[RFC PATCH v2 21/41] arm64/sve: Enable SVE on demand for userspace

[mark.rutland@arm.com (Mark Rutland)]

On Thu, Mar 23, 2017 at 12:07:26PM +0000, Dave Martin wrote:
> On Thu, Mar 23, 2017 at 11:52:27AM +0000, Mark Rutland wrote:
> > On Thu, Mar 23, 2017 at 11:30:16AM +0000, Suzuki K Poulose wrote:
> > > On 23/03/17 11:24, Dave Martin wrote:
> > > >On Wed, Mar 22, 2017 at 04:48:10PM +0000, Mark Rutland wrote:
> > 
> > > >>>+	asm ("mrs %0, cpacr_el1" : "=r" (tmp));
> > > >>>+	asm volatile ("msr cpacr_el1, %0" :: "r" (tmp | (1 << 17)));
> > 
> > > >>Please also use {read,write}_sysreg(), e.g.
> > > >
> > > >TBH, I was confused about the status of these macros at the time I
> > > >wrote this code.
> > > >
> > > >The naming clash with the cpufeature functions is unfortunate.  In my
> > > >head these names all became associated with "do something behind the
> > > >scenes that may or may not really read the underlying system reg".
> > > >
> > > >Would it be reasonable to rename read_system_reg() to something more
> > > >different, like read_kernel_sysreg(), read_system_id(),
> > > >read_sanitised_id_reg(), etc.?
> > > 
> > > I agree. read_system_reg() is not quite obvious name given all the other
> > > similar names. We could go with either read_sanitised_id_reg() or read_system_safe_reg() ?
> > 
> > I think read_sanitised_id_reg() sounds best, since "safe" cound mean a
> > few things, and having "id" in the name makes it clear that it's not a
> > general purpose sysreg accessor.
> 
> OK, I'll write a separate patch proposing that.
> 
> Are we comfortable with _id_ in the name here?
> 
> My logic was that sanitisation makes no sense for anything except
> read-only ID registers, and _system_ sounds too universal.
> 
> There is a sting in the tail here -- I add ZCR as an "id" reg so that we
> can treat the maximum configurable length in the LEN field as if it were
> an ID field.  See patch 38 (apologies Suzuki, missed your CC).
> 
> In fact, ZCR is neither read-only nor an ID register -- but the view of
> it through cpufeatures does have those properties.
> 
> It might be better to rename it, but I considered it an OK compromise.
> Let me know if you have concerns.

FWIW, I'm fine with read_sanitised_id_reg(), even considering the ZCR
case, given we're using it for feature identification information.

At best, we only need a comment over the ZCR field definitions in
arch/arm64/kernel/cpufeature.c, explaining what's going on.

Perhaps read_sanitised_ftr_reg() covers that better? That might align
better with the existing *_ftr_reg() naming.

Thanks,
Mark.