]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/parisc/mm/init.c | |
3 | * | |
4 | * Copyright (C) 1995 Linus Torvalds | |
5 | * Copyright 1999 SuSE GmbH | |
6 | * changed by Philipp Rumpf | |
7 | * Copyright 1999 Philipp Rumpf (prumpf@tux.org) | |
8 | * Copyright 2004 Randolph Chung (tausq@debian.org) | |
a8f44e38 | 9 | * Copyright 2006-2007 Helge Deller (deller@gmx.de) |
1da177e4 LT |
10 | * |
11 | */ | |
12 | ||
1da177e4 LT |
13 | |
14 | #include <linux/module.h> | |
15 | #include <linux/mm.h> | |
16 | #include <linux/bootmem.h> | |
5a0e3ad6 | 17 | #include <linux/gfp.h> |
1da177e4 LT |
18 | #include <linux/delay.h> |
19 | #include <linux/init.h> | |
20 | #include <linux/pci.h> /* for hppa_dma_ops and pcxl_dma_ops */ | |
21 | #include <linux/initrd.h> | |
22 | #include <linux/swap.h> | |
23 | #include <linux/unistd.h> | |
24 | #include <linux/nodemask.h> /* for node_online_map */ | |
25 | #include <linux/pagemap.h> /* for release_pages and page_cache_release */ | |
26 | ||
27 | #include <asm/pgalloc.h> | |
ce8420bb | 28 | #include <asm/pgtable.h> |
1da177e4 LT |
29 | #include <asm/tlb.h> |
30 | #include <asm/pdc_chassis.h> | |
31 | #include <asm/mmzone.h> | |
a581c2a4 | 32 | #include <asm/sections.h> |
1da177e4 | 33 | |
1da177e4 | 34 | extern int data_start; |
161bd3bf | 35 | extern void parisc_kernel_start(void); /* Kernel entry point in head.S */ |
1da177e4 | 36 | |
f24ffde4 | 37 | #if CONFIG_PGTABLE_LEVELS == 3 |
c39f52a9 TG |
38 | /* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout |
39 | * with the first pmd adjacent to the pgd and below it. gcc doesn't actually | |
40 | * guarantee that global objects will be laid out in memory in the same order | |
41 | * as the order of declaration, so put these in different sections and use | |
42 | * the linker script to order them. */ | |
43 | pmd_t pmd0[PTRS_PER_PMD] __attribute__ ((__section__ (".data..vm0.pmd"), aligned(PAGE_SIZE))); | |
44 | #endif | |
45 | ||
46 | pgd_t swapper_pg_dir[PTRS_PER_PGD] __attribute__ ((__section__ (".data..vm0.pgd"), aligned(PAGE_SIZE))); | |
47 | pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __attribute__ ((__section__ (".data..vm0.pte"), aligned(PAGE_SIZE))); | |
48 | ||
1da177e4 | 49 | #ifdef CONFIG_DISCONTIGMEM |
8039de10 | 50 | struct node_map_data node_data[MAX_NUMNODES] __read_mostly; |
91ea8207 | 51 | signed char pfnnid_map[PFNNID_MAP_MAX] __read_mostly; |
1da177e4 LT |
52 | #endif |
53 | ||
54 | static struct resource data_resource = { | |
55 | .name = "Kernel data", | |
56 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM, | |
57 | }; | |
58 | ||
59 | static struct resource code_resource = { | |
60 | .name = "Kernel code", | |
61 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM, | |
62 | }; | |
63 | ||
64 | static struct resource pdcdata_resource = { | |
65 | .name = "PDC data (Page Zero)", | |
66 | .start = 0, | |
67 | .end = 0x9ff, | |
68 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM, | |
69 | }; | |
70 | ||
8039de10 | 71 | static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly; |
1da177e4 LT |
72 | |
73 | /* The following array is initialized from the firmware specific | |
74 | * information retrieved in kernel/inventory.c. | |
75 | */ | |
76 | ||
8039de10 HD |
77 | physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly; |
78 | int npmem_ranges __read_mostly; | |
1da177e4 | 79 | |
a8f44e38 | 80 | #ifdef CONFIG_64BIT |
1da177e4 | 81 | #define MAX_MEM (~0UL) |
a8f44e38 | 82 | #else /* !CONFIG_64BIT */ |
1da177e4 | 83 | #define MAX_MEM (3584U*1024U*1024U) |
a8f44e38 | 84 | #endif /* !CONFIG_64BIT */ |
1da177e4 | 85 | |
8039de10 | 86 | static unsigned long mem_limit __read_mostly = MAX_MEM; |
1da177e4 LT |
87 | |
88 | static void __init mem_limit_func(void) | |
89 | { | |
90 | char *cp, *end; | |
91 | unsigned long limit; | |
1da177e4 LT |
92 | |
93 | /* We need this before __setup() functions are called */ | |
94 | ||
95 | limit = MAX_MEM; | |
668f9931 | 96 | for (cp = boot_command_line; *cp; ) { |
1da177e4 LT |
97 | if (memcmp(cp, "mem=", 4) == 0) { |
98 | cp += 4; | |
99 | limit = memparse(cp, &end); | |
100 | if (end != cp) | |
101 | break; | |
102 | cp = end; | |
103 | } else { | |
104 | while (*cp != ' ' && *cp) | |
105 | ++cp; | |
106 | while (*cp == ' ') | |
107 | ++cp; | |
108 | } | |
109 | } | |
110 | ||
111 | if (limit < mem_limit) | |
112 | mem_limit = limit; | |
113 | } | |
114 | ||
115 | #define MAX_GAP (0x40000000UL >> PAGE_SHIFT) | |
116 | ||
117 | static void __init setup_bootmem(void) | |
118 | { | |
119 | unsigned long bootmap_size; | |
120 | unsigned long mem_max; | |
121 | unsigned long bootmap_pages; | |
122 | unsigned long bootmap_start_pfn; | |
123 | unsigned long bootmap_pfn; | |
124 | #ifndef CONFIG_DISCONTIGMEM | |
125 | physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1]; | |
126 | int npmem_holes; | |
127 | #endif | |
128 | int i, sysram_resource_count; | |
129 | ||
130 | disable_sr_hashing(); /* Turn off space register hashing */ | |
131 | ||
132 | /* | |
133 | * Sort the ranges. Since the number of ranges is typically | |
134 | * small, and performance is not an issue here, just do | |
135 | * a simple insertion sort. | |
136 | */ | |
137 | ||
138 | for (i = 1; i < npmem_ranges; i++) { | |
139 | int j; | |
140 | ||
141 | for (j = i; j > 0; j--) { | |
142 | unsigned long tmp; | |
143 | ||
144 | if (pmem_ranges[j-1].start_pfn < | |
145 | pmem_ranges[j].start_pfn) { | |
146 | ||
147 | break; | |
148 | } | |
149 | tmp = pmem_ranges[j-1].start_pfn; | |
150 | pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn; | |
151 | pmem_ranges[j].start_pfn = tmp; | |
152 | tmp = pmem_ranges[j-1].pages; | |
153 | pmem_ranges[j-1].pages = pmem_ranges[j].pages; | |
154 | pmem_ranges[j].pages = tmp; | |
155 | } | |
156 | } | |
157 | ||
158 | #ifndef CONFIG_DISCONTIGMEM | |
159 | /* | |
160 | * Throw out ranges that are too far apart (controlled by | |
161 | * MAX_GAP). | |
162 | */ | |
163 | ||
164 | for (i = 1; i < npmem_ranges; i++) { | |
165 | if (pmem_ranges[i].start_pfn - | |
166 | (pmem_ranges[i-1].start_pfn + | |
167 | pmem_ranges[i-1].pages) > MAX_GAP) { | |
168 | npmem_ranges = i; | |
169 | printk("Large gap in memory detected (%ld pages). " | |
170 | "Consider turning on CONFIG_DISCONTIGMEM\n", | |
171 | pmem_ranges[i].start_pfn - | |
172 | (pmem_ranges[i-1].start_pfn + | |
173 | pmem_ranges[i-1].pages)); | |
174 | break; | |
175 | } | |
176 | } | |
177 | #endif | |
178 | ||
179 | if (npmem_ranges > 1) { | |
180 | ||
181 | /* Print the memory ranges */ | |
182 | ||
183 | printk(KERN_INFO "Memory Ranges:\n"); | |
184 | ||
185 | for (i = 0; i < npmem_ranges; i++) { | |
186 | unsigned long start; | |
187 | unsigned long size; | |
188 | ||
189 | size = (pmem_ranges[i].pages << PAGE_SHIFT); | |
190 | start = (pmem_ranges[i].start_pfn << PAGE_SHIFT); | |
191 | printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n", | |
192 | i,start, start + (size - 1), size >> 20); | |
193 | } | |
194 | } | |
195 | ||
196 | sysram_resource_count = npmem_ranges; | |
197 | for (i = 0; i < sysram_resource_count; i++) { | |
198 | struct resource *res = &sysram_resources[i]; | |
199 | res->name = "System RAM"; | |
200 | res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT; | |
201 | res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1; | |
202 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; | |
203 | request_resource(&iomem_resource, res); | |
204 | } | |
205 | ||
206 | /* | |
207 | * For 32 bit kernels we limit the amount of memory we can | |
208 | * support, in order to preserve enough kernel address space | |
209 | * for other purposes. For 64 bit kernels we don't normally | |
210 | * limit the memory, but this mechanism can be used to | |
211 | * artificially limit the amount of memory (and it is written | |
212 | * to work with multiple memory ranges). | |
213 | */ | |
214 | ||
215 | mem_limit_func(); /* check for "mem=" argument */ | |
216 | ||
217 | mem_max = 0; | |
1da177e4 LT |
218 | for (i = 0; i < npmem_ranges; i++) { |
219 | unsigned long rsize; | |
220 | ||
221 | rsize = pmem_ranges[i].pages << PAGE_SHIFT; | |
222 | if ((mem_max + rsize) > mem_limit) { | |
223 | printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20); | |
224 | if (mem_max == mem_limit) | |
225 | npmem_ranges = i; | |
226 | else { | |
227 | pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT) | |
228 | - (mem_max >> PAGE_SHIFT); | |
229 | npmem_ranges = i + 1; | |
230 | mem_max = mem_limit; | |
231 | } | |
1da177e4 LT |
232 | break; |
233 | } | |
1da177e4 LT |
234 | mem_max += rsize; |
235 | } | |
236 | ||
237 | printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20); | |
238 | ||
239 | #ifndef CONFIG_DISCONTIGMEM | |
240 | /* Merge the ranges, keeping track of the holes */ | |
241 | ||
242 | { | |
243 | unsigned long end_pfn; | |
244 | unsigned long hole_pages; | |
245 | ||
246 | npmem_holes = 0; | |
247 | end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages; | |
248 | for (i = 1; i < npmem_ranges; i++) { | |
249 | ||
250 | hole_pages = pmem_ranges[i].start_pfn - end_pfn; | |
251 | if (hole_pages) { | |
252 | pmem_holes[npmem_holes].start_pfn = end_pfn; | |
253 | pmem_holes[npmem_holes++].pages = hole_pages; | |
254 | end_pfn += hole_pages; | |
255 | } | |
256 | end_pfn += pmem_ranges[i].pages; | |
257 | } | |
258 | ||
259 | pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn; | |
260 | npmem_ranges = 1; | |
261 | } | |
262 | #endif | |
263 | ||
264 | bootmap_pages = 0; | |
265 | for (i = 0; i < npmem_ranges; i++) | |
266 | bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages); | |
267 | ||
268 | bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT; | |
269 | ||
270 | #ifdef CONFIG_DISCONTIGMEM | |
271 | for (i = 0; i < MAX_PHYSMEM_RANGES; i++) { | |
272 | memset(NODE_DATA(i), 0, sizeof(pg_data_t)); | |
b61bfa3c | 273 | NODE_DATA(i)->bdata = &bootmem_node_data[i]; |
1da177e4 LT |
274 | } |
275 | memset(pfnnid_map, 0xff, sizeof(pfnnid_map)); | |
276 | ||
d9b41e0b DR |
277 | for (i = 0; i < npmem_ranges; i++) { |
278 | node_set_state(i, N_NORMAL_MEMORY); | |
1da177e4 | 279 | node_set_online(i); |
d9b41e0b | 280 | } |
1da177e4 LT |
281 | #endif |
282 | ||
283 | /* | |
284 | * Initialize and free the full range of memory in each range. | |
285 | * Note that the only writing these routines do are to the bootmap, | |
286 | * and we've made sure to locate the bootmap properly so that they | |
287 | * won't be writing over anything important. | |
288 | */ | |
289 | ||
290 | bootmap_pfn = bootmap_start_pfn; | |
291 | max_pfn = 0; | |
292 | for (i = 0; i < npmem_ranges; i++) { | |
293 | unsigned long start_pfn; | |
294 | unsigned long npages; | |
295 | ||
296 | start_pfn = pmem_ranges[i].start_pfn; | |
297 | npages = pmem_ranges[i].pages; | |
298 | ||
299 | bootmap_size = init_bootmem_node(NODE_DATA(i), | |
300 | bootmap_pfn, | |
301 | start_pfn, | |
302 | (start_pfn + npages) ); | |
303 | free_bootmem_node(NODE_DATA(i), | |
304 | (start_pfn << PAGE_SHIFT), | |
305 | (npages << PAGE_SHIFT) ); | |
306 | bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
307 | if ((start_pfn + npages) > max_pfn) | |
308 | max_pfn = start_pfn + npages; | |
309 | } | |
310 | ||
5cdb8205 GG |
311 | /* IOMMU is always used to access "high mem" on those boxes |
312 | * that can support enough mem that a PCI device couldn't | |
313 | * directly DMA to any physical addresses. | |
314 | * ISA DMA support will need to revisit this. | |
315 | */ | |
316 | max_low_pfn = max_pfn; | |
317 | ||
8980a7ba HD |
318 | /* bootmap sizing messed up? */ |
319 | BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages); | |
1da177e4 LT |
320 | |
321 | /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */ | |
322 | ||
323 | #define PDC_CONSOLE_IO_IODC_SIZE 32768 | |
324 | ||
325 | reserve_bootmem_node(NODE_DATA(0), 0UL, | |
72a7fe39 BW |
326 | (unsigned long)(PAGE0->mem_free + |
327 | PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT); | |
161bd3bf HD |
328 | reserve_bootmem_node(NODE_DATA(0), __pa(KERNEL_BINARY_TEXT_START), |
329 | (unsigned long)(_end - KERNEL_BINARY_TEXT_START), | |
330 | BOOTMEM_DEFAULT); | |
1da177e4 | 331 | reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT), |
72a7fe39 BW |
332 | ((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT), |
333 | BOOTMEM_DEFAULT); | |
1da177e4 LT |
334 | |
335 | #ifndef CONFIG_DISCONTIGMEM | |
336 | ||
337 | /* reserve the holes */ | |
338 | ||
339 | for (i = 0; i < npmem_holes; i++) { | |
340 | reserve_bootmem_node(NODE_DATA(0), | |
341 | (pmem_holes[i].start_pfn << PAGE_SHIFT), | |
72a7fe39 BW |
342 | (pmem_holes[i].pages << PAGE_SHIFT), |
343 | BOOTMEM_DEFAULT); | |
1da177e4 LT |
344 | } |
345 | #endif | |
346 | ||
347 | #ifdef CONFIG_BLK_DEV_INITRD | |
348 | if (initrd_start) { | |
349 | printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end); | |
350 | if (__pa(initrd_start) < mem_max) { | |
351 | unsigned long initrd_reserve; | |
352 | ||
353 | if (__pa(initrd_end) > mem_max) { | |
354 | initrd_reserve = mem_max - __pa(initrd_start); | |
355 | } else { | |
356 | initrd_reserve = initrd_end - initrd_start; | |
357 | } | |
358 | initrd_below_start_ok = 1; | |
359 | printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max); | |
360 | ||
72a7fe39 BW |
361 | reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start), |
362 | initrd_reserve, BOOTMEM_DEFAULT); | |
1da177e4 LT |
363 | } |
364 | } | |
365 | #endif | |
366 | ||
367 | data_resource.start = virt_to_phys(&data_start); | |
c51d476a KM |
368 | data_resource.end = virt_to_phys(_end) - 1; |
369 | code_resource.start = virt_to_phys(_text); | |
1da177e4 LT |
370 | code_resource.end = virt_to_phys(&data_start)-1; |
371 | ||
372 | /* We don't know which region the kernel will be in, so try | |
373 | * all of them. | |
374 | */ | |
375 | for (i = 0; i < sysram_resource_count; i++) { | |
376 | struct resource *res = &sysram_resources[i]; | |
377 | request_resource(res, &code_resource); | |
378 | request_resource(res, &data_resource); | |
379 | } | |
380 | request_resource(&sysram_resources[0], &pdcdata_resource); | |
381 | } | |
382 | ||
161bd3bf HD |
383 | static int __init parisc_text_address(unsigned long vaddr) |
384 | { | |
385 | static unsigned long head_ptr __initdata; | |
386 | ||
387 | if (!head_ptr) | |
388 | head_ptr = PAGE_MASK & (unsigned long) | |
389 | dereference_function_descriptor(&parisc_kernel_start); | |
390 | ||
391 | return core_kernel_text(vaddr) || vaddr == head_ptr; | |
392 | } | |
393 | ||
d7dd2ff1 JB |
394 | static void __init map_pages(unsigned long start_vaddr, |
395 | unsigned long start_paddr, unsigned long size, | |
396 | pgprot_t pgprot, int force) | |
397 | { | |
398 | pgd_t *pg_dir; | |
399 | pmd_t *pmd; | |
400 | pte_t *pg_table; | |
401 | unsigned long end_paddr; | |
402 | unsigned long start_pmd; | |
403 | unsigned long start_pte; | |
404 | unsigned long tmp1; | |
405 | unsigned long tmp2; | |
406 | unsigned long address; | |
407 | unsigned long vaddr; | |
408 | unsigned long ro_start; | |
409 | unsigned long ro_end; | |
410 | unsigned long fv_addr; | |
411 | unsigned long gw_addr; | |
412 | extern const unsigned long fault_vector_20; | |
413 | extern void * const linux_gateway_page; | |
414 | ||
415 | ro_start = __pa((unsigned long)_text); | |
416 | ro_end = __pa((unsigned long)&data_start); | |
417 | fv_addr = __pa((unsigned long)&fault_vector_20) & PAGE_MASK; | |
418 | gw_addr = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK; | |
419 | ||
420 | end_paddr = start_paddr + size; | |
421 | ||
422 | pg_dir = pgd_offset_k(start_vaddr); | |
423 | ||
424 | #if PTRS_PER_PMD == 1 | |
425 | start_pmd = 0; | |
426 | #else | |
427 | start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1)); | |
428 | #endif | |
429 | start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)); | |
430 | ||
431 | address = start_paddr; | |
432 | vaddr = start_vaddr; | |
433 | while (address < end_paddr) { | |
434 | #if PTRS_PER_PMD == 1 | |
435 | pmd = (pmd_t *)__pa(pg_dir); | |
436 | #else | |
437 | pmd = (pmd_t *)pgd_address(*pg_dir); | |
438 | ||
439 | /* | |
440 | * pmd is physical at this point | |
441 | */ | |
442 | ||
443 | if (!pmd) { | |
444 | pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE << PMD_ORDER); | |
445 | pmd = (pmd_t *) __pa(pmd); | |
446 | } | |
447 | ||
448 | pgd_populate(NULL, pg_dir, __va(pmd)); | |
449 | #endif | |
450 | pg_dir++; | |
451 | ||
452 | /* now change pmd to kernel virtual addresses */ | |
453 | ||
454 | pmd = (pmd_t *)__va(pmd) + start_pmd; | |
455 | for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) { | |
456 | ||
457 | /* | |
458 | * pg_table is physical at this point | |
459 | */ | |
460 | ||
461 | pg_table = (pte_t *)pmd_address(*pmd); | |
462 | if (!pg_table) { | |
463 | pg_table = (pte_t *) | |
464 | alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE); | |
465 | pg_table = (pte_t *) __pa(pg_table); | |
466 | } | |
467 | ||
468 | pmd_populate_kernel(NULL, pmd, __va(pg_table)); | |
469 | ||
470 | /* now change pg_table to kernel virtual addresses */ | |
471 | ||
472 | pg_table = (pte_t *) __va(pg_table) + start_pte; | |
473 | for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) { | |
474 | pte_t pte; | |
475 | ||
476 | /* | |
477 | * Map the fault vector writable so we can | |
478 | * write the HPMC checksum. | |
479 | */ | |
480 | if (force) | |
481 | pte = __mk_pte(address, pgprot); | |
161bd3bf | 482 | else if (parisc_text_address(vaddr) && |
d7dd2ff1 JB |
483 | address != fv_addr) |
484 | pte = __mk_pte(address, PAGE_KERNEL_EXEC); | |
485 | else | |
486 | #if defined(CONFIG_PARISC_PAGE_SIZE_4KB) | |
487 | if (address >= ro_start && address < ro_end | |
488 | && address != fv_addr | |
489 | && address != gw_addr) | |
490 | pte = __mk_pte(address, PAGE_KERNEL_RO); | |
491 | else | |
492 | #endif | |
493 | pte = __mk_pte(address, pgprot); | |
494 | ||
495 | if (address >= end_paddr) { | |
496 | if (force) | |
497 | break; | |
498 | else | |
499 | pte_val(pte) = 0; | |
500 | } | |
501 | ||
502 | set_pte(pg_table, pte); | |
503 | ||
504 | address += PAGE_SIZE; | |
505 | vaddr += PAGE_SIZE; | |
506 | } | |
507 | start_pte = 0; | |
508 | ||
509 | if (address >= end_paddr) | |
510 | break; | |
511 | } | |
512 | start_pmd = 0; | |
513 | } | |
514 | } | |
515 | ||
1da177e4 LT |
516 | void free_initmem(void) |
517 | { | |
4fb11781 KM |
518 | unsigned long init_begin = (unsigned long)__init_begin; |
519 | unsigned long init_end = (unsigned long)__init_end; | |
1da177e4 | 520 | |
d7dd2ff1 JB |
521 | /* The init text pages are marked R-X. We have to |
522 | * flush the icache and mark them RW- | |
523 | * | |
524 | * This is tricky, because map_pages is in the init section. | |
525 | * Do a dummy remap of the data section first (the data | |
526 | * section is already PAGE_KERNEL) to pull in the TLB entries | |
527 | * for map_kernel */ | |
528 | map_pages(init_begin, __pa(init_begin), init_end - init_begin, | |
529 | PAGE_KERNEL_RWX, 1); | |
530 | /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute | |
531 | * map_pages */ | |
532 | map_pages(init_begin, __pa(init_begin), init_end - init_begin, | |
533 | PAGE_KERNEL, 1); | |
534 | ||
535 | /* force the kernel to see the new TLB entries */ | |
536 | __flush_tlb_range(0, init_begin, init_end); | |
1da177e4 LT |
537 | /* Attempt to catch anyone trying to execute code here |
538 | * by filling the page with BRK insns. | |
1da177e4 | 539 | */ |
20dbc9f7 | 540 | memset((void *)init_begin, 0x00, init_end - init_begin); |
d7dd2ff1 JB |
541 | /* finally dump all the instructions which were cached, since the |
542 | * pages are no-longer executable */ | |
4fb11781 | 543 | flush_icache_range(init_begin, init_end); |
1da177e4 | 544 | |
7d2c7747 | 545 | free_initmem_default(-1); |
1da177e4 LT |
546 | |
547 | /* set up a new led state on systems shipped LED State panel */ | |
548 | pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE); | |
1da177e4 LT |
549 | } |
550 | ||
1bcdd854 HD |
551 | |
552 | #ifdef CONFIG_DEBUG_RODATA | |
553 | void mark_rodata_ro(void) | |
554 | { | |
1bcdd854 HD |
555 | /* rodata memory was already mapped with KERNEL_RO access rights by |
556 | pagetable_init() and map_pages(). No need to do additional stuff here */ | |
557 | printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n", | |
a581c2a4 | 558 | (unsigned long)(__end_rodata - __start_rodata) >> 10); |
1bcdd854 HD |
559 | } |
560 | #endif | |
561 | ||
562 | ||
1da177e4 LT |
563 | /* |
564 | * Just an arbitrary offset to serve as a "hole" between mapping areas | |
565 | * (between top of physical memory and a potential pcxl dma mapping | |
566 | * area, and below the vmalloc mapping area). | |
567 | * | |
568 | * The current 32K value just means that there will be a 32K "hole" | |
569 | * between mapping areas. That means that any out-of-bounds memory | |
570 | * accesses will hopefully be caught. The vmalloc() routines leaves | |
571 | * a hole of 4kB between each vmalloced area for the same reason. | |
572 | */ | |
573 | ||
574 | /* Leave room for gateway page expansion */ | |
575 | #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE | |
576 | #error KERNEL_MAP_START is in gateway reserved region | |
577 | #endif | |
578 | #define MAP_START (KERNEL_MAP_START) | |
579 | ||
580 | #define VM_MAP_OFFSET (32*1024) | |
581 | #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \ | |
582 | & ~(VM_MAP_OFFSET-1))) | |
583 | ||
4255f0d2 HD |
584 | void *parisc_vmalloc_start __read_mostly; |
585 | EXPORT_SYMBOL(parisc_vmalloc_start); | |
1da177e4 LT |
586 | |
587 | #ifdef CONFIG_PA11 | |
8039de10 | 588 | unsigned long pcxl_dma_start __read_mostly; |
1da177e4 LT |
589 | #endif |
590 | ||
591 | void __init mem_init(void) | |
592 | { | |
48d27cb2 HD |
593 | /* Do sanity checks on page table constants */ |
594 | BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t)); | |
595 | BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t)); | |
596 | BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t)); | |
597 | BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD | |
598 | > BITS_PER_LONG); | |
599 | ||
1da177e4 | 600 | high_memory = __va((max_pfn << PAGE_SHIFT)); |
d5c017dd | 601 | set_max_mapnr(page_to_pfn(virt_to_page(high_memory - 1)) + 1); |
0c988534 | 602 | free_all_bootmem(); |
1da177e4 | 603 | |
1da177e4 LT |
604 | #ifdef CONFIG_PA11 |
605 | if (hppa_dma_ops == &pcxl_dma_ops) { | |
606 | pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START); | |
4255f0d2 HD |
607 | parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start |
608 | + PCXL_DMA_MAP_SIZE); | |
1da177e4 LT |
609 | } else { |
610 | pcxl_dma_start = 0; | |
4255f0d2 | 611 | parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START); |
1da177e4 LT |
612 | } |
613 | #else | |
4255f0d2 | 614 | parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START); |
1da177e4 LT |
615 | #endif |
616 | ||
7d2c7747 | 617 | mem_init_print_info(NULL); |
ce8420bb HD |
618 | #ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */ |
619 | printk("virtual kernel memory layout:\n" | |
620 | " vmalloc : 0x%p - 0x%p (%4ld MB)\n" | |
53faf291 | 621 | " memory : 0x%p - 0x%p (%4ld MB)\n" |
ce8420bb HD |
622 | " .init : 0x%p - 0x%p (%4ld kB)\n" |
623 | " .data : 0x%p - 0x%p (%4ld kB)\n" | |
624 | " .text : 0x%p - 0x%p (%4ld kB)\n", | |
625 | ||
626 | (void*)VMALLOC_START, (void*)VMALLOC_END, | |
627 | (VMALLOC_END - VMALLOC_START) >> 20, | |
628 | ||
629 | __va(0), high_memory, | |
630 | ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, | |
631 | ||
53faf291 KM |
632 | __init_begin, __init_end, |
633 | ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10, | |
ce8420bb | 634 | |
53faf291 KM |
635 | _etext, _edata, |
636 | ((unsigned long)_edata - (unsigned long)_etext) >> 10, | |
ce8420bb | 637 | |
53faf291 KM |
638 | _text, _etext, |
639 | ((unsigned long)_etext - (unsigned long)_text) >> 10); | |
ce8420bb | 640 | #endif |
1da177e4 LT |
641 | } |
642 | ||
8039de10 | 643 | unsigned long *empty_zero_page __read_mostly; |
22febf1f | 644 | EXPORT_SYMBOL(empty_zero_page); |
1da177e4 | 645 | |
b2b755b5 | 646 | void show_mem(unsigned int filter) |
1da177e4 | 647 | { |
aec6a888 MG |
648 | int total = 0,reserved = 0; |
649 | pg_data_t *pgdat; | |
1da177e4 LT |
650 | |
651 | printk(KERN_INFO "Mem-info:\n"); | |
7bf02ea2 | 652 | show_free_areas(filter); |
1da177e4 | 653 | |
aec6a888 MG |
654 | for_each_online_pgdat(pgdat) { |
655 | unsigned long flags; | |
656 | int zoneid; | |
657 | ||
658 | pgdat_resize_lock(pgdat, &flags); | |
659 | for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) { | |
660 | struct zone *zone = &pgdat->node_zones[zoneid]; | |
661 | if (!populated_zone(zone)) | |
662 | continue; | |
663 | ||
664 | total += zone->present_pages; | |
665 | reserved = zone->present_pages - zone->managed_pages; | |
666 | } | |
667 | pgdat_resize_unlock(pgdat, &flags); | |
1da177e4 | 668 | } |
aec6a888 | 669 | |
1da177e4 LT |
670 | printk(KERN_INFO "%d pages of RAM\n", total); |
671 | printk(KERN_INFO "%d reserved pages\n", reserved); | |
1da177e4 LT |
672 | |
673 | #ifdef CONFIG_DISCONTIGMEM | |
674 | { | |
675 | struct zonelist *zl; | |
54a6eb5c | 676 | int i, j; |
1da177e4 LT |
677 | |
678 | for (i = 0; i < npmem_ranges; i++) { | |
4413a0f6 | 679 | zl = node_zonelist(i, 0); |
1da177e4 | 680 | for (j = 0; j < MAX_NR_ZONES; j++) { |
dd1a239f | 681 | struct zoneref *z; |
54a6eb5c | 682 | struct zone *zone; |
1da177e4 LT |
683 | |
684 | printk("Zone list for zone %d on node %d: ", j, i); | |
54a6eb5c MG |
685 | for_each_zone_zonelist(zone, z, zl, j) |
686 | printk("[%d/%s] ", zone_to_nid(zone), | |
687 | zone->name); | |
1da177e4 LT |
688 | printk("\n"); |
689 | } | |
690 | } | |
691 | } | |
692 | #endif | |
693 | } | |
694 | ||
1da177e4 LT |
695 | /* |
696 | * pagetable_init() sets up the page tables | |
697 | * | |
698 | * Note that gateway_init() places the Linux gateway page at page 0. | |
699 | * Since gateway pages cannot be dereferenced this has the desirable | |
700 | * side effect of trapping those pesky NULL-reference errors in the | |
701 | * kernel. | |
702 | */ | |
703 | static void __init pagetable_init(void) | |
704 | { | |
705 | int range; | |
706 | ||
707 | /* Map each physical memory range to its kernel vaddr */ | |
708 | ||
709 | for (range = 0; range < npmem_ranges; range++) { | |
710 | unsigned long start_paddr; | |
711 | unsigned long end_paddr; | |
712 | unsigned long size; | |
713 | ||
714 | start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT; | |
715 | end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT); | |
716 | size = pmem_ranges[range].pages << PAGE_SHIFT; | |
717 | ||
718 | map_pages((unsigned long)__va(start_paddr), start_paddr, | |
d7dd2ff1 | 719 | size, PAGE_KERNEL, 0); |
1da177e4 LT |
720 | } |
721 | ||
722 | #ifdef CONFIG_BLK_DEV_INITRD | |
723 | if (initrd_end && initrd_end > mem_limit) { | |
1bcdd854 | 724 | printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end); |
1da177e4 | 725 | map_pages(initrd_start, __pa(initrd_start), |
d7dd2ff1 | 726 | initrd_end - initrd_start, PAGE_KERNEL, 0); |
1da177e4 LT |
727 | } |
728 | #endif | |
729 | ||
730 | empty_zero_page = alloc_bootmem_pages(PAGE_SIZE); | |
1da177e4 LT |
731 | } |
732 | ||
733 | static void __init gateway_init(void) | |
734 | { | |
735 | unsigned long linux_gateway_page_addr; | |
736 | /* FIXME: This is 'const' in order to trick the compiler | |
737 | into not treating it as DP-relative data. */ | |
738 | extern void * const linux_gateway_page; | |
739 | ||
740 | linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK; | |
741 | ||
742 | /* | |
743 | * Setup Linux Gateway page. | |
744 | * | |
745 | * The Linux gateway page will reside in kernel space (on virtual | |
746 | * page 0), so it doesn't need to be aliased into user space. | |
747 | */ | |
748 | ||
749 | map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page), | |
d7dd2ff1 | 750 | PAGE_SIZE, PAGE_GATEWAY, 1); |
1da177e4 LT |
751 | } |
752 | ||
1da177e4 LT |
753 | void __init paging_init(void) |
754 | { | |
755 | int i; | |
756 | ||
757 | setup_bootmem(); | |
758 | pagetable_init(); | |
759 | gateway_init(); | |
760 | flush_cache_all_local(); /* start with known state */ | |
ce33941f | 761 | flush_tlb_all_local(NULL); |
1da177e4 LT |
762 | |
763 | for (i = 0; i < npmem_ranges; i++) { | |
f06a9684 | 764 | unsigned long zones_size[MAX_NR_ZONES] = { 0, }; |
1da177e4 | 765 | |
00592837 | 766 | zones_size[ZONE_NORMAL] = pmem_ranges[i].pages; |
1da177e4 LT |
767 | |
768 | #ifdef CONFIG_DISCONTIGMEM | |
769 | /* Need to initialize the pfnnid_map before we can initialize | |
770 | the zone */ | |
771 | { | |
772 | int j; | |
773 | for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT); | |
774 | j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT); | |
775 | j++) { | |
776 | pfnnid_map[j] = i; | |
777 | } | |
778 | } | |
779 | #endif | |
780 | ||
9109fb7b | 781 | free_area_init_node(i, zones_size, |
1da177e4 LT |
782 | pmem_ranges[i].start_pfn, NULL); |
783 | } | |
784 | } | |
785 | ||
786 | #ifdef CONFIG_PA20 | |
787 | ||
788 | /* | |
7022672e | 789 | * Currently, all PA20 chips have 18 bit protection IDs, which is the |
1da177e4 LT |
790 | * limiting factor (space ids are 32 bits). |
791 | */ | |
792 | ||
793 | #define NR_SPACE_IDS 262144 | |
794 | ||
795 | #else | |
796 | ||
797 | /* | |
7022672e SA |
798 | * Currently we have a one-to-one relationship between space IDs and |
799 | * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only | |
800 | * support 15 bit protection IDs, so that is the limiting factor. | |
801 | * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's | |
1da177e4 LT |
802 | * probably not worth the effort for a special case here. |
803 | */ | |
804 | ||
805 | #define NR_SPACE_IDS 32768 | |
806 | ||
807 | #endif /* !CONFIG_PA20 */ | |
808 | ||
809 | #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2) | |
810 | #define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long))) | |
811 | ||
812 | static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */ | |
813 | static unsigned long dirty_space_id[SID_ARRAY_SIZE]; | |
814 | static unsigned long space_id_index; | |
815 | static unsigned long free_space_ids = NR_SPACE_IDS - 1; | |
816 | static unsigned long dirty_space_ids = 0; | |
817 | ||
818 | static DEFINE_SPINLOCK(sid_lock); | |
819 | ||
820 | unsigned long alloc_sid(void) | |
821 | { | |
822 | unsigned long index; | |
823 | ||
824 | spin_lock(&sid_lock); | |
825 | ||
826 | if (free_space_ids == 0) { | |
827 | if (dirty_space_ids != 0) { | |
828 | spin_unlock(&sid_lock); | |
829 | flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */ | |
830 | spin_lock(&sid_lock); | |
831 | } | |
2fd83038 | 832 | BUG_ON(free_space_ids == 0); |
1da177e4 LT |
833 | } |
834 | ||
835 | free_space_ids--; | |
836 | ||
837 | index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index); | |
838 | space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1))); | |
839 | space_id_index = index; | |
840 | ||
841 | spin_unlock(&sid_lock); | |
842 | ||
843 | return index << SPACEID_SHIFT; | |
844 | } | |
845 | ||
846 | void free_sid(unsigned long spaceid) | |
847 | { | |
848 | unsigned long index = spaceid >> SPACEID_SHIFT; | |
849 | unsigned long *dirty_space_offset; | |
850 | ||
851 | dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG); | |
852 | index &= (BITS_PER_LONG - 1); | |
853 | ||
854 | spin_lock(&sid_lock); | |
855 | ||
2fd83038 | 856 | BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */ |
1da177e4 LT |
857 | |
858 | *dirty_space_offset |= (1L << index); | |
859 | dirty_space_ids++; | |
860 | ||
861 | spin_unlock(&sid_lock); | |
862 | } | |
863 | ||
864 | ||
865 | #ifdef CONFIG_SMP | |
866 | static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array) | |
867 | { | |
868 | int i; | |
869 | ||
870 | /* NOTE: sid_lock must be held upon entry */ | |
871 | ||
872 | *ndirtyptr = dirty_space_ids; | |
873 | if (dirty_space_ids != 0) { | |
874 | for (i = 0; i < SID_ARRAY_SIZE; i++) { | |
875 | dirty_array[i] = dirty_space_id[i]; | |
876 | dirty_space_id[i] = 0; | |
877 | } | |
878 | dirty_space_ids = 0; | |
879 | } | |
880 | ||
881 | return; | |
882 | } | |
883 | ||
884 | static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array) | |
885 | { | |
886 | int i; | |
887 | ||
888 | /* NOTE: sid_lock must be held upon entry */ | |
889 | ||
890 | if (ndirty != 0) { | |
891 | for (i = 0; i < SID_ARRAY_SIZE; i++) { | |
892 | space_id[i] ^= dirty_array[i]; | |
893 | } | |
894 | ||
895 | free_space_ids += ndirty; | |
896 | space_id_index = 0; | |
897 | } | |
898 | } | |
899 | ||
900 | #else /* CONFIG_SMP */ | |
901 | ||
902 | static void recycle_sids(void) | |
903 | { | |
904 | int i; | |
905 | ||
906 | /* NOTE: sid_lock must be held upon entry */ | |
907 | ||
908 | if (dirty_space_ids != 0) { | |
909 | for (i = 0; i < SID_ARRAY_SIZE; i++) { | |
910 | space_id[i] ^= dirty_space_id[i]; | |
911 | dirty_space_id[i] = 0; | |
912 | } | |
913 | ||
914 | free_space_ids += dirty_space_ids; | |
915 | dirty_space_ids = 0; | |
916 | space_id_index = 0; | |
917 | } | |
918 | } | |
919 | #endif | |
920 | ||
921 | /* | |
922 | * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is | |
923 | * purged, we can safely reuse the space ids that were released but | |
924 | * not flushed from the tlb. | |
925 | */ | |
926 | ||
927 | #ifdef CONFIG_SMP | |
928 | ||
929 | static unsigned long recycle_ndirty; | |
930 | static unsigned long recycle_dirty_array[SID_ARRAY_SIZE]; | |
2fd83038 | 931 | static unsigned int recycle_inuse; |
1da177e4 LT |
932 | |
933 | void flush_tlb_all(void) | |
934 | { | |
935 | int do_recycle; | |
936 | ||
416821d3 | 937 | __inc_irq_stat(irq_tlb_count); |
1da177e4 LT |
938 | do_recycle = 0; |
939 | spin_lock(&sid_lock); | |
940 | if (dirty_space_ids > RECYCLE_THRESHOLD) { | |
2fd83038 | 941 | BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */ |
1da177e4 LT |
942 | get_dirty_sids(&recycle_ndirty,recycle_dirty_array); |
943 | recycle_inuse++; | |
944 | do_recycle++; | |
945 | } | |
946 | spin_unlock(&sid_lock); | |
15c8b6c1 | 947 | on_each_cpu(flush_tlb_all_local, NULL, 1); |
1da177e4 LT |
948 | if (do_recycle) { |
949 | spin_lock(&sid_lock); | |
950 | recycle_sids(recycle_ndirty,recycle_dirty_array); | |
951 | recycle_inuse = 0; | |
952 | spin_unlock(&sid_lock); | |
953 | } | |
954 | } | |
955 | #else | |
956 | void flush_tlb_all(void) | |
957 | { | |
416821d3 | 958 | __inc_irq_stat(irq_tlb_count); |
1da177e4 | 959 | spin_lock(&sid_lock); |
1b2425e3 | 960 | flush_tlb_all_local(NULL); |
1da177e4 LT |
961 | recycle_sids(); |
962 | spin_unlock(&sid_lock); | |
963 | } | |
964 | #endif | |
965 | ||
966 | #ifdef CONFIG_BLK_DEV_INITRD | |
967 | void free_initrd_mem(unsigned long start, unsigned long end) | |
968 | { | |
7d2c7747 | 969 | free_reserved_area((void *)start, (void *)end, -1, "initrd"); |
1da177e4 LT |
970 | } |
971 | #endif |