Re: [PATCH i-g-t v2] scripts: Add new script for verbatim sync with kernel

[Gustavo Sousa <gustavo.sousa@intel.com>]

Jani Nikula <jani.nikula@intel.com> writes:

> On Fri, 15 May 2026, Gustavo Sousa <gustavo.sousa@intel.com> wrote:
>> Kamil Konieczny <kamil.konieczny@linux.intel.com> writes:
>>
>>> From: Jani Nikula <jani.nikula@intel.com>
>>>
>>> This is a helper script used in cases when a verbatim kernel
>>> file should be copied into igt-dev.
>>>
>>> Cc: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
>>> Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
>>> Cc: Karthik B S <karthik.b.s@intel.com>
>>> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
>>> Signed-off-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>
>>> ---
>>>  scripts/sync_with_kernel.sh | 57 +++++++++++++++++++++++++++++++++++++
>>>  1 file changed, 57 insertions(+)
>>>  create mode 100755 scripts/sync_with_kernel.sh
>>>
>>> diff --git a/scripts/sync_with_kernel.sh b/scripts/sync_with_kernel.sh
>>> new file mode 100755
>>> index 000000000..65753677c
>>> --- /dev/null
>>> +++ b/scripts/sync_with_kernel.sh
>>> @@ -0,0 +1,57 @@
>>> +#!/bin/sh
>>> +# SPDX-License-Identifier: MIT
>>> +#
>>> +# Copyright © 2026 Intel Corporation
>>> +#
>>> +# Author: Jani Nikula <jani.nikula@intel.com>
>>> +#
>>> +# Use this for verbatim copies from kernel into igt-dev-tools
>>> +# Change paths to yours, update the repos, cd to IGT, and run.
>>> +
>>> +SINCE=962601ac4c78
>>
>> So, the plan is for us to keep updating the SINCE value manually? I
>> wonder if we could do better.
>
> It's just a starting point to not go through the entire history. It'll
> find the matching commit after that. It's sufficient to update SINCE
> when it gets too slow.

Makes sense. I guess we could add a comment so that people are aware
that we don't need to update this constantly.

>
>> One idea is to try to infer it from the files in the IGT repository: for
>> each file, we can try to find the latest IGT commit in the format "...:
>> sync ... with kernel commit ...". In case we do not find any, I guess we
>> can fallback to using ..HEAD as the revision range when searching for
>> $last down below.
>>
>>> +KERNEL=$HOME/src/linux
>>> +IGT=$HOME/src/intel-gpu-tools
>>
>> I think the script should not assume specific locations for the
>> repositories. I think:
>>
>> - The location for the kernel repository should be given as an argument;
>> - We should assume that scripts/sync_with_kernel.sh is called from the
>>   IGT repository and we can infer the repository toplevel directory
>>   based on "$(realpath "$(dirname "${BASH_SOURCE[0]}")")" and "git
>>   rev-parse --show-toplevel".
>
> Fair; I originally wrote this for myself.
>
>>> +
>>> +declare -g -A map
>>> +map["drivers/gpu/drm/i915/display/intel_vbt_defs.h"]="tools/intel_vbt_defs.h"
>>> +map["drivers/gpu/drm/i915/display/intel_dsi_vbt_defs.h"]="tools/intel_dsi_vbt_defs.h"
>>> +map["include/drm/intel/pciids.h"]="lib/pciids.h"
>>> +
>>> +cd $KERNEL
>>> +
>>> +for infile in "${!map[@]}"; do
>>> +	outfile=${map[$infile]}
>>> +
>>> +	echo "processing $infile -> $outfile"
>>> +
>>> +	last=
>>> +	for commit in $(git log --reverse --pretty=%h $SINCE..HEAD -- $infile); do
>>> +		if ! git cat-file -e $commit:$infile 2>/dev/null; then
>>> +			continue
>>> +		fi
>>> +		if git show $commit:$infile | diff -q $IGT/$outfile - >/dev/null; then
>>> +			last=$commit
>>> +			break
>>> +		fi
>>> +	done
>
> This is the bit that finds the last matching commit.
>
>>> +
>>> +	if [[ -z "$last" ]]; then
>>> +		echo "last update to $infile not found"
>>> +		continue
>>> +	fi
>>> +
>>> +	for commit in $(git log --reverse --pretty=%h $last..HEAD -- $infile); do
>>
>> Do we really need the intermediate commits when synchronizing?  What
>> would be the value in adding those?
>
> I wrote this at a time when I'd done a lot of refactoring kernel
> side. When each of those kernel changes needed corresponding IGT changes
> *outside* of the files being synced, it was tremendously helpful to
> split it to smaller changes.

Ah. Got it.

>
> If there's no such need, you can trivially manually squash them with git
> rebase -i, but I think it's helpful for the script to do individual
> changes.

Sounds good.

--
Gustavo Sousa

>
>>
>>> +		prefix="${outfile%.*}"
>>> +		base="$(basename $outfile)"
>>> +		# ref=$(git cite $commit)
>>> +		# alias: cite = log -1 --abbrev=12 '--format=%h (\"%s\")'
>>> +		ref=$(git log -1 --abbrev=12 '--format=%h (\"%s\")' $commit)
>>> +		git show $commit:$infile > $IGT/$outfile
>>> +		cd $IGT
>>> +		git commit -as \
>>
>> We can avoid changing directories with git -C "$IGT" commit ...
>>
>> Also, perhaps using "git commit ... -- $IGT/$outfile" is safer than
>> using "git commit -a ...".
>
> Fair.
>
> BR,
> Jani.
>
>>
>> --
>> Gustavo Sousa
>>
>>> +		    -m "${prefix}: sync ${base} with kernel commit $commit" \
>>> +		    -m "Synchronize ${base} with kernel commit:" \
>>> +		    -m "$ref"
>>> +		cd -
>>> +	done
>>> +done
>>> -- 
>>> 2.54.0
>
> -- 
> Jani Nikula, Intel