Re: [PATCH 2/3] KVM: x86 emulator: move string instruction completion check into separate function

[Avi Kivity <avi@redhat.com>]

  On 08/24/2010 04:20 PM, Gleb Natapov wrote:
>>
>>> +{
>>> +	struct decode_cache *c =&ctxt->decode;
>>> +
>>> +	/* All REP prefixes have the same first termination condition */
>>> +	if (address_mask(c, c->regs[VCPU_REGS_RCX]) == 0)
>>> +		return true;
>> This is checked during the beginning of the instruction, not after
>> completion.  Why is it here?  it will just be duplicated.
>>
> SDM describes REP instruction algorithm this way:
>
> WHILE CountReg ≠ 0
>    DO
>         Service pending interrupts (if any);
>         Execute associated string instruction;
>         CountReg ← (CountReg – 1);
>         IF CountReg = 0
>              THEN exit WHILE loop; FI;
>         IF (Repeat prefix is REPZ or REPE) and (ZF = 0)
>         or (Repeat prefix is REPNZ or REPNE) and (ZF = 1)
>              THEN exit WHILE loop; FI;
>    OD;
>
> So CountReg is checked at the beginning and after each iteration.

The second check is meaningless (and ZF checks should be qualified with 
the actual instruction).

> Practically it will save us one return to a guest and exit back
> to emulator at the end of rep instruction (not a big deal).

Not even that - if we reenter to the beginning of the rep instruction 
the cpu will skip over it without exiting (unless in big real mode with 
eigs=1).

-- 
error compiling committee.c: too many arguments to function