static UCS_F_ALWAYS_INLINE void
uct_rc_mlx5_post_send(uct_rc_mlx5_ep_t *ep, struct mlx5_wqe_ctrl_seg *ctrl,
uint8_t opcode, uint8_t opmod, unsigned sig_flag, unsigned wqe_size)
{
unsigned n, num_seg, num_bb;
void *src, *dst;
uint16_t sw_pi;
num_seg = ucs_div_round_up(wqe_size, UCT_IB_MLX5_WQE_SEG_SIZE);
num_bb = ucs_div_round_up(wqe_size, MLX5_SEND_WQE_BB);
sw_pi = ep->tx.sw_pi;
uct_rc_mlx5_set_ctrl_seg(ctrl, sw_pi, opcode, opmod, ep->qp_num, sig_flag,
num_seg);
uct_ib_mlx5_log_tx(IBV_QPT_RC, ctrl, ep->tx.qstart, ep->tx.qend,
(opcode == MLX5_OPCODE_SEND) ? uct_rc_ep_am_packet_dump : NULL);
/* TODO Put memory store fence here too, to prevent WC being flushed after DBrec */
ucs_memory_cpu_store_fence();
/* Write doorbell record */
ep->tx.prev_sw_pi = sw_pi;
*ep->tx.dbrec = htonl(sw_pi += num_bb);
/* Make sure that doorbell record is written before ringing the doorbell */
ucs_memory_bus_store_fence();
/* Set up copy pointers */
dst = ep->tx.bf_reg;
src = ctrl;
/* BF copy */
/* TODO support DB without BF */
ucs_assert(wqe_size <= ep->tx.bf_size);
ucs_assert(num_bb <= UCT_RC_MLX5_MAX_BB);
for (n = 0; n < num_bb; ++n) {
uct_rc_mlx5_bf_copy_bb(dst, src);
dst += MLX5_SEND_WQE_BB;
src += MLX5_SEND_WQE_BB;
if (ucs_unlikely(src == ep->tx.qend)) {
src = ep->tx.qstart;
}
}
/* We don't want the compiler to reorder instructions and hurt latency */
ucs_compiler_fence();
/* Advance queue pointer */
ucs_assert(ctrl == ep->tx.seg);
ep->tx.seg = src;
ep->tx.sw_pi = sw_pi;
/* Flip BF register */
ep->tx.bf_reg = (void*) ((uintptr_t) ep->tx.bf_reg ^ ep->tx.bf_size);
uct_rc_ep_tx_posted(&ep->super, sig_flag & MLX5_WQE_CTRL_CQ_UPDATE);
}
static ucs_status_t uct_ib_mlx5dv_create_ksm(uct_ib_md_t *ibmd,
uct_ib_mem_t *ib_memh,
off_t offset)
{
uct_ib_mlx5_mem_t *memh = ucs_derived_of(ib_memh, uct_ib_mlx5_mem_t);
uct_ib_mlx5_md_t *md = ucs_derived_of(ibmd, uct_ib_mlx5_md_t);
uint32_t out[UCT_IB_MLX5DV_ST_SZ_DW(create_mkey_out)] = {};
struct ibv_mr *mr = memh->super.mr;
ucs_status_t status = UCS_OK;
struct mlx5dv_pd dvpd = {};
struct mlx5dv_obj dv = {};
size_t reg_length, length, inlen;
int list_size, i;
void *mkc, *klm;
uint32_t *in;
intptr_t addr;
if (!(md->flags & UCT_IB_MLX5_MD_FLAG_KSM)) {
return UCS_ERR_UNSUPPORTED;
}
reg_length = UCT_IB_MD_MAX_MR_SIZE;
addr = (intptr_t)mr->addr & ~(reg_length - 1);
length = mr->length + (intptr_t)mr->addr - addr;
list_size = ucs_div_round_up(length, reg_length);
inlen = UCT_IB_MLX5DV_ST_SZ_BYTES(create_mkey_in) +
UCT_IB_MLX5DV_ST_SZ_BYTES(klm) * list_size;
in = ucs_calloc(1, inlen, "mkey mailbox");
if (in == NULL) {
return UCS_ERR_NO_MEMORY;
}
dv.pd.in = md->super.pd;
dv.pd.out = &dvpd;
mlx5dv_init_obj(&dv, MLX5DV_OBJ_PD);
UCT_IB_MLX5DV_SET(create_mkey_in, in, opcode, UCT_IB_MLX5_CMD_OP_CREATE_MKEY);
mkc = UCT_IB_MLX5DV_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
UCT_IB_MLX5DV_SET(mkc, mkc, access_mode_1_0, UCT_IB_MLX5_MKC_ACCESS_MODE_KSM);
UCT_IB_MLX5DV_SET(mkc, mkc, a, 1);
UCT_IB_MLX5DV_SET(mkc, mkc, rw, 1);
UCT_IB_MLX5DV_SET(mkc, mkc, rr, 1);
UCT_IB_MLX5DV_SET(mkc, mkc, lw, 1);
UCT_IB_MLX5DV_SET(mkc, mkc, lr, 1);
UCT_IB_MLX5DV_SET(mkc, mkc, pd, dvpd.pdn);
UCT_IB_MLX5DV_SET(mkc, mkc, translations_octword_size, list_size);
UCT_IB_MLX5DV_SET(mkc, mkc, log_entity_size, ucs_ilog2(reg_length));
UCT_IB_MLX5DV_SET(mkc, mkc, qpn, 0xffffff);
UCT_IB_MLX5DV_SET(mkc, mkc, mkey_7_0, offset & 0xff);
UCT_IB_MLX5DV_SET64(mkc, mkc, start_addr, addr + offset);
UCT_IB_MLX5DV_SET64(mkc, mkc, len, length);
UCT_IB_MLX5DV_SET(create_mkey_in, in, translations_octword_actual_size, list_size);
klm = UCT_IB_MLX5DV_ADDR_OF(create_mkey_in, in, klm_pas_mtt);
for (i = 0; i < list_size; i++) {
if (i == list_size - 1) {
UCT_IB_MLX5DV_SET(klm, klm, byte_count, length % reg_length);
} else {
UCT_IB_MLX5DV_SET(klm, klm, byte_count, reg_length);
}
UCT_IB_MLX5DV_SET(klm, klm, mkey, mr->lkey);
UCT_IB_MLX5DV_SET64(klm, klm, address, addr + (i * reg_length));
klm += UCT_IB_MLX5DV_ST_SZ_BYTES(klm);
}
memh->atomic_dvmr = mlx5dv_devx_obj_create(md->super.dev.ibv_context, in, inlen,
out, sizeof(out));
if (memh->atomic_dvmr == NULL) {
ucs_debug("CREATE_MKEY KSM failed: %m");
status = UCS_ERR_UNSUPPORTED;
md->flags &= ~UCT_IB_MLX5_MD_FLAG_KSM;
goto out;
}
memh->super.atomic_rkey =
(UCT_IB_MLX5DV_GET(create_mkey_out, out, mkey_index) << 8) |
(offset & 0xff);
ucs_debug("KSM registered memory %p..%p offset 0x%lx on %s rkey 0x%x",
mr->addr, mr->addr + mr->length, offset, uct_ib_device_name(&md->super.dev),
memh->super.atomic_rkey);
out:
ucs_free(in);
return status;
}