// swap_out_vma - try unmap pte & move pages into swap active list.
static int
swap_out_vma(struct mm_struct *mm, struct vma_struct *vma, uintptr_t addr, size_t require) {
if (require == 0 || !(addr >= vma->vm_start && addr < vma->vm_end)) {
return 0;
}
uintptr_t end;
size_t free_count = 0;
addr = ROUNDDOWN(addr, PGSIZE), end = ROUNDUP(vma->vm_end, PGSIZE);
while (addr < end && require != 0) {
pte_t *ptep = get_pte(mm->pgdir, addr, 0);
if (ptep == NULL) {
if (get_pud(mm->pgdir, addr, 0) == NULL) {
addr = ROUNDDOWN(addr + PUSIZE, PUSIZE);
}
else if (get_pmd(mm->pgdir, addr, 0) == NULL) {
addr = ROUNDDOWN(addr + PMSIZE, PMSIZE);
}
else {
addr = ROUNDDOWN(addr + PTSIZE, PTSIZE);
}
continue ;
}
if (ptep_present(ptep)) {
struct Page *page = pte2page(*ptep);
assert(!PageReserved(page));
if (ptep_accessed(ptep)) {
ptep_unset_accessed(ptep);
mp_tlb_invalidate(mm->pgdir, addr);
goto try_next_entry;
}
if (!PageSwap(page)) {
if (!swap_page_add(page, 0)) {
goto try_next_entry;
}
swap_active_list_add(page);
}
else if (ptep_dirty(ptep)) {
SetPageDirty(page);
}
swap_entry_t entry = page->index;
swap_duplicate(entry);
page_ref_dec(page);
ptep_copy(ptep, &entry);
mp_tlb_invalidate(mm->pgdir, addr);
mm->swap_address = addr + PGSIZE;
free_count ++, require --;
if ((vma->vm_flags & VM_SHARE) && page_ref(page) == 1) {
uintptr_t shmem_addr = addr - vma->vm_start + vma->shmem_off;
pte_t *sh_ptep = shmem_get_entry(vma->shmem, shmem_addr, 0);
assert(sh_ptep != NULL && ! ptep_invalid(sh_ptep));
if (ptep_present(sh_ptep)) {
shmem_insert_entry(vma->shmem, shmem_addr, entry);
}
}
}
try_next_entry:
addr += PGSIZE;
}
return free_count;
}
int
shmem_insert_entry(struct shmem_struct *shmem, uintptr_t addr, pte_t entry) {
pte_t *ptep = shmem_get_entry(shmem, addr, 1);
if (ptep == NULL) {
return -E_NO_MEM;
}
if (! ptep_invalid(ptep)) {
shmem_remove_entry_pte(ptep);
}
if (ptep_present(&entry)) {
page_ref_inc(pte2page(entry));
}
else if (! ptep_invalid(&entry)) {
swap_duplicate(entry);
}
ptep_copy(ptep, &entry);
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");
}
/* ucore use copy-on-write when forking a new process,
* thus copy_range only copy pdt/pte and set their permission to
* READONLY, a write will be handled in pgfault
*/
int
copy_range(pgd_t *to, pgd_t *from, uintptr_t start, uintptr_t end, bool share) {
assert(start % PGSIZE == 0 && end % PGSIZE == 0);
assert(USER_ACCESS(start, end));
do {
pte_t *ptep = get_pte(from, start, 0), *nptep;
if (ptep == NULL) {
if (get_pud(from, start, 0) == NULL) {
start = ROUNDDOWN(start + PUSIZE, PUSIZE);
}
else if (get_pmd(from, start, 0) == NULL) {
start = ROUNDDOWN(start + PMSIZE, PMSIZE);
}
else {
start = ROUNDDOWN(start + PTSIZE, PTSIZE);
}
continue ;
}
if (*ptep != 0) {
if ((nptep = get_pte(to, start, 1)) == NULL) {
return -E_NO_MEM;
}
int ret;
//kprintf("%08x %08x %08x\n", nptep, *nptep, start);
assert(*ptep != 0 && *nptep == 0);
#ifdef ARCH_ARM
//TODO add code to handle swap
if (ptep_present(ptep)){
//no body should be able to write this page
//before a W-pgfault
pte_perm_t perm = PTE_P;
if(ptep_u_read(ptep))
perm |= PTE_U;
if(!share){
//Original page should be set to readonly!
//because Copy-on-write may happen
//after the current proccess modifies its page
ptep_set_perm(ptep, perm);
}else{
if(ptep_u_write(ptep)){
perm |= PTE_W;
}
}
struct Page *page = pte2page(*ptep);
ret = page_insert(to, page, start, perm);
}
#else /* ARCH_ARM */
if (ptep_present(ptep)) {
pte_perm_t perm = ptep_get_perm(ptep, PTE_USER);
struct Page *page = pte2page(*ptep);
if (!share && ptep_s_write(ptep)) {
ptep_unset_s_write(&perm);
pte_perm_t perm_with_swap_stat = ptep_get_perm(ptep, PTE_SWAP);
ptep_set_perm(&perm_with_swap_stat, perm);
page_insert(from, page, start, perm_with_swap_stat);
}
ret = page_insert(to, page, start, perm);
assert(ret == 0);
}
#endif /* ARCH_ARM */
else {
#ifdef CONFIG_NO_SWAP
assert(0);
#endif
swap_entry_t entry;
ptep_copy(&entry, ptep);
swap_duplicate(entry);
ptep_copy(nptep, &entry);
}
}
start += PGSIZE;
} while (start != 0 && start < end);
#ifdef ARCH_ARM
/* we have modified the PTE of the original
* process, so invalidate TLB */
tlb_invalidate_all();
#endif
return 0;
}
int do_pgfault(struct mm_struct *mm, machine_word_t error_code, uintptr_t addr)
{
if (mm == NULL) {
assert(current != NULL);
/* Chen Yuheng
* give handler a chance to deal with it
*/
kprintf
("page fault in kernel thread: pid = %d, name = %s, %d %08x.\n",
current->pid, current->name, error_code, addr);
return -E_KILLED;
}
bool need_unlock = 1;
if (!try_lock_mm(mm)) {
if (current != NULL && mm->locked_by == current->pid) {
need_unlock = 0;
} else {
lock_mm(mm);
}
}
int ret = -E_INVAL;
struct vma_struct *vma = find_vma(mm, addr);
if (vma == NULL || vma->vm_start > addr) {
goto failed;
}
if (vma->vm_flags & VM_STACK) {
if (addr < vma->vm_start + PGSIZE) {
goto failed;
}
}
//kprintf("@ %x %08x\n", vma->vm_flags, vma->vm_start);
//assert((vma->vm_flags & VM_IO)==0);
if (vma->vm_flags & VM_IO) {
ret = -E_INVAL;
goto failed;
}
switch (error_code & 3) {
default:
/* default is 3: write, present */
case 2: /* write, not present */
if (!(vma->vm_flags & VM_WRITE)) {
goto failed;
}
break;
case 1: /* read, present */
goto failed;
case 0: /* read, not present */
if (!(vma->vm_flags & (VM_READ | VM_EXEC))) {
goto failed;
}
}
pte_perm_t perm, nperm;
#ifdef ARCH_ARM
/* ARM9 software emulated PTE_xxx */
perm = PTE_P | PTE_U;
if (vma->vm_flags & VM_WRITE) {
perm |= PTE_W;
}
#else
ptep_unmap(&perm);
ptep_set_u_read(&perm);
if (vma->vm_flags & VM_WRITE) {
ptep_set_u_write(&perm);
}
#endif
addr = ROUNDDOWN(addr, PGSIZE);
ret = -E_NO_MEM;
pte_t *ptep;
if ((ptep = get_pte(mm->pgdir, addr, 1)) == NULL) {
goto failed;
}
if (ptep_invalid(ptep)) {
#ifdef UCONFIG_BIONIC_LIBC
if (vma->mfile.file != NULL) {
struct file *file = vma->mfile.file;
off_t old_pos = file->pos, new_pos =
vma->mfile.offset + addr - vma->vm_start;
#ifdef SHARE_MAPPED_FILE
struct mapped_addr *maddr =
find_maddr(file, new_pos, NULL);
if (maddr == NULL) {
#endif // SHARE_MAPPED_FILE
struct Page *page;
if ((page = alloc_page()) == NULL) {
assert(false);
goto failed;
}
nperm = perm;
#ifdef ARCH_ARM
/* ARM9 software emulated PTE_xxx */
nperm &= ~PTE_W;
#else
ptep_unset_s_write(&nperm);
#endif
page_insert_pte(mm->pgdir, page, ptep, addr,
nperm);
if ((ret =
filestruct_setpos(file, new_pos)) != 0) {
assert(false);
goto failed;
}
filestruct_read(file, page2kva(page), PGSIZE);
if ((ret =
filestruct_setpos(file, old_pos)) != 0) {
assert(false);
goto failed;
}
#ifdef SHARE_MAPPED_FILE
if ((maddr = (struct mapped_addr *)
kmalloc(sizeof(struct mapped_addr))) !=
NULL) {
maddr->page = page;
maddr->offset = new_pos;
page->maddr = maddr;
list_add(&
(file->node->mapped_addr_list),
&(maddr->list));
} else {
assert(false);
}
} else {
nperm = perm;
#ifdef ARCH_ARM
/* ARM9 software emulated PTE_xxx */
nperm &= ~PTE_W;
#else
ptep_unset_s_write(&nperm);
#endif
page_insert_pte(mm->pgdir, maddr->page, ptep,
addr, nperm);
}
#endif //SHARE_MAPPED_FILE
} else
#endif //UCONFIG_BIONIC_LIBC
if (!(vma->vm_flags & VM_SHARE)) {
if (pgdir_alloc_page(mm->pgdir, addr, perm) == NULL) {
goto failed;
}
#ifdef UCONFIG_BIONIC_LIBC
if (vma->vm_flags & VM_ANONYMOUS) {
memset((void *)addr, 0, PGSIZE);
}
#endif //UCONFIG_BIONIC_LIBC
} else { //shared mem
lock_shmem(vma->shmem);
uintptr_t shmem_addr =
addr - vma->vm_start + vma->shmem_off;
pte_t *sh_ptep =
shmem_get_entry(vma->shmem, shmem_addr, 1);
if (sh_ptep == NULL || ptep_invalid(sh_ptep)) {
unlock_shmem(vma->shmem);
goto failed;
}
unlock_shmem(vma->shmem);
if (ptep_present(sh_ptep)) {
page_insert(mm->pgdir, pa2page(*sh_ptep), addr,
perm);
} else {
#ifdef UCONFIG_SWAP
swap_duplicate(*ptep);
ptep_copy(ptep, sh_ptep);
#else
panic("NO SWAP\n");
#endif
}
}
} else { //a present page, handle copy-on-write (cow)
struct Page *page, *newpage = NULL;
bool cow =
((vma->vm_flags & (VM_SHARE | VM_WRITE)) == VM_WRITE),
may_copy = 1;
#if 1
if (!(!ptep_present(ptep)
|| ((error_code & 2) && !ptep_u_write(ptep) && cow))) {
//assert(PADDR(mm->pgdir) == rcr3());
kprintf("%p %p %d %d %x\n", *ptep, addr, error_code,
cow, vma->vm_flags);
assert(0);
}
#endif
if (cow) {
newpage = alloc_page();
}
if (ptep_present(ptep)) {
page = pte2page(*ptep);
} else {
#ifdef UCONFIG_SWAP
if ((ret = swap_in_page(*ptep, &page)) != 0) {
if (newpage != NULL) {
free_page(newpage);
}
goto failed;
}
#else
assert(0);
#endif
if (!(error_code & 2) && cow) {
#ifdef ARCH_ARM
//#warning ARM9 software emulated PTE_xxx
perm &= ~PTE_W;
#else
ptep_unset_s_write(&perm);
#endif
may_copy = 0;
}
}
if (cow && may_copy) {
#ifdef UCONFIG_SWAP
if (page_ref(page) + swap_page_count(page) > 1) {
#else
if (page_ref(page) > 1) {
#endif
if (newpage == NULL) {
goto failed;
}
memcpy(page2kva(newpage), page2kva(page),
PGSIZE);
//kprintf("COW!\n");
page = newpage, newpage = NULL;
}
}
#ifdef UCONFIG_BIONIC_LIBC
else if (vma->mfile.file != NULL) {
#ifdef UCONFIG_SWAP
assert(page_reg(page) + swap_page_count(page) == 1);
#else
assert(page_ref(page) == 1);
#endif
#ifdef SHARE_MAPPED_FILE
off_t offset = vma->mfile.offset + addr - vma->vm_start;
struct mapped_addr *maddr =
find_maddr(vma->mfile.file, offset, page);
if (maddr != NULL) {
list_del(&(maddr->list));
kfree(maddr);
page->maddr = NULL;
assert(find_maddr(vma->mfile.file, offset, page)
== NULL);
} else {
}
#endif //SHARE_MAPPED_FILE
}
#endif //UCONFIG_BIONIC_LIBC
else {
}
page_insert(mm->pgdir, page, addr, perm);
if (newpage != NULL) {
free_page(newpage);
}
}
ret = 0;
failed:
if (need_unlock) {
unlock_mm(mm);
}
return ret;
}