transmitter is wedged. If so, try to unwedge it, process any descriptors
that might need to be free()d, then proceed.
- Disable the 'background' autonegotiation performed during bootstrap.
What happens currently is that the driver starts an autoneg session,
the sets a timeout in the ifnet structure and returns. Later, when the
timer expires, the watchdog routine calls the autoneg handler to check
the results of the session. The problem with this is that the session
may not complete until some point after we have started to mount NFS
filesystems, which can cause the mounts to fail. This is especially
troublesome if booting with an NFS rootfs: we need the interface up
and running before reaching the mountroot() code.
The default behavior now is to do the autoneg synchronously, i.e. wait
5 seconds for the autoneg to complete before exiting the driver attach
routine. People who want the old behavior can compile the driver with
XL_BACKGROUND_AUTONEG #defined. This has no effect on autoneg sessions
initiated by 'ifconfig xl0 media autoselect.'
This slows the probe down a little, but it's either that or botching
NFS mounts at bootup.
- If xl_setmode_mii() is called and there's an autoneg session in progress,
cancel it, _then_ set the modes.
and use this when masking/unmasking interrupts.
Maintain a mapping from (iopaic number, int pin) tuple to irq number,
and use this when configuring devices and programming the ioapics.
Previous code assumed that irq number was equal to int pin number, and
that the ioapic number was 0.
Don't let an AP enter _cpu_switch before all local apics are initialized.
generating a trap 12 panic. The code blindly assumed that in the event
of a transmit error, the packet that caused the error would still be
at the head of the driver's transmit queue (sc->xl_cdata.xl_tx_head).
However in the case of error 82 (which indicates that a transmit error
occurred after part of the transmit FIFO memory has been reclaimed)
this is not true: the TX queue has already been flushed, and the
pointer to the head of the queue is NULL, so trying to dereference
the pointer to find the transmit descriptor address causes a crash.
The code now checks for a NULL pointer before trying to reload the
chip's download pointer register. There may still be error messages
printed warning of the transmit error, but no panic should occur.
Note that this eror code is only generated with "cyclone" chipsets
(3c900B, 3c905B, and presumeably the 3c980 server adapter). It should
only appear during periods of heavy traffic, probably only on
non-switched networks.
Problem reported by: Darcy Buskermolen <darcy@ok-connect.com>
XCVR value read from the EEPROM is completely wrong. I've had one report
of a 3c900 card that returns an xcvr value of 14, which is impossible
(the manual states that all vales above 8 are reserved). If the value
is out of the expe
Add PCI vendor ID for the 3c980-TX server adapter card, which apparently
also uses the cyclone chip. Graciously supplied Mats O Jansson
<maja@cntw.com>.
Also noted by Mats, the 10mpbs cyclone adapters should be named 3c900B,
not 3c905B. I haven't actually encountered a 10mbps only cyclone adapter
yet, nor anybody who has one, but this makes sense given the naming
scheme used for the older boomerang adapters.
Removed Hauppauge EEPROM 0x10 detection as I think 0x10 should be a
PAL tuner, not NTSC.
Reinstated some Tuner Guesswork code from 1.27
Submitted by: Roger Hardiman <roger@cs.strath.ac.uk>
Added PR kern/7177 for SECAM Video Highway Xtreme with single crystal
PLL configuration submitted by Vsevolod Lobko <seva@alex-ua.com>.
In kernel configuration file add
options OVERRIDE_CARD=2
options OVERRIDE_TUNER=11
options BKTR_USE_PLL
Submitted by: Roger Hardiman <roger@cs.strath.ac.uk>
Normally the full 640x480 (768x576 PAL) image is grabbed. This ioctl
allows a smaller area from anywhere within the video image to be
grabbed, eg a 400x300 image from (50,10).
See restrictions in BT848SCAPAREA.
Submitted by: Roger Hardiman <roger@cs.strath.ac.uk>
cluster from the RX descriptor is passed up to the higher layers and
replaced with an empty buffer for the next time the descriptor comes
up in the RX ring. The xl_newbuf() routine returns ENOBUFS if it can't
obtain a new mbuf cluster, but this return value was being ignored.
Now, if buffer allocation fails, we leave the old one in place and
drop the packet. This is rude, but there's not much else that can be
done in this situation.
Without this, the driver can cause a panic if the system runs out of
MBUF clusters. Now it will complain loudly, but it shouldn't cause a
panic.
Also added another pair of missing newlines to some printf()s.
delay controls how long the driver waits for autonegotiation to
complete after setting the 'autoneg restart bit' in a PHY. In some
cases, it seems 3 seconds is not long enough: with 3c905-TX cards
(external PHY), you sometimes see 'autoneg not complete; no carrier'
errors due to the timeout being too short. (3c905B adapters seem to
be happy with 3 seconds though.)
of associated mbuf clusters) in the RX ring from 4 to 16. On my
really fast PI 400Mhz test machines, 4 descriptors (and associated
mbuf clusters) is enough to achieve decent performance without any
RX overruns. However, one person reported problems with the following
scenario:
- P90 system running FreeBSD with a 3c905B-TX adapter, slow IDE hard
disk (Quantum Bigfoot?)
- PII 266 with SCSI disks running LoseNT and also with a 3c905B-TX
- Both machines connected together via crossover cable at 100Mbps
full-duplex
- LoseNT machine writing largs amounts of data (2.5 GB work of
files each in the neighborhood of 1 to 2 MB in size) via samba to
the FreeBSD machine
In this case, the LoseNT machine is sending data very fast. Apparently
there weren't any problems initially because the user was writing to
one particular disk which was relatively fast, however after this disk
filled up and the user started writing to the second slower disk, RX
overruns would occur and sometimes the RX DMA engine would stall after
a 100 to 500MB had been transfered. The xl_rxeof() handler is supposed
to detect this condition and restart the upload engine; I'm not sure
why it doesn't, unless interrupts are being lost and the rx handler
isn't getting called.
This is still an improvement over the Linux driver, which uses 32
descriptors in its receive ring. :)
Problem reported by: Heiko Schaefer <hschaefer@fto.de>
If I'm reading the manual correctly, the 3c905B actually loses its
PCI configuration during the transition from D3(hot) back to D0, not
during the transition from D0 to D3(hot). This means it should be possible
to save the existing PCI settings, restet the power state, then restore
the PCI settings afterwards. Changed xl_attach() to attempt this first
thing before the normal PCI setup. I'm not certain this will work correctly,
but it shouldn't hurt.
If xl_init() is called while an autoneg session is in progress, the
autoneg timeout and chip state will get clobbered. Try to avoid this
by checking sc->xl_autoneg at the start of xl_init() and defer
the initialization until later if it's set. (xl_init() is always called
at the end of an autoneg session by xl_autoneg_mii().)
Problem pointed out by: Larry Baird <lab@gta.com>
stability now. ALso modify /sys/conf/files, /sys/i386/conf/GENERIC
and /sys/i386/conf/LINT to add entries for the XL driver. Deactivate
support for the XL adapters in the vortex driver. LAstly, add a man
page.
(Also added an MLINKS entry for the ThunderLAN man page which I forgot
previously.)
in ddb) which I broke by changing %8[l]x to %8p. Hacked the central
printf routine to not add an "0x" prefix for %p formats if the field
width is nonzero. The tables are still horribly misformatted on
64-bit machines.
Use %p instead of %8p to print pointers when the field width isn't
important.
interrupts which now defers them until the transmit queue if filled
up with completed buffers. This has two advantages: first, it reduces
the number of transmitter interrupts to just 1/120th of the rate
that they occured previously, and two, running down many buffers
at once has much improved cache effects.
- probe for PHYs by checking the BMSR (phy status) register instead
of the vendor ID register.
- fix the autonegotiation routine so that it figures out the autonegotiated
modes correctly.
- add tweaks to support the Olicom OC-2326 now that I've actually had
a chance to test one
o Olicom appears to encode the ethernet address in the EEPROM
in 16-bit chunks in network byte order. If we detect an
Olicom card (based on the PCI vendor ID), byte-swap the station
address accordingly.
XXX The Linux driver does not do this. I find this odd since
the README from the Linux driver indicates that patches to
support the Olicom cards came from somebody at Olicom; you'd
think if anyone would get that right, it'd be them. Regardless,
I accepted the word of the disgnoatic program that came bundled
with the card as gospel and fixed the attach routine to make
the station address match what it says.
o The version of the 2326 card that I got for testing is a
strange beast: the card does not look like the picture on
the box in which it was packed. For one thing, the picture
shows what looks like an external NS 83840A PHY, but the
actual card doesn't have one. The card has a TNETE100APCM
chip, which appears to have not only the usual internal
tlan 10Mbps PHY at MII address 32, but also a 10/100 PHY
at MII address 0. Curiously, this PHY's vendor and device ID
registers always return 0x0000. I suspect that this is
a mutant version of the ThunderLAN chip with 100Mbps support.
This combination behaves a little strangely and required the
following changes:
- The internal PHY has to be enabled in tl_softreset().
- The internal PHY doesn't seem to come to life after
detecting the 100Mbps PHY unless it's reset twice.
- If you want to use 100Mbps modes, you have to isolate
the internal PHY.
- If you want to use 10Mbps modes, you have to un-isolate
the internal PHY.
The latter two changes are handled at the end of tl_init(): if
the PHY vendor ID is 0x0000 (which should not be possible if we
have a real external PHY), then tl_init() forces the internal
PHY's BMCR register to the proper values.
Change the port address argument to pci_map_port to pci_port_t* which is
defined as u_int on the alpha, u_short on i386. This is a stopgap with a
hopefully limited lifetime.
Discussed with: Stefan Esser <se@freebsd.org>
respectively. Most of the longs should probably have been
u_longs, but this changes is just to prevent warnings about
casts between pointers and integers of different sizes, not
to fix poorly chosen types.
cure the problems I was having with interrupts not being acknowledged
on time. This fixes a problem I observed where starting two ping -f
processes at 10Mbps would cause an adapter check due to TX GO commands
being issued before TXEOC interrupts were being acked.
Also fix a small problem with tl_start(): the mechanism I was using
to queue new packets onto the TX chain was bogus.
Change adapter check handler so that it resets card state after
tl_softreset() is stored.
Moved all EEPROM-related macro definitions into if_tlreg.h.
Don't allow an autoneg session to start until after the TX queue has
been drained, and don't transmit anything until after the autoneg
session is complete.
Also add support for two more Compaq ThunderLAN-based cards, and three
cards from Olicom which also use the ThunderLAN chip. The only thing
different about the Olicom cards is that they store the station address
at a different location within the EEPROM.
hidden). Now "ticks" are used, which are 4 byte, not 8 byte in size.
The size mismatch did not matter due to sufficient padding at the end
of the structure that holds time stamps (there is an unused member).
The fix suggested by Bruce Evans used "sizeof (ticks_t)", but I prefer
to use "sizeof ticks", and didn't seem to object in his last mail on
this topic.
Submitted by: bde
`#if defined(ONE_THING)' is a style bug, and i386 instead of __i386__
is a bug, since i386 is never defined when the kernel is compiled
by with the default flags (`gcc -ansi ...'). Here the bug disabled
the call to pmap_setvidram(), so ISA video memory was not mapped
WC on 686's. The bug may have been masked by bugs in the committer's
version of gcc - `gcc -ansi' incorrectly defines i386 for gcc = the
version of egcs on the 2.2.6 cdrom.
