Jiaxun Yang <jiaxun.yang@flygoat.com> says:
Hi all,
This series enabled qemu-xtensa board.
For dc232b CPU it needs to be built with toolchain[1].
This is a side product of me investigating architectures
physical address != virtual address in U-Boot. Now we can
get it covered under CI and regular tests.
VirtIO devices are not working as expected, due to U-Boot's
assumption on VA == PA everywhere, I'm going to get this fixed
later.
My Xtensa knowledge is pretty limited, Xtensa people please
feel free to point out if I got anything wrong.
Thanks
[1]: https://github.com/foss-xtensa/toolchain/releases/download/2020.07/x86_64-2020.07-xtensa-dc232b-elf.tar.gz
Introduce the board and provide instructions on how to get
it work.
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Implement various CPU related functions.
I'm actually just using it to get cpu clock frequency.
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Add xtensa semihosting driver.
It can't use regular semihosting driver as Xtensa's has it's own
semihosting ABI.
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
For PTP_MMU our physical address is not directly mapped
into virtual address space, we need to access physical
memory from those fixed map segments.
Implement phys_to_virt and virt_to_phys hook to reflect
this setting.
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
So U-Boot is using _end symbol to detect location of devicetree
appended at the end of the ROM.
It needs to be calculated based on end of .data load address,
as in our lds .current address is address in RAM.
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Both GitLab and Azure (and other CI systems) have native support for
displaying JUnitXML test report results. The pytest framework that we
use can generate these reports. Change our CI tests so that they will
generate these reports and then have the respective CI platform pick
them up. We write to different locations because of where each CI is
(and isn't) able to easily pass things along.
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
On tests which require "tftpboot" we need to depend not on cmd_net but
rather cmd_tftpboot. And on tests which require cmd_pxe we do not need
to also depend on cmd_net as this should be handled already via Kconfig
logic.
Signed-off-by: Tom Rini <trini@konsulko.com>
Simon Glass <sjg@chromium.org> says:
This series includes a number of mostly unrelated changes which are in
service of running U-Boot on a lab using Labgrid.
Since write_smbios_table() returns an address, we cannot use it to
return and error number. Also, failing on sysinfo_detect() breaks
existing boards, e.g. chromebook_link
Correct this by logging and swallowing the error.
Signed-off-by: Simon Glass <sjg@chromium.org>
Fixes: a5a5756285 ("lib: smbios: Detect system properties via...")
The current test doesn't check anything about the output. If a bug
results in junk before the output, this is not currently detected.
Add a check for the first line being the one expected.
Signed-off-by: Simon Glass <sjg@chromium.org>
When a real board fails we don't want to decode the exception. Reserve
that behaviour for sandbox. Also avoid raising a new exception on
failure - just re-raise the existing one.
Signed-off-by: Simon Glass <sjg@chromium.org>
When a driver is not registered properly it is not clear which one it
is. Adjust test_dm_compat() to show this.
Signed-off-by: Simon Glass <sjg@chromium.org>
Tests for standard boot need disks to be set up, which can only be done
on sandbox, since adjusting disks on real hardware is not currently
supported. Mark the init function as sandbox-only.
Signed-off-by: Simon Glass <sjg@chromium.org>
This feature is not present on older Chromebooks, so disable the
setting.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
This binary does not prevent the system from booting. Mark it optional
so that U-Boot can be built without it.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Since commit 0dba45864b ("arm: Init the debug UART") the debug UART is
set up in _main() before early_system_init() is called.
Add a suitable board_debug_uart_init() function to set up the UART in
SPL.
Signed-off-by: Simon Glass <sjg@chromium.org>
The relocation offset can change in some initcall sequences. Handle
this and make sure it is used for all debugging statements in
init_run_list()
Update the trace test to match.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Caleb Connolly <caleb.connolly@linaro.org>
Heinrich Schuchardt <heinrich.schuchardt@canonical.com> says:
[trini: Patch 1/3 was replaced by my series merged just prior to this,
Heinrich's cover letter is lightly edited and any mistakes are my own]
If we have multiple weak implementations of functions, the linker might
choose any of these.
The EFI sub-systems uses invalidate_icache_all() after loading binaries.
Both the EFI sub-system and cmd/cache.c provide a weak
invalidate_icache_all() function. Remove the EFI instance.
For ARM11 functional implementation of invalidate_icache_all is missing.
Add it.
If multiple weak implementations of a weak function exist, it is unclear
which one the linker should chose. cmd/cache.c already defines a weak
invalidate_icache_all().
We don't need a call to invalidate_icache_all() on x86.
ARM, RISC-V, and Sandbox provide an implementation.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
In EFI sub-system we rely on invalidate_icache_all() to invalidate the
instruction cache after loading binaries. Add the missing implementation on
ARM1136, ARM1176.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Tom Rini <trini@konsulko.com> says:
Prior to this series we had some de-facto required cache functions that
were either unimplemented on some architectures or differently named.
This would lead in some cases to having multiple "weak" functions
available as well. Rework things so that an architecture must provide
these functions and it is up to that architecture if a "weak" default
function makes sense, or not.
It should be up to an architecture to decide how to implement cache
functions, and if they need to use weak functions or not. Allowing the
cache command to be built without cache functionality implemented is
unhelpful. Further, guard the call to noncached_set_region with
CONFIG_SYS_NONCACHED_MEMORY as that's when it's implemented and again is
an architecture specific detail.
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
Implement a weak default version of flush_dcache_all which is based on
the ARM default, which is to flush the entire range via
flush_dcache_range(...).
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
Implement a weak default version of flush_dcache_all which is based on
the ARM default, which is to flush the entire range via
flush_dcache_range(...).
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
The implementation of icache_invalid appears to be doing what other
architectures call invalidate_icache_all so rename to match.
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
Implement a weak default version of flush_dcache_all which is based on
the ARM default, which is to flush the entire range via
flush_dcache_range(...).
Acked-by: Angelo Dureghello <angelo@kernel-space.org>
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
When running within a Python venv we must use the 'coverage' tool (which
is within the venv) so that the venv packages are used in preference to
system packages. Otherwise the coverage tests run in a different
environment from the normal tests and may fail due to missing packages.
Handle this by detecting the venv and changing the tool name.
Signed-off-by: Simon Glass <sjg@chromium.org>
The scan_part() function uses a struct uuid to store the little-endian
partition type GUID, but this structure should be used only to contain a
big-endian UUID. Use an efi_guid_t instead.
Signed-off-by: Vincent Stehlé <vincent.stehle@arm.com>
Cc: Simon Glass <sjg@chromium.org>
Cc: Tom Rini <trini@konsulko.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This is used by some Binman entry types, so add it to allow more tests
to pass.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
Add this package so we can run code-coverage tests for Binman.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
Buildman uses all available CPUs by default, so running more than one or
two concurrent processes is not normally useful.
However in some CI cases we want to be able to run several jobs at once
to save time. For example, in a lab situation we may want to run a test
on 20 boards at a time, since only the build step actually takes much
CPU.
Add an option which allows such a limit. When buildman starts up, it
waits until the number of running processes goes below the limit, then
claims a spot in the list. The list is maintained with a temporary file.
Note that the temp file is user-specific, since it is hard to create a
locked temporary file which can be accessed by any user. In most cases,
only one user is running jobs on a machine, so this should not matter.
Signed-off-by: Simon Glass <sjg@chromium.org>
When a file is removed by a commit (e.g. include/common.h yay!) it can
cause incremental build failures since one of the dependency files from
a previous build may mention the file.
Add an option to run 'make mrproper' automatically when a build fails.
This can be used to automatically resolve the problem, without always
adding the large overhead of 'make mrproper' to every build.
Signed-off-by: Simon Glass <sjg@chromium.org>
