Sinovoip_BPI_M2_Plus target was renamed to bananapi_m2_plus_h3
am335x_boneblack target replaced by am335x_evm
build rockpro64-rk3399 target as a few people have hardware (untested)
Add back flushing/disabling caches for armv7 efi which was removed in
2019.04 and later and is required to load kernels on i.MX 6.
The UEFI specification states that non-architectural caches
(not controlled by cp15 operations) should be disabled during boot time
services. On i.MX 6 there is a PL310 L2 cache which U-Boot leaves on,
which breaks loading kernels even after we disable all the cp15
controlled caches in bootarm.efi.
the page mask that removes the lower bits was based on 1UL, which is
only 32-bit on ARMv7. Thus on machines with 4GiB memory or more the
addresses are truncated and never end up in the EFI memory map. The
second issue is that on 32-bit systems a function that returns a 64-
bit integer through a pointer was given a pointer to an address that
can only hold 32-bit values, causing a data abort. The second issue
was fixed in U-Boot 2019.04.
ok phessler@ naddy@
until it could be upstreamed. Well, the support has been upstreamed, so
we can retire these in favour of arm-trusted-firmware and u-boot-aarch64.
tested by me on an original pinebook, and by patrick@ on a 1080p pinebook.
discussed with patrick@ and jsg@
OK sthen@
trees for multiple am335x boards including the BeagleBone Black.
The am335x_boneblack target has been removed in the U-Boot repository
and will not be in the next major release. Continue building it till
then.
Add pinebook and pine64-lts targets. This conflicts with
sysutils/u-boot-pinebook though that port can be removed now the
pinebook changes are finally in a mainline U-Boot release.
having to set device tree address in config.txt as the dtb address
passed via atags is used. Patch from Alexander Graf in openSUSE U-Boot
tree. Discussed with kettenis@
efi_loader: allocate configuration table array
After this commit, the ConfigurationTable field of the system table
was no longer translated from a physical address into a virtual address
when the SerVirtualAddressMap() runtime service got called.
Broke boot on Orange Pi PC 2 as reported by kevlo@ and tracked down
by kettenis@.
Includes changes to add Ethernet to more Allwinner device trees and
finally includes the proper device tree for sopine.
Also add a patch from kettenis@ to enable Ethernet on the NanoPi A64.
- correct clock frequency used in bcm2835 sdhost driver to allow using
raspberrypi-firmware 7fdcd00e00a42a1c91e8bd6f5eb8352fe9358557 and
later without hangs
- detect rpi3 b+ and build lan78xx driver for it
- distro boot support for mvebu 8k
- add workaround for realtek phy on pine64
also remove use of MODPY_ADJ_FILES as suggested by sthen@ a while back
Tested on CuBox-i4Pro, BeagleBone Black, PandaBoard ES and
Raspberry Pi 3 Model B.
This corrects a regression on rpi3 when booting off USB where the EFI
loader would stop iterating over block devices when loading from SD card
failed reported by 'R0me0' on bugs@.
Revert commit 25877d4e4c45451c5398aec3de50e0d5befe0e9f to add
CONFIG_OF_EMBED=y back to rpi configs so rpi3 will continue to work.
Tracked down by kettenis@
While Allwinner A64/H5 targets are built an Allwinner specific build of
ARM Trusted Firmware will need to become a build prequisite in future,
so U-Boot can build complete images after a BL31 environment variable is
set at build time with the path to a prebuilt bl31.bin.
espie and naddy directed me to packages-specs(7), an rc suffix
is allowed only if there is no hyphen.
This commit doesn't include bumping EPOCH, but that may be needed.
Patch out 'python' in a Makefile for ${MODPY_BIN} and add swig to
BUILD_DEPENDS so any related problems will be caught in future.
Problem reported by naddy and espie.
Fixes data aborts reported by Daniel Bolgheroni when reading a dtb off
mmc on two different allwinner systems (Banana Pi and Orange Pi One)
with 2017.01.
All of the Xunlong Orange Pi configs (Allwinner A20 and H3)
FriendlyARM NanoPi NEO (Allwinner H3)
Inverse Path USB Armory (Freescale/NXP i.MX53)
Discussed with kettenis.
There is a use after free in kwbimage, found by building u-boot with the
use after free detection enabled in malloc. When building the clearfog target:
MKIMAGE u-boot-spl.kwb
Segmentation fault (core dumped)
kwbimage_generate -> image_version_file (alloc and free image_cfg global)
kwbimage_generate -> image_headersz_v1 -> image_count_options (image_cfg used)
It isn't clear to me if image_version_file should be inlined or another
approach taken, but as it stands it is clearly wrong.
The result of image_version_file is also never checked for -1 which multiple
paths in the function return.
I reported this on the u-boot list a week or so ago but no one has responded
so far.
