Re: Interrupt storm

From: Dan Cojocar (dan.cojocar_at_gmail.com)
Date: 06/30/05

  • Next message: Philip Paeps: "Re: bin/82762 [patch] adding a virtual horizontal scroll feature to moused"
    Date: Thu, 30 Jun 2005 09:42:05 +0300
    To: freebsd-current@freebsd.org
    
    

    On 6/29/05, John Baldwin <jhb@freebsd.org> wrote:
    > On Tuesday 05 April 2005 03:46 pm, Dan Cojocar wrote:
    > > On Apr 5, 2005 9:23 PM, John Baldwin <jhb@freebsd.org> wrote:
    > > > On Tuesday 05 April 2005 12:44 pm, Dan Cojocar wrote:
    > > > > On Apr 5, 2005 5:30 PM, John Baldwin <jhb@freebsd.org> wrote:
    > > > > > On Tuesday 29 March 2005 01:35 pm, Jung-uk Kim wrote:
    > > > > > > On Tuesday 29 March 2005 01:22 pm, Dan Cojocar wrote:
    > > > > > > > Hello all,
    > > > > > > >
    > > > > > > > I noticed the following interrupt storm in my dmesg:
    > > > > > > >
    > > > > > > > Interrupt storm detected on "irq11: rl1"; throttling interrupt
    > > > > > > > source Interrupt storm detected on "irq5: rl0"; throttling
    > > > > > > > interrupt source
    > > > > > > >
    > > > > > > > Here is my vmstat -i:
    > > > > > > > interrupt total rate
    > > > > > > > irq0: clk 128600 199
    > > > > > > > irq1: atkbd0 2499 3
    > > > > > > > irq4: sio0 2 0
    > > > > > > > irq5: rl0 128901 200
    > > > > > > > irq8: rtc 82294 127
    > > > > > > > irq11: rl1 128902 200
    > > > > > > > irq12: psm0 16152 25
    > > > > > > > irq14: ata0 12600 19
    > > > > > > > irq15: ata1 71 0
    > > > > > > > Total 500021 776
    > > > > > > >
    > > > > > > > and here is my dmesg:
    > > > > > > > http://cs.ubbcluj.ro/~dan/dmesg.txt
    > > > > > > >
    > > > > > > > I have options DEVICE_POLLING and options HZ=200, i have tested
    > > > > > > > with greater HZ values like 1000, and without DEVICE_POLLING but
    > > > > > > > i have the same storm.
    > > > > > >
    > > > > > > Try this:
    > > > > > >
    > > > > > > http://docs.freebsd.org/cgi/mid.cgi?200503161154.04555.jkim
    > > > > >
    > > > > > He's not using an APIC so that won't make any difference.
    > > > >
    > > > > If I use APIC, I will get "watchdog timeout" as i posted in an old
    > > >
    > > > email:
    > > > > >rl0: watchdog timeout
    > > > > >rl1: watchdog timeout
    > > > > >rl0: watchdog timeout
    > > > > >rl1: watchdog timeout
    > > > > >rl1: watchdog timeout
    > > > > >rl1: watchdog timeout
    > > > > >rl1: watchdog timeout
    > > > >
    > > > > Thanks,
    > > > > Dan
    > > >
    > > > That's indicative of other interrupt routing problems. This is with APIC
    > > > + ACPI, yes? If you disable ACPI, does APIC work or is it not found?
    > > > Also, if you disable ACPI, does non-APIC work ok?
    > >
    > > Yes i get "watchdog timeout" if i use ACPI+APIC.
    > > Here is my dmesg for current+acpi+apic.
    > > http://cs.ubbcluj.ro/~dan/current+acpi+apic.txt
    > > If i disable acpi, the system is not able to boot to the login
    > > prompt. I did a break
    > > into debugger after waiting for 15 minutes and i noticed that the
    > > interrupts for rl0 are increasing very quickly, so i think that is why
    > > the system is so slow.
    > > Here is a dmesg for current without ACPI and with APIC, but only in
    > > single mode
    > > without if_rl module and many other modules.
    > > http://cs.ubbcluj.ro/~dan/current+noacpi+apic+singlemode.txt
    > > And if I disable ACPI i don't see the interrupts storms but i don't
    > > have ACPI :(
    > > http://cs.ubbcluj.ro/~dan/current+noacpi+noapic.txt
    > > And here is current with ACPI and without APIC:
    > > http://cs.ubbcluj.ro/~dan/current+acpi+noapic.txt
    > > Thanks,
    > > Dan
    >
    > I haven't forgotten about this, I've just been busy. So, I'm still trying to
    > understand. Correct me if I'm wrong: APIC always gets interrupt storms; it
    > doesn't matter if ACPI is enabled or disabled. Without APIC, ACPI gives you
    > interrupt storms, but no ACPI works fine. Is that correct? Do you get
    > interrupt storms if ACPI is enabled but APIC is not on both current and
    > RELENG_5?

        Hello John,

        With apic and acpi i was getting watchdog timeout on rl0 and rl1.
    If i disabled apic, and use only acpi i was getting interrupt storms,
    the same i get if i use only apic, and the system was ok without acpi
    and apic, but in this version i didn't have acpi features :(. The
    links with my dmesg from previous message are valid.
      I was getting this behaviour only if i used a second network card,
    tested only with realteck.
      Unfortunately now i don't have the second nic :( in my system, but
    if you need more info i will get a new one.
      Now with recent -current and only one realteck nic and with acpi
    enabled and apic disabled the system is not having interrupt storms
    and no watchdog timeouts, but i noticed that my downloads rate is
    dropping to half of my bandwidth, no mather if i use device_polling or
    not. I was thinking that this is related to my realteck nic but now
    just tested my modem on usb with cdce driver and i get the same
    behaviour.
      Last time I tested this only on current, I don't know if in recent
    RELENG_5 is the same.
    Thank you John for your time,
         Dan
    _______________________________________________
    freebsd-current@freebsd.org mailing list
    http://lists.freebsd.org/mailman/listinfo/freebsd-current
    To unsubscribe, send any mail to "freebsd-current-unsubscribe@freebsd.org"


  • Next message: Philip Paeps: "Re: bin/82762 [patch] adding a virtual horizontal scroll feature to moused"

    Relevant Pages

    • Re: Interrupt storm
      ... >> Have you tried booting this machine using GENERIC with ACPI turned off? ... > timeout, no interrupt storms: ... > And finally i tried my kernel without apic and without acpi, ... It looks like your motherboard's DSDT is doing something hokey. ...
      (freebsd-current)
    • Re: Interrupt storm
      ... >> That's indicative of other interrupt routing problems. ... > If i disable acpi, the system is not able to boot to the login ... APIC always gets interrupt storms; ...
      (freebsd-current)
    • Re: jdk14 build times?
      ... >> arrea? ... "interrupt storms" on some old machines due to ACPI. ... that this was due to these "interrupt storms" and that booting with ACPI ...
      (comp.unix.bsd.freebsd.misc)
    • Re: [Bugme-new] [Bug 11543] New: kernel panic: softlockup in tick_periodic() ???
      ... and attach the output of the 'dmesg' command. ... ACPI: ... Initializing cgroup subsys ns ... registered new interface driver usbfs ...
      (Linux-Kernel)
    • Re: latest -git: hibernate: possible circular locking dependency detected
      ... The only suggestion I can make is that SysRq can be triggered remotely ... with serial console, because I know that the serial line ... boot-log comes with such a setting (ACPI was enabled; ... in the dmesg log in that case and they are not present in your dmesg. ...
      (Linux-Kernel)