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 * PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
10 *
11 * Derived from "arch/i386/mm/init.c"
12 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
13 */
14
15 #include <linux/export.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/gfp.h>
21 #include <linux/types.h>
22 #include <linux/mm.h>
23 #include <linux/stddef.h>
24 #include <linux/init.h>
25 #include <linux/memblock.h>
26 #include <linux/highmem.h>
27 #include <linux/initrd.h>
28 #include <linux/pagemap.h>
29 #include <linux/suspend.h>
30 #include <linux/hugetlb.h>
31 #include <linux/slab.h>
32 #include <linux/vmalloc.h>
33 #include <linux/memremap.h>
34
35 #include <asm/pgalloc.h>
36 #include <asm/prom.h>
37 #include <asm/io.h>
38 #include <asm/mmu_context.h>
39 #include <asm/pgtable.h>
40 #include <asm/mmu.h>
41 #include <asm/smp.h>
42 #include <asm/machdep.h>
43 #include <asm/btext.h>
44 #include <asm/tlb.h>
45 #include <asm/sections.h>
46 #include <asm/sparsemem.h>
47 #include <asm/vdso.h>
48 #include <asm/fixmap.h>
49 #include <asm/swiotlb.h>
50 #include <asm/rtas.h>
51
52 #include <mm/mmu_decl.h>
53
54 #ifndef CPU_FTR_COHERENT_ICACHE
55 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
56 #define CPU_FTR_NOEXECUTE 0
57 #endif
58
59 unsigned long long memory_limit;
60 bool init_mem_is_free;
61
62 #ifdef CONFIG_HIGHMEM
63 pte_t *kmap_pte;
64 EXPORT_SYMBOL(kmap_pte);
65 pgprot_t kmap_prot;
66 EXPORT_SYMBOL(kmap_prot);
67
virt_to_kpte(unsigned long vaddr)68 static inline pte_t *virt_to_kpte(unsigned long vaddr)
69 {
70 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
71 vaddr), vaddr), vaddr);
72 }
73 #endif
74
phys_mem_access_prot(struct file * file,unsigned long pfn,unsigned long size,pgprot_t vma_prot)75 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
76 unsigned long size, pgprot_t vma_prot)
77 {
78 if (ppc_md.phys_mem_access_prot)
79 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
80
81 if (!page_is_ram(pfn))
82 vma_prot = pgprot_noncached(vma_prot);
83
84 return vma_prot;
85 }
86 EXPORT_SYMBOL(phys_mem_access_prot);
87
88 #ifdef CONFIG_MEMORY_HOTPLUG
89
90 #ifdef CONFIG_NUMA
memory_add_physaddr_to_nid(u64 start)91 int memory_add_physaddr_to_nid(u64 start)
92 {
93 return hot_add_scn_to_nid(start);
94 }
95 #endif
96
create_section_mapping(unsigned long start,unsigned long end,int nid)97 int __weak create_section_mapping(unsigned long start, unsigned long end, int nid)
98 {
99 return -ENODEV;
100 }
101
remove_section_mapping(unsigned long start,unsigned long end)102 int __weak remove_section_mapping(unsigned long start, unsigned long end)
103 {
104 return -ENODEV;
105 }
106
107 #define FLUSH_CHUNK_SIZE SZ_1G
108 /**
109 * flush_dcache_range_chunked(): Write any modified data cache blocks out to
110 * memory and invalidate them, in chunks of up to FLUSH_CHUNK_SIZE
111 * Does not invalidate the corresponding instruction cache blocks.
112 *
113 * @start: the start address
114 * @stop: the stop address (exclusive)
115 * @chunk: the max size of the chunks
116 */
flush_dcache_range_chunked(unsigned long start,unsigned long stop,unsigned long chunk)117 static void flush_dcache_range_chunked(unsigned long start, unsigned long stop,
118 unsigned long chunk)
119 {
120 unsigned long i;
121
122 for (i = start; i < stop; i += chunk) {
123 flush_dcache_range(i, min(stop, i + chunk));
124 cond_resched();
125 }
126 }
127
arch_add_memory(int nid,u64 start,u64 size,struct mhp_restrictions * restrictions)128 int __ref arch_add_memory(int nid, u64 start, u64 size,
129 struct mhp_restrictions *restrictions)
130 {
131 unsigned long start_pfn = start >> PAGE_SHIFT;
132 unsigned long nr_pages = size >> PAGE_SHIFT;
133 int rc;
134
135 resize_hpt_for_hotplug(memblock_phys_mem_size());
136
137 start = (unsigned long)__va(start);
138 rc = create_section_mapping(start, start + size, nid);
139 if (rc) {
140 pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
141 start, start + size, rc);
142 return -EFAULT;
143 }
144
145 flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
146
147 return __add_pages(nid, start_pfn, nr_pages, restrictions);
148 }
149
arch_remove_memory(int nid,u64 start,u64 size,struct vmem_altmap * altmap)150 void __ref arch_remove_memory(int nid, u64 start, u64 size,
151 struct vmem_altmap *altmap)
152 {
153 unsigned long start_pfn = start >> PAGE_SHIFT;
154 unsigned long nr_pages = size >> PAGE_SHIFT;
155 int ret;
156
157 __remove_pages(start_pfn, nr_pages, altmap);
158
159 /* Remove htab bolted mappings for this section of memory */
160 start = (unsigned long)__va(start);
161 flush_dcache_range_chunked(start, start + size, FLUSH_CHUNK_SIZE);
162
163 ret = remove_section_mapping(start, start + size);
164 WARN_ON_ONCE(ret);
165
166 /* Ensure all vmalloc mappings are flushed in case they also
167 * hit that section of memory
168 */
169 vm_unmap_aliases();
170
171 if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
172 pr_warn("Hash collision while resizing HPT\n");
173 }
174 #endif
175
176 #ifndef CONFIG_NEED_MULTIPLE_NODES
mem_topology_setup(void)177 void __init mem_topology_setup(void)
178 {
179 max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
180 min_low_pfn = MEMORY_START >> PAGE_SHIFT;
181 #ifdef CONFIG_HIGHMEM
182 max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
183 #endif
184
185 /* Place all memblock_regions in the same node and merge contiguous
186 * memblock_regions
187 */
188 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
189 }
190
initmem_init(void)191 void __init initmem_init(void)
192 {
193 /* XXX need to clip this if using highmem? */
194 sparse_memory_present_with_active_regions(0);
195 sparse_init();
196 }
197
198 /* mark pages that don't exist as nosave */
mark_nonram_nosave(void)199 static int __init mark_nonram_nosave(void)
200 {
201 struct memblock_region *reg, *prev = NULL;
202
203 for_each_memblock(memory, reg) {
204 if (prev &&
205 memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
206 register_nosave_region(memblock_region_memory_end_pfn(prev),
207 memblock_region_memory_base_pfn(reg));
208 prev = reg;
209 }
210 return 0;
211 }
212 #else /* CONFIG_NEED_MULTIPLE_NODES */
mark_nonram_nosave(void)213 static int __init mark_nonram_nosave(void)
214 {
215 return 0;
216 }
217 #endif
218
219 /*
220 * Zones usage:
221 *
222 * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
223 * everything else. GFP_DMA32 page allocations automatically fall back to
224 * ZONE_DMA.
225 *
226 * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
227 * inform the generic DMA mapping code. 32-bit only devices (if not handled
228 * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
229 * otherwise served by ZONE_DMA.
230 */
231 static unsigned long max_zone_pfns[MAX_NR_ZONES];
232
233 /*
234 * paging_init() sets up the page tables - in fact we've already done this.
235 */
paging_init(void)236 void __init paging_init(void)
237 {
238 unsigned long long total_ram = memblock_phys_mem_size();
239 phys_addr_t top_of_ram = memblock_end_of_DRAM();
240
241 #ifdef CONFIG_PPC32
242 unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
243 unsigned long end = __fix_to_virt(FIX_HOLE);
244
245 for (; v < end; v += PAGE_SIZE)
246 map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
247 #endif
248
249 #ifdef CONFIG_HIGHMEM
250 map_kernel_page(PKMAP_BASE, 0, __pgprot(0)); /* XXX gross */
251 pkmap_page_table = virt_to_kpte(PKMAP_BASE);
252
253 kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
254 kmap_prot = PAGE_KERNEL;
255 #endif /* CONFIG_HIGHMEM */
256
257 printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
258 (unsigned long long)top_of_ram, total_ram);
259 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
260 (long int)((top_of_ram - total_ram) >> 20));
261
262 #ifdef CONFIG_ZONE_DMA
263 max_zone_pfns[ZONE_DMA] = min(max_low_pfn,
264 1UL << (ARCH_ZONE_DMA_BITS - PAGE_SHIFT));
265 #endif
266 max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
267 #ifdef CONFIG_HIGHMEM
268 max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
269 #endif
270
271 free_area_init_nodes(max_zone_pfns);
272
273 mark_nonram_nosave();
274 }
275
mem_init(void)276 void __init mem_init(void)
277 {
278 /*
279 * book3s is limited to 16 page sizes due to encoding this in
280 * a 4-bit field for slices.
281 */
282 BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
283
284 #ifdef CONFIG_SWIOTLB
285 /*
286 * Some platforms (e.g. 85xx) limit DMA-able memory way below
287 * 4G. We force memblock to bottom-up mode to ensure that the
288 * memory allocated in swiotlb_init() is DMA-able.
289 * As it's the last memblock allocation, no need to reset it
290 * back to to-down.
291 */
292 memblock_set_bottom_up(true);
293 swiotlb_init(0);
294 #endif
295
296 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
297 set_max_mapnr(max_pfn);
298 memblock_free_all();
299
300 #ifdef CONFIG_HIGHMEM
301 {
302 unsigned long pfn, highmem_mapnr;
303
304 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
305 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
306 phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
307 struct page *page = pfn_to_page(pfn);
308 if (!memblock_is_reserved(paddr))
309 free_highmem_page(page);
310 }
311 }
312 #endif /* CONFIG_HIGHMEM */
313
314 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
315 /*
316 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
317 * functions.... do it here for the non-smp case.
318 */
319 per_cpu(next_tlbcam_idx, smp_processor_id()) =
320 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
321 #endif
322
323 mem_init_print_info(NULL);
324 #ifdef CONFIG_PPC32
325 pr_info("Kernel virtual memory layout:\n");
326 #ifdef CONFIG_KASAN
327 pr_info(" * 0x%08lx..0x%08lx : kasan shadow mem\n",
328 KASAN_SHADOW_START, KASAN_SHADOW_END);
329 #endif
330 pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START, FIXADDR_TOP);
331 #ifdef CONFIG_HIGHMEM
332 pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n",
333 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
334 #endif /* CONFIG_HIGHMEM */
335 if (ioremap_bot != IOREMAP_TOP)
336 pr_info(" * 0x%08lx..0x%08lx : early ioremap\n",
337 ioremap_bot, IOREMAP_TOP);
338 pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n",
339 VMALLOC_START, VMALLOC_END);
340 #endif /* CONFIG_PPC32 */
341 }
342
free_initmem(void)343 void free_initmem(void)
344 {
345 ppc_md.progress = ppc_printk_progress;
346 mark_initmem_nx();
347 init_mem_is_free = true;
348 free_initmem_default(POISON_FREE_INITMEM);
349 }
350
351 /*
352 * This is called when a page has been modified by the kernel.
353 * It just marks the page as not i-cache clean. We do the i-cache
354 * flush later when the page is given to a user process, if necessary.
355 */
flush_dcache_page(struct page * page)356 void flush_dcache_page(struct page *page)
357 {
358 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
359 return;
360 /* avoid an atomic op if possible */
361 if (test_bit(PG_arch_1, &page->flags))
362 clear_bit(PG_arch_1, &page->flags);
363 }
364 EXPORT_SYMBOL(flush_dcache_page);
365
flush_dcache_icache_page(struct page * page)366 void flush_dcache_icache_page(struct page *page)
367 {
368 #ifdef CONFIG_HUGETLB_PAGE
369 if (PageCompound(page)) {
370 flush_dcache_icache_hugepage(page);
371 return;
372 }
373 #endif
374 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
375 /* On 8xx there is no need to kmap since highmem is not supported */
376 __flush_dcache_icache(page_address(page));
377 #else
378 if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
379 void *start = kmap_atomic(page);
380 __flush_dcache_icache(start);
381 kunmap_atomic(start);
382 } else {
383 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
384 }
385 #endif
386 }
387 EXPORT_SYMBOL(flush_dcache_icache_page);
388
clear_user_page(void * page,unsigned long vaddr,struct page * pg)389 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
390 {
391 clear_page(page);
392
393 /*
394 * We shouldn't have to do this, but some versions of glibc
395 * require it (ld.so assumes zero filled pages are icache clean)
396 * - Anton
397 */
398 flush_dcache_page(pg);
399 }
400 EXPORT_SYMBOL(clear_user_page);
401
copy_user_page(void * vto,void * vfrom,unsigned long vaddr,struct page * pg)402 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
403 struct page *pg)
404 {
405 copy_page(vto, vfrom);
406
407 /*
408 * We should be able to use the following optimisation, however
409 * there are two problems.
410 * Firstly a bug in some versions of binutils meant PLT sections
411 * were not marked executable.
412 * Secondly the first word in the GOT section is blrl, used
413 * to establish the GOT address. Until recently the GOT was
414 * not marked executable.
415 * - Anton
416 */
417 #if 0
418 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
419 return;
420 #endif
421
422 flush_dcache_page(pg);
423 }
424
flush_icache_user_range(struct vm_area_struct * vma,struct page * page,unsigned long addr,int len)425 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
426 unsigned long addr, int len)
427 {
428 unsigned long maddr;
429
430 maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
431 flush_icache_range(maddr, maddr + len);
432 kunmap(page);
433 }
434 EXPORT_SYMBOL(flush_icache_user_range);
435
436 /*
437 * System memory should not be in /proc/iomem but various tools expect it
438 * (eg kdump).
439 */
add_system_ram_resources(void)440 static int __init add_system_ram_resources(void)
441 {
442 struct memblock_region *reg;
443
444 for_each_memblock(memory, reg) {
445 struct resource *res;
446 unsigned long base = reg->base;
447 unsigned long size = reg->size;
448
449 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
450 WARN_ON(!res);
451
452 if (res) {
453 res->name = "System RAM";
454 res->start = base;
455 res->end = base + size - 1;
456 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
457 WARN_ON(request_resource(&iomem_resource, res) < 0);
458 }
459 }
460
461 return 0;
462 }
463 subsys_initcall(add_system_ram_resources);
464
465 #ifdef CONFIG_STRICT_DEVMEM
466 /*
467 * devmem_is_allowed(): check to see if /dev/mem access to a certain address
468 * is valid. The argument is a physical page number.
469 *
470 * Access has to be given to non-kernel-ram areas as well, these contain the
471 * PCI mmio resources as well as potential bios/acpi data regions.
472 */
devmem_is_allowed(unsigned long pfn)473 int devmem_is_allowed(unsigned long pfn)
474 {
475 if (page_is_rtas_user_buf(pfn))
476 return 1;
477 if (iomem_is_exclusive(PFN_PHYS(pfn)))
478 return 0;
479 if (!page_is_ram(pfn))
480 return 1;
481 return 0;
482 }
483 #endif /* CONFIG_STRICT_DEVMEM */
484
485 /*
486 * This is defined in kernel/resource.c but only powerpc needs to export it, for
487 * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
488 */
489 EXPORT_SYMBOL_GPL(walk_system_ram_range);
490