xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 19b438592238b3b40c3f945bb5f9c4ca971c0c45)
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/sparsemem.h>
24 #include <asm/svm.h>
25 
26 #include <mm/mmu_decl.h>
27 
28 unsigned long long memory_limit;
29 bool init_mem_is_free;
30 
31 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
32 			      unsigned long size, pgprot_t vma_prot)
33 {
34 	if (ppc_md.phys_mem_access_prot)
35 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
36 
37 	if (!page_is_ram(pfn))
38 		vma_prot = pgprot_noncached(vma_prot);
39 
40 	return vma_prot;
41 }
42 EXPORT_SYMBOL(phys_mem_access_prot);
43 
44 #ifdef CONFIG_MEMORY_HOTPLUG
45 static DEFINE_MUTEX(linear_mapping_mutex);
46 
47 #ifdef CONFIG_NUMA
48 int memory_add_physaddr_to_nid(u64 start)
49 {
50 	return hot_add_scn_to_nid(start);
51 }
52 #endif
53 
54 int __weak create_section_mapping(unsigned long start, unsigned long end,
55 				  int nid, pgprot_t prot)
56 {
57 	return -ENODEV;
58 }
59 
60 int __weak remove_section_mapping(unsigned long start, unsigned long end)
61 {
62 	return -ENODEV;
63 }
64 
65 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
66 				     struct mhp_params *params)
67 {
68 	int rc;
69 
70 	start = (unsigned long)__va(start);
71 	mutex_lock(&linear_mapping_mutex);
72 	rc = create_section_mapping(start, start + size, nid,
73 				    params->pgprot);
74 	mutex_unlock(&linear_mapping_mutex);
75 	if (rc) {
76 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
77 			start, start + size, rc);
78 		return -EFAULT;
79 	}
80 	return 0;
81 }
82 
83 void __ref arch_remove_linear_mapping(u64 start, u64 size)
84 {
85 	int ret;
86 
87 	/* Remove htab bolted mappings for this section of memory */
88 	start = (unsigned long)__va(start);
89 
90 	mutex_lock(&linear_mapping_mutex);
91 	ret = remove_section_mapping(start, start + size);
92 	mutex_unlock(&linear_mapping_mutex);
93 	if (ret)
94 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
95 			start, start + size, ret);
96 
97 	/* Ensure all vmalloc mappings are flushed in case they also
98 	 * hit that section of memory
99 	 */
100 	vm_unmap_aliases();
101 }
102 
103 int __ref arch_add_memory(int nid, u64 start, u64 size,
104 			  struct mhp_params *params)
105 {
106 	unsigned long start_pfn = start >> PAGE_SHIFT;
107 	unsigned long nr_pages = size >> PAGE_SHIFT;
108 	int rc;
109 
110 	rc = arch_create_linear_mapping(nid, start, size, params);
111 	if (rc)
112 		return rc;
113 	rc = __add_pages(nid, start_pfn, nr_pages, params);
114 	if (rc)
115 		arch_remove_linear_mapping(start, size);
116 	return rc;
117 }
118 
119 void __ref arch_remove_memory(int nid, u64 start, u64 size,
120 			      struct vmem_altmap *altmap)
121 {
122 	unsigned long start_pfn = start >> PAGE_SHIFT;
123 	unsigned long nr_pages = size >> PAGE_SHIFT;
124 
125 	__remove_pages(start_pfn, nr_pages, altmap);
126 	arch_remove_linear_mapping(start, size);
127 }
128 #endif
129 
130 #ifndef CONFIG_NUMA
131 void __init mem_topology_setup(void)
132 {
133 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
134 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
135 #ifdef CONFIG_HIGHMEM
136 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
137 #endif
138 
139 	/* Place all memblock_regions in the same node and merge contiguous
140 	 * memblock_regions
141 	 */
142 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
143 }
144 
145 void __init initmem_init(void)
146 {
147 	sparse_init();
148 }
149 
150 /* mark pages that don't exist as nosave */
151 static int __init mark_nonram_nosave(void)
152 {
153 	unsigned long spfn, epfn, prev = 0;
154 	int i;
155 
156 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
157 		if (prev && prev < spfn)
158 			register_nosave_region(prev, spfn);
159 
160 		prev = epfn;
161 	}
162 
163 	return 0;
164 }
165 #else /* CONFIG_NUMA */
166 static int __init mark_nonram_nosave(void)
167 {
168 	return 0;
169 }
170 #endif
171 
172 /*
173  * Zones usage:
174  *
175  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
176  * everything else. GFP_DMA32 page allocations automatically fall back to
177  * ZONE_DMA.
178  *
179  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
180  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
181  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
182  * ZONE_DMA.
183  */
184 static unsigned long max_zone_pfns[MAX_NR_ZONES];
185 
186 /*
187  * paging_init() sets up the page tables - in fact we've already done this.
188  */
189 void __init paging_init(void)
190 {
191 	unsigned long long total_ram = memblock_phys_mem_size();
192 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
193 
194 #ifdef CONFIG_HIGHMEM
195 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
196 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
197 
198 	for (; v < end; v += PAGE_SIZE)
199 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
200 
201 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
202 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
203 #endif /* CONFIG_HIGHMEM */
204 
205 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
206 	       (unsigned long long)top_of_ram, total_ram);
207 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
208 	       (long int)((top_of_ram - total_ram) >> 20));
209 
210 	/*
211 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
212 	 * powerbooks.
213 	 */
214 	if (IS_ENABLED(CONFIG_PPC32))
215 		zone_dma_bits = 30;
216 	else
217 		zone_dma_bits = 31;
218 
219 #ifdef CONFIG_ZONE_DMA
220 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
221 				      1UL << (zone_dma_bits - PAGE_SHIFT));
222 #endif
223 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
224 #ifdef CONFIG_HIGHMEM
225 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
226 #endif
227 
228 	free_area_init(max_zone_pfns);
229 
230 	mark_nonram_nosave();
231 }
232 
233 void __init mem_init(void)
234 {
235 	/*
236 	 * book3s is limited to 16 page sizes due to encoding this in
237 	 * a 4-bit field for slices.
238 	 */
239 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
240 
241 #ifdef CONFIG_SWIOTLB
242 	/*
243 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
244 	 * 4G. We force memblock to bottom-up mode to ensure that the
245 	 * memory allocated in swiotlb_init() is DMA-able.
246 	 * As it's the last memblock allocation, no need to reset it
247 	 * back to to-down.
248 	 */
249 	memblock_set_bottom_up(true);
250 	if (is_secure_guest())
251 		svm_swiotlb_init();
252 	else
253 		swiotlb_init(0);
254 #endif
255 
256 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
257 	set_max_mapnr(max_pfn);
258 
259 	kasan_late_init();
260 
261 	memblock_free_all();
262 
263 #ifdef CONFIG_HIGHMEM
264 	{
265 		unsigned long pfn, highmem_mapnr;
266 
267 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
268 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
269 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
270 			struct page *page = pfn_to_page(pfn);
271 			if (!memblock_is_reserved(paddr))
272 				free_highmem_page(page);
273 		}
274 	}
275 #endif /* CONFIG_HIGHMEM */
276 
277 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
278 	/*
279 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
280 	 * functions.... do it here for the non-smp case.
281 	 */
282 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
283 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
284 #endif
285 
286 #ifdef CONFIG_PPC32
287 	pr_info("Kernel virtual memory layout:\n");
288 #ifdef CONFIG_KASAN
289 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
290 		KASAN_SHADOW_START, KASAN_SHADOW_END);
291 #endif
292 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
293 #ifdef CONFIG_HIGHMEM
294 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
295 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
296 #endif /* CONFIG_HIGHMEM */
297 	if (ioremap_bot != IOREMAP_TOP)
298 		pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
299 			ioremap_bot, IOREMAP_TOP);
300 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
301 		VMALLOC_START, VMALLOC_END);
302 #endif /* CONFIG_PPC32 */
303 }
304 
305 void free_initmem(void)
306 {
307 	ppc_md.progress = ppc_printk_progress;
308 	mark_initmem_nx();
309 	init_mem_is_free = true;
310 	free_initmem_default(POISON_FREE_INITMEM);
311 }
312 
313 /*
314  * System memory should not be in /proc/iomem but various tools expect it
315  * (eg kdump).
316  */
317 static int __init add_system_ram_resources(void)
318 {
319 	phys_addr_t start, end;
320 	u64 i;
321 
322 	for_each_mem_range(i, &start, &end) {
323 		struct resource *res;
324 
325 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
326 		WARN_ON(!res);
327 
328 		if (res) {
329 			res->name = "System RAM";
330 			res->start = start;
331 			/*
332 			 * In memblock, end points to the first byte after
333 			 * the range while in resourses, end points to the
334 			 * last byte in the range.
335 			 */
336 			res->end = end - 1;
337 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
338 			WARN_ON(request_resource(&iomem_resource, res) < 0);
339 		}
340 	}
341 
342 	return 0;
343 }
344 subsys_initcall(add_system_ram_resources);
345 
346 #ifdef CONFIG_STRICT_DEVMEM
347 /*
348  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
349  * is valid. The argument is a physical page number.
350  *
351  * Access has to be given to non-kernel-ram areas as well, these contain the
352  * PCI mmio resources as well as potential bios/acpi data regions.
353  */
354 int devmem_is_allowed(unsigned long pfn)
355 {
356 	if (page_is_rtas_user_buf(pfn))
357 		return 1;
358 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
359 		return 0;
360 	if (!page_is_ram(pfn))
361 		return 1;
362 	return 0;
363 }
364 #endif /* CONFIG_STRICT_DEVMEM */
365 
366 /*
367  * This is defined in kernel/resource.c but only powerpc needs to export it, for
368  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
369  */
370 EXPORT_SYMBOL_GPL(walk_system_ram_range);
371