The original names are confusing; the symbols' addresses point to the start and end of the
sbi executable ram area. This may also reside in l2lim and not in ddr, depending on the
configuration, and this is defined in the linker script.
Signed-off-by: Jukka Laitinen <jukkax@ssrc.tii.ae>
Version 1.3.1 is the latest tagged version as of November
the 21st, 2023. This patch prepares the required changes
to make v1.3.1 work.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
Fix:
```
riscv-none-elf-ld: sotest.o: ABI is incompatible with that of the selected emulation:
target emulation `elf64-littleriscv' does not match `elf32-littleriscv'
riscv-none-elf-ld: failed to merge target specific data of file sotest.o
```
Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
SiFive document: "ECC Error Handling Guide" states:
"Any SRAM block or cache memory containing ECC functionality needs to be
initialized prior to use. ECC will correct defective bits based on memory
contents, so if memory is not first initialized to a known state, then ECC
will not operate as expected. It is recommended to use a DMA, if available,
to write the entire SRAM or cache to zeros prior to enabling ECC reporting.
If no DMA is present, use store instructions issued from the processor."
Clean the cache at this early stage so no ECC errors will be flooding later.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
When I try to set priorities in certain programs, such as init_priority(HIGH_PRIORITY), I've noticed that during linking, there's no guarantee that the programs will be compiled in the sequence I've specified based on priority. This has led to some runtime errors in my program.
I realized that in the ld file, when initializing dynamic arrays, there's no assurance of initializing init_array.* before init_array. This has resulted in runtime errors in the program. Consequently, I've rearranged the init_array.* in the ld file of NuttX to be placed before init_array and added a SORT operation to init_array.* to ensure accurate initialization based on priorities during linking.
Currently the IHC (Inter Hart Communication) depends on OpenAMP and
rptun. However, the bootloader portion of the IHC doesn't need
either of them. Now they are wasting a lot of bootloader space.
Reorganize the bootloader portion into a separate file 'mpfs_ihc_sbi.c'.
This file contains the OpenSBI vendor extensions, or the only required
functionalities for the bootloader. On the other hand, 'mpfs_ihc.c'
contains the non-bootloader code.
This patch also makes it possible to utilize 2 RPMSG channels. This
has been tested so that 2 separate NuttXs on harts 1 and 2 communicate
with Linux kernel that runs on harts 3 and 4.
New configuration files are added as well:
- rpmsg-ch1: sample config for RPMSG
- rpmsg-ch2: sample config for another RPMSG channel
- rpmsg-sbi: sample bootloader config for RPMSG/OpenSBI
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
It is not really needed; g_hart_stacks is only used during SBI init as
a temporary stack area. We can use the scratch area buffers for this, as
the scratch areas define almost 4K of extra space, which is used for
exception stacks anyway.
This provides an example of Asymmetric Multiprocessing (AMP). The
master from Linux sends pings that this NuttX echoes back. The system
uses RPMsg from OpenAMP.
The Inter-Hart Communication module is present in the vendor's software
stack with the tag "2021.11". The software is present on github at the
polarfire-soc project. The following conditions must be met:
1. FPGA programmed with 2021.11 software
2. HSS (Vendor bootloader) with 2021.11 software
3. U-boot and Linux kernel from 2011.11 software
Currently the IHC works as a slave only on the hart number 4.
On the NuttX side, this patch uses rptun that incorporates rpmsg and
virtio. If it used only rpmsg and virtio, the future maintenance would
likely be much heavier. Using rptun also simplifies many things.
Upon success, the master side from Linux may issue an example test:
root@icicle-kit-es-amp:/opt/microchip/amp/rpmsg-pingpong# ./rpmsg-pingpong
However, the rpmsg-pingpong.c (compiled on target with gcc), may need to
be modified as seen below to match the device id:
- char *rpmsg_dev="virtio0.rpmsg-amp-demo-channel.-1.0";
+ char *rpmsg_dev="virtio0.rpmsg-amp-demo-channel.-1.1024";
This work uses a separate linker script. Due to a bug yet unknown to date,
a small NuttX, when loaded by the vendor HSS bootloader, will cause the
Linux kernel to hang at boot. Thus, the binary size is increased with
a section 'filler_area' whose only purpose is to increase the image size
so that the Linux kernel will boot up.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
Mappings are done with vaddr=paddr.
- I/O space mapped with two gigapages
- Kernel space mapped to statically allocated page tables. 2MB of kernel
memory is supported.
- Page pool is mapped to the kernel space, to allow virtual memory access
for the kernel e.g. to initialize the page memory when it is allocated.
Enable this option to optimization the unused input sections with the
linker by compiling with " -ffunction-sections -fdata-sections ", and
linking with " --gc-sections ".
Signed-off-by: chao.an <anchao@xiaomi.com>
## Summary
A lot of linker scripts were listed twice, once for unix, once for windows.
This PR cleans up the logic so they're only listed once.
## Impact
Any opportunity to use a single source of truth and reduce lines of code is a win!
## Testing
CI will test all build
Summary:
- Apply the same style as sabre-6quad
Impact:
- None
- NOTE: esp32c3-devkit still remains old style because of link errors
Testing:
- Build only
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
NOTE: THIS ONLY WORKS WHEN KERNEL RUNS IN M-MODE FOR NOW
This frees the PMP for other use, e.g. HART memory separation.
The page tables are statically allocated, 1 per level.
This feature is now behind CONFIG_MPFS_USE_MMU_AS_MPU, because
only the MPFS target supports this (others are not tested).
If the MMU is used for memory separation within a HART, the PMP must
still be configured to allow user access to the memory mapped for the
HART, because PMP *rekoves* access by default. At this point all of
the user memory as well as the kernel RAM are opened.
A more flexible solution for PMP configuration will follow.
The old implementation used the default ld.script for the kernel side
which did not obey the memory.ld limits whatsoever.
Also, provide the user space addresses from the linker script to get rid
of the pre-processor macros that define (incorrect) default values for
the user space composition.
There is no such section. Instead, place the object mpfs_head.o at the start of
the text.
Put mpfs_head.o directly into the arch library; there is no need to define
it separately in HEAD_ASRC.
Signed-off-by: Jukka Laitinen <jukkax@ssrc.tii.ae>
Add the rest of the OpenSBI code to .text.sbi -section. They belong
to there. This frees up some space in the very limited eNVM.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
Polarfire Icicle board has only (128K - 256) bytes for the bootloader
in the non-volatile eNVM. This space is barely enough for running NuttX.
If OpenSBI is selected, it will be placed in DDR. This all means the
nuttx.bin file grows into gigabyte size, filling the unused space (ddr -
envm) with zeroes.
The memory layout is as follows:
MEMORY
{
ddr (rx) : ORIGIN = 0x80000000, LENGTH = 4M
envm (rx) : ORIGIN = 0x20220100, LENGTH = 128K - 256
l2lim (rwx) : ORIGIN = 0x08000000, LENGTH = 1024k
l2zerodevice (rwx) : ORIGIN = 0x0A000000, LENGTH = 512k
}
OpenSBI library is used as a separate binary, which is stored into
eMMC or SD-card. It is then loaded into its precise location in DDR.
Thus, we separate OpenSBI from NuttX and end up with two images
by utilizing the objcopy options:
--only-section=sectionpattern (-j in short)
--remove-section=sectionpattern (-R in short)
This is only valid when CONFIG_MPFS_OPENSBI is set.
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
OpenSBI may be compiled as an external library. OpenSBI commit d249d65
(Dec. 11, 2021) needs to be reverted as it causes memcpy / memcmp to
end up in the wrong section. That issue has yet no known workaround.
OpenSBI may be lauched from the hart0 (e51). It will start the U-Boot
and eventually the Linux kernel on harts 1-4.
OpenSBI, once initialized properly, will trap and handle illegal
instructions (for example, CSR time) and unaligned address accesses
among other things.
Due to size size limitations for the mpfs eNVM area where the NuttX
is located, we actually set up the OpenSBI on its own section which
is in the bottom of the DDR memory. Special care must be taken so that
the kernel doesn't override the OpenSBI. For example, the Linux device
tree may reserve some space from the beginning:
opensbi_reserved: opensbi@80000000 {
reg = <0x80000000 0x200000>;
label = "opensbi-reserved";
};
The resulting nuttx.bin file is very large, but objcopy is used to
create the final binary images for the regions (eNVM and DDR) using
the nuttx elf file.
Co-authored-by: Petro Karashchenko <petro.karashchenko@gmail.com>
Signed-off-by: Eero Nurkkala <eero.nurkkala@offcode.fi>
