Received: from eggs.gnu.org ([209.51.188.92]:39225) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1hAx9p-0001MB-BD for qemu-devel@nongnu.org; Mon, 01 Apr 2019 09:39:26 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1hAx9o-0005O8-5n for qemu-devel@nongnu.org; Mon, 01 Apr 2019 09:39:25 -0400 Received: from mx1.redhat.com ([209.132.183.28]:57972) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1hAx9n-0005NU-QQ for qemu-devel@nongnu.org; Mon, 01 Apr 2019 09:39:24 -0400 References: <20190401133659.20421-1-vkuznets@redhat.com> From: Paolo Bonzini Message-ID: Date: Mon, 1 Apr 2019 15:39:15 +0200 MIME-Version: 1.0 In-Reply-To: <20190401133659.20421-1-vkuznets@redhat.com> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH] ioapic: allow buggy guests mishandling level-triggered interrupts to make progress List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Vitaly Kuznetsov , qemu-devel@nongnu.org Cc: "Michael S. Tsirkin" , Marcel Apfelbaum , Liran Alon On 01/04/19 15:36, Vitaly Kuznetsov wrote: > It was found that Hyper-V 2016 on KVM in some configurations (q35 machine + > piix4-usb-uhci) hangs on boot. Trace analysis led us to the conclusion that > it is mishandling level-triggered interrupt performing EOI without fixing > the root cause. This causes immediate re-assertion and L2 VM (which is > supposedly expected to fix the cause of the interrupt) is not making any > progress. > > Gory details: https://www.spinics.net/lists/kvm/msg184484.html > > Turns out we were dealing with similar issues before; in-kernel IOAPIC > implementation has commit 184564efae4d ("kvm: ioapic: conditionally delay > irq delivery duringeoi broadcast") which describes a very similar issue. > > Steal the idea from the above mentioned commit for IOAPIC implementation in > QEMU. SUCCESSIVE_IRQ_MAX_COUNT, delay and the comment are borrowed as well. > > Signed-off-by: Vitaly Kuznetsov > --- > hw/intc/ioapic.c | 43 ++++++++++++++++++++++++++++++- > hw/intc/trace-events | 1 + > include/hw/i386/ioapic_internal.h | 3 +++ > 3 files changed, 46 insertions(+), 1 deletion(-) > > diff --git a/hw/intc/ioapic.c b/hw/intc/ioapic.c > index 9d75f84d3b..daf45cc8a8 100644 > --- a/hw/intc/ioapic.c > +++ b/hw/intc/ioapic.c > @@ -139,6 +139,15 @@ static void ioapic_service(IOAPICCommonState *s) > } > } > > +#define SUCCESSIVE_IRQ_MAX_COUNT 10000 > + > +static void ioapic_timer(void *opaque) > +{ > + IOAPICCommonState *s = opaque; > + > + ioapic_service(s); > +} > + > static void ioapic_set_irq(void *opaque, int vector, int level) > { > IOAPICCommonState *s = opaque; > @@ -227,7 +236,28 @@ void ioapic_eoi_broadcast(int vector) > trace_ioapic_clear_remote_irr(n, vector); > s->ioredtbl[n] = entry & ~IOAPIC_LVT_REMOTE_IRR; > if (!(entry & IOAPIC_LVT_MASKED) && (s->irr & (1 << n))) { > - ioapic_service(s); > + bool level = ((entry >> IOAPIC_LVT_TRIGGER_MODE_SHIFT) & 1) > + == IOAPIC_TRIGGER_LEVEL; > + > + ++s->irq_reassert[vector]; > + if (!level || > + s->irq_reassert[vector] < SUCCESSIVE_IRQ_MAX_COUNT) { > + ioapic_service(s); > + } else { > + /* > + * Real hardware does not deliver the interrupt > + * immediately during eoi broadcast, and this lets a > + * buggy guest make slow progress even if it does not > + * correctly handle a level-triggered interrupt. Emulate > + * this behavior if we detect an interrupt storm. > + */ > + trace_ioapic_eoi_delayed_reassert(vector); > + timer_mod(s->timer, > + qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + > + NANOSECONDS_PER_SECOND / 100); Should this be done only if the timer isn't pending? Paolo