On 25-Oct-23 10:07 AM, David Marchand wrote:
On Wed, Oct 25, 2023 at 11:02 AM Radu Nicolau <[email protected]> wrote:
On 25-Oct-23 12:30 AM, Zhang, Qi Z wrote:
-----Original Message-----
From: Nicolau, Radu <[email protected]>
Sent: Tuesday, October 24, 2023 10:49 PM
To: Zhang, Qi Z <[email protected]>; Marchand, David
<[email protected]>
Cc: Wu, Jingjing <[email protected]>; Xing, Beilei <[email protected]>;
[email protected]; [email protected]
Subject: Re: [PATCH] net/iavf: fix IAVF_TX_OFFLOAD_MASK definition
On 24-Oct-23 12:24 PM, Zhang, Qi Z wrote:
-----Original Message-----
From: Radu Nicolau <[email protected]>
Sent: Tuesday, October 24, 2023 6:23 PM
To: Marchand, David <[email protected]>
Cc: Wu, Jingjing <[email protected]>; Xing, Beilei
<[email protected]>; [email protected]; [email protected]
Subject: Re: [PATCH] net/iavf: fix IAVF_TX_OFFLOAD_MASK definition
On 24-Oct-23 10:49 AM, David Marchand wrote:
On Tue, Oct 24, 2023 at 11:13 AM Radu Nicolau
<[email protected]>
wrote:
IAVF_TX_OFFLOAD_MASK definition contained
RTE_ETH_TX_OFFLOAD_SECURITY
instead of RTE_MBUF_F_TX_SEC_OFFLOAD.
Fixes: 6bc987ecb860 ("net/iavf: support IPsec inline crypto")
Cc: [email protected]
Signed-off-by: Radu Nicolau <[email protected]>
Something is not clear to me.
How was the IPsec inline crypto feature supposed to work with this
driver so far?
Any packet with the RTE_MBUF_F_TX_SEC_OFFLOAD flag should have
been
refused in iavf_prep_pkts.
It worked because the IPsec sample app doesn't call
rte_eth_tx_prepare, and from what I can see no other sample app does.
To keep consistent, its better to refine the
IAVF_TX_OFFLOAD_NOTSUP_MASK definition.
You mean like this?
#define IAVF_TX_OFFLOAD_NOTSUP_MASK ( \
RTE_MBUF_F_TX_OFFLOAD_MASK ^ ( \
RTE_MBUF_F_TX_OUTER_IPV6 | \
RTE_MBUF_F_TX_OUTER_IPV4 | \
RTE_MBUF_F_TX_IPV6 | \
RTE_MBUF_F_TX_IPV4 | \
RTE_MBUF_F_TX_VLAN | \
RTE_MBUF_F_TX_IP_CKSUM | \
RTE_MBUF_F_TX_L4_MASK | \
RTE_MBUF_F_TX_TCP_SEG | \
RTE_MBUF_F_TX_UDP_SEG | \
RTE_MBUF_F_TX_TUNNEL_MASK | \
RTE_MBUF_F_TX_OUTER_IP_CKSUM | \
RTE_MBUF_F_TX_OUTER_UDP_CKSUM | \
RTE_MBUF_F_TX_SEC_OFFLOAD))
Sorry, I miss understanding this code change, actually you didn't remove a
flag, but just replace it, NOTSUP_MASK no need to be changed
Then I don't understand why "Any packet with the RTE_MBUF_F_TX_SEC_OFFLOAD flag
should have refused in iavf_prep_pkts"
But I assume tx_pkt_prepare should reject only invalid packets while still
functioning correctly with inline IPsec.
rte_eth_tx_prepare would have rejected the packets before this fix, but
no app calls rte_eth_tx_prepare. The only app that calls it is testpmd.
From my understanding, applications that want checksum offload are
required to call rte_eth_tx_prepare.
TBH I don't understand much about it and looking at the implementation
actually made things worse: for example from what I can see calling it
when RTE_MBUF_F_TX_TCP_CKSUM is set will result in having the TCP
checksum being computed (in software) in the prepare function.
