xref: /openbmc/linux/arch/parisc/mm/init.c (revision e5ef93d0)
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 #include <asm/asm-offsets.h>
36 
37 extern int  data_start;
38 extern void parisc_kernel_start(void);	/* Kernel entry point in head.S */
39 
40 #if CONFIG_PGTABLE_LEVELS == 3
41 pmd_t pmd0[PTRS_PER_PMD] __section(".data..vm0.pmd") __attribute__ ((aligned(PAGE_SIZE)));
42 #endif
43 
44 pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".data..vm0.pgd") __attribute__ ((aligned(PAGE_SIZE)));
45 pte_t pg0[PT_INITIAL * PTRS_PER_PTE] __section(".data..vm0.pte") __attribute__ ((aligned(PAGE_SIZE)));
46 
47 static struct resource data_resource = {
48 	.name	= "Kernel data",
49 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
50 };
51 
52 static struct resource code_resource = {
53 	.name	= "Kernel code",
54 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
55 };
56 
57 static struct resource pdcdata_resource = {
58 	.name	= "PDC data (Page Zero)",
59 	.start	= 0,
60 	.end	= 0x9ff,
61 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM,
62 };
63 
64 static struct resource sysram_resources[MAX_PHYSMEM_RANGES] __ro_after_init;
65 
66 /* The following array is initialized from the firmware specific
67  * information retrieved in kernel/inventory.c.
68  */
69 
70 physmem_range_t pmem_ranges[MAX_PHYSMEM_RANGES] __initdata;
71 int npmem_ranges __initdata;
72 
73 #ifdef CONFIG_64BIT
74 #define MAX_MEM         (1UL << MAX_PHYSMEM_BITS)
75 #else /* !CONFIG_64BIT */
76 #define MAX_MEM         (3584U*1024U*1024U)
77 #endif /* !CONFIG_64BIT */
78 
79 static unsigned long mem_limit __read_mostly = MAX_MEM;
80 
mem_limit_func(void)81 static void __init mem_limit_func(void)
82 {
83 	char *cp, *end;
84 	unsigned long limit;
85 
86 	/* We need this before __setup() functions are called */
87 
88 	limit = MAX_MEM;
89 	for (cp = boot_command_line; *cp; ) {
90 		if (memcmp(cp, "mem=", 4) == 0) {
91 			cp += 4;
92 			limit = memparse(cp, &end);
93 			if (end != cp)
94 				break;
95 			cp = end;
96 		} else {
97 			while (*cp != ' ' && *cp)
98 				++cp;
99 			while (*cp == ' ')
100 				++cp;
101 		}
102 	}
103 
104 	if (limit < mem_limit)
105 		mem_limit = limit;
106 }
107 
108 #define MAX_GAP (0x40000000UL >> PAGE_SHIFT)
109 
setup_bootmem(void)110 static void __init setup_bootmem(void)
111 {
112 	unsigned long mem_max;
113 #ifndef CONFIG_SPARSEMEM
114 	physmem_range_t pmem_holes[MAX_PHYSMEM_RANGES - 1];
115 	int npmem_holes;
116 #endif
117 	int i, sysram_resource_count;
118 
119 	disable_sr_hashing(); /* Turn off space register hashing */
120 
121 	/*
122 	 * Sort the ranges. Since the number of ranges is typically
123 	 * small, and performance is not an issue here, just do
124 	 * a simple insertion sort.
125 	 */
126 
127 	for (i = 1; i < npmem_ranges; i++) {
128 		int j;
129 
130 		for (j = i; j > 0; j--) {
131 			if (pmem_ranges[j-1].start_pfn <
132 			    pmem_ranges[j].start_pfn) {
133 
134 				break;
135 			}
136 			swap(pmem_ranges[j-1], pmem_ranges[j]);
137 		}
138 	}
139 
140 #ifndef CONFIG_SPARSEMEM
141 	/*
142 	 * Throw out ranges that are too far apart (controlled by
143 	 * MAX_GAP).
144 	 */
145 
146 	for (i = 1; i < npmem_ranges; i++) {
147 		if (pmem_ranges[i].start_pfn -
148 			(pmem_ranges[i-1].start_pfn +
149 			 pmem_ranges[i-1].pages) > MAX_GAP) {
150 			npmem_ranges = i;
151 			printk("Large gap in memory detected (%ld pages). "
152 			       "Consider turning on CONFIG_SPARSEMEM\n",
153 			       pmem_ranges[i].start_pfn -
154 			       (pmem_ranges[i-1].start_pfn +
155 			        pmem_ranges[i-1].pages));
156 			break;
157 		}
158 	}
159 #endif
160 
161 	/* Print the memory ranges */
162 	pr_info("Memory Ranges:\n");
163 
164 	for (i = 0; i < npmem_ranges; i++) {
165 		struct resource *res = &sysram_resources[i];
166 		unsigned long start;
167 		unsigned long size;
168 
169 		size = (pmem_ranges[i].pages << PAGE_SHIFT);
170 		start = (pmem_ranges[i].start_pfn << PAGE_SHIFT);
171 		pr_info("%2d) Start 0x%016lx End 0x%016lx Size %6ld MB\n",
172 			i, start, start + (size - 1), size >> 20);
173 
174 		/* request memory resource */
175 		res->name = "System RAM";
176 		res->start = start;
177 		res->end = start + size - 1;
178 		res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
179 		request_resource(&iomem_resource, res);
180 	}
181 
182 	sysram_resource_count = npmem_ranges;
183 
184 	/*
185 	 * For 32 bit kernels we limit the amount of memory we can
186 	 * support, in order to preserve enough kernel address space
187 	 * for other purposes. For 64 bit kernels we don't normally
188 	 * limit the memory, but this mechanism can be used to
189 	 * artificially limit the amount of memory (and it is written
190 	 * to work with multiple memory ranges).
191 	 */
192 
193 	mem_limit_func();       /* check for "mem=" argument */
194 
195 	mem_max = 0;
196 	for (i = 0; i < npmem_ranges; i++) {
197 		unsigned long rsize;
198 
199 		rsize = pmem_ranges[i].pages << PAGE_SHIFT;
200 		if ((mem_max + rsize) > mem_limit) {
201 			printk(KERN_WARNING "Memory truncated to %ld MB\n", mem_limit >> 20);
202 			if (mem_max == mem_limit)
203 				npmem_ranges = i;
204 			else {
205 				pmem_ranges[i].pages =   (mem_limit >> PAGE_SHIFT)
206 						       - (mem_max >> PAGE_SHIFT);
207 				npmem_ranges = i + 1;
208 				mem_max = mem_limit;
209 			}
210 			break;
211 		}
212 		mem_max += rsize;
213 	}
214 
215 	printk(KERN_INFO "Total Memory: %ld MB\n",mem_max >> 20);
216 
217 #ifndef CONFIG_SPARSEMEM
218 	/* Merge the ranges, keeping track of the holes */
219 	{
220 		unsigned long end_pfn;
221 		unsigned long hole_pages;
222 
223 		npmem_holes = 0;
224 		end_pfn = pmem_ranges[0].start_pfn + pmem_ranges[0].pages;
225 		for (i = 1; i < npmem_ranges; i++) {
226 
227 			hole_pages = pmem_ranges[i].start_pfn - end_pfn;
228 			if (hole_pages) {
229 				pmem_holes[npmem_holes].start_pfn = end_pfn;
230 				pmem_holes[npmem_holes++].pages = hole_pages;
231 				end_pfn += hole_pages;
232 			}
233 			end_pfn += pmem_ranges[i].pages;
234 		}
235 
236 		pmem_ranges[0].pages = end_pfn - pmem_ranges[0].start_pfn;
237 		npmem_ranges = 1;
238 	}
239 #endif
240 
241 	/*
242 	 * Initialize and free the full range of memory in each range.
243 	 */
244 
245 	max_pfn = 0;
246 	for (i = 0; i < npmem_ranges; i++) {
247 		unsigned long start_pfn;
248 		unsigned long npages;
249 		unsigned long start;
250 		unsigned long size;
251 
252 		start_pfn = pmem_ranges[i].start_pfn;
253 		npages = pmem_ranges[i].pages;
254 
255 		start = start_pfn << PAGE_SHIFT;
256 		size = npages << PAGE_SHIFT;
257 
258 		/* add system RAM memblock */
259 		memblock_add(start, size);
260 
261 		if ((start_pfn + npages) > max_pfn)
262 			max_pfn = start_pfn + npages;
263 	}
264 
265 	/*
266 	 * We can't use memblock top-down allocations because we only
267 	 * created the initial mapping up to KERNEL_INITIAL_SIZE in
268 	 * the assembly bootup code.
269 	 */
270 	memblock_set_bottom_up(true);
271 
272 	/* IOMMU is always used to access "high mem" on those boxes
273 	 * that can support enough mem that a PCI device couldn't
274 	 * directly DMA to any physical addresses.
275 	 * ISA DMA support will need to revisit this.
276 	 */
277 	max_low_pfn = max_pfn;
278 
279 	/* reserve PAGE0 pdc memory, kernel text/data/bss & bootmap */
280 
281 #define PDC_CONSOLE_IO_IODC_SIZE 32768
282 
283 	memblock_reserve(0UL, (unsigned long)(PAGE0->mem_free +
284 				PDC_CONSOLE_IO_IODC_SIZE));
285 	memblock_reserve(__pa(KERNEL_BINARY_TEXT_START),
286 			(unsigned long)(_end - KERNEL_BINARY_TEXT_START));
287 
288 #ifndef CONFIG_SPARSEMEM
289 
290 	/* reserve the holes */
291 
292 	for (i = 0; i < npmem_holes; i++) {
293 		memblock_reserve((pmem_holes[i].start_pfn << PAGE_SHIFT),
294 				(pmem_holes[i].pages << PAGE_SHIFT));
295 	}
296 #endif
297 
298 #ifdef CONFIG_BLK_DEV_INITRD
299 	if (initrd_start) {
300 		printk(KERN_INFO "initrd: %08lx-%08lx\n", initrd_start, initrd_end);
301 		if (__pa(initrd_start) < mem_max) {
302 			unsigned long initrd_reserve;
303 
304 			if (__pa(initrd_end) > mem_max) {
305 				initrd_reserve = mem_max - __pa(initrd_start);
306 			} else {
307 				initrd_reserve = initrd_end - initrd_start;
308 			}
309 			initrd_below_start_ok = 1;
310 			printk(KERN_INFO "initrd: reserving %08lx-%08lx (mem_max %08lx)\n", __pa(initrd_start), __pa(initrd_start) + initrd_reserve, mem_max);
311 
312 			memblock_reserve(__pa(initrd_start), initrd_reserve);
313 		}
314 	}
315 #endif
316 
317 	data_resource.start =  virt_to_phys(&data_start);
318 	data_resource.end = virt_to_phys(_end) - 1;
319 	code_resource.start = virt_to_phys(_text);
320 	code_resource.end = virt_to_phys(&data_start)-1;
321 
322 	/* We don't know which region the kernel will be in, so try
323 	 * all of them.
324 	 */
325 	for (i = 0; i < sysram_resource_count; i++) {
326 		struct resource *res = &sysram_resources[i];
327 		request_resource(res, &code_resource);
328 		request_resource(res, &data_resource);
329 	}
330 	request_resource(&sysram_resources[0], &pdcdata_resource);
331 
332 	/* Initialize Page Deallocation Table (PDT) and check for bad memory. */
333 	pdc_pdt_init();
334 
335 	memblock_allow_resize();
336 	memblock_dump_all();
337 }
338 
339 static bool kernel_set_to_readonly;
340 
map_pages(unsigned long start_vaddr,unsigned long start_paddr,unsigned long size,pgprot_t pgprot,int force)341 static void __ref map_pages(unsigned long start_vaddr,
342 			    unsigned long start_paddr, unsigned long size,
343 			    pgprot_t pgprot, int force)
344 {
345 	pmd_t *pmd;
346 	pte_t *pg_table;
347 	unsigned long end_paddr;
348 	unsigned long start_pmd;
349 	unsigned long start_pte;
350 	unsigned long tmp1;
351 	unsigned long tmp2;
352 	unsigned long address;
353 	unsigned long vaddr;
354 	unsigned long ro_start;
355 	unsigned long ro_end;
356 	unsigned long kernel_start, kernel_end;
357 
358 	ro_start = __pa((unsigned long)_text);
359 	ro_end   = __pa((unsigned long)&data_start);
360 	kernel_start = __pa((unsigned long)&__init_begin);
361 	kernel_end  = __pa((unsigned long)&_end);
362 
363 	end_paddr = start_paddr + size;
364 
365 	/* for 2-level configuration PTRS_PER_PMD is 0 so start_pmd will be 0 */
366 	start_pmd = ((start_vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1));
367 	start_pte = ((start_vaddr >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
368 
369 	address = start_paddr;
370 	vaddr = start_vaddr;
371 	while (address < end_paddr) {
372 		pgd_t *pgd = pgd_offset_k(vaddr);
373 		p4d_t *p4d = p4d_offset(pgd, vaddr);
374 		pud_t *pud = pud_offset(p4d, vaddr);
375 
376 #if CONFIG_PGTABLE_LEVELS == 3
377 		if (pud_none(*pud)) {
378 			pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
379 					     PAGE_SIZE << PMD_TABLE_ORDER);
380 			if (!pmd)
381 				panic("pmd allocation failed.\n");
382 			pud_populate(NULL, pud, pmd);
383 		}
384 #endif
385 
386 		pmd = pmd_offset(pud, vaddr);
387 		for (tmp1 = start_pmd; tmp1 < PTRS_PER_PMD; tmp1++, pmd++) {
388 			if (pmd_none(*pmd)) {
389 				pg_table = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
390 				if (!pg_table)
391 					panic("page table allocation failed\n");
392 				pmd_populate_kernel(NULL, pmd, pg_table);
393 			}
394 
395 			pg_table = pte_offset_kernel(pmd, vaddr);
396 			for (tmp2 = start_pte; tmp2 < PTRS_PER_PTE; tmp2++, pg_table++) {
397 				pte_t pte;
398 				pgprot_t prot;
399 				bool huge = false;
400 
401 				if (force) {
402 					prot = pgprot;
403 				} else if (address < kernel_start || address >= kernel_end) {
404 					/* outside kernel memory */
405 					prot = PAGE_KERNEL;
406 				} else if (!kernel_set_to_readonly) {
407 					/* still initializing, allow writing to RO memory */
408 					prot = PAGE_KERNEL_RWX;
409 					huge = true;
410 				} else if (address >= ro_start) {
411 					/* Code (ro) and Data areas */
412 					prot = (address < ro_end) ?
413 						PAGE_KERNEL_EXEC : PAGE_KERNEL;
414 					huge = true;
415 				} else {
416 					prot = PAGE_KERNEL;
417 				}
418 
419 				pte = __mk_pte(address, prot);
420 				if (huge)
421 					pte = pte_mkhuge(pte);
422 
423 				if (address >= end_paddr)
424 					break;
425 
426 				set_pte(pg_table, pte);
427 
428 				address += PAGE_SIZE;
429 				vaddr += PAGE_SIZE;
430 			}
431 			start_pte = 0;
432 
433 			if (address >= end_paddr)
434 			    break;
435 		}
436 		start_pmd = 0;
437 	}
438 }
439 
set_kernel_text_rw(int enable_read_write)440 void __init set_kernel_text_rw(int enable_read_write)
441 {
442 	unsigned long start = (unsigned long) __init_begin;
443 	unsigned long end   = (unsigned long) &data_start;
444 
445 	map_pages(start, __pa(start), end-start,
446 		PAGE_KERNEL_RWX, enable_read_write ? 1:0);
447 
448 	/* force the kernel to see the new page table entries */
449 	flush_cache_all();
450 	flush_tlb_all();
451 }
452 
free_initmem(void)453 void free_initmem(void)
454 {
455 	unsigned long init_begin = (unsigned long)__init_begin;
456 	unsigned long init_end = (unsigned long)__init_end;
457 	unsigned long kernel_end  = (unsigned long)&_end;
458 
459 	/* Remap kernel text and data, but do not touch init section yet. */
460 	kernel_set_to_readonly = true;
461 	map_pages(init_end, __pa(init_end), kernel_end - init_end,
462 		  PAGE_KERNEL, 0);
463 
464 	/* The init text pages are marked R-X.  We have to
465 	 * flush the icache and mark them RW-
466 	 *
467 	 * Do a dummy remap of the data section first (the data
468 	 * section is already PAGE_KERNEL) to pull in the TLB entries
469 	 * for map_kernel */
470 	map_pages(init_begin, __pa(init_begin), init_end - init_begin,
471 		  PAGE_KERNEL_RWX, 1);
472 	/* now remap at PAGE_KERNEL since the TLB is pre-primed to execute
473 	 * map_pages */
474 	map_pages(init_begin, __pa(init_begin), init_end - init_begin,
475 		  PAGE_KERNEL, 1);
476 
477 	/* force the kernel to see the new TLB entries */
478 	__flush_tlb_range(0, init_begin, kernel_end);
479 
480 	/* finally dump all the instructions which were cached, since the
481 	 * pages are no-longer executable */
482 	flush_icache_range(init_begin, init_end);
483 
484 	free_initmem_default(POISON_FREE_INITMEM);
485 
486 	/* set up a new led state on systems shipped LED State panel */
487 	pdc_chassis_send_status(PDC_CHASSIS_DIRECT_BCOMPLETE);
488 }
489 
490 
491 #ifdef CONFIG_STRICT_KERNEL_RWX
mark_rodata_ro(void)492 void mark_rodata_ro(void)
493 {
494 	/* rodata memory was already mapped with KERNEL_RO access rights by
495            pagetable_init() and map_pages(). No need to do additional stuff here */
496 	unsigned long roai_size = __end_ro_after_init - __start_ro_after_init;
497 
498 	pr_info("Write protected read-only-after-init data: %luk\n", roai_size >> 10);
499 }
500 #endif
501 
502 
503 /*
504  * Just an arbitrary offset to serve as a "hole" between mapping areas
505  * (between top of physical memory and a potential pcxl dma mapping
506  * area, and below the vmalloc mapping area).
507  *
508  * The current 32K value just means that there will be a 32K "hole"
509  * between mapping areas. That means that  any out-of-bounds memory
510  * accesses will hopefully be caught. The vmalloc() routines leaves
511  * a hole of 4kB between each vmalloced area for the same reason.
512  */
513 
514  /* Leave room for gateway page expansion */
515 #if KERNEL_MAP_START < GATEWAY_PAGE_SIZE
516 #error KERNEL_MAP_START is in gateway reserved region
517 #endif
518 #define MAP_START (KERNEL_MAP_START)
519 
520 #define VM_MAP_OFFSET  (32*1024)
521 #define SET_MAP_OFFSET(x) ((void *)(((unsigned long)(x) + VM_MAP_OFFSET) \
522 				     & ~(VM_MAP_OFFSET-1)))
523 
524 void *parisc_vmalloc_start __ro_after_init;
525 EXPORT_SYMBOL(parisc_vmalloc_start);
526 
mem_init(void)527 void __init mem_init(void)
528 {
529 	/* Do sanity checks on IPC (compat) structures */
530 	BUILD_BUG_ON(sizeof(struct ipc64_perm) != 48);
531 #ifndef CONFIG_64BIT
532 	BUILD_BUG_ON(sizeof(struct semid64_ds) != 80);
533 	BUILD_BUG_ON(sizeof(struct msqid64_ds) != 104);
534 	BUILD_BUG_ON(sizeof(struct shmid64_ds) != 104);
535 #endif
536 #ifdef CONFIG_COMPAT
537 	BUILD_BUG_ON(sizeof(struct compat_ipc64_perm) != sizeof(struct ipc64_perm));
538 	BUILD_BUG_ON(sizeof(struct compat_semid64_ds) != 80);
539 	BUILD_BUG_ON(sizeof(struct compat_msqid64_ds) != 104);
540 	BUILD_BUG_ON(sizeof(struct compat_shmid64_ds) != 104);
541 #endif
542 
543 	/* Do sanity checks on page table constants */
544 	BUILD_BUG_ON(PTE_ENTRY_SIZE != sizeof(pte_t));
545 	BUILD_BUG_ON(PMD_ENTRY_SIZE != sizeof(pmd_t));
546 	BUILD_BUG_ON(PGD_ENTRY_SIZE != sizeof(pgd_t));
547 	BUILD_BUG_ON(PAGE_SHIFT + BITS_PER_PTE + BITS_PER_PMD + BITS_PER_PGD
548 			> BITS_PER_LONG);
549 #if CONFIG_PGTABLE_LEVELS == 3
550 	BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PMD);
551 #else
552 	BUILD_BUG_ON(PT_INITIAL > PTRS_PER_PGD);
553 #endif
554 
555 #ifdef CONFIG_64BIT
556 	/* avoid ldil_%L() asm statements to sign-extend into upper 32-bits */
557 	BUILD_BUG_ON(__PAGE_OFFSET >= 0x80000000);
558 	BUILD_BUG_ON(TMPALIAS_MAP_START >= 0x80000000);
559 #endif
560 
561 	high_memory = __va((max_pfn << PAGE_SHIFT));
562 	set_max_mapnr(max_low_pfn);
563 	memblock_free_all();
564 
565 #ifdef CONFIG_PA11
566 	if (boot_cpu_data.cpu_type == pcxl2 || boot_cpu_data.cpu_type == pcxl) {
567 		pcxl_dma_start = (unsigned long)SET_MAP_OFFSET(MAP_START);
568 		parisc_vmalloc_start = SET_MAP_OFFSET(pcxl_dma_start
569 						+ PCXL_DMA_MAP_SIZE);
570 	} else
571 #endif
572 		parisc_vmalloc_start = SET_MAP_OFFSET(MAP_START);
573 
574 #if 0
575 	/*
576 	 * Do not expose the virtual kernel memory layout to userspace.
577 	 * But keep code for debugging purposes.
578 	 */
579 	printk("virtual kernel memory layout:\n"
580 	       "     vmalloc : 0x%px - 0x%px   (%4ld MB)\n"
581 	       "     fixmap  : 0x%px - 0x%px   (%4ld kB)\n"
582 	       "     memory  : 0x%px - 0x%px   (%4ld MB)\n"
583 	       "       .init : 0x%px - 0x%px   (%4ld kB)\n"
584 	       "       .data : 0x%px - 0x%px   (%4ld kB)\n"
585 	       "       .text : 0x%px - 0x%px   (%4ld kB)\n",
586 
587 	       (void*)VMALLOC_START, (void*)VMALLOC_END,
588 	       (VMALLOC_END - VMALLOC_START) >> 20,
589 
590 	       (void *)FIXMAP_START, (void *)(FIXMAP_START + FIXMAP_SIZE),
591 	       (unsigned long)(FIXMAP_SIZE / 1024),
592 
593 	       __va(0), high_memory,
594 	       ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
595 
596 	       __init_begin, __init_end,
597 	       ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10,
598 
599 	       _etext, _edata,
600 	       ((unsigned long)_edata - (unsigned long)_etext) >> 10,
601 
602 	       _text, _etext,
603 	       ((unsigned long)_etext - (unsigned long)_text) >> 10);
604 #endif
605 }
606 
607 unsigned long *empty_zero_page __ro_after_init;
608 EXPORT_SYMBOL(empty_zero_page);
609 
610 /*
611  * pagetable_init() sets up the page tables
612  *
613  * Note that gateway_init() places the Linux gateway page at page 0.
614  * Since gateway pages cannot be dereferenced this has the desirable
615  * side effect of trapping those pesky NULL-reference errors in the
616  * kernel.
617  */
pagetable_init(void)618 static void __init pagetable_init(void)
619 {
620 	int range;
621 
622 	/* Map each physical memory range to its kernel vaddr */
623 
624 	for (range = 0; range < npmem_ranges; range++) {
625 		unsigned long start_paddr;
626 		unsigned long size;
627 
628 		start_paddr = pmem_ranges[range].start_pfn << PAGE_SHIFT;
629 		size = pmem_ranges[range].pages << PAGE_SHIFT;
630 
631 		map_pages((unsigned long)__va(start_paddr), start_paddr,
632 			  size, PAGE_KERNEL, 0);
633 	}
634 
635 #ifdef CONFIG_BLK_DEV_INITRD
636 	if (initrd_end && initrd_end > mem_limit) {
637 		printk(KERN_INFO "initrd: mapping %08lx-%08lx\n", initrd_start, initrd_end);
638 		map_pages(initrd_start, __pa(initrd_start),
639 			  initrd_end - initrd_start, PAGE_KERNEL, 0);
640 	}
641 #endif
642 
643 	empty_zero_page = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
644 	if (!empty_zero_page)
645 		panic("zero page allocation failed.\n");
646 
647 }
648 
gateway_init(void)649 static void __init gateway_init(void)
650 {
651 	unsigned long linux_gateway_page_addr;
652 	/* FIXME: This is 'const' in order to trick the compiler
653 	   into not treating it as DP-relative data. */
654 	extern void * const linux_gateway_page;
655 
656 	linux_gateway_page_addr = LINUX_GATEWAY_ADDR & PAGE_MASK;
657 
658 	/*
659 	 * Setup Linux Gateway page.
660 	 *
661 	 * The Linux gateway page will reside in kernel space (on virtual
662 	 * page 0), so it doesn't need to be aliased into user space.
663 	 */
664 
665 	map_pages(linux_gateway_page_addr, __pa(&linux_gateway_page),
666 		  PAGE_SIZE, PAGE_GATEWAY, 1);
667 }
668 
fixmap_init(void)669 static void __init fixmap_init(void)
670 {
671 	unsigned long addr = FIXMAP_START;
672 	unsigned long end = FIXMAP_START + FIXMAP_SIZE;
673 	pgd_t *pgd = pgd_offset_k(addr);
674 	p4d_t *p4d = p4d_offset(pgd, addr);
675 	pud_t *pud = pud_offset(p4d, addr);
676 	pmd_t *pmd;
677 
678 	BUILD_BUG_ON(FIXMAP_SIZE > PMD_SIZE);
679 
680 #if CONFIG_PGTABLE_LEVELS == 3
681 	if (pud_none(*pud)) {
682 		pmd = memblock_alloc(PAGE_SIZE << PMD_TABLE_ORDER,
683 				     PAGE_SIZE << PMD_TABLE_ORDER);
684 		if (!pmd)
685 			panic("fixmap: pmd allocation failed.\n");
686 		pud_populate(NULL, pud, pmd);
687 	}
688 #endif
689 
690 	pmd = pmd_offset(pud, addr);
691 	do {
692 		pte_t *pte = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
693 		if (!pte)
694 			panic("fixmap: pte allocation failed.\n");
695 
696 		pmd_populate_kernel(&init_mm, pmd, pte);
697 
698 		addr += PAGE_SIZE;
699 	} while (addr < end);
700 }
701 
parisc_bootmem_free(void)702 static void __init parisc_bootmem_free(void)
703 {
704 	unsigned long max_zone_pfn[MAX_NR_ZONES] = { 0, };
705 
706 	max_zone_pfn[0] = memblock_end_of_DRAM();
707 
708 	free_area_init(max_zone_pfn);
709 }
710 
paging_init(void)711 void __init paging_init(void)
712 {
713 	setup_bootmem();
714 	pagetable_init();
715 	gateway_init();
716 	fixmap_init();
717 	flush_cache_all_local(); /* start with known state */
718 	flush_tlb_all_local(NULL);
719 
720 	sparse_init();
721 	parisc_bootmem_free();
722 }
723 
alloc_btlb(unsigned long start,unsigned long end,int * slot,unsigned long entry_info)724 static void alloc_btlb(unsigned long start, unsigned long end, int *slot,
725 			unsigned long entry_info)
726 {
727 	const int slot_max = btlb_info.fixed_range_info.num_comb;
728 	int min_num_pages = btlb_info.min_size;
729 	unsigned long size;
730 
731 	/* map at minimum 4 pages */
732 	if (min_num_pages < 4)
733 		min_num_pages = 4;
734 
735 	size = HUGEPAGE_SIZE;
736 	while (start < end && *slot < slot_max && size >= PAGE_SIZE) {
737 		/* starting address must have same alignment as size! */
738 		/* if correctly aligned and fits in double size, increase */
739 		if (((start & (2 * size - 1)) == 0) &&
740 		    (end - start) >= (2 * size)) {
741 			size <<= 1;
742 			continue;
743 		}
744 		/* if current size alignment is too big, try smaller size */
745 		if ((start & (size - 1)) != 0) {
746 			size >>= 1;
747 			continue;
748 		}
749 		if ((end - start) >= size) {
750 			if ((size >> PAGE_SHIFT) >= min_num_pages)
751 				pdc_btlb_insert(start >> PAGE_SHIFT, __pa(start) >> PAGE_SHIFT,
752 					size >> PAGE_SHIFT, entry_info, *slot);
753 			(*slot)++;
754 			start += size;
755 			continue;
756 		}
757 		size /= 2;
758 		continue;
759 	}
760 }
761 
btlb_init_per_cpu(void)762 void btlb_init_per_cpu(void)
763 {
764 	unsigned long s, t, e;
765 	int slot;
766 
767 	/* BTLBs are not available on 64-bit CPUs */
768 	if (IS_ENABLED(CONFIG_PA20))
769 		return;
770 	else if (pdc_btlb_info(&btlb_info) < 0) {
771 		memset(&btlb_info, 0, sizeof btlb_info);
772 	}
773 
774 	/* insert BLTLBs for code and data segments */
775 	s = (uintptr_t) dereference_function_descriptor(&_stext);
776 	e = (uintptr_t) dereference_function_descriptor(&_etext);
777 	t = (uintptr_t) dereference_function_descriptor(&_sdata);
778 	BUG_ON(t != e);
779 
780 	/* code segments */
781 	slot = 0;
782 	alloc_btlb(s, e, &slot, 0x13800000);
783 
784 	/* sanity check */
785 	t = (uintptr_t) dereference_function_descriptor(&_edata);
786 	e = (uintptr_t) dereference_function_descriptor(&__bss_start);
787 	BUG_ON(t != e);
788 
789 	/* data segments */
790 	s = (uintptr_t) dereference_function_descriptor(&_sdata);
791 	e = (uintptr_t) dereference_function_descriptor(&__bss_stop);
792 	alloc_btlb(s, e, &slot, 0x11800000);
793 }
794 
795 #ifdef CONFIG_PA20
796 
797 /*
798  * Currently, all PA20 chips have 18 bit protection IDs, which is the
799  * limiting factor (space ids are 32 bits).
800  */
801 
802 #define NR_SPACE_IDS 262144
803 
804 #else
805 
806 /*
807  * Currently we have a one-to-one relationship between space IDs and
808  * protection IDs. Older parisc chips (PCXS, PCXT, PCXL, PCXL2) only
809  * support 15 bit protection IDs, so that is the limiting factor.
810  * PCXT' has 18 bit protection IDs, but only 16 bit spaceids, so it's
811  * probably not worth the effort for a special case here.
812  */
813 
814 #define NR_SPACE_IDS 32768
815 
816 #endif  /* !CONFIG_PA20 */
817 
818 #define RECYCLE_THRESHOLD (NR_SPACE_IDS / 2)
819 #define SID_ARRAY_SIZE  (NR_SPACE_IDS / (8 * sizeof(long)))
820 
821 static unsigned long space_id[SID_ARRAY_SIZE] = { 1 }; /* disallow space 0 */
822 static unsigned long dirty_space_id[SID_ARRAY_SIZE];
823 static unsigned long space_id_index;
824 static unsigned long free_space_ids = NR_SPACE_IDS - 1;
825 static unsigned long dirty_space_ids;
826 
827 static DEFINE_SPINLOCK(sid_lock);
828 
alloc_sid(void)829 unsigned long alloc_sid(void)
830 {
831 	unsigned long index;
832 
833 	spin_lock(&sid_lock);
834 
835 	if (free_space_ids == 0) {
836 		if (dirty_space_ids != 0) {
837 			spin_unlock(&sid_lock);
838 			flush_tlb_all(); /* flush_tlb_all() calls recycle_sids() */
839 			spin_lock(&sid_lock);
840 		}
841 		BUG_ON(free_space_ids == 0);
842 	}
843 
844 	free_space_ids--;
845 
846 	index = find_next_zero_bit(space_id, NR_SPACE_IDS, space_id_index);
847 	space_id[BIT_WORD(index)] |= BIT_MASK(index);
848 	space_id_index = index;
849 
850 	spin_unlock(&sid_lock);
851 
852 	return index << SPACEID_SHIFT;
853 }
854 
free_sid(unsigned long spaceid)855 void free_sid(unsigned long spaceid)
856 {
857 	unsigned long index = spaceid >> SPACEID_SHIFT;
858 	unsigned long *dirty_space_offset, mask;
859 
860 	dirty_space_offset = &dirty_space_id[BIT_WORD(index)];
861 	mask = BIT_MASK(index);
862 
863 	spin_lock(&sid_lock);
864 
865 	BUG_ON(*dirty_space_offset & mask); /* attempt to free space id twice */
866 
867 	*dirty_space_offset |= mask;
868 	dirty_space_ids++;
869 
870 	spin_unlock(&sid_lock);
871 }
872 
873 
874 #ifdef CONFIG_SMP
get_dirty_sids(unsigned long * ndirtyptr,unsigned long * dirty_array)875 static void get_dirty_sids(unsigned long *ndirtyptr,unsigned long *dirty_array)
876 {
877 	int i;
878 
879 	/* NOTE: sid_lock must be held upon entry */
880 
881 	*ndirtyptr = dirty_space_ids;
882 	if (dirty_space_ids != 0) {
883 	    for (i = 0; i < SID_ARRAY_SIZE; i++) {
884 		dirty_array[i] = dirty_space_id[i];
885 		dirty_space_id[i] = 0;
886 	    }
887 	    dirty_space_ids = 0;
888 	}
889 
890 	return;
891 }
892 
recycle_sids(unsigned long ndirty,unsigned long * dirty_array)893 static void recycle_sids(unsigned long ndirty,unsigned long *dirty_array)
894 {
895 	int i;
896 
897 	/* NOTE: sid_lock must be held upon entry */
898 
899 	if (ndirty != 0) {
900 		for (i = 0; i < SID_ARRAY_SIZE; i++) {
901 			space_id[i] ^= dirty_array[i];
902 		}
903 
904 		free_space_ids += ndirty;
905 		space_id_index = 0;
906 	}
907 }
908 
909 #else /* CONFIG_SMP */
910 
recycle_sids(void)911 static void recycle_sids(void)
912 {
913 	int i;
914 
915 	/* NOTE: sid_lock must be held upon entry */
916 
917 	if (dirty_space_ids != 0) {
918 		for (i = 0; i < SID_ARRAY_SIZE; i++) {
919 			space_id[i] ^= dirty_space_id[i];
920 			dirty_space_id[i] = 0;
921 		}
922 
923 		free_space_ids += dirty_space_ids;
924 		dirty_space_ids = 0;
925 		space_id_index = 0;
926 	}
927 }
928 #endif
929 
930 /*
931  * flush_tlb_all() calls recycle_sids(), since whenever the entire tlb is
932  * purged, we can safely reuse the space ids that were released but
933  * not flushed from the tlb.
934  */
935 
936 #ifdef CONFIG_SMP
937 
938 static unsigned long recycle_ndirty;
939 static unsigned long recycle_dirty_array[SID_ARRAY_SIZE];
940 static unsigned int recycle_inuse;
941 
flush_tlb_all(void)942 void flush_tlb_all(void)
943 {
944 	int do_recycle;
945 
946 	do_recycle = 0;
947 	spin_lock(&sid_lock);
948 	__inc_irq_stat(irq_tlb_count);
949 	if (dirty_space_ids > RECYCLE_THRESHOLD) {
950 	    BUG_ON(recycle_inuse);  /* FIXME: Use a semaphore/wait queue here */
951 	    get_dirty_sids(&recycle_ndirty,recycle_dirty_array);
952 	    recycle_inuse++;
953 	    do_recycle++;
954 	}
955 	spin_unlock(&sid_lock);
956 	on_each_cpu(flush_tlb_all_local, NULL, 1);
957 	if (do_recycle) {
958 	    spin_lock(&sid_lock);
959 	    recycle_sids(recycle_ndirty,recycle_dirty_array);
960 	    recycle_inuse = 0;
961 	    spin_unlock(&sid_lock);
962 	}
963 }
964 #else
flush_tlb_all(void)965 void flush_tlb_all(void)
966 {
967 	spin_lock(&sid_lock);
968 	__inc_irq_stat(irq_tlb_count);
969 	flush_tlb_all_local(NULL);
970 	recycle_sids();
971 	spin_unlock(&sid_lock);
972 }
973 #endif
974 
975 static const pgprot_t protection_map[16] = {
976 	[VM_NONE]					= PAGE_NONE,
977 	[VM_READ]					= PAGE_READONLY,
978 	[VM_WRITE]					= PAGE_NONE,
979 	[VM_WRITE | VM_READ]				= PAGE_READONLY,
980 	[VM_EXEC]					= PAGE_EXECREAD,
981 	[VM_EXEC | VM_READ]				= PAGE_EXECREAD,
982 	[VM_EXEC | VM_WRITE]				= PAGE_EXECREAD,
983 	[VM_EXEC | VM_WRITE | VM_READ]			= PAGE_EXECREAD,
984 	[VM_SHARED]					= PAGE_NONE,
985 	[VM_SHARED | VM_READ]				= PAGE_READONLY,
986 	[VM_SHARED | VM_WRITE]				= PAGE_WRITEONLY,
987 	[VM_SHARED | VM_WRITE | VM_READ]		= PAGE_SHARED,
988 	[VM_SHARED | VM_EXEC]				= PAGE_EXECREAD,
989 	[VM_SHARED | VM_EXEC | VM_READ]			= PAGE_EXECREAD,
990 	[VM_SHARED | VM_EXEC | VM_WRITE]		= PAGE_RWX,
991 	[VM_SHARED | VM_EXEC | VM_WRITE | VM_READ]	= PAGE_RWX
992 };
993 DECLARE_VM_GET_PAGE_PROT
994