Removing the complete __fixups__ and __local_fixups__ tree might delete
data that should better be retained. See the added test for a situation
that was broken before.
Note that without removing /__fixups__ and /__local_fixups__ in
generate_fixups_tree() and generate_local_fixups_tree() respectively
calling build_and_name_child_node() isn't safe as the nodes might
already exist and then a duplicate would be added. So build_root_node()
has to be used which copes correctly here.
Fixes: 915daadbb6 ("Start with empty __local_fixups__ and __fixups__ nodes")
Closes: https://github.com/dgibson/dtc/issues/170
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@baylibre.com>
Message-ID: <b061ee57157fafbb9d5b9b0b86af760d13a04eda.1755512759.git.u.kleine-koenig@baylibre.com>
[dwg: Use -1 instead of 1 as an error return]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The dtc graph_child_address check can't distinguish between bindings
where there can only be a single endpoint, and cases where there can be
multiple endpoints.
In cases where the bindings allow for multiple endpoints but only one is
described false warnings about unnecessary #address-cells/#size-cells
can be generated, but only if the endpoint described have an address of
0 (A), for single endpoints with a non-zero address (B) no warnings are
generated.
A)
ports {
#address-cells = <1>;
#size-cells = <0>;
port@0 {
#address-cells = <1>;
#size-cells = <0>;
sourceA: endpoint@0 {
reg = <0>
};
};
};
B)
ports {
#address-cells = <1>;
#size-cells = <0>;
port@0 {
#address-cells = <1>;
#size-cells = <0>;
sourceB: endpoint@1 {
reg = <1>
};
};
};
Remove the check as it is somewhat redundant now that we can use schemas
to validate the full node.
Signed-off-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Reviewed-by: Rob Herring (Arm) <robh@kernel.org>
Message-ID: <20250817133733.3483922-1-niklas.soderlund+renesas@ragnatech.se>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
In some places run_tsets.sh needs to get the size of files, which it does
with stat(1). However the syntax to do this is different between GNU
coreutils stat(1) and BSD's stat(1). We have some logic that looks for
"GNU" in the version string to figure out the correct version.
This will break upcoming Ubuntu versions which are now using uutils, a Rust
reimplementation of coreutils. These support the same GNU syntax, but
don't have the "GNU" in the version string.
Update the detection to simply try the GNU version and otherwise assume
BSD.
Link: https://github.com/dgibson/dtc/issues/166
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Discovered with the "codespell" utility.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-ID: <20250313191607.1556384-1-thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
For certain tests, we take the output from dtc -O asm and build it into
a .so shared library which we then dlopen() for further tests. Because we
don't mark it otherwise, it's treated as requiring an executable stack,
which dlopen() refuses to open as of glibc-2.41.
Of course, the library is pure data, no code, so it certainly doesn't need
an executable stack. Add the -znoexecstack linker option to avoid the
error.
Fixes: https://github.com/dgibson/dtc/issues/163
Reported-by: Xi Ruoyao <xry111@xry111.site>
Signed-off-by: David Gibson <david@gibson.dropbear.id.a>
User @sharkcz noted that when the '--quiet' flag is removed from
./setup.py, the following can be seen from the `./setup.py install`
stage.
Running custom install script 'dtc/g/pylibfdt/../setup.py --top-builddir \
dtc/g/redhat-linux-build install --prefix=/usr/local --root=$DESTDIR'
running install
...
building '_libfdt' extension
swigging dtc/g/pylibfdt/../pylibfdt/libfdt.i to \
dtc/g/pylibfdt/../pylibfdt/libfdt_wrap.c
swig -python -Idtc/g/pylibfdt/../libfdt -o \
dtc/g/pylibfdt/../pylibfdt/libfdt_wrap.c dtc/g/pylibfdt/../pylibfdt/libfdt.i
gcc -fno-strict-overflow -Wsign-compare -DDYNAMIC_ANNOTATIONS_ENABLED=1 \
-DNDEBUG -fexceptions -fexceptions -fexceptions -fPIC -DPY_SSIZE_T_CLEAN \
-Idtc/g/pylibfdt/../libfdt -I/usr/include/python3.12 -c \
dtc/g/pylibfdt/../pylibfdt/libfdt_wrap.c -o \
build/temp.linux-ppc64le-cpython-312dtc/g/pylibfdt/../pylibfdt/libfdt_wrap.o
creating build/lib.linux-ppc64le-cpython-312
gcc -shared build/temp.linux-ppc64le-cpython-312dtc/g/pylibfdt/../pylibfdt/libfdt_wrap.o \
-Ldtc/g/redhat-linux-build/libfdt -L/usr/lib64 -lfdt -o \
build/lib.linux-ppc64le-cpython-312/_libfdt.cpython-312-powerpc64le-linux-gnu.so
copying dtc/g/pylibfdt/../pylibfdt/libfdt.py -> build/lib.linux-ppc64le-cpython-312
Meaning the python library is getting recompiled during the `meson
install` phase. This causes build issues as Meson does not set the
compiler and linker flags during the install phase.
The reason the library is getting rebuilt is during the normal build
with "build_ext", the `--build-lib` flag gets passed which changes the
default output build directory. But there is no equivalent option for
the "install" command. Install instead looks in the default directory
"./build" and so does not find the previously built library.
Since we can't fix the "install" command, drop the --build-lib flag.
This causes setup.py to compile the libraries at
`<meson-build>/build/lib.linux-x86_64-cpython-312/`. We must also then
fix run_tests.sh to find the library build directory as it's
machine-dependent.
Fixes: #135
Signed-off-by: Brandon Maier <brandon.maier@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Meson automatically passes in LD_LIBRARY_PATH pointing at the correct
build directory for libfdt.so. So preserve LD_LIBRARY_PATH if it's
already set.
Signed-off-by: Brandon Maier <brandon.maier@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Changed the target DTS from overlay_base_no_symbols.test.dtb to overlay_base_manual_symbols.test.dtb.
This ensures that the test case doesn't exit prematurely due to the absence of label-linked phandle in the symbols node.
The update guarantees that the test case appropriately checks the validity of the fixup string linked to the label, as intended.
Signed-off-by: Zheng Guangyuan <1628513611@qq.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Tests running under Meson run from a different working directory then
under Makefile. Some of these tests had not been fixed to work from a
different directory because the tests were testing for an error
condition which is indistinguishable from a missing file.
Signed-off-by: Brandon Maier <brandon.maier@collins.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
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>
Message-ID: <20240225175422.156393-2-u.kleine-koenig@pengutronix.de>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
By convention, the PKG_CONFIG environment variable is used to tell
build systems which pkg-config executable should be used. This is
often used when cross compiling, where it might be set to something
like "aarch64-unknown-linux-gnu-pkg-config".
Signed-off-by: Alyssa Ross <hi@alyssa.is>
Message-ID: <20240123130409.181128-2-hi@alyssa.is>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When the DTS output has no type markers, we have to guess the type. Prior
to commit 32b9c61307 ("Preserve datatype markers when emitting dts
format"), instances of string lists would be delimited. Since then, a
single string with embedded "\0"s are emitted. An embedded "\0" is valid
for DTS files, but that's a rare exception and lists of strings are the
overwhelming majority. Restore the prior behavior.
stringlist.dts is reused for testing this, but needs a couple of tweaks
in order to match the dts output.
Reported-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Rob Herring <robh@kernel.org>
Message-ID: <20231027142901.2536622-1-robh@kernel.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When running under Meson, check_tests() is generating dtb build files in
the source directory. This is because dtb is named by appending
".test.dtb" to the full source file name.
Use basename to extract just the source filename and write it to the
working directory which is the build directory.
Signed-off-by: Brandon Maier <brandon.maier@collins.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
* bad-graph-child-address.dts: additional child address test since the
one in bad-graph.dts is now shadowed by its prerequisites also
failing.
* bad-graph-reg-cells.dts: test warnings produced by check_graph_reg().
Signed-off-by: Johannes Beisswenger <johannes.beisswenger@cetitec.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
libyaml before 0.2.3 expects non-const string parameters. Supporting
both variants would require either cpp magic or ignoring
"discarded-qualifiers" compiler warnings. For the sake of simplicity
just support libyaml 0.2.3 and newer.
Note that NO_YAML can be overwritten on the make command line.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Because meson always builds out-of-tree we need to reference things in the
original source tree via $SRCDIR from run_tests.sh. We forgot a couple of
cases for the cell overflow tests. Fix them.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We always evaluate integer values in cell arrays as 64-bit quantities, then
truncate to the size of the array cells (32-bit by default). However to
detect accidental truncation of meaningful values, we give an error if the
truncated portion isn't either all 0 or all 1 bits. However, this can
still give counterintuitive errors. For if the user is thinking in 2's
complement 32-bit arithmetic (which would be quite natural), then they'd
expect the expression (-0xffffffff-2) to evaluate to -1 (0xffffffff).
However in 64-bit it evaluates to 0xfffffffeffffffff which does truncate
to the expected value but trips this error message.
Because of this reduce the error to only a warnings, with a somewhat more
helpful message.
Fixes: https://github.com/dgibson/dtc/issues/74
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Add a new test get_next_tag_invalid_prop_len, which covers
fdt_next_tag(), when it is passed an corrupted blob, with
invalid property len values. The test runs twice, on a blob
in sw and finished state.
Signed-off-by: Tadeusz Struk <tadeusz.struk@linaro.org>
Message-Id: <20221011182611.116011-2-tadeusz.struk@linaro.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The dtb overlay format only permits (non local) fixups to reference labels,
not paths. That's because the fixup target goes into the property name in
the overlay, and property names aren't permitted to include '/' characters.
Stop erroneously generating such fixups, because we didn't check for this
case.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Set STATIC_BUILD=1 environment variable to enable static building
when using makefiles.
Signed-off-by: Tero Tervala <tero.tervala@unikie.com>
Message-Id: <20220629163531.932281-1-tero.tervala@unikie.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
/bin/sh points to dash instead of bash in some linux distros.
One test would fail if dash was used, this fix will allow all tests
to run properly on dash too.
dash built-in printf does not support "\xNN" -hex escape format.
"\NNN" -octal escape format is supported by both bash and dash printf.
Replaced "$(echo "$expect")" because this actually runs /bin/echo
instead of shell internal echo and in some cases causes "\NNN" escapes
to be printed as the actual characters they represent instead of the
escape sequence itself.
Cosmetic quotes added to make printout a bit clearer.
Signed-off-by: Tero Tervala <tero.tervala@unikie.com>
Message-Id: <20220704073722.1075849-1-tero.tervala@unikie.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
If the corresponding '#xxx-cells' value is much too large, an integer
overflow can prevent the checks in check_property_phandle_args() from
correctly determining that the checked property is too short for the
given cells value. This leads to an infinite loops.
This patch fixes the bug, and adds a testcase for it. Further
information in https://github.com/dgibson/dtc/issues/64
Reported-by: Anciety <anciety@pku.edu.cn>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
FT is sometimes used for storing raw data. That is quite common for
U-Boot FIT images.
Extracting such data is not trivial currently. Using type 's' (string)
will replace every 0x00 (NUL) with 0x20 (space). Using type 'x' will
print bytes but in xxd incompatible format.
This commit adds support for 'r' (raw) format. Example usage:
fdtget -t r firmware.itb /images/foo data > image.raw
Support for encoding isn't added as there isn't any clean way of passing
binary data as command line argument.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Message-Id: <20211209061420.29466-1-zajec5@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Add a check for parsing 'interrupt-map' properties. The check primarily
tests parsing 'interrupt-map' properties which depends on and the parent
interrupt controller (or another map) node.
Note that this does not require '#address-cells' in the interrupt-map
parent, but treats missing '#address-cells' as 0 which is how the Linux
kernel parses it. There's numerous cases that expect this behavior.
Cc: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Message-Id: <20211015213527.2237774-1-robh@kernel.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
With the prior commit, this check is now redundant.
Signed-off-by: Rob Herring <robh@kernel.org>
Message-Id: <20210526010335.860787-4-robh@kernel.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The root node is supposed to have an empty name, but at present this is
not checked. The behaviour of such a tree is not well defined. Most
software rightly assumes that the root node is at offset 0 and does not
check the name. This oddity was discovered as part of a security
investigation into U-Boot verified boot.
Add a check for this to fdt_check_full().
Signed-off-by: Simon Glass <sjg@chromium.org>
Reported-by: Arie Haenel <arie.haenel@intel.com>
Reported-by: Julien Lenoir <julien.lenoir@intel.com>
Message-Id: <20210323010410.3222701-2-sjg@chromium.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
At present it is possible to have two root nodes and even access nodes
in the 'second' root. Such trees should not be considered valid. This
was discovered as part of a security investigation into U-Boot verified
boot.
Add a check for this to fdt_check_full().
Signed-off-by: Simon Glass <sjg@chromium.org>
Reported-by: Arie Haenel <arie.haenel@intel.com>
Reported-by: Julien Lenoir <julien.lenoir@intel.com>
Message-Id: <20210323000926.3210733-1-sjg@chromium.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This test was accidentally skipped as the wrong test dts file was built.
The fragment numbering in this sugar-free test case needed adjusting to
match the numbering generated by dtc for overlay_overlay.dts.
Signed-off-by: Paul Barker <pbarker@konsulko.com>
Message-Id: <20201219143521.2118-1-pbarker@konsulko.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
An interrupt provider (an actual interrupt-controller node or an
interrupt nexus) should have both #address-cells and #interrupt-cells
properties explicitly defined.
Add an extra test for this. We check for the #interrupt-cells property
already, but this does not cover every controller so far, only those that
get referenced by an interrupts property in some node. Also we miss
interrupt nexus nodes.
A missing #address-cells property is less critical, but creates
ambiguities when used in interrupt-map properties, so warn about this as
well now.
This removes the now redundant warning in the existing interrupts test.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Message-Id: <20200515141827.27957-2-andre.przywara@arm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Currently the test runner script always expects to be run from within the
tests/ subdirectory of a dtc source tree: it looks for dtc and other
binaries in the parent of the current directory and for the libfdt shared
library in ../libfdt.
That works great with make check and for testing a build you've just made.
However, sometimes it's useful to test a dtc & libfdt which have already
been installed on the system, or which for whatever reason are located
somewhere else.
This patch allows the test runner script to do this when TEST_BINDIR and/or
TEST_LIBDIR environment variables are set.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Support for YAML output in dtc is optional (to cover systems that don't
have libyaml). Currently the tests for yaml output test if the libyaml
package is locally installed.
That duplicates similar logic in the Makefile, and worse it will cause
failed tests if the user explicitly disables YAML support, rather than
simply not having libyaml installed.
Fix this by having the test script use the NO_YAML variable exported by
make. Fall back to the current test if the variable isn't set, such as
when running the script manually.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Currently the test script bases whether to run the Python tests on whether
it can see a built Python module. That can easily be fooled if there is
a stale module there.
Instead, have it actually look at the NO_PYTHON variable exported from the
Makefile. If the variable doesn't exist (such as if we're running the
script manually) fall back on the old logic.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Currently run_tests.sh needs the $PYTHON environment variable set to
correctly run pylibfdt tests. The Makefile does this for make check, but
it breaks if the script is run manually. Add a fallback to handle that
case.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Generalize the existing 'ranges' check to also work for 'dma-ranges'
which has the same parsing requirements.
Signed-off-by: Rob Herring <robh@kernel.org>
Message-Id: <20200303193931.1653-1-robh@kernel.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
By default FreeBSD does not have 'gcc' in the PATH (on common platforms).
As on Linux 'cc' is available as a link to the default compiler (Clang or
GCC), so just use 'cc'.
Signed-off-by: Ed Maste <emaste@freebsd.org>
Message-Id: <20191115155108.39488-1-emaste@freefall.freebsd.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
If there is trailing zero, fdtget adds extra chacarter to the
property value. Thus comparing the expected with the actual
value, an error is emitted.
Signed-off-by: Stefan Mavrodiev <stefan@olimex.com>
Message-Id: <20191111080444.9819-3-stefan@olimex.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
BSD sed requires that an extension is provided to the -i (in-place edit)
flag, which may immediately follow the -i or may be separated by a space -
sed -i .bak and sed -i.bak are equivalent. The extension is optional with
GNU sed, but if provided must immediately follow the -i. Thus, sed -i.bak
behaves identically with both GNU and BSD sed.
Signed-off-by: Ed Maste <emaste@freebsd.org>
Message-Id: <20191114203615.2866-1-emaste@freefall.freebsd.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When applying overlays, we merge symbols from the overlay into the target
tree. At the moment the logic for this assumes all symbols in the overlay
are attached to a node of the form:
/fragment@XXX/__overlay__/relative/path
And will end up applied to the relative/path node under the fragment's
target.
However, this disallows the case of a symbol in the form just:
/fragment@XXX/__overlay__
This does have a pretty obvious sensible meaning: attach the new symbol
directly to the fragment's target, but we don't currently do that.
It's pretty easy to workaround this limitation in one's overlays, but it's
also easy to handle in the overlay applying code, so we might as well
extend it to cover this case.
Reported-by: Christophe Braillon
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This adds a test case to demonstrate some issue seen when applying
overlays using 'fdtoverlay'. It fails with FDT_ERR_NOSPACE:
- with long target path
- symbols in order to use these nodes in possible subsequent overlay.
This is seen with this patch, by running:
$ make check # Reports a failed test
$ ./fdtoverlay -i tests/overlay_base.test.dtb -o out.dtb \
tests/overlay_overlay_long_path.fdoverlay.test.dtb
Failed to apply tests/overlay_overlay_long_path.fdoverlay.test.dtb (-3)
This overlay fails to apply, because dtb size is close to modulo 1024
bytes chunk: utilfdt_read() -> utilfdt_read_err() -> bufsize = 1024.
As there is not much extra space in the blob to resolve symbols (long
target path), it fails with FDT_ERR_NOSPACE. In fdtoverlay, size is :
/* grow the blob to worst case */
blob_len = fdt_totalsize(blob) + total_len;
I can see assumption is made that result should be lower than:
- base fdt size + overlay size. Is there a simple way to find to know
what the final size is?
I'm not sure what the correct fix might be, for such (worst) case?
Similar issue is also seen in u-boot/common/image-fit.c that implements
similar approach (e.g. base fdt size + overlay size).
Signed-off-by: Fabrice Gasnier <fabrice.gasnier@st.com>
Message-Id: <1538553302-1353-1-git-send-email-fabrice.gasnier@st.com>
[dwg: To avoid breaking bisection, I committed this after a fix, so
the "failed" description is no longer accurate]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The typos have been discovered with the "codespell" utility.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20190520081209.20415-1-thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Searching for duplicate names scales O(n^2) with the number of names
added to a fdt, which can cause a noticable slowdown with larger device
trees and very slow CPU cores.
Add FDT_CREATE_FLAG_NO_NAME_DEDUP that allow the caller to trade fdt size
for speed in the creation process.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20190509094122.834-4-npiggin@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
If fdt_add_property or fdt_property_placeholder fail after allocating
a string for the name, they return without freeing that string. This
does not change the structure of the tree, but in very specific cases
it could lead to undesirable space consumption.
Fix this by rolling back the string allocation in this situation.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Message-Id: <20190509094122.834-2-npiggin@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Currently the libfdt based tools (fdtput, fdtget, etc.) and all the
test binaries using libfdt are linked against the static version of libfdt.
That's made it very easy in the past to forget to properly update the
version.lds file which is needed to make functions publicaly accessible
from the shared library.
To avoid problems like that in future, alter the build so that we link and
run the tests against the shared library version of libfdt.
That immediately points out several important symbols that are still
missing from the version.lds, so fix those as well.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
tests.sh has a bunch of shell setup that's sourced in a number of other
scripts. It _doesn't_ actually run a bunch of tests, which is kind of what
the name suggests. So rename it to be more obvious.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This function will append an address range property using parent node's
"#address-cells" and "#size-cells" properties.
It will be used in implementing kdump with kexec_file_load system call
at linux kernel for arm64 once it is merged into kernel tree.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Message-Id: <20190327061552.17170-2-takahiro.akashi@linaro.org>
[dwg: Correct a SEGV error in the testcase]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Python 3 C extensions have suffix containing platform, Python version
and another details in the name so the condition has to be extended.
Signed-off-by: Lumir Balhar <lbalhar@redhat.com>
Message-Id: <20190218164856.23861-5-frenzy@frenzy.cz>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
