remove unneeded includes; ok miod@

Make ifq and ifiq interface MP safe.

Rename ifq_set_maxlen() to ifq_init_maxlen(). This function neither
uses WRITE_ONCE() nor a mutex and is called before the ifq mutex
is initialized. The new name expresses that it should be used only
during interface attach when there is no concurrency.

Protect ifq_len(), ifq_empty(), ifiq_len(), and ifiq_empty() with
READ_ONCE(). They can be used without lock as they only read a
single integer.

OK dlg@

Constify struct cfattach.

Rename the macro MCLGETI to MCLGETL and removes the dead parameter ifp.

OK dlg@, bluhm@
No Opinion mpi@
Not against it claudio@

Change users of IFQ_SET_MAXLEN() and IFQ_IS_EMPTY() to use the "new" API.

ok dlg@ tobhe@

Argument order fix for MCLGETI.

ok claudio@

If you use sys/param.h, you don't need sys/types.h

move counting if_opackets next to counting if_obytes in if_enqueue.

this means packets are consistently counted in one place, unlike the
many and various ways that drivers thought they should do it.

ok mpi@ deraadt@

m_free() and m_freem() test for NULL. Simplify callers which had their own
NULL tests.

ok mpi@

G/C IFQ_SET_READY().

replace IFF_OACTIVE manipulation with mpsafe operations.

there are two things shared between the network stack and drivers
in the send path: the send queue and the IFF_OACTIVE flag. the send
queue is now protected by a mutex. this diff makes the oactive
functionality mpsafe too.

IFF_OACTIVE is part of if_flags. there are two problems with that.
firstly, if_flags is a short and we dont have any MI atomic operations
to manipulate a short. secondly, while we could make the IFF_OACTIVE
operates mpsafe, all changes to other flags would have to be made
safe at the same time, otherwise a read-modify-write cycle on their
updates could clobber the oactive change.

instead, this moves the oactive mark into struct ifqueue and provides
an API for changing it. there's ifq_set_oactive, ifq_clr_oactive,
and ifq_is_oactive. these are modelled on ifsq_set_oactive,
ifsq_clr_oactive, and ifsq_is_oactive in dragonflybsd.

this diff includes changes to all the drivers manipulating IFF_OACTIVE
to now use the ifsq_{set,clr_is}_oactive API too.

ok kettenis@ mpi@ jmatthew@ deraadt@

You only need <net/if_dl.h> if you're using LLADDR() or a sockaddr_dl.

shuffle struct ifqueue so in flight mbufs are protected by a mutex.

the code is refactored so the IFQ macros call newly implemented ifq
functions. the ifq code is split so each discipline (priq and hfsc
in our case) is an opaque set of operations that the common ifq
code can call. the common code does the locking, accounting (ifq_len
manipulation), and freeing of the mbuf if the disciplines enqueue
function rejects it. theyre kind of like bufqs in the block layer
with their fifo and nscan disciplines.

the new api also supports atomic switching of disciplines at runtime.
the hfsc setup in pf_ioctl.c has been tweaked to build a complete
hfsc_if structure which it attaches to the send queue in a single
operation, rather than attaching to the interface up front and
building up a list of queues.

the send queue is now mutexed, which raises the expectation that
packets can be enqueued or purged on one cpu while another cpu is
dequeueing them in a driver for transmission. a lot of drivers use
IFQ_POLL to peek at an mbuf and attempt to fit it on the ring before
committing to it with a later IFQ_DEQUEUE operation. if the mbuf
gets freed in between the POLL and DEQUEUE operations, fireworks
will ensue.

to avoid this, the ifq api introduces ifq_deq_begin, ifq_deq_rollback,
and ifq_deq_commit. ifq_deq_begin allows a driver to take the ifq
mutex and get a reference to the mbuf they wish to try and tx. if
there's space, they can ifq_deq_commit it to remove the mbuf and
release the mutex. if there's no space, ifq_deq_rollback simply
releases the mutex. this api was developed to make updating the
drivers using IFQ_POLL easy, instead of having to do significant
semantic changes to avoid POLL that we cannot test on all the
hardware.

the common code has been tested pretty hard, and all the driver
modifications are straightforward except for de(4). if that breaks
it can be dealt with later.

ok mpi@ jmatthew@

Do not include <net/if_vlan_var.h> when it's not necessary.

Because of the VLAN hacks in mpw(4) this file still contains the definition
of "struct ifvlan" which depends on <sys/refcnt.h> which in turns pull
<sys/atomic.h>...

arp_ifinit() is no longer needed.

Increment if_ipackets in if_input().

Note that pseudo-drivers not using if_input() are not affected by this
conversion.

ok mikeb@, kettenis@, claudio@, dlg@

Convert to if_input(), thanks to krw@ for testing.

Remove some includes include-what-you-use claims don't
have any direct symbols used. Tested for indirect use by compiling
amd64/i386/sparc64 kernels.

ok tedu@ deraadt@

unifdef INET

after allocating an mbuf and cluster you still need to init the length
fields.

found by steven roberts, who also tested this fix for me

remove the custom jumbo allocator. its never been enabled or used.

putting this into the tree to make it easier to test.

Fewer <netinet/in_systm.h>

The few network drivers that called their children's (ie. mii PHY
drivers) activate functions at DVACT_RESUME time do not need to do
so, since their PHYs are repaired by IFF_UP.

Add a DVACT_WAKEUP op to the *_activate() API. This is called after the
kernel resumes normal (non-cold, able to run processes, etc) operation.
Previously we were relying on specific DVACT_RESUME op's in drivers
creating callback/threads themselves, but that has become too common,
indicating the need for a built-in mechanism.
ok dlg kettenis, tested by a sufficient amount of people

Most network drivers include netinet/in_var.h, but apparently they
don't have to. Just remove these include lines.
Compiled on amd64 i386 sparc64; OK henning@ mikeb@

Rewrite receive filter and ioctl handling code.

ok jsg@

Remove setting an initial assumed baudrate upon driver attach which is not
necessarily correct, there might not even be a link when attaching.

ok mikeb@ reyk@

Wake on LAN support for nfe(4). Tested by me, russel on misc@, and jsg.
ok jsg

mechanic rename M_{TCP|UDP}V4_CSUM_OUT -> M_{TCP|UDP}_CSUM_OUT
ok claudio krw

Some nfe(4)/rlphy(4) combos don't work, because the PHY responds to all
addresses on the mii bus. As a countereasure, only attach the first PHY we
encounter. It is very unlikely we're going to ever see nfe(4) with multiple
PHYs. The same is probably true for any modern NIC.

ok mikeb@, deraadt@

remove the powerhook code. All architectures now use the ca_activate tree
traversal code to suspend/resume
ok oga kettenis blambert

Add DVACT_QUIECE support. This is called before splhigh() and before
DVACT_SUSPEND, therefore DVACT_QUIECE can do standard sleeping operations
to get ready.
Discussed quite a while back with kettenis and jakemsr, oga suddenly needed
it as well and wrote half of it, so it was time to finish it.
proofread by miod.

activate function should return result of config_activate_children

Make the powerhook use the activate functions, which actually do the full
job.

remove the unused if_init callback in struct ifnet
ok deraadt@ henning@ claudio@

ca_activate function for nfe(4) for suspend/resume
tested on nvidia mcp51

ok deraadt@

BUS_DMA_ZERO instead of alloc, map, bzero.

ok krw@

Remove support for cancelled NVIDIA MCP7B and add initial support for
MCP89.

From Brad based on information from Peer Chen @ NVIDIA via Linux.

make various strings ("can't map mem space" and similar) more consistent
between instances, saving space in the kernel. feedback from many (some
incorporated, some left for future work).

ok deraadt, kettenis, "why not" miod.

Eliminate the redundant bits of code for MTU and multicast handling
from the individual drivers now that ether_ioctl() handles this.

Shrinks the i386 kernels by..
RAMDISK - 2176 bytes
RAMDISKB - 1504 bytes
RAMDISKC - 736 bytes

Tested by naddy@/okan@/sthen@/brad@/todd@/jmc@ and lots of users.
Build tested on almost all archs by todd@/brad@

ok naddy@

Introduce bpf_mtap_ether(), which for the benefit of bpf listeners
creates the VLAN encapsulation from the tag stored in the mbuf
header. Idea from FreeBSD, input from claudio@ and canacar@.

Switch all hardware VLAN enabled drivers to the new function.

ok claudio@

Remove #if NVLAN.

Re-add support for RX VLAN tag stripping.

Switch the existing TX VLAN hardware support over to having the
tag in the header. Convert TX tagging in the drivers.

Help and ok brad@

First step towards cleaning up the Ethernet driver ioctl handling.
Move calling ether_ioctl() from the top of the ioctl function, which
at the moment does absolutely nothing, to the default switch case.
Thus allowing drivers to define their own ioctl handlers and then
falling back on ether_ioctl(). The only functional change this results
in at the moment is having all Ethernet drivers returning the proper
errno of ENOTTY instead of EINVAL/ENXIO when encountering unknown
ioctl's.

Shrinks the i386 kernels by..
RAMDISK - 1024 bytes
RAMDISKB - 1120 bytes
RAMDISKC - 832 bytes

Tested by martin@/jsing@/todd@/brad@
Build tested on almost all archs by todd@/brad@

ok jsing@

Convert timeout_add() calls using multiples of hz to timeout_add_sec()

Really just the low-hanging fruit of (hopefully) forthcoming timeout
conversions.

ok art@, krw@

MCP79 are also capable of Jumbo frames. Add the Jumbo support flag.

From: Linux forcedeth

Simplify the combination use of pci_mapreg_type()/pci_mapreg_map() as
suggested by dlg@ awhile ago.

ok dlg@

correct format after %x -> %b convertion. deraadt jsg ok.

Add the NFE_CORRECT_MACADDR flag for MCP77/79 chipsets.

ok jsg@ dlg@

add the NVIDIA MCP77/79 ids.

ok jsg@ dlg@

Fix check for pending interrupts as some other device is causing
the status register to change.

From form@.

ok jsg

Make sure newer adapters are not in powerdown mode.
From Oleg Safiullin <form@pdp-11.org.ru> similiar to changes
in Linux/FreeBSD driver.

Disable jumbo allocator until the low number of buffers case
is better handled.

This will likely resolve stalls some people are seeing under high
load.

ok damien@

In nfe_start() do a fast return if IFF_OACTIVE is set, in
this case we need a Tx interrupt to clean up the DMA ring
before if_start can be properly called.

Diff based on a comment by & ok claudio@

use new malloc M_ZERO flag to shrink kernel.
remove <malloc.h> from files where malloc is not used.

Correct the backwards ethernet address that some NVidia MACs have.
diff from brad. "commit this" jsg

MCP65 has no hardware checksum support.
Diff from brad after he noticed recent changes
in the Linux driver by an NVIDIA employee.

Add support for HW TCP/IP checksum offload for adapters that support it.

Tested by many (IP/UDP/TCP):

Jason McIntyre <jmc@>
Chris Kuethe <chris.kuethe AT gmail.com>
Alf Schlichting <a.schlichting AT lemarit.com>
Rodolfo Gouveia <rgouveia AT cosmico.net>
Peter Stromberg <wilfried@>

Has been in snaps for weeks too with noone complaining so far.

Correct the WOL magic value and rename NFE_WOL_MAGIC to NFE_WOL_ENABLE.

WOL magic value from Peer Chen@NVIDIA via FreeBSD.

ok jsg@ damien@

Defer setting of the valid bit in the first TX descriptor after
all descriptors have been setup. Otherwise, hardware may start
processing descriptors faster than us and crap out.
Fixes "watchdog timeout" errors.

Original idea from Matthew Dillon @DragonFly.

ok deraadt@ jsg@ wim@
tested by many

add the NVIDIA MCP67 LAN PCI ids.

branches: 1.64.2;
add NVidia MCP61/65 ids.

add sys/timeout.h

garbage collect NFE_NO_JUMBO.

- remove ETHER_MAX_LEN_JUMBO and ETHERMTU_JUMBO.
- use if_hardmtu for MTU ioctl handlers.

ok reyk@

unknown ioctl is ENOTTY not EINVAL

remove IFCAP_JUMBO_MTU interface capabilities flag and set if_hardmtu in a few
more drivers.

ok reyk@

set if_jumbo_mtu and the IFCAP_JUMBO_MTU capabilities flag where
appropriate.

ok reyk@

Use %b in error flag printfs to describe meaning of error bits.
requested by deraadt@

And commit the diff from the right system.

check for IFF_RUNNING being set before calling nfe_init().

ok damien@ jsg@

Add work around for mbuf leak in the tx path until we
can come up with a better guess as to how the hardware works.
From Chuck Silvers.
ok damien@

allow bpf(4) to ignore packets based on their direction (inbound or
outbound), using a new BIOCSDIRFILT ioctl;
guidance, feedback and ok canacar@

branches: 1.52.2;
Put the PHY update/reset call in nfe_init back to the spot it was at
before Rx/Tx/Interrupts are enabled.
This makes <fredd at cse.sc.edu>'s nfe+icsphy setup work again.

Tested on nfe+eephy by otto@ and myself, nfe+ciphy by otto@ and
nfe+rlphy by wilfried@

ok deraadt@

use sc->sc_dev.dv_xname consistently.

Let if_nfe.c compile again by putting declaration of ifp inside #ifdef
NFE_DEBUG.

ok dlg@

Don't define NFE_DEBUG by default.
Set default debug level to 0.
Make sure to include interface or function name in debug strings.
Print Tx errors from the MAC when debug is not on.

check for M_PKTHDR.

fix nfe_txeof() to reset the wathdog timeout only when a full tx frame
has been sent.

re-enable interrupt mitigation and mask out NFE_IRQ_TIMER that was causing
interrupts flood.

update the media settings after MAC setup.

disable interrupt mitigation until i figure out why we're spending 10% of
CPU time in interrupts with it.

- remove mbuf linearization code. it is broken in this context and it is very
unlikely that we will see mbuf chains with 62 fragments anytime soon
(anyway, it would not crash, it would just freeze TX).
- fix max scatter value so we don't end up filling the ring with one mbuf
chain.

- fix h/w VLAN tagging and enable it for adapters that support it (VLAN tag
stripping job is left to the network stack).
- enable interrupt mitigation by default.
- add some magic to the initialization sequence in the hope that it will fix
TX issues seen on some adapters.

- stop enabling/disabling interrupts in nfe_intr().
- store RX/TX settings in nfe_softc so we don't recompute them all the time.
- fix h/w VLAN tagging flags.

move setting IFCAP_VLAN_MTU back to where it is supposed to be.

add support for 64bit rings base addresses.

actually disable h/w VLAN tagging for now.

- enable jumbo frames for adapters that support it.
- extend TX DMA mappings size from MCLBYTES to NFE_JBYTES.
- add initial (disabled) bits for interrupts mitigation.
- add initial (disabled) bits for h/w VLAN tagging.
- did some consistency tweaks while i'm here.

use pci_mapreg_type().

ok damien@

nVidia/nvidia -> NVIDIA

fix a printf

don't print shared interrupts when debugging is on.

use BUS_DMA_READ flag when mapping RX buffers.

reduce the number of DMA sync operations by sincing multiple TX descriptors
at once.

initial jumbo frames support (disabled for now).
#define'ing NFE_NO_JUMBO can save a few hundred KB of wired memory.

set IFCAP_VLAN_MTU in the capabilites field.

ok damien@

- call nfe_ifmedia_upd() early in nfe_init() to reset the PHY.
- set RX buffer size register properly

fix nfe_free_tx_ring() and nfe_reset_tx_ring() to unmap the good items.

don't free the mbuf in nfe_encap() if it can't be DMA'mapped since it is
left in if_snd. prettify nfe_start() a bit while i'm here.

MCP51 boards don't support jumbo frames..

force a wakeup of the MAC in nfe_init().
this makes my MCP51 board working.

committed over a nfe.

Only kick the TX engine once in nfe_start() and only if there are
packets to be transmitted, after iterating through the queue and
queuing up as many packets as possible as opposed to kicking
it each time through nfe_encap().

ok damien@ jsg@

Add back minimal debugging to help track down tx errors some MACs
seem to be reporting.

Use __LP64__ instead of __amd64__ for portability.
Pointed out by miod@.

Disable checksum offload for the moment as it appears to not
work for some people.

Simplify SIOCSIFADDR switch case a bit.

Eliminate a reset when configuring the IP address.

ok jsg@

Don't force NFE_DEBUG, set default debug level to 0.
Committed via an nfe.

quiet nfe_intr().

unbreak the tree.

pointed at by Peter Stromberg.

fix nfe_txeof().
with these changes, the connection seems to work just fine.

with help from jsg@

- start MII timer in nfe_init() and move timer initialization in nfe_attach()
- set RX filter before enabling RX in nfe_init()
- call mii_down() in nfe_stop()
- fix setting of full/half-duplex mode
- call mii_phy_reset() for each PHY attached and call mii_mediachg() in
nfe_ifmedia_upd()
- some cleaning while i'm here

fix clearing of h/w csum flags in all but the first fragment

handle link state change interrupts but do nothing for now

- add support for multicast filters.
- fix setting of if_capabilities flags for chips supporting checksum offload.
- fix dmesg output in case we can't establish the intr handler.
- fix a call to bus_dmamap_unload() in nfe_encap().

fixes miibus_{read,write}reg routines.
this resolves the ghost ukphy problem.
did some cleanup while i'm here.

tested by and ok jsg@

remove redundant code from nfe_attach().

ok jsg@

initial Tx/Rx bits. not working yet.

joint work with jsg@

fix dma mapping of tx ring.

Make sure interrupt is properly ack'd.
From damien@

Don't dynamically assign ring size, simplifies ring structures
and code that deals with rings. We can get away with this
as we only have one type of rx and one type of tx ring at a time
unlike ral(4).

Suggested by and ok damien@

Split length and flags up into seperate variables in
descriptors and make use of MCLBYTES for length setting.
Sugested by damien@

We aren't likely to need to keep track of crypted/decrypted ring
bits any time soon... pointed out by damien@

Initial bits for an nvidia nforce Ethernet driver.
bus_dma usage modelled after ral. Does not yet see rx interrupts
when testing with ck804.

Nvidia won't give out documentation for this, various "free" operating
systems include a closed source driver, and the Linux people who reverse
engineered it to create a specification won't give it out.

Make ifq and ifiq interface MP safe.

Rename ifq_set_maxlen() to ifq_init_maxlen(). This function neither
uses WRITE_ONCE() nor a mutex and is called before the ifq mutex
is initialized. The new name expresses that it should be used only
during interface attach when there is no concurrency.

Protect ifq_len(), ifq_empty(), ifiq_len(), and ifiq_empty() with
READ_ONCE(). They can be used without lock as they only read a
single integer.

OK dlg@

Constify struct cfattach.

Rename the macro MCLGETI to MCLGETL and removes the dead parameter ifp.

OK dlg@, bluhm@
No Opinion mpi@
Not against it claudio@

Change users of IFQ_SET_MAXLEN() and IFQ_IS_EMPTY() to use the "new" API.

ok dlg@ tobhe@

Argument order fix for MCLGETI.

ok claudio@

If you use sys/param.h, you don't need sys/types.h

move counting if_opackets next to counting if_obytes in if_enqueue.

this means packets are consistently counted in one place, unlike the
many and various ways that drivers thought they should do it.

ok mpi@ deraadt@

m_free() and m_freem() test for NULL. Simplify callers which had their own
NULL tests.

ok mpi@

G/C IFQ_SET_READY().

replace IFF_OACTIVE manipulation with mpsafe operations.

there are two things shared between the network stack and drivers
in the send path: the send queue and the IFF_OACTIVE flag. the send
queue is now protected by a mutex. this diff makes the oactive
functionality mpsafe too.

IFF_OACTIVE is part of if_flags. there are two problems with that.
firstly, if_flags is a short and we dont have any MI atomic operations
to manipulate a short. secondly, while we could make the IFF_OACTIVE
operates mpsafe, all changes to other flags would have to be made
safe at the same time, otherwise a read-modify-write cycle on their
updates could clobber the oactive change.

instead, this moves the oactive mark into struct ifqueue and provides
an API for changing it. there's ifq_set_oactive, ifq_clr_oactive,
and ifq_is_oactive. these are modelled on ifsq_set_oactive,
ifsq_clr_oactive, and ifsq_is_oactive in dragonflybsd.

this diff includes changes to all the drivers manipulating IFF_OACTIVE
to now use the ifsq_{set,clr_is}_oactive API too.

ok kettenis@ mpi@ jmatthew@ deraadt@

You only need <net/if_dl.h> if you're using LLADDR() or a sockaddr_dl.

shuffle struct ifqueue so in flight mbufs are protected by a mutex.

the code is refactored so the IFQ macros call newly implemented ifq
functions. the ifq code is split so each discipline (priq and hfsc
in our case) is an opaque set of operations that the common ifq
code can call. the common code does the locking, accounting (ifq_len
manipulation), and freeing of the mbuf if the disciplines enqueue
function rejects it. theyre kind of like bufqs in the block layer
with their fifo and nscan disciplines.

the new api also supports atomic switching of disciplines at runtime.
the hfsc setup in pf_ioctl.c has been tweaked to build a complete
hfsc_if structure which it attaches to the send queue in a single
operation, rather than attaching to the interface up front and
building up a list of queues.

the send queue is now mutexed, which raises the expectation that
packets can be enqueued or purged on one cpu while another cpu is
dequeueing them in a driver for transmission. a lot of drivers use
IFQ_POLL to peek at an mbuf and attempt to fit it on the ring before
committing to it with a later IFQ_DEQUEUE operation. if the mbuf
gets freed in between the POLL and DEQUEUE operations, fireworks
will ensue.

to avoid this, the ifq api introduces ifq_deq_begin, ifq_deq_rollback,
and ifq_deq_commit. ifq_deq_begin allows a driver to take the ifq
mutex and get a reference to the mbuf they wish to try and tx. if
there's space, they can ifq_deq_commit it to remove the mbuf and
release the mutex. if there's no space, ifq_deq_rollback simply
releases the mutex. this api was developed to make updating the
drivers using IFQ_POLL easy, instead of having to do significant
semantic changes to avoid POLL that we cannot test on all the
hardware.

the common code has been tested pretty hard, and all the driver
modifications are straightforward except for de(4). if that breaks
it can be dealt with later.

ok mpi@ jmatthew@

Do not include <net/if_vlan_var.h> when it's not necessary.

Because of the VLAN hacks in mpw(4) this file still contains the definition
of "struct ifvlan" which depends on <sys/refcnt.h> which in turns pull
<sys/atomic.h>...

arp_ifinit() is no longer needed.

Increment if_ipackets in if_input().

Note that pseudo-drivers not using if_input() are not affected by this
conversion.

ok mikeb@, kettenis@, claudio@, dlg@

Convert to if_input(), thanks to krw@ for testing.

Remove some includes include-what-you-use claims don't
have any direct symbols used. Tested for indirect use by compiling
amd64/i386/sparc64 kernels.

ok tedu@ deraadt@

unifdef INET

after allocating an mbuf and cluster you still need to init the length
fields.

found by steven roberts, who also tested this fix for me

remove the custom jumbo allocator. its never been enabled or used.

putting this into the tree to make it easier to test.

Fewer <netinet/in_systm.h>

The few network drivers that called their children's (ie. mii PHY
drivers) activate functions at DVACT_RESUME time do not need to do
so, since their PHYs are repaired by IFF_UP.

Add a DVACT_WAKEUP op to the *_activate() API. This is called after the
kernel resumes normal (non-cold, able to run processes, etc) operation.
Previously we were relying on specific DVACT_RESUME op's in drivers
creating callback/threads themselves, but that has become too common,
indicating the need for a built-in mechanism.
ok dlg kettenis, tested by a sufficient amount of people

Most network drivers include netinet/in_var.h, but apparently they
don't have to. Just remove these include lines.
Compiled on amd64 i386 sparc64; OK henning@ mikeb@

Rewrite receive filter and ioctl handling code.

ok jsg@

Remove setting an initial assumed baudrate upon driver attach which is not
necessarily correct, there might not even be a link when attaching.

ok mikeb@ reyk@

Wake on LAN support for nfe(4). Tested by me, russel on misc@, and jsg.
ok jsg

mechanic rename M_{TCP|UDP}V4_CSUM_OUT -> M_{TCP|UDP}_CSUM_OUT
ok claudio krw

Some nfe(4)/rlphy(4) combos don't work, because the PHY responds to all
addresses on the mii bus. As a countereasure, only attach the first PHY we
encounter. It is very unlikely we're going to ever see nfe(4) with multiple
PHYs. The same is probably true for any modern NIC.

ok mikeb@, deraadt@

remove the powerhook code. All architectures now use the ca_activate tree
traversal code to suspend/resume
ok oga kettenis blambert

Add DVACT_QUIECE support. This is called before splhigh() and before
DVACT_SUSPEND, therefore DVACT_QUIECE can do standard sleeping operations
to get ready.
Discussed quite a while back with kettenis and jakemsr, oga suddenly needed
it as well and wrote half of it, so it was time to finish it.
proofread by miod.

activate function should return result of config_activate_children

Make the powerhook use the activate functions, which actually do the full
job.

remove the unused if_init callback in struct ifnet
ok deraadt@ henning@ claudio@

ca_activate function for nfe(4) for suspend/resume
tested on nvidia mcp51

ok deraadt@

BUS_DMA_ZERO instead of alloc, map, bzero.

ok krw@

Remove support for cancelled NVIDIA MCP7B and add initial support for
MCP89.

From Brad based on information from Peer Chen @ NVIDIA via Linux.

make various strings ("can't map mem space" and similar) more consistent
between instances, saving space in the kernel. feedback from many (some
incorporated, some left for future work).

ok deraadt, kettenis, "why not" miod.

Eliminate the redundant bits of code for MTU and multicast handling
from the individual drivers now that ether_ioctl() handles this.

Shrinks the i386 kernels by..
RAMDISK - 2176 bytes
RAMDISKB - 1504 bytes
RAMDISKC - 736 bytes

Tested by naddy@/okan@/sthen@/brad@/todd@/jmc@ and lots of users.
Build tested on almost all archs by todd@/brad@

ok naddy@

Introduce bpf_mtap_ether(), which for the benefit of bpf listeners
creates the VLAN encapsulation from the tag stored in the mbuf
header. Idea from FreeBSD, input from claudio@ and canacar@.

Switch all hardware VLAN enabled drivers to the new function.

ok claudio@

Remove #if NVLAN.

Re-add support for RX VLAN tag stripping.

Switch the existing TX VLAN hardware support over to having the
tag in the header. Convert TX tagging in the drivers.

Help and ok brad@

First step towards cleaning up the Ethernet driver ioctl handling.
Move calling ether_ioctl() from the top of the ioctl function, which
at the moment does absolutely nothing, to the default switch case.
Thus allowing drivers to define their own ioctl handlers and then
falling back on ether_ioctl(). The only functional change this results
in at the moment is having all Ethernet drivers returning the proper
errno of ENOTTY instead of EINVAL/ENXIO when encountering unknown
ioctl's.

Shrinks the i386 kernels by..
RAMDISK - 1024 bytes
RAMDISKB - 1120 bytes
RAMDISKC - 832 bytes

Tested by martin@/jsing@/todd@/brad@
Build tested on almost all archs by todd@/brad@

ok jsing@

Convert timeout_add() calls using multiples of hz to timeout_add_sec()

Really just the low-hanging fruit of (hopefully) forthcoming timeout
conversions.

ok art@, krw@

MCP79 are also capable of Jumbo frames. Add the Jumbo support flag.

From: Linux forcedeth

Simplify the combination use of pci_mapreg_type()/pci_mapreg_map() as
suggested by dlg@ awhile ago.

ok dlg@

correct format after %x -> %b convertion. deraadt jsg ok.

Add the NFE_CORRECT_MACADDR flag for MCP77/79 chipsets.

ok jsg@ dlg@

add the NVIDIA MCP77/79 ids.

ok jsg@ dlg@

Fix check for pending interrupts as some other device is causing
the status register to change.

From form@.

ok jsg

Make sure newer adapters are not in powerdown mode.
From Oleg Safiullin <form@pdp-11.org.ru> similiar to changes
in Linux/FreeBSD driver.

Disable jumbo allocator until the low number of buffers case
is better handled.

This will likely resolve stalls some people are seeing under high
load.

ok damien@

In nfe_start() do a fast return if IFF_OACTIVE is set, in
this case we need a Tx interrupt to clean up the DMA ring
before if_start can be properly called.

Diff based on a comment by & ok claudio@

use new malloc M_ZERO flag to shrink kernel.
remove <malloc.h> from files where malloc is not used.

Correct the backwards ethernet address that some NVidia MACs have.
diff from brad. "commit this" jsg

MCP65 has no hardware checksum support.
Diff from brad after he noticed recent changes
in the Linux driver by an NVIDIA employee.

Add support for HW TCP/IP checksum offload for adapters that support it.

Tested by many (IP/UDP/TCP):

Jason McIntyre <jmc@>
Chris Kuethe <chris.kuethe AT gmail.com>
Alf Schlichting <a.schlichting AT lemarit.com>
Rodolfo Gouveia <rgouveia AT cosmico.net>
Peter Stromberg <wilfried@>

Has been in snaps for weeks too with noone complaining so far.

Correct the WOL magic value and rename NFE_WOL_MAGIC to NFE_WOL_ENABLE.

WOL magic value from Peer Chen@NVIDIA via FreeBSD.

ok jsg@ damien@

Defer setting of the valid bit in the first TX descriptor after
all descriptors have been setup. Otherwise, hardware may start
processing descriptors faster than us and crap out.
Fixes "watchdog timeout" errors.

Original idea from Matthew Dillon @DragonFly.

ok deraadt@ jsg@ wim@
tested by many

add the NVIDIA MCP67 LAN PCI ids.

branches: 1.64.2;
add NVidia MCP61/65 ids.

add sys/timeout.h

garbage collect NFE_NO_JUMBO.

- remove ETHER_MAX_LEN_JUMBO and ETHERMTU_JUMBO.
- use if_hardmtu for MTU ioctl handlers.

ok reyk@

unknown ioctl is ENOTTY not EINVAL

remove IFCAP_JUMBO_MTU interface capabilities flag and set if_hardmtu in a few
more drivers.

ok reyk@

set if_jumbo_mtu and the IFCAP_JUMBO_MTU capabilities flag where
appropriate.

ok reyk@

Use %b in error flag printfs to describe meaning of error bits.
requested by deraadt@

And commit the diff from the right system.

check for IFF_RUNNING being set before calling nfe_init().

ok damien@ jsg@

Add work around for mbuf leak in the tx path until we
can come up with a better guess as to how the hardware works.
From Chuck Silvers.
ok damien@

allow bpf(4) to ignore packets based on their direction (inbound or
outbound), using a new BIOCSDIRFILT ioctl;
guidance, feedback and ok canacar@

branches: 1.52.2;
Put the PHY update/reset call in nfe_init back to the spot it was at
before Rx/Tx/Interrupts are enabled.
This makes <fredd at cse.sc.edu>'s nfe+icsphy setup work again.

Tested on nfe+eephy by otto@ and myself, nfe+ciphy by otto@ and
nfe+rlphy by wilfried@

ok deraadt@

use sc->sc_dev.dv_xname consistently.

Let if_nfe.c compile again by putting declaration of ifp inside #ifdef
NFE_DEBUG.

ok dlg@

Don't define NFE_DEBUG by default.
Set default debug level to 0.
Make sure to include interface or function name in debug strings.
Print Tx errors from the MAC when debug is not on.

check for M_PKTHDR.

fix nfe_txeof() to reset the wathdog timeout only when a full tx frame
has been sent.

re-enable interrupt mitigation and mask out NFE_IRQ_TIMER that was causing
interrupts flood.

update the media settings after MAC setup.

disable interrupt mitigation until i figure out why we're spending 10% of
CPU time in interrupts with it.

- remove mbuf linearization code. it is broken in this context and it is very
unlikely that we will see mbuf chains with 62 fragments anytime soon
(anyway, it would not crash, it would just freeze TX).
- fix max scatter value so we don't end up filling the ring with one mbuf
chain.

- fix h/w VLAN tagging and enable it for adapters that support it (VLAN tag
stripping job is left to the network stack).
- enable interrupt mitigation by default.
- add some magic to the initialization sequence in the hope that it will fix
TX issues seen on some adapters.

- stop enabling/disabling interrupts in nfe_intr().
- store RX/TX settings in nfe_softc so we don't recompute them all the time.
- fix h/w VLAN tagging flags.

move setting IFCAP_VLAN_MTU back to where it is supposed to be.

add support for 64bit rings base addresses.

actually disable h/w VLAN tagging for now.

- enable jumbo frames for adapters that support it.
- extend TX DMA mappings size from MCLBYTES to NFE_JBYTES.
- add initial (disabled) bits for interrupts mitigation.
- add initial (disabled) bits for h/w VLAN tagging.
- did some consistency tweaks while i'm here.

use pci_mapreg_type().

ok damien@

nVidia/nvidia -> NVIDIA

fix a printf

don't print shared interrupts when debugging is on.

use BUS_DMA_READ flag when mapping RX buffers.

reduce the number of DMA sync operations by sincing multiple TX descriptors
at once.

initial jumbo frames support (disabled for now).
#define'ing NFE_NO_JUMBO can save a few hundred KB of wired memory.

set IFCAP_VLAN_MTU in the capabilites field.

ok damien@

- call nfe_ifmedia_upd() early in nfe_init() to reset the PHY.
- set RX buffer size register properly

fix nfe_free_tx_ring() and nfe_reset_tx_ring() to unmap the good items.

don't free the mbuf in nfe_encap() if it can't be DMA'mapped since it is
left in if_snd. prettify nfe_start() a bit while i'm here.

MCP51 boards don't support jumbo frames..

force a wakeup of the MAC in nfe_init().
this makes my MCP51 board working.

committed over a nfe.

Only kick the TX engine once in nfe_start() and only if there are
packets to be transmitted, after iterating through the queue and
queuing up as many packets as possible as opposed to kicking
it each time through nfe_encap().

ok damien@ jsg@

Add back minimal debugging to help track down tx errors some MACs
seem to be reporting.

Use __LP64__ instead of __amd64__ for portability.
Pointed out by miod@.

Disable checksum offload for the moment as it appears to not
work for some people.

Simplify SIOCSIFADDR switch case a bit.

Eliminate a reset when configuring the IP address.

ok jsg@

Don't force NFE_DEBUG, set default debug level to 0.
Committed via an nfe.

quiet nfe_intr().

unbreak the tree.

pointed at by Peter Stromberg.

fix nfe_txeof().
with these changes, the connection seems to work just fine.

with help from jsg@

- start MII timer in nfe_init() and move timer initialization in nfe_attach()
- set RX filter before enabling RX in nfe_init()
- call mii_down() in nfe_stop()
- fix setting of full/half-duplex mode
- call mii_phy_reset() for each PHY attached and call mii_mediachg() in
nfe_ifmedia_upd()
- some cleaning while i'm here

fix clearing of h/w csum flags in all but the first fragment

handle link state change interrupts but do nothing for now

- add support for multicast filters.
- fix setting of if_capabilities flags for chips supporting checksum offload.
- fix dmesg output in case we can't establish the intr handler.
- fix a call to bus_dmamap_unload() in nfe_encap().

fixes miibus_{read,write}reg routines.
this resolves the ghost ukphy problem.
did some cleanup while i'm here.

tested by and ok jsg@

remove redundant code from nfe_attach().

ok jsg@

initial Tx/Rx bits. not working yet.

joint work with jsg@

fix dma mapping of tx ring.

Make sure interrupt is properly ack'd.
From damien@

Don't dynamically assign ring size, simplifies ring structures
and code that deals with rings. We can get away with this
as we only have one type of rx and one type of tx ring at a time
unlike ral(4).

Suggested by and ok damien@

Split length and flags up into seperate variables in
descriptors and make use of MCLBYTES for length setting.
Sugested by damien@

We aren't likely to need to keep track of crypted/decrypted ring
bits any time soon... pointed out by damien@

Initial bits for an nvidia nforce Ethernet driver.
bus_dma usage modelled after ral. Does not yet see rx interrupts
when testing with ck804.

Nvidia won't give out documentation for this, various "free" operating
systems include a closed source driver, and the Linux people who reverse
engineered it to create a specification won't give it out.

Constify struct cfattach.

Rename the macro MCLGETI to MCLGETL and removes the dead parameter ifp.

OK dlg@, bluhm@
No Opinion mpi@
Not against it claudio@

Change users of IFQ_SET_MAXLEN() and IFQ_IS_EMPTY() to use the "new" API.

ok dlg@ tobhe@

Argument order fix for MCLGETI.

ok claudio@

If you use sys/param.h, you don't need sys/types.h

move counting if_opackets next to counting if_obytes in if_enqueue.

this means packets are consistently counted in one place, unlike the
many and various ways that drivers thought they should do it.

ok mpi@ deraadt@

m_free() and m_freem() test for NULL. Simplify callers which had their own
NULL tests.

ok mpi@

G/C IFQ_SET_READY().

replace IFF_OACTIVE manipulation with mpsafe operations.

there are two things shared between the network stack and drivers
in the send path: the send queue and the IFF_OACTIVE flag. the send
queue is now protected by a mutex. this diff makes the oactive
functionality mpsafe too.

IFF_OACTIVE is part of if_flags. there are two problems with that.
firstly, if_flags is a short and we dont have any MI atomic operations
to manipulate a short. secondly, while we could make the IFF_OACTIVE
operates mpsafe, all changes to other flags would have to be made
safe at the same time, otherwise a read-modify-write cycle on their
updates could clobber the oactive change.

instead, this moves the oactive mark into struct ifqueue and provides
an API for changing it. there's ifq_set_oactive, ifq_clr_oactive,
and ifq_is_oactive. these are modelled on ifsq_set_oactive,
ifsq_clr_oactive, and ifsq_is_oactive in dragonflybsd.

this diff includes changes to all the drivers manipulating IFF_OACTIVE
to now use the ifsq_{set,clr_is}_oactive API too.

ok kettenis@ mpi@ jmatthew@ deraadt@

You only need <net/if_dl.h> if you're using LLADDR() or a sockaddr_dl.

shuffle struct ifqueue so in flight mbufs are protected by a mutex.

the code is refactored so the IFQ macros call newly implemented ifq
functions. the ifq code is split so each discipline (priq and hfsc
in our case) is an opaque set of operations that the common ifq
code can call. the common code does the locking, accounting (ifq_len
manipulation), and freeing of the mbuf if the disciplines enqueue
function rejects it. theyre kind of like bufqs in the block layer
with their fifo and nscan disciplines.

the new api also supports atomic switching of disciplines at runtime.
the hfsc setup in pf_ioctl.c has been tweaked to build a complete
hfsc_if structure which it attaches to the send queue in a single
operation, rather than attaching to the interface up front and
building up a list of queues.

the send queue is now mutexed, which raises the expectation that
packets can be enqueued or purged on one cpu while another cpu is
dequeueing them in a driver for transmission. a lot of drivers use
IFQ_POLL to peek at an mbuf and attempt to fit it on the ring before
committing to it with a later IFQ_DEQUEUE operation. if the mbuf
gets freed in between the POLL and DEQUEUE operations, fireworks
will ensue.

to avoid this, the ifq api introduces ifq_deq_begin, ifq_deq_rollback,
and ifq_deq_commit. ifq_deq_begin allows a driver to take the ifq
mutex and get a reference to the mbuf they wish to try and tx. if
there's space, they can ifq_deq_commit it to remove the mbuf and
release the mutex. if there's no space, ifq_deq_rollback simply
releases the mutex. this api was developed to make updating the
drivers using IFQ_POLL easy, instead of having to do significant
semantic changes to avoid POLL that we cannot test on all the
hardware.

the common code has been tested pretty hard, and all the driver
modifications are straightforward except for de(4). if that breaks
it can be dealt with later.

ok mpi@ jmatthew@

Do not include <net/if_vlan_var.h> when it's not necessary.

Because of the VLAN hacks in mpw(4) this file still contains the definition
of "struct ifvlan" which depends on <sys/refcnt.h> which in turns pull
<sys/atomic.h>...

arp_ifinit() is no longer needed.

Increment if_ipackets in if_input().

Note that pseudo-drivers not using if_input() are not affected by this
conversion.

ok mikeb@, kettenis@, claudio@, dlg@

Convert to if_input(), thanks to krw@ for testing.

Remove some includes include-what-you-use claims don't
have any direct symbols used. Tested for indirect use by compiling
amd64/i386/sparc64 kernels.

ok tedu@ deraadt@

unifdef INET

after allocating an mbuf and cluster you still need to init the length
fields.

found by steven roberts, who also tested this fix for me

remove the custom jumbo allocator. its never been enabled or used.

putting this into the tree to make it easier to test.

Fewer <netinet/in_systm.h>

The few network drivers that called their children's (ie. mii PHY
drivers) activate functions at DVACT_RESUME time do not need to do
so, since their PHYs are repaired by IFF_UP.

Add a DVACT_WAKEUP op to the *_activate() API. This is called after the
kernel resumes normal (non-cold, able to run processes, etc) operation.
Previously we were relying on specific DVACT_RESUME op's in drivers
creating callback/threads themselves, but that has become too common,
indicating the need for a built-in mechanism.
ok dlg kettenis, tested by a sufficient amount of people

Most network drivers include netinet/in_var.h, but apparently they
don't have to. Just remove these include lines.
Compiled on amd64 i386 sparc64; OK henning@ mikeb@

Rewrite receive filter and ioctl handling code.

ok jsg@

Remove setting an initial assumed baudrate upon driver attach which is not
necessarily correct, there might not even be a link when attaching.

ok mikeb@ reyk@

Wake on LAN support for nfe(4). Tested by me, russel on misc@, and jsg.
ok jsg

mechanic rename M_{TCP|UDP}V4_CSUM_OUT -> M_{TCP|UDP}_CSUM_OUT
ok claudio krw

Some nfe(4)/rlphy(4) combos don't work, because the PHY responds to all
addresses on the mii bus. As a countereasure, only attach the first PHY we
encounter. It is very unlikely we're going to ever see nfe(4) with multiple
PHYs. The same is probably true for any modern NIC.

ok mikeb@, deraadt@

remove the powerhook code. All architectures now use the ca_activate tree
traversal code to suspend/resume
ok oga kettenis blambert

Add DVACT_QUIECE support. This is called before splhigh() and before
DVACT_SUSPEND, therefore DVACT_QUIECE can do standard sleeping operations
to get ready.
Discussed quite a while back with kettenis and jakemsr, oga suddenly needed
it as well and wrote half of it, so it was time to finish it.
proofread by miod.

activate function should return result of config_activate_children

Make the powerhook use the activate functions, which actually do the full
job.

remove the unused if_init callback in struct ifnet
ok deraadt@ henning@ claudio@

ca_activate function for nfe(4) for suspend/resume
tested on nvidia mcp51

ok deraadt@

BUS_DMA_ZERO instead of alloc, map, bzero.

ok krw@

Remove support for cancelled NVIDIA MCP7B and add initial support for
MCP89.

From Brad based on information from Peer Chen @ NVIDIA via Linux.

make various strings ("can't map mem space" and similar) more consistent
between instances, saving space in the kernel. feedback from many (some
incorporated, some left for future work).

ok deraadt, kettenis, "why not" miod.

Eliminate the redundant bits of code for MTU and multicast handling
from the individual drivers now that ether_ioctl() handles this.

Shrinks the i386 kernels by..
RAMDISK - 2176 bytes
RAMDISKB - 1504 bytes
RAMDISKC - 736 bytes

Tested by naddy@/okan@/sthen@/brad@/todd@/jmc@ and lots of users.
Build tested on almost all archs by todd@/brad@

ok naddy@

Introduce bpf_mtap_ether(), which for the benefit of bpf listeners
creates the VLAN encapsulation from the tag stored in the mbuf
header. Idea from FreeBSD, input from claudio@ and canacar@.

Switch all hardware VLAN enabled drivers to the new function.

ok claudio@

Remove #if NVLAN.

Re-add support for RX VLAN tag stripping.

Switch the existing TX VLAN hardware support over to having the
tag in the header. Convert TX tagging in the drivers.

Help and ok brad@

First step towards cleaning up the Ethernet driver ioctl handling.
Move calling ether_ioctl() from the top of the ioctl function, which
at the moment does absolutely nothing, to the default switch case.
Thus allowing drivers to define their own ioctl handlers and then
falling back on ether_ioctl(). The only functional change this results
in at the moment is having all Ethernet drivers returning the proper
errno of ENOTTY instead of EINVAL/ENXIO when encountering unknown
ioctl's.

Shrinks the i386 kernels by..
RAMDISK - 1024 bytes
RAMDISKB - 1120 bytes
RAMDISKC - 832 bytes

Tested by martin@/jsing@/todd@/brad@
Build tested on almost all archs by todd@/brad@

ok jsing@

Convert timeout_add() calls using multiples of hz to timeout_add_sec()

Really just the low-hanging fruit of (hopefully) forthcoming timeout
conversions.

ok art@, krw@

MCP79 are also capable of Jumbo frames. Add the Jumbo support flag.

From: Linux forcedeth

Simplify the combination use of pci_mapreg_type()/pci_mapreg_map() as
suggested by dlg@ awhile ago.

ok dlg@

correct format after %x -> %b convertion. deraadt jsg ok.

Add the NFE_CORRECT_MACADDR flag for MCP77/79 chipsets.

ok jsg@ dlg@

add the NVIDIA MCP77/79 ids.

ok jsg@ dlg@

Fix check for pending interrupts as some other device is causing
the status register to change.

From form@.

ok jsg

Make sure newer adapters are not in powerdown mode.
From Oleg Safiullin <form@pdp-11.org.ru> similiar to changes
in Linux/FreeBSD driver.

Disable jumbo allocator until the low number of buffers case
is better handled.

This will likely resolve stalls some people are seeing under high
load.

ok damien@

In nfe_start() do a fast return if IFF_OACTIVE is set, in
this case we need a Tx interrupt to clean up the DMA ring
before if_start can be properly called.

Diff based on a comment by & ok claudio@

use new malloc M_ZERO flag to shrink kernel.
remove <malloc.h> from files where malloc is not used.

Correct the backwards ethernet address that some NVidia MACs have.
diff from brad. "commit this" jsg

MCP65 has no hardware checksum support.
Diff from brad after he noticed recent changes
in the Linux driver by an NVIDIA employee.

Add support for HW TCP/IP checksum offload for adapters that support it.

Tested by many (IP/UDP/TCP):

Jason McIntyre <jmc@>
Chris Kuethe <chris.kuethe AT gmail.com>
Alf Schlichting <a.schlichting AT lemarit.com>
Rodolfo Gouveia <rgouveia AT cosmico.net>
Peter Stromberg <wilfried@>

Has been in snaps for weeks too with noone complaining so far.

Correct the WOL magic value and rename NFE_WOL_MAGIC to NFE_WOL_ENABLE.

WOL magic value from Peer Chen@NVIDIA via FreeBSD.

ok jsg@ damien@

Defer setting of the valid bit in the first TX descriptor after
all descriptors have been setup. Otherwise, hardware may start
processing descriptors faster than us and crap out.
Fixes "watchdog timeout" errors.

Original idea from Matthew Dillon @DragonFly.

ok deraadt@ jsg@ wim@
tested by many

add the NVIDIA MCP67 LAN PCI ids.

branches: 1.64.2;
add NVidia MCP61/65 ids.

add sys/timeout.h

garbage collect NFE_NO_JUMBO.

- remove ETHER_MAX_LEN_JUMBO and ETHERMTU_JUMBO.
- use if_hardmtu for MTU ioctl handlers.

ok reyk@

unknown ioctl is ENOTTY not EINVAL

remove IFCAP_JUMBO_MTU interface capabilities flag and set if_hardmtu in a few
more drivers.

ok reyk@

set if_jumbo_mtu and the IFCAP_JUMBO_MTU capabilities flag where
appropriate.

ok reyk@

Use %b in error flag printfs to describe meaning of error bits.
requested by deraadt@

And commit the diff from the right system.

check for IFF_RUNNING being set before calling nfe_init().

ok damien@ jsg@

Add work around for mbuf leak in the tx path until we
can come up with a better guess as to how the hardware works.
From Chuck Silvers.
ok damien@

allow bpf(4) to ignore packets based on their direction (inbound or
outbound), using a new BIOCSDIRFILT ioctl;
guidance, feedback and ok canacar@

branches: 1.52.2;
Put the PHY update/reset call in nfe_init back to the spot it was at
before Rx/Tx/Interrupts are enabled.
This makes <fredd at cse.sc.edu>'s nfe+icsphy setup work again.

Tested on nfe+eephy by otto@ and myself, nfe+ciphy by otto@ and
nfe+rlphy by wilfried@

ok deraadt@

use sc->sc_dev.dv_xname consistently.

Let if_nfe.c compile again by putting declaration of ifp inside #ifdef
NFE_DEBUG.

ok dlg@

Don't define NFE_DEBUG by default.
Set default debug level to 0.
Make sure to include interface or function name in debug strings.
Print Tx errors from the MAC when debug is not on.

check for M_PKTHDR.

fix nfe_txeof() to reset the wathdog timeout only when a full tx frame
has been sent.

re-enable interrupt mitigation and mask out NFE_IRQ_TIMER that was causing
interrupts flood.

update the media settings after MAC setup.

disable interrupt mitigation until i figure out why we're spending 10% of
CPU time in interrupts with it.

- remove mbuf linearization code. it is broken in this context and it is very
unlikely that we will see mbuf chains with 62 fragments anytime soon
(anyway, it would not crash, it would just freeze TX).
- fix max scatter value so we don't end up filling the ring with one mbuf
chain.

- fix h/w VLAN tagging and enable it for adapters that support it (VLAN tag
stripping job is left to the network stack).
- enable interrupt mitigation by default.
- add some magic to the initialization sequence in the hope that it will fix
TX issues seen on some adapters.

- stop enabling/disabling interrupts in nfe_intr().
- store RX/TX settings in nfe_softc so we don't recompute them all the time.
- fix h/w VLAN tagging flags.

move setting IFCAP_VLAN_MTU back to where it is supposed to be.

add support for 64bit rings base addresses.

actually disable h/w VLAN tagging for now.

- enable jumbo frames for adapters that support it.
- extend TX DMA mappings size from MCLBYTES to NFE_JBYTES.
- add initial (disabled) bits for interrupts mitigation.
- add initial (disabled) bits for h/w VLAN tagging.
- did some consistency tweaks while i'm here.

use pci_mapreg_type().

ok damien@

nVidia/nvidia -> NVIDIA

fix a printf

don't print shared interrupts when debugging is on.

use BUS_DMA_READ flag when mapping RX buffers.

reduce the number of DMA sync operations by sincing multiple TX descriptors
at once.

initial jumbo frames support (disabled for now).
#define'ing NFE_NO_JUMBO can save a few hundred KB of wired memory.

set IFCAP_VLAN_MTU in the capabilites field.

ok damien@

- call nfe_ifmedia_upd() early in nfe_init() to reset the PHY.
- set RX buffer size register properly

fix nfe_free_tx_ring() and nfe_reset_tx_ring() to unmap the good items.

don't free the mbuf in nfe_encap() if it can't be DMA'mapped since it is
left in if_snd. prettify nfe_start() a bit while i'm here.

MCP51 boards don't support jumbo frames..

force a wakeup of the MAC in nfe_init().
this makes my MCP51 board working.

committed over a nfe.

Only kick the TX engine once in nfe_start() and only if there are
packets to be transmitted, after iterating through the queue and
queuing up as many packets as possible as opposed to kicking
it each time through nfe_encap().

ok damien@ jsg@

Add back minimal debugging to help track down tx errors some MACs
seem to be reporting.

Use __LP64__ instead of __amd64__ for portability.
Pointed out by miod@.

Disable checksum offload for the moment as it appears to not
work for some people.

Simplify SIOCSIFADDR switch case a bit.

Eliminate a reset when configuring the IP address.

ok jsg@

Don't force NFE_DEBUG, set default debug level to 0.
Committed via an nfe.

quiet nfe_intr().

unbreak the tree.

pointed at by Peter Stromberg.

fix nfe_txeof().
with these changes, the connection seems to work just fine.

with help from jsg@

- start MII timer in nfe_init() and move timer initialization in nfe_attach()
- set RX filter before enabling RX in nfe_init()
- call mii_down() in nfe_stop()
- fix setting of full/half-duplex mode
- call mii_phy_reset() for each PHY attached and call mii_mediachg() in
nfe_ifmedia_upd()
- some cleaning while i'm here

fix clearing of h/w csum flags in all but the first fragment

handle link state change interrupts but do nothing for now

- add support for multicast filters.
- fix setting of if_capabilities flags for chips supporting checksum offload.
- fix dmesg output in case we can't establish the intr handler.
- fix a call to bus_dmamap_unload() in nfe_encap().

fixes miibus_{read,write}reg routines.
this resolves the ghost ukphy problem.
did some cleanup while i'm here.

tested by and ok jsg@

remove redundant code from nfe_attach().

ok jsg@

initial Tx/Rx bits. not working yet.

joint work with jsg@

fix dma mapping of tx ring.

Make sure interrupt is properly ack'd.
From damien@

Don't dynamically assign ring size, simplifies ring structures
and code that deals with rings. We can get away with this
as we only have one type of rx and one type of tx ring at a time
unlike ral(4).

Suggested by and ok damien@

Split length and flags up into seperate variables in
descriptors and make use of MCLBYTES for length setting.
Sugested by damien@

We aren't likely to need to keep track of crypted/decrypted ring
bits any time soon... pointed out by damien@

Initial bits for an nvidia nforce Ethernet driver.
bus_dma usage modelled after ral. Does not yet see rx interrupts
when testing with ck804.

Nvidia won't give out documentation for this, various "free" operating
systems include a closed source driver, and the Linux people who reverse
engineered it to create a specification won't give it out.

Rename the macro MCLGETI to MCLGETL and removes the dead parameter ifp.

OK dlg@, bluhm@
No Opinion mpi@
Not against it claudio@

Change users of IFQ_SET_MAXLEN() and IFQ_IS_EMPTY() to use the "new" API.

ok dlg@ tobhe@

Argument order fix for MCLGETI.

ok claudio@

If you use sys/param.h, you don't need sys/types.h

move counting if_opackets next to counting if_obytes in if_enqueue.

this means packets are consistently counted in one place, unlike the
many and various ways that drivers thought they should do it.

ok mpi@ deraadt@

m_free() and m_freem() test for NULL. Simplify callers which had their own
NULL tests.

ok mpi@

G/C IFQ_SET_READY().

replace IFF_OACTIVE manipulation with mpsafe operations.

there are two things shared between the network stack and drivers
in the send path: the send queue and the IFF_OACTIVE flag. the send
queue is now protected by a mutex. this diff makes the oactive
functionality mpsafe too.

IFF_OACTIVE is part of if_flags. there are two problems with that.
firstly, if_flags is a short and we dont have any MI atomic operations
to manipulate a short. secondly, while we could make the IFF_OACTIVE
operates mpsafe, all changes to other flags would have to be made
safe at the same time, otherwise a read-modify-write cycle on their
updates could clobber the oactive change.

instead, this moves the oactive mark into struct ifqueue and provides
an API for changing it. there's ifq_set_oactive, ifq_clr_oactive,
and ifq_is_oactive. these are modelled on ifsq_set_oactive,
ifsq_clr_oactive, and ifsq_is_oactive in dragonflybsd.

this diff includes changes to all the drivers manipulating IFF_OACTIVE
to now use the ifsq_{set,clr_is}_oactive API too.

ok kettenis@ mpi@ jmatthew@ deraadt@

You only need <net/if_dl.h> if you're using LLADDR() or a sockaddr_dl.

shuffle struct ifqueue so in flight mbufs are protected by a mutex.

the code is refactored so the IFQ macros call newly implemented ifq
functions. the ifq code is split so each discipline (priq and hfsc
in our case) is an opaque set of operations that the common ifq
code can call. the common code does the locking, accounting (ifq_len
manipulation), and freeing of the mbuf if the disciplines enqueue
function rejects it. theyre kind of like bufqs in the block layer
with their fifo and nscan disciplines.

the new api also supports atomic switching of disciplines at runtime.
the hfsc setup in pf_ioctl.c has been tweaked to build a complete
hfsc_if structure which it attaches to the send queue in a single
operation, rather than attaching to the interface up front and
building up a list of queues.

the send queue is now mutexed, which raises the expectation that
packets can be enqueued or purged on one cpu while another cpu is
dequeueing them in a driver for transmission. a lot of drivers use
IFQ_POLL to peek at an mbuf and attempt to fit it on the ring before
committing to it with a later IFQ_DEQUEUE operation. if the mbuf
gets freed in between the POLL and DEQUEUE operations, fireworks
will ensue.

to avoid this, the ifq api introduces ifq_deq_begin, ifq_deq_rollback,
and ifq_deq_commit. ifq_deq_begin allows a driver to take the ifq
mutex and get a reference to the mbuf they wish to try and tx. if
there's space, they can ifq_deq_commit it to remove the mbuf and
release the mutex. if there's no space, ifq_deq_rollback simply
releases the mutex. this api was developed to make updating the
drivers using IFQ_POLL easy, instead of having to do significant
semantic changes to avoid POLL that we cannot test on all the
hardware.

the common code has been tested pretty hard, and all the driver
modifications are straightforward except for de(4). if that breaks
it can be dealt with later.

ok mpi@ jmatthew@

Do not include <net/if_vlan_var.h> when it's not necessary.

Because of the VLAN hacks in mpw(4) this file still contains the definition
of "struct ifvlan" which depends on <sys/refcnt.h> which in turns pull
<sys/atomic.h>...

arp_ifinit() is no longer needed.

Increment if_ipackets in if_input().

Note that pseudo-drivers not using if_input() are not affected by this
conversion.

ok mikeb@, kettenis@, claudio@, dlg@

Convert to if_input(), thanks to krw@ for testing.

Remove some includes include-what-you-use claims don't
have any direct symbols used. Tested for indirect use by compiling
amd64/i386/sparc64 kernels.

ok tedu@ deraadt@

unifdef INET

after allocating an mbuf and cluster you still need to init the length
fields.

found by steven roberts, who also tested this fix for me

remove the custom jumbo allocator. its never been enabled or used.

putting this into the tree to make it easier to test.

Fewer <netinet/in_systm.h>

The few network drivers that called their children's (ie. mii PHY
drivers) activate functions at DVACT_RESUME time do not need to do
so, since their PHYs are repaired by IFF_UP.

Add a DVACT_WAKEUP op to the *_activate() API. This is called after the
kernel resumes normal (non-cold, able to run processes, etc) operation.
Previously we were relying on specific DVACT_RESUME op's in drivers
creating callback/threads themselves, but that has become too common,
indicating the need for a built-in mechanism.
ok dlg kettenis, tested by a sufficient amount of people

Most network drivers include netinet/in_var.h, but apparently they
don't have to. Just remove these include lines.
Compiled on amd64 i386 sparc64; OK henning@ mikeb@

Rewrite receive filter and ioctl handling code.

ok jsg@

Remove setting an initial assumed baudrate upon driver attach which is not
necessarily correct, there might not even be a link when attaching.

ok mikeb@ reyk@

Wake on LAN support for nfe(4). Tested by me, russel on misc@, and jsg.
ok jsg

mechanic rename M_{TCP|UDP}V4_CSUM_OUT -> M_{TCP|UDP}_CSUM_OUT
ok claudio krw

Some nfe(4)/rlphy(4) combos don't work, because the PHY responds to all
addresses on the mii bus. As a countereasure, only attach the first PHY we
encounter. It is very unlikely we're going to ever see nfe(4) with multiple
PHYs. The same is probably true for any modern NIC.

ok mikeb@, deraadt@

remove the powerhook code. All architectures now use the ca_activate tree
traversal code to suspend/resume
ok oga kettenis blambert

Add DVACT_QUIECE support. This is called before splhigh() and before
DVACT_SUSPEND, therefore DVACT_QUIECE can do standard sleeping operations
to get ready.
Discussed quite a while back with kettenis and jakemsr, oga suddenly needed
it as well and wrote half of it, so it was time to finish it.
proofread by miod.

activate function should return result of config_activate_children

Make the powerhook use the activate functions, which actually do the full
job.

remove the unused if_init callback in struct ifnet
ok deraadt@ henning@ claudio@

ca_activate function for nfe(4) for suspend/resume
tested on nvidia mcp51

ok deraadt@

BUS_DMA_ZERO instead of alloc, map, bzero.

ok krw@

Remove support for cancelled NVIDIA MCP7B and add initial support for
MCP89.

From Brad based on information from Peer Chen @ NVIDIA via Linux.

make various strings ("can't map mem space" and similar) more consistent
between instances, saving space in the kernel. feedback from many (some
incorporated, some left for future work).

ok deraadt, kettenis, "why not" miod.

Eliminate the redundant bits of code for MTU and multicast handling
from the individual drivers now that ether_ioctl() handles this.

Shrinks the i386 kernels by..
RAMDISK - 2176 bytes
RAMDISKB - 1504 bytes
RAMDISKC - 736 bytes

Tested by naddy@/okan@/sthen@/brad@/todd@/jmc@ and lots of users.
Build tested on almost all archs by todd@/brad@

ok naddy@

Introduce bpf_mtap_ether(), which for the benefit of bpf listeners
creates the VLAN encapsulation from the tag stored in the mbuf
header. Idea from FreeBSD, input from claudio@ and canacar@.

Switch all hardware VLAN enabled drivers to the new function.

ok claudio@

Remove #if NVLAN.

Re-add support for RX VLAN tag stripping.

Switch the existing TX VLAN hardware support over to having the
tag in the header. Convert TX tagging in the drivers.

Help and ok brad@

First step towards cleaning up the Ethernet driver ioctl handling.
Move calling ether_ioctl() from the top of the ioctl function, which
at the moment does absolutely nothing, to the default switch case.
Thus allowing drivers to define their own ioctl handlers and then
falling back on ether_ioctl(). The only functional change this results
in at the moment is having all Ethernet drivers returning the proper
errno of ENOTTY instead of EINVAL/ENXIO when encountering unknown
ioctl's.

Shrinks the i386 kernels by..
RAMDISK - 1024 bytes
RAMDISKB - 1120 bytes
RAMDISKC - 832 bytes

Tested by martin@/jsing@/todd@/brad@
Build tested on almost all archs by todd@/brad@

ok jsing@

Convert timeout_add() calls using multiples of hz to timeout_add_sec()

Really just the low-hanging fruit of (hopefully) forthcoming timeout
conversions.

ok art@, krw@

MCP79 are also capable of Jumbo frames. Add the Jumbo support flag.

From: Linux forcedeth

Simplify the combination use of pci_mapreg_type()/pci_mapreg_map() as
suggested by dlg@ awhile ago.

ok dlg@

correct format after %x -> %b convertion. deraadt jsg ok.

Add the NFE_CORRECT_MACADDR flag for MCP77/79 chipsets.

ok jsg@ dlg@

add the NVIDIA MCP77/79 ids.

ok jsg@ dlg@