From 68d5cd786a054bc1e84b1ee6a0deeefbfe2c33b9 Mon Sep 17 00:00:00 2001
From: Bandan Das <bsd@redhat.com>
Date: Fri, 14 Dec 2018 19:35:46 +0000
Subject: [PATCH 5/5] slow train kvm: clear out
 KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT for older machine types

RH-Author: Bandan Das <bsd@redhat.com>
Message-id: <jpgftuz9a0d.fsf@linux.bootlegged.copy>
Patchwork-id: 83524
O-Subject: [RHEL8 qemu-kvm PATCH] slow train kvm: clear out KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT for older machine types
Bugzilla: 1656829
RH-Acked-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
RH-Acked-by: Eduardo Habkost <ehabkost@redhat.com>
RH-Acked-by: Paolo Bonzini <pbonzini@redhat.com>

Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1656829
Brew: https://brewweb.engineering.redhat.com/brew/taskinfo?taskID=19521977
Upstream: Not applicable
Branch: rhel8/master-2.12.0

After the addition of support for async pf injection to L1, newer
hypervisors  advertise the feature using bit 2 of the
MSR_KVM_ASYNC_PF_EN msr. However, this was reserved in older
hypervisors which results in an error during migration like so:

qemu-kvm: error: failed to set MSR 0x4b564d02 to 0x27fc13285
qemu-kvm: /builddir/build/BUILD/qemu-2.12.0/target/i386/kvm.c:1940: kvm_put_msrs: Assertion `ret == cpu->kvm_msr_buf->nmsrs' failed.
Aborted (core dumped)

This patch introduces a new bool that is set for older machine types.
When set, Qemu's stored value clears out bit 2. This should be safe
because the guest can still enable it by writing to the MSR after
checking for support. A reset/migration for <7.6 machine type would
reset the bit though.

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Danilo C. L. de Paula <ddepaula@redhat.com>
---
 hw/i386/pc.c        | 5 +++++
 include/hw/boards.h | 2 ++
 target/i386/kvm.c   | 4 ++++
 3 files changed, 11 insertions(+)

diff --git a/hw/i386/pc.c b/hw/i386/pc.c
index 9034f02..9e1e6ae 100644
--- a/hw/i386/pc.c
+++ b/hw/i386/pc.c
@@ -2363,6 +2363,11 @@ static void pc_machine_class_init(ObjectClass *oc, void *data)
     pcmc->save_tsc_khz = true;
     pcmc->linuxboot_dma_enabled = true;
     pcmc->pc_rom_ro = true;
+    /*
+     * This might have to be set to false
+     * when and if there's a new machine type
+     */
+    mc->async_pf_vmexit_disable = true;
     mc->get_hotplug_handler = pc_get_hotpug_handler;
     mc->cpu_index_to_instance_props = pc_cpu_index_to_props;
     mc->get_default_cpu_node_id = pc_get_default_cpu_node_id;
diff --git a/include/hw/boards.h b/include/hw/boards.h
index a609239..9b4a69b 100644
--- a/include/hw/boards.h
+++ b/include/hw/boards.h
@@ -203,6 +203,8 @@ struct MachineClass {
     const char **valid_cpu_types;
     strList *allowed_dynamic_sysbus_devices;
     bool auto_enable_numa_with_memhp;
+    /* RHEL only */
+    bool async_pf_vmexit_disable;
     void (*numa_auto_assign_ram)(MachineClass *mc, NodeInfo *nodes,
                                  int nb_nodes, ram_addr_t size);
 
diff --git a/target/i386/kvm.c b/target/i386/kvm.c
index 00f2141..87e4771 100644
--- a/target/i386/kvm.c
+++ b/target/i386/kvm.c
@@ -2072,6 +2072,7 @@ static int kvm_get_msrs(X86CPU *cpu)
     struct kvm_msr_entry *msrs = cpu->kvm_msr_buf->entries;
     int ret, i;
     uint64_t mtrr_top_bits;
+    MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
 
     kvm_msr_buf_reset(cpu);
 
@@ -2364,6 +2365,9 @@ static int kvm_get_msrs(X86CPU *cpu)
             break;
         case MSR_KVM_ASYNC_PF_EN:
             env->async_pf_en_msr = msrs[i].data;
+            if (mc->async_pf_vmexit_disable) {
+                env->async_pf_en_msr &= ~(1ULL << 2);
+            }
             break;
         case MSR_KVM_PV_EOI_EN:
             env->pv_eoi_en_msr = msrs[i].data;
-- 
1.8.3.1