Timur Tabi wrote:
Kumar Gala wrote:
The specific issue I'm talking about is the addition of new nodes that
might break old device trees.
New nodes or new properties? The CPM nodes are not new. On some device
trees, the original versions did not have a compatible property for the
CPM nodes (e.g. commit 0b5cf10691eb2c95a9126bf25f5e084d83d5d743).
As I said earlier, old-style CPM trees are a completely different
binding. They are not supported. They have not been supported for a
long time. They were only supported very briefly at the introduction of
CPM hardware to arch/powerpc, and were full of problems.
Like I said earlier, if you can demonstrate that *all* of these device
tree would be broken with the latest kernel anyway, then we don't need
to worry about backwards compatibility.
That is indeed the case. A "demonstration" of that for the tree you
quote is that the "reg" address changed -- if you tried feeding the old
tree into the new kernel, it would not find the CPM command register.
There is no code in the kernel that looks for the command-proc property
anymore.
I'm tired of debugging customer issues where the kernel is updated but
the firmware and device tree aren't. IMHO, Kernel developers are
generally too lax when it comes to firmware and device tree backwards
compatibility, and I think that's wrong.
I understand and agree (and it would be easier to get the backwards
compatibility right if we didn't have the attitude of "we'll fix the dts
later if we decide we actually care about that aspect of the hardware"
when adding it in the first place) -- it just isn't an issue in this
particular case.
-Scott
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev
