Will talk about 2.6.25 kernel.
I had recently a problem with marvell pata controllers. Exactly speaking
of "IDE interface: Marvell Technology Group Ltd. 88SE6145 SATA II PCI-E
controller", pci-id 11ab:6145.
There is a (pseudo) driver in the kernel for handling that controller,
called pata_marvell, but it doesn't claim the device.
Instead, ahci driver claims that device, and then just does nothing.
Devices connected to the marvell are not seen by kernel.
After looking into the source code I found out that both ahci.c and
pata_marvell.c list that device in their pci_device_id structure (which
wouldn't be problem, if both drivers worked with the controller, but
ahci.c just doesn't)
The first solution to this was found very quickly by using google: just
add kernel parameter all-generic-ide, and the marvell controller will be
recognized by ide-generic driver, and HDD's, CDROM's on that marvell
controller are also found (as hda, hdb).
The second solution is to remove few lines from ahci.c which lists the
marvell 6145 device (and probably also marvell 6121).
Short patch included below.
After that, kernel finally uses the pata_marvell driver, and my pata
devices are registered with kernel (as sda, sdb).
The third solution is to not use ahci driver when using pata_marvell.
But this is not possible with lots of motherboards which have Intel
corp. SATA controller + marvell sata / pata controller.
AFAIK, the second idea (patch the source) would solve problems for lots
of people asking for the solution. (the first idea with the
all-generic-ide doesn't work for everybody)
I would suggest patching the ahci.c, and I appreciate any
For the sake of completness, here are relevant parts of dmesg:
(0000:03:00.0 is the mentioned marvell 6145 sata/pata controller)
scenario 1: ahci takes the controller, no devices are detected
ACPI: PCI Interrupt 0000:03:00.0[A] -> GSI 16 (level, low) -> IRQ 16
Yes. I know or to teach the AHCI driver how to drive the PATA part of the
device. Unfortunately I've been trying - privately and now publically to
get Jeff Garzik to fix that one for *months*.
I also don't have the docs for it and requests to Jeff for the relevant
And you can have my ack for it, but don't expect anything to happen
(and yes I *am* cross about this)
I don't think this issue is on the regression list. Maybe getting it
there would get some results.
BTW would it be possible for the ahci driver to determine if the PATA
ports are used and fail the probe in case they are?