The default format for arm64 Linux kernels is the "Image" format,
described in Documentation/arm64/booting.txt. This, along with an
optional gzip compression on top is all that is generated by default.
The Image format has a magic number within the header for verification,
a text_offset where the Image must be run from, an image_size that
includes the BSS and reserved fields.
This does not support automatic detection of a gzip compressed image.
Signed-off-by: Tom Rini <trini@ti.com>
This file has code in three different categories:
- Command processing
- OS-specific boot code
- Locating images and setting up to boot
Only the first category really belongs in a file called cmd_bootm.c.
Leave the command processing code where it is. Split out the OS-specific
boot code into bootm_os.c. Split out the other code into bootm.c
Header files and extern declarations are tidied but otherwise no code
changes are made, to make it easier to review.
Signed-off-by: Simon Glass <sjg@chromium.org>
run_command() returns 0 for success, 1 for failure. Fix places which
assume that failure is indicated by a negative return code.
Signed-off-by: Thomas Betker <thomas.betker@rohde-schwarz.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Tested-by: Stefan Roese <sr@denx.de>
make the use of legacy image format configurable through
the config define CONFIG_IMAGE_FORMAT_LEGACY.
When relying on signed FIT images with required signature check
the legacy image format should be disabled. Therefore introduce
this new define and enable legacy image format if CONFIG_FIT_SIGNATURE
is not set. If CONFIG_FIT_SIGNATURE is set disable per default
the legacy image format.
Signed-off-by: Heiko Schocher <hs@denx.de>
Cc: Simon Glass <sjg@chromium.org>
Cc: Lars Steubesand <lars.steubesand@philips.com>
Cc: Mike Pearce <mike@kaew.be>
Cc: Wolfgang Denk <wd@denx.de>
Cc: Tom Rini <trini@ti.com>
Cc: Michal Simek <monstr@monstr.eu>
Acked-by: Simon Glass <sjg@chromium.org>
We say we have an XIP (in this case, image loaded at desired execution
address) when the image header has been offset in the load. It's
possible that in some cases executing the header is non-fatal but that's
not true in many other cases.
Signed-off-by: Tom Rini <trini@ti.com>
This patch adds support for the Android boot-image format. The header
file is from the Android project and got slightly alterted so the struct +
its defines are not generic but have something like a namespace. The
header file is from bootloader/legacy/include/boot/bootimg.h. The header
parsing has been written from scratch and I looked at
bootloader/legacy/usbloader/usbloader.c for some details.
The image contains the physical address (load address) of the kernel and
ramdisk. This address is considered only for the kernel image.
The "second image" defined in the image header is currently not
supported. I haven't found anything that is creating this.
v3 (Rob Herring):
This is based on http://patchwork.ozlabs.org/patch/126797/ with the
following changes:
- Rebased to current mainline
- Moved android image handling to separate functions in
common/image-android.c
- s/u8/char/ in header to fix string function warnings
- Use SPDX identifiers for licenses
- Cleaned-up file source information:
android_image.h is from file include/boot/bootimg.h in repository:
https://android.googlesource.com/platform/bootable/bootloader/legacy
The git commit hash is 4205b865141ff2e255fe1d3bd16de18e217ef06a
usbloader.c would be from the same commit, but it does not appear
to have been used for any actual code.
v4:
- s/andriod/android/
- Use a separate flag ep_found to track if the entry point has been set
rather than using a magic value.
Cc: Wolfgang Denk <wd@denx.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Rob Herring <robh@kernel.org>
Reviewed-by: Tom Rini <trini@ti.com>
Reviewed-by: Lukasz Majewski <l.majewski@samsung.com>
The LZO decompressor wasn't initializing the maximum output size, which
meant it would fail to decompress most of the time.
Reported-by: Matthias Weißer <weisserm@arcor.de>
Signed-off-by: Kees Cook <keescook@chromium.org>
Tested-by: Matthias Weißer <weisserm@arcor.de>
Acked-by: Simon Glass <sjg@chromium.org>
do_bootm_standanlone() calls ntohl(images->ep) which is wrong because
endianess conversion has already been done:
do_bootm()
\-do_bootm_states()
+-bootm_find_os()
| \-images.ep = image_get_ep();
| \-uimage_to_cpu(hdr->ih_ep);
\-boot_selected_os()
\-do_bootm_standanlone()
Without this conversion the code works correctly at least on AT91SAM9G45.
On big endian systems there should be no difference after applying this
patch because uimage_to_cpu(x) and ntohl(x) both expand to 'x'.
Signed-off-by: Christian Eggers <ceggers@gmx.de>
For standalone images, bootm had a special case where the OS boot function
was NULL but did actually exist. It was just called manually.
This was removed by commit 35fc84fa which checks for the non-existence of
this function before the special case is examined.
There is no obvious reason why standalone is handled with a special case.
Adjust the code so that standalone has a normal OS boot function. We still
need a special case for when the function returns, but at least we can
avoid the main problem.
This is intended to fix the reported:
ERROR: booting os 'U-Boot' (17) is not supported
but needs testing.
Signed-off-by: Simon Glass <sjg@chromium.org>
The next version VxWorks adopts device tree (for PowerPC and ARM) as its hardware
description mechanism. For PowerPC, the boot interface conforms to
the ePAPR standard, which is:
void (*kernel_entry)(ulong fdt_addr,
ulong r4 /* 0 */,
ulong r5 /* 0 */,
ulong r6 /* EPAPR_MAGIC */, ulong r7 /* IMA size */,
ulong r8 /* 0 */, ulong r9 /* 0 */)
For ARM, the boot interface is:
void (*kernel_entry)(void *fdt_addr)
Signed-off-by: Miao Yan <miao.yan@windriver.com>
[trini: Fix build error when !CONFIG_OF_FDT is set, typo on PowerPC,
missing extern ft_fixup_num_cores]
Signed-off-by: Tom Rini <trini@ti.com>
do_bootm_vxworks now is available under the configuration option
CONFIG_BOOTM_VXWORKS, thus aligned with other operating systems
that supported by bootm command. The bootvx command still depneds
on CONFIG_CMD_ELF.
Signed-off-by: Miao Yan <miao.yan@windriver.com>
The documentation suggested the arguments where passed over r3-r6
while the code below simply does that over r0-r3.
Cc: Kumar Gala <galak@kernel.crashing.org>
In commit "5c427e4: use BOOTM_STATE_OS_CMDLINE flag for plain bootm"
and "3d187b3: Only pass BOOTM_STATE_OS_CMDLINE on PowerPC/MIPS",
BOOTM_STATE_OS_CMDLINE was added to do_bootm for PowerPC and MIPS. This
breaks other OSes (vxworks, netbsd, plan9,...) that don't support
subcommand processing, e.g. they all contain the following code in their
do_bootm_xxx():
if (flag & BOOTM_STATE_OS_PREP)
return 0;
if ((flag != 0) && (flag != BOOTM_STATE_OS_GO))
return 1;
which will result a "subcommand not supported" error.
This patch changes the above logic to:
/* if not go command, pretend everything to be OK */
if (flag != BOOTM_STATE_OS_GO)
return 0;
Signed-off-by: Miao Yan <miao.yan@windriver.com>
In 5c427e4 we pass BOOTM_STATE_OS_CMDLINE as part of the bootm states to
run, on all arches. However, this is only valid / useful on PowerPC and
MIPS, and causes a problem on ARM where we specifically do not use it.
Rather than make this state fake pass like we do for GO on some arches
(which need updating to use the GO state), we should just not pass
CMDLINE except when it may be used, like before.
Tested-by: Dan Murphy <dmurphy@ti.com>
Signed-off-by: Tom Rini <trini@ti.com>
Netconsole calls eth_halt() before giving control to another operating
system.
But the state machine of netconsole don't take it into account.
Thus, netconsole calls network functions of an halted network device,
making the whole system freeze.
Rather than modifying the state machine of netconsole, we just unregister
the current network device before booting. It does work because
nc_send_packet() verifies that the current network device is not null.
Signed-off-by: Frédéric Leroy <fredo@starox.org>
A plain bootm used to call the architecture specific boot function with
no flags, but was modified by commit 35fc84fa "Refactor the bootm
command to reduce code duplication" to call the architecture specific
boot function multiple times with various flags in sequence. The
BOOTM_STATE_OS_CMDLINE flag was not used, indeed it seems that at least
ARM prepares the command line on BOOTM_STATE_OS_PREP. However on MIPS
since commit 59e8cbdb "MIPS: bootm: refactor initialisation of kernel
cmdline" the command line is not prepared in response to a
BOOTM_STATE_OS_PREP flag, only on BOOTM_STATE_OS_CMDLINE or a call with
no flags. The end result is that a combination of those 2 commits leads
to MIPS boards booting kernels with no command line arguments.
An extra invocation of the architecture specific boot function with
BOOTM_STATE_OS_CMDLINE fixes this.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
While nothing presently examines the destination size, it should at
least be correct so that future users of sys_mapmem() will not be
surprised. Without this, it might be possible to overflow memory.
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Simon Glass <sjg@chromium.org>
Correct the following warnings found with sandbox when compression
is enabled.
cmd_bootm.c: In function 'bootm_load_os':
cmd_bootm.c:443:11: warning: passing argument 4 of 'lzop_decompress' from incompatible pointer type [enabled by default]
/usr/local/google/c/cosarm/src/third_party/u-boot/files/include/linux/lzo.h:31:5: note: expected 'size_t *' but argument is of type 'uint *'
cmd_ximg.c: In function 'do_imgextract':
cmd_ximg.c:225:6: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
cmd_ximg.c:225:14: warning: 'hdr' may be used uninitialized in this function [-Wuninitialized]
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Kees Cook <keescook@chromium.org>
This function is only defined if CONFIG_SILENT_CONSOLE is set and
CONFIG_SILENT_U_BOOT_ONLY is not set, the call to it should be based
on the same conditions.
Signed-off-by: Paul B. Henson <henson@acm.org>
Acked-by: Simon Glass <sjg@chromium.org>
At present the console for linux is silent if the U-Boot console is silent,
unless CONFIG_SILENT_U_BOOT_ONLY is set. I wonder if a better way would be
to have an environment variable to control this? Then we can control the
verbosity from scripts, and set the variable to 'no' for those boards that
want Linux to boot with console output.
Signed-off-by: Simon Glass <sjg@chromium.org>
Before the bootm refactor, fixup_silent_linux() was done only in the
monolithic bootm case, not in the subcommand case. With the refactor, it
is done always, which is good. Unfortunately it is done too late, since it
is the PREP or CMDLINE stages that set up the command line for Linux.
Move fixup_silent_linux() into the LOADOS stage, which is where we find
out the OS being used, and can thus decide whether to perform this step.
Signed-off-by: Simon Glass <sjg@chromium.org>
A recent bootm fix left the error path incomplete. If CONFIG_TRACE is
set it may still not be a supported command, so cover that with the
unsupported subcommand print. Once we handle BOOTM_STATE_OS_GO, we can
just move into the error handler itself, no need for a goto there.
Signed-off-by: Simon Glass <sjg@chromium.org>
[trini: Update slightly based on Simon's changes to also cover
CONFIG_TRACE/BOOTM_STATE_FAKE_OS_GO]
Signed-off-by: Tom Rini <trini@ti.com>
If a specific configuraion is selected by the bootm command, e.g. with
'bootm 84000000#recoveryconf' we must honour this for not just the kernel,
but also the ramdisk and FDT.
In the conversion to using a common fit_image_load() function for loading
images from FITs (commits a51ec63 and 53f375f) this feature was lost.
Reinstate it by passing the selected configuration back from
fit_image_load() to boot_get_kernel(), then use this configuration
(which is stored in images->fit_uname_cfg) in both boot_get_ramdisk()
and boot_get_fdt().
Signed-off-by: Simon Glass <sjg@chromium.org>
The OS function is now always called with the PREP stage. Adjust the
remaining bootm OS functions to deal with this correctly.
Signed-off-by: Simon Glass <sjg@chromium.org>
With ELDK4.2 we were getting a warning that load_end may be used
uninitialized in calling lmb_reserve. This could not be the case,
however. If we re-order the checks (and make them slightly clearer as
well) the warning goes away. bootm_load_os may only return 0 on
success, BOOTM_ERR_OVERLAP in a non-fatal overlap (already covered in
comments) or a fatal BOOTM_ERR that is covered in the error handler.
Signed-off-by: Tom Rini <trini@ti.com>
As a zImage does not have a U-Boot header, we cannot really do what
BOOTM_STATE_FINDOTHER does, exactly. Break the ramdisk/fdt portions of
bootm_find_other into bootm_find_ramdisk/fdt which can be called in both
cases.
Signed-off-by: Tom Rini <trini@ti.com>
In the recent bootm refactor, the PREP stage was missing in the bootz
command. This causes unpredictable behaviour.
The use of a local variable means that the reset of cmd_bootm.c does not
in fact use the same image structure, so remove this.
Also manually set the OS type to Linux, since this is the only possibility
at present, and we need to select the right boot function.
Signed-off-by: Simon Glass <sjg@chromium.org>
This function has no prototype in the headers and passes void * around, thus
requiring several casts. Tidy this up.
- Add new patch to clean up bootz_setup() function
Signed-off-by: Simon Glass <sjg@chromium.org>
The original bootm code (before commit 35fc84f) did not check for a valid
boot function in the subcommand case, which was incorrect.
This check was introduced in all cases, but in fact we should only check
for the function when we need it. Otherwise in some cases the check fires
before the OS type is known.
Signed-off-by: Simon Glass <sjg@chromium.org>
With the move of the interrupt code to earlier in the sequence, we
exposed a problem where the interrupts are disabled at each bootm
stage. This is not correct - it should be done only once. Let's disable
interrupts in the LOAD stage. Put the code in a function for clarity.
Also, bootz lost its interrupt code altogether, so reinstate it.
Signed-off-by: Simon Glass <sjg@chromium.org>
With the do_bootm_states re-organization, we have the call to any
potential sub-commands in a single spot. If one fails, we can then stop
right there and return to the caller. Prior to these calls we have
already ensured that ret is zero so we will not be returning this error
for some other case.
Signed-off-by: Tom Rini <trini@ti.com>
Tested-by: Andreas Bießmann <andreas.devel@googlemail.com>
This restores the ordering of interrupt disable to what it what before
commit 35fc84fa. It seems that on some archiectures (e.g. PowerPC) the
OS is loaded into an interrupt region, which can cause problems if
interrupts are still running.
Tested-by: Stefan Roese <sr@denx.de>
Signed-off-by: Simon Glass <sjg@chromium.org>
With 35fc84fa1 [Refactor the bootm command to reduce code duplication]
we stopped checking the return value of bootm_load_os (unintentionally!)
and simply returned if we had a non-zero return value from the function.
This broke the valid case of a legacy image file of a single kernel
loaded into an overlapping memory area (the default way of booting
nearly all TI platforms).
The best way to fix this problem in the new code is to make
bootm_load_os be the one to see if we have a problem with this, and if
it's fatal return BOOTM_ERR_RESET and if it's not BOOTM_ERR_OVERLAP, so
that we can avoid calling lmb_reserve() but continue with booting. We
however still need to handle the other BOOTM_ERR values so re-work
do_bootm_states so that we have an error handler at the bottom we can
goto for problems from bootm_load_os, or problems from the other callers
(as the code was before). Add a comment to do_bootm_states noting the
existing restriction on negative return values.
Signed-off-by: Tom Rini <trini@ti.com>
---
Changes in v2:
- Rework so that only bootm_load_os and boot_selected_os head down into
the err case code, and other errors simply return back to the caller.
Fixes 'spl export'.
This patch introduces support for command line arguments to Plan 9.
Plan 9 generally dedicates a small region of kernel memory (known
as CONFADDR) for runtime configuration. A new environment variable
named confaddr was introduced to indicate this location when copying
arguments.
Signed-off-by: Steven Stallion <sstallion@gmail.com>
[trini: Adapt for Simon's changes about correcting argc, no need to bump
by 2 now]
Signed-off-by: Tom Rini <trini@ti.com>
For tracing it is useful to run as much of U-Boot as possible so as to get
a complete picture. Quite a bit of work happens in bootm, and we don't want
to have to stop tracing before bootm starts.
Add a way of doing a 'fake' boot of the OS - which does everything up to
the point where U-Boot is about to jump to the OS image. This allows
tracing to record right until the end.
This requires arch support to work.
Signed-off-by: Simon Glass <sjg@chromium.org>
At present the bootm code is mostly duplicated for the plain 'bootm'
command and its sub-command variant. This makes the code harder to
maintain and means that changes must be made to several places.
Introduce do_bootm_states() which performs selected portions of the bootm
work, so that both plain 'bootm' and 'bootm <sub_command>' can use the
same code.
Additional duplication exists in bootz, so tidy that up as well. This
is not intended to change behaviour, apart from minor fixes where the
previously-duplicated code missed some chunks of code.
Signed-off-by: Simon Glass <sjg@chromium.org>
At present the arguments to bootm are processed in a somewhat confusing
way. Sub-functions must know how many arguments their calling functions
have processed, and the OS boot function must also have this information.
Also it isn't obvious that 'bootm' and 'bootm start' provide arguments in
the same way.
Adjust the code so that arguments are removed from the list before calling
a sub-function. This means that all functions can know that argv[0] is the
first argument of which they need to take notice.
Signed-off-by: Simon Glass <sjg@chromium.org>
Use the new common code to load a flat device tree. Also fix up a few casts
so that this code works with sandbox. Other than that the functionality
should not change.
Signed-off-by: Simon Glass <sjg@chromium.org>
This makes fixup_silent_linux() use malloc() to allocate its
working space, meaning that our maximum kernel command line
should only be limited by malloc(). Previously it was silently
overflowing the stack.
Note that nothing about this change increases the kernel's maximum
command line length. If you have a command line that is >256
bytes it's up to you to make sure that kernel can handle it.
Signed-off-by: Doug Anderson <dianders@chromium.org>
Acked-by: Mike Frysinger <vapier@gentoo.org>
With fewer #ifdefs the code is more readable and more of the code is
compiled for all boards. Add defines in the header file to control
what features are enabled, and then use if() instead of #ifdef.
Signed-off-by: Simon Glass <sjg@chromium.org>
Much of the image code uses addresses as ulongs and pointers interchangeably,
casting between the two forms as needed.
This doesn't work with sandbox, which has a U-Boot RAM buffer which is
separate from the host machine's memory.
Adjust the cost so that translating from a U-Boot address to a pointer uses
map_sysmem(). This allows bootm to work correctly on sandbox.
Note that there are no exhaustive tests for this code on sandbox, so it is
possible that some dark corners remain.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Marek Vasut <marex@denx.de> (v1)
This is the main entry point to the FIT image verification code. We will
be using it to handle image verification with signatures, so rename the
function.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Marek Vasut <marex@denx.de>
