2024-05-11 12:28:18

by Elad Yifee

[permalink] [raw]
Subject: [PATCH net-next v4] net: ethernet: mtk_eth_soc: ppe: add support for multiple PPEs

Add the missing pieces to allow multiple PPEs units, one for each GMAC.
mtk_gdm_config has been modified to work on targted mac ID,
the inner loop moved outside of the function to allow unrelated
operations like setting the MAC's PPE index.

Signed-off-by: Elad Yifee <[email protected]>
---
v4:
- applied changes suggested by Jakub Kicinski
- modified flow_offload_replace to get the correct PPE index from flow
- add gdma_to_ppe[x] registers to mtk_reg_map instead of using defines
v3: applied changes suggested by Daniel Golle
v2: fixed CI warnings
---
drivers/net/ethernet/mediatek/mtk_eth_soc.c | 106 +++++++++++-------
drivers/net/ethernet/mediatek/mtk_eth_soc.h | 15 ++-
.../net/ethernet/mediatek/mtk_ppe_offload.c | 16 ++-
3 files changed, 91 insertions(+), 46 deletions(-)

diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
index 179c0230655a..67e19bd25f7a 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
@@ -80,7 +80,9 @@ static const struct mtk_reg_map mtk_reg_map = {
.fq_blen = 0x1b2c,
},
.gdm1_cnt = 0x2400,
- .gdma_to_ppe = 0x4444,
+ .gdma_to_ppe = {
+ [0] = 0x4444,
+ },
.ppe_base = 0x0c00,
.wdma_base = {
[0] = 0x2800,
@@ -144,7 +146,10 @@ static const struct mtk_reg_map mt7986_reg_map = {
.tx_sch_rate = 0x4798,
},
.gdm1_cnt = 0x1c00,
- .gdma_to_ppe = 0x3333,
+ .gdma_to_ppe = {
+ [0] = 0x3333,
+ [1] = 0x4444,
+ },
.ppe_base = 0x2000,
.wdma_base = {
[0] = 0x4800,
@@ -192,7 +197,11 @@ static const struct mtk_reg_map mt7988_reg_map = {
.tx_sch_rate = 0x4798,
},
.gdm1_cnt = 0x1c00,
- .gdma_to_ppe = 0x3333,
+ .gdma_to_ppe = {
+ [0] = 0x3333,
+ [1] = 0x4444,
+ [2] = 0xcccc,
+ },
.ppe_base = 0x2000,
.wdma_base = {
[0] = 0x4800,
@@ -2009,6 +2018,7 @@ static int mtk_poll_rx(struct napi_struct *napi, int budget,
struct mtk_rx_dma_v2 *rxd, trxd;
int done = 0, bytes = 0;
dma_addr_t dma_addr = DMA_MAPPING_ERROR;
+ int ppe_idx = 0;

while (done < budget) {
unsigned int pktlen, *rxdcsum;
@@ -2052,6 +2062,7 @@ static int mtk_poll_rx(struct napi_struct *napi, int budget,
goto release_desc;

netdev = eth->netdev[mac];
+ ppe_idx = eth->mac[mac]->ppe_idx;

if (unlikely(test_bit(MTK_RESETTING, &eth->state)))
goto release_desc;
@@ -2175,7 +2186,7 @@ static int mtk_poll_rx(struct napi_struct *napi, int budget,
}

if (reason == MTK_PPE_CPU_REASON_HIT_UNBIND_RATE_REACHED)
- mtk_ppe_check_skb(eth->ppe[0], skb, hash);
+ mtk_ppe_check_skb(eth->ppe[ppe_idx], skb, hash);

skb_record_rx_queue(skb, 0);
napi_gro_receive(napi, skb);
@@ -3266,37 +3277,27 @@ static int mtk_start_dma(struct mtk_eth *eth)
return 0;
}

-static void mtk_gdm_config(struct mtk_eth *eth, u32 config)
+static void mtk_gdm_config(struct mtk_eth *eth, u32 id, u32 config)
{
- int i;
+ u32 val;

if (MTK_HAS_CAPS(eth->soc->caps, MTK_SOC_MT7628))
return;

- for (i = 0; i < MTK_MAX_DEVS; i++) {
- u32 val;
-
- if (!eth->netdev[i])
- continue;
+ val = mtk_r32(eth, MTK_GDMA_FWD_CFG(id));

- val = mtk_r32(eth, MTK_GDMA_FWD_CFG(i));
+ /* default setup the forward port to send frame to PDMA */
+ val &= ~0xffff;

- /* default setup the forward port to send frame to PDMA */
- val &= ~0xffff;
+ /* Enable RX checksum */
+ val |= MTK_GDMA_ICS_EN | MTK_GDMA_TCS_EN | MTK_GDMA_UCS_EN;

- /* Enable RX checksum */
- val |= MTK_GDMA_ICS_EN | MTK_GDMA_TCS_EN | MTK_GDMA_UCS_EN;
+ val |= config;

- val |= config;
+ if (eth->netdev[id] && netdev_uses_dsa(eth->netdev[id]))
+ val |= MTK_GDMA_SPECIAL_TAG;

- if (netdev_uses_dsa(eth->netdev[i]))
- val |= MTK_GDMA_SPECIAL_TAG;
-
- mtk_w32(eth, val, MTK_GDMA_FWD_CFG(i));
- }
- /* Reset and enable PSE */
- mtk_w32(eth, RST_GL_PSE, MTK_RST_GL);
- mtk_w32(eth, 0, MTK_RST_GL);
+ mtk_w32(eth, val, MTK_GDMA_FWD_CFG(id));
}


@@ -3356,7 +3357,10 @@ static int mtk_open(struct net_device *dev)
{
struct mtk_mac *mac = netdev_priv(dev);
struct mtk_eth *eth = mac->hw;
- int i, err;
+ struct mtk_mac *target_mac;
+ int i, err, ppe_num;
+
+ ppe_num = mtk_get_ppe_num(eth);

err = phylink_of_phy_connect(mac->phylink, mac->of_node, 0);
if (err) {
@@ -3380,18 +3384,38 @@ static int mtk_open(struct net_device *dev)
for (i = 0; i < ARRAY_SIZE(eth->ppe); i++)
mtk_ppe_start(eth->ppe[i]);

- gdm_config = soc->offload_version ? soc->reg_map->gdma_to_ppe
- : MTK_GDMA_TO_PDMA;
- mtk_gdm_config(eth, gdm_config);
+ for (i = 0; i < MTK_MAX_DEVS; i++) {
+ if (!eth->netdev[i])
+ break;
+
+ target_mac = netdev_priv(eth->netdev[i]);
+ if (!soc->offload_version) {
+ target_mac->ppe_idx = 0;
+ gdm_config = MTK_GDMA_TO_PDMA;
+ } else if (ppe_num >= 3 && target_mac->id == 2) {
+ target_mac->ppe_idx = 2;
+ gdm_config = soc->reg_map->gdma_to_ppe[2];
+ } else if (ppe_num >= 2 && target_mac->id == 1) {
+ target_mac->ppe_idx = 1;
+ gdm_config = soc->reg_map->gdma_to_ppe[1];
+ } else {
+ target_mac->ppe_idx = 0;
+ gdm_config = soc->reg_map->gdma_to_ppe[0];
+ }
+ mtk_gdm_config(eth, target_mac->id, gdm_config);
+ }
+ /* Reset and enable PSE */
+ mtk_w32(eth, RST_GL_PSE, MTK_RST_GL);
+ mtk_w32(eth, 0, MTK_RST_GL);

napi_enable(&eth->tx_napi);
napi_enable(&eth->rx_napi);
mtk_tx_irq_enable(eth, MTK_TX_DONE_INT);
mtk_rx_irq_enable(eth, soc->txrx.rx_irq_done_mask);
refcount_set(&eth->dma_refcnt, 1);
- }
- else
+ } else {
refcount_inc(&eth->dma_refcnt);
+ }

phylink_start(mac->phylink);
netif_tx_start_all_queues(dev);
@@ -3468,7 +3492,8 @@ static int mtk_stop(struct net_device *dev)
if (!refcount_dec_and_test(&eth->dma_refcnt))
return 0;

- mtk_gdm_config(eth, MTK_GDMA_DROP_ALL);
+ for (i = 0; i < MTK_MAX_DEVS; i++)
+ mtk_gdm_config(eth, i, MTK_GDMA_DROP_ALL);

mtk_tx_irq_disable(eth, MTK_TX_DONE_INT);
mtk_rx_irq_disable(eth, eth->soc->txrx.rx_irq_done_mask);
@@ -4949,23 +4974,24 @@ static int mtk_probe(struct platform_device *pdev)
}

if (eth->soc->offload_version) {
- u32 num_ppe = mtk_is_netsys_v2_or_greater(eth) ? 2 : 1;
+ u8 ppe_num = mtk_get_ppe_num(eth);

- num_ppe = min_t(u32, ARRAY_SIZE(eth->ppe), num_ppe);
- for (i = 0; i < num_ppe; i++) {
- u32 ppe_addr = eth->soc->reg_map->ppe_base + i * 0x400;
+ ppe_num = min_t(u8, ARRAY_SIZE(eth->ppe), ppe_num);
+ for (i = 0; i < ppe_num; i++) {
+ u32 ppe_addr = eth->soc->reg_map->ppe_base;

+ ppe_addr += (i == 2 ? 0xc00 : i * 0x400);
eth->ppe[i] = mtk_ppe_init(eth, eth->base + ppe_addr, i);

if (!eth->ppe[i]) {
err = -ENOMEM;
goto err_deinit_ppe;
}
- }
+ err = mtk_eth_offload_init(eth, i);

- err = mtk_eth_offload_init(eth);
- if (err)
- goto err_deinit_ppe;
+ if (err)
+ goto err_deinit_ppe;
+ }
}

for (i = 0; i < MTK_MAX_DEVS; i++) {
diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.h b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
index 723fc637027c..e4c8c8bc7e7b 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.h
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
@@ -1130,7 +1130,7 @@ struct mtk_reg_map {
u32 tx_sch_rate; /* tx scheduler rate control registers */
} qdma;
u32 gdm1_cnt;
- u32 gdma_to_ppe;
+ u32 gdma_to_ppe[3];
u32 ppe_base;
u32 wdma_base[3];
u32 pse_iq_sta;
@@ -1286,7 +1286,7 @@ struct mtk_eth {

struct metadata_dst *dsa_meta[MTK_MAX_DSA_PORTS];

- struct mtk_ppe *ppe[2];
+ struct mtk_ppe *ppe[3];
struct rhashtable flow_table;

struct bpf_prog __rcu *prog;
@@ -1311,6 +1311,7 @@ struct mtk_eth {
struct mtk_mac {
int id;
phy_interface_t interface;
+ u8 ppe_idx;
int speed;
struct device_node *of_node;
struct phylink *phylink;
@@ -1421,6 +1422,14 @@ static inline u32 mtk_get_ib2_multicast_mask(struct mtk_eth *eth)
return MTK_FOE_IB2_MULTICAST;
}

+static inline u8 mtk_get_ppe_num(struct mtk_eth *eth)
+{
+ if (!eth->soc->offload_version)
+ return 0;
+
+ return eth->soc->version;
+}
+
/* read the hardware status register */
void mtk_stats_update_mac(struct mtk_mac *mac);

@@ -1432,7 +1441,7 @@ int mtk_gmac_sgmii_path_setup(struct mtk_eth *eth, int mac_id);
int mtk_gmac_gephy_path_setup(struct mtk_eth *eth, int mac_id);
int mtk_gmac_rgmii_path_setup(struct mtk_eth *eth, int mac_id);

-int mtk_eth_offload_init(struct mtk_eth *eth);
+int mtk_eth_offload_init(struct mtk_eth *eth, u8 id);
int mtk_eth_setup_tc(struct net_device *dev, enum tc_setup_type type,
void *type_data);
int mtk_flow_offload_cmd(struct mtk_eth *eth, struct flow_cls_offload *cls,
diff --git a/drivers/net/ethernet/mediatek/mtk_ppe_offload.c b/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
index aa262e6f4b85..fdad16c16193 100644
--- a/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
+++ b/drivers/net/ethernet/mediatek/mtk_ppe_offload.c
@@ -245,10 +245,10 @@ mtk_flow_offload_replace(struct mtk_eth *eth, struct flow_cls_offload *f,
int ppe_index)
{
struct flow_rule *rule = flow_cls_offload_flow_rule(f);
+ struct net_device *idev = NULL, *odev = NULL;
struct flow_action_entry *act;
struct mtk_flow_data data = {};
struct mtk_foe_entry foe;
- struct net_device *odev = NULL;
struct mtk_flow_entry *entry;
int offload_type = 0;
int wed_index = -1;
@@ -264,6 +264,14 @@ mtk_flow_offload_replace(struct mtk_eth *eth, struct flow_cls_offload *f,
struct flow_match_meta match;

flow_rule_match_meta(rule, &match);
+ if (mtk_is_netsys_v2_or_greater(eth)) {
+ idev = __dev_get_by_index(&init_net, match.key->ingress_ifindex);
+ if (idev) {
+ struct mtk_mac *mac = netdev_priv(idev);
+
+ ppe_index = mac->ppe_idx;
+ }
+ }
} else {
return -EOPNOTSUPP;
}
@@ -574,7 +582,7 @@ mtk_eth_setup_tc_block_cb(enum tc_setup_type type, void *type_data, void *cb_pri
if (type != TC_SETUP_CLSFLOWER)
return -EOPNOTSUPP;

- return mtk_flow_offload_cmd(eth, cls, 0);
+ return mtk_flow_offload_cmd(eth, cls, mac->ppe_idx);
}

static int
@@ -637,7 +645,9 @@ int mtk_eth_setup_tc(struct net_device *dev, enum tc_setup_type type,
}
}

-int mtk_eth_offload_init(struct mtk_eth *eth)
+int mtk_eth_offload_init(struct mtk_eth *eth, u8 id)
{
+ if (!eth->ppe[id] || !eth->ppe[id]->foe_table)
+ return 0;
return rhashtable_init(&eth->flow_table, &mtk_flow_ht_params);
}
--
2.44.0



2024-05-17 18:50:24

by Jakub Kicinski

[permalink] [raw]
Subject: Re: [PATCH net-next v4] net: ethernet: mtk_eth_soc: ppe: add support for multiple PPEs

On Sat, 11 May 2024 15:26:53 +0300 Elad Yifee wrote:
> Add the missing pieces to allow multiple PPEs units, one for each GMAC.
> mtk_gdm_config has been modified to work on targted mac ID,
> the inner loop moved outside of the function to allow unrelated
> operations like setting the MAC's PPE index.

## Form letter - net-next-closed

The merge window for v6.10 has begun and we have already posted our pull
request. Therefore net-next is closed for new drivers, features, code
refactoring and optimizations. We are currently accepting bug fixes only.

Please repost when net-next reopens after May 26th.

RFC patches sent for review only are obviously welcome at any time.

See: https://www.kernel.org/doc/html/next/process/maintainer-netdev.html#development-cycle
--
pw-bot: defer


2024-05-29 00:43:01

by Daniel Golle

[permalink] [raw]
Subject: Re: [PATCH net-next v4] net: ethernet: mtk_eth_soc: ppe: add support for multiple PPEs

Hi Elad,

now that net-next is opened again I finally also had some time to take
a deeper look at your patch ;)

On Sat, May 11, 2024 at 03:26:53PM GMT, Elad Yifee wrote:
> Add the missing pieces to allow multiple PPEs units, one for each GMAC.
> [...]
> diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> index 179c0230655a..67e19bd25f7a 100644
> --- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> +++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
> [...]
> @@ -1311,6 +1311,7 @@ struct mtk_eth {
> struct mtk_mac {
> int id;
> phy_interface_t interface;
> + u8 ppe_idx;

I think it would be smarter to just use
ifindex % mtk_get_ppe_num(eth)

Reasons:
- no need for an additional field in mtk_mac
- works for all net_device, not just mtk_eth type

> int speed;
> struct device_node *of_node;
> struct phylink *phylink;
> @@ -1421,6 +1422,14 @@ static inline u32 mtk_get_ib2_multicast_mask(struct mtk_eth *eth)
> return MTK_FOE_IB2_MULTICAST;
> }
>
> +static inline u8 mtk_get_ppe_num(struct mtk_eth *eth)
> +{
> + if (!eth->soc->offload_version)
> + return 0;
> +
> + return eth->soc->version;

Overloading the coincidentally identical hw version and number of PPEs
doesn't feel quite right. Better extend struct mtk_reg_map and turn the
gdma_to_ppe field into an array. Then the number of PPEs for a specific
SoC is the size of that array.

Reasons:
- we already got that mtk_reg_map structure for hw abstraction, no
need to introduce more SoC-specific macros like MTK_GDMA_TO_PPEx.
- no need for mtk_get_ppe_num inline function, or at least no
overloading of offload_version meaning.


Cheers


Daniel