Re: [PATCH v7 12/14] arm/cpu: Add sysreg generation scripts

[Cornelia Huck <cohuck@redhat.com>]

On Thu, Jun 12 2025, Peter Maydell <peter.maydell@linaro.org> wrote:

> On Thu, 15 May 2025 at 16:40, Cornelia Huck <cohuck@redhat.com> wrote:
>>
>> From: Eric Auger <eric.auger@redhat.com>
>>
>> Introduce scripts that automate the generation of system register
>> definitions from a given linux source tree arch/arm64/tools/sysreg.
>>
>> Invocation of
>> ./update-aarch64-sysreg-code.sh $PATH_TO_LINUX_SOURCE_TREE
>> in scripts directory generates target/arm/cpu-sysregs.h.inc
>> containing defines for all system registers.
>>
>> [CH: update to handle current kernel sysregs structure, and to emit
>>      the re-worked register structures; cpu properties will be added
>>      later]
>> Reviewed-by: Sebastian Ott <sebott@redhat.com>
>> Signed-off-by: Eric Auger <eric.auger@redhat.com>
>> Signed-off-by: Cornelia Huck <cohuck@redhat.com>
>> ---
>
>> diff --git a/scripts/gen-cpu-sysregs-header.awk b/scripts/gen-cpu-sysregs-header.awk
>> new file mode 100755
>> index 000000000000..452e51035d52
>> --- /dev/null
>> +++ b/scripts/gen-cpu-sysregs-header.awk
>
> This should probably have 'arm' in the filename somewhere, unless
> it's really architecture-agnostic, which I don't think it is.

Nod.

>
>> @@ -0,0 +1,35 @@
>> +#!/bin/awk -f
>
> Does this really need to be in awk? Our standard scripting
> language is python. This isn't run at build time so there's
> no "extra dependency" issue, but we currently use awk only
> minimally and almost always for trivial stuff.

Much of this originally was inspired by the equivalent tooling in Linux
that creates definitions from the same sysreg file. (There's another,
more complicated awk script in the cpu model series that creates
definitions for the individual fields.) I can try to rewrite that in
python, but I think awk is actually not a bad tool to use here.

>
>> +# SPDX-License-Identifier: GPL-2.0-or-later
>> +# gen-cpu-sysregs-header.awk: arm64 sysreg header include generator
>> +#
>> +# Usage: awk -f gen-cpu-sysregs-header.awk $LINUX_PATH/arch/arm64/tools/sysreg
>> +
>> +BEGIN {
>> +    print ""
>> +} END {
>> +    print ""
>> +}
>> +
>> +# skip blank lines and comment lines
>> +/^$/ { next }
>> +/^[\t ]*#/ { next }
>> +
>> +/^Sysreg\t/ || /^Sysreg /{
>> +
>> +       reg = $2
>> +       op0 = $3
>> +       op1 = $4
>> +       crn = $5
>> +       crm = $6
>> +       op2 = $7
>> +
>> +       if (op0 == 3 && (op1>=0 && op1<=3) && crn==0 && (crm>=0 && crm<=7) && (op2>=0 && op2<=7)) {
>> +           print "DEF("reg", "op0", "op1", "crn", "crm", "op2")"
>> +       }
>> +       next
>> +}
>> +
>> +{
>> +       /* skip all other lines */
>> +       next
>> +}
>
> Generated source files should have a comment at the top
> noting that they're autogenerated and how to regenerate
> them, please.

ok

>
>> diff --git a/scripts/update-aarch64-sysreg-code.sh b/scripts/update-aarch64-sysreg-code.sh
>> new file mode 100755
>> index 000000000000..e34538887dd8
>> --- /dev/null
>> +++ b/scripts/update-aarch64-sysreg-code.sh
>> @@ -0,0 +1,26 @@
>> +#!/bin/sh -e
>> +#
>> +# SPDX-License-Identifier: GPL-2.0-or-later
>> +# Update target/arm/cpu-sysregs.h
>> +# from a linux source tree (arch/arm64/tools/sysreg)
>> +#
>> +# Copyright Red Hat, Inc. 2024
>> +#
>> +# Authors:
>> +#          Eric Auger <eric.auger@redhat.com>
>> +#
>> +
>> +linux="$1"
>> +output="$PWD"
>
> We seem to set this but then not use it.

ok

>
>> +
>> +if [ -z "$linux" ] || ! [ -d "$linux" ]; then
>> +    cat << EOF
>> +usage: update-aarch64-sysreg-code.sh LINUX_PATH
>> +
>> +LINUX_PATH      Linux kernel directory to obtain the headers from
>> +EOF
>> +    exit 1
>> +fi
>> +
>> +awk -f gen-cpu-sysregs-header.awk \
>> +    $linux/arch/arm64/tools/sysreg > ../target/arm/cpu-sysregs.h.inc
>
> This looks like it is assuming that the user runs the script
> from within the scripts/ directory ?
>
> Following the same command line conventions as our existing
> scripts/update-linux-headers.sh is probably the least
> surprising thing.

Ok, will do.