multi-cpu scheduling WIP

kernel/amd64/apic.c

@@ -82,8 +82,6 @@ static struct ioapic* amd64_ioapic_find (uint32_t irq) {
 
   for (size_t i = 0; i < ioapic_entries; i++) {
     ioapic = &ioapics[i];
-    /* uint32_t version = amd64_ioapic_read ((uintptr_t)hhdm->offset +
-     * (uintptr_t)ioapic->table_data.address, 1); */
     uint32_t version = amd64_ioapic_read (ioapic, 1);
     uint32_t max = ((version >> 16) & 0xFF);
 
@@ -233,7 +231,7 @@ static uint32_t amd64_lapic_calibrate (uint32_t us) {
 static void amd64_lapic_start (uint32_t ticks) {
   amd64_lapic_write (LAPIC_DCR, DIVIDER_VALUE);
   amd64_lapic_write (LAPIC_TIMICT, ticks);
-  amd64_lapic_write (LAPIC_LVTTR, SCHED_PREEMPT_TIMER | (1 << 17) | (1 << 16));
+  amd64_lapic_write (LAPIC_LVTTR, SCHED_PREEMPT_TIMER | (1 << 17));
 }
 
 /*

kernel/amd64/bootmain.c

@@ -9,6 +9,7 @@
 #include <irq/irq.h>
 #include <libk/std.h>
 #include <limine/limine.h>
+#include <limine/requests.h>
 #include <mm/liballoc.h>
 #include <mm/pmm.h>
 #include <proc/proc.h>
@@ -29,7 +30,9 @@ ALIGNED (16) static uint8_t uacpi_memory_buffer[UACPI_MEMORY_BUFFER_MAX];
  * the necessary platform-dependent subsystems/drivers and jump into the init app.
  */
 void bootmain (void) {
-  struct cpu* bsp_cpu = cpu_make ();
+  struct limine_mp_response* mp = limine_mp_request.response;
+
+  struct cpu* bsp_cpu = cpu_make (mp->bsp_lapic_id);
 
   amd64_init (bsp_cpu, false);
   syscall_init ();
@@ -44,9 +47,9 @@ void bootmain (void) {
   amd64_ioapic_init ();
   amd64_hpet_init ();
 
-  smp_init ();
-
   mm_init2 ();
+  
+  smp_init ();
 
   proc_init ();
 

kernel/amd64/intr.c

@@ -213,15 +213,3 @@ void irq_save (spin_lock_ctx_t* ctx) { *ctx = amd64_irq_save_flags (); }
 
 /* Restore interrupt state */
 void irq_restore (spin_lock_ctx_t* ctx) { amd64_irq_restore_flags (*ctx); }
-
-/* Map custom IRQ mappings to legacy IRQs */
-uint32_t amd64_resolve_irq (uint32_t irq) {
-  static const uint32_t mappings[] = {
-      [SCHED_PREEMPT_TIMER] = 0,
-      [TLB_SHOOTDOWN] = 6,
-      [CPU_REQUEST_SCHED] = 3,
-      [CPU_SPURIOUS] = 5,
-  };
-
-  return mappings[irq];
-}

kernel/amd64/intr.h

@@ -32,7 +32,6 @@ struct saved_regs {
 } PACKED;
 
 void amd64_load_idt (void);
-uint32_t amd64_resolve_irq (uint32_t irq);
 void amd64_intr_init (void);
 
 #endif // _KERNEL_AMD64_INTR_H

kernel/amd64/sched1.c

@@ -5,12 +5,18 @@
 #include <proc/proc.h>
 #include <sys/mm.h>
 #include <sys/smp.h>
+#include <sync/spin_lock.h>
 
-void do_sched (struct proc* proc) {
-  __asm__ volatile ("cli");
+void do_sched (struct proc* proc, spin_lock_t* cpu_lock, spin_lock_ctx_t* ctxcpu) {
+  spin_lock_ctx_t ctxpr;
+
+  spin_lock (&proc->lock, &ctxpr);
 
   thiscpu->tss.rsp0 = proc->pdata.kernel_stack;
   thiscpu->syscall_kernel_stack = proc->pdata.kernel_stack;
+  
+  spin_unlock (&proc->lock, &ctxpr);
+  spin_unlock (cpu_lock, ctxcpu);
 
   amd64_do_sched ((void*)&proc->pdata.regs, (void*)proc->pd->cr3_paddr);
 }

kernel/amd64/smp.c

@@ -13,6 +13,7 @@
 #include <sys/debug.h>
 #include <sys/smp.h>
 #include <sys/syscall.h>
+#include <sys/sched.h>
 
 /// Cpu ID counter
 static atomic_uint cpu_counter = 0;
@@ -22,7 +23,7 @@ static struct cpu cpus[CPUS_MAX];
 static atomic_int cpu_init_count;
 
 /// Allocate a CPU structure
-struct cpu* cpu_make (void) {
+struct cpu* cpu_make (uint64_t lapic_id) {
   int id = atomic_fetch_add (&cpu_counter, 1);
 
   struct cpu* cpu = &cpus[id];
@@ -30,6 +31,7 @@ struct cpu* cpu_make (void) {
   memset (cpu, 0, sizeof (*cpu));
   cpu->lock = SPIN_LOCK_INIT;
   cpu->id = id;
+  cpu->lapic_id = lapic_id;
 
   amd64_wrmsr (MSR_GS_BASE, (uint64_t)cpu);
 
@@ -47,21 +49,31 @@ void cpu_request_sched (struct cpu* cpu) {
     return;
   }
 
-  struct limine_mp_response* mp = limine_mp_request.response;
+  amd64_lapic_ipi (cpu->lapic_id, CPU_REQUEST_SCHED);
+}
 
-  for (size_t i = 0; i < mp->cpu_count; i++) {
-    if (cpu->id == i) {
-      amd64_lapic_ipi (mp->cpus[i]->lapic_id, CPU_REQUEST_SCHED);
-      break;
+struct cpu* cpu_find_lightest (void) {
+  struct cpu* cpu = &cpus[0];
+
+  int load = atomic_load (&cpu->proc_run_q_count);
+
+  for (unsigned int i = 1; i < cpu_counter; i++) {
+    struct cpu* new_cpu = &cpus[i];
+    int new_load = atomic_load (&new_cpu->proc_run_q_count);
+    if (new_load < load) {
+      load = new_load;
+      cpu = new_cpu;
     }
   }
+
+  return cpu;
 }
 
 /// Bootstrap code for non-BSP CPUs
 static void amd64_smp_bootstrap (struct limine_mp_info* mp_info) {
   amd64_load_kernel_cr3 ();
 
-  struct cpu* cpu = cpu_make ();
+  struct cpu* cpu = cpu_make (mp_info->lapic_id);
 
   amd64_init (cpu, true); /* gdt + idt */
   syscall_init ();
@@ -70,12 +82,14 @@ static void amd64_smp_bootstrap (struct limine_mp_info* mp_info) {
 
   DEBUG ("CPU %u is online!\n", thiscpu->id);
 
-  __asm__ volatile ("sti");
-
   atomic_fetch_sub (&cpu_init_count, 1);
+  
+  struct proc* spin_proc = proc_spawn_rd ("spin.exe");
+  proc_register (spin_proc, thiscpu);
 
-  for (;;)
-    ;
+  spin_lock_ctx_t ctxcpu;
+  spin_lock (&spin_proc->cpu->lock, &ctxcpu);
+  do_sched (spin_proc, &spin_proc->cpu->lock, &ctxcpu);
 }
 
 /// Initialize SMP subsystem for AMD64. Start AP CPUs
@@ -87,7 +101,7 @@ void smp_init (void) {
   cpu_init_count = mp->cpu_count - 1; /* Don't include BSP */
 
   for (size_t i = 0; i < mp->cpu_count; i++) {
-    if (mp->cpus[i]->lapic_id != thiscpu->id) {
+    if (mp->cpus[i]->lapic_id != thiscpu->lapic_id) {
       DEBUG ("Trying CPU %u\n", mp->cpus[i]->lapic_id);
       mp->cpus[i]->goto_address = &amd64_smp_bootstrap;
     }

kernel/amd64/smp.h

@@ -25,17 +25,20 @@ struct cpu {
 
   uintptr_t lapic_mmio_base;
   uint64_t lapic_ticks;
+  uint64_t lapic_id;
   uint32_t id;
 
   spin_lock_t lock;
 
   struct list_node_link* proc_run_q;
   struct proc* proc_current;
+  atomic_int proc_run_q_count;
 };
 
-struct cpu* cpu_make (void);
+struct cpu* cpu_make (uint64_t lapic_id);
 struct cpu* cpu_get (void);
 void cpu_request_sched (struct cpu* cpu);
+struct cpu* cpu_find_lightest (void);
 
 #define thiscpu (cpu_get ())
 

kernel/amd64/syscall.c

@@ -15,25 +15,27 @@ extern void amd64_syscall_entry (void);
 
 int amd64_syscall_dispatch (void* stack_ptr) {
   spin_lock_ctx_t ctxcpu;
+  
+  spin_lock (&thiscpu->lock, &ctxcpu);
 
   amd64_load_kernel_cr3 ();
-
   struct saved_regs* regs = stack_ptr;
 
-  spin_lock (&thiscpu->lock, &ctxcpu);
   memcpy (&thiscpu->regs, regs, sizeof (struct saved_regs));
-  spin_unlock (&thiscpu->lock, &ctxcpu);
 
   int syscall_num = regs->rax;
   syscall_handler_func_t func = syscall_find_handler (syscall_num);
 
-  if (func == NULL)
+  if (func == NULL) {
+    spin_unlock (&thiscpu->lock, &ctxcpu);
     return -ST_SYSCALL_NOT_FOUND;
-
+  }
+  
   struct proc* caller = thiscpu->proc_current;
+  
+  spin_unlock (&thiscpu->lock, &ctxcpu);
 
   int result = func (caller, regs, regs->rdi, regs->rsi, regs->rdx, regs->r10, regs->r8, regs->r9);
-
   return result;
 }
 

kernel/amd64/syscallentry.S

@@ -4,6 +4,8 @@
 
 .global amd64_syscall_entry
 amd64_syscall_entry:
+  cli
+
   movq %rsp, %gs:0
   movq %gs:8, %rsp