xref: /linux/arch/powerpc/mm/mem.c (revision 45d8b572fac3aa8b49d53c946b3685eaf78a2824)
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 
20 #include <asm/swiotlb.h>
21 #include <asm/machdep.h>
22 #include <asm/rtas.h>
23 #include <asm/kasan.h>
24 #include <asm/svm.h>
25 #include <asm/mmzone.h>
26 #include <asm/ftrace.h>
27 #include <asm/code-patching.h>
28 #include <asm/setup.h>
29 #include <asm/fixmap.h>
30 
31 #include <mm/mmu_decl.h>
32 
33 unsigned long long memory_limit;
34 
35 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
36 EXPORT_SYMBOL(empty_zero_page);
37 
38 pgprot_t __phys_mem_access_prot(unsigned long pfn, unsigned long size,
39 				pgprot_t vma_prot)
40 {
41 	if (ppc_md.phys_mem_access_prot)
42 		return ppc_md.phys_mem_access_prot(pfn, size, vma_prot);
43 
44 	if (!page_is_ram(pfn))
45 		vma_prot = pgprot_noncached(vma_prot);
46 
47 	return vma_prot;
48 }
49 EXPORT_SYMBOL(__phys_mem_access_prot);
50 
51 #ifdef CONFIG_MEMORY_HOTPLUG
52 static DEFINE_MUTEX(linear_mapping_mutex);
53 
54 #ifdef CONFIG_NUMA
55 int memory_add_physaddr_to_nid(u64 start)
56 {
57 	return hot_add_scn_to_nid(start);
58 }
59 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
60 #endif
61 
62 int __weak create_section_mapping(unsigned long start, unsigned long end,
63 				  int nid, pgprot_t prot)
64 {
65 	return -ENODEV;
66 }
67 
68 int __weak remove_section_mapping(unsigned long start, unsigned long end)
69 {
70 	return -ENODEV;
71 }
72 
73 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
74 				     struct mhp_params *params)
75 {
76 	int rc;
77 
78 	start = (unsigned long)__va(start);
79 	mutex_lock(&linear_mapping_mutex);
80 	rc = create_section_mapping(start, start + size, nid,
81 				    params->pgprot);
82 	mutex_unlock(&linear_mapping_mutex);
83 	if (rc) {
84 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
85 			start, start + size, rc);
86 		return -EFAULT;
87 	}
88 	return 0;
89 }
90 
91 void __ref arch_remove_linear_mapping(u64 start, u64 size)
92 {
93 	int ret;
94 
95 	/* Remove htab bolted mappings for this section of memory */
96 	start = (unsigned long)__va(start);
97 
98 	mutex_lock(&linear_mapping_mutex);
99 	ret = remove_section_mapping(start, start + size);
100 	mutex_unlock(&linear_mapping_mutex);
101 	if (ret)
102 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
103 			start, start + size, ret);
104 
105 	/* Ensure all vmalloc mappings are flushed in case they also
106 	 * hit that section of memory
107 	 */
108 	vm_unmap_aliases();
109 }
110 
111 /*
112  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
113  * updating.
114  */
115 static void update_end_of_memory_vars(u64 start, u64 size)
116 {
117 	unsigned long end_pfn = PFN_UP(start + size);
118 
119 	if (end_pfn > max_pfn) {
120 		max_pfn = end_pfn;
121 		max_low_pfn = end_pfn;
122 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
123 	}
124 }
125 
126 int __ref add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
127 		    struct mhp_params *params)
128 {
129 	int ret;
130 
131 	ret = __add_pages(nid, start_pfn, nr_pages, params);
132 	if (ret)
133 		return ret;
134 
135 	/* update max_pfn, max_low_pfn and high_memory */
136 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
137 				  nr_pages << PAGE_SHIFT);
138 
139 	return ret;
140 }
141 
142 int __ref arch_add_memory(int nid, u64 start, u64 size,
143 			  struct mhp_params *params)
144 {
145 	unsigned long start_pfn = start >> PAGE_SHIFT;
146 	unsigned long nr_pages = size >> PAGE_SHIFT;
147 	int rc;
148 
149 	rc = arch_create_linear_mapping(nid, start, size, params);
150 	if (rc)
151 		return rc;
152 	rc = add_pages(nid, start_pfn, nr_pages, params);
153 	if (rc)
154 		arch_remove_linear_mapping(start, size);
155 	return rc;
156 }
157 
158 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
159 {
160 	unsigned long start_pfn = start >> PAGE_SHIFT;
161 	unsigned long nr_pages = size >> PAGE_SHIFT;
162 
163 	__remove_pages(start_pfn, nr_pages, altmap);
164 	arch_remove_linear_mapping(start, size);
165 }
166 #endif
167 
168 #ifndef CONFIG_NUMA
169 void __init mem_topology_setup(void)
170 {
171 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
172 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
173 #ifdef CONFIG_HIGHMEM
174 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
175 #endif
176 
177 	/* Place all memblock_regions in the same node and merge contiguous
178 	 * memblock_regions
179 	 */
180 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
181 }
182 
183 void __init initmem_init(void)
184 {
185 	sparse_init();
186 }
187 
188 /* mark pages that don't exist as nosave */
189 static int __init mark_nonram_nosave(void)
190 {
191 	unsigned long spfn, epfn, prev = 0;
192 	int i;
193 
194 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
195 		if (prev && prev < spfn)
196 			register_nosave_region(prev, spfn);
197 
198 		prev = epfn;
199 	}
200 
201 	return 0;
202 }
203 #else /* CONFIG_NUMA */
204 static int __init mark_nonram_nosave(void)
205 {
206 	return 0;
207 }
208 #endif
209 
210 /*
211  * Zones usage:
212  *
213  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
214  * everything else. GFP_DMA32 page allocations automatically fall back to
215  * ZONE_DMA.
216  *
217  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
218  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
219  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
220  * ZONE_DMA.
221  */
222 static unsigned long max_zone_pfns[MAX_NR_ZONES];
223 
224 /*
225  * paging_init() sets up the page tables - in fact we've already done this.
226  */
227 void __init paging_init(void)
228 {
229 	unsigned long long total_ram = memblock_phys_mem_size();
230 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
231 
232 #ifdef CONFIG_HIGHMEM
233 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
234 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
235 
236 	for (; v < end; v += PAGE_SIZE)
237 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
238 
239 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
240 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
241 #endif /* CONFIG_HIGHMEM */
242 
243 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
244 	       (unsigned long long)top_of_ram, total_ram);
245 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
246 	       (long int)((top_of_ram - total_ram) >> 20));
247 
248 	/*
249 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
250 	 * powerbooks.
251 	 */
252 	if (IS_ENABLED(CONFIG_PPC32))
253 		zone_dma_bits = 30;
254 	else
255 		zone_dma_bits = 31;
256 
257 #ifdef CONFIG_ZONE_DMA
258 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
259 				      1UL << (zone_dma_bits - PAGE_SHIFT));
260 #endif
261 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
262 #ifdef CONFIG_HIGHMEM
263 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
264 #endif
265 
266 	free_area_init(max_zone_pfns);
267 
268 	mark_nonram_nosave();
269 }
270 
271 void __init mem_init(void)
272 {
273 	/*
274 	 * book3s is limited to 16 page sizes due to encoding this in
275 	 * a 4-bit field for slices.
276 	 */
277 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
278 
279 #ifdef CONFIG_SWIOTLB
280 	/*
281 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
282 	 * 4G. We force memblock to bottom-up mode to ensure that the
283 	 * memory allocated in swiotlb_init() is DMA-able.
284 	 * As it's the last memblock allocation, no need to reset it
285 	 * back to to-down.
286 	 */
287 	memblock_set_bottom_up(true);
288 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
289 #endif
290 
291 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
292 
293 	kasan_late_init();
294 
295 	memblock_free_all();
296 
297 #ifdef CONFIG_HIGHMEM
298 	{
299 		unsigned long pfn, highmem_mapnr;
300 
301 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
302 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
303 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
304 			struct page *page = pfn_to_page(pfn);
305 			if (memblock_is_memory(paddr) && !memblock_is_reserved(paddr))
306 				free_highmem_page(page);
307 		}
308 	}
309 #endif /* CONFIG_HIGHMEM */
310 
311 #if defined(CONFIG_PPC_E500) && !defined(CONFIG_SMP)
312 	/*
313 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
314 	 * functions.... do it here for the non-smp case.
315 	 */
316 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
317 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
318 #endif
319 
320 #ifdef CONFIG_PPC32
321 	pr_info("Kernel virtual memory layout:\n");
322 #ifdef CONFIG_KASAN
323 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
324 		KASAN_SHADOW_START, KASAN_SHADOW_END);
325 #endif
326 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
327 #ifdef CONFIG_HIGHMEM
328 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
329 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
330 #endif /* CONFIG_HIGHMEM */
331 	if (ioremap_bot != IOREMAP_TOP)
332 		pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
333 			ioremap_bot, IOREMAP_TOP);
334 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
335 		VMALLOC_START, VMALLOC_END);
336 #ifdef MODULES_VADDR
337 	pr_info("  * 0x%08lx..0x%08lx  : modules\n",
338 		MODULES_VADDR, MODULES_END);
339 #endif
340 #endif /* CONFIG_PPC32 */
341 }
342 
343 void free_initmem(void)
344 {
345 	ppc_md.progress = ppc_printk_progress;
346 	mark_initmem_nx();
347 	free_initmem_default(POISON_FREE_INITMEM);
348 	ftrace_free_init_tramp();
349 }
350 
351 /*
352  * System memory should not be in /proc/iomem but various tools expect it
353  * (eg kdump).
354  */
355 static int __init add_system_ram_resources(void)
356 {
357 	phys_addr_t start, end;
358 	u64 i;
359 
360 	for_each_mem_range(i, &start, &end) {
361 		struct resource *res;
362 
363 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
364 		WARN_ON(!res);
365 
366 		if (res) {
367 			res->name = "System RAM";
368 			res->start = start;
369 			/*
370 			 * In memblock, end points to the first byte after
371 			 * the range while in resourses, end points to the
372 			 * last byte in the range.
373 			 */
374 			res->end = end - 1;
375 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
376 			WARN_ON(request_resource(&iomem_resource, res) < 0);
377 		}
378 	}
379 
380 	return 0;
381 }
382 subsys_initcall(add_system_ram_resources);
383 
384 #ifdef CONFIG_STRICT_DEVMEM
385 /*
386  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
387  * is valid. The argument is a physical page number.
388  *
389  * Access has to be given to non-kernel-ram areas as well, these contain the
390  * PCI mmio resources as well as potential bios/acpi data regions.
391  */
392 int devmem_is_allowed(unsigned long pfn)
393 {
394 	if (page_is_rtas_user_buf(pfn))
395 		return 1;
396 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
397 		return 0;
398 	if (!page_is_ram(pfn))
399 		return 1;
400 	return 0;
401 }
402 #endif /* CONFIG_STRICT_DEVMEM */
403 
404 /*
405  * This is defined in kernel/resource.c but only powerpc needs to export it, for
406  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
407  */
408 EXPORT_SYMBOL_GPL(walk_system_ram_range);
409