From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 27 Sep 2022 13:36:11 -0700
Subject: net/mlx5e: Use runtime values of striding RQ parameters in datapath
Patch-mainline: v6.1-rc1
Git-commit: 997ce6affe264e80c2fd1ab6f09d22c00d7bfa38
References: jsc#PED-1549
Some of the parameters of striding RQ are compile-time constants, but
they are going to become dynamically calculated at runtime in a
following commit. This commit prepares the datapath to take cached
runtime parameters, prefilled at queue creation.
New fields added to struct mlx5e_rq fit into an existing 7-byte hole.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
drivers/net/ethernet/mellanox/mlx5/core/en.h | 12 +----
drivers/net/ethernet/mellanox/mlx5/core/en/params.c | 14 ++++++
drivers/net/ethernet/mellanox/mlx5/core/en/params.h | 5 ++
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 41 +++++++++++---------
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 35 +++++++++--------
5 files changed, 65 insertions(+), 42 deletions(-)
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -107,7 +107,6 @@ struct page_pool;
* dropped by the driver at a later stage.
*/
#define MLX5E_REQUIRED_WQE_MTTS (MLX5_ALIGN_MTTS(MLX5_MPWRQ_PAGES_PER_WQE + 1))
-#define MLX5E_REQUIRED_MTTS(wqes) (wqes * MLX5E_REQUIRED_WQE_MTTS)
#define MLX5E_MAX_RQ_NUM_MTTS \
(ALIGN_DOWN(U16_MAX, 4) * 2) /* So that MLX5_MTT_OCTW(num_mtts) fits into u16 */
#define MLX5E_ORDER2_MAX_PACKET_MTU (order_base_2(10 * 1024))
@@ -150,13 +149,6 @@ struct page_pool;
#define MLX5E_TX_XSK_POLL_BUDGET 64
#define MLX5E_SQ_RECOVER_MIN_INTERVAL 500 /* msecs */
-#define MLX5E_UMR_WQE_INLINE_SZ \
- (sizeof(struct mlx5e_umr_wqe) + \
- ALIGN(MLX5_MPWRQ_PAGES_PER_WQE * sizeof(struct mlx5_mtt), \
- MLX5_UMR_MTT_ALIGNMENT))
-#define MLX5E_UMR_WQEBBS \
- (DIV_ROUND_UP(MLX5E_UMR_WQE_INLINE_SZ, MLX5_SEND_WQE_BB))
-
#define MLX5E_KLM_UMR_WQE_SZ(sgl_len)\
(sizeof(struct mlx5e_umr_wqe) +\
(sizeof(struct mlx5_klm) * (sgl_len)))
@@ -712,6 +704,10 @@ struct mlx5e_rq {
u8 umr_last_bulk;
u8 umr_completed;
u8 min_wqe_bulk;
+ u8 page_shift;
+ u8 pages_per_wqe;
+ u8 umr_wqebbs;
+ u8 mtts_per_wqe;
struct mlx5e_shampo_hd *shampo;
} mpwqe;
};
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
@@ -7,6 +7,17 @@
#include "en_accel/en_accel.h"
#include "en_accel/ipsec.h"
+u16 mlx5e_mpwrq_umr_wqe_sz(u8 pages_per_wqe)
+{
+ return sizeof(struct mlx5e_umr_wqe) +
+ ALIGN(pages_per_wqe * sizeof(struct mlx5_mtt), MLX5_UMR_MTT_ALIGNMENT);
+}
+
+u8 mlx5e_mpwrq_umr_wqebbs(u8 pages_per_wqe)
+{
+ return DIV_ROUND_UP(mlx5e_mpwrq_umr_wqe_sz(pages_per_wqe), MLX5_SEND_WQE_BB);
+}
+
u16 mlx5e_get_linear_rq_headroom(struct mlx5e_params *params,
struct mlx5e_xsk_param *xsk)
{
@@ -786,7 +797,8 @@ static u8 mlx5e_build_icosq_log_wq_sz(st
if (params->rq_wq_type != MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
return MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE;
- wqebbs = MLX5E_UMR_WQEBBS * BIT(mlx5e_get_rq_log_wq_sz(rqp->rqc));
+ wqebbs = mlx5e_mpwrq_umr_wqebbs(MLX5_MPWRQ_PAGES_PER_WQE) *
+ (1 << mlx5e_get_rq_log_wq_sz(rqp->rqc));
/* If XDP program is attached, XSK may be turned on at any time without
* restarting the channel. ICOSQ must be big enough to fit UMR WQEs of
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/params.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/params.h
@@ -84,6 +84,11 @@ static inline bool mlx5e_qid_validate(co
return qid < params->num_channels * profile->rq_groups;
}
+/* Striding RQ dynamic parameters */
+
+u16 mlx5e_mpwrq_umr_wqe_sz(u8 pages_per_wqe);
+u8 mlx5e_mpwrq_umr_wqebbs(u8 pages_per_wqe);
+
/* Parameter calculations */
void mlx5e_reset_tx_moderation(struct mlx5e_params *params, u8 cq_period_mode);
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -71,17 +71,20 @@
bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
{
bool striding_rq_umr, inline_umr;
- u16 max_wqe_sz_cap;
+ u16 max_wqebbs;
+ u16 umr_wqebbs;
striding_rq_umr = MLX5_CAP_GEN(mdev, striding_rq) && MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
MLX5_CAP_ETH(mdev, reg_umr_sq);
- max_wqe_sz_cap = mlx5e_get_max_sq_aligned_wqebbs(mdev) * MLX5_SEND_WQE_BB;
- inline_umr = max_wqe_sz_cap >= MLX5E_UMR_WQE_INLINE_SZ;
+ max_wqebbs = mlx5e_get_max_sq_aligned_wqebbs(mdev);
+ umr_wqebbs = mlx5e_mpwrq_umr_wqebbs(MLX5_MPWRQ_PAGES_PER_WQE);
+ inline_umr = umr_wqebbs <= max_wqebbs;
if (!striding_rq_umr)
return false;
if (!inline_umr) {
- mlx5_core_warn(mdev, "Cannot support Striding RQ: UMR WQE size (%d) exceeds maximum supported (%d).\n",
- (int)MLX5E_UMR_WQE_INLINE_SZ, max_wqe_sz_cap);
+ mlx5_core_warn(mdev, "Cannot support Striding RQ: UMR WQE size (%u) exceeds maximum supported (%u).\n",
+ umr_wqebbs * MLX5_SEND_WQE_BB,
+ max_wqebbs * MLX5_SEND_WQE_BB);
return false;
}
return true;
@@ -206,7 +209,10 @@ static inline void mlx5e_build_umr_wqe(s
{
struct mlx5_wqe_ctrl_seg *cseg = &wqe->ctrl;
struct mlx5_wqe_umr_ctrl_seg *ucseg = &wqe->uctrl;
- u8 ds_cnt = DIV_ROUND_UP(MLX5E_UMR_WQE_INLINE_SZ, MLX5_SEND_WQE_DS);
+ u8 ds_cnt;
+
+ ds_cnt = DIV_ROUND_UP(mlx5e_mpwrq_umr_wqe_sz(rq->mpwqe.pages_per_wqe),
+ MLX5_SEND_WQE_DS);
cseg->qpn_ds = cpu_to_be32((sq->sqn << MLX5_WQE_CTRL_QPN_SHIFT) |
ds_cnt);
@@ -214,7 +220,7 @@ static inline void mlx5e_build_umr_wqe(s
ucseg->flags = MLX5_UMR_TRANSLATION_OFFSET_EN | MLX5_UMR_INLINE;
ucseg->xlt_octowords =
- cpu_to_be16(MLX5_MTT_OCTW(MLX5_MPWRQ_PAGES_PER_WQE));
+ cpu_to_be16(MLX5_MTT_OCTW(rq->mpwqe.pages_per_wqe));
ucseg->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
}
@@ -263,7 +269,7 @@ static int mlx5e_rq_alloc_mpwqe_info(str
size_t alloc_size;
alloc_size = array_size(wq_sz, struct_size(rq->mpwqe.info, dma_info,
- MLX5_MPWRQ_PAGES_PER_WQE));
+ rq->mpwqe.pages_per_wqe));
rq->mpwqe.info = kvzalloc_node(alloc_size, GFP_KERNEL, node);
if (!rq->mpwqe.info)
@@ -359,9 +365,9 @@ static int mlx5e_create_umr_klm_mkey(str
static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev *mdev, struct mlx5e_rq *rq)
{
- u64 num_mtts = MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq->mpwqe.wq));
+ u64 num_mtts = mlx5_wq_ll_get_size(&rq->mpwqe.wq) * rq->mpwqe.mtts_per_wqe;
- return mlx5e_create_umr_mtt_mkey(mdev, num_mtts, PAGE_SHIFT,
+ return mlx5e_create_umr_mtt_mkey(mdev, num_mtts, rq->mpwqe.page_shift,
&rq->umr_mkey, rq->wqe_overflow.addr);
}
@@ -379,11 +385,6 @@ static int mlx5e_create_rq_hd_umr_mkey(s
&rq->mpwqe.shampo->mkey);
}
-static u64 mlx5e_get_mpwqe_offset(u16 wqe_ix)
-{
- return MLX5E_REQUIRED_MTTS(wqe_ix) << PAGE_SHIFT;
-}
-
static void mlx5e_init_frags_partition(struct mlx5e_rq *rq)
{
struct mlx5e_wqe_frag_info next_frag = {};
@@ -590,7 +591,12 @@ static int mlx5e_alloc_rq(struct mlx5e_p
wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
- pool_size = MLX5_MPWRQ_PAGES_PER_WQE <<
+ rq->mpwqe.page_shift = PAGE_SHIFT;
+ rq->mpwqe.pages_per_wqe = MLX5_MPWRQ_PAGES_PER_WQE;
+ rq->mpwqe.umr_wqebbs = mlx5e_mpwrq_umr_wqebbs(rq->mpwqe.pages_per_wqe);
+ rq->mpwqe.mtts_per_wqe = MLX5E_REQUIRED_WQE_MTTS;
+
+ pool_size = rq->mpwqe.pages_per_wqe <<
mlx5e_mpwqe_get_log_rq_size(params, xsk);
rq->mpwqe.log_stride_sz = mlx5e_mpwqe_get_log_stride_size(mdev, params, xsk);
@@ -680,7 +686,8 @@ static int mlx5e_alloc_rq(struct mlx5e_p
mlx5_wq_ll_get_wqe(&rq->mpwqe.wq, i);
u32 byte_count =
rq->mpwqe.num_strides << rq->mpwqe.log_stride_sz;
- u64 dma_offset = mlx5e_get_mpwqe_offset(i);
+ u64 dma_offset = mul_u32_u32(i, rq->mpwqe.mtts_per_wqe) <<
+ rq->mpwqe.page_shift;
u16 headroom = test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state) ?
0 : rq->buff.headroom;
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -77,7 +77,7 @@ const struct mlx5e_rx_handlers mlx5e_rx_
static struct mlx5e_mpw_info *mlx5e_get_mpw_info(struct mlx5e_rq *rq, int i)
{
- size_t isz = struct_size(rq->mpwqe.info, dma_info, MLX5_MPWRQ_PAGES_PER_WQE);
+ size_t isz = struct_size(rq->mpwqe.info, dma_info, rq->mpwqe.pages_per_wqe);
return (struct mlx5e_mpw_info *)((char *)rq->mpwqe.info + array_size(i, isz));
}
@@ -272,6 +272,7 @@ static inline bool mlx5e_rx_cache_get(st
stats->cache_reuse++;
dma_sync_single_for_device(rq->pdev, dma_info->addr,
+ /* Non-XSK always uses PAGE_SIZE. */
PAGE_SIZE,
DMA_FROM_DEVICE);
return true;
@@ -287,6 +288,7 @@ static inline int mlx5e_page_alloc_pool(
if (unlikely(!dma_info->page))
return -ENOMEM;
+ /* Non-XSK always uses PAGE_SIZE. */
dma_info->addr = dma_map_page_attrs(rq->pdev, dma_info->page, 0, PAGE_SIZE,
rq->buff.map_dir, DMA_ATTR_SKIP_CPU_SYNC);
if (unlikely(dma_mapping_error(rq->pdev, dma_info->addr))) {
@@ -489,13 +491,12 @@ mlx5e_free_rx_mpwqe(struct mlx5e_rq *rq,
int i;
/* A common case for AF_XDP. */
- if (bitmap_full(wi->xdp_xmit_bitmap, MLX5_MPWRQ_PAGES_PER_WQE))
+ if (bitmap_full(wi->xdp_xmit_bitmap, rq->mpwqe.pages_per_wqe))
return;
- no_xdp_xmit = bitmap_empty(wi->xdp_xmit_bitmap,
- MLX5_MPWRQ_PAGES_PER_WQE);
+ no_xdp_xmit = bitmap_empty(wi->xdp_xmit_bitmap, rq->mpwqe.pages_per_wqe);
- for (i = 0; i < MLX5_MPWRQ_PAGES_PER_WQE; i++)
+ for (i = 0; i < rq->mpwqe.pages_per_wqe; i++)
if (no_xdp_xmit || !test_bit(i, wi->xdp_xmit_bitmap))
mlx5e_page_release(rq, &dma_info[i], recycle);
}
@@ -680,7 +681,7 @@ static int mlx5e_alloc_rx_mpwqe(struct m
* one-by-one, failing and moving frames to the Reuse Ring.
*/
if (rq->xsk_pool &&
- unlikely(!xsk_buff_can_alloc(rq->xsk_pool, MLX5_MPWRQ_PAGES_PER_WQE))) {
+ unlikely(!xsk_buff_can_alloc(rq->xsk_pool, rq->mpwqe.pages_per_wqe))) {
err = -ENOMEM;
goto err;
}
@@ -691,33 +692,33 @@ static int mlx5e_alloc_rx_mpwqe(struct m
goto err;
}
- pi = mlx5e_icosq_get_next_pi(sq, MLX5E_UMR_WQEBBS);
+ pi = mlx5e_icosq_get_next_pi(sq, rq->mpwqe.umr_wqebbs);
umr_wqe = mlx5_wq_cyc_get_wqe(wq, pi);
memcpy(umr_wqe, &rq->mpwqe.umr_wqe, offsetof(struct mlx5e_umr_wqe, inline_mtts));
- for (i = 0; i < MLX5_MPWRQ_PAGES_PER_WQE; i++, dma_info++) {
+ for (i = 0; i < rq->mpwqe.pages_per_wqe; i++, dma_info++) {
err = mlx5e_page_alloc(rq, dma_info);
if (unlikely(err))
goto err_unmap;
umr_wqe->inline_mtts[i].ptag = cpu_to_be64(dma_info->addr | MLX5_EN_WR);
}
- bitmap_zero(wi->xdp_xmit_bitmap, MLX5_MPWRQ_PAGES_PER_WQE);
+ bitmap_zero(wi->xdp_xmit_bitmap, rq->mpwqe.pages_per_wqe);
wi->consumed_strides = 0;
umr_wqe->ctrl.opmod_idx_opcode =
cpu_to_be32((sq->pc << MLX5_WQE_CTRL_WQE_INDEX_SHIFT) |
MLX5_OPCODE_UMR);
umr_wqe->uctrl.xlt_offset =
- cpu_to_be16(MLX5_ALIGNED_MTTS_OCTW(MLX5E_REQUIRED_MTTS(ix)));
+ cpu_to_be16(MLX5_ALIGNED_MTTS_OCTW(ix * rq->mpwqe.mtts_per_wqe));
sq->db.wqe_info[pi] = (struct mlx5e_icosq_wqe_info) {
.wqe_type = MLX5E_ICOSQ_WQE_UMR_RX,
- .num_wqebbs = MLX5E_UMR_WQEBBS,
+ .num_wqebbs = rq->mpwqe.umr_wqebbs,
.umr.rq = rq,
};
- sq->pc += MLX5E_UMR_WQEBBS;
+ sq->pc += rq->mpwqe.umr_wqebbs;
sq->doorbell_cseg = &umr_wqe->ctrl;
@@ -1805,8 +1806,8 @@ static void mlx5e_handle_rx_cqe_mpwrq_re
struct mlx5e_mpw_info *wi = mlx5e_get_mpw_info(rq, wqe_id);
u16 stride_ix = mpwrq_get_cqe_stride_index(cqe);
u32 wqe_offset = stride_ix << rq->mpwqe.log_stride_sz;
- u32 head_offset = wqe_offset & (PAGE_SIZE - 1);
- u32 page_idx = wqe_offset >> PAGE_SHIFT;
+ u32 head_offset = wqe_offset & ((1 << rq->mpwqe.page_shift) - 1);
+ u32 page_idx = wqe_offset >> rq->mpwqe.page_shift;
struct mlx5e_rx_wqe_ll *wqe;
struct mlx5_wq_ll *wq;
struct sk_buff *skb;
@@ -1863,6 +1864,7 @@ mlx5e_fill_skb_data(struct sk_buff *skb,
net_prefetchw(skb->data);
while (data_bcnt) {
+ /* Non-linear mode, hence non-XSK, which always uses PAGE_SIZE. */
u32 pg_consumed_bytes = min_t(u32, PAGE_SIZE - data_offset, data_bcnt);
unsigned int truesize;
@@ -1900,6 +1902,7 @@ mlx5e_skb_from_cqe_mpwrq_nonlinear(struc
net_prefetchw(skb->data);
+ /* Non-linear mode, hence non-XSK, which always uses PAGE_SIZE. */
if (unlikely(frag_offset >= PAGE_SIZE)) {
di++;
frag_offset -= PAGE_SIZE;
@@ -2157,8 +2160,8 @@ static void mlx5e_handle_rx_cqe_mpwrq(st
struct mlx5e_mpw_info *wi = mlx5e_get_mpw_info(rq, wqe_id);
u16 stride_ix = mpwrq_get_cqe_stride_index(cqe);
u32 wqe_offset = stride_ix << rq->mpwqe.log_stride_sz;
- u32 head_offset = wqe_offset & (PAGE_SIZE - 1);
- u32 page_idx = wqe_offset >> PAGE_SHIFT;
+ u32 head_offset = wqe_offset & ((1 << rq->mpwqe.page_shift) - 1);
+ u32 page_idx = wqe_offset >> rq->mpwqe.page_shift;
struct mlx5e_rx_wqe_ll *wqe;
struct mlx5_wq_ll *wq;
struct sk_buff *skb;