Update prestera driver to v3.0.0

[taraschornyiplv]

Switchdev Driver v3.0.0 features included:

• New support for Marvell 98DX3500 Packet Processor (AC5X)
• ACL Scale/Performance/goto
• NAT v4-v4
• Port Isolation
• Policer / Storm Control - Data plane
• OSPF/ISIS
• NAT on VLAN tagged port (bridge)
• Dev Link - local Trap/Hard Drop counters

List of commit in this PR:

• ab99e586 Enable prestera v3.0.0 driver for arm64 builds
• fc7b21a5 Bumps prestera driver to v3.0.0
• 899936e7 fix infinite loop when trying to create tc rule (planed for upstream)
• 417aa916 fix for tc action xt (dentos specific)
• c9d32d03 Backport patches from upsteran iproute2 related to json output
• 8c44ff21 Enable prestera driver on as4564 platform (dentos specific)
• 65e39dab netfilter: nf_flow_table_free(): flush flows before cleanup (planed for upstream)
• 87197712 Change HW OFFLOAD timeout to 60 sec (no need to upstream as it is configurable via sysctl in latest kernel)
• a3261c21 TC statistics does not account for NATed traffic (planed for upstream)
• 6cd1fad2 Enable BR_ISOLATED flag to be offloaded (already in upstream)
• bc29829a Modify RX loss active high for CPLD RX loss definition correction (dentos specific)
• 0767de5e arm64: kernel-5.10: Add NAT related config(dentos specific)
• 6716f37e arm64: kernel-5.10: Add Acl related config(dentos specific)
• 57b7aa13 Kernel has no interface to publish the devlink port parameters (upsteam in progress)
• a7fd0ed4 net: core: devlink: use right genl user_ptr when handling (upstreamed)
• b8223421 net: core: devlink: add 'dropped' stats field for traps(upsteam in progress)
• 2b0c0ba7 tc_nat: add action offload(planed for upstream)
• cc7a0021 net: phylink: Treat speed change as a major change (planed for upstream)
• 1b2705cb net: sfp: add quirk for Finisar FTLF8536P4BCL (planed for upstream)
• eb46eb80 net: phylink: Add SFP remove handler. (planed for upstream)
• 09c6bb82 Align PCIe reset patch with kernel 5.10(align patch provided by Marvell IPBU team)

[robimarko]

@taraschornyiplv Would it make more sense to pull v3.0.1 instead?

[paulmenzel]

The hashes in the merge/pull request summary do not match the hashes in the branch (after the latest pushes).

[jwagantall]

Closed/reopened this PR since DCO was stuck

[jwagantall]

I don't think Closing/Reopening this PR worked.

Taras, you might need to try force pushing a new update in this PR but I have seen this issue and I don't think that would work. From past experiences, whenever this happens due to a DCO downtime, GH team always asks us to create a new PR.

[taraschornyiplv]

ok I will force push

[paulmenzel]

@eladnachman, why did you merge it, despite changes were still requested? Please, do thorough reviews and keep in mind, that maintaining patches on top is a lot of work and decreases quality as upstream reviews are missing.

[eladnachman]

Merge was done following multiple reviews of multiple people (including myself) of the the kernel patches. I have reviewed the patches myself at least 3 times. I am not aware of any open outstanding changes. Upstreaming is a very long process and should be done in parallel in order to prevent delays to the development cycle.

[paulmenzel]

Upstreaming is a very long process and should be done in parallel in order to prevent delays to the development cycle.

Some of the commits were authored over half a year ago in February 2021, so it’s not happening in parallel. The network subsystem (net) changes look to me like, could have been easily sent upstream already. It looks to me, some processes are not set up optimally, if upstreaming seems to be such a problem.