not allow multicast filter programming when controller is busy to
send/receive frames. So it used to mark need_mcsetup bit and defer
multicast filter programming until controller becomes idle state.
To detect when the controller is idle fxp(4) relied on Tx
completion interrupt with NOP command and fxp_start_body and
fxp_intr_body had to see whether pending multicast filter
programming was requested. This resulted in very complex logic and
sometimes it did not work as expected.
Since the controller should be in idle state before any multicast
filter modifications I changed it to reinitialize the controller
whenever multicast filter programming is required. This is the same
way what OpenBSD and NetBSD does. Also I added IFF_DRV_RUNNING
check in ioctl handler so controller would be reinitialized only if
it is absolutely needed.
With this change I guess we can remove fxp(4) DELAY hack in ifioctl
for IPv6 case.
common mbuf dma tag for both Tx and Rx path but Rx buffer should
have single DMA segment and maximum buffer size of the segment
should be less than MCLBYTES.
fxp(4) also have to check Tx completion status which was updated by
DMA so we need BUS_DMASYNC_PREREAD and BUS_DMASYNC_POSTWRITE
synchronization in Tx path. Fix all misuse of bus_dmamap_sync(9) in
fxp(4). I guess this change shall fix occasional driver breakage in
PAE environments.
While I'm here add error messages of dma tag/buffer creation and
correct messages.
controllers. ICH based controllers are treated as 82559. 82557,
earlier revision of 82558 and 82559ER have no WOL capability.
o WOL support requires help of a firmware so add check whether
hardware is capable of handling magic frames by reading EEPROM.
o Enable accepting WOL frames only when hardware is about to
suspend or shutdown. Previously fxp(4) used to allow receipt of
magic frame under normal operation mode which could cause
hardware hang if magic frame is received by hardware. Datasheet
clearly states driver should not allow WOL frames under normal
operation mode.
o Disable WOL frame reception in device attach so have fxp(4)
immunize against system hang which can be triggered by magic
packets when the hardware is not in fully initialized state.
o Don't reset all hardware configuration data in fxp_stop()
otherwise important configuration data is lost and this would
reset WOL configuration to default state which in turn cause
hardware hang on receipt of magic frames. To fix the issue,
preserve hardware configuration data by issuing a selective
reset.
o Explicitly disable interrupts after issuing selective reset as
reset may unmask interrupts.
Tested by: Alexey Shuvaev < shuvaev <> physik DOT uni-wuerzburg DOT de >
o Configure controller to use dynamic TBD as TSO requires that
operation mode.
o Add a dummy TBD to tx_cb_u as TSO can access one more TBD in TSO
operation.
o Increase a DMA segment size to 4096 to hold a full IP segment
with link layer header.
o Unlike other TSO capable controllers, 82550/82551 does not
modify the first IP packet in TSO operation so driver should
create an IP packet with proper header. Subsequent IP packets
are generated from the header information in the first IP packet
header. Likewise pseudo checksum also should be computed by
driver for the first packet.
o TSO requires one more TBD to hold total TCP payload. To make
code simple for TSO/non-TSO case, increase the index of the
first available TBD array.
o Remove KASSERT that checks the size of a DMA segment should be
less than or equal to MCLBYTES as it's no longer valid in TSO.
o Tx threshold and number of TBDs field is used to store MSS in
TSO. So don't set the Tx threshold in TSO case.
82559 or later controllers added simple checksum calculation logic
in RU. For backward compatibility the computed checksum is appended
at the end of the data posted to Rx buffer. This type of simple
checksum calculation support had been used on several vendors such
as Sun HME/GEM, SysKonnect GENESIS and Marvell Yukon controllers.
Because this type of checksum offload support requires parsing of
received frame and pseudo checksum calculation with software
routine it still consumes more CPU cycles than that of full-fledged
checksum offload controller. But it's still better than software
checksum calculation.
IFF_DRV_OACTIVE to note resource shortage to upper stack.
- Don't count number of mbuf chains. Default 32 DMA segments for a
frame is enough for most cases. If bus_dmamap_mbuf_sg fails use
m_collapse(9) to collapse the mbuf chain instead of relying on
expensive m_defrag(9).
- Move bpf handling to fxp_start_body() which is supposed to be
more appropriate place.
- Always arm watchdog timer whenever a new Tx request is made.
Previously fxp(4) used to arm watchdog timer only when
FXP_CXINT_THRESH-th Tx request is made. Because fxp(4) does not
rely on Tx interrupt to reclaim transmitted mbufs it's better to
arm watchdog timer to detect potential lockups.
- Add more aggresive Tx buffer reclaiming in fxp_start_body to make
room for new Tx requests. Since fxp(4) does not request Tx
completion interrupt for every frames it's necessary to clean
TXCBs in advance to saturate link.
- Make fxp(4) try to start more packets transmitting regardless of
interrupt type in fxp_intr_body.
the timeout routine.
- Fix locking in detach.
- Add locking in shutdown.
- Don't mess with the PCI command register in resume, the PCI bus driver
already does this for us.
- Add locking to the non-serial ifmedia routines.
- Fix locking in ioctl.
- Remove spls and support for 4.x.
MFC after: 1 week
struct ifnet or the layer 2 common structure it was embedded in have
been replaced with a struct ifnet pointer to be filled by a call to the
new function, if_alloc(). The layer 2 common structure is also allocated
via if_alloc() based on the interface type. It is hung off the new
struct ifnet member, if_l2com.
This change removes the size of these structures from the kernel ABI and
will allow us to better manage them as interfaces come and go.
Other changes of note:
- Struct arpcom is no longer referenced in normal interface code.
Instead the Ethernet address is accessed via the IFP2ENADDR() macro.
To enforce this ac_enaddr has been renamed to _ac_enaddr.
- The second argument to ether_ifattach is now always the mac address
from driver private storage rather than sometimes being ac_enaddr.
Reviewed by: sobomax, sam
from an mbuf into the fxp_encap() function, as done in other drivers.
- Don't waste time calling bus_dmamap_load_mbuf() if we know the mbuf
chain is too long to fit in a TX descriptor, call m_defrag() first.
- Convert fxp(4) to use bus_dmamap_load_mbuf_sg().
sysctls were global (hw.fxp_rnr and hw.fxp_noflow), all of them are
now per-device. Sample output of "sysctl dev.fxp0" with this patch,
with the standard %foo nodes removed :
dev.fxp0.int_delay: 1000
dev.fxp0.bundle_max: 6
dev.fxp0.rnr: 0
dev.fxp0.noflow: 0
Check for suspend before the device polling, rather than after it.
Check to see if the current thread owns the lock in ioctl and return
EBUSY if it does.
This advances the locking to the point that I can eject my fxp card 10
times in a row, but I agree with Jeff Hsu that we need to get the
network layer locking finished before chasing more of the races here
(actually, he doesn't think this set is worth it even). There's a
number of races between FXP_LOCK in detach and all other users of
FXP_LOCK, and this gets back to the 'device with sleepers being
forcibly detached' problem as well...
1) always call fxp_stop in fxp_detach. Since we don't read from
the card, there's no need to carefully look at things with
bus_child_present.
2) Call FXP_UNLOCK() before calling bus_teardown_intr to avoid
a possible deadlock reported by jhb.
3) add gone to the softc. Set it to true in detach.
4) Return immediately if gone is true in fxp_ioctl
5) Return immediately if gone is true in fxp_intr
- Add fxp_start_body() and change fxp_start() to just acquire locks and
then call fxp_start_body(). Places that would call fxp_start() with
locks held (mutex recursion) now call fxp_start_body() directly.
Remove MTX_RECURSE flag from sc_mtx. [gallatin]
- Change fxp_attach() to work without the softc lock, saving interrupt
hooking until the head of fxp_attach().
- Call ether_ifattach() before overriding ifp parameters. This reverts
part of 1.155.
- Remove multiple error paths in fxp_attach().
- Teardown interrupt in fxp_detach() before unlocking the softc.
- Make sure mutex is not held in fxp_release()
- Delete the miibus instance and/or self in fxp_release(), not in
fxp_detach(). This can happen if attach fails partway through.
- Move ifmedia_removeall to fxp_release() since attach may fail after
media have been allocated.
- Add locking to fxp_suspend, fxp_resume, fxp_start, fxp_intr,
fxp_poll, fxp_tick, fxp_ioctl, fxp_watchdog.
- Pass in ifp to fxp_intr_body since its callers sometimes already use
it.
- Add compatibility define for INTR_MPSAFE for 4.x. [gallatin]
- You don't need to bzero softc.
Ideas from: gallatin, mux
Tested by: >400M packets of dd/ssh, NFS, ping on i386 UP
This patch is rather big because I had to significantly redesign
the driver to make the busdma conversion possible. Most notably,
hardware and software structures were carefully splitted to get
rid of all the structs overlapping evilness.
Special thanks to phk and Richard Puga <puga@mauibuilt.com> for
providing me with fxp(4) hardware to do this work.
Thanks to marcel for testing this on ia64, and to Fred Clift
<fclift@verio.net> for testing this on alpha.
Tested on: i386, ia64, alpha
the fxp driver. This is enabled only for the 82550/82551 chips
(PCI revision code 12 or 13). RX and TX checksum offload are
both supported. Transmit offload is limited to TCP and UDP only
right now: there seems to be a problem with IP header checksumming
on transmit in some cases.
This chip has hardware VLAN support as well. I hope to enable
support for this eventually.
just limited to the DEVICE_POLLING case. This removes the FXP_RFA_RNRMARK
hack, and replaces it with a softc flag that is used to record when
the handling of a no-resource condition was deferred due to running
out of DEVICE_POLLING cycles. This was tested on -stable, but the
code is essentially the same as in -current. It should only affect
the case where DEVICE_POLLING is defined.
The details of the mechanism behind the crashes are still uncertain
but the most likely cause seems to be some kind of hardware confusion
when the no-resource recovery code is accidentally invoked while
the receiver is still active. This could have happened if the
hardware left the 0x4000 bit of the RFA status word set. The comments
in the commit log for revision 1.142 stating that the driver could
clash with the hardware writing to this status word were not correct.
Tested by: Guy Helmer <ghelmer@palisadesys.com>
most cases NULL is passed, but in some cases such as network driver locks
(which use the MTX_NETWORK_LOCK macro) and UMA zone locks, a name is used.
Tested on: i386, alpha, sparc64
Non-SMP, i386-only, no polling in the idle loop at the moment.
To use this code you must compile a kernel with
options DEVICE_POLLING
and at runtime enable polling with
sysctl kern.polling.enable=1
The percentage of CPU reserved to userland can be set with
sysctl kern.polling.user_frac=NN (default is 50)
while the remainder is used by polling device drivers and netisr's.
These are the only two variables that you should need to touch. There
are a few more parameters in kern.polling but the default values
are adequate for all purposes. See the code in kern_poll.c for
more details on them.
Polling in the idle loop will be implemented shortly by introducing
a kernel thread which does the job. Until then, the amount of CPU
dedicated to polling will never exceed (100-user_frac).
The equivalent (actually, better) code for -stable is at
http://info.iet.unipi.it/~luigi/polling/
and also supports polling in the idle loop.
NOTE to Alpha developers:
There is really nothing in this code that is i386-specific.
If you move the 2 lines supporting the new option from
sys/conf/{files,options}.i386 to sys/conf/{files,options} I am
pretty sure that this should work on the Alpha as well, just that
I do not have a suitable test box to try it. If someone feels like
trying it, I would appreciate it.
NOTE to other developers:
sure some things could be done better, and as always I am open to
constructive criticism, which a few of you have already given and
I greatly appreciated.
However, before proposing radical architectural changes, please
take some time to possibly try out this code, or at the very least
read the comments in kern_poll.c, especially re. the reason why I
am using a soft netisr and cannot (I believe) replace it with a
simple timeout.
Quick description of files touched by this commit:
sys/conf/files.i386
new file kern/kern_poll.c
sys/conf/options.i386
new option
sys/i386/i386/trap.c
poll in trap (disabled by default)
sys/kern/kern_clock.c
initialization and hardclock hooks.
sys/kern/kern_intr.c
minor swi_net changes
sys/kern/kern_poll.c
the bulk of the code.
sys/net/if.h
new flag
sys/net/if_var.h
declaration for functions used in device drivers.
sys/net/netisr.h
NETISR_POLL
sys/dev/fxp/if_fxp.c
sys/dev/fxp/if_fxpvar.h
sys/pci/if_dc.c
sys/pci/if_dcreg.h
sys/pci/if_sis.c
sys/pci/if_sisreg.h
device driver modifications
a 82557 (e.g.: a newer chip) then:
+ enable MWI, if the PCI configuration indicates the system supports it
+ enable usage of extended TxCB, for better performance
+ enable hardware flow control. FC frames will be passed up to the
host only if promiscuous mode is enabled.
mtx_enter(lock, type) becomes:
mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks)
mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized)
similarily, for releasing a lock, we now have:
mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN.
We change the caller interface for the two different types of locks
because the semantics are entirely different for each case, and this
makes it explicitly clear and, at the same time, it rids us of the
extra `type' argument.
The enter->lock and exit->unlock change has been made with the idea
that we're "locking data" and not "entering locked code" in mind.
Further, remove all additional "flags" previously passed to the
lock acquire/release routines with the exception of two:
MTX_QUIET and MTX_NOSWITCH
The functionality of these flags is preserved and they can be passed
to the lock/unlock routines by calling the corresponding wrappers:
mtx_{lock, unlock}_flags(lock, flag(s)) and
mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN
locks, respectively.
Re-inline some lock acq/rel code; in the sleep lock case, we only
inline the _obtain_lock()s in order to ensure that the inlined code
fits into a cache line. In the spin lock case, we inline recursion and
actually only perform a function call if we need to spin. This change
has been made with the idea that we generally tend to avoid spin locks
and that also the spin locks that we do have and are heavily used
(i.e. sched_lock) do recurse, and therefore in an effort to reduce
function call overhead for some architectures (such as alpha), we
inline recursion for this case.
Create a new malloc type for the witness code and retire from using
the M_DEV type. The new type is called M_WITNESS and is only declared
if WITNESS is enabled.
Begin cleaning up some machdep/mutex.h code - specifically updated the
"optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN
and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently
need those.
Finally, caught up to the interface changes in all sys code.
Contributors: jake, jhb, jasone (in no particular order)
an override as a loader settable variable (fxp_iomap). fxp_iomap is
a bitmap of fxp units that should be configured to use PCI I/O space
in stead of PCI Memory space.
Reviewed by: Kees Jan Koster <dutchman@tccn.cs.kun.nl>, dg@freebsd.org
is enabling as all entries are still called with Giant being held.
Maintaining compatability with NetBSD makes what should be very simple
kinda ugly.
Reviewed by: Jason Evans
address size that is different than the standard 6bits. This fixes
support for the Compaq NC3121 card, certain newer Intel Pro/100+
cards, and should also fix integrated NICs on SuperMicro and Compaq
motherboards.
The auto-sizing algorithm was taken from NetBSD (thanks!), which I
think got it from Linux originally.
Thanks also to Andrew Sparrow <spadger@best.com> and Joe Moore
<jomor@ahpcns.com> for supplying me with unworking Compaq and Intel
cards to develop and test the fixes with.
and/or when using the card.
o Convert the driver to using bus_space. This allows alphas with
fxp's to boot, rather than panic'ing because rman_get_virtual()
doesn't really return a virtual address on alphas.
o Fix an alpha unaligned access error caused by some misfeature of
gcc/egcs: if link_addr & rbd_addr in the fxp_rfa struct are 32 bit
quantities, egcs will assume they are naturally aligned. So it will do
a ldl & some shifty/masky to twiddle 16 bit values in fxp_lwcopy().
However, if they are 16-bit aligned, the ldl will actually be done on
a 16-bit aligned value & we will panic with an unaligned access
error... Changing their definition to an array of chars seems to fix
this. I obtained this from NetBSD.
I've tested this on both i386 & alpha.
i386 platform boots, it is no longer ISA-centric, and is fully dynamic.
Most old drivers compile and run without modification via 'compatability
shims' to enable a smoother transition. eisa, isapnp and pccard* are
not yet using the new resource manager. Once fully converted, all drivers
will be loadable, including PCI and ISA.
(Some other changes appear to have snuck in, including a port of Soren's
ATA driver to the Alpha. Soren, back this out if you need to.)
This is a checkpoint of work-in-progress, but is quite functional.
The bulk of the work was done over the last few years by Doug Rabson and
Garrett Wollman.
Approved by: core
FreeBSD/alpha. The most significant item is to change the command
argument to ioctl functions from int to u_long. This change brings us
inline with various other BSD versions. Driver writers may like to
use (__FreeBSD_version == 300003) to detect this change.
The prototype FreeBSD/alpha machdep will follow in a couple of days
time.
if it is in 10Mbps mode and gets certain types of garbage prior to
the packet header. The work-around involves reprogramming the
multicast filter if nothing is received in some number of seconds
(currently set at 15). As a side effect, implemented complete support
for multicasting rather than the previous 'receive all multicasts'
hack, since we now have the ability to program the filter table.
Fixed a serious bug which crept in with the timeout() changes;
the cookie was only saved on the first timeout() call in fxp_init()
and wasn't updated in the most common place in fxp_stats_update()
when the timeout was rescheduled. This bug would have resulted in
an eventual panic if fxp_stop() was called (which happens when any
interface flags are changed, for example).
Fixed a bug in Alpha support that would have caused the TxCB
descriptor chain to span a page boundry, causing serious problems
if the pages didn't happen to be contiguous.
Removed some gratuitous bit masking that was left over from an
older implementation.
Fixed a bug where too much was copied from the configuration
template, spilling over into memory that followed it.
Fixed handling of if_timer...it was cleared too early in some cases.
