Too many devices using IRQ10?
Peter Arremann
loony at loonybin.org
Mon Nov 14 07:01:19 UTC 2005
On Monday 14 November 2005 01:47, Steffen Kluge wrote:
> Hi all,
> I've got this niggling problem that I revisit every time a new kernel is
> released for FC4, however, it doesn't appear to change:
>
> On my notebook it seems that by default most devices are sharing IRQ10,
> including the ones I would normally associate with performance (disk,
> net):
>
> $ cat /proc/interrupts
> CPU0
> 0: 34756 XT-PIC timer
> 1: 91 XT-PIC i8042
> 2: 0 XT-PIC cascade
> 8: 1 XT-PIC rtc
> 9: 1138 XT-PIC acpi
> 10: 10853 XT-PIC libata, ipw2200, Intel ICH6,
> ehci_hcd:usb1, uhci_hcd:usb2, uhci_hcd:usb3, uhci_hcd:usb4, uhci_hcd:usb5,
> yenta, ohci1394, eth0 12: 814 XT-PIC i8042
> 14: 454 XT-PIC ide0
> NMI: 0
> ERR: 0
>
> The interrupts are assigned by ACPI. Since I need ACPI for cpuspeed I
> don't want to disable it altogether, but instead fiddled with boot
> parameters to keep it from doing the IRQ assignment. For example, my
> current boot command line contains "pci=noacpi,routeirq", which seems to
> improve the situation:
>
> $ cat /proc/interrupts
> CPU0
> 0: 54122 XT-PIC timer
> 1: 73 XT-PIC i8042
> 2: 0 XT-PIC cascade
> 5: 399 XT-PIC uhci_hcd:usb3, ohci1394, eth0
> 7: 2 XT-PIC ehci_hcd:usb1, uhci_hcd:usb2
> 8: 1 XT-PIC rtc
> 9: 1727 XT-PIC acpi
> 10: 41 XT-PIC ipw2200, Intel ICH6, uhci_hcd:usb4, yenta
> 11: 10615 XT-PIC libata, uhci_hcd:usb5
> 12: 118 XT-PIC i8042
> 14: 610 XT-PIC ide0
> NMI: 0
> ERR: 0
>
> With this, the interrupts are much more evenly used.
>
> Now, I guess my question is, does this even matter? Is sharing a single
> IRQ among many devices a potential performance issue? Am I losing
> anything by stopping ACPI from messing with IRQ's?
>
> Cheers
> Steffen.
The issue you're seeing is actually more common than you think - Especially on
laptops. Most ACPI implementations still suck - many people say that before
you could learn writing good DTSDs ACPI will be outdated ;)
Seriously though - I ran into a very similar scenario on some HP desktops
fairly recently and running a gigabit lan, I saw about 4% performance drop
when completely saturation the network with ftp accesses. That's the result
of disks and ethernet competing. For each interrupt the system needs to query
each device if it caused the interrupt.
For your laptop (slower disk) its probably even less. My current laptop looks
like this:
11: 5409254 XT-PIC Intel 82801DB-ICH4, Intel 82801DB-ICH4 Modem,
ehci_hcd, uhci_hcd, uhci_hcd, uhci_hcd, yenta, ath0, nvidia
and I've never seen any issues. If you really care about it, try update your
bios or search the web for a better DSDT. Other than that, I'd not worry.
NoACPI will btw, seriously decrease your battery lifetime and on most laptops
I've seen cause fan trigger/speed issues.
Peter.
More information about the fedora-list
mailing list