xref: /linux/arch/powerpc/mm/mem.c (revision bba2c3615bd6cfee7456d1130f2e6b01b3f4e9ba)
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/memblock.h>
16 #include <linux/highmem.h>
17 #include <linux/suspend.h>
18 #include <linux/dma-direct.h>
19 #include <linux/execmem.h>
20 #include <linux/vmalloc.h>
21 
22 #include <asm/swiotlb.h>
23 #include <asm/machdep.h>
24 #include <asm/rtas.h>
25 #include <asm/kasan.h>
26 #include <asm/svm.h>
27 #include <asm/mmzone.h>
28 #include <asm/ftrace.h>
29 #include <asm/text-patching.h>
30 #include <asm/setup.h>
31 #include <asm/fixmap.h>
32 
33 #include <asm/fadump.h>
34 #include <asm/kexec.h>
35 #include <asm/kvm_ppc.h>
36 
37 #include <mm/mmu_decl.h>
38 
39 unsigned long long memory_limit __initdata;
40 
41 pgprot_t __phys_mem_access_prot(unsigned long pfn, unsigned long size,
42 				pgprot_t vma_prot)
43 {
44 	if (ppc_md.phys_mem_access_prot)
45 		return ppc_md.phys_mem_access_prot(pfn, size, vma_prot);
46 
47 	if (!page_is_ram(pfn))
48 		vma_prot = pgprot_noncached(vma_prot);
49 
50 	return vma_prot;
51 }
52 EXPORT_SYMBOL(__phys_mem_access_prot);
53 
54 #ifdef CONFIG_MEMORY_HOTPLUG
55 static DEFINE_MUTEX(linear_mapping_mutex);
56 
57 #ifdef CONFIG_NUMA
58 int memory_add_physaddr_to_nid(u64 start)
59 {
60 	return hot_add_scn_to_nid(start);
61 }
62 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
63 #endif
64 
65 int __weak create_section_mapping(unsigned long start, unsigned long end,
66 				  int nid, pgprot_t prot)
67 {
68 	return -ENODEV;
69 }
70 
71 int __weak remove_section_mapping(unsigned long start, unsigned long end)
72 {
73 	return -ENODEV;
74 }
75 
76 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
77 				     struct mhp_params *params)
78 {
79 	int rc;
80 
81 	start = (unsigned long)__va(start);
82 	mutex_lock(&linear_mapping_mutex);
83 	rc = create_section_mapping(start, start + size, nid,
84 				    params->pgprot);
85 	mutex_unlock(&linear_mapping_mutex);
86 	if (rc) {
87 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
88 			start, start + size, rc);
89 		return -EFAULT;
90 	}
91 	return 0;
92 }
93 
94 void __ref arch_remove_linear_mapping(u64 start, u64 size)
95 {
96 	int ret;
97 
98 	/* Remove htab bolted mappings for this section of memory */
99 	start = (unsigned long)__va(start);
100 
101 	mutex_lock(&linear_mapping_mutex);
102 	ret = remove_section_mapping(start, start + size);
103 	mutex_unlock(&linear_mapping_mutex);
104 	if (ret)
105 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
106 			start, start + size, ret);
107 
108 	/* Ensure all vmalloc mappings are flushed in case they also
109 	 * hit that section of memory
110 	 */
111 	vm_unmap_aliases();
112 }
113 
114 /*
115  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
116  * updating.
117  */
118 static void update_end_of_memory_vars(u64 start, u64 size)
119 {
120 	unsigned long end_pfn = PFN_UP(start + size);
121 
122 	if (end_pfn > max_pfn) {
123 		max_pfn = end_pfn;
124 		max_low_pfn = end_pfn;
125 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
126 	}
127 }
128 
129 int __ref add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
130 		    struct mhp_params *params)
131 {
132 	int ret;
133 
134 	ret = __add_pages(nid, start_pfn, nr_pages, params);
135 	if (ret)
136 		return ret;
137 
138 	/* update max_pfn, max_low_pfn and high_memory */
139 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
140 				  nr_pages << PAGE_SHIFT);
141 
142 	return ret;
143 }
144 
145 int __ref arch_add_memory(int nid, u64 start, u64 size,
146 			  struct mhp_params *params)
147 {
148 	unsigned long start_pfn = start >> PAGE_SHIFT;
149 	unsigned long nr_pages = size >> PAGE_SHIFT;
150 	int rc;
151 
152 	rc = arch_create_linear_mapping(nid, start, size, params);
153 	if (rc)
154 		return rc;
155 	rc = add_pages(nid, start_pfn, nr_pages, params);
156 	if (rc)
157 		arch_remove_linear_mapping(start, size);
158 	return rc;
159 }
160 
161 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap,
162 			      struct dev_pagemap *pgmap)
163 {
164 	unsigned long start_pfn = start >> PAGE_SHIFT;
165 	unsigned long nr_pages = size >> PAGE_SHIFT;
166 
167 	__remove_pages(start_pfn, nr_pages, altmap, pgmap);
168 	arch_remove_linear_mapping(start, size);
169 }
170 #endif
171 
172 #ifndef CONFIG_NUMA
173 void __init mem_topology_setup(void)
174 {
175 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
176 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
177 #ifdef CONFIG_HIGHMEM
178 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
179 #endif
180 
181 	/* Place all memblock_regions in the same node and merge contiguous
182 	 * memblock_regions
183 	 */
184 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
185 }
186 
187 /* mark pages that don't exist as nosave */
188 static int __init mark_nonram_nosave(void)
189 {
190 	unsigned long spfn, epfn, prev = 0;
191 	int i;
192 
193 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
194 		if (prev && prev < spfn)
195 			register_nosave_region(prev, spfn);
196 
197 		prev = epfn;
198 	}
199 
200 	return 0;
201 }
202 #else /* CONFIG_NUMA */
203 static int __init mark_nonram_nosave(void)
204 {
205 	return 0;
206 }
207 #endif
208 
209 /*
210  * Zones usage:
211  *
212  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
213  * everything else. GFP_DMA32 page allocations automatically fall back to
214  * ZONE_DMA.
215  *
216  * By using 31-bit unconditionally, we can exploit zone_dma_limit to inform the
217  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
218  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
219  * ZONE_DMA.
220  */
221 void __init arch_zone_limits_init(unsigned long *max_zone_pfns)
222 {
223 #ifdef CONFIG_ZONE_DMA
224 	max_zone_pfns[ZONE_DMA] = min((zone_dma_limit >> PAGE_SHIFT) + 1, max_low_pfn);
225 #endif
226 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
227 #ifdef CONFIG_HIGHMEM
228 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
229 #endif
230 }
231 
232 /*
233  * paging_init() sets up the page tables - in fact we've already done this.
234  */
235 void __init paging_init(void)
236 {
237 	unsigned long long total_ram = memblock_phys_mem_size();
238 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
239 	int zone_dma_bits;
240 
241 #ifdef CONFIG_HIGHMEM
242 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
243 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
244 
245 	for (; v < end; v += PAGE_SIZE)
246 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
247 
248 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
249 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
250 #endif /* CONFIG_HIGHMEM */
251 
252 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
253 	       (unsigned long long)top_of_ram, total_ram);
254 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
255 	       (long int)((top_of_ram - total_ram) >> 20));
256 
257 	/*
258 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
259 	 * powerbooks.
260 	 */
261 	if (IS_ENABLED(CONFIG_PPC32))
262 		zone_dma_bits = 30;
263 	else
264 		zone_dma_bits = 31;
265 
266 	zone_dma_limit = DMA_BIT_MASK(zone_dma_bits);
267 
268 	mark_nonram_nosave();
269 }
270 
271 void __init arch_mm_preinit(void)
272 {
273 
274 	/*
275 	 * Reserve large chunks of memory for use by CMA for kdump, fadump, KVM
276 	 * and hugetlb. These must be called after pageblock_order is
277 	 * initialised.
278 	 */
279 	fadump_cma_init();
280 	kdump_cma_reserve();
281 	kvm_cma_reserve();
282 
283 	/*
284 	 * book3s is limited to 16 page sizes due to encoding this in
285 	 * a 4-bit field for slices.
286 	 */
287 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
288 
289 #ifdef CONFIG_SWIOTLB
290 	/*
291 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
292 	 * 4G. We force memblock to bottom-up mode to ensure that the
293 	 * memory allocated in swiotlb_init() is DMA-able.
294 	 * As it's the last memblock allocation, no need to reset it
295 	 * back to to-down.
296 	 */
297 	memblock_set_bottom_up(true);
298 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
299 #endif
300 
301 	kasan_late_init();
302 
303 #if defined(CONFIG_PPC_E500) && !defined(CONFIG_SMP)
304 	/*
305 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
306 	 * functions.... do it here for the non-smp case.
307 	 */
308 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
309 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
310 #endif
311 }
312 
313 void free_initmem(void)
314 {
315 	ppc_md.progress = ppc_printk_progress;
316 	mark_initmem_nx();
317 	free_initmem_default(POISON_FREE_INITMEM);
318 	ftrace_free_init_tramp();
319 }
320 
321 /*
322  * System memory should not be in /proc/iomem but various tools expect it
323  * (eg kdump).
324  */
325 static int __init add_system_ram_resources(void)
326 {
327 	phys_addr_t start, end;
328 	u64 i;
329 
330 	for_each_mem_range(i, &start, &end) {
331 		struct resource *res;
332 
333 		res = kzalloc_obj(struct resource);
334 		WARN_ON(!res);
335 
336 		if (res) {
337 			res->name = "System RAM";
338 			res->start = start;
339 			/*
340 			 * In memblock, end points to the first byte after
341 			 * the range while in resourses, end points to the
342 			 * last byte in the range.
343 			 */
344 			res->end = end - 1;
345 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
346 			WARN_ON(insert_resource(&iomem_resource, res) < 0);
347 		}
348 	}
349 
350 	return 0;
351 }
352 subsys_initcall(add_system_ram_resources);
353 
354 #ifdef CONFIG_STRICT_DEVMEM
355 /*
356  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
357  * is valid. The argument is a physical page number.
358  *
359  * Access has to be given to non-kernel-ram areas as well, these contain the
360  * PCI mmio resources as well as potential bios/acpi data regions.
361  */
362 int devmem_is_allowed(unsigned long pfn)
363 {
364 	if (page_is_rtas_user_buf(pfn))
365 		return 1;
366 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
367 		return 0;
368 	if (!page_is_ram(pfn))
369 		return 1;
370 	return 0;
371 }
372 #endif /* CONFIG_STRICT_DEVMEM */
373 
374 /*
375  * This is defined in kernel/resource.c but only powerpc needs to export it, for
376  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
377  */
378 EXPORT_SYMBOL_GPL(walk_system_ram_range);
379 
380 #ifdef CONFIG_EXECMEM
381 static struct execmem_info execmem_info __ro_after_init;
382 
383 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC_BOOK3S_603)
384 static void prealloc_execmem_pgtable(void)
385 {
386 	unsigned long va;
387 
388 	for (va = ALIGN_DOWN(MODULES_VADDR, PGDIR_SIZE); va < MODULES_END; va += PGDIR_SIZE)
389 		pte_alloc_kernel(pmd_off_k(va), va);
390 }
391 #else
392 static void prealloc_execmem_pgtable(void) { }
393 #endif
394 
395 struct execmem_info __init *execmem_arch_setup(void)
396 {
397 	pgprot_t kprobes_prot = strict_module_rwx_enabled() ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
398 	pgprot_t prot = strict_module_rwx_enabled() ? PAGE_KERNEL : PAGE_KERNEL_EXEC;
399 	unsigned long fallback_start = 0, fallback_end = 0;
400 	unsigned long start, end;
401 
402 	/*
403 	 * BOOK3S_32 and 8xx define MODULES_VADDR for text allocations and
404 	 * allow allocating data in the entire vmalloc space
405 	 */
406 #ifdef MODULES_VADDR
407 	unsigned long limit = (unsigned long)_etext - SZ_32M;
408 
409 	/* First try within 32M limit from _etext to avoid branch trampolines */
410 	if (MODULES_VADDR < PAGE_OFFSET && MODULES_END > limit) {
411 		start = limit;
412 		fallback_start = MODULES_VADDR;
413 		fallback_end = MODULES_END;
414 	} else {
415 		start = MODULES_VADDR;
416 	}
417 
418 	end = MODULES_END;
419 #else
420 	start = VMALLOC_START;
421 	end = VMALLOC_END;
422 #endif
423 
424 	prealloc_execmem_pgtable();
425 
426 	execmem_info = (struct execmem_info){
427 		.ranges = {
428 			[EXECMEM_DEFAULT] = {
429 				.start	= start,
430 				.end	= end,
431 				.pgprot	= prot,
432 				.alignment = 1,
433 				.fallback_start	= fallback_start,
434 				.fallback_end	= fallback_end,
435 			},
436 			[EXECMEM_KPROBES] = {
437 				.start	= VMALLOC_START,
438 				.end	= VMALLOC_END,
439 				.pgprot	= kprobes_prot,
440 				.alignment = 1,
441 			},
442 			[EXECMEM_MODULE_DATA] = {
443 				.start	= VMALLOC_START,
444 				.end	= VMALLOC_END,
445 				.pgprot	= PAGE_KERNEL,
446 				.alignment = 1,
447 			},
448 		},
449 	};
450 
451 	return &execmem_info;
452 }
453 #endif /* CONFIG_EXECMEM */
454