Still IRQ routing problems with bridged devices.

Bruce Evans bde at zeta.org.au
Tue Jan 6 00:26:31 PST 2004


On Tue, 6 Jan 2004, Bernd Walter wrote:

> On Tue, Jan 06, 2004 at 12:39:42AM +0100, Bernd Walter wrote:
> > On Mon, Jan 05, 2004 at 04:33:45PM -0700, M. Warner Losh wrote:
> > > In message: <20040105233138.GR17023 at cicely12.cicely.de>
> > >             Bernd Walter <ticso at cicely12.cicely.de> writes:
> > > : On Mon, Jan 05, 2004 at 04:24:27PM -0700, M. Warner Losh wrote:
> > > : > In message: <20040105231533.GQ17023 at cicely12.cicely.de>
> > > : >             Bernd Walter <ticso at cicely12.cicely.de> writes:
> > > : > : The point is that it shouldn't take an IRQ for PCI which is configured
> > > : > : for an ISA device in device.hints.
> > > : >
> > > : > We don't do that.
> > > :
> > > : We do!
> > > :
> > > : /boot/device.hints:
> > > : hint.sio.0.irq="4"
> > > :
> > > : pci_cfgintr_virgin: using routable interrupt 4
> > > : pci_cfgintr: 0:4 INTD routed to irq 4
> > >
> > > Ah, I see what you are saying.  That would be hard to implement.

Presumably pci is configured first, so the isa hint is ignored.

> After rethinking - it wouldn't be a good idea either, because
> GENERIC.hints would block to many possible IRQs :(
> Too bad that ISA components have to be attached after PCI.

Hints (including configurations read from the BIOS) should be mostly
just hints, and accidental ordering of the probes should have no effect
on which hint is preferred.  Accidental ordering also causes longstanding
problems for configuring fast interrupts -- it is not know at
bus_setup_intr() time whether an interrupt can be fast, since subsequent
bus_setup_intr()'s on the same irq may require it to be non fast.
Drivers could handle this by delaying bus_setup_intr() until the device
is active, but most don't (exception for a non-fast intr: lpt repeats
bus_setup_intr() on every write()!), but they shouldn't have to.  Even
this is inadequate if devices are attached after boot time (other
devices that use the same interrupt may be active, and the interrupt
may need significant reorganization).

Bruce


More information about the freebsd-current mailing list