mlx4/mlx4_ib/mlx4_ib_qp.c

218 	__be32 *wqe;
233 			wqe = buf + (i & ((1 << qp->sq.wqe_shift) - 1));
234 			*wqe = stamp;
240 			wqe = buf + i;
241 			*wqe = cpu_to_be32(0xffffffff);
250 	void *wqe;
253 	ctrl = wqe = get_send_wqe(qp, n & (qp->sq.wqe_cnt - 1));
257 		struct mlx4_wqe_datagram_seg *dgram = wqe + sizeof *ctrl;
266 		inl = wqe + s;
2299 				  void *wqe, unsigned *mlx_seg_len)
2303 	struct mlx4_wqe_mlx_seg *mlx = wqe;
2304 	struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
2428 			    void *wqe, unsigned *mlx_seg_len)
2431 	struct mlx4_wqe_mlx_seg *mlx = wqe;
2432 	struct mlx4_wqe_ctrl_seg *ctrl = wqe;
2433 	struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
2806 static void build_tunnel_header(const struct ib_ud_wr *wr, void *wqe, unsigned *mlx_seg_len)
2808 	struct mlx4_wqe_inline_seg *inl = wqe;
2889 static int build_lso_seg(struct mlx4_wqe_lso_seg *wqe, const struct ib_ud_wr *wr,
2893 	unsigned halign = ALIGN(sizeof *wqe + wr->hlen, 16);
2902 	memcpy(wqe->header, wr->header, wr->hlen);
2924 static void add_zero_len_inline(void *wqe)
2926 	struct mlx4_wqe_inline_seg *inl = wqe;
2927 	memset(wqe, 0, 16);
2935 	void *wqe;
3001 		ctrl = wqe = get_send_wqe(qp, ind & (qp->sq.wqe_cnt - 1));
3016 		wqe += sizeof *ctrl;
3026 				set_raddr_seg(wqe, atomic_wr(wr)->remote_addr,
3028 				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
3030 				set_atomic_seg(wqe, atomic_wr(wr));
3031 				wqe  += sizeof (struct mlx4_wqe_atomic_seg);
3039 				set_raddr_seg(wqe, atomic_wr(wr)->remote_addr,
3041 				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
3043 				set_masked_atomic_seg(wqe, atomic_wr(wr));
3044 				wqe  += sizeof (struct mlx4_wqe_masked_atomic_seg);
3054 				set_raddr_seg(wqe, rdma_wr(wr)->remote_addr,
3056 				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
3063 				set_local_inv_seg(wqe, wr->ex.invalidate_rkey);
3064 				wqe  += sizeof (struct mlx4_wqe_local_inval_seg);
3071 				set_reg_seg(wqe, reg_wr(wr));
3072 				wqe  += sizeof(struct mlx4_wqe_fmr_seg);
3089 			wqe  += seglen;
3095 			set_datagram_seg(wqe, ud_wr(wr));
3097 			*(__be32 *) wqe |= cpu_to_be32(0x80000000);
3098 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
3102 			set_datagram_seg(wqe, ud_wr(wr));
3103 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
3107 				err = build_lso_seg(wqe, ud_wr(wr), qp, &seglen,
3113 				lso_wqe = (__be32 *) wqe;
3114 				wqe  += seglen;
3126 			wqe  += seglen;
3129 			add_zero_len_inline(wqe);
3130 			wqe += 16;
3132 			build_tunnel_header(ud_wr(wr), wqe, &seglen);
3133 			wqe  += seglen;
3142 			set_tunnel_datagram_seg(to_mdev(ibqp->device), wqe,
3145 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
3147 			build_tunnel_header(ud_wr(wr), wqe, &seglen);
3148 			wqe  += seglen;
3160 			wqe  += seglen;
3175 		dseg = wqe;
3226 		 * Same optimization applies to padding with NOP wqe