On 25/01/2023 00:36, Marijn Suijten wrote:
On 2023-01-24 09:55:24, Kuogee Hsieh wrote:
<snip>
This timing engine code is derived from our downstream code directly and
it has been used at many mobile devices by many vendors for many years
already.
On the other words, it had been tested very thorough and works on
dsi/dp/hdmi/dsc/widebus applications.
And the code already in mainline has seen 12 rounds of review, with a
focus on inter-SoC compatibility. Regardless of that, we have processes
to make changes on mainline: formatting changes (when actually making an
improvement) go separate from semantic changes. Bugfixes are clearly
described in individual patches with Fixes: tags. If you really think
the code has to be as proposed in this patch, follow Dmitry's advice and
split this accordingly.
When i brought dsc v1.2 over, I just merged it over and did not consider
too much.
And that is exactly what is wrong with this *entire* series: copying
over downstream code without "considering too much", stomping over
previous review and even reverting bugfixes [1] [2] without giving it
ANY ATTENTION in your patch description. That's unacceptable and
insulting to contributors and reviewers. Full stop. Or did you expect
us to turn a blind eye? This is mainline, not some techpack playground.
[1]: https://lore.kernel.org/linux-arm-msm/20230123201133.zzt2zbyaw3pfkzi6@xxxxxxxxxxxxxx/
[2]: https://lore.kernel.org/linux-arm-msm/20221026182824.876933-10-marijn.suijten@xxxxxxxxxxxxxx/
Can we adapt this code so that both upstream and down stream shared same
timing engine programming so that easier to maintain?
Easy, I've said this before in IRC and will state it again: stop this
techpack nonsense and focus on upstream-first. When something passes
mainline review (and please don't bother maintainers and reviewers with
series like this) it is inevitably good enough to be copied to
techpack... at which point techpack becomes worthless as you can just
backport a mainline patch or use a recent-enough kernel.
tl;dr: it seems like you nor anyone involved in pre-reviewing/vetting
this series is familiar with upstream guidelines. Follow the global
advice from Dmitry [3] to reach a more efficient v2, and please don't
let this run to v10 (or beyond) again.
One suggestion to improve efficiency: split off the DPU v1.2 hardware
block addition (and related changes) into a separate series. A smaller
series (and properly split patches!) will give everyone less moving
parts to worry about, and paves the way for DSI support without blocking
on DP.
Yes to split DSC 1.2 integration and DP+DSC in 2 patchsets, with the various
fixes not necessary to make DP+DSC work in separate patches.
Be aware the rule is to make sure each single change doesn't break boot and builds
without warning, the rule is to make sure each single kernel change can be built
and doesn't break booting. And build the code with "W=1" to the make parameter to
trigger advanced GCC warnings.
This rule exists to permit running a git bisect to determine which commit introduces
a regression.
And the second most important rule is: a single change per patch, and a clear description
of the change in the commit message.
If your message needs a "change this and also change this" then it's wrong and it must be reworked.
Do incremental changes, like introduce a new struct, then use it afterwards.
Neil
[3]: https://lore.kernel.org/linux-arm-msm/47c83e8c-09f1-d1dd-ca79-574122638256@xxxxxxxxxx/
- Marijn
