rdma: fix for-loop initialization in scalar path

When n_rx_packets is less then 16(VEC256) or 8(VEC128), code execution
will fall into scalar path of processing packets. But with a wrong
initialization value for n_left set to zero, i in the for-loop will
equal to n_rx_packets. This leads to the bypass of required ip4 checksum
validation and byte count endianness conversion in scalar path.
Besides, refactor the code using while instead of for-loop to keep
consistency with VPP code style.

Type: fix
Fixes: bf93670c515d ("rdma: fix ipv4 checksum check in rdma-input node")

Signed-off-by: Lijian Zhang <lijian.zhang@arm.com>
Signed-off-by: Jieqiang Wang <jieqiang.wang@arm.com>
Change-Id: Ib4e8cb5202735f8b060c99caddf26035657551e1

1 files changed, 35 insertions, 20 deletions

diff --git a/src/plugins/rdma/input.c b/src/plugins/rdma/input.c
index 5f2c0bbf31f..827fc577484 100644
--- a/src/plugins/rdma/input.c
+++ b/src/plugins/rdma/input.c
@@ -681,51 +681,66 @@ rdma_device_mlx5dv_l3_validate_and_swap_bc (rdma_per_thread_data_t
   /* verify that all ip4 packets have l3_ok flag set and convert packet
      length from network to host byte order */
   int skip_ip4_cksum = 1;
-  int n_left = 0;
+  int n_left = n_rx_packets;
+  u16 *cqe_flags = ptd->cqe_flags;
 
 #if defined CLIB_HAVE_VEC256
-  if (n_rx_packets >= 16)
+  if (n_left >= 16)
     {
       u16x16 mask16 = u16x16_splat (mask);
       u16x16 match16 = u16x16_splat (match);
       u16x16 r16 = {};
 
-      n_left = n_rx_packets % 16;
+      while (n_left >= 16)
+	{
+	  r16 |= (*(u16x16 *) cqe_flags & mask16) != match16;
+
+	  *(u32x8 *) bc = u32x8_byte_swap (*(u32x8 *) bc);
+	  *(u32x8 *) (bc + 8) = u32x8_byte_swap (*(u32x8 *) (bc + 8));
 
-      for (int i = 0; i < n_rx_packets / 16; i++)
-	r16 |= (ptd->cqe_flags16[i] & mask16) != match16;
+	  cqe_flags += 16;
+	  bc += 16;
+	  n_left -= 16;
+	}
 
       if (!u16x16_is_all_zero (r16))
 	skip_ip4_cksum = 0;
-
-      for (int i = 0; i < (n_rx_packets - n_left); i += 8)
-	*(u32x8 *) (bc + i) = u32x8_byte_swap (*(u32x8 *) (bc + i));
     }
 #elif defined CLIB_HAVE_VEC128
-  if (n_rx_packets >= 8)
+  if (n_left >= 8)
     {
       u16x8 mask8 = u16x8_splat (mask);
       u16x8 match8 = u16x8_splat (match);
       u16x8 r8 = {};
 
-      n_left = n_rx_packets % 8;
+      while (n_left >= 8)
+	{
+	  r8 |= (*(u16x8 *) cqe_flags & mask8) != match8;
+
+	  *(u32x4 *) bc = u32x4_byte_swap (*(u32x4 *) bc);
+	  *(u32x4 *) (bc + 4) = u32x4_byte_swap (*(u32x4 *) (bc + 4));
 
-      for (int i = 0; i < n_rx_packets / 8; i++)
-	r8 |= (ptd->cqe_flags8[i] & mask8) != match8;
+	  cqe_flags += 8;
+	  bc += 8;
+	  n_left -= 8;
+	}
 
       if (!u16x8_is_all_zero (r8))
 	skip_ip4_cksum = 0;
-
-      for (int i = 0; i < (n_rx_packets - n_left); i += 4)
-	*(u32x4 *) (bc + i) = u32x4_byte_swap (*(u32x4 *) (bc + i));
     }
 #endif
-  for (int i = (n_rx_packets - n_left); i < n_rx_packets; i++)
-    if ((ptd->cqe_flags[i] & mask) != match)
-      skip_ip4_cksum = 0;
 
-  for (int i = (n_rx_packets - n_left); i < n_rx_packets; i++)
-    bc[i] = clib_net_to_host_u32 (bc[i]);
+  while (n_left >= 1)
+    {
+      if ((cqe_flags[0] & mask) != match)
+	skip_ip4_cksum = 0;
+
+      bc[0] = clib_net_to_host_u32 (bc[0]);
+
+      cqe_flags += 1;
+      bc += 1;
+      n_left -= 1;
+    }
 
   return skip_ip4_cksum;
 }