Loading...
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#include <linux/execmem.h>
28
29#include <asm/pgalloc.h>
30#include <asm/tlb.h>
31#include <asm/pdc_chassis.h>
32#include <asm/mmzone.h>
33#include <asm/sections.h>
34#include <asm/msgbuf.h>
35#include <asm/sparsemem.h>
36#include <asm/asm-offsets.h>
37#include <asm/shmbuf.h>
38
39extern int data_start;
40extern void parisc_kernel_start(void); /* Kernel entry point in head.S */
41
42#if CONFIG_PGTABLE_LEVELS == 3
43pmd_t pmd0[PTRS_PER_PMD] __section(".data..vm0.pmd") __attribute__ ((aligned(PAGE_SIZE)));
44#endif
45
46pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".data..vm0.pgd") __attribute__ ((aligned(PAGE_SIZE)));
47pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __section(".data..vm0.pte") __attribute__ ((aligned(PAGE_SIZE)));
48
49static struct resource data_resource = {
50 .name = "Kernel data",
51 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
52};
53
54static struct resource code_resource = {
55 .name = "Kernel code",
56 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
57};
58
59static struct resource pdcdata_resource = {
60 .name = "PDC data (Page Zero)",
61 .start = 0,
62 .end = 0x9ff,
63 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
64};
65
66static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __ro_after_init;
67
68/* The following array is initialized from the firmware specific
69 * information retrieved in kernel/inventory.c.
70 */
71
72physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __initdata;
73int npmem_ranges __initdata;
74
75#ifdef CONFIG_64BIT
76#define MAX_MEM (1UL << MAX_PHYSMEM_BITS)
77#else /* !CONFIG_64BIT */
78#define MAX_MEM (3584U*1024U*1024U)
79#endif /* !CONFIG_64BIT */
80
81static unsigned long mem_limit __read_mostly = MAX_MEM;
82
83static void __init mem_limit_func(void)
84{
85 char *cp, *end;
86 unsigned long limit;
87
88 /* We need this before __setup() functions are called */
89
90 limit = MAX_MEM;
91 for (cp = boot_command_line; *cp; ) {
92 if (memcmp(cp, "mem=", 4) == 0) {
93 cp += 4;
94 limit = memparse(cp, &end);
95 if (end != cp)
96 break;
97 cp = end;
98 } else {
99 while (*cp != ' ' && *cp)
100 ++cp;
101 while (*cp == ' ')
102 ++cp;
103 }
104 }
105
106 if (limit < mem_limit)
107 mem_limit = limit;
108}
109
110#define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
111
112static void __init setup_bootmem(void)
113{
114 unsigned long mem_max;
115#ifndef CONFIG_SPARSEMEM
116 physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
117 int npmem_holes;
118#endif
119 int i, sysram_resource_count;
120
121 disable_sr_hashing(); /* Turn off space register hashing */
122
123 /*
124 * Sort the ranges. Since the number of ranges is typically
125 * small, and performance is not an issue here, just do
126 * a simple insertion sort.
127 */
128
129 for (i = 1; i < npmem_ranges; i++) {
130 int j;
131
132 for (j = i; j > 0; j--) {
133 if (pmem_ranges[j-1].start_pfn <
134 pmem_ranges[j].start_pfn) {
135
136 break;
137 }
138 swap(pmem_ranges[j-1], pmem_ranges[j]);
139 }
140 }
141
142#ifndef CONFIG_SPARSEMEM
143 /*
144 * Throw out ranges that are too far apart (controlled by
145 * MAX_GAP).
146 */
147
148 for (i = 1; i < npmem_ranges; i++) {
149 if (pmem_ranges[i].start_pfn -
150 (pmem_ranges[i-1].start_pfn +
151 pmem_ranges[i-1].pages) > MAX_GAP) {
152 npmem_ranges = i;
153 printk("Large gap in memory detected (%ld pages). "
154 "Consider turning on CONFIG_SPARSEMEM\n",
155 pmem_ranges[i].start_pfn -
156 (pmem_ranges[i-1].start_pfn +
157 pmem_ranges[i-1].pages));
158 break;
159 }
160 }
161#endif
162
163 /* Print the memory ranges */
164 pr_info("Memory Ranges:\n");
165
166 for (i = 0; i < npmem_ranges; i++) {
167 struct resource *res = &sysram_resources[i];
168 unsigned long start;
169 unsigned long size;
170
171 size = (pmem_ranges[i].pages << PAGE_SHIFT);
172 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
173 pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
174 i, start, start + (size - 1), size >> 20);
175
176 /* request memory resource */
177 res->name = "System RAM";
178 res->start = start;
179 res->end = start + size - 1;
180 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
181 request_resource(&iomem_resource, res);
182 }
183
184 sysram_resource_count = npmem_ranges;
185
186 /*
187 * For 32 bit kernels we limit the amount of memory we can
188 * support, in order to preserve enough kernel address space
189 * for other purposes. For 64 bit kernels we don't normally
190 * limit the memory, but this mechanism can be used to
191 * artificially limit the amount of memory (and it is written
192 * to work with multiple memory ranges).
193 */
194
195 mem_limit_func(); /* check for "mem=" argument */
196
197 mem_max = 0;
198 for (i = 0; i < npmem_ranges; i++) {
199 unsigned long rsize;
200
201 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
202 if ((mem_max + rsize) > mem_limit) {
203 printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
204 if (mem_max == mem_limit)
205 npmem_ranges = i;
206 else {
207 pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT)
208 - (mem_max >> PAGE_SHIFT);
209 npmem_ranges = i + 1;
210 mem_max = mem_limit;
211 }
212 break;
213 }
214 mem_max += rsize;
215 }
216
217 printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
218
219#ifndef CONFIG_SPARSEMEM
220 /* Merge the ranges, keeping track of the holes */
221 {
222 unsigned long end_pfn;
223 unsigned long hole_pages;
224
225 npmem_holes = 0;
226 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
227 for (i = 1; i < npmem_ranges; i++) {
228
229 hole_pages = pmem_ranges[i].start_pfn - end_pfn;
230 if (hole_pages) {
231 pmem_holes[npmem_holes].start_pfn = end_pfn;
232 pmem_holes[npmem_holes++].pages = hole_pages;
233 end_pfn += hole_pages;
234 }
235 end_pfn += pmem_ranges[i].pages;
236 }
237
238 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
239 npmem_ranges = 1;
240 }
241#endif
242
243 /*
244 * Initialize and free the full range of memory in each range.
245 */
246
247 max_pfn = 0;
248 for (i = 0; i < npmem_ranges; i++) {
249 unsigned long start_pfn;
250 unsigned long npages;
251 unsigned long start;
252 unsigned long size;
253
254 start_pfn = pmem_ranges[i].start_pfn;
255 npages = pmem_ranges[i].pages;
256
257 start = start_pfn << PAGE_SHIFT;
258 size = npages << PAGE_SHIFT;
259
260 /* add system RAM memblock */
261 memblock_add(start, size);
262
263 if ((start_pfn + npages) > max_pfn)
264 max_pfn = start_pfn + npages;
265 }
266
267 /*
268 * We can't use memblock top-down allocations because we only
269 * created the initial mapping up to KERNEL_INITIAL_SIZE in
270 * the assembly bootup code.
271 */
272 memblock_set_bottom_up(true);
273
274 /* IOMMU is always used to access "high mem" on those boxes
275 * that can support enough mem that a PCI device couldn't
276 * directly DMA to any physical addresses.
277 * ISA DMA support will need to revisit this.
278 */
279 max_low_pfn = max_pfn;
280
281 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
282
283#define PDC_CONSOLE_IO_IODC_SIZE 32768
284
285 memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
286 PDC_CONSOLE_IO_IODC_SIZE));
287 memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
288 (unsigned long)(_end - KERNEL_BINARY_TEXT_START));
289
290#ifndef CONFIG_SPARSEMEM
291
292 /* reserve the holes */
293
294 for (i = 0; i < npmem_holes; i++) {
295 memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
296 (pmem_holes[i].pages << PAGE_SHIFT));
297 }
298#endif
299
300#ifdef CONFIG_BLK_DEV_INITRD
301 if (initrd_start) {
302 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
303 if (__pa(initrd_start) < mem_max) {
304 unsigned long initrd_reserve;
305
306 if (__pa(initrd_end) > mem_max) {
307 initrd_reserve = mem_max - __pa(initrd_start);
308 } else {
309 initrd_reserve = initrd_end - initrd_start;
310 }
311 initrd_below_start_ok = 1;
312 printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
313
314 memblock_reserve(__pa(initrd_start), initrd_reserve);
315 }
316 }
317#endif
318
319 data_resource.start = virt_to_phys(&data_start);
320 data_resource.end = virt_to_phys(_end) - 1;
321 code_resource.start = virt_to_phys(_text);
322 code_resource.end = virt_to_phys(&data_start)-1;
323
324 /* We don't know which region the kernel will be in, so try
325 * all of them.
326 */
327 for (i = 0; i < sysram_resource_count; i++) {
328 struct resource *res = &sysram_resources[i];
329 request_resource(res, &code_resource);
330 request_resource(res, &data_resource);
331 }
332 request_resource(&sysram_resources[0], &pdcdata_resource);
333
334 /* Initialize Page Deallocation Table (PDT) and check for bad memory. */
335 pdc_pdt_init();
336
337 memblock_allow_resize();
338 memblock_dump_all();
339}
340
341static bool kernel_set_to_readonly;
342
343static void __ref map_pages(unsigned long start_vaddr,
344 unsigned long start_paddr, unsigned long size,
345 pgprot_t pgprot, int force)
346{
347 pmd_t *pmd;
348 pte_t *pg_table;
349 unsigned long end_paddr;
350 unsigned long start_pmd;
351 unsigned long start_pte;
352 unsigned long tmp1;
353 unsigned long tmp2;
354 unsigned long address;
355 unsigned long vaddr;
356 unsigned long ro_start;
357 unsigned long ro_end;
358 unsigned long kernel_start, kernel_end;
359
360 ro_start = __pa((unsigned long)_text);
361 ro_end = __pa((unsigned long)&data_start);
362 kernel_start = __pa((unsigned long)&__init_begin);
363 kernel_end = __pa((unsigned long)&_end);
364
365 end_paddr = start_paddr + size;
366
367 /* for 2-level configuration PTRS_PER_PMD is 0 so start_pmd will be 0 */
368 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
369 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
370
371 address = start_paddr;
372 vaddr = start_vaddr;
373 while (address < end_paddr) {
374 pgd_t *pgd = pgd_offset_k(vaddr);
375 p4d_t *p4d = p4d_offset(pgd, vaddr);
376 pud_t *pud = pud_offset(p4d, vaddr);
377
378#if CONFIG_PGTABLE_LEVELS == 3
379 if (pud_none(*pud)) {
380 pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
381 PAGE_SIZE << PMD_TABLE_ORDER);
382 if (!pmd)
383 panic("pmd allocation failed.\n");
384 pud_populate(NULL, pud, pmd);
385 }
386#endif
387
388 pmd = pmd_offset(pud, vaddr);
389 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
390 if (pmd_none(*pmd)) {
391 pg_table = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
392 if (!pg_table)
393 panic("page table allocation failed\n");
394 pmd_populate_kernel(NULL, pmd, pg_table);
395 }
396
397 pg_table = pte_offset_kernel(pmd, vaddr);
398 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
399 pte_t pte;
400 pgprot_t prot;
401 bool huge = false;
402
403 if (force) {
404 prot = pgprot;
405 } else if (address < kernel_start || address >= kernel_end) {
406 /* outside kernel memory */
407 prot = PAGE_KERNEL;
408 } else if (!kernel_set_to_readonly) {
409 /* still initializing, allow writing to RO memory */
410 prot = PAGE_KERNEL_RWX;
411 huge = true;
412 } else if (address >= ro_start) {
413 /* Code (ro) and Data areas */
414 prot = (address < ro_end) ?
415 PAGE_KERNEL_EXEC : PAGE_KERNEL;
416 huge = true;
417 } else {
418 prot = PAGE_KERNEL;
419 }
420
421 pte = __mk_pte(address, prot);
422 if (huge)
423 pte = pte_mkhuge(pte);
424
425 if (address >= end_paddr)
426 break;
427
428 set_pte(pg_table, pte);
429
430 address += PAGE_SIZE;
431 vaddr += PAGE_SIZE;
432 }
433 start_pte = 0;
434
435 if (address >= end_paddr)
436 break;
437 }
438 start_pmd = 0;
439 }
440}
441
442void __init set_kernel_text_rw(int enable_read_write)
443{
444 unsigned long start = (unsigned long) __init_begin;
445 unsigned long end = (unsigned long) &data_start;
446
447 map_pages(start, __pa(start), end-start,
448 PAGE_KERNEL_RWX, enable_read_write ? 1:0);
449
450 /* force the kernel to see the new page table entries */
451 flush_cache_all();
452 flush_tlb_all();
453}
454
455void free_initmem(void)
456{
457 unsigned long init_begin = (unsigned long)__init_begin;
458 unsigned long init_end = (unsigned long)__init_end;
459 unsigned long kernel_end = (unsigned long)&_end;
460
461 /* Remap kernel text and data, but do not touch init section yet. */
462 map_pages(init_end, __pa(init_end), kernel_end - init_end,
463 PAGE_KERNEL, 0);
464
465 /* The init text pages are marked R-X. We have to
466 * flush the icache and mark them RW-
467 *
468 * Do a dummy remap of the data section first (the data
469 * section is already PAGE_KERNEL) to pull in the TLB entries
470 * for map_kernel */
471 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
472 PAGE_KERNEL_RWX, 1);
473 /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
474 * map_pages */
475 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
476 PAGE_KERNEL, 1);
477
478 /* force the kernel to see the new TLB entries */
479 __flush_tlb_range(0, init_begin, kernel_end);
480
481 /* finally dump all the instructions which were cached, since the
482 * pages are no-longer executable */
483 flush_icache_range(init_begin, init_end);
484
485 free_initmem_default(POISON_FREE_INITMEM);
486
487 /* set up a new led state on systems shipped LED State panel */
488 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
489}
490
491
492#ifdef CONFIG_STRICT_KERNEL_RWX
493void mark_rodata_ro(void)
494{
495 unsigned long start = (unsigned long) &__start_rodata;
496 unsigned long end = (unsigned long) &__end_rodata;
497
498 pr_info("Write protecting the kernel read-only data: %luk\n",
499 (end - start) >> 10);
500
501 kernel_set_to_readonly = true;
502 map_pages(start, __pa(start), end - start, PAGE_KERNEL, 0);
503
504 /* force the kernel to see the new page table entries */
505 flush_cache_all();
506 flush_tlb_all();
507}
508#endif
509
510
511/*
512 * Just an arbitrary offset to serve as a "hole" between mapping areas
513 * (between top of physical memory and a potential pcxl dma mapping
514 * area, and below the vmalloc mapping area).
515 *
516 * The current 32K value just means that there will be a 32K "hole"
517 * between mapping areas. That means that any out-of-bounds memory
518 * accesses will hopefully be caught. The vmalloc() routines leaves
519 * a hole of 4kB between each vmalloced area for the same reason.
520 */
521
522 /* Leave room for gateway page expansion */
523#if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
524#error KERNEL_MAP_START is in gateway reserved region
525#endif
526#define MAP_START (KERNEL_MAP_START)
527
528#define VM_MAP_OFFSET (32*1024)
529#define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
530 & ~(VM_MAP_OFFSET-1)))
531
532void *parisc_vmalloc_start __ro_after_init;
533EXPORT_SYMBOL(parisc_vmalloc_start);
534
535void __init mem_init(void)
536{
537 /* Do sanity checks on IPC (compat) structures */
538 BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
539#ifndef CONFIG_64BIT
540 BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
541 BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
542 BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
543#endif
544#ifdef CONFIG_COMPAT
545 BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
546 BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
547 BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
548 BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
549#endif
550
551 /* Do sanity checks on page table constants */
552 BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
553 BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
554 BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
555 BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
556 > BITS_PER_LONG);
557#if CONFIG_PGTABLE_LEVELS == 3
558 BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PMD);
559#else
560 BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PGD);
561#endif
562
563#ifdef CONFIG_64BIT
564 /* avoid ldil_%L() asm statements to sign-extend into upper 32-bits */
565 BUILD_BUG_ON(__PAGE_OFFSET >= 0x80000000);
566 BUILD_BUG_ON(TMPALIAS_MAP_START >= 0x80000000);
567#endif
568
569 high_memory = __va((max_pfn << PAGE_SHIFT));
570 set_max_mapnr(max_low_pfn);
571 memblock_free_all();
572
573#ifdef CONFIG_PA11
574 if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
575 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
576 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
577 + PCXL_DMA_MAP_SIZE);
578 } else
579#endif
580 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
581
582#if 0
583 /*
584 * Do not expose the virtual kernel memory layout to userspace.
585 * But keep code for debugging purposes.
586 */
587 printk("virtual kernel memory layout:\n"
588 " vmalloc : 0x%px - 0x%px (%4ld MB)\n"
589 " fixmap : 0x%px - 0x%px (%4ld kB)\n"
590 " memory : 0x%px - 0x%px (%4ld MB)\n"
591 " .init : 0x%px - 0x%px (%4ld kB)\n"
592 " .data : 0x%px - 0x%px (%4ld kB)\n"
593 " .text : 0x%px - 0x%px (%4ld kB)\n",
594
595 (void*)VMALLOC_START, (void*)VMALLOC_END,
596 (VMALLOC_END - VMALLOC_START) >> 20,
597
598 (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE),
599 (unsigned long)(FIXMAP_SIZE / 1024),
600
601 __va(0), high_memory,
602 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
603
604 __init_begin, __init_end,
605 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
606
607 _etext, _edata,
608 ((unsigned long)_edata - (unsigned long)_etext) >> 10,
609
610 _text, _etext,
611 ((unsigned long)_etext - (unsigned long)_text) >> 10);
612#endif
613}
614
615unsigned long *empty_zero_page __ro_after_init;
616EXPORT_SYMBOL(empty_zero_page);
617
618/*
619 * pagetable_init() sets up the page tables
620 *
621 * Note that gateway_init() places the Linux gateway page at page 0.
622 * Since gateway pages cannot be dereferenced this has the desirable
623 * side effect of trapping those pesky NULL-reference errors in the
624 * kernel.
625 */
626static void __init pagetable_init(void)
627{
628 int range;
629
630 /* Map each physical memory range to its kernel vaddr */
631
632 for (range = 0; range < npmem_ranges; range++) {
633 unsigned long start_paddr;
634 unsigned long size;
635
636 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
637 size = pmem_ranges[range].pages << PAGE_SHIFT;
638
639 map_pages((unsigned long)__va(start_paddr), start_paddr,
640 size, PAGE_KERNEL, 0);
641 }
642
643#ifdef CONFIG_BLK_DEV_INITRD
644 if (initrd_end && initrd_end > mem_limit) {
645 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
646 map_pages(initrd_start, __pa(initrd_start),
647 initrd_end - initrd_start, PAGE_KERNEL, 0);
648 }
649#endif
650
651 empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
652 if (!empty_zero_page)
653 panic("zero page allocation failed.\n");
654
655}
656
657static void __init gateway_init(void)
658{
659 unsigned long linux_gateway_page_addr;
660 /* FIXME: This is 'const' in order to trick the compiler
661 into not treating it as DP-relative data. */
662 extern void * const linux_gateway_page;
663
664 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
665
666 /*
667 * Setup Linux Gateway page.
668 *
669 * The Linux gateway page will reside in kernel space (on virtual
670 * page 0), so it doesn't need to be aliased into user space.
671 */
672
673 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
674 PAGE_SIZE, PAGE_GATEWAY, 1);
675}
676
677static void __init fixmap_init(void)
678{
679 unsigned long addr = FIXMAP_START;
680 unsigned long end = FIXMAP_START + FIXMAP_SIZE;
681 pgd_t *pgd = pgd_offset_k(addr);
682 p4d_t *p4d = p4d_offset(pgd, addr);
683 pud_t *pud = pud_offset(p4d, addr);
684 pmd_t *pmd;
685
686 BUILD_BUG_ON(FIXMAP_SIZE > PMD_SIZE);
687
688#if CONFIG_PGTABLE_LEVELS == 3
689 if (pud_none(*pud)) {
690 pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
691 PAGE_SIZE << PMD_TABLE_ORDER);
692 if (!pmd)
693 panic("fixmap: pmd allocation failed.\n");
694 pud_populate(NULL, pud, pmd);
695 }
696#endif
697
698 pmd = pmd_offset(pud, addr);
699 do {
700 pte_t *pte = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
701 if (!pte)
702 panic("fixmap: pte allocation failed.\n");
703
704 pmd_populate_kernel(&init_mm, pmd, pte);
705
706 addr += PAGE_SIZE;
707 } while (addr < end);
708}
709
710static void __init parisc_bootmem_free(void)
711{
712 unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
713
714 max_zone_pfn[0] = memblock_end_of_DRAM();
715
716 free_area_init(max_zone_pfn);
717}
718
719void __init paging_init(void)
720{
721 setup_bootmem();
722 pagetable_init();
723 gateway_init();
724 fixmap_init();
725 flush_cache_all_local(); /* start with known state */
726 flush_tlb_all_local(NULL);
727
728 sparse_init();
729 parisc_bootmem_free();
730}
731
732static void alloc_btlb(unsigned long start, unsigned long end, int *slot,
733 unsigned long entry_info)
734{
735 const int slot_max = btlb_info.fixed_range_info.num_comb;
736 int min_num_pages = btlb_info.min_size;
737 unsigned long size;
738
739 /* map at minimum 4 pages */
740 if (min_num_pages < 4)
741 min_num_pages = 4;
742
743 size = HUGEPAGE_SIZE;
744 while (start < end && *slot < slot_max && size >= PAGE_SIZE) {
745 /* starting address must have same alignment as size! */
746 /* if correctly aligned and fits in double size, increase */
747 if (((start & (2 * size - 1)) == 0) &&
748 (end - start) >= (2 * size)) {
749 size <<= 1;
750 continue;
751 }
752 /* if current size alignment is too big, try smaller size */
753 if ((start & (size - 1)) != 0) {
754 size >>= 1;
755 continue;
756 }
757 if ((end - start) >= size) {
758 if ((size >> PAGE_SHIFT) >= min_num_pages)
759 pdc_btlb_insert(start >> PAGE_SHIFT, __pa(start) >> PAGE_SHIFT,
760 size >> PAGE_SHIFT, entry_info, *slot);
761 (*slot)++;
762 start += size;
763 continue;
764 }
765 size /= 2;
766 continue;
767 }
768}
769
770void btlb_init_per_cpu(void)
771{
772 unsigned long s, t, e;
773 int slot;
774
775 /* BTLBs are not available on 64-bit CPUs */
776 if (IS_ENABLED(CONFIG_PA20))
777 return;
778 else if (pdc_btlb_info(&btlb_info) < 0) {
779 memset(&btlb_info, 0, sizeof btlb_info);
780 }
781
782 /* insert BLTLBs for code and data segments */
783 s = (uintptr_t) dereference_function_descriptor(&_stext);
784 e = (uintptr_t) dereference_function_descriptor(&_etext);
785 t = (uintptr_t) dereference_function_descriptor(&_sdata);
786 BUG_ON(t != e);
787
788 /* code segments */
789 slot = 0;
790 alloc_btlb(s, e, &slot, 0x13800000);
791
792 /* sanity check */
793 t = (uintptr_t) dereference_function_descriptor(&_edata);
794 e = (uintptr_t) dereference_function_descriptor(&__bss_start);
795 BUG_ON(t != e);
796
797 /* data segments */
798 s = (uintptr_t) dereference_function_descriptor(&_sdata);
799 e = (uintptr_t) dereference_function_descriptor(&__bss_stop);
800 alloc_btlb(s, e, &slot, 0x11800000);
801}
802
803#ifdef CONFIG_PA20
804
805/*
806 * Currently, all PA20 chips have 18 bit protection IDs, which is the
807 * limiting factor (space ids are 32 bits).
808 */
809
810#define NR_SPACE_IDS 262144
811
812#else
813
814/*
815 * Currently we have a one-to-one relationship between space IDs and
816 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
817 * support 15 bit protection IDs, so that is the limiting factor.
818 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
819 * probably not worth the effort for a special case here.
820 */
821
822#define NR_SPACE_IDS 32768
823
824#endif /* !CONFIG_PA20 */
825
826#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
827#define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
828
829static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
830static unsigned long dirty_space_id[SID_ARRAY_SIZE];
831static unsigned long space_id_index;
832static unsigned long free_space_ids = NR_SPACE_IDS - 1;
833static unsigned long dirty_space_ids;
834
835static DEFINE_SPINLOCK(sid_lock);
836
837unsigned long alloc_sid(void)
838{
839 unsigned long index;
840
841 spin_lock(&sid_lock);
842
843 if (free_space_ids == 0) {
844 if (dirty_space_ids != 0) {
845 spin_unlock(&sid_lock);
846 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
847 spin_lock(&sid_lock);
848 }
849 BUG_ON(free_space_ids == 0);
850 }
851
852 free_space_ids--;
853
854 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
855 space_id[BIT_WORD(index)] |= BIT_MASK(index);
856 space_id_index = index;
857
858 spin_unlock(&sid_lock);
859
860 return index << SPACEID_SHIFT;
861}
862
863void free_sid(unsigned long spaceid)
864{
865 unsigned long index = spaceid >> SPACEID_SHIFT;
866 unsigned long *dirty_space_offset, mask;
867
868 dirty_space_offset = &dirty_space_id[BIT_WORD(index)];
869 mask = BIT_MASK(index);
870
871 spin_lock(&sid_lock);
872
873 BUG_ON(*dirty_space_offset & mask); /* attempt to free space id twice */
874
875 *dirty_space_offset |= mask;
876 dirty_space_ids++;
877
878 spin_unlock(&sid_lock);
879}
880
881
882#ifdef CONFIG_SMP
883static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
884{
885 int i;
886
887 /* NOTE: sid_lock must be held upon entry */
888
889 *ndirtyptr = dirty_space_ids;
890 if (dirty_space_ids != 0) {
891 for (i = 0; i < SID_ARRAY_SIZE; i++) {
892 dirty_array[i] = dirty_space_id[i];
893 dirty_space_id[i] = 0;
894 }
895 dirty_space_ids = 0;
896 }
897
898 return;
899}
900
901static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
902{
903 int i;
904
905 /* NOTE: sid_lock must be held upon entry */
906
907 if (ndirty != 0) {
908 for (i = 0; i < SID_ARRAY_SIZE; i++) {
909 space_id[i] ^= dirty_array[i];
910 }
911
912 free_space_ids += ndirty;
913 space_id_index = 0;
914 }
915}
916
917#else /* CONFIG_SMP */
918
919static void recycle_sids(void)
920{
921 int i;
922
923 /* NOTE: sid_lock must be held upon entry */
924
925 if (dirty_space_ids != 0) {
926 for (i = 0; i < SID_ARRAY_SIZE; i++) {
927 space_id[i] ^= dirty_space_id[i];
928 dirty_space_id[i] = 0;
929 }
930
931 free_space_ids += dirty_space_ids;
932 dirty_space_ids = 0;
933 space_id_index = 0;
934 }
935}
936#endif
937
938/*
939 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
940 * purged, we can safely reuse the space ids that were released but
941 * not flushed from the tlb.
942 */
943
944#ifdef CONFIG_SMP
945
946static unsigned long recycle_ndirty;
947static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
948static unsigned int recycle_inuse;
949
950void flush_tlb_all(void)
951{
952 int do_recycle;
953
954 do_recycle = 0;
955 spin_lock(&sid_lock);
956 __inc_irq_stat(irq_tlb_count);
957 if (dirty_space_ids > RECYCLE_THRESHOLD) {
958 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
959 get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
960 recycle_inuse++;
961 do_recycle++;
962 }
963 spin_unlock(&sid_lock);
964 on_each_cpu(flush_tlb_all_local, NULL, 1);
965 if (do_recycle) {
966 spin_lock(&sid_lock);
967 recycle_sids(recycle_ndirty,recycle_dirty_array);
968 recycle_inuse = 0;
969 spin_unlock(&sid_lock);
970 }
971}
972#else
973void flush_tlb_all(void)
974{
975 spin_lock(&sid_lock);
976 __inc_irq_stat(irq_tlb_count);
977 flush_tlb_all_local(NULL);
978 recycle_sids();
979 spin_unlock(&sid_lock);
980}
981#endif
982
983static const pgprot_t protection_map[16] = {
984 [VM_NONE] = PAGE_NONE,
985 [VM_READ] = PAGE_READONLY,
986 [VM_WRITE] = PAGE_NONE,
987 [VM_WRITE | VM_READ] = PAGE_READONLY,
988 [VM_EXEC] = PAGE_EXECREAD,
989 [VM_EXEC | VM_READ] = PAGE_EXECREAD,
990 [VM_EXEC | VM_WRITE] = PAGE_EXECREAD,
991 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_EXECREAD,
992 [VM_SHARED] = PAGE_NONE,
993 [VM_SHARED | VM_READ] = PAGE_READONLY,
994 [VM_SHARED | VM_WRITE] = PAGE_WRITEONLY,
995 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
996 [VM_SHARED | VM_EXEC] = PAGE_EXECREAD,
997 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_EXECREAD,
998 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_RWX,
999 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_RWX
1000};
1001DECLARE_VM_GET_PAGE_PROT
1002
1003#ifdef CONFIG_EXECMEM
1004static struct execmem_info execmem_info __ro_after_init;
1005
1006struct execmem_info __init *execmem_arch_setup(void)
1007{
1008 execmem_info = (struct execmem_info){
1009 .ranges = {
1010 [EXECMEM_DEFAULT] = {
1011 .start = VMALLOC_START,
1012 .end = VMALLOC_END,
1013 .pgprot = PAGE_KERNEL_RWX,
1014 .alignment = 1,
1015 },
1016 },
1017 };
1018
1019 return &execmem_info;
1020}
1021#endif /* CONFIG_EXECMEM */
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
36extern int data_start;
37extern void parisc_kernel_start(void); /* Kernel entry point in head.S */
38
39#if CONFIG_PGTABLE_LEVELS == 3
40pmd_t pmd0[PTRS_PER_PMD] __section(".data..vm0.pmd") __attribute__ ((aligned(PAGE_SIZE)));
41#endif
42
43pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".data..vm0.pgd") __attribute__ ((aligned(PAGE_SIZE)));
44pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __section(".data..vm0.pte") __attribute__ ((aligned(PAGE_SIZE)));
45
46static struct resource data_resource = {
47 .name = "Kernel data",
48 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
49};
50
51static struct resource code_resource = {
52 .name = "Kernel code",
53 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
54};
55
56static struct resource pdcdata_resource = {
57 .name = "PDC data (Page Zero)",
58 .start = 0,
59 .end = 0x9ff,
60 .flags = IORESOURCE_BUSY | IORESOURCE_MEM,
61};
62
63static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __ro_after_init;
64
65/* The following array is initialized from the firmware specific
66 * information retrieved in kernel/inventory.c.
67 */
68
69physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __initdata;
70int npmem_ranges __initdata;
71
72#ifdef CONFIG_64BIT
73#define MAX_MEM (1UL << MAX_PHYSMEM_BITS)
74#else /* !CONFIG_64BIT */
75#define MAX_MEM (3584U*1024U*1024U)
76#endif /* !CONFIG_64BIT */
77
78static unsigned long mem_limit __read_mostly = MAX_MEM;
79
80static void __init mem_limit_func(void)
81{
82 char *cp, *end;
83 unsigned long limit;
84
85 /* We need this before __setup() functions are called */
86
87 limit = MAX_MEM;
88 for (cp = boot_command_line; *cp; ) {
89 if (memcmp(cp, "mem=", 4) == 0) {
90 cp += 4;
91 limit = memparse(cp, &end);
92 if (end != cp)
93 break;
94 cp = end;
95 } else {
96 while (*cp != ' ' && *cp)
97 ++cp;
98 while (*cp == ' ')
99 ++cp;
100 }
101 }
102
103 if (limit < mem_limit)
104 mem_limit = limit;
105}
106
107#define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
108
109static void __init setup_bootmem(void)
110{
111 unsigned long mem_max;
112#ifndef CONFIG_SPARSEMEM
113 physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
114 int npmem_holes;
115#endif
116 int i, sysram_resource_count;
117
118 disable_sr_hashing(); /* Turn off space register hashing */
119
120 /*
121 * Sort the ranges. Since the number of ranges is typically
122 * small, and performance is not an issue here, just do
123 * a simple insertion sort.
124 */
125
126 for (i = 1; i < npmem_ranges; i++) {
127 int j;
128
129 for (j = i; j > 0; j--) {
130 if (pmem_ranges[j-1].start_pfn <
131 pmem_ranges[j].start_pfn) {
132
133 break;
134 }
135 swap(pmem_ranges[j-1], pmem_ranges[j]);
136 }
137 }
138
139#ifndef CONFIG_SPARSEMEM
140 /*
141 * Throw out ranges that are too far apart (controlled by
142 * MAX_GAP).
143 */
144
145 for (i = 1; i < npmem_ranges; i++) {
146 if (pmem_ranges[i].start_pfn -
147 (pmem_ranges[i-1].start_pfn +
148 pmem_ranges[i-1].pages) > MAX_GAP) {
149 npmem_ranges = i;
150 printk("Large gap in memory detected (%ld pages). "
151 "Consider turning on CONFIG_SPARSEMEM\n",
152 pmem_ranges[i].start_pfn -
153 (pmem_ranges[i-1].start_pfn +
154 pmem_ranges[i-1].pages));
155 break;
156 }
157 }
158#endif
159
160 /* Print the memory ranges */
161 pr_info("Memory Ranges:\n");
162
163 for (i = 0; i < npmem_ranges; i++) {
164 struct resource *res = &sysram_resources[i];
165 unsigned long start;
166 unsigned long size;
167
168 size = (pmem_ranges[i].pages << PAGE_SHIFT);
169 start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
170 pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
171 i, start, start + (size - 1), size >> 20);
172
173 /* request memory resource */
174 res->name = "System RAM";
175 res->start = start;
176 res->end = start + size - 1;
177 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
178 request_resource(&iomem_resource, res);
179 }
180
181 sysram_resource_count = npmem_ranges;
182
183 /*
184 * For 32 bit kernels we limit the amount of memory we can
185 * support, in order to preserve enough kernel address space
186 * for other purposes. For 64 bit kernels we don't normally
187 * limit the memory, but this mechanism can be used to
188 * artificially limit the amount of memory (and it is written
189 * to work with multiple memory ranges).
190 */
191
192 mem_limit_func(); /* check for "mem=" argument */
193
194 mem_max = 0;
195 for (i = 0; i < npmem_ranges; i++) {
196 unsigned long rsize;
197
198 rsize = pmem_ranges[i].pages << PAGE_SHIFT;
199 if ((mem_max + rsize) > mem_limit) {
200 printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
201 if (mem_max == mem_limit)
202 npmem_ranges = i;
203 else {
204 pmem_ranges[i].pages = (mem_limit >> PAGE_SHIFT)
205 - (mem_max >> PAGE_SHIFT);
206 npmem_ranges = i + 1;
207 mem_max = mem_limit;
208 }
209 break;
210 }
211 mem_max += rsize;
212 }
213
214 printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
215
216#ifndef CONFIG_SPARSEMEM
217 /* Merge the ranges, keeping track of the holes */
218 {
219 unsigned long end_pfn;
220 unsigned long hole_pages;
221
222 npmem_holes = 0;
223 end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
224 for (i = 1; i < npmem_ranges; i++) {
225
226 hole_pages = pmem_ranges[i].start_pfn - end_pfn;
227 if (hole_pages) {
228 pmem_holes[npmem_holes].start_pfn = end_pfn;
229 pmem_holes[npmem_holes++].pages = hole_pages;
230 end_pfn += hole_pages;
231 }
232 end_pfn += pmem_ranges[i].pages;
233 }
234
235 pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
236 npmem_ranges = 1;
237 }
238#endif
239
240 /*
241 * Initialize and free the full range of memory in each range.
242 */
243
244 max_pfn = 0;
245 for (i = 0; i < npmem_ranges; i++) {
246 unsigned long start_pfn;
247 unsigned long npages;
248 unsigned long start;
249 unsigned long size;
250
251 start_pfn = pmem_ranges[i].start_pfn;
252 npages = pmem_ranges[i].pages;
253
254 start = start_pfn << PAGE_SHIFT;
255 size = npages << PAGE_SHIFT;
256
257 /* add system RAM memblock */
258 memblock_add(start, size);
259
260 if ((start_pfn + npages) > max_pfn)
261 max_pfn = start_pfn + npages;
262 }
263
264 /*
265 * We can't use memblock top-down allocations because we only
266 * created the initial mapping up to KERNEL_INITIAL_SIZE in
267 * the assembly bootup code.
268 */
269 memblock_set_bottom_up(true);
270
271 /* IOMMU is always used to access "high mem" on those boxes
272 * that can support enough mem that a PCI device couldn't
273 * directly DMA to any physical addresses.
274 * ISA DMA support will need to revisit this.
275 */
276 max_low_pfn = max_pfn;
277
278 /* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
279
280#define PDC_CONSOLE_IO_IODC_SIZE 32768
281
282 memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
283 PDC_CONSOLE_IO_IODC_SIZE));
284 memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
285 (unsigned long)(_end - KERNEL_BINARY_TEXT_START));
286
287#ifndef CONFIG_SPARSEMEM
288
289 /* reserve the holes */
290
291 for (i = 0; i < npmem_holes; i++) {
292 memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
293 (pmem_holes[i].pages << PAGE_SHIFT));
294 }
295#endif
296
297#ifdef CONFIG_BLK_DEV_INITRD
298 if (initrd_start) {
299 printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
300 if (__pa(initrd_start) < mem_max) {
301 unsigned long initrd_reserve;
302
303 if (__pa(initrd_end) > mem_max) {
304 initrd_reserve = mem_max - __pa(initrd_start);
305 } else {
306 initrd_reserve = initrd_end - initrd_start;
307 }
308 initrd_below_start_ok = 1;
309 printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
310
311 memblock_reserve(__pa(initrd_start), initrd_reserve);
312 }
313 }
314#endif
315
316 data_resource.start = virt_to_phys(&data_start);
317 data_resource.end = virt_to_phys(_end) - 1;
318 code_resource.start = virt_to_phys(_text);
319 code_resource.end = virt_to_phys(&data_start)-1;
320
321 /* We don't know which region the kernel will be in, so try
322 * all of them.
323 */
324 for (i = 0; i < sysram_resource_count; i++) {
325 struct resource *res = &sysram_resources[i];
326 request_resource(res, &code_resource);
327 request_resource(res, &data_resource);
328 }
329 request_resource(&sysram_resources[0], &pdcdata_resource);
330
331 /* Initialize Page Deallocation Table (PDT) and check for bad memory. */
332 pdc_pdt_init();
333
334 memblock_allow_resize();
335 memblock_dump_all();
336}
337
338static bool kernel_set_to_readonly;
339
340static void __ref map_pages(unsigned long start_vaddr,
341 unsigned long start_paddr, unsigned long size,
342 pgprot_t pgprot, int force)
343{
344 pmd_t *pmd;
345 pte_t *pg_table;
346 unsigned long end_paddr;
347 unsigned long start_pmd;
348 unsigned long start_pte;
349 unsigned long tmp1;
350 unsigned long tmp2;
351 unsigned long address;
352 unsigned long vaddr;
353 unsigned long ro_start;
354 unsigned long ro_end;
355 unsigned long kernel_start, kernel_end;
356
357 ro_start = __pa((unsigned long)_text);
358 ro_end = __pa((unsigned long)&data_start);
359 kernel_start = __pa((unsigned long)&__init_begin);
360 kernel_end = __pa((unsigned long)&_end);
361
362 end_paddr = start_paddr + size;
363
364 /* for 2-level configuration PTRS_PER_PMD is 0 so start_pmd will be 0 */
365 start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
366 start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
367
368 address = start_paddr;
369 vaddr = start_vaddr;
370 while (address < end_paddr) {
371 pgd_t *pgd = pgd_offset_k(vaddr);
372 p4d_t *p4d = p4d_offset(pgd, vaddr);
373 pud_t *pud = pud_offset(p4d, vaddr);
374
375#if CONFIG_PGTABLE_LEVELS == 3
376 if (pud_none(*pud)) {
377 pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
378 PAGE_SIZE << PMD_TABLE_ORDER);
379 if (!pmd)
380 panic("pmd allocation failed.\n");
381 pud_populate(NULL, pud, pmd);
382 }
383#endif
384
385 pmd = pmd_offset(pud, vaddr);
386 for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
387 if (pmd_none(*pmd)) {
388 pg_table = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
389 if (!pg_table)
390 panic("page table allocation failed\n");
391 pmd_populate_kernel(NULL, pmd, pg_table);
392 }
393
394 pg_table = pte_offset_kernel(pmd, vaddr);
395 for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
396 pte_t pte;
397 pgprot_t prot;
398 bool huge = false;
399
400 if (force) {
401 prot = pgprot;
402 } else if (address < kernel_start || address >= kernel_end) {
403 /* outside kernel memory */
404 prot = PAGE_KERNEL;
405 } else if (!kernel_set_to_readonly) {
406 /* still initializing, allow writing to RO memory */
407 prot = PAGE_KERNEL_RWX;
408 huge = true;
409 } else if (address >= ro_start) {
410 /* Code (ro) and Data areas */
411 prot = (address < ro_end) ?
412 PAGE_KERNEL_EXEC : PAGE_KERNEL;
413 huge = true;
414 } else {
415 prot = PAGE_KERNEL;
416 }
417
418 pte = __mk_pte(address, prot);
419 if (huge)
420 pte = pte_mkhuge(pte);
421
422 if (address >= end_paddr)
423 break;
424
425 set_pte(pg_table, pte);
426
427 address += PAGE_SIZE;
428 vaddr += PAGE_SIZE;
429 }
430 start_pte = 0;
431
432 if (address >= end_paddr)
433 break;
434 }
435 start_pmd = 0;
436 }
437}
438
439void __init set_kernel_text_rw(int enable_read_write)
440{
441 unsigned long start = (unsigned long) __init_begin;
442 unsigned long end = (unsigned long) &data_start;
443
444 map_pages(start, __pa(start), end-start,
445 PAGE_KERNEL_RWX, enable_read_write ? 1:0);
446
447 /* force the kernel to see the new page table entries */
448 flush_cache_all();
449 flush_tlb_all();
450}
451
452void free_initmem(void)
453{
454 unsigned long init_begin = (unsigned long)__init_begin;
455 unsigned long init_end = (unsigned long)__init_end;
456 unsigned long kernel_end = (unsigned long)&_end;
457
458 /* Remap kernel text and data, but do not touch init section yet. */
459 kernel_set_to_readonly = true;
460 map_pages(init_end, __pa(init_end), kernel_end - init_end,
461 PAGE_KERNEL, 0);
462
463 /* The init text pages are marked R-X. We have to
464 * flush the icache and mark them RW-
465 *
466 * Do a dummy remap of the data section first (the data
467 * section is already PAGE_KERNEL) to pull in the TLB entries
468 * for map_kernel */
469 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
470 PAGE_KERNEL_RWX, 1);
471 /* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
472 * map_pages */
473 map_pages(init_begin, __pa(init_begin), init_end - init_begin,
474 PAGE_KERNEL, 1);
475
476 /* force the kernel to see the new TLB entries */
477 __flush_tlb_range(0, init_begin, kernel_end);
478
479 /* finally dump all the instructions which were cached, since the
480 * pages are no-longer executable */
481 flush_icache_range(init_begin, init_end);
482
483 free_initmem_default(POISON_FREE_INITMEM);
484
485 /* set up a new led state on systems shipped LED State panel */
486 pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
487}
488
489
490#ifdef CONFIG_STRICT_KERNEL_RWX
491void mark_rodata_ro(void)
492{
493 /* rodata memory was already mapped with KERNEL_RO access rights by
494 pagetable_init() and map_pages(). No need to do additional stuff here */
495 unsigned long roai_size = __end_ro_after_init - __start_ro_after_init;
496
497 pr_info("Write protected read-only-after-init data: %luk\n", roai_size >> 10);
498}
499#endif
500
501
502/*
503 * Just an arbitrary offset to serve as a "hole" between mapping areas
504 * (between top of physical memory and a potential pcxl dma mapping
505 * area, and below the vmalloc mapping area).
506 *
507 * The current 32K value just means that there will be a 32K "hole"
508 * between mapping areas. That means that any out-of-bounds memory
509 * accesses will hopefully be caught. The vmalloc() routines leaves
510 * a hole of 4kB between each vmalloced area for the same reason.
511 */
512
513 /* Leave room for gateway page expansion */
514#if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
515#error KERNEL_MAP_START is in gateway reserved region
516#endif
517#define MAP_START (KERNEL_MAP_START)
518
519#define VM_MAP_OFFSET (32*1024)
520#define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
521 & ~(VM_MAP_OFFSET-1)))
522
523void *parisc_vmalloc_start __ro_after_init;
524EXPORT_SYMBOL(parisc_vmalloc_start);
525
526#ifdef CONFIG_PA11
527unsigned long pcxl_dma_start __ro_after_init;
528#endif
529
530void __init mem_init(void)
531{
532 /* Do sanity checks on IPC (compat) structures */
533 BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
534#ifndef CONFIG_64BIT
535 BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
536 BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
537 BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
538#endif
539#ifdef CONFIG_COMPAT
540 BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
541 BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
542 BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
543 BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
544#endif
545
546 /* Do sanity checks on page table constants */
547 BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
548 BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
549 BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
550 BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
551 > BITS_PER_LONG);
552#if CONFIG_PGTABLE_LEVELS == 3
553 BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PMD);
554#else
555 BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PGD);
556#endif
557
558#ifdef CONFIG_64BIT
559 /* avoid ldil_%L() asm statements to sign-extend into upper 32-bits */
560 BUILD_BUG_ON(__PAGE_OFFSET >= 0x80000000);
561 BUILD_BUG_ON(TMPALIAS_MAP_START >= 0x80000000);
562#endif
563
564 high_memory = __va((max_pfn << PAGE_SHIFT));
565 set_max_mapnr(max_low_pfn);
566 memblock_free_all();
567
568#ifdef CONFIG_PA11
569 if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
570 pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
571 parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
572 + PCXL_DMA_MAP_SIZE);
573 } else
574#endif
575 parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
576
577#if 0
578 /*
579 * Do not expose the virtual kernel memory layout to userspace.
580 * But keep code for debugging purposes.
581 */
582 printk("virtual kernel memory layout:\n"
583 " vmalloc : 0x%px - 0x%px (%4ld MB)\n"
584 " fixmap : 0x%px - 0x%px (%4ld kB)\n"
585 " memory : 0x%px - 0x%px (%4ld MB)\n"
586 " .init : 0x%px - 0x%px (%4ld kB)\n"
587 " .data : 0x%px - 0x%px (%4ld kB)\n"
588 " .text : 0x%px - 0x%px (%4ld kB)\n",
589
590 (void*)VMALLOC_START, (void*)VMALLOC_END,
591 (VMALLOC_END - VMALLOC_START) >> 20,
592
593 (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE),
594 (unsigned long)(FIXMAP_SIZE / 1024),
595
596 __va(0), high_memory,
597 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
598
599 __init_begin, __init_end,
600 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
601
602 _etext, _edata,
603 ((unsigned long)_edata - (unsigned long)_etext) >> 10,
604
605 _text, _etext,
606 ((unsigned long)_etext - (unsigned long)_text) >> 10);
607#endif
608}
609
610unsigned long *empty_zero_page __ro_after_init;
611EXPORT_SYMBOL(empty_zero_page);
612
613/*
614 * pagetable_init() sets up the page tables
615 *
616 * Note that gateway_init() places the Linux gateway page at page 0.
617 * Since gateway pages cannot be dereferenced this has the desirable
618 * side effect of trapping those pesky NULL-reference errors in the
619 * kernel.
620 */
621static void __init pagetable_init(void)
622{
623 int range;
624
625 /* Map each physical memory range to its kernel vaddr */
626
627 for (range = 0; range < npmem_ranges; range++) {
628 unsigned long start_paddr;
629 unsigned long end_paddr;
630 unsigned long size;
631
632 start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
633 size = pmem_ranges[range].pages << PAGE_SHIFT;
634 end_paddr = start_paddr + size;
635
636 map_pages((unsigned long)__va(start_paddr), start_paddr,
637 size, PAGE_KERNEL, 0);
638 }
639
640#ifdef CONFIG_BLK_DEV_INITRD
641 if (initrd_end && initrd_end > mem_limit) {
642 printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
643 map_pages(initrd_start, __pa(initrd_start),
644 initrd_end - initrd_start, PAGE_KERNEL, 0);
645 }
646#endif
647
648 empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
649 if (!empty_zero_page)
650 panic("zero page allocation failed.\n");
651
652}
653
654static void __init gateway_init(void)
655{
656 unsigned long linux_gateway_page_addr;
657 /* FIXME: This is 'const' in order to trick the compiler
658 into not treating it as DP-relative data. */
659 extern void * const linux_gateway_page;
660
661 linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
662
663 /*
664 * Setup Linux Gateway page.
665 *
666 * The Linux gateway page will reside in kernel space (on virtual
667 * page 0), so it doesn't need to be aliased into user space.
668 */
669
670 map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
671 PAGE_SIZE, PAGE_GATEWAY, 1);
672}
673
674static void __init parisc_bootmem_free(void)
675{
676 unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
677
678 max_zone_pfn[0] = memblock_end_of_DRAM();
679
680 free_area_init(max_zone_pfn);
681}
682
683void __init paging_init(void)
684{
685 setup_bootmem();
686 pagetable_init();
687 gateway_init();
688 flush_cache_all_local(); /* start with known state */
689 flush_tlb_all_local(NULL);
690
691 sparse_init();
692 parisc_bootmem_free();
693}
694
695#ifdef CONFIG_PA20
696
697/*
698 * Currently, all PA20 chips have 18 bit protection IDs, which is the
699 * limiting factor (space ids are 32 bits).
700 */
701
702#define NR_SPACE_IDS 262144
703
704#else
705
706/*
707 * Currently we have a one-to-one relationship between space IDs and
708 * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
709 * support 15 bit protection IDs, so that is the limiting factor.
710 * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
711 * probably not worth the effort for a special case here.
712 */
713
714#define NR_SPACE_IDS 32768
715
716#endif /* !CONFIG_PA20 */
717
718#define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
719#define SID_ARRAY_SIZE (NR_SPACE_IDS / (8 * sizeof(long)))
720
721static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
722static unsigned long dirty_space_id[SID_ARRAY_SIZE];
723static unsigned long space_id_index;
724static unsigned long free_space_ids = NR_SPACE_IDS - 1;
725static unsigned long dirty_space_ids;
726
727static DEFINE_SPINLOCK(sid_lock);
728
729unsigned long alloc_sid(void)
730{
731 unsigned long index;
732
733 spin_lock(&sid_lock);
734
735 if (free_space_ids == 0) {
736 if (dirty_space_ids != 0) {
737 spin_unlock(&sid_lock);
738 flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
739 spin_lock(&sid_lock);
740 }
741 BUG_ON(free_space_ids == 0);
742 }
743
744 free_space_ids--;
745
746 index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
747 space_id[BIT_WORD(index)] |= BIT_MASK(index);
748 space_id_index = index;
749
750 spin_unlock(&sid_lock);
751
752 return index << SPACEID_SHIFT;
753}
754
755void free_sid(unsigned long spaceid)
756{
757 unsigned long index = spaceid >> SPACEID_SHIFT;
758 unsigned long *dirty_space_offset, mask;
759
760 dirty_space_offset = &dirty_space_id[BIT_WORD(index)];
761 mask = BIT_MASK(index);
762
763 spin_lock(&sid_lock);
764
765 BUG_ON(*dirty_space_offset & mask); /* attempt to free space id twice */
766
767 *dirty_space_offset |= mask;
768 dirty_space_ids++;
769
770 spin_unlock(&sid_lock);
771}
772
773
774#ifdef CONFIG_SMP
775static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
776{
777 int i;
778
779 /* NOTE: sid_lock must be held upon entry */
780
781 *ndirtyptr = dirty_space_ids;
782 if (dirty_space_ids != 0) {
783 for (i = 0; i < SID_ARRAY_SIZE; i++) {
784 dirty_array[i] = dirty_space_id[i];
785 dirty_space_id[i] = 0;
786 }
787 dirty_space_ids = 0;
788 }
789
790 return;
791}
792
793static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
794{
795 int i;
796
797 /* NOTE: sid_lock must be held upon entry */
798
799 if (ndirty != 0) {
800 for (i = 0; i < SID_ARRAY_SIZE; i++) {
801 space_id[i] ^= dirty_array[i];
802 }
803
804 free_space_ids += ndirty;
805 space_id_index = 0;
806 }
807}
808
809#else /* CONFIG_SMP */
810
811static void recycle_sids(void)
812{
813 int i;
814
815 /* NOTE: sid_lock must be held upon entry */
816
817 if (dirty_space_ids != 0) {
818 for (i = 0; i < SID_ARRAY_SIZE; i++) {
819 space_id[i] ^= dirty_space_id[i];
820 dirty_space_id[i] = 0;
821 }
822
823 free_space_ids += dirty_space_ids;
824 dirty_space_ids = 0;
825 space_id_index = 0;
826 }
827}
828#endif
829
830/*
831 * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
832 * purged, we can safely reuse the space ids that were released but
833 * not flushed from the tlb.
834 */
835
836#ifdef CONFIG_SMP
837
838static unsigned long recycle_ndirty;
839static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
840static unsigned int recycle_inuse;
841
842void flush_tlb_all(void)
843{
844 int do_recycle;
845
846 do_recycle = 0;
847 spin_lock(&sid_lock);
848 __inc_irq_stat(irq_tlb_count);
849 if (dirty_space_ids > RECYCLE_THRESHOLD) {
850 BUG_ON(recycle_inuse); /* FIXME: Use a semaphore/wait queue here */
851 get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
852 recycle_inuse++;
853 do_recycle++;
854 }
855 spin_unlock(&sid_lock);
856 on_each_cpu(flush_tlb_all_local, NULL, 1);
857 if (do_recycle) {
858 spin_lock(&sid_lock);
859 recycle_sids(recycle_ndirty,recycle_dirty_array);
860 recycle_inuse = 0;
861 spin_unlock(&sid_lock);
862 }
863}
864#else
865void flush_tlb_all(void)
866{
867 spin_lock(&sid_lock);
868 __inc_irq_stat(irq_tlb_count);
869 flush_tlb_all_local(NULL);
870 recycle_sids();
871 spin_unlock(&sid_lock);
872}
873#endif
874
875static const pgprot_t protection_map[16] = {
876 [VM_NONE] = PAGE_NONE,
877 [VM_READ] = PAGE_READONLY,
878 [VM_WRITE] = PAGE_NONE,
879 [VM_WRITE | VM_READ] = PAGE_READONLY,
880 [VM_EXEC] = PAGE_EXECREAD,
881 [VM_EXEC | VM_READ] = PAGE_EXECREAD,
882 [VM_EXEC | VM_WRITE] = PAGE_EXECREAD,
883 [VM_EXEC | VM_WRITE | VM_READ] = PAGE_EXECREAD,
884 [VM_SHARED] = PAGE_NONE,
885 [VM_SHARED | VM_READ] = PAGE_READONLY,
886 [VM_SHARED | VM_WRITE] = PAGE_WRITEONLY,
887 [VM_SHARED | VM_WRITE | VM_READ] = PAGE_SHARED,
888 [VM_SHARED | VM_EXEC] = PAGE_EXECREAD,
889 [VM_SHARED | VM_EXEC | VM_READ] = PAGE_EXECREAD,
890 [VM_SHARED | VM_EXEC | VM_WRITE] = PAGE_RWX,
891 [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = PAGE_RWX
892};
893DECLARE_VM_GET_PAGE_PROT