SchOS/kernel/core/main.c

#include "types.h"
#include "defs.h"
#include "param.h"
#include "memlayout.h"
#include "mmu.h"
#include "proc.h"
#include "x86.h"
#include "irq.h"
#include "dma.h"

static void startothers(void);
static void mpmain(void)  __attribute__((noreturn));
extern char end[]; // first address after kernel loaded from ELF file

// Set by stage2main.c (boot/bios/stage2main.c, SCHBOOT_VERBOSE_FLAG) when the
// user picks "Boot SchOS (verbose)" from the F8 boot menu; read once here,
// before anything else, so subsystem init can log its own steps.
int boot_verbose;

// Must match SCHBOOT_VERBOSE_FLAG in boot/bios/stage2main.c exactly: just past
// the VBE info block, not 0x7000 (startothers() below uses that address for
// the AP boot trampoline).
#include "schboot.h"

// Bootstrap processor starts running C code here.
// Allocate a real stack and switch to it, first
// doing some setup required for memory allocator to work.
int
kmain(uint32 fb_addr)
{
  boot_verbose = *(uchar*)P2V(SCHBOOT_VERBOSE_FLAG);

  uartearlyinit();
  if (boot_verbose) cprintf("verbose boot: uartearlyinit done\n");
  // The DMA arena is carved off the top of physical memory *before* kalloc is
  // given the rest, because a device needs physically contiguous memory and
  // kalloc's free list makes no such promise -- two pages from it are adjacent
  // only by chance. Taking it first is the only way to be sure the region
  // exists; asking for it later would mean asking kalloc for something it
  // fundamentally cannot provide. See include/dma.h.
  // Before anything touches PHYSTOP, which is now a variable read from the
  // firmware's memory map rather than a constant asserting 224 MB exists.
  e820_init();
  kinit1(end, P2V(PHYSTOP - DMA_ARENA_BYTES));
  if (boot_verbose) cprintf("verbose boot: kinit1 done\n");
  dma_init(P2V(PHYSTOP - DMA_ARENA_BYTES), P2V(PHYSTOP));
  if (boot_verbose) cprintf("verbose boot: dma_init done\n");
  kvmalloc();
  if (boot_verbose) cprintf("verbose boot: kvmalloc done\n");
  // ACPI first, the legacy MP table only if it finds nothing. Modern firmware
  // publishes a complete MADT and an MP table with one processor in it, so
  // asking the older interface first means never seeing more than one CPU.
  // See kernel/arch/x86_64/acpi.c.
  if (!acpi_init())
    mpinit();
  if (boot_verbose) cprintf("verbose boot: cpu enumeration done\n");
  lapicinit();
  if (boot_verbose) cprintf("verbose boot: lapicinit done\n");
  tvinit();
  if (boot_verbose) cprintf("verbose boot: tvinit done\n");
  seginit();
  if (boot_verbose) cprintf("verbose boot: seginit done\n");
  ioapicinit();
  if (boot_verbose) cprintf("verbose boot: ioapicinit done\n");
  irq_init();       // must precede any driver init that registers a handler
  if (boot_verbose) cprintf("verbose boot: irq_init done\n");
  // Before the keyboard and mouse, which start producing events the moment
  // their handlers are registered.
  inputinit();
  if (boot_verbose) cprintf("verbose boot: inputinit done\n");
  consoleinit();
  if (boot_verbose) cprintf("verbose boot: consoleinit done\n");
  uartinit();
  if (boot_verbose) cprintf("verbose boot: uartinit done\n");
  mouseinit();
  if (boot_verbose) cprintf("verbose boot: mouseinit done\n");
  vga_init(fb_addr);
  if (boot_verbose) cprintf("verbose boot: vga_init done (fb=0x%x)\n", fb_addr);
  // No splash drawing here on purpose. The loader owns the splash, logo and
  // all, and vga_init() deliberately does not clear -- so it is still on the
  // display right through to console_takeover(). Painting it again from here
  // would land in the last few hundred milliseconds of boot and change nothing.
  // After vga_init, because the display driver cross-checks the framebuffer
  // address it finds on the bus against the one VBE reported -- and it can
  // only do that once vga.c has read the VBE block.
  // The devices that are simply known to be on a PC rather than enumerated by
  // any bus. Registered before pci_init() so that /proc/devices lists them
  // first, which puts the machine's fixed furniture above whatever happened to
  // be plugged into it.
  platform_register_devices();
  pci_init();
  if (boot_verbose) cprintf("verbose boot: pci_init done\n");
  pinit();
  if (boot_verbose) cprintf("verbose boot: pinit done\n");
  binit();
  if (boot_verbose) cprintf("verbose boot: binit done\n");
  fileinit();
  if (boot_verbose) cprintf("verbose boot: fileinit done\n");
  sockinit();
  if (boot_verbose) cprintf("verbose boot: sockinit done\n");
  // After the timer is running, so boot_epoch can subtract the uptime so far,
  // and before userspace, so the very first timestamp any program takes is
  // already real. Reading the RTC is slow, which is exactly why it happens once
  // here rather than on every gettimeofday().
  time_init();
  ideinit();
  if (boot_verbose) cprintf("verbose boot: ideinit done\n");
  ramdiskinit();
  if (boot_verbose) cprintf("verbose boot: ramdiskinit done\n");
  // Start the application processors.
  //
  // This was broken from the beginning of the project and is fixed as of
  // 2026-08-08. Five distinct bugs, each hidden behind the one before it, and
  // all of them invisible under QEMU because it publishes only one processor in
  // the legacy MP table -- see kernel/arch/x86_64/acpi.c. VMware published the
  // real count, which is why the triple fault only ever appeared there.
  //
  // 1. entry64high had no AP branch at all; the check was a comment saying it
  //    had never been written. Every AP fell into the boot processor's path,
  //    took the BSP's stack, and ran kmain() a second time.
  // 2. seginit() called kalloc() for its per-CPU page. kalloc locks, locking
  //    reads `cpu`, and `cpu` is what seginit() was about to set up.
  // 3. kernel_fs_base was one global for all CPUs, and trapasm.S read it on
  //    every entry from user mode, so both CPUs shared one per-CPU block.
  // 4. `cpu` and `proc` were __thread. That means per *thread*, so the compiler
  //    cached the thread pointer across calls -- but %fs here is per *CPU*, and
  //    a kernel thread changes CPU at every sched(). See include/proc.h.
  // 5. The new per-CPU slots collided with syscall_entry's scratch word at
  //    %fs:(8), which is the same memory as %gs:8.
  //
  int smp_start_aps = 1;
  if (smp_start_aps) {
    startothers();
    if (boot_verbose) cprintf("verbose boot: startothers done\n");
  } else if (ncpu > 1) {
    cprintf("smp: %d cpus found, running on 1 (AP startup is broken)\n", ncpu);
  }
  kinit2();
  if (boot_verbose) cprintf("verbose boot: kinit2 done\n");
  sinit();
  if (boot_verbose) cprintf("verbose boot: sinit done\n");
  // The TCP/IP stack, after sinit because it starts a kernel thread and
  // kthread_create needs the process table to be usable. It returns
  // immediately: everything it does happens in that thread, which cannot run
  // until mpmain() below enters the scheduler.
  netinit();
  if (boot_verbose) cprintf("verbose boot: netinit done\n");
  mpmain();
}

// Other CPUs jump here from entryother.S.
void
mpenter(void)
{
  switchkvm();
  seginit();
  lapicinit();
  mpmain();
}

// Enable SSE/FXSAVE so user binaries can use XMM registers.
static void
enable_sse(void)
{
  uint64 cr0, cr4;
  asm volatile("mov %%cr0, %0" : "=r"(cr0));
  cr0 &= ~(1UL << 2);  // clear CR0.EM (no FP emulation)
  cr0 |=  (1UL << 1);  // set   CR0.MP (monitor coprocessor)
  asm volatile("mov %0, %%cr0" :: "r"(cr0));
  asm volatile("mov %%cr4, %0" : "=r"(cr4));
  cr4 |= (1UL << 9);   // CR4.OSFXSR    — OS handles FXSAVE/FXRSTOR
  cr4 |= (1UL << 10);  // CR4.OSXMMEXCPT — OS handles SSE exceptions
  asm volatile("mov %0, %%cr4" :: "r"(cr4));
}

// Common CPU setup code.
static void
mpmain(void)
{
  cprintf("cpu%d: starting\n", cpunum());
  enable_sse();    // allow user XMM/SSE instructions
  idtinit();       // load idt register
  syscallinit();   // syscall set up
  xchg(&mycpu()->started, 1); // tell startothers() we're up
  scheduler();     // start running processes
}

void entry32mp(void);

// Start the non-boot (AP) processors.
static void
startothers(void)
{
  extern uchar _binary_entryother_start[], _binary_entryother_size[];
  uchar *code;
  struct cpu *c;
  char *stack;

  // Write entry code to unused memory at 0x7000.
  // The linker has placed the image of entryother.S in
  // _binary_entryother_start.
  code = P2V(0x7000);
  memmove(code, _binary_entryother_start,
          (addr_t)_binary_entryother_size);

  for(c = cpus; c < cpus+ncpu; c++){
    if(c == cpus+cpunum())  // We've started already.
      continue;

    // Tell entryother.S what stack to use, where to enter, and what
    // pgdir to use. We cannot use kpgdir yet, because the AP processor
    // is running in low  memory, so we use entrypgdir for the APs too.
    // The AP's per-CPU page, allocated here rather than by the AP itself.
    // seginit() explains why: an AP cannot call kalloc() before %fs is set up,
    // because kalloc locks and locking reads the per-CPU pointer.
    if (c->local == 0) {
      c->local = kalloc();
      memset(c->local, 0, PGSIZE);
    }

    stack = kalloc();
    *(uint32*)(code-4) = 0x8000; // enough stack to get us to entry64mp
    *(uint32*)(code-8) = v2p(entry32mp);
    *(uint64*)(code-16) = (uint64) (stack + KSTACKSIZE);

    lapicstartap(c->apicid, V2P(code));

    // wait for cpu to finish mpmain()
    while(c->started == 0)
      ;

    // Counted here, on the boot processor, after the AP has actually reported
    // in -- not by the AP itself. This is serialised by the loop, so it needs
    // no atomic, and it cannot overcount a processor that started and died.
    ncpu_online++;
  }
}