xref: /linux/arch/powerpc/mm/mem.c (revision 4f3df2e5ea69f5717d2721922aff263c31957548)
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 #include <linux/execmem.h>
20 #include <linux/vmalloc.h>
21 
22 #include <asm/swiotlb.h>
23 #include <asm/machdep.h>
24 #include <asm/rtas.h>
25 #include <asm/kasan.h>
26 #include <asm/svm.h>
27 #include <asm/mmzone.h>
28 #include <asm/ftrace.h>
29 #include <asm/text-patching.h>
30 #include <asm/setup.h>
31 #include <asm/fixmap.h>
32 
33 #include <asm/fadump.h>
34 #include <asm/kexec.h>
35 #include <asm/kvm_ppc.h>
36 
37 #include <mm/mmu_decl.h>
38 
39 unsigned long long memory_limit __initdata;
40 
41 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
42 EXPORT_SYMBOL(empty_zero_page);
43 
__phys_mem_access_prot(unsigned long pfn,unsigned long size,pgprot_t vma_prot)44 pgprot_t __phys_mem_access_prot(unsigned long pfn, unsigned long size,
45 				pgprot_t vma_prot)
46 {
47 	if (ppc_md.phys_mem_access_prot)
48 		return ppc_md.phys_mem_access_prot(pfn, size, vma_prot);
49 
50 	if (!page_is_ram(pfn))
51 		vma_prot = pgprot_noncached(vma_prot);
52 
53 	return vma_prot;
54 }
55 EXPORT_SYMBOL(__phys_mem_access_prot);
56 
57 #ifdef CONFIG_MEMORY_HOTPLUG
58 static DEFINE_MUTEX(linear_mapping_mutex);
59 
60 #ifdef CONFIG_NUMA
memory_add_physaddr_to_nid(u64 start)61 int memory_add_physaddr_to_nid(u64 start)
62 {
63 	return hot_add_scn_to_nid(start);
64 }
65 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
66 #endif
67 
create_section_mapping(unsigned long start,unsigned long end,int nid,pgprot_t prot)68 int __weak create_section_mapping(unsigned long start, unsigned long end,
69 				  int nid, pgprot_t prot)
70 {
71 	return -ENODEV;
72 }
73 
remove_section_mapping(unsigned long start,unsigned long end)74 int __weak remove_section_mapping(unsigned long start, unsigned long end)
75 {
76 	return -ENODEV;
77 }
78 
arch_create_linear_mapping(int nid,u64 start,u64 size,struct mhp_params * params)79 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size,
80 				     struct mhp_params *params)
81 {
82 	int rc;
83 
84 	start = (unsigned long)__va(start);
85 	mutex_lock(&linear_mapping_mutex);
86 	rc = create_section_mapping(start, start + size, nid,
87 				    params->pgprot);
88 	mutex_unlock(&linear_mapping_mutex);
89 	if (rc) {
90 		pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n",
91 			start, start + size, rc);
92 		return -EFAULT;
93 	}
94 	return 0;
95 }
96 
arch_remove_linear_mapping(u64 start,u64 size)97 void __ref arch_remove_linear_mapping(u64 start, u64 size)
98 {
99 	int ret;
100 
101 	/* Remove htab bolted mappings for this section of memory */
102 	start = (unsigned long)__va(start);
103 
104 	mutex_lock(&linear_mapping_mutex);
105 	ret = remove_section_mapping(start, start + size);
106 	mutex_unlock(&linear_mapping_mutex);
107 	if (ret)
108 		pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n",
109 			start, start + size, ret);
110 
111 	/* Ensure all vmalloc mappings are flushed in case they also
112 	 * hit that section of memory
113 	 */
114 	vm_unmap_aliases();
115 }
116 
117 /*
118  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
119  * updating.
120  */
update_end_of_memory_vars(u64 start,u64 size)121 static void update_end_of_memory_vars(u64 start, u64 size)
122 {
123 	unsigned long end_pfn = PFN_UP(start + size);
124 
125 	if (end_pfn > max_pfn) {
126 		max_pfn = end_pfn;
127 		max_low_pfn = end_pfn;
128 		high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
129 	}
130 }
131 
add_pages(int nid,unsigned long start_pfn,unsigned long nr_pages,struct mhp_params * params)132 int __ref add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
133 		    struct mhp_params *params)
134 {
135 	int ret;
136 
137 	ret = __add_pages(nid, start_pfn, nr_pages, params);
138 	if (ret)
139 		return ret;
140 
141 	/* update max_pfn, max_low_pfn and high_memory */
142 	update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
143 				  nr_pages << PAGE_SHIFT);
144 
145 	return ret;
146 }
147 
arch_add_memory(int nid,u64 start,u64 size,struct mhp_params * params)148 int __ref arch_add_memory(int nid, u64 start, u64 size,
149 			  struct mhp_params *params)
150 {
151 	unsigned long start_pfn = start >> PAGE_SHIFT;
152 	unsigned long nr_pages = size >> PAGE_SHIFT;
153 	int rc;
154 
155 	rc = arch_create_linear_mapping(nid, start, size, params);
156 	if (rc)
157 		return rc;
158 	rc = add_pages(nid, start_pfn, nr_pages, params);
159 	if (rc)
160 		arch_remove_linear_mapping(start, size);
161 	return rc;
162 }
163 
arch_remove_memory(u64 start,u64 size,struct vmem_altmap * altmap)164 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
165 {
166 	unsigned long start_pfn = start >> PAGE_SHIFT;
167 	unsigned long nr_pages = size >> PAGE_SHIFT;
168 
169 	__remove_pages(start_pfn, nr_pages, altmap);
170 	arch_remove_linear_mapping(start, size);
171 }
172 #endif
173 
174 #ifndef CONFIG_NUMA
mem_topology_setup(void)175 void __init mem_topology_setup(void)
176 {
177 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
178 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
179 #ifdef CONFIG_HIGHMEM
180 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
181 #endif
182 
183 	/* Place all memblock_regions in the same node and merge contiguous
184 	 * memblock_regions
185 	 */
186 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
187 }
188 
189 /* mark pages that don't exist as nosave */
mark_nonram_nosave(void)190 static int __init mark_nonram_nosave(void)
191 {
192 	unsigned long spfn, epfn, prev = 0;
193 	int i;
194 
195 	for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) {
196 		if (prev && prev < spfn)
197 			register_nosave_region(prev, spfn);
198 
199 		prev = epfn;
200 	}
201 
202 	return 0;
203 }
204 #else /* CONFIG_NUMA */
mark_nonram_nosave(void)205 static int __init mark_nonram_nosave(void)
206 {
207 	return 0;
208 }
209 #endif
210 
211 /*
212  * Zones usage:
213  *
214  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
215  * everything else. GFP_DMA32 page allocations automatically fall back to
216  * ZONE_DMA.
217  *
218  * By using 31-bit unconditionally, we can exploit zone_dma_limit to inform the
219  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
220  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
221  * ZONE_DMA.
222  */
arch_zone_limits_init(unsigned long * max_zone_pfns)223 void __init arch_zone_limits_init(unsigned long *max_zone_pfns)
224 {
225 #ifdef CONFIG_ZONE_DMA
226 	max_zone_pfns[ZONE_DMA] = min((zone_dma_limit >> PAGE_SHIFT) + 1, max_low_pfn);
227 #endif
228 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
229 #ifdef CONFIG_HIGHMEM
230 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
231 #endif
232 }
233 
234 /*
235  * paging_init() sets up the page tables - in fact we've already done this.
236  */
paging_init(void)237 void __init paging_init(void)
238 {
239 	unsigned long long total_ram = memblock_phys_mem_size();
240 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
241 	int zone_dma_bits;
242 
243 #ifdef CONFIG_HIGHMEM
244 	unsigned long v = __fix_to_virt(FIX_KMAP_END);
245 	unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN);
246 
247 	for (; v < end; v += PAGE_SIZE)
248 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
249 
250 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
251 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
252 #endif /* CONFIG_HIGHMEM */
253 
254 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
255 	       (unsigned long long)top_of_ram, total_ram);
256 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
257 	       (long int)((top_of_ram - total_ram) >> 20));
258 
259 	/*
260 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
261 	 * powerbooks.
262 	 */
263 	if (IS_ENABLED(CONFIG_PPC32))
264 		zone_dma_bits = 30;
265 	else
266 		zone_dma_bits = 31;
267 
268 	zone_dma_limit = DMA_BIT_MASK(zone_dma_bits);
269 
270 	mark_nonram_nosave();
271 }
272 
arch_mm_preinit(void)273 void __init arch_mm_preinit(void)
274 {
275 
276 	/*
277 	 * Reserve large chunks of memory for use by CMA for kdump, fadump, KVM
278 	 * and hugetlb. These must be called after pageblock_order is
279 	 * initialised.
280 	 */
281 	fadump_cma_init();
282 	kdump_cma_reserve();
283 	kvm_cma_reserve();
284 
285 	/*
286 	 * book3s is limited to 16 page sizes due to encoding this in
287 	 * a 4-bit field for slices.
288 	 */
289 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
290 
291 #ifdef CONFIG_SWIOTLB
292 	/*
293 	 * Some platforms (e.g. 85xx) limit DMA-able memory way below
294 	 * 4G. We force memblock to bottom-up mode to ensure that the
295 	 * memory allocated in swiotlb_init() is DMA-able.
296 	 * As it's the last memblock allocation, no need to reset it
297 	 * back to to-down.
298 	 */
299 	memblock_set_bottom_up(true);
300 	swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags);
301 #endif
302 
303 	kasan_late_init();
304 
305 #if defined(CONFIG_PPC_E500) && !defined(CONFIG_SMP)
306 	/*
307 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
308 	 * functions.... do it here for the non-smp case.
309 	 */
310 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
311 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
312 #endif
313 }
314 
free_initmem(void)315 void free_initmem(void)
316 {
317 	ppc_md.progress = ppc_printk_progress;
318 	mark_initmem_nx();
319 	free_initmem_default(POISON_FREE_INITMEM);
320 	ftrace_free_init_tramp();
321 }
322 
323 /*
324  * System memory should not be in /proc/iomem but various tools expect it
325  * (eg kdump).
326  */
add_system_ram_resources(void)327 static int __init add_system_ram_resources(void)
328 {
329 	phys_addr_t start, end;
330 	u64 i;
331 
332 	for_each_mem_range(i, &start, &end) {
333 		struct resource *res;
334 
335 		res = kzalloc_obj(struct resource);
336 		WARN_ON(!res);
337 
338 		if (res) {
339 			res->name = "System RAM";
340 			res->start = start;
341 			/*
342 			 * In memblock, end points to the first byte after
343 			 * the range while in resourses, end points to the
344 			 * last byte in the range.
345 			 */
346 			res->end = end - 1;
347 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
348 			WARN_ON(insert_resource(&iomem_resource, res) < 0);
349 		}
350 	}
351 
352 	return 0;
353 }
354 subsys_initcall(add_system_ram_resources);
355 
356 #ifdef CONFIG_STRICT_DEVMEM
357 /*
358  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
359  * is valid. The argument is a physical page number.
360  *
361  * Access has to be given to non-kernel-ram areas as well, these contain the
362  * PCI mmio resources as well as potential bios/acpi data regions.
363  */
devmem_is_allowed(unsigned long pfn)364 int devmem_is_allowed(unsigned long pfn)
365 {
366 	if (page_is_rtas_user_buf(pfn))
367 		return 1;
368 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
369 		return 0;
370 	if (!page_is_ram(pfn))
371 		return 1;
372 	return 0;
373 }
374 #endif /* CONFIG_STRICT_DEVMEM */
375 
376 /*
377  * This is defined in kernel/resource.c but only powerpc needs to export it, for
378  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
379  */
380 EXPORT_SYMBOL_GPL(walk_system_ram_range);
381 
382 #ifdef CONFIG_EXECMEM
383 static struct execmem_info execmem_info __ro_after_init;
384 
385 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC_BOOK3S_603)
prealloc_execmem_pgtable(void)386 static void prealloc_execmem_pgtable(void)
387 {
388 	unsigned long va;
389 
390 	for (va = ALIGN_DOWN(MODULES_VADDR, PGDIR_SIZE); va < MODULES_END; va += PGDIR_SIZE)
391 		pte_alloc_kernel(pmd_off_k(va), va);
392 }
393 #else
prealloc_execmem_pgtable(void)394 static void prealloc_execmem_pgtable(void) { }
395 #endif
396 
execmem_arch_setup(void)397 struct execmem_info __init *execmem_arch_setup(void)
398 {
399 	pgprot_t kprobes_prot = strict_module_rwx_enabled() ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
400 	pgprot_t prot = strict_module_rwx_enabled() ? PAGE_KERNEL : PAGE_KERNEL_EXEC;
401 	unsigned long fallback_start = 0, fallback_end = 0;
402 	unsigned long start, end;
403 
404 	/*
405 	 * BOOK3S_32 and 8xx define MODULES_VADDR for text allocations and
406 	 * allow allocating data in the entire vmalloc space
407 	 */
408 #ifdef MODULES_VADDR
409 	unsigned long limit = (unsigned long)_etext - SZ_32M;
410 
411 	/* First try within 32M limit from _etext to avoid branch trampolines */
412 	if (MODULES_VADDR < PAGE_OFFSET && MODULES_END > limit) {
413 		start = limit;
414 		fallback_start = MODULES_VADDR;
415 		fallback_end = MODULES_END;
416 	} else {
417 		start = MODULES_VADDR;
418 	}
419 
420 	end = MODULES_END;
421 #else
422 	start = VMALLOC_START;
423 	end = VMALLOC_END;
424 #endif
425 
426 	prealloc_execmem_pgtable();
427 
428 	execmem_info = (struct execmem_info){
429 		.ranges = {
430 			[EXECMEM_DEFAULT] = {
431 				.start	= start,
432 				.end	= end,
433 				.pgprot	= prot,
434 				.alignment = 1,
435 				.fallback_start	= fallback_start,
436 				.fallback_end	= fallback_end,
437 			},
438 			[EXECMEM_KPROBES] = {
439 				.start	= VMALLOC_START,
440 				.end	= VMALLOC_END,
441 				.pgprot	= kprobes_prot,
442 				.alignment = 1,
443 			},
444 			[EXECMEM_MODULE_DATA] = {
445 				.start	= VMALLOC_START,
446 				.end	= VMALLOC_END,
447 				.pgprot	= PAGE_KERNEL,
448 				.alignment = 1,
449 			},
450 		},
451 	};
452 
453 	return &execmem_info;
454 }
455 #endif /* CONFIG_EXECMEM */
456