Re: [PATCH v4 4/8] arm64: Add sysreg header generation scripting

[Mark Rutland <mark.rutland@arm.com>]

On Thu, Apr 21, 2022 at 03:50:41PM +0100, Mark Brown wrote:
> On Thu, Apr 21, 2022 at 03:16:52PM +0100, Mark Rutland wrote:
> > On Thu, Apr 21, 2022 at 02:00:17PM +0100, Mark Brown wrote:
> 
> > > > > The script requires that all bits in the register be specified and that
> > > > > there be no overlapping fields. This helps the script spot errors in the
> > > > > input but means that the few registers which change layout at runtime
> > > > > depending on things like virtualisation settings will need some manual
> > > > > handling. No actual register conversions are done here but a header for
> > > > > the register data with some documention of the format is provided.
> 
> > > > It would be good to see an example of how we'd handle one of those, in case
> > > > that means we need to play around with naming or structure of the definitions a
> > > > bit.
> 
> > > My thinking here was that we might not want to handle those registers
> > > through the automated stuff at all.  I haven't yet come up with
> > > something that seems tasteful and viable for them, if I had a firm idea
> > > for what that should look like I'd probably have implemented it.
> 
> > Sure, and I'm not expecting that we automate all of that, just that we
> > have an idea of how the manual bits would work with the automatic bits.
> > If the odd cases looks simple enough, we might be able to get away with
> > a couple of small additions to the scripting.
> 
> > For example, for ESR_EL{1,2,3} today we define ESR_ELx_field
> > definitions rather than duplicate ESR_EL1_field / ESR_EL2_field /
> > ESR_EL3_field definitions. If the scripting has a mechanism to handle
> > that, then that might be good enough for the other odd cases.
> 
> Ah, that's a separate issue to the registers which change layout which
> was what was being mentioned above.

Sure; what I was getting at is that the same mechanism might help there
too -- see below.

> > For example, I think we could do something like:
> 
> > # Define a set of fields without a specific register encoding, using the
> > # name `ESR_ELx`
> > SysregFields	ESR_ELx
> > 	Res0	63:37
> > 	Field	36:32	ISS2
> > 	Field	31:26	EC
> > 	Field	25	IL
> > 	Field	24:0	ISS
> > EndSysregFields
> 
> Yes, I'd been thinking of something like this - it seemed an obvious
> enough extension.
> 
> > # This could instead be SysregEncoding
> > Sysreg	ESR_EL1		3	0	5	2	0
> > 	Comment		"See ESR_ELx for fields"
> > 	# Don't create any field definitions for this reg, and don't
> > 	# bother with the associated sanity checks
> > 	NoFields	
> > EndSysreg
> 
> I think we're going to end up wanting to still generate the numbered
> versions for use in macros so we should probably have that comment and
> NoFields be a SharedLayout (or whatever bikeshedded name) statement.

That's fair. My point was just that we *don't* do that today for ESR_ELx
and having this might simplify the conversion and/or avoid repetition.

Naming wise I'm fine with NoFields/SharedLayout/WhateverYouWant, and
structurally I'm fine with doing other things, I just think we need to
have an idea of what our get-out-of-jail-free-card looks like.

> Until we need it that can just be equivalent to a comment, ready to kick
> in once someone needs it.  I can update the existing (trivial but meh)
> TTBRn register conversion to do that.

FWIW, I'm happy either way (i.e. TTBRn to staty as-is, or be converted),
given the duplication is trivial.

> > ... and the `SysregFields` `NoFields`, and `Comment` mechanisms might be
> > good enough to cover the other odd cases we have (e.g. aliased
> > GIC registers, or different "views" for the same register).
> 
> > Does that make sense, or have I misunderstood the point you were making?
> 
> I was talking about a completely different issue, things like CPTR_EL2
> where the register layout changes depending on some runtime
> configuration.  You'd need two separate register layouts within a single
> Sysreg which isn't too bad until you get to the point of having to name
> things like RES0/RES1 and any name collisons in fields (I don't *think*
> we have name collisons...).

I understood that was a distinct problem; what I was suggesting is the
same mechanism might help there, as e.g. it would allow us to do:

	# Creates CPTR_EL2__E2H1_RES0, etc
	SysregFields	CPTR_EL2__E2H1
		Comment	"CPTR_EL2 layout when HCR_EL2.E2H == 1"
		...
	EndSysregFields

	# Creates CPTR_EL2__E2H0_RES0, etc
	SysregFields	CPTR_EL2__E2H0
		Comment	"CPTR_EL2 layout when HCR_EL2.E2H == 0"
		...
	EndSysregFields

	# Doesn't create CPTR_EL2_RES0, etc
	Sysreg	CPTR_EL2	3	0	1	0	2
		Comment "See CPTR_EL2__E2H1_* and CPTR_EL2__E2H0_*"
		NoFields
	EndSysreg


... or where we have two names for the same register encoding (which is
really silly, but does exist):

	Sysreg	ICV_EOIR0_EL1	3	0	12	8	1
		Comment	"See ICx_EOIR0_EL1_*"
		NoFields
	EndSysreg

	Sysreg	ICH_EOIR0_EL1	3	0	12	8	1
		Comment	"See ICx_EOIR0_EL1_*"
		NoFields
	EndSysreg

	SysregFields	ICx_EOIR0_EL1
		Comment	"Shared layout for ICV_EOIR0_EL1 and ICH_EOIR0_EL1"
		Res0	63:24
		Field	23:0	INTID
	EndSysregFields

I can believe there are problems with that, and/or maybe it's too ugly.

Thanks,
Mark.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel