Prior to this patch the maximal value for max WQEBBs (WQE Basic Blocks,
where WQE is a Work Queue Element) on the TX side was assumed to be 16
(fixed value). All firmware versions till today comply to this. In order
to be more flexible and resilient, read from FW the corresponding:
max_wqe_sz_sq. This value describes the maximum WQE size given in bytes,
thus max WQEBBs is given by the division in WQEBB's byte size. The
driver uses the top between 16 and the division result. This ensures
synchronization between driver and firmware and avoids unexpected
behavior. Store this value on the different SQs (Send Queues) for easy
access.
Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
min_t(int, mlx5_comp_vectors_count(mdev), MLX5E_MAX_NUM_CHANNELS);
}
min_t(int, mlx5_comp_vectors_count(mdev), MLX5E_MAX_NUM_CHANNELS);
}
+/* The maximum WQE size can be retrieved by max_wqe_sz_sq in
+ * bytes units. Driver hardens the limitation to 1KB (16
+ * WQEBBs), unless firmware capability is stricter.
+ */
+static inline u16 mlx5e_get_max_sq_wqebbs(struct mlx5_core_dev *mdev)
+{
+ return min_t(u16, MLX5_SEND_WQE_MAX_WQEBBS,
+ MLX5_CAP_GEN(mdev, max_wqe_sz_sq) / MLX5_SEND_WQE_BB);
+}
+
struct mlx5e_tx_wqe {
struct mlx5_wqe_ctrl_seg ctrl;
struct mlx5_wqe_eth_seg eth;
struct mlx5e_tx_wqe {
struct mlx5_wqe_ctrl_seg ctrl;
struct mlx5_wqe_eth_seg eth;
struct work_struct recover_work;
struct mlx5e_ptpsq *ptpsq;
cqe_ts_to_ns ptp_cyc2time;
struct work_struct recover_work;
struct mlx5e_ptpsq *ptpsq;
cqe_ts_to_ns ptp_cyc2time;
} ____cacheline_aligned_in_smp;
struct mlx5e_dma_info {
} ____cacheline_aligned_in_smp;
struct mlx5e_dma_info {
u32 sqn;
struct device *pdev;
__be32 mkey_be;
u32 sqn;
struct device *pdev;
__be32 mkey_be;
u8 min_inline_mode;
unsigned long state;
unsigned int hw_mtu;
u8 min_inline_mode;
unsigned long state;
unsigned int hw_mtu;
/* control path */
struct mlx5_wq_ctrl wq_ctrl;
struct mlx5e_channel *channel;
/* control path */
struct mlx5_wq_ctrl wq_ctrl;
struct mlx5e_channel *channel;
} ____cacheline_aligned_in_smp;
struct mlx5e_ktls_resync_resp;
} ____cacheline_aligned_in_smp;
struct mlx5e_ktls_resync_resp;
/* control path */
struct mlx5_wq_ctrl wq_ctrl;
struct mlx5e_channel *channel;
/* control path */
struct mlx5_wq_ctrl wq_ctrl;
struct mlx5e_channel *channel;
struct work_struct recover_work;
} ____cacheline_aligned_in_smp;
struct work_struct recover_work;
} ____cacheline_aligned_in_smp;
u16 stop_room;
stop_room = mlx5e_tls_get_stop_room(mdev, params);
u16 stop_room;
stop_room = mlx5e_tls_get_stop_room(mdev, params);
- stop_room += mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+ stop_room += mlx5e_stop_room_for_max_wqe(mdev);
if (is_mpwqe)
/* A MPWQE can take up to the maximum-sized WQE + all the normal
* stop room can be taken if a new packet breaks the active
* MPWQE session and allocates its WQEs right away.
*/
if (is_mpwqe)
/* A MPWQE can take up to the maximum-sized WQE + all the normal
* stop room can be taken if a new packet breaks the active
* MPWQE session and allocates its WQEs right away.
*/
- stop_room += mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+ stop_room += mlx5e_stop_room_for_max_wqe(mdev);
void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
mlx5e_build_sq_param_common(mdev, param);
void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
mlx5e_build_sq_param_common(mdev, param);
- param->stop_room = mlx5e_stop_room_for_wqe(1); /* for XSK NOP */
+ param->stop_room = mlx5e_stop_room_for_wqe(mdev, 1); /* for XSK NOP */
param->is_tls = mlx5e_accel_is_ktls_rx(mdev);
if (param->is_tls)
param->is_tls = mlx5e_accel_is_ktls_rx(mdev);
if (param->is_tls)
- param->stop_room += mlx5e_stop_room_for_wqe(1); /* for TLS RX resync NOP */
+ param->stop_room += mlx5e_stop_room_for_wqe(mdev, 1); /* for TLS RX resync NOP */
MLX5_SET(sqc, sqc, reg_umr, MLX5_CAP_ETH(mdev, reg_umr_sq));
MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
mlx5e_build_ico_cq_param(mdev, log_wq_size, ¶m->cqp);
MLX5_SET(sqc, sqc, reg_umr, MLX5_CAP_ETH(mdev, reg_umr_sq));
MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
mlx5e_build_ico_cq_param(mdev, log_wq_size, ¶m->cqp);
wq = MLX5_ADDR_OF(sqc, sqc, wq);
MLX5_SET(wq, wq, log_wq_sz, params->log_sq_size);
wq = MLX5_ADDR_OF(sqc, sqc, wq);
MLX5_SET(wq, wq, log_wq_sz, params->log_sq_size);
- param->stop_room = mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+ param->stop_room = mlx5e_stop_room_for_max_wqe(mdev);
mlx5e_build_tx_cq_param(mdev, params, ¶m->cqp);
}
mlx5e_build_tx_cq_param(mdev, params, ¶m->cqp);
}
-static inline u16 mlx5e_stop_room_for_wqe(u16 wqe_size)
-{
- BUILD_BUG_ON(PAGE_SIZE / MLX5_SEND_WQE_BB < MLX5_SEND_WQE_MAX_WQEBBS);
+#define MLX5E_STOP_ROOM(wqebbs) ((wqebbs) * 2 - 1)
+static inline u16 mlx5e_stop_room_for_wqe(struct mlx5_core_dev *mdev, u16 wqe_size)
+{
/* A WQE must not cross the page boundary, hence two conditions:
* 1. Its size must not exceed the page size.
* 2. If the WQE size is X, and the space remaining in a page is less
/* A WQE must not cross the page boundary, hence two conditions:
* 1. Its size must not exceed the page size.
* 2. If the WQE size is X, and the space remaining in a page is less
* stop room of X-1 + X.
* WQE size is also limited by the hardware limit.
*/
* stop room of X-1 + X.
* WQE size is also limited by the hardware limit.
*/
+ WARN_ONCE(wqe_size > mlx5e_get_max_sq_wqebbs(mdev),
+ "wqe_size %u is greater than max SQ WQEBBs %u",
+ wqe_size, mlx5e_get_max_sq_wqebbs(mdev));
- if (__builtin_constant_p(wqe_size))
- BUILD_BUG_ON(wqe_size > MLX5_SEND_WQE_MAX_WQEBBS);
- else
- WARN_ON_ONCE(wqe_size > MLX5_SEND_WQE_MAX_WQEBBS);
- return wqe_size * 2 - 1;
+ return MLX5E_STOP_ROOM(wqe_size);
+}
+
+static inline u16 mlx5e_stop_room_for_max_wqe(struct mlx5_core_dev *mdev)
+{
+ return MLX5E_STOP_ROOM(mlx5e_get_max_sq_wqebbs(mdev));
}
static inline bool mlx5e_icosq_can_post_wqe(struct mlx5e_icosq *sq, u16 wqe_size)
{
}
static inline bool mlx5e_icosq_can_post_wqe(struct mlx5e_icosq *sq, u16 wqe_size)
{
- u16 room = sq->reserved_room + mlx5e_stop_room_for_wqe(wqe_size);
+ u16 room = sq->reserved_room;
+
+ WARN_ONCE(wqe_size > sq->max_sq_wqebbs,
+ "wqe_size %u is greater than max SQ WQEBBs %u",
+ wqe_size, sq->max_sq_wqebbs);
+
+ room += MLX5E_STOP_ROOM(wqe_size);
return mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc, room);
}
return mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc, room);
}
INDIRECT_CALLABLE_SCOPE int mlx5e_xmit_xdp_frame_check_mpwqe(struct mlx5e_xdpsq *sq)
{
if (unlikely(!sq->mpwqe.wqe)) {
INDIRECT_CALLABLE_SCOPE int mlx5e_xmit_xdp_frame_check_mpwqe(struct mlx5e_xdpsq *sq)
{
if (unlikely(!sq->mpwqe.wqe)) {
- const u16 stop_room = mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
-
if (unlikely(!mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc,
if (unlikely(!mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc,
/* SQ is full, ring doorbell */
mlx5e_xmit_xdp_doorbell(sq);
sq->stats->full++;
/* SQ is full, ring doorbell */
mlx5e_xmit_xdp_doorbell(sq);
sq->stats->full++;
num_dumps = mlx5e_ktls_dumps_num_wqes(params, MAX_SKB_FRAGS, TLS_MAX_PAYLOAD_SIZE);
num_dumps = mlx5e_ktls_dumps_num_wqes(params, MAX_SKB_FRAGS, TLS_MAX_PAYLOAD_SIZE);
- stop_room += mlx5e_stop_room_for_wqe(MLX5E_TLS_SET_STATIC_PARAMS_WQEBBS);
- stop_room += mlx5e_stop_room_for_wqe(MLX5E_TLS_SET_PROGRESS_PARAMS_WQEBBS);
- stop_room += num_dumps * mlx5e_stop_room_for_wqe(MLX5E_KTLS_DUMP_WQEBBS);
+ stop_room += mlx5e_stop_room_for_wqe(mdev, MLX5E_TLS_SET_STATIC_PARAMS_WQEBBS);
+ stop_room += mlx5e_stop_room_for_wqe(mdev, MLX5E_TLS_SET_PROGRESS_PARAMS_WQEBBS);
+ stop_room += num_dumps * mlx5e_stop_room_for_wqe(mdev, MLX5E_KTLS_DUMP_WQEBBS);
/* FPGA */
/* Resync SKB. */
/* FPGA */
/* Resync SKB. */
- return mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+ return mlx5e_stop_room_for_max_wqe(mdev);
bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
{
bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
{
- bool striding_rq_umr = MLX5_CAP_GEN(mdev, striding_rq) &&
- MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
- MLX5_CAP_ETH(mdev, reg_umr_sq);
- u16 max_wqe_sz_cap = MLX5_CAP_GEN(mdev, max_wqe_sz_sq);
- bool inline_umr = MLX5E_UMR_WQE_INLINE_SZ <= max_wqe_sz_cap;
+ bool striding_rq_umr, inline_umr;
+ u16 max_wqe_sz_cap;
+ 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_wqebbs(mdev) * MLX5_SEND_WQE_BB;
+ inline_umr = max_wqe_sz_cap >= MLX5E_UMR_WQE_INLINE_SZ;
if (!striding_rq_umr)
return false;
if (!inline_umr) {
if (!striding_rq_umr)
return false;
if (!inline_umr) {
is_redirect ?
&c->priv->channel_stats[c->ix]->xdpsq :
&c->priv->channel_stats[c->ix]->rq_xdpsq;
is_redirect ?
&c->priv->channel_stats[c->ix]->xdpsq :
&c->priv->channel_stats[c->ix]->rq_xdpsq;
+ sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
+ sq->stop_room = MLX5E_STOP_ROOM(sq->max_sq_wqebbs);
param->wq.db_numa_node = cpu_to_node(c->cpu);
err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
param->wq.db_numa_node = cpu_to_node(c->cpu);
err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
sq->channel = c;
sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
sq->reserved_room = param->stop_room;
sq->channel = c;
sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
sq->reserved_room = param->stop_room;
+ sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
param->wq.db_numa_node = cpu_to_node(c->cpu);
err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
param->wq.db_numa_node = cpu_to_node(c->cpu);
err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, wq, &sq->wq_ctrl);
sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
sq->min_inline_mode = params->tx_min_inline_mode;
sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
sq->uar_map = mdev->mlx5e_res.hw_objs.bfreg.map;
sq->min_inline_mode = params->tx_min_inline_mode;
sq->hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
+ sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);