xref: /openbmc/linux/arch/s390/mm/init.c (revision be2b81b5)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  S390 version
4  *    Copyright IBM Corp. 1999
5  *    Author(s): Hartmut Penner (hp@de.ibm.com)
6  *
7  *  Derived from "arch/i386/mm/init.c"
8  *    Copyright (C) 1995  Linus Torvalds
9  */
10 
11 #include <linux/signal.h>
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/errno.h>
15 #include <linux/string.h>
16 #include <linux/types.h>
17 #include <linux/ptrace.h>
18 #include <linux/mman.h>
19 #include <linux/mm.h>
20 #include <linux/swap.h>
21 #include <linux/swiotlb.h>
22 #include <linux/smp.h>
23 #include <linux/init.h>
24 #include <linux/pagemap.h>
25 #include <linux/memblock.h>
26 #include <linux/memory.h>
27 #include <linux/pfn.h>
28 #include <linux/poison.h>
29 #include <linux/initrd.h>
30 #include <linux/export.h>
31 #include <linux/cma.h>
32 #include <linux/gfp.h>
33 #include <linux/dma-direct.h>
34 #include <linux/percpu.h>
35 #include <asm/processor.h>
36 #include <linux/uaccess.h>
37 #include <asm/pgalloc.h>
38 #include <asm/kfence.h>
39 #include <asm/ptdump.h>
40 #include <asm/dma.h>
41 #include <asm/abs_lowcore.h>
42 #include <asm/tlb.h>
43 #include <asm/tlbflush.h>
44 #include <asm/sections.h>
45 #include <asm/ctl_reg.h>
46 #include <asm/sclp.h>
47 #include <asm/set_memory.h>
48 #include <asm/kasan.h>
49 #include <asm/dma-mapping.h>
50 #include <asm/uv.h>
51 #include <linux/virtio_anchor.h>
52 #include <linux/virtio_config.h>
53 
54 pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".bss..swapper_pg_dir");
55 static pgd_t invalid_pg_dir[PTRS_PER_PGD] __section(".bss..invalid_pg_dir");
56 
57 unsigned long s390_invalid_asce;
58 
59 unsigned long empty_zero_page, zero_page_mask;
60 EXPORT_SYMBOL(empty_zero_page);
61 EXPORT_SYMBOL(zero_page_mask);
62 
63 static void __init setup_zero_pages(void)
64 {
65 	unsigned int order;
66 	struct page *page;
67 	int i;
68 
69 	/* Latest machines require a mapping granularity of 512KB */
70 	order = 7;
71 
72 	/* Limit number of empty zero pages for small memory sizes */
73 	while (order > 2 && (totalram_pages() >> 10) < (1UL << order))
74 		order--;
75 
76 	empty_zero_page = __get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
77 	if (!empty_zero_page)
78 		panic("Out of memory in setup_zero_pages");
79 
80 	page = virt_to_page((void *) empty_zero_page);
81 	split_page(page, order);
82 	for (i = 1 << order; i > 0; i--) {
83 		mark_page_reserved(page);
84 		page++;
85 	}
86 
87 	zero_page_mask = ((PAGE_SIZE << order) - 1) & PAGE_MASK;
88 }
89 
90 /*
91  * paging_init() sets up the page tables
92  */
93 void __init paging_init(void)
94 {
95 	unsigned long max_zone_pfns[MAX_NR_ZONES];
96 	unsigned long pgd_type, asce_bits;
97 	psw_t psw;
98 
99 	s390_invalid_asce  = (unsigned long)invalid_pg_dir;
100 	s390_invalid_asce |= _ASCE_TYPE_REGION3 | _ASCE_TABLE_LENGTH;
101 	crst_table_init((unsigned long *)invalid_pg_dir, _REGION3_ENTRY_EMPTY);
102 	init_mm.pgd = swapper_pg_dir;
103 	if (VMALLOC_END > _REGION2_SIZE) {
104 		asce_bits = _ASCE_TYPE_REGION2 | _ASCE_TABLE_LENGTH;
105 		pgd_type = _REGION2_ENTRY_EMPTY;
106 	} else {
107 		asce_bits = _ASCE_TYPE_REGION3 | _ASCE_TABLE_LENGTH;
108 		pgd_type = _REGION3_ENTRY_EMPTY;
109 	}
110 	init_mm.context.asce = (__pa(init_mm.pgd) & PAGE_MASK) | asce_bits;
111 	S390_lowcore.kernel_asce = init_mm.context.asce;
112 	S390_lowcore.user_asce = s390_invalid_asce;
113 	crst_table_init((unsigned long *) init_mm.pgd, pgd_type);
114 	vmem_map_init();
115 	kasan_copy_shadow_mapping();
116 
117 	/* enable virtual mapping in kernel mode */
118 	__ctl_load(S390_lowcore.kernel_asce, 1, 1);
119 	__ctl_load(S390_lowcore.user_asce, 7, 7);
120 	__ctl_load(S390_lowcore.kernel_asce, 13, 13);
121 	psw.mask = __extract_psw();
122 	psw_bits(psw).dat = 1;
123 	psw_bits(psw).as = PSW_BITS_AS_HOME;
124 	__load_psw_mask(psw.mask);
125 	kasan_free_early_identity();
126 
127 	sparse_init();
128 	zone_dma_bits = 31;
129 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
130 	max_zone_pfns[ZONE_DMA] = PFN_DOWN(MAX_DMA_ADDRESS);
131 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
132 	free_area_init(max_zone_pfns);
133 }
134 
135 void mark_rodata_ro(void)
136 {
137 	unsigned long size = __end_ro_after_init - __start_ro_after_init;
138 
139 	set_memory_ro((unsigned long)__start_ro_after_init, size >> PAGE_SHIFT);
140 	pr_info("Write protected read-only-after-init data: %luk\n", size >> 10);
141 	debug_checkwx();
142 }
143 
144 int set_memory_encrypted(unsigned long vaddr, int numpages)
145 {
146 	int i;
147 
148 	/* make specified pages unshared, (swiotlb, dma_free) */
149 	for (i = 0; i < numpages; ++i) {
150 		uv_remove_shared(virt_to_phys((void *)vaddr));
151 		vaddr += PAGE_SIZE;
152 	}
153 	return 0;
154 }
155 
156 int set_memory_decrypted(unsigned long vaddr, int numpages)
157 {
158 	int i;
159 	/* make specified pages shared (swiotlb, dma_alloca) */
160 	for (i = 0; i < numpages; ++i) {
161 		uv_set_shared(virt_to_phys((void *)vaddr));
162 		vaddr += PAGE_SIZE;
163 	}
164 	return 0;
165 }
166 
167 /* are we a protected virtualization guest? */
168 bool force_dma_unencrypted(struct device *dev)
169 {
170 	return is_prot_virt_guest();
171 }
172 
173 /* protected virtualization */
174 static void pv_init(void)
175 {
176 	if (!is_prot_virt_guest())
177 		return;
178 
179 	virtio_set_mem_acc_cb(virtio_require_restricted_mem_acc);
180 
181 	/* make sure bounce buffers are shared */
182 	swiotlb_init(true, SWIOTLB_FORCE | SWIOTLB_VERBOSE);
183 	swiotlb_update_mem_attributes();
184 }
185 
186 void __init mem_init(void)
187 {
188 	cpumask_set_cpu(0, &init_mm.context.cpu_attach_mask);
189 	cpumask_set_cpu(0, mm_cpumask(&init_mm));
190 
191 	set_max_mapnr(max_low_pfn);
192         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
193 
194 	pv_init();
195 	kfence_split_mapping();
196 	/* Setup guest page hinting */
197 	cmma_init();
198 
199 	/* this will put all low memory onto the freelists */
200 	memblock_free_all();
201 	setup_zero_pages();	/* Setup zeroed pages. */
202 
203 	cmma_init_nodat();
204 }
205 
206 void free_initmem(void)
207 {
208 	__set_memory((unsigned long)_sinittext,
209 		     (unsigned long)(_einittext - _sinittext) >> PAGE_SHIFT,
210 		     SET_MEMORY_RW | SET_MEMORY_NX);
211 	free_initmem_default(POISON_FREE_INITMEM);
212 }
213 
214 unsigned long memory_block_size_bytes(void)
215 {
216 	/*
217 	 * Make sure the memory block size is always greater
218 	 * or equal than the memory increment size.
219 	 */
220 	return max_t(unsigned long, MIN_MEMORY_BLOCK_SIZE, sclp.rzm);
221 }
222 
223 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
224 EXPORT_SYMBOL(__per_cpu_offset);
225 
226 static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
227 {
228 	return LOCAL_DISTANCE;
229 }
230 
231 static int __init pcpu_cpu_to_node(int cpu)
232 {
233 	return 0;
234 }
235 
236 void __init setup_per_cpu_areas(void)
237 {
238 	unsigned long delta;
239 	unsigned int cpu;
240 	int rc;
241 
242 	/*
243 	 * Always reserve area for module percpu variables.  That's
244 	 * what the legacy allocator did.
245 	 */
246 	rc = pcpu_embed_first_chunk(PERCPU_MODULE_RESERVE,
247 				    PERCPU_DYNAMIC_RESERVE, PAGE_SIZE,
248 				    pcpu_cpu_distance,
249 				    pcpu_cpu_to_node);
250 	if (rc < 0)
251 		panic("Failed to initialize percpu areas.");
252 
253 	delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
254 	for_each_possible_cpu(cpu)
255 		__per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
256 }
257 
258 #ifdef CONFIG_MEMORY_HOTPLUG
259 
260 #ifdef CONFIG_CMA
261 
262 /* Prevent memory blocks which contain cma regions from going offline */
263 
264 struct s390_cma_mem_data {
265 	unsigned long start;
266 	unsigned long end;
267 };
268 
269 static int s390_cma_check_range(struct cma *cma, void *data)
270 {
271 	struct s390_cma_mem_data *mem_data;
272 	unsigned long start, end;
273 
274 	mem_data = data;
275 	start = cma_get_base(cma);
276 	end = start + cma_get_size(cma);
277 	if (end < mem_data->start)
278 		return 0;
279 	if (start >= mem_data->end)
280 		return 0;
281 	return -EBUSY;
282 }
283 
284 static int s390_cma_mem_notifier(struct notifier_block *nb,
285 				 unsigned long action, void *data)
286 {
287 	struct s390_cma_mem_data mem_data;
288 	struct memory_notify *arg;
289 	int rc = 0;
290 
291 	arg = data;
292 	mem_data.start = arg->start_pfn << PAGE_SHIFT;
293 	mem_data.end = mem_data.start + (arg->nr_pages << PAGE_SHIFT);
294 	if (action == MEM_GOING_OFFLINE)
295 		rc = cma_for_each_area(s390_cma_check_range, &mem_data);
296 	return notifier_from_errno(rc);
297 }
298 
299 static struct notifier_block s390_cma_mem_nb = {
300 	.notifier_call = s390_cma_mem_notifier,
301 };
302 
303 static int __init s390_cma_mem_init(void)
304 {
305 	return register_memory_notifier(&s390_cma_mem_nb);
306 }
307 device_initcall(s390_cma_mem_init);
308 
309 #endif /* CONFIG_CMA */
310 
311 int arch_add_memory(int nid, u64 start, u64 size,
312 		    struct mhp_params *params)
313 {
314 	unsigned long start_pfn = PFN_DOWN(start);
315 	unsigned long size_pages = PFN_DOWN(size);
316 	int rc;
317 
318 	if (WARN_ON_ONCE(params->altmap))
319 		return -EINVAL;
320 
321 	if (WARN_ON_ONCE(params->pgprot.pgprot != PAGE_KERNEL.pgprot))
322 		return -EINVAL;
323 
324 	VM_BUG_ON(!mhp_range_allowed(start, size, true));
325 	rc = vmem_add_mapping(start, size);
326 	if (rc)
327 		return rc;
328 
329 	rc = __add_pages(nid, start_pfn, size_pages, params);
330 	if (rc)
331 		vmem_remove_mapping(start, size);
332 	return rc;
333 }
334 
335 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
336 {
337 	unsigned long start_pfn = start >> PAGE_SHIFT;
338 	unsigned long nr_pages = size >> PAGE_SHIFT;
339 
340 	__remove_pages(start_pfn, nr_pages, altmap);
341 	vmem_remove_mapping(start, size);
342 }
343 #endif /* CONFIG_MEMORY_HOTPLUG */
344