Re: [PATCH] drm/lima: Mark simple_ondemand governor as softdep

From: Dragan Simic
Date: Thu Jul 25 2024 - 04:22:21 EST

Hello Qiang,

On 2024-06-26 08:49, Dragan Simic wrote:
On 2024-06-26 03:11, Qiang Yu wrote:
On Wed, Jun 26, 2024 at 2:15 AM Dragan Simic <dsimic@xxxxxxxxxxx> wrote:
Just checking, any further thoughts about this patch?

I'm OK with this as a temp workaround because it's simple and do no harm
even it's not perfect. If no other better suggestion for short term, I'll submit
this at weekend.

Thanks. Just as you described it, it's far from perfect, but it's still
fine until there's a better solution, such as harddeps. I'll continue my
research about the possibility for adding harddeps, which would hopefully
replace quite a few instances of the softdep (ab)use.

Another option has become available for expressing additional module
dependencies, weakdeps. [1][2] Long story short, weakdeps are similar
to softdeps, in the sense of telling the initial ramdisk utilities to
include additional kernel modules, but weakdeps result in no module
loading being performed by userspace.

Maybe "weak" isn't the best possible word choice (arguably, "soft" also
wasn't the best word choice), but weakdeps should be a better choice for
use with Lima and governor_simpleondemand, because weakdeps provide the
required information to the utilities used to generate initial ramdisk,
while the actual module loading is left to the kernel.

The recent addition of weakdeps renders the previously mentioned harddeps
obsolete, because weakdeps actually do what we need. Obviously, "weak"
doesn't go along very well with the actual nature of the dependency between
Lima and governor_simpleondemand, but it's pretty much just the somewhat
unfortunate word choice.

The support for weakdeps has been already added to the kmod [3][4] and
Dracut [5] userspace utilities. I'll hopefully add support for weakdeps
to mkinitcpio [6] rather soon.

Maybe we could actually add MODULE_HARDDEP() as some kind of syntactic
sugar, which would currently be an alias for MODULE_WEAKDEP(), so the
actual hard module dependencies could be expressed properly, and possibly
handled differently in the future, with no need to go back and track all
such instances of hard module dependencies.

With all this in mind, here's what I'm going to do:

1) Submit a patch that adds MODULE_HARDDEP() as syntactic sugar
2) Implement support for weakdeps in Arch Linux's mkinitcpio [6]
3) Depending on what kind of feedback the MODULE_HARDDEP() patch receives,
I'll submit follow-up patches for Lima and Panfrost, which will swap
uses of MODULE_SOFTDEP() with MODULE_HARDDEP() or MODULE_WEAKDEP()

Looking forward to your thoughts.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/include/linux/module.h?id=61842868de13aa7fd7391c626e889f4d6f1450bf
[2] https://lore.kernel.org/linux-kernel/20240724102349.430078-1-jtornosm@xxxxxxxxxx/T/#u
[3] https://github.com/kmod-project/kmod/commit/05828b4a6e9327a63ef94df544a042b5e9ce4fe7
[4] https://github.com/kmod-project/kmod/commit/d06712b51404061eef92cb275b8303814fca86ec
[5] https://github.com/dracut-ng/dracut-ng/commit/8517a6be5e20f4a6d87e55fce35ee3e29e2a1150
[6] https://gitlab.archlinux.org/archlinux/mkinitcpio/mkinitcpio


On 2024-06-18 21:22, Dragan Simic wrote:
> On 2024-06-18 12:33, Dragan Simic wrote:
>> On 2024-06-18 10:13, Maxime Ripard wrote:
>>> On Tue, Jun 18, 2024 at 04:01:26PM GMT, Qiang Yu wrote:
>>>> On Tue, Jun 18, 2024 at 12:33 PM Qiang Yu <yuq825@xxxxxxxxx> wrote:
>>>> >
>>>> > I see the problem that initramfs need to build a module dependency chain,
>>>> > but lima does not call any symbol from simpleondemand governor module.
>>>> > softdep module seems to be optional while our dependency is hard one,
>>>> > can we just add MODULE_INFO(depends, _depends), or create a new
>>>> > macro called MODULE_DEPENDS()?
>>
>> I had the same thoughts, because softdeps are for optional module
>> dependencies, while in this case it's a hard dependency. Though,
>> I went with adding a softdep, simply because I saw no better option
>> available.
>>
>>>> This doesn't work on my side because depmod generates modules.dep
>>>> by symbol lookup instead of modinfo section. So softdep may be our
>>>> only
>>>> choice to add module dependency manually. I can accept the softdep
>>>> first, then make PM optional later.
>>
>> I also thought about making devfreq optional in the Lima driver,
>> which would make this additional softdep much more appropriate.
>> Though, I'm not really sure that's a good approach, because not
>> having working devfreq for Lima might actually cause issues on
>> some devices, such as increased power consumption.
>>
>> In other words, it might be better to have Lima probing fail if
>> devfreq can't be initialized, rather than having probing succeed
>> with no working devfreq. Basically, failed probing is obvious,
>> while a warning in the kernel log about no devfreq might easily
>> be overlooked, causing regressions on some devices.
>>
>>> It's still super fragile, and depends on the user not changing the
>>> policy. It should be solved in some other, more robust way.
>>
>> I see, but I'm not really sure how to make it more robust? In
>> the end, some user can blacklist the simple_ondemand governor
>> module, and we can't do much about it.
>>
>> Introducing harddeps alongside softdeps would make sense from
>> the design standpoint, but the amount of required changes wouldn't
>> be trivial at all, on various levels.
>
> After further investigation, it seems that the softdeps have
> already seen a fair amount of abuse for what they actually aren't
> intended, i.e. resolving hard dependencies. For example, have
> a look at the commit d5178578bcd4 (btrfs: directly call into
> crypto framework for checksumming) [1] and the lines containing
> MODULE_SOFTDEP() at the very end of fs/btrfs/super.c. [2]
>
> If a filesystem driver can rely on the abuse of softdeps, which
> admittedly are a bit fragile, I think we can follow the same
> approach, at least for now.
>
> With all that in mind, I think that accepting this patch, as well
> as the related Panfrost patch, [3] should be warranted. I'd keep
> investigating the possibility of introducing harddeps in form
> of MODULE_HARDDEP() and the related support in kmod project,
> similar to the already existing softdep support, [4] but that
> will inevitably take a lot of time, both for implementing it
> and for reaching various Linux distributions, which is another
> reason why accepting these patches seems reasonable.
>
> [1]
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d5178578bcd4
> [2]
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/fs/btrfs/super.c#n2593
> [3]
> https://lore.kernel.org/dri-devel/4e1e00422a14db4e2a80870afb704405da16fd1b.1718655077.git.dsimic@xxxxxxxxxxx/
> [4]
> https://git.kernel.org/pub/scm/utils/kernel/kmod/kmod.git/commit/?id=49d8e0b59052999de577ab732b719cfbeb89504d