Hi,
On 3/18/21 12:30 PM, Enrico Weigelt, metux IT consult wrote:
> On 17.03.21 21:03, Hans de Goede wrote:
>
> Hi,
>
>>> It just identifies the box and tells subsequent drivers which one it
>>> is, which watchdog and LED path to take. Moving the knowledge of which
>>> box has which LED/watchdog into the respective drivers seems to be the
>>> better way to go.
>>>
>>> So we would end up with a LED and a watchdog driver both
>>> MODULE_ALIAS("dmi:*:svnSIEMENSAG:*");
>
> Uh, isn't that a bit too broad ? This basically implies that Siemens
> will never produce boards with different configurations.
There is a further check done in probe() based on some Siemens specific
DMI table entries.
>>> and doing the identification with the inline dmi from that header,
>>> doing p2sb with the support to come ... possibly a "//TODO\ninline" in
>>> the meantime.
>>>
>>> So no "main platform" driver anymore, but still central platform
>>> headers.
>>>
>>> Not sure how this sounds, but i think making that change should be
>>> possible. And that is what i will try and go for in v3.
>>
>> Dropping the main drivers/platform/x86 driver sounds good to me,
>> I was already wondering a bit about its function since it just
>> instantiates devs to which the other ones bind to then instantiate
>> more devs (in the LED case).
>
> hmm, IMHO that depends on whether the individual sub-devices can be
> more generic than just that specific machine. (@Hanning: could you
> tell us more about that ?).
>
> Another question is how they're actually probed .. only dmi or maybe
> also pci dev ? (i've seen some refs to pci stuff in the led driver, but
> missed the other code thats called here).
>
> IMHO, if the whole thing lives on some PCI device (which can be probed
> via pci ID), and that device has the knowledge, where the LED registers
> actually are (eg. based on device ID, pci mmio mapping, ...) then there
> should be some parent driver that instantiates the led devices (and
> possibly other board specific stuff). That would be a clear separation,
> modularization. In that case, maybe this LED driver could even be
> replaced by some really generic "register-based-LED" driver, which just
> needs to be fed with some parameters like register ranges, bitmasks, etc.
>
> OTOH, if everything can be derived entirely from DMI match, w/o things
> like pci mappings involved (IOW: behaves like directly wired to the
> cpu's mem/io bus, no other "intelligent" bus involved), and it's all
> really board specific logic (no generic led or gpio controllers
> involved), then it might be better to have entirely separate drivers.
FWIW I'm fine with either solution, and if we go the "parent driver"
route I'm happy to have that driver sit in drivers/platform/x86
(once all the discussions surrounding this are resolved).
My reply was because I noticed that the Led driver seemed to sort of
also act as a parent driver (last time I looked) and instantiated a
bunch of stuff, so then we have 2 parent(ish) drivers. If things stay
that way then having 2 levels of parent drivers seems a bit too much
to me, esp. if it can all be done cleanly in e.g. the LED driver.
But as said I'm fine either way as long as the code is reasonably
clean and dealing with this sort of platform specific warts happens
a lot in drivers/platform/x86 .
Regards,
Hans
