Re: [Qemu-devel] [PATCH 2/3] virtio: postpone the execution of event_notifier_cleanup function

[Marc-André Lureau]

Hi

On Mon, Jan 22, 2018 at 12:53 PM, Michal Privoznik <address@hidden> wrote:
> On 01/14/2018 11:06 AM, Gal Hammer wrote:
>> Use the EventNotifier's cleanup callback function to execute the
>> event_notifier_cleanup function after kvm unregistered the eventfd.
>>
>> This change supports running the virtio_bus_set_host_notifier
>> function inside a memory region transaction. Otherwise, a closed
>> fd is sent to kvm, which results in a failure.
>>
>> Signed-off-by: Gal Hammer <address@hidden>
>> ---
>>  accel/kvm/kvm-all.c    |  4 ++++
>>  hw/virtio/virtio-bus.c | 19 +++++++++++--------
>>  2 files changed, 15 insertions(+), 8 deletions(-)
>>
>> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
>> index f290f48..071f4f5 100644
>> --- a/accel/kvm/kvm-all.c
>> +++ b/accel/kvm/kvm-all.c
>> @@ -812,6 +812,10 @@ static void kvm_mem_ioeventfd_del(MemoryListener 
>> *listener,
>>      if (r < 0) {
>>          abort();
>>      }
>> +
>> +    if (e->cleanup) {
>> +        e->cleanup(e);
>> +    }
>>  }
>>
>>  static void kvm_io_ioeventfd_add(MemoryListener *listener,
>> diff --git a/hw/virtio/virtio-bus.c b/hw/virtio/virtio-bus.c
>> index 3042232..8106346 100644
>> --- a/hw/virtio/virtio-bus.c
>> +++ b/hw/virtio/virtio-bus.c
>> @@ -256,6 +256,15 @@ bool virtio_bus_ioeventfd_enabled(VirtioBusState *bus)
>>      return k->ioeventfd_assign && k->ioeventfd_enabled(proxy);
>>  }
>>
>> +static void virtio_bus_cleanup_event_notifier(EventNotifier *notifier)
>> +{
>> +    /* Test and clear notifier after disabling event,
>> +     * in case poll callback didn't have time to run.
>> +     */
>> +    virtio_queue_host_notifier_read(notifier);
>> +    event_notifier_cleanup(notifier);
>> +}
>> +
>>  /*
>>   * This function switches ioeventfd on/off in the device.
>>   * The caller must set or clear the handlers for the EventNotifier.
>> @@ -283,19 +292,13 @@ int virtio_bus_set_host_notifier(VirtioBusState *bus, 
>> int n, bool assign)
>>          r = k->ioeventfd_assign(proxy, notifier, n, true);
>>          if (r < 0) {
>>              error_report("%s: unable to assign ioeventfd: %d", __func__, r);
>> -            goto cleanup_event_notifier;
>> +            virtio_bus_cleanup_event_notifier(notifier);
>>          }
>> -        return 0;
>>      } else {
>> +        notifier->cleanup = virtio_bus_cleanup_event_notifier;
>>          k->ioeventfd_assign(proxy, notifier, n, false);
>>      }
>>
>> -cleanup_event_notifier:
>> -    /* Test and clear notifier after disabling event,
>> -     * in case poll callback didn't have time to run.
>> -     */
>> -    virtio_queue_host_notifier_read(notifier);
>> -    event_notifier_cleanup(notifier);
>>      return r;
>>  }
>>
>>
>
> This causes abort() to me:
>
> Thread 7 "CPU 3/KVM" received signal SIGABRT, Aborted.
> [Switching to Thread 0x7f78fa1f4700 (LWP 13737)]
> __GI_raise (address@hidden) at ../sysdeps/unix/sysv/linux/raise.c:51
> 51      }
> __GI_raise 7 # bt
> #0  __GI_raise (address@hidden) at ../sysdeps/unix/sysv/linux/raise.c:51
> #1  0x00007f7a09ff5c1a in __GI_abort () at abort.c:89
> #2  0x000055f2a70bce9d in kvm_io_ioeventfd_del (listener=0x55f2a7d19200 
> <kvm_io_listener>, section=0x7f78fa1f10f0, match_data=true, data=0, 
> e=0x7f79fc81f080) at /home/zippy/work/qemu/qemu.git/accel/kvm/kvm-all.c:852
> #3  0x000055f2a70a6fac in address_space_add_del_ioeventfds (as=0x55f2a7e86620 
> <address_space_io>, fds_new=0x7f78e40b9e80, fds_new_nb=2, 
> fds_old=0x7f78e40b9f10, fds_old_nb=3) at 
> /home/zippy/work/qemu/qemu.git/memory.c:832
> #4  0x000055f2a70a744d in address_space_update_ioeventfds (as=0x55f2a7e86620 
> <address_space_io>) at /home/zippy/work/qemu/qemu.git/memory.c:895
> #5  0x000055f2a70a7f04 in memory_region_transaction_commit () at 
> /home/zippy/work/qemu/qemu.git/memory.c:1097
> #6  0x000055f2a70ab0d1 in memory_region_del_eventfd (mr=0x55f2a9422250, 
> addr=16, size=2, match_data=true, data=0, e=0x7f79fc81f080) at 
> /home/zippy/work/qemu/qemu.git/memory.c:2273
> #7  0x000055f2a7411f18 in virtio_pci_ioeventfd_assign (d=0x55f2a9421970, 
> notifier=0x7f79fc81f080, n=0, assign=false) at hw/virtio/virtio-pci.c:280
> #8  0x000055f2a7418e6f in virtio_bus_set_host_notifier (bus=0x55f2a9429a68, 
> n=0, assign=false) at hw/virtio/virtio-bus.c:299
> #9  0x000055f2a710236a in virtio_scsi_dataplane_stop (vdev=0x55f2a9429ae0) at 
> /home/zippy/work/qemu/qemu.git/hw/scsi/virtio-scsi-dataplane.c:215
> #10 0x000055f2a7418bb3 in virtio_bus_stop_ioeventfd (bus=0x55f2a9429a68) at 
> hw/virtio/virtio-bus.c:246
> #11 0x000055f2a7411f63 in virtio_pci_stop_ioeventfd (proxy=0x55f2a9421970) at 
> hw/virtio/virtio-pci.c:294
> #12 0x000055f2a74144ab in virtio_pci_common_write (opaque=0x55f2a9421970, 
> addr=20, val=0, size=1) at hw/virtio/virtio-pci.c:1262
> #13 0x000055f2a70a6016 in memory_region_write_accessor (mr=0x55f2a9422340, 
> addr=20, value=0x7f78fa1f1558, size=1, shift=0, mask=255, attrs=...) at 
> /home/zippy/work/qemu/qemu.git/memory.c:560
> #14 0x000055f2a70a622e in access_with_adjusted_size (addr=20, 
> value=0x7f78fa1f1558, size=1, access_size_min=1, access_size_max=4, 
> access_fn=0x55f2a70a5f2c <memory_region_write_accessor>, mr=0x55f2a9422340, 
> attrs=...) at /home/zippy/work/qemu/qemu.git/memory.c:627
> #15 0x000055f2a70a8eb0 in memory_region_dispatch_write (mr=0x55f2a9422340, 
> addr=20, data=0, size=1, attrs=...) at 
> /home/zippy/work/qemu/qemu.git/memory.c:1503
> #16 0x000055f2a7058b2b in flatview_write_continue (fv=0x7f78e405eba0, 
> addr=4269834260, attrs=..., buf=0x7f7a26c36028 "", len=1, addr1=20, l=1, 
> mr=0x55f2a9422340) at /home/zippy/work/qemu/qemu.git/exec.c:3038
> #17 0x000055f2a7058c79 in flatview_write (fv=0x7f78e405eba0, addr=4269834260, 
> attrs=..., buf=0x7f7a26c36028 "", len=1) at 
> /home/zippy/work/qemu/qemu.git/exec.c:3083
> #18 0x000055f2a7059050 in flatview_rw (fv=0x7f78e405eba0, addr=4269834260, 
> attrs=..., buf=0x7f7a26c36028 "", len=1, is_write=true) at 
> /home/zippy/work/qemu/qemu.git/exec.c:3192
> #19 0x000055f2a7059108 in address_space_rw (as=0x55f2a7e86680 
> <address_space_memory>, addr=4269834260, attrs=..., buf=0x7f7a26c36028 "", 
> len=1, is_write=true) at /home/zippy/work/qemu/qemu.git/exec.c:3202
> #20 0x000055f2a70bf42c in kvm_cpu_exec (cpu=0x55f2a850cda0) at 
> /home/zippy/work/qemu/qemu.git/accel/kvm/kvm-all.c:1941
> #21 0x000055f2a708d455 in qemu_kvm_cpu_thread_fn (arg=0x55f2a850cda0) at 
> /home/zippy/work/qemu/qemu.git/cpus.c:1196
> #22 0x000055f2a75ce829 in qemu_thread_start (args=0x55f2a852e260) at 
> util/qemu-thread-posix.c:504
> #23 0x00007f7a0a375887 in start_thread (arg=0x7f78fa1f4700) at 
> pthread_create.c:456
> #24 0x00007f7a0a0b7d4f in clone () at 
> ../sysdeps/unix/sysv/linux/x86_64/clone.S:97
> __GI_raise 7 # up
> #1  0x00007f7a09ff5c1a in __GI_abort () at abort.c:89
> 89            raise (SIGABRT);
> __GI_abort 7 #
> #2  0x000055f2a70bce9d in kvm_io_ioeventfd_del (listener=0x55f2a7d19200 
> <kvm_io_listener>, section=0x7f78fa1f10f0, match_data=true, data=0, 
> e=0x7f79fc81f080) at /home/zippy/work/qemu/qemu.git/accel/kvm/kvm-all.c:852
> 852             abort();
> kvm_io_ioeventfd_del 7 # p *e
> $1 = {rfd = -1, wfd = -1, cleanup = 0x0}
>
>
> I think the problem is that event_notifier_cleanup() is called before
> kvm_io_ioeventfd_del() and both are called over the same EventNotifier.
> This is supported by another data I was able to collect in a different
> run:
>
> Thread 4 "CPU 0/KVM" hit Breakpoint 1, event_notifier_cleanup 
> (e=0x7fc6ac26f080) at util/event_notifier-posix.c:83
> 83          if (e->rfd != e->wfd) {
> event_notifier_cleanup 4 # p *e
> $1 = {rfd = 48, wfd = 48, cleanup = 0x55c91c76ac87 
> <virtio_bus_cleanup_event_notifier>}
> event_notifier_cleanup 4 # c
> Continuing.
>
> Thread 4 "CPU 0/KVM" received signal SIGABRT, Aborted.
> __GI_raise (address@hidden) at ../sysdeps/unix/sysv/linux/raise.c:51
> 51      }
> __GI_raise 4 # up
> #1  0x00007fc6b9a45c1a in __GI_abort () at abort.c:89
> 89            raise (SIGABRT);
> __GI_abort 4 #
> #2  0x000055c91c40ee9d in kvm_io_ioeventfd_del (listener=0x55c91d06b200 
> <kvm_io_listener>, section=0x7fc5ab1fa0f0, match_data=true, data=0, 
> e=0x7fc6ac26f080) at /home/zippy/work/qemu/qemu.git/accel/kvm/kvm-all.c:852
> 852             abort();
> kvm_io_ioeventfd_del 4 # p *e
> $2 = {rfd = -1, wfd = -1, cleanup = 0x0}
>
>
> So I drafted dirty workaround:
>
> index 071f4f57c0..bf15344533 100644
> --- i/accel/kvm/kvm-all.c
> +++ w/accel/kvm/kvm-all.c
> @@ -845,6 +845,9 @@ static void kvm_io_ioeventfd_del(MemoryListener *listener,
>      int fd = event_notifier_get_fd(e);
>      int r;
>
> +    if (fd == -1)
> +        return;
> +
>      r = kvm_set_ioeventfd_pio(fd, section->offset_within_address_space,
>                                data, false, int128_get64(section->size),
>                                match_data);
>
> But this time I'm getting following error:
>
> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>
> __GI_raise 7 # up
> #1  0x00007f3ccd579c1a in __GI_abort () at abort.c:89
> 89            raise (SIGABRT);
> __GI_abort 7 #
> #2  0x000055c68ac84e08 in kvm_io_ioeventfd_add (listener=0x55c68b8e1200 
> <kvm_io_listener>, section=0x7f3bbd7f1090, match_data=true, data=0, 
> e=0x7f3bbcf8d080) at /home/zippy/work/qemu/qemu.git/accel/kvm/kvm-all.c:835
> 835             abort();
> kvm_io_ioeventfd_add 7 # p *e
> $1 = {rfd = 48, wfd = 48, cleanup = 0x0}
>

I was about to report the same error when booting an uefi guest, I
bisected down to the same first bad commit.

Thanks Michal

-- 
Marc-André Lureau