The command completion temporary buffer seems to be only
used by the argv tokenizer, move it to stack. This saves
2 kiB from the binary size (depends on configuration) per:
$ aarch64-linux-gnu-readelf -s u-boot | sort -n -k 3
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
The history buffer is currently a static array which can be some
10-40 kiB depending on configuration, and so adds considerably to
the U-Boot binary size. Allocate it dynamically instead to reduce
the U-Boot binary size.
Signed-off-by: Marek Vasut <marek.vasut+renesas@mailbox.org>
To quote the author:
It would be useful to be able to boot an OS when CONFIG_CMDLINE is
disabled. This could allow reduced code size.
Standard boot provides a way to handle programmatic boot, without
scripts, so such a feature is possible. The main impediment is the
inability to use the booting features of U-Boot without a command line.
So the solution is to avoid passing command arguments and the like to
code in boot/
A similar process has taken place with filesystems, for example, where
we have (somewhat) separate Kconfig options for the filesystem commands
and the filesystems themselves.
This series starts the process of refactoring the bootm logic so that
it can be called from standard boot without using the command line.
Mostly it removes the use of argc, argv and cmdtbl from the internal
logic.
Some limited tidy-up is included, but this is kept to smaller patches,
rather than trying to remove all #ifdefs etc. Some function comments
are added, however.
A simple programmatic boot is provided as a starting point.
This work will likely take many series, so this is just the start.
Size growth with this series for firefly-rk3288 (Thumb2) is:
arm: (for 1/1 boards) all +23.0 rodata -49.0 text +72.0
This should be removed by:
https://source.denx.de/u-boot/custodians/u-boot-dm/-/issues/11
but it is not included in this series as it is already large enough.
No functional change is intended in this series.
Changes in v3:
- Add a panic if programmatic boot fails
- Drop RFC tag
Changes in v2:
- Add new patch to adjust position of unmap_sysmem() in boot_get_kernel()
- Add new patch to obtain command arguments
- Fix 'boot_find_os' typo
- Pass in the command name
- Use the command table to provide the command name, instead of "bootm"
At present bootstd requires CONFIG_CMDLINE to operate. Add a new
'programmatic' boot which can be used when no command line is available.
For now it does almost nothing, since most bootmeths require the
command line.
Signed-off-by: Simon Glass <sjg@chromium.org>
When using the USB OTG FEL mode on the Allwinner H616, the BootROM
stores some data at the end of SRAM C. This is also the location where
we place the initial SPL stack, so it will overwrite this data.
We still need the BROM code after running the SPL, so should leave that
area alone.
Interestingly this does not seem to have an adverse effect, I guess on
the "way out" (when we return to FEL after the SPL has run), this data
is not needed by the BROM, for just the trailing end of the USB operation.
However this is still wrong, and we should not clobber BROM data.
Lower the SPL stack address to be situated right below the swap buffers
we use in sunxi-fel: that should be out of the way of everyone else.
This obsoletes a previous commit (eb53e7743c) with the same name:
that one was changing the address in an *unused* macro in sunxi_common.h,
so the previous patch didn't have any effect at all.
Fixes: eb53e7743c ("sunxi: h616: lower SPL stack address to avoid BROM data")
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
USB UFI uses fixed 12-byte commands (as does RBC, which is not
supported), but SCSI does not have this limitation. Use the correct
command block lengths depending on the subclass.
Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>
Some devices like YubiKeys need more time before SET_ADDRESS. The spec
says we need to wait 10ms.
Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>
On powerpc, stack protector expects a function called
__stack_chk_fail_local() instead of __stack_chk_fail()
And some versions of GCC for powerpc default to TLS canary
instead of global canary, so always force GCC to use global
canary with -mstack-protector-guard=global
Cc: Joel Peshkin <joel.peshkin@broadcom.com>
Fixes: 4e9bce1243 ("Add support for stack-protector")
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
To quote the author:
The show_board_info() function was adjusted to weak so that it could be
entirely replaced with a board-specific implementation.
The intended way for boards to provide their own information is via a
sysinfo driver. But currently there is no way to show anything other
than the model name.
This series adds support for showing a few more items, in a way that is
easy for boards to extend.
Since there is already a weak checkboard() function, we don't need to
have two levels of weak function here. So this series drops the weak
attribute from show_board_info()
Existing boards will see a slight change in output, in that the model
name will appear first, before any custom output. If that is a problem,
then the solution is to implement a sysinfo driver for the board.
At present only the model name is shown on start. Some boards want to
display more information. Add some more options to allow display of the
manufacturer as well as the version and date of any prior-stage
firmware.
This is useful for coreboot, at least. If other boards have more
information to display, it is easy to add it, now.
Signed-off-by: Simon Glass <sjg@chromium.org>
We already have:
- a sysinfo driver-interface which can obtain the model
- a weak function called checkboard() for showing board info
The current implementation has a weak function on top of a weak
function. Now that all boards have been updated to use checkboard()
instead, drop the __weak on show_board_info()
This reverts commit f7637cc014.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
- squashfs improvements, remove common.h in some places, assorted code
fixes, fix a few CONFIG symbol names in Kconfig files, bring in
linux's <linux/time.h> conversion functions, poplar updates, bcb
improvements.
If we don't DMA-align buffers we pass to FAT, it will align them itself.
This behaviour likely should be deprecated in favor of
CONFIG_BOUNCE_BUFFER, but that's a task for another series. For the
meantime, don't bother aligning the buffer unless we had been doing so in
the past.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
spi_load_image_os performs almost the same steps as the non-falcon-boot
path of spl_spi_load_image. The load address is different, and it also
loads a device tree, but that's it. Refactor the boot process so that
they can both use the same load function.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the spi load method to use spl_load. The address used for
LOAD_FIT_FULL may be different, but there are no in-tree users of that
config. Since payload_offs is only used without OS_BOOT, we defer its
initialization.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the semihosting load method to use spl_load. As a result, it
also adds support for LOAD_FIT_FULL and IMX images.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
This converts the blk load method (used exclusively by NVMe) to use
spl_load. As a consequence, it also adds support for LOAD_FIT_FULL and
IMX images.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the nor load method to use spl_load. As a result it also
adds support for LOAD_FIT_FULL. Since this is the last caller of
spl_load_legacy_img, it has been removed.
We can't load FITs with external data with SPL_LOAD_FIT_FULL, so disable the
test in that case. No boards enable SPL_NOR_SUPPORT and SPL_LOAD_FIT_FULL, so
this is not a regression.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the net load method to use spl_load. As a result, it also
adds support for LOAD_FIT_FULL and IMX images.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the nand load method to use spl_load. nand_page_size may not
be valid until after nand_spl_load_image is called (see e.g. fsl_ifc_spl),
so we set bl_len in spl_nand_read. Since spl_load reads the header for us,
we can remove that argument from spl_nand_load_element.
There are two possible regressions which could result from this commit.
First, we ask for a negative address from spl_get_load_buffer. That is,
instead of
header = spl_get_load_buffer(0, sizeof(*header));
we do
header = spl_get_load_buffer(-sizeof(*header), sizeof(*header));
this could cause a problem if spl_get_load_buffer does not return valid
memory for negative offsets. Second, we now set bl_len for legacy images.
This can cause memory up to a bl_len - 1 before the image load address to
be written, which might not have been the case before. If this turns out to
be a problem, we can add an option for a bounce buffer.
We can't load FITs with external data with SPL_LOAD_FIT_FULL, so disable the
test in that case. No boards enable SPL_NAND_SUPPORT and SPL_LOAD_FIT_FULL, so
this is not a regression.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the mmc loader to spl_load. Legacy images are handled by
spl_load (via spl_parse_image_header), so mmc_load_legacy can be
omitted. To accurately determine whether mmc_load_image_raw_sector is used
(which might not be the case if SYS_MMCSD_FS_BOOT is enabled), we introduce
a helper config SYS_MMCSD_RAW_MODE. This ensures we can inline spl_load
correctly when a board only boots from filesystems. We still need to check
for SPL_MMC, since some boards enable configure raw mode even without MMC
support.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the fat loader to use spl_load. Some platforms are very
tight on space, so we take care to only include the code we really need.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This converts the ext load method to use spl_load. As a consequence, it
also adds support for FIT and IMX images.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Implementers of SPL_LOAD_IMAGE_METHOD have to correctly determine what
type of image is being loaded and then call the appropriate image load
function correctly. This is tricky, because some image load functions
expect the whole image to already be loaded (CONFIG_SPL_LOAD_FIT_FULL),
some will load the image automatically using spl_load_info.read()
(CONFIG_SPL_LOAD_FIT/CONFIG_SPL_LOAD_IMX_CONTAINER), and some just parse
the header and expect the caller to do the actual loading afterwards
(legacy/raw images). Load methods often only support a subset of the
above methods, meaning that not all image types can be used with all
load methods. Further, the code to invoke these functions is
duplicated between different load functions.
To address this problem, this commit introduces a "spl_load" function.
It aims to handle image detection and correct invocation of each of the
parse/load functions.
Although this function generally results in a size reduction with
several users, it tends to bloat boards with only a single user.
This is generally because programmers open-coding the contents of this
function can make optimizations based on the specific loader. For
example, NOR flash is memory-mapped, so it never bothers calling
load->read. The compiler can't really make these optimizations across
translation units. LTO solves this, but it is only available on some
arches. To address this, perform "pseudo-LTO" by inlining spl_load when
there are one or fewer users. At the moment, there are no users, so
define SPL_LOAD_USERS to be 0.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
To allow for easier reuse of this functionality, split it off into its
own function.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Now that spl_nand_fit_read works in units of bytes, it can be combined with
spl_nand_legacy_read. Rename the resulting function spl_nand_read, since it
is no longer FIT-specific.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Aligning addresses and sizes causes overhead which is unnecessary when we
are not loading from block devices. Remove bl_len when it is not needed.
For example, on iot2050 we save 144 bytes with this patch (once the rest of
this series is applied):
add/remove: 0/0 grow/shrink: 0/3 up/down: 0/-144 (-144)
Function old new delta
spl_load_simple_fit 920 904 -16
load_simple_fit 496 444 -52
spl_spi_load_image 384 308 -76
Total: Before=87431, After=87287, chg -0.16%
We use panic() instead of BUILD_BUG_ON in spl_set_bl_len because we still
need to be able to compile it for things like mmc_load_image_raw_sector,
even if that function will not be used.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
For filesystems, filename serves the same purpose as priv. However,
spl_load_fit_image also uses it to determine whether to use a DMA-aligned
buffer. This is beneficial for FAT, which uses a bounce-buffer if the
destination is not DMA-aligned. However, this is unnecessary now that
filesystems set bl_len to ARCH_DMA_MINALIGN instead. With this done, we can
remove filename entirely.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Instead of relying on the presence of filename to determine whether we are
dealing with a FAT filesystem (and should DMA-align the buffer), have FAT set
bl_len to ARCH_DMA_MINALIGN instead. With this done, we can remove the
special-case logic checking for the presence of filename.
Because filesystems are not block-based, we may read less than the size passed
to spl_load_info.read. This can happen if the file size is not DMA-aligned. This
is fine as long as we read the amount we originally wanted to. Modify the
conditions for callers of spl_load_info.read to check against the original,
unaligned size to avoid failing spuriously.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Simplify things a bit for callers of spl_load_info->read by refactoring it
to use units of bytes instead of bl_len. This generally simplifies the
logic, as MMC is the only loader which actually works in sectors. It will
also allow further refactoring to remove the special-case handling of
filename. spl_load_legacy_img already works in units of bytes (oops) so it
doesn't need to be changed.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
bl_len must be a power of two, so we can use ALIGN instead of roundup and
similar tricks to avoid divisions.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
dev and priv serve the same purpose, and are never set at the same time.
Remove dev and convert all users to priv. While we're at it, reorder bl_len
to be last for better alignment.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Remove NULL assignments to fields in spl_load_info when .load doesn't
reference these fields. This can result in more efficient code. filename
must stay even if it is unused, since load_simple_fit uses it.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
On real hardware, semihosting calls tend to have a large constant
overhead (on the order of tens of milliseconds). Reduce the number of
calls by one by reusing the existing fd in smh_fit_read, and closing it
at the end of spl_smh_load_image.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
The purpose of SHOW_ERRORS is to print extra information. Make it depend
on LIBCOMMON to avoid having to check for two configs.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Initialize ret to avoid returning garbage if blk_get_devnum_by_uclass_id
fails.
Fixes: 8ce6a2e175 ("spl: blk: Support loading images from fs")
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
To quote the author:
This series tests raw nand flash in sandbox and fixes various bugs discovered in
the process. I've tried to do things in a contemporary manner, avoiding the
(numerous) variations present on only a few boards. The test is pretty minimal.
Future work could test the rest of the nand API as well as the MTD API.
Bloat (for v1) at [1] (for boards with SPL_NAND_SUPPORT enabled). Almost
everything grows by a few bytes due to nand_page_size. A few boards grow more,
mostly those using nand_spl_loaders.c. CI at [2].
[1] https://gist.github.com/Forty-Bot/9694f3401893c9e706ccc374922de6c2
[2] https://source.denx.de/u-boot/custodians/u-boot-clk/-/pipelines/18443
Rename SPL_MTD_SUPPORT to SPL_MTD in order to match MTD. This allows using
CONFIG_IS_ENABLED to test for MTD support.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Since commit 34793598c8 ("mtd: nand: mxs_nand_spl: Remove the page aligned
access") there are no longer any users of nand_get_mtd. However, it is
still important to know what the page size is so we can allocate a
large-enough buffer. If the image size is not page-aligned, we will go off
the end of the buffer and clobber some memory.
Introduce a new function nand_page_size which returns the page size. For
most drivers it is easy to determine the page size. However, a few need to
be modified since they only keep the page size around temporarily.
It's possible that this patch could cause a regression on some platforms if
the offset is non-aligned and there is invalid address space immediately
before the load address. spl_load_legacy_img does not (except when
compressing) respect bl_len, so only boards with SPL_LOAD_FIT (8 boards) or
SPL_LOAD_IMX_CONTAINER (none in tree) would be affected.
defconfig CONFIG_TEXT_BASE
======================= ================
am335x_evm 0x80800000
am43xx_evm 0x80800000
am43xx_evm_rtconly 0x80800000
am43xx_evm_usbhost_boot 0x80800000
am43xx_hs_evm 0x80800000
dra7xx_evm 0x80800000
gwventana_nand 0x17800000
imx8mn_bsh_smm_s2 0x40200000
All the sitara boards have DDR mapped at 0x80000000. gwventana is an i.MX6Q
which has DDR at 0x10000000. I don't have the IMX8MNRM handy, but on the
i.MX8M DDR starts at 0x40000000. Therefore all of these boards can handle a
little underflow.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
When allocating a buffer to load compressed data into, we need to ensure we
have enough space for over- and under-flow due to alignment. Otherwise we
will clobber the malloc bookkeeping data. Calculate the correct amount of
overhead and use it when determining the size.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
spl_nand_fit_read unconditionally accesses load->priv. Ensure it is set.
Fixes: 00e180cc51 ("spl: nand: support loading i.MX container format file")
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Correct this typo in the warning message shown when no more bootstage
records can be added.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Patrick Delaunay <patrick.delaunay@foss.st.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
