xref: /linux/arch/powerpc/mm/mem.c (revision 987b741c52c7c6c68d46fbaeb95b8d1087f10b7f)
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/machdep.h>
21 #include <asm/rtas.h>
22 #include <asm/kasan.h>
23 #include <asm/svm.h>
24 
25 #include <mm/mmu_decl.h>
26 
27 unsigned long long memory_limit;
28 bool init_mem_is_free;
29 
30 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
31 			      unsigned long size, pgprot_t vma_prot)
32 {
33 	if (ppc_md.phys_mem_access_prot)
34 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
35 
36 	if (!page_is_ram(pfn))
37 		vma_prot = pgprot_noncached(vma_prot);
38 
39 	return vma_prot;
40 }
41 EXPORT_SYMBOL(phys_mem_access_prot);
42 
43 #ifdef CONFIG_MEMORY_HOTPLUG
44 static DEFINE_MUTEX(linear_mapping_mutex);
45 
46 #ifdef CONFIG_NUMA
47 int memory_add_physaddr_to_nid(u64 start)
48 {
49 	return hot_add_scn_to_nid(start);
50 }
51 #endif
52 
53 int __weak create_section_mapping(unsigned long start, unsigned long end,
54 				  int nid, pgprot_t prot)
55 {
56 	return -ENODEV;
57 }
58 
59 int __weak remove_section_mapping(unsigned long start, unsigned long end)
60 {
61 	return -ENODEV;
62 }
63 
64 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
65 				     struct mhp_params *params)
66 {
67 	int rc;
68 
69 	start = (unsigned long)__va(start);
70 	mutex_lock(&linear_mapping_mutex);
71 	rc = create_section_mapping(start, start + size, nid,
72 				    params->pgprot);
73 	mutex_unlock(&linear_mapping_mutex);
74 	if (rc) {
75 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
76 			start, start + size, rc);
77 		return -EFAULT;
78 	}
79 	return 0;
80 }
81 
82 void __ref arch_remove_linear_mapping(u64 start, u64 size)
83 {
84 	int ret;
85 
86 	/* Remove htab bolted mappings for this section of memory */
87 	start = (unsigned long)__va(start);
88 
89 	mutex_lock(&linear_mapping_mutex);
90 	ret = remove_section_mapping(start, start + size);
91 	mutex_unlock(&linear_mapping_mutex);
92 	if (ret)
93 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
94 			start, start + size, ret);
95 
96 	/* Ensure all vmalloc mappings are flushed in case they also
97 	 * hit that section of memory
98 	 */
99 	vm_unmap_aliases();
100 }
101 
102 int __ref arch_add_memory(int nid, u64 start, u64 size,
103 			  struct mhp_params *params)
104 {
105 	unsigned long start_pfn = start >> PAGE_SHIFT;
106 	unsigned long nr_pages = size >> PAGE_SHIFT;
107 	int rc;
108 
109 	rc = arch_create_linear_mapping(nid, start, size, params);
110 	if (rc)
111 		return rc;
112 	rc = __add_pages(nid, start_pfn, nr_pages, params);
113 	if (rc)
114 		arch_remove_linear_mapping(start, size);
115 	return rc;
116 }
117 
118 void __ref arch_remove_memory(int nid, u64 start, u64 size,
119 			      struct vmem_altmap *altmap)
120 {
121 	unsigned long start_pfn = start >> PAGE_SHIFT;
122 	unsigned long nr_pages = size >> PAGE_SHIFT;
123 
124 	__remove_pages(start_pfn, nr_pages, altmap);
125 	arch_remove_linear_mapping(start, size);
126 }
127 #endif
128 
129 #ifndef CONFIG_NEED_MULTIPLE_NODES
130 void __init mem_topology_setup(void)
131 {
132 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
133 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
134 #ifdef CONFIG_HIGHMEM
135 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
136 #endif
137 
138 	/* Place all memblock_regions in the same node and merge contiguous
139 	 * memblock_regions
140 	 */
141 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
142 }
143 
144 void __init initmem_init(void)
145 {
146 	sparse_init();
147 }
148 
149 /* mark pages that don't exist as nosave */
150 static int __init mark_nonram_nosave(void)
151 {
152 	unsigned long spfn, epfn, prev = 0;
153 	int i;
154 
155 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
156 		if (prev && prev < spfn)
157 			register_nosave_region(prev, spfn);
158 
159 		prev = epfn;
160 	}
161 
162 	return 0;
163 }
164 #else /* CONFIG_NEED_MULTIPLE_NODES */
165 static int __init mark_nonram_nosave(void)
166 {
167 	return 0;
168 }
169 #endif
170 
171 /*
172  * Zones usage:
173  *
174  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
175  * everything else. GFP_DMA32 page allocations automatically fall back to
176  * ZONE_DMA.
177  *
178  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
179  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
180  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
181  * ZONE_DMA.
182  */
183 static unsigned long max_zone_pfns[MAX_NR_ZONES];
184 
185 /*
186  * paging_init() sets up the page tables - in fact we've already done this.
187  */
188 void __init paging_init(void)
189 {
190 	unsigned long long total_ram = memblock_phys_mem_size();
191 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
192 
193 #ifdef CONFIG_HIGHMEM
194 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
195 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
196 
197 	for (; v < end; v += PAGE_SIZE)
198 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
199 
200 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
201 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
202 #endif /* CONFIG_HIGHMEM */
203 
204 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
205 	       (unsigned long long)top_of_ram, total_ram);
206 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
207 	       (long int)((top_of_ram - total_ram) >> 20));
208 
209 	/*
210 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
211 	 * powerbooks.
212 	 */
213 	if (IS_ENABLED(CONFIG_PPC32))
214 		zone_dma_bits = 30;
215 	else
216 		zone_dma_bits = 31;
217 
218 #ifdef CONFIG_ZONE_DMA
219 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
220 				      1UL << (zone_dma_bits - PAGE_SHIFT));
221 #endif
222 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
223 #ifdef CONFIG_HIGHMEM
224 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
225 #endif
226 
227 	free_area_init(max_zone_pfns);
228 
229 	mark_nonram_nosave();
230 }
231 
232 void __init mem_init(void)
233 {
234 	/*
235 	 * book3s is limited to 16 page sizes due to encoding this in
236 	 * a 4-bit field for slices.
237 	 */
238 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
239 
240 #ifdef CONFIG_SWIOTLB
241 	/*
242 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
243 	 * 4G. We force memblock to bottom-up mode to ensure that the
244 	 * memory allocated in swiotlb_init() is DMA-able.
245 	 * As it's the last memblock allocation, no need to reset it
246 	 * back to to-down.
247 	 */
248 	memblock_set_bottom_up(true);
249 	if (is_secure_guest())
250 		svm_swiotlb_init();
251 	else
252 		swiotlb_init(0);
253 #endif
254 
255 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
256 	set_max_mapnr(max_pfn);
257 
258 	kasan_late_init();
259 
260 	memblock_free_all();
261 
262 #ifdef CONFIG_HIGHMEM
263 	{
264 		unsigned long pfn, highmem_mapnr;
265 
266 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
267 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
268 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
269 			struct page *page = pfn_to_page(pfn);
270 			if (!memblock_is_reserved(paddr))
271 				free_highmem_page(page);
272 		}
273 	}
274 #endif /* CONFIG_HIGHMEM */
275 
276 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
277 	/*
278 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
279 	 * functions.... do it here for the non-smp case.
280 	 */
281 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
282 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
283 #endif
284 
285 #ifdef CONFIG_PPC32
286 	pr_info("Kernel virtual memory layout:\n");
287 #ifdef CONFIG_KASAN
288 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
289 		KASAN_SHADOW_START, KASAN_SHADOW_END);
290 #endif
291 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
292 #ifdef CONFIG_HIGHMEM
293 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
294 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
295 #endif /* CONFIG_HIGHMEM */
296 	if (ioremap_bot != IOREMAP_TOP)
297 		pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
298 			ioremap_bot, IOREMAP_TOP);
299 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
300 		VMALLOC_START, VMALLOC_END);
301 #endif /* CONFIG_PPC32 */
302 }
303 
304 void free_initmem(void)
305 {
306 	ppc_md.progress = ppc_printk_progress;
307 	mark_initmem_nx();
308 	init_mem_is_free = true;
309 	free_initmem_default(POISON_FREE_INITMEM);
310 }
311 
312 /*
313  * System memory should not be in /proc/iomem but various tools expect it
314  * (eg kdump).
315  */
316 static int __init add_system_ram_resources(void)
317 {
318 	phys_addr_t start, end;
319 	u64 i;
320 
321 	for_each_mem_range(i, &start, &end) {
322 		struct resource *res;
323 
324 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
325 		WARN_ON(!res);
326 
327 		if (res) {
328 			res->name = "System RAM";
329 			res->start = start;
330 			/*
331 			 * In memblock, end points to the first byte after
332 			 * the range while in resourses, end points to the
333 			 * last byte in the range.
334 			 */
335 			res->end = end - 1;
336 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
337 			WARN_ON(request_resource(&iomem_resource, res) < 0);
338 		}
339 	}
340 
341 	return 0;
342 }
343 subsys_initcall(add_system_ram_resources);
344 
345 #ifdef CONFIG_STRICT_DEVMEM
346 /*
347  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
348  * is valid. The argument is a physical page number.
349  *
350  * Access has to be given to non-kernel-ram areas as well, these contain the
351  * PCI mmio resources as well as potential bios/acpi data regions.
352  */
353 int devmem_is_allowed(unsigned long pfn)
354 {
355 	if (page_is_rtas_user_buf(pfn))
356 		return 1;
357 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
358 		return 0;
359 	if (!page_is_ram(pfn))
360 		return 1;
361 	return 0;
362 }
363 #endif /* CONFIG_STRICT_DEVMEM */
364 
365 /*
366  * This is defined in kernel/resource.c but only powerpc needs to export it, for
367  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
368  */
369 EXPORT_SYMBOL_GPL(walk_system_ram_range);
370