Re: [PATCH QEMU v25 07/17] vfio: Register SaveVMHandlers for VFIO device

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Alex Williamson (alex.williamson@redhat.com) wrote:
> On Wed, 24 Jun 2020 00:51:06 +0530
> Kirti Wankhede <kwankhede@nvidia.com> wrote:
> 
> > On 6/23/2020 4:20 AM, Alex Williamson wrote:
> > > On Sun, 21 Jun 2020 01:51:16 +0530
> > > Kirti Wankhede <kwankhede@nvidia.com> wrote:
> > >   
> > >> Define flags to be used as delimeter in migration file stream.
> > >> Added .save_setup and .save_cleanup functions. Mapped & unmapped migration
> > >> region from these functions at source during saving or pre-copy phase.
> > >> Set VFIO device state depending on VM's state. During live migration, VM is
> > >> running when .save_setup is called, _SAVING | _RUNNING state is set for VFIO
> > >> device. During save-restore, VM is paused, _SAVING state is set for VFIO device.
> > >>
> > >> Signed-off-by: Kirti Wankhede <kwankhede@nvidia.com>
> > >> Reviewed-by: Neo Jia <cjia@nvidia.com>
> > >> ---
> > >>   hw/vfio/migration.c  | 92 ++++++++++++++++++++++++++++++++++++++++++++++++++++
> > >>   hw/vfio/trace-events |  2 ++
> > >>   2 files changed, 94 insertions(+)
> > >>
> > >> diff --git a/hw/vfio/migration.c b/hw/vfio/migration.c
> > >> index e30bd8768701..133bb5b1b3b2 100644
> > >> --- a/hw/vfio/migration.c
> > >> +++ b/hw/vfio/migration.c
> > >> @@ -8,12 +8,15 @@
> > >>    */
> > >>   
> > >>   #include "qemu/osdep.h"
> > >> +#include "qemu/main-loop.h"
> > >> +#include "qemu/cutils.h"
> > >>   #include <linux/vfio.h>
> > >>   
> > >>   #include "sysemu/runstate.h"
> > >>   #include "hw/vfio/vfio-common.h"
> > >>   #include "cpu.h"
> > >>   #include "migration/migration.h"
> > >> +#include "migration/vmstate.h"
> > >>   #include "migration/qemu-file.h"
> > >>   #include "migration/register.h"
> > >>   #include "migration/blocker.h"
> > >> @@ -24,6 +27,17 @@
> > >>   #include "pci.h"
> > >>   #include "trace.h"
> > >>   
> > >> +/*
> > >> + * Flags used as delimiter:
> > >> + * 0xffffffff => MSB 32-bit all 1s
> > >> + * 0xef10     => emulated (virtual) function IO
> > >> + * 0x0000     => 16-bits reserved for flags
> > >> + */
> > >> +#define VFIO_MIG_FLAG_END_OF_STATE      (0xffffffffef100001ULL)
> > >> +#define VFIO_MIG_FLAG_DEV_CONFIG_STATE  (0xffffffffef100002ULL)
> > >> +#define VFIO_MIG_FLAG_DEV_SETUP_STATE   (0xffffffffef100003ULL)
> > >> +#define VFIO_MIG_FLAG_DEV_DATA_STATE    (0xffffffffef100004ULL)
> > >> +
> > >>   static void vfio_migration_region_exit(VFIODevice *vbasedev)
> > >>   {
> > >>       VFIOMigration *migration = vbasedev->migration;
> > >> @@ -126,6 +140,65 @@ static int vfio_migration_set_state(VFIODevice *vbasedev, uint32_t mask,
> > >>       return 0;
> > >>   }
> > >>   
> > >> +/* ---------------------------------------------------------------------- */
> > >> +
> > >> +static int vfio_save_setup(QEMUFile *f, void *opaque)
> > >> +{
> > >> +    VFIODevice *vbasedev = opaque;
> > >> +    VFIOMigration *migration = vbasedev->migration;
> > >> +    int ret;
> > >> +
> > >> +    trace_vfio_save_setup(vbasedev->name);
> > >> +
> > >> +    qemu_put_be64(f, VFIO_MIG_FLAG_DEV_SETUP_STATE);
> > >> +
> > >> +    if (migration->region.mmaps) {
> > >> +        qemu_mutex_lock_iothread();
> > >> +        ret = vfio_region_mmap(&migration->region);
> > >> +        qemu_mutex_unlock_iothread();
> > >> +        if (ret) {
> > >> +            error_report("%s: Failed to mmap VFIO migration region %d: %s",
> > >> +                         vbasedev->name, migration->region.nr,
> > >> +                         strerror(-ret));
> > >> +            return ret;  
> > > 
> > > OTOH to my previous comments, this shouldn't be fatal, right?  mmaps
> > > are optional anyway so it should be sufficient to push an error report
> > > to explain why this might be slower than normal, but we can still
> > > proceed.
> > >   
> > 
> > Right, defining region to be sparse mmap is optional.
> > migration->region.mmaps is set if vendor driver defines sparse mmapable 
> > regions and VFIO_REGION_INFO_FLAG_MMAP flag is set. If this flag is set 
> > then error on mmap() should be fatal.
> > 
> > If there is not mmapable region, then migration will proceed.
> 
> It's both optional for the vendor to define sparse mmap support (or any
> mmap support) and optional for the user to make use of it.  The user
> can recover from an mmap failure by using read/write accesses.  The
> vendor MUST support this.  It doesn't make sense to worry about
> aborting the VM in replying to comments for 05/17, where it's not clear
> how we proceed, yet intentionally cause a fatal error here when there
> is a very clear path to proceed.
> 
> > >> +        }
> > >> +    }
> > >> +
> > >> +    ret = vfio_migration_set_state(vbasedev, VFIO_DEVICE_STATE_MASK,
> > >> +                                   VFIO_DEVICE_STATE_SAVING);
> > >> +    if (ret) {
> > >> +        error_report("%s: Failed to set state SAVING", vbasedev->name);
> > >> +        return ret;
> > >> +    }  
> > > 
> > > We seem to be lacking support in the callers for detecting if the
> > > device is in an error state.  I'm not sure what our options are
> > > though, maybe only a hw_error().
> > >   
> > 
> > Returning error here fails migration process. And if device is in error 
> > state, any application running inside VM using this device would fail.
> > I think, there is no need to take any special action here by detecting 
> > device error state.
> 
> If QEMU knows a device has failed, it seems like it would make sense to
> stop the VM, otherwise we risk an essentially endless assortment of
> ways that the user might notice the guest isn't behaving normally, some
> maybe even causing the user to lose data.  Thanks,

With GPUs especially though we can get into messy states where only the
GPU is toast;  for example you might get this if your GUI
starts/exits/crashes during a migration - that's a bit too common to
kill a VM that might have useful data on it.

Dave

> Alex
>  
> > >> +
> > >> +    qemu_put_be64(f, VFIO_MIG_FLAG_END_OF_STATE);
> > >> +
> > >> +    ret = qemu_file_get_error(f);
> > >> +    if (ret) {
> > >> +        return ret;
> > >> +    }
> > >> +
> > >> +    return 0;
> > >> +}
> > >> +
> > >> +static void vfio_save_cleanup(void *opaque)
> > >> +{
> > >> +    VFIODevice *vbasedev = opaque;
> > >> +    VFIOMigration *migration = vbasedev->migration;
> > >> +
> > >> +    if (migration->region.mmaps) {
> > >> +        vfio_region_unmap(&migration->region);
> > >> +    }
> > >> +    trace_vfio_save_cleanup(vbasedev->name);
> > >> +}
> > >> +
> > >> +static SaveVMHandlers savevm_vfio_handlers = {
> > >> +    .save_setup = vfio_save_setup,
> > >> +    .save_cleanup = vfio_save_cleanup,
> > >> +};
> > >> +
> > >> +/* ---------------------------------------------------------------------- */
> > >> +
> > >>   static void vfio_vmstate_change(void *opaque, int running, RunState state)
> > >>   {
> > >>       VFIODevice *vbasedev = opaque;
> > >> @@ -180,6 +253,7 @@ static int vfio_migration_init(VFIODevice *vbasedev,
> > >>                                  struct vfio_region_info *info)
> > >>   {
> > >>       int ret;
> > >> +    char id[256] = "";
> > >>   
> > >>       vbasedev->migration = g_new0(VFIOMigration, 1);
> > >>   
> > >> @@ -192,6 +266,24 @@ static int vfio_migration_init(VFIODevice *vbasedev,
> > >>           return ret;
> > >>       }
> > >>   
> > >> +    if (vbasedev->ops->vfio_get_object) {  
> > > 
> > > Nit, vfio_migration_region_init() would have failed already if this were
> > > not available.  Perhaps do the test once at the start of this function
> > > instead?  Thanks,
> > >   
> > 
> > Ok, will do that.
> > 
> > Thanks,
> > Kirti
> > 
> > 
> > > Alex
> > >   
> > >> +        Object *obj = vbasedev->ops->vfio_get_object(vbasedev);
> > >> +
> > >> +        if (obj) {
> > >> +            DeviceState *dev = DEVICE(obj);
> > >> +            char *oid = vmstate_if_get_id(VMSTATE_IF(dev));
> > >> +
> > >> +            if (oid) {
> > >> +                pstrcpy(id, sizeof(id), oid);
> > >> +                pstrcat(id, sizeof(id), "/");
> > >> +                g_free(oid);
> > >> +            }
> > >> +        }
> > >> +    }
> > >> +    pstrcat(id, sizeof(id), "vfio");
> > >> +
> > >> +    register_savevm_live(id, VMSTATE_INSTANCE_ID_ANY, 1, &savevm_vfio_handlers,
> > >> +                         vbasedev);
> > >>       vbasedev->vm_state = qemu_add_vm_change_state_handler(vfio_vmstate_change,
> > >>                                                             vbasedev);
> > >>       vbasedev->migration_state.notify = vfio_migration_state_notifier;
> > >> diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events
> > >> index bd3d47b005cb..86c18def016e 100644
> > >> --- a/hw/vfio/trace-events
> > >> +++ b/hw/vfio/trace-events
> > >> @@ -149,3 +149,5 @@ vfio_migration_probe(const char *name, uint32_t index) " (%s) Region %d"
> > >>   vfio_migration_set_state(const char *name, uint32_t state) " (%s) state %d"
> > >>   vfio_vmstate_change(const char *name, int running, const char *reason, uint32_t dev_state) " (%s) running %d reason %s device state %d"
> > >>   vfio_migration_state_notifier(const char *name, const char *state) " (%s) state %s"
> > >> +vfio_save_setup(const char *name) " (%s)"
> > >> +vfio_save_cleanup(const char *name) " (%s)"  
> > >   
> > 
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK