ramips: implement hardware NAT offload for MT7621

Supports IPv4 flow offloading on MT7621 for Routing, SNAT and DNAT

Supported are regular ethernet->ethernet connections, including one
802.1q VLAN and/or PPPoE encapsulation

Signed-off-by: John Crispin <john@phrozen.org>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This commit is contained in:
John Crispin 2018-03-23 13:42:56 +01:00 committed by Felix Fietkau
parent dea9922acd
commit 424a9ae128
8 changed files with 985 additions and 3 deletions

View file

@ -378,6 +378,16 @@
#size-cells = <1>;
};
hnat: hnat@1e100000 {
compatible = "mediatek,mt7623-hnat";
reg = <0x1e100000 0x10000>;
mtketh-ppd = "eth0";
mtketh-lan = "eth0";
mtketh-wan = "eth0";
resets = <&rstctrl 0>;
reset-names = "mtketh";
};
ethernet: ethernet@1e100000 {
compatible = "mediatek,mt7621-eth";
reg = <0x1e100000 0x10000>;

View file

@ -33,6 +33,14 @@ config NET_MEDIATEK_MT7621
endchoice
config NET_MEDIATEK_OFFLOAD
def_bool NET_MEDIATEK_SOC
depends on NET_MEDIATEK_MT7621
config NET_MEDIATEK_HW_QOS
def_bool NET_MEDIATEK_SOC
depends on NET_MEDIATEK_MT7623
config NET_MEDIATEK_MDIO
def_bool NET_MEDIATEK_SOC
depends on (NET_MEDIATEK_RT2880 || NET_MEDIATEK_RT3883 || NET_MEDIATEK_MT7620 || NET_MEDIATEK_MT7621)

View file

@ -8,6 +8,8 @@ mtk-eth-soc-$(CONFIG_NET_MEDIATEK_MDIO) += mdio.o
mtk-eth-soc-$(CONFIG_NET_MEDIATEK_MDIO_RT2880) += mdio_rt2880.o
mtk-eth-soc-$(CONFIG_NET_MEDIATEK_MDIO_MT7620) += mdio_mt7620.o
mtk-eth-soc-$(CONFIG_NET_MEDIATEK_OFFLOAD) += mtk_offload.o mtk_debugfs.o
mtk-eth-soc-$(CONFIG_NET_MEDIATEK_RT2880) += soc_rt2880.o
mtk-eth-soc-$(CONFIG_NET_MEDIATEK_RT3050) += soc_rt3050.o
mtk-eth-soc-$(CONFIG_NET_MEDIATEK_RT3883) += soc_rt3883.o

View file

@ -0,0 +1,115 @@
/* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; version 2 of the License
*
* 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.
*
* Copyright (C) 2014-2016 Sean Wang <sean.wang@mediatek.com>
* Copyright (C) 2016-2017 John Crispin <blogic@openwrt.org>
*/
#include "mtk_offload.h"
static const char *mtk_foe_entry_state_str[] = {
"INVALID",
"UNBIND",
"BIND",
"FIN"
};
static const char *mtk_foe_packet_type_str[] = {
"IPV4_HNAPT",
"IPV4_HNAT",
"IPV6_1T_ROUTE",
"IPV4_DSLITE",
"IPV6_3T_ROUTE",
"IPV6_5T_ROUTE",
"IPV6_6RD",
};
#define IPV4_HNAPT 0
#define IPV4_HNAT 1
#define IS_IPV4_HNAPT(x) (((x)->bfib1.pkt_type == IPV4_HNAPT) ? 1: 0)
struct mtk_eth *_eth;
#define es(entry) (mtk_foe_entry_state_str[entry->bfib1.state])
//#define ei(entry, end) (MTK_PPE_TBL_SZ - (int)(end - entry))
#define ei(entry, end) (MTK_PPE_ENTRY_CNT - (int)(end - entry))
#define pt(entry) (mtk_foe_packet_type_str[entry->ipv4_hnapt.bfib1.pkt_type])
static int mtk_ppe_debugfs_foe_show(struct seq_file *m, void *private)
{
struct mtk_eth *eth = _eth;
struct mtk_foe_entry *entry, *end;
int i = 0;
entry = eth->foe_table;
end = eth->foe_table + MTK_PPE_ENTRY_CNT;
while (entry < end) {
if (IS_IPV4_HNAPT(entry)) {
__be32 saddr = htonl(entry->ipv4_hnapt.sip);
__be32 daddr = htonl(entry->ipv4_hnapt.dip);
__be32 nsaddr = htonl(entry->ipv4_hnapt.new_sip);
__be32 ndaddr = htonl(entry->ipv4_hnapt.new_dip);
unsigned char h_dest[ETH_ALEN];
unsigned char h_source[ETH_ALEN];
*((u32*) h_source) = swab32(entry->ipv4_hnapt.smac_hi);
*((u16*) &h_source[4]) = swab16(entry->ipv4_hnapt.smac_lo);
*((u32*) h_dest) = swab32(entry->ipv4_hnapt.dmac_hi);
*((u16*) &h_dest[4]) = swab16(entry->ipv4_hnapt.dmac_lo);
seq_printf(m,
"(%x)0x%05x|state=%s|type=%s|"
"%pI4:%d->%pI4:%d=>%pI4:%d->%pI4:%d|%pM=>%pM|"
"etype=0x%04x|info1=0x%x|info2=0x%x|"
"vlan1=%d|vlan2=%d\n",
i,
ei(entry, end), es(entry), pt(entry),
&saddr, entry->ipv4_hnapt.sport,
&daddr, entry->ipv4_hnapt.dport,
&nsaddr, entry->ipv4_hnapt.new_sport,
&ndaddr, entry->ipv4_hnapt.new_dport, h_source,
h_dest, ntohs(entry->ipv4_hnapt.etype),
entry->ipv4_hnapt.info_blk1,
entry->ipv4_hnapt.info_blk2,
entry->ipv4_hnapt.vlan1,
entry->ipv4_hnapt.vlan2);
} else
seq_printf(m, "0x%05x state=%s\n",
ei(entry, end), es(entry));
entry++;
i++;
}
return 0;
}
static int mtk_ppe_debugfs_foe_open(struct inode *inode, struct file *file)
{
return single_open(file, mtk_ppe_debugfs_foe_show, file->private_data);
}
static const struct file_operations mtk_ppe_debugfs_foe_fops = {
.open = mtk_ppe_debugfs_foe_open,
.read = seq_read,
.llseek = seq_lseek,
.release = single_release,
};
int mtk_ppe_debugfs_init(struct mtk_eth *eth)
{
struct dentry *root;
_eth = eth;
root = debugfs_create_dir("mtk_ppe", NULL);
if (!root)
return -ENOMEM;
debugfs_create_file("all_entry", S_IRUGO, root, eth, &mtk_ppe_debugfs_foe_fops);
return 0;
}

View file

@ -30,6 +30,8 @@
#include <linux/tcp.h>
#include <linux/io.h>
#include <linux/bug.h>
#include <linux/netfilter.h>
#include <net/netfilter/nf_flow_table.h>
#include <asm/mach-ralink/ralink_regs.h>
@ -110,6 +112,18 @@ u32 fe_reg_r32(enum fe_reg reg)
return fe_r32(fe_reg_table[reg]);
}
void fe_m32(struct fe_priv *eth, u32 clear, u32 set, unsigned reg)
{
u32 val;
spin_lock(&eth->page_lock);
val = __raw_readl(fe_base + reg);
val &= ~clear;
val |= set;
__raw_writel(val, fe_base + reg);
spin_unlock(&eth->page_lock);
}
void fe_reset(u32 reset_bits)
{
u32 t;
@ -865,10 +879,14 @@ static int fe_poll_rx(struct napi_struct *napi, int budget,
skb_checksum_none_assert(skb);
skb->protocol = eth_type_trans(skb, netdev);
stats->rx_packets++;
stats->rx_bytes += pktlen;
if (mtk_offload_check_rx(priv, skb, trxd.rxd4) == 0) {
stats->rx_packets++;
stats->rx_bytes += pktlen;
napi_gro_receive(napi, skb);
napi_gro_receive(napi, skb);
} else {
dev_kfree_skb(skb);
}
ring->rx_data[idx] = new_data;
rxd->rxd1 = (unsigned int)dma_addr;
@ -1207,6 +1225,9 @@ static int fe_open(struct net_device *dev)
napi_enable(&priv->rx_napi);
fe_int_enable(priv->soc->tx_int | priv->soc->rx_int);
netif_start_queue(dev);
#ifdef CONFIG_NET_MEDIATEK_OFFLOAD
mtk_ppe_probe(priv);
#endif
return 0;
}
@ -1243,6 +1264,10 @@ static int fe_stop(struct net_device *dev)
fe_free_dma(priv);
#ifdef CONFIG_NET_MEDIATEK_OFFLOAD
mtk_ppe_remove(priv);
#endif
return 0;
}
@ -1390,6 +1415,23 @@ static int fe_change_mtu(struct net_device *dev, int new_mtu)
return fe_open(dev);
}
#ifdef CONFIG_NET_MEDIATEK_OFFLOAD
static int
fe_flow_offload(enum flow_offload_type type, struct flow_offload *flow,
struct flow_offload_hw_path *src,
struct flow_offload_hw_path *dest)
{
struct fe_priv *priv;
if (src->dev != dest->dev)
return -EINVAL;
priv = netdev_priv(src->dev);
return mtk_flow_offload(priv, type, flow, src, dest);
}
#endif
static const struct net_device_ops fe_netdev_ops = {
.ndo_init = fe_init,
.ndo_uninit = fe_uninit,
@ -1407,6 +1449,9 @@ static const struct net_device_ops fe_netdev_ops = {
#ifdef CONFIG_NET_POLL_CONTROLLER
.ndo_poll_controller = fe_poll_controller,
#endif
#ifdef CONFIG_NET_MEDIATEK_OFFLOAD
.ndo_flow_offload = fe_flow_offload,
#endif
};
static void fe_reset_pending(struct fe_priv *priv)

View file

@ -499,11 +499,17 @@ struct fe_priv {
unsigned long vlan_map;
struct work_struct pending_work;
DECLARE_BITMAP(pending_flags, FE_FLAG_MAX);
struct reset_control *rst_ppe;
struct mtk_foe_entry *foe_table;
dma_addr_t foe_table_phys;
struct flow_offload __rcu **foe_flow_table;
};
extern const struct of_device_id of_fe_match[];
void fe_w32(u32 val, unsigned reg);
void fe_m32(struct fe_priv *priv, u32 clear, u32 set, unsigned reg);
u32 fe_r32(unsigned reg);
int fe_set_clock_cycle(struct fe_priv *priv);
@ -520,4 +526,14 @@ static inline void *priv_netdev(struct fe_priv *priv)
return (char *)priv - ALIGN(sizeof(struct net_device), NETDEV_ALIGN);
}
int mtk_ppe_probe(struct fe_priv *eth);
void mtk_ppe_remove(struct fe_priv *eth);
int mtk_flow_offload(struct fe_priv *eth,
enum flow_offload_type type,
struct flow_offload *flow,
struct flow_offload_hw_path *src,
struct flow_offload_hw_path *dest);
int mtk_offload_check_rx(struct fe_priv *eth, struct sk_buff *skb, u32 rxd4);
#endif /* FE_ETH_H */

View file

@ -0,0 +1,526 @@
/* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; version 2 of the License
*
* 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.
*
* Copyright (C) 2018 John Crispin <john@phrozen.org>
*/
#include "mtk_offload.h"
#define INVALID 0
#define UNBIND 1
#define BIND 2
#define FIN 3
#define IPV4_HNAPT 0
#define IPV4_HNAT 1
static u32
mtk_flow_hash_v4(struct flow_offload_tuple *tuple)
{
u32 ports = ntohs(tuple->src_port) << 16 | ntohs(tuple->dst_port);
u32 src = ntohl(tuple->dst_v4.s_addr);
u32 dst = ntohl(tuple->src_v4.s_addr);
u32 hash = (ports & src) | ((~ports) & dst);
u32 hash_23_0 = hash & 0xffffff;
u32 hash_31_24 = hash & 0xff000000;
hash = ports ^ src ^ dst ^ ((hash_23_0 << 8) | (hash_31_24 >> 24));
hash = ((hash & 0xffff0000) >> 16 ) ^ (hash & 0xfffff);
hash &= 0x7ff;
hash *= 2;;
return hash;
}
static int
mtk_foe_prepare_v4(struct mtk_foe_entry *entry,
struct flow_offload_tuple *tuple,
struct flow_offload_tuple *dest_tuple,
struct flow_offload_hw_path *src,
struct flow_offload_hw_path *dest)
{
int is_mcast = !!is_multicast_ether_addr(dest->eth_dest);
if (tuple->l4proto == IPPROTO_UDP)
entry->ipv4_hnapt.bfib1.udp = 1;
entry->ipv4_hnapt.etype = htons(ETH_P_IP);
entry->ipv4_hnapt.bfib1.pkt_type = IPV4_HNAPT;
entry->ipv4_hnapt.iblk2.fqos = 0;
entry->ipv4_hnapt.bfib1.ttl = 1;
entry->ipv4_hnapt.bfib1.cah = 1;
entry->ipv4_hnapt.bfib1.ka = 1;
entry->ipv4_hnapt.iblk2.mcast = is_mcast;
entry->ipv4_hnapt.iblk2.dscp = 0;
entry->ipv4_hnapt.iblk2.port_mg = 0x3f;
entry->ipv4_hnapt.iblk2.port_ag = 0x1f;
#ifdef CONFIG_NET_MEDIATEK_HW_QOS
entry->ipv4_hnapt.iblk2.qid = 1;
entry->ipv4_hnapt.iblk2.fqos = 1;
#endif
#ifdef CONFIG_RALINK
entry->ipv4_hnapt.iblk2.dp = 1;
if ((dest->flags & FLOW_OFFLOAD_PATH_VLAN) && (dest->vlan_id > 1))
entry->ipv4_hnapt.iblk2.qid += 8;
#else
entry->ipv4_hnapt.iblk2.dp = (dest->dev->name[3] - '0') + 1;
#endif
entry->ipv4_hnapt.sip = ntohl(tuple->src_v4.s_addr);
entry->ipv4_hnapt.dip = ntohl(tuple->dst_v4.s_addr);
entry->ipv4_hnapt.sport = ntohs(tuple->src_port);
entry->ipv4_hnapt.dport = ntohs(tuple->dst_port);
entry->ipv4_hnapt.new_sip = ntohl(dest_tuple->dst_v4.s_addr);
entry->ipv4_hnapt.new_dip = ntohl(dest_tuple->src_v4.s_addr);
entry->ipv4_hnapt.new_sport = ntohs(dest_tuple->dst_port);
entry->ipv4_hnapt.new_dport = ntohs(dest_tuple->src_port);
entry->bfib1.state = BIND;
if (dest->flags & FLOW_OFFLOAD_PATH_PPPOE) {
entry->bfib1.psn = 1;
entry->ipv4_hnapt.etype = htons(ETH_P_PPP_SES);
entry->ipv4_hnapt.pppoe_id = dest->pppoe_sid;
}
if (dest->flags & FLOW_OFFLOAD_PATH_VLAN) {
entry->ipv4_hnapt.vlan1 = dest->vlan_id;
entry->bfib1.vlan_layer = 1;
switch (dest->vlan_proto) {
case htons(ETH_P_8021Q):
entry->ipv4_hnapt.bfib1.vpm = 1;
break;
case htons(ETH_P_8021AD):
entry->ipv4_hnapt.bfib1.vpm = 2;
break;
default:
return -EINVAL;
}
}
return 0;
}
static void
mtk_foe_set_mac(struct mtk_foe_entry *entry, u8 *smac, u8 *dmac)
{
entry->ipv4_hnapt.dmac_hi = swab32(*((u32*) dmac));
entry->ipv4_hnapt.dmac_lo = swab16(*((u16*) &dmac[4]));
entry->ipv4_hnapt.smac_hi = swab32(*((u32*) smac));
entry->ipv4_hnapt.smac_lo = swab16(*((u16*) &smac[4]));
}
static void
mtk_foe_write(struct mtk_eth *eth, u32 hash,
struct mtk_foe_entry *entry)
{
struct mtk_foe_entry *table = (struct mtk_foe_entry *)eth->foe_table;
memcpy(&table[hash], entry, sizeof(*entry));
}
int mtk_flow_offload(struct mtk_eth *eth,
enum flow_offload_type type,
struct flow_offload *flow,
struct flow_offload_hw_path *src,
struct flow_offload_hw_path *dest)
{
struct flow_offload_tuple *otuple = &flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple;
struct flow_offload_tuple *rtuple = &flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple;
u32 time_stamp = mtk_r32(eth, 0x0010) & (0x7fff);
u32 ohash, rhash;
struct mtk_foe_entry orig = {
.bfib1.time_stamp = time_stamp,
.bfib1.psn = 0,
};
struct mtk_foe_entry reply = {
.bfib1.time_stamp = time_stamp,
.bfib1.psn = 0,
};
if (otuple->l4proto != IPPROTO_TCP && otuple->l4proto != IPPROTO_UDP)
return -EINVAL;
switch (otuple->l3proto) {
case AF_INET:
if (mtk_foe_prepare_v4(&orig, otuple, rtuple, src, dest) ||
mtk_foe_prepare_v4(&reply, rtuple, otuple, dest, src))
return -EINVAL;
ohash = mtk_flow_hash_v4(otuple);
rhash = mtk_flow_hash_v4(rtuple);
break;
case AF_INET6:
return -EINVAL;
default:
return -EINVAL;
}
if (type == FLOW_OFFLOAD_DEL) {
orig.bfib1.state = INVALID;
reply.bfib1.state = INVALID;
flow = NULL;
goto write;
}
mtk_foe_set_mac(&orig, dest->eth_src, dest->eth_dest);
mtk_foe_set_mac(&reply, src->eth_src, src->eth_dest);
write:
mtk_foe_write(eth, ohash, &orig);
mtk_foe_write(eth, rhash, &reply);
rcu_assign_pointer(eth->foe_flow_table[ohash], flow);
rcu_assign_pointer(eth->foe_flow_table[rhash], flow);
if (type == FLOW_OFFLOAD_DEL)
synchronize_rcu();
return 0;
}
#ifdef CONFIG_NET_MEDIATEK_HW_QOS
#define QDMA_TX_SCH_TX 0x1a14
static void mtk_ppe_scheduler(struct mtk_eth *eth, int id, u32 rate)
{
int exp = 0, shift = 0;
u32 reg = mtk_r32(eth, QDMA_TX_SCH_TX);
u32 val = 0;
if (rate)
val = BIT(11);
while (rate > 127) {
rate /= 10;
exp++;
}
val |= (rate & 0x7f) << 4;
val |= exp & 0xf;
if (id)
shift = 16;
reg &= ~(0xffff << shift);
reg |= val << shift;
mtk_w32(eth, val, QDMA_TX_SCH_TX);
}
#define QTX_CFG(x) (0x1800 + (x * 0x10))
#define QTX_SCH(x) (0x1804 + (x * 0x10))
static void mtk_ppe_queue(struct mtk_eth *eth, int id, int sched, int weight, int resv, u32 min_rate, u32 max_rate)
{
int max_exp = 0, min_exp = 0;
u32 reg;
if (id >= 16)
return;
reg = mtk_r32(eth, QTX_SCH(id));
reg &= 0x70000000;
if (sched)
reg |= BIT(31);
if (min_rate)
reg |= BIT(27);
if (max_rate)
reg |= BIT(11);
while (max_rate > 127) {
max_rate /= 10;
max_exp++;
}
while (min_rate > 127) {
min_rate /= 10;
min_exp++;
}
reg |= (min_rate & 0x7f) << 20;
reg |= (min_exp & 0xf) << 16;
reg |= (weight & 0xf) << 12;
reg |= (max_rate & 0x7f) << 4;
reg |= max_exp & 0xf;
mtk_w32(eth, reg, QTX_SCH(id));
resv &= 0xff;
reg = mtk_r32(eth, QTX_CFG(id));
reg &= 0xffff0000;
reg |= (resv << 8) | resv;
mtk_w32(eth, reg, QTX_CFG(id));
}
#endif
static int mtk_init_foe_table(struct mtk_eth *eth)
{
if (eth->foe_table)
return 0;
eth->foe_flow_table = devm_kcalloc(eth->dev, MTK_PPE_ENTRY_CNT,
sizeof(*eth->foe_flow_table),
GFP_KERNEL);
if (!eth->foe_flow_table)
return -EINVAL;
/* map the FOE table */
eth->foe_table = dmam_alloc_coherent(eth->dev, MTK_PPE_TBL_SZ,
&eth->foe_table_phys, GFP_KERNEL);
if (!eth->foe_table) {
dev_err(eth->dev, "failed to allocate foe table\n");
kfree(eth->foe_flow_table);
return -ENOMEM;
}
return 0;
}
static int mtk_ppe_start(struct mtk_eth *eth)
{
int ret;
ret = mtk_init_foe_table(eth);
if (ret)
return ret;
/* tell the PPE about the tables base address */
mtk_w32(eth, eth->foe_table_phys, MTK_REG_PPE_TB_BASE);
/* flush the table */
memset(eth->foe_table, 0, MTK_PPE_TBL_SZ);
/* setup hashing */
mtk_m32(eth,
MTK_PPE_TB_CFG_HASH_MODE_MASK | MTK_PPE_TB_CFG_TBL_SZ_MASK,
MTK_PPE_TB_CFG_HASH_MODE1 | MTK_PPE_TB_CFG_TBL_SZ_4K,
MTK_REG_PPE_TB_CFG);
/* set the default hashing seed */
mtk_w32(eth, MTK_PPE_HASH_SEED, MTK_REG_PPE_HASH_SEED);
/* each foe entry is 64bytes and is setup by cpu forwarding*/
mtk_m32(eth, MTK_PPE_CAH_CTRL_X_MODE | MTK_PPE_TB_CFG_ENTRY_SZ_MASK |
MTK_PPE_TB_CFG_SMA_MASK,
MTK_PPE_TB_CFG_ENTRY_SZ_64B | MTK_PPE_TB_CFG_SMA_FWD_CPU,
MTK_REG_PPE_TB_CFG);
/* set ip proto */
mtk_w32(eth, 0xFFFFFFFF, MTK_REG_PPE_IP_PROT_CHK);
/* setup caching */
mtk_m32(eth, 0, MTK_PPE_CAH_CTRL_X_MODE, MTK_REG_PPE_CAH_CTRL);
mtk_m32(eth, MTK_PPE_CAH_CTRL_X_MODE, MTK_PPE_CAH_CTRL_EN,
MTK_REG_PPE_CAH_CTRL);
/* enable FOE */
mtk_m32(eth, 0, MTK_PPE_FLOW_CFG_IPV4_NAT_FRAG_EN |
MTK_PPE_FLOW_CFG_IPV4_NAPT_EN | MTK_PPE_FLOW_CFG_IPV4_NAT_EN |
MTK_PPE_FLOW_CFG_IPV4_GREK_EN,
MTK_REG_PPE_FLOW_CFG);
/* setup flow entry un/bind aging */
mtk_m32(eth, 0,
MTK_PPE_TB_CFG_UNBD_AGE | MTK_PPE_TB_CFG_NTU_AGE |
MTK_PPE_TB_CFG_FIN_AGE | MTK_PPE_TB_CFG_UDP_AGE |
MTK_PPE_TB_CFG_TCP_AGE,
MTK_REG_PPE_TB_CFG);
mtk_m32(eth, MTK_PPE_UNB_AGE_MNP_MASK | MTK_PPE_UNB_AGE_DLTA_MASK,
MTK_PPE_UNB_AGE_MNP | MTK_PPE_UNB_AGE_DLTA,
MTK_REG_PPE_UNB_AGE);
mtk_m32(eth, MTK_PPE_BND_AGE0_NTU_DLTA_MASK |
MTK_PPE_BND_AGE0_UDP_DLTA_MASK,
MTK_PPE_BND_AGE0_NTU_DLTA | MTK_PPE_BND_AGE0_UDP_DLTA,
MTK_REG_PPE_BND_AGE0);
mtk_m32(eth, MTK_PPE_BND_AGE1_FIN_DLTA_MASK |
MTK_PPE_BND_AGE1_TCP_DLTA_MASK,
MTK_PPE_BND_AGE1_FIN_DLTA | MTK_PPE_BND_AGE1_TCP_DLTA,
MTK_REG_PPE_BND_AGE1);
/* setup flow entry keep alive */
mtk_m32(eth, MTK_PPE_TB_CFG_KA_MASK, MTK_PPE_TB_CFG_KA,
MTK_REG_PPE_TB_CFG);
mtk_w32(eth, MTK_PPE_KA_UDP | MTK_PPE_KA_TCP | MTK_PPE_KA_T, MTK_REG_PPE_KA);
/* setup flow entry rate limit */
mtk_w32(eth, (0x3fff << 16) | 0x3fff, MTK_REG_PPE_BIND_LMT_0);
mtk_w32(eth, MTK_PPE_NTU_KA | 0x3fff, MTK_REG_PPE_BIND_LMT_1);
mtk_m32(eth, MTK_PPE_BNDR_RATE_MASK, 1, MTK_REG_PPE_BNDR);
/* enable the PPE */
mtk_m32(eth, 0, MTK_PPE_GLO_CFG_EN, MTK_REG_PPE_GLO_CFG);
#ifdef CONFIG_RALINK
/* set the default forwarding port to QDMA */
mtk_w32(eth, 0x0, MTK_REG_PPE_DFT_CPORT);
#else
/* set the default forwarding port to QDMA */
mtk_w32(eth, 0x55555555, MTK_REG_PPE_DFT_CPORT);
#endif
/* drop packets with TTL=0 */
mtk_m32(eth, 0, MTK_PPE_GLO_CFG_TTL0_DROP, MTK_REG_PPE_GLO_CFG);
/* send all traffic from gmac to the ppe */
mtk_m32(eth, 0xffff, 0x4444, MTK_GDMA_FWD_CFG(0));
mtk_m32(eth, 0xffff, 0x4444, MTK_GDMA_FWD_CFG(1));
dev_info(eth->dev, "PPE started\n");
#ifdef CONFIG_NET_MEDIATEK_HW_QOS
mtk_ppe_scheduler(eth, 0, 500000);
mtk_ppe_scheduler(eth, 1, 500000);
mtk_ppe_queue(eth, 0, 0, 7, 32, 250000, 0);
mtk_ppe_queue(eth, 1, 0, 7, 32, 250000, 0);
mtk_ppe_queue(eth, 8, 1, 7, 32, 250000, 0);
mtk_ppe_queue(eth, 9, 1, 7, 32, 250000, 0);
#endif
return 0;
}
static int mtk_ppe_busy_wait(struct mtk_eth *eth)
{
unsigned long t_start = jiffies;
u32 r = 0;
while (1) {
r = mtk_r32(eth, MTK_REG_PPE_GLO_CFG);
if (!(r & MTK_PPE_GLO_CFG_BUSY))
return 0;
if (time_after(jiffies, t_start + HZ))
break;
usleep_range(10, 20);
}
dev_err(eth->dev, "ppe: table busy timeout - resetting\n");
reset_control_reset(eth->rst_ppe);
return -ETIMEDOUT;
}
static int mtk_ppe_stop(struct mtk_eth *eth)
{
u32 r1 = 0, r2 = 0;
int i;
/* discard all traffic while we disable the PPE */
mtk_m32(eth, 0xffff, 0x7777, MTK_GDMA_FWD_CFG(0));
mtk_m32(eth, 0xffff, 0x7777, MTK_GDMA_FWD_CFG(1));
if (mtk_ppe_busy_wait(eth))
return -ETIMEDOUT;
/* invalidate all flow table entries */
for (i = 0; i < MTK_PPE_ENTRY_CNT; i++)
eth->foe_table[i].bfib1.state = FOE_STATE_INVALID;
/* disable caching */
mtk_m32(eth, 0, MTK_PPE_CAH_CTRL_X_MODE, MTK_REG_PPE_CAH_CTRL);
mtk_m32(eth, MTK_PPE_CAH_CTRL_X_MODE | MTK_PPE_CAH_CTRL_EN, 0,
MTK_REG_PPE_CAH_CTRL);
/* flush cache has to be ahead of hnat diable --*/
mtk_m32(eth, MTK_PPE_GLO_CFG_EN, 0, MTK_REG_PPE_GLO_CFG);
/* disable FOE */
mtk_m32(eth,
MTK_PPE_FLOW_CFG_IPV4_NAT_FRAG_EN |
MTK_PPE_FLOW_CFG_IPV4_NAPT_EN | MTK_PPE_FLOW_CFG_IPV4_NAT_EN |
MTK_PPE_FLOW_CFG_FUC_FOE | MTK_PPE_FLOW_CFG_FMC_FOE,
0, MTK_REG_PPE_FLOW_CFG);
/* disable FOE aging */
mtk_m32(eth, 0,
MTK_PPE_TB_CFG_FIN_AGE | MTK_PPE_TB_CFG_UDP_AGE |
MTK_PPE_TB_CFG_TCP_AGE | MTK_PPE_TB_CFG_UNBD_AGE |
MTK_PPE_TB_CFG_NTU_AGE, MTK_REG_PPE_TB_CFG);
r1 = mtk_r32(eth, 0x100);
r2 = mtk_r32(eth, 0x10c);
dev_info(eth->dev, "0x100 = 0x%x, 0x10c = 0x%x\n", r1, r2);
if (((r1 & 0xff00) >> 0x8) >= (r1 & 0xff) ||
((r1 & 0xff00) >> 0x8) >= (r2 & 0xff)) {
dev_info(eth->dev, "reset pse\n");
mtk_w32(eth, 0x1, 0x4);
}
/* set the foe entry base address to 0 */
mtk_w32(eth, 0, MTK_REG_PPE_TB_BASE);
if (mtk_ppe_busy_wait(eth))
return -ETIMEDOUT;
/* send all traffic back to the DMA engine */
#ifdef CONFIG_RALINK
mtk_m32(eth, 0xffff, 0x0, MTK_GDMA_FWD_CFG(0));
mtk_m32(eth, 0xffff, 0x0, MTK_GDMA_FWD_CFG(1));
#else
mtk_m32(eth, 0xffff, 0x5555, MTK_GDMA_FWD_CFG(0));
mtk_m32(eth, 0xffff, 0x5555, MTK_GDMA_FWD_CFG(1));
#endif
return 0;
}
static void mtk_offload_keepalive(struct fe_priv *eth, unsigned int hash)
{
struct flow_offload *flow;
rcu_read_lock();
flow = rcu_dereference(eth->foe_flow_table[hash]);
if (flow)
flow->timeout = jiffies + 30 * HZ;
rcu_read_unlock();
}
int mtk_offload_check_rx(struct fe_priv *eth, struct sk_buff *skb, u32 rxd4)
{
unsigned int hash;
switch (FIELD_GET(MTK_RXD4_CPU_REASON, rxd4)) {
case MTK_CPU_REASON_KEEPALIVE_UC_OLD_HDR:
case MTK_CPU_REASON_KEEPALIVE_MC_NEW_HDR:
case MTK_CPU_REASON_KEEPALIVE_DUP_OLD_HDR:
hash = FIELD_GET(MTK_RXD4_FOE_ENTRY, rxd4);
mtk_offload_keepalive(eth, hash);
return -1;
case MTK_CPU_REASON_PACKET_SAMPLING:
return -1;
default:
return 0;
}
}
int mtk_ppe_probe(struct mtk_eth *eth)
{
int err;
err = mtk_ppe_start(eth);
if (err)
return err;
err = mtk_ppe_debugfs_init(eth);
if (err)
return err;
return 0;
}
void mtk_ppe_remove(struct mtk_eth *eth)
{
mtk_ppe_stop(eth);
}

View file

@ -0,0 +1,260 @@
/* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; version 2 of the License
*
* 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.
*
* Copyright (C) 2014-2016 Sean Wang <sean.wang@mediatek.com>
* Copyright (C) 2016-2017 John Crispin <blogic@openwrt.org>
*/
#include <linux/dma-mapping.h>
#include <linux/delay.h>
#include <linux/if.h>
#include <linux/io.h>
#include <linux/module.h>
#include <linux/of_device.h>
#include <linux/platform_device.h>
#include <linux/reset.h>
#include <linux/netfilter.h>
#include <linux/netdevice.h>
#include <net/netfilter/nf_flow_table.h>
#include <linux/debugfs.h>
#include <linux/etherdevice.h>
#include <linux/bitfield.h>
#include "mtk_eth_soc.h"
#ifdef CONFIG_RALINK
/* ramips compat */
#define mtk_eth fe_priv
#define MTK_GDMA_FWD_CFG(x) (0x500 + (x * 0x1000))
#define mtk_m32 fe_m32
static inline u32
mtk_r32(struct mtk_eth *eth, u32 reg)
{
return fe_r32(reg);
}
static inline void
mtk_w32(struct mtk_eth *eth, u32 val, u32 reg)
{
fe_w32(val, reg);
}
#endif
#define MTK_REG_PPE_GLO_CFG 0xe00
#define MTK_PPE_GLO_CFG_BUSY BIT(31)
#define MTK_PPE_GLO_CFG_TTL0_DROP BIT(4)
#define MTK_PPE_GLO_CFG_EN BIT(0)
#define MTK_REG_PPE_FLOW_CFG 0xe04
#define MTK_PPE_FLOW_CFG_IPV4_GREK_EN BIT(19)
#define MTK_PPE_FLOW_CFG_IPV4_NAT_FRAG_EN BIT(17)
#define MTK_PPE_FLOW_CFG_IPV4_NAPT_EN BIT(13)
#define MTK_PPE_FLOW_CFG_IPV4_NAT_EN BIT(12)
#define MTK_PPE_FLOW_CFG_FUC_FOE BIT(2)
#define MTK_PPE_FLOW_CFG_FMC_FOE BIT(1)
#define MTK_REG_PPE_IP_PROT_CHK 0xe08
#define MTK_REG_PPE_TB_BASE 0xe20
#define MTK_REG_PPE_BNDR 0xe28
#define MTK_PPE_BNDR_RATE_MASK 0xffff
#define MTK_REG_PPE_BIND_LMT_0 0xe2C
#define MTK_REG_PPE_BIND_LMT_1 0xe30
#define MTK_PPE_NTU_KA BIT(16)
#define MTK_REG_PPE_KA 0xe34
#define MTK_PPE_KA_T BIT(0)
#define MTK_PPE_KA_TCP BIT(16)
#define MTK_PPE_KA_UDP BIT(24)
#define MTK_REG_PPE_UNB_AGE 0xe38
#define MTK_PPE_UNB_AGE_MNP_MASK (0xffff << 16)
#define MTK_PPE_UNB_AGE_MNP (1000 << 16)
#define MTK_PPE_UNB_AGE_DLTA_MASK 0xff
#define MTK_PPE_UNB_AGE_DLTA 3
#define MTK_REG_PPE_BND_AGE0 0xe3c
#define MTK_PPE_BND_AGE0_NTU_DLTA_MASK (0xffff << 16)
#define MTK_PPE_BND_AGE0_NTU_DLTA (5 << 16)
#define MTK_PPE_BND_AGE0_UDP_DLTA_MASK 0xffff
#define MTK_PPE_BND_AGE0_UDP_DLTA 5
#define MTK_REG_PPE_BND_AGE1 0xe40
#define MTK_PPE_BND_AGE1_FIN_DLTA_MASK (0xffff << 16)
#define MTK_PPE_BND_AGE1_FIN_DLTA (5 << 16)
#define MTK_PPE_BND_AGE1_TCP_DLTA_MASK 0xffff
#define MTK_PPE_BND_AGE1_TCP_DLTA 5
#define MTK_REG_PPE_DFT_CPORT 0xe48
#define MTK_REG_PPE_TB_CFG 0xe1c
#define MTK_PPE_TB_CFG_X_MODE_MASK (3 << 18)
#define MTK_PPE_TB_CFG_HASH_MODE1 BIT(14)
#define MTK_PPE_TB_CFG_HASH_MODE_MASK (0x3 << 14)
#define MTK_PPE_TB_CFG_KA (3 << 12)
#define MTK_PPE_TB_CFG_KA_MASK (0x3 << 12)
#define MTK_PPE_TB_CFG_FIN_AGE BIT(11)
#define MTK_PPE_TB_CFG_UDP_AGE BIT(10)
#define MTK_PPE_TB_CFG_TCP_AGE BIT(9)
#define MTK_PPE_TB_CFG_UNBD_AGE BIT(8)
#define MTK_PPE_TB_CFG_NTU_AGE BIT(7)
#define MTK_PPE_TB_CFG_SMA_FWD_CPU (0x3 << 4)
#define MTK_PPE_TB_CFG_SMA_MASK (0x3 << 4)
#define MTK_PPE_TB_CFG_ENTRY_SZ_64B 0
#define MTK_PPE_TB_CFG_ENTRY_SZ_MASK BIT(3)
#define MTK_PPE_TB_CFG_TBL_SZ_4K 2
#define MTK_PPE_TB_CFG_TBL_SZ_MASK 0x7
#define MTK_REG_PPE_HASH_SEED 0xe44
#define MTK_PPE_HASH_SEED 0x12345678
#define MTK_REG_PPE_CAH_CTRL 0xf20
#define MTK_PPE_CAH_CTRL_X_MODE BIT(9)
#define MTK_PPE_CAH_CTRL_EN BIT(0)
struct mtk_foe_unbind_info_blk {
u32 time_stamp:8;
u32 pcnt:16; /* packet count */
u32 preb:1;
u32 pkt_type:3;
u32 state:2;
u32 udp:1;
u32 sta:1; /* static entry */
} __attribute__ ((packed));
struct mtk_foe_bind_info_blk {
u32 time_stamp:15;
u32 ka:1; /* keep alive */
u32 vlan_layer:3;
u32 psn:1; /* egress packet has PPPoE session */
#ifdef CONFIG_RALINK
u32 vpm:2; /* 0:ethertype remark, 1:0x8100(CR default) */
#else
u32 vpm:1; /* 0:ethertype remark, 1:0x8100(CR default) */
u32 ps:1; /* packet sampling */
#endif
u32 cah:1; /* cacheable flag */
u32 rmt:1; /* remove tunnel ip header (6rd/dslite only) */
u32 ttl:1;
u32 pkt_type:3;
u32 state:2;
u32 udp:1;
u32 sta:1; /* static entry */
} __attribute__ ((packed));
struct mtk_foe_info_blk2 {
u32 qid:4; /* QID in Qos Port */
u32 fqos:1; /* force to PSE QoS port */
u32 dp:3; /* force to PSE port x
0:PSE,1:GSW, 2:GMAC,4:PPE,5:QDMA,7=DROP */
u32 mcast:1; /* multicast this packet to CPU */
u32 pcpl:1; /* OSBN */
u32 mlen:1; /* 0:post 1:pre packet length in meter */
u32 alen:1; /* 0:post 1:pre packet length in accounting */
u32 port_mg:6; /* port meter group */
u32 port_ag:6; /* port account group */
u32 dscp:8; /* DSCP value */
} __attribute__ ((packed));
struct mtk_foe_ipv4_hnapt {
union {
struct mtk_foe_bind_info_blk bfib1;
struct mtk_foe_unbind_info_blk udib1;
u32 info_blk1;
};
u32 sip;
u32 dip;
u16 dport;
u16 sport;
union {
struct mtk_foe_info_blk2 iblk2;
u32 info_blk2;
};
u32 new_sip;
u32 new_dip;
u16 new_dport;
u16 new_sport;
u32 resv1;
u32 resv2;
u32 resv3:26;
u32 act_dp:6; /* UDF */
u16 vlan1;
u16 etype;
u32 dmac_hi;
u16 vlan2;
u16 dmac_lo;
u32 smac_hi;
u16 pppoe_id;
u16 smac_lo;
} __attribute__ ((packed));
struct mtk_foe_entry {
union {
struct mtk_foe_unbind_info_blk udib1;
struct mtk_foe_bind_info_blk bfib1;
struct mtk_foe_ipv4_hnapt ipv4_hnapt;
};
};
enum mtk_foe_entry_state {
FOE_STATE_INVALID = 0,
FOE_STATE_UNBIND = 1,
FOE_STATE_BIND = 2,
FOE_STATE_FIN = 3
};
#define MTK_RXD4_FOE_ENTRY GENMASK(13, 0)
#define MTK_RXD4_CPU_REASON GENMASK(18, 14)
#define MTK_RXD4_SRC_PORT GENMASK(21, 19)
#define MTK_RXD4_ALG GENMASK(31, 22)
enum mtk_foe_cpu_reason {
MTK_CPU_REASON_TTL_EXCEEDED = 0x02,
MTK_CPU_REASON_OPTION_HEADER = 0x03,
MTK_CPU_REASON_NO_FLOW = 0x07,
MTK_CPU_REASON_IPV4_FRAG = 0x08,
MTK_CPU_REASON_IPV4_DSLITE_FRAG = 0x09,
MTK_CPU_REASON_IPV4_DSLITE_NO_TCP_UDP = 0x0a,
MTK_CPU_REASON_IPV6_6RD_NO_TCP_UDP = 0x0b,
MTK_CPU_REASON_TCP_FIN_SYN_RST = 0x0c,
MTK_CPU_REASON_UN_HIT = 0x0d,
MTK_CPU_REASON_HIT_UNBIND = 0x0e,
MTK_CPU_REASON_HIT_UNBIND_RATE_REACHED = 0x0f,
MTK_CPU_REASON_HIT_BIND_TCP_FIN = 0x10,
MTK_CPU_REASON_HIT_TTL_1 = 0x11,
MTK_CPU_REASON_HIT_BIND_VLAN_VIOLATION = 0x12,
MTK_CPU_REASON_KEEPALIVE_UC_OLD_HDR = 0x13,
MTK_CPU_REASON_KEEPALIVE_MC_NEW_HDR = 0x14,
MTK_CPU_REASON_KEEPALIVE_DUP_OLD_HDR = 0x15,
MTK_CPU_REASON_HIT_BIND_FORCE_CPU = 0x16,
MTK_CPU_REASON_TUNNEL_OPTION_HEADER = 0x17,
MTK_CPU_REASON_MULTICAST_TO_CPU = 0x18,
MTK_CPU_REASON_MULTICAST_TO_GMAC1_CPU = 0x19,
MTK_CPU_REASON_HIT_PRE_BIND = 0x1a,
MTK_CPU_REASON_PACKET_SAMPLING = 0x1b,
MTK_CPU_REASON_EXCEED_MTU = 0x1c,
MTK_CPU_REASON_PPE_BYPASS = 0x1e,
MTK_CPU_REASON_INVALID = 0x1f,
};
/* our table size is 4K */
#define MTK_PPE_ENTRY_CNT 0x1000
#define MTK_PPE_TBL_SZ \
(MTK_PPE_ENTRY_CNT * sizeof(struct mtk_foe_entry))
int mtk_ppe_debugfs_init(struct mtk_eth *eth);