Rasmus Villemoes <ravi@prevas.dk> says:

There are quite a few places where we allocate X+1 bytes, initialize
the first X bytes via memcpy() and then set the last byte to 0.

The kernel has a helper for that, kmemdup_nul(). Introduce a similar
one, and start making use of it in a few places. Also the existing
memdup() helper can be put to more use.

There are lots more places one could modify. But for code shared with
host tools, one would need to do some refactoring, putting memdup()
and memdup_nul() in their own str-util.c TU which could then also be
included in the tools build.

Link: https://lore.kernel.org/r/20260421075439.16696-1-ravi@prevas.dk

Use memdup() instead of open-coding it.

In the dm_setup_inst() case, there was never any reason to use
calloc(), as the whole allocation is definitely initialized via the
immediately following memcpy().

Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Rasmus Villemoes <ravi@prevas.dk>

Casey Connolly <casey.connolly@linaro.org> says:

This series implements various improvements to Linux header
compatibility, largely in preparation for a full port of Linux CCF but
many of these changes would also be helpful when porting other drivers.

Beside the basic header/compat stuff there are a few larger patches:

Patch 1 adds the "%pOF" format specifier to vsprintf, this behaves the
same as it does in Linux printing the name of the ofnode, but notably it
expects an ofnode pointer rather than a device_node.

Patch 2 adds an option to skip doing a full DM scan pre-relocation.
Some platforms like Qualcomm don't actually need devices to be probed
prior to relocation, it is also quite slow to scan the entire FDT before
caches are up. This option gets us to main loop 30-50% faster.

Unfortunately it isn't possible to totally skip DM since U-Boot will
panic if it can't find a serial port, but the serial uclass code will
bind the serial port itself by reading /chosen/stdout-path, however any
dependencies like clocks won't be found so this should only be enabled
if the serial driver gracefully handles missing clocks.

Patch 3 adds [k]strdup_const(), this works the same as the Linux version
saving a small amount of memory by avoiding duplicating strings stored
in .rodata, this is particularly useful for CCF.

Patch 4 adds 64-bit versions of some 32-bit ofnode utilities functions,
making it possible to parse 64-bit arrays.

Patch 6 provides a simple implementation of kref, this will be used
by CCF.

Patch 9 adds devm_krealloc() support to devres, it relies on storing
allocation sizes in the devres struct which will add a small overhead.

Link: https://lore.kernel.org/r/20260401-casey-ccf-compat-v2-0-414d5b7f040b@linaro.org

These are similar to their Linux counterparts, adding helpers
for reading arrays of 64-bit values with of_access and fdtdec
implementations.

Signed-off-by: Casey Connolly <casey.connolly@linaro.org>

Rework dev_phys_to_bus() and dev_bus_to_phys() to respect the size
of the area specified in dma-ranges DT property. The area outside
of ranges is remapped 1:1, while the area in the ranges is remapped
according to the description in the dma-ranges property.

Adjust the test to test the area within the remapped range, not area
outside the remapped range, which was incorrect.

Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Reviewed-by: Markus Schneider-Pargmann <msp@baylibre.com>

As exposed by "make randconfig", we have an issue with the dependencies
for REGMAP (and xPL variants). As this is a library function, it should
always be selected and not depended on by other functionality. This is
largely done correctly today, so just correct the few outliers.

Acked-by: Anshul Dalal <anshuld@ti.com>
Signed-off-by: Tom Rini <trini@konsulko.com>

Simon Glass <sjg@chromium.org> says:

Each arch does something slightly different before booting the OS. Some
archs even do different things depending on the CPU type.

It is quite hard to know what actually happens in the final milliseconds
before the OS boot.

This series attempts to start cleaning up U-Boot in this area.

The basic intent is to create a new bootm_final() function which can be
called by all archs. It provides some flags for a couple of necessary
variations but otherwise it is generic.

All architectures are converted over to use this new function.

board_quiesce_devices() is moved into bootm_final() so that all archs
benefit from it.

This series fixes a bug in device_remove() is fixed where removing a
parent with specialised flags (e.g. DM_REMOVE_ACTIVE_ALL) could leave
children activated, since they do not match the flags. This fixes is
needed to avoid bootm_final() causing test failures on sandbox.

Future work could take this a little further:
- Convert EFI loader to use the same function
- Improve comments for cleanup_before_linux() across architectures
- Support fake-run tracing on all archs

Link: https://lore.kernel.org/r/20260306023638.2678886-1-sjg@chromium.org

When dm_remove_devices_active() removes devices using specialised flags
like DM_REMOVE_ACTIVE_ALL, a parent device may match (e.g. MMC has
DM_FLAG_OS_PREPARE) while its children do not. This deactivates the
parent but leaves children activated, an inconsistent state.

Later, when uclass_destroy() calls device_remove() with DM_REMOVE_NORMAL
on the already-deactivated parent, it returns early without touching the
children. The subsequent device_unbind() then fails because the children
are still activated.

Fix this by dropping only the DM_REMOVE_ACTIVE_ALL requirement for child
removal when the parent is being removed. This ensures children are
removed along with their parent, while still preserving other flags like
DM_REMOVE_NON_VITAL so that vital devices remain protected.

Signed-off-by: Simon Glass <simon.glass@canonical.com>

Move the flags_remove() call before device_chld_remove() and save the
result in a separate variable. This is just a refactoring with no
behaviour change, preparing for the next commit which needs to know
whether the parent will be removed before deciding how to remove its
children.

Signed-off-by: Simon Glass <simon.glass@canonical.com>

The ofnode_to_fdt() function may return a NULL pointer in multiple cases.
Or, this function's return value is often passed directly to functions such
as fdt_getprop() which end up dereferencing it, thus causing a NULL pointer
exception.

Don't allow ofnode_to_fdt() to return NULL, to avoid a NULL pointer
dereference.

Reviewed-by: Raphaël Gallais-Pou <raphael.gallais-pou@foss.st.com>
Signed-off-by: Romain Gantois <romain.gantois@bootlin.com>
Reviewed-by: Simon Glass <simon.glass@canonical.com>