xref: /openbmc/linux/arch/parisc/mm/init.c (revision b9dd2add)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/arch/parisc/mm/init.c
4  *
5  *  Copyright (C) 1995	Linus Torvalds
6  *  Copyright 1999 SuSE GmbH
7  *    changed by Philipp Rumpf
8  *  Copyright 1999 Philipp Rumpf (prumpf@tux.org)
9  *  Copyright 2004 Randolph Chung (tausq@debian.org)
10  *  Copyright 2006-2007 Helge Deller (deller@gmx.de)
11  *
12  */
13 
14 
15 #include <linux/module.h>
16 #include <linux/mm.h>
17 #include <linux/memblock.h>
18 #include <linux/gfp.h>
19 #include <linux/delay.h>
20 #include <linux/init.h>
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 */
26 #include <linux/compat.h>
27 
28 #include <asm/pgalloc.h>
29 #include <asm/tlb.h>
30 #include <asm/pdc_chassis.h>
31 #include <asm/mmzone.h>
32 #include <asm/sections.h>
33 #include <asm/msgbuf.h>
34 #include <asm/sparsemem.h>
35 
36 extern int  data_start;
37 extern void parisc_kernel_start(void);	/* Kernel entry point in head.S */
38 
39 #if CONFIG_PGTABLE_LEVELS == 3
40 /* NOTE: This layout exactly conforms to the hybrid L2/L3 page table layout
41  * with the first pmd adjacent to the pgd and below it. gcc doesn't actually
42  * guarantee that global objects will be laid out in memory in the same order
43  * as the order of declaration, so put these in different sections and use
44  * the linker script to order them. */
45 pmd_t pmd0[PTRS_PER_PMD] __section(".data..vm0.pmd") __attribute__ ((aligned(PAGE_SIZE)));
46 #endif
47 
48 pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".data..vm0.pgd") __attribute__ ((aligned(PAGE_SIZE)));
49 pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __section(".data..vm0.pte") __attribute__ ((aligned(PAGE_SIZE)));
50 
51 static struct resource data_resource = {
52 	.name	= "Kernel data",
53 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
54 };
55 
56 static struct resource code_resource = {
57 	.name	= "Kernel code",
58 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
59 };
60 
61 static struct resource pdcdata_resource = {
62 	.name	= "PDC data (Page Zero)",
63 	.start	= 0,
64 	.end	= 0x9ff,
65 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM,
66 };
67 
68 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __ro_after_init;
69 
70 /* The following array is initialized from the firmware specific
71  * information retrieved in kernel/inventory.c.
72  */
73 
74 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __initdata;
75 int npmem_ranges __initdata;
76 
77 #ifdef CONFIG_64BIT
78 #define MAX_MEM         (1UL << MAX_PHYSMEM_BITS)
79 #else /* !CONFIG_64BIT */
80 #define MAX_MEM         (3584U*1024U*1024U)
81 #endif /* !CONFIG_64BIT */
82 
83 static unsigned long mem_limit __read_mostly = MAX_MEM;
84 
85 static void __init mem_limit_func(void)
86 {
87 	char *cp, *end;
88 	unsigned long limit;
89 
90 	/* We need this before __setup() functions are called */
91 
92 	limit = MAX_MEM;
93 	for (cp = boot_command_line; *cp; ) {
94 		if (memcmp(cp, "mem=", 4) == 0) {
95 			cp += 4;
96 			limit = memparse(cp, &end);
97 			if (end != cp)
98 				break;
99 			cp = end;
100 		} else {
101 			while (*cp != ' ' && *cp)
102 				++cp;
103 			while (*cp == ' ')
104 				++cp;
105 		}
106 	}
107 
108 	if (limit < mem_limit)
109 		mem_limit = limit;
110 }
111 
112 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
113 
114 static void __init setup_bootmem(void)
115 {
116 	unsigned long mem_max;
117 #ifndef CONFIG_SPARSEMEM
118 	physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
119 	int npmem_holes;
120 #endif
121 	int i, sysram_resource_count;
122 
123 	disable_sr_hashing(); /* Turn off space register hashing */
124 
125 	/*
126 	 * Sort the ranges. Since the number of ranges is typically
127 	 * small, and performance is not an issue here, just do
128 	 * a simple insertion sort.
129 	 */
130 
131 	for (i = 1; i < npmem_ranges; i++) {
132 		int j;
133 
134 		for (j = i; j > 0; j--) {
135 			physmem_range_t tmp;
136 
137 			if (pmem_ranges[j-1].start_pfn <
138 			    pmem_ranges[j].start_pfn) {
139 
140 				break;
141 			}
142 			tmp = pmem_ranges[j-1];
143 			pmem_ranges[j-1] = pmem_ranges[j];
144 			pmem_ranges[j] = tmp;
145 		}
146 	}
147 
148 #ifndef CONFIG_SPARSEMEM
149 	/*
150 	 * Throw out ranges that are too far apart (controlled by
151 	 * MAX_GAP).
152 	 */
153 
154 	for (i = 1; i < npmem_ranges; i++) {
155 		if (pmem_ranges[i].start_pfn -
156 			(pmem_ranges[i-1].start_pfn +
157 			 pmem_ranges[i-1].pages) > MAX_GAP) {
158 			npmem_ranges = i;
159 			printk("Large gap in memory detected (%ld pages). "
160 			       "Consider turning on CONFIG_SPARSEMEM\n",
161 			       pmem_ranges[i].start_pfn -
162 			       (pmem_ranges[i-1].start_pfn +
163 			        pmem_ranges[i-1].pages));
164 			break;
165 		}
166 	}
167 #endif
168 
169 	/* Print the memory ranges */
170 	pr_info("Memory Ranges:\n");
171 
172 	for (i = 0; i < npmem_ranges; i++) {
173 		struct resource *res = &sysram_resources[i];
174 		unsigned long start;
175 		unsigned long size;
176 
177 		size = (pmem_ranges[i].pages << PAGE_SHIFT);
178 		start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
179 		pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
180 			i, start, start + (size - 1), size >> 20);
181 
182 		/* request memory resource */
183 		res->name = "System RAM";
184 		res->start = start;
185 		res->end = start + size - 1;
186 		res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
187 		request_resource(&iomem_resource, res);
188 	}
189 
190 	sysram_resource_count = npmem_ranges;
191 
192 	/*
193 	 * For 32 bit kernels we limit the amount of memory we can
194 	 * support, in order to preserve enough kernel address space
195 	 * for other purposes. For 64 bit kernels we don't normally
196 	 * limit the memory, but this mechanism can be used to
197 	 * artificially limit the amount of memory (and it is written
198 	 * to work with multiple memory ranges).
199 	 */
200 
201 	mem_limit_func();       /* check for "mem=" argument */
202 
203 	mem_max = 0;
204 	for (i = 0; i < npmem_ranges; i++) {
205 		unsigned long rsize;
206 
207 		rsize = pmem_ranges[i].pages << PAGE_SHIFT;
208 		if ((mem_max + rsize) > mem_limit) {
209 			printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
210 			if (mem_max == mem_limit)
211 				npmem_ranges = i;
212 			else {
213 				pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
214 						       - (mem_max >> PAGE_SHIFT);
215 				npmem_ranges = i + 1;
216 				mem_max = mem_limit;
217 			}
218 			break;
219 		}
220 		mem_max += rsize;
221 	}
222 
223 	printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
224 
225 #ifndef CONFIG_SPARSEMEM
226 	/* Merge the ranges, keeping track of the holes */
227 	{
228 		unsigned long end_pfn;
229 		unsigned long hole_pages;
230 
231 		npmem_holes = 0;
232 		end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
233 		for (i = 1; i < npmem_ranges; i++) {
234 
235 			hole_pages = pmem_ranges[i].start_pfn - end_pfn;
236 			if (hole_pages) {
237 				pmem_holes[npmem_holes].start_pfn = end_pfn;
238 				pmem_holes[npmem_holes++].pages = hole_pages;
239 				end_pfn += hole_pages;
240 			}
241 			end_pfn += pmem_ranges[i].pages;
242 		}
243 
244 		pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
245 		npmem_ranges = 1;
246 	}
247 #endif
248 
249 	/*
250 	 * Initialize and free the full range of memory in each range.
251 	 */
252 
253 	max_pfn = 0;
254 	for (i = 0; i < npmem_ranges; i++) {
255 		unsigned long start_pfn;
256 		unsigned long npages;
257 		unsigned long start;
258 		unsigned long size;
259 
260 		start_pfn = pmem_ranges[i].start_pfn;
261 		npages = pmem_ranges[i].pages;
262 
263 		start = start_pfn << PAGE_SHIFT;
264 		size = npages << PAGE_SHIFT;
265 
266 		/* add system RAM memblock */
267 		memblock_add(start, size);
268 
269 		if ((start_pfn + npages) > max_pfn)
270 			max_pfn = start_pfn + npages;
271 	}
272 
273 	/*
274 	 * We can't use memblock top-down allocations because we only
275 	 * created the initial mapping up to KERNEL_INITIAL_SIZE in
276 	 * the assembly bootup code.
277 	 */
278 	memblock_set_bottom_up(true);
279 
280 	/* IOMMU is always used to access "high mem" on those boxes
281 	 * that can support enough mem that a PCI device couldn't
282 	 * directly DMA to any physical addresses.
283 	 * ISA DMA support will need to revisit this.
284 	 */
285 	max_low_pfn = max_pfn;
286 
287 	/* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
288 
289 #define PDC_CONSOLE_IO_IODC_SIZE 32768
290 
291 	memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
292 				PDC_CONSOLE_IO_IODC_SIZE));
293 	memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
294 			(unsigned long)(_end - KERNEL_BINARY_TEXT_START));
295 
296 #ifndef CONFIG_SPARSEMEM
297 
298 	/* reserve the holes */
299 
300 	for (i = 0; i < npmem_holes; i++) {
301 		memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
302 				(pmem_holes[i].pages << PAGE_SHIFT));
303 	}
304 #endif
305 
306 #ifdef CONFIG_BLK_DEV_INITRD
307 	if (initrd_start) {
308 		printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
309 		if (__pa(initrd_start) < mem_max) {
310 			unsigned long initrd_reserve;
311 
312 			if (__pa(initrd_end) > mem_max) {
313 				initrd_reserve = mem_max - __pa(initrd_start);
314 			} else {
315 				initrd_reserve = initrd_end - initrd_start;
316 			}
317 			initrd_below_start_ok = 1;
318 			printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
319 
320 			memblock_reserve(__pa(initrd_start), initrd_reserve);
321 		}
322 	}
323 #endif
324 
325 	data_resource.start =  virt_to_phys(&data_start);
326 	data_resource.end = virt_to_phys(_end) - 1;
327 	code_resource.start = virt_to_phys(_text);
328 	code_resource.end = virt_to_phys(&data_start)-1;
329 
330 	/* We don't know which region the kernel will be in, so try
331 	 * all of them.
332 	 */
333 	for (i = 0; i < sysram_resource_count; i++) {
334 		struct resource *res = &sysram_resources[i];
335 		request_resource(res, &code_resource);
336 		request_resource(res, &data_resource);
337 	}
338 	request_resource(&sysram_resources[0], &pdcdata_resource);
339 
340 	/* Initialize Page Deallocation Table (PDT) and check for bad memory. */
341 	pdc_pdt_init();
342 
343 	memblock_allow_resize();
344 	memblock_dump_all();
345 }
346 
347 static bool kernel_set_to_readonly;
348 
349 static void __init map_pages(unsigned long start_vaddr,
350 			     unsigned long start_paddr, unsigned long size,
351 			     pgprot_t pgprot, int force)
352 {
353 	pmd_t *pmd;
354 	pte_t *pg_table;
355 	unsigned long end_paddr;
356 	unsigned long start_pmd;
357 	unsigned long start_pte;
358 	unsigned long tmp1;
359 	unsigned long tmp2;
360 	unsigned long address;
361 	unsigned long vaddr;
362 	unsigned long ro_start;
363 	unsigned long ro_end;
364 	unsigned long kernel_start, kernel_end;
365 
366 	ro_start = __pa((unsigned long)_text);
367 	ro_end   = __pa((unsigned long)&data_start);
368 	kernel_start = __pa((unsigned long)&__init_begin);
369 	kernel_end  = __pa((unsigned long)&_end);
370 
371 	end_paddr = start_paddr + size;
372 
373 	/* for 2-level configuration PTRS_PER_PMD is 0 so start_pmd will be 0 */
374 	start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
375 	start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
376 
377 	address = start_paddr;
378 	vaddr = start_vaddr;
379 	while (address < end_paddr) {
380 		pgd_t *pgd = pgd_offset_k(vaddr);
381 		p4d_t *p4d = p4d_offset(pgd, vaddr);
382 		pud_t *pud = pud_offset(p4d, vaddr);
383 
384 #if CONFIG_PGTABLE_LEVELS == 3
385 		if (pud_none(*pud)) {
386 			pmd = memblock_alloc(PAGE_SIZE << PMD_ORDER,
387 					     PAGE_SIZE << PMD_ORDER);
388 			if (!pmd)
389 				panic("pmd allocation failed.\n");
390 			pud_populate(NULL, pud, pmd);
391 		}
392 #endif
393 
394 		pmd = pmd_offset(pud, vaddr);
395 		for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
396 			if (pmd_none(*pmd)) {
397 				pg_table = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
398 				if (!pg_table)
399 					panic("page table allocation failed\n");
400 				pmd_populate_kernel(NULL, pmd, pg_table);
401 			}
402 
403 			pg_table = pte_offset_kernel(pmd, vaddr);
404 			for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
405 				pte_t pte;
406 				pgprot_t prot;
407 				bool huge = false;
408 
409 				if (force) {
410 					prot = pgprot;
411 				} else if (address < kernel_start || address >= kernel_end) {
412 					/* outside kernel memory */
413 					prot = PAGE_KERNEL;
414 				} else if (!kernel_set_to_readonly) {
415 					/* still initializing, allow writing to RO memory */
416 					prot = PAGE_KERNEL_RWX;
417 					huge = true;
418 				} else if (address >= ro_start) {
419 					/* Code (ro) and Data areas */
420 					prot = (address < ro_end) ?
421 						PAGE_KERNEL_EXEC : PAGE_KERNEL;
422 					huge = true;
423 				} else {
424 					prot = PAGE_KERNEL;
425 				}
426 
427 				pte = __mk_pte(address, prot);
428 				if (huge)
429 					pte = pte_mkhuge(pte);
430 
431 				if (address >= end_paddr)
432 					break;
433 
434 				set_pte(pg_table, pte);
435 
436 				address += PAGE_SIZE;
437 				vaddr += PAGE_SIZE;
438 			}
439 			start_pte = 0;
440 
441 			if (address >= end_paddr)
442 			    break;
443 		}
444 		start_pmd = 0;
445 	}
446 }
447 
448 void __init set_kernel_text_rw(int enable_read_write)
449 {
450 	unsigned long start = (unsigned long) __init_begin;
451 	unsigned long end   = (unsigned long) &data_start;
452 
453 	map_pages(start, __pa(start), end-start,
454 		PAGE_KERNEL_RWX, enable_read_write ? 1:0);
455 
456 	/* force the kernel to see the new page table entries */
457 	flush_cache_all();
458 	flush_tlb_all();
459 }
460 
461 void __ref free_initmem(void)
462 {
463 	unsigned long init_begin = (unsigned long)__init_begin;
464 	unsigned long init_end = (unsigned long)__init_end;
465 	unsigned long kernel_end  = (unsigned long)&_end;
466 
467 	/* Remap kernel text and data, but do not touch init section yet. */
468 	kernel_set_to_readonly = true;
469 	map_pages(init_end, __pa(init_end), kernel_end - init_end,
470 		  PAGE_KERNEL, 0);
471 
472 	/* The init text pages are marked R-X.  We have to
473 	 * flush the icache and mark them RW-
474 	 *
475 	 * This is tricky, because map_pages is in the init section.
476 	 * Do a dummy remap of the data section first (the data
477 	 * section is already PAGE_KERNEL) to pull in the TLB entries
478 	 * for map_kernel */
479 	map_pages(init_begin, __pa(init_begin), init_end - init_begin,
480 		  PAGE_KERNEL_RWX, 1);
481 	/* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
482 	 * map_pages */
483 	map_pages(init_begin, __pa(init_begin), init_end - init_begin,
484 		  PAGE_KERNEL, 1);
485 
486 	/* force the kernel to see the new TLB entries */
487 	__flush_tlb_range(0, init_begin, kernel_end);
488 
489 	/* finally dump all the instructions which were cached, since the
490 	 * pages are no-longer executable */
491 	flush_icache_range(init_begin, init_end);
492 
493 	free_initmem_default(POISON_FREE_INITMEM);
494 
495 	/* set up a new led state on systems shipped LED State panel */
496 	pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
497 }
498 
499 
500 #ifdef CONFIG_STRICT_KERNEL_RWX
501 void mark_rodata_ro(void)
502 {
503 	/* rodata memory was already mapped with KERNEL_RO access rights by
504            pagetable_init() and map_pages(). No need to do additional stuff here */
505 	unsigned long roai_size = __end_ro_after_init - __start_ro_after_init;
506 
507 	pr_info("Write protected read-only-after-init data: %luk\n", roai_size >> 10);
508 }
509 #endif
510 
511 
512 /*
513  * Just an arbitrary offset to serve as a "hole" between mapping areas
514  * (between top of physical memory and a potential pcxl dma mapping
515  * area, and below the vmalloc mapping area).
516  *
517  * The current 32K value just means that there will be a 32K "hole"
518  * between mapping areas. That means that  any out-of-bounds memory
519  * accesses will hopefully be caught. The vmalloc() routines leaves
520  * a hole of 4kB between each vmalloced area for the same reason.
521  */
522 
523  /* Leave room for gateway page expansion */
524 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
525 #error KERNEL_MAP_START is in gateway reserved region
526 #endif
527 #define MAP_START (KERNEL_MAP_START)
528 
529 #define VM_MAP_OFFSET  (32*1024)
530 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
531 				     & ~(VM_MAP_OFFSET-1)))
532 
533 void *parisc_vmalloc_start __ro_after_init;
534 EXPORT_SYMBOL(parisc_vmalloc_start);
535 
536 #ifdef CONFIG_PA11
537 unsigned long pcxl_dma_start __ro_after_init;
538 #endif
539 
540 void __init mem_init(void)
541 {
542 	/* Do sanity checks on IPC (compat) structures */
543 	BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
544 #ifndef CONFIG_64BIT
545 	BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
546 	BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
547 	BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
548 #endif
549 #ifdef CONFIG_COMPAT
550 	BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
551 	BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
552 	BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
553 	BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
554 #endif
555 
556 	/* Do sanity checks on page table constants */
557 	BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
558 	BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
559 	BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
560 	BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
561 			> BITS_PER_LONG);
562 
563 	high_memory = __va((max_pfn << PAGE_SHIFT));
564 	set_max_mapnr(max_low_pfn);
565 	memblock_free_all();
566 
567 #ifdef CONFIG_PA11
568 	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
569 		pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
570 		parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
571 						+ PCXL_DMA_MAP_SIZE);
572 	} else
573 #endif
574 		parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
575 
576 	mem_init_print_info(NULL);
577 
578 #if 0
579 	/*
580 	 * Do not expose the virtual kernel memory layout to userspace.
581 	 * But keep code for debugging purposes.
582 	 */
583 	printk("virtual kernel memory layout:\n"
584 	       "     vmalloc : 0x%px - 0x%px   (%4ld MB)\n"
585 	       "     fixmap  : 0x%px - 0x%px   (%4ld kB)\n"
586 	       "     memory  : 0x%px - 0x%px   (%4ld MB)\n"
587 	       "       .init : 0x%px - 0x%px   (%4ld kB)\n"
588 	       "       .data : 0x%px - 0x%px   (%4ld kB)\n"
589 	       "       .text : 0x%px - 0x%px   (%4ld kB)\n",
590 
591 	       (void*)VMALLOC_START, (void*)VMALLOC_END,
592 	       (VMALLOC_END - VMALLOC_START) >> 20,
593 
594 	       (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE),
595 	       (unsigned long)(FIXMAP_SIZE / 1024),
596 
597 	       __va(0), high_memory,
598 	       ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
599 
600 	       __init_begin, __init_end,
601 	       ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
602 
603 	       _etext, _edata,
604 	       ((unsigned long)_edata - (unsigned long)_etext) >> 10,
605 
606 	       _text, _etext,
607 	       ((unsigned long)_etext - (unsigned long)_text) >> 10);
608 #endif
609 }
610 
611 unsigned long *empty_zero_page __ro_after_init;
612 EXPORT_SYMBOL(empty_zero_page);
613 
614 /*
615  * pagetable_init() sets up the page tables
616  *
617  * Note that gateway_init() places the Linux gateway page at page 0.
618  * Since gateway pages cannot be dereferenced this has the desirable
619  * side effect of trapping those pesky NULL-reference errors in the
620  * kernel.
621  */
622 static void __init pagetable_init(void)
623 {
624 	int range;
625 
626 	/* Map each physical memory range to its kernel vaddr */
627 
628 	for (range = 0; range < npmem_ranges; range++) {
629 		unsigned long start_paddr;
630 		unsigned long end_paddr;
631 		unsigned long size;
632 
633 		start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
634 		size = pmem_ranges[range].pages << PAGE_SHIFT;
635 		end_paddr = start_paddr + size;
636 
637 		map_pages((unsigned long)__va(start_paddr), start_paddr,
638 			  size, PAGE_KERNEL, 0);
639 	}
640 
641 #ifdef CONFIG_BLK_DEV_INITRD
642 	if (initrd_end && initrd_end > mem_limit) {
643 		printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
644 		map_pages(initrd_start, __pa(initrd_start),
645 			  initrd_end - initrd_start, PAGE_KERNEL, 0);
646 	}
647 #endif
648 
649 	empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
650 	if (!empty_zero_page)
651 		panic("zero page allocation failed.\n");
652 
653 }
654 
655 static void __init gateway_init(void)
656 {
657 	unsigned long linux_gateway_page_addr;
658 	/* FIXME: This is 'const' in order to trick the compiler
659 	   into not treating it as DP-relative data. */
660 	extern void * const linux_gateway_page;
661 
662 	linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
663 
664 	/*
665 	 * Setup Linux Gateway page.
666 	 *
667 	 * The Linux gateway page will reside in kernel space (on virtual
668 	 * page 0), so it doesn't need to be aliased into user space.
669 	 */
670 
671 	map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
672 		  PAGE_SIZE, PAGE_GATEWAY, 1);
673 }
674 
675 static void __init parisc_bootmem_free(void)
676 {
677 	unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
678 
679 	max_zone_pfn[0] = memblock_end_of_DRAM();
680 
681 	free_area_init(max_zone_pfn);
682 }
683 
684 void __init paging_init(void)
685 {
686 	setup_bootmem();
687 	pagetable_init();
688 	gateway_init();
689 	flush_cache_all_local(); /* start with known state */
690 	flush_tlb_all_local(NULL);
691 
692 	sparse_init();
693 	parisc_bootmem_free();
694 }
695 
696 #ifdef CONFIG_PA20
697 
698 /*
699  * Currently, all PA20 chips have 18 bit protection IDs, which is the
700  * limiting factor (space ids are 32 bits).
701  */
702 
703 #define NR_SPACE_IDS 262144
704 
705 #else
706 
707 /*
708  * Currently we have a one-to-one relationship between space IDs and
709  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
710  * support 15 bit protection IDs, so that is the limiting factor.
711  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
712  * probably not worth the effort for a special case here.
713  */
714 
715 #define NR_SPACE_IDS 32768
716 
717 #endif  /* !CONFIG_PA20 */
718 
719 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
720 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
721 
722 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
723 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
724 static unsigned long space_id_index;
725 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
726 static unsigned long dirty_space_ids = 0;
727 
728 static DEFINE_SPINLOCK(sid_lock);
729 
730 unsigned long alloc_sid(void)
731 {
732 	unsigned long index;
733 
734 	spin_lock(&sid_lock);
735 
736 	if (free_space_ids == 0) {
737 		if (dirty_space_ids != 0) {
738 			spin_unlock(&sid_lock);
739 			flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
740 			spin_lock(&sid_lock);
741 		}
742 		BUG_ON(free_space_ids == 0);
743 	}
744 
745 	free_space_ids--;
746 
747 	index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
748 	space_id[BIT_WORD(index)] |= BIT_MASK(index);
749 	space_id_index = index;
750 
751 	spin_unlock(&sid_lock);
752 
753 	return index << SPACEID_SHIFT;
754 }
755 
756 void free_sid(unsigned long spaceid)
757 {
758 	unsigned long index = spaceid >> SPACEID_SHIFT;
759 	unsigned long *dirty_space_offset, mask;
760 
761 	dirty_space_offset = &dirty_space_id[BIT_WORD(index)];
762 	mask = BIT_MASK(index);
763 
764 	spin_lock(&sid_lock);
765 
766 	BUG_ON(*dirty_space_offset & mask); /* attempt to free space id twice */
767 
768 	*dirty_space_offset |= mask;
769 	dirty_space_ids++;
770 
771 	spin_unlock(&sid_lock);
772 }
773 
774 
775 #ifdef CONFIG_SMP
776 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
777 {
778 	int i;
779 
780 	/* NOTE: sid_lock must be held upon entry */
781 
782 	*ndirtyptr = dirty_space_ids;
783 	if (dirty_space_ids != 0) {
784 	    for (i = 0; i < SID_ARRAY_SIZE; i++) {
785 		dirty_array[i] = dirty_space_id[i];
786 		dirty_space_id[i] = 0;
787 	    }
788 	    dirty_space_ids = 0;
789 	}
790 
791 	return;
792 }
793 
794 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
795 {
796 	int i;
797 
798 	/* NOTE: sid_lock must be held upon entry */
799 
800 	if (ndirty != 0) {
801 		for (i = 0; i < SID_ARRAY_SIZE; i++) {
802 			space_id[i] ^= dirty_array[i];
803 		}
804 
805 		free_space_ids += ndirty;
806 		space_id_index = 0;
807 	}
808 }
809 
810 #else /* CONFIG_SMP */
811 
812 static void recycle_sids(void)
813 {
814 	int i;
815 
816 	/* NOTE: sid_lock must be held upon entry */
817 
818 	if (dirty_space_ids != 0) {
819 		for (i = 0; i < SID_ARRAY_SIZE; i++) {
820 			space_id[i] ^= dirty_space_id[i];
821 			dirty_space_id[i] = 0;
822 		}
823 
824 		free_space_ids += dirty_space_ids;
825 		dirty_space_ids = 0;
826 		space_id_index = 0;
827 	}
828 }
829 #endif
830 
831 /*
832  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
833  * purged, we can safely reuse the space ids that were released but
834  * not flushed from the tlb.
835  */
836 
837 #ifdef CONFIG_SMP
838 
839 static unsigned long recycle_ndirty;
840 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
841 static unsigned int recycle_inuse;
842 
843 void flush_tlb_all(void)
844 {
845 	int do_recycle;
846 
847 	__inc_irq_stat(irq_tlb_count);
848 	do_recycle = 0;
849 	spin_lock(&sid_lock);
850 	if (dirty_space_ids > RECYCLE_THRESHOLD) {
851 	    BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
852 	    get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
853 	    recycle_inuse++;
854 	    do_recycle++;
855 	}
856 	spin_unlock(&sid_lock);
857 	on_each_cpu(flush_tlb_all_local, NULL, 1);
858 	if (do_recycle) {
859 	    spin_lock(&sid_lock);
860 	    recycle_sids(recycle_ndirty,recycle_dirty_array);
861 	    recycle_inuse = 0;
862 	    spin_unlock(&sid_lock);
863 	}
864 }
865 #else
866 void flush_tlb_all(void)
867 {
868 	__inc_irq_stat(irq_tlb_count);
869 	spin_lock(&sid_lock);
870 	flush_tlb_all_local(NULL);
871 	recycle_sids();
872 	spin_unlock(&sid_lock);
873 }
874 #endif
875