xref: /linux/arch/powerpc/mm/pgtable_64.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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/processor.h>
51 #include <asm/cputable.h>
52 #include <asm/sections.h>
53 #include <asm/firmware.h>
54 #include <asm/dma.h>
55 #include <asm/powernv.h>
56 
57 #include "mmu_decl.h"
58 
59 #ifdef CONFIG_PPC_STD_MMU_64
60 #if TASK_SIZE_USER64 > (1UL << (ESID_BITS + SID_SHIFT))
61 #error TASK_SIZE_USER64 exceeds user VSID range
62 #endif
63 #endif
64 
65 #ifdef CONFIG_PPC_BOOK3S_64
66 /*
67  * partition table and process table for ISA 3.0
68  */
69 struct prtb_entry *process_tb;
70 struct patb_entry *partition_tb;
71 /*
72  * page table size
73  */
74 unsigned long __pte_index_size;
75 EXPORT_SYMBOL(__pte_index_size);
76 unsigned long __pmd_index_size;
77 EXPORT_SYMBOL(__pmd_index_size);
78 unsigned long __pud_index_size;
79 EXPORT_SYMBOL(__pud_index_size);
80 unsigned long __pgd_index_size;
81 EXPORT_SYMBOL(__pgd_index_size);
82 unsigned long __pmd_cache_index;
83 EXPORT_SYMBOL(__pmd_cache_index);
84 unsigned long __pte_table_size;
85 EXPORT_SYMBOL(__pte_table_size);
86 unsigned long __pmd_table_size;
87 EXPORT_SYMBOL(__pmd_table_size);
88 unsigned long __pud_table_size;
89 EXPORT_SYMBOL(__pud_table_size);
90 unsigned long __pgd_table_size;
91 EXPORT_SYMBOL(__pgd_table_size);
92 unsigned long __pmd_val_bits;
93 EXPORT_SYMBOL(__pmd_val_bits);
94 unsigned long __pud_val_bits;
95 EXPORT_SYMBOL(__pud_val_bits);
96 unsigned long __pgd_val_bits;
97 EXPORT_SYMBOL(__pgd_val_bits);
98 unsigned long __kernel_virt_start;
99 EXPORT_SYMBOL(__kernel_virt_start);
100 unsigned long __kernel_virt_size;
101 EXPORT_SYMBOL(__kernel_virt_size);
102 unsigned long __vmalloc_start;
103 EXPORT_SYMBOL(__vmalloc_start);
104 unsigned long __vmalloc_end;
105 EXPORT_SYMBOL(__vmalloc_end);
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 #if defined(CONFIG_PPC_BOOK3S_64)
245 	flags |= _PAGE_PRIVILEGED;
246 #else
247 	flags &= ~_PAGE_USER;
248 #endif
249 
250 
251 #ifdef _PAGE_BAP_SR
252 	/* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
253 	 * which means that we just cleared supervisor access... oops ;-) This
254 	 * restores it
255 	 */
256 	flags |= _PAGE_BAP_SR;
257 #endif
258 
259 	if (ppc_md.ioremap)
260 		return ppc_md.ioremap(addr, size, flags, caller);
261 	return __ioremap_caller(addr, size, flags, caller);
262 }
263 
264 
265 /*
266  * Unmap an IO region and remove it from imalloc'd list.
267  * Access to IO memory should be serialized by driver.
268  */
269 void __iounmap(volatile void __iomem *token)
270 {
271 	void *addr;
272 
273 	if (!slab_is_available())
274 		return;
275 
276 	addr = (void *) ((unsigned long __force)
277 			 PCI_FIX_ADDR(token) & PAGE_MASK);
278 	if ((unsigned long)addr < ioremap_bot) {
279 		printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
280 		       " at 0x%p\n", addr);
281 		return;
282 	}
283 	vunmap(addr);
284 }
285 
286 void iounmap(volatile void __iomem *token)
287 {
288 	if (ppc_md.iounmap)
289 		ppc_md.iounmap(token);
290 	else
291 		__iounmap(token);
292 }
293 
294 EXPORT_SYMBOL(ioremap);
295 EXPORT_SYMBOL(ioremap_wc);
296 EXPORT_SYMBOL(ioremap_prot);
297 EXPORT_SYMBOL(__ioremap);
298 EXPORT_SYMBOL(__ioremap_at);
299 EXPORT_SYMBOL(iounmap);
300 EXPORT_SYMBOL(__iounmap);
301 EXPORT_SYMBOL(__iounmap_at);
302 
303 #ifndef __PAGETABLE_PUD_FOLDED
304 /* 4 level page table */
305 struct page *pgd_page(pgd_t pgd)
306 {
307 	if (pgd_huge(pgd))
308 		return pte_page(pgd_pte(pgd));
309 	return virt_to_page(pgd_page_vaddr(pgd));
310 }
311 #endif
312 
313 struct page *pud_page(pud_t pud)
314 {
315 	if (pud_huge(pud))
316 		return pte_page(pud_pte(pud));
317 	return virt_to_page(pud_page_vaddr(pud));
318 }
319 
320 /*
321  * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
322  * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
323  */
324 struct page *pmd_page(pmd_t pmd)
325 {
326 	if (pmd_trans_huge(pmd) || pmd_huge(pmd))
327 		return pte_page(pmd_pte(pmd));
328 	return virt_to_page(pmd_page_vaddr(pmd));
329 }
330 
331 #ifdef CONFIG_PPC_64K_PAGES
332 static pte_t *get_from_cache(struct mm_struct *mm)
333 {
334 	void *pte_frag, *ret;
335 
336 	spin_lock(&mm->page_table_lock);
337 	ret = mm->context.pte_frag;
338 	if (ret) {
339 		pte_frag = ret + PTE_FRAG_SIZE;
340 		/*
341 		 * If we have taken up all the fragments mark PTE page NULL
342 		 */
343 		if (((unsigned long)pte_frag & ~PAGE_MASK) == 0)
344 			pte_frag = NULL;
345 		mm->context.pte_frag = pte_frag;
346 	}
347 	spin_unlock(&mm->page_table_lock);
348 	return (pte_t *)ret;
349 }
350 
351 static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel)
352 {
353 	void *ret = NULL;
354 	struct page *page = alloc_page(GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
355 	if (!page)
356 		return NULL;
357 	if (!kernel && !pgtable_page_ctor(page)) {
358 		__free_page(page);
359 		return NULL;
360 	}
361 
362 	ret = page_address(page);
363 	spin_lock(&mm->page_table_lock);
364 	/*
365 	 * If we find pgtable_page set, we return
366 	 * the allocated page with single fragement
367 	 * count.
368 	 */
369 	if (likely(!mm->context.pte_frag)) {
370 		set_page_count(page, PTE_FRAG_NR);
371 		mm->context.pte_frag = ret + PTE_FRAG_SIZE;
372 	}
373 	spin_unlock(&mm->page_table_lock);
374 
375 	return (pte_t *)ret;
376 }
377 
378 pte_t *pte_fragment_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
379 {
380 	pte_t *pte;
381 
382 	pte = get_from_cache(mm);
383 	if (pte)
384 		return pte;
385 
386 	return __alloc_for_cache(mm, kernel);
387 }
388 #endif /* CONFIG_PPC_64K_PAGES */
389 
390 void pte_fragment_free(unsigned long *table, int kernel)
391 {
392 	struct page *page = virt_to_page(table);
393 	if (put_page_testzero(page)) {
394 		if (!kernel)
395 			pgtable_page_dtor(page);
396 		free_hot_cold_page(page, 0);
397 	}
398 }
399 
400 #ifdef CONFIG_SMP
401 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
402 {
403 	unsigned long pgf = (unsigned long)table;
404 
405 	BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
406 	pgf |= shift;
407 	tlb_remove_table(tlb, (void *)pgf);
408 }
409 
410 void __tlb_remove_table(void *_table)
411 {
412 	void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
413 	unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
414 
415 	if (!shift)
416 		/* PTE page needs special handling */
417 		pte_fragment_free(table, 0);
418 	else {
419 		BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
420 		kmem_cache_free(PGT_CACHE(shift), table);
421 	}
422 }
423 #else
424 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
425 {
426 	if (!shift) {
427 		/* PTE page needs special handling */
428 		pte_fragment_free(table, 0);
429 	} else {
430 		BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
431 		kmem_cache_free(PGT_CACHE(shift), table);
432 	}
433 }
434 #endif
435 
436 #ifdef CONFIG_PPC_BOOK3S_64
437 void __init mmu_partition_table_init(void)
438 {
439 	unsigned long patb_size = 1UL << PATB_SIZE_SHIFT;
440 	unsigned long ptcr;
441 
442 	BUILD_BUG_ON_MSG((PATB_SIZE_SHIFT > 36), "Partition table size too large.");
443 	partition_tb = __va(memblock_alloc_base(patb_size, patb_size,
444 						MEMBLOCK_ALLOC_ANYWHERE));
445 
446 	/* Initialize the Partition Table with no entries */
447 	memset((void *)partition_tb, 0, patb_size);
448 
449 	/*
450 	 * update partition table control register,
451 	 * 64 K size.
452 	 */
453 	ptcr = __pa(partition_tb) | (PATB_SIZE_SHIFT - 12);
454 	mtspr(SPRN_PTCR, ptcr);
455 	powernv_set_nmmu_ptcr(ptcr);
456 }
457 
458 void mmu_partition_table_set_entry(unsigned int lpid, unsigned long dw0,
459 				   unsigned long dw1)
460 {
461 	unsigned long old = be64_to_cpu(partition_tb[lpid].patb0);
462 
463 	partition_tb[lpid].patb0 = cpu_to_be64(dw0);
464 	partition_tb[lpid].patb1 = cpu_to_be64(dw1);
465 
466 	/*
467 	 * Global flush of TLBs and partition table caches for this lpid.
468 	 * The type of flush (hash or radix) depends on what the previous
469 	 * use of this partition ID was, not the new use.
470 	 */
471 	asm volatile("ptesync" : : : "memory");
472 	if (old & PATB_HR)
473 		asm volatile(PPC_TLBIE_5(%0,%1,2,0,1) : :
474 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
475 	else
476 		asm volatile(PPC_TLBIE_5(%0,%1,2,0,0) : :
477 			     "r" (TLBIEL_INVAL_SET_LPID), "r" (lpid));
478 	asm volatile("eieio; tlbsync; ptesync" : : : "memory");
479 }
480 EXPORT_SYMBOL_GPL(mmu_partition_table_set_entry);
481 #endif /* CONFIG_PPC_BOOK3S_64 */
482