xref: /linux/arch/powerpc/mm/mem.c (revision b5bee6ced21ca98389000b7017dd41b0cc37fa50)
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 
30 #include <mm/mmu_decl.h>
31 
32 unsigned long long memory_limit;
33 
34 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
35 EXPORT_SYMBOL(empty_zero_page);
36 
37 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
38 			      unsigned long size, pgprot_t vma_prot)
39 {
40 	if (ppc_md.phys_mem_access_prot)
41 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
42 
43 	if (!page_is_ram(pfn))
44 		vma_prot = pgprot_noncached(vma_prot);
45 
46 	return vma_prot;
47 }
48 EXPORT_SYMBOL(phys_mem_access_prot);
49 
50 #ifdef CONFIG_MEMORY_HOTPLUG
51 static DEFINE_MUTEX(linear_mapping_mutex);
52 
53 #ifdef CONFIG_NUMA
54 int memory_add_physaddr_to_nid(u64 start)
55 {
56 	return hot_add_scn_to_nid(start);
57 }
58 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
59 #endif
60 
61 int __weak create_section_mapping(unsigned long start, unsigned long end,
62 				  int nid, pgprot_t prot)
63 {
64 	return -ENODEV;
65 }
66 
67 int __weak remove_section_mapping(unsigned long start, unsigned long end)
68 {
69 	return -ENODEV;
70 }
71 
72 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
73 				     struct mhp_params *params)
74 {
75 	int rc;
76 
77 	start = (unsigned long)__va(start);
78 	mutex_lock(&linear_mapping_mutex);
79 	rc = create_section_mapping(start, start + size, nid,
80 				    params->pgprot);
81 	mutex_unlock(&linear_mapping_mutex);
82 	if (rc) {
83 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
84 			start, start + size, rc);
85 		return -EFAULT;
86 	}
87 	return 0;
88 }
89 
90 void __ref arch_remove_linear_mapping(u64 start, u64 size)
91 {
92 	int ret;
93 
94 	/* Remove htab bolted mappings for this section of memory */
95 	start = (unsigned long)__va(start);
96 
97 	mutex_lock(&linear_mapping_mutex);
98 	ret = remove_section_mapping(start, start + size);
99 	mutex_unlock(&linear_mapping_mutex);
100 	if (ret)
101 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
102 			start, start + size, ret);
103 
104 	/* Ensure all vmalloc mappings are flushed in case they also
105 	 * hit that section of memory
106 	 */
107 	vm_unmap_aliases();
108 }
109 
110 /*
111  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
112  * updating.
113  */
114 static void update_end_of_memory_vars(u64 start, u64 size)
115 {
116 	unsigned long end_pfn = PFN_UP(start + size);
117 
118 	if (end_pfn > max_pfn) {
119 		max_pfn = end_pfn;
120 		max_low_pfn = end_pfn;
121 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
122 	}
123 }
124 
125 int __ref add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
126 		    struct mhp_params *params)
127 {
128 	int ret;
129 
130 	ret = __add_pages(nid, start_pfn, nr_pages, params);
131 	if (ret)
132 		return ret;
133 
134 	/* update max_pfn, max_low_pfn and high_memory */
135 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
136 				  nr_pages << PAGE_SHIFT);
137 
138 	return ret;
139 }
140 
141 int __ref arch_add_memory(int nid, u64 start, u64 size,
142 			  struct mhp_params *params)
143 {
144 	unsigned long start_pfn = start >> PAGE_SHIFT;
145 	unsigned long nr_pages = size >> PAGE_SHIFT;
146 	int rc;
147 
148 	rc = arch_create_linear_mapping(nid, start, size, params);
149 	if (rc)
150 		return rc;
151 	rc = add_pages(nid, start_pfn, nr_pages, params);
152 	if (rc)
153 		arch_remove_linear_mapping(start, size);
154 	return rc;
155 }
156 
157 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
158 {
159 	unsigned long start_pfn = start >> PAGE_SHIFT;
160 	unsigned long nr_pages = size >> PAGE_SHIFT;
161 
162 	__remove_pages(start_pfn, nr_pages, altmap);
163 	arch_remove_linear_mapping(start, size);
164 }
165 #endif
166 
167 #ifndef CONFIG_NUMA
168 void __init mem_topology_setup(void)
169 {
170 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
171 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
172 #ifdef CONFIG_HIGHMEM
173 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
174 #endif
175 
176 	/* Place all memblock_regions in the same node and merge contiguous
177 	 * memblock_regions
178 	 */
179 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
180 }
181 
182 void __init initmem_init(void)
183 {
184 	sparse_init();
185 }
186 
187 /* mark pages that don't exist as nosave */
188 static int __init mark_nonram_nosave(void)
189 {
190 	unsigned long spfn, epfn, prev = 0;
191 	int i;
192 
193 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
194 		if (prev && prev < spfn)
195 			register_nosave_region(prev, spfn);
196 
197 		prev = epfn;
198 	}
199 
200 	return 0;
201 }
202 #else /* CONFIG_NUMA */
203 static int __init mark_nonram_nosave(void)
204 {
205 	return 0;
206 }
207 #endif
208 
209 /*
210  * Zones usage:
211  *
212  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
213  * everything else. GFP_DMA32 page allocations automatically fall back to
214  * ZONE_DMA.
215  *
216  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
217  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
218  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
219  * ZONE_DMA.
220  */
221 static unsigned long max_zone_pfns[MAX_NR_ZONES];
222 
223 /*
224  * paging_init() sets up the page tables - in fact we've already done this.
225  */
226 void __init paging_init(void)
227 {
228 	unsigned long long total_ram = memblock_phys_mem_size();
229 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
230 
231 #ifdef CONFIG_HIGHMEM
232 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
233 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
234 
235 	for (; v < end; v += PAGE_SIZE)
236 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
237 
238 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
239 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
240 #endif /* CONFIG_HIGHMEM */
241 
242 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
243 	       (unsigned long long)top_of_ram, total_ram);
244 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
245 	       (long int)((top_of_ram - total_ram) >> 20));
246 
247 	/*
248 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
249 	 * powerbooks.
250 	 */
251 	if (IS_ENABLED(CONFIG_PPC32))
252 		zone_dma_bits = 30;
253 	else
254 		zone_dma_bits = 31;
255 
256 #ifdef CONFIG_ZONE_DMA
257 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
258 				      1UL << (zone_dma_bits - PAGE_SHIFT));
259 #endif
260 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
261 #ifdef CONFIG_HIGHMEM
262 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
263 #endif
264 
265 	free_area_init(max_zone_pfns);
266 
267 	mark_nonram_nosave();
268 }
269 
270 void __init mem_init(void)
271 {
272 	/*
273 	 * book3s is limited to 16 page sizes due to encoding this in
274 	 * a 4-bit field for slices.
275 	 */
276 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
277 
278 #ifdef CONFIG_SWIOTLB
279 	/*
280 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
281 	 * 4G. We force memblock to bottom-up mode to ensure that the
282 	 * memory allocated in swiotlb_init() is DMA-able.
283 	 * As it's the last memblock allocation, no need to reset it
284 	 * back to to-down.
285 	 */
286 	memblock_set_bottom_up(true);
287 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
288 #endif
289 
290 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
291 	set_max_mapnr(max_pfn);
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_reserved(paddr))
306 				free_highmem_page(page);
307 		}
308 	}
309 #endif /* CONFIG_HIGHMEM */
310 
311 #if defined(CONFIG_PPC_FSL_BOOK3E) && !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 	static_branch_enable(&init_mem_is_free);
348 	free_initmem_default(POISON_FREE_INITMEM);
349 	ftrace_free_init_tramp();
350 }
351 
352 /*
353  * System memory should not be in /proc/iomem but various tools expect it
354  * (eg kdump).
355  */
356 static int __init add_system_ram_resources(void)
357 {
358 	phys_addr_t start, end;
359 	u64 i;
360 
361 	for_each_mem_range(i, &start, &end) {
362 		struct resource *res;
363 
364 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
365 		WARN_ON(!res);
366 
367 		if (res) {
368 			res->name = "System RAM";
369 			res->start = start;
370 			/*
371 			 * In memblock, end points to the first byte after
372 			 * the range while in resourses, end points to the
373 			 * last byte in the range.
374 			 */
375 			res->end = end - 1;
376 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
377 			WARN_ON(request_resource(&iomem_resource, res) < 0);
378 		}
379 	}
380 
381 	return 0;
382 }
383 subsys_initcall(add_system_ram_resources);
384 
385 #ifdef CONFIG_STRICT_DEVMEM
386 /*
387  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
388  * is valid. The argument is a physical page number.
389  *
390  * Access has to be given to non-kernel-ram areas as well, these contain the
391  * PCI mmio resources as well as potential bios/acpi data regions.
392  */
393 int devmem_is_allowed(unsigned long pfn)
394 {
395 	if (page_is_rtas_user_buf(pfn))
396 		return 1;
397 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
398 		return 0;
399 	if (!page_is_ram(pfn))
400 		return 1;
401 	return 0;
402 }
403 #endif /* CONFIG_STRICT_DEVMEM */
404 
405 /*
406  * This is defined in kernel/resource.c but only powerpc needs to export it, for
407  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
408  */
409 EXPORT_SYMBOL_GPL(walk_system_ram_range);
410