Create proper FDT attachment for GICv2m

[zbb]

Summary

Avoid probing GICv2m to any parent bus/driver. Instead, match
GICv2m driver with FDT complatible strings.
The intention behind this is to have two different drivers named
"gic" (for example GICv2 and GICv3). This is not possible if
GICv2m always probes.
In addition, not every GIC has a MSI controller in the form of
GICv2m extension.

Test Plan

We don't have any machine with GICv2m so please test.

[andrew]

Why do we want two drivers with the same name?

[zbb]

In D5015#106471, @andrew wrote:

Why do we want two drivers with the same name?

We want one (type of) PIC driver and one MSI controller matched by the compatible string or whatever.
The name is strictly for the convenience reasons :-). For example for all PCI bridges you have 'pcib' name and still only one driver attaches.
In addition, this commit prevents from attaching GICv2m on platforms that do not have GICv2m.

[andrew]

In D5015#106472, @zbb wrote:

In D5015#106471, @andrew wrote:

Why do we want two drivers with the same name?

We want one (type of) PIC driver and one MSI controller matched by the compatible string or whatever.
The name is strictly for the convenience reasons :-). For example for all PCI bridges you have 'pcib' name and still only one driver attaches.
In addition, this commit prevents from attaching GICv2m on platforms that do not have GICv2m.

But with pcib we have a common driver that attaches to it, this is not the case with the gic driver.

I'm not opposed to the change, querying the justification given in the summary, and why renaming the v3 driver is needed.

[zbb]

In D5015#106473, @andrew wrote:

In D5015#106472, @zbb wrote:

In D5015#106471, @andrew wrote:

Why do we want two drivers with the same name?

We want one (type of) PIC driver and one MSI controller matched by the compatible string or whatever.
The name is strictly for the convenience reasons :-). For example for all PCI bridges you have 'pcib' name and still only one driver attaches.
In addition, this commit prevents from attaching GICv2m on platforms that do not have GICv2m.

But with pcib we have a common driver that attaches to it, this is not the case with the gic driver.

I'm not opposed to the change, querying the justification given in the summary, and why renaming the v3 driver is needed.

OK. So the justification for the change is due to two reasons:

1. To not attach giv2m on platforms that do not have v2m
2. Change the driver's name to something more convenient.

Regarding 2. reason: when you attach gic_v3 you get something like this:

gic_v30
gic_v31
...

which may be somehow misleading (apart from the length of both names gic_v3 and gic_v3_its).
That is why we want to change the names to be short and clear. What do you think?

[zbb]

No objections here? Should I commit?