diff mbox series

[RFC,4/4] kvm: Implement atomic memory region resizes via region_resize()

Message ID 20200303141939.352319-5-david@redhat.com (mailing list archive)
State New, archived
Headers show
Series None | expand

Commit Message

David Hildenbrand March 3, 2020, 2:19 p.m. UTC
virtio-mem wants to resize (esp. grow) ram memory regions while the guest
is already aware of them and makes use of them. Resizing a KVM slot can
only currently be done by removing it and re-adding it. While the kvm slot
is temporarily removed, VCPUs that try to read from these slots will fault.
But also, other ioctls might depend on all slots being in place.

Let's inhibit most KVM ioctls while performing the resize. Once we have an
ioctl that can perform atomic resizes (e.g., KVM_SET_USER_MEMORY_REGION
extensions), we can make inhibiting optional at runtime.

Also, make sure to hold the kvm_slots_lock while performing both
actions (removing+re-adding).

Note: Resizes of memory regions currently seems to happen during bootup
only, so I don't think any existing RT users should be affected.

Cc: Richard Henderson <rth@twiddle.net>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: kvm@vger.kernel.org
Signed-off-by: David Hildenbrand <david@redhat.com>
---
 accel/kvm/kvm-all.c   | 121 +++++++++++++++++++++++++++++++++++++++---
 include/hw/core/cpu.h |   3 ++
 2 files changed, 117 insertions(+), 7 deletions(-)

Comments

Paolo Bonzini March 6, 2020, 9:50 a.m. UTC | #1
On 03/03/20 15:19, David Hildenbrand wrote:
> virtio-mem wants to resize (esp. grow) ram memory regions while the guest
> is already aware of them and makes use of them. Resizing a KVM slot can
> only currently be done by removing it and re-adding it. While the kvm slot
> is temporarily removed, VCPUs that try to read from these slots will fault.

Only fetches I think?  Data reads and write would be treated as MMIO
accesses and they should just work (using either the old or new FlatView).

> But also, other ioctls might depend on all slots being in place.
> 
> Let's inhibit most KVM ioctls while performing the resize. Once we have an
> ioctl that can perform atomic resizes (e.g., KVM_SET_USER_MEMORY_REGION
> extensions), we can make inhibiting optional at runtime.
> 
> Also, make sure to hold the kvm_slots_lock while performing both
> actions (removing+re-adding).
>
> Note: Resizes of memory regions currently seems to happen during bootup
> only, so I don't think any existing RT users should be affected.

rwlocks are not efficient, they cause cache line contention.  For
MMIO-heavy workloads the impact will be very large (well, not that large
because right now they all take the BQL, but one can always hope).

I would very much prefer to add a KVM_SET_USER_MEMORY_REGION extension
right away.

Paolo

> Cc: Richard Henderson <rth@twiddle.net>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
> Cc: Eduardo Habkost <ehabkost@redhat.com>
> Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
> Cc: Igor Mammedov <imammedo@redhat.com>
> Cc: kvm@vger.kernel.org
> Signed-off-by: David Hildenbrand <david@redhat.com>
> ---
>  accel/kvm/kvm-all.c   | 121 +++++++++++++++++++++++++++++++++++++++---
>  include/hw/core/cpu.h |   3 ++
>  2 files changed, 117 insertions(+), 7 deletions(-)
> 
> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
> index 439a4efe52..bba58db098 100644
> --- a/accel/kvm/kvm-all.c
> +++ b/accel/kvm/kvm-all.c
> @@ -149,6 +149,21 @@ bool kvm_msi_use_devid;
>  static bool kvm_immediate_exit;
>  static hwaddr kvm_max_slot_size = ~0;
>  
> +/*
> + * While holding this lock in write, no new KVM ioctls can be started, but
> + * kvm ioctl inhibitors will have to wait for existing ones to finish
> + * (indicated by cpu->in_ioctl and kvm_in_ioctl, both updated with this lock
> + * held in read when entering the ioctl).
> + */
> +pthread_rwlock_t kvm_ioctl_lock;
> +/*
> + * Atomic counter of active KVM ioctls except
> + * - The KVM ioctl inhibitor is doing an ioctl
> + * - kvm_ioctl(): Harmless and not interesting for inhibitors.
> + * - kvm_vcpu_ioctl(): Tracked via cpu->in_ioctl.
> + */
> +static int kvm_in_ioctl;
> +
>  static const KVMCapabilityInfo kvm_required_capabilites[] = {
>      KVM_CAP_INFO(USER_MEMORY),
>      KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
> @@ -1023,6 +1038,7 @@ void kvm_set_max_memslot_size(hwaddr max_slot_size)
>      kvm_max_slot_size = max_slot_size;
>  }
>  
> +/* Called with KVMMemoryListener.slots_lock held */
>  static void kvm_set_phys_mem(KVMMemoryListener *kml,
>                               MemoryRegionSection *section, bool add)
>  {
> @@ -1052,14 +1068,12 @@ static void kvm_set_phys_mem(KVMMemoryListener *kml,
>      ram = memory_region_get_ram_ptr(mr) + section->offset_within_region +
>            (start_addr - section->offset_within_address_space);
>  
> -    kvm_slots_lock(kml);
> -
>      if (!add) {
>          do {
>              slot_size = MIN(kvm_max_slot_size, size);
>              mem = kvm_lookup_matching_slot(kml, start_addr, slot_size);
>              if (!mem) {
> -                goto out;
> +                return;
>              }
>              if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
>                  kvm_physical_sync_dirty_bitmap(kml, section);
> @@ -1079,7 +1093,7 @@ static void kvm_set_phys_mem(KVMMemoryListener *kml,
>              start_addr += slot_size;
>              size -= slot_size;
>          } while (size);
> -        goto out;
> +        return;
>      }
>  
>      /* register the new slot */
> @@ -1108,9 +1122,6 @@ static void kvm_set_phys_mem(KVMMemoryListener *kml,
>          ram += slot_size;
>          size -= slot_size;
>      } while (size);
> -
> -out:
> -    kvm_slots_unlock(kml);
>  }
>  
>  static void kvm_region_add(MemoryListener *listener,
> @@ -1119,7 +1130,9 @@ static void kvm_region_add(MemoryListener *listener,
>      KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
>  
>      memory_region_ref(section->mr);
> +    kvm_slots_lock(kml);
>      kvm_set_phys_mem(kml, section, true);
> +    kvm_slots_unlock(kml);
>  }
>  
>  static void kvm_region_del(MemoryListener *listener,
> @@ -1127,10 +1140,68 @@ static void kvm_region_del(MemoryListener *listener,
>  {
>      KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
>  
> +    kvm_slots_lock(kml);
>      kvm_set_phys_mem(kml, section, false);
> +    kvm_slots_unlock(kml);
>      memory_region_unref(section->mr);
>  }
>  
> +/*
> + * Certain updates (e.g., resizing memory regions) require temporarily removing
> + * kvm memory slots. Make sure any ioctl sees a consistent memory slot state.
> + */
> +static void kvm_ioctl_inhibit_begin(void)
> +{
> +    CPUState *cpu;
> +
> +    /*
> +     * We allow to inhibit only when holding the BQL, so we can identify
> +     * when an inhibitor wants to issue an ioctl easily.
> +     */
> +    g_assert(qemu_mutex_iothread_locked());
> +
> +    pthread_rwlock_wrlock(&kvm_ioctl_lock);
> +
> +    /* Inhibiting happens rarely, we can keep things simple and spin here. */
> +    while (true) {
> +        bool any_cpu_in_ioctl = false;
> +
> +        CPU_FOREACH(cpu) {
> +            if (atomic_read(&cpu->in_ioctl)) {
> +                any_cpu_in_ioctl = true;
> +                qemu_cpu_kick(cpu);
> +            }
> +        }
> +        if (!any_cpu_in_ioctl && !atomic_read(&kvm_in_ioctl)) {
> +            break;
> +        }
> +        g_usleep(100);
> +    }
> +}
> +
> +static void kvm_ioctl_inhibit_end(void)
> +{
> +    pthread_rwlock_unlock(&kvm_ioctl_lock);
> +}
> +
> +static void kvm_region_resize(MemoryListener *listener,
> +                              MemoryRegionSection *section, Int128 new)
> +{
> +    KVMMemoryListener *kml = container_of(listener, KVMMemoryListener,
> +                                          listener);
> +    MemoryRegionSection new_section = *section;
> +
> +    new_section.size = new;
> +
> +    kvm_slots_lock(kml);
> +    /* Inhibit KVM ioctls while temporarily removing slots. */
> +    kvm_ioctl_inhibit_begin();
> +    kvm_set_phys_mem(kml, section, false);
> +    kvm_set_phys_mem(kml, &new_section, true);
> +    kvm_ioctl_inhibit_end();
> +    kvm_slots_unlock(kml);
> +}
> +
>  static void kvm_log_sync(MemoryListener *listener,
>                           MemoryRegionSection *section)
>  {
> @@ -1249,6 +1320,7 @@ void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
>  
>      kml->listener.region_add = kvm_region_add;
>      kml->listener.region_del = kvm_region_del;
> +    kml->listener.region_resize = kvm_region_resize;
>      kml->listener.log_start = kvm_log_start;
>      kml->listener.log_stop = kvm_log_stop;
>      kml->listener.log_sync = kvm_log_sync;
> @@ -1894,6 +1966,7 @@ static int kvm_init(MachineState *ms)
>      assert(TARGET_PAGE_SIZE <= qemu_real_host_page_size);
>  
>      s->sigmask_len = 8;
> +    pthread_rwlock_init(&kvm_ioctl_lock, NULL);
>  
>  #ifdef KVM_CAP_SET_GUEST_DEBUG
>      QTAILQ_INIT(&s->kvm_sw_breakpoints);
> @@ -2304,6 +2377,34 @@ static void kvm_eat_signals(CPUState *cpu)
>      } while (sigismember(&chkset, SIG_IPI));
>  }
>  
> +static void kvm_cpu_set_in_ioctl(CPUState *cpu, bool in_ioctl)
> +{
> +    if (unlikely(qemu_mutex_iothread_locked())) {
> +        return;
> +    }
> +    if (in_ioctl) {
> +        pthread_rwlock_rdlock(&kvm_ioctl_lock);
> +        atomic_set(&cpu->in_ioctl, true);
> +        pthread_rwlock_unlock(&kvm_ioctl_lock);
> +    } else {
> +        atomic_set(&cpu->in_ioctl, false);
> +    }
> +}
> +
> +static void kvm_set_in_ioctl(bool in_ioctl)
> +{
> +    if (likely(qemu_mutex_iothread_locked())) {
> +        return;
> +    }
> +    if (in_ioctl) {
> +        pthread_rwlock_rdlock(&kvm_ioctl_lock);
> +        atomic_inc(&kvm_in_ioctl);
> +        pthread_rwlock_unlock(&kvm_ioctl_lock);
> +    } else {
> +        atomic_dec(&kvm_in_ioctl);
> +    }
> +}
> +
>  int kvm_cpu_exec(CPUState *cpu)
>  {
>      struct kvm_run *run = cpu->kvm_run;
> @@ -2488,7 +2589,9 @@ int kvm_vm_ioctl(KVMState *s, int type, ...)
>      va_end(ap);
>  
>      trace_kvm_vm_ioctl(type, arg);
> +    kvm_set_in_ioctl(true);
>      ret = ioctl(s->vmfd, type, arg);
> +    kvm_set_in_ioctl(false);
>      if (ret == -1) {
>          ret = -errno;
>      }
> @@ -2506,7 +2609,9 @@ int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
>      va_end(ap);
>  
>      trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
> +    kvm_cpu_set_in_ioctl(cpu, true);
>      ret = ioctl(cpu->kvm_fd, type, arg);
> +    kvm_cpu_set_in_ioctl(cpu, false);
>      if (ret == -1) {
>          ret = -errno;
>      }
> @@ -2524,7 +2629,9 @@ int kvm_device_ioctl(int fd, int type, ...)
>      va_end(ap);
>  
>      trace_kvm_device_ioctl(fd, type, arg);
> +    kvm_set_in_ioctl(true);
>      ret = ioctl(fd, type, arg);
> +    kvm_set_in_ioctl(false);
>      if (ret == -1) {
>          ret = -errno;
>      }
> diff --git a/include/hw/core/cpu.h b/include/hw/core/cpu.h
> index 73e9a869a4..4fbff6f3d7 100644
> --- a/include/hw/core/cpu.h
> +++ b/include/hw/core/cpu.h
> @@ -431,6 +431,9 @@ struct CPUState {
>      /* shared by kvm, hax and hvf */
>      bool vcpu_dirty;
>  
> +    /* kvm only for now: CPU is in kvm_vcpu_ioctl() (esp. KVM_RUN) */
> +    bool in_ioctl;
> +
>      /* Used to keep track of an outstanding cpu throttle thread for migration
>       * autoconverge
>       */
>
David Hildenbrand March 6, 2020, 10:20 a.m. UTC | #2
On 06.03.20 10:50, Paolo Bonzini wrote:
> On 03/03/20 15:19, David Hildenbrand wrote:
>> virtio-mem wants to resize (esp. grow) ram memory regions while the guest
>> is already aware of them and makes use of them. Resizing a KVM slot can
>> only currently be done by removing it and re-adding it. While the kvm slot
>> is temporarily removed, VCPUs that try to read from these slots will fault.
> 

s/try to read/try to access/

> Only fetches I think?  Data reads and write would be treated as MMIO
> accesses and they should just work (using either the old or new FlatView).

On x86-64, I saw KVM fault printks getting printed (it was about 1-2
years ago, though, when I realized this was a problem). Could be that
these were fetches. At least the guest eventually crashed :)

On other archs (esp. s390x) guests will directly receive a
PGM_ADDRESSING from KVM if they stumble over memory that is not covered
by a kvm slot.

> 
>> But also, other ioctls might depend on all slots being in place.
>>
>> Let's inhibit most KVM ioctls while performing the resize. Once we have an
>> ioctl that can perform atomic resizes (e.g., KVM_SET_USER_MEMORY_REGION
>> extensions), we can make inhibiting optional at runtime.
>>
>> Also, make sure to hold the kvm_slots_lock while performing both
>> actions (removing+re-adding).
>>
>> Note: Resizes of memory regions currently seems to happen during bootup
>> only, so I don't think any existing RT users should be affected.
> 
> rwlocks are not efficient, they cause cache line contention.  For
> MMIO-heavy workloads the impact will be very large (well, not that large
> because right now they all take the BQL, but one can always hope).

Yeah, rwlocks are not optimal and I am still looking for better
alternatives (suggestions welcome :) ). Using RCU might not work,
because the rcu_read region might be too big (esp. while in KVM_RUN).

I had a prototype which used a bunch of atomics + qemu_cond_wait. But it
was quite elaborate and buggy.

(I assume only going into KVM_RUN is really affected, and I do wonder if
it will be noticeable at all. Doing an ioctl is always already an
expensive operation.)

I can look into per-cpu locks instead of the rwlock.

> 
> I would very much prefer to add a KVM_SET_USER_MEMORY_REGION extension
> right away.
> 

I really want to avoid dependencies on kernel features to at least make
it work for now. Especially, resizing memory slots in KVM (especially
while dirty bitmaps, rmaps, etc. are active) is non-trivial.

Thanks Paolo!
Paolo Bonzini March 6, 2020, 11:38 a.m. UTC | #3
On 06/03/20 11:20, David Hildenbrand wrote:
> Yeah, rwlocks are not optimal and I am still looking for better
> alternatives (suggestions welcome :) ). Using RCU might not work,
> because the rcu_read region might be too big (esp. while in KVM_RUN).
> 
> I had a prototype which used a bunch of atomics + qemu_cond_wait. But it
> was quite elaborate and buggy.
> 
> (I assume only going into KVM_RUN is really affected, and I do wonder if
> it will be noticeable at all. Doing an ioctl is always already an
> expensive operation.)
> 
> I can look into per-cpu locks instead of the rwlock.

Assuming we're only talking about CPU ioctls (seems like a good
approximation) maybe you could use start_exclusive/end_exclusive?  The
current_cpu->in_exclusive_context assignments can be made conditional on
"if (current_cpu)".

However that means you have to drop the BQL, see
process_queued_cpu_work.  It may be a problem.


Paolo
David Hildenbrand March 6, 2020, 12:18 p.m. UTC | #4
On 06.03.20 12:38, Paolo Bonzini wrote:
> On 06/03/20 11:20, David Hildenbrand wrote:
>> Yeah, rwlocks are not optimal and I am still looking for better
>> alternatives (suggestions welcome :) ). Using RCU might not work,
>> because the rcu_read region might be too big (esp. while in KVM_RUN).
>>
>> I had a prototype which used a bunch of atomics + qemu_cond_wait. But it
>> was quite elaborate and buggy.
>>
>> (I assume only going into KVM_RUN is really affected, and I do wonder if
>> it will be noticeable at all. Doing an ioctl is always already an
>> expensive operation.)
>>
>> I can look into per-cpu locks instead of the rwlock.
> 
> Assuming we're only talking about CPU ioctls (seems like a good

Yeah, I guess most !CPU ioctls are done under the BQL.

> approximation) maybe you could use start_exclusive/end_exclusive?  The
> current_cpu->in_exclusive_context assignments can be made conditional on
> "if (current_cpu)".
> 
> However that means you have to drop the BQL, see
> process_queued_cpu_work.  It may be a problem.

Thanks, I'll look into that. I currently have a simple cpu->ioctl_mutex.

Cheers!
David Hildenbrand March 6, 2020, 2:30 p.m. UTC | #5
On 06.03.20 12:38, Paolo Bonzini wrote:
> On 06/03/20 11:20, David Hildenbrand wrote:
>> Yeah, rwlocks are not optimal and I am still looking for better
>> alternatives (suggestions welcome :) ). Using RCU might not work,
>> because the rcu_read region might be too big (esp. while in KVM_RUN).
>>
>> I had a prototype which used a bunch of atomics + qemu_cond_wait. But it
>> was quite elaborate and buggy.
>>
>> (I assume only going into KVM_RUN is really affected, and I do wonder if
>> it will be noticeable at all. Doing an ioctl is always already an
>> expensive operation.)
>>
>> I can look into per-cpu locks instead of the rwlock.
> 
> Assuming we're only talking about CPU ioctls (seems like a good
> approximation) maybe you could use start_exclusive/end_exclusive?  The
> current_cpu->in_exclusive_context assignments can be made conditional on
> "if (current_cpu)".
> 
> However that means you have to drop the BQL, see
> process_queued_cpu_work.  It may be a problem.
> 

Yeah, start_exclusive() is expected to be called without the BQL,
otherwise the other CPUs would not be able to make progress and can
eventually be "caught".

It's essentially the same reason why I can't use high-level
pause_all_vcpus()/resume_all_vcpus(). Will drop the BQL which is very
bad for resizing code.
Paolo Bonzini March 6, 2020, 2:39 p.m. UTC | #6
On 06/03/20 15:30, David Hildenbrand wrote:
>> Assuming we're only talking about CPU ioctls (seems like a good
>> approximation) maybe you could use start_exclusive/end_exclusive?  The
>> current_cpu->in_exclusive_context assignments can be made conditional on
>> "if (current_cpu)".
>>
>> However that means you have to drop the BQL, see
>> process_queued_cpu_work.  It may be a problem.
>>
> Yeah, start_exclusive() is expected to be called without the BQL,
> otherwise the other CPUs would not be able to make progress and can
> eventually be "caught".
> 
> It's essentially the same reason why I can't use high-level
> pause_all_vcpus()/resume_all_vcpus(). Will drop the BQL which is very
> bad for resizing code.

But any other synchronization primitive that you do which blocks all
vCPUs will have the same issue, otherwise you get a deadlock.

Paolo
David Hildenbrand March 6, 2020, 2:44 p.m. UTC | #7
On 06.03.20 15:39, Paolo Bonzini wrote:
> On 06/03/20 15:30, David Hildenbrand wrote:
>>> Assuming we're only talking about CPU ioctls (seems like a good
>>> approximation) maybe you could use start_exclusive/end_exclusive?  The
>>> current_cpu->in_exclusive_context assignments can be made conditional on
>>> "if (current_cpu)".
>>>
>>> However that means you have to drop the BQL, see
>>> process_queued_cpu_work.  It may be a problem.
>>>
>> Yeah, start_exclusive() is expected to be called without the BQL,
>> otherwise the other CPUs would not be able to make progress and can
>> eventually be "caught".
>>
>> It's essentially the same reason why I can't use high-level
>> pause_all_vcpus()/resume_all_vcpus(). Will drop the BQL which is very
>> bad for resizing code.
> 
> But any other synchronization primitive that you do which blocks all
> vCPUs will have the same issue, otherwise you get a deadlock.

This is essentially what this patch solves.

The lock essentially blocks anybody from entering, but not leaving a KVM
ioctl. An inhibitor only waits for all IOCTLs to be left. No other lock
prohibits that, so I don't think there can ever be a deadlock.
diff mbox series

Patch

diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
index 439a4efe52..bba58db098 100644
--- a/accel/kvm/kvm-all.c
+++ b/accel/kvm/kvm-all.c
@@ -149,6 +149,21 @@  bool kvm_msi_use_devid;
 static bool kvm_immediate_exit;
 static hwaddr kvm_max_slot_size = ~0;
 
+/*
+ * While holding this lock in write, no new KVM ioctls can be started, but
+ * kvm ioctl inhibitors will have to wait for existing ones to finish
+ * (indicated by cpu->in_ioctl and kvm_in_ioctl, both updated with this lock
+ * held in read when entering the ioctl).
+ */
+pthread_rwlock_t kvm_ioctl_lock;
+/*
+ * Atomic counter of active KVM ioctls except
+ * - The KVM ioctl inhibitor is doing an ioctl
+ * - kvm_ioctl(): Harmless and not interesting for inhibitors.
+ * - kvm_vcpu_ioctl(): Tracked via cpu->in_ioctl.
+ */
+static int kvm_in_ioctl;
+
 static const KVMCapabilityInfo kvm_required_capabilites[] = {
     KVM_CAP_INFO(USER_MEMORY),
     KVM_CAP_INFO(DESTROY_MEMORY_REGION_WORKS),
@@ -1023,6 +1038,7 @@  void kvm_set_max_memslot_size(hwaddr max_slot_size)
     kvm_max_slot_size = max_slot_size;
 }
 
+/* Called with KVMMemoryListener.slots_lock held */
 static void kvm_set_phys_mem(KVMMemoryListener *kml,
                              MemoryRegionSection *section, bool add)
 {
@@ -1052,14 +1068,12 @@  static void kvm_set_phys_mem(KVMMemoryListener *kml,
     ram = memory_region_get_ram_ptr(mr) + section->offset_within_region +
           (start_addr - section->offset_within_address_space);
 
-    kvm_slots_lock(kml);
-
     if (!add) {
         do {
             slot_size = MIN(kvm_max_slot_size, size);
             mem = kvm_lookup_matching_slot(kml, start_addr, slot_size);
             if (!mem) {
-                goto out;
+                return;
             }
             if (mem->flags & KVM_MEM_LOG_DIRTY_PAGES) {
                 kvm_physical_sync_dirty_bitmap(kml, section);
@@ -1079,7 +1093,7 @@  static void kvm_set_phys_mem(KVMMemoryListener *kml,
             start_addr += slot_size;
             size -= slot_size;
         } while (size);
-        goto out;
+        return;
     }
 
     /* register the new slot */
@@ -1108,9 +1122,6 @@  static void kvm_set_phys_mem(KVMMemoryListener *kml,
         ram += slot_size;
         size -= slot_size;
     } while (size);
-
-out:
-    kvm_slots_unlock(kml);
 }
 
 static void kvm_region_add(MemoryListener *listener,
@@ -1119,7 +1130,9 @@  static void kvm_region_add(MemoryListener *listener,
     KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
 
     memory_region_ref(section->mr);
+    kvm_slots_lock(kml);
     kvm_set_phys_mem(kml, section, true);
+    kvm_slots_unlock(kml);
 }
 
 static void kvm_region_del(MemoryListener *listener,
@@ -1127,10 +1140,68 @@  static void kvm_region_del(MemoryListener *listener,
 {
     KVMMemoryListener *kml = container_of(listener, KVMMemoryListener, listener);
 
+    kvm_slots_lock(kml);
     kvm_set_phys_mem(kml, section, false);
+    kvm_slots_unlock(kml);
     memory_region_unref(section->mr);
 }
 
+/*
+ * Certain updates (e.g., resizing memory regions) require temporarily removing
+ * kvm memory slots. Make sure any ioctl sees a consistent memory slot state.
+ */
+static void kvm_ioctl_inhibit_begin(void)
+{
+    CPUState *cpu;
+
+    /*
+     * We allow to inhibit only when holding the BQL, so we can identify
+     * when an inhibitor wants to issue an ioctl easily.
+     */
+    g_assert(qemu_mutex_iothread_locked());
+
+    pthread_rwlock_wrlock(&kvm_ioctl_lock);
+
+    /* Inhibiting happens rarely, we can keep things simple and spin here. */
+    while (true) {
+        bool any_cpu_in_ioctl = false;
+
+        CPU_FOREACH(cpu) {
+            if (atomic_read(&cpu->in_ioctl)) {
+                any_cpu_in_ioctl = true;
+                qemu_cpu_kick(cpu);
+            }
+        }
+        if (!any_cpu_in_ioctl && !atomic_read(&kvm_in_ioctl)) {
+            break;
+        }
+        g_usleep(100);
+    }
+}
+
+static void kvm_ioctl_inhibit_end(void)
+{
+    pthread_rwlock_unlock(&kvm_ioctl_lock);
+}
+
+static void kvm_region_resize(MemoryListener *listener,
+                              MemoryRegionSection *section, Int128 new)
+{
+    KVMMemoryListener *kml = container_of(listener, KVMMemoryListener,
+                                          listener);
+    MemoryRegionSection new_section = *section;
+
+    new_section.size = new;
+
+    kvm_slots_lock(kml);
+    /* Inhibit KVM ioctls while temporarily removing slots. */
+    kvm_ioctl_inhibit_begin();
+    kvm_set_phys_mem(kml, section, false);
+    kvm_set_phys_mem(kml, &new_section, true);
+    kvm_ioctl_inhibit_end();
+    kvm_slots_unlock(kml);
+}
+
 static void kvm_log_sync(MemoryListener *listener,
                          MemoryRegionSection *section)
 {
@@ -1249,6 +1320,7 @@  void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
 
     kml->listener.region_add = kvm_region_add;
     kml->listener.region_del = kvm_region_del;
+    kml->listener.region_resize = kvm_region_resize;
     kml->listener.log_start = kvm_log_start;
     kml->listener.log_stop = kvm_log_stop;
     kml->listener.log_sync = kvm_log_sync;
@@ -1894,6 +1966,7 @@  static int kvm_init(MachineState *ms)
     assert(TARGET_PAGE_SIZE <= qemu_real_host_page_size);
 
     s->sigmask_len = 8;
+    pthread_rwlock_init(&kvm_ioctl_lock, NULL);
 
 #ifdef KVM_CAP_SET_GUEST_DEBUG
     QTAILQ_INIT(&s->kvm_sw_breakpoints);
@@ -2304,6 +2377,34 @@  static void kvm_eat_signals(CPUState *cpu)
     } while (sigismember(&chkset, SIG_IPI));
 }
 
+static void kvm_cpu_set_in_ioctl(CPUState *cpu, bool in_ioctl)
+{
+    if (unlikely(qemu_mutex_iothread_locked())) {
+        return;
+    }
+    if (in_ioctl) {
+        pthread_rwlock_rdlock(&kvm_ioctl_lock);
+        atomic_set(&cpu->in_ioctl, true);
+        pthread_rwlock_unlock(&kvm_ioctl_lock);
+    } else {
+        atomic_set(&cpu->in_ioctl, false);
+    }
+}
+
+static void kvm_set_in_ioctl(bool in_ioctl)
+{
+    if (likely(qemu_mutex_iothread_locked())) {
+        return;
+    }
+    if (in_ioctl) {
+        pthread_rwlock_rdlock(&kvm_ioctl_lock);
+        atomic_inc(&kvm_in_ioctl);
+        pthread_rwlock_unlock(&kvm_ioctl_lock);
+    } else {
+        atomic_dec(&kvm_in_ioctl);
+    }
+}
+
 int kvm_cpu_exec(CPUState *cpu)
 {
     struct kvm_run *run = cpu->kvm_run;
@@ -2488,7 +2589,9 @@  int kvm_vm_ioctl(KVMState *s, int type, ...)
     va_end(ap);
 
     trace_kvm_vm_ioctl(type, arg);
+    kvm_set_in_ioctl(true);
     ret = ioctl(s->vmfd, type, arg);
+    kvm_set_in_ioctl(false);
     if (ret == -1) {
         ret = -errno;
     }
@@ -2506,7 +2609,9 @@  int kvm_vcpu_ioctl(CPUState *cpu, int type, ...)
     va_end(ap);
 
     trace_kvm_vcpu_ioctl(cpu->cpu_index, type, arg);
+    kvm_cpu_set_in_ioctl(cpu, true);
     ret = ioctl(cpu->kvm_fd, type, arg);
+    kvm_cpu_set_in_ioctl(cpu, false);
     if (ret == -1) {
         ret = -errno;
     }
@@ -2524,7 +2629,9 @@  int kvm_device_ioctl(int fd, int type, ...)
     va_end(ap);
 
     trace_kvm_device_ioctl(fd, type, arg);
+    kvm_set_in_ioctl(true);
     ret = ioctl(fd, type, arg);
+    kvm_set_in_ioctl(false);
     if (ret == -1) {
         ret = -errno;
     }
diff --git a/include/hw/core/cpu.h b/include/hw/core/cpu.h
index 73e9a869a4..4fbff6f3d7 100644
--- a/include/hw/core/cpu.h
+++ b/include/hw/core/cpu.h
@@ -431,6 +431,9 @@  struct CPUState {
     /* shared by kvm, hax and hvf */
     bool vcpu_dirty;
 
+    /* kvm only for now: CPU is in kvm_vcpu_ioctl() (esp. KVM_RUN) */
+    bool in_ioctl;
+
     /* Used to keep track of an outstanding cpu throttle thread for migration
      * autoconverge
      */