xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision d9d193de)
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 int __ref arch_add_memory(int nid, u64 start, u64 size,
109 			  struct mhp_params *params)
110 {
111 	unsigned long start_pfn = start >> PAGE_SHIFT;
112 	unsigned long nr_pages = size >> PAGE_SHIFT;
113 	int rc;
114 
115 	rc = arch_create_linear_mapping(nid, start, size, params);
116 	if (rc)
117 		return rc;
118 	rc = __add_pages(nid, start_pfn, nr_pages, params);
119 	if (rc)
120 		arch_remove_linear_mapping(start, size);
121 	return rc;
122 }
123 
124 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
125 {
126 	unsigned long start_pfn = start >> PAGE_SHIFT;
127 	unsigned long nr_pages = size >> PAGE_SHIFT;
128 
129 	__remove_pages(start_pfn, nr_pages, altmap);
130 	arch_remove_linear_mapping(start, size);
131 }
132 #endif
133 
134 #ifndef CONFIG_NUMA
135 void __init mem_topology_setup(void)
136 {
137 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
138 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
139 #ifdef CONFIG_HIGHMEM
140 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
141 #endif
142 
143 	/* Place all memblock_regions in the same node and merge contiguous
144 	 * memblock_regions
145 	 */
146 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
147 }
148 
149 void __init initmem_init(void)
150 {
151 	sparse_init();
152 }
153 
154 /* mark pages that don't exist as nosave */
155 static int __init mark_nonram_nosave(void)
156 {
157 	unsigned long spfn, epfn, prev = 0;
158 	int i;
159 
160 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
161 		if (prev && prev < spfn)
162 			register_nosave_region(prev, spfn);
163 
164 		prev = epfn;
165 	}
166 
167 	return 0;
168 }
169 #else /* CONFIG_NUMA */
170 static int __init mark_nonram_nosave(void)
171 {
172 	return 0;
173 }
174 #endif
175 
176 /*
177  * Zones usage:
178  *
179  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
180  * everything else. GFP_DMA32 page allocations automatically fall back to
181  * ZONE_DMA.
182  *
183  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
184  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
185  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
186  * ZONE_DMA.
187  */
188 static unsigned long max_zone_pfns[MAX_NR_ZONES];
189 
190 /*
191  * paging_init() sets up the page tables - in fact we've already done this.
192  */
193 void __init paging_init(void)
194 {
195 	unsigned long long total_ram = memblock_phys_mem_size();
196 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
197 
198 #ifdef CONFIG_HIGHMEM
199 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
200 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
201 
202 	for (; v < end; v += PAGE_SIZE)
203 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
204 
205 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
206 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
207 #endif /* CONFIG_HIGHMEM */
208 
209 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
210 	       (unsigned long long)top_of_ram, total_ram);
211 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
212 	       (long int)((top_of_ram - total_ram) >> 20));
213 
214 	/*
215 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
216 	 * powerbooks.
217 	 */
218 	if (IS_ENABLED(CONFIG_PPC32))
219 		zone_dma_bits = 30;
220 	else
221 		zone_dma_bits = 31;
222 
223 #ifdef CONFIG_ZONE_DMA
224 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
225 				      1UL << (zone_dma_bits - PAGE_SHIFT));
226 #endif
227 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
228 #ifdef CONFIG_HIGHMEM
229 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
230 #endif
231 
232 	free_area_init(max_zone_pfns);
233 
234 	mark_nonram_nosave();
235 }
236 
237 void __init mem_init(void)
238 {
239 	/*
240 	 * book3s is limited to 16 page sizes due to encoding this in
241 	 * a 4-bit field for slices.
242 	 */
243 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
244 
245 #ifdef CONFIG_SWIOTLB
246 	/*
247 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
248 	 * 4G. We force memblock to bottom-up mode to ensure that the
249 	 * memory allocated in swiotlb_init() is DMA-able.
250 	 * As it's the last memblock allocation, no need to reset it
251 	 * back to to-down.
252 	 */
253 	memblock_set_bottom_up(true);
254 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
255 #endif
256 
257 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
258 	set_max_mapnr(max_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 	static_branch_enable(&init_mem_is_free);
315 	free_initmem_default(POISON_FREE_INITMEM);
316 	ftrace_free_init_tramp();
317 }
318 
319 /*
320  * System memory should not be in /proc/iomem but various tools expect it
321  * (eg kdump).
322  */
323 static int __init add_system_ram_resources(void)
324 {
325 	phys_addr_t start, end;
326 	u64 i;
327 
328 	for_each_mem_range(i, &start, &end) {
329 		struct resource *res;
330 
331 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
332 		WARN_ON(!res);
333 
334 		if (res) {
335 			res->name = "System RAM";
336 			res->start = start;
337 			/*
338 			 * In memblock, end points to the first byte after
339 			 * the range while in resourses, end points to the
340 			 * last byte in the range.
341 			 */
342 			res->end = end - 1;
343 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
344 			WARN_ON(request_resource(&iomem_resource, res) < 0);
345 		}
346 	}
347 
348 	return 0;
349 }
350 subsys_initcall(add_system_ram_resources);
351 
352 #ifdef CONFIG_STRICT_DEVMEM
353 /*
354  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
355  * is valid. The argument is a physical page number.
356  *
357  * Access has to be given to non-kernel-ram areas as well, these contain the
358  * PCI mmio resources as well as potential bios/acpi data regions.
359  */
360 int devmem_is_allowed(unsigned long pfn)
361 {
362 	if (page_is_rtas_user_buf(pfn))
363 		return 1;
364 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
365 		return 0;
366 	if (!page_is_ram(pfn))
367 		return 1;
368 	return 0;
369 }
370 #endif /* CONFIG_STRICT_DEVMEM */
371 
372 /*
373  * This is defined in kernel/resource.c but only powerpc needs to export it, for
374  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
375  */
376 EXPORT_SYMBOL_GPL(walk_system_ram_range);
377