In SMP all harts will register themself in available_hart
during start up. Then main hart will send IPI to other harts
according to this variables. But this mechanism may not
guarantee that all other harts can jump to next stage.
When main hart is sending IPI to other hart according to
available_harts, but other harts maybe still not finish the
registration. Then the SMP booting will miss some harts finally.
So let it become an option and it will be enabled by default.
Please refer to the discussion:
https://www.mail-archive.com/u-boot@lists.denx.de/msg449997.html
Signed-off-by: Rick Chen <rick@andestech.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
U-Boot and SPL don't necessary share the same location, so we might end
with U-Boot SPL in read-only memory (XIP) and U-Boot in read-write memory.
In case of non XIP boot mode, we rely on such variables as "hart_lottery"
and "available_harts_lock" which we use as atomics.
The problem is that CONFIG_XIP also propagate to main U-Boot, not only SPL,
so we need CONFIG_SPL_XIP to distinguish SPL XIP from other XIP modes.
This adds an option special for SPL to behave it in XIP manner and we don't
use hart_lottery and available_harts_lock, during start proccess.
Signed-off-by: Nikita Shubin <n.shubin@yadro.com>
Reviewed-by: Rick Chen <rick@andestech.com>
The booting flow is SPL -> OpenSBI -> U-Boot.
The boot hart may change after OpenSBI and may not always be hart0,
so wrap the related branch instruction with M-MODE.
Current DTB setup for XIP is not valid.
There is no chance for CONFIG_SYS_FDT_BASE, the DTB address used
in XIP mode, to be returned. Fix this.
Fixes: 2e8d2f8843 ("riscv: Remove OF_PRIOR_STAGE from RISC-V boards")
Signed-off-by: Rick Chen <rick@andestech.com>
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Restore global pointer before board_init_f_init_reserve call,
as "a0" can be set in harts_early_init call and we end up with
invalid global pointer.
Signed-off-by: Nikita Shubin <n.shubin@yadro.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Rename the sections used to implement linker lists so they begin with
'__u_boot_list' rather than '.u_boot_list'. The double underscore at the
start is still distinct from the single underscore used by the symbol
names.
Having a '.' in the section names conflicts with clang's ASAN
instrumentation which tries to add redzones between the linker list
elements, causing expected accesses to fail. However, clang doesn't try
to add redzones to user sections, which are names with all alphanumeric
and underscore characters.
Signed-off-by: Andrew Scull <ascull@google.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
- Make all users of CUSTOM_SYS_INIT_SP_ADDR reference SYS_INIT_SP_ADDR
- Introduce HAS_CUSTOM_SYS_INIT_SP_ADDR to allow for setting the stack
pointer directly, otherwise we use the common calculation.
- On some platforms that were using the standard calculation but did not
set CONFIG_SYS_INIT_RAM_SIZE / CONFIG_SYS_INIT_RAM_ADDR, set them.
- On a small number of platforms that were not subtracting
GENERATED_GBL_DATA_SIZE do so now via the standard calculation.
- CONFIG_SYS_INIT_SP_OFFSET is now widely unused, so remove it from most
board config header files.
Signed-off-by: Tom Rini <trini@konsulko.com>
Instead of a special function, send an event after driver model is inited
and adjust the boards which use this function.
Signed-off-by: Simon Glass <sjg@chromium.org>
Enable saving environment to SPI flash memory on SiFive
Unmatched.
Signed-off-by: Thomas Skibo <thomas-git@skibo.net>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
At some point back in 2018 prior_stage_fdt_address and OF_PRIOR_STAGE got
introduced, in order to support a DTB handed over by an earlier stage boo
loader. However we have another option in the Kconfig (OF_BOARD) which has
identical semantics.
On RISC-V some of the boards pick up the DTB from a1 and copy it in their
private gd_t. Apart from that they copy it to prior_stage_fdt_address, if
the Kconfig option is selected, which is unnecessary.
So let's switch the config option for those boards to OF_BOARD and define
the required board_fdt_blob_setup() for them.
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
If Coherence Manager were not set in the beginning,
u-boot-spl would sometimes fail to boot to u-boot proper.
Enable CM and I/D cache at the same time in harts_early_init
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Provide sysreset driver using the SBI system reset extension.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Samuel Holland <samuel@sholland.org>
Invokes the common cache_init function to initialize ccache.
Signed-off-by: Zong Li <zong.li@sifive.com>
Reviewed-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Rename these options so that CONFIG_IS_ENABLED can be used with them.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Enable SPL_I2C_SUPPORT for fu740, and add 'u-boot,dm-spl' property in
i2c node.
Signed-off-by: Zong Li <zong.li@sifive.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
This commit does the same thing as Linux commit 33def8498fdd.
Use a more generic form for __section that requires quotes to avoid
complications with clang and gcc differences.
Remove the quote operator # from compiler_attributes.h __section macro.
Convert all unquoted __section(foo) uses to quoted __section("foo").
Also convert __attribute__((section("foo"))) uses to __section("foo")
even if the __attribute__ has multiple list entry forms.
Signed-off-by: Marek Behún <marek.behun@nic.cz>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
By utilizing the newly introduced BINMAN_STANDALONE_FDT option, along
with a new dedicated device tree source file for the QEMU virt target
used for binman only, we can now use binman to generate u-boot.itb.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
At present there is only one Kconfig option CONFIG_SIFIVE_CLINT to
control the enabling of SiFive CLINT support in both SPL (M-mode)
and U-Boot proper (S-mode). So for a typical SPL config that the
SiFive CLINT driver is enabled in both SPL and U-Boot proper, that
means the S-mode U-Boot tries to access the memory-mapped CLINT
registers directly, instead of the normal 'rdtime' instruction.
This was not a problem before, as the hardware does not forbid the
access from S-mode. However this becomes an issue now with OpenSBI
commit 8b569803475e ("lib: utils/sys: Add CLINT memregion in the root domain")
that the SiFive CLINT register space is protected by PMP for M-mode
access only. U-Boot proper does not boot any more with the latest
OpenSBI, that access exceptions are fired forever from U-Boot when
trying to read the timer value via the SiFive CLINT driver in U-Boot.
To solve this, we need to split current SiFive CLINT support between
SPL and U-Boot proper, using 2 separate Kconfig options.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Sean Anderson <seanga2@gmail.com>
This reverts commit bc8bbb77f7.
This commit breaks U-Boot booting on SiFive Unleashed board, as
there is no such CSR on U54 core.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Add a callback harts_early_init() to start.S to allow different riscv
hart perform setup code for each hart as early as possible. Since all
the harts enter the callback, they must be able to run the same
setup.
Signed-off-by: Green Wan <green.wan@sifive.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
The _SUPPORT suffix is from an earlier time and interferes with use of
the CONFIG_IS_ENABLED() macro. Rename the option to drop the suffix.
Tidy up the TODO that prompted this.
Signed-off-by: Simon Glass <sjg@chromium.org>
When testing QEMU RISC-V 'virt' machine with a 2 GiB memory
configuration, it was discovered gd->ram_top is assigned to
value zero in setup_dest_addr().
While gd->ram_top should not be declared as type `unsigned long`,
which will be updated in a future patch, the current logic in
board_get_usable_ram_top() can be updated to cover both 64-bit
and 32-bit RISC-V.
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Move this out of the common header and include it only where needed. In
a number of cases this requires adding "struct udevice;" to avoid adding
another large header or in other cases replacing / adding missing header
files that had been pulled in, very indirectly. Finally, we have a few
cases where we did not need to include <asm/global_data.h> at all, so
remove that include.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
Not s2 register, t1 register is correct
Fortunately, it works because t1 register has a garbage value
Signed-off-by: Brad Kim <brad.kim@semifive.com>
Reviewed-by: Lukas Auer <lukas@auer.io>
Reviewed-by: Leo Liang <ycliang@andestech.com>
This matches the naming scheme of other timer drivers.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Rick Chen <rick@andestech.com>
This adds comments regarding the ordering and purpose of certain
instructions as I understand them.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Reviewed-by: Leo Liang <ycliang@andestech.com>
This ensures constructs like `if (gd & gd->...) { ... }` work when
accessing the global data pointer. Without this change, it was possible for
a very early trap to cause _exit_trap to directly or indirectly (through
printf) to read arbitrary memory. This could cause a second trap,
preventing show_regs from being printed.
printf (and specifically puts) uses gd to determine what function to print
with. These functions in turn use gd to find the serial device, etc.
However, before accessing gd, puts first checks to see if it is non-NULL.
This indicates an existing (perhaps undocumented) assumption that either gd
is NULL or it is completely valid.
Before this patch, gd either points to unexpected data (because it retains
the value it did from the prior-stage) or points to uninitialized data
(because it has not yet been initialized by board_init_f_init_reserve)
until the hart has acquired available_harts_lock. This can cause two
problems, depending on the value of gd->flags. If GD_FLG_SERIAL_READY is
unset, then some garbage data will be printed to stdout, but there will not
be a second trap. However, if GD_FLG_SERIAL_READY is set, then puts will
try to print with serial_puts, which will likely cause a second trap.
After this patch, gd is zero up until either a hart has set it in
wait_for_gd_init, or until it is set by arch_init_gd. This prevents its
usage before its data is initialized because both handle_trap and puts
ensure that gd is nonzero before using it. After gd has been set, it is OK
to access it because its data has been cleared (and so flags is valid).
XIP cannot use locks because flash is not writable. This leaves it
vulnerable to the same class of bugs regarding already-pending IPIs as
before this series. Fixing that would require finding another method of
synchronization, which is outside the scope of this series.
Fixes: 7c6ca03eae ("riscv: additional crash information")
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
We can reduce the number of instructions needed to use available_harts_lock
by using the aq and rl suffixes for AMOs.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Even though we no longer call smp_function if an IPI was not sent by
U-Boot, we still need to clear any IPIs which were pending from the
execution environment. Otherwise, secondary harts will busy-wait in
secondary_hart_loop, instead of relaxing.
Along with the previous commit ("riscv: Use a valid bit to ignore
already-pending IPIs"), this fixes SMP booting on the Kendryte K210.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Clearing MIP.MSIP is not guaranteed to do anything by the spec. In
addition, most existing RISC-V hardware does nothing when this bit is set.
The following commits "riscv: Use a valid bit to ignore already-pending
IPIs" and "riscv: Clear pending IPIs on initialization" should implement
the original intent of the reverted commit in a more robust manner.
This reverts commit 9472630337.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
The riscv-timer driver currently serves as a shim for several riscv timer
drivers. This is not too desirable because it bypasses the usual timer
selection via the driver model. There is no easy way to specify an
alternate timing driver, or have the tick rate depend on the cpu's
configured frequency. The timer drivers also do not have device structs,
and so have to rely on storing parameters in gd_t. Lastly, there is no
initialization call, so driver init is done in the same function which
reads the time. This can result in confusing error messages. To a user, it
looks like the driver failed when trying to read the time, whereas it may
have failed while initializing.
This patch removes the shim functionality from the riscv-timer driver, and
has it instead implement the former rdtime.c timer driver. This is because
existing u-boot users who pass in a device tree (e.g. qemu) do not create a
timer device for S-mode u-boot. The existing behavior of creating the
riscv-timer device in the riscv cpu driver must be kept. The actual reading
of the CSRs has been redone in the style of Linux's get_cycles64.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
At present fdtdec_get_addr() is used to get L2 cache controller base
address. This only works for a fixed #address-cells and #size-cells.
Change to use fdtdec_get_addr_size_auto_parent() instead.
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Reviewed-by: Pragnesh Patel <pragnesh.patel@openfive.com>
We should not initialize a variable if the value is overwritten before
being read.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Pragnesh Patel <pragnesh.patel@sifive.com>
Tested-by: Pragnesh Patel <pragnesh.patel@sifive.com>
Reviewed-by: Rick Chen <rick@andestech.com>
All FU540 driver related options should be in the SoC level Kconfig.
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Reviewed-by: Pragnesh Patel <pragnesh.patel@sifive.com>
Tested-by: Pragnesh Patel <pragnesh.patel@sifive.com>
spl_soc_init() seems to be a better name, as all SPL functions
names start from the spl_ prefix.
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Reviewed-by: Pragnesh Patel <pragnesh.patel@sifive.com>
Tested-by: Pragnesh Patel <pragnesh.patel@sifive.com>
Switch off SMP support when building u-boot-spl would cause linking error as follow:
undefined reference to 'secondary hart relocate' and 'smp_call_function'.
Add macro to wrap up proper code region that needs SMP configuration on.
Signed-off by: Leo Liang <ycliang@andestech.com>
Cc: rick@andestech.com
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Commit 40686c394e ("riscv: Clean up IPI initialization code")
caused U-Boot failed to boot on SiFive HiFive Unleashed board.
The codes inside arch_cpu_init_dm() may call U-Boot timer APIs
before the call to riscv_init_ipi(). At that time the timer register
base (e.g.: the SiFive CLINT device in this case) is unknown yet.
It might be the name riscv_init_ipi() that misleads people to only
consider it is related to IPI, but in fact the timer capability is
provided by the same SiFive CLINT device that provides the IPI.
Timer capability is needed for both UP and SMP.
Considering that the original refactor does have benefits, that it
makes the IPI code more similar to U-Boot initialization idioms.
It also removes some quite ugly macros. Let's do the minimal revert
instead of a complete revert, plus a fixes to arch_cpu_init_dm() to
consider the SPL case.
Fixes: 40686c394e ("riscv: Clean up IPI initialization code")
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Sean Anderson <seanga2@gmail.com>
Tested-by: Leo Liang <ycliang@andestech.com>
Add L2 cache node to enable all cache ways from U-Boot proper.
Signed-off-by: Pragnesh Patel <pragnesh.patel@sifive.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Some older processors (notably the Kendryte K210) use an older version of
the RISC-V privileged specification. The primary changes between the old
and new are in virtual memory, and in the merging of three separate counter
enable CSRs. Using the new CSR on an old processor causes an illegal
instruction exception. This patch adds an option to use the old CSRs
instead of the new one.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
The previous IPI code initialized the device whenever the first call was
made to a riscv_*_ipi function. This made it difficult to determine when
the IPI device was initialized. This patch introduces a new function
riscv_init_ipi. It is called once during arch_cpu_init_dm. In SPL, it is
called in spl_invoke_opensbi. Before this point, no riscv_*_ipi functions
should be called.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Rick Chen <rick@andestech.com>
On some platforms (k210), the previous stage bootloader may have not
cleared pending IPIs before transferring control to U-Boot. This can cause
race conditions, as multiple harts all attempt to initialize the IPI
controller at once. This patch clears IPIs before enabling them, ensuring
that only one hart modifies shared memory at once.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Rick Chen <rick@andestech.com>
