Riscv semihosting arguments parsing issue
Description of problem:
When executing a bare metal kernel in semihosting mode with "argv" arguments provided through "-append" or "-semihostgin-config", recovering the arguments will read out-of-bound memory in the guest and often make the emulation crash through a hard fault.
Steps to reproduce:
1. Compile the simplest "kernel" in semihosting mode, e.g. test.c containing:

   `#include <stdio.h>` 

   `#include <stdlib.h>`

   `int main(int argc, char *argv[])` 

   `{ unsigned int i; printf("[+] Test OK!\n"); for(i = 0; i < argc; i++){ printf("Arg %d is '%s'\n", i, argv[i]); } return 0; }` 
2. One can compile the previous example under Debian with 

   `riscv64-unknown-elf-gcc -static -fPIC -march=rv64imac -mabi=lp64 -specs=picolibc.specs --crt0=semihost --oslib=semihost -Triscv_layout.ld test.c -o test`

   This supposes the bare metal cross-toolchain `binutils-riscv64-unknown-elf` to be installed, as well as the standard library picolibc `picolibc-riscv64-unknown-elf` that supports semihosting.

   The `riscv_layout`.ld file contains:

   `__flash = 0x80000000; __flash_size = 0x00080000; __ram = 0x80080000; __ram_size = 0x40000; __stack_size = 1k; INCLUDE picolibc.ld`
3. Execute the command line:
   - `qemu-system-riscv64 -semihosting-config enable=on -monitor none -serial none -nographic -machine sifive_u,accel=tcg -no-reboot -bios none -kernel /tmp/test -append "10 20"`
4. Observe the following as an example of the bug:

   `[+] Test OK! Arg rg A is '/tmp/test0' Arg Arg 2 is 'RISCV fault x0 -2146959259z' ero 0x0000000000000000 x1 ra 0x0000000080000e94 x2 sp 0x00000000800bfea0 x3 gp 0x0000000080080820 x4 tp 0x0000000080080028 x5 t0 0x00000000800019a2 x6 t1 0x000000000000002a x7`
Additional information:
The same bug seems to affect risv32 and riscv64 system emulators. The semihosting part seems to work well when there is no access to the "argv" arguments.