[06/22] aarch64: Add GCS instructions

  From: Szabolcs Nagy <szabolcs.nagy@arm.com>

Add instructions for the Guarded Control Stack extension.

GCSSS1 and GCSSS2 are modelled as a single GCSSS unspec, because they
are always used together in the compiler.

Before GCSPOPM and GCSSS2 an extra "mov xn, 0" is added to clear the
output register, this is needed to get reasonable result when GCS is
disabled, when the instructions are NOPs. Since the instructions are
expected to be used behind runtime feature checks, this is mainly
relevant if GCS can be disabled asynchronously.

The output of GCSPOPM is usually not needed, so a separate gcspopm_xzr
was added to model that. Did not do the same for GCSSS as it is a less
common operation.

The used mnemonics do not depend on updated assembler since these
instructions can be used without new -march setting behind a runtime
check.

Reading the GCSPR is modelled as unspec_volatile so it does not get
reordered wrt the other instructions changing the GCSPR.

gcc/ChangeLog:

	* config/aarch64/aarch64.md (aarch64_load_gcspr): New.
	(aarch64_gcspopm): New.
	(aarch64_gcspopm_xzr): New.
	(aarch64_gcsss): New.
---
 gcc/config/aarch64/aarch64.md | 35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)
  

Yury Khrustalev <yury.khrustalev@arm.com> writes:
> From: Szabolcs Nagy <szabolcs.nagy@arm.com>
>
> Add instructions for the Guarded Control Stack extension.
>
> GCSSS1 and GCSSS2 are modelled as a single GCSSS unspec, because they
> are always used together in the compiler.
>
> Before GCSPOPM and GCSSS2 an extra "mov xn, 0" is added to clear the
> output register, this is needed to get reasonable result when GCS is
> disabled, when the instructions are NOPs. Since the instructions are
> expected to be used behind runtime feature checks, this is mainly
> relevant if GCS can be disabled asynchronously.
>
> The output of GCSPOPM is usually not needed, so a separate gcspopm_xzr
> was added to model that. Did not do the same for GCSSS as it is a less
> common operation.
>
> The used mnemonics do not depend on updated assembler since these
> instructions can be used without new -march setting behind a runtime
> check.
>
> Reading the GCSPR is modelled as unspec_volatile so it does not get
> reordered wrt the other instructions changing the GCSPR.

Sorry to be awkward, but I think we should still use one define_insn
per instruction, with no embedded moves.  E.g.:

>
> gcc/ChangeLog:
>
> 	* config/aarch64/aarch64.md (aarch64_load_gcspr): New.
> 	(aarch64_gcspopm): New.
> 	(aarch64_gcspopm_xzr): New.
> 	(aarch64_gcsss): New.
> ---
>  gcc/config/aarch64/aarch64.md | 35 +++++++++++++++++++++++++++++++++++
>  1 file changed, 35 insertions(+)
>
> diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
> index 43bed0ce10f..e4e11e35b5b 100644
> --- a/gcc/config/aarch64/aarch64.md
> +++ b/gcc/config/aarch64/aarch64.md
> @@ -382,6 +382,9 @@ (define_c_enum "unspecv" [
>      UNSPECV_BTI_J		; Represent BTI j.
>      UNSPECV_BTI_JC		; Represent BTI jc.
>      UNSPECV_CHKFEAT		; Represent CHKFEAT X16.
> +    UNSPECV_GCSPR		; Represent MRS Xn, GCSPR_EL0
> +    UNSPECV_GCSPOPM		; Represent GCSPOPM.
> +    UNSPECV_GCSSS		; Represent GCSSS1 and GCSSS2.
>      UNSPECV_TSTART		; Represent transaction start.
>      UNSPECV_TCOMMIT		; Represent transaction commit.
>      UNSPECV_TCANCEL		; Represent transaction cancel.
> @@ -8321,6 +8324,38 @@ (define_insn "aarch64_chkfeat"
>    "hint\\t40 // chkfeat x16"
>  )
>  
> +;; Guarded Control Stack (GCS) instructions
> +(define_insn "aarch64_load_gcspr"
> +  [(set (match_operand:DI 0 "register_operand" "=r")
> +	(unspec_volatile:DI [(const_int 0)] UNSPECV_GCSPR))]
> +  ""
> +  "mrs\\t%0, s3_3_c2_c5_1 // gcspr_el0"
> +  [(set_attr "type" "mrs")]
> +)
> +
> +(define_insn "aarch64_gcspopm"
> +  [(set (match_operand:DI 0 "register_operand" "=r")
> +	(unspec_volatile:DI [(const_int 0)] UNSPECV_GCSPOPM))]
> +  ""
> +  "mov\\t%0, 0\;sysl\\t%0, #3, c7, c7, #1 // gcspopm"
> +  [(set_attr "length" "8")]
> +)

...this would be:

(define_insn "aarch64_gcspopm"
  [(set (match_operand:DI 0 "register_operand" "=r")
	(unspec_volatile:DI [(match_operand:DI 1 "register_operand" "0")] UNSPECV_GCSPOPM))]
  ""
  "sysl\\t%0, #3, c7, c7, #1 // gcspopm"
)

with the code that emits the instruction first emitting a zeroing
of operand 1.

Thanks,
Richard

> +
> +(define_insn "aarch64_gcspopm_xzr"
> +  [(unspec_volatile [(const_int 0)] UNSPECV_GCSPOPM)]
> +  ""
> +  "sysl\\txzr, #3, c7, c7, #1 // gcspopm"
> +)
> +
> +(define_insn "aarch64_gcsss"
> +  [(set (match_operand:DI 0 "register_operand" "=r")
> +	(unspec_volatile:DI [(match_operand:DI 1 "register_operand" "r")]
> +	  UNSPECV_GCSSS))]
> +  ""
> +  "sys\\t#3, c7, c7, #2, %1 // gcsss1\;mov\\t%0, 0\;sysl\\t%0, #3, c7, c7, #3 // gcsss2"
> +  [(set_attr "length" "12")]
> +)
> +
>  ;; AdvSIMD Stuff
>  (include "aarch64-simd.md")