Since MCR bit 7 is used for the prescaler, we need to update the
cached copy of the MCR after a call to SERDEV_ADJUST_RCLK.

Clean up some bad bits in the past update.

Fix up puc(4) abuse... add new method to serdev, update the termios info.
Adjust rclk again when using tcsetattr() so we don't get stuck with a
weird rclk.
Add a TIOCGFBAUD to get the current exact baudrate.
Use a struct instead of an array of integers.
Various bug fixes.

One other thing maybe not related to this but what happened to TSO4 option? It is no longer visible in ifconfig output...

In D16300#346609, @johalun0_gmail.com wrote:

OK, switching TX/RX between I128LM and I219LM makes no difference, still 5 Kpps. iperf between the two machine measure +900 Mbps.

Use the same callout reset logic in the admin task as the txq timer.

Restore const declaration for head pointer in txsync

I'm not actually sure where the best place to document this new sysctl is... I'm open to suggestions.

In D15720#333069, @marius wrote:

On the bottom line, a front-end driver has to supply both the maximum TSO size and the maximum DMA transfer size to iflib(9). The cleanest approach for that would be to have a separate member for the maximum DMA transfer size in struct if_softc_ctx.

I think that if the size of the DMA tag needs to be larger than the MACs max, iflib should handle that itself rather than have every driver need to add a non-obvious value for non-obvious reasons.

Looks good.

Looks good, but not forcing CSUM_TCP_IPV6 for TSO6 may need a closer look.

Update patch to latest head

Remove lagg fix

Use the new epoch based reclamation API. Now the hot paths will not
block at all, and the sx lock is used for the softc data.

Leave the same group we joined.

Add ume since he's the last one to actually touch this code (14 years ago).

Destroy rm lock, clean up spurious changes.

Address feedback.

Use bool and true/false rather than _Bool and 1/0

Protect against multiple slowpath entries.

Ressurect the rwlock, but only to protect the new sc_slowpath boolean.
If the slowpath is currently running, DELAY(1) until it's not. Still a
bit of a hack, but now it doesn't (shouldn't) cause LORs.

Yeah, all the LAGG_RLOCK -> LAGG_SLOCK conversions need to be fixed up... right now it's a LOR generator for sure.

Rather than remove LAGG_WLOCK_ASSERT()s, convert to LAGG_XLOCK_ASSERT()

Are there some bits missing from the review?

/usr/home/shurd/src/fbsd/sys/net/iflib.c:5308:92: error: passing 'const char *' to parameter of type 'char *' discards qualifiers [-Werror,-Wincompatible-pointer-types-discards-qualifiers]
err = iflib_irq_set_affinity(ctx, rman_get_start(irq->ii_res), type, qid, gtask, tqg, q, name);
^~~~
/usr/home/shurd/src/fbsd/sys/net/iflib.c:5215:72: note: passing argument to parameter 'name' here
struct grouptask *gtask, struct taskqgroup *tqg, void *uniq, char *name)
^
/usr/home/shurd/src/fbsd/sys/net/iflib.c:5356:72: error: passing 'const char *' to parameter of type 'char *' discards qualifiers [-Werror,-Wincompatible-pointer-types-discards-qualifiers]
err = iflib_irq_set_affinity(ctx, irq_num, type, qid, gtask, tqg, q, name);
^~~~
/usr/home/shurd/src/fbsd/sys/net/iflib.c:5215:72: note: passing argument to parameter 'name' here
struct grouptask *gtask, struct taskqgroup *tqg, void *uniq, char *name)
^
/usr/home/shurd/src/fbsd/sys/net/iflib.c:5376:1: error: conflicting types for 'iflib_legacy_setup'
iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, const char *name)
^
/usr/home/shurd/src/fbsd/sys/net/iflib.c:723:12: note: previous declaration is here
static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, char *str);
^

Add r333373

Add Jacob Keller.333366

Add r333366

Add erj@ as a subscriber as he's asked about MFCs.

There doesn't seem to be any will to move this forward, leaving review.

committed as r333364.

May as well abandon it until/unless there's interest. I don't think there's much gain for the added complexity.

Changes committed did not make it into this review due to issues updating Phabricator from a git source. See rS333175 for final version.

SPARC and PowerPC get this:

Arm ports fail as well:

--- kernel.full ---
linking kernel.full
in_mcast.o: In function `inm_init':
/usr/home/admin/src/freebsd-base/sys/netinet/in_mcast.c:236: undefined reference to `iflib_config_gtask_init'
in6_mcast.o: In function `in6m_init':
/usr/home/admin/src/freebsd-base/sys/netinet6/in6_mcast.c:557: undefined reference to `iflib_config_gtask_init'
*** [kernel.full] Error code 1