Re: [patch v2 2/4] x86, ptrace: regset extensions to support xstate

[Oleg Nesterov <oleg@redhat.com>]

On 02/09, Suresh Siddha wrote:
>
> +int xstateregs_get(struct task_struct *target, const struct user_regset *regset,
> +		unsigned int pos, unsigned int count,
> +		void *kbuf, void __user *ubuf)
> +{
> +	int ret;
> +	int size = regset->n * regset->size;
> +	struct xsave_hdr_struct *xsave_hdr =
> +				&target->thread.xstate->xsave.xsave_hdr;
> +
> +	if (!cpu_has_xsave)
> +		return -ENODEV;
> +
> +	ret = init_fpu(target);
> +	if (ret)
> +		return ret;
> +
> +	/*
> +	 * First copy the fxsave bytes 0..463
> +	 */
> +	ret = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
> +				  &target->thread.xstate->xsave, 0,
> +				  offsetof(struct i387_fxsave_struct,
> +					   sw_reserved));
> +	if (!ret)
> +		/*
> +		 * Copy the 48bytes defined by software
> +		 */
> +		ret = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
> +					  xstate_fx_sw_bytes,
> +					  offsetof(struct i387_fxsave_struct,
> +						   sw_reserved),
> +					  offsetof(struct xsave_struct,
> +						   xsave_hdr));

Hmm. Suresh, could you confirm these offsetof's are correct?

We are copying xstate_fx_sw_bytes array which is u64[6], but
start_pos == sizeof(i387_fxsave_struct) - padding ?

While we are here. Perhaps it would be more symmetrical to do

	ret = user_regset_copyout();
	if (ret)
		return ret;

instead of "if (!ret)" which needs a tab, but this is up to you.

> +	if (!ret)
> +		/*
> +		 * Copy the rest of xstate memory layout
> +		 */
> +		ret = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
> +					  xsave_hdr,

Another very minor nit, this is the only user of xsave_hdr, but this
var was defined at the top. Perhaps it would be a bit more clean to
have

	struct xsave_struct *xsave = target->thread.xstate->xsave;

and use it in 1st and 3rd copyout?

Speaking about the first user_regset_copyout(), perhaps xsave->i387
would be more clear than xsave?

Oleg.