1/*
2 * This file contains the routines setting up the linux page tables.
3 *  -- paulus
4 *
5 *  Derived from arch/ppc/mm/init.c:
6 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
7 *
8 *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
9 *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
10 *    Copyright (C) 1996 Paul Mackerras
11 *
12 *  Derived from "arch/i386/mm/init.c"
13 *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
14 *
15 *  This program is free software; you can redistribute it and/or
16 *  modify it under the terms of the GNU General Public License
17 *  as published by the Free Software Foundation; either version
18 *  2 of the License, or (at your option) any later version.
19 *
20 */
21
22#include <linux/kernel.h>
23#include <linux/module.h>
24#include <linux/types.h>
25#include <linux/mm.h>
26#include <linux/vmalloc.h>
27#include <linux/init.h>
28#include <linux/highmem.h>
29#include <linux/memblock.h>
30#include <linux/slab.h>
31
32#include <asm/pgtable.h>
33#include <asm/pgalloc.h>
34#include <asm/fixmap.h>
35#include <asm/io.h>
36#include <asm/setup.h>
37
38#include "mmu_decl.h"
39
40unsigned long ioremap_base;
41unsigned long ioremap_bot;
42EXPORT_SYMBOL(ioremap_bot);	/* aka VMALLOC_END */
43
44#ifdef CONFIG_6xx
45#define HAVE_BATS	1
46#endif
47
48#if defined(CONFIG_FSL_BOOKE)
49#define HAVE_TLBCAM	1
50#endif
51
52extern char etext[], _stext[];
53
54#ifdef HAVE_BATS
55extern phys_addr_t v_mapped_by_bats(unsigned long va);
56extern unsigned long p_mapped_by_bats(phys_addr_t pa);
57#else /* !HAVE_BATS */
58#define v_mapped_by_bats(x)	(0UL)
59#define p_mapped_by_bats(x)	(0UL)
60#endif /* HAVE_BATS */
61
62#ifdef HAVE_TLBCAM
63extern phys_addr_t v_mapped_by_tlbcam(unsigned long va);
64extern unsigned long p_mapped_by_tlbcam(phys_addr_t pa);
65#else /* !HAVE_TLBCAM */
66#define v_mapped_by_tlbcam(x)	(0UL)
67#define p_mapped_by_tlbcam(x)	(0UL)
68#endif /* HAVE_TLBCAM */
69
70#define PGDIR_ORDER	(32 + PGD_T_LOG2 - PGDIR_SHIFT)
71
72#ifndef CONFIG_PPC_4K_PAGES
73static struct kmem_cache *pgtable_cache;
74
75void pgtable_cache_init(void)
76{
77	pgtable_cache = kmem_cache_create("PGDIR cache", 1 << PGDIR_ORDER,
78					  1 << PGDIR_ORDER, 0, NULL);
79	if (pgtable_cache == NULL)
80		panic("Couldn't allocate pgtable caches");
81}
82#endif
83
84pgd_t *pgd_alloc(struct mm_struct *mm)
85{
86	pgd_t *ret;
87
88	/* pgdir take page or two with 4K pages and a page fraction otherwise */
89#ifndef CONFIG_PPC_4K_PAGES
90	ret = kmem_cache_alloc(pgtable_cache, GFP_KERNEL | __GFP_ZERO);
91#else
92	ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
93			PGDIR_ORDER - PAGE_SHIFT);
94#endif
95	return ret;
96}
97
98void pgd_free(struct mm_struct *mm, pgd_t *pgd)
99{
100#ifndef CONFIG_PPC_4K_PAGES
101	kmem_cache_free(pgtable_cache, (void *)pgd);
102#else
103	free_pages((unsigned long)pgd, PGDIR_ORDER - PAGE_SHIFT);
104#endif
105}
106
107__init_refok pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
108{
109	pte_t *pte;
110
111	if (slab_is_available()) {
112		pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
113	} else {
114		pte = __va(memblock_alloc(PAGE_SIZE, PAGE_SIZE));
115		if (pte)
116			clear_page(pte);
117	}
118	return pte;
119}
120
121pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
122{
123	struct page *ptepage;
124
125	gfp_t flags = GFP_KERNEL | __GFP_REPEAT | __GFP_ZERO;
126
127	ptepage = alloc_pages(flags, 0);
128	if (!ptepage)
129		return NULL;
130	if (!pgtable_page_ctor(ptepage)) {
131		__free_page(ptepage);
132		return NULL;
133	}
134	return ptepage;
135}
136
137void __iomem *
138ioremap(phys_addr_t addr, unsigned long size)
139{
140	return __ioremap_caller(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED,
141				__builtin_return_address(0));
142}
143EXPORT_SYMBOL(ioremap);
144
145void __iomem *
146ioremap_wc(phys_addr_t addr, unsigned long size)
147{
148	return __ioremap_caller(addr, size, _PAGE_NO_CACHE,
149				__builtin_return_address(0));
150}
151EXPORT_SYMBOL(ioremap_wc);
152
153void __iomem *
154ioremap_prot(phys_addr_t addr, unsigned long size, unsigned long flags)
155{
156	/* writeable implies dirty for kernel addresses */
157	if ((flags & (_PAGE_RW | _PAGE_RO)) != _PAGE_RO)
158		flags |= _PAGE_DIRTY | _PAGE_HWWRITE;
159
160	/* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
161	flags &= ~(_PAGE_USER | _PAGE_EXEC);
162
163#ifdef _PAGE_BAP_SR
164	/* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
165	 * which means that we just cleared supervisor access... oops ;-) This
166	 * restores it
167	 */
168	flags |= _PAGE_BAP_SR;
169#endif
170
171	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
172}
173EXPORT_SYMBOL(ioremap_prot);
174
175void __iomem *
176__ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
177{
178	return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
179}
180
181void __iomem *
182__ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
183		 void *caller)
184{
185	unsigned long v, i;
186	phys_addr_t p;
187	int err;
188
189	/* Make sure we have the base flags */
190	if ((flags & _PAGE_PRESENT) == 0)
191		flags |= pgprot_val(PAGE_KERNEL);
192
193	/* Non-cacheable page cannot be coherent */
194	if (flags & _PAGE_NO_CACHE)
195		flags &= ~_PAGE_COHERENT;
196
197	/*
198	 * Choose an address to map it to.
199	 * Once the vmalloc system is running, we use it.
200	 * Before then, we use space going down from ioremap_base
201	 * (ioremap_bot records where we're up to).
202	 */
203	p = addr & PAGE_MASK;
204	size = PAGE_ALIGN(addr + size) - p;
205
206	/*
207	 * If the address lies within the first 16 MB, assume it's in ISA
208	 * memory space
209	 */
210	if (p < 16*1024*1024)
211		p += _ISA_MEM_BASE;
212
213#ifndef CONFIG_CRASH_DUMP
214	/*
215	 * Don't allow anybody to remap normal RAM that we're using.
216	 * mem_init() sets high_memory so only do the check after that.
217	 */
218	if (slab_is_available() && (p < virt_to_phys(high_memory)) &&
219	    !(__allow_ioremap_reserved && memblock_is_region_reserved(p, size))) {
220		printk("__ioremap(): phys addr 0x%llx is RAM lr %ps\n",
221		       (unsigned long long)p, __builtin_return_address(0));
222		return NULL;
223	}
224#endif
225
226	if (size == 0)
227		return NULL;
228
229	/*
230	 * Is it already mapped?  Perhaps overlapped by a previous
231	 * BAT mapping.  If the whole area is mapped then we're done,
232	 * otherwise remap it since we want to keep the virt addrs for
233	 * each request contiguous.
234	 *
235	 * We make the assumption here that if the bottom and top
236	 * of the range we want are mapped then it's mapped to the
237	 * same virt address (and this is contiguous).
238	 *  -- Cort
239	 */
240	if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
241		goto out;
242
243	if ((v = p_mapped_by_tlbcam(p)))
244		goto out;
245
246	if (slab_is_available()) {
247		struct vm_struct *area;
248		area = get_vm_area_caller(size, VM_IOREMAP, caller);
249		if (area == 0)
250			return NULL;
251		area->phys_addr = p;
252		v = (unsigned long) area->addr;
253	} else {
254		v = (ioremap_bot -= size);
255	}
256
257	/*
258	 * Should check if it is a candidate for a BAT mapping
259	 */
260
261	err = 0;
262	for (i = 0; i < size && err == 0; i += PAGE_SIZE)
263		err = map_page(v+i, p+i, flags);
264	if (err) {
265		if (slab_is_available())
266			vunmap((void *)v);
267		return NULL;
268	}
269
270out:
271	return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
272}
273EXPORT_SYMBOL(__ioremap);
274
275void iounmap(volatile void __iomem *addr)
276{
277	/*
278	 * If mapped by BATs then there is nothing to do.
279	 * Calling vfree() generates a benign warning.
280	 */
281	if (v_mapped_by_bats((unsigned long)addr)) return;
282
283	if (addr > high_memory && (unsigned long) addr < ioremap_bot)
284		vunmap((void *) (PAGE_MASK & (unsigned long)addr));
285}
286EXPORT_SYMBOL(iounmap);
287
288int map_page(unsigned long va, phys_addr_t pa, int flags)
289{
290	pmd_t *pd;
291	pte_t *pg;
292	int err = -ENOMEM;
293
294	/* Use upper 10 bits of VA to index the first level map */
295	pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
296	/* Use middle 10 bits of VA to index the second-level map */
297	pg = pte_alloc_kernel(pd, va);
298	if (pg != 0) {
299		err = 0;
300		/* The PTE should never be already set nor present in the
301		 * hash table
302		 */
303		BUG_ON((pte_val(*pg) & (_PAGE_PRESENT | _PAGE_HASHPTE)) &&
304		       flags);
305		set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT,
306						     __pgprot(flags)));
307	}
308	smp_wmb();
309	return err;
310}
311
312/*
313 * Map in a chunk of physical memory starting at start.
314 */
315void __init __mapin_ram_chunk(unsigned long offset, unsigned long top)
316{
317	unsigned long v, s, f;
318	phys_addr_t p;
319	int ktext;
320
321	s = offset;
322	v = PAGE_OFFSET + s;
323	p = memstart_addr + s;
324	for (; s < top; s += PAGE_SIZE) {
325		ktext = ((char *) v >= _stext && (char *) v < etext);
326		f = ktext ? pgprot_val(PAGE_KERNEL_TEXT) : pgprot_val(PAGE_KERNEL);
327		map_page(v, p, f);
328#ifdef CONFIG_PPC_STD_MMU_32
329		if (ktext)
330			hash_preload(&init_mm, v, 0, 0x300);
331#endif
332		v += PAGE_SIZE;
333		p += PAGE_SIZE;
334	}
335}
336
337void __init mapin_ram(void)
338{
339	unsigned long s, top;
340
341#ifndef CONFIG_WII
342	top = total_lowmem;
343	s = mmu_mapin_ram(top);
344	__mapin_ram_chunk(s, top);
345#else
346	if (!wii_hole_size) {
347		s = mmu_mapin_ram(total_lowmem);
348		__mapin_ram_chunk(s, total_lowmem);
349	} else {
350		top = wii_hole_start;
351		s = mmu_mapin_ram(top);
352		__mapin_ram_chunk(s, top);
353
354		top = memblock_end_of_DRAM();
355		s = wii_mmu_mapin_mem2(top);
356		__mapin_ram_chunk(s, top);
357	}
358#endif
359}
360
361/* Scan the real Linux page tables and return a PTE pointer for
362 * a virtual address in a context.
363 * Returns true (1) if PTE was found, zero otherwise.  The pointer to
364 * the PTE pointer is unmodified if PTE is not found.
365 */
366int
367get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep, pmd_t **pmdp)
368{
369        pgd_t	*pgd;
370	pud_t	*pud;
371        pmd_t	*pmd;
372        pte_t	*pte;
373        int     retval = 0;
374
375        pgd = pgd_offset(mm, addr & PAGE_MASK);
376        if (pgd) {
377		pud = pud_offset(pgd, addr & PAGE_MASK);
378		if (pud && pud_present(*pud)) {
379			pmd = pmd_offset(pud, addr & PAGE_MASK);
380			if (pmd_present(*pmd)) {
381				pte = pte_offset_map(pmd, addr & PAGE_MASK);
382				if (pte) {
383					retval = 1;
384					*ptep = pte;
385					if (pmdp)
386						*pmdp = pmd;
387					/* XXX caller needs to do pte_unmap, yuck */
388				}
389			}
390		}
391        }
392        return(retval);
393}
394
395#ifdef CONFIG_DEBUG_PAGEALLOC
396
397static int __change_page_attr(struct page *page, pgprot_t prot)
398{
399	pte_t *kpte;
400	pmd_t *kpmd;
401	unsigned long address;
402
403	BUG_ON(PageHighMem(page));
404	address = (unsigned long)page_address(page);
405
406	if (v_mapped_by_bats(address) || v_mapped_by_tlbcam(address))
407		return 0;
408	if (!get_pteptr(&init_mm, address, &kpte, &kpmd))
409		return -EINVAL;
410	__set_pte_at(&init_mm, address, kpte, mk_pte(page, prot), 0);
411	wmb();
412	flush_tlb_page(NULL, address);
413	pte_unmap(kpte);
414
415	return 0;
416}
417
418/*
419 * Change the page attributes of an page in the linear mapping.
420 *
421 * THIS CONFLICTS WITH BAT MAPPINGS, DEBUG USE ONLY
422 */
423static int change_page_attr(struct page *page, int numpages, pgprot_t prot)
424{
425	int i, err = 0;
426	unsigned long flags;
427
428	local_irq_save(flags);
429	for (i = 0; i < numpages; i++, page++) {
430		err = __change_page_attr(page, prot);
431		if (err)
432			break;
433	}
434	local_irq_restore(flags);
435	return err;
436}
437
438
439void __kernel_map_pages(struct page *page, int numpages, int enable)
440{
441	if (PageHighMem(page))
442		return;
443
444	change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
445}
446#endif /* CONFIG_DEBUG_PAGEALLOC */
447
448static int fixmaps;
449
450void __set_fixmap (enum fixed_addresses idx, phys_addr_t phys, pgprot_t flags)
451{
452	unsigned long address = __fix_to_virt(idx);
453
454	if (idx >= __end_of_fixed_addresses) {
455		BUG();
456		return;
457	}
458
459	map_page(address, phys, pgprot_val(flags));
460	fixmaps++;
461}
462
463void __this_fixmap_does_not_exist(void)
464{
465	WARN_ON(1);
466}
467