down, the dc driver and receiver can fall out of sync with one another,
resulting in a condition where the chip continues to receive packets
but the driver never notices. Normally, the receive handler checks each
descriptor starting from the current producer index to see if the chip
has relinquished ownership, indicating that a packet has been received.
The driver hands the packet off to ether_input() and then prepares the
descriptor to receive another frame before moving on to the next
descriptor in the ring. But sometimes, the chip appears to skip a
descriptor. This leaves the driver testing the status word in a descriptor
that never gets updated. The driver still gets "RX done" interrupts but
never advances further into the RX ring, until the ring fills up and the
chip interrupts again to signal an error condition. Sometimes, the
driver will remain in this desynchronized state, resulting in spotty
performance until the interface is reset.
Fortunately, it's fairly simple to detect this condition: if we call
the rxeof routine but the number of received packets doesn't increase,
we suspect that there could be a problem. In this case, we call a new
routine called dc_rx_resync(), which scans ahead in the RX ring to see
if there's a frame waiting for us somewhere beyond that the driver thinks
is the current producer index. If it finds one, it bumps up the index
and calls the rxeof handler again to snarf up the packet and bring the
driver back in sync with the chip. (It may actually do this several times
in the event that there's more than one "hole" in the ring.)
So far the only card supported by if_dc which has exhibited this problem
is a LinkSys LNE100TX v2.0 (82c115 PNIC II), and it only seems to happen
on one particular system, however the fix is general enough and has low
enough overhead that we may as well apply it for all supported chipsets.
I also implemented the same fix for the 3Com xl driver, which is apparently
vulnerable to the same problem.
Problem originally noted and patch tested by: Matt Dillon
the 3c450-TX HomeConnect. Like the 3cSOHO100-TX OfficeConnect, this NIC
uses the same ASIC as the 3c905B/3c905C but is targeted for a particular
market segment (home users). It is somewhat less expensive than the
3c905B/3c905C ($49, according to the 3Com web site), comes with its
own custom driver kit and is bundled with various goofy Windows software
packages designed to demonstrate the niftyness of home networking (networked
game demos, etc...).
Changes are:
- Add PCI ID to list in if_xlreg.h.
- Update xl_devs table in if_xl.c.
- Update xl_choose_xcvr() to consider the HomeConnect the
same as all the other 10baseT/100baseTX cards.
- When setting/clearing promisc mode, just update the filter, don't
reset the whole interface.
- Call xl_init() in xl_ifmedia_upd() when setting miibus media modes. This
fixes a problem with the 3c905B-COMBO where switching from 10base5/AUI
or 10base2/BNC to a 10/100 mode doesn't always work right.
- Attempt to reset the interface in xl_init() so that we know we're getting
the receive and transmit rings reset properly.
there are stubs compiled into the kernel if BPF support is not enabled,
there aren't any problems with unresolved symbols. The modules in /modules
are compiled with BPF support enabled anyway, so the most this will do is
bloat GENERIC a little.
declaration for the interface driver from "foo" to "if_foo" but leave the
declaration for the miibus attached to the interface driver alone. This
lets the internal module name be "if_foo" while still allowing the miibus
instances to attach to "foo."
This should allow ifconfig to autoload driver modules again without
breaking the miibus attach.
submitter, who *still* hasn't bothered to answer me back.
The thing which the submitter completely failed to mention is that
his 3c900B-TPO card has the transceiver selection in the EEPROM set
to "auto." You can tweak the setting using the 3C90XCFG.EXE utility
that 3Com provides with the card. I'm not sure if it's supposed to
default to auto or if the user fiddled with it. Currently, the xl
driver only does autoselection for 10/100 NICs (i.e. those with NWAY
autonegotiation capabilities). For the 10baseT, 10base5, 10base2,
10baseFL and 100baseFX cards, the driver sets the default media to
whatever the EEPROM transceiver selector says. The problem is that
the "auto" selection is mistakenly identified as "10/100 NWAY
autoselection mode" and this is not handled correctly: the default
media ends up being chosen as 100baseTX, which doesn't work because
we've only added 10baseT media types to the ifmedia word. This leads
to a panic in ifmedia_set() (something else which the submitter never
bothered to mention).
A workaround for this is to re-run the 3C90XCFG.EXE utility and change
the transceiver selection to something besides "auto." I have also
patched the driver to watch for the "auto" setting in the non-miibus
case and select a reasonable default based on the card type instead of
falling through to 100baseTX and exploding.
PR: misc/13665
This whole idea isn't going to work until somebody makes the bus/kld
code smarter. The idea here is to change the module's internal name
from "foo" to "if_foo" so that ifconfig can tell a network driver from
a non-network one. However doing this doesn't work correctly no matter
how you slice it. For everything to work, you have to change the name
in both the driver_t struct and the DRIVER_MODULE() declaration. The
problems are:
- If you change the name in both places, then the kernel thinks that
the device's name is now "if_foo", so you get things like:
if_foo0: <FOO ethernet> irq foo at device foo on pcifoo
if_foo0: Ethernet address: foo:foo:foo:foo:foo:foo
This is bogus. Now the device name doesn't agree with the logical
interface name. There's no reason for this, and it violates the
principle of least astonishment.
- If you leave the name in the driver_t struct as "foo" and only
change the names in the DRIVER_MODULE() declaration to "if_foo" then
attaching drivers to child devices doesn't work because the names don't
agree. This breaks miibus: drivers that need to have miibuses and PHY
drivers attached never get them.
In other words: damned if you do, damned if you don't.
This needs to be thought through some more. Since the drivers that
use miibus are broken, I have to change these all back in order to
make them work again. Yes this will stop ifconfig from being able
to demand load driver modules. On the whole, I'd rather have that
than having the drivers not work at all.
strategy used in the 3Com Linux driver. The new strategy is to use transmit
descriptor polling -- that is, the NIC polls the descriptors to see when
new packets are available for transmission. The advantage to the new scheme
is that no register accesses are needed in the transmit routine. The old
scheme requires several register accesses to stall the TX engine, update the
TX DMA list pointer register, then unstall the TX engine. Hopefully the new
scheme will provide improved transmit performance with less CPU overhead.
This only affects the 3c90xB or 3c90xC cards, not the 3c90x cards. This
means the original 3c900 and 3c905 cards are unaffected. Newer cards include
the 3c900B series, the 3c905B, 3c980, 3c980B, 3c905C and 3c905C, and the
3cSOHO100-TX OfficeConnect.
It's GPL'ed of course, but looking over it tonight I learned of Yet Another
Fast EtherLink XL Adapter: the 3c980C server adapter. This is basically
an updated version of the 3c980 that uses the Tornado ASIC instead of the
earlier Hurricane ASIC. The only change here is to add the new PCI device
ID (0x9805) and corresponding table entries.
external NatSemi PHY chip was programmed to respond to MII address 24.
In the 3c905B ASICs, the transceiver is internal but it's still mapped
to MII address 24. But *some* 3Com 3c905B ASIC revisions map the
transceiver control registers to *all* MII addresses (0 through 31).
The miibus code probes for PHYs at all MII addresses and because of
this unusual behavior, it will attempt to map the same PHY registers
several times over, which doesn't work.
Naturally, the 3c905B NIC that I tested happened not to exhibit this
behavior.
The fix is to tweak xl_miibus_readreg() and xl_miibus_writereg()
to only respond when attempting to read from MII address 24. This
is safe to do since the 3Com documentation indicates that the PHY
and/or internal transceiver will always be mapped to address 24,
and there are no 3Com XL NICs with more than one PHY.
due to the fact that there are non-MII cards supported by the same
driver and I don't have all of the cards available for testing. There's
also the 3c905B-COMBO which has MII, AUI and BNC media ports all in one
package. Supporting the COMBO is difficult because we have to add the
10base5 and 10base2 media types to the same ifmedia struct as the
MII-attached types, however there is no way to force the miibus and
child PHYs into existence before xl_attach() completes, so there is
no ifmedia struct available in xl_attach(). What we do inistead is
use the mediainit method as a callback: when a child PHY is attached,
it calls the miibus mediainit routine which selects a default media.
This routing also calls the NIC driver's mediainit method (if it
implements one) at which point we can safely add the other media
types.
reset command.
I observed some anomalous behavior while testing a 3c905C with a
Dell PowerEdge 4300/500 dual PIII 500Mhz system. The NIC would seem
to work correctly most of the time but would sometimes fail to receive
certain packets, in particular NFS create requests. I could mount
an NFS filesystem from the PowerEdge and do an ls on it, but trying
to do a "touch foo" would hang. Monitoring traffic from another host
revealed that the client was properly sending an NFS create request
but the server was not receiving it. It *did* receive it when I
ran the same test with an Intel fxp card.
I don't understand the exact mechanics of this strange behavior, but
resetting the receiver and transmitter seems to get rid of it. I used
to perform an RX and TX reset in xl_init(), but stopped doing it there
because on 3c905B and later cards this causes the autoneg session to
restart, which would lead to the NIC waiting a long time before exchanging
traffic after being brought up the first time. Apparently the receiver
and transmitter resets should be performed at least once when initializing
the card.
Hopefully this will cure problems that people have been having with the
3c905C -- this was the only strange behavior that I have observed with
the 3c905C so far which does not appear with the 3c905B or 3c905.
been booted works too -- very neat. However I don't want the system to
stop for 5 seconds when the MII autoprobe is triggered in the xl and
tl drivers since that's lame. Instead, only use the hard delay when
we've been cold booted. If not, use the timeout mechanism instead.
(The SysKonnect driver doesn't use the same autonegotiation scheme, so
no change is required there.)
into a loadable module, and all of the platform dependencies are gone
(except for the alpha_XXX_dmamap() thing, which is another issue -- I
still don't know how to use the busdma stuff with a network driver).
Also increase the delay in xl_reset(); testing on a 486/66 with a 3c905C
shows that reading the EEPROM fails immediately after a reset. Waiting
a little longer after the reset completes seems to fix it.
clears out the transmit queue and zeroes the downlist pointer register,
but xl_txeof() isn't called before xl_start() tries to queue more packets,
xl_start() will think that the DMA is still in progress and not update
the downlist register again, thus causing packets to sit in the transmit
queue forever.
Patch provided by: Russell T Hunt <alaric@MIT.EDU>
compiles cleanly on the Alpha. (On the alpha, the port type is an int,
not a short).
Cast a couple of pointers to ints via 'uintptr_t' rather than 'unsigned
int' since uintptr_t is long (64 bit) on Alpha, as are pointers.
happens if you have a BIOS with a 'Plug & Play OS' setting and you leave
it set to 'Yes.' This is wrong for FreeBSD (and LoseNT): it should be set
to 'No.' Apparently it's still possible to map the iobase of the NIC and
have the card work by reading the config space manually (which is what
the driver does if pci_map_port() fails) but we need to warn the user to
do fix their machine anyway. Anyway, warn the user to check the 'Plug &
Play OS' setting in their BIOS if mapping the io space fails.
chipset. First you thrilled to the 3c905, then you trembled at the
3c905B, now gaze in wonder at: the 3c905C! This appears to be another
3c90X series chip called the Tornado (PCI ID 0x10B7/0x9200) and should
be equivalent (from the driver API perspective) to the 3c905B, so all
we have to do is add the PCI ID to the list.
#define COMPAT_PCI_DRIVER(name,data) DATA_SET(pcidevice_set,data)
.. to 2.2.x and 3.x if people think it's worth it. Driver writers can do
this if it's not defined. (The reason for this is that I'm trying to
progressively eliminate use of linker_sets where it hurts modularity and
runtime load capability, and these DATA_SET's keep getting in the way.)
uses the AUI port with an on-board AUI to 10baseFL transceiver, not the
10baseT port like I had earlier suspected. The 3c900B-FL should be properly
supported now.
- Try to unbreak what I broke by screwing with the tx queuing again.
I'm waiting for a few more people to test out this code and report back
before I move it into current. Hopefully it will be soon. Basically I
reverted to the old TX queuing strategy.
- Add experimental support for the 3c900B-FL (10mbps ST fiber). The card
should be detected properly and the 10baseFL mode supported, but again
I'm still waiting for word from a tester to see if this actually works.
It shouldn't affect the other cards though; all the differences are in
media selection.
- Set the TX start threshold register to get better performance.
- Increase the size of the RX and TX rings. UDP performance was pretty
bad because the TX ring was too small. Should be substantially better
now (I can saturate the link with either TCP or UDP now).
- Change some of the #defines to reflect proper 3Com ASIC names (boomerang,
cyclone, krakatoa, hurricane).
- Simplify and reorganize interrupt handler; ack all interrupts right
away and then process them. This avoids a potential race condition.
(Noted by Matt Dillon.)
- Reorganize the bridging code to eliminate using a goto to jump into
the middle of an if() {} clause. Sorry, that just made my brain itch.
- Use m_adj() in xl_rxeof().
- Make the payload alignment in xl_newbuf() the default (instead of
just conditionally defined for the alpha) to improve NFS performance
(avoids need for nfs_realign()).
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
3c900B-TPC (twisted pair and coax). Treated similarly to the
3c900B-COMBO, except no AUI port.
- Fix media selection so that it's possible to select the AUI and BNC
ports on the 3c905B-COMBO. This board is now fully supported.
- Change TX queueing strategy to hopefully be more efficient by avoiding
register accesses in xl_start(). Should provide small performance
improvement and a little better reliability.
(cut-down version of the "cyclone" for the small office/home office
"cheap bastard" market). Basically the same as a 3c905B but without
Wake-on-LAN, ROM socket, etc...
- Wait longer for the reset to complete in xl_attach() to try and avoid
'command never completed' warnings.
- Clean up a few odds and ends in xl_attach().
- Add PCI ID for the 3c905B-COMBO (a new card). Right now this is
treated as a 3c905B; I need to dig up one of these cards for testing
before I can make the AUI and BNC ports work.
- Add a hack to force reading the I/O address directly from the PCI
registers if pci_map_port() fails. I SHOULD NOT HAVE TO DO THIS:
SOMEBODY WITH MORE PCI CLUES THAN I SHOULD INVESTIGATE WHY THIS
HAPPENS.
sys/alpha/conf/GENERIC.
Note: the PNIC ignores the lower few bits of the RX buffer DMA address,
which means we have to add yet another kludge to make it happy. Since
we can't offset the packet data, we copy the first few bytes of the
received data into a separate mbuf with proper alignment. This puts
the IP header where it needs to be to prevent unaligned accesses.
Also modified the PNIC driver to use a non-interrupt driven TX
strategy. This improves performance somewhat on x86/SMP systems where
interrupt delivery doesn't seem to be as fast with an SMP kernel as
with a UP kernel.
3c905B, the RX and TX reset commands also reset the cyclone chip's internal
PHY, which causes it to restart its autonegotiation session. This takes a
second or two to complete, which makes the interface seem to stop responding
for a few seconds every time you do something that reinitializes it.
Issuing the RX and TX resets on the older 3c905 boomerang adapters doesn't
cause any delay because the boomerang chip requires an external PHY.
This should fix the problem where people doing network installs via 3c905B
cards experience a delay after the interface is first initialized, among
other things.
