Re: [Qemu-devel] [PATCH v2 07/12] i2c:pm_smbus: Fix state transfer

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

* Corey Minyard (minyard@acm.org) wrote:
> On 11/26/18 12:24 PM, Corey Minyard wrote:
> > On 11/26/18 11:20 AM, Dr. David Alan Gilbert wrote:
> > > * minyard@acm.org (minyard@acm.org) wrote:
> > > > From: Corey Minyard <cminyard@mvista.com>
> > > > 
> > > > Transfer the state information for the SMBus registers and
> > > > internal data so it will work on a VM transfer.
> > > > 
> > > > Signed-off-by: Corey Minyard <cminyard@mvista.com>
> > > > Cc: Michael S. Tsirkin <mst@redhat.com>
> > > > Cc: Paolo Bonzini <pbonzini@redhat.com>
> > > > Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
> > > > ---
> > > >   hw/acpi/piix4.c           |  3 ++-
> > > >   hw/i2c/pm_smbus.c         | 31 +++++++++++++++++++++++++++++++
> > > >   hw/i2c/smbus_ich9.c       |  6 ++++--
> > > >   include/hw/i2c/pm_smbus.h |  2 ++
> > > >   4 files changed, 39 insertions(+), 3 deletions(-)
> > > > 
> > > > diff --git a/hw/acpi/piix4.c b/hw/acpi/piix4.c
> > > > index e330f24c71..313305f5a0 100644
> > > > --- a/hw/acpi/piix4.c
> > > > +++ b/hw/acpi/piix4.c
> > > > @@ -309,7 +309,7 @@ static const VMStateDescription
> > > > vmstate_cpuhp_state = {
> > > >    */
> > > >   static const VMStateDescription vmstate_acpi = {
> > > >       .name = "piix4_pm",
> > > > -    .version_id = 3,
> > > > +    .version_id = 4,
> > > OK, so do we need to bump this version ?  Ideally you'd keep the version
> > > as is and let the needed do the work.
> > 
> > 
> > Got it, that makes sense.  Same for the comments below, I'll get all
> > those.
> > 
> Well, maybe not.  the .needed function is only called on the save side, it
> is
> not called on the load side  So a 2.12 to 3.0 transfer fails.  So I've
> exported
> the migration needed function and I'm using the VMSTATE_STRUCT_TEST()
> function to transfer it in each case.  With that I can migrate from 2.12 to
> 3.0 and back without issue.

Ah OK, that's an interesting observation; I hadn't realised it wasn't
symmetric like that, but I can kind of see why thinking about how the
code got that way.

Dave

> -corey
> 
> 
> > Thanks,
> > 
> > -corey
> > 
> > 
> > > >       .minimum_version_id = 3,
> > > >       .minimum_version_id_old = 1,
> > > >       .load_state_old = acpi_load_old,
> > > > @@ -320,6 +320,7 @@ static const VMStateDescription vmstate_acpi = {
> > > >           VMSTATE_UINT16(ar.pm1.evt.en, PIIX4PMState),
> > > >           VMSTATE_UINT16(ar.pm1.cnt.cnt, PIIX4PMState),
> > > >           VMSTATE_STRUCT(apm, PIIX4PMState, 0, vmstate_apm, APMState),
> > > > +        VMSTATE_STRUCT(smb, PIIX4PMState, 4, pmsmb_vmstate, PMSMBus),
> > > >           VMSTATE_TIMER_PTR(ar.tmr.timer, PIIX4PMState),
> > > >           VMSTATE_INT64(ar.tmr.overflow_time, PIIX4PMState),
> > > >           VMSTATE_STRUCT(ar.gpe, PIIX4PMState, 2, vmstate_gpe,
> > > > ACPIGPE),
> > > > diff --git a/hw/i2c/pm_smbus.c b/hw/i2c/pm_smbus.c
> > > > index 8793113c25..75907e1c22 100644
> > > > --- a/hw/i2c/pm_smbus.c
> > > > +++ b/hw/i2c/pm_smbus.c
> > > > @@ -19,6 +19,7 @@
> > > >    */
> > > >   #include "qemu/osdep.h"
> > > >   #include "hw/hw.h"
> > > > +#include "hw/boards.h"
> > > >   #include "hw/i2c/pm_smbus.h"
> > > >   #include "hw/i2c/smbus_master.h"
> > > >   @@ -450,6 +451,36 @@ static const MemoryRegionOps pm_smbus_ops = {
> > > >       .endianness = DEVICE_LITTLE_ENDIAN,
> > > >   };
> > > >   +static bool pm_smbus_vmstate_needed(void *opaque)
> > > > +{
> > > > +    MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
> > > > +
> > > > +    return !mc->smbus_no_migration_support;
> > > > +}
> > > OK
> > > 
> > > > +const VMStateDescription pmsmb_vmstate = {
> > > > +    .name = "pmsmb",
> > > > +    .version_id = 1,
> > > > +    .minimum_version_id = 1,
> > > > +    .needed = pm_smbus_vmstate_needed,
> > > > +    .fields = (VMStateField[]) {
> > > > +        VMSTATE_UINT8(smb_stat, PMSMBus),
> > > > +        VMSTATE_UINT8(smb_ctl, PMSMBus),
> > > > +        VMSTATE_UINT8(smb_cmd, PMSMBus),
> > > > +        VMSTATE_UINT8(smb_addr, PMSMBus),
> > > > +        VMSTATE_UINT8(smb_data0, PMSMBus),
> > > > +        VMSTATE_UINT8(smb_data1, PMSMBus),
> > > > +        VMSTATE_UINT32(smb_index, PMSMBus),
> > > > +        VMSTATE_UINT8_ARRAY(smb_data, PMSMBus, PM_SMBUS_MAX_MSG_SIZE),
> > > > +        VMSTATE_UINT8(smb_auxctl, PMSMBus),
> > > > +        VMSTATE_BOOL(i2c_enable, PMSMBus),
> > > > +        VMSTATE_BOOL(op_done, PMSMBus),
> > > > +        VMSTATE_BOOL(in_i2c_block_read, PMSMBus),
> > > > +        VMSTATE_BOOL(start_transaction_on_status_read, PMSMBus),
> > > > +        VMSTATE_END_OF_LIST()
> > > > +    }
> > > > +};
> > > > +
> > > >   void pm_smbus_init(DeviceState *parent, PMSMBus *smb, bool
> > > > force_aux_blk)
> > > >   {
> > > >       smb->op_done = true;
> > > > diff --git a/hw/i2c/smbus_ich9.c b/hw/i2c/smbus_ich9.c
> > > > index e6d8d28194..c9b7482a54 100644
> > > > --- a/hw/i2c/smbus_ich9.c
> > > > +++ b/hw/i2c/smbus_ich9.c
> > > > @@ -45,10 +45,12 @@ typedef struct ICH9SMBState {
> > > >     static const VMStateDescription vmstate_ich9_smbus = {
> > > >       .name = "ich9_smb",
> > > > -    .version_id = 1,
> > > > +    .version_id = 2,
> > > Again, do we need to bump this?
> > > 
> > > >       .minimum_version_id = 1,
> > > >       .fields = (VMStateField[]) {
> > > > -        VMSTATE_PCI_DEVICE(dev, struct ICH9SMBState),
> > > > +        VMSTATE_PCI_DEVICE(dev, ICH9SMBState),
> > > > +        VMSTATE_BOOL_V(irq_enabled, ICH9SMBState, 2),
> > > Can we make this a VMSTATE_BOOL_TEST (see VMSTATE_UINT64_TEST for the
> > > pattern) tied to the same .neede, and again we don't need to bump the
> > > version.
> > > 
> > > > +        VMSTATE_STRUCT(smb, ICH9SMBState, 2, pmsmb_vmstate, PMSMBus),
> > > So that's taken care of by the .needed.
> > > 
> > > Dave
> > > 
> > > >           VMSTATE_END_OF_LIST()
> > > >       }
> > > >   };
> > > > diff --git a/include/hw/i2c/pm_smbus.h b/include/hw/i2c/pm_smbus.h
> > > > index 7bcca97672..5075fc64fa 100644
> > > > --- a/include/hw/i2c/pm_smbus.h
> > > > +++ b/include/hw/i2c/pm_smbus.h
> > > > @@ -43,4 +43,6 @@ typedef struct PMSMBus {
> > > >     void pm_smbus_init(DeviceState *parent, PMSMBus *smb, bool
> > > > force_aux_blk);
> > > >   +extern const VMStateDescription pmsmb_vmstate;
> > > > +
> > > >   #endif /* PM_SMBUS_H */
> > > > -- 
> > > > 2.17.1
> > > > 
> > > -- 
> > > Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK
> > 
> > 
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK