[pve-devel] [PATCH kernel 3/7] add KVM cherry-pick for Windows live-migration

[Fabian Grünbichler]

the following three cherry-picks from 4.6, first introduced
in Ubuntu-4.4.0-63.84 to fix LP#1649718 break Windows live
migration:

commit 296cb660d22c3732e08464f456d9336392b94c60
    kvm: x86: Check dest_map->vector to match eoi signals for rtc
    (cherry picked from commit 4d99ba898dd0c521ca6cdfdde55c9b58aea3cb3d)

commit f05aa1ca76109353c9421dfe1421d2fa42bd1605
    kvm: x86: Track irq vectors in ioapic->rtc_status.dest_map
    (cherry picked from commit 9daa50076f585854f0040aa8403eac020d6f5d64)

commit aee4b7a8ebc1496f55299241ba96590f80acdc19
    kvm: x86: Convert ioapic->rtc_status.dest_map to a struct
    (cherry picked from commit 9e4aabe2bb3454c83dac8139cf9974503ee044db)

cherry-pick a follow up commit from Linux 4.8, which fixes
the issue:

b0eaf4506f5f95d15d6731d72c0ddf4a2179eefa
    kvm: x86: correctly reset dest_map->vector when restoring LAPIC state

    When userspace sends KVM_SET_LAPIC, KVM schedules a check between
    the vCPU's IRR and ISR and the IOAPIC redirection table, in order
    to re-establish the IOAPIC's dest_map (the list of CPUs servicing
    the real-time clock interrupt with the corresponding vectors).

    However, __rtc_irq_eoi_tracking_restore_one was forgetting to
    set dest_map->vectors.  Because of this, the IOAPIC did not process
    the real-time clock interrupt EOI, ioapic->rtc_status.pending_eoi
    got stuck at a non-zero value, and further RTC interrupts were
    reported to userspace as coalesced.

    Fixes: 9e4aabe2bb3454c83dac8139cf9974503ee044db
    Fixes: 4d99ba898dd0c521ca6cdfdde55c9b58aea3cb3d

Signed-off-by: Fabian Grünbichler <f.gruenbichler at proxmox.com>
---
 Makefile                                           |  1 +
 ...rectly-reset-dest_map-vector-when-restori.patch | 68 ++++++++++++++++++++++
 2 files changed, 69 insertions(+)
 create mode 100644 0001-kvm-x86-correctly-reset-dest_map-vector-when-restori.patch

diff --git a/Makefile b/Makefile
index 7fa9bc2..866db24 100644
--- a/Makefile
+++ b/Makefile
@@ -265,6 +265,7 @@ ${KERNEL_SRC}/README ${KERNEL_CFG_ORG}: ${KERNELSRCTAR}
 	# IPoIB performance regression fix
 	cd ${KERNEL_SRC}; patch -p1 < ../IB-ipoib-move-back-the-IB-LL-address-into-the-hard-header.patch
 	cd ${KERNEL_SRC}; patch -p1 < ../cgroup-cpuset-add-cpuset.remap_cpus.patch
+	cd ${KERNEL_SRC}; patch -p1 < ../0001-kvm-x86-correctly-reset-dest_map-vector-when-restori.patch
 	cd ${KERNEL_SRC}; patch -p1 < ../CVE-2017-2596-kvm-page-reference-leakage-in-handle_vmon.patch
 	cd ${KERNEL_SRC}; patch -p1 < ../CVE-2017-6074-dccp-fix-freeing-skb-too-early-for-IPV6_RECVPKTINFO.patch
 	sed -i ${KERNEL_SRC}/Makefile -e 's/^EXTRAVERSION.*$$/EXTRAVERSION=${EXTRAVERSION}/'
diff --git a/0001-kvm-x86-correctly-reset-dest_map-vector-when-restori.patch b/0001-kvm-x86-correctly-reset-dest_map-vector-when-restori.patch
new file mode 100644
index 0000000..da9f4f8
--- /dev/null
+++ b/0001-kvm-x86-correctly-reset-dest_map-vector-when-restori.patch
@@ -0,0 +1,68 @@
+From b0eaf4506f5f95d15d6731d72c0ddf4a2179eefa Mon Sep 17 00:00:00 2001
+From: Paolo Bonzini <pbonzini at redhat.com>
+Date: Wed, 14 Sep 2016 23:39:12 +0200
+Subject: [PATCH] kvm: x86: correctly reset dest_map->vector when restoring
+ LAPIC state
+MIME-Version: 1.0
+Content-Type: text/plain; charset=UTF-8
+Content-Transfer-Encoding: 8bit
+
+When userspace sends KVM_SET_LAPIC, KVM schedules a check between
+the vCPU's IRR and ISR and the IOAPIC redirection table, in order
+to re-establish the IOAPIC's dest_map (the list of CPUs servicing
+the real-time clock interrupt with the corresponding vectors).
+
+However, __rtc_irq_eoi_tracking_restore_one was forgetting to
+set dest_map->vectors.  Because of this, the IOAPIC did not process
+the real-time clock interrupt EOI, ioapic->rtc_status.pending_eoi
+got stuck at a non-zero value, and further RTC interrupts were
+reported to userspace as coalesced.
+
+Fixes: 9e4aabe2bb3454c83dac8139cf9974503ee044db
+Fixes: 4d99ba898dd0c521ca6cdfdde55c9b58aea3cb3d
+Cc: stable at vger.kernel.org
+Cc: Joerg Roedel <jroedel at suse.de>
+Cc: David Gilbert <dgilbert at redhat.com>
+Reviewed-by: Radim Krčmář <rkrcmar at redhat.com>
+Signed-off-by: Paolo Bonzini <pbonzini at redhat.com>
+Signed-off-by: Fabian Grünbichler <f.gruenbichler at proxmox.com>
+---
+ arch/x86/kvm/ioapic.c | 8 +++++---
+ 1 file changed, 5 insertions(+), 3 deletions(-)
+
+diff --git a/arch/x86/kvm/ioapic.c b/arch/x86/kvm/ioapic.c
+index 5f42d03..c7220ba 100644
+--- a/arch/x86/kvm/ioapic.c
++++ b/arch/x86/kvm/ioapic.c
+@@ -109,6 +109,7 @@ static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
+ {
+ 	bool new_val, old_val;
+ 	struct kvm_ioapic *ioapic = vcpu->kvm->arch.vioapic;
++	struct dest_map *dest_map = &ioapic->rtc_status.dest_map;
+ 	union kvm_ioapic_redirect_entry *e;
+ 
+ 	e = &ioapic->redirtbl[RTC_GSI];
+@@ -117,16 +118,17 @@ static void __rtc_irq_eoi_tracking_restore_one(struct kvm_vcpu *vcpu)
+ 		return;
+ 
+ 	new_val = kvm_apic_pending_eoi(vcpu, e->fields.vector);
+-	old_val = test_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map.map);
++	old_val = test_bit(vcpu->vcpu_id, dest_map->map);
+ 
+ 	if (new_val == old_val)
+ 		return;
+ 
+ 	if (new_val) {
+-		__set_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map.map);
++		__set_bit(vcpu->vcpu_id, dest_map->map);
++		dest_map->vectors[vcpu->vcpu_id] = e->fields.vector;
+ 		ioapic->rtc_status.pending_eoi++;
+ 	} else {
+-		__clear_bit(vcpu->vcpu_id, ioapic->rtc_status.dest_map.map);
++		__clear_bit(vcpu->vcpu_id, dest_map->map);
+ 		ioapic->rtc_status.pending_eoi--;
+ 		rtc_status_pending_eoi_check_valid(ioapic);
+ 	}
+-- 
+2.1.4
+
-- 
2.1.4