issue: 1792164 Introduce flags to process zero copy send

There flags are added: VMA_TX_PACKET_ZEROCOPY - to use on sockinfo/dst_entry layers TCP_WRITE_ZEROCOPY - to use inside lwip tcp_write TF_SEG_OPTS_ZEROCOPY - to mark tcp segment with zero copy attribute Signed-off-by: Igor Ivanov <[email protected]>
Mellanox · Jun 11, 2020 · 687271f · 687271f
1 parent a99fbf2
commit 687271f
Show file tree

Hide file tree

Showing 5 changed files with 32 additions and 5 deletions.
diff --git a/src/vma/lwip/tcp.h b/src/vma/lwip/tcp.h
@@ -528,6 +528,7 @@ err_t            tcp_shutdown(struct tcp_pcb *pcb, int shut_rx, int shut_tx);
 #define TCP_WRITE_DUMMY     0x10
 #define TCP_WRITE_TSO       0x20
 #define TCP_WRITE_FILE      0x40
+#define TCP_WRITE_ZEROCOPY  0x80
 
 err_t            tcp_write   (struct tcp_pcb *pcb, const void *dataptr, u32_t len,
                               u8_t apiflags);

diff --git a/src/vma/lwip/tcp_impl.h b/src/vma/lwip/tcp_impl.h
@@ -305,6 +305,7 @@ struct tcp_seg {
 #define TF_SEG_OPTS_WNDSCALE	(u8_t)0x08U /* Include window scaling option */
 #define TF_SEG_OPTS_DUMMY_MSG	(u8_t)TCP_WRITE_DUMMY /* Include dummy send option */
 #define TF_SEG_OPTS_TSO         (u8_t)TCP_WRITE_TSO /* Use TSO send mode */
+#define TF_SEG_OPTS_ZEROCOPY    (u8_t)TCP_WRITE_ZEROCOPY /* Use zerocopy send mode */
 
   struct tcp_hdr *tcphdr;  /* the TCP header */
 };

diff --git a/src/vma/proto/dst_entry_tcp.cpp b/src/vma/proto/dst_entry_tcp.cpp
@@ -81,6 +81,11 @@ ssize_t dst_entry_tcp::fast_send(const iovec* p_iov, const ssize_t sz_iov, vma_s
 
 	p_tcp_iov = (tcp_iovec*)p_iov;
 
+	/* Suppress flags that should not be used anymore
+	 * to avoid conflicts with VMA_TX_PACKET_L3_CSUM and VMA_TX_PACKET_L4_CSUM
+	 */
+	attr.flags = (vma_wr_tx_packet_attr)(attr.flags & ~(VMA_TX_PACKET_ZEROCOPY | VMA_TX_FILE));
+
 	attr.flags = (vma_wr_tx_packet_attr)(attr.flags | VMA_TX_PACKET_L3_CSUM | VMA_TX_PACKET_L4_CSUM);
 
 	/* Supported scenarios:

diff --git a/src/vma/proto/dst_entry_udp.cpp b/src/vma/proto/dst_entry_udp.cpp
@@ -308,6 +308,12 @@ ssize_t dst_entry_udp::fast_send(const iovec* p_iov, const ssize_t sz_iov, vma_s
 {
 	// Calc user data payload size
 	ssize_t sz_data_payload = 0;
+
+	/* Suppress flags that should not be used anymore
+	 * to avoid conflicts with VMA_TX_PACKET_L3_CSUM and VMA_TX_PACKET_L4_CSUM
+	 */
+	attr.flags = (vma_wr_tx_packet_attr)(attr.flags & ~(VMA_TX_PACKET_ZEROCOPY | VMA_TX_FILE));
+
 	for (ssize_t i = 0; i < sz_iov; i++)
 		sz_data_payload += p_iov[i].iov_len;
 
@@ -324,7 +330,7 @@ ssize_t dst_entry_udp::fast_send(const iovec* p_iov, const ssize_t sz_iov, vma_s
 		attr.flags = (vma_wr_tx_packet_attr)(attr.flags | VMA_TX_PACKET_L3_CSUM | VMA_TX_PACKET_L4_CSUM);
 		return fast_send_not_fragmented(p_iov, sz_iov, attr.flags, sz_udp_payload, sz_data_payload);
 	} else {
-                attr.flags = (vma_wr_tx_packet_attr)(attr.flags | VMA_TX_PACKET_L3_CSUM);
+		attr.flags = (vma_wr_tx_packet_attr)(attr.flags | VMA_TX_PACKET_L3_CSUM);
 		return fast_send_fragmented(p_iov, sz_iov, attr.flags, sz_udp_payload, sz_data_payload);
 	}
 }

diff --git a/src/vma/proto/vma_lwip.h b/src/vma/proto/vma_lwip.h
@@ -51,11 +51,25 @@ typedef enum vma_wr_tx_packet_attr {
 	VMA_TX_PACKET_TSO  = TCP_WRITE_TSO, /* 0x20 */
 	/* sendfile operation. */
 	VMA_TX_FILE  = TCP_WRITE_FILE, /* 0x40 */
+	/* zcopy write operation (MSG_ZEROCOPY). */
+	VMA_TX_PACKET_ZEROCOPY = TCP_WRITE_ZEROCOPY, /* 0x80 */
 
-	/* MLX5_ETH_WQE_L3_CSUM offload to HW L3 (IP) header checksum */
-	VMA_TX_PACKET_L3_CSUM = (1 << 6), /* hardcoded values. It is the same as VMA_TX_FILE but there is no conflict */
-	/* MLX5_ETH_WQE_L4_CSUM offload to HW L4 (TCP/UDP) header checksum */
-	VMA_TX_PACKET_L4_CSUM = (1 << 7), /* hardcoded values */
+	/* MLX5_ETH_WQE_L3_CSUM offload to HW L3 (IP) header checksum
+	 * Important:
+	 *  - hardcoded value used directly to program send to wire
+	 *  - it is the same as VMA_TX_FILE but there is no conflict as far as
+	 *    VMA_TX_FILE is passed into dst_entry::fast_send() operation
+	 *    and it is not needed later doing send to wire
+	 */
+	VMA_TX_PACKET_L3_CSUM = (1 << 6),
+	/* MLX5_ETH_WQE_L4_CSUM offload to HW L4 (TCP/UDP) header checksum
+	 * Important:
+	 *  - hardcoded value used directly to program send to wire
+	 *  - it is the same as TCP_WRITE_ZEROCOPY but there is no conflict as far as
+	 *    TCP_WRITE_ZEROCOPY is passed into dst_entry::fast_send() operation
+	 *    and it is not needed later doing send to wire
+	 */
+	VMA_TX_PACKET_L4_CSUM = (1 << 7),
 	/* blocking send operation */
 	VMA_TX_PACKET_BLOCK   = (1 << 8),
 	/* Force SW checksum */