Re: [PATCH v3 2/2] x86/fpu: Split fpu_setup_fpu() in three

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 13/08/2024 3:21 pm, Alejandro Vallejo wrote:
> It was trying to do too many things at once and there was no clear way of
> defining what it was meant to do. This commit splits the function in three.
>
>   1. A function to return the FPU to power-on reset values.
>   2. A function to return the FPU to default values.
>   3. A x87/SSE state loader (equivalent to the old function when it took a 
> data
>      pointer).
>
> While at it, make sure the abridged tag is consistent with the manuals and
> start as 0xFF.
>
> Signed-off-by: Alejandro Vallejo <alejandro.vallejo@xxxxxxxxx>

fpu_setup_fpu() definitely needs splitting, and it was doing many things
before, but...

> diff --git a/xen/arch/x86/include/asm/i387.h b/xen/arch/x86/include/asm/i387.h
> index a783549db991..7a69577de45b 100644
> --- a/xen/arch/x86/include/asm/i387.h
> +++ b/xen/arch/x86/include/asm/i387.h
> @@ -31,10 +31,30 @@ void vcpu_restore_fpu_nonlazy(struct vcpu *v, bool 
> need_stts);
>  void vcpu_restore_fpu_lazy(struct vcpu *v);
>  void vcpu_save_fpu(struct vcpu *v);
>  void save_fpu_enable(void);
> -
>  int vcpu_init_fpu(struct vcpu *v);
> -struct xsave_struct;
> -void vcpu_setup_fpu(struct vcpu *v, struct xsave_struct *xsave_area,
> -                    const void *data, unsigned int fcw_default);
>  void vcpu_destroy_fpu(struct vcpu *v);
> +
> +/*
> + * Restore v's FPU to power-on reset values
> + *
> + * @param v vCPU containing the FPU
> + */
> +void vcpu_reset_fpu(struct vcpu *v);
> +
> +/*
> + * Restore v's FPU to default values
> + *
> + * @param v vCPU containing the FPU
> + */
> +void vcpu_default_fpu(struct vcpu *v);

... whether this is an accurate transform of the logic or not, we
oughtn't to have both.

AFAICT, these two functions differ only in the choice of FCW constant,
and whether FTW gets a nonzero value or not.


The x86 architecture has #RESET and #INIT states (and Xen is especially
bad at these right now).  They're well defined in the SDM/APM;  #RESET
zeroes most things but leaves FCW=0x40 and FTW=0x5555 (for which the
FXSAVE mapping is 0xFF).  #INIT leaves everything unmodified.

[After a very long manual-diving session.  Thankyou Christian]


The 80287's #RESET and F(N)INIT instructions were broadly similar.  They
differed on whether they changed the addressing mode[1] but both cases
set FCW=0x37f and FTW=0x5555.

The 80387 intentionally diverged #RESET and FNINIT, with #RESET setting
FCW=0x40.  This had a side effect of asserting #ERROR, and software is
required to execute FNINIT which sets FCW=0x37f and de-asserts #ERROR

Why?  From the 387's programmers reference:

---%<---
6.2.2  Hardware Recognition of the NPX

The 80386 identifies the type of its coprocessor (80287 or 80387) by
sampling its ERROR# input some time after the falling edge of RESET and
before executing the first ESC instruction. The 80287 keeps its ERROR#
output in inactive state after hardware reset; the 80387 keeps its
ERROR# output in active state after hardware reset. The 80386 records
this difference in the ET bit of control register zero (CR0). The 80386
subsequently uses ET to control its interface with the coprocessor. If
ET is set, it employs the 32-bit protocol of the 80387; if ET is not
set, it employs the 16-bit protocol of the 80287.

<snip>

6.2.5  Initializing the 80387

Initializing the 80387 NPX simply means placing the NPX in a known state
unaffected by any activity performed earlier. A single FNINIT
instruction performs this initialization. All the error masks are set,
all registers are tagged empty, TOP is set to zero, and default rounding
and precision controls are set. Table 6-1 shows the state of the 80387
NPX following FINIT or FNINIT. This state is compatible with that of the
80287 after FINIT or after hardware RESET.
---%<---

The 486 gets even more complicated, but I've been writing this email for
long enough.

So, 0x40 is the correct reset value for the 386/387 and later, hence why
that's what the SDM/APM say these days.


As for Xen.  For HVM guests, setting FCW=0x37f is definitely wrong.

For PV guests, things are ill-defined, but software has been required to
issue FNINIT for 4 decades already (even MiniOS does!) so I find myself
dis-interested in trying to maintain compatibility for coprocessor which
predates Xen by ~15y.

You're already playing with FTW too (vs the old logic), so please drop
vcpu_default_fpu() and just use vcpu_reset_fpu() instead.

As a bonus, it lets you delete yet-more code, but it also needs to come
with a rework of the commit message.  I'd aim for less "split in 3" and
more "delete and write something sane". 


> +
> +/*
> + * Load x87/SSE state into v's FPU
> + *
> + * Overrides the XSAVE header to set the state components to be x87 and SSE.
> + *
> + * @param v    vCPU containing the FPU
> + * @param data 512-octet blob for x87/SSE state
> + */
> +void vcpu_setup_fpu(struct vcpu *v, const void *data);
>  #endif /* __ASM_I386_I387_H */
> diff --git a/xen/arch/x86/include/asm/xstate.h 
> b/xen/arch/x86/include/asm/xstate.h
> index ebeb2a3dcaf9..6144ed6f8551 100644
> --- a/xen/arch/x86/include/asm/xstate.h
> +++ b/xen/arch/x86/include/asm/xstate.h
> @@ -14,6 +14,7 @@
>  
>  #define FCW_DEFAULT               0x037f
>  #define FCW_RESET                 0x0040
> +#define FTW_RESET                 0xFF

This isn't a tag word reset value.  The value for that would be 0x5555.

I'd suggest:

#define FXSAVE_FTW_RESET   0xff /* Abridged Tag Word format */

for want of anything better.  At least this name makes it clear it's
specifically for the FXSAVE format.

~Andrew

[1] the 286, (in)famously couldn't leave protected mode, and neither
could the 287 it seems.