| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
|
|
|
|
| |
the recent changes to the pcibios read/write functions and ioc3-eth.c
changes.
|
| |
|
|
|
|
|
|
| |
of 6 unrecognized IOC3 interfaces on posix0. The leftover two are
missing due to IOC3 PCI interface bugs which I'll fight next. After
a decent nap ...
|
| |
|
| |
|
|
|
|
|
| |
hardcode the 0xFF00 value into subsystem_vendor, instead of relying on
PCI commands to ioc3 to do this.
|
|
|
|
| |
addresses for ports. Serial driver changes are being sent to Linus.
|
| |
|
| |
|
|
|
|
|
| |
address for all IOC3 devices. We will now be able to
reach IOC3 cards on other nodes than the first one.
|
|
|
|
|
| |
that all others do. Further cleanup to make ioc3-eth.c start looking
like a driver.
|
|
|
|
|
| |
Instead of encoding bus/slot numbers in the IRQ, have seperate arrays
to store that information.
|
|
|
|
|
|
|
| |
pci irq, we need to bump up NR_IRQs to handle more than 4 PCI busses.
io.h: Port numbers are really "unsigned long", since they need to
contain the nasids too.
ip27-pci.c: Encode the nasid into the port number for isp1020 for pio.
|
|
|
|
|
|
|
| |
all PCI irqs above that. Fix pci_map_irq to not just have 2 bits
for slot number, else irq numbers will not be unique. Include the
bus number as part of the irq for now, so that scsi controllers
on various PCI busses get unique irq numbers.
|
|
|
|
|
|
|
| |
nasid and widget id get incoded in pci_dev->irq. pcibr_setup() now does
some rudimentary probing for bridge widgets hanging off its xbow.
We're not seeing interrupts from remote devices at the cpu,
so mscsi card's are turned off for now.
|
|
|
|
|
|
| |
the device's config space registers and turn on byte swizzling in the
bridge for this device (needs to be done before the driver sees it).
Again, hardwired to slot 5. Will generalize later.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
| |
platforms. At this time I've only verified that IP22 support compiles
and IP27 actually works.
|
|
|
|
|
|
|
|
|
|
| |
style access instead of memory-mapped style access. Older qlogic drivers
only support io-mapped style access, the 2.3.40 driver supports both; it
is just easier to get things working with io-mapped access.
2. To achieve above, get CF0_WRITE_PCI_CFG to actually program the
registers ... properly.
3. Hack CF0_WRITE_PCI_CFG so that it still only does real work in the
above case ... else things hangs. To be investigated.
|
|
|
|
|
| |
ioport values to prevent gratuitous check_region failures from qlogic
driver.
|
|
Unlike the IRIX functions they only have one instruction overhead per
access, don't involve subroutine calls or any spinlocks, so are
implicitly threadsafe.
Add PCI infrastructure for SN0 + the necessary fixups for IOC3
brokeness.
|