Messages in this thread | | | From | Ingo Molnar <> | Date | Wed, 6 May 2015 06:20:29 +0200 | Subject | Re: [PATCH 198/208] x86/fpu: Document the various fpregs state formats |
| |
* Dave Hansen <dave.hansen@linux.intel.com> wrote:
> On 05/05/2015 10:58 AM, Ingo Molnar wrote: > > +/* > > + * This is our most modern FPU state format, as saved by the XSAVE > > + * and restored by the XRSTOR instructions. > > + * > > + * It consists of a legacy fxregs portion, an xstate header and > > + * subsequent fixed size areas as defined by the xstate header. > > + * Not all CPUs support all the extensions. > > + */ > > struct xregs_state { > > struct fxregs_state i387; > > struct xstate_header header; > > @@ -150,6 +169,13 @@ struct xregs_state { > > /* New processor state extensions will go here. */ > > } __attribute__ ((packed, aligned (64))); > > Fenghua has a "fix" for this, but I think this misses a pretty big point. > > This structure includes only the "legacy" state, followed by the header. > The remainder of the layout here is enumerated in CPUID leaves and can > not be laid out in a structure because we do not know what it looks like > until we run CPUID. > > There is logically a variable length array at the end of this > sucker.
Yes, exactly, that is where we want to go, and this direction is what I tried to cover with this bit of the series:
struct xregs_state { struct fxregs_state i387; struct xstate_header header; u8 __reserved[XSTATE_RESERVE]; } __attribute__ ((packed, aligned (64)));
Note how it's now opaque after the xstate header, because there's no guarantee of what's in that area.
The only 'fixed' aspect of the xstates is the feature bit enumeration:
enum xfeature_bit { XSTATE_BIT_FP, XSTATE_BIT_SSE, XSTATE_BIT_YMM, XSTATE_BIT_BNDREGS, XSTATE_BIT_BNDCSR, XSTATE_BIT_OPMASK, XSTATE_BIT_ZMM_Hi256, XSTATE_BIT_Hi16_ZMM,
XFEATURES_NR_MAX,
Plus with point #4 of the announcement I wanted to signal that I think we should allocate the variable part dynamically:
4)
task->thread.fpu->state got embedded again, as task->thread.fpu.state. This eliminated a lot of awkward late dynamic memory allocation of FPU state and the problematic handling of failures.
Note that while the allocation is static right now, this is a WIP interim state: we can still do dynamic allocation of FPU state, by moving the FPU state last in task_struct and then allocating task_struct accordingly.
I.e. we can put the variable size state array at the end of task_struct, make task_struct size per boot variable and still have essentially a single static allocation for all fundamental task state.
But I first wanted to see people test this series - it's ambitious enough as-is already!
Thanks,
Ingo
| |