Hi!
This updated patch against current CVS implements TCP segmentation offloading for RTL8139 in C+ mode.
I fixed a couple of problems in implementation (wrong sequence number calculation), and now TCP performance seem to be normal.
Dependency on slirp.h header is now gone.
Again tested with linux (ethtool -K eth0 tx on sg on tso on); please apply.
It turned out that a couple of variable types were left dependent on slirp.h
and transmit/receive pointers were not reset in right places (required for freebsd re driver)
This is an incremental update to fix these problems.
--- hw/rtl8139-20060709-with-working-tso-2.c 2006-07-09 12:41:47.000000000 +0400
+++ hw/rtl8139.c 2006-07-11 01:15:27.000000000 +0400
@@ -40,6 +40,7 @@
* 2006-Jul-09 Igor Kovalenko : Fixed TCP header length calculation while processing
* segmentation offloading
* Removed slirp.h dependency
+ * Added rx/tx buffer reset when enabling rx/tx operation
*/
#include "vl.h"
@@ -1368,10 +1369,14 @@
if (val & CmdRxEnb)
{
DEBUG_PRINT(("RTL8139: ChipCmd enable receiver\n"));
+
+ s->currCPlusRxDesc = 0;
}
if (val & CmdTxEnb)
{
DEBUG_PRINT(("RTL8139: ChipCmd enable transmitter\n"));
+
+ s->currCPlusTxDesc = 0;
}
/* mask unwriteable bits */
@@ -2078,8 +2083,8 @@
/* ip packet header */
ip_header *ip = 0;
int hlen = 0;
- u_int8_t ip_protocol = 0;
- u_int16_t ip_data_len = 0;
+ uint8_t ip_protocol = 0;
+ uint16_t ip_data_len = 0;
uint8_t *eth_payload_data = 0;
size_t eth_payload_len = 0;
@@ -2450,8 +2455,6 @@
DEBUG_PRINT(("RTL8139: TxAddr write offset=0x%x val=0x%08x\n", txAddrOffset, val));
s->TxAddr[txAddrOffset/4] = le32_to_cpu(val);
-
- s->currCPlusTxDesc = 0;
}
static uint32_t rtl8139_TxAddr_read(RTL8139State *s, uint32_t txAddrOffset)
_______________________________________________ Qemu-devel mailing list Qemu-devel@nongnu.org http://lists.nongnu.org/mailman/listinfo/qemu-devel
