Re: [PATCHv4 3/3] x86/tdx: Handle load_unaligned_zeropad() page-cross to a shared page

[Dave Hansen <dave.hansen@intel.com>]

On 6/14/22 05:01, Kirill A. Shutemov wrote:
> load_unaligned_zeropad() can lead to unwanted loads across page boundaries.
> The unwanted loads are typically harmless. But, they might be made to
> totally unrelated or even unmapped memory. load_unaligned_zeropad()
> relies on exception fixup (#PF, #GP and now #VE) to recover from these
> unwanted loads.
> 
> In TDX guests, the second page can be shared page and VMM may configure
> it to trigger #VE.
> 
> Kernel assumes that #VE on a shared page is MMIO access and tries to
> decode instruction to handle it. In case of load_unaligned_zeropad() it
> may result in confusion as it is not MMIO access.
> 
> Fix it by detecting split page MMIO accesses and fail them.
> load_unaligned_zeropad() will recover using exception fixups.
> 
> The issue was discovered by analysis. It was not triggered during the
> testing.
> 
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> ---
>  arch/x86/coco/tdx/tdx.c | 15 ++++++++++++++-
>  1 file changed, 14 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
> index 7d6d484a6d28..3bcaf2170ede 100644
> --- a/arch/x86/coco/tdx/tdx.c
> +++ b/arch/x86/coco/tdx/tdx.c
> @@ -333,8 +333,8 @@ static bool mmio_write(int size, unsigned long addr, unsigned long val)
>  
>  static int handle_mmio(struct pt_regs *regs, struct ve_info *ve)
>  {
> +	unsigned long *reg, val, vaddr;
>  	char buffer[MAX_INSN_SIZE];
> -	unsigned long *reg, val;
>  	struct insn insn = {};
>  	enum mmio_type mmio;
>  	int size, extend_size;
> @@ -360,6 +360,19 @@ static int handle_mmio(struct pt_regs *regs, struct ve_info *ve)
>  			return -EINVAL;
>  	}
>  
> +	/*
> +	 * Reject EPT violation #VEs that split pages.
> +	 *
> +	 * MMIO accesses suppose to be naturally aligned and therefore never
> +	 * cross a page boundary. Seeing split page accesses indicates a bug
> +	 * or load_unaligned_zeropad() that steps into unmapped shared page.

Isn't this "unmapped" thing a rather superfluous implementation detail?

For the guest, it just needs to know that it *CAN* #VE on access to MMIO
and that it needs to be prepared.  The fact that MMIO is implemented
with TDX shared memory *AND* that "unmapped shared pages" can cause
#VE's seems like too much detail.

Also, is this all precise?  Are literal unmapped shared pages the *ONLY*
thing that a hypervisor can do do case a #VE?  What about, say, reserved
bits being set in a shared EPT entry?

I was thinking a comment like this might be better:

>         /*
>          * Reject EPT violation #VEs that split pages.
>          *
>          * MMIO accesses are supposed to be naturally aligned and therefore
>          * never cross page boundaries. Seeing split page accesses indicates
>          * a bug or a load_unaligned_zeropad() that stepped into an MMIO page.
>          *
>          * load_unaligned_zeropad() will recover using exception fixups.
>          */