When one xPL phase wants to jump to the next, the next phase must be
loaded into its required address. This means that the TEXT_BASE for the
two phases must be different and there cannot be any memory overlap
between the code used by the two phases. It also can mean that phases
need to be moved around to accommodate any size growth.
Having two xPL phases in SRAM at the same time can be tricky if SRAM
is limited, which it often is. It would be better if the second phase
could be loaded somewhere else, then decompressed into place over the
top of the first phase.
Introduce a relocating jump for xPL to support this. This selects a
suitable place to load the (typically compressed) next phase, copies
some decompression code out of the first phase, then jumps to this code
to decompress and start the next phase.
This feature makes it much easier to support Verified Boot for Embedded
(VBE) on RK3399 boards, which have 192KB of SRAM.
Signed-off-by: Simon Glass <sjg@chromium.org>
This function will be used by the relocating jumper too, so add a
typedef to the header file to avoid mismatches.
Signed-off-by: Simon Glass <sjg@chromium.org>
In many cases the FIT includes a devicetree. Add support for loading
this into a suitable place in memory.
Signed-off-by: Simon Glass <sjg@chromium.org>
In some cases only the 'loadable' property is present in the FIT.
Handle this by loading the first such image.
Signed-off-by: Simon Glass <sjg@chromium.org>
There is no guarantee that an FIT image starts on a block boundary. When
it doesn't, the image starts part-way through the first block.
Add logic to detect this and copy the image down into place.
Signed-off-by: Simon Glass <sjg@chromium.org>
At present the VBE implementation is limited to sandbox only. Adjust the
call to fit_image_load() to remove this limitation.
Signed-off-by: Simon Glass <sjg@chromium.org>
It is convenient to use TEXT_BASE as a place to hold the FIT header, but
this does not work in VPL, since SDRAM is not inited yet.
Allocate the memory instead. Ensure the size is aligned to the media
block-size so that it can be read in directly. Improve the
error-checking for blk_read() and add some more debugging.
Keep the existing TEXT_BASE mechanism in sandbox to avoid an
'Exec format error' when trying to run the image.
Signed-off-by: Simon Glass <sjg@chromium.org>
Loading a FIT is useful for other VBE methods, such as ABrec. Create a
new function to handling reading it.
Signed-off-by: Simon Glass <sjg@chromium.org>
VBE is currently quite careful with function arguments because it is
used in VPL which cannot be updated after manufacture. Bugs can cause
security holes.
Unfortunately this adds to code size.
In several cases we are reading values from a devicetree which is part
of U-Boot (or at least VPL) and so known to be good. Also, in several
places, getting bad values does not matter.
So change a few checks to assert() to reduce code size.
Signed-off-by: Simon Glass <sjg@chromium.org>
Pass the private data instead of the device, to help the compiler
optimise better. This saves 16 bytes of code on pinecube (rk3288)
Signed-off-by: Simon Glass <sjg@chromium.org>
Pass a struct udevice instead of the descriptor structure, since this is
the native argument for blk_read()
Signed-off-by: Simon Glass <sjg@chromium.org>
Move a few things into a new, common header file so that vbe-simple can
share code with the upcoming abrec.
Put struct simple_nvdata in it and rename it.
Signed-off-by: Simon Glass <sjg@chromium.org>
We should not be using the old blk_d...() interface, is only there to
aid migration to driver model.
Move to blk_read() instead.
Changes in v2:
- Split patch into several pieces
Signed-off-by: Simon Glass <sjg@chromium.org>
commit fc37a73e66 ("fdt: Swap the signature for
board_fdt_blob_setup()") introduced a subtle change to the Snapdragon
implementation, removing the assignment to gd->fdt_blob partway through
the function.
This breaks qcom_parse_memory() which was also called during
board_fdt_blob_setup().
The underlying issue here is that qcom_parse_memory is using the of_ api
to traverse a devicetree, which relies on the fdt_blob in global data.
Rather than relying on this subtle behaviour, explicitly pass the FDT
that should be consulted for a /memory node.
Using the OF API is typically preferable because it's easier to read,
but using the lower level fdt_ methods instead here doesn't add too much
complexity, I think.
Finally, a minor tweak was made to board_fdt_blob_setup to use the
passed fdt blob pointer instead of gd->fdt_blob, which removes the last
of the references to global data in this area.
Fixes: fc37a73e66 (fdt: Swap the signature for board_fdt_blob_setup())
Reviewed-by: Caleb Connolly <caleb.connolly@linaro.org>
Signed-off-by: Sam Day <me@samcday.com>
Link: https://lore.kernel.org/r/20250122-qcom-parse-memory-updates-v2-1-98dfcac821d7@samcday.com
Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
On 32-bit architectures, LAST_LINE (_LAST_LINE - 1UL) is 64 bits long,
but size_t (from ARRAY_SIZE(...)) is 32 bits. This results in a warning
because the max() macro expects the same type on both sides:
cmd/ufetch.c: In function ‘do_ufetch’:
include/linux/kernel.h:179:24: warning: comparison of distinct pointer types lacks a cast [-Wcompare-distinct-pointer-types]
179 | (void) (&_max1 == &_max2); \
| ^~
cmd/ufetch.c:92:25: note: in expansion of macro ‘max’
92 | int num_lines = max(LAST_LINE + 1, ARRAY_SIZE(logo_lines));
| ^~~
Fix this by casting LAST_LINE to size_t.
Reviewed-by: Caleb Connolly <caleb.connolly@linaro.org>
Signed-off-by: J. Neuschäfer <j.ne@posteo.net>
Link: https://lore.kernel.org/r/20241211-ufetch-v2-1-2b5432ffaeb1@posteo.net
Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
Add support for the PCIe busses on Qualcomm platforms,
by using the pcie_dw_common infrastructure.
The driver is based on the Linux driver but only supporting
the "1_9_0" and compatible platforms like:
- sa8540p
- sc7280
- sc8180x
- sc8280xp
- sdm845
- sdx55
- sm8150
- sm8250
- sm8350
- sm8450
- sm8550
- sm8650
- x1e80100
But it has only been tested on:
- sc7280
- sm8550
- sm8650
- x1e80100
It supports setting the IOMMU SID table for supported platforms.
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://lore.kernel.org/r/20241125-topic-pcie-controller-v1-2-45c20070dd53@linaro.org
Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
Replace delay_ns by delay_us which is the field name used into
struct cyclic_info.
Signed-off-by: Patrice Chotard <patrice.chotard@foss.st.com>
Reviewed-by: Stefan Roese <sr@denx.de>
On 32 bits platforms, timer_get_us() returns an unsigned long which
is a 32 bits. timer_get_us() wraps around every 72 minutes
(2 ^ 32 / 1000000 =~ 4295 sec =~ 72 min).
So the test "if time_after_eq64(now, cyclic->next_call)" is no more
true when cyclic->next_call becomes above 32 bits max value (4294967295).
At this point after 72 min, no more cyclic function are
executed included watchdog one.
Instead of using timer_get_us(), use get_timer_us() which returns a
uint64_t, this allows a rollover every 584942 years.
Signed-off-by: Patrice Chotard <patrice.chotard@foss.st.com>
Reviewed-by: Stefan Roese <sr@denx.de>
