]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame_incremental - arch/powerpc/mm/init_64.c
UBUNTU: Ubuntu-5.3.0-29.31
[mirror_ubuntu-eoan-kernel.git] / arch / powerpc / mm / init_64.c
... / ...
CommitLineData
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * PowerPC version
4 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
5 *
6 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
7 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
8 * Copyright (C) 1996 Paul Mackerras
9 *
10 * Derived from "arch/i386/mm/init.c"
11 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
12 *
13 * Dave Engebretsen <engebret@us.ibm.com>
14 * Rework for PPC64 port.
15 */
16
17#undef DEBUG
18
19#include <linux/signal.h>
20#include <linux/sched.h>
21#include <linux/kernel.h>
22#include <linux/errno.h>
23#include <linux/string.h>
24#include <linux/types.h>
25#include <linux/mman.h>
26#include <linux/mm.h>
27#include <linux/swap.h>
28#include <linux/stddef.h>
29#include <linux/vmalloc.h>
30#include <linux/init.h>
31#include <linux/delay.h>
32#include <linux/highmem.h>
33#include <linux/idr.h>
34#include <linux/nodemask.h>
35#include <linux/module.h>
36#include <linux/poison.h>
37#include <linux/memblock.h>
38#include <linux/hugetlb.h>
39#include <linux/slab.h>
40#include <linux/of_fdt.h>
41#include <linux/libfdt.h>
42#include <linux/memremap.h>
43
44#include <asm/pgalloc.h>
45#include <asm/page.h>
46#include <asm/prom.h>
47#include <asm/rtas.h>
48#include <asm/io.h>
49#include <asm/mmu_context.h>
50#include <asm/pgtable.h>
51#include <asm/mmu.h>
52#include <linux/uaccess.h>
53#include <asm/smp.h>
54#include <asm/machdep.h>
55#include <asm/tlb.h>
56#include <asm/eeh.h>
57#include <asm/processor.h>
58#include <asm/mmzone.h>
59#include <asm/cputable.h>
60#include <asm/sections.h>
61#include <asm/iommu.h>
62#include <asm/vdso.h>
63
64#include <mm/mmu_decl.h>
65
66phys_addr_t memstart_addr = ~0;
67EXPORT_SYMBOL_GPL(memstart_addr);
68phys_addr_t kernstart_addr;
69EXPORT_SYMBOL_GPL(kernstart_addr);
70
71#ifdef CONFIG_SPARSEMEM_VMEMMAP
72/*
73 * Given an address within the vmemmap, determine the pfn of the page that
74 * represents the start of the section it is within. Note that we have to
75 * do this by hand as the proffered address may not be correctly aligned.
76 * Subtraction of non-aligned pointers produces undefined results.
77 */
78static unsigned long __meminit vmemmap_section_start(unsigned long page)
79{
80 unsigned long offset = page - ((unsigned long)(vmemmap));
81
82 /* Return the pfn of the start of the section. */
83 return (offset / sizeof(struct page)) & PAGE_SECTION_MASK;
84}
85
86/*
87 * Check if this vmemmap page is already initialised. If any section
88 * which overlaps this vmemmap page is initialised then this page is
89 * initialised already.
90 */
91static int __meminit vmemmap_populated(unsigned long start, int page_size)
92{
93 unsigned long end = start + page_size;
94 start = (unsigned long)(pfn_to_page(vmemmap_section_start(start)));
95
96 for (; start < end; start += (PAGES_PER_SECTION * sizeof(struct page)))
97 if (pfn_valid(page_to_pfn((struct page *)start)))
98 return 1;
99
100 return 0;
101}
102
103/*
104 * vmemmap virtual address space management does not have a traditonal page
105 * table to track which virtual struct pages are backed by physical mapping.
106 * The virtual to physical mappings are tracked in a simple linked list
107 * format. 'vmemmap_list' maintains the entire vmemmap physical mapping at
108 * all times where as the 'next' list maintains the available
109 * vmemmap_backing structures which have been deleted from the
110 * 'vmemmap_global' list during system runtime (memory hotplug remove
111 * operation). The freed 'vmemmap_backing' structures are reused later when
112 * new requests come in without allocating fresh memory. This pointer also
113 * tracks the allocated 'vmemmap_backing' structures as we allocate one
114 * full page memory at a time when we dont have any.
115 */
116struct vmemmap_backing *vmemmap_list;
117static struct vmemmap_backing *next;
118
119/*
120 * The same pointer 'next' tracks individual chunks inside the allocated
121 * full page during the boot time and again tracks the freeed nodes during
122 * runtime. It is racy but it does not happen as they are separated by the
123 * boot process. Will create problem if some how we have memory hotplug
124 * operation during boot !!
125 */
126static int num_left;
127static int num_freed;
128
129static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node)
130{
131 struct vmemmap_backing *vmem_back;
132 /* get from freed entries first */
133 if (num_freed) {
134 num_freed--;
135 vmem_back = next;
136 next = next->list;
137
138 return vmem_back;
139 }
140
141 /* allocate a page when required and hand out chunks */
142 if (!num_left) {
143 next = vmemmap_alloc_block(PAGE_SIZE, node);
144 if (unlikely(!next)) {
145 WARN_ON(1);
146 return NULL;
147 }
148 num_left = PAGE_SIZE / sizeof(struct vmemmap_backing);
149 }
150
151 num_left--;
152
153 return next++;
154}
155
156static __meminit void vmemmap_list_populate(unsigned long phys,
157 unsigned long start,
158 int node)
159{
160 struct vmemmap_backing *vmem_back;
161
162 vmem_back = vmemmap_list_alloc(node);
163 if (unlikely(!vmem_back)) {
164 WARN_ON(1);
165 return;
166 }
167
168 vmem_back->phys = phys;
169 vmem_back->virt_addr = start;
170 vmem_back->list = vmemmap_list;
171
172 vmemmap_list = vmem_back;
173}
174
175static bool altmap_cross_boundary(struct vmem_altmap *altmap, unsigned long start,
176 unsigned long page_size)
177{
178 unsigned long nr_pfn = page_size / sizeof(struct page);
179 unsigned long start_pfn = page_to_pfn((struct page *)start);
180
181 if ((start_pfn + nr_pfn) > altmap->end_pfn)
182 return true;
183
184 if (start_pfn < altmap->base_pfn)
185 return true;
186
187 return false;
188}
189
190int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
191 struct vmem_altmap *altmap)
192{
193 unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
194
195 /* Align to the page size of the linear mapping. */
196 start = _ALIGN_DOWN(start, page_size);
197
198 pr_debug("vmemmap_populate %lx..%lx, node %d\n", start, end, node);
199
200 for (; start < end; start += page_size) {
201 void *p = NULL;
202 int rc;
203
204 if (vmemmap_populated(start, page_size))
205 continue;
206
207 /*
208 * Allocate from the altmap first if we have one. This may
209 * fail due to alignment issues when using 16MB hugepages, so
210 * fall back to system memory if the altmap allocation fail.
211 */
212 if (altmap && !altmap_cross_boundary(altmap, start, page_size)) {
213 p = altmap_alloc_block_buf(page_size, altmap);
214 if (!p)
215 pr_debug("altmap block allocation failed, falling back to system memory");
216 }
217 if (!p)
218 p = vmemmap_alloc_block_buf(page_size, node);
219 if (!p)
220 return -ENOMEM;
221
222 vmemmap_list_populate(__pa(p), start, node);
223
224 pr_debug(" * %016lx..%016lx allocated at %p\n",
225 start, start + page_size, p);
226
227 rc = vmemmap_create_mapping(start, page_size, __pa(p));
228 if (rc < 0) {
229 pr_warn("%s: Unable to create vmemmap mapping: %d\n",
230 __func__, rc);
231 return -EFAULT;
232 }
233 }
234
235 return 0;
236}
237
238#ifdef CONFIG_MEMORY_HOTPLUG
239static unsigned long vmemmap_list_free(unsigned long start)
240{
241 struct vmemmap_backing *vmem_back, *vmem_back_prev;
242
243 vmem_back_prev = vmem_back = vmemmap_list;
244
245 /* look for it with prev pointer recorded */
246 for (; vmem_back; vmem_back = vmem_back->list) {
247 if (vmem_back->virt_addr == start)
248 break;
249 vmem_back_prev = vmem_back;
250 }
251
252 if (unlikely(!vmem_back)) {
253 WARN_ON(1);
254 return 0;
255 }
256
257 /* remove it from vmemmap_list */
258 if (vmem_back == vmemmap_list) /* remove head */
259 vmemmap_list = vmem_back->list;
260 else
261 vmem_back_prev->list = vmem_back->list;
262
263 /* next point to this freed entry */
264 vmem_back->list = next;
265 next = vmem_back;
266 num_freed++;
267
268 return vmem_back->phys;
269}
270
271void __ref vmemmap_free(unsigned long start, unsigned long end,
272 struct vmem_altmap *altmap)
273{
274 unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift;
275 unsigned long page_order = get_order(page_size);
276 unsigned long alt_start = ~0, alt_end = ~0;
277 unsigned long base_pfn;
278
279 start = _ALIGN_DOWN(start, page_size);
280 if (altmap) {
281 alt_start = altmap->base_pfn;
282 alt_end = altmap->base_pfn + altmap->reserve +
283 altmap->free + altmap->alloc + altmap->align;
284 }
285
286 pr_debug("vmemmap_free %lx...%lx\n", start, end);
287
288 for (; start < end; start += page_size) {
289 unsigned long nr_pages, addr;
290 struct page *page;
291
292 /*
293 * the section has already be marked as invalid, so
294 * vmemmap_populated() true means some other sections still
295 * in this page, so skip it.
296 */
297 if (vmemmap_populated(start, page_size))
298 continue;
299
300 addr = vmemmap_list_free(start);
301 if (!addr)
302 continue;
303
304 page = pfn_to_page(addr >> PAGE_SHIFT);
305 nr_pages = 1 << page_order;
306 base_pfn = PHYS_PFN(addr);
307
308 if (base_pfn >= alt_start && base_pfn < alt_end) {
309 vmem_altmap_free(altmap, nr_pages);
310 } else if (PageReserved(page)) {
311 /* allocated from bootmem */
312 if (page_size < PAGE_SIZE) {
313 /*
314 * this shouldn't happen, but if it is
315 * the case, leave the memory there
316 */
317 WARN_ON_ONCE(1);
318 } else {
319 while (nr_pages--)
320 free_reserved_page(page++);
321 }
322 } else {
323 free_pages((unsigned long)(__va(addr)), page_order);
324 }
325
326 vmemmap_remove_mapping(start, page_size);
327 }
328}
329#endif
330void register_page_bootmem_memmap(unsigned long section_nr,
331 struct page *start_page, unsigned long size)
332{
333}
334
335#endif /* CONFIG_SPARSEMEM_VMEMMAP */
336
337#ifdef CONFIG_PPC_BOOK3S_64
338static bool disable_radix = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
339
340static int __init parse_disable_radix(char *p)
341{
342 bool val;
343
344 if (!p)
345 val = true;
346 else if (kstrtobool(p, &val))
347 return -EINVAL;
348
349 disable_radix = val;
350
351 return 0;
352}
353early_param("disable_radix", parse_disable_radix);
354
355/*
356 * If we're running under a hypervisor, we need to check the contents of
357 * /chosen/ibm,architecture-vec-5 to see if the hypervisor is willing to do
358 * radix. If not, we clear the radix feature bit so we fall back to hash.
359 */
360static void __init early_check_vec5(void)
361{
362 unsigned long root, chosen;
363 int size;
364 const u8 *vec5;
365 u8 mmu_supported;
366
367 root = of_get_flat_dt_root();
368 chosen = of_get_flat_dt_subnode_by_name(root, "chosen");
369 if (chosen == -FDT_ERR_NOTFOUND) {
370 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX;
371 return;
372 }
373 vec5 = of_get_flat_dt_prop(chosen, "ibm,architecture-vec-5", &size);
374 if (!vec5) {
375 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX;
376 return;
377 }
378 if (size <= OV5_INDX(OV5_MMU_SUPPORT)) {
379 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX;
380 return;
381 }
382
383 /* Check for supported configuration */
384 mmu_supported = vec5[OV5_INDX(OV5_MMU_SUPPORT)] &
385 OV5_FEAT(OV5_MMU_SUPPORT);
386 if (mmu_supported == OV5_FEAT(OV5_MMU_RADIX)) {
387 /* Hypervisor only supports radix - check enabled && GTSE */
388 if (!early_radix_enabled()) {
389 pr_warn("WARNING: Ignoring cmdline option disable_radix\n");
390 }
391 if (!(vec5[OV5_INDX(OV5_RADIX_GTSE)] &
392 OV5_FEAT(OV5_RADIX_GTSE))) {
393 pr_warn("WARNING: Hypervisor doesn't support RADIX with GTSE\n");
394 }
395 /* Do radix anyway - the hypervisor said we had to */
396 cur_cpu_spec->mmu_features |= MMU_FTR_TYPE_RADIX;
397 } else if (mmu_supported == OV5_FEAT(OV5_MMU_HASH)) {
398 /* Hypervisor only supports hash - disable radix */
399 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX;
400 }
401}
402
403void __init mmu_early_init_devtree(void)
404{
405 /* Disable radix mode based on kernel command line. */
406 if (disable_radix)
407 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX;
408
409 /*
410 * Check /chosen/ibm,architecture-vec-5 if running as a guest.
411 * When running bare-metal, we can use radix if we like
412 * even though the ibm,architecture-vec-5 property created by
413 * skiboot doesn't have the necessary bits set.
414 */
415 if (!(mfmsr() & MSR_HV))
416 early_check_vec5();
417
418 if (early_radix_enabled())
419 radix__early_init_devtree();
420 else
421 hash__early_init_devtree();
422}
423#endif /* CONFIG_PPC_BOOK3S_64 */