Known problems and limitations for the 3.1.22 PCMCIA release
============================================================
Bug summaries:
o CardBus devices with 2.0.* kernels are not recommended
o PCI interrupt routing issues for CardBus bridges on some systems
o PCI configuration problems with 2.3.* kernels
o O2Micro and ToPIC CardBus bridge configuration problems
o On the Sony VAIO PCG-N505VE, interrupts break after a suspend
o SCSI drivers are not hot swap safe
o Iomega Clik! drives require a kernel patch
o Token ring memory allocation issues
o Interrupt lossage with Megahertz multifunction cards
o The aic7xxx/apa1480_cb driver misbehaves if the cable is detached
o Ositech Jack of Diamonds firmware issue
o Serial interrupt sharing bug in certain 2.2.*, 2.3.* kernels
o aha152x interrupt bug in certain 2.2.* kernels
o IDE driver shutdown bug in certain 2.2.*, 2.3.* kernels
o Tedious IDE probes for nonexistent slave devices
o IDE driver does not share PCI interrupts properly
o SuSE 6.4 IDE driver problem
o parport_cs doesn't work in 2.3.*
o Xircom CardBus driver instability
o NE2000 compatible cards based on the Asix AX88190 are unsupported
Bug details:
o Use of CardBus cards with 2.0.* kernels is discouraged. It may work
on some systems, but not on others, due to PCI BIOS limitations.
Also, it is harder to diagnose problems, because /proc/bus/pccard is
not available with these kernels.
o With some PCI host bridges, the PCMCIA subsystem is not able to
determine the PCI interrupt routing for CardBus bridges. For some
types of CardBus bridges, this means that we can't configure
interrupts for CardBus cards at all. When the PCMCIA drivers are
loaded, they may complain about an "unknown interrupt router".
Prognosis: see the discussion in the PCMCIA-HOWTO.
o Some 2.3.* kernels may make mistakes when autoconfiguring CardBus
bridges. Specifically, the kernel may memory map bridges at invalid
addresses. The i82365 driver reports a "Bad bridge mapping". In
other cases, the PCI subsystem may misconfigure PCI interrupts for
CardBus bridges.
Prognosis: either upgrade to a newer 2.3.* kernel if possible, or
revert to a 2.2.* kernel.
o Interrupt routing on O2Micro CardBus bridges seems to have problems.
Toshiba ToPIC97 bridges also seem to have problems, particularly
with Cardbus cards.
Prognosis: I think the O2Micro problems should now be fixed. For
the ToPIC problem, Toshiba does not seem willing and/or able to
provide adequate help, so I've mostly given up on it. For both the
O2Micro and ToPIC problems, fixes would require someone with device
driver experience and the relevant hardware to work on it: data
sheets are available, and I can make suggestions of things to try,
but I can't debug the problems by email.
With ToPIC chipsets, some systems seem to work better if the bridge
mode is changed to either "PCIC" or "CardBus", rather than "Auto",
in the BIOS setup menu.
In some cases, ToPIC chipsets generate bogus eject/insert sequences
when a card is first powered up. It may be useful to increase the
vcc_settle and/or setup_time parameters for the pcmcia_core module
to prevent this.
o On the Sony VAIO PCG-N505VE, after a suspend, no interrupts are
delivered by the CardBus bridge until the system is rebooted.
Prognosis: I've spent a lot of time trying to track this down, but
I'm completely stumped. The PCMCIA drivers appear to restore the
state of the CardBus bridge correctly, and the PCI interrupt router
is also configured properly. But no interrupts get through.
o All of the SCSI drivers, and most of the CardBus drivers, do not
implement suspend/resume handling. The only workaround now is to
eject these cards (or do "cardctl eject") before suspending.
Prognosis: CardBus Network cards will probably be fixed eventually,
but it has not been a high priority. SCSI drivers are less likely
to be fixed since we're more dependent on kernel code.
o The Iomega Clik! drive is incompatible with the kernel ide-floppy
driver. A kernel patch for 2.2.14 and later 2.2 kernels is
available at http://paulbristow.net/linux/clik.html.
o The token ring driver tweaks a problem in the memory management
code. To work around the problem, remove all high memory windows
from /etc/pcmcia/config.opts. The driver is also completely broken
for late 2.1 and early 2.2 kernels. A fix is in 2.2.7.
o Megahertz EM1144, EM3288, and EM3336 cards drop interrupts if the
modem and ethernet are used simultaneously.
Prognosis: Unlikely to be fixed, since these cards are old and we
are unlikely to ever get more complete tech info.
o The kernel aic7xxx driver, which is linked into the apa1480_cb
driver, can generate spurious interrupts when a card is initialized,
which can cause system lockups. This will usually happen if the
card is inserted with no SCSI cable attached.
Prognosis: I've sent a patch to the driver maintainer. The problem
can be mostly mitigated by disabling use of PCI interrupts for
CardBus cards, by setting PCIC_OPTS="pci_int=0". This setting does
not work on some newer laptops; in those cases, you'll have to wait
for a kernel update to fix the problem.
o Some Ositech Jack of Diamonds 33.6K modem/ethernet cards don't work
because of a firmware issue. With these cards, the smc91c92_cs
driver reports "Bad chip signature".
A DOS program to update the card firmware to v8.1B is available from
Ositech's web site at ftp://www.ositech.com/pub/jod/JDCEL422.EXE
o The 2.2.*/2.3.* serial driver had a bug that interfered with
interrupt sharing for multifunction cards. The effect is that
opening a serial port on a multifunction card fails, giving an IO
error. It was fixed in 2.2.11 and 2.3.9.
The bug can be fixed by editing linux/drivers/char/serial.c and
changing each use of IRQ_T(info) to IRQ_T(state).
o The kernel aha152x driver, used for Adaptec 16-bit SCSI adapters,
had a PCMCIA compatibility problem in 2.2.* that was fixed in 2.2.9.
The effect was that interrupts were ignored, unless the card
happened to be configured for irq 9..12.
Either upgrade to a 2.2.9 or later kernel, or if you have an
appropriate interrupt available, add to /etc/pcmcia/config.opts:
module "aha152x_cs" opts "irq_list=9,10,11,12"
o The kernel IDE driver had a bug that causes shutdown of some PCMCIA
IDE cards to cause a kernel trap. It was introduced in 2.2.9/2.3.1
and fixed in 2.2.10/2.3.4.
o For some ATA/IDE devices, the IDE driver will lock up the system for
up to 15 seconds while probing for (non-existent) slave devices.
I've told the IDE maintainer about the issue and it is just a matter
of getting the kernel driver updated. There are two aspects to the
fix; one is to improve automatic detection of flash memory cards,
and the other is to change the probe to sleep instead of freezing
the system during the probe. The 2.4 driver is fixed.
o The linux IDE driver generates spurious interrupts when it probes
for new devices. This is ok at boot time because the IDE probe runs
before almost all other drivers. But it causes lockups if the probe
is done when another driver is using the same PCI interrupt. This
happens when the PCMCIA subsystem is configured to use only PCI
interrupts for card status changes as well as card interrupts.
Prognosis: the IDE device probe needs to be rewritten; I don't know
when that might happen. In some situations, you can work around the
issue by using startup options like:
PCIC_OPTS="irq_mode=0 pci_csc=0"
which will prevent the i82365 driver from sharing the PCI interrupt
for monitoring card insert/eject events; this will not help if other
PCI devices also need to share.
o The SuSE 6.4 version of the 2.2.14 kernel has a broken IDE probe
that messes up at least some PCMCIA devices. The result is that
the ide_cs driver reports "ide_register(...) failed".
Prognosis: substitute drivers/block/ide-probe.c from SuSE's 2.2.13
kernel or from a virgin 2.2.14 source tree.
o The 2.3.6 kernel update broke the PCMCIA parport_cs driver.
Prognosis: it should be fixable in newer 2.3.* kernels, but I just
have not gotten around to doing it.
o Xircom CBEM support in the tulip_cb driver seems to be extremely
unreliable. A wide range of symptoms are reported, ranging from no
packet reception, to correct operation at certain speed/duplex
combinations but not others, to frequent missed interrupts. Also,
some people have reported kernel faults when a Xircom card is shut
down with "ifconfig down".
Prognosis: I used to think that the Xircom inconsistencies had to do
with multiple chipset revisions. After more study, it seems that
the receive filter setup on the Xircom chipset is quirky; different
network setups kick the filter differently, so many reported
differences in card behavior actually reflect different network
setups.
Temporarily setting the card to promiscuous mode ("ifconfig eth0
promisc") or all-multicast mode ("ifconfig eth0 allmulti") seems
to help with the packet reception issue in some cases. The current
driver is broken for more than 6 multicast addresses, unless
"allmulti" is used.
o A number of "NE2000 compatible" cards use the Asix AX88190 chipset,
which has several serious bugs and incompatibilities that render the
regular pcnet_cs driver unusable. The pcnet_cs driver will identify
and refuse to deal with these cards.
Prognosis: a Linux driver is available from the Asix web site at
http://www.asix.com.tw/driver2.htm and I'm investigating how to best
deal with the issue in the regular pcnet_cs driver. It is a tricky
issue and I'm not sure when a better fix will be available.