We should only perform additional iteration steps when needed to
initialize the parent of a device. Other binding errors (such as a
missing driver) should not lead to additional iteration steps.

Unnecessary iteration steps can cause issues when memory is tightly
constrained (such as in the TPL/SPL) since device_bind_by_name()
unconditionally allocates memory for a struct udevice. On the SanCloud
BBE this led to boot failure caused by memory exhaustion in the SPL
when booting from SPI flash.

Signed-off-by: Paul Barker <paul.barker@sancloud.com>

Use dot instead of comma. The fix doesn't affect anything but it is good to
be aligned with used pattern. The first is used only for string size
calculation and the second change is in the comment.

Fixes: db681d4929ca ("net: phy: Add new read ethernet phy id function")
Signed-off-by: Michal Simek <michal.simek@amd.com>

When probing a device fails NULL pointer is returned, and following
devices in uclass list cannot be iterated. Skip to next device on error
instead.

With that the only condition under which these simple iteration
functions return error is when the dm is not initialized at uclass_get
time. This is not all that interesting, change return type to void.

Fixes: 6494d708bf ("dm: Add base driver model support")
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Reviewed-by: Simon Glass <sjg@chromium.org>

This is needed in the platforms that use "ranges" node property for
address translation in their dts for TPL.

Signed-off-by: William Zhang <william.zhang@broadcom.com>
Reviewed-by: Simon Glass <sjg@chromium.org>

The call to device_bind_with_driver_data() passes id->data but if
the entry has no of_match then the id has not been set by the selected
driver.

Normally this passes unnoticed since a previous driver likely had an
of_match value, so the id is set to that. Of course it is not correct
to pass the id->data from a different driver.

With clang-14 the driver ordering is such that the id is never actually
set in the 'bind /usb@1 usb_ether' line in test_bind_unbind_with_node()
thus causing a crash.

Fix this by passing 0 if the of_match for a driver does not exist.

Signed-off-by: Simon Glass <sjg@chromium.org>

There are no longer any platforms which do not enable DM, move this to a
def_bool y and remove the check in the Makefile.

Cc: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Simon Glass <sjg@chromium.org>

The _err variant iterators use the simple iterators without suffix as
basis.

However, there is no user that uclass_next_device_err for iteration,
many users of uclass_first_device_err use it to get the first and
(assumed) only device of an uclass, and a couple that use
uclass_next_device_err to get the device following a known device in the
uclass list.

While there are some truly singleton device classes in which more than
one device cannot exist these are quite rare, and most classes can have
multiple devices even if it is not the case on the SoC's EVB.

In a later patch the simple iterators will be updated to not stop on
error and return next device instead. With this in many cases the code
that expects the first device or an error if it fails to probe may get
the next device instead. Use the _check iterators as the basis of _err
iterators to preserve the old behavior.

Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Reviewed-by: Simon Glass <sjg@chromium.org>

uclass_probe_all uses uclass_first_device/uclass_next_device assigning
the return value.

The interface for getting meaningful error is
uclass_first_device_check/uclass_next_device_check, use it.

Also do not stop iteration when an error is encountered. Probing all
devices includes those that happen to be after a failing device in the
uclass order.

Fixes: a59153dfeb ("dm: core: add function uclass_probe_all() to probe all devices")
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Reviewed-by: Simon Glass <sjg@chromium.org>

A recent change to regmap breaks building of phycore-rk3288 for me. The
difference is only a few bytes. Somehow CI seems to pass, even though it
fails when I run docker locally. But it prevents me from sending any more
pull requests.

In any case this board is clearly near the limit. We could revert the
offending change, but it is needed for sandbox tests.

Instead, add a way to drop the range checks in SPL, since they end up
doing nothing if everything is working as expected.

This makes phycore-rk3288 build again for me and reduces the size of SPL
slightly for a number of boards.

Signed-off-by: Simon Glass <sjg@chromium.org>
Fixes: 947d4f132b4 ("regmap: fix range checks")

On the 32bit ARM sandbox 'dm ut dm_test_devm_regmap' fails with an abort.
This is due to incorrect range checks.

On 32-bit systems the size of size_t and int is both 32 bit. The expression
(offset + val_len) is bound to overflow if offset == -1. Add an overflow
check.

Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Simon Glass <sjg@chromium.org>