[deliverable/linux.git] / drivers / net / ethernet / chelsio / cxgb4 / l2t.c

/*
 * This file is part of the Chelsio T4 Ethernet driver for Linux.
 *
 * Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/skbuff.h>
#include <linux/netdevice.h>
#include <linux/if.h>
#include <linux/if_vlan.h>
#include <linux/jhash.h>
#include <linux/module.h>
#include <linux/debugfs.h>
#include <linux/seq_file.h>
#include <net/neighbour.h>
#include "cxgb4.h"
#include "l2t.h"
#include "t4_msg.h"
#include "t4fw_api.h"
#include "t4_regs.h"

#define VLAN_NONE 0xfff

/* identifies sync vs async L2T_WRITE_REQs */
#define F_SYNC_WR    (1 << 12)

enum {
	L2T_STATE_VALID,      /* entry is up to date */
	L2T_STATE_STALE,      /* entry may be used but needs revalidation */
	L2T_STATE_RESOLVING,  /* entry needs address resolution */
	L2T_STATE_SYNC_WRITE, /* synchronous write of entry underway */

	/* when state is one of the below the entry is not hashed */
	L2T_STATE_SWITCHING,  /* entry is being used by a switching filter */
	L2T_STATE_UNUSED      /* entry not in use */
};

struct l2t_data {
	rwlock_t lock;
	atomic_t nfree;             /* number of free entries */
	struct l2t_entry *rover;    /* starting point for next allocation */
	struct l2t_entry l2tab[L2T_SIZE];
};

static inline unsigned int vlan_prio(const struct l2t_entry *e)
{
	return e->vlan >> 13;
}

static inline void l2t_hold(struct l2t_data *d, struct l2t_entry *e)
{
	if (atomic_add_return(1, &e->refcnt) == 1)  /* 0 -> 1 transition */
		atomic_dec(&d->nfree);
}

/*
 * To avoid having to check address families we do not allow v4 and v6
 * neighbors to be on the same hash chain.  We keep v4 entries in the first
 * half of available hash buckets and v6 in the second.
 */
enum {
	L2T_SZ_HALF = L2T_SIZE / 2,
	L2T_HASH_MASK = L2T_SZ_HALF - 1
};

static inline unsigned int arp_hash(const u32 *key, int ifindex)
{
	return jhash_2words(*key, ifindex, 0) & L2T_HASH_MASK;
}

static inline unsigned int ipv6_hash(const u32 *key, int ifindex)
{
	u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3];

	return L2T_SZ_HALF + (jhash_2words(xor, ifindex, 0) & L2T_HASH_MASK);
}

static unsigned int addr_hash(const u32 *addr, int addr_len, int ifindex)
{
	return addr_len == 4 ? arp_hash(addr, ifindex) :
			       ipv6_hash(addr, ifindex);
}

/*
 * Checks if an L2T entry is for the given IP/IPv6 address.  It does not check
 * whether the L2T entry and the address are of the same address family.
 * Callers ensure an address is only checked against L2T entries of the same
 * family, something made trivial by the separation of IP and IPv6 hash chains
 * mentioned above.  Returns 0 if there's a match,
 */
static int addreq(const struct l2t_entry *e, const u32 *addr)
{
	if (e->v6)
		return (e->addr[0] ^ addr[0]) | (e->addr[1] ^ addr[1]) |
		       (e->addr[2] ^ addr[2]) | (e->addr[3] ^ addr[3]);
	return e->addr[0] ^ addr[0];
}

static void neigh_replace(struct l2t_entry *e, struct neighbour *n)
{
	neigh_hold(n);
	if (e->neigh)
		neigh_release(e->neigh);
	e->neigh = n;
}

/*
 * Write an L2T entry.  Must be called with the entry locked.
 * The write may be synchronous or asynchronous.
 */
static int write_l2e(struct adapter *adap, struct l2t_entry *e, int sync)
{
	struct sk_buff *skb;
	struct cpl_l2t_write_req *req;

	skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
	if (!skb)
		return -ENOMEM;

	req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
	INIT_TP_WR(req, 0);

	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ,
					e->idx | (sync ? F_SYNC_WR : 0) |
					TID_QID(adap->sge.fw_evtq.abs_id)));
	req->params = htons(L2T_W_PORT(e->lport) | L2T_W_NOREPLY(!sync));
	req->l2t_idx = htons(e->idx);
	req->vlan = htons(e->vlan);
	if (e->neigh)
		memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
	memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));

	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
	t4_ofld_send(adap, skb);

	if (sync && e->state != L2T_STATE_SWITCHING)
		e->state = L2T_STATE_SYNC_WRITE;
	return 0;
}

/*
 * Send packets waiting in an L2T entry's ARP queue.  Must be called with the
 * entry locked.
 */
static void send_pending(struct adapter *adap, struct l2t_entry *e)
{
	while (e->arpq_head) {
		struct sk_buff *skb = e->arpq_head;

		e->arpq_head = skb->next;
		skb->next = NULL;
		t4_ofld_send(adap, skb);
	}
	e->arpq_tail = NULL;
}

/*
 * Process a CPL_L2T_WRITE_RPL.  Wake up the ARP queue if it completes a
 * synchronous L2T_WRITE.  Note that the TID in the reply is really the L2T
 * index it refers to.
 */
void do_l2t_write_rpl(struct adapter *adap, const struct cpl_l2t_write_rpl *rpl)
{
	unsigned int tid = GET_TID(rpl);
	unsigned int idx = tid & (L2T_SIZE - 1);

	if (unlikely(rpl->status != CPL_ERR_NONE)) {
		dev_err(adap->pdev_dev,
			"Unexpected L2T_WRITE_RPL status %u for entry %u\n",
			rpl->status, idx);
		return;
	}

	if (tid & F_SYNC_WR) {
		struct l2t_entry *e = &adap->l2t->l2tab[idx];

		spin_lock(&e->lock);
		if (e->state != L2T_STATE_SWITCHING) {
			send_pending(adap, e);
			e->state = (e->neigh->nud_state & NUD_STALE) ?
					L2T_STATE_STALE : L2T_STATE_VALID;
		}
		spin_unlock(&e->lock);
	}
}

/*
 * Add a packet to an L2T entry's queue of packets awaiting resolution.
 * Must be called with the entry's lock held.
 */
static inline void arpq_enqueue(struct l2t_entry *e, struct sk_buff *skb)
{
	skb->next = NULL;
	if (e->arpq_head)
		e->arpq_tail->next = skb;
	else
		e->arpq_head = skb;
	e->arpq_tail = skb;
}

int cxgb4_l2t_send(struct net_device *dev, struct sk_buff *skb,
		   struct l2t_entry *e)
{
	struct adapter *adap = netdev2adap(dev);

again:
	switch (e->state) {
	case L2T_STATE_STALE:     /* entry is stale, kick off revalidation */
		neigh_event_send(e->neigh, NULL);
		spin_lock_bh(&e->lock);
		if (e->state == L2T_STATE_STALE)
			e->state = L2T_STATE_VALID;
		spin_unlock_bh(&e->lock);
	case L2T_STATE_VALID:     /* fast-path, send the packet on */
		return t4_ofld_send(adap, skb);
	case L2T_STATE_RESOLVING:
	case L2T_STATE_SYNC_WRITE:
		spin_lock_bh(&e->lock);
		if (e->state != L2T_STATE_SYNC_WRITE &&
		    e->state != L2T_STATE_RESOLVING) {
			spin_unlock_bh(&e->lock);
			goto again;
		}
		arpq_enqueue(e, skb);
		spin_unlock_bh(&e->lock);

		if (e->state == L2T_STATE_RESOLVING &&
		    !neigh_event_send(e->neigh, NULL)) {
			spin_lock_bh(&e->lock);
			if (e->state == L2T_STATE_RESOLVING && e->arpq_head)
				write_l2e(adap, e, 1);
			spin_unlock_bh(&e->lock);
		}
	}
	return 0;
}
EXPORT_SYMBOL(cxgb4_l2t_send);

/*
 * Allocate a free L2T entry.  Must be called with l2t_data.lock held.
 */
static struct l2t_entry *alloc_l2e(struct l2t_data *d)
{
	struct l2t_entry *end, *e, **p;

	if (!atomic_read(&d->nfree))
		return NULL;

	/* there's definitely a free entry */
	for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e)
		if (atomic_read(&e->refcnt) == 0)
			goto found;

	for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
		;
found:
	d->rover = e + 1;
	atomic_dec(&d->nfree);

	/*
	 * The entry we found may be an inactive entry that is
	 * presently in the hash table.  We need to remove it.
	 */
	if (e->state < L2T_STATE_SWITCHING)
		for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next)
			if (*p == e) {
				*p = e->next;
				e->next = NULL;
				break;
			}

	e->state = L2T_STATE_UNUSED;
	return e;
}

/*
 * Called when an L2T entry has no more users.
 */
static void t4_l2e_free(struct l2t_entry *e)
{
	struct l2t_data *d;

	spin_lock_bh(&e->lock);
	if (atomic_read(&e->refcnt) == 0) {  /* hasn't been recycled */
		if (e->neigh) {
			neigh_release(e->neigh);
			e->neigh = NULL;
		}
		while (e->arpq_head) {
			struct sk_buff *skb = e->arpq_head;

			e->arpq_head = skb->next;
			kfree_skb(skb);
		}
		e->arpq_tail = NULL;
	}
	spin_unlock_bh(&e->lock);

	d = container_of(e, struct l2t_data, l2tab[e->idx]);
	atomic_inc(&d->nfree);
}

void cxgb4_l2t_release(struct l2t_entry *e)
{
	if (atomic_dec_and_test(&e->refcnt))
		t4_l2e_free(e);
}
EXPORT_SYMBOL(cxgb4_l2t_release);

/*
 * Update an L2T entry that was previously used for the same next hop as neigh.
 * Must be called with softirqs disabled.
 */
static void reuse_entry(struct l2t_entry *e, struct neighbour *neigh)
{
	unsigned int nud_state;

	spin_lock(&e->lock);                /* avoid race with t4_l2t_free */
	if (neigh != e->neigh)
		neigh_replace(e, neigh);
	nud_state = neigh->nud_state;
	if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) ||
	    !(nud_state & NUD_VALID))
		e->state = L2T_STATE_RESOLVING;
	else if (nud_state & NUD_CONNECTED)
		e->state = L2T_STATE_VALID;
	else
		e->state = L2T_STATE_STALE;
	spin_unlock(&e->lock);
}

struct l2t_entry *cxgb4_l2t_get(struct l2t_data *d, struct neighbour *neigh,
				const struct net_device *physdev,
				unsigned int priority)
{
	u8 lport;
	u16 vlan;
	struct l2t_entry *e;
	int addr_len = neigh->tbl->key_len;
	u32 *addr = (u32 *)neigh->primary_key;
	int ifidx = neigh->dev->ifindex;
	int hash = addr_hash(addr, addr_len, ifidx);

	if (neigh->dev->flags & IFF_LOOPBACK)
		lport = netdev2pinfo(physdev)->tx_chan + 4;
	else
		lport = netdev2pinfo(physdev)->lport;

	if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
		vlan = vlan_dev_vlan_id(neigh->dev);
	else
		vlan = VLAN_NONE;

	write_lock_bh(&d->lock);
	for (e = d->l2tab[hash].first; e; e = e->next)
		if (!addreq(e, addr) && e->ifindex == ifidx &&
		    e->vlan == vlan && e->lport == lport) {
			l2t_hold(d, e);
			if (atomic_read(&e->refcnt) == 1)
				reuse_entry(e, neigh);
			goto done;
		}

	/* Need to allocate a new entry */
	e = alloc_l2e(d);
	if (e) {
		spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
		e->state = L2T_STATE_RESOLVING;
		memcpy(e->addr, addr, addr_len);
		e->ifindex = ifidx;
		e->hash = hash;
		e->lport = lport;
		e->v6 = addr_len == 16;
		atomic_set(&e->refcnt, 1);
		neigh_replace(e, neigh);
		e->vlan = vlan;
		e->next = d->l2tab[hash].first;
		d->l2tab[hash].first = e;
		spin_unlock(&e->lock);
	}
done:
	write_unlock_bh(&d->lock);
	return e;
}
EXPORT_SYMBOL(cxgb4_l2t_get);

u64 cxgb4_select_ntuple(struct net_device *dev,
			const struct l2t_entry *l2t)
{
	struct adapter *adap = netdev2adap(dev);
	struct tp_params *tp = &adap->params.tp;
	u64 ntuple = 0;

	/* Initialize each of the fields which we care about which are present
	 * in the Compressed Filter Tuple.
	 */
	if (tp->vlan_shift >= 0 && l2t->vlan != VLAN_NONE)
		ntuple |= (F_FT_VLAN_VLD | l2t->vlan) << tp->vlan_shift;

	if (tp->port_shift >= 0)
		ntuple |= (u64)l2t->lport << tp->port_shift;

	if (tp->protocol_shift >= 0)
		ntuple |= (u64)IPPROTO_TCP << tp->protocol_shift;

	if (tp->vnic_shift >= 0) {
		u32 viid = cxgb4_port_viid(dev);
		u32 vf = FW_VIID_VIN_GET(viid);
		u32 pf = FW_VIID_PFN_GET(viid);
		u32 vld = FW_VIID_VIVLD_GET(viid);

		ntuple |= (u64)(V_FT_VNID_ID_VF(vf) |
				V_FT_VNID_ID_PF(pf) |
				V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift;
	}

	return ntuple;
}
EXPORT_SYMBOL(cxgb4_select_ntuple);

/*
 * Called when address resolution fails for an L2T entry to handle packets
 * on the arpq head.  If a packet specifies a failure handler it is invoked,
 * otherwise the packet is sent to the device.
 */
static void handle_failed_resolution(struct adapter *adap, struct sk_buff *arpq)
{
	while (arpq) {
		struct sk_buff *skb = arpq;
		const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);

		arpq = skb->next;
		skb->next = NULL;
		if (cb->arp_err_handler)
			cb->arp_err_handler(cb->handle, skb);
		else
			t4_ofld_send(adap, skb);
	}
}

/*
 * Called when the host's neighbor layer makes a change to some entry that is
 * loaded into the HW L2 table.
 */
void t4_l2t_update(struct adapter *adap, struct neighbour *neigh)
{
	struct l2t_entry *e;
	struct sk_buff *arpq = NULL;
	struct l2t_data *d = adap->l2t;
	int addr_len = neigh->tbl->key_len;
	u32 *addr = (u32 *) neigh->primary_key;
	int ifidx = neigh->dev->ifindex;
	int hash = addr_hash(addr, addr_len, ifidx);

	read_lock_bh(&d->lock);
	for (e = d->l2tab[hash].first; e; e = e->next)
		if (!addreq(e, addr) && e->ifindex == ifidx) {
			spin_lock(&e->lock);
			if (atomic_read(&e->refcnt))
				goto found;
			spin_unlock(&e->lock);
			break;
		}
	read_unlock_bh(&d->lock);
	return;

 found:
	read_unlock(&d->lock);

	if (neigh != e->neigh)
		neigh_replace(e, neigh);

	if (e->state == L2T_STATE_RESOLVING) {
		if (neigh->nud_state & NUD_FAILED) {
			arpq = e->arpq_head;
			e->arpq_head = e->arpq_tail = NULL;
		} else if ((neigh->nud_state & (NUD_CONNECTED | NUD_STALE)) &&
			   e->arpq_head) {
			write_l2e(adap, e, 1);
		}
	} else {
		e->state = neigh->nud_state & NUD_CONNECTED ?
			L2T_STATE_VALID : L2T_STATE_STALE;
		if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
			write_l2e(adap, e, 0);
	}

	spin_unlock_bh(&e->lock);

	if (arpq)
		handle_failed_resolution(adap, arpq);
}

/* Allocate an L2T entry for use by a switching rule.  Such need to be
 * explicitly freed and while busy they are not on any hash chain, so normal
 * address resolution updates do not see them.
 */
struct l2t_entry *t4_l2t_alloc_switching(struct l2t_data *d)
{
	struct l2t_entry *e;

	write_lock_bh(&d->lock);
	e = alloc_l2e(d);
	if (e) {
		spin_lock(&e->lock);          /* avoid race with t4_l2t_free */
		e->state = L2T_STATE_SWITCHING;
		atomic_set(&e->refcnt, 1);
		spin_unlock(&e->lock);
	}
	write_unlock_bh(&d->lock);
	return e;
}

/* Sets/updates the contents of a switching L2T entry that has been allocated
 * with an earlier call to @t4_l2t_alloc_switching.
 */
int t4_l2t_set_switching(struct adapter *adap, struct l2t_entry *e, u16 vlan,
		u8 port, u8 *eth_addr)
{
	e->vlan = vlan;
	e->lport = port;
	memcpy(e->dmac, eth_addr, ETH_ALEN);
	return write_l2e(adap, e, 0);
}

struct l2t_data *t4_init_l2t(void)
{
	int i;
	struct l2t_data *d;

	d = t4_alloc_mem(sizeof(*d));
	if (!d)
		return NULL;

	d->rover = d->l2tab;
	atomic_set(&d->nfree, L2T_SIZE);
	rwlock_init(&d->lock);

	for (i = 0; i < L2T_SIZE; ++i) {
		d->l2tab[i].idx = i;
		d->l2tab[i].state = L2T_STATE_UNUSED;
		spin_lock_init(&d->l2tab[i].lock);
		atomic_set(&d->l2tab[i].refcnt, 0);
	}
	return d;
}

static inline void *l2t_get_idx(struct seq_file *seq, loff_t pos)
{
	struct l2t_entry *l2tab = seq->private;

	return pos >= L2T_SIZE ? NULL : &l2tab[pos];
}

static void *l2t_seq_start(struct seq_file *seq, loff_t *pos)
{
	return *pos ? l2t_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
}

static void *l2t_seq_next(struct seq_file *seq, void *v, loff_t *pos)
{
	v = l2t_get_idx(seq, *pos);
	if (v)
		++*pos;
	return v;
}

static void l2t_seq_stop(struct seq_file *seq, void *v)
{
}

static char l2e_state(const struct l2t_entry *e)
{
	switch (e->state) {
	case L2T_STATE_VALID: return 'V';
	case L2T_STATE_STALE: return 'S';
	case L2T_STATE_SYNC_WRITE: return 'W';
	case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R';
	case L2T_STATE_SWITCHING: return 'X';
	default:
		return 'U';
	}
}

static int l2t_seq_show(struct seq_file *seq, void *v)
{
	if (v == SEQ_START_TOKEN)
		seq_puts(seq, " Idx IP address                "
			 "Ethernet address  VLAN/P LP State Users Port\n");
	else {
		char ip[60];
		struct l2t_entry *e = v;

		spin_lock_bh(&e->lock);
		if (e->state == L2T_STATE_SWITCHING)
			ip[0] = '\0';
		else
			sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
		seq_printf(seq, "%4u %-25s %17pM %4d %u %2u   %c   %5u %s\n",
			   e->idx, ip, e->dmac,
			   e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
			   l2e_state(e), atomic_read(&e->refcnt),
			   e->neigh ? e->neigh->dev->name : "");
		spin_unlock_bh(&e->lock);
	}
	return 0;
}

static const struct seq_operations l2t_seq_ops = {
	.start = l2t_seq_start,
	.next = l2t_seq_next,
	.stop = l2t_seq_stop,
	.show = l2t_seq_show
};

static int l2t_seq_open(struct inode *inode, struct file *file)
{
	int rc = seq_open(file, &l2t_seq_ops);

	if (!rc) {
		struct adapter *adap = inode->i_private;
		struct seq_file *seq = file->private_data;

		seq->private = adap->l2t->l2tab;
	}
	return rc;
}

const struct file_operations t4_l2t_fops = {
	.owner = THIS_MODULE,
	.open = l2t_seq_open,
	.read = seq_read,
	.llseek = seq_lseek,
	.release = seq_release,
};
Commit	Line	Data
625ba2c2 DM	1	/*
	2	* This file is part of the Chelsio T4 Ethernet driver for Linux.
	3	*
	4	* Copyright (c) 2003-2010 Chelsio Communications, Inc. All rights reserved.
	5	*
	6	* This software is available to you under a choice of one of two
	7	* licenses. You may choose to be licensed under the terms of the GNU
	8	* General Public License (GPL) Version 2, available from the file
	9	* COPYING in the main directory of this source tree, or the
	10	* OpenIB.org BSD license below:
	11	*
	12	* Redistribution and use in source and binary forms, with or
	13	* without modification, are permitted provided that the following
	14	* conditions are met:
	15	*
	16	* - Redistributions of source code must retain the above
	17	* copyright notice, this list of conditions and the following
	18	* disclaimer.
	19	*
	20	* - Redistributions in binary form must reproduce the above
	21	* copyright notice, this list of conditions and the following
	22	* disclaimer in the documentation and/or other materials
	23	* provided with the distribution.
	24	*
	25	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	26	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	27	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	28	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	29	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	30	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	31	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	32	* SOFTWARE.
	33	*/
	34
	35	#include <linux/skbuff.h>
	36	#include <linux/netdevice.h>
	37	#include <linux/if.h>
	38	#include <linux/if_vlan.h>
	39	#include <linux/jhash.h>
310587c3 PG	40	#include <linux/module.h>
	41	#include <linux/debugfs.h>
	42	#include <linux/seq_file.h>
625ba2c2 DM	43	#include <net/neighbour.h>
	44	#include "cxgb4.h"
	45	#include "l2t.h"
	46	#include "t4_msg.h"
	47	#include "t4fw_api.h"
dcf7b6f5	48	#include "t4_regs.h"
625ba2c2 DM	49
	50	#define VLAN_NONE 0xfff
	51
	52	/* identifies sync vs async L2T_WRITE_REQs */
	53	#define F_SYNC_WR (1 << 12)
	54
	55	enum {
	56	L2T_STATE_VALID, /* entry is up to date */
	57	L2T_STATE_STALE, /* entry may be used but needs revalidation */
	58	L2T_STATE_RESOLVING, /* entry needs address resolution */
	59	L2T_STATE_SYNC_WRITE, /* synchronous write of entry underway */
	60
	61	/* when state is one of the below the entry is not hashed */
	62	L2T_STATE_SWITCHING, /* entry is being used by a switching filter */
	63	L2T_STATE_UNUSED /* entry not in use */
	64	};
	65
	66	struct l2t_data {
	67	rwlock_t lock;
	68	atomic_t nfree; /* number of free entries */
	69	struct l2t_entry rover; / starting point for next allocation */
	70	struct l2t_entry l2tab[L2T_SIZE];
	71	};
	72
	73	static inline unsigned int vlan_prio(const struct l2t_entry *e)
	74	{
	75	return e->vlan >> 13;
	76	}
	77
	78	static inline void l2t_hold(struct l2t_data d, struct l2t_entry e)
	79	{
	80	if (atomic_add_return(1, &e->refcnt) == 1) /* 0 -> 1 transition */
	81	atomic_dec(&d->nfree);
	82	}
	83
	84	/*
	85	* To avoid having to check address families we do not allow v4 and v6
	86	* neighbors to be on the same hash chain. We keep v4 entries in the first
	87	* half of available hash buckets and v6 in the second.
	88	*/
	89	enum {
	90	L2T_SZ_HALF = L2T_SIZE / 2,
	91	L2T_HASH_MASK = L2T_SZ_HALF - 1
	92	};
	93
	94	static inline unsigned int arp_hash(const u32 *key, int ifindex)
	95	{
	96	return jhash_2words(*key, ifindex, 0) & L2T_HASH_MASK;
	97	}
	98
	99	static inline unsigned int ipv6_hash(const u32 *key, int ifindex)
	100	{
	101	u32 xor = key[0] ^ key[1] ^ key[2] ^ key[3];
	102
	103	return L2T_SZ_HALF + (jhash_2words(xor, ifindex, 0) & L2T_HASH_MASK);
	104	}
	105
	106	static unsigned int addr_hash(const u32 *addr, int addr_len, int ifindex)
	107	{
	108	return addr_len == 4 ? arp_hash(addr, ifindex) :
	109	ipv6_hash(addr, ifindex);
	110	}
	111
	112	/*
113	* Checks if an L2T entry is for the given IP/IPv6 address. It does not check
114	* whether the L2T entry and the address are of the same address family.
115	* Callers ensure an address is only checked against L2T entries of the same
116	* family, something made trivial by the separation of IP and IPv6 hash chains
117	* mentioned above. Returns 0 if there's a match,
118	*/
119	static int addreq(const struct l2t_entry e, const u32 addr)
120	{
121	if (e->v6)
122	return (e->addr[0] ^ addr[0]) \| (e->addr[1] ^ addr[1]) \|
123	(e->addr[2] ^ addr[2]) \| (e->addr[3] ^ addr[3]);
124	return e->addr[0] ^ addr[0];
125	}
126
127	static void neigh_replace(struct l2t_entry e, struct neighbour n)
128	{
129	neigh_hold(n);
130	if (e->neigh)
131	neigh_release(e->neigh);
132	e->neigh = n;
133	}
134
135	/*
136	* Write an L2T entry. Must be called with the entry locked.
137	* The write may be synchronous or asynchronous.
138	*/
139	static int write_l2e(struct adapter adap, struct l2t_entry e, int sync)
140	{
141	struct sk_buff *skb;
142	struct cpl_l2t_write_req *req;
143
144	skb = alloc_skb(sizeof(*req), GFP_ATOMIC);
145	if (!skb)
146	return -ENOMEM;
147
148	req = (struct cpl_l2t_write_req )__skb_put(skb, sizeof(req));
149	INIT_TP_WR(req, 0);
150
151	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ,
152	e->idx \| (sync ? F_SYNC_WR : 0) \|
153	TID_QID(adap->sge.fw_evtq.abs_id)));
154	req->params = htons(L2T_W_PORT(e->lport) \| L2T_W_NOREPLY(!sync));
155	req->l2t_idx = htons(e->idx);
156	req->vlan = htons(e->vlan);
157	if (e->neigh)
158	memcpy(e->dmac, e->neigh->ha, sizeof(e->dmac));
159	memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
160
161	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
162	t4_ofld_send(adap, skb);
163
164	if (sync && e->state != L2T_STATE_SWITCHING)
165	e->state = L2T_STATE_SYNC_WRITE;
166	return 0;
167	}
168
169	/*
170	* Send packets waiting in an L2T entry's ARP queue. Must be called with the
171	* entry locked.
172	*/
173	static void send_pending(struct adapter adap, struct l2t_entry e)
174	{
175	while (e->arpq_head) {
176	struct sk_buff *skb = e->arpq_head;
177
178	e->arpq_head = skb->next;
179	skb->next = NULL;
180	t4_ofld_send(adap, skb);
181	}
182	e->arpq_tail = NULL;
183	}
184
185	/*
186	* Process a CPL_L2T_WRITE_RPL. Wake up the ARP queue if it completes a
187	* synchronous L2T_WRITE. Note that the TID in the reply is really the L2T
188	* index it refers to.
189	*/
190	void do_l2t_write_rpl(struct adapter adap, const struct cpl_l2t_write_rpl rpl)
191	{
192	unsigned int tid = GET_TID(rpl);
193	unsigned int idx = tid & (L2T_SIZE - 1);
194
195	if (unlikely(rpl->status != CPL_ERR_NONE)) {
196	dev_err(adap->pdev_dev,
197	"Unexpected L2T_WRITE_RPL status %u for entry %u\n",
198	rpl->status, idx);
199	return;
200	}
201
202	if (tid & F_SYNC_WR) {
203	struct l2t_entry *e = &adap->l2t->l2tab[idx];
204
205	spin_lock(&e->lock);
206	if (e->state != L2T_STATE_SWITCHING) {
207	send_pending(adap, e);
208	e->state = (e->neigh->nud_state & NUD_STALE) ?
209	L2T_STATE_STALE : L2T_STATE_VALID;
210	}
211	spin_unlock(&e->lock);
212	}
213	}
214
215	/*
216	* Add a packet to an L2T entry's queue of packets awaiting resolution.
217	* Must be called with the entry's lock held.
218	*/
219	static inline void arpq_enqueue(struct l2t_entry e, struct sk_buff skb)
220	{
221	skb->next = NULL;
222	if (e->arpq_head)
223	e->arpq_tail->next = skb;
224	else
225	e->arpq_head = skb;
226	e->arpq_tail = skb;
227	}
228
229	int cxgb4_l2t_send(struct net_device dev, struct sk_buff skb,
230	struct l2t_entry *e)
231	{
232	struct adapter *adap = netdev2adap(dev);
233
234	again:
235	switch (e->state) {
236	case L2T_STATE_STALE: /* entry is stale, kick off revalidation */
237	neigh_event_send(e->neigh, NULL);
238	spin_lock_bh(&e->lock);
239	if (e->state == L2T_STATE_STALE)
240	e->state = L2T_STATE_VALID;
241	spin_unlock_bh(&e->lock);
242	case L2T_STATE_VALID: /* fast-path, send the packet on */
243	return t4_ofld_send(adap, skb);
244	case L2T_STATE_RESOLVING:
245	case L2T_STATE_SYNC_WRITE:
246	spin_lock_bh(&e->lock);
247	if (e->state != L2T_STATE_SYNC_WRITE &&
248	e->state != L2T_STATE_RESOLVING) {
249	spin_unlock_bh(&e->lock);
250	goto again;
251	}
252	arpq_enqueue(e, skb);
253	spin_unlock_bh(&e->lock);
254
255	if (e->state == L2T_STATE_RESOLVING &&
256	!neigh_event_send(e->neigh, NULL)) {
257	spin_lock_bh(&e->lock);
258	if (e->state == L2T_STATE_RESOLVING && e->arpq_head)
259	write_l2e(adap, e, 1);
260	spin_unlock_bh(&e->lock);
261	}
262	}
263	return 0;
264	}
265	EXPORT_SYMBOL(cxgb4_l2t_send);
266
267	/*
268	* Allocate a free L2T entry. Must be called with l2t_data.lock held.
269	*/
270	static struct l2t_entry alloc_l2e(struct l2t_data d)
271	{
272	struct l2t_entry end, e, **p;
273
274	if (!atomic_read(&d->nfree))
275	return NULL;
276
277	/* there's definitely a free entry */
278	for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e)
279	if (atomic_read(&e->refcnt) == 0)
280	goto found;
281
282	for (e = d->l2tab; atomic_read(&e->refcnt); ++e)
283	;
284	found:
285	d->rover = e + 1;
286	atomic_dec(&d->nfree);
287
288	/*
289	* The entry we found may be an inactive entry that is
290	* presently in the hash table. We need to remove it.
291	*/
292	if (e->state < L2T_STATE_SWITCHING)
293	for (p = &d->l2tab[e->hash].first; p; p = &(p)->next)
294	if (*p == e) {
295	*p = e->next;
296	e->next = NULL;
297	break;
298	}
299
300	e->state = L2T_STATE_UNUSED;
301	return e;
302	}
303
304	/*
305	* Called when an L2T entry has no more users.
306	*/
307	static void t4_l2e_free(struct l2t_entry *e)
308	{
309	struct l2t_data *d;
310
311	spin_lock_bh(&e->lock);
312	if (atomic_read(&e->refcnt) == 0) { /* hasn't been recycled */
313	if (e->neigh) {
314	neigh_release(e->neigh);
315	e->neigh = NULL;
316	}
204dc3c0 DM	317	while (e->arpq_head) {
	318	struct sk_buff *skb = e->arpq_head;
	319
	320	e->arpq_head = skb->next;
05eda04b	321	kfree_skb(skb);
204dc3c0 DM	322	}
204dc3c0 DM	323	e->arpq_tail = NULL;
625ba2c2 DM	324	}
	325	spin_unlock_bh(&e->lock);
	326
	327	d = container_of(e, struct l2t_data, l2tab[e->idx]);
	328	atomic_inc(&d->nfree);
	329	}
	330
	331	void cxgb4_l2t_release(struct l2t_entry *e)
	332	{
	333	if (atomic_dec_and_test(&e->refcnt))
	334	t4_l2e_free(e);
	335	}
	336	EXPORT_SYMBOL(cxgb4_l2t_release);
	337
	338	/*
	339	* Update an L2T entry that was previously used for the same next hop as neigh.
	340	* Must be called with softirqs disabled.
	341	*/
	342	static void reuse_entry(struct l2t_entry e, struct neighbour neigh)
	343	{
	344	unsigned int nud_state;
	345
	346	spin_lock(&e->lock); /* avoid race with t4_l2t_free */
	347	if (neigh != e->neigh)
	348	neigh_replace(e, neigh);
	349	nud_state = neigh->nud_state;
	350	if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)) \|\|
	351	!(nud_state & NUD_VALID))
	352	e->state = L2T_STATE_RESOLVING;
	353	else if (nud_state & NUD_CONNECTED)
	354	e->state = L2T_STATE_VALID;
	355	else
	356	e->state = L2T_STATE_STALE;
	357	spin_unlock(&e->lock);
	358	}
	359
	360	struct l2t_entry cxgb4_l2t_get(struct l2t_data d, struct neighbour *neigh,
	361	const struct net_device *physdev,
	362	unsigned int priority)
	363	{
	364	u8 lport;
	365	u16 vlan;
	366	struct l2t_entry *e;
	367	int addr_len = neigh->tbl->key_len;
	368	u32 addr = (u32 )neigh->primary_key;
	369	int ifidx = neigh->dev->ifindex;
	370	int hash = addr_hash(addr, addr_len, ifidx);
	371
	372	if (neigh->dev->flags & IFF_LOOPBACK)
	373	lport = netdev2pinfo(physdev)->tx_chan + 4;
	374	else
	375	lport = netdev2pinfo(physdev)->lport;
	376
	377	if (neigh->dev->priv_flags & IFF_802_1Q_VLAN)
	378	vlan = vlan_dev_vlan_id(neigh->dev);
	379	else
	380	vlan = VLAN_NONE;
	381
	382	write_lock_bh(&d->lock);
	383	for (e = d->l2tab[hash].first; e; e = e->next)
	384	if (!addreq(e, addr) && e->ifindex == ifidx &&
	385	e->vlan == vlan && e->lport == lport) {
	386	l2t_hold(d, e);
	387	if (atomic_read(&e->refcnt) == 1)
388	reuse_entry(e, neigh);
389	goto done;
390	}
391
392	/* Need to allocate a new entry */
393	e = alloc_l2e(d);
394	if (e) {
395	spin_lock(&e->lock); /* avoid race with t4_l2t_free */
396	e->state = L2T_STATE_RESOLVING;
397	memcpy(e->addr, addr, addr_len);
398	e->ifindex = ifidx;
399	e->hash = hash;
400	e->lport = lport;
401	e->v6 = addr_len == 16;
402	atomic_set(&e->refcnt, 1);
403	neigh_replace(e, neigh);
404	e->vlan = vlan;
405	e->next = d->l2tab[hash].first;
406	d->l2tab[hash].first = e;
407	spin_unlock(&e->lock);
408	}
409	done:
410	write_unlock_bh(&d->lock);
411	return e;
412	}
413	EXPORT_SYMBOL(cxgb4_l2t_get);
414
dcf7b6f5 KS	415	u64 cxgb4_select_ntuple(struct net_device *dev,
	416	const struct l2t_entry *l2t)
	417	{
	418	struct adapter *adap = netdev2adap(dev);
	419	struct tp_params *tp = &adap->params.tp;
	420	u64 ntuple = 0;
	421
	422	/* Initialize each of the fields which we care about which are present
	423	* in the Compressed Filter Tuple.
	424	*/
	425	if (tp->vlan_shift >= 0 && l2t->vlan != VLAN_NONE)
	426	ntuple \|= (F_FT_VLAN_VLD \| l2t->vlan) << tp->vlan_shift;
	427
	428	if (tp->port_shift >= 0)
	429	ntuple \|= (u64)l2t->lport << tp->port_shift;
	430
	431	if (tp->protocol_shift >= 0)
	432	ntuple \|= (u64)IPPROTO_TCP << tp->protocol_shift;
	433
	434	if (tp->vnic_shift >= 0) {
	435	u32 viid = cxgb4_port_viid(dev);
	436	u32 vf = FW_VIID_VIN_GET(viid);
	437	u32 pf = FW_VIID_PFN_GET(viid);
	438	u32 vld = FW_VIID_VIVLD_GET(viid);
	439
	440	ntuple \|= (u64)(V_FT_VNID_ID_VF(vf) \|
	441	V_FT_VNID_ID_PF(pf) \|
	442	V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift;
	443	}
	444
	445	return ntuple;
	446	}
	447	EXPORT_SYMBOL(cxgb4_select_ntuple);
	448
625ba2c2 DM	449	/*
	450	* Called when address resolution fails for an L2T entry to handle packets
	451	* on the arpq head. If a packet specifies a failure handler it is invoked,
	452	* otherwise the packet is sent to the device.
	453	*/
	454	static void handle_failed_resolution(struct adapter adap, struct sk_buff arpq)
	455	{
	456	while (arpq) {
	457	struct sk_buff *skb = arpq;
	458	const struct l2t_skb_cb *cb = L2T_SKB_CB(skb);
	459
	460	arpq = skb->next;
	461	skb->next = NULL;
	462	if (cb->arp_err_handler)
	463	cb->arp_err_handler(cb->handle, skb);
	464	else
	465	t4_ofld_send(adap, skb);
	466	}
	467	}
	468
	469	/*
	470	* Called when the host's neighbor layer makes a change to some entry that is
	471	* loaded into the HW L2 table.
	472	*/
	473	void t4_l2t_update(struct adapter adap, struct neighbour neigh)
	474	{
	475	struct l2t_entry *e;
	476	struct sk_buff *arpq = NULL;
	477	struct l2t_data *d = adap->l2t;
	478	int addr_len = neigh->tbl->key_len;
	479	u32 addr = (u32 ) neigh->primary_key;
	480	int ifidx = neigh->dev->ifindex;
	481	int hash = addr_hash(addr, addr_len, ifidx);
	482
	483	read_lock_bh(&d->lock);
	484	for (e = d->l2tab[hash].first; e; e = e->next)
	485	if (!addreq(e, addr) && e->ifindex == ifidx) {
	486	spin_lock(&e->lock);
	487	if (atomic_read(&e->refcnt))
	488	goto found;
	489	spin_unlock(&e->lock);
	490	break;
	491	}
	492	read_unlock_bh(&d->lock);
	493	return;
	494
	495	found:
	496	read_unlock(&d->lock);
	497
	498	if (neigh != e->neigh)
	499	neigh_replace(e, neigh);
	500
	501	if (e->state == L2T_STATE_RESOLVING) {
	502	if (neigh->nud_state & NUD_FAILED) {
	503	arpq = e->arpq_head;
	504	e->arpq_head = e->arpq_tail = NULL;
	505	} else if ((neigh->nud_state & (NUD_CONNECTED \| NUD_STALE)) &&
	506	e->arpq_head) {
	507	write_l2e(adap, e, 1);
	508	}
	509	} else {
	510	e->state = neigh->nud_state & NUD_CONNECTED ?
	511	L2T_STATE_VALID : L2T_STATE_STALE;
	512	if (memcmp(e->dmac, neigh->ha, sizeof(e->dmac)))
513	write_l2e(adap, e, 0);
514	}
515
516	spin_unlock_bh(&e->lock);
517
518	if (arpq)
519	handle_failed_resolution(adap, arpq);
520	}
521
f2b7e78d VP	522	/* Allocate an L2T entry for use by a switching rule. Such need to be
	523	* explicitly freed and while busy they are not on any hash chain, so normal
	524	* address resolution updates do not see them.
	525	*/
	526	struct l2t_entry t4_l2t_alloc_switching(struct l2t_data d)
	527	{
	528	struct l2t_entry *e;
	529
	530	write_lock_bh(&d->lock);
	531	e = alloc_l2e(d);
	532	if (e) {
	533	spin_lock(&e->lock); /* avoid race with t4_l2t_free */
	534	e->state = L2T_STATE_SWITCHING;
	535	atomic_set(&e->refcnt, 1);
	536	spin_unlock(&e->lock);
	537	}
	538	write_unlock_bh(&d->lock);
	539	return e;
	540	}
	541
	542	/* Sets/updates the contents of a switching L2T entry that has been allocated
	543	* with an earlier call to @t4_l2t_alloc_switching.
	544	*/
	545	int t4_l2t_set_switching(struct adapter adap, struct l2t_entry e, u16 vlan,
	546	u8 port, u8 *eth_addr)
	547	{
	548	e->vlan = vlan;
	549	e->lport = port;
	550	memcpy(e->dmac, eth_addr, ETH_ALEN);
	551	return write_l2e(adap, e, 0);
	552	}
	553
625ba2c2 DM	554	struct l2t_data *t4_init_l2t(void)
	555	{
	556	int i;
	557	struct l2t_data *d;
	558
	559	d = t4_alloc_mem(sizeof(*d));
	560	if (!d)
	561	return NULL;
	562
	563	d->rover = d->l2tab;
	564	atomic_set(&d->nfree, L2T_SIZE);
	565	rwlock_init(&d->lock);
	566
	567	for (i = 0; i < L2T_SIZE; ++i) {
	568	d->l2tab[i].idx = i;
	569	d->l2tab[i].state = L2T_STATE_UNUSED;
	570	spin_lock_init(&d->l2tab[i].lock);
	571	atomic_set(&d->l2tab[i].refcnt, 0);
	572	}
	573	return d;
	574	}
	575
625ba2c2 DM	576	static inline void l2t_get_idx(struct seq_file seq, loff_t pos)
	577	{
	578	struct l2t_entry *l2tab = seq->private;
	579
	580	return pos >= L2T_SIZE ? NULL : &l2tab[pos];
	581	}
	582
	583	static void l2t_seq_start(struct seq_file seq, loff_t *pos)
	584	{
	585	return pos ? l2t_get_idx(seq, pos - 1) : SEQ_START_TOKEN;
	586	}
	587
	588	static void l2t_seq_next(struct seq_file seq, void v, loff_t pos)
	589	{
	590	v = l2t_get_idx(seq, *pos);
	591	if (v)
	592	++*pos;
	593	return v;
	594	}
	595
	596	static void l2t_seq_stop(struct seq_file seq, void v)
	597	{
	598	}
	599
	600	static char l2e_state(const struct l2t_entry *e)
	601	{
	602	switch (e->state) {
	603	case L2T_STATE_VALID: return 'V';
	604	case L2T_STATE_STALE: return 'S';
	605	case L2T_STATE_SYNC_WRITE: return 'W';
	606	case L2T_STATE_RESOLVING: return e->arpq_head ? 'A' : 'R';
	607	case L2T_STATE_SWITCHING: return 'X';
	608	default:
	609	return 'U';
	610	}
	611	}
	612
	613	static int l2t_seq_show(struct seq_file seq, void v)
	614	{
	615	if (v == SEQ_START_TOKEN)
	616	seq_puts(seq, " Idx IP address "
	617	"Ethernet address VLAN/P LP State Users Port\n");
	618	else {
	619	char ip[60];
	620	struct l2t_entry *e = v;
	621
	622	spin_lock_bh(&e->lock);
	623	if (e->state == L2T_STATE_SWITCHING)
	624	ip[0] = '\0';
	625	else
	626	sprintf(ip, e->v6 ? "%pI6c" : "%pI4", e->addr);
	627	seq_printf(seq, "%4u %-25s %17pM %4d %u %2u %c %5u %s\n",
	628	e->idx, ip, e->dmac,
	629	e->vlan & VLAN_VID_MASK, vlan_prio(e), e->lport,
	630	l2e_state(e), atomic_read(&e->refcnt),
	631	e->neigh ? e->neigh->dev->name : "");
	632	spin_unlock_bh(&e->lock);
	633	}
	634	return 0;
	635	}
	636
	637	static const struct seq_operations l2t_seq_ops = {
	638	.start = l2t_seq_start,
	639	.next = l2t_seq_next,
640	.stop = l2t_seq_stop,
641	.show = l2t_seq_show
642	};
643
644	static int l2t_seq_open(struct inode inode, struct file file)
645	{
646	int rc = seq_open(file, &l2t_seq_ops);
647
648	if (!rc) {
649	struct adapter *adap = inode->i_private;
650	struct seq_file *seq = file->private_data;
651
652	seq->private = adap->l2t->l2tab;
653	}
654	return rc;
655	}
656
657	const struct file_operations t4_l2t_fops = {
658	.owner = THIS_MODULE,
659	.open = l2t_seq_open,
660	.read = seq_read,
661	.llseek = seq_lseek,
662	.release = seq_release,
663	};