Re: [PATCH 2/2] KVM: Document KVM_SET_TSS_ADDR

[Pekka Enberg <penberg@cs.helsinki.fi>]

Avi Kivity kirjoitti:
> On 03/25/2010 12:36 PM, Pekka Enberg wrote:
>>> +4.35 KVM_SET_TSS_ADDR
>>> +
>>> +Capability: KVM_CAP_SET_TSS_ADDR
>>> +Architectures: x86
>>> +Type: vm ioctl
>>> +Parameters: unsigned long tss_address (in)
>>> +Returns: 0 on success, -1 on error
>>> +
>>> +This ioctl defines the physical address of a three-page region in 
>>> the guest
>>> +physical address space.  The region must be within the first 4GB of the
>>> +guest physical address space and must not conflict with any memory slot
>>> +or any mmio address.  The guest may malfunction if it accesses this 
>>> memory
>>> +region.
>>> +
>>> +This ioctl is required on Intel-based hosts.
>>
>> I don't quite understand what it's _used for_ from the above 
>> description. I assume it's about task state segment...?
> 
> It's a quirk in the Intel implementation of hardware virtualization 
> extensions.  You cannot enter guest mode in vmx with the guest cr0.pe 
> cleared (i.e. real mode), so kvm enters the guest in vm86 mode which is 
> fairly similar and tries to massage things so it looks to the guest as 
> if it is running in real mode.  Unfortunately, vm86 mode requires a task 
> state segment in the address space, and there is no way for us to hide 
> it.  kvm doesn't know anything about the guest physical memory map, so 
> it has to rely on userspace to supply an unused region.
> 
> I don't think such a technical description of an implementation detail 
> has a place in the API reference; maybe in internal documentation.

Sure but it would be nice to have something along the lines of "This is 
needed on Intel hardware because of a quirk in the virtualization 
implementation" and maybe point the reader to a more appropriate 
document (internals document, Intel manuals, ...).

			Pekka