xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 8ede5890)
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 #include <asm/mmzone.h>
25 
26 #include <mm/mmu_decl.h>
27 
28 unsigned long long memory_limit;
29 
30 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
31 EXPORT_SYMBOL(empty_zero_page);
32 
33 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
34 			      unsigned long size, pgprot_t vma_prot)
35 {
36 	if (ppc_md.phys_mem_access_prot)
37 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
38 
39 	if (!page_is_ram(pfn))
40 		vma_prot = pgprot_noncached(vma_prot);
41 
42 	return vma_prot;
43 }
44 EXPORT_SYMBOL(phys_mem_access_prot);
45 
46 #ifdef CONFIG_MEMORY_HOTPLUG
47 static DEFINE_MUTEX(linear_mapping_mutex);
48 
49 #ifdef CONFIG_NUMA
50 int memory_add_physaddr_to_nid(u64 start)
51 {
52 	return hot_add_scn_to_nid(start);
53 }
54 #endif
55 
56 int __weak create_section_mapping(unsigned long start, unsigned long end,
57 				  int nid, pgprot_t prot)
58 {
59 	return -ENODEV;
60 }
61 
62 int __weak remove_section_mapping(unsigned long start, unsigned long end)
63 {
64 	return -ENODEV;
65 }
66 
67 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
68 				     struct mhp_params *params)
69 {
70 	int rc;
71 
72 	start = (unsigned long)__va(start);
73 	mutex_lock(&linear_mapping_mutex);
74 	rc = create_section_mapping(start, start + size, nid,
75 				    params->pgprot);
76 	mutex_unlock(&linear_mapping_mutex);
77 	if (rc) {
78 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
79 			start, start + size, rc);
80 		return -EFAULT;
81 	}
82 	return 0;
83 }
84 
85 void __ref arch_remove_linear_mapping(u64 start, u64 size)
86 {
87 	int ret;
88 
89 	/* Remove htab bolted mappings for this section of memory */
90 	start = (unsigned long)__va(start);
91 
92 	mutex_lock(&linear_mapping_mutex);
93 	ret = remove_section_mapping(start, start + size);
94 	mutex_unlock(&linear_mapping_mutex);
95 	if (ret)
96 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
97 			start, start + size, ret);
98 
99 	/* Ensure all vmalloc mappings are flushed in case they also
100 	 * hit that section of memory
101 	 */
102 	vm_unmap_aliases();
103 }
104 
105 int __ref arch_add_memory(int nid, u64 start, u64 size,
106 			  struct mhp_params *params)
107 {
108 	unsigned long start_pfn = start >> PAGE_SHIFT;
109 	unsigned long nr_pages = size >> PAGE_SHIFT;
110 	int rc;
111 
112 	rc = arch_create_linear_mapping(nid, start, size, params);
113 	if (rc)
114 		return rc;
115 	rc = __add_pages(nid, start_pfn, nr_pages, params);
116 	if (rc)
117 		arch_remove_linear_mapping(start, size);
118 	return rc;
119 }
120 
121 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
122 {
123 	unsigned long start_pfn = start >> PAGE_SHIFT;
124 	unsigned long nr_pages = size >> PAGE_SHIFT;
125 
126 	__remove_pages(start_pfn, nr_pages, altmap);
127 	arch_remove_linear_mapping(start, size);
128 }
129 #endif
130 
131 #ifndef CONFIG_NUMA
132 void __init mem_topology_setup(void)
133 {
134 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
135 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
136 #ifdef CONFIG_HIGHMEM
137 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
138 #endif
139 
140 	/* Place all memblock_regions in the same node and merge contiguous
141 	 * memblock_regions
142 	 */
143 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
144 }
145 
146 void __init initmem_init(void)
147 {
148 	sparse_init();
149 }
150 
151 /* mark pages that don't exist as nosave */
152 static int __init mark_nonram_nosave(void)
153 {
154 	unsigned long spfn, epfn, prev = 0;
155 	int i;
156 
157 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
158 		if (prev && prev < spfn)
159 			register_nosave_region(prev, spfn);
160 
161 		prev = epfn;
162 	}
163 
164 	return 0;
165 }
166 #else /* CONFIG_NUMA */
167 static int __init mark_nonram_nosave(void)
168 {
169 	return 0;
170 }
171 #endif
172 
173 /*
174  * Zones usage:
175  *
176  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
177  * everything else. GFP_DMA32 page allocations automatically fall back to
178  * ZONE_DMA.
179  *
180  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
181  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
182  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
183  * ZONE_DMA.
184  */
185 static unsigned long max_zone_pfns[MAX_NR_ZONES];
186 
187 /*
188  * paging_init() sets up the page tables - in fact we've already done this.
189  */
190 void __init paging_init(void)
191 {
192 	unsigned long long total_ram = memblock_phys_mem_size();
193 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
194 
195 #ifdef CONFIG_HIGHMEM
196 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
197 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
198 
199 	for (; v < end; v += PAGE_SIZE)
200 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
201 
202 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
203 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
204 #endif /* CONFIG_HIGHMEM */
205 
206 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
207 	       (unsigned long long)top_of_ram, total_ram);
208 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
209 	       (long int)((top_of_ram - total_ram) >> 20));
210 
211 	/*
212 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
213 	 * powerbooks.
214 	 */
215 	if (IS_ENABLED(CONFIG_PPC32))
216 		zone_dma_bits = 30;
217 	else
218 		zone_dma_bits = 31;
219 
220 #ifdef CONFIG_ZONE_DMA
221 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
222 				      1UL << (zone_dma_bits - PAGE_SHIFT));
223 #endif
224 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
225 #ifdef CONFIG_HIGHMEM
226 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
227 #endif
228 
229 	free_area_init(max_zone_pfns);
230 
231 	mark_nonram_nosave();
232 }
233 
234 void __init mem_init(void)
235 {
236 	/*
237 	 * book3s is limited to 16 page sizes due to encoding this in
238 	 * a 4-bit field for slices.
239 	 */
240 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
241 
242 #ifdef CONFIG_SWIOTLB
243 	/*
244 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
245 	 * 4G. We force memblock to bottom-up mode to ensure that the
246 	 * memory allocated in swiotlb_init() is DMA-able.
247 	 * As it's the last memblock allocation, no need to reset it
248 	 * back to to-down.
249 	 */
250 	memblock_set_bottom_up(true);
251 	if (is_secure_guest())
252 		svm_swiotlb_init();
253 	else
254 		swiotlb_init(0);
255 #endif
256 
257 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
258 	set_max_mapnr(max_low_pfn);
259 
260 	kasan_late_init();
261 
262 	memblock_free_all();
263 
264 #ifdef CONFIG_HIGHMEM
265 	{
266 		unsigned long pfn, highmem_mapnr;
267 
268 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
269 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
270 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
271 			struct page *page = pfn_to_page(pfn);
272 			if (!memblock_is_reserved(paddr))
273 				free_highmem_page(page);
274 		}
275 	}
276 #endif /* CONFIG_HIGHMEM */
277 
278 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
279 	/*
280 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
281 	 * functions.... do it here for the non-smp case.
282 	 */
283 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
284 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
285 #endif
286 
287 #ifdef CONFIG_PPC32
288 	pr_info("Kernel virtual memory layout:\n");
289 #ifdef CONFIG_KASAN
290 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
291 		KASAN_SHADOW_START, KASAN_SHADOW_END);
292 #endif
293 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
294 #ifdef CONFIG_HIGHMEM
295 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
296 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
297 #endif /* CONFIG_HIGHMEM */
298 	if (ioremap_bot != IOREMAP_TOP)
299 		pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
300 			ioremap_bot, IOREMAP_TOP);
301 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
302 		VMALLOC_START, VMALLOC_END);
303 #ifdef MODULES_VADDR
304 	pr_info("  * 0x%08lx..0x%08lx  : modules\n",
305 		MODULES_VADDR, MODULES_END);
306 #endif
307 #endif /* CONFIG_PPC32 */
308 }
309 
310 void free_initmem(void)
311 {
312 	ppc_md.progress = ppc_printk_progress;
313 	mark_initmem_nx();
314 	free_initmem_default(POISON_FREE_INITMEM);
315 }
316 
317 /*
318  * System memory should not be in /proc/iomem but various tools expect it
319  * (eg kdump).
320  */
321 static int __init add_system_ram_resources(void)
322 {
323 	phys_addr_t start, end;
324 	u64 i;
325 
326 	for_each_mem_range(i, &start, &end) {
327 		struct resource *res;
328 
329 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
330 		WARN_ON(!res);
331 
332 		if (res) {
333 			res->name = "System RAM";
334 			res->start = start;
335 			/*
336 			 * In memblock, end points to the first byte after
337 			 * the range while in resourses, end points to the
338 			 * last byte in the range.
339 			 */
340 			res->end = end - 1;
341 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
342 			WARN_ON(request_resource(&iomem_resource, res) < 0);
343 		}
344 	}
345 
346 	return 0;
347 }
348 subsys_initcall(add_system_ram_resources);
349 
350 #ifdef CONFIG_STRICT_DEVMEM
351 /*
352  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
353  * is valid. The argument is a physical page number.
354  *
355  * Access has to be given to non-kernel-ram areas as well, these contain the
356  * PCI mmio resources as well as potential bios/acpi data regions.
357  */
358 int devmem_is_allowed(unsigned long pfn)
359 {
360 	if (page_is_rtas_user_buf(pfn))
361 		return 1;
362 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
363 		return 0;
364 	if (!page_is_ram(pfn))
365 		return 1;
366 	return 0;
367 }
368 #endif /* CONFIG_STRICT_DEVMEM */
369 
370 /*
371  * This is defined in kernel/resource.c but only powerpc needs to export it, for
372  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
373  */
374 EXPORT_SYMBOL_GPL(walk_system_ram_range);
375