On 19.12.14 18:29, Paolo Bonzini wrote:
>
>
> On 19/12/2014 18:16, Alexander Graf wrote:
>>
>>
>> On 19.12.14 17:51, Mark Cave-Ayland wrote:
>>> On 19/12/14 16:35, Peter Maydell wrote:
>>>
>>>> On 19 December 2014 at 16:15, Mark Cave-Ayland
>>>> <mark.cave-ayl...@ilande.co.uk> wrote:
>>>>> It looks like after restoring the CPU state with loadvm env->htab_base
>>>>> is still at 0x0 rather than its pre-migration value of 0x7e000000.
>>>>
>>>> Odd -- there is code that is at least trying to do that:
>>>> machine.c calls ppc_store_sdr1() which should set htab_base/mask
>>>> according to the migrated value of the register...
>>>
>>> Stepping through ppc_store_sdr1() I see the problem is the if() statement:
>>>
>>> if (env->spr[SPR_SDR1] != value) {
>>> ....
>>> }
>>>
>>> It looks like env->spr[SPR_SDR1] has already been set to 0x7e000000
>>> before the call to ppc_store_sdr1() and so as the values are already
>>> equal then the code to setup env->htab_mask and env->htab_base is bypassed.
>>>
>>> The quick fix is to comment out the above if() statement which allows me
>>> to restore my OpenBIOS image successfully with -loadvm but sadly not my
>>> OS image (I guess there are probably a few more state bugs still lying
>>> around). Does this seem the right thing to do or is there a better way?
>>
>> How about something like this instead? Then we at least don't lose the
>> acceleration we get from not invalidating the TLB on identical SDR1 writes.
>>
>>
>> Alex
>>
>>
>> diff --git a/target-ppc/machine.c b/target-ppc/machine.c
>> index c801b82..6f347bf 100644
>> --- a/target-ppc/machine.c
>> +++ b/target-ppc/machine.c
>> @@ -71,6 +71,7 @@ static int cpu_load_old(QEMUFile *f, void *opaque, int
>> version_id)
>> for (i = 0; i < 1024; i++)
>> qemu_get_betls(f, &env->spr[i]);
>> if (!env->external_htab) {
>> + env->spr[SPR_SDR1] = 0;
>> ppc_store_sdr1(env, sdr1);
>> }
>> qemu_get_be32s(f, &env->vscr);
>
> Or even move the tlb_flush and if to the caller, g. It
> is only needed there, and would get in the way if one day PPC were to
> support --disable-tcg (which removes cputlb.c from the executable).
Even better, yes.
Alex
