Devices which use of-platdata have their own platdata. However, in many
cases the driver will have its own auto-alloced platdata, for use with the
device tree. The ofdata_to_platdata() method converts the device tree
settings to platdata.
With of-platdata we would not normally allocate the platdata since it is
provided by the U_BOOT_DEVICE() declaration. However this is inconvenient
since the of-platdata struct is closely tied to the device tree properties.
It is unlikely to exactly match the platdata needed by the driver.
In fact a useful approach is to declare platdata in the driver like this:
struct r3288_mmc_platdata {
struct dtd_rockchip_rk3288_dw_mshc of_platdata;
/* the 'normal' fields go here */
};
In this case we have dt_platadata available, but the normal fields are not
present, since ofdata_to_platdata() is never called. In fact driver model
doesn't allocate any space for the 'normal' fields, since it sees that there
is already platform data attached to the device.
To make this easier, adjust driver model to allocate the full size of the
struct (i.e. platdata_auto_alloc_size from the driver) and copy in the
of-platdata. This means that when the driver's bind() method is called,
the of-platdata will be present, followed by zero bytes for the empty
'normal field' portion.
A new DM_FLAG_OF_PLATDATA flag is available that indicates that the platdata
came from of-platdata. When the allocation/copy happens, the
DM_FLAG_ALLOC_PDATA flag will be set as well. The dtoc tool is updated to
output the platdata_size field, since U-Boot has no other way of knowing
the size of the of-platdata struct.
Signed-off-by: Simon Glass <sjg@chromium.org>
When swig is not available, we can still build correctly. So make this
optional. Add a comment about how to enable this build.
Signed-off-by: Simon Glass <sjg@chromium.org>
Add a Python version of the libfdt library which contains enough features to
support the dtoc tool. This is only a very bare-bones implementation. It
requires the 'swig' to build.
Signed-off-by: Simon Glass <sjg@chromium.org>
This tool can produce C struct definitions and C platform data tables.
This is used to support the of-platdata feature.
Signed-off-by: Simon Glass <sjg@chromium.org>
This Python library provides a way to access the contents of the device
tree. It uses fdtget, so is inefficient for larger device tree files.
Signed-off-by: Simon Glass <sjg@chromium.org>
We should not be returning -1 as an error code. This can mask a situation
where we run out of space adding things to the FIT. By returning the correct
error in this case (-ENOSPC) it can be handled by the higher-level code.
This may fix the error reported by Tom Van Deun here:
https://www.mail-archive.com/u-boot@lists.denx.de/msg217417.html
although I am not sure as I cannot actually repeat it.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reported-by: Tom Van Deun <tom.vandeun@wapice.com>
Reviewed-by: Teddy Reed <teddy.reed@gmail.com>
The error code may provide useful information for debugging. Add it to the
error string.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Teddy Reed <teddy.reed@gmail.com>
Update the error-handling code for -A, -C and -O to show a list of valid
options when an invalid one is provided.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reported-by: Vinoth Eswaran <evinoth1206@gmail.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
The existing error code only displays image types which are claimed by a
particular U_BOOT_IMAGE_TYPE() driver. But this does not seem correct. The
mkimage tool should support all image types, so it makes sense to allow
creation of images of any type with the tool.
When an incorrect image type is provided, use generic code to display the
error.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
Add a generic function which can display a list of items in any category.
This will allow displaying of images for the -A, -C, -O and -T flags. At
present only -T is supported.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
The fit_write_images() function incorrectly uses the long name for the
architecture. This cannot be parsed with the FIT is read. Fix this by using
the short name instead.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
There is no need to set params.fit_image_type while parsing the arguments.
It is set up later anyway.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
When auto-fit is used, it is not valid to create a FIT without an image
file. Add a check for this to avoid a very confusing error message later
("Can't open (null): Bad address").
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
There is a special case in the code when auto-fit is used. Add a comment to
make it easier to understand why this is needed.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Joe Hershberger <joe.hershberger@ni.com>
The default image type is supposed to be IH_TYPE_KERNEL, as set in the
'params' variable. Honour this with auto-fit also.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
The following python error:
Traceback (most recent call last):
File "./tools/patman/patman", line 144, in <module>
series = patchstream.FixPatches(series, args)
File "./tools/patman/patchstream.py", line 477, in FixPatches
commit = series.commits[count]
IndexError: list index out of range
is seen when:
- 'END' is missing in those tags
- those tags are put in the last part in a commit message
- the commit is not the last commit of the series
Add testing logic to see if a new commit starts.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
'Series-changes' uses blank line to indicate its end. If that is
missing, series internal state variable 'in_change' may be wrong.
Correct its state.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
If 'END' is missing in a 'Cover-letter' section, and that section
happens to show up at the very end of the commit message, and the
commit is the last commit of the series, patman fails to generate
cover letter for us. Handle this in CloseCommit of patchstream.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
'Cover-letter', 'Series-notes' and 'Commit-notes' tags require an
'END' to be put at the end of its section. If we forget to put an
'END' in those sections, and these sections are followed by another
patman tag, patman generates incorrect patches. This adds codes to
handle such scenario.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Like other patman tags, use a new variable cover_match to indicate
a match for 'Cover-letter'.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Acked-by: Simon Glass <sjg@chromium.org>
Because a gpimage cannot be detected, a false
GP header is printed instead of checking
for further image types.
Move gpimage as last to be linked, letting check
all other image types and printing a GP header just
in case no image is detected.
Signed-off-by: Stefano Babic <sbabic@denx.de>
When building a FIT with external data (-E), U-Boot proper may require
absolute positioning for executing the external firmware. To acheive this
use the (-p) switch, which will replace the amended 'data-offset' with
'data-position' indicating the absolute position of external data.
It is considered an error if the requested absolute position overlaps with the
initial data required for the compact FIT.
Signed-off-by: Teddy Reed <teddy.reed@gmail.com>
There are some cases where config options are moved, but they are
ripped off at the final savedefconfig stage:
- The moved option is not user-configurable, for example, due to
a missing prompt in the Kconfig entry
- The config was not defined in the original config header despite
the Kconfig specifies it as non-bool type
- The config define in the header contains reference to another
macro, for example:
#define CONFIG_CONS_INDEX (CONFIG_SYS_LPC32XX_UART - 2)
The current moveconfig does not support recursive macro expansion.
In these cases, the conversion is very likely to be an unexpected
result. That is why I decided to display the log in yellow color
in commit 5da4f857be ("tools: moveconfig: report when CONFIGs are
removed by savedefconfig").
It would be nice to display the list of suspicious boards when the
tool finishes processing. It is highly recommended to check the
defconfigs once again when this message is displayed.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Since commit 1d085568b3 ("tools: moveconfig: display log atomically
in more readable format"), the function color_text() is clever enough
to exclude LF from escape sequences. Exploit it for removing the
"for" loops from Slots.show_failed_boards().
Also, display "(the list has been saved in moveconfig.failed)" if
there are failed boards.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
The subprocess.Popen() does not change the child process's working
directory if cwd=None is given. Let's exploit this fact to refactor
the source directory handling.
We no longer have to pass "-C <reference_src_dir>" to the sub-process
because self.current_src_dir tracks the source tree against which we
want to run defconfig/autoconf.
The flag self.use_git_ref is not necessary either because we can know
the current state by checking whether the self.current_src_dir is a
valid string or None.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
The class WorkDir can be used in a very generic way, but currently
it is only used for containing a reference source directory.
This commit changes it for a more dedicated use. The move_config
function can be more readable by enclosing the git-clone and git-
checkout in the class constructor.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
When moving an integer type option with default value 1, the tool
moves configs with the same value as the default (, and then removed
by the later savedefconfig). This is a needless operation.
The KconfigParser.parse_one_config() should compare the config after
the "=y -> =1" fixup.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
The sed script, tools/scripts/define2mk.sed, converts config defines
from C headers into include/autoconf.mk for the use in Makefiles.
I found the tool adds quotes around negative integer values.
For example, at the point of the v2016.07-rc1 tag,
include/configs/microblaze-generic.h defines
#define CONFIG_BOOTDELAY -1 /* -1 disables auto-boot */
Because it is an integer option, it should be converted to:
CONFIG_BOOTDELAY=-1
But, the script actually converts it to:
CONFIG_BOOTDELAY="-1"
This is a fatal problem for the tools/moveconfig.py because it parses
include/autoconf.mk for the config defines from the board headers.
CONFIG_BOOTDELAY="-1" is considered as a string type option and it
is dropped due to the type mismatch from the entry in Kconfig.
This commit fixes the script so that the tools/moveconfig.py can
correctly convert integer options with a negative value.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
This option allows the 'make autoconf.mk' step to run against a former
repo state, while the savedefconfig step runs against the current repo
state. This is convenient for the case where something in the Kconfig
has changed such that the defconfig is no longer complete with the new
Kconfigs. This feature allows the .config to be built assuming those old
Kconfigs, but then savedefconfig based on the new state of the Kconfigs.
If in doubt, always specify this switch. It will always do the right
thing even if not required, but if it was required and you don't use it,
the moved configs will be incorrect. When not using this switch, you
must very carefully evaluate that all moved configs are correct.
Signed-off-by: Joe Hershberger <joe.hershberger@ni.com>
Reviewed-by: Masahiro Yamada <yamada.masahiro@socionext.com>
The old color blends in with similar messages and makes them not stand
out.
Signed-off-by: Joe Hershberger <joe.hershberger@ni.com>
Reviewed-by: Masahiro Yamada <yamada.masahiro@socionext.com>
The Slot.poll() method is already complicated and a new feature
we are going to add will make it more difficult to understand
the execution flow.
Refactor it with helper methods, .handle_error(), .do_defconfig(),
.do_autoconf(), .do_savedefconfig, and .update_defconfig().
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
I found "tools/moveconfig -s" might be useful for defconfig re-sync.
I could optimize it for re-sync if I wanted, but I do not want to
make the code complex for this feature.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Now, this tools invokes "make savedefconfig" only when it needs to
do so, but there might be cases where a user wants the tool to do
savedefconfig forcibly, for example, some defconfigs were already
out of sync and the user wants to fix it as well.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
There are various factors that determine if the given defconfig is
updated, and it is probably what users are more interested in.
Show the log when the defconfig is updated. Also, copy the file
only when the file content was really updated to avoid changing
the time stamp needlessly.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
This is a rare case, but there is still possibility that some CONFIG
is moved to the .config, but it is removed by "make savedefconfig".
(For example, it happens when the specified CONFIG has no prompt in
the Kconfig entry, i.e. it is not user-configurable.)
It might be an unexpected case. So, display the log in this case
(in yellow color to gain user's attention if --color option is given).
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Now, "make savedefconfig" does not always happen. Display the log
when it happens.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
If no CONFIG option is moved to the .config, no need to sync the
defconfig file. This accelerates the processing by skipping
needless "make savedefconfig".
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Move similar code to finish() function.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Before this commit, the log was displayed in the format:
<defconfig_name> : <action1>
<defconfig_name> : <action2>
<defconfig_name> : <action3>
When we move multiple CONFIGs at the same time, we see as many
<defconfig_name> strings as actions for every defconfig, which is
redundant information.
Moreover, since normal log and error log are displayed separately,
Messages from different threads could be mixed, like this:
<foo> : <action1>
<foo> : <action2>
<bar> : <action1>
<bar> : <action2>
<foo> : <error_log>
This commit makes sure to call "print" once a defconfig, which
enables atomic logging for each defconfig. It also makes it
possible to refactor the log format as follows:
<foo_defconfig>
<action1>
<action2>
<error_log>
<bar_defconfig>
<action1>
<action2>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
This will help further improvement/clean-up.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
The paths to .config, include/autoconf.mk, include/config/auto.conf
are not changed during the defconfig walk. Compute them only once
when a new class instance is created.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
We still pass the input file with CONFIG name, type, default value
in each line, but the last two fields are just ignored by the tool.
So, let's deprecate the input file and allow users to give CONFIG
names directly from the command line. The types and default values
are automatically detected and handled nicely by the tool.
Going forward, we can use this tool more easily like:
tools/moveconfig.py CONFIG_FOO CONFIG_BAR
Update the documentation and fix some typos I noticed while I was
working on.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Now types and defalut values given by the input file are just
ignored. Delete unnecessary code.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>
Commit 96464badc7 ("moveconfig: Always run savedefconfig on the
moved config") changed the work flow of this tool a lot from the
original intention when this tool was designed first.
Since then, before running this tool, users must edit the Kconfig to
add the menu entries for the configs they are moving. It means users
had already specified the type and the default value for each CONFIG
via its Kconfig entry. Nevertheless, users are still required to
dictate the same type and the default value in the input file. This
is tedious to use. So, my idea here is to deprecate the latter.
Before moving forward with it, there is one issue worth mentioning;
since the savedefconfig re-sync was introduced, this tool has not
been able to move bool options with "default y". Joe sent a patch
to solve this problem about a year ago, but it was not applied for
some reasons. Now, he came back with an updated patch, so this
problem will be fixed soon.
For other use cases, I see no reason to require redundant dictation
in the input file. Instead, the tool can know the types and default
values by parsing the .config file.
This commit changes the tool to use the CONFIG names, but ignore the
types and default values given by the input file.
This commit also fixes one bug. Prior to this commit, it could not
move an integer-typed CONFIG with value 1.
For example, assume we are moving CONFIG_CONS_INDEX. Please note
this is an integer type option.
Many board headers define this CONFIG as 1.
#define CONFIG_CONS_INDEX 1
It will be converted to
CONFIG_CONS_INDEX=y
and moved to include/autoconf.mk, by the tools/scripts/define2mk.sed.
It will cause "make savedefconfig" to fail due to the type conflict.
This commit takes care of it by detecting the type and converting the
CONFIG value correctly.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Joe Hershberger <joe.hershberger@ni.com>