]> git.proxmox.com Git - mirror_ubuntu-kernels.git/blame - arch/powerpc/mm/hash_utils_64.c
Merge branches 'for-5.1/upstream-fixes', 'for-5.2/core', 'for-5.2/ish', 'for-5.2...
[mirror_ubuntu-kernels.git] / arch / powerpc / mm / hash_utils_64.c
CommitLineData
1da177e4
LT
1/*
2 * PowerPC64 port by Mike Corrigan and Dave Engebretsen
3 * {mikejc|engebret}@us.ibm.com
4 *
5 * Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com>
6 *
7 * SMP scalability work:
8 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
9 *
10 * Module name: htab.c
11 *
12 * Description:
13 * PowerPC Hashed Page Table functions
14 *
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
19 */
20
21#undef DEBUG
3c726f8d 22#undef DEBUG_LOW
1da177e4 23
7f142661 24#define pr_fmt(fmt) "hash-mmu: " fmt
1da177e4
LT
25#include <linux/spinlock.h>
26#include <linux/errno.h>
589ee628 27#include <linux/sched/mm.h>
1da177e4
LT
28#include <linux/proc_fs.h>
29#include <linux/stat.h>
30#include <linux/sysctl.h>
66b15db6 31#include <linux/export.h>
1da177e4
LT
32#include <linux/ctype.h>
33#include <linux/cache.h>
34#include <linux/init.h>
35#include <linux/signal.h>
95f72d1e 36#include <linux/memblock.h>
ba12eede 37#include <linux/context_tracking.h>
5556ecf5 38#include <linux/libfdt.h>
92e3da3c 39#include <linux/pkeys.h>
1da177e4 40
7644d581 41#include <asm/debugfs.h>
1da177e4
LT
42#include <asm/processor.h>
43#include <asm/pgtable.h>
44#include <asm/mmu.h>
45#include <asm/mmu_context.h>
46#include <asm/page.h>
47#include <asm/types.h>
7c0f6ba6 48#include <linux/uaccess.h>
1da177e4 49#include <asm/machdep.h>
d9b2b2a2 50#include <asm/prom.h>
1da177e4
LT
51#include <asm/io.h>
52#include <asm/eeh.h>
53#include <asm/tlb.h>
54#include <asm/cacheflush.h>
55#include <asm/cputable.h>
1da177e4 56#include <asm/sections.h>
be3ebfe8 57#include <asm/copro.h>
aa39be09 58#include <asm/udbg.h>
b68a70c4 59#include <asm/code-patching.h>
3ccc00a7 60#include <asm/fadump.h>
f5339277 61#include <asm/firmware.h>
bc2a9408 62#include <asm/tm.h>
cfcb3d80 63#include <asm/trace.h>
166dd7d3 64#include <asm/ps3.h>
94171b19 65#include <asm/pte-walk.h>
eacbb218 66#include <asm/asm-prototypes.h>
1da177e4
LT
67
68#ifdef DEBUG
69#define DBG(fmt...) udbg_printf(fmt)
70#else
71#define DBG(fmt...)
72#endif
73
3c726f8d
BH
74#ifdef DEBUG_LOW
75#define DBG_LOW(fmt...) udbg_printf(fmt)
76#else
77#define DBG_LOW(fmt...)
78#endif
79
80#define KB (1024)
81#define MB (1024*KB)
658013e9 82#define GB (1024L*MB)
3c726f8d 83
1da177e4
LT
84/*
85 * Note: pte --> Linux PTE
86 * HPTE --> PowerPC Hashed Page Table Entry
87 *
88 * Execution context:
89 * htab_initialize is called with the MMU off (of course), but
90 * the kernel has been copied down to zero so it can directly
91 * reference global data. At this point it is very difficult
92 * to print debug info.
93 *
94 */
95
799d6046
PM
96static unsigned long _SDR1;
97struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
e1802b06 98EXPORT_SYMBOL_GPL(mmu_psize_defs);
799d6046 99
0eeede0c
PM
100u8 hpte_page_sizes[1 << LP_BITS];
101EXPORT_SYMBOL_GPL(hpte_page_sizes);
102
8e561e7e 103struct hash_pte *htab_address;
337a7128 104unsigned long htab_size_bytes;
96e28449 105unsigned long htab_hash_mask;
4ab79aa8 106EXPORT_SYMBOL_GPL(htab_hash_mask);
3c726f8d 107int mmu_linear_psize = MMU_PAGE_4K;
8ca7a82f 108EXPORT_SYMBOL_GPL(mmu_linear_psize);
3c726f8d 109int mmu_virtual_psize = MMU_PAGE_4K;
bf72aeba 110int mmu_vmalloc_psize = MMU_PAGE_4K;
cec08e7a
BH
111#ifdef CONFIG_SPARSEMEM_VMEMMAP
112int mmu_vmemmap_psize = MMU_PAGE_4K;
113#endif
bf72aeba 114int mmu_io_psize = MMU_PAGE_4K;
1189be65 115int mmu_kernel_ssize = MMU_SEGSIZE_256M;
8ca7a82f 116EXPORT_SYMBOL_GPL(mmu_kernel_ssize);
1189be65 117int mmu_highuser_ssize = MMU_SEGSIZE_256M;
584f8b71 118u16 mmu_slb_size = 64;
4ab79aa8 119EXPORT_SYMBOL_GPL(mmu_slb_size);
bf72aeba
PM
120#ifdef CONFIG_PPC_64K_PAGES
121int mmu_ci_restrictions;
122#endif
370a908d
BH
123#ifdef CONFIG_DEBUG_PAGEALLOC
124static u8 *linear_map_hash_slots;
125static unsigned long linear_map_hash_count;
ed166692 126static DEFINE_SPINLOCK(linear_map_hash_lock);
370a908d 127#endif /* CONFIG_DEBUG_PAGEALLOC */
7025776e
BH
128struct mmu_hash_ops mmu_hash_ops;
129EXPORT_SYMBOL(mmu_hash_ops);
1da177e4 130
3c726f8d
BH
131/* There are definitions of page sizes arrays to be used when none
132 * is provided by the firmware.
133 */
1da177e4 134
471d7ff8
NP
135/*
136 * Fallback (4k pages only)
3c726f8d 137 */
471d7ff8 138static struct mmu_psize_def mmu_psize_defaults[] = {
3c726f8d
BH
139 [MMU_PAGE_4K] = {
140 .shift = 12,
141 .sllp = 0,
b1022fbd 142 .penc = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1},
3c726f8d
BH
143 .avpnm = 0,
144 .tlbiel = 0,
145 },
146};
147
148/* POWER4, GPUL, POWER5
149 *
150 * Support for 16Mb large pages
151 */
09de9ff8 152static struct mmu_psize_def mmu_psize_defaults_gp[] = {
3c726f8d
BH
153 [MMU_PAGE_4K] = {
154 .shift = 12,
155 .sllp = 0,
b1022fbd 156 .penc = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1},
3c726f8d
BH
157 .avpnm = 0,
158 .tlbiel = 1,
159 },
160 [MMU_PAGE_16M] = {
161 .shift = 24,
162 .sllp = SLB_VSID_L,
b1022fbd
AK
163 .penc = {[0 ... MMU_PAGE_16M - 1] = -1, [MMU_PAGE_16M] = 0,
164 [MMU_PAGE_16M + 1 ... MMU_PAGE_COUNT - 1] = -1 },
3c726f8d
BH
165 .avpnm = 0x1UL,
166 .tlbiel = 0,
167 },
168};
169
dc47c0c1
AK
170/*
171 * 'R' and 'C' update notes:
172 * - Under pHyp or KVM, the updatepp path will not set C, thus it *will*
173 * create writeable HPTEs without C set, because the hcall H_PROTECT
174 * that we use in that case will not update C
175 * - The above is however not a problem, because we also don't do that
176 * fancy "no flush" variant of eviction and we use H_REMOVE which will
177 * do the right thing and thus we don't have the race I described earlier
178 *
179 * - Under bare metal, we do have the race, so we need R and C set
180 * - We make sure R is always set and never lost
181 * - C is _PAGE_DIRTY, and *should* always be set for a writeable mapping
182 */
c6a3c495 183unsigned long htab_convert_pte_flags(unsigned long pteflags)
bc033b63 184{
c6a3c495 185 unsigned long rflags = 0;
bc033b63
BH
186
187 /* _PAGE_EXEC -> NOEXEC */
188 if ((pteflags & _PAGE_EXEC) == 0)
189 rflags |= HPTE_R_N;
c6a3c495 190 /*
e58e87ad 191 * PPP bits:
1ec3f937 192 * Linux uses slb key 0 for kernel and 1 for user.
e58e87ad
AK
193 * kernel RW areas are mapped with PPP=0b000
194 * User area is mapped with PPP=0b010 for read/write
195 * or PPP=0b011 for read-only (including writeable but clean pages).
bc033b63 196 */
e58e87ad
AK
197 if (pteflags & _PAGE_PRIVILEGED) {
198 /*
199 * Kernel read only mapped with ppp bits 0b110
200 */
984d7a1e
AK
201 if (!(pteflags & _PAGE_WRITE)) {
202 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
203 rflags |= (HPTE_R_PP0 | 0x2);
204 else
205 rflags |= 0x3;
206 }
e58e87ad 207 } else {
c7d54842
AK
208 if (pteflags & _PAGE_RWX)
209 rflags |= 0x2;
210 if (!((pteflags & _PAGE_WRITE) && (pteflags & _PAGE_DIRTY)))
c6a3c495
AK
211 rflags |= 0x1;
212 }
c8c06f5a 213 /*
dc47c0c1
AK
214 * We can't allow hardware to update hpte bits. Hence always
215 * set 'R' bit and set 'C' if it is a write fault
c8c06f5a 216 */
e568006b 217 rflags |= HPTE_R_R;
dc47c0c1
AK
218
219 if (pteflags & _PAGE_DIRTY)
220 rflags |= HPTE_R_C;
40e8550a
AK
221 /*
222 * Add in WIG bits
223 */
30bda41a
AK
224
225 if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_TOLERANT)
40e8550a 226 rflags |= HPTE_R_I;
e568006b 227 else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_NON_IDEMPOTENT)
30bda41a 228 rflags |= (HPTE_R_I | HPTE_R_G);
e568006b
AK
229 else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_SAO)
230 rflags |= (HPTE_R_W | HPTE_R_I | HPTE_R_M);
231 else
232 /*
233 * Add memory coherence if cache inhibited is not set
234 */
235 rflags |= HPTE_R_M;
40e8550a 236
a6590ca5 237 rflags |= pte_to_hpte_pkey_bits(pteflags);
40e8550a 238 return rflags;
bc033b63 239}
3c726f8d
BH
240
241int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
bc033b63 242 unsigned long pstart, unsigned long prot,
1189be65 243 int psize, int ssize)
1da177e4 244{
3c726f8d
BH
245 unsigned long vaddr, paddr;
246 unsigned int step, shift;
3c726f8d 247 int ret = 0;
1da177e4 248
3c726f8d
BH
249 shift = mmu_psize_defs[psize].shift;
250 step = 1 << shift;
1da177e4 251
bc033b63
BH
252 prot = htab_convert_pte_flags(prot);
253
254 DBG("htab_bolt_mapping(%lx..%lx -> %lx (%lx,%d,%d)\n",
255 vstart, vend, pstart, prot, psize, ssize);
256
3c726f8d
BH
257 for (vaddr = vstart, paddr = pstart; vaddr < vend;
258 vaddr += step, paddr += step) {
370a908d 259 unsigned long hash, hpteg;
1189be65 260 unsigned long vsid = get_kernel_vsid(vaddr, ssize);
5524a27d 261 unsigned long vpn = hpt_vpn(vaddr, vsid, ssize);
9e88ba4e
PM
262 unsigned long tprot = prot;
263
c60ac569
AK
264 /*
265 * If we hit a bad address return error.
266 */
267 if (!vsid)
268 return -1;
9e88ba4e 269 /* Make kernel text executable */
549e8152 270 if (overlaps_kernel_text(vaddr, vaddr + step))
9e88ba4e 271 tprot &= ~HPTE_R_N;
1da177e4 272
b18db0b8
AG
273 /* Make kvm guest trampolines executable */
274 if (overlaps_kvm_tmp(vaddr, vaddr + step))
275 tprot &= ~HPTE_R_N;
276
429d2e83
MS
277 /*
278 * If relocatable, check if it overlaps interrupt vectors that
279 * are copied down to real 0. For relocatable kernel
280 * (e.g. kdump case) we copy interrupt vectors down to real
281 * address 0. Mark that region as executable. This is
282 * because on p8 system with relocation on exception feature
283 * enabled, exceptions are raised with MMU (IR=DR=1) ON. Hence
284 * in order to execute the interrupt handlers in virtual
285 * mode the vector region need to be marked as executable.
286 */
287 if ((PHYSICAL_START > MEMORY_START) &&
288 overlaps_interrupt_vector_text(vaddr, vaddr + step))
289 tprot &= ~HPTE_R_N;
290
5524a27d 291 hash = hpt_hash(vpn, shift, ssize);
1da177e4
LT
292 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
293
7025776e
BH
294 BUG_ON(!mmu_hash_ops.hpte_insert);
295 ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
296 HPTE_V_BOLTED, psize, psize,
297 ssize);
c30a4df3 298
3c726f8d
BH
299 if (ret < 0)
300 break;
e7df0d88 301
370a908d 302#ifdef CONFIG_DEBUG_PAGEALLOC
e7df0d88
JK
303 if (debug_pagealloc_enabled() &&
304 (paddr >> PAGE_SHIFT) < linear_map_hash_count)
370a908d
BH
305 linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80;
306#endif /* CONFIG_DEBUG_PAGEALLOC */
3c726f8d
BH
307 }
308 return ret < 0 ? ret : 0;
309}
1da177e4 310
ed5694a8 311int htab_remove_mapping(unsigned long vstart, unsigned long vend,
f8c8803b
BP
312 int psize, int ssize)
313{
314 unsigned long vaddr;
315 unsigned int step, shift;
27828f98
DG
316 int rc;
317 int ret = 0;
f8c8803b
BP
318
319 shift = mmu_psize_defs[psize].shift;
320 step = 1 << shift;
321
7025776e 322 if (!mmu_hash_ops.hpte_removebolted)
abd0a0e7 323 return -ENODEV;
f8c8803b 324
27828f98 325 for (vaddr = vstart; vaddr < vend; vaddr += step) {
7025776e 326 rc = mmu_hash_ops.hpte_removebolted(vaddr, psize, ssize);
27828f98
DG
327 if (rc == -ENOENT) {
328 ret = -ENOENT;
329 continue;
330 }
331 if (rc < 0)
332 return rc;
333 }
52db9b44 334
27828f98 335 return ret;
f8c8803b
BP
336}
337
faf78829
OH
338static bool disable_1tb_segments = false;
339
340static int __init parse_disable_1tb_segments(char *p)
341{
342 disable_1tb_segments = true;
343 return 0;
344}
345early_param("disable_1tb_segments", parse_disable_1tb_segments);
346
1189be65
PM
347static int __init htab_dt_scan_seg_sizes(unsigned long node,
348 const char *uname, int depth,
349 void *data)
350{
9d0c4dfe
RH
351 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
352 const __be32 *prop;
353 int size = 0;
1189be65
PM
354
355 /* We are scanning "cpu" nodes only */
356 if (type == NULL || strcmp(type, "cpu") != 0)
357 return 0;
358
12f04f2b 359 prop = of_get_flat_dt_prop(node, "ibm,processor-segment-sizes", &size);
1189be65
PM
360 if (prop == NULL)
361 return 0;
362 for (; size >= 4; size -= 4, ++prop) {
12f04f2b 363 if (be32_to_cpu(prop[0]) == 40) {
1189be65 364 DBG("1T segment support detected\n");
faf78829
OH
365
366 if (disable_1tb_segments) {
367 DBG("1T segments disabled by command line\n");
368 break;
369 }
370
44ae3ab3 371 cur_cpu_spec->mmu_features |= MMU_FTR_1T_SEGMENT;
f5534004 372 return 1;
1189be65 373 }
1189be65 374 }
44ae3ab3 375 cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
1189be65
PM
376 return 0;
377}
378
b1022fbd
AK
379static int __init get_idx_from_shift(unsigned int shift)
380{
381 int idx = -1;
382
383 switch (shift) {
384 case 0xc:
385 idx = MMU_PAGE_4K;
386 break;
387 case 0x10:
388 idx = MMU_PAGE_64K;
389 break;
390 case 0x14:
391 idx = MMU_PAGE_1M;
392 break;
393 case 0x18:
394 idx = MMU_PAGE_16M;
395 break;
396 case 0x22:
397 idx = MMU_PAGE_16G;
398 break;
399 }
400 return idx;
401}
402
3c726f8d
BH
403static int __init htab_dt_scan_page_sizes(unsigned long node,
404 const char *uname, int depth,
405 void *data)
406{
9d0c4dfe
RH
407 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
408 const __be32 *prop;
409 int size = 0;
3c726f8d
BH
410
411 /* We are scanning "cpu" nodes only */
412 if (type == NULL || strcmp(type, "cpu") != 0)
413 return 0;
414
12f04f2b 415 prop = of_get_flat_dt_prop(node, "ibm,segment-page-sizes", &size);
9e34992a
ME
416 if (!prop)
417 return 0;
418
419 pr_info("Page sizes from device-tree:\n");
420 size /= 4;
421 cur_cpu_spec->mmu_features &= ~(MMU_FTR_16M_PAGE);
422 while(size > 0) {
423 unsigned int base_shift = be32_to_cpu(prop[0]);
424 unsigned int slbenc = be32_to_cpu(prop[1]);
425 unsigned int lpnum = be32_to_cpu(prop[2]);
426 struct mmu_psize_def *def;
427 int idx, base_idx;
428
429 size -= 3; prop += 3;
430 base_idx = get_idx_from_shift(base_shift);
431 if (base_idx < 0) {
432 /* skip the pte encoding also */
433 prop += lpnum * 2; size -= lpnum * 2;
434 continue;
435 }
436 def = &mmu_psize_defs[base_idx];
437 if (base_idx == MMU_PAGE_16M)
438 cur_cpu_spec->mmu_features |= MMU_FTR_16M_PAGE;
439
440 def->shift = base_shift;
441 if (base_shift <= 23)
442 def->avpnm = 0;
443 else
444 def->avpnm = (1 << (base_shift - 23)) - 1;
445 def->sllp = slbenc;
446 /*
447 * We don't know for sure what's up with tlbiel, so
448 * for now we only set it for 4K and 64K pages
449 */
450 if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K)
451 def->tlbiel = 1;
452 else
453 def->tlbiel = 0;
454
455 while (size > 0 && lpnum) {
456 unsigned int shift = be32_to_cpu(prop[0]);
457 int penc = be32_to_cpu(prop[1]);
458
459 prop += 2; size -= 2;
460 lpnum--;
461
462 idx = get_idx_from_shift(shift);
463 if (idx < 0)
b1022fbd 464 continue;
9e34992a
ME
465
466 if (penc == -1)
467 pr_err("Invalid penc for base_shift=%d "
468 "shift=%d\n", base_shift, shift);
469
470 def->penc[idx] = penc;
471 pr_info("base_shift=%d: shift=%d, sllp=0x%04lx,"
472 " avpnm=0x%08lx, tlbiel=%d, penc=%d\n",
473 base_shift, shift, def->sllp,
474 def->avpnm, def->tlbiel, def->penc[idx]);
1da177e4 475 }
3c726f8d 476 }
9e34992a
ME
477
478 return 1;
3c726f8d
BH
479}
480
e16a9c09 481#ifdef CONFIG_HUGETLB_PAGE
658013e9
JT
482/* Scan for 16G memory blocks that have been set aside for huge pages
483 * and reserve those blocks for 16G huge pages.
484 */
485static int __init htab_dt_scan_hugepage_blocks(unsigned long node,
486 const char *uname, int depth,
487 void *data) {
9d0c4dfe
RH
488 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
489 const __be64 *addr_prop;
490 const __be32 *page_count_prop;
658013e9
JT
491 unsigned int expected_pages;
492 long unsigned int phys_addr;
493 long unsigned int block_size;
494
495 /* We are scanning "memory" nodes only */
496 if (type == NULL || strcmp(type, "memory") != 0)
497 return 0;
498
499 /* This property is the log base 2 of the number of virtual pages that
500 * will represent this memory block. */
501 page_count_prop = of_get_flat_dt_prop(node, "ibm,expected#pages", NULL);
502 if (page_count_prop == NULL)
503 return 0;
12f04f2b 504 expected_pages = (1 << be32_to_cpu(page_count_prop[0]));
658013e9
JT
505 addr_prop = of_get_flat_dt_prop(node, "reg", NULL);
506 if (addr_prop == NULL)
507 return 0;
12f04f2b
AB
508 phys_addr = be64_to_cpu(addr_prop[0]);
509 block_size = be64_to_cpu(addr_prop[1]);
658013e9
JT
510 if (block_size != (16 * GB))
511 return 0;
512 printk(KERN_INFO "Huge page(16GB) memory: "
513 "addr = 0x%lX size = 0x%lX pages = %d\n",
514 phys_addr, block_size, expected_pages);
23493c12 515 if (phys_addr + block_size * expected_pages <= memblock_end_of_DRAM()) {
95f72d1e 516 memblock_reserve(phys_addr, block_size * expected_pages);
79cc38de 517 pseries_add_gpage(phys_addr, block_size, expected_pages);
4792adba 518 }
658013e9
JT
519 return 0;
520}
e16a9c09 521#endif /* CONFIG_HUGETLB_PAGE */
658013e9 522
b1022fbd
AK
523static void mmu_psize_set_default_penc(void)
524{
525 int bpsize, apsize;
526 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
527 for (apsize = 0; apsize < MMU_PAGE_COUNT; apsize++)
528 mmu_psize_defs[bpsize].penc[apsize] = -1;
529}
530
9048e648
AG
531#ifdef CONFIG_PPC_64K_PAGES
532
533static bool might_have_hea(void)
534{
535 /*
536 * The HEA ethernet adapter requires awareness of the
537 * GX bus. Without that awareness we can easily assume
538 * we will never see an HEA ethernet device.
539 */
540#ifdef CONFIG_IBMEBUS
2b4e3ad8 541 return !cpu_has_feature(CPU_FTR_ARCH_207S) &&
08bf75ba 542 firmware_has_feature(FW_FEATURE_SPLPAR);
9048e648
AG
543#else
544 return false;
545#endif
546}
547
548#endif /* #ifdef CONFIG_PPC_64K_PAGES */
549
bacf9cf8 550static void __init htab_scan_page_sizes(void)
3c726f8d
BH
551{
552 int rc;
553
b1022fbd
AK
554 /* se the invalid penc to -1 */
555 mmu_psize_set_default_penc();
556
3c726f8d 557 /* Default to 4K pages only */
471d7ff8
NP
558 memcpy(mmu_psize_defs, mmu_psize_defaults,
559 sizeof(mmu_psize_defaults));
3c726f8d
BH
560
561 /*
562 * Try to find the available page sizes in the device-tree
563 */
564 rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
b8f1b4f8 565 if (rc == 0 && early_mmu_has_feature(MMU_FTR_16M_PAGE)) {
bacf9cf8
ME
566 /*
567 * Nothing in the device-tree, but the CPU supports 16M pages,
568 * so let's fallback on a known size list for 16M capable CPUs.
569 */
3c726f8d
BH
570 memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
571 sizeof(mmu_psize_defaults_gp));
bacf9cf8
ME
572 }
573
574#ifdef CONFIG_HUGETLB_PAGE
85975387
HB
575 if (!hugetlb_disabled) {
576 /* Reserve 16G huge page memory sections for huge pages */
577 of_scan_flat_dt(htab_dt_scan_hugepage_blocks, NULL);
578 }
bacf9cf8
ME
579#endif /* CONFIG_HUGETLB_PAGE */
580}
581
0eeede0c
PM
582/*
583 * Fill in the hpte_page_sizes[] array.
584 * We go through the mmu_psize_defs[] array looking for all the
585 * supported base/actual page size combinations. Each combination
586 * has a unique pagesize encoding (penc) value in the low bits of
587 * the LP field of the HPTE. For actual page sizes less than 1MB,
588 * some of the upper LP bits are used for RPN bits, meaning that
589 * we need to fill in several entries in hpte_page_sizes[].
590 *
591 * In diagrammatic form, with r = RPN bits and z = page size bits:
592 * PTE LP actual page size
593 * rrrr rrrz >=8KB
594 * rrrr rrzz >=16KB
595 * rrrr rzzz >=32KB
596 * rrrr zzzz >=64KB
597 * ...
598 *
599 * The zzzz bits are implementation-specific but are chosen so that
600 * no encoding for a larger page size uses the same value in its
601 * low-order N bits as the encoding for the 2^(12+N) byte page size
602 * (if it exists).
603 */
604static void init_hpte_page_sizes(void)
605{
606 long int ap, bp;
607 long int shift, penc;
608
609 for (bp = 0; bp < MMU_PAGE_COUNT; ++bp) {
610 if (!mmu_psize_defs[bp].shift)
611 continue; /* not a supported page size */
612 for (ap = bp; ap < MMU_PAGE_COUNT; ++ap) {
613 penc = mmu_psize_defs[bp].penc[ap];
10527e80 614 if (penc == -1 || !mmu_psize_defs[ap].shift)
0eeede0c
PM
615 continue;
616 shift = mmu_psize_defs[ap].shift - LP_SHIFT;
617 if (shift <= 0)
618 continue; /* should never happen */
619 /*
620 * For page sizes less than 1MB, this loop
621 * replicates the entry for all possible values
622 * of the rrrr bits.
623 */
624 while (penc < (1 << LP_BITS)) {
625 hpte_page_sizes[penc] = (ap << 4) | bp;
626 penc += 1 << shift;
627 }
628 }
629 }
630}
631
bacf9cf8
ME
632static void __init htab_init_page_sizes(void)
633{
0eeede0c
PM
634 init_hpte_page_sizes();
635
e7df0d88
JK
636 if (!debug_pagealloc_enabled()) {
637 /*
638 * Pick a size for the linear mapping. Currently, we only
639 * support 16M, 1M and 4K which is the default
640 */
641 if (mmu_psize_defs[MMU_PAGE_16M].shift)
642 mmu_linear_psize = MMU_PAGE_16M;
643 else if (mmu_psize_defs[MMU_PAGE_1M].shift)
644 mmu_linear_psize = MMU_PAGE_1M;
645 }
3c726f8d 646
bf72aeba 647#ifdef CONFIG_PPC_64K_PAGES
3c726f8d
BH
648 /*
649 * Pick a size for the ordinary pages. Default is 4K, we support
bf72aeba
PM
650 * 64K for user mappings and vmalloc if supported by the processor.
651 * We only use 64k for ioremap if the processor
652 * (and firmware) support cache-inhibited large pages.
653 * If not, we use 4k and set mmu_ci_restrictions so that
654 * hash_page knows to switch processes that use cache-inhibited
655 * mappings to 4k pages.
3c726f8d 656 */
bf72aeba 657 if (mmu_psize_defs[MMU_PAGE_64K].shift) {
3c726f8d 658 mmu_virtual_psize = MMU_PAGE_64K;
bf72aeba 659 mmu_vmalloc_psize = MMU_PAGE_64K;
370a908d
BH
660 if (mmu_linear_psize == MMU_PAGE_4K)
661 mmu_linear_psize = MMU_PAGE_64K;
44ae3ab3 662 if (mmu_has_feature(MMU_FTR_CI_LARGE_PAGE)) {
cfe666b1 663 /*
9048e648
AG
664 * When running on pSeries using 64k pages for ioremap
665 * would stop us accessing the HEA ethernet. So if we
666 * have the chance of ever seeing one, stay at 4k.
cfe666b1 667 */
2b4e3ad8 668 if (!might_have_hea())
cfe666b1
PM
669 mmu_io_psize = MMU_PAGE_64K;
670 } else
bf72aeba
PM
671 mmu_ci_restrictions = 1;
672 }
370a908d 673#endif /* CONFIG_PPC_64K_PAGES */
3c726f8d 674
cec08e7a
BH
675#ifdef CONFIG_SPARSEMEM_VMEMMAP
676 /* We try to use 16M pages for vmemmap if that is supported
677 * and we have at least 1G of RAM at boot
678 */
679 if (mmu_psize_defs[MMU_PAGE_16M].shift &&
95f72d1e 680 memblock_phys_mem_size() >= 0x40000000)
cec08e7a
BH
681 mmu_vmemmap_psize = MMU_PAGE_16M;
682 else if (mmu_psize_defs[MMU_PAGE_64K].shift)
683 mmu_vmemmap_psize = MMU_PAGE_64K;
684 else
685 mmu_vmemmap_psize = MMU_PAGE_4K;
686#endif /* CONFIG_SPARSEMEM_VMEMMAP */
687
bf72aeba 688 printk(KERN_DEBUG "Page orders: linear mapping = %d, "
cec08e7a
BH
689 "virtual = %d, io = %d"
690#ifdef CONFIG_SPARSEMEM_VMEMMAP
691 ", vmemmap = %d"
692#endif
693 "\n",
3c726f8d 694 mmu_psize_defs[mmu_linear_psize].shift,
bf72aeba 695 mmu_psize_defs[mmu_virtual_psize].shift,
cec08e7a
BH
696 mmu_psize_defs[mmu_io_psize].shift
697#ifdef CONFIG_SPARSEMEM_VMEMMAP
698 ,mmu_psize_defs[mmu_vmemmap_psize].shift
699#endif
700 );
3c726f8d
BH
701}
702
703static int __init htab_dt_scan_pftsize(unsigned long node,
704 const char *uname, int depth,
705 void *data)
706{
9d0c4dfe
RH
707 const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
708 const __be32 *prop;
3c726f8d
BH
709
710 /* We are scanning "cpu" nodes only */
711 if (type == NULL || strcmp(type, "cpu") != 0)
712 return 0;
713
12f04f2b 714 prop = of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
3c726f8d
BH
715 if (prop != NULL) {
716 /* pft_size[0] is the NUMA CEC cookie */
12f04f2b 717 ppc64_pft_size = be32_to_cpu(prop[1]);
3c726f8d 718 return 1;
1da177e4 719 }
3c726f8d 720 return 0;
1da177e4
LT
721}
722
5c3c7ede 723unsigned htab_shift_for_mem_size(unsigned long mem_size)
3eac8c69 724{
5c3c7ede
DG
725 unsigned memshift = __ilog2(mem_size);
726 unsigned pshift = mmu_psize_defs[mmu_virtual_psize].shift;
727 unsigned pteg_shift;
728
729 /* round mem_size up to next power of 2 */
730 if ((1UL << memshift) < mem_size)
731 memshift += 1;
3eac8c69 732
5c3c7ede
DG
733 /* aim for 2 pages / pteg */
734 pteg_shift = memshift - (pshift + 1);
3eac8c69 735
5c3c7ede
DG
736 /*
737 * 2^11 PTEGS of 128 bytes each, ie. 2^18 bytes is the minimum htab
738 * size permitted by the architecture.
739 */
740 return max(pteg_shift + 7, 18U);
741}
742
743static unsigned long __init htab_get_table_size(void)
744{
3c726f8d 745 /* If hash size isn't already provided by the platform, we try to
943ffb58 746 * retrieve it from the device-tree. If it's not there neither, we
3c726f8d 747 * calculate it now based on the total RAM size
3eac8c69 748 */
3c726f8d
BH
749 if (ppc64_pft_size == 0)
750 of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
3eac8c69
PM
751 if (ppc64_pft_size)
752 return 1UL << ppc64_pft_size;
753
5c3c7ede 754 return 1UL << htab_shift_for_mem_size(memblock_phys_mem_size());
3eac8c69
PM
755}
756
54b79248 757#ifdef CONFIG_MEMORY_HOTPLUG
438cc81a
DG
758void resize_hpt_for_hotplug(unsigned long new_mem_size)
759{
760 unsigned target_hpt_shift;
761
762 if (!mmu_hash_ops.resize_hpt)
763 return;
764
765 target_hpt_shift = htab_shift_for_mem_size(new_mem_size);
766
767 /*
768 * To avoid lots of HPT resizes if memory size is fluctuating
769 * across a boundary, we deliberately have some hysterisis
770 * here: we immediately increase the HPT size if the target
771 * shift exceeds the current shift, but we won't attempt to
772 * reduce unless the target shift is at least 2 below the
773 * current shift
774 */
775 if ((target_hpt_shift > ppc64_pft_size)
776 || (target_hpt_shift < (ppc64_pft_size - 1))) {
777 int rc;
778
779 rc = mmu_hash_ops.resize_hpt(target_hpt_shift);
7339390d 780 if (rc && (rc != -ENODEV))
438cc81a
DG
781 printk(KERN_WARNING
782 "Unable to resize hash page table to target order %d: %d\n",
783 target_hpt_shift, rc);
784 }
785}
786
29ab6c47 787int hash__create_section_mapping(unsigned long start, unsigned long end, int nid)
54b79248 788{
1dace6c6
DG
789 int rc = htab_bolt_mapping(start, end, __pa(start),
790 pgprot_val(PAGE_KERNEL), mmu_linear_psize,
791 mmu_kernel_ssize);
792
793 if (rc < 0) {
794 int rc2 = htab_remove_mapping(start, end, mmu_linear_psize,
795 mmu_kernel_ssize);
796 BUG_ON(rc2 && (rc2 != -ENOENT));
797 }
798 return rc;
54b79248 799}
f8c8803b 800
32b53c01 801int hash__remove_section_mapping(unsigned long start, unsigned long end)
f8c8803b 802{
abd0a0e7
DG
803 int rc = htab_remove_mapping(start, end, mmu_linear_psize,
804 mmu_kernel_ssize);
805 WARN_ON(rc < 0);
806 return rc;
f8c8803b 807}
54b79248
MK
808#endif /* CONFIG_MEMORY_HOTPLUG */
809
50de596d 810static void __init hash_init_partition_table(phys_addr_t hash_table,
4b7a3504 811 unsigned long htab_size)
50de596d 812{
9d661958 813 mmu_partition_table_init();
50de596d
AK
814
815 /*
9d661958
PM
816 * PS field (VRMA page size) is not used for LPID 0, hence set to 0.
817 * For now, UPRT is 0 and we have no segment table.
50de596d 818 */
4b7a3504 819 htab_size = __ilog2(htab_size) - 18;
9d661958 820 mmu_partition_table_set_entry(0, hash_table | htab_size, 0);
56547411 821 pr_info("Partition table %p\n", partition_tb);
50de596d
AK
822}
823
757c74d2 824static void __init htab_initialize(void)
1da177e4 825{
337a7128 826 unsigned long table;
1da177e4 827 unsigned long pteg_count;
9e88ba4e 828 unsigned long prot;
5556ecf5 829 unsigned long base = 0, size = 0;
28be7072 830 struct memblock_region *reg;
3c726f8d 831
1da177e4
LT
832 DBG(" -> htab_initialize()\n");
833
44ae3ab3 834 if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) {
1189be65
PM
835 mmu_kernel_ssize = MMU_SEGSIZE_1T;
836 mmu_highuser_ssize = MMU_SEGSIZE_1T;
837 printk(KERN_INFO "Using 1TB segments\n");
838 }
839
1da177e4
LT
840 /*
841 * Calculate the required size of the htab. We want the number of
842 * PTEGs to equal one half the number of real pages.
843 */
3c726f8d 844 htab_size_bytes = htab_get_table_size();
1da177e4
LT
845 pteg_count = htab_size_bytes >> 7;
846
1da177e4
LT
847 htab_hash_mask = pteg_count - 1;
848
5556ecf5
BH
849 if (firmware_has_feature(FW_FEATURE_LPAR) ||
850 firmware_has_feature(FW_FEATURE_PS3_LV1)) {
1da177e4
LT
851 /* Using a hypervisor which owns the htab */
852 htab_address = NULL;
853 _SDR1 = 0;
dbfcf3cb
PM
854 /*
855 * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
856 * to inform the hypervisor that we wish to use the HPT.
857 */
858 if (cpu_has_feature(CPU_FTR_ARCH_300))
859 register_process_table(0, 0, 0);
3ccc00a7
MS
860#ifdef CONFIG_FA_DUMP
861 /*
862 * If firmware assisted dump is active firmware preserves
863 * the contents of htab along with entire partition memory.
864 * Clear the htab if firmware assisted dump is active so
865 * that we dont end up using old mappings.
866 */
7025776e
BH
867 if (is_fadump_active() && mmu_hash_ops.hpte_clear_all)
868 mmu_hash_ops.hpte_clear_all();
3ccc00a7 869#endif
1da177e4 870 } else {
5556ecf5
BH
871 unsigned long limit = MEMBLOCK_ALLOC_ANYWHERE;
872
873#ifdef CONFIG_PPC_CELL
874 /*
875 * Cell may require the hash table down low when using the
876 * Axon IOMMU in order to fit the dynamic region over it, see
877 * comments in cell/iommu.c
1da177e4 878 */
5556ecf5 879 if (fdt_subnode_offset(initial_boot_params, 0, "axon") > 0) {
31bf1119 880 limit = 0x80000000;
5556ecf5
BH
881 pr_info("Hash table forced below 2G for Axon IOMMU\n");
882 }
883#endif /* CONFIG_PPC_CELL */
41d824bf 884
0ba9e6ed
MR
885 table = memblock_phys_alloc_range(htab_size_bytes,
886 htab_size_bytes,
887 0, limit);
888 if (!table)
889 panic("ERROR: Failed to allocate %pa bytes below %pa\n",
890 &htab_size_bytes, &limit);
1da177e4
LT
891
892 DBG("Hash table allocated at %lx, size: %lx\n", table,
893 htab_size_bytes);
894
70267a7f 895 htab_address = __va(table);
1da177e4
LT
896
897 /* htab absolute addr + encoded htabsize */
4b7a3504 898 _SDR1 = table + __ilog2(htab_size_bytes) - 18;
1da177e4
LT
899
900 /* Initialize the HPT with no entries */
901 memset((void *)table, 0, htab_size_bytes);
799d6046 902
50de596d
AK
903 if (!cpu_has_feature(CPU_FTR_ARCH_300))
904 /* Set SDR1 */
905 mtspr(SPRN_SDR1, _SDR1);
906 else
4b7a3504 907 hash_init_partition_table(table, htab_size_bytes);
1da177e4
LT
908 }
909
f5ea64dc 910 prot = pgprot_val(PAGE_KERNEL);
1da177e4 911
370a908d 912#ifdef CONFIG_DEBUG_PAGEALLOC
e7df0d88
JK
913 if (debug_pagealloc_enabled()) {
914 linear_map_hash_count = memblock_end_of_DRAM() >> PAGE_SHIFT;
f806714f
MR
915 linear_map_hash_slots = memblock_alloc_try_nid(
916 linear_map_hash_count, 1, MEMBLOCK_LOW_LIMIT,
917 ppc64_rma_size, NUMA_NO_NODE);
8a7f97b9
MR
918 if (!linear_map_hash_slots)
919 panic("%s: Failed to allocate %lu bytes max_addr=%pa\n",
920 __func__, linear_map_hash_count, &ppc64_rma_size);
e7df0d88 921 }
370a908d
BH
922#endif /* CONFIG_DEBUG_PAGEALLOC */
923
1da177e4 924 /* create bolted the linear mapping in the hash table */
28be7072
BH
925 for_each_memblock(memory, reg) {
926 base = (unsigned long)__va(reg->base);
927 size = reg->size;
1da177e4 928
5c339919 929 DBG("creating mapping for region: %lx..%lx (prot: %lx)\n",
9e88ba4e 930 base, size, prot);
1da177e4 931
caf80e57 932 BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
9e88ba4e 933 prot, mmu_linear_psize, mmu_kernel_ssize));
e63075a3
BH
934 }
935 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
1da177e4
LT
936
937 /*
938 * If we have a memory_limit and we've allocated TCEs then we need to
939 * explicitly map the TCE area at the top of RAM. We also cope with the
940 * case that the TCEs start below memory_limit.
941 * tce_alloc_start/end are 16MB aligned so the mapping should work
942 * for either 4K or 16MB pages.
943 */
944 if (tce_alloc_start) {
b5666f70
ME
945 tce_alloc_start = (unsigned long)__va(tce_alloc_start);
946 tce_alloc_end = (unsigned long)__va(tce_alloc_end);
1da177e4
LT
947
948 if (base + size >= tce_alloc_start)
949 tce_alloc_start = base + size + 1;
950
caf80e57 951 BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
bc033b63 952 __pa(tce_alloc_start), prot,
1189be65 953 mmu_linear_psize, mmu_kernel_ssize));
1da177e4
LT
954 }
955
7d0daae4 956
1da177e4
LT
957 DBG(" <- htab_initialize()\n");
958}
959#undef KB
960#undef MB
1da177e4 961
bacf9cf8
ME
962void __init hash__early_init_devtree(void)
963{
964 /* Initialize segment sizes */
965 of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL);
966
967 /* Initialize page sizes */
968 htab_scan_page_sizes();
969}
970
756d08d1 971void __init hash__early_init_mmu(void)
799d6046 972{
9d2edb18 973#ifndef CONFIG_PPC_64K_PAGES
6aa59f51 974 /*
9d2edb18 975 * We have code in __hash_page_4K() and elsewhere, which assumes it can
6aa59f51
AK
976 * do the following:
977 * new_pte |= (slot << H_PAGE_F_GIX_SHIFT) & (H_PAGE_F_SECOND | H_PAGE_F_GIX);
978 *
979 * Where the slot number is between 0-15, and values of 8-15 indicate
980 * the secondary bucket. For that code to work H_PAGE_F_SECOND and
981 * H_PAGE_F_GIX must occupy four contiguous bits in the PTE, and
982 * H_PAGE_F_SECOND must be placed above H_PAGE_F_GIX. Assert that here
983 * with a BUILD_BUG_ON().
984 */
985 BUILD_BUG_ON(H_PAGE_F_SECOND != (1ul << (H_PAGE_F_GIX_SHIFT + 3)));
9d2edb18 986#endif /* CONFIG_PPC_64K_PAGES */
6aa59f51 987
bacf9cf8
ME
988 htab_init_page_sizes();
989
dd1842a2
AK
990 /*
991 * initialize page table size
992 */
5ed7ecd0
AK
993 __pte_frag_nr = H_PTE_FRAG_NR;
994 __pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT;
8a6c697b
AK
995 __pmd_frag_nr = H_PMD_FRAG_NR;
996 __pmd_frag_size_shift = H_PMD_FRAG_SIZE_SHIFT;
5ed7ecd0 997
dd1842a2
AK
998 __pte_index_size = H_PTE_INDEX_SIZE;
999 __pmd_index_size = H_PMD_INDEX_SIZE;
1000 __pud_index_size = H_PUD_INDEX_SIZE;
1001 __pgd_index_size = H_PGD_INDEX_SIZE;
fae22116 1002 __pud_cache_index = H_PUD_CACHE_INDEX;
dd1842a2
AK
1003 __pte_table_size = H_PTE_TABLE_SIZE;
1004 __pmd_table_size = H_PMD_TABLE_SIZE;
1005 __pud_table_size = H_PUD_TABLE_SIZE;
1006 __pgd_table_size = H_PGD_TABLE_SIZE;
a2f41eb9
AK
1007 /*
1008 * 4k use hugepd format, so for hash set then to
1009 * zero
1010 */
da7ad366
AK
1011 __pmd_val_bits = HASH_PMD_VAL_BITS;
1012 __pud_val_bits = HASH_PUD_VAL_BITS;
1013 __pgd_val_bits = HASH_PGD_VAL_BITS;
d6a9996e
AK
1014
1015 __kernel_virt_start = H_KERN_VIRT_START;
1016 __kernel_virt_size = H_KERN_VIRT_SIZE;
1017 __vmalloc_start = H_VMALLOC_START;
1018 __vmalloc_end = H_VMALLOC_END;
63ee9b2f 1019 __kernel_io_start = H_KERN_IO_START;
d6a9996e
AK
1020 vmemmap = (struct page *)H_VMEMMAP_BASE;
1021 ioremap_bot = IOREMAP_BASE;
1022
bfa37087
DS
1023#ifdef CONFIG_PCI
1024 pci_io_base = ISA_IO_BASE;
1025#endif
1026
166dd7d3
BH
1027 /* Select appropriate backend */
1028 if (firmware_has_feature(FW_FEATURE_PS3_LV1))
1029 ps3_early_mm_init();
1030 else if (firmware_has_feature(FW_FEATURE_LPAR))
6364e84e 1031 hpte_init_pseries();
fbef66f0 1032 else if (IS_ENABLED(CONFIG_PPC_NATIVE))
166dd7d3
BH
1033 hpte_init_native();
1034
7353644f
ME
1035 if (!mmu_hash_ops.hpte_insert)
1036 panic("hash__early_init_mmu: No MMU hash ops defined!\n");
1037
757c74d2 1038 /* Initialize the MMU Hash table and create the linear mapping
376af594
ME
1039 * of memory. Has to be done before SLB initialization as this is
1040 * currently where the page size encoding is obtained.
757c74d2
BH
1041 */
1042 htab_initialize();
1043
56547411 1044 pr_info("Initializing hash mmu with SLB\n");
376af594 1045 /* Initialize SLB management */
13b3d13b 1046 slb_initialize();
d4748276
NP
1047
1048 if (cpu_has_feature(CPU_FTR_ARCH_206)
1049 && cpu_has_feature(CPU_FTR_HVMODE))
1050 tlbiel_all();
757c74d2
BH
1051}
1052
1053#ifdef CONFIG_SMP
756d08d1 1054void hash__early_init_mmu_secondary(void)
757c74d2
BH
1055{
1056 /* Initialize hash table for that CPU */
b5dcc609 1057 if (!firmware_has_feature(FW_FEATURE_LPAR)) {
cac4a185 1058
b5dcc609
AK
1059 if (!cpu_has_feature(CPU_FTR_ARCH_300))
1060 mtspr(SPRN_SDR1, _SDR1);
1061 else
1062 mtspr(SPRN_PTCR,
1063 __pa(partition_tb) | (PATB_SIZE_SHIFT - 12));
1064 }
376af594 1065 /* Initialize SLB */
13b3d13b 1066 slb_initialize();
d4748276
NP
1067
1068 if (cpu_has_feature(CPU_FTR_ARCH_206)
1069 && cpu_has_feature(CPU_FTR_HVMODE))
1070 tlbiel_all();
799d6046 1071}
757c74d2 1072#endif /* CONFIG_SMP */
799d6046 1073
1da177e4
LT
1074/*
1075 * Called by asm hashtable.S for doing lazy icache flush
1076 */
1077unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
1078{
1079 struct page *page;
1080
76c8e25b
BH
1081 if (!pfn_valid(pte_pfn(pte)))
1082 return pp;
1083
1da177e4
LT
1084 page = pte_page(pte);
1085
1086 /* page is dirty */
1087 if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
1088 if (trap == 0x400) {
0895ecda 1089 flush_dcache_icache_page(page);
1da177e4
LT
1090 set_bit(PG_arch_1, &page->flags);
1091 } else
3c726f8d 1092 pp |= HPTE_R_N;
1da177e4
LT
1093 }
1094 return pp;
1095}
1096
3a8247cc 1097#ifdef CONFIG_PPC_MM_SLICES
54be0b9c 1098static unsigned int get_paca_psize(unsigned long addr)
3a8247cc 1099{
15472423 1100 unsigned char *psizes;
7aa0727f 1101 unsigned long index, mask_index;
3a8247cc
PM
1102
1103 if (addr < SLICE_LOW_TOP) {
54be0b9c 1104 psizes = get_paca()->mm_ctx_low_slices_psize;
3a8247cc 1105 index = GET_LOW_SLICE_INDEX(addr);
15472423 1106 } else {
54be0b9c 1107 psizes = get_paca()->mm_ctx_high_slices_psize;
15472423 1108 index = GET_HIGH_SLICE_INDEX(addr);
3a8247cc 1109 }
7aa0727f 1110 mask_index = index & 0x1;
15472423 1111 return (psizes[index >> 1] >> (mask_index * 4)) & 0xF;
3a8247cc
PM
1112}
1113
1114#else
54be0b9c 1115unsigned int get_paca_psize(unsigned long addr)
3a8247cc 1116{
54be0b9c 1117 return get_paca()->mm_ctx_user_psize;
3a8247cc
PM
1118}
1119#endif
1120
721151d0
PM
1121/*
1122 * Demote a segment to using 4k pages.
1123 * For now this makes the whole process use 4k pages.
1124 */
721151d0 1125#ifdef CONFIG_PPC_64K_PAGES
fa28237c 1126void demote_segment_4k(struct mm_struct *mm, unsigned long addr)
16f1c746 1127{
54be0b9c 1128 if (get_slice_psize(mm, addr) == MMU_PAGE_4K)
721151d0 1129 return;
3a8247cc 1130 slice_set_range_psize(mm, addr, 1, MMU_PAGE_4K);
be3ebfe8 1131 copro_flush_all_slbs(mm);
54be0b9c
ME
1132 if ((get_paca_psize(addr) != MMU_PAGE_4K) && (current->mm == mm)) {
1133
1134 copy_mm_to_paca(mm);
94ee4272 1135 slb_flush_and_restore_bolted();
54be0b9c 1136 }
721151d0 1137}
16f1c746 1138#endif /* CONFIG_PPC_64K_PAGES */
721151d0 1139
fa28237c
PM
1140#ifdef CONFIG_PPC_SUBPAGE_PROT
1141/*
1142 * This looks up a 2-bit protection code for a 4k subpage of a 64k page.
1143 * Userspace sets the subpage permissions using the subpage_prot system call.
1144 *
1145 * Result is 0: full permissions, _PAGE_RW: read-only,
73a1441a 1146 * _PAGE_RWX: no access.
fa28237c 1147 */
d28513bc 1148static int subpage_protection(struct mm_struct *mm, unsigned long ea)
fa28237c 1149{
d28513bc 1150 struct subpage_prot_table *spt = &mm->context.spt;
fa28237c
PM
1151 u32 spp = 0;
1152 u32 **sbpm, *sbpp;
1153
1154 if (ea >= spt->maxaddr)
1155 return 0;
b0d436c7 1156 if (ea < 0x100000000UL) {
fa28237c
PM
1157 /* addresses below 4GB use spt->low_prot */
1158 sbpm = spt->low_prot;
1159 } else {
1160 sbpm = spt->protptrs[ea >> SBP_L3_SHIFT];
1161 if (!sbpm)
1162 return 0;
1163 }
1164 sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
1165 if (!sbpp)
1166 return 0;
1167 spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)];
1168
1169 /* extract 2-bit bitfield for this 4k subpage */
1170 spp >>= 30 - 2 * ((ea >> 12) & 0xf);
1171
73a1441a
AK
1172 /*
1173 * 0 -> full premission
1174 * 1 -> Read only
1175 * 2 -> no access.
1176 * We return the flag that need to be cleared.
1177 */
1178 spp = ((spp & 2) ? _PAGE_RWX : 0) | ((spp & 1) ? _PAGE_WRITE : 0);
fa28237c
PM
1179 return spp;
1180}
1181
1182#else /* CONFIG_PPC_SUBPAGE_PROT */
d28513bc 1183static inline int subpage_protection(struct mm_struct *mm, unsigned long ea)
fa28237c
PM
1184{
1185 return 0;
1186}
1187#endif
1188
4b8692c0
BH
1189void hash_failure_debug(unsigned long ea, unsigned long access,
1190 unsigned long vsid, unsigned long trap,
d8139ebf 1191 int ssize, int psize, int lpsize, unsigned long pte)
4b8692c0
BH
1192{
1193 if (!printk_ratelimit())
1194 return;
1195 pr_info("mm: Hashing failure ! EA=0x%lx access=0x%lx current=%s\n",
1196 ea, access, current->comm);
d8139ebf
AK
1197 pr_info(" trap=0x%lx vsid=0x%lx ssize=%d base psize=%d psize %d pte=0x%lx\n",
1198 trap, vsid, ssize, psize, lpsize, pte);
4b8692c0
BH
1199}
1200
54be0b9c
ME
1201static void check_paca_psize(unsigned long ea, struct mm_struct *mm,
1202 int psize, bool user_region)
1203{
1204 if (user_region) {
1205 if (psize != get_paca_psize(ea)) {
1206 copy_mm_to_paca(mm);
94ee4272 1207 slb_flush_and_restore_bolted();
54be0b9c
ME
1208 }
1209 } else if (get_paca()->vmalloc_sllp !=
1210 mmu_psize_defs[mmu_vmalloc_psize].sllp) {
1211 get_paca()->vmalloc_sllp =
1212 mmu_psize_defs[mmu_vmalloc_psize].sllp;
1213 slb_vmalloc_update();
1214 }
1215}
1216
1da177e4
LT
1217/* Result code is:
1218 * 0 - handled
1219 * 1 - normal page fault
1220 * -1 - critical hash insertion error
fa28237c 1221 * -2 - access not permitted by subpage protection mechanism
1da177e4 1222 */
aefa5688
AK
1223int hash_page_mm(struct mm_struct *mm, unsigned long ea,
1224 unsigned long access, unsigned long trap,
1225 unsigned long flags)
1da177e4 1226{
891121e6 1227 bool is_thp;
ba12eede 1228 enum ctx_state prev_state = exception_enter();
a1128f8f 1229 pgd_t *pgdir;
1da177e4 1230 unsigned long vsid;
1da177e4 1231 pte_t *ptep;
a4fe3ce7 1232 unsigned hugeshift;
aefa5688 1233 int rc, user_region = 0;
1189be65 1234 int psize, ssize;
1da177e4 1235
3c726f8d
BH
1236 DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
1237 ea, access, trap);
cfcb3d80 1238 trace_hash_fault(ea, access, trap);
1f8d419e 1239
3c726f8d 1240 /* Get region & vsid */
1da177e4
LT
1241 switch (REGION_ID(ea)) {
1242 case USER_REGION_ID:
1243 user_region = 1;
3c726f8d
BH
1244 if (! mm) {
1245 DBG_LOW(" user region with no mm !\n");
ba12eede
LZ
1246 rc = 1;
1247 goto bail;
3c726f8d 1248 }
54be0b9c 1249 psize = get_slice_psize(mm, ea);
1189be65 1250 ssize = user_segment_size(ea);
f384796c 1251 vsid = get_user_vsid(&mm->context, ea, ssize);
1da177e4 1252 break;
1da177e4 1253 case VMALLOC_REGION_ID:
1189be65 1254 vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
bf72aeba
PM
1255 if (ea < VMALLOC_END)
1256 psize = mmu_vmalloc_psize;
1257 else
1258 psize = mmu_io_psize;
1189be65 1259 ssize = mmu_kernel_ssize;
1da177e4 1260 break;
1da177e4
LT
1261 default:
1262 /* Not a valid range
1263 * Send the problem up to do_page_fault
1264 */
ba12eede
LZ
1265 rc = 1;
1266 goto bail;
1da177e4 1267 }
3c726f8d 1268 DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
1da177e4 1269
c60ac569
AK
1270 /* Bad address. */
1271 if (!vsid) {
1272 DBG_LOW("Bad address!\n");
ba12eede
LZ
1273 rc = 1;
1274 goto bail;
c60ac569 1275 }
3c726f8d 1276 /* Get pgdir */
1da177e4 1277 pgdir = mm->pgd;
ba12eede
LZ
1278 if (pgdir == NULL) {
1279 rc = 1;
1280 goto bail;
1281 }
1da177e4 1282
3c726f8d 1283 /* Check CPU locality */
b426e4bd 1284 if (user_region && mm_is_thread_local(mm))
aefa5688 1285 flags |= HPTE_LOCAL_UPDATE;
1da177e4 1286
16c2d476 1287#ifndef CONFIG_PPC_64K_PAGES
a4fe3ce7
DG
1288 /* If we use 4K pages and our psize is not 4K, then we might
1289 * be hitting a special driver mapping, and need to align the
1290 * address before we fetch the PTE.
1291 *
1292 * It could also be a hugepage mapping, in which case this is
1293 * not necessary, but it's not harmful, either.
16c2d476
BH
1294 */
1295 if (psize != MMU_PAGE_4K)
1296 ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
1297#endif /* CONFIG_PPC_64K_PAGES */
1298
3c726f8d 1299 /* Get PTE and page size from page tables */
94171b19 1300 ptep = find_linux_pte(pgdir, ea, &is_thp, &hugeshift);
3c726f8d
BH
1301 if (ptep == NULL || !pte_present(*ptep)) {
1302 DBG_LOW(" no PTE !\n");
ba12eede
LZ
1303 rc = 1;
1304 goto bail;
3c726f8d
BH
1305 }
1306
ca91e6c0
BH
1307 /* Add _PAGE_PRESENT to the required access perm */
1308 access |= _PAGE_PRESENT;
1309
1310 /* Pre-check access permissions (will be re-checked atomically
1311 * in __hash_page_XX but this pre-check is a fast path
1312 */
ac29c640 1313 if (!check_pte_access(access, pte_val(*ptep))) {
ca91e6c0 1314 DBG_LOW(" no access !\n");
ba12eede
LZ
1315 rc = 1;
1316 goto bail;
ca91e6c0
BH
1317 }
1318
ba12eede 1319 if (hugeshift) {
891121e6 1320 if (is_thp)
6d492ecc 1321 rc = __hash_page_thp(ea, access, vsid, (pmd_t *)ptep,
aefa5688 1322 trap, flags, ssize, psize);
6d492ecc
AK
1323#ifdef CONFIG_HUGETLB_PAGE
1324 else
1325 rc = __hash_page_huge(ea, access, vsid, ptep, trap,
aefa5688 1326 flags, ssize, hugeshift, psize);
6d492ecc
AK
1327#else
1328 else {
1329 /*
1330 * if we have hugeshift, and is not transhuge with
1331 * hugetlb disabled, something is really wrong.
1332 */
1333 rc = 1;
1334 WARN_ON(1);
1335 }
1336#endif
54be0b9c
ME
1337 if (current->mm == mm)
1338 check_paca_psize(ea, mm, psize, user_region);
1339
ba12eede
LZ
1340 goto bail;
1341 }
a4fe3ce7 1342
3c726f8d
BH
1343#ifndef CONFIG_PPC_64K_PAGES
1344 DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
1345#else
1346 DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
1347 pte_val(*(ptep + PTRS_PER_PTE)));
1348#endif
3c726f8d 1349 /* Do actual hashing */
16c2d476 1350#ifdef CONFIG_PPC_64K_PAGES
945537df
AK
1351 /* If H_PAGE_4K_PFN is set, make sure this is a 4k segment */
1352 if ((pte_val(*ptep) & H_PAGE_4K_PFN) && psize == MMU_PAGE_64K) {
721151d0
PM
1353 demote_segment_4k(mm, ea);
1354 psize = MMU_PAGE_4K;
1355 }
1356
16f1c746
BH
1357 /* If this PTE is non-cacheable and we have restrictions on
1358 * using non cacheable large pages, then we switch to 4k
1359 */
30bda41a 1360 if (mmu_ci_restrictions && psize == MMU_PAGE_64K && pte_ci(*ptep)) {
16f1c746
BH
1361 if (user_region) {
1362 demote_segment_4k(mm, ea);
1363 psize = MMU_PAGE_4K;
1364 } else if (ea < VMALLOC_END) {
1365 /*
1366 * some driver did a non-cacheable mapping
1367 * in vmalloc space, so switch vmalloc
1368 * to 4k pages
1369 */
1370 printk(KERN_ALERT "Reducing vmalloc segment "
1371 "to 4kB pages because of "
1372 "non-cacheable mapping\n");
1373 psize = mmu_vmalloc_psize = MMU_PAGE_4K;
be3ebfe8 1374 copro_flush_all_slbs(mm);
bf72aeba 1375 }
16f1c746 1376 }
09567e7f 1377
0863d7f2
AK
1378#endif /* CONFIG_PPC_64K_PAGES */
1379
54be0b9c
ME
1380 if (current->mm == mm)
1381 check_paca_psize(ea, mm, psize, user_region);
1382
73b341ef 1383#ifdef CONFIG_PPC_64K_PAGES
bf72aeba 1384 if (psize == MMU_PAGE_64K)
aefa5688
AK
1385 rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1386 flags, ssize);
3c726f8d 1387 else
73b341ef 1388#endif /* CONFIG_PPC_64K_PAGES */
fa28237c 1389 {
a1128f8f 1390 int spp = subpage_protection(mm, ea);
fa28237c
PM
1391 if (access & spp)
1392 rc = -2;
1393 else
1394 rc = __hash_page_4K(ea, access, vsid, ptep, trap,
aefa5688 1395 flags, ssize, spp);
fa28237c 1396 }
3c726f8d 1397
4b8692c0
BH
1398 /* Dump some info in case of hash insertion failure, they should
1399 * never happen so it is really useful to know if/when they do
1400 */
1401 if (rc == -1)
1402 hash_failure_debug(ea, access, vsid, trap, ssize, psize,
d8139ebf 1403 psize, pte_val(*ptep));
3c726f8d
BH
1404#ifndef CONFIG_PPC_64K_PAGES
1405 DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
1406#else
1407 DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
1408 pte_val(*(ptep + PTRS_PER_PTE)));
1409#endif
1410 DBG_LOW(" -> rc=%d\n", rc);
ba12eede
LZ
1411
1412bail:
1413 exception_exit(prev_state);
3c726f8d 1414 return rc;
1da177e4 1415}
a1dca346
IM
1416EXPORT_SYMBOL_GPL(hash_page_mm);
1417
aefa5688
AK
1418int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
1419 unsigned long dsisr)
a1dca346 1420{
aefa5688 1421 unsigned long flags = 0;
a1dca346
IM
1422 struct mm_struct *mm = current->mm;
1423
1424 if (REGION_ID(ea) == VMALLOC_REGION_ID)
1425 mm = &init_mm;
1426
aefa5688
AK
1427 if (dsisr & DSISR_NOHPTE)
1428 flags |= HPTE_NOHPTE_UPDATE;
1429
1430 return hash_page_mm(mm, ea, access, trap, flags);
a1dca346 1431}
67207b96 1432EXPORT_SYMBOL_GPL(hash_page);
1da177e4 1433
106713a1
AK
1434int __hash_page(unsigned long ea, unsigned long msr, unsigned long trap,
1435 unsigned long dsisr)
1436{
c7d54842 1437 unsigned long access = _PAGE_PRESENT | _PAGE_READ;
106713a1
AK
1438 unsigned long flags = 0;
1439 struct mm_struct *mm = current->mm;
1440
1441 if (REGION_ID(ea) == VMALLOC_REGION_ID)
1442 mm = &init_mm;
1443
1444 if (dsisr & DSISR_NOHPTE)
1445 flags |= HPTE_NOHPTE_UPDATE;
1446
1447 if (dsisr & DSISR_ISSTORE)
c7d54842 1448 access |= _PAGE_WRITE;
106713a1 1449 /*
ac29c640
AK
1450 * We set _PAGE_PRIVILEGED only when
1451 * kernel mode access kernel space.
1452 *
1453 * _PAGE_PRIVILEGED is NOT set
1454 * 1) when kernel mode access user space
1455 * 2) user space access kernel space.
106713a1 1456 */
ac29c640 1457 access |= _PAGE_PRIVILEGED;
106713a1 1458 if ((msr & MSR_PR) || (REGION_ID(ea) == USER_REGION_ID))
ac29c640 1459 access &= ~_PAGE_PRIVILEGED;
106713a1
AK
1460
1461 if (trap == 0x400)
1462 access |= _PAGE_EXEC;
1463
1464 return hash_page_mm(mm, ea, access, trap, flags);
1465}
1466
8bbc9b7b
ME
1467#ifdef CONFIG_PPC_MM_SLICES
1468static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1469{
54be0b9c 1470 int psize = get_slice_psize(mm, ea);
aac55d75 1471
8bbc9b7b 1472 /* We only prefault standard pages for now */
aac55d75
ME
1473 if (unlikely(psize != mm->context.user_psize))
1474 return false;
1475
1476 /*
1477 * Don't prefault if subpage protection is enabled for the EA.
1478 */
1479 if (unlikely((psize == MMU_PAGE_4K) && subpage_protection(mm, ea)))
8bbc9b7b
ME
1480 return false;
1481
1482 return true;
1483}
1484#else
1485static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1486{
1487 return true;
1488}
1489#endif
1490
3c726f8d 1491void hash_preload(struct mm_struct *mm, unsigned long ea,
34eb138e 1492 bool is_exec, unsigned long trap)
1da177e4 1493{
12bc9f6f 1494 int hugepage_shift;
3c726f8d 1495 unsigned long vsid;
0b97fee0 1496 pgd_t *pgdir;
3c726f8d 1497 pte_t *ptep;
3c726f8d 1498 unsigned long flags;
aefa5688 1499 int rc, ssize, update_flags = 0;
34eb138e 1500 unsigned long access = _PAGE_PRESENT | _PAGE_READ | (is_exec ? _PAGE_EXEC : 0);
3c726f8d 1501
d0f13e3c
BH
1502 BUG_ON(REGION_ID(ea) != USER_REGION_ID);
1503
8bbc9b7b 1504 if (!should_hash_preload(mm, ea))
3c726f8d
BH
1505 return;
1506
1507 DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
1508 " trap=%lx\n", mm, mm->pgd, ea, access, trap);
1da177e4 1509
16f1c746 1510 /* Get Linux PTE if available */
3c726f8d
BH
1511 pgdir = mm->pgd;
1512 if (pgdir == NULL)
1513 return;
0ac52dd7
AK
1514
1515 /* Get VSID */
1516 ssize = user_segment_size(ea);
f384796c 1517 vsid = get_user_vsid(&mm->context, ea, ssize);
0ac52dd7
AK
1518 if (!vsid)
1519 return;
1520 /*
1521 * Hash doesn't like irqs. Walking linux page table with irq disabled
1522 * saves us from holding multiple locks.
1523 */
1524 local_irq_save(flags);
1525
12bc9f6f
AK
1526 /*
1527 * THP pages use update_mmu_cache_pmd. We don't do
1528 * hash preload there. Hence can ignore THP here
1529 */
94171b19 1530 ptep = find_current_mm_pte(pgdir, ea, NULL, &hugepage_shift);
3c726f8d 1531 if (!ptep)
0ac52dd7 1532 goto out_exit;
16f1c746 1533
12bc9f6f 1534 WARN_ON(hugepage_shift);
16f1c746 1535#ifdef CONFIG_PPC_64K_PAGES
945537df 1536 /* If either H_PAGE_4K_PFN or cache inhibited is set (and we are on
16f1c746
BH
1537 * a 64K kernel), then we don't preload, hash_page() will take
1538 * care of it once we actually try to access the page.
1539 * That way we don't have to duplicate all of the logic for segment
1540 * page size demotion here
1541 */
945537df 1542 if ((pte_val(*ptep) & H_PAGE_4K_PFN) || pte_ci(*ptep))
0ac52dd7 1543 goto out_exit;
16f1c746
BH
1544#endif /* CONFIG_PPC_64K_PAGES */
1545
16c2d476 1546 /* Is that local to this CPU ? */
b426e4bd 1547 if (mm_is_thread_local(mm))
aefa5688 1548 update_flags |= HPTE_LOCAL_UPDATE;
16c2d476
BH
1549
1550 /* Hash it in */
73b341ef 1551#ifdef CONFIG_PPC_64K_PAGES
bf72aeba 1552 if (mm->context.user_psize == MMU_PAGE_64K)
aefa5688
AK
1553 rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1554 update_flags, ssize);
1da177e4 1555 else
73b341ef 1556#endif /* CONFIG_PPC_64K_PAGES */
aefa5688
AK
1557 rc = __hash_page_4K(ea, access, vsid, ptep, trap, update_flags,
1558 ssize, subpage_protection(mm, ea));
4b8692c0
BH
1559
1560 /* Dump some info in case of hash insertion failure, they should
1561 * never happen so it is really useful to know if/when they do
1562 */
1563 if (rc == -1)
1564 hash_failure_debug(ea, access, vsid, trap, ssize,
d8139ebf
AK
1565 mm->context.user_psize,
1566 mm->context.user_psize,
1567 pte_val(*ptep));
0ac52dd7 1568out_exit:
3c726f8d
BH
1569 local_irq_restore(flags);
1570}
1571
087003e9
RP
1572#ifdef CONFIG_PPC_MEM_KEYS
1573/*
1574 * Return the protection key associated with the given address and the
1575 * mm_struct.
1576 */
1577u16 get_mm_addr_key(struct mm_struct *mm, unsigned long address)
1578{
1579 pte_t *ptep;
1580 u16 pkey = 0;
1581 unsigned long flags;
1582
1583 if (!mm || !mm->pgd)
1584 return 0;
1585
1586 local_irq_save(flags);
1587 ptep = find_linux_pte(mm->pgd, address, NULL, NULL);
1588 if (ptep)
1589 pkey = pte_to_pkey_bits(pte_val(READ_ONCE(*ptep)));
1590 local_irq_restore(flags);
1591
1592 return pkey;
1593}
1594#endif /* CONFIG_PPC_MEM_KEYS */
1595
f1a55ce0
RT
1596#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1597static inline void tm_flush_hash_page(int local)
1598{
1599 /*
1600 * Transactions are not aborted by tlbiel, only tlbie. Without, syncing a
1601 * page back to a block device w/PIO could pick up transactional data
1602 * (bad!) so we force an abort here. Before the sync the page will be
1603 * made read-only, which will flush_hash_page. BIG ISSUE here: if the
1604 * kernel uses a page from userspace without unmapping it first, it may
1605 * see the speculated version.
1606 */
1607 if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
1608 MSR_TM_ACTIVE(current->thread.regs->msr)) {
1609 tm_enable();
1610 tm_abort(TM_CAUSE_TLBI);
1611 }
1612}
1613#else
1614static inline void tm_flush_hash_page(int local)
1615{
1616}
1617#endif
1618
318995b4
RP
1619/*
1620 * Return the global hash slot, corresponding to the given PTE, which contains
1621 * the HPTE.
1622 */
1623unsigned long pte_get_hash_gslot(unsigned long vpn, unsigned long shift,
1624 int ssize, real_pte_t rpte, unsigned int subpg_index)
1625{
1626 unsigned long hash, gslot, hidx;
1627
1628 hash = hpt_hash(vpn, shift, ssize);
1629 hidx = __rpte_to_hidx(rpte, subpg_index);
1630 if (hidx & _PTEIDX_SECONDARY)
1631 hash = ~hash;
1632 gslot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1633 gslot += hidx & _PTEIDX_GROUP_IX;
1634 return gslot;
1635}
1636
f6ab0b92
BH
1637/* WARNING: This is called from hash_low_64.S, if you change this prototype,
1638 * do not forget to update the assembly call site !
1639 */
5524a27d 1640void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize,
aefa5688 1641 unsigned long flags)
3c726f8d 1642{
a8548686 1643 unsigned long index, shift, gslot;
aefa5688 1644 int local = flags & HPTE_LOCAL_UPDATE;
3c726f8d 1645
5524a27d
AK
1646 DBG_LOW("flush_hash_page(vpn=%016lx)\n", vpn);
1647 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
a8548686
RP
1648 gslot = pte_get_hash_gslot(vpn, shift, ssize, pte, index);
1649 DBG_LOW(" sub %ld: gslot=%lx\n", index, gslot);
db3d8534
AK
1650 /*
1651 * We use same base page size and actual psize, because we don't
1652 * use these functions for hugepage
1653 */
a8548686 1654 mmu_hash_ops.hpte_invalidate(gslot, vpn, psize, psize,
7025776e 1655 ssize, local);
3c726f8d 1656 } pte_iterate_hashed_end();
bc2a9408 1657
f1a55ce0 1658 tm_flush_hash_page(local);
1da177e4
LT
1659}
1660
f1581bf1
AK
1661#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1662void flush_hash_hugepage(unsigned long vsid, unsigned long addr,
aefa5688
AK
1663 pmd_t *pmdp, unsigned int psize, int ssize,
1664 unsigned long flags)
f1581bf1
AK
1665{
1666 int i, max_hpte_count, valid;
1667 unsigned long s_addr;
1668 unsigned char *hpte_slot_array;
1669 unsigned long hidx, shift, vpn, hash, slot;
aefa5688 1670 int local = flags & HPTE_LOCAL_UPDATE;
f1581bf1
AK
1671
1672 s_addr = addr & HPAGE_PMD_MASK;
1673 hpte_slot_array = get_hpte_slot_array(pmdp);
1674 /*
1675 * IF we try to do a HUGE PTE update after a withdraw is done.
1676 * we will find the below NULL. This happens when we do
1677 * split_huge_page_pmd
1678 */
1679 if (!hpte_slot_array)
1680 return;
1681
7025776e
BH
1682 if (mmu_hash_ops.hugepage_invalidate) {
1683 mmu_hash_ops.hugepage_invalidate(vsid, s_addr, hpte_slot_array,
1684 psize, ssize, local);
d557b098
AK
1685 goto tm_abort;
1686 }
f1581bf1
AK
1687 /*
1688 * No bluk hpte removal support, invalidate each entry
1689 */
1690 shift = mmu_psize_defs[psize].shift;
1691 max_hpte_count = HPAGE_PMD_SIZE >> shift;
1692 for (i = 0; i < max_hpte_count; i++) {
1693 /*
1694 * 8 bits per each hpte entries
1695 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit]
1696 */
1697 valid = hpte_valid(hpte_slot_array, i);
1698 if (!valid)
1699 continue;
1700 hidx = hpte_hash_index(hpte_slot_array, i);
1701
1702 /* get the vpn */
1703 addr = s_addr + (i * (1ul << shift));
1704 vpn = hpt_vpn(addr, vsid, ssize);
1705 hash = hpt_hash(vpn, shift, ssize);
1706 if (hidx & _PTEIDX_SECONDARY)
1707 hash = ~hash;
1708
1709 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1710 slot += hidx & _PTEIDX_GROUP_IX;
7025776e
BH
1711 mmu_hash_ops.hpte_invalidate(slot, vpn, psize,
1712 MMU_PAGE_16M, ssize, local);
d557b098
AK
1713 }
1714tm_abort:
f1a55ce0 1715 tm_flush_hash_page(local);
f1581bf1
AK
1716}
1717#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1718
61b1a942 1719void flush_hash_range(unsigned long number, int local)
1da177e4 1720{
7025776e
BH
1721 if (mmu_hash_ops.flush_hash_range)
1722 mmu_hash_ops.flush_hash_range(number, local);
3c726f8d 1723 else {
1da177e4 1724 int i;
61b1a942 1725 struct ppc64_tlb_batch *batch =
69111bac 1726 this_cpu_ptr(&ppc64_tlb_batch);
1da177e4
LT
1727
1728 for (i = 0; i < number; i++)
5524a27d 1729 flush_hash_page(batch->vpn[i], batch->pte[i],
1189be65 1730 batch->psize, batch->ssize, local);
1da177e4
LT
1731 }
1732}
1733
1da177e4
LT
1734/*
1735 * low_hash_fault is called when we the low level hash code failed
1736 * to instert a PTE due to an hypervisor error
1737 */
fa28237c 1738void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc)
1da177e4 1739{
ba12eede
LZ
1740 enum ctx_state prev_state = exception_enter();
1741
1da177e4 1742 if (user_mode(regs)) {
fa28237c
PM
1743#ifdef CONFIG_PPC_SUBPAGE_PROT
1744 if (rc == -2)
1745 _exception(SIGSEGV, regs, SEGV_ACCERR, address);
1746 else
1747#endif
1748 _exception(SIGBUS, regs, BUS_ADRERR, address);
1749 } else
1750 bad_page_fault(regs, address, SIGBUS);
ba12eede
LZ
1751
1752 exception_exit(prev_state);
1da177e4 1753}
370a908d 1754
b170bd3d
LZ
1755long hpte_insert_repeating(unsigned long hash, unsigned long vpn,
1756 unsigned long pa, unsigned long rflags,
1757 unsigned long vflags, int psize, int ssize)
1758{
1759 unsigned long hpte_group;
1760 long slot;
1761
1762repeat:
1531cff4 1763 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
b170bd3d
LZ
1764
1765 /* Insert into the hash table, primary slot */
7025776e
BH
1766 slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, vflags,
1767 psize, psize, ssize);
b170bd3d
LZ
1768
1769 /* Primary is full, try the secondary */
1770 if (unlikely(slot == -1)) {
1531cff4 1771 hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
7025776e
BH
1772 slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags,
1773 vflags | HPTE_V_SECONDARY,
1774 psize, psize, ssize);
b170bd3d
LZ
1775 if (slot == -1) {
1776 if (mftb() & 0x1)
1531cff4
AK
1777 hpte_group = (hash & htab_hash_mask) *
1778 HPTES_PER_GROUP;
b170bd3d 1779
7025776e 1780 mmu_hash_ops.hpte_remove(hpte_group);
b170bd3d
LZ
1781 goto repeat;
1782 }
1783 }
1784
1785 return slot;
1786}
1787
370a908d
BH
1788#ifdef CONFIG_DEBUG_PAGEALLOC
1789static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi)
1790{
016af59f 1791 unsigned long hash;
1189be65 1792 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
5524a27d 1793 unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
09f3f326 1794 unsigned long mode = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL));
016af59f 1795 long ret;
370a908d 1796
5524a27d 1797 hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
370a908d 1798
c60ac569
AK
1799 /* Don't create HPTE entries for bad address */
1800 if (!vsid)
1801 return;
016af59f
LZ
1802
1803 ret = hpte_insert_repeating(hash, vpn, __pa(vaddr), mode,
1804 HPTE_V_BOLTED,
1805 mmu_linear_psize, mmu_kernel_ssize);
1806
370a908d
BH
1807 BUG_ON (ret < 0);
1808 spin_lock(&linear_map_hash_lock);
1809 BUG_ON(linear_map_hash_slots[lmi] & 0x80);
1810 linear_map_hash_slots[lmi] = ret | 0x80;
1811 spin_unlock(&linear_map_hash_lock);
1812}
1813
1814static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
1815{
1189be65
PM
1816 unsigned long hash, hidx, slot;
1817 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
5524a27d 1818 unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
370a908d 1819
5524a27d 1820 hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
370a908d
BH
1821 spin_lock(&linear_map_hash_lock);
1822 BUG_ON(!(linear_map_hash_slots[lmi] & 0x80));
1823 hidx = linear_map_hash_slots[lmi] & 0x7f;
1824 linear_map_hash_slots[lmi] = 0;
1825 spin_unlock(&linear_map_hash_lock);
1826 if (hidx & _PTEIDX_SECONDARY)
1827 hash = ~hash;
1828 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1829 slot += hidx & _PTEIDX_GROUP_IX;
7025776e
BH
1830 mmu_hash_ops.hpte_invalidate(slot, vpn, mmu_linear_psize,
1831 mmu_linear_psize,
1832 mmu_kernel_ssize, 0);
370a908d
BH
1833}
1834
031bc574 1835void __kernel_map_pages(struct page *page, int numpages, int enable)
370a908d
BH
1836{
1837 unsigned long flags, vaddr, lmi;
1838 int i;
1839
1840 local_irq_save(flags);
1841 for (i = 0; i < numpages; i++, page++) {
1842 vaddr = (unsigned long)page_address(page);
1843 lmi = __pa(vaddr) >> PAGE_SHIFT;
1844 if (lmi >= linear_map_hash_count)
1845 continue;
1846 if (enable)
1847 kernel_map_linear_page(vaddr, lmi);
1848 else
1849 kernel_unmap_linear_page(vaddr, lmi);
1850 }
1851 local_irq_restore(flags);
1852}
1853#endif /* CONFIG_DEBUG_PAGEALLOC */
cd3db0c4 1854
756d08d1 1855void hash__setup_initial_memory_limit(phys_addr_t first_memblock_base,
cd3db0c4
BH
1856 phys_addr_t first_memblock_size)
1857{
1858 /* We don't currently support the first MEMBLOCK not mapping 0
1859 * physical on those processors
1860 */
1861 BUG_ON(first_memblock_base != 0);
1862
1513c33d
NP
1863 /*
1864 * On virtualized systems the first entry is our RMA region aka VRMA,
1865 * non-virtualized 64-bit hash MMU systems don't have a limitation
1866 * on real mode access.
1867 *
c610d65c
NP
1868 * For guests on platforms before POWER9, we clamp the it limit to 1G
1869 * to avoid some funky things such as RTAS bugs etc...
cd3db0c4 1870 */
1513c33d 1871 if (!early_cpu_has_feature(CPU_FTR_HVMODE)) {
c610d65c
NP
1872 ppc64_rma_size = first_memblock_size;
1873 if (!early_cpu_has_feature(CPU_FTR_ARCH_300))
1874 ppc64_rma_size = min_t(u64, ppc64_rma_size, 0x40000000);
cd3db0c4 1875
1513c33d
NP
1876 /* Finally limit subsequent allocations */
1877 memblock_set_current_limit(ppc64_rma_size);
1878 } else {
1879 ppc64_rma_size = ULONG_MAX;
1880 }
cd3db0c4 1881}
dbcf929c
DG
1882
1883#ifdef CONFIG_DEBUG_FS
1884
1885static int hpt_order_get(void *data, u64 *val)
1886{
1887 *val = ppc64_pft_size;
1888 return 0;
1889}
1890
1891static int hpt_order_set(void *data, u64 val)
1892{
1893 if (!mmu_hash_ops.resize_hpt)
1894 return -ENODEV;
1895
1896 return mmu_hash_ops.resize_hpt(val);
1897}
1898
7cd4774f 1899DEFINE_DEBUGFS_ATTRIBUTE(fops_hpt_order, hpt_order_get, hpt_order_set, "%llu\n");
dbcf929c
DG
1900
1901static int __init hash64_debugfs(void)
1902{
7cd4774f
Y
1903 if (!debugfs_create_file_unsafe("hpt_order", 0600, powerpc_debugfs_root,
1904 NULL, &fops_hpt_order)) {
dbcf929c
DG
1905 pr_err("lpar: unable to create hpt_order debugsfs file\n");
1906 }
1907
1908 return 0;
1909}
1910machine_device_initcall(pseries, hash64_debugfs);
dbcf929c 1911#endif /* CONFIG_DEBUG_FS */