Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.os.linux.hardware > #1294

Re: Assigned interrupts ... what's correct ?

Show key headers only | View raw

Armin wrote:
> I have tried to find out what IRQ has been assigned to the devices of
> my SuSE/Intel machine ... and I got different infos.

There are multiple ways of routing and handling interrupts, with various
degrees of backwards compatibility.

In the good^Wbad ol' times of the ISA bus, the interrupt controller (the
PIC) was able to handle 16 interrupts, numbered 0 to 15.  Modern
interrupt controllers are still compatible with that, and the BIOS must
initialize interrupts to use that because it doesn't know if the OS is
able to use anything else.

All PCI devices have an interrupt number register, which does not affect
the device itself but is just used by the BIOS or the OS to tell the
driver to which interrupt the chip's interrupt pin is routed.  This
register is what is shown by lspci -H1:

> lspci -vvv -H1
> 02:00.0 Ethernet controller: Atheros Communications L2 Fast Ethernet (rev a0)
>         Interrupt: pin A routed to IRQ 10 <=======

This is the ISA interrupt that was set by the BIOS.

Most ISA interrupts are reserved for certain ISA devices, so you often
end up with all PCI devices sharing only one or two interrupts.
Therefore, a better interrupt controller was introduced, the IOAPIC.
Typically, it has eight additional interrupts lines (16-23) for PCI
devices.

When a driver attaches to a PCI device, the kernel automatically
configures the interrupt to be routed through the IOAPIC, if possible.
This is what you see in the kernel log:

> dmesg | grep IRQ
> atl2 0000:02:00.0: PCI INT A -> GSI 17 (level, low) -> IRQ 17  <========

There is still the problem that with eight interrupt lines, modern PCs
still have many shared PCI interrupts.  So PCI Express has a new
interrupt delivery mechanism (MSI) that does not require interrupt lines
at all and allows arbitrarily many interrupts.

MSIs behave not exactly the same as old-style PCI interrupts, so the
kernel cannot enable them automatically but does so only when a driver
requests it.  You can see this happen with "dmesg | grep -i IRQ", or
by using lspci without -H which shows the kernel-internal data structure:

> lspci -vvv
> 02:00.0 Ethernet controller: Atheros Communications L2 Fast Ethernet (rev a0)
>         Interrupt: pin A routed to IRQ 43  <=======

This is also shown in /proc/interrupts.


To answer your question "what's correct?" -- all of them are, at
different times.  :-)


Regards,
Clemens

Back to comp.os.linux.hardware | Previous | Next — Previous in thread | Find similar | Unroll thread

Thread

Assigned interrupts ... what's correct ? Armin <a-steinhoff@web.de> - 2012-06-30 12:42 +0200
  Re: Assigned interrupts ... what's correct ? Armin <a-steinhoff@web.de> - 2012-06-30 13:12 +0200
  Re: Assigned interrupts ... what's correct ? Clemens Ladisch <clemens@ladisch.de> - 2012-06-30 13:37 +0200

csiph-web