AFAIK the PAPR document which defines the virtual device interface used by
the ibmveth driver doesn't specify a specific maximum MTU. So, in the
ibmveth driver, the maximum allowed MTU is determined by the maximum
allocated buffer size of 64k (corresponding to one page in the common case)
minus the per-buffer overhead IBMVETH_BUFF_OH (which has value 22 for 14
bytes of ethernet header, plus 8 bytes for an opaque handle).
This suggests a maximum allowable MTU of 65514 bytes, but in fact the
driver only permits a maximum MTU of 65513. This is because there is a <
instead of an <= in ibmveth_change_mtu(), which only permits an MTU which
is strictly smaller than the buffer size, rather than allowing the buffer
to be completely filled.
This patch fixes the buglet.
Signed-off-by: David Gibson <[email protected]>1
---
drivers/net/ethernet/ibm/ibmveth.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
Changes since v1:
* Fixed a second instance of the same off-by-one error. Thanks to
Thomas Falcon for spotting this.
diff --git a/drivers/net/ethernet/ibm/ibmveth.c b/drivers/net/ethernet/ibm/ibmveth.c
index cd7675a..1813476 100644
--- a/drivers/net/ethernet/ibm/ibmveth.c
+++ b/drivers/net/ethernet/ibm/ibmveth.c
@@ -1238,7 +1238,7 @@ static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
return -EINVAL;
for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
- if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
+ if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
break;
if (i == IBMVETH_NUM_BUFF_POOLS)
@@ -1257,7 +1257,7 @@ static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
adapter->rx_buff_pool[i].active = 1;
- if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
+ if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
dev->mtu = new_mtu;
vio_cmo_set_dev_desired(viodev,
ibmveth_get_desired_dma
--
2.1.0
On 04/20/2015 08:07 PM, David Gibson wrote:
> AFAIK the PAPR document which defines the virtual device interface used by
> the ibmveth driver doesn't specify a specific maximum MTU. So, in the
> ibmveth driver, the maximum allowed MTU is determined by the maximum
> allocated buffer size of 64k (corresponding to one page in the common case)
> minus the per-buffer overhead IBMVETH_BUFF_OH (which has value 22 for 14
> bytes of ethernet header, plus 8 bytes for an opaque handle).
>
> This suggests a maximum allowable MTU of 65514 bytes, but in fact the
> driver only permits a maximum MTU of 65513. This is because there is a <
> instead of an <= in ibmveth_change_mtu(), which only permits an MTU which
> is strictly smaller than the buffer size, rather than allowing the buffer
> to be completely filled.
>
> This patch fixes the buglet.
Thanks!
Acked-by: Thomas Falcon <[email protected]>
>
> Signed-off-by: David Gibson <[email protected]>1
> ---
> drivers/net/ethernet/ibm/ibmveth.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> Changes since v1:
> * Fixed a second instance of the same off-by-one error. Thanks to
> Thomas Falcon for spotting this.
>
> diff --git a/drivers/net/ethernet/ibm/ibmveth.c b/drivers/net/ethernet/ibm/ibmveth.c
> index cd7675a..1813476 100644
> --- a/drivers/net/ethernet/ibm/ibmveth.c
> +++ b/drivers/net/ethernet/ibm/ibmveth.c
> @@ -1238,7 +1238,7 @@ static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
> return -EINVAL;
>
> for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
> - if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
> + if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
> break;
>
> if (i == IBMVETH_NUM_BUFF_POOLS)
> @@ -1257,7 +1257,7 @@ static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
> for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
> adapter->rx_buff_pool[i].active = 1;
>
> - if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
> + if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
> dev->mtu = new_mtu;
> vio_cmo_set_dev_desired(viodev,
> ibmveth_get_desired_dma
On 04/22/2015 02:42 PM, Thomas Falcon wrote:
> On 04/20/2015 08:07 PM, David Gibson wrote:
>> AFAIK the PAPR document which defines the virtual device interface used by
>> the ibmveth driver doesn't specify a specific maximum MTU. So, in the
PAPR itself doesn't, but the max-frame-size property, which I understand
to be synonymous with max MTU, is required by PAPR to be present in the
Logical LAN OF node corresponding to the ibmveth device.
-Tyrel
>> ibmveth driver, the maximum allowed MTU is determined by the maximum
>> allocated buffer size of 64k (corresponding to one page in the common case)
>> minus the per-buffer overhead IBMVETH_BUFF_OH (which has value 22 for 14
>> bytes of ethernet header, plus 8 bytes for an opaque handle).
>>
>> This suggests a maximum allowable MTU of 65514 bytes, but in fact the
>> driver only permits a maximum MTU of 65513. This is because there is a <
>> instead of an <= in ibmveth_change_mtu(), which only permits an MTU which
>> is strictly smaller than the buffer size, rather than allowing the buffer
>> to be completely filled.
>>
>> This patch fixes the buglet.
>
> Thanks!
>
> Acked-by: Thomas Falcon <[email protected]>
>
>>
>> Signed-off-by: David Gibson <[email protected]>1
>> ---
>> drivers/net/ethernet/ibm/ibmveth.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> Changes since v1:
>> * Fixed a second instance of the same off-by-one error. Thanks to
>> Thomas Falcon for spotting this.
>>
>> diff --git a/drivers/net/ethernet/ibm/ibmveth.c b/drivers/net/ethernet/ibm/ibmveth.c
>> index cd7675a..1813476 100644
>> --- a/drivers/net/ethernet/ibm/ibmveth.c
>> +++ b/drivers/net/ethernet/ibm/ibmveth.c
>> @@ -1238,7 +1238,7 @@ static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
>> return -EINVAL;
>>
>> for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++)
>> - if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size)
>> + if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size)
>> break;
>>
>> if (i == IBMVETH_NUM_BUFF_POOLS)
>> @@ -1257,7 +1257,7 @@ static int ibmveth_change_mtu(struct net_device *dev, int new_mtu)
>> for (i = 0; i < IBMVETH_NUM_BUFF_POOLS; i++) {
>> adapter->rx_buff_pool[i].active = 1;
>>
>> - if (new_mtu_oh < adapter->rx_buff_pool[i].buff_size) {
>> + if (new_mtu_oh <= adapter->rx_buff_pool[i].buff_size) {
>> dev->mtu = new_mtu;
>> vio_cmo_set_dev_desired(viodev,
>> ibmveth_get_desired_dma
>
> _______________________________________________
> Linuxppc-dev mailing list
> [email protected]
> https://lists.ozlabs.org/listinfo/linuxppc-dev
>