Commit | Line | Data |
---|---|---|
14cf11af PM |
1 | /* |
2 | * PowerPC version | |
3 | * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) | |
4 | * | |
5 | * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au) | |
6 | * and Cort Dougan (PReP) (cort@cs.nmt.edu) | |
7 | * Copyright (C) 1996 Paul Mackerras | |
14cf11af PM |
8 | * |
9 | * Derived from "arch/i386/mm/init.c" | |
10 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
11 | * | |
12 | * Dave Engebretsen <engebret@us.ibm.com> | |
13 | * Rework for PPC64 port. | |
14 | * | |
15 | * This program is free software; you can redistribute it and/or | |
16 | * modify it under the terms of the GNU General Public License | |
17 | * as published by the Free Software Foundation; either version | |
18 | * 2 of the License, or (at your option) any later version. | |
19 | * | |
20 | */ | |
21 | ||
cec08e7a BH |
22 | #undef DEBUG |
23 | ||
14cf11af PM |
24 | #include <linux/signal.h> |
25 | #include <linux/sched.h> | |
26 | #include <linux/kernel.h> | |
27 | #include <linux/errno.h> | |
28 | #include <linux/string.h> | |
29 | #include <linux/types.h> | |
30 | #include <linux/mman.h> | |
31 | #include <linux/mm.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/stddef.h> | |
34 | #include <linux/vmalloc.h> | |
35 | #include <linux/init.h> | |
36 | #include <linux/delay.h> | |
14cf11af PM |
37 | #include <linux/highmem.h> |
38 | #include <linux/idr.h> | |
39 | #include <linux/nodemask.h> | |
40 | #include <linux/module.h> | |
c9cf5528 | 41 | #include <linux/poison.h> |
95f72d1e | 42 | #include <linux/memblock.h> |
a4fe3ce7 | 43 | #include <linux/hugetlb.h> |
5a0e3ad6 | 44 | #include <linux/slab.h> |
18569c1f PM |
45 | #include <linux/of_fdt.h> |
46 | #include <linux/libfdt.h> | |
b584c254 | 47 | #include <linux/memremap.h> |
14cf11af PM |
48 | |
49 | #include <asm/pgalloc.h> | |
50 | #include <asm/page.h> | |
51 | #include <asm/prom.h> | |
14cf11af PM |
52 | #include <asm/rtas.h> |
53 | #include <asm/io.h> | |
54 | #include <asm/mmu_context.h> | |
55 | #include <asm/pgtable.h> | |
56 | #include <asm/mmu.h> | |
7c0f6ba6 | 57 | #include <linux/uaccess.h> |
14cf11af PM |
58 | #include <asm/smp.h> |
59 | #include <asm/machdep.h> | |
60 | #include <asm/tlb.h> | |
61 | #include <asm/eeh.h> | |
62 | #include <asm/processor.h> | |
63 | #include <asm/mmzone.h> | |
64 | #include <asm/cputable.h> | |
14cf11af | 65 | #include <asm/sections.h> |
14cf11af | 66 | #include <asm/iommu.h> |
14cf11af | 67 | #include <asm/vdso.h> |
800fc3ee DG |
68 | |
69 | #include "mmu_decl.h" | |
14cf11af | 70 | |
4e003747 | 71 | #ifdef CONFIG_PPC_BOOK3S_64 |
dd1842a2 | 72 | #if H_PGTABLE_RANGE > USER_VSID_RANGE |
14cf11af PM |
73 | #warning Limited user VSID range means pagetable space is wasted |
74 | #endif | |
4e003747 | 75 | #endif /* CONFIG_PPC_BOOK3S_64 */ |
14cf11af | 76 | |
37dd2bad | 77 | phys_addr_t memstart_addr = ~0; |
79c3095f | 78 | EXPORT_SYMBOL_GPL(memstart_addr); |
37dd2bad | 79 | phys_addr_t kernstart_addr; |
79c3095f | 80 | EXPORT_SYMBOL_GPL(kernstart_addr); |
d7917ba7 | 81 | |
d29eff7b AW |
82 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
83 | /* | |
84 | * Given an address within the vmemmap, determine the pfn of the page that | |
85 | * represents the start of the section it is within. Note that we have to | |
86 | * do this by hand as the proffered address may not be correctly aligned. | |
87 | * Subtraction of non-aligned pointers produces undefined results. | |
88 | */ | |
09de9ff8 | 89 | static unsigned long __meminit vmemmap_section_start(unsigned long page) |
d29eff7b AW |
90 | { |
91 | unsigned long offset = page - ((unsigned long)(vmemmap)); | |
92 | ||
93 | /* Return the pfn of the start of the section. */ | |
94 | return (offset / sizeof(struct page)) & PAGE_SECTION_MASK; | |
95 | } | |
96 | ||
97 | /* | |
98 | * Check if this vmemmap page is already initialised. If any section | |
99 | * which overlaps this vmemmap page is initialised then this page is | |
100 | * initialised already. | |
101 | */ | |
09de9ff8 | 102 | static int __meminit vmemmap_populated(unsigned long start, int page_size) |
d29eff7b AW |
103 | { |
104 | unsigned long end = start + page_size; | |
16a05bff | 105 | start = (unsigned long)(pfn_to_page(vmemmap_section_start(start))); |
d29eff7b AW |
106 | |
107 | for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page))) | |
16a05bff | 108 | if (pfn_valid(page_to_pfn((struct page *)start))) |
d29eff7b AW |
109 | return 1; |
110 | ||
111 | return 0; | |
112 | } | |
113 | ||
39e46751 AK |
114 | /* |
115 | * vmemmap virtual address space management does not have a traditonal page | |
116 | * table to track which virtual struct pages are backed by physical mapping. | |
117 | * The virtual to physical mappings are tracked in a simple linked list | |
118 | * format. 'vmemmap_list' maintains the entire vmemmap physical mapping at | |
119 | * all times where as the 'next' list maintains the available | |
120 | * vmemmap_backing structures which have been deleted from the | |
121 | * 'vmemmap_global' list during system runtime (memory hotplug remove | |
122 | * operation). The freed 'vmemmap_backing' structures are reused later when | |
123 | * new requests come in without allocating fresh memory. This pointer also | |
124 | * tracks the allocated 'vmemmap_backing' structures as we allocate one | |
125 | * full page memory at a time when we dont have any. | |
126 | */ | |
91eea67c | 127 | struct vmemmap_backing *vmemmap_list; |
bd8cb03d | 128 | static struct vmemmap_backing *next; |
39e46751 AK |
129 | |
130 | /* | |
131 | * The same pointer 'next' tracks individual chunks inside the allocated | |
132 | * full page during the boot time and again tracks the freeed nodes during | |
133 | * runtime. It is racy but it does not happen as they are separated by the | |
134 | * boot process. Will create problem if some how we have memory hotplug | |
135 | * operation during boot !! | |
136 | */ | |
bd8cb03d LZ |
137 | static int num_left; |
138 | static int num_freed; | |
91eea67c MN |
139 | |
140 | static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node) | |
141 | { | |
bd8cb03d LZ |
142 | struct vmemmap_backing *vmem_back; |
143 | /* get from freed entries first */ | |
144 | if (num_freed) { | |
145 | num_freed--; | |
146 | vmem_back = next; | |
147 | next = next->list; | |
148 | ||
149 | return vmem_back; | |
150 | } | |
91eea67c MN |
151 | |
152 | /* allocate a page when required and hand out chunks */ | |
bd8cb03d | 153 | if (!num_left) { |
91eea67c MN |
154 | next = vmemmap_alloc_block(PAGE_SIZE, node); |
155 | if (unlikely(!next)) { | |
156 | WARN_ON(1); | |
157 | return NULL; | |
158 | } | |
159 | num_left = PAGE_SIZE / sizeof(struct vmemmap_backing); | |
160 | } | |
161 | ||
162 | num_left--; | |
163 | ||
164 | return next++; | |
165 | } | |
166 | ||
167 | static __meminit void vmemmap_list_populate(unsigned long phys, | |
168 | unsigned long start, | |
169 | int node) | |
170 | { | |
171 | struct vmemmap_backing *vmem_back; | |
172 | ||
173 | vmem_back = vmemmap_list_alloc(node); | |
174 | if (unlikely(!vmem_back)) { | |
175 | WARN_ON(1); | |
176 | return; | |
177 | } | |
178 | ||
179 | vmem_back->phys = phys; | |
180 | vmem_back->virt_addr = start; | |
181 | vmem_back->list = vmemmap_list; | |
182 | ||
183 | vmemmap_list = vmem_back; | |
184 | } | |
185 | ||
7b73d978 CH |
186 | int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, |
187 | struct vmem_altmap *altmap) | |
71b0bfe4 LZ |
188 | { |
189 | unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift; | |
190 | ||
191 | /* Align to the page size of the linear mapping. */ | |
192 | start = _ALIGN_DOWN(start, page_size); | |
193 | ||
194 | pr_debug("vmemmap_populate %lx..%lx, node %d\n", start, end, node); | |
195 | ||
196 | for (; start < end; start += page_size) { | |
197 | void *p; | |
1dace6c6 | 198 | int rc; |
71b0bfe4 LZ |
199 | |
200 | if (vmemmap_populated(start, page_size)) | |
201 | continue; | |
202 | ||
b584c254 | 203 | p = __vmemmap_alloc_block_buf(page_size, node, altmap); |
71b0bfe4 LZ |
204 | if (!p) |
205 | return -ENOMEM; | |
206 | ||
207 | vmemmap_list_populate(__pa(p), start, node); | |
208 | ||
209 | pr_debug(" * %016lx..%016lx allocated at %p\n", | |
210 | start, start + page_size, p); | |
211 | ||
1dace6c6 DG |
212 | rc = vmemmap_create_mapping(start, page_size, __pa(p)); |
213 | if (rc < 0) { | |
214 | pr_warning( | |
215 | "vmemmap_populate: Unable to create vmemmap mapping: %d\n", | |
216 | rc); | |
217 | return -EFAULT; | |
218 | } | |
71b0bfe4 LZ |
219 | } |
220 | ||
221 | return 0; | |
222 | } | |
223 | ||
224 | #ifdef CONFIG_MEMORY_HOTPLUG | |
bd8cb03d LZ |
225 | static unsigned long vmemmap_list_free(unsigned long start) |
226 | { | |
227 | struct vmemmap_backing *vmem_back, *vmem_back_prev; | |
228 | ||
229 | vmem_back_prev = vmem_back = vmemmap_list; | |
230 | ||
231 | /* look for it with prev pointer recorded */ | |
232 | for (; vmem_back; vmem_back = vmem_back->list) { | |
233 | if (vmem_back->virt_addr == start) | |
234 | break; | |
235 | vmem_back_prev = vmem_back; | |
236 | } | |
237 | ||
238 | if (unlikely(!vmem_back)) { | |
239 | WARN_ON(1); | |
240 | return 0; | |
241 | } | |
242 | ||
243 | /* remove it from vmemmap_list */ | |
244 | if (vmem_back == vmemmap_list) /* remove head */ | |
245 | vmemmap_list = vmem_back->list; | |
246 | else | |
247 | vmem_back_prev->list = vmem_back->list; | |
248 | ||
249 | /* next point to this freed entry */ | |
250 | vmem_back->list = next; | |
251 | next = vmem_back; | |
252 | num_freed++; | |
253 | ||
254 | return vmem_back->phys; | |
255 | } | |
256 | ||
71b0bfe4 | 257 | void __ref vmemmap_free(unsigned long start, unsigned long end) |
d29eff7b | 258 | { |
cec08e7a | 259 | unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift; |
d7d9b612 | 260 | unsigned long page_order = get_order(page_size); |
d29eff7b | 261 | |
d29eff7b AW |
262 | start = _ALIGN_DOWN(start, page_size); |
263 | ||
71b0bfe4 | 264 | pr_debug("vmemmap_free %lx...%lx\n", start, end); |
32a74949 | 265 | |
d29eff7b | 266 | for (; start < end; start += page_size) { |
d7d9b612 | 267 | unsigned long nr_pages, addr; |
b584c254 OH |
268 | struct vmem_altmap *altmap; |
269 | struct page *section_base; | |
d7d9b612 | 270 | struct page *page; |
d29eff7b | 271 | |
71b0bfe4 LZ |
272 | /* |
273 | * the section has already be marked as invalid, so | |
274 | * vmemmap_populated() true means some other sections still | |
275 | * in this page, so skip it. | |
276 | */ | |
d29eff7b AW |
277 | if (vmemmap_populated(start, page_size)) |
278 | continue; | |
279 | ||
71b0bfe4 | 280 | addr = vmemmap_list_free(start); |
d7d9b612 OH |
281 | if (!addr) |
282 | continue; | |
283 | ||
284 | page = pfn_to_page(addr >> PAGE_SHIFT); | |
b584c254 | 285 | section_base = pfn_to_page(vmemmap_section_start(start)); |
d7d9b612 OH |
286 | nr_pages = 1 << page_order; |
287 | ||
b584c254 OH |
288 | altmap = to_vmem_altmap((unsigned long) section_base); |
289 | if (altmap) { | |
290 | vmem_altmap_free(altmap, nr_pages); | |
291 | } else if (PageReserved(page)) { | |
d7d9b612 OH |
292 | /* allocated from bootmem */ |
293 | if (page_size < PAGE_SIZE) { | |
294 | /* | |
295 | * this shouldn't happen, but if it is | |
296 | * the case, leave the memory there | |
297 | */ | |
298 | WARN_ON_ONCE(1); | |
299 | } else { | |
300 | while (nr_pages--) | |
301 | free_reserved_page(page++); | |
302 | } | |
303 | } else { | |
304 | free_pages((unsigned long)(__va(addr)), page_order); | |
71b0bfe4 | 305 | } |
d7d9b612 OH |
306 | |
307 | vmemmap_remove_mapping(start, page_size); | |
d29eff7b | 308 | } |
0197518c | 309 | } |
71b0bfe4 | 310 | #endif |
f7e3334a NF |
311 | void register_page_bootmem_memmap(unsigned long section_nr, |
312 | struct page *start_page, unsigned long size) | |
313 | { | |
314 | } | |
cd3db0c4 | 315 | |
8e0861fa AK |
316 | /* |
317 | * We do not have access to the sparsemem vmemmap, so we fallback to | |
318 | * walking the list of sparsemem blocks which we already maintain for | |
319 | * the sake of crashdump. In the long run, we might want to maintain | |
320 | * a tree if performance of that linear walk becomes a problem. | |
321 | * | |
322 | * realmode_pfn_to_page functions can fail due to: | |
323 | * 1) As real sparsemem blocks do not lay in RAM continously (they | |
324 | * are in virtual address space which is not available in the real mode), | |
325 | * the requested page struct can be split between blocks so get_page/put_page | |
326 | * may fail. | |
327 | * 2) When huge pages are used, the get_page/put_page API will fail | |
328 | * in real mode as the linked addresses in the page struct are virtual | |
329 | * too. | |
330 | */ | |
331 | struct page *realmode_pfn_to_page(unsigned long pfn) | |
332 | { | |
333 | struct vmemmap_backing *vmem_back; | |
334 | struct page *page; | |
335 | unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift; | |
336 | unsigned long pg_va = (unsigned long) pfn_to_page(pfn); | |
337 | ||
338 | for (vmem_back = vmemmap_list; vmem_back; vmem_back = vmem_back->list) { | |
339 | if (pg_va < vmem_back->virt_addr) | |
340 | continue; | |
341 | ||
bd8cb03d LZ |
342 | /* After vmemmap_list entry free is possible, need check all */ |
343 | if ((pg_va + sizeof(struct page)) <= | |
344 | (vmem_back->virt_addr + page_size)) { | |
345 | page = (struct page *) (vmem_back->phys + pg_va - | |
8e0861fa | 346 | vmem_back->virt_addr); |
bd8cb03d LZ |
347 | return page; |
348 | } | |
8e0861fa AK |
349 | } |
350 | ||
bd8cb03d | 351 | /* Probably that page struct is split between real pages */ |
8e0861fa AK |
352 | return NULL; |
353 | } | |
354 | EXPORT_SYMBOL_GPL(realmode_pfn_to_page); | |
355 | ||
7e7dc66a | 356 | #else |
8e0861fa AK |
357 | |
358 | struct page *realmode_pfn_to_page(unsigned long pfn) | |
359 | { | |
360 | struct page *page = pfn_to_page(pfn); | |
361 | return page; | |
362 | } | |
363 | EXPORT_SYMBOL_GPL(realmode_pfn_to_page); | |
364 | ||
7e7dc66a | 365 | #endif /* CONFIG_SPARSEMEM_VMEMMAP */ |
1a01dc87 | 366 | |
4e003747 | 367 | #ifdef CONFIG_PPC_BOOK3S_64 |
1fd6c022 ME |
368 | static bool disable_radix = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT); |
369 | ||
c610ec60 ME |
370 | static int __init parse_disable_radix(char *p) |
371 | { | |
1fd6c022 ME |
372 | bool val; |
373 | ||
374 | if (strlen(p) == 0) | |
375 | val = true; | |
376 | else if (kstrtobool(p, &val)) | |
377 | return -EINVAL; | |
378 | ||
379 | disable_radix = val; | |
380 | ||
c610ec60 ME |
381 | return 0; |
382 | } | |
383 | early_param("disable_radix", parse_disable_radix); | |
384 | ||
18569c1f | 385 | /* |
cc3d2940 PM |
386 | * If we're running under a hypervisor, we need to check the contents of |
387 | * /chosen/ibm,architecture-vec-5 to see if the hypervisor is willing to do | |
388 | * radix. If not, we clear the radix feature bit so we fall back to hash. | |
18569c1f | 389 | */ |
7559952e | 390 | static void __init early_check_vec5(void) |
18569c1f PM |
391 | { |
392 | unsigned long root, chosen; | |
393 | int size; | |
394 | const u8 *vec5; | |
014d02cb | 395 | u8 mmu_supported; |
18569c1f PM |
396 | |
397 | root = of_get_flat_dt_root(); | |
398 | chosen = of_get_flat_dt_subnode_by_name(root, "chosen"); | |
014d02cb SJS |
399 | if (chosen == -FDT_ERR_NOTFOUND) { |
400 | cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; | |
18569c1f | 401 | return; |
014d02cb | 402 | } |
18569c1f | 403 | vec5 = of_get_flat_dt_prop(chosen, "ibm,architecture-vec-5", &size); |
014d02cb SJS |
404 | if (!vec5) { |
405 | cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; | |
18569c1f | 406 | return; |
014d02cb SJS |
407 | } |
408 | if (size <= OV5_INDX(OV5_MMU_SUPPORT)) { | |
cc3d2940 | 409 | cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; |
014d02cb SJS |
410 | return; |
411 | } | |
412 | ||
413 | /* Check for supported configuration */ | |
414 | mmu_supported = vec5[OV5_INDX(OV5_MMU_SUPPORT)] & | |
415 | OV5_FEAT(OV5_MMU_SUPPORT); | |
416 | if (mmu_supported == OV5_FEAT(OV5_MMU_RADIX)) { | |
417 | /* Hypervisor only supports radix - check enabled && GTSE */ | |
418 | if (!early_radix_enabled()) { | |
419 | pr_warn("WARNING: Ignoring cmdline option disable_radix\n"); | |
420 | } | |
421 | if (!(vec5[OV5_INDX(OV5_RADIX_GTSE)] & | |
422 | OV5_FEAT(OV5_RADIX_GTSE))) { | |
423 | pr_warn("WARNING: Hypervisor doesn't support RADIX with GTSE\n"); | |
424 | } | |
425 | /* Do radix anyway - the hypervisor said we had to */ | |
426 | cur_cpu_spec->mmu_features |= MMU_FTR_TYPE_RADIX; | |
427 | } else if (mmu_supported == OV5_FEAT(OV5_MMU_HASH)) { | |
428 | /* Hypervisor only supports hash - disable radix */ | |
429 | cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; | |
430 | } | |
18569c1f PM |
431 | } |
432 | ||
1a01dc87 ME |
433 | void __init mmu_early_init_devtree(void) |
434 | { | |
c610ec60 | 435 | /* Disable radix mode based on kernel command line. */ |
fc36a903 | 436 | if (disable_radix) |
5a25b6f5 | 437 | cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; |
bacf9cf8 | 438 | |
18569c1f PM |
439 | /* |
440 | * Check /chosen/ibm,architecture-vec-5 if running as a guest. | |
441 | * When running bare-metal, we can use radix if we like | |
442 | * even though the ibm,architecture-vec-5 property created by | |
443 | * skiboot doesn't have the necessary bits set. | |
444 | */ | |
014d02cb | 445 | if (!(mfmsr() & MSR_HV)) |
18569c1f PM |
446 | early_check_vec5(); |
447 | ||
b8f1b4f8 | 448 | if (early_radix_enabled()) |
2537b09c ME |
449 | radix__early_init_devtree(); |
450 | else | |
bacf9cf8 | 451 | hash__early_init_devtree(); |
1a01dc87 | 452 | } |
4e003747 | 453 | #endif /* CONFIG_PPC_BOOK3S_64 */ |