xref: /openbmc/linux/arch/riscv/mm/init.c (revision 165f2d28)
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/pgtable.h>
21 #include <asm/io.h>
22 
23 #include "../kernel/head.h"
24 
25 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
26 							__page_aligned_bss;
27 EXPORT_SYMBOL(empty_zero_page);
28 
29 extern char _start[];
30 void *dtb_early_va;
31 
32 static void __init zone_sizes_init(void)
33 {
34 	unsigned long max_zone_pfns[MAX_NR_ZONES] = { 0, };
35 
36 #ifdef CONFIG_ZONE_DMA32
37 	max_zone_pfns[ZONE_DMA32] = PFN_DOWN(min(4UL * SZ_1G,
38 			(unsigned long) PFN_PHYS(max_low_pfn)));
39 #endif
40 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
41 
42 	free_area_init_nodes(max_zone_pfns);
43 }
44 
45 static void setup_zero_page(void)
46 {
47 	memset((void *)empty_zero_page, 0, PAGE_SIZE);
48 }
49 
50 #ifdef CONFIG_DEBUG_VM
51 static inline void print_mlk(char *name, unsigned long b, unsigned long t)
52 {
53 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld kB)\n", name, b, t,
54 		  (((t) - (b)) >> 10));
55 }
56 
57 static inline void print_mlm(char *name, unsigned long b, unsigned long t)
58 {
59 	pr_notice("%12s : 0x%08lx - 0x%08lx   (%4ld MB)\n", name, b, t,
60 		  (((t) - (b)) >> 20));
61 }
62 
63 static void print_vm_layout(void)
64 {
65 	pr_notice("Virtual kernel memory layout:\n");
66 	print_mlk("fixmap", (unsigned long)FIXADDR_START,
67 		  (unsigned long)FIXADDR_TOP);
68 	print_mlm("pci io", (unsigned long)PCI_IO_START,
69 		  (unsigned long)PCI_IO_END);
70 	print_mlm("vmemmap", (unsigned long)VMEMMAP_START,
71 		  (unsigned long)VMEMMAP_END);
72 	print_mlm("vmalloc", (unsigned long)VMALLOC_START,
73 		  (unsigned long)VMALLOC_END);
74 	print_mlm("lowmem", (unsigned long)PAGE_OFFSET,
75 		  (unsigned long)high_memory);
76 }
77 #else
78 static void print_vm_layout(void) { }
79 #endif /* CONFIG_DEBUG_VM */
80 
81 void __init mem_init(void)
82 {
83 #ifdef CONFIG_FLATMEM
84 	BUG_ON(!mem_map);
85 #endif /* CONFIG_FLATMEM */
86 
87 	high_memory = (void *)(__va(PFN_PHYS(max_low_pfn)));
88 	memblock_free_all();
89 
90 	mem_init_print_info(NULL);
91 	print_vm_layout();
92 }
93 
94 #ifdef CONFIG_BLK_DEV_INITRD
95 static void __init setup_initrd(void)
96 {
97 	unsigned long size;
98 
99 	if (initrd_start >= initrd_end) {
100 		pr_info("initrd not found or empty");
101 		goto disable;
102 	}
103 	if (__pa_symbol(initrd_end) > PFN_PHYS(max_low_pfn)) {
104 		pr_err("initrd extends beyond end of memory");
105 		goto disable;
106 	}
107 
108 	size = initrd_end - initrd_start;
109 	memblock_reserve(__pa_symbol(initrd_start), size);
110 	initrd_below_start_ok = 1;
111 
112 	pr_info("Initial ramdisk at: 0x%p (%lu bytes)\n",
113 		(void *)(initrd_start), size);
114 	return;
115 disable:
116 	pr_cont(" - disabling initrd\n");
117 	initrd_start = 0;
118 	initrd_end = 0;
119 }
120 #endif /* CONFIG_BLK_DEV_INITRD */
121 
122 static phys_addr_t dtb_early_pa __initdata;
123 
124 void __init setup_bootmem(void)
125 {
126 	struct memblock_region *reg;
127 	phys_addr_t mem_size = 0;
128 	phys_addr_t vmlinux_end = __pa_symbol(&_end);
129 	phys_addr_t vmlinux_start = __pa_symbol(&_start);
130 
131 	/* Find the memory region containing the kernel */
132 	for_each_memblock(memory, reg) {
133 		phys_addr_t end = reg->base + reg->size;
134 
135 		if (reg->base <= vmlinux_start && vmlinux_end <= end) {
136 			mem_size = min(reg->size, (phys_addr_t)-PAGE_OFFSET);
137 
138 			/*
139 			 * Remove memblock from the end of usable area to the
140 			 * end of region
141 			 */
142 			if (reg->base + mem_size < end)
143 				memblock_remove(reg->base + mem_size,
144 						end - reg->base - mem_size);
145 		}
146 	}
147 	BUG_ON(mem_size == 0);
148 
149 	/* Reserve from the start of the kernel to the end of the kernel */
150 	memblock_reserve(vmlinux_start, vmlinux_end - vmlinux_start);
151 
152 	set_max_mapnr(PFN_DOWN(mem_size));
153 	max_pfn = PFN_DOWN(memblock_end_of_DRAM());
154 	max_low_pfn = max_pfn;
155 
156 #ifdef CONFIG_BLK_DEV_INITRD
157 	setup_initrd();
158 #endif /* CONFIG_BLK_DEV_INITRD */
159 
160 	/*
161 	 * Avoid using early_init_fdt_reserve_self() since __pa() does
162 	 * not work for DTB pointers that are fixmap addresses
163 	 */
164 	memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
165 
166 	early_init_fdt_scan_reserved_mem();
167 	memblock_allow_resize();
168 	memblock_dump_all();
169 
170 	for_each_memblock(memory, reg) {
171 		unsigned long start_pfn = memblock_region_memory_base_pfn(reg);
172 		unsigned long end_pfn = memblock_region_memory_end_pfn(reg);
173 
174 		memblock_set_node(PFN_PHYS(start_pfn),
175 				  PFN_PHYS(end_pfn - start_pfn),
176 				  &memblock.memory, 0);
177 	}
178 }
179 
180 #ifdef CONFIG_MMU
181 unsigned long va_pa_offset;
182 EXPORT_SYMBOL(va_pa_offset);
183 unsigned long pfn_base;
184 EXPORT_SYMBOL(pfn_base);
185 
186 pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
187 pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss;
188 pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
189 static bool mmu_enabled;
190 
191 #define MAX_EARLY_MAPPING_SIZE	SZ_128M
192 
193 pgd_t early_pg_dir[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE);
194 
195 void __set_fixmap(enum fixed_addresses idx, phys_addr_t phys, pgprot_t prot)
196 {
197 	unsigned long addr = __fix_to_virt(idx);
198 	pte_t *ptep;
199 
200 	BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
201 
202 	ptep = &fixmap_pte[pte_index(addr)];
203 
204 	if (pgprot_val(prot)) {
205 		set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, prot));
206 	} else {
207 		pte_clear(&init_mm, addr, ptep);
208 		local_flush_tlb_page(addr);
209 	}
210 }
211 
212 static pte_t *__init get_pte_virt(phys_addr_t pa)
213 {
214 	if (mmu_enabled) {
215 		clear_fixmap(FIX_PTE);
216 		return (pte_t *)set_fixmap_offset(FIX_PTE, pa);
217 	} else {
218 		return (pte_t *)((uintptr_t)pa);
219 	}
220 }
221 
222 static phys_addr_t __init alloc_pte(uintptr_t va)
223 {
224 	/*
225 	 * We only create PMD or PGD early mappings so we
226 	 * should never reach here with MMU disabled.
227 	 */
228 	BUG_ON(!mmu_enabled);
229 
230 	return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
231 }
232 
233 static void __init create_pte_mapping(pte_t *ptep,
234 				      uintptr_t va, phys_addr_t pa,
235 				      phys_addr_t sz, pgprot_t prot)
236 {
237 	uintptr_t pte_index = pte_index(va);
238 
239 	BUG_ON(sz != PAGE_SIZE);
240 
241 	if (pte_none(ptep[pte_index]))
242 		ptep[pte_index] = pfn_pte(PFN_DOWN(pa), prot);
243 }
244 
245 #ifndef __PAGETABLE_PMD_FOLDED
246 
247 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
248 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
249 
250 #if MAX_EARLY_MAPPING_SIZE < PGDIR_SIZE
251 #define NUM_EARLY_PMDS		1UL
252 #else
253 #define NUM_EARLY_PMDS		(1UL + MAX_EARLY_MAPPING_SIZE / PGDIR_SIZE)
254 #endif
255 pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
256 
257 static pmd_t *__init get_pmd_virt(phys_addr_t pa)
258 {
259 	if (mmu_enabled) {
260 		clear_fixmap(FIX_PMD);
261 		return (pmd_t *)set_fixmap_offset(FIX_PMD, pa);
262 	} else {
263 		return (pmd_t *)((uintptr_t)pa);
264 	}
265 }
266 
267 static phys_addr_t __init alloc_pmd(uintptr_t va)
268 {
269 	uintptr_t pmd_num;
270 
271 	if (mmu_enabled)
272 		return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
273 
274 	pmd_num = (va - PAGE_OFFSET) >> PGDIR_SHIFT;
275 	BUG_ON(pmd_num >= NUM_EARLY_PMDS);
276 	return (uintptr_t)&early_pmd[pmd_num * PTRS_PER_PMD];
277 }
278 
279 static void __init create_pmd_mapping(pmd_t *pmdp,
280 				      uintptr_t va, phys_addr_t pa,
281 				      phys_addr_t sz, pgprot_t prot)
282 {
283 	pte_t *ptep;
284 	phys_addr_t pte_phys;
285 	uintptr_t pmd_index = pmd_index(va);
286 
287 	if (sz == PMD_SIZE) {
288 		if (pmd_none(pmdp[pmd_index]))
289 			pmdp[pmd_index] = pfn_pmd(PFN_DOWN(pa), prot);
290 		return;
291 	}
292 
293 	if (pmd_none(pmdp[pmd_index])) {
294 		pte_phys = alloc_pte(va);
295 		pmdp[pmd_index] = pfn_pmd(PFN_DOWN(pte_phys), PAGE_TABLE);
296 		ptep = get_pte_virt(pte_phys);
297 		memset(ptep, 0, PAGE_SIZE);
298 	} else {
299 		pte_phys = PFN_PHYS(_pmd_pfn(pmdp[pmd_index]));
300 		ptep = get_pte_virt(pte_phys);
301 	}
302 
303 	create_pte_mapping(ptep, va, pa, sz, prot);
304 }
305 
306 #define pgd_next_t		pmd_t
307 #define alloc_pgd_next(__va)	alloc_pmd(__va)
308 #define get_pgd_next_virt(__pa)	get_pmd_virt(__pa)
309 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
310 	create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
311 #define fixmap_pgd_next		fixmap_pmd
312 #else
313 #define pgd_next_t		pte_t
314 #define alloc_pgd_next(__va)	alloc_pte(__va)
315 #define get_pgd_next_virt(__pa)	get_pte_virt(__pa)
316 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
317 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
318 #define fixmap_pgd_next		fixmap_pte
319 #endif
320 
321 static void __init create_pgd_mapping(pgd_t *pgdp,
322 				      uintptr_t va, phys_addr_t pa,
323 				      phys_addr_t sz, pgprot_t prot)
324 {
325 	pgd_next_t *nextp;
326 	phys_addr_t next_phys;
327 	uintptr_t pgd_index = pgd_index(va);
328 
329 	if (sz == PGDIR_SIZE) {
330 		if (pgd_val(pgdp[pgd_index]) == 0)
331 			pgdp[pgd_index] = pfn_pgd(PFN_DOWN(pa), prot);
332 		return;
333 	}
334 
335 	if (pgd_val(pgdp[pgd_index]) == 0) {
336 		next_phys = alloc_pgd_next(va);
337 		pgdp[pgd_index] = pfn_pgd(PFN_DOWN(next_phys), PAGE_TABLE);
338 		nextp = get_pgd_next_virt(next_phys);
339 		memset(nextp, 0, PAGE_SIZE);
340 	} else {
341 		next_phys = PFN_PHYS(_pgd_pfn(pgdp[pgd_index]));
342 		nextp = get_pgd_next_virt(next_phys);
343 	}
344 
345 	create_pgd_next_mapping(nextp, va, pa, sz, prot);
346 }
347 
348 static uintptr_t __init best_map_size(phys_addr_t base, phys_addr_t size)
349 {
350 	/* Upgrade to PMD_SIZE mappings whenever possible */
351 	if ((base & (PMD_SIZE - 1)) || (size & (PMD_SIZE - 1)))
352 		return PAGE_SIZE;
353 
354 	return PMD_SIZE;
355 }
356 
357 /*
358  * setup_vm() is called from head.S with MMU-off.
359  *
360  * Following requirements should be honoured for setup_vm() to work
361  * correctly:
362  * 1) It should use PC-relative addressing for accessing kernel symbols.
363  *    To achieve this we always use GCC cmodel=medany.
364  * 2) The compiler instrumentation for FTRACE will not work for setup_vm()
365  *    so disable compiler instrumentation when FTRACE is enabled.
366  *
367  * Currently, the above requirements are honoured by using custom CFLAGS
368  * for init.o in mm/Makefile.
369  */
370 
371 #ifndef __riscv_cmodel_medany
372 #error "setup_vm() is called from head.S before relocate so it should not use absolute addressing."
373 #endif
374 
375 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
376 {
377 	uintptr_t va, end_va;
378 	uintptr_t load_pa = (uintptr_t)(&_start);
379 	uintptr_t load_sz = (uintptr_t)(&_end) - load_pa;
380 	uintptr_t map_size = best_map_size(load_pa, MAX_EARLY_MAPPING_SIZE);
381 
382 	va_pa_offset = PAGE_OFFSET - load_pa;
383 	pfn_base = PFN_DOWN(load_pa);
384 
385 	/*
386 	 * Enforce boot alignment requirements of RV32 and
387 	 * RV64 by only allowing PMD or PGD mappings.
388 	 */
389 	BUG_ON(map_size == PAGE_SIZE);
390 
391 	/* Sanity check alignment and size */
392 	BUG_ON((PAGE_OFFSET % PGDIR_SIZE) != 0);
393 	BUG_ON((load_pa % map_size) != 0);
394 	BUG_ON(load_sz > MAX_EARLY_MAPPING_SIZE);
395 
396 	/* Setup early PGD for fixmap */
397 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
398 			   (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
399 
400 #ifndef __PAGETABLE_PMD_FOLDED
401 	/* Setup fixmap PMD */
402 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
403 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
404 	/* Setup trampoline PGD and PMD */
405 	create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
406 			   (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
407 	create_pmd_mapping(trampoline_pmd, PAGE_OFFSET,
408 			   load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
409 #else
410 	/* Setup trampoline PGD */
411 	create_pgd_mapping(trampoline_pg_dir, PAGE_OFFSET,
412 			   load_pa, PGDIR_SIZE, PAGE_KERNEL_EXEC);
413 #endif
414 
415 	/*
416 	 * Setup early PGD covering entire kernel which will allows
417 	 * us to reach paging_init(). We map all memory banks later
418 	 * in setup_vm_final() below.
419 	 */
420 	end_va = PAGE_OFFSET + load_sz;
421 	for (va = PAGE_OFFSET; va < end_va; va += map_size)
422 		create_pgd_mapping(early_pg_dir, va,
423 				   load_pa + (va - PAGE_OFFSET),
424 				   map_size, PAGE_KERNEL_EXEC);
425 
426 	/* Create fixed mapping for early FDT parsing */
427 	end_va = __fix_to_virt(FIX_FDT) + FIX_FDT_SIZE;
428 	for (va = __fix_to_virt(FIX_FDT); va < end_va; va += PAGE_SIZE)
429 		create_pte_mapping(fixmap_pte, va,
430 				   dtb_pa + (va - __fix_to_virt(FIX_FDT)),
431 				   PAGE_SIZE, PAGE_KERNEL);
432 
433 	/* Save pointer to DTB for early FDT parsing */
434 	dtb_early_va = (void *)fix_to_virt(FIX_FDT) + (dtb_pa & ~PAGE_MASK);
435 	/* Save physical address for memblock reservation */
436 	dtb_early_pa = dtb_pa;
437 }
438 
439 static void __init setup_vm_final(void)
440 {
441 	uintptr_t va, map_size;
442 	phys_addr_t pa, start, end;
443 	struct memblock_region *reg;
444 
445 	/* Set mmu_enabled flag */
446 	mmu_enabled = true;
447 
448 	/* Setup swapper PGD for fixmap */
449 	create_pgd_mapping(swapper_pg_dir, FIXADDR_START,
450 			   __pa_symbol(fixmap_pgd_next),
451 			   PGDIR_SIZE, PAGE_TABLE);
452 
453 	/* Map all memory banks */
454 	for_each_memblock(memory, reg) {
455 		start = reg->base;
456 		end = start + reg->size;
457 
458 		if (start >= end)
459 			break;
460 		if (memblock_is_nomap(reg))
461 			continue;
462 		if (start <= __pa(PAGE_OFFSET) &&
463 		    __pa(PAGE_OFFSET) < end)
464 			start = __pa(PAGE_OFFSET);
465 
466 		map_size = best_map_size(start, end - start);
467 		for (pa = start; pa < end; pa += map_size) {
468 			va = (uintptr_t)__va(pa);
469 			create_pgd_mapping(swapper_pg_dir, va, pa,
470 					   map_size, PAGE_KERNEL_EXEC);
471 		}
472 	}
473 
474 	/* Clear fixmap PTE and PMD mappings */
475 	clear_fixmap(FIX_PTE);
476 	clear_fixmap(FIX_PMD);
477 
478 	/* Move to swapper page table */
479 	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
480 	local_flush_tlb_all();
481 }
482 
483 void free_initmem(void)
484 {
485 	unsigned long init_begin = (unsigned long)__init_begin;
486 	unsigned long init_end = (unsigned long)__init_end;
487 
488 	/* Make the region as non-execuatble. */
489 	set_memory_nx(init_begin, (init_end - init_begin) >> PAGE_SHIFT);
490 	free_initmem_default(POISON_FREE_INITMEM);
491 }
492 
493 #else
494 asmlinkage void __init setup_vm(uintptr_t dtb_pa)
495 {
496 	dtb_early_va = (void *)dtb_pa;
497 }
498 
499 static inline void setup_vm_final(void)
500 {
501 }
502 #endif /* CONFIG_MMU */
503 
504 #ifdef CONFIG_STRICT_KERNEL_RWX
505 void mark_rodata_ro(void)
506 {
507 	unsigned long text_start = (unsigned long)_text;
508 	unsigned long text_end = (unsigned long)_etext;
509 	unsigned long rodata_start = (unsigned long)__start_rodata;
510 	unsigned long data_start = (unsigned long)_data;
511 	unsigned long max_low = (unsigned long)(__va(PFN_PHYS(max_low_pfn)));
512 
513 	set_memory_ro(text_start, (text_end - text_start) >> PAGE_SHIFT);
514 	set_memory_ro(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
515 	set_memory_nx(rodata_start, (data_start - rodata_start) >> PAGE_SHIFT);
516 	set_memory_nx(data_start, (max_low - data_start) >> PAGE_SHIFT);
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