> -----Original Message----- > From: Ananyev, Konstantin > Sent: Tuesday, March 26, 2019 10:18 PM > To: Zhang, Qi Z <qi.z.zh...@intel.com>; Yigit, Ferruh > <ferruh.yi...@intel.com>; > Stokes, Ian <ian.sto...@intel.com>; dev@dpdk.org > Cc: step...@networkplumber.org; Lu, Wenzhuo <wenzhuo...@intel.com>; Xing, > Beilei <beilei.x...@intel.com> > Subject: RE: [dpdk-dev] [PATCH v2 6/7] net/e1000: set min and max MTU for igb > devices > > > > > > > > Hi Qi, > > > > > > > > > > > > > On 3/22/2019 1:01 PM, Ian Stokes wrote: > > > > > > This commit sets the min and max supported MTU values for igb > > > > > > devices via the eth_igb_info_get() function. Min MTU supported > > > > > > is set to ETHER_MIN_MTU and max mtu is calculated as the max > > > > > > packet length supported minus the transport overhead. To aid > > > > > > in these calculations a new MACRO 'E1000_ETH_OVERHEAD' has > > > > > > been introduced to consolidate overhead calculation and avoid > duplication. > > > > > > > > > > > > Signed-off-by: Ian Stokes <ian.sto...@intel.com> > > > > > > --- > > > > > > drivers/net/e1000/e1000_ethdev.h | 6 ++++++ > > > > > > drivers/net/e1000/igb_ethdev.c | 7 +++++-- > > > > > > 2 files changed, 11 insertions(+), 2 deletions(-) > > > > > > > > > > > > diff --git a/drivers/net/e1000/e1000_ethdev.h > > > > > > b/drivers/net/e1000/e1000_ethdev.h > > > > > > index 94edff08e..3e74cd8fe 100644 > > > > > > --- a/drivers/net/e1000/e1000_ethdev.h > > > > > > +++ b/drivers/net/e1000/e1000_ethdev.h > > > > > > @@ -89,6 +89,12 @@ > > > > > > ETH_RSS_IPV6_UDP_EX) > > > > > > > > > > > > /* > > > > > > + * The overhead from MTU to max frame size. > > > > > > + * Considering VLAN so a tag needs to be counted. > > > > > > + */ > > > > > > +#define E1000_ETH_OVERHEAD (ETHER_HDR_LEN + ETHER_CRC_LEN + > > > > > > +VLAN_TAG_SIZE) > > > > > > > > > > As an overhead, following drivers set: > > > > > i40e: HDR + CRC + 2 * VLAN > > > > > ixgbe: HDR + CRC > > > > > e1000: HDR + CRC + VLAN > > > > > > > > > > I wonder if this difference is HW limitation, or driver > > > > > limitation or just implementation inconsistency. > > > > > > > > I think this is implementation inconsistency > > > > > > > > The NIC only accept Max Frame Size. > > > > > > > > The problem here is seems all of three setup are not perfect. > > > > > > > > HDR + CRC + 2 * VLAN - it may allow non vlan or single vlan packet > > > > that exceed > > > mtu. > > > > > > Hmm, wonder how? > > > > I'm talking about the case: > > > > Assume mtu = 1500, we will set max frame size to 1500 + 14 + 4 + 2*4 > > = 1526 Let's assume a non vlan packet with 1522 size, so its l2 > > payload will be 1504 that exceed the mtu, but it will still be accepted, > > does it > break the configure? > > Of course it would, but as I can read the mail, we discussing overhead to > subtract > from max_rx_pkt_len to report max allowable mtu. > From that perspective bigger overhead is more conservative and makes sure our > tx packet will never be bigger than max_rx_pkt_len. > Konstantin
I'm OK to choose HDR + CRC + 2 * VLAN as MTU overhead to keep all driver consistent. Qi > > > > > > > > > > HDR + CRC - it may reject vlan or double vlan packet that follow mtu. > > > > HDR + CRC + VLAN , it may reject double vlan packet that follow > > > > mtu > > > > > > > > I agree it's better to keep consistent on all drivers, but before > > > > this, we may need to decide which one we should take :) > > > > > > > > Regards > > > > Qi > > > > > > > > > > > > > > Better to confirm it that it is not implementation inconsistency. > > > > > > > > > > Wenzhuo, Konstantin, Beilei, Qi, > > > > > > > > > > Can you please comment? > > > > > > > > > > Thanks, > > > > > ferruh
