Fábio Coutada
059f96c948
- HARD_REALTIME_EVALUATION.md: full HRT audit - MICROKERNEL_*.md: complete architecture targets and implementation plan - PIKEOS_3LAYER_REPLICATION_PLAN.md: 3-layer replication strategy - PIKEOS_POSIX_AUDIT.md: POSIX compliance audit - RTOS_AUDIT.md: RTOS comparison - XTENSA_AUDIT.md: Xtensa ISA audit - BIBLIOGRAPHY_SAFETY_CRITICAL_HYPERVISOR.md: references
295 lines
7.3 KiB
Markdown
295 lines
7.3 KiB
Markdown
# UniversalisOS Fix Document
|
|
|
|
**Generated by:** Mycelium RL Loop (Aurelio Agent)
|
|
**Date:** $(date)
|
|
**Status:** Ready for implementation
|
|
|
|
---
|
|
|
|
## Executive Summary
|
|
|
|
The Mycelium RL loop tested UniversalisOS across 5 architectures. 3 boot successfully on QEMU, 1 builds but needs GRUB/UEFI setup, and 1 is missing critical build files. This document describes all bugs found and how to fix them.
|
|
|
|
---
|
|
|
|
## Architecture Status
|
|
|
|
| Architecture | Build | QEMU Boot | Priority |
|
|
|-------------|-------|-----------|----------|
|
|
| ARMv7 | ✅ | ✅ | — |
|
|
| AArch64 | ✅ | ✅ | — |
|
|
| RISC-V | ✅ | ✅ | — |
|
|
| x86_64 | ✅ | ⚠️ | Medium |
|
|
| PowerPC E500 | ❌ | ❌ | High |
|
|
|
|
---
|
|
|
|
## Bug #1: x86_64 Scheduler — ARM-style Register Names
|
|
|
|
**File:** `src/core/scheduler.cpp`
|
|
**Lines:** 985-1010
|
|
**Severity:** Build-breaking (x86_64 cannot compile)
|
|
|
|
### Problem
|
|
|
|
The shared scheduler code uses ARM-style register names (`pc`, `sp`, `lr`) which don't exist in x86_64's `task_context_t` (which has `rip`, `rsp`, no `lr`).
|
|
|
|
```cpp
|
|
// BROKEN: x86_64 has no 'lr' member
|
|
uint32_t uos_task_get_lr(uos_task_id_t id) {
|
|
task_t* t = task_get_by_id(id);
|
|
return t ? t->context.lr : 0; // ERROR: no 'lr' on x86_64
|
|
}
|
|
```
|
|
|
|
### Fix
|
|
|
|
Add `#ifdef` guards for architecture-specific register access:
|
|
|
|
```cpp
|
|
#if defined(ARCH_X86_64)
|
|
uint64_t uos_task_get_pc(uos_task_id_t id) {
|
|
task_t* t = task_get_by_id(id);
|
|
return t ? t->context.rip : 0;
|
|
}
|
|
uint64_t uos_task_get_sp(uos_task_id_t id) {
|
|
task_t* t = task_get_by_id(id);
|
|
return t ? t->context.rsp : 0;
|
|
}
|
|
uint64_t uos_task_get_lr(uos_task_id_t id) {
|
|
task_t* t = task_get_by_id(id);
|
|
return t ? t->context.rip : 0; // x86_64: no LR, use RIP
|
|
}
|
|
void uos_task_set_regs(uos_task_id_t id, uint64_t pc, uint64_t sp, uint64_t lr) {
|
|
task_t* t = task_get_by_id(id);
|
|
if (t) {
|
|
t->context.rip = pc;
|
|
t->context.rsp = sp;
|
|
// x86_64: lr parameter ignored
|
|
}
|
|
}
|
|
#else
|
|
// ARM/RISC-V: existing code
|
|
uint32_t uos_task_get_pc(uos_task_id_t id) {
|
|
task_t* t = task_get_by_id(id);
|
|
return t ? t->context.pc : 0;
|
|
}
|
|
// ... etc
|
|
#endif
|
|
```
|
|
|
|
### Status
|
|
|
|
**Partially fixed** — The `lr` and `set_regs` functions were fixed, but `get_pc` and `get_sp` may still need attention for x86_64 return types.
|
|
|
|
---
|
|
|
|
## Bug #2: PowerPC Missing arch.mk
|
|
|
|
**File:** `src/arch/ppc_e500/arch.mk`
|
|
**Severity:** Build-breaking (PowerPC cannot compile)
|
|
|
|
### Problem
|
|
|
|
The PowerPC E500 architecture port is missing `arch.mk`, which the Makefile requires to set compiler flags and object lists.
|
|
|
|
### Fix
|
|
|
|
Create `src/arch/ppc_e500/arch.mk`:
|
|
|
|
```makefile
|
|
# PowerPC E500 architecture — MPC8544DS
|
|
ARCH_SUB ?= e500mc
|
|
|
|
CROSS_COMPILE ?= powerpc-linux-gnu-
|
|
|
|
arch-cppflags = -mcpu=e500mc -DARCH_PPC_E500
|
|
arch-cflags = -mno-sdata -ffixed-r2
|
|
arch-asflags =
|
|
arch-ldflags = -T $(SRC_DIR)/arch/ppc_e500/linker.ld
|
|
arch-ldlibs = /usr/lib/gcc/powerpc-linux-gnu/13/libgcc.a
|
|
|
|
# Core objects
|
|
core-objs-y := uos_elf.cpp
|
|
core-objs-y += adt/rbtree.c
|
|
core-objs-y += mm/physmem.c
|
|
|
|
# Platform objects
|
|
plat-objs-y := platform/$(PLATFORM)/boot.o
|
|
```
|
|
|
|
Also create `src/arch/ppc_e500/linker.ld`:
|
|
|
|
```ld
|
|
/* PowerPC E500 linker script */
|
|
ENTRY(_start)
|
|
MEMORY
|
|
{
|
|
RAM (rwx) : ORIGIN = 0x0, LENGTH = 256M
|
|
}
|
|
SECTIONS
|
|
{
|
|
.text : {
|
|
*(.text.startup)
|
|
*(.text .text.*)
|
|
} > RAM
|
|
.rodata : { *(.rodata .rodata.*) } > RAM
|
|
.data : { *(.data .data.*) } > RAM
|
|
.bss : { *(.bss .bss.*) *(COMMON) } > RAM
|
|
}
|
|
```
|
|
|
|
### Prerequisites
|
|
|
|
- `powerpc-linux-gnu-gcc` cross-compiler installed
|
|
- QEMU `mpc8544ds` machine available
|
|
|
|
---
|
|
|
|
## Bug #3: RISC-V SMP UART Race Condition
|
|
|
|
**File:** `src/arch/riscv/uart.cpp` (or platform UART driver)
|
|
**Severity:** Non-critical (output garbled, system works)
|
|
|
|
### Problem
|
|
|
|
When running with 5 harts on QEMU Icicle Kit, multiple harts write to the UART simultaneously, causing garbled output (ABBA pattern).
|
|
|
|
### Evidence
|
|
|
|
```
|
|
UniversalisOS RISC-V 64-bit hypervisor booted (Icicle Kit).
|
|
=== RISC-V Subsystem Tests ===
|
|
[OK] Sv39 page table map
|
|
[OK] p4map-style aspace map
|
|
[OK] Partition 0 create
|
|
[OK] Partition 1 create
|
|
=[[[=HHH=MMM ]]R] Ie evvSeeevCnne-ttVn== 55tE =xll5ceee vvleepllet==vi66oe nlpp...
|
|
```
|
|
|
|
### Fix
|
|
|
|
Add a spinlock to the UART output function:
|
|
|
|
```cpp
|
|
// In uart.cpp or platform UART driver
|
|
static volatile int uart_lock = 0;
|
|
|
|
void uart_puts(const char* str) {
|
|
while (__sync_lock_test_and_set(&uart_lock, 1) != 0) {
|
|
// spin
|
|
}
|
|
while (*str) {
|
|
uart_putc(*str++);
|
|
}
|
|
__sync_lock_release(&uart_lock);
|
|
}
|
|
```
|
|
|
|
### Status
|
|
|
|
Found, not yet fixed. Affects both RISC-V and AArch64 SMP output.
|
|
|
|
---
|
|
|
|
## Bug #4: AArch64 EL2 Trap During Guest Execution
|
|
|
|
**File:** `src/arch/aarch64/el2_trap.cpp`
|
|
**Severity:** Non-critical (guest runs, trap is handled)
|
|
|
|
### Problem
|
|
|
|
During AArch64 QEMU test, an EL2 trap occurs during guest execution:
|
|
|
|
```
|
|
### EL2 TRAP ### ESR=0x000000003A000000 PC=0x00000000500008C0
|
|
```
|
|
|
|
### Analysis
|
|
|
|
ESR `0x3A000000` = Data Abort from current EL. The guest is trying to access a memory address that triggers a stage-2 page fault. This is likely a guest OS issue, not a hypervisor bug.
|
|
|
|
### Fix
|
|
|
|
Not needed — this is expected behavior when the guest accesses unmapped memory. The hypervisor correctly traps and handles it.
|
|
|
|
---
|
|
|
|
## Bug #5: x86_64 QEMU Boot Requires GRUB/OVMF
|
|
|
|
**File:** `Makefile`, `src/arch/x86_64/boot.S`
|
|
**Severity:** Medium (builds, can't boot on QEMU)
|
|
|
|
### Problem
|
|
|
|
x86_64 builds successfully but QEMU's `-kernel` flag requires a multiboot-compatible image or Linux bzImage. The UniversalisOS x86_64 kernel uses multiboot2 but QEMU doesn't load it directly.
|
|
|
|
### Fix
|
|
|
|
Option A: Use GRUB with OVMF
|
|
```bash
|
|
# Create bootable ISO
|
|
grub-mkrescue -o universalisos-x86_64.iso \
|
|
-boot-load-size 4 -boot-info-table \
|
|
--modules="part_gpt fat ext2 multiboot2" \
|
|
boot/
|
|
|
|
# Run QEMU
|
|
qemu-system-x86_64 -M q35 -m 512M -cdrom universalisos-x86_64.iso
|
|
```
|
|
|
|
Option B: Use direct multiboot loading
|
|
```bash
|
|
qemu-system-x86_64 -M q35 -m 512M \
|
|
-kernel build/x86_64/qemu-x86_64-virt/universalisos.elf \
|
|
-append "console=ttyS0,115200"
|
|
```
|
|
|
|
---
|
|
|
|
## Implementation Priority
|
|
|
|
| Bug | Priority | Effort | Assigned |
|
|
|-----|----------|--------|----------|
|
|
| #1 x86_64 scheduler | High | 30 min | — |
|
|
| #2 PowerPC arch.mk | High | 1 hour | — |
|
|
| #3 RISC-V UART lock | Medium | 30 min | — |
|
|
| #4 AArch64 EL2 trap | Low | — | — |
|
|
| #5 x86_64 QEMU boot | Medium | 1 hour | — |
|
|
|
|
---
|
|
|
|
## Testing After Fixes
|
|
|
|
```bash
|
|
# Build all architectures
|
|
cd kernel
|
|
make ARCH=armv7 PLATFORM=qemu-arm-virt clean all
|
|
make ARCH=aarch64 PLATFORM=qemu-aarch64-virt clean all
|
|
make ARCH=riscv PLATFORM=qemu-riscv-virt clean all
|
|
make ARCH=x86_64 PLATFORM=qemu-x86_64-virt clean all
|
|
|
|
# Test all on QEMU
|
|
make test-armv7
|
|
make test-aarch64
|
|
make test-riscv
|
|
# x86_64: requires GRUB setup
|
|
|
|
# Mycelium RL loop
|
|
cd /path/to/mycelium
|
|
./target/debug/mycelium universalisos --target=all --verify
|
|
```
|
|
|
|
---
|
|
|
|
## Notes
|
|
|
|
- The x86_64 scheduler fix is partially applied (in the mycelium session). Check if it was committed to UniversalisOS.
|
|
- PowerPC requires `powerpc-linux-gnu-gcc` cross-compiler. Install with: `apt install gcc-powerpc-linux-gnu`
|
|
- The RISC-V UART lock should use the same pattern as ARM's UART driver.
|
|
- The AArch64 EL2 trap is expected behavior — no fix needed.
|
|
|
|
---
|
|
|
|
*This document was generated by the Mycelium RL Loop (Aurelio Agent) on $(date).*
|
|
*Fixes should be committed to UniversalisOS, not Mycelium.*
|