openwrt-mirror/openwrt
1
0
Fork 0
mirror of https://github.com/openwrt/openwrt.git synced 2026-06-17 12:40:16 +04:00
Code Issues Packages Projects Releases Wiki Activity
Files
028dc3f57a6f9430181587a5af34d0cb7e9a442e
openwrt/target/linux/generic/pending-6.18/732-03-net-ethernet-mtk_eth_soc-optimize-dma-ring-address-i.patch
T
John Audia ed4b6ad372 kernel: bump 6.18 to 6.18.25 
Changelog: https://cdn.kernel.org/pub/linux/kernel/v6.x/ChangeLog-6.18.25

Manually rebased:
  generic/pending-6.18/795-09-net-ethernet-mtk_ppe-offload-flows-to-MxL862xx-switc.patch

All other patches automatically rebased via update_kernel.sh

Build system: x86/64
Build-tested: x86/64-glibc
Run-tested: x86/64-glibc

Signed-off-by: John Audia <therealgraysky@proton.me>
Link: https://github.com/openwrt/openwrt/pull/22890
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
2026-04-30 01:15:19 +02:00

484 lines
16 KiB
Diff
Raw Blame History

From: Felix Fietkau <nbd@nbd.name>
Date: Tue, 15 Oct 2024 12:52:56 +0200
Subject: [PATCH] net: ethernet: mtk_eth_soc: optimize dma ring address/index
calculation
Since DMA descriptor sizes are all power of 2, we can avoid costly integer
division in favor or simple shifts.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
---
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
@@ -44,6 +44,11 @@ MODULE_PARM_DESC(msg_level, "Message lev
offsetof(struct mtk_hw_stats, xdp_stats.x) / \
sizeof(u64) }
+#define RX_DESC_OFS(eth, i) \
+ ((i) << (eth)->soc->rx.desc_shift)
+#define TX_DESC_OFS(eth, i) \
+ ((i) << (eth)->soc->tx.desc_shift)
+
static const struct mtk_reg_map mtk_reg_map = {
.tx_irq_mask = 0x1a1c,
.tx_irq_status = 0x1a18,
@@ -1306,13 +1311,13 @@ static int mtk_init_fq_dma(struct mtk_et
dma_addr_t dma_addr;
int i, j, len;
- eth->scratch_ring = mtk_dma_ring_alloc(eth, cnt * soc->tx.desc_size,
+ eth->scratch_ring = mtk_dma_ring_alloc(eth, TX_DESC_OFS(eth, cnt),
&eth->phy_scratch_ring, true);
if (unlikely(!eth->scratch_ring))
return -ENOMEM;
- phy_ring_tail = eth->phy_scratch_ring + soc->tx.desc_size * (cnt - 1);
+ phy_ring_tail = eth->phy_scratch_ring + TX_DESC_OFS(eth, cnt - 1);
for (j = 0; j < DIV_ROUND_UP(soc->tx.fq_dma_size, MTK_FQ_DMA_LENGTH); j++) {
len = min_t(int, cnt - j * MTK_FQ_DMA_LENGTH, MTK_FQ_DMA_LENGTH);
@@ -1331,11 +1336,11 @@ static int mtk_init_fq_dma(struct mtk_et
for (i = 0; i < len; i++) {
struct mtk_tx_dma_v2 *txd;
- txd = eth->scratch_ring + (j * MTK_FQ_DMA_LENGTH + i) * soc->tx.desc_size;
+ txd = eth->scratch_ring + TX_DESC_OFS(eth, j * MTK_FQ_DMA_LENGTH + i);
txd->txd1 = dma_addr + i * MTK_QDMA_PAGE_SIZE;
if (j * MTK_FQ_DMA_LENGTH + i < cnt)
txd->txd2 = eth->phy_scratch_ring +
- (j * MTK_FQ_DMA_LENGTH + i + 1) * soc->tx.desc_size;
+ TX_DESC_OFS(eth, j * MTK_FQ_DMA_LENGTH + i + 1);
txd->txd3 = TX_DMA_PLEN0(MTK_QDMA_PAGE_SIZE);
if (MTK_HAS_CAPS(soc->caps, MTK_36BIT_DMA))
@@ -1365,9 +1370,9 @@ static void *mtk_qdma_phys_to_virt(struc
}
static struct mtk_tx_buf *mtk_desc_to_tx_buf(struct mtk_tx_ring *ring,
- void *txd, u32 txd_size)
+ void *txd, u32 txd_shift)
{
- int idx = (txd - ring->dma) / txd_size;
+ int idx = (txd - ring->dma) >> txd_shift;
return &ring->buf[idx];
}
@@ -1378,9 +1383,9 @@ static struct mtk_tx_dma *qdma_to_pdma(s
return ring->dma_pdma - (struct mtk_tx_dma *)ring->dma + dma;
}
-static int txd_to_idx(struct mtk_tx_ring *ring, void *dma, u32 txd_size)
+static int txd_to_idx(struct mtk_tx_ring *ring, void *dma, u32 txd_shift)
{
- return (dma - ring->dma) / txd_size;
+ return (dma - ring->dma) >> txd_shift;
}
static void mtk_tx_unmap(struct mtk_eth *eth, struct mtk_tx_buf *tx_buf,
@@ -1588,7 +1593,7 @@ static int mtk_tx_map(struct sk_buff *sk
if (itxd == ring->last_free)
return -ENOMEM;
- itx_buf = mtk_desc_to_tx_buf(ring, itxd, soc->tx.desc_size);
+ itx_buf = mtk_desc_to_tx_buf(ring, itxd, soc->tx.desc_shift);
memset(itx_buf, 0, sizeof(*itx_buf));
txd_info.addr = dma_map_single(eth->dma_dev, skb->data, txd_info.size,
@@ -1642,7 +1647,7 @@ static int mtk_tx_map(struct sk_buff *sk
mtk_tx_set_dma_desc(dev, txd, &txd_info);
tx_buf = mtk_desc_to_tx_buf(ring, txd,
- soc->tx.desc_size);
+ soc->tx.desc_shift);
if (new_desc)
memset(tx_buf, 0, sizeof(*tx_buf));
tx_buf->data = (void *)MTK_DMA_DUMMY_DESC;
@@ -1685,7 +1690,7 @@ static int mtk_tx_map(struct sk_buff *sk
} else {
int next_idx;
- next_idx = NEXT_DESP_IDX(txd_to_idx(ring, txd, soc->tx.desc_size),
+ next_idx = NEXT_DESP_IDX(txd_to_idx(ring, txd, soc->tx.desc_shift),
ring->dma_size);
mtk_w32(eth, next_idx, MT7628_TX_CTX_IDX0);
}
@@ -1694,7 +1699,7 @@ static int mtk_tx_map(struct sk_buff *sk
err_dma:
do {
- tx_buf = mtk_desc_to_tx_buf(ring, itxd, soc->tx.desc_size);
+ tx_buf = mtk_desc_to_tx_buf(ring, itxd, soc->tx.desc_shift);
/* unmap dma */
mtk_tx_unmap(eth, tx_buf, NULL, false);
@@ -1868,7 +1873,7 @@ static struct mtk_rx_ring *mtk_get_rx_ri
ring = &eth->rx_ring[i];
idx = NEXT_DESP_IDX(ring->calc_idx, ring->dma_size);
- rxd = ring->dma + idx * eth->soc->rx.desc_size;
+ rxd = ring->dma + RX_DESC_OFS(eth, idx);
if (rxd->rxd2 & RX_DMA_DONE) {
ring->calc_idx_update = true;
return ring;
@@ -2036,7 +2041,7 @@ static int mtk_xdp_submit_frame(struct m
}
htxd = txd;
- tx_buf = mtk_desc_to_tx_buf(ring, txd, soc->tx.desc_size);
+ tx_buf = mtk_desc_to_tx_buf(ring, txd, soc->tx.desc_shift);
memset(tx_buf, 0, sizeof(*tx_buf));
htx_buf = tx_buf;
@@ -2055,7 +2060,7 @@ static int mtk_xdp_submit_frame(struct m
goto unmap;
tx_buf = mtk_desc_to_tx_buf(ring, txd,
- soc->tx.desc_size);
+ soc->tx.desc_shift);
memset(tx_buf, 0, sizeof(*tx_buf));
n_desc++;
}
@@ -2093,7 +2098,7 @@ static int mtk_xdp_submit_frame(struct m
} else {
int idx;
- idx = txd_to_idx(ring, txd, soc->tx.desc_size);
+ idx = txd_to_idx(ring, txd, soc->tx.desc_shift);
mtk_w32(eth, NEXT_DESP_IDX(idx, ring->dma_size),
MT7628_TX_CTX_IDX0);
}
@@ -2104,7 +2109,7 @@ static int mtk_xdp_submit_frame(struct m
unmap:
while (htxd != txd) {
- tx_buf = mtk_desc_to_tx_buf(ring, htxd, soc->tx.desc_size);
+ tx_buf = mtk_desc_to_tx_buf(ring, htxd, soc->tx.desc_shift);
mtk_tx_unmap(eth, tx_buf, NULL, false);
htxd->txd3 = TX_DMA_LS0 | TX_DMA_OWNER_CPU;
@@ -2236,7 +2241,7 @@ static int mtk_poll_rx(struct napi_struc
goto rx_done;
idx = NEXT_DESP_IDX(ring->calc_idx, ring->dma_size);
- rxd = ring->dma + idx * eth->soc->rx.desc_size;
+ rxd = ring->dma + RX_DESC_OFS(eth, idx);
data = ring->data[idx];
if (!mtk_rx_get_desc(eth, &trxd, rxd))
@@ -2503,7 +2508,7 @@ static int mtk_poll_tx_qdma(struct mtk_e
break;
tx_buf = mtk_desc_to_tx_buf(ring, desc,
- eth->soc->tx.desc_size);
+ eth->soc->tx.desc_shift);
if (!tx_buf->data)
break;
@@ -2554,7 +2559,7 @@ static int mtk_poll_tx_pdma(struct mtk_e
}
mtk_tx_unmap(eth, tx_buf, &bq, true);
- desc = ring->dma + cpu * eth->soc->tx.desc_size;
+ desc = ring->dma + TX_DESC_OFS(eth, cpu);
ring->last_free = desc;
atomic_inc(&ring->free_count);
@@ -2672,7 +2677,7 @@ static int mtk_tx_alloc(struct mtk_eth *
{
const struct mtk_soc_data *soc = eth->soc;
struct mtk_tx_ring *ring = &eth->tx_ring;
- int i, sz = soc->tx.desc_size;
+ int i, sz = TX_DESC_OFS(eth, 1);
struct mtk_tx_dma_v2 *txd;
int ring_size;
u32 ofs, val;
@@ -2712,7 +2717,7 @@ static int mtk_tx_alloc(struct mtk_eth *
* descriptors in ring->dma_pdma.
*/
if (!MTK_HAS_CAPS(soc->caps, MTK_QDMA)) {
- ring->dma_pdma = dma_alloc_coherent(eth->dma_dev, ring_size * sz,
+ ring->dma_pdma = dma_alloc_coherent(eth->dma_dev, TX_DESC_OFS(eth, ring_size),
&ring->phys_pdma, GFP_KERNEL);
if (!ring->dma_pdma)
goto no_tx_mem;
@@ -2727,7 +2732,7 @@ static int mtk_tx_alloc(struct mtk_eth *
atomic_set(&ring->free_count, ring_size - 2);
ring->next_free = ring->dma;
ring->last_free = (void *)txd;
- ring->last_free_ptr = (u32)(ring->phys + ((ring_size - 1) * sz));
+ ring->last_free_ptr = (u32)(ring->phys + TX_DESC_OFS(eth, ring_size - 1));
ring->thresh = MAX_SKB_FRAGS;
/* make sure that all changes to the dma ring are flushed before we
@@ -2739,7 +2744,7 @@ static int mtk_tx_alloc(struct mtk_eth *
mtk_w32(eth, ring->phys, soc->reg_map->qdma.ctx_ptr);
mtk_w32(eth, ring->phys, soc->reg_map->qdma.dtx_ptr);
mtk_w32(eth,
- ring->phys + ((ring_size - 1) * sz),
+ ring->phys + TX_DESC_OFS(eth, ring_size - 1),
soc->reg_map->qdma.crx_ptr);
mtk_w32(eth, ring->last_free_ptr, soc->reg_map->qdma.drx_ptr);
@@ -2776,7 +2781,6 @@ no_tx_mem:
static void mtk_tx_clean(struct mtk_eth *eth)
{
- const struct mtk_soc_data *soc = eth->soc;
struct mtk_tx_ring *ring = &eth->tx_ring;
int i;
@@ -2788,14 +2792,14 @@ static void mtk_tx_clean(struct mtk_eth
}
if (ring->dma) {
- mtk_dma_ring_free(eth, ring->dma_size * soc->tx.desc_size,
+ mtk_dma_ring_free(eth, TX_DESC_OFS(eth, ring->dma_size),
ring->dma, ring->phys, true);
ring->dma = NULL;
}
if (ring->dma_pdma) {
dma_free_coherent(eth->dma_dev,
- ring->dma_size * soc->tx.desc_size,
+ TX_DESC_OFS(eth, ring->dma_size),
ring->dma_pdma, ring->phys_pdma);
ring->dma_pdma = NULL;
}
@@ -2844,7 +2848,7 @@ static int mtk_rx_alloc(struct mtk_eth *
}
ring->dma = mtk_dma_ring_alloc(eth,
- rx_dma_size * eth->soc->rx.desc_size,
+ RX_DESC_OFS(eth, rx_dma_size),
&ring->phys,
rx_flag == MTK_RX_FLAGS_NORMAL);
if (!ring->dma)
@@ -2855,7 +2859,7 @@ static int mtk_rx_alloc(struct mtk_eth *
dma_addr_t dma_addr;
void *data;
- rxd = ring->dma + i * eth->soc->rx.desc_size;
+ rxd = ring->dma + RX_DESC_OFS(eth, i);
if (ring->page_pool) {
data = mtk_page_pool_get_buff(ring->page_pool,
&dma_addr, GFP_KERNEL);
@@ -2946,7 +2950,7 @@ static void mtk_rx_clean(struct mtk_eth
if (!ring->data[i])
continue;
- rxd = ring->dma + i * eth->soc->rx.desc_size;
+ rxd = ring->dma + RX_DESC_OFS(eth, i);
if (!rxd->rxd1)
continue;
@@ -2962,7 +2966,7 @@ static void mtk_rx_clean(struct mtk_eth
}
if (ring->dma) {
- mtk_dma_ring_free(eth, ring->dma_size * eth->soc->rx.desc_size,
+ mtk_dma_ring_free(eth, RX_DESC_OFS(eth, ring->dma_size),
ring->dma, ring->phys, in_sram);
ring->dma = NULL;
}
@@ -3332,7 +3336,7 @@ static void mtk_dma_free(struct mtk_eth
}
if (eth->scratch_ring) {
- mtk_dma_ring_free(eth, soc->tx.fq_dma_size * soc->tx.desc_size,
+ mtk_dma_ring_free(eth, TX_DESC_OFS(eth, soc->tx.fq_dma_size),
eth->scratch_ring, eth->phy_scratch_ring,
true);
eth->scratch_ring = NULL;
@@ -5403,6 +5407,9 @@ static void mtk_remove(struct platform_d
mtk_mdio_cleanup(eth);
}
+#define DESC_SIZE(struct_name) \
+ .desc_shift = const_ilog2(sizeof(struct_name))
+
static const struct mtk_soc_data mt2701_data = {
.reg_map = &mtk_reg_map,
.caps = MT7623_CAPS | MTK_HWLRO,
@@ -5411,14 +5418,14 @@ static const struct mtk_soc_data mt2701_
.required_pctl = true,
.version = 1,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma),
+ DESC_SIZE(struct mtk_tx_dma),
.dma_max_len = MTK_TX_DMA_BUF_LEN,
.dma_len_offset = 16,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID,
.dma_size = MTK_DMA_SIZE(2K),
@@ -5439,14 +5446,14 @@ static const struct mtk_soc_data mt7621_
.hash_offset = 2,
.foe_entry_size = MTK_FOE_ENTRY_V1_SIZE,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma),
+ DESC_SIZE(struct mtk_tx_dma),
.dma_max_len = MTK_TX_DMA_BUF_LEN,
.dma_len_offset = 16,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID,
.dma_size = MTK_DMA_SIZE(2K),
@@ -5469,14 +5476,14 @@ static const struct mtk_soc_data mt7622_
.has_accounting = true,
.foe_entry_size = MTK_FOE_ENTRY_V1_SIZE,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma),
+ DESC_SIZE(struct mtk_tx_dma),
.dma_max_len = MTK_TX_DMA_BUF_LEN,
.dma_len_offset = 16,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID,
.dma_size = MTK_DMA_SIZE(2K),
@@ -5498,14 +5505,14 @@ static const struct mtk_soc_data mt7623_
.foe_entry_size = MTK_FOE_ENTRY_V1_SIZE,
.disable_pll_modes = true,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma),
+ DESC_SIZE(struct mtk_tx_dma),
.dma_max_len = MTK_TX_DMA_BUF_LEN,
.dma_len_offset = 16,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID,
.dma_size = MTK_DMA_SIZE(2K),
@@ -5524,14 +5531,14 @@ static const struct mtk_soc_data mt7629_
.has_accounting = true,
.version = 1,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma),
+ DESC_SIZE(struct mtk_tx_dma),
.dma_max_len = MTK_TX_DMA_BUF_LEN,
.dma_len_offset = 16,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID,
.dma_size = MTK_DMA_SIZE(2K),
@@ -5554,14 +5561,14 @@ static const struct mtk_soc_data mt7981_
.has_accounting = true,
.foe_entry_size = MTK_FOE_ENTRY_V2_SIZE,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma_v2),
+ DESC_SIZE(struct mtk_tx_dma_v2),
.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
.dma_len_offset = 8,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID_V2,
.dma_max_len = MTK_TX_DMA_BUF_LEN,
@@ -5584,14 +5591,14 @@ static const struct mtk_soc_data mt7986_
.has_accounting = true,
.foe_entry_size = MTK_FOE_ENTRY_V2_SIZE,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma_v2),
+ DESC_SIZE(struct mtk_tx_dma_v2),
.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
.dma_len_offset = 8,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID_V2,
.dma_max_len = MTK_TX_DMA_BUF_LEN,
@@ -5614,14 +5621,14 @@ static const struct mtk_soc_data mt7988_
.has_accounting = true,
.foe_entry_size = MTK_FOE_ENTRY_V3_SIZE,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma_v2),
+ DESC_SIZE(struct mtk_tx_dma_v2),
.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
.dma_len_offset = 8,
.dma_size = MTK_DMA_SIZE(2K),
.fq_dma_size = MTK_DMA_SIZE(4K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma_v2),
+ DESC_SIZE(struct mtk_rx_dma_v2),
.irq_done_mask = MTK_RX_DONE_INT_V2,
.dma_l4_valid = RX_DMA_L4_VALID_V2,
.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
@@ -5638,13 +5645,13 @@ static const struct mtk_soc_data rt5350_
.required_pctl = false,
.version = 1,
.tx = {
- .desc_size = sizeof(struct mtk_tx_dma),
+ DESC_SIZE(struct mtk_tx_dma),
.dma_max_len = MTK_TX_DMA_BUF_LEN,
.dma_len_offset = 16,
.dma_size = MTK_DMA_SIZE(2K),
},
.rx = {
- .desc_size = sizeof(struct mtk_rx_dma),
+ DESC_SIZE(struct mtk_rx_dma),
.irq_done_mask = MTK_RX_DONE_INT,
.dma_l4_valid = RX_DMA_L4_VALID_PDMA,
.dma_max_len = MTK_TX_DMA_BUF_LEN,
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.h
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
@@ -1200,7 +1200,7 @@ struct mtk_reg_map {
* @foe_entry_size Foe table entry size.
* @has_accounting Bool indicating support for accounting of
* offloaded flows.
- * @desc_size Tx/Rx DMA descriptor size.
+ * @desc_shift Tx/Rx DMA descriptor size (in power-of-2).
* @irq_done_mask Rx irq done register mask.
* @dma_l4_valid Rx DMA valid register mask.
* @dma_max_len Max DMA tx/rx buffer length.
@@ -1221,14 +1221,14 @@ struct mtk_soc_data {
bool has_accounting;
bool disable_pll_modes;
struct {
- u32 desc_size;
+ u32 desc_shift;
u32 dma_max_len;
u32 dma_len_offset;
u32 dma_size;
u32 fq_dma_size;
} tx;
struct {
- u32 desc_size;
+ u32 desc_shift;
u32 irq_done_mask;
u32 dma_l4_valid;
u32 dma_max_len;
Reference in New Issue View Git Blame Copy Permalink