2 * Copyright IBM Corp. 2007,2009
3 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
6 #include <linux/sched.h>
7 #include <linux/kernel.h>
8 #include <linux/errno.h>
10 #include <linux/swap.h>
11 #include <linux/smp.h>
12 #include <linux/highmem.h>
13 #include <linux/slab.h>
14 #include <linux/pagemap.h>
15 #include <linux/spinlock.h>
16 #include <linux/module.h>
17 #include <linux/quicklist.h>
19 #include <asm/system.h>
20 #include <asm/pgtable.h>
21 #include <asm/pgalloc.h>
23 #include <asm/tlbflush.h>
24 #include <asm/mmu_context.h>
28 #define TABLES_PER_PAGE 4
29 #define FRAG_MASK 15UL
30 #define SECOND_HALVES 10UL
32 void clear_table_pgstes(unsigned long *table)
34 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/4);
35 memset(table + 256, 0, PAGE_SIZE/4);
36 clear_table(table + 512, _PAGE_TYPE_EMPTY, PAGE_SIZE/4);
37 memset(table + 768, 0, PAGE_SIZE/4);
42 #define TABLES_PER_PAGE 2
44 #define SECOND_HALVES 2UL
46 void clear_table_pgstes(unsigned long *table)
48 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
49 memset(table + 256, 0, PAGE_SIZE/2);
54 unsigned long VMALLOC_START = VMALLOC_END - VMALLOC_SIZE;
55 EXPORT_SYMBOL(VMALLOC_START);
57 static int __init parse_vmalloc(char *arg)
61 VMALLOC_START = (VMALLOC_END - memparse(arg, &arg)) & PAGE_MASK;
64 early_param("vmalloc", parse_vmalloc);
66 unsigned long *crst_table_alloc(struct mm_struct *mm, int noexec)
68 struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
74 struct page *shadow = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
76 __free_pages(page, ALLOC_ORDER);
79 page->index = page_to_phys(shadow);
81 spin_lock(&mm->page_table_lock);
82 list_add(&page->lru, &mm->context.crst_list);
83 spin_unlock(&mm->page_table_lock);
84 return (unsigned long *) page_to_phys(page);
87 void crst_table_free(struct mm_struct *mm, unsigned long *table)
89 unsigned long *shadow = get_shadow_table(table);
90 struct page *page = virt_to_page(table);
92 spin_lock(&mm->page_table_lock);
94 spin_unlock(&mm->page_table_lock);
96 free_pages((unsigned long) shadow, ALLOC_ORDER);
97 free_pages((unsigned long) table, ALLOC_ORDER);
101 int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
103 unsigned long *table, *pgd;
106 BUG_ON(limit > (1UL << 53));
108 table = crst_table_alloc(mm, mm->context.noexec);
111 spin_lock(&mm->page_table_lock);
112 if (mm->context.asce_limit < limit) {
113 pgd = (unsigned long *) mm->pgd;
114 if (mm->context.asce_limit <= (1UL << 31)) {
115 entry = _REGION3_ENTRY_EMPTY;
116 mm->context.asce_limit = 1UL << 42;
117 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
121 entry = _REGION2_ENTRY_EMPTY;
122 mm->context.asce_limit = 1UL << 53;
123 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
127 crst_table_init(table, entry);
128 pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
129 mm->pgd = (pgd_t *) table;
130 mm->task_size = mm->context.asce_limit;
133 spin_unlock(&mm->page_table_lock);
135 crst_table_free(mm, table);
136 if (mm->context.asce_limit < limit)
138 update_mm(mm, current);
142 void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
146 if (mm->context.asce_limit <= limit)
149 while (mm->context.asce_limit > limit) {
151 switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
152 case _REGION_ENTRY_TYPE_R2:
153 mm->context.asce_limit = 1UL << 42;
154 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
158 case _REGION_ENTRY_TYPE_R3:
159 mm->context.asce_limit = 1UL << 31;
160 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
167 mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
168 mm->task_size = mm->context.asce_limit;
169 crst_table_free(mm, (unsigned long *) pgd);
171 update_mm(mm, current);
176 * page table entry allocation/free routines.
178 unsigned long *page_table_alloc(struct mm_struct *mm)
181 unsigned long *table;
184 bits = (mm->context.noexec || mm->context.has_pgste) ? 3UL : 1UL;
185 spin_lock(&mm->page_table_lock);
187 if (!list_empty(&mm->context.pgtable_list)) {
188 page = list_first_entry(&mm->context.pgtable_list,
190 if ((page->flags & FRAG_MASK) == ((1UL << TABLES_PER_PAGE) - 1))
194 spin_unlock(&mm->page_table_lock);
195 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
198 pgtable_page_ctor(page);
199 page->flags &= ~FRAG_MASK;
200 table = (unsigned long *) page_to_phys(page);
201 if (mm->context.has_pgste)
202 clear_table_pgstes(table);
204 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE);
205 spin_lock(&mm->page_table_lock);
206 list_add(&page->lru, &mm->context.pgtable_list);
208 table = (unsigned long *) page_to_phys(page);
209 while (page->flags & bits) {
214 if ((page->flags & FRAG_MASK) == ((1UL << TABLES_PER_PAGE) - 1))
215 list_move_tail(&page->lru, &mm->context.pgtable_list);
216 spin_unlock(&mm->page_table_lock);
220 void page_table_free(struct mm_struct *mm, unsigned long *table)
225 bits = (mm->context.noexec || mm->context.has_pgste) ? 3UL : 1UL;
226 bits <<= (__pa(table) & (PAGE_SIZE - 1)) / 256 / sizeof(unsigned long);
227 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
228 spin_lock(&mm->page_table_lock);
230 if (page->flags & FRAG_MASK) {
231 /* Page now has some free pgtable fragments. */
232 list_move(&page->lru, &mm->context.pgtable_list);
235 /* All fragments of the 4K page have been freed. */
236 list_del(&page->lru);
237 spin_unlock(&mm->page_table_lock);
239 pgtable_page_dtor(page);
244 void disable_noexec(struct mm_struct *mm, struct task_struct *tsk)
248 spin_lock(&mm->page_table_lock);
249 /* Free shadow region and segment tables. */
250 list_for_each_entry(page, &mm->context.crst_list, lru)
252 free_pages((unsigned long) page->index, ALLOC_ORDER);
255 /* "Free" second halves of page tables. */
256 list_for_each_entry(page, &mm->context.pgtable_list, lru)
257 page->flags &= ~SECOND_HALVES;
258 spin_unlock(&mm->page_table_lock);
259 mm->context.noexec = 0;
264 * switch on pgstes for its userspace process (for kvm)
266 int s390_enable_sie(void)
268 struct task_struct *tsk = current;
269 struct mm_struct *mm, *old_mm;
271 /* Do we have switched amode? If no, we cannot do sie */
275 /* Do we have pgstes? if yes, we are done */
276 if (tsk->mm->context.has_pgste)
279 /* lets check if we are allowed to replace the mm */
281 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
282 tsk->mm != tsk->active_mm || !hlist_empty(&tsk->mm->ioctx_list)) {
288 /* we copy the mm and let dup_mm create the page tables with_pgstes */
289 tsk->mm->context.alloc_pgste = 1;
291 tsk->mm->context.alloc_pgste = 0;
295 /* Now lets check again if something happened */
297 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
298 tsk->mm != tsk->active_mm || !hlist_empty(&tsk->mm->ioctx_list)) {
304 /* ok, we are alone. No ptrace, no threads, etc. */
306 tsk->mm = tsk->active_mm = mm;
309 cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
315 EXPORT_SYMBOL_GPL(s390_enable_sie);