xref: /openbmc/linux/arch/powerpc/mm/pgtable_64.c (revision 2359ccdd)
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 
61 #ifdef CONFIG_PPC_BOOK3S_64
62 /*
63  * partition table and process table for ISA 3.0
64  */
65 struct prtb_entry *process_tb;
66 struct patb_entry *partition_tb;
67 /*
68  * page table size
69  */
70 unsigned long __pte_index_size;
71 EXPORT_SYMBOL(__pte_index_size);
72 unsigned long __pmd_index_size;
73 EXPORT_SYMBOL(__pmd_index_size);
74 unsigned long __pud_index_size;
75 EXPORT_SYMBOL(__pud_index_size);
76 unsigned long __pgd_index_size;
77 EXPORT_SYMBOL(__pgd_index_size);
78 unsigned long __pmd_cache_index;
79 EXPORT_SYMBOL(__pmd_cache_index);
80 unsigned long __pud_cache_index;
81 EXPORT_SYMBOL(__pud_cache_index);
82 unsigned long __pte_table_size;
83 EXPORT_SYMBOL(__pte_table_size);
84 unsigned long __pmd_table_size;
85 EXPORT_SYMBOL(__pmd_table_size);
86 unsigned long __pud_table_size;
87 EXPORT_SYMBOL(__pud_table_size);
88 unsigned long __pgd_table_size;
89 EXPORT_SYMBOL(__pgd_table_size);
90 unsigned long __pmd_val_bits;
91 EXPORT_SYMBOL(__pmd_val_bits);
92 unsigned long __pud_val_bits;
93 EXPORT_SYMBOL(__pud_val_bits);
94 unsigned long __pgd_val_bits;
95 EXPORT_SYMBOL(__pgd_val_bits);
96 unsigned long __kernel_virt_start;
97 EXPORT_SYMBOL(__kernel_virt_start);
98 unsigned long __kernel_virt_size;
99 EXPORT_SYMBOL(__kernel_virt_size);
100 unsigned long __vmalloc_start;
101 EXPORT_SYMBOL(__vmalloc_start);
102 unsigned long __vmalloc_end;
103 EXPORT_SYMBOL(__vmalloc_end);
104 unsigned long __kernel_io_start;
105 EXPORT_SYMBOL(__kernel_io_start);
106 struct page *vmemmap;
107 EXPORT_SYMBOL(vmemmap);
108 unsigned long __pte_frag_nr;
109 EXPORT_SYMBOL(__pte_frag_nr);
110 unsigned long __pte_frag_size_shift;
111 EXPORT_SYMBOL(__pte_frag_size_shift);
112 unsigned long ioremap_bot;
113 #else /* !CONFIG_PPC_BOOK3S_64 */
114 unsigned long ioremap_bot = IOREMAP_BASE;
115 #endif
116 
117 /**
118  * __ioremap_at - Low level function to establish the page tables
119  *                for an IO mapping
120  */
121 void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size,
122 			    unsigned long flags)
123 {
124 	unsigned long i;
125 
126 	/* Make sure we have the base flags */
127 	if ((flags & _PAGE_PRESENT) == 0)
128 		flags |= pgprot_val(PAGE_KERNEL);
129 
130 	/* We don't support the 4K PFN hack with ioremap */
131 	if (flags & H_PAGE_4K_PFN)
132 		return NULL;
133 
134 	WARN_ON(pa & ~PAGE_MASK);
135 	WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
136 	WARN_ON(size & ~PAGE_MASK);
137 
138 	for (i = 0; i < size; i += PAGE_SIZE)
139 		if (map_kernel_page((unsigned long)ea+i, pa+i, flags))
140 			return NULL;
141 
142 	return (void __iomem *)ea;
143 }
144 
145 /**
146  * __iounmap_from - Low level function to tear down the page tables
147  *                  for an IO mapping. This is used for mappings that
148  *                  are manipulated manually, like partial unmapping of
149  *                  PCI IOs or ISA space.
150  */
151 void __iounmap_at(void *ea, unsigned long size)
152 {
153 	WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
154 	WARN_ON(size & ~PAGE_MASK);
155 
156 	unmap_kernel_range((unsigned long)ea, size);
157 }
158 
159 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
160 				unsigned long flags, void *caller)
161 {
162 	phys_addr_t paligned;
163 	void __iomem *ret;
164 
165 	/*
166 	 * Choose an address to map it to.
167 	 * Once the imalloc system is running, we use it.
168 	 * Before that, we map using addresses going
169 	 * up from ioremap_bot.  imalloc will use
170 	 * the addresses from ioremap_bot through
171 	 * IMALLOC_END
172 	 *
173 	 */
174 	paligned = addr & PAGE_MASK;
175 	size = PAGE_ALIGN(addr + size) - paligned;
176 
177 	if ((size == 0) || (paligned == 0))
178 		return NULL;
179 
180 	if (slab_is_available()) {
181 		struct vm_struct *area;
182 
183 		area = __get_vm_area_caller(size, VM_IOREMAP,
184 					    ioremap_bot, IOREMAP_END,
185 					    caller);
186 		if (area == NULL)
187 			return NULL;
188 
189 		area->phys_addr = paligned;
190 		ret = __ioremap_at(paligned, area->addr, size, flags);
191 		if (!ret)
192 			vunmap(area->addr);
193 	} else {
194 		ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags);
195 		if (ret)
196 			ioremap_bot += size;
197 	}
198 
199 	if (ret)
200 		ret += addr & ~PAGE_MASK;
201 	return ret;
202 }
203 
204 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
205 			 unsigned long flags)
206 {
207 	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
208 }
209 
210 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
211 {
212 	unsigned long flags = pgprot_val(pgprot_noncached(__pgprot(0)));
213 	void *caller = __builtin_return_address(0);
214 
215 	if (ppc_md.ioremap)
216 		return ppc_md.ioremap(addr, size, flags, caller);
217 	return __ioremap_caller(addr, size, flags, caller);
218 }
219 
220 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
221 {
222 	unsigned long flags = pgprot_val(pgprot_noncached_wc(__pgprot(0)));
223 	void *caller = __builtin_return_address(0);
224 
225 	if (ppc_md.ioremap)
226 		return ppc_md.ioremap(addr, size, flags, caller);
227 	return __ioremap_caller(addr, size, flags, caller);
228 }
229 
230 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
231 			     unsigned long flags)
232 {
233 	void *caller = __builtin_return_address(0);
234 
235 	/* writeable implies dirty for kernel addresses */
236 	if (flags & _PAGE_WRITE)
237 		flags |= _PAGE_DIRTY;
238 
239 	/* we don't want to let _PAGE_EXEC leak out */
240 	flags &= ~_PAGE_EXEC;
241 	/*
242 	 * Force kernel mapping.
243 	 */
244 	flags &= ~_PAGE_USER;
245 	flags |= _PAGE_PRIVILEGED;
246 
247 	if (ppc_md.ioremap)
248 		return ppc_md.ioremap(addr, size, flags, caller);
249 	return __ioremap_caller(addr, size, flags, caller);
250 }
251 
252 
253 /*
254  * Unmap an IO region and remove it from imalloc'd list.
255  * Access to IO memory should be serialized by driver.
256  */
257 void __iounmap(volatile void __iomem *token)
258 {
259 	void *addr;
260 
261 	if (!slab_is_available())
262 		return;
263 
264 	addr = (void *) ((unsigned long __force)
265 			 PCI_FIX_ADDR(token) & PAGE_MASK);
266 	if ((unsigned long)addr < ioremap_bot) {
267 		printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
268 		       " at 0x%p\n", addr);
269 		return;
270 	}
271 	vunmap(addr);
272 }
273 
274 void iounmap(volatile void __iomem *token)
275 {
276 	if (ppc_md.iounmap)
277 		ppc_md.iounmap(token);
278 	else
279 		__iounmap(token);
280 }
281 
282 EXPORT_SYMBOL(ioremap);
283 EXPORT_SYMBOL(ioremap_wc);
284 EXPORT_SYMBOL(ioremap_prot);
285 EXPORT_SYMBOL(__ioremap);
286 EXPORT_SYMBOL(__ioremap_at);
287 EXPORT_SYMBOL(iounmap);
288 EXPORT_SYMBOL(__iounmap);
289 EXPORT_SYMBOL(__iounmap_at);
290 
291 #ifndef __PAGETABLE_PUD_FOLDED
292 /* 4 level page table */
293 struct page *pgd_page(pgd_t pgd)
294 {
295 	if (pgd_huge(pgd))
296 		return pte_page(pgd_pte(pgd));
297 	return virt_to_page(pgd_page_vaddr(pgd));
298 }
299 #endif
300 
301 struct page *pud_page(pud_t pud)
302 {
303 	if (pud_huge(pud))
304 		return pte_page(pud_pte(pud));
305 	return virt_to_page(pud_page_vaddr(pud));
306 }
307 
308 /*
309  * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
310  * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
311  */
312 struct page *pmd_page(pmd_t pmd)
313 {
314 	if (pmd_trans_huge(pmd) || pmd_huge(pmd) || pmd_devmap(pmd))
315 		return pte_page(pmd_pte(pmd));
316 	return virt_to_page(pmd_page_vaddr(pmd));
317 }
318 
319 #ifdef CONFIG_PPC_64K_PAGES
320 static pte_t *get_from_cache(struct mm_struct *mm)
321 {
322 	void *pte_frag, *ret;
323 
324 	spin_lock(&mm->page_table_lock);
325 	ret = mm->context.pte_frag;
326 	if (ret) {
327 		pte_frag = ret + PTE_FRAG_SIZE;
328 		/*
329 		 * If we have taken up all the fragments mark PTE page NULL
330 		 */
331 		if (((unsigned long)pte_frag & ~PAGE_MASK) == 0)
332 			pte_frag = NULL;
333 		mm->context.pte_frag = pte_frag;
334 	}
335 	spin_unlock(&mm->page_table_lock);
336 	return (pte_t *)ret;
337 }
338 
339 static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel)
340 {
341 	void *ret = NULL;
342 	struct page *page;
343 
344 	if (!kernel) {
345 		page = alloc_page(PGALLOC_GFP | __GFP_ACCOUNT);
346 		if (!page)
347 			return NULL;
348 		if (!pgtable_page_ctor(page)) {
349 			__free_page(page);
350 			return NULL;
351 		}
352 	} else {
353 		page = alloc_page(PGALLOC_GFP);
354 		if (!page)
355 			return NULL;
356 	}
357 
358 	ret = page_address(page);
359 	spin_lock(&mm->page_table_lock);
360 	/*
361 	 * If we find pgtable_page set, we return
362 	 * the allocated page with single fragement
363 	 * count.
364 	 */
365 	if (likely(!mm->context.pte_frag)) {
366 		set_page_count(page, PTE_FRAG_NR);
367 		mm->context.pte_frag = ret + PTE_FRAG_SIZE;
368 	}
369 	spin_unlock(&mm->page_table_lock);
370 
371 	return (pte_t *)ret;
372 }
373 
374 pte_t *pte_fragment_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
375 {
376 	pte_t *pte;
377 
378 	pte = get_from_cache(mm);
379 	if (pte)
380 		return pte;
381 
382 	return __alloc_for_cache(mm, kernel);
383 }
384 #endif /* CONFIG_PPC_64K_PAGES */
385 
386 void pte_fragment_free(unsigned long *table, int kernel)
387 {
388 	struct page *page = virt_to_page(table);
389 	if (put_page_testzero(page)) {
390 		if (!kernel)
391 			pgtable_page_dtor(page);
392 		free_unref_page(page);
393 	}
394 }
395 
396 #ifdef CONFIG_SMP
397 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
398 {
399 	unsigned long pgf = (unsigned long)table;
400 
401 	BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
402 	pgf |= shift;
403 	tlb_remove_table(tlb, (void *)pgf);
404 }
405 
406 void __tlb_remove_table(void *_table)
407 {
408 	void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
409 	unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
410 
411 	if (!shift)
412 		/* PTE page needs special handling */
413 		pte_fragment_free(table, 0);
414 	else {
415 		BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
416 		kmem_cache_free(PGT_CACHE(shift), table);
417 	}
418 }
419 #else
420 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
421 {
422 	if (!shift) {
423 		/* PTE page needs special handling */
424 		pte_fragment_free(table, 0);
425 	} else {
426 		BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
427 		kmem_cache_free(PGT_CACHE(shift), table);
428 	}
429 }
430 #endif
431 
432 #ifdef CONFIG_PPC_BOOK3S_64
433 void __init mmu_partition_table_init(void)
434 {
435 	unsigned long patb_size = 1UL << PATB_SIZE_SHIFT;
436 	unsigned long ptcr;
437 
438 	BUILD_BUG_ON_MSG((PATB_SIZE_SHIFT > 36), "Partition table size too large.");
439 	partition_tb = __va(memblock_alloc_base(patb_size, patb_size,
440 						MEMBLOCK_ALLOC_ANYWHERE));
441 
442 	/* Initialize the Partition Table with no entries */
443 	memset((void *)partition_tb, 0, patb_size);
444 
445 	/*
446 	 * update partition table control register,
447 	 * 64 K size.
448 	 */
449 	ptcr = __pa(partition_tb) | (PATB_SIZE_SHIFT - 12);
450 	mtspr(SPRN_PTCR, ptcr);
451 	powernv_set_nmmu_ptcr(ptcr);
452 }
453 
454 void mmu_partition_table_set_entry(unsigned int lpid, unsigned long dw0,
455 				   unsigned long dw1)
456 {
457 	unsigned long old = be64_to_cpu(partition_tb[lpid].patb0);
458 
459 	partition_tb[lpid].patb0 = cpu_to_be64(dw0);
460 	partition_tb[lpid].patb1 = cpu_to_be64(dw1);
461 
462 	/*
463 	 * Global flush of TLBs and partition table caches for this lpid.
464 	 * The type of flush (hash or radix) depends on what the previous
465 	 * use of this partition ID was, not the new use.
466 	 */
467 	asm volatile("ptesync" : : : "memory");
468 	if (old & PATB_HR) {
469 		asm volatile(PPC_TLBIE_5(%0,%1,2,0,1) : :
470 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
471 		asm volatile(PPC_TLBIE_5(%0,%1,2,1,1) : :
472 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
473 		trace_tlbie(lpid, 0, TLBIEL_INVAL_SET_LPID, lpid, 2, 0, 1);
474 	} else {
475 		asm volatile(PPC_TLBIE_5(%0,%1,2,0,0) : :
476 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
477 		trace_tlbie(lpid, 0, TLBIEL_INVAL_SET_LPID, lpid, 2, 0, 0);
478 	}
479 	/* do we need fixup here ?*/
480 	asm volatile("eieio; tlbsync; ptesync" : : : "memory");
481 }
482 EXPORT_SYMBOL_GPL(mmu_partition_table_set_entry);
483 #endif /* CONFIG_PPC_BOOK3S_64 */
484 
485 #ifdef CONFIG_STRICT_KERNEL_RWX
486 void mark_rodata_ro(void)
487 {
488 	if (!mmu_has_feature(MMU_FTR_KERNEL_RO)) {
489 		pr_warn("Warning: Unable to mark rodata read only on this CPU.\n");
490 		return;
491 	}
492 
493 	if (radix_enabled())
494 		radix__mark_rodata_ro();
495 	else
496 		hash__mark_rodata_ro();
497 }
498 
499 void mark_initmem_nx(void)
500 {
501 	if (radix_enabled())
502 		radix__mark_initmem_nx();
503 	else
504 		hash__mark_initmem_nx();
505 }
506 #endif
507