xref: /linux/arch/powerpc/mm/mem.c (revision 6215d9f4470fbb48245ffdfade821685e2728c65)
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 {
163 	unsigned long start_pfn = start >> PAGE_SHIFT;
164 	unsigned long nr_pages = size >> PAGE_SHIFT;
165 
166 	__remove_pages(start_pfn, nr_pages, altmap);
167 	arch_remove_linear_mapping(start, size);
168 }
169 #endif
170 
171 #ifndef CONFIG_NUMA
172 void __init mem_topology_setup(void)
173 {
174 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
175 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
176 #ifdef CONFIG_HIGHMEM
177 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
178 #endif
179 
180 	/* Place all memblock_regions in the same node and merge contiguous
181 	 * memblock_regions
182 	 */
183 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
184 }
185 
186 /* mark pages that don't exist as nosave */
187 static int __init mark_nonram_nosave(void)
188 {
189 	unsigned long spfn, epfn, prev = 0;
190 	int i;
191 
192 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
193 		if (prev && prev < spfn)
194 			register_nosave_region(prev, spfn);
195 
196 		prev = epfn;
197 	}
198 
199 	return 0;
200 }
201 #else /* CONFIG_NUMA */
202 static int __init mark_nonram_nosave(void)
203 {
204 	return 0;
205 }
206 #endif
207 
208 /*
209  * Zones usage:
210  *
211  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
212  * everything else. GFP_DMA32 page allocations automatically fall back to
213  * ZONE_DMA.
214  *
215  * By using 31-bit unconditionally, we can exploit zone_dma_limit to inform the
216  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
217  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
218  * ZONE_DMA.
219  */
220 void __init arch_zone_limits_init(unsigned long *max_zone_pfns)
221 {
222 #ifdef CONFIG_ZONE_DMA
223 	max_zone_pfns[ZONE_DMA] = min((zone_dma_limit >> PAGE_SHIFT) + 1, max_low_pfn);
224 #endif
225 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
226 #ifdef CONFIG_HIGHMEM
227 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
228 #endif
229 }
230 
231 /*
232  * paging_init() sets up the page tables - in fact we've already done this.
233  */
234 void __init paging_init(void)
235 {
236 	unsigned long long total_ram = memblock_phys_mem_size();
237 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
238 	int zone_dma_bits;
239 
240 #ifdef CONFIG_HIGHMEM
241 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
242 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
243 
244 	for (; v < end; v += PAGE_SIZE)
245 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
246 
247 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
248 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
249 #endif /* CONFIG_HIGHMEM */
250 
251 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
252 	       (unsigned long long)top_of_ram, total_ram);
253 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
254 	       (long int)((top_of_ram - total_ram) >> 20));
255 
256 	/*
257 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
258 	 * powerbooks.
259 	 */
260 	if (IS_ENABLED(CONFIG_PPC32))
261 		zone_dma_bits = 30;
262 	else
263 		zone_dma_bits = 31;
264 
265 	zone_dma_limit = DMA_BIT_MASK(zone_dma_bits);
266 
267 	mark_nonram_nosave();
268 }
269 
270 void __init arch_mm_preinit(void)
271 {
272 
273 	/*
274 	 * Reserve large chunks of memory for use by CMA for kdump, fadump, KVM
275 	 * and hugetlb. These must be called after pageblock_order is
276 	 * initialised.
277 	 */
278 	fadump_cma_init();
279 	kdump_cma_reserve();
280 	kvm_cma_reserve();
281 
282 	/*
283 	 * book3s is limited to 16 page sizes due to encoding this in
284 	 * a 4-bit field for slices.
285 	 */
286 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
287 
288 #ifdef CONFIG_SWIOTLB
289 	/*
290 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
291 	 * 4G. We force memblock to bottom-up mode to ensure that the
292 	 * memory allocated in swiotlb_init() is DMA-able.
293 	 * As it's the last memblock allocation, no need to reset it
294 	 * back to to-down.
295 	 */
296 	memblock_set_bottom_up(true);
297 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
298 #endif
299 
300 	kasan_late_init();
301 
302 #if defined(CONFIG_PPC_E500) && !defined(CONFIG_SMP)
303 	/*
304 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
305 	 * functions.... do it here for the non-smp case.
306 	 */
307 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
308 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
309 #endif
310 }
311 
312 void free_initmem(void)
313 {
314 	ppc_md.progress = ppc_printk_progress;
315 	mark_initmem_nx();
316 	free_initmem_default(POISON_FREE_INITMEM);
317 	ftrace_free_init_tramp();
318 }
319 
320 /*
321  * System memory should not be in /proc/iomem but various tools expect it
322  * (eg kdump).
323  */
324 static int __init add_system_ram_resources(void)
325 {
326 	phys_addr_t start, end;
327 	u64 i;
328 
329 	for_each_mem_range(i, &start, &end) {
330 		struct resource *res;
331 
332 		res = kzalloc_obj(struct resource);
333 		WARN_ON(!res);
334 
335 		if (res) {
336 			res->name = "System RAM";
337 			res->start = start;
338 			/*
339 			 * In memblock, end points to the first byte after
340 			 * the range while in resourses, end points to the
341 			 * last byte in the range.
342 			 */
343 			res->end = end - 1;
344 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
345 			WARN_ON(insert_resource(&iomem_resource, res) < 0);
346 		}
347 	}
348 
349 	return 0;
350 }
351 subsys_initcall(add_system_ram_resources);
352 
353 #ifdef CONFIG_STRICT_DEVMEM
354 /*
355  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
356  * is valid. The argument is a physical page number.
357  *
358  * Access has to be given to non-kernel-ram areas as well, these contain the
359  * PCI mmio resources as well as potential bios/acpi data regions.
360  */
361 int devmem_is_allowed(unsigned long pfn)
362 {
363 	if (page_is_rtas_user_buf(pfn))
364 		return 1;
365 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
366 		return 0;
367 	if (!page_is_ram(pfn))
368 		return 1;
369 	return 0;
370 }
371 #endif /* CONFIG_STRICT_DEVMEM */
372 
373 /*
374  * This is defined in kernel/resource.c but only powerpc needs to export it, for
375  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
376  */
377 EXPORT_SYMBOL_GPL(walk_system_ram_range);
378 
379 #ifdef CONFIG_EXECMEM
380 static struct execmem_info execmem_info __ro_after_init;
381 
382 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC_BOOK3S_603)
383 static void prealloc_execmem_pgtable(void)
384 {
385 	unsigned long va;
386 
387 	for (va = ALIGN_DOWN(MODULES_VADDR, PGDIR_SIZE); va < MODULES_END; va += PGDIR_SIZE)
388 		pte_alloc_kernel(pmd_off_k(va), va);
389 }
390 #else
391 static void prealloc_execmem_pgtable(void) { }
392 #endif
393 
394 struct execmem_info __init *execmem_arch_setup(void)
395 {
396 	pgprot_t kprobes_prot = strict_module_rwx_enabled() ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
397 	pgprot_t prot = strict_module_rwx_enabled() ? PAGE_KERNEL : PAGE_KERNEL_EXEC;
398 	unsigned long fallback_start = 0, fallback_end = 0;
399 	unsigned long start, end;
400 
401 	/*
402 	 * BOOK3S_32 and 8xx define MODULES_VADDR for text allocations and
403 	 * allow allocating data in the entire vmalloc space
404 	 */
405 #ifdef MODULES_VADDR
406 	unsigned long limit = (unsigned long)_etext - SZ_32M;
407 
408 	/* First try within 32M limit from _etext to avoid branch trampolines */
409 	if (MODULES_VADDR < PAGE_OFFSET && MODULES_END > limit) {
410 		start = limit;
411 		fallback_start = MODULES_VADDR;
412 		fallback_end = MODULES_END;
413 	} else {
414 		start = MODULES_VADDR;
415 	}
416 
417 	end = MODULES_END;
418 #else
419 	start = VMALLOC_START;
420 	end = VMALLOC_END;
421 #endif
422 
423 	prealloc_execmem_pgtable();
424 
425 	execmem_info = (struct execmem_info){
426 		.ranges = {
427 			[EXECMEM_DEFAULT] = {
428 				.start	= start,
429 				.end	= end,
430 				.pgprot	= prot,
431 				.alignment = 1,
432 				.fallback_start	= fallback_start,
433 				.fallback_end	= fallback_end,
434 			},
435 			[EXECMEM_KPROBES] = {
436 				.start	= VMALLOC_START,
437 				.end	= VMALLOC_END,
438 				.pgprot	= kprobes_prot,
439 				.alignment = 1,
440 			},
441 			[EXECMEM_MODULE_DATA] = {
442 				.start	= VMALLOC_START,
443 				.end	= VMALLOC_END,
444 				.pgprot	= PAGE_KERNEL,
445 				.alignment = 1,
446 			},
447 		},
448 	};
449 
450 	return &execmem_info;
451 }
452 #endif /* CONFIG_EXECMEM */
453