aboutsummaryrefslogtreecommitdiffstats
path: root/arch/ppc64/mm/init.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/ppc64/mm/init.c')
-rw-r--r--arch/ppc64/mm/init.c950
1 files changed, 0 insertions, 950 deletions
diff --git a/arch/ppc64/mm/init.c b/arch/ppc64/mm/init.c
deleted file mode 100644
index e2bd7776622f..000000000000
--- a/arch/ppc64/mm/init.c
+++ /dev/null
@@ -1,950 +0,0 @@
1/*
2 * PowerPC version
3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4 *
5 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
6 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
7 * Copyright (C) 1996 Paul Mackerras
8 * Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
9 *
10 * Derived from "arch/i386/mm/init.c"
11 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
12 *
13 * Dave Engebretsen <engebret@us.ibm.com>
14 * Rework for PPC64 port.
15 *
16 * This program is free software; you can redistribute it and/or
17 * modify it under the terms of the GNU General Public License
18 * as published by the Free Software Foundation; either version
19 * 2 of the License, or (at your option) any later version.
20 *
21 */
22
23#include <linux/config.h>
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/types.h>
30#include <linux/mman.h>
31#include <linux/mm.h>
32#include <linux/swap.h>
33#include <linux/stddef.h>
34#include <linux/vmalloc.h>
35#include <linux/init.h>
36#include <linux/delay.h>
37#include <linux/bootmem.h>
38#include <linux/highmem.h>
39#include <linux/idr.h>
40#include <linux/nodemask.h>
41#include <linux/module.h>
42
43#include <asm/pgalloc.h>
44#include <asm/page.h>
45#include <asm/prom.h>
46#include <asm/lmb.h>
47#include <asm/rtas.h>
48#include <asm/io.h>
49#include <asm/mmu_context.h>
50#include <asm/pgtable.h>
51#include <asm/mmu.h>
52#include <asm/uaccess.h>
53#include <asm/smp.h>
54#include <asm/machdep.h>
55#include <asm/tlb.h>
56#include <asm/eeh.h>
57#include <asm/processor.h>
58#include <asm/mmzone.h>
59#include <asm/cputable.h>
60#include <asm/ppcdebug.h>
61#include <asm/sections.h>
62#include <asm/system.h>
63#include <asm/iommu.h>
64#include <asm/abs_addr.h>
65#include <asm/vdso.h>
66#include <asm/imalloc.h>
67
68#if PGTABLE_RANGE > USER_VSID_RANGE
69#warning Limited user VSID range means pagetable space is wasted
70#endif
71
72#if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
73#warning TASK_SIZE is smaller than it needs to be.
74#endif
75
76int mem_init_done;
77unsigned long ioremap_bot = IMALLOC_BASE;
78static unsigned long phbs_io_bot = PHBS_IO_BASE;
79
80extern pgd_t swapper_pg_dir[];
81extern struct task_struct *current_set[NR_CPUS];
82
83unsigned long klimit = (unsigned long)_end;
84
85unsigned long _SDR1=0;
86unsigned long _ASR=0;
87
88/* max amount of RAM to use */
89unsigned long __max_memory;
90
91/* info on what we think the IO hole is */
92unsigned long io_hole_start;
93unsigned long io_hole_size;
94
95void show_mem(void)
96{
97 unsigned long total = 0, reserved = 0;
98 unsigned long shared = 0, cached = 0;
99 struct page *page;
100 pg_data_t *pgdat;
101 unsigned long i;
102
103 printk("Mem-info:\n");
104 show_free_areas();
105 printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
106 for_each_pgdat(pgdat) {
107 unsigned long flags;
108 pgdat_resize_lock(pgdat, &flags);
109 for (i = 0; i < pgdat->node_spanned_pages; i++) {
110 page = pgdat_page_nr(pgdat, i);
111 total++;
112 if (PageReserved(page))
113 reserved++;
114 else if (PageSwapCache(page))
115 cached++;
116 else if (page_count(page))
117 shared += page_count(page) - 1;
118 }
119 pgdat_resize_unlock(pgdat, &flags);
120 }
121 printk("%ld pages of RAM\n", total);
122 printk("%ld reserved pages\n", reserved);
123 printk("%ld pages shared\n", shared);
124 printk("%ld pages swap cached\n", cached);
125}
126
127#ifdef CONFIG_PPC_ISERIES
128
129void __iomem *ioremap(unsigned long addr, unsigned long size)
130{
131 return (void __iomem *)addr;
132}
133
134extern void __iomem *__ioremap(unsigned long addr, unsigned long size,
135 unsigned long flags)
136{
137 return (void __iomem *)addr;
138}
139
140void iounmap(volatile void __iomem *addr)
141{
142 return;
143}
144
145#else
146
147/*
148 * map_io_page currently only called by __ioremap
149 * map_io_page adds an entry to the ioremap page table
150 * and adds an entry to the HPT, possibly bolting it
151 */
152static int map_io_page(unsigned long ea, unsigned long pa, int flags)
153{
154 pgd_t *pgdp;
155 pud_t *pudp;
156 pmd_t *pmdp;
157 pte_t *ptep;
158 unsigned long vsid;
159
160 if (mem_init_done) {
161 pgdp = pgd_offset_k(ea);
162 pudp = pud_alloc(&init_mm, pgdp, ea);
163 if (!pudp)
164 return -ENOMEM;
165 pmdp = pmd_alloc(&init_mm, pudp, ea);
166 if (!pmdp)
167 return -ENOMEM;
168 ptep = pte_alloc_kernel(pmdp, ea);
169 if (!ptep)
170 return -ENOMEM;
171 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
172 __pgprot(flags)));
173 } else {
174 unsigned long va, vpn, hash, hpteg;
175
176 /*
177 * If the mm subsystem is not fully up, we cannot create a
178 * linux page table entry for this mapping. Simply bolt an
179 * entry in the hardware page table.
180 */
181 vsid = get_kernel_vsid(ea);
182 va = (vsid << 28) | (ea & 0xFFFFFFF);
183 vpn = va >> PAGE_SHIFT;
184
185 hash = hpt_hash(vpn, 0);
186
187 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
188
189 /* Panic if a pte grpup is full */
190 if (ppc_md.hpte_insert(hpteg, va, pa >> PAGE_SHIFT,
191 HPTE_V_BOLTED,
192 _PAGE_NO_CACHE|_PAGE_GUARDED|PP_RWXX)
193 == -1) {
194 panic("map_io_page: could not insert mapping");
195 }
196 }
197 return 0;
198}
199
200
201static void __iomem * __ioremap_com(unsigned long addr, unsigned long pa,
202 unsigned long ea, unsigned long size,
203 unsigned long flags)
204{
205 unsigned long i;
206
207 if ((flags & _PAGE_PRESENT) == 0)
208 flags |= pgprot_val(PAGE_KERNEL);
209
210 for (i = 0; i < size; i += PAGE_SIZE)
211 if (map_io_page(ea+i, pa+i, flags))
212 return NULL;
213
214 return (void __iomem *) (ea + (addr & ~PAGE_MASK));
215}
216
217
218void __iomem *
219ioremap(unsigned long addr, unsigned long size)
220{
221 return __ioremap(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED);
222}
223
224void __iomem * __ioremap(unsigned long addr, unsigned long size,
225 unsigned long flags)
226{
227 unsigned long pa, ea;
228 void __iomem *ret;
229
230 /*
231 * Choose an address to map it to.
232 * Once the imalloc system is running, we use it.
233 * Before that, we map using addresses going
234 * up from ioremap_bot. imalloc will use
235 * the addresses from ioremap_bot through
236 * IMALLOC_END
237 *
238 */
239 pa = addr & PAGE_MASK;
240 size = PAGE_ALIGN(addr + size) - pa;
241
242 if (size == 0)
243 return NULL;
244
245 if (mem_init_done) {
246 struct vm_struct *area;
247 area = im_get_free_area(size);
248 if (area == NULL)
249 return NULL;
250 ea = (unsigned long)(area->addr);
251 ret = __ioremap_com(addr, pa, ea, size, flags);
252 if (!ret)
253 im_free(area->addr);
254 } else {
255 ea = ioremap_bot;
256 ret = __ioremap_com(addr, pa, ea, size, flags);
257 if (ret)
258 ioremap_bot += size;
259 }
260 return ret;
261}
262
263#define IS_PAGE_ALIGNED(_val) ((_val) == ((_val) & PAGE_MASK))
264
265int __ioremap_explicit(unsigned long pa, unsigned long ea,
266 unsigned long size, unsigned long flags)
267{
268 struct vm_struct *area;
269 void __iomem *ret;
270
271 /* For now, require page-aligned values for pa, ea, and size */
272 if (!IS_PAGE_ALIGNED(pa) || !IS_PAGE_ALIGNED(ea) ||
273 !IS_PAGE_ALIGNED(size)) {
274 printk(KERN_ERR "unaligned value in %s\n", __FUNCTION__);
275 return 1;
276 }
277
278 if (!mem_init_done) {
279 /* Two things to consider in this case:
280 * 1) No records will be kept (imalloc, etc) that the region
281 * has been remapped
282 * 2) It won't be easy to iounmap() the region later (because
283 * of 1)
284 */
285 ;
286 } else {
287 area = im_get_area(ea, size,
288 IM_REGION_UNUSED|IM_REGION_SUBSET|IM_REGION_EXISTS);
289 if (area == NULL) {
290 /* Expected when PHB-dlpar is in play */
291 return 1;
292 }
293 if (ea != (unsigned long) area->addr) {
294 printk(KERN_ERR "unexpected addr return from "
295 "im_get_area\n");
296 return 1;
297 }
298 }
299
300 ret = __ioremap_com(pa, pa, ea, size, flags);
301 if (ret == NULL) {
302 printk(KERN_ERR "ioremap_explicit() allocation failure !\n");
303 return 1;
304 }
305 if (ret != (void *) ea) {
306 printk(KERN_ERR "__ioremap_com() returned unexpected addr\n");
307 return 1;
308 }
309
310 return 0;
311}
312
313/*
314 * Unmap an IO region and remove it from imalloc'd list.
315 * Access to IO memory should be serialized by driver.
316 * This code is modeled after vmalloc code - unmap_vm_area()
317 *
318 * XXX what about calls before mem_init_done (ie python_countermeasures())
319 */
320void iounmap(volatile void __iomem *token)
321{
322 void *addr;
323
324 if (!mem_init_done)
325 return;
326
327 addr = (void *) ((unsigned long __force) token & PAGE_MASK);
328
329 im_free(addr);
330}
331
332static int iounmap_subset_regions(unsigned long addr, unsigned long size)
333{
334 struct vm_struct *area;
335
336 /* Check whether subsets of this region exist */
337 area = im_get_area(addr, size, IM_REGION_SUPERSET);
338 if (area == NULL)
339 return 1;
340
341 while (area) {
342 iounmap((void __iomem *) area->addr);
343 area = im_get_area(addr, size,
344 IM_REGION_SUPERSET);
345 }
346
347 return 0;
348}
349
350int iounmap_explicit(volatile void __iomem *start, unsigned long size)
351{
352 struct vm_struct *area;
353 unsigned long addr;
354 int rc;
355
356 addr = (unsigned long __force) start & PAGE_MASK;
357
358 /* Verify that the region either exists or is a subset of an existing
359 * region. In the latter case, split the parent region to create
360 * the exact region
361 */
362 area = im_get_area(addr, size,
363 IM_REGION_EXISTS | IM_REGION_SUBSET);
364 if (area == NULL) {
365 /* Determine whether subset regions exist. If so, unmap */
366 rc = iounmap_subset_regions(addr, size);
367 if (rc) {
368 printk(KERN_ERR
369 "%s() cannot unmap nonexistent range 0x%lx\n",
370 __FUNCTION__, addr);
371 return 1;
372 }
373 } else {
374 iounmap((void __iomem *) area->addr);
375 }
376 /*
377 * FIXME! This can't be right:
378 iounmap(area->addr);
379 * Maybe it should be "iounmap(area);"
380 */
381 return 0;
382}
383
384#endif
385
386EXPORT_SYMBOL(ioremap);
387EXPORT_SYMBOL(__ioremap);
388EXPORT_SYMBOL(iounmap);
389
390void free_initmem(void)
391{
392 unsigned long addr;
393
394 addr = (unsigned long)__init_begin;
395 for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
396 memset((void *)addr, 0xcc, PAGE_SIZE);
397 ClearPageReserved(virt_to_page(addr));
398 set_page_count(virt_to_page(addr), 1);
399 free_page(addr);
400 totalram_pages++;
401 }
402 printk ("Freeing unused kernel memory: %luk freed\n",
403 ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
404}
405
406#ifdef CONFIG_BLK_DEV_INITRD
407void free_initrd_mem(unsigned long start, unsigned long end)
408{
409 if (start < end)
410 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
411 for (; start < end; start += PAGE_SIZE) {
412 ClearPageReserved(virt_to_page(start));
413 set_page_count(virt_to_page(start), 1);
414 free_page(start);
415 totalram_pages++;
416 }
417}
418#endif
419
420static DEFINE_SPINLOCK(mmu_context_lock);
421static DEFINE_IDR(mmu_context_idr);
422
423int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
424{
425 int index;
426 int err;
427
428again:
429 if (!idr_pre_get(&mmu_context_idr, GFP_KERNEL))
430 return -ENOMEM;
431
432 spin_lock(&mmu_context_lock);
433 err = idr_get_new_above(&mmu_context_idr, NULL, 1, &index);
434 spin_unlock(&mmu_context_lock);
435
436 if (err == -EAGAIN)
437 goto again;
438 else if (err)
439 return err;
440
441 if (index > MAX_CONTEXT) {
442 idr_remove(&mmu_context_idr, index);
443 return -ENOMEM;
444 }
445
446 mm->context.id = index;
447
448 return 0;
449}
450
451void destroy_context(struct mm_struct *mm)
452{
453 spin_lock(&mmu_context_lock);
454 idr_remove(&mmu_context_idr, mm->context.id);
455 spin_unlock(&mmu_context_lock);
456
457 mm->context.id = NO_CONTEXT;
458}
459
460/*
461 * Do very early mm setup.
462 */
463void __init mm_init_ppc64(void)
464{
465#ifndef CONFIG_PPC_ISERIES
466 unsigned long i;
467#endif
468
469 ppc64_boot_msg(0x100, "MM Init");
470
471 /* This is the story of the IO hole... please, keep seated,
472 * unfortunately, we are out of oxygen masks at the moment.
473 * So we need some rough way to tell where your big IO hole
474 * is. On pmac, it's between 2G and 4G, on POWER3, it's around
475 * that area as well, on POWER4 we don't have one, etc...
476 * We need that as a "hint" when sizing the TCE table on POWER3
477 * So far, the simplest way that seem work well enough for us it
478 * to just assume that the first discontinuity in our physical
479 * RAM layout is the IO hole. That may not be correct in the future
480 * (and isn't on iSeries but then we don't care ;)
481 */
482
483#ifndef CONFIG_PPC_ISERIES
484 for (i = 1; i < lmb.memory.cnt; i++) {
485 unsigned long base, prevbase, prevsize;
486
487 prevbase = lmb.memory.region[i-1].base;
488 prevsize = lmb.memory.region[i-1].size;
489 base = lmb.memory.region[i].base;
490 if (base > (prevbase + prevsize)) {
491 io_hole_start = prevbase + prevsize;
492 io_hole_size = base - (prevbase + prevsize);
493 break;
494 }
495 }
496#endif /* CONFIG_PPC_ISERIES */
497 if (io_hole_start)
498 printk("IO Hole assumed to be %lx -> %lx\n",
499 io_hole_start, io_hole_start + io_hole_size - 1);
500
501 ppc64_boot_msg(0x100, "MM Init Done");
502}
503
504/*
505 * This is called by /dev/mem to know if a given address has to
506 * be mapped non-cacheable or not
507 */
508int page_is_ram(unsigned long pfn)
509{
510 int i;
511 unsigned long paddr = (pfn << PAGE_SHIFT);
512
513 for (i=0; i < lmb.memory.cnt; i++) {
514 unsigned long base;
515
516 base = lmb.memory.region[i].base;
517
518 if ((paddr >= base) &&
519 (paddr < (base + lmb.memory.region[i].size))) {
520 return 1;
521 }
522 }
523
524 return 0;
525}
526EXPORT_SYMBOL(page_is_ram);
527
528/*
529 * Initialize the bootmem system and give it all the memory we
530 * have available.
531 */
532#ifndef CONFIG_NEED_MULTIPLE_NODES
533void __init do_init_bootmem(void)
534{
535 unsigned long i;
536 unsigned long start, bootmap_pages;
537 unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
538 int boot_mapsize;
539
540 /*
541 * Find an area to use for the bootmem bitmap. Calculate the size of
542 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
543 * Add 1 additional page in case the address isn't page-aligned.
544 */
545 bootmap_pages = bootmem_bootmap_pages(total_pages);
546
547 start = lmb_alloc(bootmap_pages<<PAGE_SHIFT, PAGE_SIZE);
548 BUG_ON(!start);
549
550 boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
551
552 max_pfn = max_low_pfn;
553
554 /* Add all physical memory to the bootmem map, mark each area
555 * present.
556 */
557 for (i=0; i < lmb.memory.cnt; i++)
558 free_bootmem(lmb.memory.region[i].base,
559 lmb_size_bytes(&lmb.memory, i));
560
561 /* reserve the sections we're already using */
562 for (i=0; i < lmb.reserved.cnt; i++)
563 reserve_bootmem(lmb.reserved.region[i].base,
564 lmb_size_bytes(&lmb.reserved, i));
565
566 for (i=0; i < lmb.memory.cnt; i++)
567 memory_present(0, lmb_start_pfn(&lmb.memory, i),
568 lmb_end_pfn(&lmb.memory, i));
569}
570
571/*
572 * paging_init() sets up the page tables - in fact we've already done this.
573 */
574void __init paging_init(void)
575{
576 unsigned long zones_size[MAX_NR_ZONES];
577 unsigned long zholes_size[MAX_NR_ZONES];
578 unsigned long total_ram = lmb_phys_mem_size();
579 unsigned long top_of_ram = lmb_end_of_DRAM();
580
581 printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
582 top_of_ram, total_ram);
583 printk(KERN_INFO "Memory hole size: %ldMB\n",
584 (top_of_ram - total_ram) >> 20);
585 /*
586 * All pages are DMA-able so we put them all in the DMA zone.
587 */
588 memset(zones_size, 0, sizeof(zones_size));
589 memset(zholes_size, 0, sizeof(zholes_size));
590
591 zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
592 zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
593
594 free_area_init_node(0, NODE_DATA(0), zones_size,
595 __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
596}
597#endif /* ! CONFIG_NEED_MULTIPLE_NODES */
598
599static struct kcore_list kcore_vmem;
600
601static int __init setup_kcore(void)
602{
603 int i;
604
605 for (i=0; i < lmb.memory.cnt; i++) {
606 unsigned long base, size;
607 struct kcore_list *kcore_mem;
608
609 base = lmb.memory.region[i].base;
610 size = lmb.memory.region[i].size;
611
612 /* GFP_ATOMIC to avoid might_sleep warnings during boot */
613 kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC);
614 if (!kcore_mem)
615 panic("mem_init: kmalloc failed\n");
616
617 kclist_add(kcore_mem, __va(base), size);
618 }
619
620 kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
621
622 return 0;
623}
624module_init(setup_kcore);
625
626void __init mem_init(void)
627{
628#ifdef CONFIG_NEED_MULTIPLE_NODES
629 int nid;
630#endif
631 pg_data_t *pgdat;
632 unsigned long i;
633 struct page *page;
634 unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
635
636 num_physpages = max_low_pfn; /* RAM is assumed contiguous */
637 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
638
639#ifdef CONFIG_NEED_MULTIPLE_NODES
640 for_each_online_node(nid) {
641 if (NODE_DATA(nid)->node_spanned_pages != 0) {
642 printk("freeing bootmem node %x\n", nid);
643 totalram_pages +=
644 free_all_bootmem_node(NODE_DATA(nid));
645 }
646 }
647#else
648 max_mapnr = num_physpages;
649 totalram_pages += free_all_bootmem();
650#endif
651
652 for_each_pgdat(pgdat) {
653 unsigned long flags;
654 pgdat_resize_lock(pgdat, &flags);
655 for (i = 0; i < pgdat->node_spanned_pages; i++) {
656 page = pgdat_page_nr(pgdat, i);
657 if (PageReserved(page))
658 reservedpages++;
659 }
660 pgdat_resize_unlock(pgdat, &flags);
661 }
662
663 codesize = (unsigned long)&_etext - (unsigned long)&_stext;
664 initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
665 datasize = (unsigned long)&_edata - (unsigned long)&__init_end;
666 bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
667
668 printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
669 "%luk reserved, %luk data, %luk bss, %luk init)\n",
670 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
671 num_physpages << (PAGE_SHIFT-10),
672 codesize >> 10,
673 reservedpages << (PAGE_SHIFT-10),
674 datasize >> 10,
675 bsssize >> 10,
676 initsize >> 10);
677
678 mem_init_done = 1;
679
680 /* Initialize the vDSO */
681 vdso_init();
682}
683
684/*
685 * This is called when a page has been modified by the kernel.
686 * It just marks the page as not i-cache clean. We do the i-cache
687 * flush later when the page is given to a user process, if necessary.
688 */
689void flush_dcache_page(struct page *page)
690{
691 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
692 return;
693 /* avoid an atomic op if possible */
694 if (test_bit(PG_arch_1, &page->flags))
695 clear_bit(PG_arch_1, &page->flags);
696}
697EXPORT_SYMBOL(flush_dcache_page);
698
699void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
700{
701 clear_page(page);
702
703 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
704 return;
705 /*
706 * We shouldnt have to do this, but some versions of glibc
707 * require it (ld.so assumes zero filled pages are icache clean)
708 * - Anton
709 */
710
711 /* avoid an atomic op if possible */
712 if (test_bit(PG_arch_1, &pg->flags))
713 clear_bit(PG_arch_1, &pg->flags);
714}
715EXPORT_SYMBOL(clear_user_page);
716
717void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
718 struct page *pg)
719{
720 copy_page(vto, vfrom);
721
722 /*
723 * We should be able to use the following optimisation, however
724 * there are two problems.
725 * Firstly a bug in some versions of binutils meant PLT sections
726 * were not marked executable.
727 * Secondly the first word in the GOT section is blrl, used
728 * to establish the GOT address. Until recently the GOT was
729 * not marked executable.
730 * - Anton
731 */
732#if 0
733 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
734 return;
735#endif
736
737 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
738 return;
739
740 /* avoid an atomic op if possible */
741 if (test_bit(PG_arch_1, &pg->flags))
742 clear_bit(PG_arch_1, &pg->flags);
743}
744
745void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
746 unsigned long addr, int len)
747{
748 unsigned long maddr;
749
750 maddr = (unsigned long)page_address(page) + (addr & ~PAGE_MASK);
751 flush_icache_range(maddr, maddr + len);
752}
753EXPORT_SYMBOL(flush_icache_user_range);
754
755/*
756 * This is called at the end of handling a user page fault, when the
757 * fault has been handled by updating a PTE in the linux page tables.
758 * We use it to preload an HPTE into the hash table corresponding to
759 * the updated linux PTE.
760 *
761 * This must always be called with the mm->page_table_lock held
762 */
763void update_mmu_cache(struct vm_area_struct *vma, unsigned long ea,
764 pte_t pte)
765{
766 unsigned long vsid;
767 void *pgdir;
768 pte_t *ptep;
769 int local = 0;
770 cpumask_t tmp;
771 unsigned long flags;
772
773 /* handle i-cache coherency */
774 if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
775 !cpu_has_feature(CPU_FTR_NOEXECUTE)) {
776 unsigned long pfn = pte_pfn(pte);
777 if (pfn_valid(pfn)) {
778 struct page *page = pfn_to_page(pfn);
779 if (!PageReserved(page)
780 && !test_bit(PG_arch_1, &page->flags)) {
781 __flush_dcache_icache(page_address(page));
782 set_bit(PG_arch_1, &page->flags);
783 }
784 }
785 }
786
787 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
788 if (!pte_young(pte))
789 return;
790
791 pgdir = vma->vm_mm->pgd;
792 if (pgdir == NULL)
793 return;
794
795 ptep = find_linux_pte(pgdir, ea);
796 if (!ptep)
797 return;
798
799 vsid = get_vsid(vma->vm_mm->context.id, ea);
800
801 local_irq_save(flags);
802 tmp = cpumask_of_cpu(smp_processor_id());
803 if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
804 local = 1;
805
806 __hash_page(ea, 0, vsid, ptep, 0x300, local);
807 local_irq_restore(flags);
808}
809
810void __iomem * reserve_phb_iospace(unsigned long size)
811{
812 void __iomem *virt_addr;
813
814 if (phbs_io_bot >= IMALLOC_BASE)
815 panic("reserve_phb_iospace(): phb io space overflow\n");
816
817 virt_addr = (void __iomem *) phbs_io_bot;
818 phbs_io_bot += size;
819
820 return virt_addr;
821}
822
823static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags)
824{
825 memset(addr, 0, kmem_cache_size(cache));
826}
827
828static const int pgtable_cache_size[2] = {
829 PTE_TABLE_SIZE, PMD_TABLE_SIZE
830};
831static const char *pgtable_cache_name[ARRAY_SIZE(pgtable_cache_size)] = {
832 "pgd_pte_cache", "pud_pmd_cache",
833};
834
835kmem_cache_t *pgtable_cache[ARRAY_SIZE(pgtable_cache_size)];
836
837void pgtable_cache_init(void)
838{
839 int i;
840
841 BUILD_BUG_ON(PTE_TABLE_SIZE != pgtable_cache_size[PTE_CACHE_NUM]);
842 BUILD_BUG_ON(PMD_TABLE_SIZE != pgtable_cache_size[PMD_CACHE_NUM]);
843 BUILD_BUG_ON(PUD_TABLE_SIZE != pgtable_cache_size[PUD_CACHE_NUM]);
844 BUILD_BUG_ON(PGD_TABLE_SIZE != pgtable_cache_size[PGD_CACHE_NUM]);
845
846 for (i = 0; i < ARRAY_SIZE(pgtable_cache_size); i++) {
847 int size = pgtable_cache_size[i];
848 const char *name = pgtable_cache_name[i];
849
850 pgtable_cache[i] = kmem_cache_create(name,
851 size, size,
852 SLAB_HWCACHE_ALIGN
853 | SLAB_MUST_HWCACHE_ALIGN,
854 zero_ctor,
855 NULL);
856 if (! pgtable_cache[i])
857 panic("pgtable_cache_init(): could not create %s!\n",
858 name);
859 }
860}
861
862pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
863 unsigned long size, pgprot_t vma_prot)
864{
865 if (ppc_md.phys_mem_access_prot)
866 return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
867
868 if (!page_is_ram(addr >> PAGE_SHIFT))
869 vma_prot = __pgprot(pgprot_val(vma_prot)
870 | _PAGE_GUARDED | _PAGE_NO_CACHE);
871 return vma_prot;
872}
873EXPORT_SYMBOL(phys_mem_access_prot);
874
875#ifdef CONFIG_MEMORY_HOTPLUG
876
877void online_page(struct page *page)
878{
879 ClearPageReserved(page);
880 free_cold_page(page);
881 totalram_pages++;
882 num_physpages++;
883}
884
885/*
886 * This works only for the non-NUMA case. Later, we'll need a lookup
887 * to convert from real physical addresses to nid, that doesn't use
888 * pfn_to_nid().
889 */
890int __devinit add_memory(u64 start, u64 size)
891{
892 struct pglist_data *pgdata = NODE_DATA(0);
893 struct zone *zone;
894 unsigned long start_pfn = start >> PAGE_SHIFT;
895 unsigned long nr_pages = size >> PAGE_SHIFT;
896
897 /* this should work for most non-highmem platforms */
898 zone = pgdata->node_zones;
899
900 return __add_pages(zone, start_pfn, nr_pages);
901
902 return 0;
903}
904
905/*
906 * First pass at this code will check to determine if the remove
907 * request is within the RMO. Do not allow removal within the RMO.
908 */
909int __devinit remove_memory(u64 start, u64 size)
910{
911 struct zone *zone;
912 unsigned long start_pfn, end_pfn, nr_pages;
913
914 start_pfn = start >> PAGE_SHIFT;
915 nr_pages = size >> PAGE_SHIFT;
916 end_pfn = start_pfn + nr_pages;
917
918 printk("%s(): Attempting to remove memoy in range "
919 "%lx to %lx\n", __func__, start, start+size);
920 /*
921 * check for range within RMO
922 */
923 zone = page_zone(pfn_to_page(start_pfn));
924
925 printk("%s(): memory will be removed from "
926 "the %s zone\n", __func__, zone->name);
927
928 /*
929 * not handling removing memory ranges that
930 * overlap multiple zones yet
931 */
932 if (end_pfn > (zone->zone_start_pfn + zone->spanned_pages))
933 goto overlap;
934
935 /* make sure it is NOT in RMO */
936 if ((start < lmb.rmo_size) || ((start+size) < lmb.rmo_size)) {
937 printk("%s(): range to be removed must NOT be in RMO!\n",
938 __func__);
939 goto in_rmo;
940 }
941
942 return __remove_pages(zone, start_pfn, nr_pages);
943
944overlap:
945 printk("%s(): memory range to be removed overlaps "
946 "multiple zones!!!\n", __func__);
947in_rmo:
948 return -1;
949}
950#endif /* CONFIG_MEMORY_HOTPLUG */
generated by cgit v1.2.2 (git 2.25.0) at 2025-03-10 21:25:31 -0400