Moved timeouts into the udp transports.

Simplified the fast path checking in the fw, but it turns out this was not the issue.
Fixed some bad bit operations with the 16sc words (dont forget sign extension).
Added some more documentation to the headers....

1 files changed, 11 insertions, 5 deletions

diff --git a/firmware/microblaze/apps/txrx.c b/firmware/microblaze/apps/txrx.c
index 9c3caa4f2..1b3299150 100644
--- a/firmware/microblaze/apps/txrx.c
+++ b/firmware/microblaze/apps/txrx.c
@@ -504,16 +504,22 @@ void handle_udp_ctrl_packet(
 static bool
 eth_pkt_inspector(dbsm_t *sm, int bufno)
 {
-  //extract buffer point and length
+  //point me to the ethernet frame
   uint32_t *buff = (uint32_t *)buffer_ram(bufno);
-  size_t len = buffer_pool_status->last_line[bufno] - 3;
 
   //treat this as fast-path data?
-  if (is_udp_packet_with_vrt(buff, len, USRP2_UDP_DATA_PORT)){
-    return false;
-  }
+  // We have to do this operation as fast as possible.
+  // Therefore, we do not check all the headers,
+  // just check that the udp port matches
+  // and that the vrt header is non zero.
+  // In the future, a hardware state machine will do this...
+  if ( //warning! magic numbers approaching....
+      (((buff + ((2 + 14 + 20)/sizeof(uint32_t)))[0] & 0xffff) == USRP2_UDP_DATA_PORT) &&
+      ((buff + ((2 + 14 + 20 + 8)/sizeof(uint32_t)))[0] != 0)
+  ) return false;
 
   //pass it to the slow-path handler
+  size_t len = buffer_pool_status->last_line[bufno] - 3;
   handle_eth_packet(buff, len);
   return true;
 }