multi-cpu scheduling WIP

init/init.c

@@ -47,6 +47,8 @@ void app_main (void) {
 
   spawn (&app_thread1);
 
+    /* for (volatile int i = 0; i < 1000*1000; i++) */
+    /*   ; */
   for (;;) {
     lock_mutex (MUTEX, RV_PRIVATE);
 
@@ -54,6 +56,9 @@ void app_main (void) {
       test ('a');
 
     unlock_mutex (MUTEX, RV_PRIVATE);
+    
+    /* for (volatile int i = 0; i < 1000*1000; i++) */
+    /*   ; */
   }
 }
 
@@ -65,6 +70,9 @@ void app_thread1 (void) {
       test ('b');
 
     unlock_mutex (MUTEX, RV_PRIVATE);
+
+    /* for (volatile int i = 0; i < 1000*1000; i++) */
+    /*   ; */
   }
 
   quit ();

init/src.mk

@@ -1,3 +1,3 @@
-S += init.S
+c += init.c
 
 o += init.o

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) {
+void do_sched (struct proc* proc, spin_lock_t* cpu_lock, spin_lock_ctx_t* ctxcpu) {
-  __asm__ volatile ("cli");
+  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++) {
+struct cpu* cpu_find_lightest (void) {
-    if (cpu->id == i) {
+  struct cpu* cpu = &cpus[0];
-      amd64_lapic_ipi (mp->cpus[i]->lapic_id, CPU_REQUEST_SCHED);
+
-      break;
+  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
  

kernel/irq/irq.c

@@ -30,11 +30,6 @@ bool irq_attach (void (*func) (void*, void*), void* arg, uint32_t irq_num) {
   irq_table[irq_num] = irq;
   rw_spin_write_unlock (&irqs_lock, &ctxiqa);
 
-#if defined(__x86_64__)
-  uint8_t resolution = amd64_resolve_irq (irq_num);
-  amd64_ioapic_route_irq (irq_num, resolution, 0, amd64_lapic_id ());
-#endif
-
   return true;
 }
 

kernel/proc/mutex.c

@@ -6,6 +6,8 @@
 #include <proc/proc.h>
 #include <sync/spin_lock.h>
 #include <sys/debug.h>
+#include <sys/smp.h>
+#include <sys/spin_lock.h>
 
 bool proc_create_resource_mutex (struct proc_mutex* mutex) {
   memset (mutex, 0, sizeof (*mutex));
@@ -34,6 +36,8 @@ static void proc_mutex_suspend (struct proc* proc, struct proc_suspension_q* sq,
   proc->suspension_q = sq;
 
   list_remove (cpu->proc_run_q, &proc->cpu_run_q_link);
+  atomic_fetch_sub (&cpu->proc_run_q_count, 1);
+
   if (cpu->proc_current == proc)
     cpu->proc_current = NULL;
 
@@ -64,6 +68,7 @@ static void proc_mutex_resume (struct proc* proc) {
     atomic_store (&proc->state, PROC_READY);
 
     list_append (cpu->proc_run_q, &proc->cpu_run_q_link);
+    atomic_fetch_add (&cpu->proc_run_q_count, 1);
 
     spin_unlock (&sq->lock, &ctxsq);
   }

kernel/proc/proc.c

@@ -181,7 +181,7 @@ struct elf_aux proc_load_segments (struct proc* proc, uint8_t* elf) {
   return aux;
 }
 
-static struct proc* proc_spawn_rd (char* name) {
+struct proc* proc_spawn_rd (char* name) {
   struct rd_file* rd_file = rd_get_file (name);
 
   bool ok = proc_check_elf (rd_file->content);
@@ -204,10 +204,13 @@ struct proc* proc_find_pid (int pid) {
   return proc;
 }
 
-void proc_register (struct proc* proc, struct cpu* cpu) {
+void proc_register (struct proc* proc, struct cpu* cpu1) {
   spin_lock_ctx_t ctxcpu, ctxprtr;
 
-  proc->cpu = cpu;
+  proc->cpu = cpu1 != NULL ? cpu1 : cpu_find_lightest ();
+  DEBUG ("Assigning CPU %d to PID %d\n", proc->cpu->id, proc->pid);
+
+  struct cpu* cpu = proc->cpu;
 
   rw_spin_write_lock (&proc_tree_lock, &ctxprtr);
   rbtree_insert (struct proc, &proc_tree, &proc->proc_tree_link, proc_tree_link, pid);
@@ -215,6 +218,7 @@ void proc_register (struct proc* proc, struct cpu* cpu) {
 
   spin_lock (&cpu->lock, &ctxcpu);
   list_append (cpu->proc_run_q, &proc->cpu_run_q_link);
+  atomic_fetch_add (&cpu->proc_run_q_count, 1);
 
   if (cpu->proc_current == NULL)
     cpu->proc_current = proc;
@@ -306,15 +310,15 @@ void proc_sched (void) {
   next = proc_find_sched (cpu);
 
   if (prev != NULL) {
+    spin_lock (&prev->lock, &ctxpr);
     memcpy (&prev->pdata.regs, &cpu->regs, sizeof (struct saved_regs));
     spin_unlock (&prev->lock, &ctxpr);
   }
 
   if (next) {
     cpu->proc_current = next;
-    spin_unlock (&cpu->lock, &ctxcpu);
 
-    do_sched (next);
+    do_sched (next, &cpu->lock, &ctxcpu);
   } else {
     cpu->proc_current = NULL;
     spin_unlock (&cpu->lock, &ctxcpu);
@@ -329,11 +333,13 @@ void proc_kill (struct proc* proc) {
 
   spin_lock (&proc->lock, &ctxpr);
   atomic_store (&proc->state, PROC_DEAD);
+  proc->cpu = NULL;
   spin_unlock (&proc->lock, &ctxpr);
 
   spin_lock (&cpu->lock, &ctxcpu);
 
   list_remove (cpu->proc_run_q, &proc->cpu_run_q_link);
+  atomic_fetch_sub (&cpu->proc_run_q_count, 1);
   if (cpu->proc_current == proc)
     cpu->proc_current = NULL;
 
@@ -409,8 +415,13 @@ void proc_init (void) {
 
   proc_kpproc_init ();
 
-  struct proc* init = proc_spawn_rd ("init.exe");
+  struct proc* spin_proc = proc_spawn_rd ("spin.exe");
-  proc_register (init, thiscpu);
+  proc_register (spin_proc, thiscpu);
 
-  do_sched (init);
+  struct proc* init = proc_spawn_rd ("init.exe");
+  proc_register (init, NULL);
+
+  spin_lock_ctx_t ctxcpu;
+  spin_lock (&init->cpu->lock, &ctxcpu);
+  do_sched (init, &init->cpu->lock, &ctxcpu);
 }

kernel/proc/proc.h

@@ -68,6 +68,7 @@ bool proc_unmap (struct proc* proc, uintptr_t start_vaddr, size_t pages);
 struct elf_aux proc_load_segments (struct proc* proc, uint8_t* elf);
 void proc_register (struct proc* proc, struct cpu* cpu);
 struct proc* proc_find_pid (int pid);
+struct proc* proc_spawn_rd (char* name);
 void proc_init (void);
 
 #endif // _KERNEL_PROC_PROC_H

kernel/sys/sched.h

@@ -4,6 +4,6 @@
 #include <libk/std.h>
 #include <proc/proc.h>
 
-void do_sched (struct proc* proc);
+void do_sched (struct proc* proc, spin_lock_t* cpu_lock, spin_lock_ctx_t* ctxcpu);
 
 #endif // _KERNEL_SYS_SCHED_H

kernel/syscall/syscall.c

@@ -160,8 +160,6 @@ DEFINE_SYSCALL (sys_clone) {
   size_t stack_size = (size_t)a2;
   uintptr_t entry = a3;
 
-  struct cpu* cpu = proc->cpu;
-
   struct proc* new = proc_clone (proc, vstack_top, stack_size, entry);
 
   DEBUG ("new=%p\n", new);
@@ -172,7 +170,7 @@ DEFINE_SYSCALL (sys_clone) {
 
   int pid = new->pid;
 
-  proc_register (new, cpu);
+  proc_register (new, NULL);
 
   return pid;
 }
@@ -234,7 +232,6 @@ DEFINE_SYSCALL (sys_lock_mutex) {
   if (mutex_resource == NULL)
     return -ST_NOT_FOUND;
 
-  DEBUG ("locking %d\n", proc->pid);
   proc_mutex_lock (proc, &mutex_resource->u.mutex);
 
   return ST_OK;
@@ -253,7 +250,6 @@ DEFINE_SYSCALL (sys_unlock_mutex) {
   if (mutex_resource == NULL)
     return -ST_NOT_FOUND;
 
-  DEBUG ("unlocking %d\n", proc->pid);
   return proc_mutex_unlock (proc, &mutex_resource->u.mutex) ? ST_OK : -ST_PERMISSION_ERROR;
 }
 

make/apps.mk

@@ -1,4 +1,4 @@
-apps := init
+apps := init spin
 
 all_apps:
 	@for d in $(apps); do make -C $$d platform=$(platform) all; done

spin/.gitignore

@@ -0,0 +1,2 @@
+*.o
+*.exe

spin/Makefile

@@ -0,0 +1 @@
+include ../make/user.mk

spin/app.mk

@@ -0,0 +1 @@
+app := spin.exe

spin/spin.c

@@ -0,0 +1,4 @@
+void app_main (void) {
+  for (;;)
+    ;
+}

spin/src.mk

@@ -0,0 +1,3 @@
+c += spin.c
+
+o += spin.o