xref: /linux/arch/parisc/mm/init.c (revision c4c11dd160a8cc98f402c4e12f94b1572e822ffd)
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)
9  *  Copyright 2006-2007 Helge Deller (deller@gmx.de)
10  *
11  */
12 
13 
14 #include <linux/module.h>
15 #include <linux/mm.h>
16 #include <linux/bootmem.h>
17 #include <linux/gfp.h>
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>
28 #include <asm/pgtable.h>
29 #include <asm/tlb.h>
30 #include <asm/pdc_chassis.h>
31 #include <asm/mmzone.h>
32 #include <asm/sections.h>
33 
34 extern int  data_start;
35 
36 #if PT_NLEVELS == 3
37 /* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout
38  * with the first pmd adjacent to the pgd and below it. gcc doesn't actually
39  * guarantee that global objects will be laid out in memory in the same order
40  * as the order of declaration, so put these in different sections and use
41  * the linker script to order them. */
42 pmd_t pmd0[PTRS_PER_PMD] __attribute__ ((__section__ (".data..vm0.pmd"), aligned(PAGE_SIZE)));
43 #endif
44 
45 pgd_t swapper_pg_dir[PTRS_PER_PGD] __attribute__ ((__section__ (".data..vm0.pgd"), aligned(PAGE_SIZE)));
46 pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __attribute__ ((__section__ (".data..vm0.pte"), aligned(PAGE_SIZE)));
47 
48 #ifdef CONFIG_DISCONTIGMEM
49 struct node_map_data node_data[MAX_NUMNODES] __read_mostly;
50 signed char pfnnid_map[PFNNID_MAP_MAX] __read_mostly;
51 #endif
52 
53 static struct resource data_resource = {
54 	.name	= "Kernel data",
55 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM,
56 };
57 
58 static struct resource code_resource = {
59 	.name	= "Kernel code",
60 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM,
61 };
62 
63 static struct resource pdcdata_resource = {
64 	.name	= "PDC data (Page Zero)",
65 	.start	= 0,
66 	.end	= 0x9ff,
67 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM,
68 };
69 
70 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __read_mostly;
71 
72 /* The following array is initialized from the firmware specific
73  * information retrieved in kernel/inventory.c.
74  */
75 
76 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __read_mostly;
77 int npmem_ranges __read_mostly;
78 
79 #ifdef CONFIG_64BIT
80 #define MAX_MEM         (~0UL)
81 #else /* !CONFIG_64BIT */
82 #define MAX_MEM         (3584U*1024U*1024U)
83 #endif /* !CONFIG_64BIT */
84 
85 static unsigned long mem_limit __read_mostly = MAX_MEM;
86 
87 static void __init mem_limit_func(void)
88 {
89 	char *cp, *end;
90 	unsigned long limit;
91 
92 	/* We need this before __setup() functions are called */
93 
94 	limit = MAX_MEM;
95 	for (cp = boot_command_line; *cp; ) {
96 		if (memcmp(cp, "mem=", 4) == 0) {
97 			cp += 4;
98 			limit = memparse(cp, &end);
99 			if (end != cp)
100 				break;
101 			cp = end;
102 		} else {
103 			while (*cp != ' ' && *cp)
104 				++cp;
105 			while (*cp == ' ')
106 				++cp;
107 		}
108 	}
109 
110 	if (limit < mem_limit)
111 		mem_limit = limit;
112 }
113 
114 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
115 
116 static void __init setup_bootmem(void)
117 {
118 	unsigned long bootmap_size;
119 	unsigned long mem_max;
120 	unsigned long bootmap_pages;
121 	unsigned long bootmap_start_pfn;
122 	unsigned long bootmap_pfn;
123 #ifndef CONFIG_DISCONTIGMEM
124 	physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
125 	int npmem_holes;
126 #endif
127 	int i, sysram_resource_count;
128 
129 	disable_sr_hashing(); /* Turn off space register hashing */
130 
131 	/*
132 	 * Sort the ranges. Since the number of ranges is typically
133 	 * small, and performance is not an issue here, just do
134 	 * a simple insertion sort.
135 	 */
136 
137 	for (i = 1; i < npmem_ranges; i++) {
138 		int j;
139 
140 		for (j = i; j > 0; j--) {
141 			unsigned long tmp;
142 
143 			if (pmem_ranges[j-1].start_pfn <
144 			    pmem_ranges[j].start_pfn) {
145 
146 				break;
147 			}
148 			tmp = pmem_ranges[j-1].start_pfn;
149 			pmem_ranges[j-1].start_pfn = pmem_ranges[j].start_pfn;
150 			pmem_ranges[j].start_pfn = tmp;
151 			tmp = pmem_ranges[j-1].pages;
152 			pmem_ranges[j-1].pages = pmem_ranges[j].pages;
153 			pmem_ranges[j].pages = tmp;
154 		}
155 	}
156 
157 #ifndef CONFIG_DISCONTIGMEM
158 	/*
159 	 * Throw out ranges that are too far apart (controlled by
160 	 * MAX_GAP).
161 	 */
162 
163 	for (i = 1; i < npmem_ranges; i++) {
164 		if (pmem_ranges[i].start_pfn -
165 			(pmem_ranges[i-1].start_pfn +
166 			 pmem_ranges[i-1].pages) > MAX_GAP) {
167 			npmem_ranges = i;
168 			printk("Large gap in memory detected (%ld pages). "
169 			       "Consider turning on CONFIG_DISCONTIGMEM\n",
170 			       pmem_ranges[i].start_pfn -
171 			       (pmem_ranges[i-1].start_pfn +
172 			        pmem_ranges[i-1].pages));
173 			break;
174 		}
175 	}
176 #endif
177 
178 	if (npmem_ranges > 1) {
179 
180 		/* Print the memory ranges */
181 
182 		printk(KERN_INFO "Memory Ranges:\n");
183 
184 		for (i = 0; i < npmem_ranges; i++) {
185 			unsigned long start;
186 			unsigned long size;
187 
188 			size = (pmem_ranges[i].pages << PAGE_SHIFT);
189 			start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
190 			printk(KERN_INFO "%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
191 				i,start, start + (size - 1), size >> 20);
192 		}
193 	}
194 
195 	sysram_resource_count = npmem_ranges;
196 	for (i = 0; i < sysram_resource_count; i++) {
197 		struct resource *res = &sysram_resources[i];
198 		res->name = "System RAM";
199 		res->start = pmem_ranges[i].start_pfn << PAGE_SHIFT;
200 		res->end = res->start + (pmem_ranges[i].pages << PAGE_SHIFT)-1;
201 		res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
202 		request_resource(&iomem_resource, res);
203 	}
204 
205 	/*
206 	 * For 32 bit kernels we limit the amount of memory we can
207 	 * support, in order to preserve enough kernel address space
208 	 * for other purposes. For 64 bit kernels we don't normally
209 	 * limit the memory, but this mechanism can be used to
210 	 * artificially limit the amount of memory (and it is written
211 	 * to work with multiple memory ranges).
212 	 */
213 
214 	mem_limit_func();       /* check for "mem=" argument */
215 
216 	mem_max = 0;
217 	for (i = 0; i < npmem_ranges; i++) {
218 		unsigned long rsize;
219 
220 		rsize = pmem_ranges[i].pages << PAGE_SHIFT;
221 		if ((mem_max + rsize) > mem_limit) {
222 			printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
223 			if (mem_max == mem_limit)
224 				npmem_ranges = i;
225 			else {
226 				pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
227 						       - (mem_max >> PAGE_SHIFT);
228 				npmem_ranges = i + 1;
229 				mem_max = mem_limit;
230 			}
231 			break;
232 		}
233 		mem_max += rsize;
234 	}
235 
236 	printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
237 
238 #ifndef CONFIG_DISCONTIGMEM
239 	/* Merge the ranges, keeping track of the holes */
240 
241 	{
242 		unsigned long end_pfn;
243 		unsigned long hole_pages;
244 
245 		npmem_holes = 0;
246 		end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
247 		for (i = 1; i < npmem_ranges; i++) {
248 
249 			hole_pages = pmem_ranges[i].start_pfn - end_pfn;
250 			if (hole_pages) {
251 				pmem_holes[npmem_holes].start_pfn = end_pfn;
252 				pmem_holes[npmem_holes++].pages = hole_pages;
253 				end_pfn += hole_pages;
254 			}
255 			end_pfn += pmem_ranges[i].pages;
256 		}
257 
258 		pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
259 		npmem_ranges = 1;
260 	}
261 #endif
262 
263 	bootmap_pages = 0;
264 	for (i = 0; i < npmem_ranges; i++)
265 		bootmap_pages += bootmem_bootmap_pages(pmem_ranges[i].pages);
266 
267 	bootmap_start_pfn = PAGE_ALIGN(__pa((unsigned long) &_end)) >> PAGE_SHIFT;
268 
269 #ifdef CONFIG_DISCONTIGMEM
270 	for (i = 0; i < MAX_PHYSMEM_RANGES; i++) {
271 		memset(NODE_DATA(i), 0, sizeof(pg_data_t));
272 		NODE_DATA(i)->bdata = &bootmem_node_data[i];
273 	}
274 	memset(pfnnid_map, 0xff, sizeof(pfnnid_map));
275 
276 	for (i = 0; i < npmem_ranges; i++) {
277 		node_set_state(i, N_NORMAL_MEMORY);
278 		node_set_online(i);
279 	}
280 #endif
281 
282 	/*
283 	 * Initialize and free the full range of memory in each range.
284 	 * Note that the only writing these routines do are to the bootmap,
285 	 * and we've made sure to locate the bootmap properly so that they
286 	 * won't be writing over anything important.
287 	 */
288 
289 	bootmap_pfn = bootmap_start_pfn;
290 	max_pfn = 0;
291 	for (i = 0; i < npmem_ranges; i++) {
292 		unsigned long start_pfn;
293 		unsigned long npages;
294 
295 		start_pfn = pmem_ranges[i].start_pfn;
296 		npages = pmem_ranges[i].pages;
297 
298 		bootmap_size = init_bootmem_node(NODE_DATA(i),
299 						bootmap_pfn,
300 						start_pfn,
301 						(start_pfn + npages) );
302 		free_bootmem_node(NODE_DATA(i),
303 				  (start_pfn << PAGE_SHIFT),
304 				  (npages << PAGE_SHIFT) );
305 		bootmap_pfn += (bootmap_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
306 		if ((start_pfn + npages) > max_pfn)
307 			max_pfn = start_pfn + npages;
308 	}
309 
310 	/* IOMMU is always used to access "high mem" on those boxes
311 	 * that can support enough mem that a PCI device couldn't
312 	 * directly DMA to any physical addresses.
313 	 * ISA DMA support will need to revisit this.
314 	 */
315 	max_low_pfn = max_pfn;
316 
317 	/* bootmap sizing messed up? */
318 	BUG_ON((bootmap_pfn - bootmap_start_pfn) != bootmap_pages);
319 
320 	/* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
321 
322 #define PDC_CONSOLE_IO_IODC_SIZE 32768
323 
324 	reserve_bootmem_node(NODE_DATA(0), 0UL,
325 			(unsigned long)(PAGE0->mem_free +
326 				PDC_CONSOLE_IO_IODC_SIZE), BOOTMEM_DEFAULT);
327 	reserve_bootmem_node(NODE_DATA(0), __pa((unsigned long)_text),
328 			(unsigned long)(_end - _text), BOOTMEM_DEFAULT);
329 	reserve_bootmem_node(NODE_DATA(0), (bootmap_start_pfn << PAGE_SHIFT),
330 			((bootmap_pfn - bootmap_start_pfn) << PAGE_SHIFT),
331 			BOOTMEM_DEFAULT);
332 
333 #ifndef CONFIG_DISCONTIGMEM
334 
335 	/* reserve the holes */
336 
337 	for (i = 0; i < npmem_holes; i++) {
338 		reserve_bootmem_node(NODE_DATA(0),
339 				(pmem_holes[i].start_pfn << PAGE_SHIFT),
340 				(pmem_holes[i].pages << PAGE_SHIFT),
341 				BOOTMEM_DEFAULT);
342 	}
343 #endif
344 
345 #ifdef CONFIG_BLK_DEV_INITRD
346 	if (initrd_start) {
347 		printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
348 		if (__pa(initrd_start) < mem_max) {
349 			unsigned long initrd_reserve;
350 
351 			if (__pa(initrd_end) > mem_max) {
352 				initrd_reserve = mem_max - __pa(initrd_start);
353 			} else {
354 				initrd_reserve = initrd_end - initrd_start;
355 			}
356 			initrd_below_start_ok = 1;
357 			printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
358 
359 			reserve_bootmem_node(NODE_DATA(0), __pa(initrd_start),
360 					initrd_reserve, BOOTMEM_DEFAULT);
361 		}
362 	}
363 #endif
364 
365 	data_resource.start =  virt_to_phys(&data_start);
366 	data_resource.end = virt_to_phys(_end) - 1;
367 	code_resource.start = virt_to_phys(_text);
368 	code_resource.end = virt_to_phys(&data_start)-1;
369 
370 	/* We don't know which region the kernel will be in, so try
371 	 * all of them.
372 	 */
373 	for (i = 0; i < sysram_resource_count; i++) {
374 		struct resource *res = &sysram_resources[i];
375 		request_resource(res, &code_resource);
376 		request_resource(res, &data_resource);
377 	}
378 	request_resource(&sysram_resources[0], &pdcdata_resource);
379 }
380 
381 static void __init map_pages(unsigned long start_vaddr,
382 			     unsigned long start_paddr, unsigned long size,
383 			     pgprot_t pgprot, int force)
384 {
385 	pgd_t *pg_dir;
386 	pmd_t *pmd;
387 	pte_t *pg_table;
388 	unsigned long end_paddr;
389 	unsigned long start_pmd;
390 	unsigned long start_pte;
391 	unsigned long tmp1;
392 	unsigned long tmp2;
393 	unsigned long address;
394 	unsigned long vaddr;
395 	unsigned long ro_start;
396 	unsigned long ro_end;
397 	unsigned long fv_addr;
398 	unsigned long gw_addr;
399 	extern const unsigned long fault_vector_20;
400 	extern void * const linux_gateway_page;
401 
402 	ro_start = __pa((unsigned long)_text);
403 	ro_end   = __pa((unsigned long)&data_start);
404 	fv_addr  = __pa((unsigned long)&fault_vector_20) & PAGE_MASK;
405 	gw_addr  = __pa((unsigned long)&linux_gateway_page) & PAGE_MASK;
406 
407 	end_paddr = start_paddr + size;
408 
409 	pg_dir = pgd_offset_k(start_vaddr);
410 
411 #if PTRS_PER_PMD == 1
412 	start_pmd = 0;
413 #else
414 	start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
415 #endif
416 	start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
417 
418 	address = start_paddr;
419 	vaddr = start_vaddr;
420 	while (address < end_paddr) {
421 #if PTRS_PER_PMD == 1
422 		pmd = (pmd_t *)__pa(pg_dir);
423 #else
424 		pmd = (pmd_t *)pgd_address(*pg_dir);
425 
426 		/*
427 		 * pmd is physical at this point
428 		 */
429 
430 		if (!pmd) {
431 			pmd = (pmd_t *) alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE << PMD_ORDER);
432 			pmd = (pmd_t *) __pa(pmd);
433 		}
434 
435 		pgd_populate(NULL, pg_dir, __va(pmd));
436 #endif
437 		pg_dir++;
438 
439 		/* now change pmd to kernel virtual addresses */
440 
441 		pmd = (pmd_t *)__va(pmd) + start_pmd;
442 		for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
443 
444 			/*
445 			 * pg_table is physical at this point
446 			 */
447 
448 			pg_table = (pte_t *)pmd_address(*pmd);
449 			if (!pg_table) {
450 				pg_table = (pte_t *)
451 					alloc_bootmem_low_pages_node(NODE_DATA(0), PAGE_SIZE);
452 				pg_table = (pte_t *) __pa(pg_table);
453 			}
454 
455 			pmd_populate_kernel(NULL, pmd, __va(pg_table));
456 
457 			/* now change pg_table to kernel virtual addresses */
458 
459 			pg_table = (pte_t *) __va(pg_table) + start_pte;
460 			for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
461 				pte_t pte;
462 
463 				/*
464 				 * Map the fault vector writable so we can
465 				 * write the HPMC checksum.
466 				 */
467 				if (force)
468 					pte =  __mk_pte(address, pgprot);
469 				else if (core_kernel_text(vaddr) &&
470 					 address != fv_addr)
471 					pte = __mk_pte(address, PAGE_KERNEL_EXEC);
472 				else
473 #if defined(CONFIG_PARISC_PAGE_SIZE_4KB)
474 				if (address >= ro_start && address < ro_end
475 							&& address != fv_addr
476 							&& address != gw_addr)
477 					pte = __mk_pte(address, PAGE_KERNEL_RO);
478 				else
479 #endif
480 					pte = __mk_pte(address, pgprot);
481 
482 				if (address >= end_paddr) {
483 					if (force)
484 						break;
485 					else
486 						pte_val(pte) = 0;
487 				}
488 
489 				set_pte(pg_table, pte);
490 
491 				address += PAGE_SIZE;
492 				vaddr += PAGE_SIZE;
493 			}
494 			start_pte = 0;
495 
496 			if (address >= end_paddr)
497 			    break;
498 		}
499 		start_pmd = 0;
500 	}
501 }
502 
503 void free_initmem(void)
504 {
505 	unsigned long init_begin = (unsigned long)__init_begin;
506 	unsigned long init_end = (unsigned long)__init_end;
507 
508 	/* The init text pages are marked R-X.  We have to
509 	 * flush the icache and mark them RW-
510 	 *
511 	 * This is tricky, because map_pages is in the init section.
512 	 * Do a dummy remap of the data section first (the data
513 	 * section is already PAGE_KERNEL) to pull in the TLB entries
514 	 * for map_kernel */
515 	map_pages(init_begin, __pa(init_begin), init_end - init_begin,
516 		  PAGE_KERNEL_RWX, 1);
517 	/* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
518 	 * map_pages */
519 	map_pages(init_begin, __pa(init_begin), init_end - init_begin,
520 		  PAGE_KERNEL, 1);
521 
522 	/* force the kernel to see the new TLB entries */
523 	__flush_tlb_range(0, init_begin, init_end);
524 	/* Attempt to catch anyone trying to execute code here
525 	 * by filling the page with BRK insns.
526 	 */
527 	memset((void *)init_begin, 0x00, init_end - init_begin);
528 	/* finally dump all the instructions which were cached, since the
529 	 * pages are no-longer executable */
530 	flush_icache_range(init_begin, init_end);
531 
532 	free_initmem_default(-1);
533 
534 	/* set up a new led state on systems shipped LED State panel */
535 	pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
536 }
537 
538 
539 #ifdef CONFIG_DEBUG_RODATA
540 void mark_rodata_ro(void)
541 {
542 	/* rodata memory was already mapped with KERNEL_RO access rights by
543            pagetable_init() and map_pages(). No need to do additional stuff here */
544 	printk (KERN_INFO "Write protecting the kernel read-only data: %luk\n",
545 		(unsigned long)(__end_rodata - __start_rodata) >> 10);
546 }
547 #endif
548 
549 
550 /*
551  * Just an arbitrary offset to serve as a "hole" between mapping areas
552  * (between top of physical memory and a potential pcxl dma mapping
553  * area, and below the vmalloc mapping area).
554  *
555  * The current 32K value just means that there will be a 32K "hole"
556  * between mapping areas. That means that  any out-of-bounds memory
557  * accesses will hopefully be caught. The vmalloc() routines leaves
558  * a hole of 4kB between each vmalloced area for the same reason.
559  */
560 
561  /* Leave room for gateway page expansion */
562 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
563 #error KERNEL_MAP_START is in gateway reserved region
564 #endif
565 #define MAP_START (KERNEL_MAP_START)
566 
567 #define VM_MAP_OFFSET  (32*1024)
568 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
569 				     & ~(VM_MAP_OFFSET-1)))
570 
571 void *parisc_vmalloc_start __read_mostly;
572 EXPORT_SYMBOL(parisc_vmalloc_start);
573 
574 #ifdef CONFIG_PA11
575 unsigned long pcxl_dma_start __read_mostly;
576 #endif
577 
578 void __init mem_init(void)
579 {
580 	/* Do sanity checks on page table constants */
581 	BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
582 	BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
583 	BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
584 	BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
585 			> BITS_PER_LONG);
586 
587 	high_memory = __va((max_pfn << PAGE_SHIFT));
588 	set_max_mapnr(page_to_pfn(virt_to_page(high_memory - 1)) + 1);
589 	free_all_bootmem();
590 
591 #ifdef CONFIG_PA11
592 	if (hppa_dma_ops == &pcxl_dma_ops) {
593 		pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
594 		parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
595 						+ PCXL_DMA_MAP_SIZE);
596 	} else {
597 		pcxl_dma_start = 0;
598 		parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
599 	}
600 #else
601 	parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
602 #endif
603 
604 	mem_init_print_info(NULL);
605 #ifdef CONFIG_DEBUG_KERNEL /* double-sanity-check paranoia */
606 	printk("virtual kernel memory layout:\n"
607 	       "    vmalloc : 0x%p - 0x%p   (%4ld MB)\n"
608 	       "    memory  : 0x%p - 0x%p   (%4ld MB)\n"
609 	       "      .init : 0x%p - 0x%p   (%4ld kB)\n"
610 	       "      .data : 0x%p - 0x%p   (%4ld kB)\n"
611 	       "      .text : 0x%p - 0x%p   (%4ld kB)\n",
612 
613 	       (void*)VMALLOC_START, (void*)VMALLOC_END,
614 	       (VMALLOC_END - VMALLOC_START) >> 20,
615 
616 	       __va(0), high_memory,
617 	       ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
618 
619 	       __init_begin, __init_end,
620 	       ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
621 
622 	       _etext, _edata,
623 	       ((unsigned long)_edata - (unsigned long)_etext) >> 10,
624 
625 	       _text, _etext,
626 	       ((unsigned long)_etext - (unsigned long)_text) >> 10);
627 #endif
628 }
629 
630 unsigned long *empty_zero_page __read_mostly;
631 EXPORT_SYMBOL(empty_zero_page);
632 
633 void show_mem(unsigned int filter)
634 {
635 	int i,free = 0,total = 0,reserved = 0;
636 	int shared = 0, cached = 0;
637 
638 	printk(KERN_INFO "Mem-info:\n");
639 	show_free_areas(filter);
640 	if (filter & SHOW_MEM_FILTER_PAGE_COUNT)
641 		return;
642 #ifndef CONFIG_DISCONTIGMEM
643 	i = max_mapnr;
644 	while (i-- > 0) {
645 		total++;
646 		if (PageReserved(mem_map+i))
647 			reserved++;
648 		else if (PageSwapCache(mem_map+i))
649 			cached++;
650 		else if (!page_count(&mem_map[i]))
651 			free++;
652 		else
653 			shared += page_count(&mem_map[i]) - 1;
654 	}
655 #else
656 	for (i = 0; i < npmem_ranges; i++) {
657 		int j;
658 
659 		for (j = node_start_pfn(i); j < node_end_pfn(i); j++) {
660 			struct page *p;
661 			unsigned long flags;
662 
663 			pgdat_resize_lock(NODE_DATA(i), &flags);
664 			p = nid_page_nr(i, j) - node_start_pfn(i);
665 
666 			total++;
667 			if (PageReserved(p))
668 				reserved++;
669 			else if (PageSwapCache(p))
670 				cached++;
671 			else if (!page_count(p))
672 				free++;
673 			else
674 				shared += page_count(p) - 1;
675 			pgdat_resize_unlock(NODE_DATA(i), &flags);
676         	}
677 	}
678 #endif
679 	printk(KERN_INFO "%d pages of RAM\n", total);
680 	printk(KERN_INFO "%d reserved pages\n", reserved);
681 	printk(KERN_INFO "%d pages shared\n", shared);
682 	printk(KERN_INFO "%d pages swap cached\n", cached);
683 
684 
685 #ifdef CONFIG_DISCONTIGMEM
686 	{
687 		struct zonelist *zl;
688 		int i, j;
689 
690 		for (i = 0; i < npmem_ranges; i++) {
691 			zl = node_zonelist(i, 0);
692 			for (j = 0; j < MAX_NR_ZONES; j++) {
693 				struct zoneref *z;
694 				struct zone *zone;
695 
696 				printk("Zone list for zone %d on node %d: ", j, i);
697 				for_each_zone_zonelist(zone, z, zl, j)
698 					printk("[%d/%s] ", zone_to_nid(zone),
699 								zone->name);
700 				printk("\n");
701 			}
702 		}
703 	}
704 #endif
705 }
706 
707 /*
708  * pagetable_init() sets up the page tables
709  *
710  * Note that gateway_init() places the Linux gateway page at page 0.
711  * Since gateway pages cannot be dereferenced this has the desirable
712  * side effect of trapping those pesky NULL-reference errors in the
713  * kernel.
714  */
715 static void __init pagetable_init(void)
716 {
717 	int range;
718 
719 	/* Map each physical memory range to its kernel vaddr */
720 
721 	for (range = 0; range < npmem_ranges; range++) {
722 		unsigned long start_paddr;
723 		unsigned long end_paddr;
724 		unsigned long size;
725 
726 		start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
727 		end_paddr = start_paddr + (pmem_ranges[range].pages << PAGE_SHIFT);
728 		size = pmem_ranges[range].pages << PAGE_SHIFT;
729 
730 		map_pages((unsigned long)__va(start_paddr), start_paddr,
731 			  size, PAGE_KERNEL, 0);
732 	}
733 
734 #ifdef CONFIG_BLK_DEV_INITRD
735 	if (initrd_end && initrd_end > mem_limit) {
736 		printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
737 		map_pages(initrd_start, __pa(initrd_start),
738 			  initrd_end - initrd_start, PAGE_KERNEL, 0);
739 	}
740 #endif
741 
742 	empty_zero_page = alloc_bootmem_pages(PAGE_SIZE);
743 	memset(empty_zero_page, 0, PAGE_SIZE);
744 }
745 
746 static void __init gateway_init(void)
747 {
748 	unsigned long linux_gateway_page_addr;
749 	/* FIXME: This is 'const' in order to trick the compiler
750 	   into not treating it as DP-relative data. */
751 	extern void * const linux_gateway_page;
752 
753 	linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
754 
755 	/*
756 	 * Setup Linux Gateway page.
757 	 *
758 	 * The Linux gateway page will reside in kernel space (on virtual
759 	 * page 0), so it doesn't need to be aliased into user space.
760 	 */
761 
762 	map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
763 		  PAGE_SIZE, PAGE_GATEWAY, 1);
764 }
765 
766 #ifdef CONFIG_HPUX
767 void
768 map_hpux_gateway_page(struct task_struct *tsk, struct mm_struct *mm)
769 {
770 	pgd_t *pg_dir;
771 	pmd_t *pmd;
772 	pte_t *pg_table;
773 	unsigned long start_pmd;
774 	unsigned long start_pte;
775 	unsigned long address;
776 	unsigned long hpux_gw_page_addr;
777 	/* FIXME: This is 'const' in order to trick the compiler
778 	   into not treating it as DP-relative data. */
779 	extern void * const hpux_gateway_page;
780 
781 	hpux_gw_page_addr = HPUX_GATEWAY_ADDR & PAGE_MASK;
782 
783 	/*
784 	 * Setup HP-UX Gateway page.
785 	 *
786 	 * The HP-UX gateway page resides in the user address space,
787 	 * so it needs to be aliased into each process.
788 	 */
789 
790 	pg_dir = pgd_offset(mm,hpux_gw_page_addr);
791 
792 #if PTRS_PER_PMD == 1
793 	start_pmd = 0;
794 #else
795 	start_pmd = ((hpux_gw_page_addr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
796 #endif
797 	start_pte = ((hpux_gw_page_addr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
798 
799 	address = __pa(&hpux_gateway_page);
800 #if PTRS_PER_PMD == 1
801 	pmd = (pmd_t *)__pa(pg_dir);
802 #else
803 	pmd = (pmd_t *) pgd_address(*pg_dir);
804 
805 	/*
806 	 * pmd is physical at this point
807 	 */
808 
809 	if (!pmd) {
810 		pmd = (pmd_t *) get_zeroed_page(GFP_KERNEL);
811 		pmd = (pmd_t *) __pa(pmd);
812 	}
813 
814 	__pgd_val_set(*pg_dir, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pmd);
815 #endif
816 	/* now change pmd to kernel virtual addresses */
817 
818 	pmd = (pmd_t *)__va(pmd) + start_pmd;
819 
820 	/*
821 	 * pg_table is physical at this point
822 	 */
823 
824 	pg_table = (pte_t *) pmd_address(*pmd);
825 	if (!pg_table)
826 		pg_table = (pte_t *) __pa(get_zeroed_page(GFP_KERNEL));
827 
828 	__pmd_val_set(*pmd, PxD_FLAG_PRESENT | PxD_FLAG_VALID | (unsigned long) pg_table);
829 
830 	/* now change pg_table to kernel virtual addresses */
831 
832 	pg_table = (pte_t *) __va(pg_table) + start_pte;
833 	set_pte(pg_table, __mk_pte(address, PAGE_GATEWAY));
834 }
835 EXPORT_SYMBOL(map_hpux_gateway_page);
836 #endif
837 
838 void __init paging_init(void)
839 {
840 	int i;
841 
842 	setup_bootmem();
843 	pagetable_init();
844 	gateway_init();
845 	flush_cache_all_local(); /* start with known state */
846 	flush_tlb_all_local(NULL);
847 
848 	for (i = 0; i < npmem_ranges; i++) {
849 		unsigned long zones_size[MAX_NR_ZONES] = { 0, };
850 
851 		zones_size[ZONE_NORMAL] = pmem_ranges[i].pages;
852 
853 #ifdef CONFIG_DISCONTIGMEM
854 		/* Need to initialize the pfnnid_map before we can initialize
855 		   the zone */
856 		{
857 		    int j;
858 		    for (j = (pmem_ranges[i].start_pfn >> PFNNID_SHIFT);
859 			 j <= ((pmem_ranges[i].start_pfn + pmem_ranges[i].pages) >> PFNNID_SHIFT);
860 			 j++) {
861 			pfnnid_map[j] = i;
862 		    }
863 		}
864 #endif
865 
866 		free_area_init_node(i, zones_size,
867 				pmem_ranges[i].start_pfn, NULL);
868 	}
869 }
870 
871 #ifdef CONFIG_PA20
872 
873 /*
874  * Currently, all PA20 chips have 18 bit protection IDs, which is the
875  * limiting factor (space ids are 32 bits).
876  */
877 
878 #define NR_SPACE_IDS 262144
879 
880 #else
881 
882 /*
883  * Currently we have a one-to-one relationship between space IDs and
884  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
885  * support 15 bit protection IDs, so that is the limiting factor.
886  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
887  * probably not worth the effort for a special case here.
888  */
889 
890 #define NR_SPACE_IDS 32768
891 
892 #endif  /* !CONFIG_PA20 */
893 
894 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
895 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
896 
897 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
898 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
899 static unsigned long space_id_index;
900 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
901 static unsigned long dirty_space_ids = 0;
902 
903 static DEFINE_SPINLOCK(sid_lock);
904 
905 unsigned long alloc_sid(void)
906 {
907 	unsigned long index;
908 
909 	spin_lock(&sid_lock);
910 
911 	if (free_space_ids == 0) {
912 		if (dirty_space_ids != 0) {
913 			spin_unlock(&sid_lock);
914 			flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
915 			spin_lock(&sid_lock);
916 		}
917 		BUG_ON(free_space_ids == 0);
918 	}
919 
920 	free_space_ids--;
921 
922 	index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
923 	space_id[index >> SHIFT_PER_LONG] |= (1L << (index & (BITS_PER_LONG - 1)));
924 	space_id_index = index;
925 
926 	spin_unlock(&sid_lock);
927 
928 	return index << SPACEID_SHIFT;
929 }
930 
931 void free_sid(unsigned long spaceid)
932 {
933 	unsigned long index = spaceid >> SPACEID_SHIFT;
934 	unsigned long *dirty_space_offset;
935 
936 	dirty_space_offset = dirty_space_id + (index >> SHIFT_PER_LONG);
937 	index &= (BITS_PER_LONG - 1);
938 
939 	spin_lock(&sid_lock);
940 
941 	BUG_ON(*dirty_space_offset & (1L << index)); /* attempt to free space id twice */
942 
943 	*dirty_space_offset |= (1L << index);
944 	dirty_space_ids++;
945 
946 	spin_unlock(&sid_lock);
947 }
948 
949 
950 #ifdef CONFIG_SMP
951 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
952 {
953 	int i;
954 
955 	/* NOTE: sid_lock must be held upon entry */
956 
957 	*ndirtyptr = dirty_space_ids;
958 	if (dirty_space_ids != 0) {
959 	    for (i = 0; i < SID_ARRAY_SIZE; i++) {
960 		dirty_array[i] = dirty_space_id[i];
961 		dirty_space_id[i] = 0;
962 	    }
963 	    dirty_space_ids = 0;
964 	}
965 
966 	return;
967 }
968 
969 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
970 {
971 	int i;
972 
973 	/* NOTE: sid_lock must be held upon entry */
974 
975 	if (ndirty != 0) {
976 		for (i = 0; i < SID_ARRAY_SIZE; i++) {
977 			space_id[i] ^= dirty_array[i];
978 		}
979 
980 		free_space_ids += ndirty;
981 		space_id_index = 0;
982 	}
983 }
984 
985 #else /* CONFIG_SMP */
986 
987 static void recycle_sids(void)
988 {
989 	int i;
990 
991 	/* NOTE: sid_lock must be held upon entry */
992 
993 	if (dirty_space_ids != 0) {
994 		for (i = 0; i < SID_ARRAY_SIZE; i++) {
995 			space_id[i] ^= dirty_space_id[i];
996 			dirty_space_id[i] = 0;
997 		}
998 
999 		free_space_ids += dirty_space_ids;
1000 		dirty_space_ids = 0;
1001 		space_id_index = 0;
1002 	}
1003 }
1004 #endif
1005 
1006 /*
1007  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
1008  * purged, we can safely reuse the space ids that were released but
1009  * not flushed from the tlb.
1010  */
1011 
1012 #ifdef CONFIG_SMP
1013 
1014 static unsigned long recycle_ndirty;
1015 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
1016 static unsigned int recycle_inuse;
1017 
1018 void flush_tlb_all(void)
1019 {
1020 	int do_recycle;
1021 
1022 	__inc_irq_stat(irq_tlb_count);
1023 	do_recycle = 0;
1024 	spin_lock(&sid_lock);
1025 	if (dirty_space_ids > RECYCLE_THRESHOLD) {
1026 	    BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
1027 	    get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
1028 	    recycle_inuse++;
1029 	    do_recycle++;
1030 	}
1031 	spin_unlock(&sid_lock);
1032 	on_each_cpu(flush_tlb_all_local, NULL, 1);
1033 	if (do_recycle) {
1034 	    spin_lock(&sid_lock);
1035 	    recycle_sids(recycle_ndirty,recycle_dirty_array);
1036 	    recycle_inuse = 0;
1037 	    spin_unlock(&sid_lock);
1038 	}
1039 }
1040 #else
1041 void flush_tlb_all(void)
1042 {
1043 	__inc_irq_stat(irq_tlb_count);
1044 	spin_lock(&sid_lock);
1045 	flush_tlb_all_local(NULL);
1046 	recycle_sids();
1047 	spin_unlock(&sid_lock);
1048 }
1049 #endif
1050 
1051 #ifdef CONFIG_BLK_DEV_INITRD
1052 void free_initrd_mem(unsigned long start, unsigned long end)
1053 {
1054 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
1055 }
1056 #endif
1057