xref: /openbmc/linux/arch/riscv/mm/init.c (revision f0838790)
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 	unsigned long size;
99 
100 	if (initrd_start >= initrd_end) {
101 		pr_info("initrd not found or empty");
102 		goto disable;
103 	}
104 	if (__pa_symbol(initrd_end) > PFN_PHYS(max_low_pfn)) {
105 		pr_err("initrd extends beyond end of memory");
106 		goto disable;
107 	}
108 
109 	size = initrd_end - initrd_start;
110 	memblock_reserve(__pa_symbol(initrd_start), size);
111 	initrd_below_start_ok = 1;
112 
113 	pr_info("Initial ramdisk at: 0x%p (%lu bytes)\n",
114 		(void *)(initrd_start), size);
115 	return;
116 disable:
117 	pr_cont(" - disabling initrd\n");
118 	initrd_start = 0;
119 	initrd_end = 0;
120 }
121 #endif /* CONFIG_BLK_DEV_INITRD */
122 
123 static phys_addr_t dtb_early_pa __initdata;
124 
125 void __init setup_bootmem(void)
126 {
127 	struct memblock_region *reg;
128 	phys_addr_t mem_size = 0;
129 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
130 	phys_addr_t vmlinux_start = __pa_symbol(&_start);
131 
132 	/* Find the memory region containing the kernel */
133 	for_each_memblock(memory, reg) {
134 		phys_addr_t end = reg->base + reg->size;
135 
136 		if (reg->base <= vmlinux_start && vmlinux_end <= end) {
137 			mem_size = min(reg->size, (phys_addr_t)-PAGE_OFFSET);
138 
139 			/*
140 			 * Remove memblock from the end of usable area to the
141 			 * end of region
142 			 */
143 			if (reg->base + mem_size < end)
144 				memblock_remove(reg->base + mem_size,
145 						end - reg->base - mem_size);
146 		}
147 	}
148 	BUG_ON(mem_size == 0);
149 
150 	/* Reserve from the start of the kernel to the end of the kernel */
151 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
152 
153 	set_max_mapnr(PFN_DOWN(mem_size));
154 	max_pfn = PFN_DOWN(memblock_end_of_DRAM());
155 	max_low_pfn = max_pfn;
156 
157 #ifdef CONFIG_BLK_DEV_INITRD
158 	setup_initrd();
159 #endif /* CONFIG_BLK_DEV_INITRD */
160 
161 	/*
162 	 * Avoid using early_init_fdt_reserve_self() since __pa() does
163 	 * not work for DTB pointers that are fixmap addresses
164 	 */
165 	memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
166 
167 	early_init_fdt_scan_reserved_mem();
168 	memblock_allow_resize();
169 	memblock_dump_all();
170 
171 	for_each_memblock(memory, reg) {
172 		unsigned long start_pfn = memblock_region_memory_base_pfn(reg);
173 		unsigned long end_pfn = memblock_region_memory_end_pfn(reg);
174 
175 		memblock_set_node(PFN_PHYS(start_pfn),
176 				  PFN_PHYS(end_pfn - start_pfn),
177 				  &memblock.memory, 0);
178 	}
179 }
180 
181 #ifdef CONFIG_MMU
182 unsigned long va_pa_offset;
183 EXPORT_SYMBOL(va_pa_offset);
184 unsigned long pfn_base;
185 EXPORT_SYMBOL(pfn_base);
186 
187 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
188 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
189 pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
190 static bool mmu_enabled;
191 
192 #define MAX_EARLY_MAPPING_SIZE	SZ_128M
193 
194 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
195 
196 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
197 {
198 	unsigned long addr = __fix_to_virt(idx);
199 	pte_t *ptep;
200 
201 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
202 
203 	ptep = &fixmap_pte[pte_index(addr)];
204 
205 	if (pgprot_val(prot)) {
206 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
207 	} else {
208 		pte_clear(&init_mm, addr, ptep);
209 		local_flush_tlb_page(addr);
210 	}
211 }
212 
213 static pte_t *__init get_pte_virt(phys_addr_t pa)
214 {
215 	if (mmu_enabled) {
216 		clear_fixmap(FIX_PTE);
217 		return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
218 	} else {
219 		return (pte_t *)((uintptr_t)pa);
220 	}
221 }
222 
223 static phys_addr_t __init alloc_pte(uintptr_t va)
224 {
225 	/*
226 	 * We only create PMD or PGD early mappings so we
227 	 * should never reach here with MMU disabled.
228 	 */
229 	BUG_ON(!mmu_enabled);
230 
231 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
232 }
233 
234 static void __init create_pte_mapping(pte_t *ptep,
235 				      uintptr_t va, phys_addr_t pa,
236 				      phys_addr_t sz, pgprot_t prot)
237 {
238 	uintptr_t pte_idx = pte_index(va);
239 
240 	BUG_ON(sz != PAGE_SIZE);
241 
242 	if (pte_none(ptep[pte_idx]))
243 		ptep[pte_idx] = pfn_pte(PFN_DOWN(pa), prot);
244 }
245 
246 #ifndef __PAGETABLE_PMD_FOLDED
247 
248 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
249 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
250 
251 #if MAX_EARLY_MAPPING_SIZE < PGDIR_SIZE
252 #define NUM_EARLY_PMDS		1UL
253 #else
254 #define NUM_EARLY_PMDS		(1UL + MAX_EARLY_MAPPING_SIZE / PGDIR_SIZE)
255 #endif
256 pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
257 
258 static pmd_t *__init get_pmd_virt(phys_addr_t pa)
259 {
260 	if (mmu_enabled) {
261 		clear_fixmap(FIX_PMD);
262 		return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
263 	} else {
264 		return (pmd_t *)((uintptr_t)pa);
265 	}
266 }
267 
268 static phys_addr_t __init alloc_pmd(uintptr_t va)
269 {
270 	uintptr_t pmd_num;
271 
272 	if (mmu_enabled)
273 		return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
274 
275 	pmd_num = (va - PAGE_OFFSET) >> PGDIR_SHIFT;
276 	BUG_ON(pmd_num >= NUM_EARLY_PMDS);
277 	return (uintptr_t)&early_pmd[pmd_num * PTRS_PER_PMD];
278 }
279 
280 static void __init create_pmd_mapping(pmd_t *pmdp,
281 				      uintptr_t va, phys_addr_t pa,
282 				      phys_addr_t sz, pgprot_t prot)
283 {
284 	pte_t *ptep;
285 	phys_addr_t pte_phys;
286 	uintptr_t pmd_idx = pmd_index(va);
287 
288 	if (sz == PMD_SIZE) {
289 		if (pmd_none(pmdp[pmd_idx]))
290 			pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pa), prot);
291 		return;
292 	}
293 
294 	if (pmd_none(pmdp[pmd_idx])) {
295 		pte_phys = alloc_pte(va);
296 		pmdp[pmd_idx] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
297 		ptep = get_pte_virt(pte_phys);
298 		memset(ptep, 0, PAGE_SIZE);
299 	} else {
300 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_idx]));
301 		ptep = get_pte_virt(pte_phys);
302 	}
303 
304 	create_pte_mapping(ptep, va, pa, sz, prot);
305 }
306 
307 #define pgd_next_t		pmd_t
308 #define alloc_pgd_next(__va)	alloc_pmd(__va)
309 #define get_pgd_next_virt(__pa)	get_pmd_virt(__pa)
310 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
311 	create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
312 #define fixmap_pgd_next		fixmap_pmd
313 #else
314 #define pgd_next_t		pte_t
315 #define alloc_pgd_next(__va)	alloc_pte(__va)
316 #define get_pgd_next_virt(__pa)	get_pte_virt(__pa)
317 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
318 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
319 #define fixmap_pgd_next		fixmap_pte
320 #endif
321 
322 static void __init create_pgd_mapping(pgd_t *pgdp,
323 				      uintptr_t va, phys_addr_t pa,
324 				      phys_addr_t sz, pgprot_t prot)
325 {
326 	pgd_next_t *nextp;
327 	phys_addr_t next_phys;
328 	uintptr_t pgd_idx = pgd_index(va);
329 
330 	if (sz == PGDIR_SIZE) {
331 		if (pgd_val(pgdp[pgd_idx]) == 0)
332 			pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(pa), prot);
333 		return;
334 	}
335 
336 	if (pgd_val(pgdp[pgd_idx]) == 0) {
337 		next_phys = alloc_pgd_next(va);
338 		pgdp[pgd_idx] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
339 		nextp = get_pgd_next_virt(next_phys);
340 		memset(nextp, 0, PAGE_SIZE);
341 	} else {
342 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_idx]));
343 		nextp = get_pgd_next_virt(next_phys);
344 	}
345 
346 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
347 }
348 
349 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
350 {
351 	/* Upgrade to PMD_SIZE mappings whenever possible */
352 	if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
353 		return PAGE_SIZE;
354 
355 	return PMD_SIZE;
356 }
357 
358 /*
359  * setup_vm() is called from head.S with MMU-off.
360  *
361  * Following requirements should be honoured for setup_vm() to work
362  * correctly:
363  * 1) It should use PC-relative addressing for accessing kernel symbols.
364  *    To achieve this we always use GCC cmodel=medany.
365  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
366  *    so disable compiler instrumentation when FTRACE is enabled.
367  *
368  * Currently, the above requirements are honoured by using custom CFLAGS
369  * for init.o in mm/Makefile.
370  */
371 
372 #ifndef __riscv_cmodel_medany
373 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
374 #endif
375 
376 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
377 {
378 	uintptr_t va, end_va;
379 	uintptr_t load_pa = (uintptr_t)(&_start);
380 	uintptr_t load_sz = (uintptr_t)(&_end) - load_pa;
381 	uintptr_t map_size = best_map_size(load_pa, MAX_EARLY_MAPPING_SIZE);
382 
383 	va_pa_offset = PAGE_OFFSET - load_pa;
384 	pfn_base = PFN_DOWN(load_pa);
385 
386 	/*
387 	 * Enforce boot alignment requirements of RV32 and
388 	 * RV64 by only allowing PMD or PGD mappings.
389 	 */
390 	BUG_ON(map_size == PAGE_SIZE);
391 
392 	/* Sanity check alignment and size */
393 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
394 	BUG_ON((load_pa % map_size) != 0);
395 	BUG_ON(load_sz > MAX_EARLY_MAPPING_SIZE);
396 
397 	/* Setup early PGD for fixmap */
398 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
399 			   (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
400 
401 #ifndef __PAGETABLE_PMD_FOLDED
402 	/* Setup fixmap PMD */
403 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
404 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
405 	/* Setup trampoline PGD and PMD */
406 	create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
407 			   (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
408 	create_pmd_mapping(trampoline_pmd, PAGE_OFFSET,
409 			   load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
410 #else
411 	/* Setup trampoline PGD */
412 	create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
413 			   load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC);
414 #endif
415 
416 	/*
417 	 * Setup early PGD covering entire kernel which will allows
418 	 * us to reach paging_init(). We map all memory banks later
419 	 * in setup_vm_final() below.
420 	 */
421 	end_va = PAGE_OFFSET + load_sz;
422 	for (va = PAGE_OFFSET; va < end_va; va += map_size)
423 		create_pgd_mapping(early_pg_dir, va,
424 				   load_pa + (va - PAGE_OFFSET),
425 				   map_size, PAGE_KERNEL_EXEC);
426 
427 	/* Create fixed mapping for early FDT parsing */
428 	end_va = __fix_to_virt(FIX_FDT) + FIX_FDT_SIZE;
429 	for (va = __fix_to_virt(FIX_FDT); va < end_va; va += PAGE_SIZE)
430 		create_pte_mapping(fixmap_pte, va,
431 				   dtb_pa + (va - __fix_to_virt(FIX_FDT)),
432 				   PAGE_SIZE, PAGE_KERNEL);
433 
434 	/* Save pointer to DTB for early FDT parsing */
435 	dtb_early_va = (void *)fix_to_virt(FIX_FDT) + (dtb_pa & ~PAGE_MASK);
436 	/* Save physical address for memblock reservation */
437 	dtb_early_pa = dtb_pa;
438 }
439 
440 static void __init setup_vm_final(void)
441 {
442 	uintptr_t va, map_size;
443 	phys_addr_t pa, start, end;
444 	struct memblock_region *reg;
445 
446 	/* Set mmu_enabled flag */
447 	mmu_enabled = true;
448 
449 	/* Setup swapper PGD for fixmap */
450 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
451 			   __pa_symbol(fixmap_pgd_next),
452 			   PGDIR_SIZE, PAGE_TABLE);
453 
454 	/* Map all memory banks */
455 	for_each_memblock(memory, reg) {
456 		start = reg->base;
457 		end = start + reg->size;
458 
459 		if (start >= end)
460 			break;
461 		if (memblock_is_nomap(reg))
462 			continue;
463 		if (start <= __pa(PAGE_OFFSET) &&
464 		    __pa(PAGE_OFFSET) < end)
465 			start = __pa(PAGE_OFFSET);
466 
467 		map_size = best_map_size(start, end - start);
468 		for (pa = start; pa < end; pa += map_size) {
469 			va = (uintptr_t)__va(pa);
470 			create_pgd_mapping(swapper_pg_dir, va, pa,
471 					   map_size, PAGE_KERNEL_EXEC);
472 		}
473 	}
474 
475 	/* Clear fixmap PTE and PMD mappings */
476 	clear_fixmap(FIX_PTE);
477 	clear_fixmap(FIX_PMD);
478 
479 	/* Move to swapper page table */
480 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
481 	local_flush_tlb_all();
482 }
483 #else
484 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
485 {
486 #ifdef CONFIG_BUILTIN_DTB
487 	dtb_early_va = soc_lookup_builtin_dtb();
488 	if (!dtb_early_va) {
489 		/* Fallback to first available DTS */
490 		dtb_early_va = (void *) __dtb_start;
491 	}
492 #else
493 	dtb_early_va = (void *)dtb_pa;
494 #endif
495 }
496 
497 static inline void setup_vm_final(void)
498 {
499 }
500 #endif /* CONFIG_MMU */
501 
502 #ifdef CONFIG_STRICT_KERNEL_RWX
503 void mark_rodata_ro(void)
504 {
505 	unsigned long text_start = (unsigned long)_text;
506 	unsigned long text_end = (unsigned long)_etext;
507 	unsigned long rodata_start = (unsigned long)__start_rodata;
508 	unsigned long data_start = (unsigned long)_data;
509 	unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
510 
511 	set_memory_ro(text_start, (text_end - text_start) >> PAGE_SHIFT);
512 	set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
513 	set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
514 	set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
515 
516 	debug_checkwx();
517 }
518 #endif
519 
520 void __init paging_init(void)
521 {
522 	setup_vm_final();
523 	memblocks_present();
524 	sparse_init();
525 	setup_zero_page();
526 	zone_sizes_init();
527 }
528 
529 #ifdef CONFIG_SPARSEMEM_VMEMMAP
530 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
531 			       struct vmem_altmap *altmap)
532 {
533 	return vmemmap_populate_basepages(start, end, node);
534 }
535 #endif
536