The struct skb_frag_struct changed since Linux 5.4
commit 8842d285bafa9 ("net: Convert skb_frag_t to bio_vec") and
commit b8b576a16f79e ("net: Rename skb_frag_t size to bv_len")

The correct skb frag size accessors have been available since Linux 3.2
commit 9e903e085262 ("net: add skb frag size accessors")

Since Linux 5.0 the __ETHTOOL_LINK_MODE_LAST macro has been replaced
with __ETHTOOL_LINK_MODE_MASK_NBITS.

Since Linux 5.0 the __ETHTOOL_LINK_MODE_LAST macro is no longer available.
commit e728fdf06289 ("net: phy: improve definition of __ETHTOOL_LINK_MODE_MASK_NBITS")
replaced it with __ETHTOOL_LINK_MODE_MASK_NBITS.

In v0.3.6.17.2 the vendor driver replaced __ETHTOOL_LINK_MODE_LAST+1
with 0xFFFFFFFF. This reverts that, because we want to use the proper
solution that upstream drivers use, namely
__ETHTOOL_LINK_MODE_MASK_NBITS.

This change provides minimal support for the SANLink3 T1 Thunderbolt3
NBase-T adapter (https://www.promise.com/Products/SANLink/SANLink3-T1),
which has the TN4025 with an AQR105 inside it.

The interface comes up, does negotiation, brings up link at 10 Gbit/s
and achieves OK throughput (8 Gbit/s).  The modest throughput might be
related to errors that are periodically logged by the driver:

  [462738.525551] rxd_err = 0x4
  [462738.670364] rxd_err = 0x4
  ...

I didn't have any NBase-T switches offhand to test 2.5G or 5G rates,
so I don't know if that works.

This change didn't work earlier on the 0.3.6.15 driver, so I suspect
it requires the 2.b.e2 AQR105 firmware mentioned in release_notes.

This driver was downloaded from:
http://www.tehutinetworks.net/?t=drivers&L1=8&L2=12&L3=26&L4=0&L5=0&L6=0&L7=0&f=tn40xx-0.3.6.17.2.tgz

The source code was extracted, but not modified.

Thanks to Jean-Christophe Heger for contributing DKMS support.

Prepare for 001 release:
* briefly describe dkms
* change version for release
* change clone url to point to release repo

move branch descriptions and vendor diffs to docs/

Removes the redundant STRING_FMT macro that vendor added in v0.3.6.16

replace DBG(...) with pr_debug(...)
replace MSG(...) with pr_info(...)
replace ERR(...) with pr_err(...)
and
remove unused DBG, MSG, ERR macros

The cleanup step for removing redundant returns at the end of void
functions (added in commit 49c0bf4a) was flawed. Redundant returns have
a space (ie. 'return ;'), so adjust the regex to match that.

upstream fixed this variable name in 0.3.16.15 and this backports the
fix to make diffs between cleanup branches smaller.

Remove redundant TN40_FTRACE, FTRACE, FTRACE_ON and FTRACE_OFF macros.

(cherry picked from commit 68c5bf29)

void functions do not need to return; at the end. The empty return was
left over after removing the RET() macro in
f7344c42 and
3e916da0

(cherry picked from commit d9e29d13)

Lindent needs to be run at the start of the cleanup process, so that the
format is known and consistent, before the regex transformations are applied.

(cherry picked from commit 66ed7f24)

The test for remaining DBG1 macros (added in
 d7b9e9ab) incorrectly exited with
non-zero status and caused the cleanup.sh script to abort.

(cherry picked from commit 76d9bb10)

Remove comments like
 /* foo(bar); */
and
 /* foo->bar = baz; */
and
 /* foo = bar->baz; */

This doesn't get rid of all dead code. As usual the result was visually
checked for false positives to ensure that no accidental content
removals occurred.

(cherry picked from commit f2d0a35a)

(cherry picked from commit 3e916da0)