]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - arch/sparc/mm/init_64.c
projects / mirror_ubuntu-eoan-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
UBUNTU: Ubuntu-5.3.0-29.31
[mirror_ubuntu-eoan-kernel.git] / arch / sparc / mm / init_64.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
b00dc837 2/*
1da177e4
LT		 3 * arch/sparc64/mm/init.c
4 *
5 * Copyright (C) 1996-1999 David S. Miller (davem@caip.rutgers.edu)
6 * Copyright (C) 1997-1999 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
7 */
8
cdd4f4c7 9#include <linux/extable.h>
1da177e4
LT		 10#include <linux/kernel.h>
11#include <linux/sched.h>
12#include <linux/string.h>
13#include <linux/init.h>
57c8a661 14#include <linux/memblock.h>
1da177e4
LT		 15#include <linux/mm.h>
16#include <linux/hugetlb.h>
1da177e4
LT		 17#include <linux/initrd.h>
18#include <linux/swap.h>
19#include <linux/pagemap.h>
c9cf5528 20#include <linux/poison.h>
1da177e4
LT		 21#include <linux/fs.h>
22#include <linux/seq_file.h>
05e14cb3 23#include <linux/kprobes.h>
1ac4f5eb 24#include <linux/cache.h>
13edad7a 25#include <linux/sort.h>
f6d4fb5c 26#include <linux/ioport.h>
5cbc3073 27#include <linux/percpu.h>
919ee677 28#include <linux/mmzone.h>
5a0e3ad6 29#include <linux/gfp.h>
1da177e4
LT		 30
31#include <asm/head.h>
1da177e4
LT		 32#include <asm/page.h>
33#include <asm/pgalloc.h>
34#include <asm/pgtable.h>
35#include <asm/oplib.h>
36#include <asm/iommu.h>
37#include <asm/io.h>
7c0f6ba6 38#include <linux/uaccess.h>
1da177e4
LT		 39#include <asm/mmu_context.h>
40#include <asm/tlbflush.h>
41#include <asm/dma.h>
42#include <asm/starfire.h>
43#include <asm/tlb.h>
44#include <asm/spitfire.h>
45#include <asm/sections.h>
517af332 46#include <asm/tsb.h>
481295f9 47#include <asm/hypervisor.h>
372b07bb 48#include <asm/prom.h>
5cbc3073 49#include <asm/mdesc.h>
3d5ae6b6 50#include <asm/cpudata.h>
59dec13b 51#include <asm/setup.h>
4f70f7a9 52#include <asm/irq.h>
1da177e4 53
27137e52 54#include "init_64.h"
9cc3a1ac 55
4f93d21d 56unsigned long kern_linear_pte_xor[4] __read_mostly;
494e5b6f 57static unsigned long page_cache4v_flag;
9cc3a1ac 58
4f93d21d
DM		 59/* A bitmap, two bits for every 256MB of physical memory. These two
60 * bits determine what page size we use for kernel linear
61 * translations. They form an index into kern_linear_pte_xor[]. The
62 * value in the indexed slot is XOR'd with the TLB miss virtual
63 * address to form the resulting TTE. The mapping is:
64 *
65 * 0 ==> 4MB
66 * 1 ==> 256MB
67 * 2 ==> 2GB
68 * 3 ==> 16GB
69 *
70 * All sun4v chips support 256MB pages. Only SPARC-T4 and later
71 * support 2GB pages, and hopefully future cpus will support the 16GB
72 * pages as well. For slots 2 and 3, we encode a 256MB TTE xor there
73 * if these larger page sizes are not supported by the cpu.
74 *
75 * It would be nice to determine this from the machine description
76 * 'cpu' properties, but we need to have this table setup before the
77 * MDESC is initialized.
9cc3a1ac 78 */
9cc3a1ac 79
d1acb421 80#ifndef CONFIG_DEBUG_PAGEALLOC
4f93d21d
DM		 81/* A special kernel TSB for 4MB, 256MB, 2GB and 16GB linear mappings.
82 * Space is allocated for this right after the trap table in
83 * arch/sparc64/kernel/head.S
2d9e2763
DM		 84 */
85extern struct tsb swapper_4m_tsb[KERNEL_TSB4M_NENTRIES];
d1acb421 86#endif
0dd5b7b0 87extern struct tsb swapper_tsb[KERNEL_TSB_NENTRIES];
d7744a09 88
ce33fdc5
DM		 89static unsigned long cpu_pgsz_mask;
90
d195b71b 91#define MAX_BANKS 1024
13edad7a 92
7c9503b8
GKH		 93static struct linux_prom64_registers pavail[MAX_BANKS];
94static int pavail_ents;
13edad7a 95
52708d69
NG		 96u64 numa_latency[MAX_NUMNODES][MAX_NUMNODES];
97
13edad7a
DM		 98static int cmp_p64(const void *a, const void *b)
99{
100 const struct linux_prom64_registers *x = a, *y = b;
101
102 if (x->phys_addr > y->phys_addr)
103 return 1;
104 if (x->phys_addr < y->phys_addr)
105 return -1;
106 return 0;
107}
108
109static void __init read_obp_memory(const char *property,
110 struct linux_prom64_registers *regs,
111 int *num_ents)
112{
8d125562 113 phandle node = prom_finddevice("/memory");
13edad7a
DM		 114 int prop_size = prom_getproplen(node, property);
115 int ents, ret, i;
116
117 ents = prop_size / sizeof(struct linux_prom64_registers);
118 if (ents > MAX_BANKS) {
119 prom_printf("The machine has more %s property entries than "
120 "this kernel can support (%d).\n",
121 property, MAX_BANKS);
122 prom_halt();
123 }
124
125 ret = prom_getproperty(node, property, (char *) regs, prop_size);
126 if (ret == -1) {
5da444aa
AM		 127 prom_printf("Couldn't get %s property from /memory.\n",
128 property);
13edad7a
DM		 129 prom_halt();
130 }
131
13edad7a
DM		 132 /* Sanitize what we got from the firmware, by page aligning
133 * everything.
134 */
135 for (i = 0; i < ents; i++) {
136 unsigned long base, size;
137
138 base = regs[i].phys_addr;
139 size = regs[i].reg_size;
10147570 140
13edad7a
DM		 141 size &= PAGE_MASK;
142 if (base & ~PAGE_MASK) {
143 unsigned long new_base = PAGE_ALIGN(base);
144
145 size -= new_base - base;
146 if ((long) size < 0L)
147 size = 0UL;
148 base = new_base;
149 }
0015d3d6
DM		 150 if (size == 0UL) {
151 /* If it is empty, simply get rid of it.
152 * This simplifies the logic of the other
153 * functions that process these arrays.
154 */
155 memmove(&regs[i], &regs[i + 1],
156 (ents - i - 1) * sizeof(regs[0]));
486ad10a 157 i--;
0015d3d6
DM		 158 ents--;
159 continue;
486ad10a 160 }
0015d3d6
DM		 161 regs[i].phys_addr = base;
162 regs[i].reg_size = size;
486ad10a
DM		 163 }
164
165 *num_ents = ents;
166
c9c10830 167 sort(regs, ents, sizeof(struct linux_prom64_registers),
13edad7a
DM		 168 cmp_p64, NULL);
169}
1da177e4 170
d1112018 171/* Kernel physical address base and size in bytes. */
1ac4f5eb
DM		 172unsigned long kern_base __read_mostly;
173unsigned long kern_size __read_mostly;
1da177e4 174
1da177e4
LT		 175/* Initial ramdisk setup */
176extern unsigned long sparc_ramdisk_image64;
177extern unsigned int sparc_ramdisk_image;
178extern unsigned int sparc_ramdisk_size;
179
1ac4f5eb 180struct page *mem_map_zero __read_mostly;
35802c0b 181EXPORT_SYMBOL(mem_map_zero);
1da177e4 182
0835ae0f
DM		 183unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly;
184
185unsigned long sparc64_kern_pri_context __read_mostly;
186unsigned long sparc64_kern_pri_nuc_bits __read_mostly;
187unsigned long sparc64_kern_sec_context __read_mostly;
188
64658743 189int num_kernel_image_mappings;
1da177e4 190
1da177e4
LT		 191#ifdef CONFIG_DEBUG_DCFLUSH
192atomic_t dcpage_flushes = ATOMIC_INIT(0);
193#ifdef CONFIG_SMP
194atomic_t dcpage_flushes_xcall = ATOMIC_INIT(0);
195#endif
196#endif
197
7a591cfe 198inline void flush_dcache_page_impl(struct page *page)
1da177e4 199{
7a591cfe 200 BUG_ON(tlb_type == hypervisor);
1da177e4
LT		 201#ifdef CONFIG_DEBUG_DCFLUSH
202 atomic_inc(&dcpage_flushes);
203#endif
204
205#ifdef DCACHE_ALIASING_POSSIBLE
206 __flush_dcache_page(page_address(page),
207 ((tlb_type == spitfire) &&
cb9f753a 208 page_mapping_file(page) != NULL));
1da177e4 209#else
cb9f753a 210 if (page_mapping_file(page) != NULL &&
1da177e4
LT		 211 tlb_type == spitfire)
212 __flush_icache_page(__pa(page_address(page)));
213#endif
214}
215
216#define PG_dcache_dirty PG_arch_1
22adb358
DM		 217#define PG_dcache_cpu_shift 32UL
218#define PG_dcache_cpu_mask \
219 ((1UL<<ilog2(roundup_pow_of_two(NR_CPUS)))-1UL)
1da177e4
LT		 220
221#define dcache_dirty_cpu(page) \
48b0e548 222 (((page)->flags >> PG_dcache_cpu_shift) & PG_dcache_cpu_mask)
1da177e4 223
d979f179 224static inline void set_dcache_dirty(struct page *page, int this_cpu)
1da177e4
LT		 225{
226 unsigned long mask = this_cpu;
48b0e548
DM		 227 unsigned long non_cpu_bits;
228
229 non_cpu_bits = ~(PG_dcache_cpu_mask << PG_dcache_cpu_shift);
230 mask = (mask << PG_dcache_cpu_shift) | (1UL << PG_dcache_dirty);
231
1da177e4
LT		 232 __asm__ __volatile__("1:\n\t"
233 "ldx [%2], %%g7\n\t"
234 "and %%g7, %1, %%g1\n\t"
235 "or %%g1, %0, %%g1\n\t"
236 "casx [%2], %%g7, %%g1\n\t"
237 "cmp %%g7, %%g1\n\t"
238 "bne,pn %%xcc, 1b\n\t"
b445e26c 239 " nop"
1da177e4
LT		 240 : /* no outputs */
241 : "r" (mask), "r" (non_cpu_bits), "r" (&page->flags)
242 : "g1", "g7");
243}
244
d979f179 245static inline void clear_dcache_dirty_cpu(struct page *page, unsigned long cpu)
1da177e4
LT		 246{
247 unsigned long mask = (1UL << PG_dcache_dirty);
248
249 __asm__ __volatile__("! test_and_clear_dcache_dirty\n"
250 "1:\n\t"
251 "ldx [%2], %%g7\n\t"
48b0e548 252 "srlx %%g7, %4, %%g1\n\t"
1da177e4
LT		 253 "and %%g1, %3, %%g1\n\t"
254 "cmp %%g1, %0\n\t"
255 "bne,pn %%icc, 2f\n\t"
256 " andn %%g7, %1, %%g1\n\t"
257 "casx [%2], %%g7, %%g1\n\t"
258 "cmp %%g7, %%g1\n\t"
259 "bne,pn %%xcc, 1b\n\t"
b445e26c 260 " nop\n"
1da177e4
LT		 261 "2:"
262 : /* no outputs */
263 : "r" (cpu), "r" (mask), "r" (&page->flags),
48b0e548
DM		 264 "i" (PG_dcache_cpu_mask),
265 "i" (PG_dcache_cpu_shift)
1da177e4
LT		 266 : "g1", "g7");
267}
268
517af332
DM		 269static inline void tsb_insert(struct tsb *ent, unsigned long tag, unsigned long pte)
270{
271 unsigned long tsb_addr = (unsigned long) ent;
272
3b3ab2eb 273 if (tlb_type == cheetah_plus || tlb_type == hypervisor)
517af332
DM		 274 tsb_addr = __pa(tsb_addr);
275
276 __tsb_insert(tsb_addr, tag, pte);
277}
278
c4bce90e 279unsigned long _PAGE_ALL_SZ_BITS __read_mostly;
c4bce90e 280
ff9aefbf 281static void flush_dcache(unsigned long pfn)
1da177e4 282{
ff9aefbf 283 struct page *page;
7a591cfe 284
ff9aefbf 285 page = pfn_to_page(pfn);
1a78cedb 286 if (page) {
7a591cfe 287 unsigned long pg_flags;
7a591cfe 288
ff9aefbf
SR		 289 pg_flags = page->flags;
290 if (pg_flags & (1UL << PG_dcache_dirty)) {
7a591cfe
DM		 291 int cpu = ((pg_flags >> PG_dcache_cpu_shift) &
292 PG_dcache_cpu_mask);
293 int this_cpu = get_cpu();
294
295 /* This is just to optimize away some function calls
296 * in the SMP case.
297 */
298 if (cpu == this_cpu)
299 flush_dcache_page_impl(page);
300 else
301 smp_flush_dcache_page_impl(page, cpu);
302
303 clear_dcache_dirty_cpu(page, cpu);
304
305 put_cpu();
306 }
1da177e4 307 }
ff9aefbf
SR		 308}
309
9e695d2e
DM		 310/* mm->context.lock must be held */
311static void __update_mmu_tsb_insert(struct mm_struct *mm, unsigned long tsb_index,
312 unsigned long tsb_hash_shift, unsigned long address,
313 unsigned long tte)
314{
315 struct tsb *tsb = mm->context.tsb_block[tsb_index].tsb;
316 unsigned long tag;
317
bcd896ba
DM		 318 if (unlikely(!tsb))
319 return;
320
9e695d2e
DM		 321 tsb += ((address >> tsb_hash_shift) &
322 (mm->context.tsb_block[tsb_index].tsb_nentries - 1UL));
323 tag = (address >> 22UL);
324 tsb_insert(tsb, tag, tte);
325}
326
c7d9f77d 327#ifdef CONFIG_HUGETLB_PAGE
8399e4b8
NG		 328static void __init add_huge_page_size(unsigned long size)
329{
330 unsigned int order;
331
332 if (size_to_hstate(size))
333 return;
334
335 order = ilog2(size) - PAGE_SHIFT;
336 hugetlb_add_hstate(order);
337}
338
339static int __init hugetlbpage_init(void)
340{
341 add_huge_page_size(1UL << HPAGE_64K_SHIFT);
342 add_huge_page_size(1UL << HPAGE_SHIFT);
343 add_huge_page_size(1UL << HPAGE_256MB_SHIFT);
344 add_huge_page_size(1UL << HPAGE_2GB_SHIFT);
345
346 return 0;
347}
348
349arch_initcall(hugetlbpage_init);
350
df7b2155
NG		 351static void __init pud_huge_patch(void)
352{
353 struct pud_huge_patch_entry *p;
354 unsigned long addr;
355
356 p = &__pud_huge_patch;
357 addr = p->addr;
358 *(unsigned int *)addr = p->insn;
359
360 __asm__ __volatile__("flush %0" : : "r" (addr));
361}
362
c7d9f77d
NG		 363static int __init setup_hugepagesz(char *string)
364{
365 unsigned long long hugepage_size;
366 unsigned int hugepage_shift;
367 unsigned short hv_pgsz_idx;
368 unsigned int hv_pgsz_mask;
369 int rc = 0;
370
371 hugepage_size = memparse(string, &string);
372 hugepage_shift = ilog2(hugepage_size);
373
374 switch (hugepage_shift) {
df7b2155
NG		 375 case HPAGE_16GB_SHIFT:
376 hv_pgsz_mask = HV_PGSZ_MASK_16GB;
377 hv_pgsz_idx = HV_PGSZ_IDX_16GB;
378 pud_huge_patch();
379 break;
85b1da7c
NG		 380 case HPAGE_2GB_SHIFT:
381 hv_pgsz_mask = HV_PGSZ_MASK_2GB;
382 hv_pgsz_idx = HV_PGSZ_IDX_2GB;
383 break;
c7d9f77d
NG		 384 case HPAGE_256MB_SHIFT:
385 hv_pgsz_mask = HV_PGSZ_MASK_256MB;
386 hv_pgsz_idx = HV_PGSZ_IDX_256MB;
387 break;
388 case HPAGE_SHIFT:
389 hv_pgsz_mask = HV_PGSZ_MASK_4MB;
390 hv_pgsz_idx = HV_PGSZ_IDX_4MB;
391 break;
dcd1912d
NG		 392 case HPAGE_64K_SHIFT:
393 hv_pgsz_mask = HV_PGSZ_MASK_64K;
394 hv_pgsz_idx = HV_PGSZ_IDX_64K;
395 break;
c7d9f77d
NG		 396 default:
397 hv_pgsz_mask = 0;
398 }
399
400 if ((hv_pgsz_mask & cpu_pgsz_mask) == 0U) {
f322980b
LH		 401 hugetlb_bad_size();
402 pr_err("hugepagesz=%llu not supported by MMU.\n",
c7d9f77d
NG		 403 hugepage_size);
404 goto out;
405 }
406
8399e4b8 407 add_huge_page_size(hugepage_size);
c7d9f77d
NG		 408 rc = 1;
409
410out:
411 return rc;
412}
413__setup("hugepagesz=", setup_hugepagesz);
414#endif /* CONFIG_HUGETLB_PAGE */
415
4b3073e1 416void update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t *ptep)
ff9aefbf
SR		 417{
418 struct mm_struct *mm;
bcd896ba 419 unsigned long flags;
df7b2155 420 bool is_huge_tsb;
4b3073e1 421 pte_t pte = *ptep;
ff9aefbf
SR		 422
423 if (tlb_type != hypervisor) {
424 unsigned long pfn = pte_pfn(pte);
425
426 if (pfn_valid(pfn))
427 flush_dcache(pfn);
428 }
bd40791e
DM		 429
430 mm = vma->vm_mm;
7a1ac526 431
18f38132
DM		 432 /* Don't insert a non-valid PTE into the TSB, we'll deadlock. */
433 if (!pte_accessible(mm, pte))
434 return;
435
7a1ac526
DM		 436 spin_lock_irqsave(&mm->context.lock, flags);
437
df7b2155 438 is_huge_tsb = false;
9e695d2e 439#if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
df7b2155
NG		 440 if (mm->context.hugetlb_pte_count || mm->context.thp_pte_count) {
441 unsigned long hugepage_size = PAGE_SIZE;
442
443 if (is_vm_hugetlb_page(vma))
444 hugepage_size = huge_page_size(hstate_vma(vma));
445
446 if (hugepage_size >= PUD_SIZE) {
447 unsigned long mask = 0x1ffc00000UL;
448
449 /* Transfer bits [32:22] from address to resolve
450 * at 4M granularity.
451 */
452 pte_val(pte) &= ~mask;
453 pte_val(pte) |= (address & mask);
454 } else if (hugepage_size >= PMD_SIZE) {
455 /* We are fabricating 8MB pages using 4MB
456 * real hw pages.
457 */
458 pte_val(pte) |= (address & (1UL << REAL_HPAGE_SHIFT));
459 }
460
461 if (hugepage_size >= PMD_SIZE) {
462 __update_mmu_tsb_insert(mm, MM_TSB_HUGE,
463 REAL_HPAGE_SHIFT, address, pte_val(pte));
464 is_huge_tsb = true;
465 }
466 }
dcc1e8dd 467#endif
df7b2155 468 if (!is_huge_tsb)
bcd896ba
DM		 469 __update_mmu_tsb_insert(mm, MM_TSB_BASE, PAGE_SHIFT,
470 address, pte_val(pte));
7a1ac526
DM		 471
472 spin_unlock_irqrestore(&mm->context.lock, flags);
1da177e4
LT		 473}
474
475void flush_dcache_page(struct page *page)
476{
a9546f59
DM		 477 struct address_space *mapping;
478 int this_cpu;
1da177e4 479
7a591cfe
DM		 480 if (tlb_type == hypervisor)
481 return;
482
a9546f59
DM		 483 /* Do not bother with the expensive D-cache flush if it
484 * is merely the zero page. The 'bigcore' testcase in GDB
485 * causes this case to run millions of times.
486 */
487 if (page == ZERO_PAGE(0))
488 return;
489
490 this_cpu = get_cpu();
491
cb9f753a 492 mapping = page_mapping_file(page);
1da177e4 493 if (mapping && !mapping_mapped(mapping)) {
a9546f59 494 int dirty = test_bit(PG_dcache_dirty, &page->flags);
1da177e4 495 if (dirty) {
a9546f59
DM		 496 int dirty_cpu = dcache_dirty_cpu(page);
497
1da177e4
LT		 498 if (dirty_cpu == this_cpu)
499 goto out;
500 smp_flush_dcache_page_impl(page, dirty_cpu);
501 }
502 set_dcache_dirty(page, this_cpu);
503 } else {
504 /* We could delay the flush for the !page_mapping
505 * case too. But that case is for exec env/arg
506 * pages and those are %99 certainly going to get
507 * faulted into the tlb (and thus flushed) anyways.
508 */
509 flush_dcache_page_impl(page);
510 }
511
512out:
513 put_cpu();
514}
917c3660 515EXPORT_SYMBOL(flush_dcache_page);
1da177e4 516
05e14cb3 517void __kprobes flush_icache_range(unsigned long start, unsigned long end)
1da177e4 518{
a43fe0e7 519 /* Cheetah and Hypervisor platform cpus have coherent I-cache. */
1da177e4
LT		 520 if (tlb_type == spitfire) {
521 unsigned long kaddr;
522
a94aa253
DM		 523 /* This code only runs on Spitfire cpus so this is
524 * why we can assume _PAGE_PADDR_4U.
525 */
526 for (kaddr = start; kaddr < end; kaddr += PAGE_SIZE) {
527 unsigned long paddr, mask = _PAGE_PADDR_4U;
528
529 if (kaddr >= PAGE_OFFSET)
530 paddr = kaddr & mask;
531 else {
532 pgd_t *pgdp = pgd_offset_k(kaddr);
533 pud_t *pudp = pud_offset(pgdp, kaddr);
534 pmd_t *pmdp = pmd_offset(pudp, kaddr);
535 pte_t *ptep = pte_offset_kernel(pmdp, kaddr);
536
537 paddr = pte_val(*ptep) & mask;
538 }
539 __flush_icache_page(paddr);
540 }
1da177e4
LT		 541 }
542}
917c3660 543EXPORT_SYMBOL(flush_icache_range);
1da177e4 544
1da177e4
LT		 545void mmu_info(struct seq_file *m)
546{
ce33fdc5
DM		 547 static const char *pgsz_strings[] = {
548 "8K", "64K", "512K", "4MB", "32MB",
549 "256MB", "2GB", "16GB",
550 };
551 int i, printed;
552
1da177e4
LT		 553 if (tlb_type == cheetah)
554 seq_printf(m, "MMU Type\t: Cheetah\n");
555 else if (tlb_type == cheetah_plus)
556 seq_printf(m, "MMU Type\t: Cheetah+\n");
557 else if (tlb_type == spitfire)
558 seq_printf(m, "MMU Type\t: Spitfire\n");
a43fe0e7
DM		 559 else if (tlb_type == hypervisor)
560 seq_printf(m, "MMU Type\t: Hypervisor (sun4v)\n");
1da177e4
LT		 561 else
562 seq_printf(m, "MMU Type\t: ???\n");
563
ce33fdc5
DM		 564 seq_printf(m, "MMU PGSZs\t: ");
565 printed = 0;
566 for (i = 0; i < ARRAY_SIZE(pgsz_strings); i++) {
567 if (cpu_pgsz_mask & (1UL << i)) {
568 seq_printf(m, "%s%s",
569 printed ? "," : "", pgsz_strings[i]);
570 printed++;
571 }
572 }
573 seq_putc(m, '\n');
574
1da177e4
LT		 575#ifdef CONFIG_DEBUG_DCFLUSH
576 seq_printf(m, "DCPageFlushes\t: %d\n",
577 atomic_read(&dcpage_flushes));
578#ifdef CONFIG_SMP
579 seq_printf(m, "DCPageFlushesXC\t: %d\n",
580 atomic_read(&dcpage_flushes_xcall));
581#endif /* CONFIG_SMP */
582#endif /* CONFIG_DEBUG_DCFLUSH */
583}
584
a94aa253
DM		 585struct linux_prom_translation prom_trans[512] __read_mostly;
586unsigned int prom_trans_ents __read_mostly;
587
1da177e4
LT		 588unsigned long kern_locked_tte_data;
589
c9c10830
DM		 590/* The obp translations are saved based on 8k pagesize, since obp can
591 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
74bf4312 592 * HI_OBP_ADDRESS range are handled in ktlb.S.
c9c10830 593 */
5085b4a5
DM		 594static inline int in_obp_range(unsigned long vaddr)
595{
596 return (vaddr >= LOW_OBP_ADDRESS &&
597 vaddr < HI_OBP_ADDRESS);
598}
599
c9c10830 600static int cmp_ptrans(const void *a, const void *b)
405599bd 601{
c9c10830 602 const struct linux_prom_translation *x = a, *y = b;
405599bd 603
c9c10830
DM		 604 if (x->virt > y->virt)
605 return 1;
606 if (x->virt < y->virt)
607 return -1;
608 return 0;
405599bd
DM		 609}
610
c9c10830 611/* Read OBP translations property into 'prom_trans[]'. */
9ad98c5b 612static void __init read_obp_translations(void)
405599bd 613{
c9c10830 614 int n, node, ents, first, last, i;
1da177e4
LT		 615
616 node = prom_finddevice("/virtual-memory");
617 n = prom_getproplen(node, "translations");
405599bd 618 if (unlikely(n == 0 || n == -1)) {
b206fc4c 619 prom_printf("prom_mappings: Couldn't get size.\n");
1da177e4
LT		 620 prom_halt();
621 }
405599bd 622 if (unlikely(n > sizeof(prom_trans))) {
5da444aa 623 prom_printf("prom_mappings: Size %d is too big.\n", n);
1da177e4
LT		 624 prom_halt();
625 }
405599bd 626
b206fc4c 627 if ((n = prom_getproperty(node, "translations",
405599bd
DM		 628 (char *)&prom_trans[0],
629 sizeof(prom_trans))) == -1) {
b206fc4c 630 prom_printf("prom_mappings: Couldn't get property.\n");
1da177e4
LT		 631 prom_halt();
632 }
9ad98c5b 633
b206fc4c 634 n = n / sizeof(struct linux_prom_translation);
9ad98c5b 635
c9c10830
DM		 636 ents = n;
637
638 sort(prom_trans, ents, sizeof(struct linux_prom_translation),
639 cmp_ptrans, NULL);
640
641 /* Now kick out all the non-OBP entries. */
642 for (i = 0; i < ents; i++) {
643 if (in_obp_range(prom_trans[i].virt))
644 break;
645 }
646 first = i;
647 for (; i < ents; i++) {
648 if (!in_obp_range(prom_trans[i].virt))
649 break;
650 }
651 last = i;
652
653 for (i = 0; i < (last - first); i++) {
654 struct linux_prom_translation *src = &prom_trans[i + first];
655 struct linux_prom_translation *dest = &prom_trans[i];
656
657 *dest = *src;
658 }
659 for (; i < ents; i++) {
660 struct linux_prom_translation *dest = &prom_trans[i];
661 dest->virt = dest->size = dest->data = 0x0UL;
662 }
663
664 prom_trans_ents = last - first;
665
666 if (tlb_type == spitfire) {
667 /* Clear diag TTE bits. */
668 for (i = 0; i < prom_trans_ents; i++)
669 prom_trans[i].data &= ~0x0003fe0000000000UL;
670 }
f4142cba
DM		 671
672 /* Force execute bit on. */
673 for (i = 0; i < prom_trans_ents; i++)
674 prom_trans[i].data |= (tlb_type == hypervisor ?
675 _PAGE_EXEC_4V : _PAGE_EXEC_4U);
405599bd 676}
1da177e4 677
d82ace7d
DM		 678static void __init hypervisor_tlb_lock(unsigned long vaddr,
679 unsigned long pte,
680 unsigned long mmu)
681{
7db35f31
DM		 682 unsigned long ret = sun4v_mmu_map_perm_addr(vaddr, 0, pte, mmu);
683
684 if (ret != 0) {
5da444aa 685 prom_printf("hypervisor_tlb_lock[%lx:%x:%lx:%lx]: "
7db35f31 686 "errors with %lx\n", vaddr, 0, pte, mmu, ret);
12e126ad
DM		 687 prom_halt();
688 }
d82ace7d
DM		 689}
690
c4bce90e
DM		 691static unsigned long kern_large_tte(unsigned long paddr);
692
898cf0ec 693static void __init remap_kernel(void)
405599bd
DM		 694{
695 unsigned long phys_page, tte_vaddr, tte_data;
64658743 696 int i, tlb_ent = sparc64_highest_locked_tlbent();
405599bd 697
1da177e4 698 tte_vaddr = (unsigned long) KERNBASE;
0eef331a 699 phys_page = (prom_boot_mapping_phys_low >> ILOG2_4MB) << ILOG2_4MB;
c4bce90e 700 tte_data = kern_large_tte(phys_page);
1da177e4
LT		 701
702 kern_locked_tte_data = tte_data;
703
d82ace7d
DM		 704 /* Now lock us into the TLBs via Hypervisor or OBP. */
705 if (tlb_type == hypervisor) {
64658743 706 for (i = 0; i < num_kernel_image_mappings; i++) {
d82ace7d
DM		 707 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_DMMU);
708 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_IMMU);
64658743
DM		 709 tte_vaddr += 0x400000;
710 tte_data += 0x400000;
d82ace7d
DM		 711 }
712 } else {
64658743
DM		 713 for (i = 0; i < num_kernel_image_mappings; i++) {
714 prom_dtlb_load(tlb_ent - i, tte_data, tte_vaddr);
715 prom_itlb_load(tlb_ent - i, tte_data, tte_vaddr);
716 tte_vaddr += 0x400000;
717 tte_data += 0x400000;
d82ace7d 718 }
64658743 719 sparc64_highest_unlocked_tlb_ent = tlb_ent - i;
1da177e4 720 }
0835ae0f
DM		 721 if (tlb_type == cheetah_plus) {
722 sparc64_kern_pri_context = (CTX_CHEETAH_PLUS_CTX0 |
723 CTX_CHEETAH_PLUS_NUC);
724 sparc64_kern_pri_nuc_bits = CTX_CHEETAH_PLUS_NUC;
725 sparc64_kern_sec_context = CTX_CHEETAH_PLUS_CTX0;
726 }
405599bd 727}
1da177e4 728
405599bd 729
c9c10830 730static void __init inherit_prom_mappings(void)
9ad98c5b 731{
405599bd 732 /* Now fixup OBP's idea about where we really are mapped. */
3c62a2d3 733 printk("Remapping the kernel... ");
405599bd 734 remap_kernel();
3c62a2d3 735 printk("done.\n");
1da177e4
LT		 736}
737
1da177e4
LT		 738void prom_world(int enter)
739{
1da177e4 740 if (!enter)
dff933da 741 set_fs(get_fs());
1da177e4 742
3487d1d4 743 __asm__ __volatile__("flushw");
1da177e4
LT		 744}
745
1da177e4
LT		 746void __flush_dcache_range(unsigned long start, unsigned long end)
747{
748 unsigned long va;
749
750 if (tlb_type == spitfire) {
751 int n = 0;
752
753 for (va = start; va < end; va += 32) {
754 spitfire_put_dcache_tag(va & 0x3fe0, 0x0);
755 if (++n >= 512)
756 break;
757 }
a43fe0e7 758 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1da177e4
LT		 759 start = __pa(start);
760 end = __pa(end);
761 for (va = start; va < end; va += 32)
762 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
763 "membar #Sync"
764 : /* no outputs */
765 : "r" (va),
766 "i" (ASI_DCACHE_INVALIDATE));
767 }
768}
917c3660 769EXPORT_SYMBOL(__flush_dcache_range);
1da177e4 770
85f1e1f6
DM		 771/* get_new_mmu_context() uses "cache + 1". */
772DEFINE_SPINLOCK(ctx_alloc_lock);
c4415235 773unsigned long tlb_context_cache = CTX_FIRST_VERSION;
85f1e1f6
DM		 774#define MAX_CTX_NR (1UL << CTX_NR_BITS)
775#define CTX_BMAP_SLOTS BITS_TO_LONGS(MAX_CTX_NR)
776DECLARE_BITMAP(mmu_context_bmap, MAX_CTX_NR);
7a5b4bbf 777DEFINE_PER_CPU(struct mm_struct *, per_cpu_secondary_mm) = {0};
85f1e1f6 778
a0582f26
PT		 779static void mmu_context_wrap(void)
780{
781 unsigned long old_ver = tlb_context_cache & CTX_VERSION_MASK;
782 unsigned long new_ver, new_ctx, old_ctx;
783 struct mm_struct *mm;
784 int cpu;
785
786 bitmap_zero(mmu_context_bmap, 1 << CTX_NR_BITS);
787
788 /* Reserve kernel context */
789 set_bit(0, mmu_context_bmap);
790
791 new_ver = (tlb_context_cache & CTX_VERSION_MASK) + CTX_FIRST_VERSION;
792 if (unlikely(new_ver == 0))
793 new_ver = CTX_FIRST_VERSION;
794 tlb_context_cache = new_ver;
795
796 /*
797 * Make sure that any new mm that are added into per_cpu_secondary_mm,
798 * are going to go through get_new_mmu_context() path.
799 */
800 mb();
801
802 /*
803 * Updated versions to current on those CPUs that had valid secondary
804 * contexts
805 */
806 for_each_online_cpu(cpu) {
807 /*
808 * If a new mm is stored after we took this mm from the array,
809 * it will go into get_new_mmu_context() path, because we
810 * already bumped the version in tlb_context_cache.
811 */
812 mm = per_cpu(per_cpu_secondary_mm, cpu);
813
814 if (unlikely(!mm || mm == &init_mm))
815 continue;
816
817 old_ctx = mm->context.sparc64_ctx_val;
818 if (likely((old_ctx & CTX_VERSION_MASK) == old_ver)) {
819 new_ctx = (old_ctx & ~CTX_VERSION_MASK) | new_ver;
820 set_bit(new_ctx & CTX_NR_MASK, mmu_context_bmap);
821 mm->context.sparc64_ctx_val = new_ctx;
822 }
823 }
824}
825
1da177e4
LT		 826/* Caller does TLB context flushing on local CPU if necessary.
827 * The caller also ensures that CTX_VALID(mm->context) is false.
828 *
829 * We must be careful about boundary cases so that we never
830 * let the user have CTX 0 (nucleus) or we ever use a CTX
831 * version of zero (and thus NO_CONTEXT would not be caught
832 * by version mis-match tests in mmu_context.h).
a0663a79
DM		 833 *
834 * Always invoked with interrupts disabled.
1da177e4
LT		 835 */
836void get_new_mmu_context(struct mm_struct *mm)
837{
838 unsigned long ctx, new_ctx;
839 unsigned long orig_pgsz_bits;
1da177e4 840
07df8418 841 spin_lock(&ctx_alloc_lock);
a0582f26
PT		 842retry:
843 /* wrap might have happened, test again if our context became valid */
844 if (unlikely(CTX_VALID(mm->context)))
845 goto out;
1da177e4
LT		 846 orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
847 ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
848 new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
849 if (new_ctx >= (1 << CTX_NR_BITS)) {
850 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1);
851 if (new_ctx >= ctx) {
a0582f26
PT		 852 mmu_context_wrap();
853 goto retry;
1da177e4
LT		 854 }
855 }
58897485
PT		 856 if (mm->context.sparc64_ctx_val)
857 cpumask_clear(mm_cpumask(mm));
1da177e4
LT		 858 mmu_context_bmap[new_ctx>>6] |= (1UL << (new_ctx & 63));
859 new_ctx |= (tlb_context_cache & CTX_VERSION_MASK);
1da177e4
LT		 860 tlb_context_cache = new_ctx;
861 mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
a0582f26 862out:
07df8418 863 spin_unlock(&ctx_alloc_lock);
1da177e4
LT		 864}
865
919ee677
DM		 866static int numa_enabled = 1;
867static int numa_debug;
868
869static int __init early_numa(char *p)
1da177e4 870{
919ee677
DM		 871 if (!p)
872 return 0;
873
874 if (strstr(p, "off"))
875 numa_enabled = 0;
d1112018 876
919ee677
DM		 877 if (strstr(p, "debug"))
878 numa_debug = 1;
d1112018 879
919ee677 880 return 0;
d1112018 881}
919ee677
DM		 882early_param("numa", early_numa);
883
884#define numadbg(f, a...) \
885do { if (numa_debug) \
886 printk(KERN_INFO f, ## a); \
887} while (0)
d1112018 888
4e82c9a6
DM		 889static void __init find_ramdisk(unsigned long phys_base)
890{
891#ifdef CONFIG_BLK_DEV_INITRD
892 if (sparc_ramdisk_image || sparc_ramdisk_image64) {
893 unsigned long ramdisk_image;
894
895 /* Older versions of the bootloader only supported a
896 * 32-bit physical address for the ramdisk image
897 * location, stored at sparc_ramdisk_image. Newer
898 * SILO versions set sparc_ramdisk_image to zero and
899 * provide a full 64-bit physical address at
900 * sparc_ramdisk_image64.
901 */
902 ramdisk_image = sparc_ramdisk_image;
903 if (!ramdisk_image)
904 ramdisk_image = sparc_ramdisk_image64;
905
906 /* Another bootloader quirk. The bootloader normalizes
907 * the physical address to KERNBASE, so we have to
908 * factor that back out and add in the lowest valid
909 * physical page address to get the true physical address.
910 */
911 ramdisk_image -= KERNBASE;
912 ramdisk_image += phys_base;
913
919ee677
DM		 914 numadbg("Found ramdisk at physical address 0x%lx, size %u\n",
915 ramdisk_image, sparc_ramdisk_size);
916
4e82c9a6
DM		 917 initrd_start = ramdisk_image;
918 initrd_end = ramdisk_image + sparc_ramdisk_size;
3b2a7e23 919
95f72d1e 920 memblock_reserve(initrd_start, sparc_ramdisk_size);
d45100f7
DM		 921
922 initrd_start += PAGE_OFFSET;
923 initrd_end += PAGE_OFFSET;
4e82c9a6
DM		 924 }
925#endif
926}
927
919ee677
DM		 928struct node_mem_mask {
929 unsigned long mask;
1537b26d 930 unsigned long match;
919ee677
DM		 931};
932static struct node_mem_mask node_masks[MAX_NUMNODES];
933static int num_node_masks;
934
48d37216
SR		 935#ifdef CONFIG_NEED_MULTIPLE_NODES
936
1537b26d
PT		 937struct mdesc_mlgroup {
938 u64 node;
939 u64 latency;
940 u64 match;
941 u64 mask;
942};
943
944static struct mdesc_mlgroup *mlgroups;
945static int num_mlgroups;
946
919ee677
DM		 947int numa_cpu_lookup_table[NR_CPUS];
948cpumask_t numa_cpumask_lookup_table[MAX_NUMNODES];
949
919ee677
DM		 950struct mdesc_mblock {
951 u64 base;
952 u64 size;
953 u64 offset; /* RA-to-PA */
954};
955static struct mdesc_mblock *mblocks;
956static int num_mblocks;
957
1537b26d 958static struct mdesc_mblock * __init addr_to_mblock(unsigned long addr)
919ee677 959{
1537b26d 960 struct mdesc_mblock *m = NULL;
919ee677
DM		 961 int i;
962
963 for (i = 0; i < num_mblocks; i++) {
1537b26d 964 m = &mblocks[i];
919ee677
DM		 965
966 if (addr >= m->base &&
967 addr < (m->base + m->size)) {
919ee677
DM		 968 break;
969 }
970 }
1537b26d
PT		 971
972 return m;
919ee677
DM		 973}
974
1537b26d 975static u64 __init memblock_nid_range_sun4u(u64 start, u64 end, int *nid)
919ee677 976{
1537b26d 977 int prev_nid, new_nid;
919ee677 978
98fa15f3 979 prev_nid = NUMA_NO_NODE;
1537b26d
PT		 980 for ( ; start < end; start += PAGE_SIZE) {
981 for (new_nid = 0; new_nid < num_node_masks; new_nid++) {
982 struct node_mem_mask *p = &node_masks[new_nid];
919ee677 983
1537b26d 984 if ((start & p->mask) == p->match) {
98fa15f3 985 if (prev_nid == NUMA_NO_NODE)
1537b26d
PT		 986 prev_nid = new_nid;
987 break;
988 }
74a5ed5c 989 }
1537b26d
PT		 990
991 if (new_nid == num_node_masks) {
992 prev_nid = 0;
993 WARN_ONCE(1, "addr[%Lx] doesn't match a NUMA node rule. Some memory will be owned by node 0.",
994 start);
995 break;
996 }
997
998 if (prev_nid != new_nid)
999 break;
74a5ed5c 1000 }
1537b26d 1001 *nid = prev_nid;
74a5ed5c 1002
1537b26d 1003 return start > end ? end : start;
919ee677
DM		 1004}
1005
87a349f9 1006static u64 __init memblock_nid_range(u64 start, u64 end, int *nid)
919ee677 1007{
1537b26d
PT		 1008 u64 ret_end, pa_start, m_mask, m_match, m_end;
1009 struct mdesc_mblock *mblock;
1010 int _nid, i;
1011
1012 if (tlb_type != hypervisor)
1013 return memblock_nid_range_sun4u(start, end, nid);
1014
1015 mblock = addr_to_mblock(start);
1016 if (!mblock) {
1017 WARN_ONCE(1, "memblock_nid_range: Can't find mblock addr[%Lx]",
1018 start);
1019
1020 _nid = 0;
1021 ret_end = end;
1022 goto done;
1023 }
1024
1025 pa_start = start + mblock->offset;
1026 m_match = 0;
1027 m_mask = 0;
919ee677 1028
1537b26d
PT		 1029 for (_nid = 0; _nid < num_node_masks; _nid++) {
1030 struct node_mem_mask *const m = &node_masks[_nid];
1031
1032 if ((pa_start & m->mask) == m->match) {
1033 m_match = m->match;
1034 m_mask = m->mask;
919ee677 1035 break;
1537b26d 1036 }
919ee677
DM		 1037 }
1038
1537b26d
PT		 1039 if (num_node_masks == _nid) {
1040 /* We could not find NUMA group, so default to 0, but lets
1041 * search for latency group, so we could calculate the correct
1042 * end address that we return
1043 */
1044 _nid = 0;
1045
1046 for (i = 0; i < num_mlgroups; i++) {
1047 struct mdesc_mlgroup *const m = &mlgroups[i];
c918dcce 1048
1537b26d
PT		 1049 if ((pa_start & m->mask) == m->match) {
1050 m_match = m->match;
1051 m_mask = m->mask;
1052 break;
1053 }
1054 }
1055
1056 if (i == num_mlgroups) {
1057 WARN_ONCE(1, "memblock_nid_range: Can't find latency group addr[%Lx]",
1058 start);
1059
1060 ret_end = end;
1061 goto done;
1062 }
1063 }
1064
1065 /*
1066 * Each latency group has match and mask, and each memory block has an
1067 * offset. An address belongs to a latency group if its address matches
1068 * the following formula: ((addr + offset) & mask) == match
1069 * It is, however, slow to check every single page if it matches a
1070 * particular latency group. As optimization we calculate end value by
1071 * using bit arithmetics.
1072 */
1073 m_end = m_match + (1ul << __ffs(m_mask)) - mblock->offset;
1074 m_end += pa_start & ~((1ul << fls64(m_mask)) - 1);
1075 ret_end = m_end > end ? end : m_end;
1076
1077done:
1078 *nid = _nid;
1079 return ret_end;
919ee677 1080}
919ee677
DM		 1081#endif
1082
1083/* This must be invoked after performing all of the necessary
2a4814df 1084 * memblock_set_node() calls for 'nid'. We need to be able to get
919ee677 1085 * correct data from get_pfn_range_for_nid().
f1cfdb55 1086 */
919ee677
DM		 1087static void __init allocate_node_data(int nid)
1088{
919ee677 1089 struct pglist_data *p;
aa6f0790 1090 unsigned long start_pfn, end_pfn;
919ee677 1091#ifdef CONFIG_NEED_MULTIPLE_NODES
aa6f0790 1092
b63a07d6
MR		 1093 NODE_DATA(nid) = memblock_alloc_node(sizeof(struct pglist_data),
1094 SMP_CACHE_BYTES, nid);
1095 if (!NODE_DATA(nid)) {
919ee677
DM		 1096 prom_printf("Cannot allocate pglist_data for nid[%d]\n", nid);
1097 prom_halt();
1098 }
919ee677 1099
625d693e 1100 NODE_DATA(nid)->node_id = nid;
919ee677
DM		 1101#endif
1102
1103 p = NODE_DATA(nid);
1104
1105 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
1106 p->node_start_pfn = start_pfn;
1107 p->node_spanned_pages = end_pfn - start_pfn;
919ee677
DM		 1108}
1109
1110static void init_node_masks_nonnuma(void)
d1112018 1111{
48d37216 1112#ifdef CONFIG_NEED_MULTIPLE_NODES
1da177e4 1113 int i;
48d37216 1114#endif
1da177e4 1115
919ee677 1116 numadbg("Initializing tables for non-numa.\n");
6fc5bae7 1117
1537b26d
PT		 1118 node_masks[0].mask = 0;
1119 node_masks[0].match = 0;
919ee677 1120 num_node_masks = 1;
d1112018 1121
48d37216 1122#ifdef CONFIG_NEED_MULTIPLE_NODES
919ee677
DM		 1123 for (i = 0; i < NR_CPUS; i++)
1124 numa_cpu_lookup_table[i] = 0;
1da177e4 1125
fb1fece5 1126 cpumask_setall(&numa_cpumask_lookup_table[0]);
48d37216 1127#endif
919ee677
DM		 1128}
1129
1130#ifdef CONFIG_NEED_MULTIPLE_NODES
1131struct pglist_data *node_data[MAX_NUMNODES];
1132
1133EXPORT_SYMBOL(numa_cpu_lookup_table);
1134EXPORT_SYMBOL(numa_cpumask_lookup_table);
1135EXPORT_SYMBOL(node_data);
1136
919ee677
DM		 1137static int scan_pio_for_cfg_handle(struct mdesc_handle *md, u64 pio,
1138 u32 cfg_handle)
1139{
1140 u64 arc;
1141
1142 mdesc_for_each_arc(arc, md, pio, MDESC_ARC_TYPE_FWD) {
1143 u64 target = mdesc_arc_target(md, arc);
1144 const u64 *val;
1145
1146 val = mdesc_get_property(md, target,
1147 "cfg-handle", NULL);
1148 if (val && *val == cfg_handle)
1149 return 0;
1150 }
1151 return -ENODEV;
1152}
1153
1154static int scan_arcs_for_cfg_handle(struct mdesc_handle *md, u64 grp,
1155 u32 cfg_handle)
1156{
1157 u64 arc, candidate, best_latency = ~(u64)0;
1158
1159 candidate = MDESC_NODE_NULL;
1160 mdesc_for_each_arc(arc, md, grp, MDESC_ARC_TYPE_FWD) {
1161 u64 target = mdesc_arc_target(md, arc);
1162 const char *name = mdesc_node_name(md, target);
1163 const u64 *val;
1164
1165 if (strcmp(name, "pio-latency-group"))
1166 continue;
1167
1168 val = mdesc_get_property(md, target, "latency", NULL);
1169 if (!val)
1170 continue;
1171
1172 if (*val < best_latency) {
1173 candidate = target;
1174 best_latency = *val;
1175 }
1176 }
1177
1178 if (candidate == MDESC_NODE_NULL)
1179 return -ENODEV;
1180
1181 return scan_pio_for_cfg_handle(md, candidate, cfg_handle);
1182}
1183
1184int of_node_to_nid(struct device_node *dp)
1185{
1186 const struct linux_prom64_registers *regs;
1187 struct mdesc_handle *md;
1188 u32 cfg_handle;
1189 int count, nid;
1190 u64 grp;
1191
072bd413
DM		 1192 /* This is the right thing to do on currently supported
1193 * SUN4U NUMA platforms as well, as the PCI controller does
1194 * not sit behind any particular memory controller.
1195 */
919ee677
DM		 1196 if (!mlgroups)
1197 return -1;
1198
1199 regs = of_get_property(dp, "reg", NULL);
1200 if (!regs)
1201 return -1;
1202
1203 cfg_handle = (regs->phys_addr >> 32UL) & 0x0fffffff;
1204
1205 md = mdesc_grab();
1206
1207 count = 0;
98fa15f3 1208 nid = NUMA_NO_NODE;
919ee677
DM		 1209 mdesc_for_each_node_by_name(md, grp, "group") {
1210 if (!scan_arcs_for_cfg_handle(md, grp, cfg_handle)) {
1211 nid = count;
1212 break;
1213 }
1214 count++;
1215 }
1216
1217 mdesc_release(md);
1218
1219 return nid;
1220}
1221
01c45381 1222static void __init add_node_ranges(void)
919ee677 1223{
08b84798 1224 struct memblock_region *reg;
cd429ce2
PT		 1225 unsigned long prev_max;
1226
1227memblock_resized:
1228 prev_max = memblock.memory.max;
919ee677 1229
08b84798
BH		 1230 for_each_memblock(memory, reg) {
1231 unsigned long size = reg->size;
919ee677
DM		 1232 unsigned long start, end;
1233
08b84798 1234 start = reg->base;
919ee677
DM		 1235 end = start + size;
1236 while (start < end) {
1237 unsigned long this_end;
1238 int nid;
1239
35a1f0bd 1240 this_end = memblock_nid_range(start, end, &nid);
919ee677 1241
2a4814df 1242 numadbg("Setting memblock NUMA node nid[%d] "
919ee677
DM		 1243 "start[%lx] end[%lx]\n",
1244 nid, start, this_end);
1245
e7e8de59
TC		 1246 memblock_set_node(start, this_end - start,
1247 &memblock.memory, nid);
cd429ce2
PT		 1248 if (memblock.memory.max != prev_max)
1249 goto memblock_resized;
919ee677
DM		 1250 start = this_end;
1251 }
1252 }
1253}
1254
1255static int __init grab_mlgroups(struct mdesc_handle *md)
1256{
1257 unsigned long paddr;
1258 int count = 0;
1259 u64 node;
1260
1261 mdesc_for_each_node_by_name(md, node, "memory-latency-group")
1262 count++;
1263 if (!count)
1264 return -ENOENT;
1265
9a8dd708
MR		 1266 paddr = memblock_phys_alloc(count * sizeof(struct mdesc_mlgroup),
1267 SMP_CACHE_BYTES);
919ee677
DM		 1268 if (!paddr)
1269 return -ENOMEM;
1270
1271 mlgroups = __va(paddr);
1272 num_mlgroups = count;
1273
1274 count = 0;
1275 mdesc_for_each_node_by_name(md, node, "memory-latency-group") {
1276 struct mdesc_mlgroup *m = &mlgroups[count++];
1277 const u64 *val;
1278
1279 m->node = node;
1280
1281 val = mdesc_get_property(md, node, "latency", NULL);
1282 m->latency = *val;
1283 val = mdesc_get_property(md, node, "address-match", NULL);
1284 m->match = *val;
1285 val = mdesc_get_property(md, node, "address-mask", NULL);
1286 m->mask = *val;
1287
90181136
SR		 1288 numadbg("MLGROUP[%d]: node[%llx] latency[%llx] "
1289 "match[%llx] mask[%llx]\n",
919ee677
DM		 1290 count - 1, m->node, m->latency, m->match, m->mask);
1291 }
1292
1293 return 0;
1294}
1295
1296static int __init grab_mblocks(struct mdesc_handle *md)
1297{
1298 unsigned long paddr;
1299 int count = 0;
1300 u64 node;
1301
1302 mdesc_for_each_node_by_name(md, node, "mblock")
1303 count++;
1304 if (!count)
1305 return -ENOENT;
1306
9a8dd708
MR		 1307 paddr = memblock_phys_alloc(count * sizeof(struct mdesc_mblock),
1308 SMP_CACHE_BYTES);
919ee677
DM		 1309 if (!paddr)
1310 return -ENOMEM;
1311
1312 mblocks = __va(paddr);
1313 num_mblocks = count;
1314
1315 count = 0;
1316 mdesc_for_each_node_by_name(md, node, "mblock") {
1317 struct mdesc_mblock *m = &mblocks[count++];
1318 const u64 *val;
1319
1320 val = mdesc_get_property(md, node, "base", NULL);
1321 m->base = *val;
1322 val = mdesc_get_property(md, node, "size", NULL);
1323 m->size = *val;
1324 val = mdesc_get_property(md, node,
1325 "address-congruence-offset", NULL);
771a37ff 1326
1327 /* The address-congruence-offset property is optional.
1328 * Explicity zero it be identifty this.
1329 */
1330 if (val)
1331 m->offset = *val;
1332 else
1333 m->offset = 0UL;
919ee677 1334
90181136 1335 numadbg("MBLOCK[%d]: base[%llx] size[%llx] offset[%llx]\n",
919ee677
DM		 1336 count - 1, m->base, m->size, m->offset);
1337 }
1338
1339 return 0;
1340}
1341
1342static void __init numa_parse_mdesc_group_cpus(struct mdesc_handle *md,
1343 u64 grp, cpumask_t *mask)
1344{
1345 u64 arc;
1346
fb1fece5 1347 cpumask_clear(mask);
919ee677
DM		 1348
1349 mdesc_for_each_arc(arc, md, grp, MDESC_ARC_TYPE_BACK) {
1350 u64 target = mdesc_arc_target(md, arc);
1351 const char *name = mdesc_node_name(md, target);
1352 const u64 *id;
1353
1354 if (strcmp(name, "cpu"))
1355 continue;
1356 id = mdesc_get_property(md, target, "id", NULL);
e305cb8f 1357 if (*id < nr_cpu_ids)
fb1fece5 1358 cpumask_set_cpu(*id, mask);
919ee677
DM		 1359 }
1360}
1361
1362static struct mdesc_mlgroup * __init find_mlgroup(u64 node)
1363{
1364 int i;
1365
1366 for (i = 0; i < num_mlgroups; i++) {
1367 struct mdesc_mlgroup *m = &mlgroups[i];
1368 if (m->node == node)
1369 return m;
1370 }
1371 return NULL;
1372}
1373
52708d69
NG		 1374int __node_distance(int from, int to)
1375{
1376 if ((from >= MAX_NUMNODES) || (to >= MAX_NUMNODES)) {
1377 pr_warn("Returning default NUMA distance value for %d->%d\n",
1378 from, to);
1379 return (from == to) ? LOCAL_DISTANCE : REMOTE_DISTANCE;
1380 }
1381 return numa_latency[from][to];
1382}
2b4792ea 1383EXPORT_SYMBOL(__node_distance);
52708d69 1384
bdf2f59e 1385static int __init find_best_numa_node_for_mlgroup(struct mdesc_mlgroup *grp)
52708d69
NG		 1386{
1387 int i;
1388
1389 for (i = 0; i < MAX_NUMNODES; i++) {
1390 struct node_mem_mask *n = &node_masks[i];
1391
1537b26d 1392 if ((grp->mask == n->mask) && (grp->match == n->match))
52708d69
NG		 1393 break;
1394 }
1395 return i;
1396}
1397
bdf2f59e
PG		 1398static void __init find_numa_latencies_for_group(struct mdesc_handle *md,
1399 u64 grp, int index)
52708d69
NG		 1400{
1401 u64 arc;
1402
1403 mdesc_for_each_arc(arc, md, grp, MDESC_ARC_TYPE_FWD) {
1404 int tnode;
1405 u64 target = mdesc_arc_target(md, arc);
1406 struct mdesc_mlgroup *m = find_mlgroup(target);
1407
1408 if (!m)
1409 continue;
1410 tnode = find_best_numa_node_for_mlgroup(m);
1411 if (tnode == MAX_NUMNODES)
1412 continue;
1413 numa_latency[index][tnode] = m->latency;
1414 }
1415}
1416
919ee677
DM		 1417static int __init numa_attach_mlgroup(struct mdesc_handle *md, u64 grp,
1418 int index)
1419{
1420 struct mdesc_mlgroup *candidate = NULL;
1421 u64 arc, best_latency = ~(u64)0;
1422 struct node_mem_mask *n;
1423
1424 mdesc_for_each_arc(arc, md, grp, MDESC_ARC_TYPE_FWD) {
1425 u64 target = mdesc_arc_target(md, arc);
1426 struct mdesc_mlgroup *m = find_mlgroup(target);
1427 if (!m)
1428 continue;
1429 if (m->latency < best_latency) {
1430 candidate = m;
1431 best_latency = m->latency;
1432 }
1433 }
1434 if (!candidate)
1435 return -ENOENT;
1436
1437 if (num_node_masks != index) {
1438 printk(KERN_ERR "Inconsistent NUMA state, "
1439 "index[%d] != num_node_masks[%d]\n",
1440 index, num_node_masks);
1441 return -EINVAL;
1442 }
1443
1444 n = &node_masks[num_node_masks++];
1445
1446 n->mask = candidate->mask;
1537b26d 1447 n->match = candidate->match;
1da177e4 1448
1537b26d
PT		 1449 numadbg("NUMA NODE[%d]: mask[%lx] match[%lx] (latency[%llx])\n",
1450 index, n->mask, n->match, candidate->latency);
1da177e4 1451
919ee677
DM		 1452 return 0;
1453}
1454
1455static int __init numa_parse_mdesc_group(struct mdesc_handle *md, u64 grp,
1456 int index)
1457{
1458 cpumask_t mask;
1459 int cpu;
1460
1461 numa_parse_mdesc_group_cpus(md, grp, &mask);
1462
fb1fece5 1463 for_each_cpu(cpu, &mask)
919ee677 1464 numa_cpu_lookup_table[cpu] = index;
fb1fece5 1465 cpumask_copy(&numa_cpumask_lookup_table[index], &mask);
919ee677
DM		 1466
1467 if (numa_debug) {
1468 printk(KERN_INFO "NUMA GROUP[%d]: cpus [ ", index);
fb1fece5 1469 for_each_cpu(cpu, &mask)
919ee677
DM		 1470 printk("%d ", cpu);
1471 printk("]\n");
1472 }
1473
1474 return numa_attach_mlgroup(md, grp, index);
1475}
1476
1477static int __init numa_parse_mdesc(void)
1478{
1479 struct mdesc_handle *md = mdesc_grab();
52708d69 1480 int i, j, err, count;
919ee677
DM		 1481 u64 node;
1482
1483 node = mdesc_node_by_name(md, MDESC_NODE_NULL, "latency-groups");
1484 if (node == MDESC_NODE_NULL) {
1485 mdesc_release(md);
1486 return -ENOENT;
1487 }
1488
1489 err = grab_mblocks(md);
1490 if (err < 0)
1491 goto out;
1492
1493 err = grab_mlgroups(md);
1494 if (err < 0)
1495 goto out;
1496
1497 count = 0;
1498 mdesc_for_each_node_by_name(md, node, "group") {
1499 err = numa_parse_mdesc_group(md, node, count);
1500 if (err < 0)
1501 break;
1502 count++;
1503 }
1504
52708d69
NG		 1505 count = 0;
1506 mdesc_for_each_node_by_name(md, node, "group") {
1507 find_numa_latencies_for_group(md, node, count);
1508 count++;
1509 }
1510
1511 /* Normalize numa latency matrix according to ACPI SLIT spec. */
1512 for (i = 0; i < MAX_NUMNODES; i++) {
1513 u64 self_latency = numa_latency[i][i];
1514
1515 for (j = 0; j < MAX_NUMNODES; j++) {
1516 numa_latency[i][j] =
1517 (numa_latency[i][j] * LOCAL_DISTANCE) /
1518 self_latency;
1519 }
1520 }
1521
919ee677
DM		 1522 add_node_ranges();
1523
1524 for (i = 0; i < num_node_masks; i++) {
1525 allocate_node_data(i);
1526 node_set_online(i);
1527 }
1528
1529 err = 0;
1530out:
1531 mdesc_release(md);
1532 return err;
1533}
1534
072bd413
DM		 1535static int __init numa_parse_jbus(void)
1536{
1537 unsigned long cpu, index;
1538
1539 /* NUMA node id is encoded in bits 36 and higher, and there is
1540 * a 1-to-1 mapping from CPU ID to NUMA node ID.
1541 */
1542 index = 0;
1543 for_each_present_cpu(cpu) {
1544 numa_cpu_lookup_table[cpu] = index;
fb1fece5 1545 cpumask_copy(&numa_cpumask_lookup_table[index], cpumask_of(cpu));
072bd413 1546 node_masks[index].mask = ~((1UL << 36UL) - 1UL);
1537b26d 1547 node_masks[index].match = cpu << 36UL;
072bd413
DM		 1548
1549 index++;
1550 }
1551 num_node_masks = index;
1552
1553 add_node_ranges();
1554
1555 for (index = 0; index < num_node_masks; index++) {
1556 allocate_node_data(index);
1557 node_set_online(index);
1558 }
1559
1560 return 0;
1561}
1562
919ee677
DM		 1563static int __init numa_parse_sun4u(void)
1564{
072bd413
DM		 1565 if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1566 unsigned long ver;
1567
1568 __asm__ ("rdpr %%ver, %0" : "=r" (ver));
1569 if ((ver >> 32UL) == __JALAPENO_ID ||
1570 (ver >> 32UL) == __SERRANO_ID)
1571 return numa_parse_jbus();
1572 }
919ee677
DM		 1573 return -1;
1574}
1575
1576static int __init bootmem_init_numa(void)
1577{
36beca65 1578 int i, j;
919ee677
DM		 1579 int err = -1;
1580
1581 numadbg("bootmem_init_numa()\n");
1582
36beca65
NG		 1583 /* Some sane defaults for numa latency values */
1584 for (i = 0; i < MAX_NUMNODES; i++) {
1585 for (j = 0; j < MAX_NUMNODES; j++)
1586 numa_latency[i][j] = (i == j) ?
1587 LOCAL_DISTANCE : REMOTE_DISTANCE;
1588 }
1589
919ee677
DM		 1590 if (numa_enabled) {
1591 if (tlb_type == hypervisor)
1592 err = numa_parse_mdesc();
1593 else
1594 err = numa_parse_sun4u();
1595 }
1596 return err;
1597}
1598
1599#else
1da177e4 1600
919ee677
DM		 1601static int bootmem_init_numa(void)
1602{
1603 return -1;
1604}
1605
1606#endif
1607
1608static void __init bootmem_init_nonnuma(void)
1609{
95f72d1e
YL		 1610 unsigned long top_of_ram = memblock_end_of_DRAM();
1611 unsigned long total_ram = memblock_phys_mem_size();
919ee677
DM		 1612
1613 numadbg("bootmem_init_nonnuma()\n");
1614
1615 printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
1616 top_of_ram, total_ram);
1617 printk(KERN_INFO "Memory hole size: %ldMB\n",
1618 (top_of_ram - total_ram) >> 20);
1619
1620 init_node_masks_nonnuma();
d7dc899a 1621 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
919ee677 1622 allocate_node_data(0);
919ee677
DM		 1623 node_set_online(0);
1624}
1625
919ee677
DM		 1626static unsigned long __init bootmem_init(unsigned long phys_base)
1627{
1628 unsigned long end_pfn;
919ee677 1629
95f72d1e 1630 end_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
919ee677
DM		 1631 max_pfn = max_low_pfn = end_pfn;
1632 min_low_pfn = (phys_base >> PAGE_SHIFT);
1633
1634 if (bootmem_init_numa() < 0)
1635 bootmem_init_nonnuma();
1636
625d693e
DM		 1637 /* Dump memblock with node info. */
1638 memblock_dump_all();
919ee677 1639
625d693e 1640 /* XXX cpu notifier XXX */
d1112018 1641
625d693e 1642 sparse_memory_present_with_active_regions(MAX_NUMNODES);
d1112018
DM		 1643 sparse_init();
1644
1da177e4
LT		 1645 return end_pfn;
1646}
1647
9cc3a1ac
DM		 1648static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
1649static int pall_ents __initdata;
1650
0dd5b7b0
DM		 1651static unsigned long max_phys_bits = 40;
1652
1653bool kern_addr_valid(unsigned long addr)
1654{
0dd5b7b0
DM		 1655 pgd_t *pgd;
1656 pud_t *pud;
1657 pmd_t *pmd;
1658 pte_t *pte;
1659
bb4e6e85 1660 if ((long)addr < 0L) {
0dd5b7b0
DM		 1661 unsigned long pa = __pa(addr);
1662
adfae8a5 1663 if ((pa >> max_phys_bits) != 0UL)
bb4e6e85
DM		 1664 return false;
1665
0dd5b7b0
DM		 1666 return pfn_valid(pa >> PAGE_SHIFT);
1667 }
1668
bb4e6e85
DM		 1669 if (addr >= (unsigned long) KERNBASE &&
1670 addr < (unsigned long)&_end)
1671 return true;
1672
0dd5b7b0
DM		 1673 pgd = pgd_offset_k(addr);
1674 if (pgd_none(*pgd))
1675 return 0;
1676
1677 pud = pud_offset(pgd, addr);
1678 if (pud_none(*pud))
1679 return 0;
1680
1681 if (pud_large(*pud))
1682 return pfn_valid(pud_pfn(*pud));
1683
1684 pmd = pmd_offset(pud, addr);
1685 if (pmd_none(*pmd))
1686 return 0;
1687
1688 if (pmd_large(*pmd))
1689 return pfn_valid(pmd_pfn(*pmd));
1690
1691 pte = pte_offset_kernel(pmd, addr);
1692 if (pte_none(*pte))
1693 return 0;
1694
1695 return pfn_valid(pte_pfn(*pte));
1696}
1697EXPORT_SYMBOL(kern_addr_valid);
1698
1699static unsigned long __ref kernel_map_hugepud(unsigned long vstart,
1700 unsigned long vend,
1701 pud_t *pud)
1702{
1703 const unsigned long mask16gb = (1UL << 34) - 1UL;
1704 u64 pte_val = vstart;
1705
1706 /* Each PUD is 8GB */
1707 if ((vstart & mask16gb) ||
1708 (vend - vstart <= mask16gb)) {
1709 pte_val ^= kern_linear_pte_xor[2];
1710 pud_val(*pud) = pte_val | _PAGE_PUD_HUGE;
1711
1712 return vstart + PUD_SIZE;
1713 }
1714
1715 pte_val ^= kern_linear_pte_xor[3];
1716 pte_val |= _PAGE_PUD_HUGE;
1717
1718 vend = vstart + mask16gb + 1UL;
1719 while (vstart < vend) {
1720 pud_val(*pud) = pte_val;
1721
1722 pte_val += PUD_SIZE;
1723 vstart += PUD_SIZE;
1724 pud++;
1725 }
1726 return vstart;
1727}
1728
1729static bool kernel_can_map_hugepud(unsigned long vstart, unsigned long vend,
1730 bool guard)
1731{
1732 if (guard && !(vstart & ~PUD_MASK) && (vend - vstart) >= PUD_SIZE)
1733 return true;
1734
1735 return false;
1736}
1737
1738static unsigned long __ref kernel_map_hugepmd(unsigned long vstart,
1739 unsigned long vend,
1740 pmd_t *pmd)
1741{
1742 const unsigned long mask256mb = (1UL << 28) - 1UL;
1743 const unsigned long mask2gb = (1UL << 31) - 1UL;
1744 u64 pte_val = vstart;
1745
1746 /* Each PMD is 8MB */
1747 if ((vstart & mask256mb) ||
1748 (vend - vstart <= mask256mb)) {
1749 pte_val ^= kern_linear_pte_xor[0];
1750 pmd_val(*pmd) = pte_val | _PAGE_PMD_HUGE;
1751
1752 return vstart + PMD_SIZE;
1753 }
1754
1755 if ((vstart & mask2gb) ||
1756 (vend - vstart <= mask2gb)) {
1757 pte_val ^= kern_linear_pte_xor[1];
1758 pte_val |= _PAGE_PMD_HUGE;
1759 vend = vstart + mask256mb + 1UL;
1760 } else {
1761 pte_val ^= kern_linear_pte_xor[2];
1762 pte_val |= _PAGE_PMD_HUGE;
1763 vend = vstart + mask2gb + 1UL;
1764 }
1765
1766 while (vstart < vend) {
1767 pmd_val(*pmd) = pte_val;
1768
1769 pte_val += PMD_SIZE;
1770 vstart += PMD_SIZE;
1771 pmd++;
1772 }
1773
1774 return vstart;
1775}
1776
1777static bool kernel_can_map_hugepmd(unsigned long vstart, unsigned long vend,
1778 bool guard)
1779{
1780 if (guard && !(vstart & ~PMD_MASK) && (vend - vstart) >= PMD_SIZE)
1781 return true;
1782
1783 return false;
1784}
1785
896aef43 1786static unsigned long __ref kernel_map_range(unsigned long pstart,
0dd5b7b0
DM		 1787 unsigned long pend, pgprot_t prot,
1788 bool use_huge)
56425306
DM		 1789{
1790 unsigned long vstart = PAGE_OFFSET + pstart;
1791 unsigned long vend = PAGE_OFFSET + pend;
1792 unsigned long alloc_bytes = 0UL;
1793
1794 if ((vstart & ~PAGE_MASK) || (vend & ~PAGE_MASK)) {
13edad7a 1795 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
56425306
DM		 1796 vstart, vend);
1797 prom_halt();
1798 }
1799
1800 while (vstart < vend) {
1801 unsigned long this_end, paddr = __pa(vstart);
1802 pgd_t *pgd = pgd_offset_k(vstart);
1803 pud_t *pud;
1804 pmd_t *pmd;
1805 pte_t *pte;
1806
ac55c768
DM		 1807 if (pgd_none(*pgd)) {
1808 pud_t *new;
1809
4fc4a09e
MR		 1810 new = memblock_alloc_from(PAGE_SIZE, PAGE_SIZE,
1811 PAGE_SIZE);
b1e1c869
MR		 1812 if (!new)
1813 goto err_alloc;
ac55c768
DM		 1814 alloc_bytes += PAGE_SIZE;
1815 pgd_populate(&init_mm, pgd, new);
1816 }
56425306
DM		 1817 pud = pud_offset(pgd, vstart);
1818 if (pud_none(*pud)) {
1819 pmd_t *new;
1820
0dd5b7b0
DM		 1821 if (kernel_can_map_hugepud(vstart, vend, use_huge)) {
1822 vstart = kernel_map_hugepud(vstart, vend, pud);
1823 continue;
1824 }
4fc4a09e
MR		 1825 new = memblock_alloc_from(PAGE_SIZE, PAGE_SIZE,
1826 PAGE_SIZE);
b1e1c869
MR		 1827 if (!new)
1828 goto err_alloc;
56425306
DM		 1829 alloc_bytes += PAGE_SIZE;
1830 pud_populate(&init_mm, pud, new);
1831 }
1832
1833 pmd = pmd_offset(pud, vstart);
0dd5b7b0 1834 if (pmd_none(*pmd)) {
56425306
DM		 1835 pte_t *new;
1836
0dd5b7b0
DM		 1837 if (kernel_can_map_hugepmd(vstart, vend, use_huge)) {
1838 vstart = kernel_map_hugepmd(vstart, vend, pmd);
1839 continue;
1840 }
4fc4a09e
MR		 1841 new = memblock_alloc_from(PAGE_SIZE, PAGE_SIZE,
1842 PAGE_SIZE);
b1e1c869
MR		 1843 if (!new)
1844 goto err_alloc;
56425306
DM		 1845 alloc_bytes += PAGE_SIZE;
1846 pmd_populate_kernel(&init_mm, pmd, new);
1847 }
1848
1849 pte = pte_offset_kernel(pmd, vstart);
1850 this_end = (vstart + PMD_SIZE) & PMD_MASK;
1851 if (this_end > vend)
1852 this_end = vend;
1853
1854 while (vstart < this_end) {
1855 pte_val(*pte) = (paddr | pgprot_val(prot));
1856
1857 vstart += PAGE_SIZE;
1858 paddr += PAGE_SIZE;
1859 pte++;
1860 }
1861 }
1862
1863 return alloc_bytes;
b1e1c869
MR		 1864
1865err_alloc:
1866 panic("%s: Failed to allocate %lu bytes align=%lx from=%lx\n",
1867 __func__, PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
1868 return -ENOMEM;
56425306
DM		 1869}
1870
0dd5b7b0 1871static void __init flush_all_kernel_tsbs(void)
4f93d21d 1872{
0dd5b7b0 1873 int i;
4f93d21d 1874
0dd5b7b0
DM		 1875 for (i = 0; i < KERNEL_TSB_NENTRIES; i++) {
1876 struct tsb *ent = &swapper_tsb[i];
4f93d21d 1877
0dd5b7b0 1878 ent->tag = (1UL << TSB_TAG_INVALID_BIT);
4f93d21d 1879 }
0dd5b7b0
DM		 1880#ifndef CONFIG_DEBUG_PAGEALLOC
1881 for (i = 0; i < KERNEL_TSB4M_NENTRIES; i++) {
1882 struct tsb *ent = &swapper_4m_tsb[i];
4f93d21d 1883
0dd5b7b0 1884 ent->tag = (1UL << TSB_TAG_INVALID_BIT);
9cc3a1ac 1885 }
0dd5b7b0 1886#endif
9cc3a1ac 1887}
56425306 1888
0dd5b7b0 1889extern unsigned int kvmap_linear_patch[1];
9cc3a1ac 1890
8f361453
DM		 1891static void __init kernel_physical_mapping_init(void)
1892{
8f361453 1893 unsigned long i, mem_alloced = 0UL;
0dd5b7b0 1894 bool use_huge = true;
8f361453 1895
0dd5b7b0
DM		 1896#ifdef CONFIG_DEBUG_PAGEALLOC
1897 use_huge = false;
1898#endif
8f361453
DM		 1899 for (i = 0; i < pall_ents; i++) {
1900 unsigned long phys_start, phys_end;
1901
1902 phys_start = pall[i].phys_addr;
1903 phys_end = phys_start + pall[i].reg_size;
1904
56425306 1905 mem_alloced += kernel_map_range(phys_start, phys_end,
0dd5b7b0 1906 PAGE_KERNEL, use_huge);
56425306
DM		 1907 }
1908
1909 printk("Allocated %ld bytes for kernel page tables.\n",
1910 mem_alloced);
1911
1912 kvmap_linear_patch[0] = 0x01000000; /* nop */
1913 flushi(&kvmap_linear_patch[0]);
1914
0dd5b7b0
DM		 1915 flush_all_kernel_tsbs();
1916
56425306
DM		 1917 __flush_tlb_all();
1918}
1919
9cc3a1ac 1920#ifdef CONFIG_DEBUG_PAGEALLOC
031bc574 1921void __kernel_map_pages(struct page *page, int numpages, int enable)
56425306
DM		 1922{
1923 unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT;
1924 unsigned long phys_end = phys_start + (numpages * PAGE_SIZE);
1925
1926 kernel_map_range(phys_start, phys_end,
0dd5b7b0 1927 (enable ? PAGE_KERNEL : __pgprot(0)), false);
56425306 1928
74bf4312
DM		 1929 flush_tsb_kernel_range(PAGE_OFFSET + phys_start,
1930 PAGE_OFFSET + phys_end);
1931
56425306
DM		 1932 /* we should perform an IPI and flush all tlbs,
1933 * but that can deadlock->flush only current cpu.
1934 */
1935 __flush_tlb_kernel_range(PAGE_OFFSET + phys_start,
1936 PAGE_OFFSET + phys_end);
1937}
1938#endif
1939
10147570
DM		 1940unsigned long __init find_ecache_flush_span(unsigned long size)
1941{
0836a0eb
DM		 1942 int i;
1943
13edad7a
DM		 1944 for (i = 0; i < pavail_ents; i++) {
1945 if (pavail[i].reg_size >= size)
1946 return pavail[i].phys_addr;
0836a0eb
DM		 1947 }
1948
13edad7a 1949 return ~0UL;
0836a0eb
DM		 1950}
1951
b2d43834
DM		 1952unsigned long PAGE_OFFSET;
1953EXPORT_SYMBOL(PAGE_OFFSET);
1954
bb4e6e85
DM		 1955unsigned long VMALLOC_END = 0x0000010000000000UL;
1956EXPORT_SYMBOL(VMALLOC_END);
1957
4397bed0
DM		 1958unsigned long sparc64_va_hole_top = 0xfffff80000000000UL;
1959unsigned long sparc64_va_hole_bottom = 0x0000080000000000UL;
1960
b2d43834
DM		 1961static void __init setup_page_offset(void)
1962{
b2d43834 1963 if (tlb_type == cheetah || tlb_type == cheetah_plus) {
4397bed0
DM		 1964 /* Cheetah/Panther support a full 64-bit virtual
1965 * address, so we can use all that our page tables
1966 * support.
1967 */
1968 sparc64_va_hole_top = 0xfff0000000000000UL;
1969 sparc64_va_hole_bottom = 0x0010000000000000UL;
1970
b2d43834
DM		 1971 max_phys_bits = 42;
1972 } else if (tlb_type == hypervisor) {
1973 switch (sun4v_chip_type) {
1974 case SUN4V_CHIP_NIAGARA1:
1975 case SUN4V_CHIP_NIAGARA2:
4397bed0
DM		 1976 /* T1 and T2 support 48-bit virtual addresses. */
1977 sparc64_va_hole_top = 0xffff800000000000UL;
1978 sparc64_va_hole_bottom = 0x0000800000000000UL;
1979
b2d43834
DM		 1980 max_phys_bits = 39;
1981 break;
1982 case SUN4V_CHIP_NIAGARA3:
4397bed0
DM		 1983 /* T3 supports 48-bit virtual addresses. */
1984 sparc64_va_hole_top = 0xffff800000000000UL;
1985 sparc64_va_hole_bottom = 0x0000800000000000UL;
1986
b2d43834
DM		 1987 max_phys_bits = 43;
1988 break;
1989 case SUN4V_CHIP_NIAGARA4:
1990 case SUN4V_CHIP_NIAGARA5:
1991 case SUN4V_CHIP_SPARC64X:
7c0fa0f2 1992 case SUN4V_CHIP_SPARC_M6:
4397bed0
DM		 1993 /* T4 and later support 52-bit virtual addresses. */
1994 sparc64_va_hole_top = 0xfff8000000000000UL;
1995 sparc64_va_hole_bottom = 0x0008000000000000UL;
b2d43834
DM		 1996 max_phys_bits = 47;
1997 break;
7c0fa0f2 1998 case SUN4V_CHIP_SPARC_M7:
c5b8b5be 1999 case SUN4V_CHIP_SPARC_SN:
7c0fa0f2
DM		 2000 /* M7 and later support 52-bit virtual addresses. */
2001 sparc64_va_hole_top = 0xfff8000000000000UL;
2002 sparc64_va_hole_bottom = 0x0008000000000000UL;
2003 max_phys_bits = 49;
2004 break;
fdaccf74
VK		 2005 case SUN4V_CHIP_SPARC_M8:
2006 default:
2007 /* M8 and later support 54-bit virtual addresses.
2008 * However, restricting M8 and above VA bits to 53
2009 * as 4-level page table cannot support more than
2010 * 53 VA bits.
2011 */
2012 sparc64_va_hole_top = 0xfff0000000000000UL;
2013 sparc64_va_hole_bottom = 0x0010000000000000UL;
2014 max_phys_bits = 51;
2015 break;
b2d43834
DM		 2016 }
2017 }
2018
2019 if (max_phys_bits > MAX_PHYS_ADDRESS_BITS) {
2020 prom_printf("MAX_PHYS_ADDRESS_BITS is too small, need %lu\n",
2021 max_phys_bits);
2022 prom_halt();
2023 }
2024
bb4e6e85
DM		 2025 PAGE_OFFSET = sparc64_va_hole_top;
2026 VMALLOC_END = ((sparc64_va_hole_bottom >> 1) +
2027 (sparc64_va_hole_bottom >> 2));
b2d43834 2028
bb4e6e85 2029 pr_info("MM: PAGE_OFFSET is 0x%016lx (max_phys_bits == %lu)\n",
b2d43834 2030 PAGE_OFFSET, max_phys_bits);
bb4e6e85
DM		 2031 pr_info("MM: VMALLOC [0x%016lx --> 0x%016lx]\n",
2032 VMALLOC_START, VMALLOC_END);
2033 pr_info("MM: VMEMMAP [0x%016lx --> 0x%016lx]\n",
2034 VMEMMAP_BASE, VMEMMAP_BASE << 1);
b2d43834
DM		 2035}
2036
517af332
DM		 2037static void __init tsb_phys_patch(void)
2038{
d257d5da 2039 struct tsb_ldquad_phys_patch_entry *pquad;
517af332
DM		 2040 struct tsb_phys_patch_entry *p;
2041
d257d5da
DM		 2042 pquad = &__tsb_ldquad_phys_patch;
2043 while (pquad < &__tsb_ldquad_phys_patch_end) {
2044 unsigned long addr = pquad->addr;
2045
2046 if (tlb_type == hypervisor)
2047 *(unsigned int *) addr = pquad->sun4v_insn;
2048 else
2049 *(unsigned int *) addr = pquad->sun4u_insn;
2050 wmb();
2051 __asm__ __volatile__("flush %0"
2052 : /* no outputs */
2053 : "r" (addr));
2054
2055 pquad++;
2056 }
2057
517af332
DM		 2058 p = &__tsb_phys_patch;
2059 while (p < &__tsb_phys_patch_end) {
2060 unsigned long addr = p->addr;
2061
2062 *(unsigned int *) addr = p->insn;
2063 wmb();
2064 __asm__ __volatile__("flush %0"
2065 : /* no outputs */
2066 : "r" (addr));
2067
2068 p++;
2069 }
2070}
2071
490384e7 2072/* Don't mark as init, we give this to the Hypervisor. */
d1acb421
DM		 2073#ifndef CONFIG_DEBUG_PAGEALLOC
2074#define NUM_KTSB_DESCR 2
2075#else
2076#define NUM_KTSB_DESCR 1
2077#endif
2078static struct hv_tsb_descr ktsb_descr[NUM_KTSB_DESCR];
490384e7 2079
8c82dc0e
DM		 2080/* The swapper TSBs are loaded with a base sequence of:
2081 *
2082 * sethi %uhi(SYMBOL), REG1
2083 * sethi %hi(SYMBOL), REG2
2084 * or REG1, %ulo(SYMBOL), REG1
2085 * or REG2, %lo(SYMBOL), REG2
2086 * sllx REG1, 32, REG1
2087 * or REG1, REG2, REG1
2088 *
2089 * When we use physical addressing for the TSB accesses, we patch the
2090 * first four instructions in the above sequence.
2091 */
2092
9076d0e7
DM		 2093static void patch_one_ktsb_phys(unsigned int *start, unsigned int *end, unsigned long pa)
2094{
8c82dc0e
DM		 2095 unsigned long high_bits, low_bits;
2096
2097 high_bits = (pa >> 32) & 0xffffffff;
2098 low_bits = (pa >> 0) & 0xffffffff;
9076d0e7
DM		 2099
2100 while (start < end) {
2101 unsigned int *ia = (unsigned int *)(unsigned long)*start;
2102
8c82dc0e 2103 ia[0] = (ia[0] & ~0x3fffff) | (high_bits >> 10);
9076d0e7
DM		 2104 __asm__ __volatile__("flush %0" : : "r" (ia));
2105
8c82dc0e 2106 ia[1] = (ia[1] & ~0x3fffff) | (low_bits >> 10);
9076d0e7
DM		 2107 __asm__ __volatile__("flush %0" : : "r" (ia + 1));
2108
8c82dc0e
DM		 2109 ia[2] = (ia[2] & ~0x1fff) | (high_bits & 0x3ff);
2110 __asm__ __volatile__("flush %0" : : "r" (ia + 2));
2111
2112 ia[3] = (ia[3] & ~0x1fff) | (low_bits & 0x3ff);
2113 __asm__ __volatile__("flush %0" : : "r" (ia + 3));
2114
9076d0e7
DM		 2115 start++;
2116 }
2117}
2118
2119static void ktsb_phys_patch(void)
2120{
2121 extern unsigned int __swapper_tsb_phys_patch;
2122 extern unsigned int __swapper_tsb_phys_patch_end;
9076d0e7
DM		 2123 unsigned long ktsb_pa;
2124
2125 ktsb_pa = kern_base + ((unsigned long)&swapper_tsb[0] - KERNBASE);
2126 patch_one_ktsb_phys(&__swapper_tsb_phys_patch,
2127 &__swapper_tsb_phys_patch_end, ktsb_pa);
2128#ifndef CONFIG_DEBUG_PAGEALLOC
0785a8e8
DM		 2129 {
2130 extern unsigned int __swapper_4m_tsb_phys_patch;
2131 extern unsigned int __swapper_4m_tsb_phys_patch_end;
9076d0e7
DM		 2132 ktsb_pa = (kern_base +
2133 ((unsigned long)&swapper_4m_tsb[0] - KERNBASE));
2134 patch_one_ktsb_phys(&__swapper_4m_tsb_phys_patch,
2135 &__swapper_4m_tsb_phys_patch_end, ktsb_pa);
0785a8e8 2136 }
9076d0e7
DM		 2137#endif
2138}
2139
490384e7
DM		 2140static void __init sun4v_ktsb_init(void)
2141{
2142 unsigned long ktsb_pa;
2143
d7744a09 2144 /* First KTSB for PAGE_SIZE mappings. */
490384e7
DM		 2145 ktsb_pa = kern_base + ((unsigned long)&swapper_tsb[0] - KERNBASE);
2146
2147 switch (PAGE_SIZE) {
2148 case 8 * 1024:
2149 default:
2150 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_8K;
2151 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_8K;
2152 break;
2153
2154 case 64 * 1024:
2155 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_64K;
2156 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_64K;
2157 break;
2158
2159 case 512 * 1024:
2160 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_512K;
2161 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_512K;
2162 break;
2163
2164 case 4 * 1024 * 1024:
2165 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_4MB;
2166 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_4MB;
2167 break;
6cb79b3f 2168 }
490384e7 2169
3f19a84e 2170 ktsb_descr[0].assoc = 1;
490384e7
DM		 2171 ktsb_descr[0].num_ttes = KERNEL_TSB_NENTRIES;
2172 ktsb_descr[0].ctx_idx = 0;
2173 ktsb_descr[0].tsb_base = ktsb_pa;
2174 ktsb_descr[0].resv = 0;
2175
d1acb421 2176#ifndef CONFIG_DEBUG_PAGEALLOC
4f93d21d 2177 /* Second KTSB for 4MB/256MB/2GB/16GB mappings. */
d7744a09
DM		 2178 ktsb_pa = (kern_base +
2179 ((unsigned long)&swapper_4m_tsb[0] - KERNBASE));
2180
2181 ktsb_descr[1].pgsz_idx = HV_PGSZ_IDX_4MB;
c69ad0a3
DM		 2182 ktsb_descr[1].pgsz_mask = ((HV_PGSZ_MASK_4MB |
2183 HV_PGSZ_MASK_256MB |
2184 HV_PGSZ_MASK_2GB |
2185 HV_PGSZ_MASK_16GB) &
2186 cpu_pgsz_mask);
d7744a09
DM		 2187 ktsb_descr[1].assoc = 1;
2188 ktsb_descr[1].num_ttes = KERNEL_TSB4M_NENTRIES;
2189 ktsb_descr[1].ctx_idx = 0;
2190 ktsb_descr[1].tsb_base = ktsb_pa;
2191 ktsb_descr[1].resv = 0;
d1acb421 2192#endif
490384e7
DM		 2193}
2194
2066aadd 2195void sun4v_ktsb_register(void)
490384e7 2196{
7db35f31 2197 unsigned long pa, ret;
490384e7
DM		 2198
2199 pa = kern_base + ((unsigned long)&ktsb_descr[0] - KERNBASE);
2200
7db35f31
DM		 2201 ret = sun4v_mmu_tsb_ctx0(NUM_KTSB_DESCR, pa);
2202 if (ret != 0) {
2203 prom_printf("hypervisor_mmu_tsb_ctx0[%lx]: "
2204 "errors with %lx\n", pa, ret);
2205 prom_halt();
2206 }
490384e7
DM		 2207}
2208
c69ad0a3
DM		 2209static void __init sun4u_linear_pte_xor_finalize(void)
2210{
2211#ifndef CONFIG_DEBUG_PAGEALLOC
2212 /* This is where we would add Panther support for
2213 * 32MB and 256MB pages.
2214 */
2215#endif
2216}
2217
2218static void __init sun4v_linear_pte_xor_finalize(void)
2219{
494e5b6f
KA		 2220 unsigned long pagecv_flag;
2221
2222 /* Bit 9 of TTE is no longer CV bit on M7 processor and it instead
2223 * enables MCD error. Do not set bit 9 on M7 processor.
2224 */
2225 switch (sun4v_chip_type) {
2226 case SUN4V_CHIP_SPARC_M7:
7d484acb 2227 case SUN4V_CHIP_SPARC_M8:
c5b8b5be 2228 case SUN4V_CHIP_SPARC_SN:
494e5b6f
KA		 2229 pagecv_flag = 0x00;
2230 break;
2231 default:
2232 pagecv_flag = _PAGE_CV_4V;
2233 break;
2234 }
c69ad0a3
DM		 2235#ifndef CONFIG_DEBUG_PAGEALLOC
2236 if (cpu_pgsz_mask & HV_PGSZ_MASK_256MB) {
2237 kern_linear_pte_xor[1] = (_PAGE_VALID | _PAGE_SZ256MB_4V) ^
922631b9 2238 PAGE_OFFSET;
494e5b6f 2239 kern_linear_pte_xor[1] |= (_PAGE_CP_4V | pagecv_flag |
c69ad0a3
DM		 2240 _PAGE_P_4V | _PAGE_W_4V);
2241 } else {
2242 kern_linear_pte_xor[1] = kern_linear_pte_xor[0];
2243 }
2244
2245 if (cpu_pgsz_mask & HV_PGSZ_MASK_2GB) {
2246 kern_linear_pte_xor[2] = (_PAGE_VALID | _PAGE_SZ2GB_4V) ^
922631b9 2247 PAGE_OFFSET;
494e5b6f 2248 kern_linear_pte_xor[2] |= (_PAGE_CP_4V | pagecv_flag |
c69ad0a3
DM		 2249 _PAGE_P_4V | _PAGE_W_4V);
2250 } else {
2251 kern_linear_pte_xor[2] = kern_linear_pte_xor[1];
2252 }
2253
2254 if (cpu_pgsz_mask & HV_PGSZ_MASK_16GB) {
2255 kern_linear_pte_xor[3] = (_PAGE_VALID | _PAGE_SZ16GB_4V) ^
922631b9 2256 PAGE_OFFSET;
494e5b6f 2257 kern_linear_pte_xor[3] |= (_PAGE_CP_4V | pagecv_flag |
c69ad0a3
DM		 2258 _PAGE_P_4V | _PAGE_W_4V);
2259 } else {
2260 kern_linear_pte_xor[3] = kern_linear_pte_xor[2];
2261 }
2262#endif
2263}
2264
1da177e4
LT		 2265/* paging_init() sets up the page tables */
2266
1da177e4 2267static unsigned long last_valid_pfn;
ac55c768 2268
c4bce90e
DM		 2269static void sun4u_pgprot_init(void);
2270static void sun4v_pgprot_init(void);
2271
494e5b6f
KA		 2272#define _PAGE_CACHE_4U (_PAGE_CP_4U | _PAGE_CV_4U)
2273#define _PAGE_CACHE_4V (_PAGE_CP_4V | _PAGE_CV_4V)
2274#define __DIRTY_BITS_4U (_PAGE_MODIFIED_4U | _PAGE_WRITE_4U | _PAGE_W_4U)
2275#define __DIRTY_BITS_4V (_PAGE_MODIFIED_4V | _PAGE_WRITE_4V | _PAGE_W_4V)
2276#define __ACCESS_BITS_4U (_PAGE_ACCESSED_4U | _PAGE_READ_4U | _PAGE_R)
2277#define __ACCESS_BITS_4V (_PAGE_ACCESSED_4V | _PAGE_READ_4V | _PAGE_R)
2278
7c21d533 2279/* We need to exclude reserved regions. This exclusion will include
2280 * vmlinux and initrd. To be more precise the initrd size could be used to
2281 * compute a new lower limit because it is freed later during initialization.
2282 */
2283static void __init reduce_memory(phys_addr_t limit_ram)
2284{
f4d9a23d
MR		 2285 limit_ram += memblock_reserved_size();
2286 memblock_enforce_memory_limit(limit_ram);
7c21d533 2287}
2288
1da177e4
LT		 2289void __init paging_init(void)
2290{
919ee677 2291 unsigned long end_pfn, shift, phys_base;
0836a0eb
DM		 2292 unsigned long real_end, i;
2293
b2d43834
DM		 2294 setup_page_offset();
2295
22adb358
DM		 2296 /* These build time checkes make sure that the dcache_dirty_cpu()
2297 * page->flags usage will work.
2298 *
2299 * When a page gets marked as dcache-dirty, we store the
2300 * cpu number starting at bit 32 in the page->flags. Also,
2301 * functions like clear_dcache_dirty_cpu use the cpu mask
2302 * in 13-bit signed-immediate instruction fields.
2303 */
9223b419
CL		 2304
2305 /*
2306 * Page flags must not reach into upper 32 bits that are used
2307 * for the cpu number
2308 */
2309 BUILD_BUG_ON(NR_PAGEFLAGS > 32);
2310
2311 /*
2312 * The bit fields placed in the high range must not reach below
2313 * the 32 bit boundary. Otherwise we cannot place the cpu field
2314 * at the 32 bit boundary.
2315 */
22adb358 2316 BUILD_BUG_ON(SECTIONS_WIDTH + NODES_WIDTH + ZONES_WIDTH +
9223b419
CL		 2317 ilog2(roundup_pow_of_two(NR_CPUS)) > 32);
2318
22adb358
DM		 2319 BUILD_BUG_ON(NR_CPUS > 4096);
2320
0eef331a 2321 kern_base = (prom_boot_mapping_phys_low >> ILOG2_4MB) << ILOG2_4MB;
481295f9
DM		 2322 kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
2323
d7744a09 2324 /* Invalidate both kernel TSBs. */
8b234274 2325 memset(swapper_tsb, 0x40, sizeof(swapper_tsb));
d1acb421 2326#ifndef CONFIG_DEBUG_PAGEALLOC
d7744a09 2327 memset(swapper_4m_tsb, 0x40, sizeof(swapper_4m_tsb));
d1acb421 2328#endif
8b234274 2329
494e5b6f
KA		 2330 /* TTE.cv bit on sparc v9 occupies the same position as TTE.mcde
2331 * bit on M7 processor. This is a conflicting usage of the same
2332 * bit. Enabling TTE.cv on M7 would turn on Memory Corruption
2333 * Detection error on all pages and this will lead to problems
2334 * later. Kernel does not run with MCD enabled and hence rest
2335 * of the required steps to fully configure memory corruption
2336 * detection are not taken. We need to ensure TTE.mcde is not
2337 * set on M7 processor. Compute the value of cacheability
2338 * flag for use later taking this into consideration.
2339 */
2340 switch (sun4v_chip_type) {
2341 case SUN4V_CHIP_SPARC_M7:
7d484acb 2342 case SUN4V_CHIP_SPARC_M8:
c5b8b5be 2343 case SUN4V_CHIP_SPARC_SN:
494e5b6f
KA		 2344 page_cache4v_flag = _PAGE_CP_4V;
2345 break;
2346 default:
2347 page_cache4v_flag = _PAGE_CACHE_4V;
2348 break;
2349 }
2350
c4bce90e
DM		 2351 if (tlb_type == hypervisor)
2352 sun4v_pgprot_init();
2353 else
2354 sun4u_pgprot_init();
2355
d257d5da 2356 if (tlb_type == cheetah_plus ||
9076d0e7 2357 tlb_type == hypervisor) {
517af332 2358 tsb_phys_patch();
9076d0e7
DM		 2359 ktsb_phys_patch();
2360 }
517af332 2361
c69ad0a3 2362 if (tlb_type == hypervisor)
d257d5da
DM		 2363 sun4v_patch_tlb_handlers();
2364
a94a172d
DM		 2365 /* Find available physical memory...
2366 *
2367 * Read it twice in order to work around a bug in openfirmware.
2368 * The call to grab this table itself can cause openfirmware to
2369 * allocate memory, which in turn can take away some space from
2370 * the list of available memory. Reading it twice makes sure
2371 * we really do get the final value.
2372 */
2373 read_obp_translations();
2374 read_obp_memory("reg", &pall[0], &pall_ents);
2375 read_obp_memory("available", &pavail[0], &pavail_ents);
13edad7a 2376 read_obp_memory("available", &pavail[0], &pavail_ents);
0836a0eb
DM		 2377
2378 phys_base = 0xffffffffffffffffUL;
3b2a7e23 2379 for (i = 0; i < pavail_ents; i++) {
13edad7a 2380 phys_base = min(phys_base, pavail[i].phys_addr);
95f72d1e 2381 memblock_add(pavail[i].phys_addr, pavail[i].reg_size);
3b2a7e23
DM		 2382 }
2383
95f72d1e 2384 memblock_reserve(kern_base, kern_size);
0836a0eb 2385
4e82c9a6
DM		 2386 find_ramdisk(phys_base);
2387
7c21d533 2388 if (cmdline_memory_size)
2389 reduce_memory(cmdline_memory_size);
25b0c659 2390
1aadc056 2391 memblock_allow_resize();
95f72d1e 2392 memblock_dump_all();
3b2a7e23 2393
1da177e4
LT		 2394 set_bit(0, mmu_context_bmap);
2395
2bdb3cb2
DM		 2396 shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE);
2397
1da177e4 2398 real_end = (unsigned long)_end;
0eef331a 2399 num_kernel_image_mappings = DIV_ROUND_UP(real_end - KERNBASE, 1 << ILOG2_4MB);
64658743
DM		 2400 printk("Kernel: Using %d locked TLB entries for main kernel image.\n",
2401 num_kernel_image_mappings);
2bdb3cb2
DM		 2402
2403 /* Set kernel pgd to upper alias so physical page computations
1da177e4
LT		 2404 * work.
2405 */
2406 init_mm.pgd += ((shift) / (sizeof(pgd_t)));
2407
d195b71b 2408 memset(swapper_pg_dir, 0, sizeof(swapper_pg_dir));
0dd5b7b0 2409
c9c10830 2410 inherit_prom_mappings();
5085b4a5 2411
a8b900d8
DM		 2412 /* Ok, we can use our TLB miss and window trap handlers safely. */
2413 setup_tba();
1da177e4 2414
c9c10830 2415 __flush_tlb_all();
9ad98c5b 2416
ad072004 2417 prom_build_devicetree();
b696fdc2 2418 of_populate_present_mask();
b99c6ebe
DM		 2419#ifndef CONFIG_SMP
2420 of_fill_in_cpu_data();
2421#endif
ad072004 2422
890db403 2423 if (tlb_type == hypervisor) {
4a283339 2424 sun4v_mdesc_init();
6ac5c610 2425 mdesc_populate_present_mask(cpu_all_mask);
b99c6ebe
DM		 2426#ifndef CONFIG_SMP
2427 mdesc_fill_in_cpu_data(cpu_all_mask);
2428#endif
ce33fdc5 2429 mdesc_get_page_sizes(cpu_all_mask, &cpu_pgsz_mask);
c69ad0a3
DM		 2430
2431 sun4v_linear_pte_xor_finalize();
2432
2433 sun4v_ktsb_init();
2434 sun4v_ktsb_register();
ce33fdc5
DM		 2435 } else {
2436 unsigned long impl, ver;
2437
2438 cpu_pgsz_mask = (HV_PGSZ_MASK_8K | HV_PGSZ_MASK_64K |
2439 HV_PGSZ_MASK_512K | HV_PGSZ_MASK_4MB);
2440
2441 __asm__ __volatile__("rdpr %%ver, %0" : "=r" (ver));
2442 impl = ((ver >> 32) & 0xffff);
2443 if (impl == PANTHER_IMPL)
2444 cpu_pgsz_mask |= (HV_PGSZ_MASK_32MB |
2445 HV_PGSZ_MASK_256MB);
c69ad0a3
DM		 2446
2447 sun4u_linear_pte_xor_finalize();
890db403 2448 }
4a283339 2449
c69ad0a3
DM		 2450 /* Flush the TLBs and the 4M TSB so that the updated linear
2451 * pte XOR settings are realized for all mappings.
2452 */
2453 __flush_tlb_all();
2454#ifndef CONFIG_DEBUG_PAGEALLOC
2455 memset(swapper_4m_tsb, 0x40, sizeof(swapper_4m_tsb));
2456#endif
2457 __flush_tlb_all();
2458
5ed56f1a
DM		 2459 /* Setup bootmem... */
2460 last_valid_pfn = end_pfn = bootmem_init(phys_base);
2461
56425306 2462 kernel_physical_mapping_init();
56425306 2463
1da177e4 2464 {
919ee677 2465 unsigned long max_zone_pfns[MAX_NR_ZONES];
1da177e4 2466
919ee677 2467 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
1da177e4 2468
919ee677 2469 max_zone_pfns[ZONE_NORMAL] = end_pfn;
1da177e4 2470
919ee677 2471 free_area_init_nodes(max_zone_pfns);
1da177e4
LT		 2472 }
2473
3c62a2d3 2474 printk("Booting Linux...\n");
1da177e4
LT		 2475}
2476
7c9503b8 2477int page_in_phys_avail(unsigned long paddr)
919ee677
DM		 2478{
2479 int i;
2480
2481 paddr &= PAGE_MASK;
2482
2483 for (i = 0; i < pavail_ents; i++) {
2484 unsigned long start, end;
2485
2486 start = pavail[i].phys_addr;
2487 end = start + pavail[i].reg_size;
2488
2489 if (paddr >= start && paddr < end)
2490 return 1;
2491 }
2492 if (paddr >= kern_base && paddr < (kern_base + kern_size))
2493 return 1;
2494#ifdef CONFIG_BLK_DEV_INITRD
2495 if (paddr >= __pa(initrd_start) &&
2496 paddr < __pa(PAGE_ALIGN(initrd_end)))
2497 return 1;
2498#endif
2499
2500 return 0;
2501}
2502
961f8fa0
YL		 2503static void __init register_page_bootmem_info(void)
2504{
2505#ifdef CONFIG_NEED_MULTIPLE_NODES
2506 int i;
2507
2508 for_each_online_node(i)
2509 if (NODE_DATA(i)->node_spanned_pages)
2510 register_page_bootmem_info_node(NODE_DATA(i));
2511#endif
2512}
1da177e4
LT		 2513void __init mem_init(void)
2514{
1da177e4
LT		 2515 high_memory = __va(last_valid_pfn << PAGE_SHIFT);
2516
c6ffc5ca 2517 memblock_free_all();
919ee677 2518
2a20aa17
PT		 2519 /*
2520 * Must be done after boot memory is put on freelist, because here we
2521 * might set fields in deferred struct pages that have not yet been
c6ffc5ca 2522 * initialized, and memblock_free_all() initializes all the reserved
2a20aa17
PT		 2523 * deferred pages for us.
2524 */
2525 register_page_bootmem_info();
2526
1da177e4
LT		 2527 /*
2528 * Set up the zero page, mark it reserved, so that page count
2529 * is not manipulated when freeing the page from user ptes.
2530 */
2531 mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
2532 if (mem_map_zero == NULL) {
2533 prom_printf("paging_init: Cannot alloc zero page.\n");
2534 prom_halt();
2535 }
70affe45 2536 mark_page_reserved(mem_map_zero);
1da177e4 2537
dceccbe9 2538 mem_init_print_info(NULL);
1da177e4
LT		 2539
2540 if (tlb_type == cheetah || tlb_type == cheetah_plus)
2541 cheetah_ecache_flush_init();
2542}
2543
898cf0ec 2544void free_initmem(void)
1da177e4
LT		 2545{
2546 unsigned long addr, initend;
f2b60794
DM		 2547 int do_free = 1;
2548
2549 /* If the physical memory maps were trimmed by kernel command
2550 * line options, don't even try freeing this initmem stuff up.
2551 * The kernel image could have been in the trimmed out region
2552 * and if so the freeing below will free invalid page structs.
2553 */
2554 if (cmdline_memory_size)
2555 do_free = 0;
1da177e4
LT		 2556
2557 /*
2558 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
2559 */
2560 addr = PAGE_ALIGN((unsigned long)(__init_begin));
2561 initend = (unsigned long)(__init_end) & PAGE_MASK;
2562 for (; addr < initend; addr += PAGE_SIZE) {
2563 unsigned long page;
1da177e4
LT		 2564
2565 page = (addr +
2566 ((unsigned long) __va(kern_base)) -
2567 ((unsigned long) KERNBASE));
c9cf5528 2568 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1da177e4 2569
70affe45
JL		 2570 if (do_free)
2571 free_reserved_page(virt_to_page(page));
1da177e4
LT		 2572 }
2573}
2574
c4bce90e
DM		 2575pgprot_t PAGE_KERNEL __read_mostly;
2576EXPORT_SYMBOL(PAGE_KERNEL);
2577
2578pgprot_t PAGE_KERNEL_LOCKED __read_mostly;
2579pgprot_t PAGE_COPY __read_mostly;
0f15952a
DM		 2580
2581pgprot_t PAGE_SHARED __read_mostly;
2582EXPORT_SYMBOL(PAGE_SHARED);
2583
c4bce90e
DM		 2584unsigned long pg_iobits __read_mostly;
2585
2586unsigned long _PAGE_IE __read_mostly;
987c74fc 2587EXPORT_SYMBOL(_PAGE_IE);
b2bef442 2588
c4bce90e 2589unsigned long _PAGE_E __read_mostly;
b2bef442
DM		 2590EXPORT_SYMBOL(_PAGE_E);
2591
c4bce90e 2592unsigned long _PAGE_CACHE __read_mostly;
b2bef442 2593EXPORT_SYMBOL(_PAGE_CACHE);
c4bce90e 2594
46644c24 2595#ifdef CONFIG_SPARSEMEM_VMEMMAP
0aad818b 2596int __meminit vmemmap_populate(unsigned long vstart, unsigned long vend,
7b73d978 2597 int node, struct vmem_altmap *altmap)
46644c24 2598{
46644c24
DM		 2599 unsigned long pte_base;
2600
2601 pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4U |
2602 _PAGE_CP_4U | _PAGE_CV_4U |
2603 _PAGE_P_4U | _PAGE_W_4U);
2604 if (tlb_type == hypervisor)
2605 pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4V |
494e5b6f 2606 page_cache4v_flag | _PAGE_P_4V | _PAGE_W_4V);
46644c24 2607
c06240c7 2608 pte_base |= _PAGE_PMD_HUGE;
46644c24 2609
c06240c7
DM		 2610 vstart = vstart & PMD_MASK;
2611 vend = ALIGN(vend, PMD_SIZE);
2612 for (; vstart < vend; vstart += PMD_SIZE) {
df8ee578 2613 pgd_t *pgd = vmemmap_pgd_populate(vstart, node);
c06240c7
DM		 2614 unsigned long pte;
2615 pud_t *pud;
2616 pmd_t *pmd;
2617
df8ee578
PT		 2618 if (!pgd)
2619 return -ENOMEM;
c06240c7 2620
df8ee578
PT		 2621 pud = vmemmap_pud_populate(pgd, vstart, node);
2622 if (!pud)
2623 return -ENOMEM;
2856cc2e 2624
c06240c7 2625 pmd = pmd_offset(pud, vstart);
c06240c7
DM		 2626 pte = pmd_val(*pmd);
2627 if (!(pte & _PAGE_VALID)) {
2628 void *block = vmemmap_alloc_block(PMD_SIZE, node);
2629
2630 if (!block)
2631 return -ENOMEM;
2632
2633 pmd_val(*pmd) = pte_base | __pa(block);
2634 }
2856cc2e 2635 }
c06240c7
DM		 2636
2637 return 0;
2856cc2e 2638}
46723bfa 2639
24b6d416
CH		 2640void vmemmap_free(unsigned long start, unsigned long end,
2641 struct vmem_altmap *altmap)
0197518c
TC		 2642{
2643}
46644c24
DM		 2644#endif /* CONFIG_SPARSEMEM_VMEMMAP */
2645
c4bce90e
DM		 2646static void prot_init_common(unsigned long page_none,
2647 unsigned long page_shared,
2648 unsigned long page_copy,
2649 unsigned long page_readonly,
2650 unsigned long page_exec_bit)
2651{
2652 PAGE_COPY = __pgprot(page_copy);
0f15952a 2653 PAGE_SHARED = __pgprot(page_shared);
c4bce90e
DM		 2654
2655 protection_map[0x0] = __pgprot(page_none);
2656 protection_map[0x1] = __pgprot(page_readonly & ~page_exec_bit);
2657 protection_map[0x2] = __pgprot(page_copy & ~page_exec_bit);
2658 protection_map[0x3] = __pgprot(page_copy & ~page_exec_bit);
2659 protection_map[0x4] = __pgprot(page_readonly);
2660 protection_map[0x5] = __pgprot(page_readonly);
2661 protection_map[0x6] = __pgprot(page_copy);
2662 protection_map[0x7] = __pgprot(page_copy);
2663 protection_map[0x8] = __pgprot(page_none);
2664 protection_map[0x9] = __pgprot(page_readonly & ~page_exec_bit);
2665 protection_map[0xa] = __pgprot(page_shared & ~page_exec_bit);
2666 protection_map[0xb] = __pgprot(page_shared & ~page_exec_bit);
2667 protection_map[0xc] = __pgprot(page_readonly);
2668 protection_map[0xd] = __pgprot(page_readonly);
2669 protection_map[0xe] = __pgprot(page_shared);
2670 protection_map[0xf] = __pgprot(page_shared);
2671}
2672
2673static void __init sun4u_pgprot_init(void)
2674{
2675 unsigned long page_none, page_shared, page_copy, page_readonly;
2676 unsigned long page_exec_bit;
4f93d21d 2677 int i;
c4bce90e
DM		 2678
2679 PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
2680 _PAGE_CACHE_4U | _PAGE_P_4U |
2681 __ACCESS_BITS_4U | __DIRTY_BITS_4U |
2682 _PAGE_EXEC_4U);
2683 PAGE_KERNEL_LOCKED = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
2684 _PAGE_CACHE_4U | _PAGE_P_4U |
2685 __ACCESS_BITS_4U | __DIRTY_BITS_4U |
2686 _PAGE_EXEC_4U | _PAGE_L_4U);
c4bce90e
DM		 2687
2688 _PAGE_IE = _PAGE_IE_4U;
2689 _PAGE_E = _PAGE_E_4U;
2690 _PAGE_CACHE = _PAGE_CACHE_4U;
2691
2692 pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4U | __DIRTY_BITS_4U |
2693 __ACCESS_BITS_4U | _PAGE_E_4U);
2694
d1acb421 2695#ifdef CONFIG_DEBUG_PAGEALLOC
922631b9 2696 kern_linear_pte_xor[0] = _PAGE_VALID ^ PAGE_OFFSET;
d1acb421 2697#else
9cc3a1ac 2698 kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4U) ^
922631b9 2699 PAGE_OFFSET;
d1acb421 2700#endif
9cc3a1ac
DM		 2701 kern_linear_pte_xor[0] |= (_PAGE_CP_4U | _PAGE_CV_4U |
2702 _PAGE_P_4U | _PAGE_W_4U);
2703
4f93d21d
DM		 2704 for (i = 1; i < 4; i++)
2705 kern_linear_pte_xor[i] = kern_linear_pte_xor[0];
c4bce90e 2706
c4bce90e
DM		 2707 _PAGE_ALL_SZ_BITS = (_PAGE_SZ4MB_4U | _PAGE_SZ512K_4U |
2708 _PAGE_SZ64K_4U | _PAGE_SZ8K_4U |
2709 _PAGE_SZ32MB_4U | _PAGE_SZ256MB_4U);
2710
2711
2712 page_none = _PAGE_PRESENT_4U | _PAGE_ACCESSED_4U | _PAGE_CACHE_4U;
2713 page_shared = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
2714 __ACCESS_BITS_4U | _PAGE_WRITE_4U | _PAGE_EXEC_4U);
2715 page_copy = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
2716 __ACCESS_BITS_4U | _PAGE_EXEC_4U);
2717 page_readonly = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
2718 __ACCESS_BITS_4U | _PAGE_EXEC_4U);
2719
2720 page_exec_bit = _PAGE_EXEC_4U;
2721
2722 prot_init_common(page_none, page_shared, page_copy, page_readonly,
2723 page_exec_bit);
2724}
2725
2726static void __init sun4v_pgprot_init(void)
2727{
2728 unsigned long page_none, page_shared, page_copy, page_readonly;
2729 unsigned long page_exec_bit;
4f93d21d 2730 int i;
c4bce90e
DM		 2731
2732 PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4V | _PAGE_VALID |
494e5b6f 2733 page_cache4v_flag | _PAGE_P_4V |
c4bce90e
DM		 2734 __ACCESS_BITS_4V | __DIRTY_BITS_4V |
2735 _PAGE_EXEC_4V);
2736 PAGE_KERNEL_LOCKED = PAGE_KERNEL;
c4bce90e
DM		 2737
2738 _PAGE_IE = _PAGE_IE_4V;
2739 _PAGE_E = _PAGE_E_4V;
494e5b6f 2740 _PAGE_CACHE = page_cache4v_flag;
c4bce90e 2741
d1acb421 2742#ifdef CONFIG_DEBUG_PAGEALLOC
922631b9 2743 kern_linear_pte_xor[0] = _PAGE_VALID ^ PAGE_OFFSET;
d1acb421 2744#else
9cc3a1ac 2745 kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4V) ^
922631b9 2746 PAGE_OFFSET;
d1acb421 2747#endif
494e5b6f
KA		 2748 kern_linear_pte_xor[0] |= (page_cache4v_flag | _PAGE_P_4V |
2749 _PAGE_W_4V);
9cc3a1ac 2750
c69ad0a3
DM		 2751 for (i = 1; i < 4; i++)
2752 kern_linear_pte_xor[i] = kern_linear_pte_xor[0];
4f93d21d 2753
c4bce90e
DM		 2754 pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4V | __DIRTY_BITS_4V |
2755 __ACCESS_BITS_4V | _PAGE_E_4V);
2756
c4bce90e
DM		 2757 _PAGE_ALL_SZ_BITS = (_PAGE_SZ16GB_4V | _PAGE_SZ2GB_4V |
2758 _PAGE_SZ256MB_4V | _PAGE_SZ32MB_4V |
2759 _PAGE_SZ4MB_4V | _PAGE_SZ512K_4V |
2760 _PAGE_SZ64K_4V | _PAGE_SZ8K_4V);
2761
494e5b6f
KA		 2762 page_none = _PAGE_PRESENT_4V | _PAGE_ACCESSED_4V | page_cache4v_flag;
2763 page_shared = (_PAGE_VALID | _PAGE_PRESENT_4V | page_cache4v_flag |
c4bce90e 2764 __ACCESS_BITS_4V | _PAGE_WRITE_4V | _PAGE_EXEC_4V);
494e5b6f 2765 page_copy = (_PAGE_VALID | _PAGE_PRESENT_4V | page_cache4v_flag |
c4bce90e 2766 __ACCESS_BITS_4V | _PAGE_EXEC_4V);
494e5b6f 2767 page_readonly = (_PAGE_VALID | _PAGE_PRESENT_4V | page_cache4v_flag |
c4bce90e
DM		 2768 __ACCESS_BITS_4V | _PAGE_EXEC_4V);
2769
2770 page_exec_bit = _PAGE_EXEC_4V;
2771
2772 prot_init_common(page_none, page_shared, page_copy, page_readonly,
2773 page_exec_bit);
2774}
2775
2776unsigned long pte_sz_bits(unsigned long sz)
2777{
2778 if (tlb_type == hypervisor) {
2779 switch (sz) {
2780 case 8 * 1024:
2781 default:
2782 return _PAGE_SZ8K_4V;
2783 case 64 * 1024:
2784 return _PAGE_SZ64K_4V;
2785 case 512 * 1024:
2786 return _PAGE_SZ512K_4V;
2787 case 4 * 1024 * 1024:
2788 return _PAGE_SZ4MB_4V;
6cb79b3f 2789 }
c4bce90e
DM		 2790 } else {
2791 switch (sz) {
2792 case 8 * 1024:
2793 default:
2794 return _PAGE_SZ8K_4U;
2795 case 64 * 1024:
2796 return _PAGE_SZ64K_4U;
2797 case 512 * 1024:
2798 return _PAGE_SZ512K_4U;
2799 case 4 * 1024 * 1024:
2800 return _PAGE_SZ4MB_4U;
6cb79b3f 2801 }
c4bce90e
DM		 2802 }
2803}
2804
2805pte_t mk_pte_io(unsigned long page, pgprot_t prot, int space, unsigned long page_size)
2806{
2807 pte_t pte;
cf627156
DM		 2808
2809 pte_val(pte) = page | pgprot_val(pgprot_noncached(prot));
c4bce90e
DM		 2810 pte_val(pte) |= (((unsigned long)space) << 32);
2811 pte_val(pte) |= pte_sz_bits(page_size);
c4bce90e 2812
cf627156 2813 return pte;
c4bce90e
DM		 2814}
2815
2816static unsigned long kern_large_tte(unsigned long paddr)
2817{
2818 unsigned long val;
2819
2820 val = (_PAGE_VALID | _PAGE_SZ4MB_4U |
2821 _PAGE_CP_4U | _PAGE_CV_4U | _PAGE_P_4U |
2822 _PAGE_EXEC_4U | _PAGE_L_4U | _PAGE_W_4U);
2823 if (tlb_type == hypervisor)
2824 val = (_PAGE_VALID | _PAGE_SZ4MB_4V |
494e5b6f 2825 page_cache4v_flag | _PAGE_P_4V |
c4bce90e
DM		 2826 _PAGE_EXEC_4V | _PAGE_W_4V);
2827
2828 return val | paddr;
2829}
2830
c4bce90e
DM		 2831/* If not locked, zap it. */
2832void __flush_tlb_all(void)
2833{
2834 unsigned long pstate;
2835 int i;
2836
2837 __asm__ __volatile__("flushw\n\t"
2838 "rdpr %%pstate, %0\n\t"
2839 "wrpr %0, %1, %%pstate"
2840 : "=r" (pstate)
2841 : "i" (PSTATE_IE));
8f361453
DM		 2842 if (tlb_type == hypervisor) {
2843 sun4v_mmu_demap_all();
2844 } else if (tlb_type == spitfire) {
c4bce90e
DM		 2845 for (i = 0; i < 64; i++) {
2846 /* Spitfire Errata #32 workaround */
2847 /* NOTE: Always runs on spitfire, so no
2848 * cheetah+ page size encodings.
2849 */
2850 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
2851 "flush %%g6"
2852 : /* No outputs */
2853 : "r" (0),
2854 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
2855
2856 if (!(spitfire_get_dtlb_data(i) & _PAGE_L_4U)) {
2857 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
2858 "membar #Sync"
2859 : /* no outputs */
2860 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
2861 spitfire_put_dtlb_data(i, 0x0UL);
2862 }
2863
2864 /* Spitfire Errata #32 workaround */
2865 /* NOTE: Always runs on spitfire, so no
2866 * cheetah+ page size encodings.
2867 */
2868 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
2869 "flush %%g6"
2870 : /* No outputs */
2871 : "r" (0),
2872 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
2873
2874 if (!(spitfire_get_itlb_data(i) & _PAGE_L_4U)) {
2875 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
2876 "membar #Sync"
2877 : /* no outputs */
2878 : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
2879 spitfire_put_itlb_data(i, 0x0UL);
2880 }
2881 }
2882 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
2883 cheetah_flush_dtlb_all();
2884 cheetah_flush_itlb_all();
2885 }
2886 __asm__ __volatile__("wrpr %0, 0, %%pstate"
2887 : : "r" (pstate));
2888}
c460bec7 2889
4cf58924 2890pte_t *pte_alloc_one_kernel(struct mm_struct *mm)
c460bec7 2891{
75f296d9 2892 struct page *page = alloc_page(GFP_KERNEL | __GFP_ZERO);
37b3a8ff 2893 pte_t *pte = NULL;
c460bec7 2894
c460bec7
DM		 2895 if (page)
2896 pte = (pte_t *) page_address(page);
2897
2898 return pte;
2899}
2900
4cf58924 2901pgtable_t pte_alloc_one(struct mm_struct *mm)
c460bec7 2902{
75f296d9 2903 struct page *page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1ae9ae5f
KS		 2904 if (!page)
2905 return NULL;
2906 if (!pgtable_page_ctor(page)) {
2d4894b5 2907 free_unref_page(page);
1ae9ae5f 2908 return NULL;
c460bec7 2909 }
1ae9ae5f 2910 return (pte_t *) page_address(page);
c460bec7
DM		 2911}
2912
2913void pte_free_kernel(struct mm_struct *mm, pte_t *pte)
2914{
37b3a8ff 2915 free_page((unsigned long)pte);
c460bec7
DM		 2916}
2917
2918static void __pte_free(pgtable_t pte)
2919{
2920 struct page *page = virt_to_page(pte);
37b3a8ff
DM		 2921
2922 pgtable_page_dtor(page);
2923 __free_page(page);
c460bec7
DM		 2924}
2925
2926void pte_free(struct mm_struct *mm, pgtable_t pte)
2927{
2928 __pte_free(pte);
2929}
2930
2931void pgtable_free(void *table, bool is_page)
2932{
2933 if (is_page)
2934 __pte_free(table);
2935 else
2936 kmem_cache_free(pgtable_cache, table);
2937}
9e695d2e
DM		 2938
2939#ifdef CONFIG_TRANSPARENT_HUGEPAGE
9e695d2e
DM		 2940void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr,
2941 pmd_t *pmd)
2942{
2943 unsigned long pte, flags;
2944 struct mm_struct *mm;
2945 pmd_t entry = *pmd;
9e695d2e
DM		 2946
2947 if (!pmd_large(entry) || !pmd_young(entry))
2948 return;
2949
a7b9403f 2950 pte = pmd_val(entry);
9e695d2e 2951
18f38132
DM		 2952 /* Don't insert a non-valid PMD into the TSB, we'll deadlock. */
2953 if (!(pte & _PAGE_VALID))
2954 return;
2955
37b3a8ff
DM		 2956 /* We are fabricating 8MB pages using 4MB real hw pages. */
2957 pte |= (addr & (1UL << REAL_HPAGE_SHIFT));
9e695d2e
DM		 2958
2959 mm = vma->vm_mm;
2960
2961 spin_lock_irqsave(&mm->context.lock, flags);
2962
2963 if (mm->context.tsb_block[MM_TSB_HUGE].tsb != NULL)
37b3a8ff 2964 __update_mmu_tsb_insert(mm, MM_TSB_HUGE, REAL_HPAGE_SHIFT,
9e695d2e
DM		 2965 addr, pte);
2966
2967 spin_unlock_irqrestore(&mm->context.lock, flags);
2968}
2969#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2970
2971#if defined(CONFIG_HUGETLB_PAGE) || defined(CONFIG_TRANSPARENT_HUGEPAGE)
2972static void context_reload(void *__data)
2973{
2974 struct mm_struct *mm = __data;
2975
2976 if (mm == current->mm)
2977 load_secondary_context(mm);
2978}
2979
0fbebed6 2980void hugetlb_setup(struct pt_regs *regs)
9e695d2e 2981{
0fbebed6
DM		 2982 struct mm_struct *mm = current->mm;
2983 struct tsb_config *tp;
9e695d2e 2984
70ffdb93 2985 if (faulthandler_disabled() || !mm) {
0fbebed6
DM		 2986 const struct exception_table_entry *entry;
2987
2988 entry = search_exception_tables(regs->tpc);
2989 if (entry) {
2990 regs->tpc = entry->fixup;
2991 regs->tnpc = regs->tpc + 4;
2992 return;
2993 }
2994 pr_alert("Unexpected HugeTLB setup in atomic context.\n");
2995 die_if_kernel("HugeTSB in atomic", regs);
2996 }
2997
2998 tp = &mm->context.tsb_block[MM_TSB_HUGE];
2999 if (likely(tp->tsb == NULL))
3000 tsb_grow(mm, MM_TSB_HUGE, 0);
9e695d2e 3001
9e695d2e
DM		 3002 tsb_context_switch(mm);
3003 smp_tsb_sync(mm);
3004
3005 /* On UltraSPARC-III+ and later, configure the second half of
3006 * the Data-TLB for huge pages.
3007 */
3008 if (tlb_type == cheetah_plus) {
9ea46abe 3009 bool need_context_reload = false;
9e695d2e
DM		 3010 unsigned long ctx;
3011
9ea46abe 3012 spin_lock_irq(&ctx_alloc_lock);
9e695d2e
DM		 3013 ctx = mm->context.sparc64_ctx_val;
3014 ctx &= ~CTX_PGSZ_MASK;
3015 ctx |= CTX_PGSZ_BASE << CTX_PGSZ0_SHIFT;
3016 ctx |= CTX_PGSZ_HUGE << CTX_PGSZ1_SHIFT;
3017
3018 if (ctx != mm->context.sparc64_ctx_val) {
3019 /* When changing the page size fields, we
3020 * must perform a context flush so that no
3021 * stale entries match. This flush must
3022 * occur with the original context register
3023 * settings.
3024 */
3025 do_flush_tlb_mm(mm);
3026
3027 /* Reload the context register of all processors
3028 * also executing in this address space.
3029 */
3030 mm->context.sparc64_ctx_val = ctx;
9ea46abe 3031 need_context_reload = true;
9e695d2e 3032 }
9ea46abe
DM		 3033 spin_unlock_irq(&ctx_alloc_lock);
3034
3035 if (need_context_reload)
3036 on_each_cpu(context_reload, mm, 0);
9e695d2e
DM		 3037 }
3038}
3039#endif
f6d4fb5c 3040
3041static struct resource code_resource = {
3042 .name = "Kernel code",
35d98e93 3043 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
f6d4fb5c 3044};
3045
3046static struct resource data_resource = {
3047 .name = "Kernel data",
35d98e93 3048 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
f6d4fb5c 3049};
3050
3051static struct resource bss_resource = {
3052 .name = "Kernel bss",
35d98e93 3053 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
f6d4fb5c 3054};
3055
3056static inline resource_size_t compute_kern_paddr(void *addr)
3057{
3058 return (resource_size_t) (addr - KERNBASE + kern_base);
3059}
3060
3061static void __init kernel_lds_init(void)
3062{
3063 code_resource.start = compute_kern_paddr(_text);
3064 code_resource.end = compute_kern_paddr(_etext - 1);
3065 data_resource.start = compute_kern_paddr(_etext);
3066 data_resource.end = compute_kern_paddr(_edata - 1);
3067 bss_resource.start = compute_kern_paddr(__bss_start);
3068 bss_resource.end = compute_kern_paddr(_end - 1);
3069}
3070
3071static int __init report_memory(void)
3072{
3073 int i;
3074 struct resource *res;
3075
3076 kernel_lds_init();
3077
3078 for (i = 0; i < pavail_ents; i++) {
3079 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
3080
3081 if (!res) {
3082 pr_warn("Failed to allocate source.\n");
3083 break;
3084 }
3085
3086 res->name = "System RAM";
3087 res->start = pavail[i].phys_addr;
3088 res->end = pavail[i].phys_addr + pavail[i].reg_size - 1;
35d98e93 3089 res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
f6d4fb5c 3090
3091 if (insert_resource(&iomem_resource, res) < 0) {
3092 pr_warn("Resource insertion failed.\n");
3093 break;
3094 }
3095
3096 insert_resource(res, &code_resource);
3097 insert_resource(res, &data_resource);
3098 insert_resource(res, &bss_resource);
3099 }
3100
3101 return 0;
3102}
3c08158e 3103arch_initcall(report_memory);
e9011d08 3104
4ca9a237
DM		 3105#ifdef CONFIG_SMP
3106#define do_flush_tlb_kernel_range smp_flush_tlb_kernel_range
3107#else
3108#define do_flush_tlb_kernel_range __flush_tlb_kernel_range
3109#endif
3110
3111void flush_tlb_kernel_range(unsigned long start, unsigned long end)
3112{
3113 if (start < HI_OBP_ADDRESS && end > LOW_OBP_ADDRESS) {
3114 if (start < LOW_OBP_ADDRESS) {
3115 flush_tsb_kernel_range(start, LOW_OBP_ADDRESS);
3116 do_flush_tlb_kernel_range(start, LOW_OBP_ADDRESS);
3117 }
3118 if (end > HI_OBP_ADDRESS) {
473ad7f4
DM		 3119 flush_tsb_kernel_range(HI_OBP_ADDRESS, end);
3120 do_flush_tlb_kernel_range(HI_OBP_ADDRESS, end);
4ca9a237
DM		 3121 }
3122 } else {
3123 flush_tsb_kernel_range(start, end);
3124 do_flush_tlb_kernel_range(start, end);
3125 }
3126}
74a04967
KA		 3127
3128void copy_user_highpage(struct page *to, struct page *from,
3129 unsigned long vaddr, struct vm_area_struct *vma)
3130{
3131 char *vfrom, *vto;
3132
3133 vfrom = kmap_atomic(from);
3134 vto = kmap_atomic(to);
3135 copy_user_page(vto, vfrom, vaddr, to);
3136 kunmap_atomic(vto);
3137 kunmap_atomic(vfrom);
3138
3139 /* If this page has ADI enabled, copy over any ADI tags
3140 * as well
3141 */
3142 if (vma->vm_flags & VM_SPARC_ADI) {
3143 unsigned long pfrom, pto, i, adi_tag;
3144
3145 pfrom = page_to_phys(from);
3146 pto = page_to_phys(to);
3147
3148 for (i = pfrom; i < (pfrom + PAGE_SIZE); i += adi_blksize()) {
3149 asm volatile("ldxa [%1] %2, %0\n\t"
3150 : "=r" (adi_tag)
3151 : "r" (i), "i" (ASI_MCD_REAL));
3152 asm volatile("stxa %0, [%1] %2\n\t"
3153 :
3154 : "r" (adi_tag), "r" (pto),
3155 "i" (ASI_MCD_REAL));
3156 pto += adi_blksize();
3157 }
3158 asm volatile("membar #Sync\n\t");
3159 }
3160}
3161EXPORT_SYMBOL(copy_user_highpage);
3162
3163void copy_highpage(struct page *to, struct page *from)
3164{
3165 char *vfrom, *vto;
3166
3167 vfrom = kmap_atomic(from);
3168 vto = kmap_atomic(to);
3169 copy_page(vto, vfrom);
3170 kunmap_atomic(vto);
3171 kunmap_atomic(vfrom);
3172
3173 /* If this platform is ADI enabled, copy any ADI tags
3174 * as well
3175 */
3176 if (adi_capable()) {
3177 unsigned long pfrom, pto, i, adi_tag;
3178
3179 pfrom = page_to_phys(from);
3180 pto = page_to_phys(to);
3181
3182 for (i = pfrom; i < (pfrom + PAGE_SIZE); i += adi_blksize()) {
3183 asm volatile("ldxa [%1] %2, %0\n\t"
3184 : "=r" (adi_tag)
3185 : "r" (i), "i" (ASI_MCD_REAL));
3186 asm volatile("stxa %0, [%1] %2\n\t"
3187 :
3188 : "r" (adi_tag), "r" (pto),
3189 "i" (ASI_MCD_REAL));
3190 pto += adi_blksize();
3191 }
3192 asm volatile("membar #Sync\n\t");
3193 }
3194}
3195EXPORT_SYMBOL(copy_highpage);
Ubuntu Eoan Linux kernel mirror