Den 06.02.2017 16.53, skrev Daniel Vetter:
On Mon, Feb 06, 2017 at 12:08:47PM +0100, Thierry Reding wrote:
On Mon, Feb 06, 2017 at 11:07:42AM +0100, Daniel Vetter wrote:
On Mon, Feb 6, 2017 at 10:35 AM, Thierry Reding thierry.reding@gmail.com wrote:
> +EXPORT_SYMBOL(tinydrm_disable_backlight); > +#endif These look like they really should be part of the backlight subsystem.
I
don't see anything DRM specific about them. Well, except for the error messages.
So this is a bit an unpopular opinion with some folks, but I don't
require
anyone to submit new code to subsystems outside of drm for new drivers. Simply because it takes months to get stuff landed, and in general it's not worth the trouble.
"Not worth the trouble" is very subjective. If you look at the Linux kernel in general, one of the reasons why it works so well is because the changes we make apply to the kernel as a whole. Yes, sometimes that makes things more difficult and time-consuming, but it also means that the end result will be much more widely usable and therefore benefits everyone else in return. In my opinion that's a large part of why the kernel is so successful.
We have piles of stuff in drm and drm drivers that should be in core but isn't.
Imo the only reasonable way is to merge as-is, then follow-up with a
patch
series to move the helper into the right subsystem. Most often unfortunately that follow-up patch series will just die.
Of course follow-up series die. That's because nobody cares to follow-up once their code has been merged.
Collecting our own helpers or variants of subsystems is a great way of isolating ourselves from the rest of the community. I don't think that's a good solution in the long run at all.
We have a bunch of patch series that we resubmit for months and they go exactly nowhere. They don't die because we stop caring, they die because they die. Some of them we even need to constantly rebase and carry around in drm-tip since our CI would Oops or spew WARNIGs all over the place. There's simply some areas of the kernel which seem overloaded under patches and no one is willing or able to fix things, and I can't fix the entire kernel. Nor expect contributors (who have much less political weight to throw around than me) to do that and succeed. And we don't end up with worse code in the drm subsystem, since we can still do the refactoring within drm helpers and end up with clean drivers.
I fully agree that it's not great for the kernel's future, but when I'm stuck with the option to get shit done or burning out playing the upstreaming game, the choice is easy. And in the end I care about open source gfx much more than the kernel, and I think for open source gfx's success it's crucial that we're welcoming to new contributors and don't throw up massive roadblocks. Open source gfx is tiny and still far away from world domination, we need _lots_ more people. If that means routing around other subsystems for them, I'm all for it.
I can't say I fully agree with that sentiment. I do see how routing around subsystems can be useful occasionally. If nobody will merge the code, or if nobody cares, then by all means, let's make them DRM- specific helpers.
But I think we need to at least try to do the right thing. If only to teach people what the right way is. If we start accepting such things by default, how can we expect contributors to even try?
I also think that contributors will often end up contributing not only to DRM but to the kernel as a whole. As such it should be part of our mentoring to teach them about how the process works as a rule, even if the occasional exception is necessary to get things done.
In this particular case, I know for a fact that both backlight and SPI maintainers are very responsive, so that's not a good excuse.
I definitely don't want that we don't attempt this. But brought from years of experience, I recommend to merge first (with pre-refactoring already applied, but helpers only extracted, not yet at the right spot), and then follow up with. Because on average, there's way too many trees with overloaded maintainers who maybe look at your patch once per kernel release cycle.
If you know that backlight and spi isn't one of these areas (anything that goes through takashi/sound is a similar good experience for us on the i915 side), then I guess we can try. But then Noralf has already written a few months worth of really great refactoring, and I'm seriously starting to feel guilty for volunteering him for all of this. Even though he seems to be really good at it, and seems to not mind, it's getting a bit silly. Given that I'd say up to Noralf.
In short, there's always a balance.
Yes, it's a balance between the perfect and not's so perfect, the professinal and the amateur, and the drm expert and the newbie.
If I knew how much time I would have to spend on tinydrm to get it merged, then I would never have started it. I wondered, a couple of months back, if I should just cut my losses and move to something else. dri-devel is a friendly place and I do get help to keep moving, but I get the feeling that drm is really a place for professionals that write kernel code 50 hours a week. And there's nothing wrong in that, drm is complex and maybe that kind of expertice and work-hours are needed to do work here.
It's not that I plan on backing out now, I'm just putting down a few words about the challenge it has been for me as a hobbyist to meet drm.
As for the specifics,
Backlight enable/disable helpers, I haven't thought about those as backlight subsystem helpers. But I see some drm panel drivers that do the same, so it makes sense. But what I'm feeling is this: If I'm expected to get everything right, then I will "never" get this merged. This thing by itself isn't much, but adding up every little thing amounts to a lot. And it's not just "make this patch", it's probably clean up the drivers as well :-)
tinydrm_spi_transfer() can print the content of the buffer. This is important at least until I have seen some real use of the drivers. I have emulation code for handling bit widths not supported by the SPI controller, so I want to see what goes over the wire. SPI core uses trace events to track what's going on, and I don't think I can get buffer dumping into that code.
Don't get me wrong Thierry, I do appreciate that you take the time to look at the code. I'm just frustrated that it takes so long to get this right. I thought that all I needed now was a DT maintainer ack :-)
Noralf.