On Thu, Jan 4, 2018 at 9:53 PM, David Gibson <david@xxxxxxxxxxxxxxxxxxxxx> wrote: > On Thu, Jan 04, 2018 at 03:08:08PM -0700, Ian Lepore wrote: >> On Thu, 2018-01-04 at 15:47 -0600, Kyle Evans wrote: >> > On Thu, Jan 4, 2018 at 3:34 PM, Kyle Evans <kevans@xxxxxxxxxxx> wrote: >> > > >> > > On Thu, Jan 4, 2018 at 2:41 PM, Kyle Evans <kevans@xxxxxxxxxxx> wrote: >> > > > >> > > > On Thu, Jan 4, 2018 at 2:33 PM, Frank Rowand wrote: >> > > > > >> > > > > [... snip ...] >> > > > > >> > > > > Does this remove the need for the proposed patch, or am I still >> > > > > missing something? >> > > > ... nope. Apparently I never tested this with this particular dtc(1) >> > > > and instead just assumed it did the same as ours- allocate phandle >> > > > sparsely, even with -@. That certainly removes the need for this >> > > > patch, and I'm somewhat upset that I hadn't previously considered >> > > > this. >> > > > >> > > > @David, Jon: Please disregard all of the patches along these lines... >> > > > I'll fix this in our dtc, where it should be fixed. >> > > > >> > > > Thanks, Frank! >> > > Actually, I'm kind of torn on whether this is useful or not. With >> > > being able to have EFI-provided FDT, it's hard to guarantee whether >> > > the FDT we're provided has been compiled with GPL dtc(1) and -@. The >> > > above solves this problem for most of my personal use-cases , though, >> > > since I can guarantee that our FDT and U-Boot provided FDT is compiled >> > > properly. >> > Apologies for the triple post; I realized that this argument is >> > inherently wrong, since we can't reference the node if there's no >> > symbol anyways. >> > >> > The only way this might still be a good idea is to support more >> > minimal cases where an implementation might prefer to not create a >> > phandle for nodes that haven't been referenced. >> > >> > In our case, we have a function [1] that walks the tree and generates >> > metadata on nodes that have phandles, under the assumption that these >> > have been referenced somewhere and provides a way to more quickly >> > reference these specifically through a separate linked link. >> > Allocating phandles for everything as GPL dtc does adds quite a bit >> > more overhead to this. >> > >> > [1] http://src.illumos.org/source/xref/freebsd-head/sys/dev/ofw/openfirm.c#119 >> >> In particular, it makes lookups more expensive as it now must traverse >> a list that includes every node in the dtb, rather than just nodes that >> are actually referenced. (It also increases the amount of storage, but >> at 20-ish bytes per node, that's not a big deal.) > > Lookups of what exactly? Aren't you unflattening the tree after you > read it in? Lookup in this context would be a lookup of the device from the xref phandle, see OF_device_from_xref [1] and its inverse OF_xref_from_device right below it. Devices, as they attach, register for the xref phandle, then consumers lookup the device associated with it and generally hold a handle to the device afterwards. [1] http://src.illumos.org/source/xref/freebsd-head/sys/dev/ofw/openfirm.c#628 -- To unsubscribe from this list: send the line "unsubscribe devicetree-compiler" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
