Skip to contentSkip to conversation

qemu: e1000 fix TOR math

PATCH Superseded Michael S. Tsirkin 2 participants 3 messages Sep 5, 2010 – Sep 6, 2010

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 qemu: e1000 fix TOR math

Patch b0b900070c7cb29bbefb732ec00397abe5de6d73 made TOR valuer incorrect: the spec says it should always include the CRC field, while size does not include CRC now. No one seems to use TOR field (which is likely why current code works fine), but better to stick to spec.

Lightly tested with a linux guest.

---
 hw/e1000.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)
hw/e1000.c +6 -2
index 80b78bc..eb9faf2 100644
@@ -345,7 +345,7 @@ is_vlan_txd(uint32_t txd_lower)
 
 /* FCS aka Ethernet CRC-32. We don't get it from backends and can't
  * fill it in, just pad descriptor length by 4 bytes unless guest
- * told us to trip it off the packet. */
+ * told us to strip it off the packet. */
 static inline int
 fcs_len(E1000State *s)
 {
@@ -690,8 +690,12 @@ e1000_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 
     s->mac_reg[GPRC]++;
     s->mac_reg[TPR]++;
+    /* TOR - Total Octets Received:
+     * This register includes bytes received in a packet from the <Destination
+     * Address> field through the <CRC> field, inclusively.
+     */
     n = s->mac_reg[TORL];
-    if ((s->mac_reg[TORL] += size) < n)
+    if ((s->mac_reg[TORL] += size + 4 /* Always include FCS length. */) < n)
         s->mac_reg[TORH]++;
 
     n = E1000_ICS_RXT0;
Signature · 1 line
1.7.2.rc0.14.g41c1c
#1 lore raw
Quoted 28 lines · Michael S. Tsirkin

Patch b0b900070c7cb29bbefb732ec00397abe5de6d73 made TOR valuer incorrect: the spec says it should always include the CRC field, while size does not include CRC now. No one seems to use TOR field (which is likely why current code works fine), but better to stick to spec.

Lightly tested with a linux guest.

---
 hw/e1000.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)
hw/e1000.c +1 -1
index 80b78bc..eb9faf2 100644
@@ -345,7 +345,7 @@ is_vlan_txd(uint32_t txd_lower)
 
 /* FCS aka Ethernet CRC-32. We don't get it from backends and can't
  * fill it in, just pad descriptor length by 4 bytes unless guest
- * told us to trip it off the packet. */
+ * told us to strip it off the packet. */
 static inline int
 fcs_len(E1000State *s)
 {
@@ -690,8 +690,12 @@ e1000_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 

hw/e1000.c e1000_receive(VLANClientState *nc, const uint8_t *buf, size_t size) · line 699

Quoted 10 lines
hw/e1000.c +5 -1
     s->mac_reg[GPRC]++;
     s->mac_reg[TPR]++;
+    /* TOR - Total Octets Received:
+     * This register includes bytes received in a packet from the <Destination
+     * Address> field through the <CRC> field, inclusively.
+     */
     n = s->mac_reg[TORL];
-    if ((s->mac_reg[TORL] += size) < n)
+    if ((s->mac_reg[TORL] += size + 4 /* Always include FCS
     length. */) < n)

once changing this, can we move the assignation out of the if? It was already complex enough, but adding a comment inside an if condition looks "too much" to me.

Later, Juan.

Quoted 3 lines
hw/e1000.c +0 -0
         s->mac_reg[TORH]++;
 
     n = E1000_ICS_RXT0;
#2 lore raw
Quoted 44 lines · Juan Quintela
Quoted 28 lines · Michael S. Tsirkin

Patch b0b900070c7cb29bbefb732ec00397abe5de6d73 made TOR valuer incorrect: the spec says it should always include the CRC field, while size does not include CRC now. No one seems to use TOR field (which is likely why current code works fine), but better to stick to spec.

Lightly tested with a linux guest.

---
 hw/e1000.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)
hw/e1000.c +1 -1
index 80b78bc..eb9faf2 100644
@@ -345,7 +345,7 @@ is_vlan_txd(uint32_t txd_lower)
 
 /* FCS aka Ethernet CRC-32. We don't get it from backends and can't
  * fill it in, just pad descriptor length by 4 bytes unless guest
- * told us to trip it off the packet. */
+ * told us to strip it off the packet. */
 static inline int
 fcs_len(E1000State *s)
 {
@@ -690,8 +690,12 @@ e1000_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 

hw/e1000.c e1000_receive(VLANClientState *nc, const uint8_t *buf, size_t size) · line 699

Quoted 10 lines
hw/e1000.c +5 -1
     s->mac_reg[GPRC]++;
     s->mac_reg[TPR]++;
+    /* TOR - Total Octets Received:
+     * This register includes bytes received in a packet from the <Destination
+     * Address> field through the <CRC> field, inclusively.
+     */
     n = s->mac_reg[TORL];
-    if ((s->mac_reg[TORL] += size) < n)
+    if ((s->mac_reg[TORL] += size + 4 /* Always include FCS
     length. */) < n)

once changing this, can we move the assignation out of the if? It was already complex enough, but adding a comment inside an if condition looks "too much" to me.

Quoted 1 line

Later, Juan.

Sure, I'll do that.

Quoted 3 lines
Quoted 3 lines
hw/e1000.c +0 -0
         s->mac_reg[TORH]++;
 
     n = E1000_ICS_RXT0;