[lkml]   [2010]   [Nov]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH 0/9] KVM: Make the instruction emulator aware of Nested Virtualization
On 11/25/2010 08:21 PM, Roedel, Joerg wrote:
> On Thu, Nov 25, 2010 at 10:15:43AM -0500, Avi Kivity wrote:
> > On 11/25/2010 01:46 PM, Roedel, Joerg wrote:
> > Eventually the emulator will be used outside kvm. We don't want to tie
> > the two together.
> Does any user outside of KVM care about nested virtualization?


> > All that's needed is to read the svm chapter in the AMD manual; you
> > don't need to understand kvm or out nested svm implementation. On the
> > other hand, some information needs to be encoded in the emulator (the
> > order of the intercept check vs exception check) or we need to duplicate
> > checks. We also do a split decode.
> Is that person also required to read through the 500 pages of VMX
> documentation when nested VMX gets merged?


> > So they get special treatment. Decode bits are for the general case.
> >
> > Let's see:
> >
> > CRx/DRx checks - need group mechanism extension, can use decode bits
> The CRx writes are mostly special because exceptions for validity of the
> values written take precedence over the intercept.

We can have three checks, controlled by the decode bits:

// decode instruction
if ((c->d & SvmMask) == SvmInterceptBefore)
... do intercept check
// do privivilge level checks
if ((c->d & SvmMask) == SvmInterceptAfterPriv)
... do intercept check
// fetch operands
if ((c->d & SvmMask) == SvmInterceptAfterMemory)
... do intercept check

> Implementing these
> checks also requires to put the intercept check into the kvm_set_crX
> functions, which, by themselves, needs to be reworked in an SVM specific
> way for this.

Add a kvm_x86_ops callback for this (vmx as usual is pretty complicated

> > Selective CR0 - special
> Needs to be handled in the write-cr0 path

In the appropriate callback

> Check for a valid address before the intercept check. Thus special too.

See above - we can regularize it by encoding where the check takes place.

> > RDTSC/RDPMC/CPUID - decode bits
> RDTSC and RDPMC check all exceptions before the intercept too.
> > PUSHF/POPF/RSM/IRET/INTn - decode bits, + flag to check before exceptions
> Should work with decode-bits.
> > INVD /HLT/INVLPG/INVLPGA - decode bits
> Exceptions are only caused on cpl> 0 and take precedence over the
> intercept. Should work with decode bits.
> > preempts exceptions)
> VMRUN/VMLOAD/VMSAVE need to check rax for a valid physical address
> before the intercept is taken.

Add an SrcPhys/DstPhys decode, it becomes regular.

> All SVM instructions are not allowed in
> real-mode which needs to be checkd too. The realmode-check may be
> generic but with the address check this is harder. So at least
> VMRUN/VMLOAD/VMSAVE are special too.
> Further the SVM instructions are not implemented in the emulator at all
> (like some other instructions which can be intercepted). Proper
> emulation of these instructions would require new callbacks.


> RDTSCP needs special handling like RDTSC.


> MONITOR is special too because
> it checks all exceptions before the intercept.
> All this can be done, but I doubt the result will look better or is
> better maintainable than the current the solution in this patch-set.

With proper infrastructure I think all the modifications needed will be
the three checks above and the decode bits (assuming the current
crx/drx/pio callbacks are in the right place).

I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

 \ /
  Last update: 2010-11-26 09:37    [from the cache]
©2003-2014 Jasper Spaans. hosted at Digital Ocean