int
kswapd_main(void *arg) {
int guard = 0;
while (1) {
if (pressure > 0) {
int needs = (pressure << 5), rounds = 16;
list_entry_t *list = &proc_mm_list;
assert(!list_empty(list));
while (needs > 0 && rounds -- > 0) {
list_entry_t *le = list_next(list);
list_del(le);
list_add_before(list, le);
struct mm_struct *mm = le2mm(le, proc_mm_link);
needs -= swap_out_mm(mm, (needs < 32) ? needs : 32);
}
}
pressure -= page_launder();
refill_inactive_scan();
if (pressure > 0) {
if ((++ guard) >= 1000) {
guard = 0;
warn("kswapd: may out of memory");
}
continue ;
}
pressure = 0, guard = 0;
kswapd_wakeup_all();
do_sleep(1000);
}
}
static int swap_out(unsigned int priority, unsigned int gfp_mask, zone_t * classzone)
{
#ifndef NO_MM
int counter, nr_pages = SWAP_CLUSTER_MAX;
struct mm_struct *mm;
/* Then, look at the other mm's */
counter = mmlist_nr;
do {
if (unlikely(current->need_resched)) {
__set_current_state(TASK_RUNNING);
schedule();
}
spin_lock(&mmlist_lock);
mm = swap_mm;
while (mm->swap_address == TASK_SIZE || mm == &init_mm) {
mm->swap_address = 0;
mm = list_entry(mm->mmlist.next, struct mm_struct, mmlist);
if (mm == swap_mm)
goto empty;
swap_mm = mm;
}
/* Make sure the mm doesn't disappear when we drop the lock.. */
atomic_inc(&mm->mm_users);
spin_unlock(&mmlist_lock);
nr_pages = swap_out_mm(mm, nr_pages, &counter, classzone);
mmput(mm);
if (!nr_pages)
return 1;
} while (--counter >= 0);
return 0;
empty:
spin_unlock(&mmlist_lock);
#endif /* NO_MM */
return 0;
}
// check_swap - check the correctness of swap & page replacement algorithm
static void
check_swap(void) {
size_t nr_used_pages_store = nr_used_pages();
size_t slab_allocated_store = slab_allocated();
size_t offset;
for (offset = 2; offset < max_swap_offset; offset ++) {
mem_map[offset] = 1;
}
struct mm_struct *mm = mm_create();
assert(mm != NULL);
extern struct mm_struct *check_mm_struct;
assert(check_mm_struct == NULL);
check_mm_struct = mm;
pgd_t *pgdir = mm->pgdir = init_pgdir_get();
assert(pgdir[PGX(TEST_PAGE)] == 0);
struct vma_struct *vma = vma_create(TEST_PAGE, TEST_PAGE + PTSIZE, VM_WRITE | VM_READ);
assert(vma != NULL);
insert_vma_struct(mm, vma);
struct Page *rp0 = alloc_page(), *rp1 = alloc_page();
assert(rp0 != NULL && rp1 != NULL);
pte_perm_t perm;
ptep_unmap (&perm);
ptep_set_u_write(&perm);
int ret = page_insert(pgdir, rp1, TEST_PAGE, perm);
assert(ret == 0 && page_ref(rp1) == 1);
page_ref_inc(rp1);
ret = page_insert(pgdir, rp0, TEST_PAGE, perm);
assert(ret == 0 && page_ref(rp1) == 1 && page_ref(rp0) == 1);
// check try_alloc_swap_entry
swap_entry_t entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1);
mem_map[1] = 1;
assert(try_alloc_swap_entry() == 0);
// set rp1, Swap, Active, add to hash_list, active_list
swap_page_add(rp1, entry);
swap_active_list_add(rp1);
assert(PageSwap(rp1));
mem_map[1] = 0;
entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1);
assert(!PageSwap(rp1));
// check swap_remove_entry
assert(swap_hash_find(entry) == NULL);
mem_map[1] = 2;
swap_remove_entry(entry);
assert(mem_map[1] == 1);
swap_page_add(rp1, entry);
swap_inactive_list_add(rp1);
swap_remove_entry(entry);
assert(PageSwap(rp1));
assert(rp1->index == entry && mem_map[1] == 0);
// check page_launder, move page from inactive_list to active_list
assert(page_ref(rp1) == 1);
assert(nr_active_pages == 0 && nr_inactive_pages == 1);
assert(list_next(&(inactive_list.swap_list)) == &(rp1->swap_link));
page_launder();
assert(nr_active_pages == 1 && nr_inactive_pages == 0);
assert(PageSwap(rp1) && PageActive(rp1));
entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1);
assert(!PageSwap(rp1) && nr_active_pages == 0);
assert(list_empty(&(active_list.swap_list)));
// set rp1 inactive again
assert(page_ref(rp1) == 1);
swap_page_add(rp1, 0);
assert(PageSwap(rp1) && swap_offset(rp1->index) == 1);
swap_inactive_list_add(rp1);
mem_map[1] = 1;
assert(nr_inactive_pages == 1);
page_ref_dec(rp1);
size_t count = nr_used_pages();
swap_remove_entry(entry);
assert(nr_inactive_pages == 0 && nr_used_pages() == count - 1);
// check swap_out_mm
pte_t *ptep0 = get_pte(pgdir, TEST_PAGE, 0), *ptep1;
assert(ptep0 != NULL && pte2page(*ptep0) == rp0);
ret = swap_out_mm(mm, 0);
assert(ret == 0);
ret = swap_out_mm(mm, 10);
assert(ret == 1 && mm->swap_address == TEST_PAGE + PGSIZE);
ret = swap_out_mm(mm, 10);
assert(ret == 0 && *ptep0 == entry && mem_map[1] == 1);
assert(PageDirty(rp0) && PageActive(rp0) && page_ref(rp0) == 0);
assert(nr_active_pages == 1 && list_next(&(active_list.swap_list)) == &(rp0->swap_link));
// check refill_inactive_scan()
refill_inactive_scan();
assert(!PageActive(rp0) && page_ref(rp0) == 0);
assert(nr_inactive_pages == 1 && list_next(&(inactive_list.swap_list)) == &(rp0->swap_link));
page_ref_inc(rp0);
page_launder();
assert(PageActive(rp0) && page_ref(rp0) == 1);
assert(nr_active_pages == 1 && list_next(&(active_list.swap_list)) == &(rp0->swap_link));
page_ref_dec(rp0);
refill_inactive_scan();
assert(!PageActive(rp0));
// save data in rp0
int i;
for (i = 0; i < PGSIZE; i ++) {
((char *)page2kva(rp0))[i] = (char)i;
}
page_launder();
assert(nr_inactive_pages == 0 && list_empty(&(inactive_list.swap_list)));
assert(mem_map[1] == 1);
rp1 = alloc_page();
assert(rp1 != NULL);
ret = swapfs_read(entry, rp1);
assert(ret == 0);
for (i = 0; i < PGSIZE; i ++) {
assert(((char *)page2kva(rp1))[i] == (char)i);
}
// page fault now
*(char *)(TEST_PAGE) = 0xEF;
rp0 = pte2page(*ptep0);
assert(page_ref(rp0) == 1);
assert(PageSwap(rp0) && PageActive(rp0));
entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1 && mem_map[1] == SWAP_UNUSED);
assert(!PageSwap(rp0) && nr_active_pages == 0 && nr_inactive_pages == 0);
// clear accessed flag
assert(rp0 == pte2page(*ptep0));
assert(!PageSwap(rp0));
ret = swap_out_mm(mm, 10);
assert(ret == 0);
assert(!PageSwap(rp0) && ptep_present(ptep0));
// change page table
ret = swap_out_mm(mm, 10);
assert(ret == 1);
assert(*ptep0 == entry && page_ref(rp0) == 0 && mem_map[1] == 1);
count = nr_used_pages();
refill_inactive_scan();
page_launder();
assert(count - 1 == nr_used_pages());
ret = swapfs_read(entry, rp1);
assert(ret == 0 && *(char *)(page2kva(rp1)) == (char)0xEF);
free_page(rp1);
// duplictate *ptep0
ptep1 = get_pte(pgdir, TEST_PAGE + PGSIZE, 0);
assert(ptep1 != NULL && ptep_invalid(ptep1));
swap_duplicate(*ptep0);
ptep_copy(ptep1, ptep0);
mp_tlb_invalidate (pgdir, TEST_PAGE + PGSIZE);
// page fault again
// update for copy on write
*(char *)(TEST_PAGE + 1) = 0x88;
*(char *)(TEST_PAGE + PGSIZE) = 0x8F;
*(char *)(TEST_PAGE + PGSIZE + 1) = 0xFF;
assert(pte2page(*ptep0) != pte2page(*ptep1));
assert(*(char *)(TEST_PAGE) == (char)0xEF);
assert(*(char *)(TEST_PAGE + 1) == (char)0x88);
assert(*(char *)(TEST_PAGE + PGSIZE) == (char)0x8F);
assert(*(char *)(TEST_PAGE + PGSIZE + 1) == (char)0xFF);
rp0 = pte2page(*ptep0);
rp1 = pte2page(*ptep1);
assert(!PageSwap(rp0) && PageSwap(rp1) && PageActive(rp1));
entry = try_alloc_swap_entry();
assert(!PageSwap(rp0) && !PageSwap(rp1));
assert(swap_offset(entry) == 1 && mem_map[1] == SWAP_UNUSED);
assert(list_empty(&(active_list.swap_list)));
assert(list_empty(&(inactive_list.swap_list)));
ptep_set_accessed(&perm);
page_insert(pgdir, rp0, TEST_PAGE + PGSIZE, perm);
// check swap_out_mm
*(char *)(TEST_PAGE) = *(char *)(TEST_PAGE + PGSIZE) = 0xEE;
mm->swap_address = TEST_PAGE + PGSIZE * 2;
ret = swap_out_mm(mm, 2);
assert(ret == 0);
assert(ptep_present(ptep0) && ! ptep_accessed(ptep0));
assert(ptep_present(ptep1) && ! ptep_accessed(ptep1));
ret = swap_out_mm(mm, 2);
assert(ret == 2);
assert(mem_map[1] == 2 && page_ref(rp0) == 0);
refill_inactive_scan();
page_launder();
assert(mem_map[1] == 2 && swap_hash_find(entry) == NULL);
// check copy entry
swap_remove_entry(entry);
ptep_unmap(ptep1);
assert(mem_map[1] == 1);
swap_entry_t store;
ret = swap_copy_entry(entry, &store);
assert(ret == -E_NO_MEM);
mem_map[2] = SWAP_UNUSED;
ret = swap_copy_entry(entry, &store);
assert(ret == 0 && swap_offset(store) == 2 && mem_map[2] == 0);
mem_map[2] = 1;
ptep_copy(ptep1, &store);
assert(*(char *)(TEST_PAGE + PGSIZE) == (char)0xEE && *(char *)(TEST_PAGE + PGSIZE + 1)== (char)0x88);
*(char *)(TEST_PAGE + PGSIZE) = 1, *(char *)(TEST_PAGE + PGSIZE + 1) = 2;
assert(*(char *)TEST_PAGE == (char)0xEE && *(char *)(TEST_PAGE + 1) == (char)0x88);
ret = swap_in_page(entry, &rp0);
assert(ret == 0);
ret = swap_in_page(store, &rp1);
assert(ret == 0);
assert(rp1 != rp0);
// free memory
swap_list_del(rp0), swap_list_del(rp1);
swap_page_del(rp0), swap_page_del(rp1);
assert(page_ref(rp0) == 1 && page_ref(rp1) == 1);
assert(nr_active_pages == 0 && list_empty(&(active_list.swap_list)));
assert(nr_inactive_pages == 0 && list_empty(&(inactive_list.swap_list)));
for (i = 0; i < HASH_LIST_SIZE; i ++) {
assert(list_empty(hash_list + i));
}
page_remove(pgdir, TEST_PAGE);
page_remove(pgdir, (TEST_PAGE + PGSIZE));
#if PMXSHIFT != PUXSHIFT
free_page(pa2page(PMD_ADDR(*get_pmd(pgdir, TEST_PAGE, 0))));
#endif
#if PUXSHIFT != PGXSHIFT
free_page(pa2page(PUD_ADDR(*get_pud(pgdir, TEST_PAGE, 0))));
#endif
free_page(pa2page(PGD_ADDR(*get_pgd(pgdir, TEST_PAGE, 0))));
pgdir[PGX(TEST_PAGE)] = 0;
mm->pgdir = NULL;
mm_destroy(mm);
check_mm_struct = NULL;
assert(nr_active_pages == 0 && nr_inactive_pages == 0);
for (offset = 0; offset < max_swap_offset; offset ++) {
mem_map[offset] = SWAP_UNUSED;
}
assert(nr_used_pages_store == nr_used_pages());
assert(slab_allocated_store == slab_allocated());
kprintf("check_swap() succeeded.\n");
}
static void
check_mm_swap(void) {
size_t nr_free_pages_store = nr_free_pages();
size_t slab_allocated_store = slab_allocated();
int ret, i, j;
for (i = 0; i < max_swap_offset; i ++) {
assert(mem_map[i] == SWAP_UNUSED);
}
extern struct mm_struct *check_mm_struct;
assert(check_mm_struct == NULL);
// step1: check mm_map
struct mm_struct *mm0 = mm_create(), *mm1;
assert(mm0 != NULL && list_empty(&(mm0->mmap_list)));
uintptr_t addr0, addr1;
addr0 = 0;
do {
ret = mm_map(mm0, addr0, PTSIZE, 0, NULL);
assert(ret == (USER_ACCESS(addr0, addr0 + PTSIZE)) ? 0 : -E_INVAL);
addr0 += PTSIZE;
} while (addr0 != 0);
addr0 = 0;
for (i = 0; i < 1024; i ++, addr0 += PTSIZE) {
ret = mm_map(mm0, addr0, PGSIZE, 0, NULL);
assert(ret == -E_INVAL);
}
mm_destroy(mm0);
mm0 = mm_create();
assert(mm0 != NULL && list_empty(&(mm0->mmap_list)));
addr0 = 0, i = 0;
do {
ret = mm_map(mm0, addr0, PTSIZE - PGSIZE, 0, NULL);
assert(ret == (USER_ACCESS(addr0, addr0 + PTSIZE)) ? 0 : -E_INVAL);
if (ret == 0) {
i ++;
}
addr0 += PTSIZE;
} while (addr0 != 0);
addr0 = 0, j = 0;
do {
addr0 += PTSIZE - PGSIZE;
ret = mm_map(mm0, addr0, PGSIZE, 0, NULL);
assert(ret == (USER_ACCESS(addr0, addr0 + PGSIZE)) ? 0 : -E_INVAL);
if (ret == 0) {
j ++;
}
addr0 += PGSIZE;
} while (addr0 != 0);
assert(j + 1 >= i);
mm_destroy(mm0);
assert(nr_free_pages_store == nr_free_pages());
assert(slab_allocated_store == slab_allocated());
cprintf("check_mm_swap: step1, mm_map ok.\n");
// step2: check page fault
mm0 = mm_create();
assert(mm0 != NULL && list_empty(&(mm0->mmap_list)));
// setup page table
struct Page *page = alloc_page();
assert(page != NULL);
pde_t *pgdir = page2kva(page);
memcpy(pgdir, boot_pgdir, PGSIZE);
pgdir[PDX(VPT)] = PADDR(pgdir) | PTE_P | PTE_W;
// prepare for page fault
mm0->pgdir = pgdir;
check_mm_struct = mm0;
lcr3(PADDR(mm0->pgdir));
uint32_t vm_flags = VM_WRITE | VM_READ;
struct vma_struct *vma;
addr0 = 0;
do {
if ((ret = mm_map(mm0, addr0, PTSIZE, vm_flags, &vma)) == 0) {
break;
}
addr0 += PTSIZE;
} while (addr0 != 0);
assert(ret == 0 && addr0 != 0 && mm0->map_count == 1);
assert(vma->vm_start == addr0 && vma->vm_end == addr0 + PTSIZE);
// check pte entry
pte_t *ptep;
for (addr1 = addr0; addr1 < addr0 + PTSIZE; addr1 += PGSIZE) {
ptep = get_pte(pgdir, addr1, 0);
assert(ptep == NULL);
}
memset((void *)addr0, 0xEF, PGSIZE * 2);
ptep = get_pte(pgdir, addr0, 0);
assert(ptep != NULL && (*ptep & PTE_P));
ptep = get_pte(pgdir, addr0 + PGSIZE, 0);
assert(ptep != NULL && (*ptep & PTE_P));
ret = mm_unmap(mm0, - PTSIZE, PTSIZE);
assert(ret == -E_INVAL);
ret = mm_unmap(mm0, addr0 + PTSIZE, PGSIZE);
assert(ret == 0);
addr1 = addr0 + PTSIZE / 2;
ret = mm_unmap(mm0, addr1, PGSIZE);
assert(ret == 0 && mm0->map_count == 2);
ret = mm_unmap(mm0, addr1 + 2 * PGSIZE, PGSIZE * 4);
assert(ret == 0 && mm0->map_count == 3);
ret = mm_map(mm0, addr1, PGSIZE * 6, 0, NULL);
assert(ret == -E_INVAL);
ret = mm_map(mm0, addr1, PGSIZE, 0, NULL);
assert(ret == 0 && mm0->map_count == 4);
ret = mm_map(mm0, addr1 + 2 * PGSIZE, PGSIZE * 4, 0, NULL);
assert(ret == 0 && mm0->map_count == 5);
ret = mm_unmap(mm0, addr1 + PGSIZE / 2, PTSIZE / 2 - PGSIZE);
assert(ret == 0 && mm0->map_count == 1);
addr1 = addr0 + PGSIZE;
for (i = 0; i < PGSIZE; i ++) {
assert(*(char *)(addr1 + i) == (char)0xEF);
}
ret = mm_unmap(mm0, addr1 + PGSIZE / 2, PGSIZE / 4);
assert(ret == 0 && mm0->map_count == 2);
ptep = get_pte(pgdir, addr0, 0);
assert(ptep != NULL && (*ptep & PTE_P));
ptep = get_pte(pgdir, addr0 + PGSIZE, 0);
assert(ptep != NULL && *ptep == 0);
ret = mm_map(mm0, addr1, PGSIZE, vm_flags, NULL);
memset((void *)addr1, 0x88, PGSIZE);
assert(*(char *)addr1 == (char)0x88 && mm0->map_count == 3);
for (i = 1; i < 16; i += 2) {
ret = mm_unmap(mm0, addr0 + PGSIZE * i, PGSIZE);
assert(ret == 0);
if (i < 8) {
ret = mm_map(mm0, addr0 + PGSIZE * i, PGSIZE, 0, NULL);
assert(ret == 0);
}
}
assert(mm0->map_count == 13);
ret = mm_unmap(mm0, addr0 + PGSIZE * 2, PTSIZE - PGSIZE * 2);
assert(ret == 0 && mm0->map_count == 2);
ret = mm_unmap(mm0, addr0, PGSIZE * 2);
assert(ret == 0 && mm0->map_count == 0);
cprintf("check_mm_swap: step2, mm_unmap ok.\n");
// step3: check exit_mmap
ret = mm_map(mm0, addr0, PTSIZE, vm_flags, NULL);
assert(ret == 0);
for (i = 0, addr1 = addr0; i < 4; i ++, addr1 += PGSIZE) {
*(char *)addr1 = (char)0xFF;
}
exit_mmap(mm0);
for (i = 0; i < PDX(KERNBASE); i ++) {
assert(pgdir[i] == 0);
}
cprintf("check_mm_swap: step3, exit_mmap ok.\n");
// step4: check dup_mmap
for (i = 0; i < max_swap_offset; i ++) {
assert(mem_map[i] == SWAP_UNUSED);
}
ret = mm_map(mm0, addr0, PTSIZE, vm_flags, NULL);
assert(ret != 0);
addr1 = addr0;
for (i = 0; i < 4; i ++, addr1 += PGSIZE) {
*(char *)addr1 = (char)(i * i);
}
ret = 0;
ret += swap_out_mm(mm0, 10);
ret += swap_out_mm(mm0, 10);
assert(ret == 4);
for (; i < 8; i ++, addr1 += PGSIZE) {
*(char *)addr1 = (char)(i * i);
}
// setup mm1
mm1 = mm_create();
assert(mm1 != NULL);
page = alloc_page();
assert(page != NULL);
pgdir = page2kva(page);
memcpy(pgdir, boot_pgdir, PGSIZE);
pgdir[PDX(VPT)] = PADDR(pgdir) | PTE_P | PTE_W;
mm1->pgdir = pgdir;
ret = dup_mmap(mm1, mm0);
assert(ret == 0);
// switch to mm1
check_mm_struct = mm1;
lcr3(PADDR(mm1->pgdir));
addr1 = addr0;
for (i = 0; i < 8; i ++, addr1 += PGSIZE) {
assert(*(char *)addr1 == (char)(i * i));
*(char *)addr1 = (char)0x88;
}
// switch to mm0
check_mm_struct = mm0;
lcr3(PADDR(mm0->pgdir));
addr1 = addr0;
for (i = 0; i < 8; i ++, addr1 += PGSIZE) {
assert(*(char *)addr1 == (char)(i * i));
}
// switch to boot_cr3
check_mm_struct = NULL;
lcr3(boot_cr3);
// free memory
exit_mmap(mm0);
exit_mmap(mm1);
free_page(kva2page(mm0->pgdir));
mm_destroy(mm0);
free_page(kva2page(mm1->pgdir));
mm_destroy(mm1);
cprintf("check_mm_swap: step4, dup_mmap ok.\n");
refill_inactive_scan();
page_launder();
for (i = 0; i < max_swap_offset; i ++) {
assert(mem_map[i] == SWAP_UNUSED);
}
assert(nr_free_pages_store == nr_free_pages());
assert(slab_allocated_store == slab_allocated());
cprintf("check_mm_swap() succeeded.\n");
}
// check_swap - check the correctness of swap & page replacement algorithm
static void
check_swap(void) {
size_t nr_free_pages_store = nr_free_pages();
size_t slab_allocated_store = slab_allocated();
size_t offset;
for (offset = 2; offset < max_swap_offset; offset ++) {
mem_map[offset] = 1;
}
struct mm_struct *mm = mm_create();
assert(mm != NULL);
extern struct mm_struct *check_mm_struct;
assert(check_mm_struct == NULL);
check_mm_struct = mm;
pde_t *pgdir = mm->pgdir = boot_pgdir;
assert(pgdir[0] == 0);
struct vma_struct *vma = vma_create(0, PTSIZE, VM_WRITE | VM_READ);
assert(vma != NULL);
insert_vma_struct(mm, vma);
struct Page *rp0 = alloc_page(), *rp1 = alloc_page();
assert(rp0 != NULL && rp1 != NULL);
uint32_t perm = PTE_U | PTE_W;
int ret = page_insert(pgdir, rp1, 0, perm);
assert(ret == 0 && page_ref(rp1) == 1);
page_ref_inc(rp1);
ret = page_insert(pgdir, rp0, 0, perm);
assert(ret == 0 && page_ref(rp1) == 1 && page_ref(rp0) == 1);
// check try_alloc_swap_entry
swap_entry_t entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1);
mem_map[1] = 1;
assert(try_alloc_swap_entry() == 0);
// set rp1, Swap, Active, add to hash_list, active_list
swap_page_add(rp1, entry);
swap_active_list_add(rp1);
assert(PageSwap(rp1));
mem_map[1] = 0;
entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1);
assert(!PageSwap(rp1));
// check swap_remove_entry
assert(swap_hash_find(entry) == NULL);
mem_map[1] = 2;
swap_remove_entry(entry);
assert(mem_map[1] == 1);
swap_page_add(rp1, entry);
swap_inactive_list_add(rp1);
swap_remove_entry(entry);
assert(PageSwap(rp1));
assert(rp1->index == entry && mem_map[1] == 0);
// check page_launder, move page from inactive_list to active_list
assert(page_ref(rp1) == 1);
assert(nr_active_pages == 0 && nr_inactive_pages == 1);
assert(list_next(&(inactive_list.swap_list)) == &(rp1->swap_link));
page_launder();
assert(nr_active_pages == 1 && nr_inactive_pages == 0);
assert(PageSwap(rp1) && PageActive(rp1));
entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1);
assert(!PageSwap(rp1) && nr_active_pages == 0);
assert(list_empty(&(active_list.swap_list)));
// set rp1 inactive again
assert(page_ref(rp1) == 1);
swap_page_add(rp1, 0);
assert(PageSwap(rp1) && swap_offset(rp1->index) == 1);
swap_inactive_list_add(rp1);
mem_map[1] = 1;
assert(nr_inactive_pages == 1);
page_ref_dec(rp1);
size_t count = nr_free_pages();
swap_remove_entry(entry);
assert(nr_inactive_pages == 0 && nr_free_pages() == count + 1);
// check swap_out_mm
pte_t *ptep0 = get_pte(pgdir, 0, 0), *ptep1;
assert(ptep0 != NULL && pte2page(*ptep0) == rp0);
ret = swap_out_mm(mm, 0);
assert(ret == 0);
ret = swap_out_mm(mm, 10);
assert(ret == 1 && mm->swap_address == PGSIZE);
ret = swap_out_mm(mm, 10);
assert(ret == 0 && *ptep0 == entry && mem_map[1] == 1);
assert(PageDirty(rp0) && PageActive(rp0) && page_ref(rp0) == 0);
assert(nr_active_pages == 1 && list_next(&(active_list.swap_list)) == &(rp0->swap_link));
// check refill_inactive_scan()
refill_inactive_scan();
assert(!PageActive(rp0) && page_ref(rp0) == 0);
assert(nr_inactive_pages == 1 && list_next(&(inactive_list.swap_list)) == &(rp0->swap_link));
page_ref_inc(rp0);
page_launder();
assert(PageActive(rp0) && page_ref(rp0) == 1);
assert(nr_active_pages == 1 && list_next(&(active_list.swap_list)) == &(rp0->swap_link));
page_ref_dec(rp0);
refill_inactive_scan();
assert(!PageActive(rp0));
// save data in rp0
int i;
for (i = 0; i < PGSIZE; i ++) {
((char *)page2kva(rp0))[i] = (char)i;
}
page_launder();
assert(nr_inactive_pages == 0 && list_empty(&(inactive_list.swap_list)));
assert(mem_map[1] == 1);
rp1 = alloc_page();
assert(rp1 != NULL);
ret = swapfs_read(entry, rp1);
assert(ret == 0);
for (i = 0; i < PGSIZE; i ++) {
assert(((char *)page2kva(rp1))[i] == (char)i);
}
// page fault now
*(char *)0 = 0xEF;
rp0 = pte2page(*ptep0);
assert(page_ref(rp0) == 1);
assert(PageSwap(rp0) && PageActive(rp0));
entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1 && mem_map[1] == SWAP_UNUSED);
assert(!PageSwap(rp0) && nr_active_pages == 0 && nr_inactive_pages == 0);
// clear accessed flag
assert(rp0 == pte2page(*ptep0));
assert(!PageSwap(rp0));
ret = swap_out_mm(mm, 10);
assert(ret == 0);
assert(!PageSwap(rp0) && (*ptep0 & PTE_P));
// change page table
ret = swap_out_mm(mm, 10);
assert(ret == 1);
assert(*ptep0 == entry && page_ref(rp0) == 0 && mem_map[1] == 1);
count = nr_free_pages();
refill_inactive_scan();
page_launder();
assert(count + 1 == nr_free_pages());
ret = swapfs_read(entry, rp1);
assert(ret == 0 && *(char *)(page2kva(rp1)) == (char)0xEF);
free_page(rp1);
// duplictate *ptep0
ptep1 = get_pte(pgdir, PGSIZE, 0);
assert(ptep1 != NULL && *ptep1 == 0);
swap_duplicate(*ptep0);
*ptep1 = *ptep0;
// page fault again
*(char *)0 = 0xFF;
*(char *)(PGSIZE + 1) = 0x88;
assert(pte2page(*ptep0) == pte2page(*ptep1));
rp0 = pte2page(*ptep0);
assert(*(char *)1 == (char)0x88 && *(char *)PGSIZE == (char)0xFF);
assert(page_ref(rp0) == 2 && rp0->index == entry && mem_map[1] == 0);
assert(PageSwap(rp0) && PageActive(rp0));
entry = try_alloc_swap_entry();
assert(swap_offset(entry) == 1 && mem_map[1] == SWAP_UNUSED);
assert(!PageSwap(rp0));
assert(list_empty(&(active_list.swap_list)));
assert(list_empty(&(inactive_list.swap_list)));
// check swap_out_mm
*(char *)0 = *(char *)PGSIZE = 0xEE;
mm->swap_address = PGSIZE * 2;
ret = swap_out_mm(mm, 2);
assert(ret == 0);
assert((*ptep0 & PTE_P) && !(*ptep0 & PTE_A));
assert((*ptep1 & PTE_P) && !(*ptep1 & PTE_A));
ret = swap_out_mm(mm, 2);
assert(ret == 2);
assert(mem_map[1] == 2 && page_ref(rp0) == 0);
refill_inactive_scan();
page_launder();
assert(mem_map[1] == 2 && swap_hash_find(entry) == NULL);
// check copy entry
swap_remove_entry(entry);
*ptep1 = 0;
assert(mem_map[1] == 1);
swap_entry_t store;
ret = swap_copy_entry(entry, &store);
assert(ret == -E_NO_MEM);
mem_map[2] = SWAP_UNUSED;
ret = swap_copy_entry(entry, &store);
assert(ret == 0 && swap_offset(store) == 2 && mem_map[2] == 0);
mem_map[2] = 1;
*ptep1 = store;
assert(*(char *)PGSIZE == (char)0xEE && *(char *)(PGSIZE + 1)== (char)0x88);
*(char *)PGSIZE = 1, *(char *)(PGSIZE + 1) = 2;
assert(*(char *)0 == (char)0xEE && *(char *)1 == (char)0x88);
ret = swap_in_page(entry, &rp0);
assert(ret == 0);
ret = swap_in_page(store, &rp1);
assert(ret == 0);
assert(rp1 != rp0);
// free memory
swap_list_del(rp0), swap_list_del(rp1);
swap_page_del(rp0), swap_page_del(rp1);
assert(page_ref(rp0) == 1 && page_ref(rp1) == 1);
assert(nr_active_pages == 0 && list_empty(&(active_list.swap_list)));
assert(nr_inactive_pages == 0 && list_empty(&(inactive_list.swap_list)));
for (i = 0; i < HASH_LIST_SIZE; i ++) {
assert(list_empty(hash_list + i));
}
page_remove(pgdir, 0);
page_remove(pgdir, PGSIZE);
free_page(pa2page(pgdir[0]));
pgdir[0] = 0;
mm->pgdir = NULL;
mm_destroy(mm);
check_mm_struct = NULL;
assert(nr_active_pages == 0 && nr_inactive_pages == 0);
for (offset = 0; offset < max_swap_offset; offset ++) {
mem_map[offset] = SWAP_UNUSED;
}
assert(nr_free_pages_store == nr_free_pages());
assert(slab_allocated_store == slab_allocated());
cprintf("check_swap() succeeded.\n");
}
static void
check_mm_shm_swap(void) {
size_t nr_free_pages_store = nr_free_pages();
size_t slab_allocated_store = slab_allocated();
int ret, i;
for (i = 0; i < max_swap_offset; i ++) {
assert(mem_map[i] == SWAP_UNUSED);
}
extern struct mm_struct *check_mm_struct;
assert(check_mm_struct == NULL);
struct mm_struct *mm0 = mm_create(), *mm1;
assert(mm0 != NULL && list_empty(&(mm0->mmap_list)));
struct Page *page = alloc_page();
assert(page != NULL);
pde_t *pgdir = page2kva(page);
memcpy(pgdir, boot_pgdir, PGSIZE);
pgdir[PDX(VPT)] = PADDR(pgdir) | PTE_P | PTE_W;
mm0->pgdir = pgdir;
check_mm_struct = mm0;
lcr3(PADDR(mm0->pgdir));
uint32_t vm_flags = VM_WRITE | VM_READ;
uintptr_t addr0, addr1;
addr0 = 0;
do {
if ((ret = mm_map(mm0, addr0, PTSIZE * 4, vm_flags, NULL)) == 0) {
break;
}
addr0 += PTSIZE;
} while (addr0 != 0);
assert(ret == 0 && addr0 != 0 && mm0->map_count == 1);
ret = mm_unmap(mm0, addr0, PTSIZE * 4);
assert(ret == 0 && mm0->map_count == 0);
struct shmem_struct *shmem = shmem_create(PTSIZE * 2);
assert(shmem != NULL && shmem_ref(shmem) == 0);
// step1: check share memory
struct vma_struct *vma;
addr1 = addr0 + PTSIZE * 2;
ret = mm_map_shmem(mm0, addr0, vm_flags, shmem, &vma);
assert(ret == 0);
assert((vma->vm_flags & VM_SHARE) && vma->shmem == shmem && shmem_ref(shmem) == 1);
ret = mm_map_shmem(mm0, addr1, vm_flags, shmem, &vma);
assert(ret == 0);
assert((vma->vm_flags & VM_SHARE) && vma->shmem == shmem && shmem_ref(shmem) == 2);
// page fault
for (i = 0; i < 4; i ++) {
*(char *)(addr0 + i * PGSIZE) = (char)(i * i);
}
for (i = 0; i < 4; i ++) {
assert(*(char *)(addr1 + i * PGSIZE) == (char)(i * i));
}
for (i = 0; i < 4; i ++) {
*(char *)(addr1 + i * PGSIZE) = (char)(- i * i);
}
for (i = 0; i < 4; i ++) {
assert(*(char *)(addr1 + i * PGSIZE) == (char)(- i * i));
}
// check swap
ret = swap_out_mm(mm0, 8) + swap_out_mm(mm0, 8);
assert(ret == 8 && nr_active_pages == 4 && nr_inactive_pages == 0);
refill_inactive_scan();
assert(nr_active_pages == 0 && nr_inactive_pages == 4);
// write & read again
memset((void *)addr0, 0x77, PGSIZE);
for (i = 0; i < PGSIZE; i ++) {
assert(*(char *)(addr1 + i) == (char)0x77);
}
// check unmap
ret = mm_unmap(mm0, addr1, PGSIZE * 4);
assert(ret == 0);
addr0 += 4 * PGSIZE, addr1 += 4 * PGSIZE;
*(char *)(addr0) = (char)(0xDC);
assert(*(char *)(addr1) == (char)(0xDC));
*(char *)(addr1 + PTSIZE) = (char)(0xDC);
assert(*(char *)(addr0 + PTSIZE) == (char)(0xDC));
cprintf("check_mm_shm_swap: step1, share memory ok.\n");
// setup mm1
mm1 = mm_create();
assert(mm1 != NULL);
page = alloc_page();
assert(page != NULL);
pgdir = page2kva(page);
memcpy(pgdir, boot_pgdir, PGSIZE);
pgdir[PDX(VPT)] = PADDR(pgdir) | PTE_P | PTE_W;
mm1->pgdir = pgdir;
ret = dup_mmap(mm1, mm0);
assert(ret == 0 && shmem_ref(shmem) == 4);
// switch to mm1
check_mm_struct = mm1;
lcr3(PADDR(mm1->pgdir));
for (i = 0; i < 4; i ++) {
*(char *)(addr0 + i * PGSIZE) = (char)(0x57 + i);
}
for (i = 0; i < 4; i ++) {
assert(*(char *)(addr1 + i * PGSIZE) == (char)(0x57 + i));
}
check_mm_struct = mm0;
lcr3(PADDR(mm0->pgdir));
for (i = 0; i < 4; i ++) {
assert(*(char *)(addr0 + i * PGSIZE) == (char)(0x57 + i));
assert(*(char *)(addr1 + i * PGSIZE) == (char)(0x57 + i));
}
swap_out_mm(mm1, 4);
exit_mmap(mm1);
free_page(kva2page(mm1->pgdir));
mm_destroy(mm1);
assert(shmem_ref(shmem) == 2);
cprintf("check_mm_shm_swap: step2, dup_mmap ok.\n");
// free memory
check_mm_struct = NULL;
lcr3(boot_cr3);
exit_mmap(mm0);
free_page(kva2page(mm0->pgdir));
mm_destroy(mm0);
refill_inactive_scan();
page_launder();
for (i = 0; i < max_swap_offset; i ++) {
assert(mem_map[i] == SWAP_UNUSED);
}
assert(nr_free_pages_store == nr_free_pages());
assert(slab_allocated_store == slab_allocated());
cprintf("check_mm_shm_swap() succeeded.\n");
}
static int swap_out(unsigned int priority, int gfp_mask, unsigned long idle_time)
{
struct task_struct * p;
int counter;
int __ret = 0;
lock_kernel();
/*
* We make one or two passes through the task list, indexed by
* assign = {0, 1}:
* Pass 1: select the swappable task with maximal RSS that has
* not yet been swapped out.
* Pass 2: re-assign rss swap_cnt values, then select as above.
*
* With this approach, there's no need to remember the last task
* swapped out. If the swap-out fails, we clear swap_cnt so the
* task won't be selected again until all others have been tried.
*
* Think of swap_cnt as a "shadow rss" - it tells us which process
* we want to page out (always try largest first).
*/
counter = (nr_threads << SWAP_SHIFT) >> priority;
if (counter < 1)
counter = 1;
for (; counter >= 0; counter--) {
unsigned long max_cnt = 0;
struct mm_struct *best = NULL;
int pid = 0;
int assign = 0;
int found_task = 0;
select:
read_lock(&tasklist_lock);
p = init_task.next_task;
for (; p != &init_task; p = p->next_task) {
struct mm_struct *mm = p->mm;
if (!p->swappable || !mm)
continue;
if (mm->rss <= 0)
continue;
/* Skip tasks which haven't slept long enough yet when idle-swapping. */
if (idle_time && !assign && (!(p->state & TASK_INTERRUPTIBLE) ||
time_after(p->sleep_time + idle_time * HZ, jiffies)))
continue;
found_task++;
/* Refresh swap_cnt? */
if (assign == 1) {
mm->swap_cnt = (mm->rss >> SWAP_SHIFT);
if (mm->swap_cnt < SWAP_MIN)
mm->swap_cnt = SWAP_MIN;
}
if (mm->swap_cnt > max_cnt) {
max_cnt = mm->swap_cnt;
best = mm;
pid = p->pid;
}
}
read_unlock(&tasklist_lock);
if (!best) {
if (!assign && found_task > 0) {
assign = 1;
goto select;
}
goto out;
} else {
int ret;
atomic_inc(&best->mm_count);
ret = swap_out_mm(best, gfp_mask);
mmdrop(best);
__ret = 1;
goto out;
}
}