Hello David,

On Thu, Mar 14, 2024 at 03:43:13PM +1100, David Gibson wrote:
> On Sun, Mar 10, 2024 at 09:30:31AM +0100, Uwe Kleine-König wrote:
> > On Mon, Feb 26, 2024 at 04:53:53PM +1100, David Gibson wrote:
> > > On Sun, Feb 25, 2024 at 06:54:23PM +0100, Uwe Kleine-König wrote:
> > > > A phandle in an overlay is not supposed to overwrite a phandle that
> > > > already exists in the base dtb as this breaks references to the
> > > > respective node in the base.
> > > > 
> > > > So add another iteration over the fdto that checks for such overwrites
> > > > and fixes the fdto phandle's value to match the fdt's.
> > > > 
> > > > A test is added that checks that newly added phandles and existing
> > > > phandles work as expected.
> > > > 
> > > > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> > > > ---
> > > > Hello,
> > > > 
> > > > here comes the next iteration of the patch that fixes overlay
> > > > application to not overwrite existing phandles.
> > > > 
> > > > It is rebased to current main branch. The changes since v2 are:
> > > > 
> > > >  - Add documentation
> > > >  - Apply the simplification from 24f60011fd43 ("libfdt: Simplify
> > > >    adjustment of values for local fixups") in the functions added here.
> > > >  - Rename functions using shorter and better names
> > > >  - Changed the test device trees to yield a hole in the phandle space
> > > >  - Checked each phandle value not being overwritten separately
> > > > 
> > > > Note I didn't switch the order of overlay_prevent_phandle_overwrite() and
> > > > overlay_fixup_phandles() because the overlay's phandles must be resolved
> > > > before I can do the recursion needed in
> > > > overlay_prevent_phandle_overwrite().
> > > 
> > > I'm not following what you mean here.  IIUC, conflicts of the sort
> > > you're handling can only arise when the overlay describes a phandle
> > > for the target node of the reference - and therefore that target is in
> > > the overlay.  In that case all references to it in the overlay should
> > > be encoded in __local_fixups__ rather than __fixups__.  __fixups__, in
> > > contrast describes references to nodes that aren't in the overlay, and
> > > so can't be filled in - even with a tentative value - until the
> > > overlay is applied.
> > > 
> > > So, I'm not seeing how fixing these conflicts depends on resolution of
> > > those "external" fixups, rather than just the "local" fixups.  Am I
> > > missing something?
> > 
> > yupp, look at the overlay dts I added in tests/. It has
> > 
> > 	&node_a {
> > 		value = <32>;
> > 	};
> > 
> > which is translated to:
> > 
> > 	fragment@1 {
> > 	    target = <0xffffffff>;
> > 	    __overlay__ {
> > 	        value = <0x00000020>;
> > 	    };
> > 	};
> > 	...
> > 	__fixups__ {
> > 	    node_a = ..., "/fragment@1:target:0"
> > 	};
> > 
> > Before I can recurse over fragment@1 and the matching base dtb node to
> > check for phandle conflicts, I need /fragment@1:target resolved;
> > otherwise I don't know where to look in the base dtb.
> > 
> > So if I switch the order, fdtoverlay reports
> > 
> > 	Failed to apply 'overlay_overlay_phandle.test.dtb': FDT_ERR_BADPHANDLE
> > 
> > in make check.
> 
> Ah, right.  It's specifically that we need to resolve the fragment
> targets (including via external symbols) before we can resolve this.
> Do you have a test case for this specific problem?  If not, I'd be
> worried, that I or someone else might forget the subtletey and try to
> re-order at some point in the future.

The test I added fails if you swap the two operations. That's how I
noticed.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |