Netdev List

* [PATCH net-next 12/12] qed: Add iWARP support for fpdu spanned over more than two tcp packets
From: Michal Kalderon @ 2017-10-02  8:23 UTC (permalink / raw)
  To: davem-fT/PcQaiUtIeIZ0/mPfg9Q
  Cc: netdev-u79uwXL29TY76Z2rM5mHXA, linux-rdma-u79uwXL29TY76Z2rM5mHXA,
	dledford-H+wXaHxf7aLQT0dZR+AlfA, Michal Kalderon, Ariel Elior
In-Reply-To: <1506932638-26268-1-git-send-email-Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>

We continue to maintain a maximum of three buffers per fpdu, to ensure
that there are enough buffers for additional unaligned mpa packets.
To support this, if a fpdu is split over more than two tcp packets, we
use an intermediate buffer to copy the data to the previous buffer, then
we can release the data. We need an intermediate buffer as the initial
buffer partial packet could be located at the end of the packet, not
leaving room for additional data. This is a corner case, and will usually
not be the case.

Signed-off-by: Michal Kalderon <Michal.Kalderon-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
Signed-off-by: Ariel Elior <Ariel.Elior-YGCgFSpz5w/QT0dZR+AlfA@public.gmane.org>
---
 drivers/net/ethernet/qlogic/qed/qed_iwarp.c | 193 ++++++++++++++++++++++++++++
 drivers/net/ethernet/qlogic/qed/qed_iwarp.h |   1 +
 2 files changed, 194 insertions(+)

diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c
index 2994942..b2b1f87 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c
@@ -1420,6 +1420,7 @@ void qed_iwarp_resc_free(struct qed_hwfn *p_hwfn)
 	qed_rdma_bmap_free(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, 1);
 	kfree(iwarp_info->mpa_bufs);
 	kfree(iwarp_info->partial_fpdus);
+	kfree(iwarp_info->mpa_intermediate_buf);
 }
 
 int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams)
@@ -1762,6 +1763,11 @@ enum qed_iwarp_mpa_pkt_type {
 	"QED_IWARP_MPA_PKT_UNALIGNED"
 };
 
+static int
+qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn,
+		      struct qed_iwarp_fpdu *fpdu,
+		      struct qed_iwarp_ll2_buff *buf);
+
 static enum qed_iwarp_mpa_pkt_type
 qed_iwarp_mpa_classify(struct qed_hwfn *p_hwfn,
 		       struct qed_iwarp_fpdu *fpdu,
@@ -1822,6 +1828,68 @@ enum qed_iwarp_mpa_pkt_type {
 	fpdu->mpa_frag_len = fpdu->fpdu_length - fpdu->incomplete_bytes;
 }
 
+static int
+qed_iwarp_cp_pkt(struct qed_hwfn *p_hwfn,
+		 struct qed_iwarp_fpdu *fpdu,
+		 struct unaligned_opaque_data *pkt_data,
+		 struct qed_iwarp_ll2_buff *buf, u16 tcp_payload_size)
+{
+	u8 *tmp_buf = p_hwfn->p_rdma_info->iwarp.mpa_intermediate_buf;
+	int rc;
+
+	/* need to copy the data from the partial packet stored in fpdu
+	 * to the new buf, for this we also need to move the data currently
+	 * placed on the buf. The assumption is that the buffer is big enough
+	 * since fpdu_length <= mss, we use an intermediate buffer since
+	 * we may need to copy the new data to an overlapping location
+	 */
+	if ((fpdu->mpa_frag_len + tcp_payload_size) > (u16)buf->buff_size) {
+		DP_ERR(p_hwfn,
+		       "MPA ALIGN: Unexpected: buffer is not large enough for split fpdu buff_size = %d mpa_frag_len = %d, tcp_payload_size = %d, incomplete_bytes = %d\n",
+		       buf->buff_size, fpdu->mpa_frag_len,
+		       tcp_payload_size, fpdu->incomplete_bytes);
+		return -EINVAL;
+	}
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+		   "MPA ALIGN Copying fpdu: [%p, %d] [%p, %d]\n",
+		   fpdu->mpa_frag_virt, fpdu->mpa_frag_len,
+		   (u8 *)(buf->data) + pkt_data->first_mpa_offset,
+		   tcp_payload_size);
+
+	memcpy(tmp_buf, fpdu->mpa_frag_virt, fpdu->mpa_frag_len);
+	memcpy(tmp_buf + fpdu->mpa_frag_len,
+	       (u8 *)(buf->data) + pkt_data->first_mpa_offset,
+	       tcp_payload_size);
+
+	rc = qed_iwarp_recycle_pkt(p_hwfn, fpdu, fpdu->mpa_buf);
+	if (rc)
+		return rc;
+
+	/* If we managed to post the buffer copy the data to the new buffer
+	 * o/w this will occur in the next round...
+	 */
+	memcpy((u8 *)(buf->data), tmp_buf,
+	       fpdu->mpa_frag_len + tcp_payload_size);
+
+	fpdu->mpa_buf = buf;
+	/* fpdu->pkt_hdr remains as is */
+	/* fpdu->mpa_frag is overridden with new buf */
+	fpdu->mpa_frag = buf->data_phys_addr;
+	fpdu->mpa_frag_virt = buf->data;
+	fpdu->mpa_frag_len += tcp_payload_size;
+
+	fpdu->incomplete_bytes -= tcp_payload_size;
+
+	DP_VERBOSE(p_hwfn,
+		   QED_MSG_RDMA,
+		   "MPA ALIGN: split fpdu buff_size = %d mpa_frag_len = %d, tcp_payload_size = %d, incomplete_bytes = %d\n",
+		   buf->buff_size, fpdu->mpa_frag_len, tcp_payload_size,
+		   fpdu->incomplete_bytes);
+
+	return 0;
+}
+
 static void
 qed_iwarp_update_fpdu_length(struct qed_hwfn *p_hwfn,
 			     struct qed_iwarp_fpdu *fpdu, u8 *mpa_data)
@@ -1843,6 +1911,90 @@ enum qed_iwarp_mpa_pkt_type {
 	}
 }
 
+#define QED_IWARP_IS_RIGHT_EDGE(_curr_pkt) \
+	(GET_FIELD((_curr_pkt)->flags,	   \
+		   UNALIGNED_OPAQUE_DATA_PKT_REACHED_WIN_RIGHT_EDGE))
+
+/* This function is used to recycle a buffer using the ll2 drop option. It
+ * uses the mechanism to ensure that all buffers posted to tx before this one
+ * were completed. The buffer sent here will be sent as a cookie in the tx
+ * completion function and can then be reposted to rx chain when done. The flow
+ * that requires this is the flow where a FPDU splits over more than 3 tcp
+ * segments. In this case the driver needs to re-post a rx buffer instead of
+ * the one received, but driver can't simply repost a buffer it copied from
+ * as there is a case where the buffer was originally a packed FPDU, and is
+ * partially posted to FW. Driver needs to ensure FW is done with it.
+ */
+static int
+qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn,
+		      struct qed_iwarp_fpdu *fpdu,
+		      struct qed_iwarp_ll2_buff *buf)
+{
+	struct qed_ll2_tx_pkt_info tx_pkt;
+	u8 ll2_handle;
+	int rc;
+
+	memset(&tx_pkt, 0, sizeof(tx_pkt));
+	tx_pkt.num_of_bds = 1;
+	tx_pkt.tx_dest = QED_LL2_TX_DEST_DROP;
+	tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2;
+	tx_pkt.first_frag = fpdu->pkt_hdr;
+	tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
+	buf->piggy_buf = NULL;
+	tx_pkt.cookie = buf;
+
+	ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
+
+	rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
+	if (rc)
+		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+			   "Can't drop packet rc=%d\n", rc);
+
+	DP_VERBOSE(p_hwfn,
+		   QED_MSG_RDMA,
+		   "MPA_ALIGN: send drop tx packet [%lx, 0x%x], buf=%p, rc=%d\n",
+		   (unsigned long int)tx_pkt.first_frag,
+		   tx_pkt.first_frag_len, buf, rc);
+
+	return rc;
+}
+
+static int
+qed_iwarp_win_right_edge(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu)
+{
+	struct qed_ll2_tx_pkt_info tx_pkt;
+	u8 ll2_handle;
+	int rc;
+
+	memset(&tx_pkt, 0, sizeof(tx_pkt));
+	tx_pkt.num_of_bds = 1;
+	tx_pkt.tx_dest = QED_LL2_TX_DEST_LB;
+	tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2;
+
+	tx_pkt.first_frag = fpdu->pkt_hdr;
+	tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
+	tx_pkt.enable_ip_cksum = true;
+	tx_pkt.enable_l4_cksum = true;
+	tx_pkt.calc_ip_len = true;
+	/* vlan overload with enum iwarp_ll2_tx_queues */
+	tx_pkt.vlan = IWARP_LL2_ALIGNED_RIGHT_TRIMMED_TX_QUEUE;
+
+	ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
+
+	rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
+	if (rc)
+		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+			   "Can't send right edge rc=%d\n", rc);
+	DP_VERBOSE(p_hwfn,
+		   QED_MSG_RDMA,
+		   "MPA_ALIGN: Sent right edge FPDU num_bds=%d [%lx, 0x%x], rc=%d\n",
+		   tx_pkt.num_of_bds,
+		   (unsigned long int)tx_pkt.first_frag,
+		   tx_pkt.first_frag_len, rc);
+
+	return rc;
+}
+
 static int
 qed_iwarp_send_fpdu(struct qed_hwfn *p_hwfn,
 		    struct qed_iwarp_fpdu *fpdu,
@@ -1971,6 +2123,20 @@ enum qed_iwarp_mpa_pkt_type {
 					    mpa_buf->tcp_payload_len,
 					    mpa_buf->placement_offset);
 
+			if (!QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) {
+				mpa_buf->tcp_payload_len = 0;
+				break;
+			}
+
+			rc = qed_iwarp_win_right_edge(p_hwfn, fpdu);
+
+			if (rc) {
+				DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+					   "Can't send FPDU:reset rc=%d\n", rc);
+				memset(fpdu, 0, sizeof(*fpdu));
+				break;
+			}
+
 			mpa_buf->tcp_payload_len = 0;
 			break;
 		case QED_IWARP_MPA_PKT_PACKED:
@@ -1994,6 +2160,28 @@ enum qed_iwarp_mpa_pkt_type {
 			break;
 		case QED_IWARP_MPA_PKT_UNALIGNED:
 			qed_iwarp_update_fpdu_length(p_hwfn, fpdu, mpa_data);
+			if (mpa_buf->tcp_payload_len < fpdu->incomplete_bytes) {
+				/* special handling of fpdu split over more
+				 * than 2 segments
+				 */
+				if (QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) {
+					rc = qed_iwarp_win_right_edge(p_hwfn,
+								      fpdu);
+					/* packet will be re-processed later */
+					if (rc)
+						return rc;
+				}
+
+				rc = qed_iwarp_cp_pkt(p_hwfn, fpdu, curr_pkt,
+						      buf,
+						      mpa_buf->tcp_payload_len);
+				if (rc) /* packet will be re-processed later */
+					return rc;
+
+				mpa_buf->tcp_payload_len = 0;
+				break;
+			}
+
 			rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf,
 						 mpa_buf->tcp_payload_len,
 						 pkt_type);
@@ -2510,6 +2698,11 @@ static int qed_iwarp_ll2_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
 		goto err;
 
 	iwarp_info->max_num_partial_fpdus = (u16)p_hwfn->p_rdma_info->num_qps;
+
+	iwarp_info->mpa_intermediate_buf = kzalloc(mpa_buff_size, GFP_KERNEL);
+	if (!iwarp_info->mpa_intermediate_buf)
+		goto err;
+
 	/* The mpa_bufs array serves for pending RX packets received on the
 	 * mpa ll2 that don't have place on the tx ring and require later
 	 * processing. We can't fail on allocation of such a struct therefore
diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h
index c58793a..c1ecd74 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h
@@ -107,6 +107,7 @@ struct qed_iwarp_info {
 	enum mpa_rtr_type rtr_type;
 	struct qed_iwarp_fpdu *partial_fpdus;
 	struct qed_iwarp_ll2_mpa_buf *mpa_bufs;
+	u8 *mpa_intermediate_buf;
 	u16 max_num_partial_fpdus;
 };
 
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply related