Use CSR_UCCTLCOMMAND instead of CSR_MCCTLCOMMAND
to do cache flush operation in M-mode and S-mode.
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Cache could be enabled in harts_early_init board-specific hook,
so remove cache enablement in start.S
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Rename spl_soc_init() to spl_dram_init() because the generic function
name does not reflect what the function actually does. Also
spl_dram_init() is commonly used for dram initialization and should be
called from board_init_f().
Signed-off-by: Lukas Funke <lukas.funke@weidmueller.com>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Rename spl_soc_init() to spl_dram_init() because the generic function
name does not reflect what the function actually does. Also
spl_dram_init() is commonly used for dram initialization and should be
called from board_init_f().
Signed-off-by: Lukas Funke <lukas.funke@weidmueller.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
The instruction and data cache line sizes of Andes core
are 64-byte. Select SYS_CACHE_SHIFT_6 for RISCV_NDS so
the SYS_CACHELINE_SIZE is enabled with a default value.
Signed-off-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
A new property has been added, with an extensive rationale at [1], that
can be used in place of "riscv,isa" to indicate what extensions are
supported by a given platform that is a list of strings rather than a
single string. There are some differences between the new property,
"riscv,isa-extensions" and the incumbent "riscv,isa" - chief among them
for the sake of parsing being the list of strings, as opposed to a
string. Another advantage is strictly defined meanings for each string
in a dt-binding, rather than deriving meaning from RVI standards. This
will likely to some divergence over time, but U-Boot's current use of
extension detection is very limited - there are just four callsites of
supports_extension() in mainline U-Boot.
These checks are limited to two checks for FPU support and two checks
for "s" and "u". "s" and "u" are not supported by the new property, but
they were also not permitted in "riscv,isa". These checks are only
meaningful (or run) in M-Mode, in which case supports_extension() does
not parse the devicetree anyway.
Add support for the new property in U-Boot, prioritising it, before
falling back to the, now deprecated, "riscv,isa" property if it is not
present.
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
cpu_get_desc() for the RISC-V CPU currently reads "riscv,isa" to get
the description, but it is no longer a required property and cannot be
assummed to always be present, as the new "riscv,isa-extensions" and
"riscv,isa-base" properties may be present instead.
On RISC-V, cpu_get_desc() has two main uses - firstly providing an
informational name for the CPU for smbios or at boot with
DISPLAY_CPUINFO etc and secondly it forms the basis of ISA extension
detection in supports_extension() as it returns (a portion of) an ISA
string.
cpu_get_desc() returns a string, which aligned with "riscv,isa" but
the new property is a list of strings. Rather than add support for
the list of strings property, which would require creating an isa
string from "riscv,isa-extensions", modify the RISC-V CPU's
implementaion of cpu_get_desc() return the first compatible as the
cpu description instead. This may be fine for the informational cases,
but it would break extension dtection, given supports_extension()
expects cpu_get_desc() to return an ISA string.
Call dev_read_string() directly in supports_extension() to get the
contents of "riscv,isa" so that extension detection remains functional.
As a knock-on affect of this change, extension detection is no longer
broken for long ISA strings. Previously if the ISA string exceeded the
32 element array that supports_extension() passed to cpu_get_desc(),
it would return ENOSPC and no extensions would be detected.
This bug probably had no impact as U-Boot does not currently do anything
meaningful with the results of supports_extension() and most SoCs
supported by U-Boot don't have anywhere near that complex of an ISA
string. The QEMU virt machine's CPUs do however, so extension detection
doesn't work there.
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Add dcache operations invalidate_dcache_range and flush_dcache_range for
cv1800b.
Signed-off-by: Kongyang Liu <seashell11234455@gmail.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Add Sophgo cv1800b SoC to support RISC-V arch.
Signed-off-by: Kongyang Liu <seashell11234455@gmail.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
When debugging, it is useful to have a backtrace to find
out what is in the call stack as the previous function (RA)
may not have been the culprit.
Since this adds size to the build, do not add it by default
and avoid putting it in the SPL build if not needed.
Signed-off-by: Ben Dooks <ben.dooks@codethink.co.uk>
Tested-by: Leo Yu-Chi Liang <ycliang@andestech.com>
The first multi-letter extension after the single-letter extensions does
not have to be preceded by an underscore, which could cause the parser
to mistakenly find a single-letter extension after the start of the
multi-letter portion of the string.
Three letters precede multi-letter extensions (s, x & z), none of which
are valid single-letter extensions. The dt-binding also allows
multi-letter extensions starting with h, but no such extension have been
frozen or ratified, and the unprivileged spec no longer uses "h" as a
prefix for multi-letter hypervisor extensions, having moved to "sh"
instead. For that reason, modify the parser to stop at s, x & z to prevent
this overrun, ignoring h.
Signed-off-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Andes CPU supports cache and TLB ECC.
Enable them by default.
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Yu Chien Peter Lin <peterlin@andestech.com>
Andes CPU has memboost feature including prefetch,
write-around and non-blocking load. Enable them by default.
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Yu Chien Peter Lin <peterlin@andestech.com>
Kconfig provides SYS_[I|D]CACHE_OFF config options to switch off caches.
Provide the corresponding implementation to the options.
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Yu Chien Peter Lin <peterlin@andestech.com>
The current do_reset() is called from a command context. Add a function
which can be used from anywhere, as is done on ARM. Adjust do_reset()
to call it.
Note that reset_cpu() is normally provided by SYSRESET so make this
declaration conditional on that being disabled.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Chanho Park <chanho61.park@samsung.com>
Tested-by: Chanho Park <chanho61.park@samsung.com>
This is required on CPUs which always operate in CLIC mode, such as the
T-HEAD E906 and E907. Per the CLIC specification: "In this mode, the
trap vector base address held in mtvec is constrained to be aligned on a
64-byte or larger power-of-two boundary."
Reported-by: Madushan Nishantha <jlmadushan@gmail.com>
Signed-off-by: Samuel Holland <samuel@sholland.org>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
We can remove common.h from most cases of the code here, and only a few
places need an additional header instead.
Signed-off-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Remove dram_init_banksize() on the architecture level.
Limiting used RAM to under 4 GiB is only necessary for CPUs which have a
DMA issue. SoC specific code already exists for FU540, FU740, JH7110.
Not all RISC-V boards will have memory below 4 GiB.
A weak implementation of dram_init_banksize() exists in common/board_f.c.
See the discussion in
https://lore.kernel.org/u-boot/545fe813-cb1e-469c-a131-0025c77aeaa2@canonical.com/T/
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
The Andes PLMT driver directly accesses the mtime MMIO region,
indicating its intended use in the M-mode boot stage. However,
since U-Boot proper (S-mode) also uses the PLMT driver, we need
to specifically mark the region as readable through PMPCFGx (or
S/U-mode read-only shared data region for Smepmp) in OpenSBI.
Granting permission for this case doesn't make sense. Instead,
we should use the generic RISC-V timer driver to read the mtime
through the TIME CSR. Therefore, we add the SPL_ANDES_PLMT_TIMER
config, which ensures that the PLMT driver is linked exclusively
against M-mode U-Boot or U-Boot SPL binaries.
Signed-off-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Samuel Holland <samuel@sholland.org>
With having both an EVENT_SPY_SIMPLE setup for both riscv_cpu_probe and
riscv_cpu_setup we do not need the latter function to call the former
function as it will already have been done in time.
Fixes: 1c55d62fb9 ("riscv: cpu: make riscv_cpu_probe to EVT_DM_POST_INIT_R callback")
Tested-by: Milan P. Stanić <mps@arvanta.net>
Signed-off-by: Tom Rini <trini@konsulko.com>
This function should now be a EVENT_SPY_SIMPLE call, update it.
Tested-by: Milan P. Stanić <mps@arvanta.net>
Reviewed-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Signed-off-by: Tom Rini <trini@konsulko.com>
Starfive JH7110 needs to clear L2 LIM to zero before use or ECC error
would be triggered. Currently, we use DDR ram for SPL malloc arena on
Visionfive 2 board in defconfig, but it's also possible to use L2 LIM as
SPL malloc arena. To avoid triggering ECC error in this scenario, we
imply SPL_SYS_MALLOC_CLEAR_ON_INIT as default.
Signed-off-by: Bo Gan <ganboing@gmail.com>
Signed-off-by: Shengyu Qu <wiagn233@outlook.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Very few of the existing event-spy records use the arguments they are
passed. Update them to use a simple spy instead, to simplify the code.
Where an adaptor function is currently used, remove it where possible.
Signed-off-by: Simon Glass <sjg@chromium.org>
Since the Patch 55171aedda, VisionFive2 booting has been broken [1].
VisionFive2 board requires to enable CONFIG_TIMER_EARLY but booting went
to panic from initr_dm_devices due to lack of a timer device.
- Error logs
initcall sequence 00000000fffd8d38 failed at call 00000000402185e4
(err=-19)
Thus, we need to move riscv_cpu_probe function in order to register
the timer earlier than initr_dm_devices.
Fixes: 7fe32b3442 ("event: Convert arch_cpu_init_dm() to use events")
Cc: Simon Glass <sjg@chromium.org>
Cc: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Chanho Park <chanho61.park@samsung.com>
Tested-by: Milan P. Stanić <mps@arvanta.net>
Tested-by: Roland Ruckerbauer <mail@ruabmbua.dev>
Tested-by: Roland Ruckerbauer <mail@ruabmbua.dev>
board_get_usable_ram_top() returns a physical address that is stored in
gd->ram_top. The return type of the function should be phys_addr_t like the
current type of gd->ram_top.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Add Kconfig item for Starfive JH7110 to select SPL_ZERO_MEM_BEFORE_USE.
Signed-off-by: Bo Gan <ganboing@gmail.com>
Signed-off-by: Shengyu Qu <wiagn233@outlook.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Add the actual support code for SPL_ZERO_MEM_BEFORE_USE and remove
existing Starfive JH7110's L2 LIM clean code, since existing code has
following issues:
1. Each hart (in the middle of a function call) overwriting its own
stack and other harts' stacks.
(data-race and data-corruption)
2. Lottery winner hart can be doing "board_init_f_init_reserve",
while other harts are in the middle of zeroing L2 LIM.
(data-race)
Signed-off-by: Bo Gan <ganboing@gmail.com>
Signed-off-by: Shengyu Qu <wiagn233@outlook.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Some device driver need SYS_CACHELINE_SIZE macro. Add StarFive
SYS_CACHE_SHIFT_6 to enable it.
Signed-off-by: Minda Chen <minda.chen@starfivetech.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
The USB 3.0 driver xhci-mem.c requires CONFIG_SYS_CACHELINE_SIZE to be set.
Define the cache line size for QEMU on RISC-V to be 64 bytes.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Harts need to use per-hart stack before any function call, even if that
function is a simple one. When the callee uses stack for register save/
restore, especially RA, if nested call, concurrent access by multiple
harts on the same stack will cause data-race.
This patch sets up SP before `board_init_f_alloc_reserve`. A side effect
of this is that the memory layout has changed as the following:
+----------------+ +----------------+ <----- SPL_STACK/
| ...... | | hart 0 stack | SYS_INIT_SP_ADDR
| malloc_base | +----------------+
+----------------+ | hart 1 stack |
| GD | +----------------+ If not SMP, N=1
+----------------+ | ...... |
| hart 0 stack | +----------------+
+----------------+ ==> | hart N-1 stack|
| hart 1 stack | +----------------+
+----------------+ | ...... |
| ...... | | malloc_base |
+----------------+ +----------------+
| hart N-1 stack| | GD |
+----------------+ +----------------+
| | | |
Signed-off-by: Bo Gan <ganboing@gmail.com>
Cc: Rick Chen <rick@andestech.com>
Cc: Leo <ycliang@andestech.com>
Cc: Sean Anderson <seanga2@gmail.com>
Cc: Bin Meng <bmeng.cn@gmail.com>
Cc: Lukas Auer <lukas.auer@aisec.fraunhofer.de>
Reviewed-by: Rick Chen <rick@andestech.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
As the RISC-V ACLINT specification is defined to be backward compatible
with the SiFive CLINT specification, we rename SiFive CLINT to RISC-V
ALINT in the source tree to be future-proof.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Rick Chen <rick@andestech.com>
StarFive VisionFive 2 has two versions, 1.2A and 1.3B, each version of
DDR capacity includes 2G/4G/8G, a DT can not support multiple
capacities, so the capacity size information is recorded to EEPROM, when
DDR initialization required capacity size information is read from
EEPROM.
If there is no information in EEPROM, it is initialized with the default
size defined in DT.
Signed-off-by: Yanhong Wang <yanhong.wang@starfivetech.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
In current linker script both .efi_runtime_rel and .rela.dyn sections
are of RELA type whose entry size is either 12 (RV32) or 24 (RV64).
These two are arranged as a continuous region on purpose so that the
prelink-riscv executable can fix up the PIE addresses in one loop.
However there is an 'ALIGN(8)' between these 2 sections which might
cause a gap to be inserted between these 2 sections to satisfy the
alignment requirement on RV32. This would break the assumption of
the prelink process and generate an unbootable image.
Fixes: 9a6569a043 ("riscv: Update alignment for some sections in linker scripts")
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Rick Chen <rick@andestech.com>
The original function was only called once, before relocation. The new
one is called again after relocation. This was not the intent of the
original call. Fix this by renaming and updating the calling logic.
With this, chromebook_link64 makes it through SPL.
Fixes: 7fe32b3442 ("event: Convert arch_cpu_init_dm() to use events")
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Some sections in the linker scripts are aligned to 4 bytes, which
may cause misaligned exception on some platforms, e.g.: clearing
the bss section on 64-bit hardware if __bss_start does not start
from a naturally 8 bytes aligned address.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
U-Boot SPL is not relocable. Drop these relocation sections.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
board_init_r does not return for U-Boot SPL hence there is no need
to update the link register when jumping to board_init_r.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
The codes currently skip the very first relocation entry, and have
an inaccurate comment "skip first reserved entry" indicating that
the first entry is reserved, but later it references the elements
in the first relocation entry using a minus offset.
Change to use a positive offset so that there is no need to skip
the first relocation entry.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
't5' already contains relocation type so don't bother reloading it.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Rick Chen <rick@andestech.com>
The __bss_start is the source end address hence load its address
directly into register 't2' for optimization.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Rick Chen <rick@andestech.com>
Add Kconfig to select the basic functions for StarFive JH7110 SoC.
Signed-off-by: Yanhong Wang <yanhong.wang@starfivetech.com>
Tested-by: Conor Dooley <conor.dooley@microchip.com>
Add StarFive JH7110 SoC to support RISC-V arch.
Signed-off-by: Yanhong Wang <yanhong.wang@starfivetech.com>
Reviewed-by: Rick Chen <rick@andestech.com>
Tested-by: Conor Dooley <conor.dooley@microchip.com>
The current ae350-related defconfigs could also
support newer Andes CPU IP, so modify the names of CPU
from ax25 to andesv5, and board name from ax25-ae350 to ae350.
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Rick Chen <rick@andestech.com>
To reduce the code size, CONFIG_V5L2_CACHE was disabled since commit:
ca06444aac
Turing on does not significantly increase the size of u-boot-spl.bin,
so we enable it by default to improve performance.
Signed-off-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Rick Chen <rick@andestech.com>
This patch refines L1 cache enable/disable and v5l2-cache enable
functions.
Signed-off-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Rick Chen <rick@andestech.com>
As the OpenSBI v1.2 does not enable the cache [0], we enable
the i/d-cache in harts_early_init() and do not disable in
cleanup_before_linux(). This patch also simplifies the logic
and moves the CSR encoding to include/asm/arch-andes/csr.h.
[0] bd7ef41398
Signed-off-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Rick Chen <rick@andestech.com>
There is no need for RISCV_NDS_CACHE config to control cache switches.
Signed-off-by: Leo Yu-Chi Liang <ycliang@andestech.com>
Reviewed-by: Yu Chien Peter Lin <peterlin@andestech.com>
Reviewed-by: Rick Chen <rick@andestech.com>
