07 Mar, 2014
13 commits
-
Kbuild brought about many advantages for us but a significant
performance regression was reported by Simon Glass.After some discussions and analysis, it turned out
its main cause is in $(call cc-option,...).Historically, U-Boot parses all config.mk
(arch/*/config.mk and board/*/config.mk)
every time descending into subdirectories.
That means cc-options are evaluated over and over again.$(call cc-option,...) is useful but costly.
So we want to evaluate them only in ./Makefile
and spl/Makefile and export compiler flags.This commit changes the build system as follows:
- Modify scripts/Makefile.build to not include config.mk
Instead, add $(PLATFORM_CPPFLAGS) to asflags-y, ccflags-y,
cppflags-y.- Export many variables
Going forward, Kbuild will not parse config.mk files
when it descends into subdirectories.
If we want to set variables in config.mk and use them
in subdirectories, they must be exported.This is the list of variables to get exported:
PLATFORM_CPPFLAGS
CPUDIR
BOARDDIR
OBJCOPYFLAGS
LDFLAGS
LDFLAGS_FINAL
(used in nand_spl/board/*/*/Makefile)
CONFIG_STANDALONE_LOAD_ADDR
(used in examples/standalone/Makefile)
SYM_PREFIX
(used in examples/standalone/Makefile)
RELFLAGS
(used in examples/standalone/Makefile)- Delete CPPFLAGS
This variable has been replaced with PLATFORM_CPPFLAGS- Copy gcclibdir from example/standalone/Makefile
to arch/sparc/config.mk
The reference in CONFIG_STANDALONE_LOAD_ADDR must be
resolved before it is exported.Signed-off-by: Masahiro Yamada
Reported-by: Simon Glass
Acked-by: Simon Glass
Tested-by: Simon Glass [on Sandbox]
Tested-by: Stephen Warren [on Tegra] -
We want to change the build system to include config.mk
only from ./Makefile and spl/Makefile.
We must prepare for that in this commit.$(src) is a moving target and not handy for our purpose.
We must replace it with a fixed path.Signed-off-by: Masahiro Yamada
-
Before this commit, USE_PRIVATE_LIBGCC was defined in
arch-specific config.mk and referenced in
arch/$(ARCH)/lib/Makefile.We are not happy about parsing config.mk again and again.
We have to keep the same behavior with a different way.By adding "CONFIG_" prefix, this macro appears
in include/autoconf.mk, include/spl-autoconf.mk.
(And treating USE_PRIVATE_LIBGCC as CONFIG macro
is reasonable enough.)Tegra SoC family defined USE_PRIVATE_LIBGCC as "yes"
in arch/arm/cpu/arm720t/tegra*/config.mk,
whereas did not define it in arch/arm/cpu/armv7/tegra*/config.mk.It means Tegra enables PRIVATE_LIBGCC only for SPL.
We can describe the same behavior by adding#ifdef CONFIG_SPL_BUILD
# define CONFIG_USE_PRIVATE_LIBGCC
#endifto include/configs/tegra-common.h.
Signed-off-by: Masahiro Yamada
Cc: Tom Warren
Cc: Simon Glass
Acked-by: Stephen Warren -
Many (but not all) of Blackfin boards give -O2 option
to compile under lib/ directory.
That means lib/ should be speed-optimized,
whereas other parts should be size-optimized.We want to keep the same behavior,
but do not want to parse board/*/config.mk again and again.
We've got no choice but to invent a new method.CONFIG_CC_OPTIMIZE_LIBS_FOR_SPEED, if it is enabled,
gives -O2 flag only for building under lib/ directory.Dirty codes which I had marked as "FIX ME"
in board/${BOARD}/config.mk have been deleted.
Instead, CONFIG_CC_OPTIMIZE_LIBS_FOR_SPEED has been
defined in include/configs/${BOARD}.h.Signed-off-by: Masahiro Yamada
Cc: Sonic Zhang -
When importing a checksummed area we need to be told how big the area in
question is so that we know that will match the size of the area which
the checksum is generated against.Reported-by: Pierre AUBERT
Signed-off-by: Tom Rini -
Signed-off-by: Vasili Galka
-
As ppc4xx currently only supports the deprecated nand_spl infrastructure
and nobody seems to have time / resources to port this over to the newer
SPL infrastructure, lets remove NAND booting completely.This should not affect the "normal", non NAND-booting ppc4xx platforms
that are currently supported.Signed-off-by: Stefan Roese
Cc: Wolfgang Denk
Cc: Tirumala Marri
Cc: Matthias Fuchs
Cc: Masahiro Yamada
Cc: Tom Rini
Tested-by: Matthias Fuchs -
There was wrong phy_mask for AR8021 device,
so the AR8021 can't be probed correctly.
Changed it from 0x4fffff to 0x4ffff0.Signed-off-by: Haijun Zhang
-
config.tmp is never generated
Signed-off-by: Masahiro Yamada
Cc: Michal Simek
Acked-by: Michal Simek -
The unit-test for hush's "test -e" currently relies upon being run in
the U-Boot build directory, because it tests for the existence of a file
that exists in that directory.Fix this by explicitly creating the file we use for the existence test,
and deleting it afterwards so that multiple successive unit-test
invocations succeed. This required adding an os.c function to erase
files.Reported-by: Simon Glass
Signed-off-by: Stephen Warren -
The following shell command fails:
if test -z "$x"; then echo "zero"; else echo "non-zero"; fi
(assuming $x does not exist, it prints "non-zero" rather than "zero").
... since "$x" expands to nothing, and the argument is completely
dropped, causing too few to be passed to -z, causing cmd_test() to
error out early.This is because when variable expansions are processed by make_string(),
the expanded results are concatenated back into a new string. However,
no quoting is applied when doing so, so any empty variables simply don't
generate any parameter when the combined string is parsed again.Fix this by explicitly replacing quoting any argument that was originally
quoted when re-generating a string from the already-parsed argument list.This also fixes loss of whitespace in commands such as:
setenv space " "
setenv var " 1${space}${space} 2 "
echo ">>${var}<
Acked-by: Simon Glass
Signed-off-by: Stephen Warren -
Delete the temporary variables that are used to save unit-test results
from the environment after running the test. This prevents polluting
the environment, or growing it too much.Signed-off-by: Stephen Warren
Acked-by: Simon Glass -
For Ethernet/USB RX packets, the ASIX HW pads odd-sized packets so that
they have an even size. Currently, asix_recv() does remove this padding,
and asic_send() adds equivalent padding in the TX path. However, the HW
does not appear to need this packing for TX packets in practical testing
with "ASIX Elec. Corp. AX88x72A 000001" Vendor: 0x0b95 Product 0x7720
Version 0.1. The Linux kernel does no such padding for the TX path.Remove the padding from the TX path:
* For consistency with the Linux kernel.
* NVIDIA has a Tegra simulator which validates that the length of USB
packets sent to an ASIX device matches the packet length value inside
the packet data. Having U-Boot and the kernel do the same thing when
creating the TX packets simplifies the simulator's validation.Cc: Lucas Stach
Cc: Marek Vasut
Cc: Simon Glass
Signed-off-by: Stephen Warren
Acked-by: Simon Glass
Acked-by: Marek Vasut
Tested-by: Marek Vasut
Tested-by: Gerhard Sittig
05 Mar, 2014
27 commits
-
omap_elm.h is a generic header used by OMAP ELM driver for all TI platfoms.
Hence this file should be present in generic folder instead of architecture
specific include folder.
Build tested using: ./MAKEALL -s am33xx -s omap3 -s omap4 -s omap5Signed-off-by: Pekon Gupta
-
omap_gpmc.h is a generic header used by OMAP NAND driver for all TI platfoms.
Hence this file should be present in generic folder instead of architecture
specific include folder.
Build tested using: ./MAKEALL -s am33xx -s omap3 -s omap4 -s omap5Signed-off-by: Pekon Gupta
-
Each SoC platform (AM33xx, OMAP3, OMAP4, OMAP5) has its own copy of GPMC related
defines and declarations scattered in SoC platform specific header files
like include/asm/arch-xx/cpu.h
However, GPMC hardware remains same across all platforms thus this patch merges
GPMC data scattered across different arch-xx specific header files into single
header file include/asm/arch/omap_gpmc.hBuild tested using: ./MAKEALL -s am33xx -s omap3 -s omap4 -s omap5
Signed-off-by: Pekon Gupta
-
OMAP NAND driver can detect Page-size and OOB-size of NAND device from ONFI
params or nand_id[] table. And based on that it defines ECC layout.
This patch
1) removes following board configs used for defining NAND ECC layout
- GPMC_NAND_ECC_LP_x16_LAYOUT (for large page x16 NAND)
- GPMC_NAND_ECC_LP_x8_LAYOUT (for large page x8 NAND)
- GPMC_NAND_ECC_SP_x16_LAYOUT (for small page x16 NAND)
- GPMC_NAND_ECC_SP_x8_LAYOUT (for small page x8 NAND)2) removes unused #defines in common omap_gpmc.h depending on above configs
Build tested using: ./MAKEALL -s am33xx -s omap3 -s omap4 -s omap5
Signed-off-by: Pekon Gupta
-
Currently there are two sets of omap_gpmc.h header files
(a) arch/arm/include/asm/omap_gpmc.h
common header file for all platforms, containing defines and declarations used
by GPMC NAND driver.(b) arch/arm/include/asm/arch-xx/omap_gpmc.h
SoC platform specific header file containing defines like ECC layout.This patch removes platform specific arch-xx/omap_gpmc.c because:
- GPMC hardware engine is common for all SoC platforms hence only (a) is enough
- ECC layout is now defined in omap_nand.c driver itself based on ecc-scheme
selected. Hence all ECC layout declarations in (b) are redundant.Build tested using: ./MAKEALL -s am33xx -s omap3 -s omap4 -s omap5
Signed-off-by: Pekon Gupta
-
README.malta referred to board/malta, but malta has now been moved
within board/imgtec/, so correct the path.Signed-off-by: James Hogan
Cc: Daniel Schwierzeck
Cc: Paul Burton -
Linux MIPS uses asm-generic/int-ll64.h in asm/types.h.
Thus u64 and __u64 are defined as unsigned long long. Port this
over to U-Boot.Signed-off-by: Daniel Schwierzeck
-
There are no source files in board/synopsys/arcangel4/
directory.Signed-off-by: Masahiro Yamada
Cc: Alexey Brodkin -
U-Boot has compelled all boards to have
board/${BOARD}/ or board/${VENDOR}/${BOARD}/ directory.Sometimes it does not seem suitable for some boards,
for example Sandbox. (Is it a board?)And arcangel4 board has nothing to compile
under the board directory.This commit makes the build system more flexible:
If '' is given to the 6th column (=Board name) of boards.cfg,
Kbuild will not descend into the board directory.Signed-off-by: Masahiro Yamada
-
LOGO_BMP was never overwritten by board-specific or
vendor-specific logos.Signed-off-by: Masahiro Yamada
Reported-by: Bo Shen
Tested-by: Bo Shen -
Programs in tools/ directory are usually built for the host.
But some of them (mkimage, dumpimge, gen_eth_addr, etc.) are
useful on the target OS too.Actually, prior to Kbuild, U-Boot could build tools for
the target like follows:$ make _config
$ export CROSS_COMPILE=
$ make HOSTCC=${CROSS_COMPILE}gcc HOSTSTRIP=${CROSS_COMPILE}strip toolsIn Kbuild, we can no longer replace HOSTCC at the command line.
In order to get back that feature, this commit adds "cross-tools" target.
Usage:
Build tools for the host
$ make CROSS_COMPILE= toolsBuild tools for the target
$ make CROSS_COMPILE= cross_toolsBesides, "make cross_tools" strip tools programs because we
generally expect smaller storages on embedded systems.Signed-off-by: Masahiro Yamada
Reported-by: Heiko Schocher
Cc: Wolfgang Denk
Cc: Tom Rini
Tested-by: Heiko Schocher
Acked-by: Heiko Schocher -
The top Makefile must export HOST_TOOLS_ALL to use it
in tools/Makefile.Signed-off-by: Masahiro Yamada
Tested-by: Heiko Schocher
Acked-by: Heiko Schocher -
config.tmp is not there.
Signed-off-by: Masahiro Yamada
-
Ignore
- board/cray/L1/bootscript.{c|image}
- board/matrix_vision/mvblm7/bootscript.img
- board/maxtir_vision/mvsmr/bootscript.imgSigned-off-by: Masahiro Yamada
Cc: Andre Schwarz -
The same outputs are generated with or without -I$(TOPDIR)/board.
I cannot understand why it is necessary. Remove.Signed-off-by: Masahiro Yamada
Cc: Daniel Hellstrom -
The syntax
CROSS_COMIPLE ?=
does not work because config.mk is parsed after
exporting CROSS_COMPILE.Like Linux Kernel's arch/$(ARCH)/Makefile,
we must write as follows:ifeq ($(CROSS_COMPILE),)
CROSS_COMPILE :=
endifSigned-off-by: Masahiro Yamada
-
We had switched to Kbuild so now we can specify
PLATFORM_LIBS/PLATFORM_LIBGCC with relative path.Signed-off-by: Masahiro Yamada
Acked-by: Tom Rini
Acked-by: Simon Glass
Tested-by: Simon Glass -
Update to v3.14-rc4's version of checkpatch.pl. In doing so we drop the
changes to top_of_kernel_tree() as we pass in --no-tree and drop our
changes about MAINTAINERS as that's for reporting checkpatch.pl problems
itself (and upstream has said they'll reword this section to be
clearer).Signed-off-by: Tom Rini
-
Add 64-bit data for memory commands, such as md, mw, mm, cmp. The new
size ".q " is introduced.For 64-bit architecture, 64-bit data is enabled by default, by detecting
compiler __LP64__. It is optional for other architectures.Signed-off-by: York Sun
-
This documentation pertains to the planned implementation of driver model
in U-Boot for each subsystem, but it has not been superseded. It is
probably better to have this documentation in the source code for each
subsystem where possible, so that docbook will pick it up. Where this does
not make sense, new documentation can be placed in some suitable file in
doc/driver-model.Signed-off-by: Simon Glass
-
Now that named GPIO banks are supported, along with a way of obtaining
the status of a GPIO (input or output), we can provide an enhanced
GPIO command for driver model. Where the driver provides its own operation
for obtaining the GPIO state, this is used, otherwise a generic version
is sufficient.Signed-off-by: Simon Glass
-
Convert sandbox over to use driver model GPIOs.
Signed-off-by: Simon Glass
-
Add driver model support for GPIOs. Since existing GPIO drivers do not use
driver model, this feature must be enabled by CONFIG_DM_GPIO. After all
GPO drivers are converted over we can perhaps remove this config.Tests are provided for the sandbox implementation, and are a sufficient
sanity check for basic operation.The GPIO uclass understands the concept of named banks of GPIOs, with each
GPIO device providing a single bank. Within each bank the GPIOs are numbered
using an offset from 0 to n-1. For example a bank named 'b' with 20
offsets will provide GPIOs named b0 to b19.Anonymous GPIO banks are also supported, and are just numbered without any
prefix.Each time a GPIO driver is added to the uclass, the GPIOs are renumbered
accordinging, so there is always a global GPIO numbering order.Signed-off-by: Simon Glass
Signed-off-by: Marek Vasut
Signed-off-by: Pavel Herrmann
Signed-off-by: Viktor Křivák
Signed-off-by: Tomas Hlavacek -
As an example of how to write a uclass and a driver, provide a demo version
of each, accessible through the 'demo' command.To use these with driver model, define CONFIG_CMD_DEMO and CONFIG_DM_DEMO.
The two demo drivers are enabled with CONFIG_DM_DEMO_SIMPLE and
CONFIG_DM_DEMO_SHAPE.Signed-off-by: Simon Glass
Signed-off-by: Marek Vasut
Signed-off-by: Pavel Herrmann
Signed-off-by: Viktor Křivák
Signed-off-by: Tomas Hlavacek -
This command is not required for driver model operation, but can be useful
for testing. It provides simple dumps of internal data structures.Signed-off-by: Simon Glass
Signed-off-by: Marek Vasut
Signed-off-by: Pavel Herrmann
Signed-off-by: Viktor Křivák
Signed-off-by: Tomas Hlavacek
