_device\n"); + return -ENOMEM; + } + + nd->dev = dev; + ch->ch_id = channel_idx; + ch->linked = true; + + dev->ml_priv = nd; + if (register_netdev(dev)) { + pr_err("registering net device failed\n"); + ch->linked = false; + free_netdev(dev); + return -EINVAL; + } + } + + ch->ch_id = channel_idx; + ch->linked = true; + + return 0; +} + +static int aim_disconnect_channel(struct most_interface *iface, + int channel_idx) +{ + struct net_dev_context *nd; + struct net_dev_channel *ch; + + nd = get_net_dev_context(iface); + if (!nd) + return -EINVAL; + + if (nd->rx.linked && channel_idx == nd->rx.ch_id) + ch = &nd->rx; + else if (nd->tx.linked && channel_idx == nd->tx.ch_id) + ch = &nd->tx; + else + return -EINVAL; + + ch->linked = false; + + /* + * do not call most_stop_channel() here, because channels are + * going to be closed in ndo_stop() after unregister_netdev() + */ + most_net_rm_netdev_safe(nd); + + if (!nd->rx.linked && !nd->tx.linked) { + spin_lock(&list_lock); + list_del(&nd->list); + spin_unlock(&list_lock); + kfree(nd); + } + + return 0; +} + +static int aim_resume_tx_channel(struct most_interface *iface, + int channel_idx) +{ + struct net_dev_context *nd; + + nd = get_net_dev_context(iface); + if (!nd || !nd->channels_opened || nd->tx.ch_id != channel_idx) + return 0; + + if (!nd->dev) + return 0; + + netif_wake_queue(nd->dev); + return 0; +} + +static int aim_rx_data(struct mbo *mbo) +{ + const u32 zero = 0; + struct net_dev_context *nd; + char *buf = mbo->virt_address; + u32 len = mbo->processed_length; + struct sk_buff *skb; + struct net_device *dev; + unsigned int skb_len; + + nd = get_net_dev_context(mbo->ifp); + if (!nd || !nd->channels_opened || nd->rx.ch_id != mbo->hdm_channel_id) + return -EIO; + + dev = nd->dev; + if (!dev) { + pr_err_once("drop packet: missing net_device\n"); + return -EIO; + } + + if (nd->is_mamac) { + if (!PMS_IS_MAMAC(buf, len)) + return -EIO; + + skb = dev_alloc_skb(len - MDP_HDR_LEN + 2 * ETH_ALEN + 2); + } else { + if (!PMS_IS_MEP(buf, len)) + return -EIO; + + skb = dev_alloc_skb(len - MEP_HDR_LEN); + } + + if (!skb) { + dev->stats.rx_dropped++; + pr_err_once("drop packet: no memory for skb\n"); + goto out; + } + + skb->dev = dev; + + if (nd->is_mamac) { + /* dest */ + ether_addr_copy(skb_put(skb, ETH_ALEN), dev->dev_addr); + + /* src */ + memcpy(skb_put(skb, 4), &zero, 4); + memcpy(skb_put(skb, 2), buf + 5, 2); + + /* eth type */ + memcpy(skb_put(skb, 2), buf + 10, 2); + + buf += MDP_HDR_LEN; + len -= MDP_HDR_LEN; + } else { + buf += MEP_HDR_LEN; + len -= MEP_HDR_LEN; + } + + memcpy(skb_put(skb, len), buf, len); + skb->protocol = eth_type_trans(skb, dev); + skb_len = skb->len; + if (netif_rx(skb) == NET_RX_SUCCESS) { + dev->stats.rx_packets++; + dev->stats.rx_bytes += skb_len; + } else { + dev->stats.rx_dropped++; + } + +out: + most_put_mbo(mbo); + return 0; +} + +static struct most_aim aim = { + .name = "networking", + .probe_channel = aim_probe_channel, + .disconnect_channel = aim_disconnect_channel, + .tx_completion = aim_resume_tx_channel, + .rx_completion = aim_rx_data, +}; + +static int __init most_net_init(void) +{ + pr_info("most_net_init()\n"); + spin_lock_init(&list_lock); + return most_register_aim(&aim); +} + +static void __exit most_net_exit(void) +{ + struct net_dev_context *nd, *tmp; + + spin_lock(&list_lock); + list_for_each_entry_safe(nd, tmp, &net_devices, list) { + list_del(&nd->list); + spin_unlock(&list_lock); + /* + * do not call most_stop_channel() here, because channels are + * going to be closed in ndo_stop() after unregister_netdev() + */ + most_net_rm_netdev_safe(nd); + kfree(nd); + spin_lock(&list_lock); + } + spin_unlock(&list_lock); + + most_deregister_aim(&aim); + pr_info("most_net_exit()\n"); +} + +/** + * most_deliver_netinfo - callback for HDM to be informed about HW's MAC + * @param iface - most interface instance + * @param link_stat - link status + * @param mac_addr - MAC address + */ +void most_deliver_netinfo(struct most_interface *iface, + unsigned char link_stat, unsigned char *mac_addr) +{ + struct net_dev_context *nd; + struct net_device *dev; + + pr_info("Received netinfo from %s\n", iface->description); + + nd = get_net_dev_context(iface); + if (!nd) + return; + + dev = nd->dev; + if (!dev) + return; + + if (mac_addr) + ether_addr_copy(dev->dev_addr, mac_addr); + + if (nd->link_stat != link_stat) { + nd->link_stat = link_stat; + if (nd->link_stat) + netif_wake_queue(dev); + else + netif_stop_queue(dev); + } +} +EXPORT_SYMBOL(most_deliver_netinfo); + +module_init(most_net_init); +module_exit(most_net_exit); +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("Andrey Shvetsov <andrey.shvetsov@k2l.de>"); +MODULE_DESCRIPTION("Networking Application Interface Module for MostCore"); diff --git a/driver/aim-network/networking.h b/driver/aim-network/networking.h new file mode 100644 index 0000000..6f346d4 --- /dev/null +++ b/driver/aim-network/networking.h @@ -0,0 +1,21 @@ +/* + * Networking AIM - Networking Application Interface Module for MostCore + * + * Copyright (C) 2015, Microchip Technology Germany II GmbH & Co. KG + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * This file is licensed under GPLv2. + */ +#ifndef _NETWORKING_H_ +#define _NETWORKING_H_ + +#include "mostcore.h" + +void most_deliver_netinfo(struct most_interface *iface, + unsigned char link_stat, unsigned char *mac_addr); + +#endif |