Syscall multitasking

[Dreaded-Gnu]

System call and multitasking stuff.

• [x] Revise assert usage to return errors instead of failing. Assertions should be made after call if necessary
• [x] Check code base against possible memory leaks
• [x] Check again whole code base ( after above completed ), so that nowhere a return of a function call is ignored
• [x] Check for further functions where error handling might be necessary
• [x] Implement virtual memory manager destroy context, which is currently only a wrapped panic to finalise system call implementations thread exit, process exit, process fork and process replace
• [x] Heap
  • [x] Fix data abort within kernel heap, which occurs, when a process is polling again and again a non existent file from vfs daemon.
  • [x] Check heap alignment calculation when a block with exact size has been found.
• [x] Extend avl implementation by lookup and cleanup callback
• [x] Extend list implementation by lookup and cleanup callback
• [x] Add cleanup and lookup helper for avl tree and lists where it is necessary
• [x] Cleanup avl tree helpers ( compare, lookup, delete ) where default functions can be used
• [x] Fix issue within switch to user mode logic
• [x] Add support for thumb mode in user processes
• [x] Rename dummy system calls to kernel putc and puts for early output until console, framebuffer and tty are existing
• [x] Console / TTY
  • [x] Add console server providing and handling stdin, stderr and stdout
  • [x] Add terminal server providing initially handling only output until input driver exists
  • [x] Add framebuffer server handling rendering requests
• [x] Add helpers for copy from / to unsafe necessary for copying stuff from kernel to user space or from user space to kernel
• [x] Replace mailbox system call in kernel by specific driver

System calls

• [x] process_exit
• [x] process_id
• [x] process_fork
• [x] process_replace
• [x] thread_create
• [x] thread_exit
• [x] thread_id
• [x] memory_acquire
• [x] memory_release
• [x] memory_shared_create
• [x] memory_shared_attach
• [x] memory_shared_detach
• [x] memory_translate_physical
• [x] rpc_set_handler
• [x] rpc_raise
• [x] rpc_ret
• [x] rpc_get_data
• [x] rpc_get_data_size
• [x] rpc_wait_for_call
• [x] rpc_set_ready
• [x] rpc_end
• [x] rpc_wait_for_ready
• [x] interrupt_acquire
• [x] interrupt_release
• [x] timer_tick_count
• [x] timer_frequency
• [x] timer_acquire
• [x] timer_release
• [x] kernel_putc
• [x] kernel_puts

Closes #18

[Dreaded-Gnu]

Testing progress / Found and fixed issues

Previous comments already addressed

From time to time the interaction hangs like at some point under special circumstances some return wasn't possible: application <=> vfs <=> console <=> terminal <=> framebuffer

Real hardware testing

• [x] Ensure that kernel and servers work on real hardware ( short descriptor without caches )
• [x] Ensure that kernel and servers work on real hardware ( short descriptor with caches )
• [x] Ensure that kernel and servers work on real hardware ( long descriptor descriptor without caches )
• [x] Ensure that kernel and servers work on real hardware ( long descriptor descriptor with caches )

Delayed / Postponed hardware testing

• [x] Check serial gdb stub with real hardware ( short descriptor without caches )
• [x] Check serial gdb stub with real hardware ( short descriptor with caches )
• [x] Check serial gdb stub with real hardware ( long descriptor without caches )
• [x] Check serial gdb stub with real hardware ( long descriptor with caches )

Emulator testing

Caches are not modelled by emulator

• [x] Ensure that kernel and servers work with qemu ( short descriptor )
• [x] Ensure that kernel and servers work with qemu ( long descriptor )

Not addressed issues

• [x] Check for further optimizations regarding the rendering to screen Related to mmu setup and cachability / sharability. Will be addressed with a separate issue

Fixed issues

• [x] Switch to thread uses unpredictable code
• [x] SPSR within thread creation is wrong / incomplete
• [x] Amount of timer interrupts per second needs to be set to a value of 1
• [x] Interrupts occurring within kernel ( e.g. shortly before return to user process from syscall ) lead to stack smashing failure
• [x] Add optimized versions of libc string implementations like memcpy, memset, ....
• [x] Really bad / slow performance on real hardware also with compile option -s. memcpy and memset needs to be optimized.
• [x] VFP register save and restore leads to an issue, when svc mode is already active. If that's the case, the return from interrupt will return to some invalid address. That behaviour is not bound to enabled VFP register save, but it can be reproduced best with that one enabled.
• [x] Short descriptor mmu doesn't work without cpu caches enabled ( real hardware )
• [x] Short descriptor mmu doesn't work with cpu caches enabled ( real hardware )
• [x] Long descriptor mmu doesn't work without cpu caches enabled ( real hardware )
• [x] Long descriptor mmu doesn't work with cpu caches enabled ( real hardware )
• [x] Fix optimization level -O3, whcih doesn't work ( emulator and real hardware produce same or similar errors )
• [x] Ensure that kernel mappings are not accessible from user mode in long descriptor mmu
• [x] MMU fork behaves differently with -O0 on real hardware and "fails" because forked process memory seems to be not properly copied