Re: [Xen-devel] linux 4.4 Regression: 100% cpu usage on idle pv guest under Xen with single vcpu

[Eric Shelton <eshelton@xxxxxxxxx>]

Please note that the same issue appears to have been introduced in the recent 4.2.7 kernel. It perhaps has to do withÂb4ff8389ed14b849354b59ce9b360bdefcdbf99c having a matching commitÂe8d097151d309eb71f750bbf34e6a7ef6256da7e in linux-stable.git. The below patch toÂarch/x86/kernel/rtc.c was also effective for 4.2.7.

Eric

On 2015-12-02 18:30,ÂSander Eikelenboom wrote:
>ÂOn 2015-12-02 15:55, David Vrabel wrote:

>Â> On 28/11/15 15:47, Sander Eikelenboom wrote:
>Â>> genirq: Flags mismatch irq 8. 00000000 (hvc_console) vs. 00000000
>Â>> (rtc0)
>Â>
>Â> We shouldn't register an rtc_cmos device because its legacy irq
>Â> conflicts with the irq needed for hvc0. For a multi VCPU guest irq 8
>Â> is
>Â> in use for the pv spinlocks and this gets requested first, preventing
>Â> the rtc device from probing.
>Â>
>Â> Does this patch fix it for you?
>Â>
>Â> David
>Â
>ÂIt does, thanks.
>Â
>ÂReported-and-tested-by: Sander Eikelenboom <linux@eikelenboom.it>
>Â
>Â--
>ÂSander
>Â
>Â> 8<--------------------
>Â> x86: rtc_cmos platform device requires legacy irqs
>Â>
>Â> Adding the rtc platform device when there are no legacy irqs (no
>Â> legacy PIC) causes a conflict with other devices that end up using the
>Â> same irq number.
>Â>
>Â> In a single VCPU PV guest we should have:
>Â>
>Â> /proc/interrupts:
>Â>Â Â Â Â Â Â CPU0
>Â> Â0:   Â4934 xen-percpu-virq   timer0
>Â> Â1:     0 xen-percpu-ipi   Âspinlock0
>Â> Â2:     0 xen-percpu-ipi   Âresched0
>Â> Â3:     0 xen-percpu-ipi   Âcallfunc0
>Â> Â4:     0 xen-percpu-virq   debug0
>Â> Â5:     0 xen-percpu-ipi   Âcallfuncsingle0
>Â> Â6:     0 xen-percpu-ipi   Âirqwork0
>Â> Â7:    321 Âxen-dyn-event  Âxenbus
>Â> Â8:    Â90 Âxen-dyn-event  Âhvc_console
>Â>Â Â...
>Â>
>Â> But hvc_console cannot get its interrupt because it is already in use
>Â> by rtc0 and the console does not work.
>Â>
>Â>Â Âgenirq: Flags mismatch irq 8. 00000000 (hvc_console) vs. 00000000
>Â> (rtc0)
>Â>
>Â> The rtc_cmos device requires a particular legacy irq so don't add it
>Â> if there are no legacy irqs.
>Â>
>Â> Signed-off-by: David Vrabel <david.vrabel@xxxxxxxxxx>
>Â> ---
>Â>Â arch/x86/kernel/rtc.c | 5 +++++
>Â>Â 1 file changed, 5 insertions(+)
>Â>
>Â> diff --git a/arch/x86/kernel/rtc.c b/arch/x86/kernel/rtc.c
>Â> index cd96852..07c70f1 100644
>Â> --- a/arch/x86/kernel/rtc.c
>Â> +++ b/arch/x86/kernel/rtc.c
>Â> @@ -14,6 +14,7 @@
>Â>Â #include <asm/time.h>
>Â>Â #include <asm/intel-mid.h>
>Â>Â #include <asm/rtc.h>
>Â> +#include <asm/i8259.h>
>Â>
>Â>Â #ifdef CONFIG_X86_32
>Â>Â /*
>Â> @@ -200,6 +201,10 @@ static __init int add_rtc_cmos(void)
>Â>Â Â Â Â}
>Â>Â #endif
>Â>
>Â> +Â Â Â/* RTC uses legacy IRQs. */
>Â> +Â Â Âif (!nr_legacy_irqs())
>Â> +Â Â Â Â Â Â Âreturn -ENODEV;
>Â> +
>Â>Â Â Â Âplatform_device_register(&rtc_device);
>Â>Â Â Â Âdev_info(&rtc_device.dev,
>Â>Â Â Â Â Â Â Â Â "registered platform RTC device (no PNP device found)\n");

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel