Re: [PATCH 1/3] efi/x86: simplify 64-bit EFI firmware call wrapper

[Andy Lutomirski <luto@amacapital.net>]

> On Dec 28, 2019, at 12:44 PM, Arvind Sankar <nivedita@alum.mit.edu> wrote:
> 

>> 
>> Does the UEFI spec explicitly state what FP state can be used by the
>> EFI functions?
> 
> For 32-bit, it requires the following for boot services and runtime
> services:
> 
> * Direction flag in EFLAGs clear

Check.

> * 4 KiB, or more, of available stack space

There ought to be.

> * The stack must be 16-byte aligned

Nope. The asm needs to do this for runtime services. The kernel runs with 8-byte stack alignment.

> * Floating-point control word must be initialized to 0x027F (all exceptions masked, double-
>  precision, round-to-nearest)

Ingo, surely kernel_fpu_begin() does this.  But I can’t find any code that does this. Please tell me I’m just missing it because I’m on my phone.

Presumably the code ought to optimize it by saving regs, then checking for unexpected values, then doing FNINIT and STMXCSR if the state is bad.

> * Multimedia-extensions control word (if supported) must be initialized to 0x1F80 (all
>  exceptions masked, round-to-nearest, flush to zero for masked underflow)

Ditto.

> * CR0.EM must be zero
> * CR0.TS must be zero

Check, on modern kernels anyway.

> 
> We don't actually align the stack for 32-bit mode before calling, do we?
> 
> No FP registers are clobbered except the floating point status register.

I don’t believe this for a minute :).

> 
> For 64-bit:
> * Direction flag in EFLAGs clear
> * 4 KiB, or more, of available stack space

Check.

> * The stack must be 16-byte aligned

Nope. We need to align it. Does the EFI runtime service code do this?

> * Floating-point control word must be initialized to 0x037F (all exceptions masked, double-
>  extended-precision, round-to-nearest)
> * Multimedia-extensions control word (if supported) must be initialized to 0x1F80 (all exceptions
>  masked, round-to-nearest, flush to zero for masked underflow)

See above.

> * CR0.EM must be zero
> * CR0.TS must be zero
> 
> xmm0-5 (and the FP status register) may be clobbered, the rest of the FP
> registers are preserved.

Seems reasonable.