xref: /linux/arch/powerpc/mm/pgtable_64.c (revision e3b9f1e81de2083f359bacd2a94bf1c024f2ede0)
1 /*
2  *  This file contains ioremap and related functions for 64-bit machines.
3  *
4  *  Derived from arch/ppc64/mm/init.c
5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6  *
7  *  Modifications by Paul Mackerras (PowerMac) (paulus@samba.org)
8  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
9  *    Copyright (C) 1996 Paul Mackerras
10  *
11  *  Derived from "arch/i386/mm/init.c"
12  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
13  *
14  *  Dave Engebretsen <engebret@us.ibm.com>
15  *      Rework for PPC64 port.
16  *
17  *  This program is free software; you can redistribute it and/or
18  *  modify it under the terms of the GNU General Public License
19  *  as published by the Free Software Foundation; either version
20  *  2 of the License, or (at your option) any later version.
21  *
22  */
23 
24 #include <linux/signal.h>
25 #include <linux/sched.h>
26 #include <linux/kernel.h>
27 #include <linux/errno.h>
28 #include <linux/string.h>
29 #include <linux/export.h>
30 #include <linux/types.h>
31 #include <linux/mman.h>
32 #include <linux/mm.h>
33 #include <linux/swap.h>
34 #include <linux/stddef.h>
35 #include <linux/vmalloc.h>
36 #include <linux/memblock.h>
37 #include <linux/slab.h>
38 #include <linux/hugetlb.h>
39 
40 #include <asm/pgalloc.h>
41 #include <asm/page.h>
42 #include <asm/prom.h>
43 #include <asm/io.h>
44 #include <asm/mmu_context.h>
45 #include <asm/pgtable.h>
46 #include <asm/mmu.h>
47 #include <asm/smp.h>
48 #include <asm/machdep.h>
49 #include <asm/tlb.h>
50 #include <asm/trace.h>
51 #include <asm/processor.h>
52 #include <asm/cputable.h>
53 #include <asm/sections.h>
54 #include <asm/firmware.h>
55 #include <asm/dma.h>
56 #include <asm/powernv.h>
57 
58 #include "mmu_decl.h"
59 
60 #ifdef CONFIG_PPC_BOOK3S_64
61 #if TASK_SIZE_USER64 > (1UL << (ESID_BITS + SID_SHIFT))
62 #error TASK_SIZE_USER64 exceeds user VSID range
63 #endif
64 #endif
65 
66 #ifdef CONFIG_PPC_BOOK3S_64
67 /*
68  * partition table and process table for ISA 3.0
69  */
70 struct prtb_entry *process_tb;
71 struct patb_entry *partition_tb;
72 /*
73  * page table size
74  */
75 unsigned long __pte_index_size;
76 EXPORT_SYMBOL(__pte_index_size);
77 unsigned long __pmd_index_size;
78 EXPORT_SYMBOL(__pmd_index_size);
79 unsigned long __pud_index_size;
80 EXPORT_SYMBOL(__pud_index_size);
81 unsigned long __pgd_index_size;
82 EXPORT_SYMBOL(__pgd_index_size);
83 unsigned long __pmd_cache_index;
84 EXPORT_SYMBOL(__pmd_cache_index);
85 unsigned long __pud_cache_index;
86 EXPORT_SYMBOL(__pud_cache_index);
87 unsigned long __pte_table_size;
88 EXPORT_SYMBOL(__pte_table_size);
89 unsigned long __pmd_table_size;
90 EXPORT_SYMBOL(__pmd_table_size);
91 unsigned long __pud_table_size;
92 EXPORT_SYMBOL(__pud_table_size);
93 unsigned long __pgd_table_size;
94 EXPORT_SYMBOL(__pgd_table_size);
95 unsigned long __pmd_val_bits;
96 EXPORT_SYMBOL(__pmd_val_bits);
97 unsigned long __pud_val_bits;
98 EXPORT_SYMBOL(__pud_val_bits);
99 unsigned long __pgd_val_bits;
100 EXPORT_SYMBOL(__pgd_val_bits);
101 unsigned long __kernel_virt_start;
102 EXPORT_SYMBOL(__kernel_virt_start);
103 unsigned long __kernel_virt_size;
104 EXPORT_SYMBOL(__kernel_virt_size);
105 unsigned long __vmalloc_start;
106 EXPORT_SYMBOL(__vmalloc_start);
107 unsigned long __vmalloc_end;
108 EXPORT_SYMBOL(__vmalloc_end);
109 unsigned long __kernel_io_start;
110 EXPORT_SYMBOL(__kernel_io_start);
111 struct page *vmemmap;
112 EXPORT_SYMBOL(vmemmap);
113 unsigned long __pte_frag_nr;
114 EXPORT_SYMBOL(__pte_frag_nr);
115 unsigned long __pte_frag_size_shift;
116 EXPORT_SYMBOL(__pte_frag_size_shift);
117 unsigned long ioremap_bot;
118 #else /* !CONFIG_PPC_BOOK3S_64 */
119 unsigned long ioremap_bot = IOREMAP_BASE;
120 #endif
121 
122 /**
123  * __ioremap_at - Low level function to establish the page tables
124  *                for an IO mapping
125  */
126 void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size,
127 			    unsigned long flags)
128 {
129 	unsigned long i;
130 
131 	/* Make sure we have the base flags */
132 	if ((flags & _PAGE_PRESENT) == 0)
133 		flags |= pgprot_val(PAGE_KERNEL);
134 
135 	/* We don't support the 4K PFN hack with ioremap */
136 	if (flags & H_PAGE_4K_PFN)
137 		return NULL;
138 
139 	WARN_ON(pa & ~PAGE_MASK);
140 	WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
141 	WARN_ON(size & ~PAGE_MASK);
142 
143 	for (i = 0; i < size; i += PAGE_SIZE)
144 		if (map_kernel_page((unsigned long)ea+i, pa+i, flags))
145 			return NULL;
146 
147 	return (void __iomem *)ea;
148 }
149 
150 /**
151  * __iounmap_from - Low level function to tear down the page tables
152  *                  for an IO mapping. This is used for mappings that
153  *                  are manipulated manually, like partial unmapping of
154  *                  PCI IOs or ISA space.
155  */
156 void __iounmap_at(void *ea, unsigned long size)
157 {
158 	WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
159 	WARN_ON(size & ~PAGE_MASK);
160 
161 	unmap_kernel_range((unsigned long)ea, size);
162 }
163 
164 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
165 				unsigned long flags, void *caller)
166 {
167 	phys_addr_t paligned;
168 	void __iomem *ret;
169 
170 	/*
171 	 * Choose an address to map it to.
172 	 * Once the imalloc system is running, we use it.
173 	 * Before that, we map using addresses going
174 	 * up from ioremap_bot.  imalloc will use
175 	 * the addresses from ioremap_bot through
176 	 * IMALLOC_END
177 	 *
178 	 */
179 	paligned = addr & PAGE_MASK;
180 	size = PAGE_ALIGN(addr + size) - paligned;
181 
182 	if ((size == 0) || (paligned == 0))
183 		return NULL;
184 
185 	if (slab_is_available()) {
186 		struct vm_struct *area;
187 
188 		area = __get_vm_area_caller(size, VM_IOREMAP,
189 					    ioremap_bot, IOREMAP_END,
190 					    caller);
191 		if (area == NULL)
192 			return NULL;
193 
194 		area->phys_addr = paligned;
195 		ret = __ioremap_at(paligned, area->addr, size, flags);
196 		if (!ret)
197 			vunmap(area->addr);
198 	} else {
199 		ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags);
200 		if (ret)
201 			ioremap_bot += size;
202 	}
203 
204 	if (ret)
205 		ret += addr & ~PAGE_MASK;
206 	return ret;
207 }
208 
209 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
210 			 unsigned long flags)
211 {
212 	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
213 }
214 
215 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
216 {
217 	unsigned long flags = pgprot_val(pgprot_noncached(__pgprot(0)));
218 	void *caller = __builtin_return_address(0);
219 
220 	if (ppc_md.ioremap)
221 		return ppc_md.ioremap(addr, size, flags, caller);
222 	return __ioremap_caller(addr, size, flags, caller);
223 }
224 
225 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
226 {
227 	unsigned long flags = pgprot_val(pgprot_noncached_wc(__pgprot(0)));
228 	void *caller = __builtin_return_address(0);
229 
230 	if (ppc_md.ioremap)
231 		return ppc_md.ioremap(addr, size, flags, caller);
232 	return __ioremap_caller(addr, size, flags, caller);
233 }
234 
235 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
236 			     unsigned long flags)
237 {
238 	void *caller = __builtin_return_address(0);
239 
240 	/* writeable implies dirty for kernel addresses */
241 	if (flags & _PAGE_WRITE)
242 		flags |= _PAGE_DIRTY;
243 
244 	/* we don't want to let _PAGE_EXEC leak out */
245 	flags &= ~_PAGE_EXEC;
246 	/*
247 	 * Force kernel mapping.
248 	 */
249 	flags &= ~_PAGE_USER;
250 	flags |= _PAGE_PRIVILEGED;
251 
252 	if (ppc_md.ioremap)
253 		return ppc_md.ioremap(addr, size, flags, caller);
254 	return __ioremap_caller(addr, size, flags, caller);
255 }
256 
257 
258 /*
259  * Unmap an IO region and remove it from imalloc'd list.
260  * Access to IO memory should be serialized by driver.
261  */
262 void __iounmap(volatile void __iomem *token)
263 {
264 	void *addr;
265 
266 	if (!slab_is_available())
267 		return;
268 
269 	addr = (void *) ((unsigned long __force)
270 			 PCI_FIX_ADDR(token) & PAGE_MASK);
271 	if ((unsigned long)addr < ioremap_bot) {
272 		printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
273 		       " at 0x%p\n", addr);
274 		return;
275 	}
276 	vunmap(addr);
277 }
278 
279 void iounmap(volatile void __iomem *token)
280 {
281 	if (ppc_md.iounmap)
282 		ppc_md.iounmap(token);
283 	else
284 		__iounmap(token);
285 }
286 
287 EXPORT_SYMBOL(ioremap);
288 EXPORT_SYMBOL(ioremap_wc);
289 EXPORT_SYMBOL(ioremap_prot);
290 EXPORT_SYMBOL(__ioremap);
291 EXPORT_SYMBOL(__ioremap_at);
292 EXPORT_SYMBOL(iounmap);
293 EXPORT_SYMBOL(__iounmap);
294 EXPORT_SYMBOL(__iounmap_at);
295 
296 #ifndef __PAGETABLE_PUD_FOLDED
297 /* 4 level page table */
298 struct page *pgd_page(pgd_t pgd)
299 {
300 	if (pgd_huge(pgd))
301 		return pte_page(pgd_pte(pgd));
302 	return virt_to_page(pgd_page_vaddr(pgd));
303 }
304 #endif
305 
306 struct page *pud_page(pud_t pud)
307 {
308 	if (pud_huge(pud))
309 		return pte_page(pud_pte(pud));
310 	return virt_to_page(pud_page_vaddr(pud));
311 }
312 
313 /*
314  * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
315  * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
316  */
317 struct page *pmd_page(pmd_t pmd)
318 {
319 	if (pmd_trans_huge(pmd) || pmd_huge(pmd) || pmd_devmap(pmd))
320 		return pte_page(pmd_pte(pmd));
321 	return virt_to_page(pmd_page_vaddr(pmd));
322 }
323 
324 #ifdef CONFIG_PPC_64K_PAGES
325 static pte_t *get_from_cache(struct mm_struct *mm)
326 {
327 	void *pte_frag, *ret;
328 
329 	spin_lock(&mm->page_table_lock);
330 	ret = mm->context.pte_frag;
331 	if (ret) {
332 		pte_frag = ret + PTE_FRAG_SIZE;
333 		/*
334 		 * If we have taken up all the fragments mark PTE page NULL
335 		 */
336 		if (((unsigned long)pte_frag & ~PAGE_MASK) == 0)
337 			pte_frag = NULL;
338 		mm->context.pte_frag = pte_frag;
339 	}
340 	spin_unlock(&mm->page_table_lock);
341 	return (pte_t *)ret;
342 }
343 
344 static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel)
345 {
346 	void *ret = NULL;
347 	struct page *page;
348 
349 	if (!kernel) {
350 		page = alloc_page(PGALLOC_GFP | __GFP_ACCOUNT);
351 		if (!page)
352 			return NULL;
353 		if (!pgtable_page_ctor(page)) {
354 			__free_page(page);
355 			return NULL;
356 		}
357 	} else {
358 		page = alloc_page(PGALLOC_GFP);
359 		if (!page)
360 			return NULL;
361 	}
362 
363 	ret = page_address(page);
364 	spin_lock(&mm->page_table_lock);
365 	/*
366 	 * If we find pgtable_page set, we return
367 	 * the allocated page with single fragement
368 	 * count.
369 	 */
370 	if (likely(!mm->context.pte_frag)) {
371 		set_page_count(page, PTE_FRAG_NR);
372 		mm->context.pte_frag = ret + PTE_FRAG_SIZE;
373 	}
374 	spin_unlock(&mm->page_table_lock);
375 
376 	return (pte_t *)ret;
377 }
378 
379 pte_t *pte_fragment_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
380 {
381 	pte_t *pte;
382 
383 	pte = get_from_cache(mm);
384 	if (pte)
385 		return pte;
386 
387 	return __alloc_for_cache(mm, kernel);
388 }
389 #endif /* CONFIG_PPC_64K_PAGES */
390 
391 void pte_fragment_free(unsigned long *table, int kernel)
392 {
393 	struct page *page = virt_to_page(table);
394 	if (put_page_testzero(page)) {
395 		if (!kernel)
396 			pgtable_page_dtor(page);
397 		free_unref_page(page);
398 	}
399 }
400 
401 #ifdef CONFIG_SMP
402 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
403 {
404 	unsigned long pgf = (unsigned long)table;
405 
406 	BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
407 	pgf |= shift;
408 	tlb_remove_table(tlb, (void *)pgf);
409 }
410 
411 void __tlb_remove_table(void *_table)
412 {
413 	void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
414 	unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
415 
416 	if (!shift)
417 		/* PTE page needs special handling */
418 		pte_fragment_free(table, 0);
419 	else {
420 		BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
421 		kmem_cache_free(PGT_CACHE(shift), table);
422 	}
423 }
424 #else
425 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
426 {
427 	if (!shift) {
428 		/* PTE page needs special handling */
429 		pte_fragment_free(table, 0);
430 	} else {
431 		BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
432 		kmem_cache_free(PGT_CACHE(shift), table);
433 	}
434 }
435 #endif
436 
437 #ifdef CONFIG_PPC_BOOK3S_64
438 void __init mmu_partition_table_init(void)
439 {
440 	unsigned long patb_size = 1UL << PATB_SIZE_SHIFT;
441 	unsigned long ptcr;
442 
443 	BUILD_BUG_ON_MSG((PATB_SIZE_SHIFT > 36), "Partition table size too large.");
444 	partition_tb = __va(memblock_alloc_base(patb_size, patb_size,
445 						MEMBLOCK_ALLOC_ANYWHERE));
446 
447 	/* Initialize the Partition Table with no entries */
448 	memset((void *)partition_tb, 0, patb_size);
449 
450 	/*
451 	 * update partition table control register,
452 	 * 64 K size.
453 	 */
454 	ptcr = __pa(partition_tb) | (PATB_SIZE_SHIFT - 12);
455 	mtspr(SPRN_PTCR, ptcr);
456 	powernv_set_nmmu_ptcr(ptcr);
457 }
458 
459 void mmu_partition_table_set_entry(unsigned int lpid, unsigned long dw0,
460 				   unsigned long dw1)
461 {
462 	unsigned long old = be64_to_cpu(partition_tb[lpid].patb0);
463 
464 	partition_tb[lpid].patb0 = cpu_to_be64(dw0);
465 	partition_tb[lpid].patb1 = cpu_to_be64(dw1);
466 
467 	/*
468 	 * Global flush of TLBs and partition table caches for this lpid.
469 	 * The type of flush (hash or radix) depends on what the previous
470 	 * use of this partition ID was, not the new use.
471 	 */
472 	asm volatile("ptesync" : : : "memory");
473 	if (old & PATB_HR) {
474 		asm volatile(PPC_TLBIE_5(%0,%1,2,0,1) : :
475 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
476 		asm volatile(PPC_TLBIE_5(%0,%1,2,1,1) : :
477 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
478 		trace_tlbie(lpid, 0, TLBIEL_INVAL_SET_LPID, lpid, 2, 0, 1);
479 	} else {
480 		asm volatile(PPC_TLBIE_5(%0,%1,2,0,0) : :
481 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
482 		trace_tlbie(lpid, 0, TLBIEL_INVAL_SET_LPID, lpid, 2, 0, 0);
483 	}
484 	asm volatile("eieio; tlbsync; ptesync" : : : "memory");
485 }
486 EXPORT_SYMBOL_GPL(mmu_partition_table_set_entry);
487 #endif /* CONFIG_PPC_BOOK3S_64 */
488 
489 #ifdef CONFIG_STRICT_KERNEL_RWX
490 void mark_rodata_ro(void)
491 {
492 	if (!mmu_has_feature(MMU_FTR_KERNEL_RO)) {
493 		pr_warn("Warning: Unable to mark rodata read only on this CPU.\n");
494 		return;
495 	}
496 
497 	if (radix_enabled())
498 		radix__mark_rodata_ro();
499 	else
500 		hash__mark_rodata_ro();
501 }
502 
503 void mark_initmem_nx(void)
504 {
505 	if (radix_enabled())
506 		radix__mark_initmem_nx();
507 	else
508 		hash__mark_initmem_nx();
509 }
510 #endif
511