From 54d636d25cda432fd411ce3d5d73a74c8a082506 Mon Sep 17 00:00:00 2001
From: Stefan Lankes <lankes@lfbs.rwth-aachen.de>
Date: Thu, 24 Feb 2011 18:32:58 +0100
Subject: [PATCH] after process termination, the PGD and its PGTs will be freed

---
 arch/x86/include/asm/page.h |  5 +++++
 arch/x86/mm/page.c          | 25 +++++++++++++++++++++++++
 include/metalsvm/mmu.h      |  2 ++
 include/metalsvm/tasks.h    |  5 ++++-
 kernel/syscall.c            |  4 +++-
 kernel/tasks.c              | 25 +++++++++++++++++--------
 kernel/tests.c              |  2 +-
 mm/memory.c                 | 21 ++++++++++++++++++++-
 8 files changed, 77 insertions(+), 12 deletions(-)

diff --git a/arch/x86/include/asm/page.h b/arch/x86/include/asm/page.h
index bf7ba801..9b00f891 100644
--- a/arch/x86/include/asm/page.h
+++ b/arch/x86/include/asm/page.h
@@ -110,6 +110,11 @@ int get_kernel_pgd(task_t* task);
  */
 int get_user_pgd(task_t* task);
 
+/*
+ * Delete page directory and its page tables
+ */
+int drop_pgd(void);
+
 /*
  * Change the page permission in the page tables of the current task
  */
diff --git a/arch/x86/mm/page.c b/arch/x86/mm/page.c
index c274ecbb..64d3af48 100644
--- a/arch/x86/mm/page.c
+++ b/arch/x86/mm/page.c
@@ -116,6 +116,31 @@ int get_user_pgd(task_t* task)
 	return 0;
 }
 
+int drop_pgd(void)
+{
+	uint32_t i;
+	page_dir_t* pgd = per_core(current_task)->pgd;
+	size_t phy_pgd = virt_to_phys((size_t) pgd);
+
+	if (BUILTIN_EXPECT(pgd == &boot_pgd, 0))
+		return -EINVAL;
+
+	spinlock_lock(per_core(current_task)->pgd_lock);
+	
+	for(i=KERNEL_SPACE/(1024*PAGE_SIZE); i<1024; i++)
+		if (pgd->entries[i] & 0xFFFFF000)
+			put_page(pgd->entries[i] & 0xFFFFF000);
+
+	// freeing the page directory
+	put_page(phy_pgd);
+
+	per_core(current_task)->pgd = NULL;
+
+	spinlock_unlock(per_core(current_task)->pgd_lock);
+
+	return 0;
+}
+
 size_t virt_to_phys(size_t viraddr)
 {
 	task_t* task = per_core(current_task);
diff --git a/include/metalsvm/mmu.h b/include/metalsvm/mmu.h
index a0cf78b5..96220b3d 100644
--- a/include/metalsvm/mmu.h
+++ b/include/metalsvm/mmu.h
@@ -34,6 +34,8 @@ extern atomic_int32_t		total_available_pages;
 
 int mmu_init(void);
 size_t get_pages(uint32_t no_pages);
+static inline size_t get_page(void) { return get_pages(1); }
+int put_page(size_t phyaddr);
 
 #ifdef __cplusplus
 }
diff --git a/include/metalsvm/tasks.h b/include/metalsvm/tasks.h
index 1aae59d4..19e13e6b 100644
--- a/include/metalsvm/tasks.h
+++ b/include/metalsvm/tasks.h
@@ -64,9 +64,12 @@ void NORETURN leave_kernel_task(void);
 /* by leaving user level task, this function will be called */
 void NORETURN leave_user_task(void);
 
-/* system call to terminate a user level task */
+/* system call to terminate a user level process */
 void NORETURN sys_exit(int);
 
+/* system call to create a new child process */
+int sys_fork(void);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/kernel/syscall.c b/kernel/syscall.c
index aadca322..d79eae6f 100644
--- a/kernel/syscall.c
+++ b/kernel/syscall.c
@@ -68,10 +68,12 @@ int syscall_handler(uint32_t sys_nr, ...)
 		ret = per_core(current_task)->id;
 		break;
 	case __NR_fork:
+		ret = sys_fork();
+		break;
 	case __NR_fstat:
 	default:
 		kputs("invalid system call\n");
-		ret = -EINVAL;
+		ret = -ENOSYS;
 		break;
 	};
 
diff --git a/kernel/tasks.c b/kernel/tasks.c
index 2e8cbe6f..7410c4d7 100644
--- a/kernel/tasks.c
+++ b/kernel/tasks.c
@@ -38,8 +38,7 @@ static spinlock_t table_lock = SPINLOCK_INIT;
 /*
  * helper function for the assembly code to determine the current task
  */
-task_t* get_current_task(void)
-{
+task_t* get_current_task(void) {
 	return per_core(current_task);
 }
 
@@ -83,6 +82,8 @@ static void NORETURN do_exit(int arg) {
 
 	kprintf("Terminate task: %u, return value %d\n", per_core(current_task)->id, arg);
 
+	wakeup_blocked_tasks(arg);
+
 	//vma_dump(per_core(current_task));
 	spinlock_lock(&(per_core(current_task)->vma_lock));
 
@@ -95,7 +96,8 @@ static void NORETURN do_exit(int arg) {
 
 	spinlock_unlock(&(per_core(current_task)->vma_lock));
 
-	wakeup_blocked_tasks(arg);
+	drop_pgd(); // delete page directory and its page tables
+	
 	if (atomic_int32_read(&per_core(current_task)->mem_usage))
 		kprintf("Memory leak! Task %d did not release %d pages\n", per_core(current_task)->id, atomic_int32_read(&per_core(current_task)->mem_usage));
 	per_core(current_task)->status = TASK_FINISHED;
@@ -114,8 +116,7 @@ void NORETURN leave_kernel_task(void) {
         do_exit(result);
 }
 
-void NORETURN sys_exit(int arg) 
-{
+void NORETURN sys_exit(int arg) {
 	do_exit(arg);
 }
 
@@ -135,10 +136,14 @@ static int create_task(tid_t* id, entry_point_t ep, void* arg, int user)
 
 	for(i=0; i<MAX_TASKS; i++) {
 		if (task_table[i].status == TASK_INVALID) {
-			if (!user) 
+			if (!user) {
 				ret = get_kernel_pgd(task_table+i);
-			else
+				atomic_int32_set(&task_table[i].mem_usage, 0);
+			} else {
 				ret = get_user_pgd(task_table+i);
+				// user-level pgd needs already a page
+				atomic_int32_set(&task_table[i].mem_usage, 1);
+			}
 			
 			if (ret != 0) {
 				ret = -ENOMEM;
@@ -147,7 +152,6 @@ static int create_task(tid_t* id, entry_point_t ep, void* arg, int user)
 
 			task_table[i].id = i;
 			task_table[i].status = TASK_READY;
-			atomic_int32_set(&task_table[i].mem_usage, 0);
 			spinlock_init(&task_table[i].vma_lock);
 			task_table[i].vma_list = NULL;	
 			memset(task_table[i].mbox, 0x00, sizeof(mailbox_int32_t*)*MAX_TASKS);
@@ -313,6 +317,11 @@ int create_user_task(tid_t* id, size_t sz, const char* fname, int argc, char** a
 	return create_task(id, user_entry, node, 1);
 }
 
+int sys_fork(void)
+{
+	return -EINVAL;
+}
+
 int join_task(tid_t id, int* result) 
 {
 	int32_t tmp;
diff --git a/kernel/tests.c b/kernel/tests.c
index 26140de1..48322438 100644
--- a/kernel/tests.c
+++ b/kernel/tests.c
@@ -104,7 +104,7 @@ int test_init(void)
 	sem_init(&consuming, 0);
 	mailbox_int32_init(&mbox);
 
-	create_kernel_task(NULL, foo, "Hello from foo1\n");
+	//create_kernel_task(NULL, foo, "Hello from foo1\n");
 	//create_kernel_task(NULL, join_test, NULL);
 	//create_kernel_task(NULL, producer, NULL);
 	//create_kernel_task(NULL, consumer, NULL);
diff --git a/mm/memory.c b/mm/memory.c
index efe616c6..d5233ad9 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -25,6 +25,7 @@
 #include <metalsvm/time.h>
 #include <metalsvm/processor.h>
 #include <metalsvm/page.h>
+#include <metalsvm/errno.h>
 #ifdef CONFIG_MULTIBOOT
 #include <asm/multiboot.h>
 #endif
@@ -235,6 +236,25 @@ oom:
 	return ret;
 }
 
+int put_page(size_t phyaddr)
+{
+	uint32_t index = phyaddr / PAGE_SIZE;
+	task_t* task = per_core(current_task);
+
+	if (BUILTIN_EXPECT(!phyaddr, 0))
+		return -EINVAL;
+
+	spinlock_lock(&bitmap_lock);
+	page_clear_mark(index);
+	spinlock_unlock(&bitmap_lock);
+
+	atomic_int32_sub(&total_allocated_pages, 1);
+	atomic_int32_add(&total_available_pages, 1);
+	atomic_int32_sub(&(task->mem_usage), 1);
+
+	return 0;
+}
+
 void* mem_allocation(size_t sz, uint32_t flags)
 {
 	size_t phyaddr, viraddr;
@@ -273,7 +293,6 @@ void kfree(void* addr, size_t sz)
 	if (sz % PAGE_SIZE)
 		npages++;
 
-
 	spinlock_lock(&bitmap_lock);
 	for(i=0; i<npages; i++) {
 		phyaddr = virt_to_phys((size_t) addr+i*PAGE_SIZE);