Re: [PATCH v2 06/13] KVM: x86: save/load state on SMM switch

["Radim Krčmář" <rkrcmar@redhat.com>]

2015-06-04 08:14+0200, Paolo Bonzini:
> On 03/06/2015 21:02, Radim Krčmář wrote:
>>> +	r = kvm_write_guest(vcpu->kvm, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
>> 
>> The state is saved in SMRAM, but we are accessing it using the non-SMM
>> address space ... how did it pass testing?
>> (Restore is using SMM address space, so I'm guessing that the mapping
>>  from QEMU wasn't really utilizing two separate address spaces.)
> 
> At this point of the series there are no separate address spaces yet.
> Patch 10 then changes it everywhere:
> 
> @@ -6558,7 +6558,7 @@ static void process_smi(struct kvm_vcpu *vcpu)

My bad, people using jackhammers at 7am are getting the better of me.

> Why did I order it this way?  Because it is already possible to test
> this code with the default SMBASE of 0x30000, and it is already
> possible to run the full firmware if you hack it not to close SMRAM
> (for this I used q35's high SMRAM).  It is not possible to test the
> code partially if you first add the two address spaces, and only
> implement the world switch second.

The ordering makes sense;  I wanted to point out the early return,
noticed this as well and missed that it was fixed later, sorry.