Re: [PATCH] mips/kvm: Fix ABI for compatibility with 64-bit guests.

[David Daney <ddaney.cavm@gmail.com>]

On 05/06/2013 04:11 PM, Sanjay Lal wrote:
>
> On May 6, 2013, at 3:39 PM, David Daney wrote:
>
>>
>> /* for KVM_GET_REGS and KVM_SET_REGS */
>> +/*
>> + * If Config[AT] is zero (32-bit CPU), the register contents are
>> + * stored in the lower 32-bits of the struct kvm_regs fields and sign
>> + * extended to 64-bits.
>> + */
>> struct kvm_regs {
>> -	__u32 gprs[32];
>> -	__u32 hi;
>> -	__u32 lo;
>> -	__u32 pc;
>> +	/* out (KVM_GET_REGS) / in (KVM_SET_REGS) */
>> +	__u64 gpr[32];
>> +	__u64 hi, lo;
>> +	__u64 pc;
>> +};
>>
>> -	__u32 cp0reg[N_MIPS_COPROC_REGS][N_MIPS_COPROC_SEL];
>
> Hi David, I'll try out the diff with QEMU and confirm that it works as expected. Could you just leave the GPR field in kvm_regs as "gprs". Its a minor change but avoids diffs that just replace "gprs" with "gpr".
>

Well, there were two changes with respect to 'gprs' vs. 'gpr'.

The change you show above only results in a small handful of diff lines.

My argument for the change is that it will be part of a public ABI, and 
should be short and concise, so I changed it to 'gpr'.

I also changed the field with the same name in struct kvm_vcpu_arch to 
match, which causes the changes in asm-offsets.c and quite a few other 
places as well.  One could argue that this one was gratuitous, but I 
thought it would be nice for them to match.  Since it is an internal 
implementation detail, it is not that important, so I could revert this 
part if there are strong objections.

David Daney