xref: /linux/arch/powerpc/mm/mem.c (revision 8f426582e0e0c9bbd58e170e1b209334eb5df79e)
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 #endif
59 
60 int __weak create_section_mapping(unsigned long start, unsigned long end,
61 				  int nid, pgprot_t prot)
62 {
63 	return -ENODEV;
64 }
65 
66 int __weak remove_section_mapping(unsigned long start, unsigned long end)
67 {
68 	return -ENODEV;
69 }
70 
71 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
72 				     struct mhp_params *params)
73 {
74 	int rc;
75 
76 	start = (unsigned long)__va(start);
77 	mutex_lock(&linear_mapping_mutex);
78 	rc = create_section_mapping(start, start + size, nid,
79 				    params->pgprot);
80 	mutex_unlock(&linear_mapping_mutex);
81 	if (rc) {
82 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
83 			start, start + size, rc);
84 		return -EFAULT;
85 	}
86 	return 0;
87 }
88 
89 void __ref arch_remove_linear_mapping(u64 start, u64 size)
90 {
91 	int ret;
92 
93 	/* Remove htab bolted mappings for this section of memory */
94 	start = (unsigned long)__va(start);
95 
96 	mutex_lock(&linear_mapping_mutex);
97 	ret = remove_section_mapping(start, start + size);
98 	mutex_unlock(&linear_mapping_mutex);
99 	if (ret)
100 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
101 			start, start + size, ret);
102 
103 	/* Ensure all vmalloc mappings are flushed in case they also
104 	 * hit that section of memory
105 	 */
106 	vm_unmap_aliases();
107 }
108 
109 /*
110  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
111  * updating.
112  */
113 static void update_end_of_memory_vars(u64 start, u64 size)
114 {
115 	unsigned long end_pfn = PFN_UP(start + size);
116 
117 	if (end_pfn > max_pfn) {
118 		max_pfn = end_pfn;
119 		max_low_pfn = end_pfn;
120 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
121 	}
122 }
123 
124 int __ref add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
125 		    struct mhp_params *params)
126 {
127 	int ret;
128 
129 	ret = __add_pages(nid, start_pfn, nr_pages, params);
130 	if (ret)
131 		return ret;
132 
133 	/* update max_pfn, max_low_pfn and high_memory */
134 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
135 				  nr_pages << PAGE_SHIFT);
136 
137 	return ret;
138 }
139 
140 int __ref arch_add_memory(int nid, u64 start, u64 size,
141 			  struct mhp_params *params)
142 {
143 	unsigned long start_pfn = start >> PAGE_SHIFT;
144 	unsigned long nr_pages = size >> PAGE_SHIFT;
145 	int rc;
146 
147 	rc = arch_create_linear_mapping(nid, start, size, params);
148 	if (rc)
149 		return rc;
150 	rc = add_pages(nid, start_pfn, nr_pages, params);
151 	if (rc)
152 		arch_remove_linear_mapping(start, size);
153 	return rc;
154 }
155 
156 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
157 {
158 	unsigned long start_pfn = start >> PAGE_SHIFT;
159 	unsigned long nr_pages = size >> PAGE_SHIFT;
160 
161 	__remove_pages(start_pfn, nr_pages, altmap);
162 	arch_remove_linear_mapping(start, size);
163 }
164 #endif
165 
166 #ifndef CONFIG_NUMA
167 void __init mem_topology_setup(void)
168 {
169 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
170 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
171 #ifdef CONFIG_HIGHMEM
172 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
173 #endif
174 
175 	/* Place all memblock_regions in the same node and merge contiguous
176 	 * memblock_regions
177 	 */
178 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
179 }
180 
181 void __init initmem_init(void)
182 {
183 	sparse_init();
184 }
185 
186 /* mark pages that don't exist as nosave */
187 static int __init mark_nonram_nosave(void)
188 {
189 	unsigned long spfn, epfn, prev = 0;
190 	int i;
191 
192 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
193 		if (prev && prev < spfn)
194 			register_nosave_region(prev, spfn);
195 
196 		prev = epfn;
197 	}
198 
199 	return 0;
200 }
201 #else /* CONFIG_NUMA */
202 static int __init mark_nonram_nosave(void)
203 {
204 	return 0;
205 }
206 #endif
207 
208 /*
209  * Zones usage:
210  *
211  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
212  * everything else. GFP_DMA32 page allocations automatically fall back to
213  * ZONE_DMA.
214  *
215  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
216  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
217  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
218  * ZONE_DMA.
219  */
220 static unsigned long max_zone_pfns[MAX_NR_ZONES];
221 
222 /*
223  * paging_init() sets up the page tables - in fact we've already done this.
224  */
225 void __init paging_init(void)
226 {
227 	unsigned long long total_ram = memblock_phys_mem_size();
228 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
229 
230 #ifdef CONFIG_HIGHMEM
231 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
232 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
233 
234 	for (; v < end; v += PAGE_SIZE)
235 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
236 
237 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
238 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
239 #endif /* CONFIG_HIGHMEM */
240 
241 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
242 	       (unsigned long long)top_of_ram, total_ram);
243 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
244 	       (long int)((top_of_ram - total_ram) >> 20));
245 
246 	/*
247 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
248 	 * powerbooks.
249 	 */
250 	if (IS_ENABLED(CONFIG_PPC32))
251 		zone_dma_bits = 30;
252 	else
253 		zone_dma_bits = 31;
254 
255 #ifdef CONFIG_ZONE_DMA
256 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
257 				      1UL << (zone_dma_bits - PAGE_SHIFT));
258 #endif
259 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
260 #ifdef CONFIG_HIGHMEM
261 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
262 #endif
263 
264 	free_area_init(max_zone_pfns);
265 
266 	mark_nonram_nosave();
267 }
268 
269 void __init mem_init(void)
270 {
271 	/*
272 	 * book3s is limited to 16 page sizes due to encoding this in
273 	 * a 4-bit field for slices.
274 	 */
275 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
276 
277 #ifdef CONFIG_SWIOTLB
278 	/*
279 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
280 	 * 4G. We force memblock to bottom-up mode to ensure that the
281 	 * memory allocated in swiotlb_init() is DMA-able.
282 	 * As it's the last memblock allocation, no need to reset it
283 	 * back to to-down.
284 	 */
285 	memblock_set_bottom_up(true);
286 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
287 #endif
288 
289 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
290 	set_max_mapnr(max_pfn);
291 
292 	kasan_late_init();
293 
294 	memblock_free_all();
295 
296 #ifdef CONFIG_HIGHMEM
297 	{
298 		unsigned long pfn, highmem_mapnr;
299 
300 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
301 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
302 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
303 			struct page *page = pfn_to_page(pfn);
304 			if (!memblock_is_reserved(paddr))
305 				free_highmem_page(page);
306 		}
307 	}
308 #endif /* CONFIG_HIGHMEM */
309 
310 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
311 	/*
312 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
313 	 * functions.... do it here for the non-smp case.
314 	 */
315 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
316 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
317 #endif
318 
319 #ifdef CONFIG_PPC32
320 	pr_info("Kernel virtual memory layout:\n");
321 #ifdef CONFIG_KASAN
322 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
323 		KASAN_SHADOW_START, KASAN_SHADOW_END);
324 #endif
325 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
326 #ifdef CONFIG_HIGHMEM
327 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
328 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
329 #endif /* CONFIG_HIGHMEM */
330 	if (ioremap_bot != IOREMAP_TOP)
331 		pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
332 			ioremap_bot, IOREMAP_TOP);
333 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
334 		VMALLOC_START, VMALLOC_END);
335 #ifdef MODULES_VADDR
336 	pr_info("  * 0x%08lx..0x%08lx  : modules\n",
337 		MODULES_VADDR, MODULES_END);
338 #endif
339 #endif /* CONFIG_PPC32 */
340 }
341 
342 void free_initmem(void)
343 {
344 	ppc_md.progress = ppc_printk_progress;
345 	mark_initmem_nx();
346 	static_branch_enable(&init_mem_is_free);
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