Re: [PATCH] whpx: Added support for saving/restoring VM state

[Paolo Bonzini]

On 5/16/22 20:44, Ivan Shcherbakov wrote:
> Passing it to x86_cpu_xrstor_all_areas()/x86_cpu_xsave_all_areas() changed the 
> following values:
>
> 0x0000001C: ff ff -> 00 00
> 0x00000208: 07 -> 00
> 0x0000020F: 80 -> 00

0x1C-0x1F is MXCSR_MASK.  There's already a field in the x86 CPUState, 
but it was forgotten in 
x86_cpu_xsave_all_areas()/x86_cpu_xrstor_all_areas().  The field should 
also be initialized to 0xffff in the CPU reset function.

0x208...0x20F is XCOMP_BV and bit 63 in there is indeed signaling 
compacted format.  First of all I'd start with your patch and hack it to 
check if Hyper-V accepts zero at 0x208..0x20F; in this specific case of 
0x208...0x20F have all low consecutive bits set plus bit 63 set, it's 
fine to do just that.  If so, x86_cpu_xrstor_all_areas() needs no 
support for compacted format.  I would be somewhat surprised if Hyper-V 
needs support in XRSTOR too.

For XSAVE, the algorithm to compute the offset (instead of just using 
x->offset) is given in the Intel manual:

If XCOMP_BV[i] = 0, state component i is not in the XSAVE area at all.

If XCOMP_BV[i] = 1, state component i is located at a byte offset  from 
the base address of the XSAVE area, which is determined by the following 
steps:

- If i is the first bit set in bits 62:2 of the XCOMP_BV, state 
component i starts at offset 576

- Otherwise, take CPUID[EAX=0DH,ECX=i].ECX[1]:

  - If it is 0, state component i starts right after the preceding state
    component whose bit is set in XCOMP_BV (where the size of component
    j is enumerated in CPUID[EAX=0DH,ECX=j].EAX).

  - If it is 1, state component i starts after the preceding state
    component whose bit is set in XCOMP_BV, but on a 64-byte aligned
    offset relative to the beginning of the XSAVE area.

Paolo