[PATCH v4 1/1] image: fit: Apply overlays using aligned writable FDT copies

[James Hilliard]

On Thu, Feb 12, 2026 at 8:26 AM Marek Vasut <marek.vasut at mailbox.org> wrote:
>
> On 2/11/26 7:06 PM, James Hilliard wrote:
>
> > +static int boot_get_fdt_fit_into_buffer(const void *src, ulong srclen,
> > +                                     ulong extra, ulong min_dstlen,
> > +                                     void **fdtdstbuf, ulong *fdtdstlenp)
> > +{
> > +     const void *fdtsrcbuf;
> > +     void *tmp = NULL;
> > +     void *dstbuf, *newdstbuf;
> > +     ulong dstlen, newdstlen;
> > +     int err;
> > +
> > +     /* Make sure the source FDT/DTO is 8-byte aligned for libfdt. */
> > +     fdtsrcbuf = src;
> > +     if (!IS_ALIGNED((uintptr_t)src, 8)) {
> > +             tmp = memalign(8, srclen);
> > +             if (!tmp)
> > +                     return -ENOMEM;
> > +
> > +             memcpy(tmp, src, srclen);
> > +             fdtsrcbuf = tmp;
> > +     }
> > +
> > +     newdstlen = ALIGN(fdt_totalsize(fdtsrcbuf) + extra, SZ_4K);
> > +     min_dstlen = ALIGN(min_dstlen, SZ_4K);
> > +     if (newdstlen < min_dstlen)
> > +             newdstlen = min_dstlen;
> > +
> > +     dstbuf = *fdtdstbuf;
> > +     dstlen = dstbuf ? *fdtdstlenp : 0;
> > +
> > +     /*
> > +      * If the caller already provided a large enough writable buffer,
> > +      * and we're not moving the FDT, nothing to do.
> > +      */
> > +     if (dstbuf && dstlen >= newdstlen && dstbuf == fdtsrcbuf) {
>
> Can $dstbuf ever be NULL ?
>
> > +             free(tmp);
> > +             return 0;
>
> You could try something like this here, to reduce the duplicate
> free(tmp) in the code:
>
> "
> err = 0;
> goto exit;
>
> ...
>
> exit:
>    free(tmp);
>    return err;
> "
>
> > +     }
> > +
>
> [...]
>
> > -     load = (ulong)of_flat_tree;
> > +     len = fdt_off_dt_strings(base_buf) + fdt_size_dt_strings(base_buf);
> How does this new length calculation work, can you please clarify that ?

>From my understanding, that len is tracking the current packed DTB data
length (up to the end of the strings block), not the allocated buffer size. We
recompute it after each overlay so need = len + ovcopylen
CONFIG_SYS_FDT_PAD grows from real used bytes instead of
accumulated slack. This is roughly the same idea as libfdt’s internal
fdt_data_size_(), and the final returned size is still taken after fdt_pack()
with fdt_totalsize().

> I don't think this will work for fitImages with external data (generated
> using mkimage -E) , but I might be wrong.

>From my understanding, this should still work with mkimage -E, external
data only changes where FIT payload bytes are stored, and
fit_image_load()/fit_image_get_data() already resolves that before overlay
application. At that point we’re operating on a normal copied DTB buffer,
and the length math is based on DTB-internal layout, not FIT
external-data placement.