Currently, the initrd is placed at 128MB, which overlaps with the kernel
when it is large (for example syzbot kernels are). From the kernel side,
there is no reason we could not push the initrd further away in memory
to accommodate large kernels, so move the initrd at 512MB when possible.
The ideal solution would have been to place the initrd based on the
kernel size but we actually can't since the bss size is not known when
the image is loaded by load_image_targphys_as() and the initrd would
then overlap with this section.
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Message-ID: <20240206154042.514698-1-alexghiti@rivosinc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
The microchip_icicle_kit, sifive_u, spike and virt boards are now doing
the same steps when '-kernel' is used:
- execute load_kernel()
- load init_rd()
- write kernel_cmdline
Let's fold everything inside riscv_load_kernel() to avoid code
repetition. To not change the behavior of boards that aren't calling
riscv_load_init(), add an 'load_initrd' flag to riscv_load_kernel() and
allow these boards to opt out from initrd loading.
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20230206140022.2748401-3-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Next patch will move all calls to riscv_load_initrd() to
riscv_load_kernel(). Machines that want to load initrd will be able to
do via an extra flag to riscv_load_kernel().
This change will expose a sign-extend behavior that is happening in
load_elf_ram_sym() when running 32 bit guests [1]. This is currently
obscured by the fact that riscv_load_initrd() is using the return of
riscv_load_kernel(), defined as target_ulong, and this return type will
crop the higher 32 bits that would be padded with 1s by the sign
extension when running in 32 bit targets. The changes to be done will
force riscv_load_initrd() to use an uint64_t instead, exposing it to the
padding when dealing with 32 bit CPUs.
There is a discussion about whether load_elf_ram_sym() should or should
not sign extend the value returned by 'lowaddr'. What we can do is to
prevent the behavior change that the next patch will end up doing.
riscv_load_initrd() wasn't dealing with 64 bit kernel entries when
running 32 bit CPUs, and we want to keep it that way.
One way of doing it is to use target_ulong in 'kernel_entry' in
riscv_load_kernel() and rely on the fact that this var will not be sign
extended for 32 bit targets. Another way is to explictly clear the
higher 32 bits when running 32 bit CPUs for all possibilities of
kernel_entry.
We opted for the later. This will allow us to be clear about the design
choices made in the function, while also allowing us to add a small
comment about what load_elf_ram_sym() is doing. With this change, the
consolation patch can do its job without worrying about unintended
behavioral changes.
[1] https://lists.gnu.org/archive/html/qemu-devel/2023-01/msg02281.html
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20230206140022.2748401-2-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
As it is now, riscv_compute_fdt_addr() is receiving a dram_base, a
mem_size (which is defaulted to MachineState::ram_size in all boards)
and the FDT pointer. And it makes a very important assumption: the DRAM
interval dram_base + mem_size is contiguous. This is indeed the case for
most boards that use a FDT.
The Icicle Kit board works with 2 distinct RAM banks that are separated
by a gap. We have a lower bank with 1GiB size, a gap follows, then at
64GiB the high memory starts. MachineClass::default_ram_size for this
board is set to 1.5Gb, and machine_init() is enforcing it as minimal RAM
size, meaning that there we'll always have at least 512 MiB in the Hi
RAM area.
Using riscv_compute_fdt_addr() in this board is weird because not only
the board has sparse RAM, and it's calling it using the base address of
the Lo RAM area, but it's also using a mem_size that we have guarantees
that it will go up to the Hi RAM. All the function assumptions doesn't
work for this board.
In fact, what makes the function works at all in this case is a
coincidence. Commit 1a475d39ef introduced a 3GB boundary for the FDT,
down from 4Gb, that is enforced if dram_base is lower than 3072 MiB. For
the Icicle Kit board, memmap[MICROCHIP_PFSOC_DRAM_LO].base is 0x80000000
(2 Gb) and it has a 1Gb size, so it will fall in the conditions to put
the FDT under a 3Gb address, which happens to be exactly at the end of
DRAM_LO. If the base address of the Lo area started later than 3Gb this
function would be unusable by the board. Changing any assumptions inside
riscv_compute_fdt_addr() can also break it by accident as well.
Let's change riscv_compute_fdt_addr() semantics to be appropriate to the
Icicle Kit board and for future boards that might have sparse RAM
topologies to worry about:
- relieve the condition that the dram_base + mem_size area is contiguous,
since this is already not the case today;
- receive an extra 'dram_size' size attribute that refers to a contiguous
RAM block that the board wants the FDT to reside on.
Together with 'mem_size' and 'fdt', which are now now being consumed by a
MachineState pointer, we're able to make clear assumptions based on the
DRAM block and total mem_size available to ensure that the FDT will be put
in a valid RAM address.
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20230201171212.1219375-4-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
A common trend in other archs is to calculate the fdt address, which is
usually straightforward, and then calling a function that loads the
fdt/dtb by using that address.
riscv_load_fdt() is doing a bit too much in comparison. It's calculating
the fdt address via an elaborated heuristic to put the FDT at the bottom
of DRAM, and "bottom of DRAM" will vary across boards and
configurations, then it's actually loading the fdt, and finally it's
returning the fdt address used to the caller.
Reduce the existing complexity of riscv_load_fdt() by splitting its code
into a new function, riscv_compute_fdt_addr(), that will take care of
all fdt address logic. riscv_load_fdt() can then be a simple function
that just loads a fdt at the given fdt address.
We're also taken the opportunity to clarify the intentions and
assumptions made by these functions. riscv_load_fdt() is now receiving a
hwaddr as fdt_addr because there is no restriction of having to load the
fdt in higher addresses that doesn't fit in an uint32_t.
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Message-Id: <20230201171212.1219375-3-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
riscv_load_initrd() returns the initrd end addr while also writing a
'start' var to mark the addr start. These informations are being used
just to write the initrd FDT node. Every existing caller of
riscv_load_initrd() is writing the FDT in the same manner.
We can simplify things by writing the FDT inside riscv_load_initrd(),
sparing callers from having to manage start/end addrs to write the FDT
themselves.
An 'if (fdt)' check is already inserted at the end of the function
because we'll end up using it later on with other boards that doesn´t
have a FDT.
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20230102115241.25733-7-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
riscv_load_firmware(), riscv_load_initrd() and riscv_load_kernel() works
under the assumption that a 'filename' parameter is always not NULL.
This is currently the case since all callers of these functions are
checking for NULL before calling them. Add an g_assert() to make sure
that a NULL value in these cases are to be considered a bug.
Suggested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20230102115241.25733-5-dbarboza@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Rename previous riscv_find_firmware() to riscv_find_bios(), and
introduce a new riscv_find_firmware() to implement the first half
part of the work done in riscv_find_and_load_firmware().
This new API is helpful for machine that wants to know the final
chosen firmware file name but does not want to load it.
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Reviewed-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20221229091828.1945072-12-bmeng@tinylab.org>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Some boards are duplicating the 'riscv_find_and_load_firmware' call
because the 32 and 64 bits images have different names. Create
a function to handle this detail instead of hardcoding it in the boards.
Ideally we would bake this logic inside riscv_find_and_load_firmware(),
or even create a riscv_load_default_firmware(), but at this moment we
cannot infer whether the machine is running 32 or 64 bits without
accessing RISCVHartArrayState, which in turn can't be accessed via the
common code from boot.c. In the end we would exchange 'firmware_name'
for a flag with riscv_is_32bit(), which isn't much better than what we
already have today.
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Signed-off-by: Daniel Henrique Barboza <dbarboza@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng@tinylab.org>
Signed-off-by: Bin Meng <bmeng@tinylab.org>
Message-Id: <20221221182300.307900-6-dbarboza@ventanamicro.com>
Message-Id: <20221229091828.1945072-11-bmeng@tinylab.org>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
To boot S-mode firmware payload like EDK2 from persistent
flash storage, qemu needs to pass the flash address as the
next_addr in fw_dynamic_info to the opensbi.
When both -kernel and -pflash options are provided in command line,
the kernel (and initrd if -initrd) will be copied to fw_cfg table.
The S-mode FW will load the kernel/initrd from fw_cfg table.
If only pflash is given but not -kernel, then it is the job of
of the S-mode firmware to locate and load the kernel.
In either case, update the kernel_entry with the flash address
so that the opensbi can jump to the entry point of the S-mode
firmware.
Signed-off-by: Sunil V L <sunilvl@ventanamicro.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20221004092351.18209-4-sunilvl@ventanamicro.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
The current riscv_load_fdt() forces fdt_load_addr to be placed at a dram address within 3GB,
but not all platforms have dram_base within 3GB.
This patch adds an exception for dram base not within 3GB,
which will place fdt at dram_end align 16MB.
riscv_setup_rom_reset_vec() also needs to be modified
Signed-off-by: Dylan Jhong <dylan@andestech.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20220419115945.37945-1-dylan@andestech.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
The original BBL boot method had the kernel embedded as an opaque blob
that was blindly jumped to, which OpenSBI implemented as fw_payload.
OpenSBI then implemented fw_jump, which allows the payload to be loaded
elsewhere, but still blindly jumps to a fixed address at which the
kernel is to be loaded. Finally, OpenSBI introduced fw_dynamic, which
allows the previous stage to inform it where to jump to, rather than
having to blindly guess like fw_jump, or embed the payload as part of
the build like fw_payload. When used with an opaque binary (i.e. the
output of objcopy -O binary), it matches the behaviour of the previous
methods. However, when used with an ELF, QEMU currently passes on the
ELF's entry point address, which causes a discrepancy compared with all
the other boot methods if that entry point is not the first instruction
in the binary.
This difference specific to fw_dynamic with an ELF is not apparent when
booting Linux, since its entry point is the first instruction in the
binary. However, FreeBSD has a separate ELF entry point, following the
calling convention used by its bootloader, that differs from the first
instruction in the binary, used for the legacy SBI entry point, and so
the specific combination of QEMU's default fw_dynamic firmware with
booting FreeBSD as an ELF rather than a raw binary does not work.
Thus, align the behaviour when loading an ELF with the behaviour when
loading a raw binary; namely, use the base address of the loaded kernel
in place of the entry point.
The uImage code is left as-is in using the U-Boot header's entry point,
since the calling convention for that entry point is the same as the SBI
one and it mirrors what U-Boot will do.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20211214032456.70203-1-jrtc27@jrtc27.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Currently, we place the DTB at 2MB from 4GB or end of DRAM which ever is
lesser. However, Linux kernel can address only 1GB of memory for RV32.
Thus, it can not map anything beyond 3GB (assuming 2GB is the starting address).
As a result, it can not process DT and panic if opensbi dynamic firmware
is used. While at it, place the DTB further away to avoid in memory placement
issues in future.
Fix this by placing the DTB at 16MB from 3GB or end of DRAM whichever is lower.
Fixes: 66b1205bc5 ("RISC-V: Copy the fdt in dram instead of ROM")
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Tested-by: Bin Meng <bin.meng@windriver.com>
Signed-off-by: Atish Patra <atish.patra@wdc.com>
Message-id: 20210107091127.3407870-1-atish.patra@wdc.com
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
OpenSBI is the default firmware in Qemu and has various firmware loading
options. Currently, qemu loader uses fw_jump which has a compile time
pre-defined address where fdt & kernel image must reside. This puts a
constraint on image size of the Linux kernel depending on the fdt location
and available memory. However, fw_dynamic allows the loader to specify
the next stage location (i.e. Linux kernel/U-Boot) in memory and other
configurable boot options available in OpenSBI.
Add support for OpenSBI dynamic firmware loading support. This doesn't
break existing setup and fw_jump will continue to work as it is. Any
other firmware will continue to work without any issues as long as it
doesn't expect anything specific from loader in "a2" register.
Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Tested-by: Bin Meng <bin.meng@windriver.com>
Message-Id: <20200701183949.398134-4-atish.patra@wdc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Currently, the fdt is copied to the ROM after the reset vector. The firmware
has to copy it to DRAM. Instead of this, directly copy the device tree to a
pre-computed dram address. The device tree load address should be as far as
possible from kernel and initrd images. That's why it is kept at the end of
the DRAM or 4GB whichever is lesser.
Signed-off-by: Atish Patra <atish.patra@wdc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Tested-by: Bin Meng <bin.meng@windriver.com>
Message-Id: <20200701183949.398134-3-atish.patra@wdc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Per QEMU deprecated doc, QEMU 4.1 introduced support for the -bios
option in QEMU for RISC-V for the virt machine and sifive_u machine.
The default behavior has been that QEMU does not automatically load
any firmware if no -bios option is included.
Now 2 releases passed, it's time to change the default behavior to
load the default OpenSBI firmware automatically. The firmware is
included with the QEMU release and no user interaction is required.
All a user needs to do is specify the kernel they want to boot with
the -kernel option.
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-id: 1588335545-649-1-git-send-email-bmeng.cn@gmail.com
Message-Id: <1588335545-649-1-git-send-email-bmeng.cn@gmail.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
This patch adds an optional function pointer, "sym_cb", to
riscv_load_firmware() which provides the possibility to access
the symbol table during kernel loading.
The pointer is ignored, if supplied with flat (non-elf) firmware image.
The Spike board requires it locate the HTIF symbols from firmware ELF
passed via "-bios" option.
Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-id: 20200427080644.168461-2-anup.patel@wdc.com
Message-Id: <20200427080644.168461-2-anup.patel@wdc.com>
Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
This patch adds an optional function pointer, "sym_cb", to
riscv_load_kernel() which provides the possibility to access the symbol
table during kernel loading.
The pointer is ignored, if supplied with Image or uImage file.
The Spike board requires the access to locate the HTIF symbols.
Fixes: 0ac24d56c5 ("hw/riscv: Split out the boot functions")
Buglink: https://bugs.launchpad.net/qemu/+bug/1835827
Signed-off-by: Siwei Zhuang <siwei.zhuang@data61.csiro.au>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
