You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
openwrt/target/linux/ramips/files/drivers/net/ethernet/ralink/mtk_offload.c

541 lines
14 KiB
C

/* 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_RALINK_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 int
mtk_check_entry_available(struct mtk_eth *eth, u32 hash)
{
struct mtk_foe_entry entry = ((struct mtk_foe_entry *)eth->foe_table)[hash];
return (entry.bfib1.state == BIND)? 0:1;
}
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;
if (type == FLOW_OFFLOAD_DEL) {
flow = NULL;
synchronize_rcu();
return 0;
}
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;
}
/* Two-way hash: when hash collision occurs, the hash value will be shifted to the next position. */
if (!mtk_check_entry_available(eth, ohash)){
if (!mtk_check_entry_available(eth, ohash + 1))
return -EINVAL;
ohash += 1;
}
if (!mtk_check_entry_available(eth, rhash)){
if (!mtk_check_entry_available(eth, rhash + 1))
return -EINVAL;
rhash += 1;
}
mtk_foe_set_mac(&orig, dest->eth_src, dest->eth_dest);
mtk_foe_set_mac(&reply, src->eth_src, src->eth_dest);
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);
return 0;
}
#ifdef CONFIG_NET_RALINK_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
/* allow packets with TTL=0 */
mtk_m32(eth, MTK_PPE_GLO_CFG_TTL0_DROP, 0, 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_RALINK_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);
}