xref: /openbmc/linux/arch/riscv/mm/init.c (revision dfc53baa)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2012 Regents of the University of California
4  * Copyright (C) 2019 Western Digital Corporation or its affiliates.
5  */
6 
7 #include <linux/init.h>
8 #include <linux/mm.h>
9 #include <linux/memblock.h>
10 #include <linux/initrd.h>
11 #include <linux/swap.h>
12 #include <linux/sizes.h>
13 #include <linux/of_fdt.h>
14 #include <linux/libfdt.h>
15 #include <linux/set_memory.h>
16 
17 #include <asm/fixmap.h>
18 #include <asm/tlbflush.h>
19 #include <asm/sections.h>
20 #include <asm/soc.h>
21 #include <asm/io.h>
22 #include <asm/ptdump.h>
23 
24 #include "../kernel/head.h"
25 
26 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
27 							__page_aligned_bss;
28 EXPORT_SYMBOL(empty_zero_page);
29 
30 extern char _start[];
31 void *dtb_early_va;
32 
33 static void __init zone_sizes_init(void)
34 {
35 	unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
36 
37 #ifdef CONFIG_ZONE_DMA32
38 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(min(4UL * SZ_1G,
39 			(unsigned long) PFN_PHYS(max_low_pfn)));
40 #endif
41 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
42 
43 	free_area_init(max_zone_pfns);
44 }
45 
46 static void setup_zero_page(void)
47 {
48 	memset((void *)empty_zero_page, 0, PAGE_SIZE);
49 }
50 
51 #if defined(CONFIG_MMU) && defined(CONFIG_DEBUG_VM)
52 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
53 {
54 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
55 		  (((t) - (b)) >> 10));
56 }
57 
58 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
59 {
60 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
61 		  (((t) - (b)) >> 20));
62 }
63 
64 static void print_vm_layout(void)
65 {
66 	pr_notice("Virtual kernel memory layout:\n");
67 	print_mlk("fixmap", (unsigned long)FIXADDR_START,
68 		  (unsigned long)FIXADDR_TOP);
69 	print_mlm("pci io", (unsigned long)PCI_IO_START,
70 		  (unsigned long)PCI_IO_END);
71 	print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
72 		  (unsigned long)VMEMMAP_END);
73 	print_mlm("vmalloc", (unsigned long)VMALLOC_START,
74 		  (unsigned long)VMALLOC_END);
75 	print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
76 		  (unsigned long)high_memory);
77 }
78 #else
79 static void print_vm_layout(void) { }
80 #endif /* CONFIG_DEBUG_VM */
81 
82 void __init mem_init(void)
83 {
84 #ifdef CONFIG_FLATMEM
85 	BUG_ON(!mem_map);
86 #endif /* CONFIG_FLATMEM */
87 
88 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
89 	memblock_free_all();
90 
91 	mem_init_print_info(NULL);
92 	print_vm_layout();
93 }
94 
95 #ifdef CONFIG_BLK_DEV_INITRD
96 static void __init setup_initrd(void)
97 {
98 	phys_addr_t start;
99 	unsigned long size;
100 
101 	/* Ignore the virtul address computed during device tree parsing */
102 	initrd_start = initrd_end = 0;
103 
104 	if (!phys_initrd_size)
105 		return;
106 	/*
107 	 * Round the memory region to page boundaries as per free_initrd_mem()
108 	 * This allows us to detect whether the pages overlapping the initrd
109 	 * are in use, but more importantly, reserves the entire set of pages
110 	 * as we don't want these pages allocated for other purposes.
111 	 */
112 	start = round_down(phys_initrd_start, PAGE_SIZE);
113 	size = phys_initrd_size + (phys_initrd_start - start);
114 	size = round_up(size, PAGE_SIZE);
115 
116 	if (!memblock_is_region_memory(start, size)) {
117 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
118 		       (u64)start, size);
119 		goto disable;
120 	}
121 
122 	if (memblock_is_region_reserved(start, size)) {
123 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
124 		       (u64)start, size);
125 		goto disable;
126 	}
127 
128 	memblock_reserve(start, size);
129 	/* Now convert initrd to virtual addresses */
130 	initrd_start = (unsigned long)__va(phys_initrd_start);
131 	initrd_end = initrd_start + phys_initrd_size;
132 	initrd_below_start_ok = 1;
133 
134 	pr_info("Initial ramdisk at: 0x%p (%lu bytes)\n",
135 		(void *)(initrd_start), size);
136 	return;
137 disable:
138 	pr_cont(" - disabling initrd\n");
139 	initrd_start = 0;
140 	initrd_end = 0;
141 }
142 #endif /* CONFIG_BLK_DEV_INITRD */
143 
144 static phys_addr_t dtb_early_pa __initdata;
145 
146 void __init setup_bootmem(void)
147 {
148 	struct memblock_region *reg;
149 	phys_addr_t mem_size = 0;
150 	phys_addr_t total_mem = 0;
151 	phys_addr_t mem_start, end = 0;
152 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
153 	phys_addr_t vmlinux_start = __pa_symbol(&_start);
154 
155 	/* Find the memory region containing the kernel */
156 	for_each_memblock(memory, reg) {
157 		end = reg->base + reg->size;
158 		if (!total_mem)
159 			mem_start = reg->base;
160 		if (reg->base <= vmlinux_start && vmlinux_end <= end)
161 			BUG_ON(reg->size == 0);
162 		total_mem = total_mem + reg->size;
163 	}
164 
165 	/*
166 	 * Remove memblock from the end of usable area to the
167 	 * end of region
168 	 */
169 	mem_size = min(total_mem, (phys_addr_t)-PAGE_OFFSET);
170 	if (mem_start + mem_size < end)
171 		memblock_remove(mem_start + mem_size,
172 				end - mem_start - mem_size);
173 
174 	/* Reserve from the start of the kernel to the end of the kernel */
175 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
176 
177 	max_pfn = PFN_DOWN(memblock_end_of_DRAM());
178 	max_low_pfn = max_pfn;
179 	set_max_mapnr(max_low_pfn);
180 
181 #ifdef CONFIG_BLK_DEV_INITRD
182 	setup_initrd();
183 #endif /* CONFIG_BLK_DEV_INITRD */
184 
185 	/*
186 	 * Avoid using early_init_fdt_reserve_self() since __pa() does
187 	 * not work for DTB pointers that are fixmap addresses
188 	 */
189 	memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
190 
191 	early_init_fdt_scan_reserved_mem();
192 	memblock_allow_resize();
193 	memblock_dump_all();
194 
195 	for_each_memblock(memory, reg) {
196 		unsigned long start_pfn = memblock_region_memory_base_pfn(reg);
197 		unsigned long end_pfn = memblock_region_memory_end_pfn(reg);
198 
199 		memblock_set_node(PFN_PHYS(start_pfn),
200 				  PFN_PHYS(end_pfn - start_pfn),
201 				  &memblock.memory, 0);
202 	}
203 }
204 
205 #ifdef CONFIG_MMU
206 unsigned long va_pa_offset;
207 EXPORT_SYMBOL(va_pa_offset);
208 unsigned long pfn_base;
209 EXPORT_SYMBOL(pfn_base);
210 
211 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
212 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
213 pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
214 static bool mmu_enabled;
215 
216 #define MAX_EARLY_MAPPING_SIZE	SZ_128M
217 
218 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
219 
220 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
221 {
222 	unsigned long addr = __fix_to_virt(idx);
223 	pte_t *ptep;
224 
225 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
226 
227 	ptep = &fixmap_pte[pte_index(addr)];
228 
229 	if (pgprot_val(prot))
230 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
231 	else
232 		pte_clear(&init_mm, addr, ptep);
233 	local_flush_tlb_page(addr);
234 }
235 
236 static pte_t *__init get_pte_virt(phys_addr_t pa)
237 {
238 	if (mmu_enabled) {
239 		clear_fixmap(FIX_PTE);
240 		return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
241 	} else {
242 		return (pte_t *)((uintptr_t)pa);
243 	}
244 }
245 
246 static phys_addr_t __init alloc_pte(uintptr_t va)
247 {
248 	/*
249 	 * We only create PMD or PGD early mappings so we
250 	 * should never reach here with MMU disabled.
251 	 */
252 	BUG_ON(!mmu_enabled);
253 
254 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
255 }
256 
257 static void __init create_pte_mapping(pte_t *ptep,
258 				      uintptr_t va, phys_addr_t pa,
259 				      phys_addr_t sz, pgprot_t prot)
260 {
261 	uintptr_t pte_idx = pte_index(va);
262 
263 	BUG_ON(sz != PAGE_SIZE);
264 
265 	if (pte_none(ptep[pte_idx]))
266 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
267 }
268 
269 #ifndef __PAGETABLE_PMD_FOLDED
270 
271 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
272 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
273 
274 #if MAX_EARLY_MAPPING_SIZE < PGDIR_SIZE
275 #define NUM_EARLY_PMDS		1UL
276 #else
277 #define NUM_EARLY_PMDS		(1UL + MAX_EARLY_MAPPING_SIZE / PGDIR_SIZE)
278 #endif
279 pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
280 
281 static pmd_t *__init get_pmd_virt(phys_addr_t pa)
282 {
283 	if (mmu_enabled) {
284 		clear_fixmap(FIX_PMD);
285 		return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
286 	} else {
287 		return (pmd_t *)((uintptr_t)pa);
288 	}
289 }
290 
291 static phys_addr_t __init alloc_pmd(uintptr_t va)
292 {
293 	uintptr_t pmd_num;
294 
295 	if (mmu_enabled)
296 		return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
297 
298 	pmd_num = (va - PAGE_OFFSET) >> PGDIR_SHIFT;
299 	BUG_ON(pmd_num >= NUM_EARLY_PMDS);
300 	return (uintptr_t)&early_pmd[pmd_num * PTRS_PER_PMD];
301 }
302 
303 static void __init create_pmd_mapping(pmd_t *pmdp,
304 				      uintptr_t va, phys_addr_t pa,
305 				      phys_addr_t sz, pgprot_t prot)
306 {
307 	pte_t *ptep;
308 	phys_addr_t pte_phys;
309 	uintptr_t pmd_idx = pmd_index(va);
310 
311 	if (sz == PMD_SIZE) {
312 		if (pmd_none(pmdp[pmd_idx]))
313 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
314 		return;
315 	}
316 
317 	if (pmd_none(pmdp[pmd_idx])) {
318 		pte_phys = alloc_pte(va);
319 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
320 		ptep = get_pte_virt(pte_phys);
321 		memset(ptep, 0, PAGE_SIZE);
322 	} else {
323 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
324 		ptep = get_pte_virt(pte_phys);
325 	}
326 
327 	create_pte_mapping(ptep, va, pa, sz, prot);
328 }
329 
330 #define pgd_next_t		pmd_t
331 #define alloc_pgd_next(__va)	alloc_pmd(__va)
332 #define get_pgd_next_virt(__pa)	get_pmd_virt(__pa)
333 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
334 	create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
335 #define fixmap_pgd_next		fixmap_pmd
336 #else
337 #define pgd_next_t		pte_t
338 #define alloc_pgd_next(__va)	alloc_pte(__va)
339 #define get_pgd_next_virt(__pa)	get_pte_virt(__pa)
340 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
341 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
342 #define fixmap_pgd_next		fixmap_pte
343 #endif
344 
345 static void __init create_pgd_mapping(pgd_t *pgdp,
346 				      uintptr_t va, phys_addr_t pa,
347 				      phys_addr_t sz, pgprot_t prot)
348 {
349 	pgd_next_t *nextp;
350 	phys_addr_t next_phys;
351 	uintptr_t pgd_idx = pgd_index(va);
352 
353 	if (sz == PGDIR_SIZE) {
354 		if (pgd_val(pgdp[pgd_idx]) == 0)
355 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
356 		return;
357 	}
358 
359 	if (pgd_val(pgdp[pgd_idx]) == 0) {
360 		next_phys = alloc_pgd_next(va);
361 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
362 		nextp = get_pgd_next_virt(next_phys);
363 		memset(nextp, 0, PAGE_SIZE);
364 	} else {
365 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
366 		nextp = get_pgd_next_virt(next_phys);
367 	}
368 
369 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
370 }
371 
372 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
373 {
374 	/* Upgrade to PMD_SIZE mappings whenever possible */
375 	if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
376 		return PAGE_SIZE;
377 
378 	return PMD_SIZE;
379 }
380 
381 /*
382  * setup_vm() is called from head.S with MMU-off.
383  *
384  * Following requirements should be honoured for setup_vm() to work
385  * correctly:
386  * 1) It should use PC-relative addressing for accessing kernel symbols.
387  *    To achieve this we always use GCC cmodel=medany.
388  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
389  *    so disable compiler instrumentation when FTRACE is enabled.
390  *
391  * Currently, the above requirements are honoured by using custom CFLAGS
392  * for init.o in mm/Makefile.
393  */
394 
395 #ifndef __riscv_cmodel_medany
396 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
397 #endif
398 
399 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
400 {
401 	uintptr_t va, end_va;
402 	uintptr_t load_pa = (uintptr_t)(&_start);
403 	uintptr_t load_sz = (uintptr_t)(&_end) - load_pa;
404 	uintptr_t map_size = best_map_size(load_pa, MAX_EARLY_MAPPING_SIZE);
405 
406 	va_pa_offset = PAGE_OFFSET - load_pa;
407 	pfn_base = PFN_DOWN(load_pa);
408 
409 	/*
410 	 * Enforce boot alignment requirements of RV32 and
411 	 * RV64 by only allowing PMD or PGD mappings.
412 	 */
413 	BUG_ON(map_size == PAGE_SIZE);
414 
415 	/* Sanity check alignment and size */
416 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
417 	BUG_ON((load_pa % map_size) != 0);
418 	BUG_ON(load_sz > MAX_EARLY_MAPPING_SIZE);
419 
420 	/* Setup early PGD for fixmap */
421 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
422 			   (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
423 
424 #ifndef __PAGETABLE_PMD_FOLDED
425 	/* Setup fixmap PMD */
426 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
427 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
428 	/* Setup trampoline PGD and PMD */
429 	create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
430 			   (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
431 	create_pmd_mapping(trampoline_pmd, PAGE_OFFSET,
432 			   load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
433 #else
434 	/* Setup trampoline PGD */
435 	create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
436 			   load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC);
437 #endif
438 
439 	/*
440 	 * Setup early PGD covering entire kernel which will allows
441 	 * us to reach paging_init(). We map all memory banks later
442 	 * in setup_vm_final() below.
443 	 */
444 	end_va = PAGE_OFFSET + load_sz;
445 	for (va = PAGE_OFFSET; va < end_va; va += map_size)
446 		create_pgd_mapping(early_pg_dir, va,
447 				   load_pa + (va - PAGE_OFFSET),
448 				   map_size, PAGE_KERNEL_EXEC);
449 
450 	/* Create fixed mapping for early FDT parsing */
451 	end_va = __fix_to_virt(FIX_FDT) + FIX_FDT_SIZE;
452 	for (va = __fix_to_virt(FIX_FDT); va < end_va; va += PAGE_SIZE)
453 		create_pte_mapping(fixmap_pte, va,
454 				   dtb_pa + (va - __fix_to_virt(FIX_FDT)),
455 				   PAGE_SIZE, PAGE_KERNEL);
456 
457 	/* Save pointer to DTB for early FDT parsing */
458 	dtb_early_va = (void *)fix_to_virt(FIX_FDT) + (dtb_pa & ~PAGE_MASK);
459 	/* Save physical address for memblock reservation */
460 	dtb_early_pa = dtb_pa;
461 }
462 
463 static void __init setup_vm_final(void)
464 {
465 	uintptr_t va, map_size;
466 	phys_addr_t pa, start, end;
467 	struct memblock_region *reg;
468 
469 	/* Set mmu_enabled flag */
470 	mmu_enabled = true;
471 
472 	/* Setup swapper PGD for fixmap */
473 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
474 			   __pa_symbol(fixmap_pgd_next),
475 			   PGDIR_SIZE, PAGE_TABLE);
476 
477 	/* Map all memory banks */
478 	for_each_memblock(memory, reg) {
479 		start = reg->base;
480 		end = start + reg->size;
481 
482 		if (start >= end)
483 			break;
484 		if (memblock_is_nomap(reg))
485 			continue;
486 		if (start <= __pa(PAGE_OFFSET) &&
487 		    __pa(PAGE_OFFSET) < end)
488 			start = __pa(PAGE_OFFSET);
489 
490 		map_size = best_map_size(start, end - start);
491 		for (pa = start; pa < end; pa += map_size) {
492 			va = (uintptr_t)__va(pa);
493 			create_pgd_mapping(swapper_pg_dir, va, pa,
494 					   map_size, PAGE_KERNEL_EXEC);
495 		}
496 	}
497 
498 	/* Clear fixmap PTE and PMD mappings */
499 	clear_fixmap(FIX_PTE);
500 	clear_fixmap(FIX_PMD);
501 
502 	/* Move to swapper page table */
503 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
504 	local_flush_tlb_all();
505 }
506 #else
507 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
508 {
509 #ifdef CONFIG_BUILTIN_DTB
510 	dtb_early_va = soc_lookup_builtin_dtb();
511 	if (!dtb_early_va) {
512 		/* Fallback to first available DTS */
513 		dtb_early_va = (void *) __dtb_start;
514 	}
515 #else
516 	dtb_early_va = (void *)dtb_pa;
517 #endif
518 }
519 
520 static inline void setup_vm_final(void)
521 {
522 }
523 #endif /* CONFIG_MMU */
524 
525 #ifdef CONFIG_STRICT_KERNEL_RWX
526 void mark_rodata_ro(void)
527 {
528 	unsigned long text_start = (unsigned long)_text;
529 	unsigned long text_end = (unsigned long)_etext;
530 	unsigned long rodata_start = (unsigned long)__start_rodata;
531 	unsigned long data_start = (unsigned long)_data;
532 	unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
533 
534 	set_memory_ro(text_start, (text_end - text_start) >> PAGE_SHIFT);
535 	set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
536 	set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
537 	set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
538 
539 	debug_checkwx();
540 }
541 #endif
542 
543 static void __init resource_init(void)
544 {
545 	struct memblock_region *region;
546 
547 	for_each_memblock(memory, region) {
548 		struct resource *res;
549 
550 		res = memblock_alloc(sizeof(struct resource), SMP_CACHE_BYTES);
551 		if (!res)
552 			panic("%s: Failed to allocate %zu bytes\n", __func__,
553 			      sizeof(struct resource));
554 
555 		if (memblock_is_nomap(region)) {
556 			res->name = "reserved";
557 			res->flags = IORESOURCE_MEM;
558 		} else {
559 			res->name = "System RAM";
560 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
561 		}
562 		res->start = __pfn_to_phys(memblock_region_memory_base_pfn(region));
563 		res->end = __pfn_to_phys(memblock_region_memory_end_pfn(region)) - 1;
564 
565 		request_resource(&iomem_resource, res);
566 	}
567 }
568 
569 void __init paging_init(void)
570 {
571 	setup_vm_final();
572 	sparse_init();
573 	setup_zero_page();
574 	zone_sizes_init();
575 	resource_init();
576 }
577 
578 #ifdef CONFIG_SPARSEMEM_VMEMMAP
579 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
580 			       struct vmem_altmap *altmap)
581 {
582 	return vmemmap_populate_basepages(start, end, node, NULL);
583 }
584 #endif
585