LinuxLists.cc - [PATCH] scripts/make_fit: Support decomposing DTBs

2024-06-05 09:49:07

Subject: [PATCH] scripts/make_fit: Support decomposing DTBs

The kernel tree builds some "composite" DTBs, where the final DTB is the
result of applying one or more DTB overlays on top of a base DTB with
fdtoverlay.

The FIT image specification already supports configurations having one
base DTB and overlays applied on top. It is then up to the bootloader to
apply said overlays and either use or pass on the final result. This
allows the FIT image builder to reuse the same FDT images for multiple
configurations, if such cases exist.

The decomposition function depends on the kernel build system, reading
back the .cmd files for the to-be-packaged DTB files to check for the
fdtoverlay command being called. This will not work outside the kernel
tree. The function is off by default to keep compatibility with possible
existing users.

To facilitate the decomposition and keep the code clean, the model and
compatitble string extraction have been moved out of the output_dtb
function. The FDT image description is replaced with the base file name
of the included image.

Signed-off-by: Chen-Yu Tsai <[email protected]>
---
This is a feature I alluded to in my replies to Simon's original
submission of the make_fit.py script [1].

This is again made a runtime argument as not all firmware out there
that boot FIT images support applying overlays. Like my previous
submission for disabling compression for included FDT images, the
bootloader found in RK3399 and MT8173 Chromebooks do not support
applying overlays. Another case of this is U-boot shipped by development
board vendors in binary form (without upstream) in an image or in
SPI flash on the board that were built with OF_LIBFDT_OVERLAY=n.
These would fail to boot FIT images with DT overlays. One such
example is my Hummingboard Pulse. In these cases the firmware is
either not upgradable or very hard to upgrade.

I believe there is value in supporting these cases. A common script
shipped with the kernel source that can be shared by distros means
the distro people don't have to reimplement this in their downstream
repos or meta-packages. For ChromeOS this means reducing the amount
of package code we have in shell script.

[1] https://lore.kernel.org/linux-kbuild/[email protected]/
[2]

scripts/Makefile.lib | 1 +
scripts/make_fit.py | 70 ++++++++++++++++++++++++++++++--------------
2 files changed, 49 insertions(+), 22 deletions(-)

diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 9f06f6aaf7fc..d78b5d38beaa 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -522,6 +522,7 @@ quiet_cmd_fit = FIT $@
cmd_fit = $(MAKE_FIT) -o $@ --arch $(UIMAGE_ARCH) --os linux \
--name '$(UIMAGE_NAME)' \
$(if $(findstring 1,$(KBUILD_VERBOSE)),-v) \
+ $(if $(FIT_DECOMPOSE_DTBS),--decompose-dtbs) \
--compress $(FIT_COMPRESSION) -k $< @$(word 2,$^)

# XZ
diff --git a/scripts/make_fit.py b/scripts/make_fit.py
index 263147df80a4..120f13e1323c 100755
--- a/scripts/make_fit.py
+++ b/scripts/make_fit.py
@@ -22,6 +22,11 @@ the entire FIT.
Use -c to compress the data, using bzip2, gzip, lz4, lzma, lzo and
zstd algorithms.

+Use -d to decompose "composite" DTBs into their base components and
+deduplicate the resulting base DTBs and DTB overlays. This requires the
+DTBs to be sourced from the kernel build directory, as the implementation
+looks at the .cmd files produced by the kernel build.
+
The resulting FIT can be booted by bootloaders which support FIT, such
as U-Boot, Linuxboot, Tianocore, etc.

@@ -64,6 +69,8 @@ def parse_args():
help='Specifies the architecture')
parser.add_argument('-c', '--compress', type=str, default='none',
help='Specifies the compression')
+ parser.add_argument('-d', '--decompose-dtbs', action='store_true',
+ help='Decompose composite DTBs into base DTB and overlays')
parser.add_argument('-E', '--external', action='store_true',
help='Convert the FIT to use external data')
parser.add_argument('-n', '--name', type=str, required=True,
@@ -140,12 +147,12 @@ def finish_fit(fsw, entries):
fsw.end_node()
seq = 0
with fsw.add_node('configurations'):
- for model, compat in entries:
+ for model, compat, files in entries:
seq += 1
with fsw.add_node(f'conf-{seq}'):
fsw.property('compatible', bytes(compat))
fsw.property_string('description', model)
- fsw.property_string('fdt', f'fdt-{seq}')
+ fsw.property('fdt', b''.join([b'fdt-%d\x00' % x for x in files]))
fsw.property_string('kernel', 'kernel')
fsw.end_node()

@@ -193,21 +200,9 @@ def output_dtb(fsw, seq, fname, arch, compress):
fname (str): Filename containing the DTB
arch: FIT architecture, e.g. 'arm64'
compress (str): Compressed algorithm, e.g. 'gzip'
-
- Returns:
- tuple:
- str: Model name
- bytes: Compatible stringlist
"""
with fsw.add_node(f'fdt-{seq}'):
- # Get the compatible / model information
- with open(fname, 'rb') as inf:
- data = inf.read()
- fdt = libfdt.FdtRo(data)
- model = fdt.getprop(0, 'model').as_str()
- compat = fdt.getprop(0, 'compatible')
-
- fsw.property_string('description', model)
+ fsw.property_string('description', os.path.basename(fname))
fsw.property_string('type', 'flat_dt')
fsw.property_string('arch', arch)
fsw.property_string('compression', compress)
@@ -215,7 +210,6 @@ def output_dtb(fsw, seq, fname, arch, compress):
with open(fname, 'rb') as inf:
compressed = compress_data(inf, compress)
fsw.property('data', compressed)
- return model, compat

def build_fit(args):
@@ -235,6 +229,7 @@ def build_fit(args):
fsw = libfdt.FdtSw()
setup_fit(fsw, args.name)
entries = []
+ fdts = collections.OrderedDict()

# Handle the kernel
with open(args.kernel, 'rb') as inf:
@@ -243,12 +238,43 @@ def build_fit(args):
write_kernel(fsw, comp_data, args)

for fname in args.dtbs:
- # Ignore overlay (.dtbo) files
- if os.path.splitext(fname)[1] == '.dtb':
- seq += 1
- size += os.path.getsize(fname)
- model, compat = output_dtb(fsw, seq, fname, args.arch, args.compress)
- entries.append([model, compat])
+ # Ignore non-DTB (*.dtb) files
+ if os.path.splitext(fname)[1] != '.dtb':
+ continue
+
+ # Get the compatible / model information
+ with open(fname, 'rb') as inf:
+ data = inf.read()
+ fdt = libfdt.FdtRo(data)
+ model = fdt.getprop(0, 'model').as_str()
+ compat = fdt.getprop(0, 'compatible')
+
+ if args.decompose_dtbs:
+ # Check if the DTB needs to be decomposed
+ path, basename = os.path.split(fname)
+ cmd_fname = os.path.join(path, f'.{basename}.cmd')
+ with open(cmd_fname, 'r', encoding='ascii') as inf:
+ cmd = inf.read()
+
+ if 'scripts/dtc/fdtoverlay' in cmd:
+ # This depends on the structure of the composite DTB command
+ files = cmd.split()
+ files = files[files.index('-i')+1:]
+ else:
+ files = [fname]
+ else:
+ files = [fname]
+
+ for fn in files:
+ if fn not in fdts:
+ seq += 1
+ size += os.path.getsize(fn)
+ output_dtb(fsw, seq, fn, args.arch, args.compress)
+ fdts[fn] = seq
+
+ files_seq = [fdts[fn] for fn in files]
+
+ entries.append([model, compat, files_seq])

finish_fit(fsw, entries)

--
2.45.1.288.g0e0cd299f1-goog

2024-06-10 16:05:38

by Simon Glass

[permalink] [raw]

On Tue, Jun 11, 2024 at 10:45 PM Masahiro Yamada <[email protected]> wrote:
>
> On Tue, Jun 11, 2024 at 5:52 PM Chen-Yu Tsai <[email protected]> wrote:
> >
> > On Mon, Jun 10, 2024 at 11:16 PM Simon Glass <[email protected]> wrote:
> > >
> > > Hi Chen-Yu,
> > >
> > > On Wed, 5 Jun 2024 at 03:48, Chen-Yu Tsai <[email protected]> wrote:
> > > >
> > > > The kernel tree builds some "composite" DTBs, where the final DTB is the
> > > > result of applying one or more DTB overlays on top of a base DTB with
> > > > fdtoverlay.
> > > >
> > > > The FIT image specification already supports configurations having one
> > > > base DTB and overlays applied on top. It is then up to the bootloader to
> > > > apply said overlays and either use or pass on the final result. This
> > > > allows the FIT image builder to reuse the same FDT images for multiple
> > > > configurations, if such cases exist.
> > > >
> > > > The decomposition function depends on the kernel build system, reading
> > > > back the .cmd files for the to-be-packaged DTB files to check for the
> > > > fdtoverlay command being called. This will not work outside the kernel
> > > > tree. The function is off by default to keep compatibility with possible
> > > > existing users.
> > > >
> > > > To facilitate the decomposition and keep the code clean, the model and
> > > > compatitble string extraction have been moved out of the output_dtb
> > > > function. The FDT image description is replaced with the base file name
> > > > of the included image.
> > > >
> > > > Signed-off-by: Chen-Yu Tsai <[email protected]>
> > > > ---
> > > > This is a feature I alluded to in my replies to Simon's original
> > > > submission of the make_fit.py script [1].
> > > >
> > > > This is again made a runtime argument as not all firmware out there
> > > > that boot FIT images support applying overlays. Like my previous
> > > > submission for disabling compression for included FDT images, the
> > > > bootloader found in RK3399 and MT8173 Chromebooks do not support
> > > > applying overlays. Another case of this is U-boot shipped by development
> > > > board vendors in binary form (without upstream) in an image or in
> > > > SPI flash on the board that were built with OF_LIBFDT_OVERLAY=n.
> > > > These would fail to boot FIT images with DT overlays. One such
> > > > example is my Hummingboard Pulse. In these cases the firmware is
> > > > either not upgradable or very hard to upgrade.
> > > >
> > > > I believe there is value in supporting these cases. A common script
> > > > shipped with the kernel source that can be shared by distros means
> > > > the distro people don't have to reimplement this in their downstream
> > > > repos or meta-packages. For ChromeOS this means reducing the amount
> > > > of package code we have in shell script.
> > > >
> > > > [1] https://lore.kernel.org/linux-kbuild/[email protected]/
> > > > [2]
> > > >
> > > > scripts/Makefile.lib | 1 +
> > > > scripts/make_fit.py | 70 ++++++++++++++++++++++++++++++--------------
> > > > 2 files changed, 49 insertions(+), 22 deletions(-)
> > >
> > > This is a clever way to discover the included files. Does it need to
> > > rely on the Linux build information, or could this information somehow
> > > be in the .dtb files? I had expected some sort of overlay scheme in
> >
> > (+CC DT folks and mailing list)
> >
> > I suppose we could make the `fdtoverlay` program embed this data during
> > the kernel build. That would keep the information together, while also
> > having one source of truth (the kernel Makefiles). Whether it belongs
> > in the DTB or not is a separate matter.
>
>
> Some time ago, I asked a similar question
> with a similar motivation.
>
> https://lore.kernel.org/devicetree-compiler/CAK7LNARV8Bo-tBXMdOu55Wg9uZRXvNiRdkDJ4LH8PwVMnMp4cA@mail.gmail.com/

I think this discussion was geared towards "unapplying" overlays. What
we would like is for metadata to be available, and preferably not tied
to the kernel build system generated metadata file, so that with the same
bunch of DTB + DTBO files, we could figure out how to decompose them and
just bundle the components.

If the metadata is embedded within the composite DTB, then given a DTB
bundle (such as installed with `make dtbs_install`) the make_fit.py
script could go and do the decomposition without the *.cmd files
from the kernel build. This is assuming all the component parts are
installed together.

ChenYu

> >
> > > the source, but perhaps people have given up on that?
> >
> > I wouldn't say given up, since we haven't agreed on anything either.
> > Elliot had some concerns when I brought this up earlier [1] though.
> >
> > ChenYu
> >
> > [1] https://lore.kernel.org/linux-mediatek/[email protected]/
> >
>
> --
> Best Regards
> Masahiro Yamada