xref: /linux/arch/powerpc/mm/mem.c (revision 5ee46bfbb65fd971b734c3972ea9cc123fc869d1)
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 
29 #include <mm/mmu_decl.h>
30 
31 unsigned long long memory_limit;
32 
33 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
34 EXPORT_SYMBOL(empty_zero_page);
35 
36 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
37 			      unsigned long size, pgprot_t vma_prot)
38 {
39 	if (ppc_md.phys_mem_access_prot)
40 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
41 
42 	if (!page_is_ram(pfn))
43 		vma_prot = pgprot_noncached(vma_prot);
44 
45 	return vma_prot;
46 }
47 EXPORT_SYMBOL(phys_mem_access_prot);
48 
49 #ifdef CONFIG_MEMORY_HOTPLUG
50 static DEFINE_MUTEX(linear_mapping_mutex);
51 
52 #ifdef CONFIG_NUMA
53 int memory_add_physaddr_to_nid(u64 start)
54 {
55 	return hot_add_scn_to_nid(start);
56 }
57 #endif
58 
59 int __weak create_section_mapping(unsigned long start, unsigned long end,
60 				  int nid, pgprot_t prot)
61 {
62 	return -ENODEV;
63 }
64 
65 int __weak remove_section_mapping(unsigned long start, unsigned long end)
66 {
67 	return -ENODEV;
68 }
69 
70 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
71 				     struct mhp_params *params)
72 {
73 	int rc;
74 
75 	start = (unsigned long)__va(start);
76 	mutex_lock(&linear_mapping_mutex);
77 	rc = create_section_mapping(start, start + size, nid,
78 				    params->pgprot);
79 	mutex_unlock(&linear_mapping_mutex);
80 	if (rc) {
81 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
82 			start, start + size, rc);
83 		return -EFAULT;
84 	}
85 	return 0;
86 }
87 
88 void __ref arch_remove_linear_mapping(u64 start, u64 size)
89 {
90 	int ret;
91 
92 	/* Remove htab bolted mappings for this section of memory */
93 	start = (unsigned long)__va(start);
94 
95 	mutex_lock(&linear_mapping_mutex);
96 	ret = remove_section_mapping(start, start + size);
97 	mutex_unlock(&linear_mapping_mutex);
98 	if (ret)
99 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
100 			start, start + size, ret);
101 
102 	/* Ensure all vmalloc mappings are flushed in case they also
103 	 * hit that section of memory
104 	 */
105 	vm_unmap_aliases();
106 }
107 
108 /*
109  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
110  * updating.
111  */
112 static void update_end_of_memory_vars(u64 start, u64 size)
113 {
114 	unsigned long end_pfn = PFN_UP(start + size);
115 
116 	if (end_pfn > max_pfn) {
117 		max_pfn = end_pfn;
118 		max_low_pfn = end_pfn;
119 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
120 	}
121 }
122 
123 int __ref add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
124 		    struct mhp_params *params)
125 {
126 	int ret;
127 
128 	ret = __add_pages(nid, start_pfn, nr_pages, params);
129 	if (ret)
130 		return ret;
131 
132 	/* update max_pfn, max_low_pfn and high_memory */
133 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
134 				  nr_pages << PAGE_SHIFT);
135 
136 	return ret;
137 }
138 
139 int __ref arch_add_memory(int nid, u64 start, u64 size,
140 			  struct mhp_params *params)
141 {
142 	unsigned long start_pfn = start >> PAGE_SHIFT;
143 	unsigned long nr_pages = size >> PAGE_SHIFT;
144 	int rc;
145 
146 	rc = arch_create_linear_mapping(nid, start, size, params);
147 	if (rc)
148 		return rc;
149 	rc = add_pages(nid, start_pfn, nr_pages, params);
150 	if (rc)
151 		arch_remove_linear_mapping(start, size);
152 	return rc;
153 }
154 
155 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
156 {
157 	unsigned long start_pfn = start >> PAGE_SHIFT;
158 	unsigned long nr_pages = size >> PAGE_SHIFT;
159 
160 	__remove_pages(start_pfn, nr_pages, altmap);
161 	arch_remove_linear_mapping(start, size);
162 }
163 #endif
164 
165 #ifndef CONFIG_NUMA
166 void __init mem_topology_setup(void)
167 {
168 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
169 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
170 #ifdef CONFIG_HIGHMEM
171 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
172 #endif
173 
174 	/* Place all memblock_regions in the same node and merge contiguous
175 	 * memblock_regions
176 	 */
177 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
178 }
179 
180 void __init initmem_init(void)
181 {
182 	sparse_init();
183 }
184 
185 /* mark pages that don't exist as nosave */
186 static int __init mark_nonram_nosave(void)
187 {
188 	unsigned long spfn, epfn, prev = 0;
189 	int i;
190 
191 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
192 		if (prev && prev < spfn)
193 			register_nosave_region(prev, spfn);
194 
195 		prev = epfn;
196 	}
197 
198 	return 0;
199 }
200 #else /* CONFIG_NUMA */
201 static int __init mark_nonram_nosave(void)
202 {
203 	return 0;
204 }
205 #endif
206 
207 /*
208  * Zones usage:
209  *
210  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
211  * everything else. GFP_DMA32 page allocations automatically fall back to
212  * ZONE_DMA.
213  *
214  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
215  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
216  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
217  * ZONE_DMA.
218  */
219 static unsigned long max_zone_pfns[MAX_NR_ZONES];
220 
221 /*
222  * paging_init() sets up the page tables - in fact we've already done this.
223  */
224 void __init paging_init(void)
225 {
226 	unsigned long long total_ram = memblock_phys_mem_size();
227 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
228 
229 #ifdef CONFIG_HIGHMEM
230 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
231 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
232 
233 	for (; v < end; v += PAGE_SIZE)
234 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
235 
236 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
237 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
238 #endif /* CONFIG_HIGHMEM */
239 
240 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
241 	       (unsigned long long)top_of_ram, total_ram);
242 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
243 	       (long int)((top_of_ram - total_ram) >> 20));
244 
245 	/*
246 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
247 	 * powerbooks.
248 	 */
249 	if (IS_ENABLED(CONFIG_PPC32))
250 		zone_dma_bits = 30;
251 	else
252 		zone_dma_bits = 31;
253 
254 #ifdef CONFIG_ZONE_DMA
255 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
256 				      1UL << (zone_dma_bits - PAGE_SHIFT));
257 #endif
258 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
259 #ifdef CONFIG_HIGHMEM
260 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
261 #endif
262 
263 	free_area_init(max_zone_pfns);
264 
265 	mark_nonram_nosave();
266 }
267 
268 void __init mem_init(void)
269 {
270 	/*
271 	 * book3s is limited to 16 page sizes due to encoding this in
272 	 * a 4-bit field for slices.
273 	 */
274 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
275 
276 #ifdef CONFIG_SWIOTLB
277 	/*
278 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
279 	 * 4G. We force memblock to bottom-up mode to ensure that the
280 	 * memory allocated in swiotlb_init() is DMA-able.
281 	 * As it's the last memblock allocation, no need to reset it
282 	 * back to to-down.
283 	 */
284 	memblock_set_bottom_up(true);
285 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
286 #endif
287 
288 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
289 	set_max_mapnr(max_pfn);
290 
291 	kasan_late_init();
292 
293 	memblock_free_all();
294 
295 #ifdef CONFIG_HIGHMEM
296 	{
297 		unsigned long pfn, highmem_mapnr;
298 
299 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
300 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
301 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
302 			struct page *page = pfn_to_page(pfn);
303 			if (!memblock_is_reserved(paddr))
304 				free_highmem_page(page);
305 		}
306 	}
307 #endif /* CONFIG_HIGHMEM */
308 
309 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
310 	/*
311 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
312 	 * functions.... do it here for the non-smp case.
313 	 */
314 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
315 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
316 #endif
317 
318 #ifdef CONFIG_PPC32
319 	pr_info("Kernel virtual memory layout:\n");
320 #ifdef CONFIG_KASAN
321 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
322 		KASAN_SHADOW_START, KASAN_SHADOW_END);
323 #endif
324 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
325 #ifdef CONFIG_HIGHMEM
326 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
327 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
328 #endif /* CONFIG_HIGHMEM */
329 	if (ioremap_bot != IOREMAP_TOP)
330 		pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
331 			ioremap_bot, IOREMAP_TOP);
332 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
333 		VMALLOC_START, VMALLOC_END);
334 #ifdef MODULES_VADDR
335 	pr_info("  * 0x%08lx..0x%08lx  : modules\n",
336 		MODULES_VADDR, MODULES_END);
337 #endif
338 #endif /* CONFIG_PPC32 */
339 }
340 
341 void free_initmem(void)
342 {
343 	ppc_md.progress = ppc_printk_progress;
344 	mark_initmem_nx();
345 	static_branch_enable(&init_mem_is_free);
346 	free_initmem_default(POISON_FREE_INITMEM);
347 	ftrace_free_init_tramp();
348 }
349 
350 /*
351  * System memory should not be in /proc/iomem but various tools expect it
352  * (eg kdump).
353  */
354 static int __init add_system_ram_resources(void)
355 {
356 	phys_addr_t start, end;
357 	u64 i;
358 
359 	for_each_mem_range(i, &start, &end) {
360 		struct resource *res;
361 
362 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
363 		WARN_ON(!res);
364 
365 		if (res) {
366 			res->name = "System RAM";
367 			res->start = start;
368 			/*
369 			 * In memblock, end points to the first byte after
370 			 * the range while in resourses, end points to the
371 			 * last byte in the range.
372 			 */
373 			res->end = end - 1;
374 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
375 			WARN_ON(request_resource(&iomem_resource, res) < 0);
376 		}
377 	}
378 
379 	return 0;
380 }
381 subsys_initcall(add_system_ram_resources);
382 
383 #ifdef CONFIG_STRICT_DEVMEM
384 /*
385  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
386  * is valid. The argument is a physical page number.
387  *
388  * Access has to be given to non-kernel-ram areas as well, these contain the
389  * PCI mmio resources as well as potential bios/acpi data regions.
390  */
391 int devmem_is_allowed(unsigned long pfn)
392 {
393 	if (page_is_rtas_user_buf(pfn))
394 		return 1;
395 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
396 		return 0;
397 	if (!page_is_ram(pfn))
398 		return 1;
399 	return 0;
400 }
401 #endif /* CONFIG_STRICT_DEVMEM */
402 
403 /*
404  * This is defined in kernel/resource.c but only powerpc needs to export it, for
405  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
406  */
407 EXPORT_SYMBOL_GPL(walk_system_ram_range);
408