Merge branch 'virtex-for-2.6.25' of git://git.secretlab.ca/git/linux-2.6-virtex into...
[sfrench/cifs-2.6.git] / arch / s390 / mm / vmem.c
1 /*
2  *  arch/s390/mm/vmem.c
3  *
4  *    Copyright IBM Corp. 2006
5  *    Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
6  */
7
8 #include <linux/bootmem.h>
9 #include <linux/pfn.h>
10 #include <linux/mm.h>
11 #include <linux/module.h>
12 #include <linux/list.h>
13 #include <asm/pgalloc.h>
14 #include <asm/pgtable.h>
15 #include <asm/setup.h>
16 #include <asm/tlbflush.h>
17
18 static DEFINE_MUTEX(vmem_mutex);
19
20 struct memory_segment {
21         struct list_head list;
22         unsigned long start;
23         unsigned long size;
24 };
25
26 static LIST_HEAD(mem_segs);
27
28 void __meminit memmap_init(unsigned long size, int nid, unsigned long zone,
29                            unsigned long start_pfn)
30 {
31         struct page *start, *end;
32         struct page *map_start, *map_end;
33         int i;
34
35         start = pfn_to_page(start_pfn);
36         end = start + size;
37
38         for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
39                 unsigned long cstart, cend;
40
41                 cstart = PFN_DOWN(memory_chunk[i].addr);
42                 cend = cstart + PFN_DOWN(memory_chunk[i].size);
43
44                 map_start = mem_map + cstart;
45                 map_end = mem_map + cend;
46
47                 if (map_start < start)
48                         map_start = start;
49                 if (map_end > end)
50                         map_end = end;
51
52                 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1))
53                         / sizeof(struct page);
54                 map_end += ((PFN_ALIGN((unsigned long) map_end)
55                              - (unsigned long) map_end)
56                             / sizeof(struct page));
57
58                 if (map_start < map_end)
59                         memmap_init_zone((unsigned long)(map_end - map_start),
60                                          nid, zone, page_to_pfn(map_start),
61                                          MEMMAP_EARLY);
62         }
63 }
64
65 static void __ref *vmem_alloc_pages(unsigned int order)
66 {
67         if (slab_is_available())
68                 return (void *)__get_free_pages(GFP_KERNEL, order);
69         return alloc_bootmem_pages((1 << order) * PAGE_SIZE);
70 }
71
72 #define vmem_pud_alloc()        ({ BUG(); ((pud_t *) NULL); })
73
74 static inline pmd_t *vmem_pmd_alloc(void)
75 {
76         pmd_t *pmd = NULL;
77
78 #ifdef CONFIG_64BIT
79         pmd = vmem_alloc_pages(2);
80         if (!pmd)
81                 return NULL;
82         clear_table((unsigned long *) pmd, _SEGMENT_ENTRY_EMPTY, PAGE_SIZE*4);
83 #endif
84         return pmd;
85 }
86
87 static inline pte_t *vmem_pte_alloc(void)
88 {
89         pte_t *pte = vmem_alloc_pages(0);
90
91         if (!pte)
92                 return NULL;
93         clear_table((unsigned long *) pte, _PAGE_TYPE_EMPTY, PAGE_SIZE);
94         return pte;
95 }
96
97 /*
98  * Add a physical memory range to the 1:1 mapping.
99  */
100 static int vmem_add_range(unsigned long start, unsigned long size)
101 {
102         unsigned long address;
103         pgd_t *pg_dir;
104         pud_t *pu_dir;
105         pmd_t *pm_dir;
106         pte_t *pt_dir;
107         pte_t  pte;
108         int ret = -ENOMEM;
109
110         for (address = start; address < start + size; address += PAGE_SIZE) {
111                 pg_dir = pgd_offset_k(address);
112                 if (pgd_none(*pg_dir)) {
113                         pu_dir = vmem_pud_alloc();
114                         if (!pu_dir)
115                                 goto out;
116                         pgd_populate_kernel(&init_mm, pg_dir, pu_dir);
117                 }
118
119                 pu_dir = pud_offset(pg_dir, address);
120                 if (pud_none(*pu_dir)) {
121                         pm_dir = vmem_pmd_alloc();
122                         if (!pm_dir)
123                                 goto out;
124                         pud_populate_kernel(&init_mm, pu_dir, pm_dir);
125                 }
126
127                 pm_dir = pmd_offset(pu_dir, address);
128                 if (pmd_none(*pm_dir)) {
129                         pt_dir = vmem_pte_alloc();
130                         if (!pt_dir)
131                                 goto out;
132                         pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
133                 }
134
135                 pt_dir = pte_offset_kernel(pm_dir, address);
136                 pte = pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL);
137                 *pt_dir = pte;
138         }
139         ret = 0;
140 out:
141         flush_tlb_kernel_range(start, start + size);
142         return ret;
143 }
144
145 /*
146  * Remove a physical memory range from the 1:1 mapping.
147  * Currently only invalidates page table entries.
148  */
149 static void vmem_remove_range(unsigned long start, unsigned long size)
150 {
151         unsigned long address;
152         pgd_t *pg_dir;
153         pud_t *pu_dir;
154         pmd_t *pm_dir;
155         pte_t *pt_dir;
156         pte_t  pte;
157
158         pte_val(pte) = _PAGE_TYPE_EMPTY;
159         for (address = start; address < start + size; address += PAGE_SIZE) {
160                 pg_dir = pgd_offset_k(address);
161                 pu_dir = pud_offset(pg_dir, address);
162                 if (pud_none(*pu_dir))
163                         continue;
164                 pm_dir = pmd_offset(pu_dir, address);
165                 if (pmd_none(*pm_dir))
166                         continue;
167                 pt_dir = pte_offset_kernel(pm_dir, address);
168                 *pt_dir = pte;
169         }
170         flush_tlb_kernel_range(start, start + size);
171 }
172
173 /*
174  * Add a backed mem_map array to the virtual mem_map array.
175  */
176 static int vmem_add_mem_map(unsigned long start, unsigned long size)
177 {
178         unsigned long address, start_addr, end_addr;
179         struct page *map_start, *map_end;
180         pgd_t *pg_dir;
181         pud_t *pu_dir;
182         pmd_t *pm_dir;
183         pte_t *pt_dir;
184         pte_t  pte;
185         int ret = -ENOMEM;
186
187         map_start = VMEM_MAP + PFN_DOWN(start);
188         map_end = VMEM_MAP + PFN_DOWN(start + size);
189
190         start_addr = (unsigned long) map_start & PAGE_MASK;
191         end_addr = PFN_ALIGN((unsigned long) map_end);
192
193         for (address = start_addr; address < end_addr; address += PAGE_SIZE) {
194                 pg_dir = pgd_offset_k(address);
195                 if (pgd_none(*pg_dir)) {
196                         pu_dir = vmem_pud_alloc();
197                         if (!pu_dir)
198                                 goto out;
199                         pgd_populate_kernel(&init_mm, pg_dir, pu_dir);
200                 }
201
202                 pu_dir = pud_offset(pg_dir, address);
203                 if (pud_none(*pu_dir)) {
204                         pm_dir = vmem_pmd_alloc();
205                         if (!pm_dir)
206                                 goto out;
207                         pud_populate_kernel(&init_mm, pu_dir, pm_dir);
208                 }
209
210                 pm_dir = pmd_offset(pu_dir, address);
211                 if (pmd_none(*pm_dir)) {
212                         pt_dir = vmem_pte_alloc();
213                         if (!pt_dir)
214                                 goto out;
215                         pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
216                 }
217
218                 pt_dir = pte_offset_kernel(pm_dir, address);
219                 if (pte_none(*pt_dir)) {
220                         unsigned long new_page;
221
222                         new_page =__pa(vmem_alloc_pages(0));
223                         if (!new_page)
224                                 goto out;
225                         pte = pfn_pte(new_page >> PAGE_SHIFT, PAGE_KERNEL);
226                         *pt_dir = pte;
227                 }
228         }
229         ret = 0;
230 out:
231         flush_tlb_kernel_range(start_addr, end_addr);
232         return ret;
233 }
234
235 static int vmem_add_mem(unsigned long start, unsigned long size)
236 {
237         int ret;
238
239         ret = vmem_add_mem_map(start, size);
240         if (ret)
241                 return ret;
242         return vmem_add_range(start, size);
243 }
244
245 /*
246  * Add memory segment to the segment list if it doesn't overlap with
247  * an already present segment.
248  */
249 static int insert_memory_segment(struct memory_segment *seg)
250 {
251         struct memory_segment *tmp;
252
253         if (seg->start + seg->size >= VMEM_MAX_PHYS ||
254             seg->start + seg->size < seg->start)
255                 return -ERANGE;
256
257         list_for_each_entry(tmp, &mem_segs, list) {
258                 if (seg->start >= tmp->start + tmp->size)
259                         continue;
260                 if (seg->start + seg->size <= tmp->start)
261                         continue;
262                 return -ENOSPC;
263         }
264         list_add(&seg->list, &mem_segs);
265         return 0;
266 }
267
268 /*
269  * Remove memory segment from the segment list.
270  */
271 static void remove_memory_segment(struct memory_segment *seg)
272 {
273         list_del(&seg->list);
274 }
275
276 static void __remove_shared_memory(struct memory_segment *seg)
277 {
278         remove_memory_segment(seg);
279         vmem_remove_range(seg->start, seg->size);
280 }
281
282 int remove_shared_memory(unsigned long start, unsigned long size)
283 {
284         struct memory_segment *seg;
285         int ret;
286
287         mutex_lock(&vmem_mutex);
288
289         ret = -ENOENT;
290         list_for_each_entry(seg, &mem_segs, list) {
291                 if (seg->start == start && seg->size == size)
292                         break;
293         }
294
295         if (seg->start != start || seg->size != size)
296                 goto out;
297
298         ret = 0;
299         __remove_shared_memory(seg);
300         kfree(seg);
301 out:
302         mutex_unlock(&vmem_mutex);
303         return ret;
304 }
305
306 int add_shared_memory(unsigned long start, unsigned long size)
307 {
308         struct memory_segment *seg;
309         struct page *page;
310         unsigned long pfn, num_pfn, end_pfn;
311         int ret;
312
313         mutex_lock(&vmem_mutex);
314         ret = -ENOMEM;
315         seg = kzalloc(sizeof(*seg), GFP_KERNEL);
316         if (!seg)
317                 goto out;
318         seg->start = start;
319         seg->size = size;
320
321         ret = insert_memory_segment(seg);
322         if (ret)
323                 goto out_free;
324
325         ret = vmem_add_mem(start, size);
326         if (ret)
327                 goto out_remove;
328
329         pfn = PFN_DOWN(start);
330         num_pfn = PFN_DOWN(size);
331         end_pfn = pfn + num_pfn;
332
333         page = pfn_to_page(pfn);
334         memset(page, 0, num_pfn * sizeof(struct page));
335
336         for (; pfn < end_pfn; pfn++) {
337                 page = pfn_to_page(pfn);
338                 init_page_count(page);
339                 reset_page_mapcount(page);
340                 SetPageReserved(page);
341                 INIT_LIST_HEAD(&page->lru);
342         }
343         goto out;
344
345 out_remove:
346         __remove_shared_memory(seg);
347 out_free:
348         kfree(seg);
349 out:
350         mutex_unlock(&vmem_mutex);
351         return ret;
352 }
353
354 /*
355  * map whole physical memory to virtual memory (identity mapping)
356  * we reserve enough space in the vmalloc area for vmemmap to hotplug
357  * additional memory segments.
358  */
359 void __init vmem_map_init(void)
360 {
361         int i;
362
363         NODE_DATA(0)->node_mem_map = VMEM_MAP;
364         for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++)
365                 vmem_add_mem(memory_chunk[i].addr, memory_chunk[i].size);
366 }
367
368 /*
369  * Convert memory chunk array to a memory segment list so there is a single
370  * list that contains both r/w memory and shared memory segments.
371  */
372 static int __init vmem_convert_memory_chunk(void)
373 {
374         struct memory_segment *seg;
375         int i;
376
377         mutex_lock(&vmem_mutex);
378         for (i = 0; i < MEMORY_CHUNKS; i++) {
379                 if (!memory_chunk[i].size)
380                         continue;
381                 seg = kzalloc(sizeof(*seg), GFP_KERNEL);
382                 if (!seg)
383                         panic("Out of memory...\n");
384                 seg->start = memory_chunk[i].addr;
385                 seg->size = memory_chunk[i].size;
386                 insert_memory_segment(seg);
387         }
388         mutex_unlock(&vmem_mutex);
389         return 0;
390 }
391
392 core_initcall(vmem_convert_memory_chunk);