Driver model currently only operates after relocation is complete. In this
state U-Boot typically has a small amount of memory available. In adding
support for driver model prior to relocation we must try to use as little
memory as possible.
In addition, on some machines the memory has not be inited and/or the CPU
is not running at full speed or the data cache is off. These can reduce
execution performance, so the less initialisation that is done before
relocation the better.
An immediately-obvious improvement is to only initialise drivers which are
actually going to be used before relocation. On many boards the only such
driver is a serial UART, so this provides a very large potential benefit.
Allow drivers to mark themselves as 'pre-reloc' which means that they will
be initialised prior to relocation. This can be done either with a driver
flag or with a 'dm,pre-reloc' device tree property.
To support this, the various dm scanning function now take a 'pre_reloc_only'
parameter which indicates that only drivers marked pre-reloc should be
bound.
Signed-off-by: Simon Glass <sjg@chromium.org>
Drivers are supposed to be able to close down cleanly. To set a good example,
make sandbox shut down its driver model drivers and remove them before exit.
It may be desirable to do the same more generally once driver model is more
widely-used. This could be done during bootm, before U-Boot jumps to the OS.
It seems far too early to make this change.
Signed-off-by: Simon Glass <sjg@chromium.org>
The root device should be probed just like any other device. The effect of
this is to mark the device as activated, so that it can be removed (along
with its children) if required.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Marek Vasut <marex@denx.de>
At present stdio device functions do not get any clue as to which stdio
device is being acted on. Some implementations go to great lengths to work
around this, such as defining a whole separate set of functions for each
possible device.
For driver model we need to associate a stdio_dev with a device. It doesn't
seem possible to continue with this work-around approach.
Instead, add a stdio_dev pointer to each of the stdio member functions.
Note: The serial drivers have the same problem, but it is not strictly
necessary to fix that to get driver model running. Also, if we convert
serial over to driver model the problem will go away.
Code size increases by 244 bytes for Thumb2 and 428 for PowerPC.
22: stdio: Pass device pointer to stdio methods
arm: (for 2/2 boards) all +244.0 bss -4.0 text +248.0
powerpc: (for 1/1 boards) all +428.0 text +428.0
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Marek Vasut <marex@denx.de>
Reviewed-by: Marek Vasut <marex@denx.de>
There is no point in setting a structure's memory to NULL when it has
already been zeroed with memset().
Also, there is no need to create a stub function for stdio to call - if the
function is NULL it will not be called.
This is a clean-up, with no change in functionality.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Marek Vasut <marex@denx.de>
Rather than reusing the 'reg' property, use an explicit property for the
expected ping value used in testing.
Signed-off-by: Simon Glass <sjg@chromium.org>
Tun on DEBUG in malloc(). This adds code space and slows things down but
for sandbox this is acceptable. We gain the ability to check for memory
leaks in tests.
Signed-off-by: Simon Glass <sjg@chromium.org>
Set up and zero global data before board_init_f() is called so that we can
remove the need for CONFIG_SYS_GENERIC_GLOBAL_DATA.
Signed-off-by: Simon Glass <sjg@chromium.org>
If we are to have driver model before relocation we need to support some
way of calling memory allocation routines.
The standard malloc() is pretty complicated:
1. It uses some BSS memory for its state, and BSS is not available before
relocation
2. It supports algorithms for reducing memory fragmentation and improving
performace of free(). Before relocation we could happily just not support
free().
3. It includes about 4KB of code (Thumb 2) and 1KB of data. However since
this has been loaded anyway this is not really a problem.
The simplest way to support pre-relocation malloc() is to reserve an area
of memory and allocate it in increasing blocks as needed. This
implementation does this.
To enable it, you need to define the size of the malloc() pool as described
in the README. It will be located above the pre-relocation stack on
supported architectures.
Note that this implementation is only useful on machines which have some
memory available before dram_init() is called - this includes those that
do no DRAM init (like tegra) and those that do it in SPL (quite a few
boards). Enabling driver model preior to relocation for the rest of the
boards is left for a later exercise.
Signed-off-by: Simon Glass <sjg@chromium.org>
At present sandbox defines CONFIG_SYS_GENERIC_GLOBAL_DATA, meaning that
the global_data pointer is set up in board_init_f().
If we set up and zero the global data before calling board_init_f() then we
don't need to define CONFIG_SYS_GENERIC_GLOBAL_DATA.
Make this change to simplify the init process.
Signed-off-by: Simon Glass <sjg@chromium.org>
These don't really serve any purpose in the modern age. On the other hand
they show up as annoying control characters in my editor, which then happily
removes them.
I believe we can drop these characters from the file.
Signed-off-by: Simon Glass <sjg@chromium.org>
Commit ddc94378d changed the definition of __kernel_size_t
from unsigned int to unsigned long.
It is true that it fixed warnings on some crosstools
but it increased warnings on the others.
The problem is that we cannot see consistency in terms of
the typedef of __kernel_size_t on M68K architecture.
However, I'd like to suggest to have __kernel_size_t to be
unsigned int again.
Rationale:
[1] Linux Kernel defines __kernel_size_t on M68K as unsigned int.
Let's stick to the Linux's way.
[2] We want to build boards with popular pre-built toolchains,
not the one locally-built by indivisuals.
I think m68-linux-gcc which can be downloaded from www.kernel.org
is the candidate for our _recommended_ toolchains.
With this patch, all the m68k boards can be built without any warnings.
Give it a try with the following crosstools:
https://www.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.6.3/
x86_64-gcc-4.6.3-nolibc_m68k-linux.tar.xz
or
https://www.kernel.org/pub/tools/crosstool/files/bin/x86_64/4.9.0/
x86_64-gcc-4.9.0-nolibc_m68k-linux.tar.xz
(The latter is newer.)
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Simon Glass <sjg@chromium.org>
Cc: Jason Jin <Jason.jin@freescale.com>
The latest GCC is so clever that it reports more warnings
than old ones did:
------------------------------>8------------------------------
board/freescale/m5253demo/flash.c: In function 'flash_get_offsets':
board/freescale/m5253demo/flash.c:65:23: warning: iteration 2047u
invokes undefined behavior [-Waggressive-loop-optimizations]
info->start[k + 1] = info->start[k] + CONFIG_SYS_SST_SECTSZ;
^
board/freescale/m5253demo/flash.c:64:3: note: containing loop
for (k = 0, j = 0; j < CONFIG_SYS_SST_SECT; j++, k++) {
^
------------------------------8<------------------------------
The cause of the warning is like this:
The for statement iterates 2048 times in flash_get_offsets() func.
(Notice CONFIG_SYS_SST_SECT is defined as 2048)
The last iteration does
info->start[2048] = info->start[2047] + CONFIG_SYS_SST_SECTSZ;
causing an undefined behavior.
(Please note the array size of info->start is 2048.
CONFIG_SYS_MAX_FLASH_SECT is defined as 2048 for this board.)
This commit fixes that so as not to overrun the info->start array.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Jason Jin <Jason.jin@freescale.com>
This has been disabled for ARM in initr_scsi since that function was
introduced. However it works fine for me on Cubieboard and Cubietruck (with the
upcoming AHCI glue patch).
I also tested on two random ARM platforms which seem to define CONFIG_CMD_SCSI:
- highbank worked fine (on midway hardware)
- omap5_uevm built OK and I confirmed using objdump that things were as
expected (i.e. the default weak scsi_init nop was used).
While there remove the mismatched comment from the #endif (omitting the comment
seems to be the prevailing style in this file).
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Acked-by: Simon Glass <sjg@chromium.org>
In 73545f75b6 "ahci: wait longer for link" I increased the
timeout to 40ms based on the observed behaviour of a WD disk on a
Cubietruck. Since then Karsten Merker and myself have both
observed timeouts with HGST disks (Karsten on Cubietruck, me on
Cubieboard2). Increasing the timeout to ~175ms fixes this, so go
to 200ms for a bit of headroom.
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Cc: Karsten Merker <merker@debian.org>
Acked-by: Hans de Goede <hdegoede@redhat.com>
There have been /errlog and /reloc_off in the top level .gitignore
since commit 1b4aaffe added it about 7 years ago.
But they are no longer generated.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Add more entries to .mailmap for the canonical names with
50 commits or more.
This commit was generated by the following command:
scripts/mailmapper > tmp; mv tmp .mailmap
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
This tool helps to create/update the mailmap file.
It runs 'git shortlog' internally and searches differently spelled author
names which share the same email address. The author name with the most
commits is asuumed to be a canonical real name. If the number of commits
from the cananonical name is equal to or greater than 'MIN_COMMITS' (=50),
the entry for the cananical name will be output. ('MIN_COMMITS' is used
here because we do not want to create a fat mailmap by adding every author
with only a few commits.)
If there exists a mailmap file specified by the mailmap.file configuration
options or '.mailmap' at the toplevel of the repository, it is used as
a base file.
The base file and the newly added entries are merged together and sorted
alphabetically (but the comment block is kept untouched), and then printed
to standard output.
Usage
-----
scripts/mailmapper
prints the mailmapping to standard output.
scripts/mailmapper > tmp; mv tmp .mailmap
will be useful for updating '.mailmap' file.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
This is the first version of .mailmap created by hand.
Please see "man git-shortlog" for what this commit is trying to do.
Without this file, for example, "git shortlog -n -s" shows as follows:
2693 Wolfgang Denk <------
1002 Stefan Roese <------
811 wdenk <------
808 Mike Frysinger
806 Simon Glass
[snip]
177 Matthias Fuchs
154 stroese <------
153 Timur Tabi
And then, with this file, it shows as follows:
3504 Wolfgang Denk <------
1156 Stefan Roese <------
808 Mike Frysinger
806 Simon Glass
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Stefan Roese <sr@denx.de>
Cc: Wolfgang Denk <wd@denx.de>
Replace the TAGS target name by the actual ctags target name.
Also, add etags target documentation.
Cc: Masahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: Igor Grinberg <grinberg@compulab.co.il>
Commit efcf861 (kbuild: use scripts/Makefile.clean)
refactored the cleaning targets and accidentially replaced the actually
generated "ctags" and "etags" files in the file list by "tags" and "TAGS".
"tags" and "TAGS" are not part of the Makefile build targets and
therefore should not be a part of the list for clean targets.
Substitute the actually generated files instead, to fix the clean
targets behavior.
Cc: Masahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: Igor Grinberg <grinberg@compulab.co.il>
Commit 9e41403 (kbuild: change out-of-tree build)
changed the U-Boot build working directory to the output tree
for the out-of-tree builds.
This broke the {c,e}tags/cscope build targets as TAG_SUBDIRS variable
collected directories based on assumption that the build working
directory is the U-Boot source tree directory.
Fix the {c,e}tags/cscope build targets by adding the $(srctree) prefix.
Also, remove the $(obj) prefix from the etags build target to finish
the $(obj) prefix removal started by the same commit.
Cc: Masahiro Yamada <yamada.m@jp.panasonic.com>
Signed-off-by: Igor Grinberg <grinberg@compulab.co.il>
When this option is enabled, CRLF is treated like LF when importing environments
from text files, which means CRs ('\r') in front of LFs ('\n') are just ignored.
Drawback of enabling this option is that (maybe exported) variables which have
a trailing CR in their content will get imported without that CR. But this
drawback is very unlikely and the big advantage of letting Windows user create
a *working* uEnv.txt too is likely more welcome.
Signed-off-by: Alexander Holler <holler@ahsoftware.de>
This board is close in binary size to one of its hard limits, so disable
SHA256 FIT image support to gain some breathing room.
Signed-off-by: Tom Rini <trini@ti.com>
As I initially suspected overflow in time handling, I took a detailed
look at lib/time.c. This adds comments about units being used, reduces
amount of type casting being done, and makes __udelay() always wait at
least one tick. (Current code could do no delaying at all for short
delays).
Signed-off-by: Pavel Machek <pavel@denx.de>
clang chokes about the concept of having an alias to an
always_inlined function. gcc likely just ignores the always
inlined since binary sizes are equal before and after this
patch. Convert the aliases to weak functions and provide
missing prototypes.
cc: Pavel Herrmann <morpheus.ibis@gmail.com>
Signed-off-by: Jeroen Hofstee <jeroen@myspectrum.nl>
Since no board defines CONFIG_TUNE_PIO this is just dead
code, so remove it.
cc: Pavel Herrmann <morpheus.ibis@gmail.com>
Signed-off-by: Jeroen Hofstee <jeroen@myspectrum.nl>
Just for type checking.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Marek Vasut <marex@denx.de>
Acked-by: Marek Vasut <marex@denx.de>
The function still returns the same value.
The comment block is no longer necessary because our intention is
clear enough by using DIV_ROUND_CLOSEST() macro.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
If CONFIG_OMAP1610 is defined, the code returning the fixed value (26)
is enabled. But this case is covered by the following code.
(CONFIG_SYS_NS16550_CLK + (gd->baudrate * (MODE_X_DIV / 2))) /
(MODE_X_DIV * gd->baudrate)
= (48000000 + (115200 * (16 / 2))) / (16 * 115200)
= 48921600 / 1843200
= 26
The "#ifdef CONFIG_OMAP1610" was added by commit 6f21347d more than
ten years ago. In those days, the divide-and-round was not used.
I guess that is why this weird code was added here.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Tom Rini <trini@ti.com>
Cc: Rishi Bhattacharya <rishi@ti.com>
For the same reason as in commit 50c8d66d, all the remaining
CONFIG_SYS_NAND_PAGE_SIZE in common/spl/spl_nand.c can be replaced
with sizeof(*header).
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Tim Harvey <tharvey@gateworks.com>
Cc: Scott Wood <scottwood@freescale.com>
Cc: Stefano Babic <sbabic@denx.de>
If filename is passed instead of address to ext2load or fatload,
u-boot silently accepts that, and uses 0 for load address and default
filename from environment. That is confusing, display help instead.
Signed-off-by: Pavel Machek <pavel@denx.de>
genimg_print_time uses time_t, but time.h is never included.
Linux gets away with this since types.h includes time.h.
Explicitly include the header file so building on e.g. FreeBSD
also works.
cc: Tom Rini <trini@ti.com>
Signed-off-by: Jeroen Hofstee <jeroen@myspectrum.nl>
upper_32_bits() and lower_32_bits() have been ported into linux/compat.h.
Start use them now in drivers/usb/host/xhci.h.
Signed-off-by: Lijun Pan <Lijun.Pan@freescale.com>
[backport from linux commit 204b885e and 218e180e7]
64 bit processors are becomming more and more popular.
lower_32_bits and upper_32_bits save our labor doing
shifts/manipulations like (u32)(n) and (u32)((n) >> 32).
They are good helpers in both little and big endian cases.
Port these two functions here from Linux:include/linux/kernel.h,
cater the comment message to little/big endian cases.
Later on, developers could include linux/compat.h if they want to
use these two functions.
Signed-off-by: Lijun Pan <Lijun.Pan@freescale.com>
There is only one line diff between LEON2 and LEON3 linker scripts:
- arch/sparc/cpu/leon2/start.o (.text)
+ arch/sparc/cpu/leon3/start.o (.text)
They can be written in the same way:
*/start.o (.text)
So, board/gaisler/grsim_leon2/u-boot.lds
and arch/sparc/cpu/leon3/u-boot.lds
can be merged into arch/sparc/cpu/u-boot.lds.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Daniel Hellstrom <daniel@gaisler.com>
The linker scripts of LEON3
- board/gaisler/grsim/u-boot.lds
- board/gaisler/gr_cpci_ax2000/u-boot.lds
- board/gaisler/gr_ep2s60/u-boot.lds
- board/gaisler/gr_xc3s_1500/u-boot.lds
are the same (except cosmetic differences such as indentation).
This commit merges them into arch/sparc/cpu/leon3/u-boot.lds.
Signed-off-by: Masahiro Yamada <yamada.m@jp.panasonic.com>
Cc: Daniel Hellstrom <daniel@gaisler.com>
Rather than calling do_source directly (which is not officially exported
from cmd_source.c) call 'source' which is exported and requires a little
less code to do so as well.
Signed-off-by: Tom Rini <trini@ti.com>
Simon Glass
Masahiro Yamada
Wolfgang Denk
Pavel Machek
Ian Campbell
Igor Grinberg
Alexander Holler
Tom Rini
Jeroen Hofstee
Lijun Pan