qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [very-WIP 1/7] migration: Add VMSTATE_WITH_TMP


From: Dr. David Alan Gilbert
Subject: Re: [Qemu-devel] [very-WIP 1/7] migration: Add VMSTATE_WITH_TMP
Date: Tue, 18 Oct 2016 09:06:24 +0100
User-agent: Mutt/1.7.1 (2016-10-04)

* Jianjun Duan (address@hidden) wrote:
> 
> 
> On 10/17/2016 12:16 PM, Dr. David Alan Gilbert wrote:
> > * Jianjun Duan (address@hidden) wrote:
> >>
> >>
> >> On 10/17/2016 11:52 AM, Dr. David Alan Gilbert wrote:
> >>> * Jianjun Duan (address@hidden) wrote:
> >>>>
> >>>>
> >>>> On 10/16/2016 08:31 PM, David Gibson wrote:
> >>>>> On Tue, Oct 11, 2016 at 06:18:30PM +0100, Dr. David Alan Gilbert (git) 
> >>>>> wrote:
> >>>>>> From: "Dr. David Alan Gilbert" <address@hidden>
> >>>>>>
> >>>>>> VMSTATE_WITH_TMP is for handling structures where some calculation
> >>>>>> or rearrangement of the data needs to be performed before the data
> >>>>>> hits the wire.
> >>>>>> For example,  where the value on the wire is an offset from a
> >>>>>> non-migrated base, but the data in the structure is the actual pointer.
> >>>>>>
> >>>>>> To use it, a temporary type is created and a vmsd used on that type.
> >>>>>> The first element of the type must be 'parent' a pointer back to the
> >>>>>> type of the main structure.  VMSTATE_WITH_TMP takes care of allocating
> >>>>>> and freeing the temporary before running the child vmsd.
> >>>>>>
> >>>>>> The post_load/pre_save on the child vmsd can copy things from the 
> >>>>>> parent
> >>>>>> to the temporary using the parent pointer and do any other calculations
> >>>>>> needed; it can then use normal VMSD entries to do the actual data
> >>>>>> storage without having to fiddle around with qemu_get_*/qemu_put_*
> >>>>>>
> >>>> If customized put/get can do transformation and dumping/loading data
> >>>> to/from the parent structure, you don't have to go through
> >>>> pre_save/post_load, and may get rid of parent pointer.
> >>>
> >>> Yes but I'd rather try and get rid of the customized put/get from
> >>> every device, because then people start using qemu_put/qemu_get in them 
> >>> all.
> >>>
> >> Then customized handling need to happen in pre_save/post_load. I think
> >> you need a way to pass TMP pointer around?
> > 
> > But then why is that better than having the parent pointer?
> > 
> IIUC, from the put_tmp, I didn't see how tmp is filled with data. I
> suppose it is to be filled by pre_save. So tmp pointer needs to find a
> way from inside pre_save to put_tmp. How does it happen?

The tmp.parent pointer is filled by the '*(void **)tmp = pv;' in the put_tmp and
get_tmp.
Only after that is the child vmsd run and it's passed the tmp pointer; it's the
child's pre_save/pre_load that has the chance to do whatever it likes to fill 
the
tmp.

Dave

> Thanks,
> Jianjun
> 
> 
> > Dave
> > 
> >>
> >> Thanks,
> >> Jianjun
> >>> Dave
> >>>
> >>>>
> >>>> Thanks,
> >>>> Jianjun
> >>>>
> >>>>>> Signed-off-by: Dr. David Alan Gilbert <address@hidden>
> >>>>>
> >>>>> The requirement for the parent pointer is a little clunky, but I don't
> >>>>> quickly see a better way, and it is compile-time verified.  As noted
> >>>>> elsewhere I think this is a really useful approach which could allow a
> >>>>> bunch of internal state cleanups while preserving migration.
> >>>>>
> >>>>> Reviewed-by: David Gibson <address@hidden>
> >>>>>
> >>>>>> ---
> >>>>>>  include/migration/vmstate.h | 20 ++++++++++++++++++++
> >>>>>>  migration/vmstate.c         | 38 
> >>>>>> ++++++++++++++++++++++++++++++++++++++
> >>>>>>  2 files changed, 58 insertions(+)
> >>>>>>
> >>>>>> diff --git a/include/migration/vmstate.h b/include/migration/vmstate.h
> >>>>>> index 9500da1..efb0e90 100644
> >>>>>> --- a/include/migration/vmstate.h
> >>>>>> +++ b/include/migration/vmstate.h
> >>>>>> @@ -259,6 +259,7 @@ extern const VMStateInfo vmstate_info_cpudouble;
> >>>>>>  extern const VMStateInfo vmstate_info_timer;
> >>>>>>  extern const VMStateInfo vmstate_info_buffer;
> >>>>>>  extern const VMStateInfo vmstate_info_unused_buffer;
> >>>>>> +extern const VMStateInfo vmstate_info_tmp;
> >>>>>>  extern const VMStateInfo vmstate_info_bitmap;
> >>>>>>  extern const VMStateInfo vmstate_info_qtailq;
> >>>>>>  
> >>>>>> @@ -651,6 +652,25 @@ extern const VMStateInfo vmstate_info_qtailq;
> >>>>>>      .offset     = offsetof(_state, _field),                          \
> >>>>>>  }
> >>>>>>  
> >>>>>> +/* Allocate a temporary of type 'tmp_type', set tmp->parent to _state
> >>>>>> + * and execute the vmsd on the temporary.  Note that we're working 
> >>>>>> with
> >>>>>> + * the whole of _state here, not a field within it.
> >>>>>> + * We compile time check that:
> >>>>>> + *    That _tmp_type contains a 'parent' member that's a pointer to 
> >>>>>> the
> >>>>>> + *        '_state' type
> >>>>>> + *    That the pointer is right at the start of _tmp_type.
> >>>>>> + */
> >>>>>> +#define VMSTATE_WITH_TMP(_state, _tmp_type, _vmsd) {                 \
> >>>>>> +    .name         = "tmp",                                           \
> >>>>>> +    .size         = sizeof(_tmp_type) +                              \
> >>>>>> +                    QEMU_BUILD_BUG_EXPR(offsetof(_tmp_type, parent) 
> >>>>>> != 0) + \
> >>>>>> +                    type_check_pointer(_state,                       \
> >>>>>> +                        typeof_field(_tmp_type, parent)),            \
> >>>>>> +    .vmsd         = &(_vmsd),                                        \
> >>>>>> +    .info         = &vmstate_info_tmp,                               \
> >>>>>> +    .flags        = VMS_LINKED,                                      \
> >>>>>> +}
> >>>>>> +
> >>>>>>  #define VMSTATE_UNUSED_BUFFER(_test, _version, _size) {              \
> >>>>>>      .name         = "unused",                                        \
> >>>>>>      .field_exists = (_test),                                         \
> >>>>>> diff --git a/migration/vmstate.c b/migration/vmstate.c
> >>>>>> index 2157997..f2563c5 100644
> >>>>>> --- a/migration/vmstate.c
> >>>>>> +++ b/migration/vmstate.c
> >>>>>> @@ -925,6 +925,44 @@ const VMStateInfo vmstate_info_unused_buffer = {
> >>>>>>      .put  = put_unused_buffer,
> >>>>>>  };
> >>>>>>  
> >>>>>> +/* vmstate_info_tmp, see VMSTATE_WITH_TMP, the idea is that we 
> >>>>>> allocate
> >>>>>> + * a temporary buffer and the pre_load/pre_save methods in the child 
> >>>>>> vmsd
> >>>>>> + * copy stuff from the parent into the child and do calculations to 
> >>>>>> fill
> >>>>>> + * in fields that don't really exist in the parent but need to be in 
> >>>>>> the
> >>>>>> + * stream.
> >>>>>> + */
> >>>>>> +static int get_tmp(QEMUFile *f, void *pv, size_t size, VMStateField 
> >>>>>> *field)
> >>>>>> +{
> >>>>>> +    int ret;
> >>>>>> +    const VMStateDescription *vmsd = field->vmsd;
> >>>>>> +    int version_id = field->version_id;
> >>>>>> +    void *tmp = g_malloc(size);
> >>>>>> +
> >>>>>> +    /* Writes the parent field which is at the start of the tmp */
> >>>>>> +    *(void **)tmp = pv;
> >>>>>> +    ret = vmstate_load_state(f, vmsd, tmp, version_id);
> >>>>>> +    g_free(tmp);
> >>>>>> +    return ret;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static void put_tmp(QEMUFile *f, void *pv, size_t size, VMStateField 
> >>>>>> *field,
> >>>>>> +                    QJSON *vmdesc)
> >>>>>> +{
> >>>>>> +    const VMStateDescription *vmsd = field->vmsd;
> >>>>>> +    void *tmp = g_malloc(size);
> >>>>>> +
> >>>>>> +    /* Writes the parent field which is at the start of the tmp */
> >>>>>> +    *(void **)tmp = pv;
> >>>>>> +    vmstate_save_state(f, vmsd, tmp, vmdesc);
> >>>>>> +    g_free(tmp);
> >>>>>> +}
> >>>>>> +
> >>>>>> +const VMStateInfo vmstate_info_tmp = {
> >>>>>> +    .name = "tmp",
> >>>>>> +    .get = get_tmp,
> >>>>>> +    .put = put_tmp,
> >>>>>> +};
> >>>>>> +
> >>>>>>  /* bitmaps (as defined by bitmap.h). Note that size here is the size
> >>>>>>   * of the bitmap in bits. The on-the-wire format of a bitmap is 64
> >>>>>>   * bit words with the bits in big endian order. The in-memory format
> >>>>>
> >>>>
> >>> --
> >>> Dr. David Alan Gilbert / address@hidden / Manchester, UK
> >>>
> >>
> > --
> > Dr. David Alan Gilbert / address@hidden / Manchester, UK
> > 
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK



reply via email to

[Prev in Thread] Current Thread [Next in Thread]