Use RW spin locks

kernel/amd64/apic.c

@@ -4,7 +4,7 @@
 #include <amd64/msr.h>
 #include <libk/std.h>
 #include <limine/requests.h>
-#include <sync/spin_lock.h>
+#include <sync/rw_spin_lock.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
 #include <sys/spin.h>
@@ -36,7 +36,7 @@
 
 struct ioapic {
   struct acpi_madt_ioapic table_data;
-  spin_lock_t lock;
+  rw_spin_lock_t lock;
   uintptr_t mmio_base;
 };
 
@@ -55,19 +55,19 @@ static uint64_t lapic_ticks;
 
 /* Read IOAPIC */
 static uint32_t amd64_ioapic_read (struct ioapic* ioapic, uint32_t reg) {
-  spin_lock (&ioapic->lock);
+  rw_spin_read_lock (&ioapic->lock);
   *(volatile uint32_t*)ioapic->mmio_base = reg;
   uint32_t ret = *(volatile uint32_t*)(ioapic->mmio_base + 0x10);
-  spin_unlock (&ioapic->lock);
+  rw_spin_read_unlock (&ioapic->lock);
   return ret;
 }
 
 /* Write IOAPIC */
 static void amd64_ioapic_write (struct ioapic* ioapic, uint32_t reg, uint32_t value) {
-  spin_lock (&ioapic->lock);
+  rw_spin_write_lock (&ioapic->lock);
   *(volatile uint32_t*)ioapic->mmio_base = reg;
   *(volatile uint32_t*)(ioapic->mmio_base + 0x10) = value;
-  spin_unlock (&ioapic->lock);
+  rw_spin_write_unlock (&ioapic->lock);
 }
 
 /* Find an IOAPIC corresposting to provided IRQ */
@@ -160,7 +160,7 @@ void amd64_ioapic_init (void) {
                           (uintptr_t)hhdm->offset + (uintptr_t)ioapic_table_data->address,
                           MM_PG_PRESENT | MM_PG_RW | MM_PD_RELOAD);
       ioapics[ioapic_entries++] = (struct ioapic){
-          .lock = SPIN_LOCK_INIT,
+          .lock = RW_SPIN_LOCK_INIT,
           .table_data = *ioapic_table_data,
           .mmio_base = ((uintptr_t)hhdm->offset + (uintptr_t)ioapic_table_data->address),
       };

kernel/amd64/smp.c

@@ -8,7 +8,7 @@
 #include <libk/string.h>
 #include <limine/requests.h>
 #include <mm/liballoc.h>
-#include <sync/spin_lock.h>
+#include <sync/rw_spin_lock.h>
 #include <sys/debug.h>
 #include <sys/smp.h>
 #include <sys/syscall.h>
@@ -27,7 +27,7 @@ struct cpu* cpu_make (void) {
   struct cpu* cpu = &cpus[id];
 
   memset (cpu, 0, sizeof (*cpu));
-  cpu->lock = SPIN_LOCK_INIT;
+  cpu->lock = RW_SPIN_LOCK_INIT;
   cpu->id = id;
   cpu->self = cpu;
 

kernel/amd64/smp.h

@@ -7,6 +7,7 @@
 #include <libk/rbtree.h>
 #include <libk/std.h>
 #include <proc/proc.h>
+#include <sync/rw_spin_lock.h>
 
 #define CPUS_MAX 32
 
@@ -29,7 +30,7 @@ struct cpu {
     atomic_int nesting;
   } irq_ctx;
 
-  spin_lock_t lock;
+  rw_spin_lock_t lock;
 
   struct rb_node_link* proc_run_q;
   struct proc* proc_current;

kernel/irq/irq.c

@@ -2,7 +2,7 @@
 #include <libk/list.h>
 #include <libk/std.h>
 #include <mm/liballoc.h>
-#include <sync/spin_lock.h>
+#include <sync/rw_spin_lock.h>
 
 #if defined(__x86_64__)
 #include <amd64/apic.h>
@@ -12,7 +12,7 @@
 /* TODO: figure out a generic way to work with IRQs */
 
 static struct list_node_link* irqs = NULL;
-static spin_lock_t irqs_lock;
+static rw_spin_lock_t irqs_lock;
 
 bool irq_attach (void (*func) (void*, void*), void* arg, uint32_t irq_num, uint32_t flags) {
   struct irq* irq = malloc (sizeof (*irq));
@@ -25,9 +25,9 @@ bool irq_attach (void (*func) (void*, void*), void* arg, uint32_t irq_num, uint3
   irq->irq_num = irq_num;
   irq->flags = flags;
 
-  spin_lock (&irqs_lock);
+  rw_spin_write_lock (&irqs_lock);
   list_append (irqs, &irq->irqs_link);
-  spin_unlock (&irqs_lock);
+  rw_spin_write_unlock (&irqs_lock);
 
 #if defined(__x86_64__)
   uint8_t resolution = amd64_resolve_irq (irq_num);
@@ -39,32 +39,40 @@ bool irq_attach (void (*func) (void*, void*), void* arg, uint32_t irq_num, uint3
 
 void irq_detach (void (*func) (void*, void*)) {
   struct list_node_link *irq_link, *irq_link_tmp;
-  spin_lock (&irqs_lock);
+  struct irq* irq = NULL;
+
+  rw_spin_read_lock (&irqs_lock);
 
   list_foreach (irqs, irq_link, irq_link_tmp) {
-    struct irq* irq = list_entry (irq_link, struct irq, irqs_link);
+    irq = list_entry (irq_link, struct irq, irqs_link);
 
     if ((uintptr_t)irq->func == (uintptr_t)func)
-      list_remove (irqs, irq_link);
+      break;
   }
 
-  spin_unlock (&irqs_lock);
+  rw_spin_read_unlock (&irqs_lock);
+
+  if (irq != NULL) {
+    rw_spin_write_lock (&irqs_lock);
+    list_remove (irqs, &irq->irqs_link);
+    rw_spin_write_unlock (&irqs_lock);
+  }
 }
 
 struct irq* irq_find (uint32_t irq_num) {
   struct list_node_link *irq_link, *irq_link_tmp;
-  spin_lock (&irqs_lock);
+  rw_spin_read_lock (&irqs_lock);
 
   list_foreach (irqs, irq_link, irq_link_tmp) {
     struct irq* irq = list_entry (irq_link, struct irq, irqs_link);
 
     if (irq->irq_num == irq_num) {
-      spin_unlock (&irqs_lock);
+      rw_spin_read_unlock (&irqs_lock);
       return irq;
     }
   }
 
-  spin_unlock (&irqs_lock);
+  rw_spin_read_unlock (&irqs_lock);
 
   return NULL;
 }

kernel/proc/proc.c

@@ -12,6 +12,7 @@
 #include <proc/proc.h>
 #include <proc/resource.h>
 #include <rd/rd.h>
+#include <sync/rw_spin_lock.h>
 #include <sync/spin_lock.h>
 #include <sys/debug.h>
 #include <sys/mm.h>
@@ -25,13 +26,15 @@
 #endif
 
 /*
- * Lock ordering:
+ * Lock hierachy:
- * 1. proc_tree_lock
+ * - proc_tree_lock
- * 2. [cpu]->lock
+ *   - cpu->lock
+ *     - proc->lock
+ *       - suspension_q->lock
  */
 
 static struct rb_node_link* proc_tree = NULL;
-static spin_lock_t proc_tree_lock = SPIN_LOCK_INIT;
+static rw_spin_lock_t proc_tree_lock = RW_SPIN_LOCK_INIT;
 
 static bool proc_check_elf (uint8_t* elf) {
   if (!((elf[0] == 0x7F) && (elf[1] == 'E') && (elf[2] == 'L') && (elf[3] == 'F')))
@@ -191,8 +194,8 @@ static struct proc* proc_spawn_rd (char* name) {
 static void proc_register (struct proc* proc, struct cpu* cpu) {
   proc->cpu = cpu;
 
-  spin_lock (&proc_tree_lock);
+  rw_spin_write_lock (&proc_tree_lock);
-  spin_lock (&cpu->lock);
+  rw_spin_write_lock (&cpu->lock);
 
   rbtree_insert (struct proc, &cpu->proc_run_q, &proc->cpu_run_q_link, cpu_run_q_link, pid);
   rbtree_insert (struct proc, &proc_tree, &proc->proc_tree_link, proc_tree_link, pid);
@@ -200,8 +203,8 @@ static void proc_register (struct proc* proc, struct cpu* cpu) {
   if (cpu->proc_current == NULL)
     cpu->proc_current = proc;
 
-  spin_unlock (&cpu->lock);
+  rw_spin_write_unlock (&cpu->lock);
-  spin_unlock (&proc_tree_lock);
+  rw_spin_write_unlock (&proc_tree_lock);
 }
 
 static struct proc* proc_find_sched (void) {
@@ -241,19 +244,22 @@ static struct proc* proc_find_sched (void) {
 void proc_sched (void) {
   struct proc* next = NULL;
 
-  spin_lock (&thiscpu->lock);
+  rw_spin_read_lock (&thiscpu->lock);
 
   if (thiscpu->proc_run_q == NULL) {
-    spin_unlock (&thiscpu->lock);
+    rw_spin_read_unlock (&thiscpu->lock);
     goto idle;
   }
 
   next = proc_find_sched ();
 
-  if (next != NULL)
+  rw_spin_read_unlock (&thiscpu->lock);
-    thiscpu->proc_current = next;
 
-  spin_unlock (&thiscpu->lock);
+  if (next != NULL) {
+    rw_spin_write_lock (&thiscpu->lock);
+    thiscpu->proc_current = next;
+    rw_spin_write_unlock (&thiscpu->lock);
+  }
 
   if (next != NULL && atomic_load (&next->state) == PROC_READY)
     do_sched (next);
@@ -265,14 +271,16 @@ idle:
 void proc_kill (struct proc* proc) {
   atomic_store (&proc->state, PROC_DEAD);
 
-  spin_lock (&proc_tree_lock);
+  rw_spin_write_lock (&proc_tree_lock);
+  spin_lock (&proc->lock);
   rbtree_delete (&proc_tree, &proc->proc_tree_link);
-  spin_unlock (&proc_tree_lock);
+  spin_unlock (&proc->lock);
+  rw_spin_write_unlock (&proc_tree_lock);
 
   struct cpu* cpu = proc->cpu;
-  spin_lock (&cpu->lock);
+  rw_spin_write_lock (&cpu->lock);
   rbtree_delete (&cpu->proc_run_q, &proc->cpu_run_q_link);
-  spin_unlock (&cpu->lock);
+  rw_spin_write_unlock (&cpu->lock);
 
   DEBUG ("killed PID %d\n", proc->pid);
 
@@ -284,6 +292,57 @@ void proc_kill (struct proc* proc) {
     cpu_request_sched (cpu);
 }
 
+void proc_suspend (struct proc* proc, struct proc_suspension_q* sq) {
+  struct cpu* cpu;
+
+  spin_lock (&proc->lock);
+
+  atomic_store (&proc->state, PROC_SUSPENDED);
+
+  cpu = proc->cpu;
+
+  /* remove from run q */
+  rw_spin_write_lock (&cpu->lock);
+  rbtree_delete (&cpu->proc_run_q, &proc->cpu_run_q_link);
+  if (cpu->proc_current == proc)
+    cpu->proc_current = NULL;
+  rw_spin_write_unlock (&cpu->lock);
+
+  proc->suspension_q = sq;
+  spin_lock (&proc->suspension_q->lock);
+  rbtree_insert (struct proc, &proc->suspension_q->proc_tree, &proc->suspension_link,
+                 suspension_link, pid);
+  spin_unlock (&proc->suspension_q->lock);
+
+  spin_unlock (&proc->lock);
+
+  cpu_request_sched (cpu);
+}
+
+void proc_wakeup (struct proc* proc) {
+  struct cpu* cpu;
+
+  spin_lock (&proc->lock);
+
+  cpu = proc->cpu;
+
+  spin_lock (&proc->suspension_q->lock);
+  rbtree_delete (&proc->suspension_q->proc_tree, &proc->suspension_link);
+  spin_unlock (&proc->suspension_q->lock);
+
+  proc->suspension_q = NULL;
+
+  rw_spin_write_lock (&cpu->lock);
+  rbtree_insert (struct proc, &cpu->proc_run_q, &proc->cpu_run_q_link, cpu_run_q_link, pid);
+  rw_spin_write_unlock (&cpu->lock);
+
+  atomic_store (&proc->state, PROC_READY);
+
+  spin_unlock (&proc->lock);
+
+  cpu_request_sched (cpu);
+}
+
 static void proc_irq_sched (void* arg, void* regs) {
   (void)arg, (void)regs;
   proc_sched ();

kernel/proc/proc.h

@@ -15,10 +15,12 @@
 #include <amd64/proc.h> /* USTACK_SIZE */
 #endif
 
-/// Process is ready to run
+/* Process is ready to run */
 #define PROC_READY 0
-/// Process marked garbage collection
+/* Process marked garbage collection */
 #define PROC_DEAD 1
+/* Process is suspended */
+#define PROC_SUSPENDED 2
 
 #define PROC_RESOURCES_MAX 1024
 
@@ -36,6 +38,7 @@ struct proc {
   int pid;
   struct rb_node_link proc_tree_link;
   struct rb_node_link cpu_run_q_link;
+  struct rb_node_link suspension_link;
 
   struct list_node_link* mappings; /* pd.lock implicitly protects this field */
   struct proc_platformdata pdata;
@@ -45,8 +48,15 @@ struct proc {
   atomic_int state;
   struct rb_node_link* resource_tree;
   atomic_int rids;
+  struct proc_suspension_q* suspension_q;
 };
 
+struct proc_suspension_q {
+  struct rb_node_link* proc_tree;
+  spin_lock_t lock;
+};
+
+void proc_suspend (struct proc* proc, struct proc_suspension_q* sq);
 void proc_sched (void);
 void proc_kill (struct proc* proc);
 bool proc_map (struct proc* proc, uintptr_t start_paddr, uintptr_t start_vaddr, size_t pages,

kernel/sync/rw_spin_lock.c

@@ -0,0 +1,59 @@
+#include <libk/assert.h>
+#include <libk/std.h>
+#include <sync/rw_spin_lock.h>
+#include <sys/debug.h>
+#include <sys/spin_lock.h>
+
+#define WRITER_WAIT (1U << 31)
+#define READER_MASK (~WRITER_WAIT)
+
+void rw_spin_read_lock (rw_spin_lock_t* rw) {
+  uint32_t value;
+
+  for (;;) {
+    value = atomic_load_explicit (rw, memory_order_relaxed);
+
+    if ((value & WRITER_WAIT) == 0) {
+      if (atomic_compare_exchange_weak_explicit (rw, &value, value + 1, memory_order_acquire,
+                                                 memory_order_relaxed)) {
+        return;
+      }
+    }
+
+    spin_lock_relax ();
+  }
+}
+
+void rw_spin_read_unlock (rw_spin_lock_t* rw) {
+  uint32_t old = atomic_fetch_sub_explicit (rw, 1, memory_order_release);
+  assert ((old & READER_MASK) > 0);
+}
+
+void rw_spin_write_lock (rw_spin_lock_t* rw) {
+  uint32_t value;
+
+  /* announce writer */
+  for (;;) {
+    value = atomic_load_explicit (rw, memory_order_relaxed);
+
+    if ((value & WRITER_WAIT) == 0) {
+      if (atomic_compare_exchange_weak_explicit (rw, &value, (value | WRITER_WAIT),
+                                                 memory_order_acquire, memory_order_relaxed))
+        break;
+    } else
+      spin_lock_relax ();
+  }
+
+  /* wait for readers */
+  for (;;) {
+    value = atomic_load_explicit (rw, memory_order_acquire);
+    if ((value & READER_MASK) == 0)
+      return;
+
+    spin_lock_relax ();
+  }
+}
+
+void rw_spin_write_unlock (rw_spin_lock_t* rw) {
+  atomic_store_explicit (rw, 0, memory_order_release);
+}

kernel/sync/rw_spin_lock.h

@@ -0,0 +1,16 @@
+#ifndef _KERNEL_SYNC_RW_SPIN_LOCK_H
+#define _KERNEL_SYNC_RW_SPIN_LOCK_H
+
+#include <libk/std.h>
+#include <sync/spin_lock.h>
+
+#define RW_SPIN_LOCK_INIT 0
+
+typedef _Atomic (uint32_t) rw_spin_lock_t;
+
+void rw_spin_read_lock (rw_spin_lock_t* rw);
+void rw_spin_read_unlock (rw_spin_lock_t* rw);
+void rw_spin_write_lock (rw_spin_lock_t* rw);
+void rw_spin_write_unlock (rw_spin_lock_t* rw);
+
+#endif // _KERNEL_SYNC_RW_SPIN_LOCK_H

kernel/sync/src.mk

@@ -1,3 +1,5 @@
-c += sync/spin_lock.c
+c += sync/spin_lock.c \
+		 sync/rw_spin_lock.c
 
-o += sync/spin_lock.o
+o += sync/spin_lock.o \
+		 sync/rw_spin_lock.o