Hugh Brown
|
Re: pci network card interrupt re-assignment
|
Hugh Brown
03/07/2012 9:27 AM
post91943
|
Re: pci network card interrupt re-assignment
There isn't a way to re-assign interrupts on an x86 system as these are
setup by the BIOS. If you use 6.5.0 with startup-apic and pci-bios-v2, you
shouldn't get any shared interrupts if the device supports MSI/X.
--
Hugh Brown
QNX Software Systems Limited
1001 Farrar Rd.,
Ottawa. ON. K2K 0B3.
Telephone: 613-591-0931
On 12-03-06 8:37 PM, "janusz ruszel" <community-noreply@qnx.com> wrote:
>Gurus,
>
>How to assign a pci network card interrupt on a x86 BIOS based system
>during a boot process? Are there any parameters which can be passed to
>pci-bios? Any idea and comments greatly appreciate.
>
>The issue i am having are shared interrupts. There is an interrupt not
>assigned to any device in the system and i would like to be that
>interrupt used by network card exclusively.
>
>Best
>Janusz
>
>
>
>
>_______________________________________________
>
>Networking Drivers
>http://community.qnx.com/sf/go/post91928
>
|
|
|
Will Miles
|
Re: pci network card interrupt re-assignment
|
Will Miles
03/07/2012 9:39 AM
post91945
|
Re: pci network card interrupt re-assignment
Hi,
We ran in to a similar issue with some of our high precision sensor interfaces which required an extremely low upper
bound on the interrupt latency. I don't know of any officially supported mechanism, but we found a solution based on
using PCI BIOS calls to retrieve the mapping of PCI interrupt lines, then directly interacting with the bus interface in
the chipset to reassign them to different IRQs. The bad news is that the interface is different for every chipset -
the good news is that most of the documentation is freely available, and you can often find examples of interacting with
the chipset registers in open-source operating systems.
If it's OK with the QNX folks I can post the source of our remapping tool - it's actually based on the pci-bios sources
from the 6.3.0 x86 BSP (because it needs to issue PCI bios calls to query the mapping and instruct the cards that
they've been reassigned - it just seemed easier to use the existing infrastructure for making those calls). If they
would prefer that I not post the whole source file, I can post a patch instead. Fair warning, it's a kluge at best -
it's not exactly well organized.
-Will
|
|
|