Commit | Line | Data |
---|---|---|
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 | |
1da177e4 LT |
24 | #include <linux/spinlock.h> |
25 | #include <linux/errno.h> | |
26 | #include <linux/sched.h> | |
27 | #include <linux/proc_fs.h> | |
28 | #include <linux/stat.h> | |
29 | #include <linux/sysctl.h> | |
66b15db6 | 30 | #include <linux/export.h> |
1da177e4 LT |
31 | #include <linux/ctype.h> |
32 | #include <linux/cache.h> | |
33 | #include <linux/init.h> | |
34 | #include <linux/signal.h> | |
95f72d1e | 35 | #include <linux/memblock.h> |
ba12eede | 36 | #include <linux/context_tracking.h> |
1da177e4 | 37 | |
1da177e4 LT |
38 | #include <asm/processor.h> |
39 | #include <asm/pgtable.h> | |
40 | #include <asm/mmu.h> | |
41 | #include <asm/mmu_context.h> | |
42 | #include <asm/page.h> | |
43 | #include <asm/types.h> | |
1da177e4 LT |
44 | #include <asm/uaccess.h> |
45 | #include <asm/machdep.h> | |
d9b2b2a2 | 46 | #include <asm/prom.h> |
1da177e4 LT |
47 | #include <asm/tlbflush.h> |
48 | #include <asm/io.h> | |
49 | #include <asm/eeh.h> | |
50 | #include <asm/tlb.h> | |
51 | #include <asm/cacheflush.h> | |
52 | #include <asm/cputable.h> | |
1da177e4 | 53 | #include <asm/sections.h> |
be3ebfe8 | 54 | #include <asm/copro.h> |
aa39be09 | 55 | #include <asm/udbg.h> |
b68a70c4 | 56 | #include <asm/code-patching.h> |
3ccc00a7 | 57 | #include <asm/fadump.h> |
f5339277 | 58 | #include <asm/firmware.h> |
bc2a9408 | 59 | #include <asm/tm.h> |
cfcb3d80 | 60 | #include <asm/trace.h> |
1da177e4 LT |
61 | |
62 | #ifdef DEBUG | |
63 | #define DBG(fmt...) udbg_printf(fmt) | |
64 | #else | |
65 | #define DBG(fmt...) | |
66 | #endif | |
67 | ||
3c726f8d BH |
68 | #ifdef DEBUG_LOW |
69 | #define DBG_LOW(fmt...) udbg_printf(fmt) | |
70 | #else | |
71 | #define DBG_LOW(fmt...) | |
72 | #endif | |
73 | ||
74 | #define KB (1024) | |
75 | #define MB (1024*KB) | |
658013e9 | 76 | #define GB (1024L*MB) |
3c726f8d | 77 | |
1da177e4 LT |
78 | /* |
79 | * Note: pte --> Linux PTE | |
80 | * HPTE --> PowerPC Hashed Page Table Entry | |
81 | * | |
82 | * Execution context: | |
83 | * htab_initialize is called with the MMU off (of course), but | |
84 | * the kernel has been copied down to zero so it can directly | |
85 | * reference global data. At this point it is very difficult | |
86 | * to print debug info. | |
87 | * | |
88 | */ | |
89 | ||
90 | #ifdef CONFIG_U3_DART | |
91 | extern unsigned long dart_tablebase; | |
92 | #endif /* CONFIG_U3_DART */ | |
93 | ||
799d6046 PM |
94 | static unsigned long _SDR1; |
95 | struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT]; | |
e1802b06 | 96 | EXPORT_SYMBOL_GPL(mmu_psize_defs); |
799d6046 | 97 | |
8e561e7e | 98 | struct hash_pte *htab_address; |
337a7128 | 99 | unsigned long htab_size_bytes; |
96e28449 | 100 | unsigned long htab_hash_mask; |
4ab79aa8 | 101 | EXPORT_SYMBOL_GPL(htab_hash_mask); |
3c726f8d | 102 | int mmu_linear_psize = MMU_PAGE_4K; |
8ca7a82f | 103 | EXPORT_SYMBOL_GPL(mmu_linear_psize); |
3c726f8d | 104 | int mmu_virtual_psize = MMU_PAGE_4K; |
bf72aeba | 105 | int mmu_vmalloc_psize = MMU_PAGE_4K; |
cec08e7a BH |
106 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
107 | int mmu_vmemmap_psize = MMU_PAGE_4K; | |
108 | #endif | |
bf72aeba | 109 | int mmu_io_psize = MMU_PAGE_4K; |
1189be65 | 110 | int mmu_kernel_ssize = MMU_SEGSIZE_256M; |
8ca7a82f | 111 | EXPORT_SYMBOL_GPL(mmu_kernel_ssize); |
1189be65 | 112 | int mmu_highuser_ssize = MMU_SEGSIZE_256M; |
584f8b71 | 113 | u16 mmu_slb_size = 64; |
4ab79aa8 | 114 | EXPORT_SYMBOL_GPL(mmu_slb_size); |
bf72aeba PM |
115 | #ifdef CONFIG_PPC_64K_PAGES |
116 | int mmu_ci_restrictions; | |
117 | #endif | |
370a908d BH |
118 | #ifdef CONFIG_DEBUG_PAGEALLOC |
119 | static u8 *linear_map_hash_slots; | |
120 | static unsigned long linear_map_hash_count; | |
ed166692 | 121 | static DEFINE_SPINLOCK(linear_map_hash_lock); |
370a908d | 122 | #endif /* CONFIG_DEBUG_PAGEALLOC */ |
1da177e4 | 123 | |
3c726f8d BH |
124 | /* There are definitions of page sizes arrays to be used when none |
125 | * is provided by the firmware. | |
126 | */ | |
1da177e4 | 127 | |
3c726f8d BH |
128 | /* Pre-POWER4 CPUs (4k pages only) |
129 | */ | |
09de9ff8 | 130 | static struct mmu_psize_def mmu_psize_defaults_old[] = { |
3c726f8d BH |
131 | [MMU_PAGE_4K] = { |
132 | .shift = 12, | |
133 | .sllp = 0, | |
b1022fbd | 134 | .penc = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1}, |
3c726f8d BH |
135 | .avpnm = 0, |
136 | .tlbiel = 0, | |
137 | }, | |
138 | }; | |
139 | ||
140 | /* POWER4, GPUL, POWER5 | |
141 | * | |
142 | * Support for 16Mb large pages | |
143 | */ | |
09de9ff8 | 144 | static struct mmu_psize_def mmu_psize_defaults_gp[] = { |
3c726f8d BH |
145 | [MMU_PAGE_4K] = { |
146 | .shift = 12, | |
147 | .sllp = 0, | |
b1022fbd | 148 | .penc = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1}, |
3c726f8d BH |
149 | .avpnm = 0, |
150 | .tlbiel = 1, | |
151 | }, | |
152 | [MMU_PAGE_16M] = { | |
153 | .shift = 24, | |
154 | .sllp = SLB_VSID_L, | |
b1022fbd AK |
155 | .penc = {[0 ... MMU_PAGE_16M - 1] = -1, [MMU_PAGE_16M] = 0, |
156 | [MMU_PAGE_16M + 1 ... MMU_PAGE_COUNT - 1] = -1 }, | |
3c726f8d BH |
157 | .avpnm = 0x1UL, |
158 | .tlbiel = 0, | |
159 | }, | |
160 | }; | |
161 | ||
c6a3c495 | 162 | unsigned long htab_convert_pte_flags(unsigned long pteflags) |
bc033b63 | 163 | { |
c6a3c495 | 164 | unsigned long rflags = 0; |
bc033b63 BH |
165 | |
166 | /* _PAGE_EXEC -> NOEXEC */ | |
167 | if ((pteflags & _PAGE_EXEC) == 0) | |
168 | rflags |= HPTE_R_N; | |
c6a3c495 AK |
169 | /* |
170 | * PP bits: | |
171 | * Linux use slb key 0 for kernel and 1 for user. | |
172 | * kernel areas are mapped by PP bits 00 | |
173 | * and and there is no kernel RO (_PAGE_KERNEL_RO). | |
174 | * User area mapped by 0x2 and read only use by | |
175 | * 0x3. | |
bc033b63 | 176 | */ |
c6a3c495 AK |
177 | if (pteflags & _PAGE_USER) { |
178 | rflags |= 0x2; | |
179 | if (!((pteflags & _PAGE_RW) && (pteflags & _PAGE_DIRTY))) | |
180 | rflags |= 0x1; | |
181 | } | |
c8c06f5a AK |
182 | /* |
183 | * Always add "C" bit for perf. Memory coherence is always enabled | |
184 | */ | |
40e8550a AK |
185 | rflags |= HPTE_R_C | HPTE_R_M; |
186 | /* | |
187 | * Add in WIG bits | |
188 | */ | |
189 | if (pteflags & _PAGE_WRITETHRU) | |
190 | rflags |= HPTE_R_W; | |
191 | if (pteflags & _PAGE_NO_CACHE) | |
192 | rflags |= HPTE_R_I; | |
193 | if (pteflags & _PAGE_GUARDED) | |
194 | rflags |= HPTE_R_G; | |
195 | ||
196 | return rflags; | |
bc033b63 | 197 | } |
3c726f8d BH |
198 | |
199 | int htab_bolt_mapping(unsigned long vstart, unsigned long vend, | |
bc033b63 | 200 | unsigned long pstart, unsigned long prot, |
1189be65 | 201 | int psize, int ssize) |
1da177e4 | 202 | { |
3c726f8d BH |
203 | unsigned long vaddr, paddr; |
204 | unsigned int step, shift; | |
3c726f8d | 205 | int ret = 0; |
1da177e4 | 206 | |
3c726f8d BH |
207 | shift = mmu_psize_defs[psize].shift; |
208 | step = 1 << shift; | |
1da177e4 | 209 | |
bc033b63 BH |
210 | prot = htab_convert_pte_flags(prot); |
211 | ||
212 | DBG("htab_bolt_mapping(%lx..%lx -> %lx (%lx,%d,%d)\n", | |
213 | vstart, vend, pstart, prot, psize, ssize); | |
214 | ||
3c726f8d BH |
215 | for (vaddr = vstart, paddr = pstart; vaddr < vend; |
216 | vaddr += step, paddr += step) { | |
370a908d | 217 | unsigned long hash, hpteg; |
1189be65 | 218 | unsigned long vsid = get_kernel_vsid(vaddr, ssize); |
5524a27d | 219 | unsigned long vpn = hpt_vpn(vaddr, vsid, ssize); |
9e88ba4e PM |
220 | unsigned long tprot = prot; |
221 | ||
c60ac569 AK |
222 | /* |
223 | * If we hit a bad address return error. | |
224 | */ | |
225 | if (!vsid) | |
226 | return -1; | |
9e88ba4e | 227 | /* Make kernel text executable */ |
549e8152 | 228 | if (overlaps_kernel_text(vaddr, vaddr + step)) |
9e88ba4e | 229 | tprot &= ~HPTE_R_N; |
1da177e4 | 230 | |
b18db0b8 AG |
231 | /* Make kvm guest trampolines executable */ |
232 | if (overlaps_kvm_tmp(vaddr, vaddr + step)) | |
233 | tprot &= ~HPTE_R_N; | |
234 | ||
429d2e83 MS |
235 | /* |
236 | * If relocatable, check if it overlaps interrupt vectors that | |
237 | * are copied down to real 0. For relocatable kernel | |
238 | * (e.g. kdump case) we copy interrupt vectors down to real | |
239 | * address 0. Mark that region as executable. This is | |
240 | * because on p8 system with relocation on exception feature | |
241 | * enabled, exceptions are raised with MMU (IR=DR=1) ON. Hence | |
242 | * in order to execute the interrupt handlers in virtual | |
243 | * mode the vector region need to be marked as executable. | |
244 | */ | |
245 | if ((PHYSICAL_START > MEMORY_START) && | |
246 | overlaps_interrupt_vector_text(vaddr, vaddr + step)) | |
247 | tprot &= ~HPTE_R_N; | |
248 | ||
5524a27d | 249 | hash = hpt_hash(vpn, shift, ssize); |
1da177e4 LT |
250 | hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP); |
251 | ||
c30a4df3 | 252 | BUG_ON(!ppc_md.hpte_insert); |
5524a27d | 253 | ret = ppc_md.hpte_insert(hpteg, vpn, paddr, tprot, |
b1022fbd | 254 | HPTE_V_BOLTED, psize, psize, ssize); |
c30a4df3 | 255 | |
3c726f8d BH |
256 | if (ret < 0) |
257 | break; | |
370a908d BH |
258 | #ifdef CONFIG_DEBUG_PAGEALLOC |
259 | if ((paddr >> PAGE_SHIFT) < linear_map_hash_count) | |
260 | linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80; | |
261 | #endif /* CONFIG_DEBUG_PAGEALLOC */ | |
3c726f8d BH |
262 | } |
263 | return ret < 0 ? ret : 0; | |
264 | } | |
1da177e4 | 265 | |
ae86f008 | 266 | #ifdef CONFIG_MEMORY_HOTPLUG |
ed5694a8 | 267 | int htab_remove_mapping(unsigned long vstart, unsigned long vend, |
f8c8803b BP |
268 | int psize, int ssize) |
269 | { | |
270 | unsigned long vaddr; | |
271 | unsigned int step, shift; | |
272 | ||
273 | shift = mmu_psize_defs[psize].shift; | |
274 | step = 1 << shift; | |
275 | ||
276 | if (!ppc_md.hpte_removebolted) { | |
52db9b44 BP |
277 | printk(KERN_WARNING "Platform doesn't implement " |
278 | "hpte_removebolted\n"); | |
279 | return -EINVAL; | |
f8c8803b BP |
280 | } |
281 | ||
282 | for (vaddr = vstart; vaddr < vend; vaddr += step) | |
283 | ppc_md.hpte_removebolted(vaddr, psize, ssize); | |
52db9b44 BP |
284 | |
285 | return 0; | |
f8c8803b | 286 | } |
ae86f008 | 287 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
f8c8803b | 288 | |
1189be65 PM |
289 | static int __init htab_dt_scan_seg_sizes(unsigned long node, |
290 | const char *uname, int depth, | |
291 | void *data) | |
292 | { | |
9d0c4dfe RH |
293 | const char *type = of_get_flat_dt_prop(node, "device_type", NULL); |
294 | const __be32 *prop; | |
295 | int size = 0; | |
1189be65 PM |
296 | |
297 | /* We are scanning "cpu" nodes only */ | |
298 | if (type == NULL || strcmp(type, "cpu") != 0) | |
299 | return 0; | |
300 | ||
12f04f2b | 301 | prop = of_get_flat_dt_prop(node, "ibm,processor-segment-sizes", &size); |
1189be65 PM |
302 | if (prop == NULL) |
303 | return 0; | |
304 | for (; size >= 4; size -= 4, ++prop) { | |
12f04f2b | 305 | if (be32_to_cpu(prop[0]) == 40) { |
1189be65 | 306 | DBG("1T segment support detected\n"); |
44ae3ab3 | 307 | cur_cpu_spec->mmu_features |= MMU_FTR_1T_SEGMENT; |
f5534004 | 308 | return 1; |
1189be65 | 309 | } |
1189be65 | 310 | } |
44ae3ab3 | 311 | cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B; |
1189be65 PM |
312 | return 0; |
313 | } | |
314 | ||
315 | static void __init htab_init_seg_sizes(void) | |
316 | { | |
317 | of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL); | |
318 | } | |
319 | ||
b1022fbd AK |
320 | static int __init get_idx_from_shift(unsigned int shift) |
321 | { | |
322 | int idx = -1; | |
323 | ||
324 | switch (shift) { | |
325 | case 0xc: | |
326 | idx = MMU_PAGE_4K; | |
327 | break; | |
328 | case 0x10: | |
329 | idx = MMU_PAGE_64K; | |
330 | break; | |
331 | case 0x14: | |
332 | idx = MMU_PAGE_1M; | |
333 | break; | |
334 | case 0x18: | |
335 | idx = MMU_PAGE_16M; | |
336 | break; | |
337 | case 0x22: | |
338 | idx = MMU_PAGE_16G; | |
339 | break; | |
340 | } | |
341 | return idx; | |
342 | } | |
343 | ||
3c726f8d BH |
344 | static int __init htab_dt_scan_page_sizes(unsigned long node, |
345 | const char *uname, int depth, | |
346 | void *data) | |
347 | { | |
9d0c4dfe RH |
348 | const char *type = of_get_flat_dt_prop(node, "device_type", NULL); |
349 | const __be32 *prop; | |
350 | int size = 0; | |
3c726f8d BH |
351 | |
352 | /* We are scanning "cpu" nodes only */ | |
353 | if (type == NULL || strcmp(type, "cpu") != 0) | |
354 | return 0; | |
355 | ||
12f04f2b | 356 | prop = of_get_flat_dt_prop(node, "ibm,segment-page-sizes", &size); |
9e34992a ME |
357 | if (!prop) |
358 | return 0; | |
359 | ||
360 | pr_info("Page sizes from device-tree:\n"); | |
361 | size /= 4; | |
362 | cur_cpu_spec->mmu_features &= ~(MMU_FTR_16M_PAGE); | |
363 | while(size > 0) { | |
364 | unsigned int base_shift = be32_to_cpu(prop[0]); | |
365 | unsigned int slbenc = be32_to_cpu(prop[1]); | |
366 | unsigned int lpnum = be32_to_cpu(prop[2]); | |
367 | struct mmu_psize_def *def; | |
368 | int idx, base_idx; | |
369 | ||
370 | size -= 3; prop += 3; | |
371 | base_idx = get_idx_from_shift(base_shift); | |
372 | if (base_idx < 0) { | |
373 | /* skip the pte encoding also */ | |
374 | prop += lpnum * 2; size -= lpnum * 2; | |
375 | continue; | |
376 | } | |
377 | def = &mmu_psize_defs[base_idx]; | |
378 | if (base_idx == MMU_PAGE_16M) | |
379 | cur_cpu_spec->mmu_features |= MMU_FTR_16M_PAGE; | |
380 | ||
381 | def->shift = base_shift; | |
382 | if (base_shift <= 23) | |
383 | def->avpnm = 0; | |
384 | else | |
385 | def->avpnm = (1 << (base_shift - 23)) - 1; | |
386 | def->sllp = slbenc; | |
387 | /* | |
388 | * We don't know for sure what's up with tlbiel, so | |
389 | * for now we only set it for 4K and 64K pages | |
390 | */ | |
391 | if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K) | |
392 | def->tlbiel = 1; | |
393 | else | |
394 | def->tlbiel = 0; | |
395 | ||
396 | while (size > 0 && lpnum) { | |
397 | unsigned int shift = be32_to_cpu(prop[0]); | |
398 | int penc = be32_to_cpu(prop[1]); | |
399 | ||
400 | prop += 2; size -= 2; | |
401 | lpnum--; | |
402 | ||
403 | idx = get_idx_from_shift(shift); | |
404 | if (idx < 0) | |
b1022fbd | 405 | continue; |
9e34992a ME |
406 | |
407 | if (penc == -1) | |
408 | pr_err("Invalid penc for base_shift=%d " | |
409 | "shift=%d\n", base_shift, shift); | |
410 | ||
411 | def->penc[idx] = penc; | |
412 | pr_info("base_shift=%d: shift=%d, sllp=0x%04lx," | |
413 | " avpnm=0x%08lx, tlbiel=%d, penc=%d\n", | |
414 | base_shift, shift, def->sllp, | |
415 | def->avpnm, def->tlbiel, def->penc[idx]); | |
1da177e4 | 416 | } |
3c726f8d | 417 | } |
9e34992a ME |
418 | |
419 | return 1; | |
3c726f8d BH |
420 | } |
421 | ||
e16a9c09 | 422 | #ifdef CONFIG_HUGETLB_PAGE |
658013e9 JT |
423 | /* Scan for 16G memory blocks that have been set aside for huge pages |
424 | * and reserve those blocks for 16G huge pages. | |
425 | */ | |
426 | static int __init htab_dt_scan_hugepage_blocks(unsigned long node, | |
427 | const char *uname, int depth, | |
428 | void *data) { | |
9d0c4dfe RH |
429 | const char *type = of_get_flat_dt_prop(node, "device_type", NULL); |
430 | const __be64 *addr_prop; | |
431 | const __be32 *page_count_prop; | |
658013e9 JT |
432 | unsigned int expected_pages; |
433 | long unsigned int phys_addr; | |
434 | long unsigned int block_size; | |
435 | ||
436 | /* We are scanning "memory" nodes only */ | |
437 | if (type == NULL || strcmp(type, "memory") != 0) | |
438 | return 0; | |
439 | ||
440 | /* This property is the log base 2 of the number of virtual pages that | |
441 | * will represent this memory block. */ | |
442 | page_count_prop = of_get_flat_dt_prop(node, "ibm,expected#pages", NULL); | |
443 | if (page_count_prop == NULL) | |
444 | return 0; | |
12f04f2b | 445 | expected_pages = (1 << be32_to_cpu(page_count_prop[0])); |
658013e9 JT |
446 | addr_prop = of_get_flat_dt_prop(node, "reg", NULL); |
447 | if (addr_prop == NULL) | |
448 | return 0; | |
12f04f2b AB |
449 | phys_addr = be64_to_cpu(addr_prop[0]); |
450 | block_size = be64_to_cpu(addr_prop[1]); | |
658013e9 JT |
451 | if (block_size != (16 * GB)) |
452 | return 0; | |
453 | printk(KERN_INFO "Huge page(16GB) memory: " | |
454 | "addr = 0x%lX size = 0x%lX pages = %d\n", | |
455 | phys_addr, block_size, expected_pages); | |
95f72d1e YL |
456 | if (phys_addr + (16 * GB) <= memblock_end_of_DRAM()) { |
457 | memblock_reserve(phys_addr, block_size * expected_pages); | |
4792adba JT |
458 | add_gpage(phys_addr, block_size, expected_pages); |
459 | } | |
658013e9 JT |
460 | return 0; |
461 | } | |
e16a9c09 | 462 | #endif /* CONFIG_HUGETLB_PAGE */ |
658013e9 | 463 | |
b1022fbd AK |
464 | static void mmu_psize_set_default_penc(void) |
465 | { | |
466 | int bpsize, apsize; | |
467 | for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) | |
468 | for (apsize = 0; apsize < MMU_PAGE_COUNT; apsize++) | |
469 | mmu_psize_defs[bpsize].penc[apsize] = -1; | |
470 | } | |
471 | ||
9048e648 AG |
472 | #ifdef CONFIG_PPC_64K_PAGES |
473 | ||
474 | static bool might_have_hea(void) | |
475 | { | |
476 | /* | |
477 | * The HEA ethernet adapter requires awareness of the | |
478 | * GX bus. Without that awareness we can easily assume | |
479 | * we will never see an HEA ethernet device. | |
480 | */ | |
481 | #ifdef CONFIG_IBMEBUS | |
482 | return !cpu_has_feature(CPU_FTR_ARCH_207S); | |
483 | #else | |
484 | return false; | |
485 | #endif | |
486 | } | |
487 | ||
488 | #endif /* #ifdef CONFIG_PPC_64K_PAGES */ | |
489 | ||
3c726f8d BH |
490 | static void __init htab_init_page_sizes(void) |
491 | { | |
492 | int rc; | |
493 | ||
b1022fbd AK |
494 | /* se the invalid penc to -1 */ |
495 | mmu_psize_set_default_penc(); | |
496 | ||
3c726f8d BH |
497 | /* Default to 4K pages only */ |
498 | memcpy(mmu_psize_defs, mmu_psize_defaults_old, | |
499 | sizeof(mmu_psize_defaults_old)); | |
500 | ||
501 | /* | |
502 | * Try to find the available page sizes in the device-tree | |
503 | */ | |
504 | rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL); | |
505 | if (rc != 0) /* Found */ | |
506 | goto found; | |
507 | ||
508 | /* | |
509 | * Not in the device-tree, let's fallback on known size | |
510 | * list for 16M capable GP & GR | |
511 | */ | |
44ae3ab3 | 512 | if (mmu_has_feature(MMU_FTR_16M_PAGE)) |
3c726f8d BH |
513 | memcpy(mmu_psize_defs, mmu_psize_defaults_gp, |
514 | sizeof(mmu_psize_defaults_gp)); | |
515 | found: | |
370a908d | 516 | #ifndef CONFIG_DEBUG_PAGEALLOC |
3c726f8d BH |
517 | /* |
518 | * Pick a size for the linear mapping. Currently, we only support | |
519 | * 16M, 1M and 4K which is the default | |
520 | */ | |
521 | if (mmu_psize_defs[MMU_PAGE_16M].shift) | |
522 | mmu_linear_psize = MMU_PAGE_16M; | |
523 | else if (mmu_psize_defs[MMU_PAGE_1M].shift) | |
524 | mmu_linear_psize = MMU_PAGE_1M; | |
370a908d | 525 | #endif /* CONFIG_DEBUG_PAGEALLOC */ |
3c726f8d | 526 | |
bf72aeba | 527 | #ifdef CONFIG_PPC_64K_PAGES |
3c726f8d BH |
528 | /* |
529 | * Pick a size for the ordinary pages. Default is 4K, we support | |
bf72aeba PM |
530 | * 64K for user mappings and vmalloc if supported by the processor. |
531 | * We only use 64k for ioremap if the processor | |
532 | * (and firmware) support cache-inhibited large pages. | |
533 | * If not, we use 4k and set mmu_ci_restrictions so that | |
534 | * hash_page knows to switch processes that use cache-inhibited | |
535 | * mappings to 4k pages. | |
3c726f8d | 536 | */ |
bf72aeba | 537 | if (mmu_psize_defs[MMU_PAGE_64K].shift) { |
3c726f8d | 538 | mmu_virtual_psize = MMU_PAGE_64K; |
bf72aeba | 539 | mmu_vmalloc_psize = MMU_PAGE_64K; |
370a908d BH |
540 | if (mmu_linear_psize == MMU_PAGE_4K) |
541 | mmu_linear_psize = MMU_PAGE_64K; | |
44ae3ab3 | 542 | if (mmu_has_feature(MMU_FTR_CI_LARGE_PAGE)) { |
cfe666b1 | 543 | /* |
9048e648 AG |
544 | * When running on pSeries using 64k pages for ioremap |
545 | * would stop us accessing the HEA ethernet. So if we | |
546 | * have the chance of ever seeing one, stay at 4k. | |
cfe666b1 | 547 | */ |
9048e648 | 548 | if (!might_have_hea() || !machine_is(pseries)) |
cfe666b1 PM |
549 | mmu_io_psize = MMU_PAGE_64K; |
550 | } else | |
bf72aeba PM |
551 | mmu_ci_restrictions = 1; |
552 | } | |
370a908d | 553 | #endif /* CONFIG_PPC_64K_PAGES */ |
3c726f8d | 554 | |
cec08e7a BH |
555 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
556 | /* We try to use 16M pages for vmemmap if that is supported | |
557 | * and we have at least 1G of RAM at boot | |
558 | */ | |
559 | if (mmu_psize_defs[MMU_PAGE_16M].shift && | |
95f72d1e | 560 | memblock_phys_mem_size() >= 0x40000000) |
cec08e7a BH |
561 | mmu_vmemmap_psize = MMU_PAGE_16M; |
562 | else if (mmu_psize_defs[MMU_PAGE_64K].shift) | |
563 | mmu_vmemmap_psize = MMU_PAGE_64K; | |
564 | else | |
565 | mmu_vmemmap_psize = MMU_PAGE_4K; | |
566 | #endif /* CONFIG_SPARSEMEM_VMEMMAP */ | |
567 | ||
bf72aeba | 568 | printk(KERN_DEBUG "Page orders: linear mapping = %d, " |
cec08e7a BH |
569 | "virtual = %d, io = %d" |
570 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
571 | ", vmemmap = %d" | |
572 | #endif | |
573 | "\n", | |
3c726f8d | 574 | mmu_psize_defs[mmu_linear_psize].shift, |
bf72aeba | 575 | mmu_psize_defs[mmu_virtual_psize].shift, |
cec08e7a BH |
576 | mmu_psize_defs[mmu_io_psize].shift |
577 | #ifdef CONFIG_SPARSEMEM_VMEMMAP | |
578 | ,mmu_psize_defs[mmu_vmemmap_psize].shift | |
579 | #endif | |
580 | ); | |
3c726f8d BH |
581 | |
582 | #ifdef CONFIG_HUGETLB_PAGE | |
658013e9 JT |
583 | /* Reserve 16G huge page memory sections for huge pages */ |
584 | of_scan_flat_dt(htab_dt_scan_hugepage_blocks, NULL); | |
3c726f8d BH |
585 | #endif /* CONFIG_HUGETLB_PAGE */ |
586 | } | |
587 | ||
588 | static int __init htab_dt_scan_pftsize(unsigned long node, | |
589 | const char *uname, int depth, | |
590 | void *data) | |
591 | { | |
9d0c4dfe RH |
592 | const char *type = of_get_flat_dt_prop(node, "device_type", NULL); |
593 | const __be32 *prop; | |
3c726f8d BH |
594 | |
595 | /* We are scanning "cpu" nodes only */ | |
596 | if (type == NULL || strcmp(type, "cpu") != 0) | |
597 | return 0; | |
598 | ||
12f04f2b | 599 | prop = of_get_flat_dt_prop(node, "ibm,pft-size", NULL); |
3c726f8d BH |
600 | if (prop != NULL) { |
601 | /* pft_size[0] is the NUMA CEC cookie */ | |
12f04f2b | 602 | ppc64_pft_size = be32_to_cpu(prop[1]); |
3c726f8d | 603 | return 1; |
1da177e4 | 604 | } |
3c726f8d | 605 | return 0; |
1da177e4 LT |
606 | } |
607 | ||
3c726f8d | 608 | static unsigned long __init htab_get_table_size(void) |
3eac8c69 | 609 | { |
13870b65 | 610 | unsigned long mem_size, rnd_mem_size, pteg_count, psize; |
3eac8c69 | 611 | |
3c726f8d | 612 | /* If hash size isn't already provided by the platform, we try to |
943ffb58 | 613 | * retrieve it from the device-tree. If it's not there neither, we |
3c726f8d | 614 | * calculate it now based on the total RAM size |
3eac8c69 | 615 | */ |
3c726f8d BH |
616 | if (ppc64_pft_size == 0) |
617 | of_scan_flat_dt(htab_dt_scan_pftsize, NULL); | |
3eac8c69 PM |
618 | if (ppc64_pft_size) |
619 | return 1UL << ppc64_pft_size; | |
620 | ||
621 | /* round mem_size up to next power of 2 */ | |
95f72d1e | 622 | mem_size = memblock_phys_mem_size(); |
799d6046 PM |
623 | rnd_mem_size = 1UL << __ilog2(mem_size); |
624 | if (rnd_mem_size < mem_size) | |
3eac8c69 PM |
625 | rnd_mem_size <<= 1; |
626 | ||
627 | /* # pages / 2 */ | |
13870b65 AB |
628 | psize = mmu_psize_defs[mmu_virtual_psize].shift; |
629 | pteg_count = max(rnd_mem_size >> (psize + 1), 1UL << 11); | |
3eac8c69 PM |
630 | |
631 | return pteg_count << 7; | |
632 | } | |
633 | ||
54b79248 | 634 | #ifdef CONFIG_MEMORY_HOTPLUG |
a1194097 | 635 | int create_section_mapping(unsigned long start, unsigned long end) |
54b79248 | 636 | { |
a1194097 | 637 | return htab_bolt_mapping(start, end, __pa(start), |
f5ea64dc | 638 | pgprot_val(PAGE_KERNEL), mmu_linear_psize, |
a1194097 | 639 | mmu_kernel_ssize); |
54b79248 | 640 | } |
f8c8803b | 641 | |
52db9b44 | 642 | int remove_section_mapping(unsigned long start, unsigned long end) |
f8c8803b | 643 | { |
52db9b44 BP |
644 | return htab_remove_mapping(start, end, mmu_linear_psize, |
645 | mmu_kernel_ssize); | |
f8c8803b | 646 | } |
54b79248 MK |
647 | #endif /* CONFIG_MEMORY_HOTPLUG */ |
648 | ||
757c74d2 | 649 | static void __init htab_initialize(void) |
1da177e4 | 650 | { |
337a7128 | 651 | unsigned long table; |
1da177e4 | 652 | unsigned long pteg_count; |
9e88ba4e | 653 | unsigned long prot; |
41d824bf | 654 | unsigned long base = 0, size = 0, limit; |
28be7072 | 655 | struct memblock_region *reg; |
3c726f8d | 656 | |
1da177e4 LT |
657 | DBG(" -> htab_initialize()\n"); |
658 | ||
1189be65 PM |
659 | /* Initialize segment sizes */ |
660 | htab_init_seg_sizes(); | |
661 | ||
3c726f8d BH |
662 | /* Initialize page sizes */ |
663 | htab_init_page_sizes(); | |
664 | ||
44ae3ab3 | 665 | if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) { |
1189be65 PM |
666 | mmu_kernel_ssize = MMU_SEGSIZE_1T; |
667 | mmu_highuser_ssize = MMU_SEGSIZE_1T; | |
668 | printk(KERN_INFO "Using 1TB segments\n"); | |
669 | } | |
670 | ||
1da177e4 LT |
671 | /* |
672 | * Calculate the required size of the htab. We want the number of | |
673 | * PTEGs to equal one half the number of real pages. | |
674 | */ | |
3c726f8d | 675 | htab_size_bytes = htab_get_table_size(); |
1da177e4 LT |
676 | pteg_count = htab_size_bytes >> 7; |
677 | ||
1da177e4 LT |
678 | htab_hash_mask = pteg_count - 1; |
679 | ||
57cfb814 | 680 | if (firmware_has_feature(FW_FEATURE_LPAR)) { |
1da177e4 LT |
681 | /* Using a hypervisor which owns the htab */ |
682 | htab_address = NULL; | |
683 | _SDR1 = 0; | |
3ccc00a7 MS |
684 | #ifdef CONFIG_FA_DUMP |
685 | /* | |
686 | * If firmware assisted dump is active firmware preserves | |
687 | * the contents of htab along with entire partition memory. | |
688 | * Clear the htab if firmware assisted dump is active so | |
689 | * that we dont end up using old mappings. | |
690 | */ | |
691 | if (is_fadump_active() && ppc_md.hpte_clear_all) | |
692 | ppc_md.hpte_clear_all(); | |
693 | #endif | |
1da177e4 LT |
694 | } else { |
695 | /* Find storage for the HPT. Must be contiguous in | |
41d824bf | 696 | * the absolute address space. On cell we want it to be |
31bf1119 | 697 | * in the first 2 Gig so we can use it for IOMMU hacks. |
1da177e4 | 698 | */ |
41d824bf | 699 | if (machine_is(cell)) |
31bf1119 | 700 | limit = 0x80000000; |
41d824bf | 701 | else |
27f574c2 | 702 | limit = MEMBLOCK_ALLOC_ANYWHERE; |
41d824bf | 703 | |
95f72d1e | 704 | table = memblock_alloc_base(htab_size_bytes, htab_size_bytes, limit); |
1da177e4 LT |
705 | |
706 | DBG("Hash table allocated at %lx, size: %lx\n", table, | |
707 | htab_size_bytes); | |
708 | ||
70267a7f | 709 | htab_address = __va(table); |
1da177e4 LT |
710 | |
711 | /* htab absolute addr + encoded htabsize */ | |
712 | _SDR1 = table + __ilog2(pteg_count) - 11; | |
713 | ||
714 | /* Initialize the HPT with no entries */ | |
715 | memset((void *)table, 0, htab_size_bytes); | |
799d6046 PM |
716 | |
717 | /* Set SDR1 */ | |
718 | mtspr(SPRN_SDR1, _SDR1); | |
1da177e4 LT |
719 | } |
720 | ||
f5ea64dc | 721 | prot = pgprot_val(PAGE_KERNEL); |
1da177e4 | 722 | |
370a908d | 723 | #ifdef CONFIG_DEBUG_PAGEALLOC |
95f72d1e YL |
724 | linear_map_hash_count = memblock_end_of_DRAM() >> PAGE_SHIFT; |
725 | linear_map_hash_slots = __va(memblock_alloc_base(linear_map_hash_count, | |
cd3db0c4 | 726 | 1, ppc64_rma_size)); |
370a908d BH |
727 | memset(linear_map_hash_slots, 0, linear_map_hash_count); |
728 | #endif /* CONFIG_DEBUG_PAGEALLOC */ | |
729 | ||
1da177e4 LT |
730 | /* On U3 based machines, we need to reserve the DART area and |
731 | * _NOT_ map it to avoid cache paradoxes as it's remapped non | |
732 | * cacheable later on | |
733 | */ | |
1da177e4 LT |
734 | |
735 | /* create bolted the linear mapping in the hash table */ | |
28be7072 BH |
736 | for_each_memblock(memory, reg) { |
737 | base = (unsigned long)__va(reg->base); | |
738 | size = reg->size; | |
1da177e4 | 739 | |
5c339919 | 740 | DBG("creating mapping for region: %lx..%lx (prot: %lx)\n", |
9e88ba4e | 741 | base, size, prot); |
1da177e4 LT |
742 | |
743 | #ifdef CONFIG_U3_DART | |
744 | /* Do not map the DART space. Fortunately, it will be aligned | |
95f72d1e | 745 | * in such a way that it will not cross two memblock regions and |
3c726f8d BH |
746 | * will fit within a single 16Mb page. |
747 | * The DART space is assumed to be a full 16Mb region even if | |
748 | * we only use 2Mb of that space. We will use more of it later | |
749 | * for AGP GART. We have to use a full 16Mb large page. | |
1da177e4 LT |
750 | */ |
751 | DBG("DART base: %lx\n", dart_tablebase); | |
752 | ||
753 | if (dart_tablebase != 0 && dart_tablebase >= base | |
754 | && dart_tablebase < (base + size)) { | |
caf80e57 | 755 | unsigned long dart_table_end = dart_tablebase + 16 * MB; |
1da177e4 | 756 | if (base != dart_tablebase) |
3c726f8d | 757 | BUG_ON(htab_bolt_mapping(base, dart_tablebase, |
9e88ba4e | 758 | __pa(base), prot, |
1189be65 PM |
759 | mmu_linear_psize, |
760 | mmu_kernel_ssize)); | |
caf80e57 | 761 | if ((base + size) > dart_table_end) |
3c726f8d | 762 | BUG_ON(htab_bolt_mapping(dart_tablebase+16*MB, |
caf80e57 ME |
763 | base + size, |
764 | __pa(dart_table_end), | |
9e88ba4e | 765 | prot, |
1189be65 PM |
766 | mmu_linear_psize, |
767 | mmu_kernel_ssize)); | |
1da177e4 LT |
768 | continue; |
769 | } | |
770 | #endif /* CONFIG_U3_DART */ | |
caf80e57 | 771 | BUG_ON(htab_bolt_mapping(base, base + size, __pa(base), |
9e88ba4e | 772 | prot, mmu_linear_psize, mmu_kernel_ssize)); |
e63075a3 BH |
773 | } |
774 | memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE); | |
1da177e4 LT |
775 | |
776 | /* | |
777 | * If we have a memory_limit and we've allocated TCEs then we need to | |
778 | * explicitly map the TCE area at the top of RAM. We also cope with the | |
779 | * case that the TCEs start below memory_limit. | |
780 | * tce_alloc_start/end are 16MB aligned so the mapping should work | |
781 | * for either 4K or 16MB pages. | |
782 | */ | |
783 | if (tce_alloc_start) { | |
b5666f70 ME |
784 | tce_alloc_start = (unsigned long)__va(tce_alloc_start); |
785 | tce_alloc_end = (unsigned long)__va(tce_alloc_end); | |
1da177e4 LT |
786 | |
787 | if (base + size >= tce_alloc_start) | |
788 | tce_alloc_start = base + size + 1; | |
789 | ||
caf80e57 | 790 | BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end, |
bc033b63 | 791 | __pa(tce_alloc_start), prot, |
1189be65 | 792 | mmu_linear_psize, mmu_kernel_ssize)); |
1da177e4 LT |
793 | } |
794 | ||
7d0daae4 | 795 | |
1da177e4 LT |
796 | DBG(" <- htab_initialize()\n"); |
797 | } | |
798 | #undef KB | |
799 | #undef MB | |
1da177e4 | 800 | |
757c74d2 | 801 | void __init early_init_mmu(void) |
799d6046 | 802 | { |
757c74d2 | 803 | /* Initialize the MMU Hash table and create the linear mapping |
376af594 ME |
804 | * of memory. Has to be done before SLB initialization as this is |
805 | * currently where the page size encoding is obtained. | |
757c74d2 BH |
806 | */ |
807 | htab_initialize(); | |
808 | ||
376af594 | 809 | /* Initialize SLB management */ |
13b3d13b | 810 | slb_initialize(); |
757c74d2 BH |
811 | } |
812 | ||
813 | #ifdef CONFIG_SMP | |
061d19f2 | 814 | void early_init_mmu_secondary(void) |
757c74d2 BH |
815 | { |
816 | /* Initialize hash table for that CPU */ | |
57cfb814 | 817 | if (!firmware_has_feature(FW_FEATURE_LPAR)) |
799d6046 | 818 | mtspr(SPRN_SDR1, _SDR1); |
757c74d2 | 819 | |
376af594 | 820 | /* Initialize SLB */ |
13b3d13b | 821 | slb_initialize(); |
799d6046 | 822 | } |
757c74d2 | 823 | #endif /* CONFIG_SMP */ |
799d6046 | 824 | |
1da177e4 LT |
825 | /* |
826 | * Called by asm hashtable.S for doing lazy icache flush | |
827 | */ | |
828 | unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap) | |
829 | { | |
830 | struct page *page; | |
831 | ||
76c8e25b BH |
832 | if (!pfn_valid(pte_pfn(pte))) |
833 | return pp; | |
834 | ||
1da177e4 LT |
835 | page = pte_page(pte); |
836 | ||
837 | /* page is dirty */ | |
838 | if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) { | |
839 | if (trap == 0x400) { | |
0895ecda | 840 | flush_dcache_icache_page(page); |
1da177e4 LT |
841 | set_bit(PG_arch_1, &page->flags); |
842 | } else | |
3c726f8d | 843 | pp |= HPTE_R_N; |
1da177e4 LT |
844 | } |
845 | return pp; | |
846 | } | |
847 | ||
3a8247cc | 848 | #ifdef CONFIG_PPC_MM_SLICES |
e51df2c1 | 849 | static unsigned int get_paca_psize(unsigned long addr) |
3a8247cc | 850 | { |
7aa0727f AK |
851 | u64 lpsizes; |
852 | unsigned char *hpsizes; | |
853 | unsigned long index, mask_index; | |
3a8247cc PM |
854 | |
855 | if (addr < SLICE_LOW_TOP) { | |
2fc251a8 | 856 | lpsizes = get_paca()->mm_ctx_low_slices_psize; |
3a8247cc | 857 | index = GET_LOW_SLICE_INDEX(addr); |
7aa0727f | 858 | return (lpsizes >> (index * 4)) & 0xF; |
3a8247cc | 859 | } |
2fc251a8 | 860 | hpsizes = get_paca()->mm_ctx_high_slices_psize; |
7aa0727f AK |
861 | index = GET_HIGH_SLICE_INDEX(addr); |
862 | mask_index = index & 0x1; | |
863 | return (hpsizes[index >> 1] >> (mask_index * 4)) & 0xF; | |
3a8247cc PM |
864 | } |
865 | ||
866 | #else | |
867 | unsigned int get_paca_psize(unsigned long addr) | |
868 | { | |
869 | return get_paca()->context.user_psize; | |
870 | } | |
871 | #endif | |
872 | ||
721151d0 PM |
873 | /* |
874 | * Demote a segment to using 4k pages. | |
875 | * For now this makes the whole process use 4k pages. | |
876 | */ | |
721151d0 | 877 | #ifdef CONFIG_PPC_64K_PAGES |
fa28237c | 878 | void demote_segment_4k(struct mm_struct *mm, unsigned long addr) |
16f1c746 | 879 | { |
3a8247cc | 880 | if (get_slice_psize(mm, addr) == MMU_PAGE_4K) |
721151d0 | 881 | return; |
3a8247cc | 882 | slice_set_range_psize(mm, addr, 1, MMU_PAGE_4K); |
be3ebfe8 | 883 | copro_flush_all_slbs(mm); |
a1dca346 | 884 | if ((get_paca_psize(addr) != MMU_PAGE_4K) && (current->mm == mm)) { |
c395465d MN |
885 | |
886 | copy_mm_to_paca(&mm->context); | |
fa28237c PM |
887 | slb_flush_and_rebolt(); |
888 | } | |
721151d0 | 889 | } |
16f1c746 | 890 | #endif /* CONFIG_PPC_64K_PAGES */ |
721151d0 | 891 | |
fa28237c PM |
892 | #ifdef CONFIG_PPC_SUBPAGE_PROT |
893 | /* | |
894 | * This looks up a 2-bit protection code for a 4k subpage of a 64k page. | |
895 | * Userspace sets the subpage permissions using the subpage_prot system call. | |
896 | * | |
897 | * Result is 0: full permissions, _PAGE_RW: read-only, | |
898 | * _PAGE_USER or _PAGE_USER|_PAGE_RW: no access. | |
899 | */ | |
d28513bc | 900 | static int subpage_protection(struct mm_struct *mm, unsigned long ea) |
fa28237c | 901 | { |
d28513bc | 902 | struct subpage_prot_table *spt = &mm->context.spt; |
fa28237c PM |
903 | u32 spp = 0; |
904 | u32 **sbpm, *sbpp; | |
905 | ||
906 | if (ea >= spt->maxaddr) | |
907 | return 0; | |
b0d436c7 | 908 | if (ea < 0x100000000UL) { |
fa28237c PM |
909 | /* addresses below 4GB use spt->low_prot */ |
910 | sbpm = spt->low_prot; | |
911 | } else { | |
912 | sbpm = spt->protptrs[ea >> SBP_L3_SHIFT]; | |
913 | if (!sbpm) | |
914 | return 0; | |
915 | } | |
916 | sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)]; | |
917 | if (!sbpp) | |
918 | return 0; | |
919 | spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)]; | |
920 | ||
921 | /* extract 2-bit bitfield for this 4k subpage */ | |
922 | spp >>= 30 - 2 * ((ea >> 12) & 0xf); | |
923 | ||
924 | /* turn 0,1,2,3 into combination of _PAGE_USER and _PAGE_RW */ | |
925 | spp = ((spp & 2) ? _PAGE_USER : 0) | ((spp & 1) ? _PAGE_RW : 0); | |
926 | return spp; | |
927 | } | |
928 | ||
929 | #else /* CONFIG_PPC_SUBPAGE_PROT */ | |
d28513bc | 930 | static inline int subpage_protection(struct mm_struct *mm, unsigned long ea) |
fa28237c PM |
931 | { |
932 | return 0; | |
933 | } | |
934 | #endif | |
935 | ||
4b8692c0 BH |
936 | void hash_failure_debug(unsigned long ea, unsigned long access, |
937 | unsigned long vsid, unsigned long trap, | |
d8139ebf | 938 | int ssize, int psize, int lpsize, unsigned long pte) |
4b8692c0 BH |
939 | { |
940 | if (!printk_ratelimit()) | |
941 | return; | |
942 | pr_info("mm: Hashing failure ! EA=0x%lx access=0x%lx current=%s\n", | |
943 | ea, access, current->comm); | |
d8139ebf AK |
944 | pr_info(" trap=0x%lx vsid=0x%lx ssize=%d base psize=%d psize %d pte=0x%lx\n", |
945 | trap, vsid, ssize, psize, lpsize, pte); | |
4b8692c0 BH |
946 | } |
947 | ||
09567e7f ME |
948 | static void check_paca_psize(unsigned long ea, struct mm_struct *mm, |
949 | int psize, bool user_region) | |
950 | { | |
951 | if (user_region) { | |
952 | if (psize != get_paca_psize(ea)) { | |
c395465d | 953 | copy_mm_to_paca(&mm->context); |
09567e7f ME |
954 | slb_flush_and_rebolt(); |
955 | } | |
956 | } else if (get_paca()->vmalloc_sllp != | |
957 | mmu_psize_defs[mmu_vmalloc_psize].sllp) { | |
958 | get_paca()->vmalloc_sllp = | |
959 | mmu_psize_defs[mmu_vmalloc_psize].sllp; | |
960 | slb_vmalloc_update(); | |
961 | } | |
962 | } | |
963 | ||
1da177e4 LT |
964 | /* Result code is: |
965 | * 0 - handled | |
966 | * 1 - normal page fault | |
967 | * -1 - critical hash insertion error | |
fa28237c | 968 | * -2 - access not permitted by subpage protection mechanism |
1da177e4 | 969 | */ |
aefa5688 AK |
970 | int hash_page_mm(struct mm_struct *mm, unsigned long ea, |
971 | unsigned long access, unsigned long trap, | |
972 | unsigned long flags) | |
1da177e4 | 973 | { |
891121e6 | 974 | bool is_thp; |
ba12eede | 975 | enum ctx_state prev_state = exception_enter(); |
a1128f8f | 976 | pgd_t *pgdir; |
1da177e4 | 977 | unsigned long vsid; |
1da177e4 | 978 | pte_t *ptep; |
a4fe3ce7 | 979 | unsigned hugeshift; |
56aa4129 | 980 | const struct cpumask *tmp; |
aefa5688 | 981 | int rc, user_region = 0; |
1189be65 | 982 | int psize, ssize; |
1da177e4 | 983 | |
3c726f8d BH |
984 | DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n", |
985 | ea, access, trap); | |
cfcb3d80 | 986 | trace_hash_fault(ea, access, trap); |
1f8d419e | 987 | |
3c726f8d | 988 | /* Get region & vsid */ |
1da177e4 LT |
989 | switch (REGION_ID(ea)) { |
990 | case USER_REGION_ID: | |
991 | user_region = 1; | |
3c726f8d BH |
992 | if (! mm) { |
993 | DBG_LOW(" user region with no mm !\n"); | |
ba12eede LZ |
994 | rc = 1; |
995 | goto bail; | |
3c726f8d | 996 | } |
16c2d476 | 997 | psize = get_slice_psize(mm, ea); |
1189be65 PM |
998 | ssize = user_segment_size(ea); |
999 | vsid = get_vsid(mm->context.id, ea, ssize); | |
1da177e4 | 1000 | break; |
1da177e4 | 1001 | case VMALLOC_REGION_ID: |
1189be65 | 1002 | vsid = get_kernel_vsid(ea, mmu_kernel_ssize); |
bf72aeba PM |
1003 | if (ea < VMALLOC_END) |
1004 | psize = mmu_vmalloc_psize; | |
1005 | else | |
1006 | psize = mmu_io_psize; | |
1189be65 | 1007 | ssize = mmu_kernel_ssize; |
1da177e4 | 1008 | break; |
1da177e4 LT |
1009 | default: |
1010 | /* Not a valid range | |
1011 | * Send the problem up to do_page_fault | |
1012 | */ | |
ba12eede LZ |
1013 | rc = 1; |
1014 | goto bail; | |
1da177e4 | 1015 | } |
3c726f8d | 1016 | DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid); |
1da177e4 | 1017 | |
c60ac569 AK |
1018 | /* Bad address. */ |
1019 | if (!vsid) { | |
1020 | DBG_LOW("Bad address!\n"); | |
ba12eede LZ |
1021 | rc = 1; |
1022 | goto bail; | |
c60ac569 | 1023 | } |
3c726f8d | 1024 | /* Get pgdir */ |
1da177e4 | 1025 | pgdir = mm->pgd; |
ba12eede LZ |
1026 | if (pgdir == NULL) { |
1027 | rc = 1; | |
1028 | goto bail; | |
1029 | } | |
1da177e4 | 1030 | |
3c726f8d | 1031 | /* Check CPU locality */ |
56aa4129 RR |
1032 | tmp = cpumask_of(smp_processor_id()); |
1033 | if (user_region && cpumask_equal(mm_cpumask(mm), tmp)) | |
aefa5688 | 1034 | flags |= HPTE_LOCAL_UPDATE; |
1da177e4 | 1035 | |
16c2d476 | 1036 | #ifndef CONFIG_PPC_64K_PAGES |
a4fe3ce7 DG |
1037 | /* If we use 4K pages and our psize is not 4K, then we might |
1038 | * be hitting a special driver mapping, and need to align the | |
1039 | * address before we fetch the PTE. | |
1040 | * | |
1041 | * It could also be a hugepage mapping, in which case this is | |
1042 | * not necessary, but it's not harmful, either. | |
16c2d476 BH |
1043 | */ |
1044 | if (psize != MMU_PAGE_4K) | |
1045 | ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1); | |
1046 | #endif /* CONFIG_PPC_64K_PAGES */ | |
1047 | ||
3c726f8d | 1048 | /* Get PTE and page size from page tables */ |
891121e6 | 1049 | ptep = __find_linux_pte_or_hugepte(pgdir, ea, &is_thp, &hugeshift); |
3c726f8d BH |
1050 | if (ptep == NULL || !pte_present(*ptep)) { |
1051 | DBG_LOW(" no PTE !\n"); | |
ba12eede LZ |
1052 | rc = 1; |
1053 | goto bail; | |
3c726f8d BH |
1054 | } |
1055 | ||
ca91e6c0 BH |
1056 | /* Add _PAGE_PRESENT to the required access perm */ |
1057 | access |= _PAGE_PRESENT; | |
1058 | ||
1059 | /* Pre-check access permissions (will be re-checked atomically | |
1060 | * in __hash_page_XX but this pre-check is a fast path | |
1061 | */ | |
1062 | if (access & ~pte_val(*ptep)) { | |
1063 | DBG_LOW(" no access !\n"); | |
ba12eede LZ |
1064 | rc = 1; |
1065 | goto bail; | |
ca91e6c0 BH |
1066 | } |
1067 | ||
ba12eede | 1068 | if (hugeshift) { |
891121e6 | 1069 | if (is_thp) |
6d492ecc | 1070 | rc = __hash_page_thp(ea, access, vsid, (pmd_t *)ptep, |
aefa5688 | 1071 | trap, flags, ssize, psize); |
6d492ecc AK |
1072 | #ifdef CONFIG_HUGETLB_PAGE |
1073 | else | |
1074 | rc = __hash_page_huge(ea, access, vsid, ptep, trap, | |
aefa5688 | 1075 | flags, ssize, hugeshift, psize); |
6d492ecc AK |
1076 | #else |
1077 | else { | |
1078 | /* | |
1079 | * if we have hugeshift, and is not transhuge with | |
1080 | * hugetlb disabled, something is really wrong. | |
1081 | */ | |
1082 | rc = 1; | |
1083 | WARN_ON(1); | |
1084 | } | |
1085 | #endif | |
a1dca346 IM |
1086 | if (current->mm == mm) |
1087 | check_paca_psize(ea, mm, psize, user_region); | |
09567e7f | 1088 | |
ba12eede LZ |
1089 | goto bail; |
1090 | } | |
a4fe3ce7 | 1091 | |
3c726f8d BH |
1092 | #ifndef CONFIG_PPC_64K_PAGES |
1093 | DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep)); | |
1094 | #else | |
1095 | DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep), | |
1096 | pte_val(*(ptep + PTRS_PER_PTE))); | |
1097 | #endif | |
3c726f8d | 1098 | /* Do actual hashing */ |
16c2d476 | 1099 | #ifdef CONFIG_PPC_64K_PAGES |
721151d0 | 1100 | /* If _PAGE_4K_PFN is set, make sure this is a 4k segment */ |
3a8247cc | 1101 | if ((pte_val(*ptep) & _PAGE_4K_PFN) && psize == MMU_PAGE_64K) { |
721151d0 PM |
1102 | demote_segment_4k(mm, ea); |
1103 | psize = MMU_PAGE_4K; | |
1104 | } | |
1105 | ||
16f1c746 BH |
1106 | /* If this PTE is non-cacheable and we have restrictions on |
1107 | * using non cacheable large pages, then we switch to 4k | |
1108 | */ | |
1109 | if (mmu_ci_restrictions && psize == MMU_PAGE_64K && | |
1110 | (pte_val(*ptep) & _PAGE_NO_CACHE)) { | |
1111 | if (user_region) { | |
1112 | demote_segment_4k(mm, ea); | |
1113 | psize = MMU_PAGE_4K; | |
1114 | } else if (ea < VMALLOC_END) { | |
1115 | /* | |
1116 | * some driver did a non-cacheable mapping | |
1117 | * in vmalloc space, so switch vmalloc | |
1118 | * to 4k pages | |
1119 | */ | |
1120 | printk(KERN_ALERT "Reducing vmalloc segment " | |
1121 | "to 4kB pages because of " | |
1122 | "non-cacheable mapping\n"); | |
1123 | psize = mmu_vmalloc_psize = MMU_PAGE_4K; | |
be3ebfe8 | 1124 | copro_flush_all_slbs(mm); |
bf72aeba | 1125 | } |
16f1c746 | 1126 | } |
09567e7f | 1127 | |
0863d7f2 AK |
1128 | #endif /* CONFIG_PPC_64K_PAGES */ |
1129 | ||
a1dca346 IM |
1130 | if (current->mm == mm) |
1131 | check_paca_psize(ea, mm, psize, user_region); | |
16f1c746 | 1132 | |
73b341ef | 1133 | #ifdef CONFIG_PPC_64K_PAGES |
bf72aeba | 1134 | if (psize == MMU_PAGE_64K) |
aefa5688 AK |
1135 | rc = __hash_page_64K(ea, access, vsid, ptep, trap, |
1136 | flags, ssize); | |
3c726f8d | 1137 | else |
73b341ef | 1138 | #endif /* CONFIG_PPC_64K_PAGES */ |
fa28237c | 1139 | { |
a1128f8f | 1140 | int spp = subpage_protection(mm, ea); |
fa28237c PM |
1141 | if (access & spp) |
1142 | rc = -2; | |
1143 | else | |
1144 | rc = __hash_page_4K(ea, access, vsid, ptep, trap, | |
aefa5688 | 1145 | flags, ssize, spp); |
fa28237c | 1146 | } |
3c726f8d | 1147 | |
4b8692c0 BH |
1148 | /* Dump some info in case of hash insertion failure, they should |
1149 | * never happen so it is really useful to know if/when they do | |
1150 | */ | |
1151 | if (rc == -1) | |
1152 | hash_failure_debug(ea, access, vsid, trap, ssize, psize, | |
d8139ebf | 1153 | psize, pte_val(*ptep)); |
3c726f8d BH |
1154 | #ifndef CONFIG_PPC_64K_PAGES |
1155 | DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep)); | |
1156 | #else | |
1157 | DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep), | |
1158 | pte_val(*(ptep + PTRS_PER_PTE))); | |
1159 | #endif | |
1160 | DBG_LOW(" -> rc=%d\n", rc); | |
ba12eede LZ |
1161 | |
1162 | bail: | |
1163 | exception_exit(prev_state); | |
3c726f8d | 1164 | return rc; |
1da177e4 | 1165 | } |
a1dca346 IM |
1166 | EXPORT_SYMBOL_GPL(hash_page_mm); |
1167 | ||
aefa5688 AK |
1168 | int hash_page(unsigned long ea, unsigned long access, unsigned long trap, |
1169 | unsigned long dsisr) | |
a1dca346 | 1170 | { |
aefa5688 | 1171 | unsigned long flags = 0; |
a1dca346 IM |
1172 | struct mm_struct *mm = current->mm; |
1173 | ||
1174 | if (REGION_ID(ea) == VMALLOC_REGION_ID) | |
1175 | mm = &init_mm; | |
1176 | ||
aefa5688 AK |
1177 | if (dsisr & DSISR_NOHPTE) |
1178 | flags |= HPTE_NOHPTE_UPDATE; | |
1179 | ||
1180 | return hash_page_mm(mm, ea, access, trap, flags); | |
a1dca346 | 1181 | } |
67207b96 | 1182 | EXPORT_SYMBOL_GPL(hash_page); |
1da177e4 | 1183 | |
106713a1 AK |
1184 | int __hash_page(unsigned long ea, unsigned long msr, unsigned long trap, |
1185 | unsigned long dsisr) | |
1186 | { | |
1187 | unsigned long access = _PAGE_PRESENT; | |
1188 | unsigned long flags = 0; | |
1189 | struct mm_struct *mm = current->mm; | |
1190 | ||
1191 | if (REGION_ID(ea) == VMALLOC_REGION_ID) | |
1192 | mm = &init_mm; | |
1193 | ||
1194 | if (dsisr & DSISR_NOHPTE) | |
1195 | flags |= HPTE_NOHPTE_UPDATE; | |
1196 | ||
1197 | if (dsisr & DSISR_ISSTORE) | |
1198 | access |= _PAGE_RW; | |
1199 | /* | |
1200 | * We need to set the _PAGE_USER bit if MSR_PR is set or if we are | |
1201 | * accessing a userspace segment (even from the kernel). We assume | |
1202 | * kernel addresses always have the high bit set. | |
1203 | */ | |
1204 | if ((msr & MSR_PR) || (REGION_ID(ea) == USER_REGION_ID)) | |
1205 | access |= _PAGE_USER; | |
1206 | ||
1207 | if (trap == 0x400) | |
1208 | access |= _PAGE_EXEC; | |
1209 | ||
1210 | return hash_page_mm(mm, ea, access, trap, flags); | |
1211 | } | |
1212 | ||
3c726f8d BH |
1213 | void hash_preload(struct mm_struct *mm, unsigned long ea, |
1214 | unsigned long access, unsigned long trap) | |
1da177e4 | 1215 | { |
12bc9f6f | 1216 | int hugepage_shift; |
3c726f8d | 1217 | unsigned long vsid; |
0b97fee0 | 1218 | pgd_t *pgdir; |
3c726f8d | 1219 | pte_t *ptep; |
3c726f8d | 1220 | unsigned long flags; |
aefa5688 | 1221 | int rc, ssize, update_flags = 0; |
3c726f8d | 1222 | |
d0f13e3c BH |
1223 | BUG_ON(REGION_ID(ea) != USER_REGION_ID); |
1224 | ||
1225 | #ifdef CONFIG_PPC_MM_SLICES | |
1226 | /* We only prefault standard pages for now */ | |
2b02d139 | 1227 | if (unlikely(get_slice_psize(mm, ea) != mm->context.user_psize)) |
3c726f8d | 1228 | return; |
d0f13e3c | 1229 | #endif |
3c726f8d BH |
1230 | |
1231 | DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx," | |
1232 | " trap=%lx\n", mm, mm->pgd, ea, access, trap); | |
1da177e4 | 1233 | |
16f1c746 | 1234 | /* Get Linux PTE if available */ |
3c726f8d BH |
1235 | pgdir = mm->pgd; |
1236 | if (pgdir == NULL) | |
1237 | return; | |
0ac52dd7 AK |
1238 | |
1239 | /* Get VSID */ | |
1240 | ssize = user_segment_size(ea); | |
1241 | vsid = get_vsid(mm->context.id, ea, ssize); | |
1242 | if (!vsid) | |
1243 | return; | |
1244 | /* | |
1245 | * Hash doesn't like irqs. Walking linux page table with irq disabled | |
1246 | * saves us from holding multiple locks. | |
1247 | */ | |
1248 | local_irq_save(flags); | |
1249 | ||
12bc9f6f AK |
1250 | /* |
1251 | * THP pages use update_mmu_cache_pmd. We don't do | |
1252 | * hash preload there. Hence can ignore THP here | |
1253 | */ | |
891121e6 | 1254 | ptep = find_linux_pte_or_hugepte(pgdir, ea, NULL, &hugepage_shift); |
3c726f8d | 1255 | if (!ptep) |
0ac52dd7 | 1256 | goto out_exit; |
16f1c746 | 1257 | |
12bc9f6f | 1258 | WARN_ON(hugepage_shift); |
16f1c746 BH |
1259 | #ifdef CONFIG_PPC_64K_PAGES |
1260 | /* If either _PAGE_4K_PFN or _PAGE_NO_CACHE is set (and we are on | |
1261 | * a 64K kernel), then we don't preload, hash_page() will take | |
1262 | * care of it once we actually try to access the page. | |
1263 | * That way we don't have to duplicate all of the logic for segment | |
1264 | * page size demotion here | |
1265 | */ | |
1266 | if (pte_val(*ptep) & (_PAGE_4K_PFN | _PAGE_NO_CACHE)) | |
0ac52dd7 | 1267 | goto out_exit; |
16f1c746 BH |
1268 | #endif /* CONFIG_PPC_64K_PAGES */ |
1269 | ||
16c2d476 | 1270 | /* Is that local to this CPU ? */ |
56aa4129 | 1271 | if (cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id()))) |
aefa5688 | 1272 | update_flags |= HPTE_LOCAL_UPDATE; |
16c2d476 BH |
1273 | |
1274 | /* Hash it in */ | |
73b341ef | 1275 | #ifdef CONFIG_PPC_64K_PAGES |
bf72aeba | 1276 | if (mm->context.user_psize == MMU_PAGE_64K) |
aefa5688 AK |
1277 | rc = __hash_page_64K(ea, access, vsid, ptep, trap, |
1278 | update_flags, ssize); | |
1da177e4 | 1279 | else |
73b341ef | 1280 | #endif /* CONFIG_PPC_64K_PAGES */ |
aefa5688 AK |
1281 | rc = __hash_page_4K(ea, access, vsid, ptep, trap, update_flags, |
1282 | ssize, subpage_protection(mm, ea)); | |
4b8692c0 BH |
1283 | |
1284 | /* Dump some info in case of hash insertion failure, they should | |
1285 | * never happen so it is really useful to know if/when they do | |
1286 | */ | |
1287 | if (rc == -1) | |
1288 | hash_failure_debug(ea, access, vsid, trap, ssize, | |
d8139ebf AK |
1289 | mm->context.user_psize, |
1290 | mm->context.user_psize, | |
1291 | pte_val(*ptep)); | |
0ac52dd7 | 1292 | out_exit: |
3c726f8d BH |
1293 | local_irq_restore(flags); |
1294 | } | |
1295 | ||
f6ab0b92 BH |
1296 | /* WARNING: This is called from hash_low_64.S, if you change this prototype, |
1297 | * do not forget to update the assembly call site ! | |
1298 | */ | |
5524a27d | 1299 | void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize, |
aefa5688 | 1300 | unsigned long flags) |
3c726f8d BH |
1301 | { |
1302 | unsigned long hash, index, shift, hidx, slot; | |
aefa5688 | 1303 | int local = flags & HPTE_LOCAL_UPDATE; |
3c726f8d | 1304 | |
5524a27d AK |
1305 | DBG_LOW("flush_hash_page(vpn=%016lx)\n", vpn); |
1306 | pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) { | |
1307 | hash = hpt_hash(vpn, shift, ssize); | |
3c726f8d BH |
1308 | hidx = __rpte_to_hidx(pte, index); |
1309 | if (hidx & _PTEIDX_SECONDARY) | |
1310 | hash = ~hash; | |
1311 | slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; | |
1312 | slot += hidx & _PTEIDX_GROUP_IX; | |
5c339919 | 1313 | DBG_LOW(" sub %ld: hash=%lx, hidx=%lx\n", index, slot, hidx); |
db3d8534 AK |
1314 | /* |
1315 | * We use same base page size and actual psize, because we don't | |
1316 | * use these functions for hugepage | |
1317 | */ | |
1318 | ppc_md.hpte_invalidate(slot, vpn, psize, psize, ssize, local); | |
3c726f8d | 1319 | } pte_iterate_hashed_end(); |
bc2a9408 MN |
1320 | |
1321 | #ifdef CONFIG_PPC_TRANSACTIONAL_MEM | |
1322 | /* Transactions are not aborted by tlbiel, only tlbie. | |
1323 | * Without, syncing a page back to a block device w/ PIO could pick up | |
1324 | * transactional data (bad!) so we force an abort here. Before the | |
1325 | * sync the page will be made read-only, which will flush_hash_page. | |
1326 | * BIG ISSUE here: if the kernel uses a page from userspace without | |
1327 | * unmapping it first, it may see the speculated version. | |
1328 | */ | |
1329 | if (local && cpu_has_feature(CPU_FTR_TM) && | |
c2fd22df | 1330 | current->thread.regs && |
bc2a9408 MN |
1331 | MSR_TM_ACTIVE(current->thread.regs->msr)) { |
1332 | tm_enable(); | |
1333 | tm_abort(TM_CAUSE_TLBI); | |
1334 | } | |
1335 | #endif | |
1da177e4 LT |
1336 | } |
1337 | ||
f1581bf1 AK |
1338 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
1339 | void flush_hash_hugepage(unsigned long vsid, unsigned long addr, | |
aefa5688 AK |
1340 | pmd_t *pmdp, unsigned int psize, int ssize, |
1341 | unsigned long flags) | |
f1581bf1 AK |
1342 | { |
1343 | int i, max_hpte_count, valid; | |
1344 | unsigned long s_addr; | |
1345 | unsigned char *hpte_slot_array; | |
1346 | unsigned long hidx, shift, vpn, hash, slot; | |
aefa5688 | 1347 | int local = flags & HPTE_LOCAL_UPDATE; |
f1581bf1 AK |
1348 | |
1349 | s_addr = addr & HPAGE_PMD_MASK; | |
1350 | hpte_slot_array = get_hpte_slot_array(pmdp); | |
1351 | /* | |
1352 | * IF we try to do a HUGE PTE update after a withdraw is done. | |
1353 | * we will find the below NULL. This happens when we do | |
1354 | * split_huge_page_pmd | |
1355 | */ | |
1356 | if (!hpte_slot_array) | |
1357 | return; | |
1358 | ||
d557b098 AK |
1359 | if (ppc_md.hugepage_invalidate) { |
1360 | ppc_md.hugepage_invalidate(vsid, s_addr, hpte_slot_array, | |
1361 | psize, ssize, local); | |
1362 | goto tm_abort; | |
1363 | } | |
f1581bf1 AK |
1364 | /* |
1365 | * No bluk hpte removal support, invalidate each entry | |
1366 | */ | |
1367 | shift = mmu_psize_defs[psize].shift; | |
1368 | max_hpte_count = HPAGE_PMD_SIZE >> shift; | |
1369 | for (i = 0; i < max_hpte_count; i++) { | |
1370 | /* | |
1371 | * 8 bits per each hpte entries | |
1372 | * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit] | |
1373 | */ | |
1374 | valid = hpte_valid(hpte_slot_array, i); | |
1375 | if (!valid) | |
1376 | continue; | |
1377 | hidx = hpte_hash_index(hpte_slot_array, i); | |
1378 | ||
1379 | /* get the vpn */ | |
1380 | addr = s_addr + (i * (1ul << shift)); | |
1381 | vpn = hpt_vpn(addr, vsid, ssize); | |
1382 | hash = hpt_hash(vpn, shift, ssize); | |
1383 | if (hidx & _PTEIDX_SECONDARY) | |
1384 | hash = ~hash; | |
1385 | ||
1386 | slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; | |
1387 | slot += hidx & _PTEIDX_GROUP_IX; | |
1388 | ppc_md.hpte_invalidate(slot, vpn, psize, | |
d557b098 AK |
1389 | MMU_PAGE_16M, ssize, local); |
1390 | } | |
1391 | tm_abort: | |
1392 | #ifdef CONFIG_PPC_TRANSACTIONAL_MEM | |
1393 | /* Transactions are not aborted by tlbiel, only tlbie. | |
1394 | * Without, syncing a page back to a block device w/ PIO could pick up | |
1395 | * transactional data (bad!) so we force an abort here. Before the | |
1396 | * sync the page will be made read-only, which will flush_hash_page. | |
1397 | * BIG ISSUE here: if the kernel uses a page from userspace without | |
1398 | * unmapping it first, it may see the speculated version. | |
1399 | */ | |
1400 | if (local && cpu_has_feature(CPU_FTR_TM) && | |
1401 | current->thread.regs && | |
1402 | MSR_TM_ACTIVE(current->thread.regs->msr)) { | |
1403 | tm_enable(); | |
1404 | tm_abort(TM_CAUSE_TLBI); | |
f1581bf1 | 1405 | } |
d557b098 | 1406 | #endif |
2e826695 | 1407 | return; |
f1581bf1 AK |
1408 | } |
1409 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1410 | ||
61b1a942 | 1411 | void flush_hash_range(unsigned long number, int local) |
1da177e4 | 1412 | { |
3c726f8d | 1413 | if (ppc_md.flush_hash_range) |
61b1a942 | 1414 | ppc_md.flush_hash_range(number, local); |
3c726f8d | 1415 | else { |
1da177e4 | 1416 | int i; |
61b1a942 | 1417 | struct ppc64_tlb_batch *batch = |
69111bac | 1418 | this_cpu_ptr(&ppc64_tlb_batch); |
1da177e4 LT |
1419 | |
1420 | for (i = 0; i < number; i++) | |
5524a27d | 1421 | flush_hash_page(batch->vpn[i], batch->pte[i], |
1189be65 | 1422 | batch->psize, batch->ssize, local); |
1da177e4 LT |
1423 | } |
1424 | } | |
1425 | ||
1da177e4 LT |
1426 | /* |
1427 | * low_hash_fault is called when we the low level hash code failed | |
1428 | * to instert a PTE due to an hypervisor error | |
1429 | */ | |
fa28237c | 1430 | void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc) |
1da177e4 | 1431 | { |
ba12eede LZ |
1432 | enum ctx_state prev_state = exception_enter(); |
1433 | ||
1da177e4 | 1434 | if (user_mode(regs)) { |
fa28237c PM |
1435 | #ifdef CONFIG_PPC_SUBPAGE_PROT |
1436 | if (rc == -2) | |
1437 | _exception(SIGSEGV, regs, SEGV_ACCERR, address); | |
1438 | else | |
1439 | #endif | |
1440 | _exception(SIGBUS, regs, BUS_ADRERR, address); | |
1441 | } else | |
1442 | bad_page_fault(regs, address, SIGBUS); | |
ba12eede LZ |
1443 | |
1444 | exception_exit(prev_state); | |
1da177e4 | 1445 | } |
370a908d | 1446 | |
b170bd3d LZ |
1447 | long hpte_insert_repeating(unsigned long hash, unsigned long vpn, |
1448 | unsigned long pa, unsigned long rflags, | |
1449 | unsigned long vflags, int psize, int ssize) | |
1450 | { | |
1451 | unsigned long hpte_group; | |
1452 | long slot; | |
1453 | ||
1454 | repeat: | |
1455 | hpte_group = ((hash & htab_hash_mask) * | |
1456 | HPTES_PER_GROUP) & ~0x7UL; | |
1457 | ||
1458 | /* Insert into the hash table, primary slot */ | |
1459 | slot = ppc_md.hpte_insert(hpte_group, vpn, pa, rflags, vflags, | |
b1022fbd | 1460 | psize, psize, ssize); |
b170bd3d LZ |
1461 | |
1462 | /* Primary is full, try the secondary */ | |
1463 | if (unlikely(slot == -1)) { | |
1464 | hpte_group = ((~hash & htab_hash_mask) * | |
1465 | HPTES_PER_GROUP) & ~0x7UL; | |
1466 | slot = ppc_md.hpte_insert(hpte_group, vpn, pa, rflags, | |
1467 | vflags | HPTE_V_SECONDARY, | |
b1022fbd | 1468 | psize, psize, ssize); |
b170bd3d LZ |
1469 | if (slot == -1) { |
1470 | if (mftb() & 0x1) | |
1471 | hpte_group = ((hash & htab_hash_mask) * | |
1472 | HPTES_PER_GROUP)&~0x7UL; | |
1473 | ||
1474 | ppc_md.hpte_remove(hpte_group); | |
1475 | goto repeat; | |
1476 | } | |
1477 | } | |
1478 | ||
1479 | return slot; | |
1480 | } | |
1481 | ||
370a908d BH |
1482 | #ifdef CONFIG_DEBUG_PAGEALLOC |
1483 | static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi) | |
1484 | { | |
016af59f | 1485 | unsigned long hash; |
1189be65 | 1486 | unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize); |
5524a27d | 1487 | unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize); |
09f3f326 | 1488 | unsigned long mode = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL)); |
016af59f | 1489 | long ret; |
370a908d | 1490 | |
5524a27d | 1491 | hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize); |
370a908d | 1492 | |
c60ac569 AK |
1493 | /* Don't create HPTE entries for bad address */ |
1494 | if (!vsid) | |
1495 | return; | |
016af59f LZ |
1496 | |
1497 | ret = hpte_insert_repeating(hash, vpn, __pa(vaddr), mode, | |
1498 | HPTE_V_BOLTED, | |
1499 | mmu_linear_psize, mmu_kernel_ssize); | |
1500 | ||
370a908d BH |
1501 | BUG_ON (ret < 0); |
1502 | spin_lock(&linear_map_hash_lock); | |
1503 | BUG_ON(linear_map_hash_slots[lmi] & 0x80); | |
1504 | linear_map_hash_slots[lmi] = ret | 0x80; | |
1505 | spin_unlock(&linear_map_hash_lock); | |
1506 | } | |
1507 | ||
1508 | static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi) | |
1509 | { | |
1189be65 PM |
1510 | unsigned long hash, hidx, slot; |
1511 | unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize); | |
5524a27d | 1512 | unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize); |
370a908d | 1513 | |
5524a27d | 1514 | hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize); |
370a908d BH |
1515 | spin_lock(&linear_map_hash_lock); |
1516 | BUG_ON(!(linear_map_hash_slots[lmi] & 0x80)); | |
1517 | hidx = linear_map_hash_slots[lmi] & 0x7f; | |
1518 | linear_map_hash_slots[lmi] = 0; | |
1519 | spin_unlock(&linear_map_hash_lock); | |
1520 | if (hidx & _PTEIDX_SECONDARY) | |
1521 | hash = ~hash; | |
1522 | slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; | |
1523 | slot += hidx & _PTEIDX_GROUP_IX; | |
db3d8534 AK |
1524 | ppc_md.hpte_invalidate(slot, vpn, mmu_linear_psize, mmu_linear_psize, |
1525 | mmu_kernel_ssize, 0); | |
370a908d BH |
1526 | } |
1527 | ||
031bc574 | 1528 | void __kernel_map_pages(struct page *page, int numpages, int enable) |
370a908d BH |
1529 | { |
1530 | unsigned long flags, vaddr, lmi; | |
1531 | int i; | |
1532 | ||
1533 | local_irq_save(flags); | |
1534 | for (i = 0; i < numpages; i++, page++) { | |
1535 | vaddr = (unsigned long)page_address(page); | |
1536 | lmi = __pa(vaddr) >> PAGE_SHIFT; | |
1537 | if (lmi >= linear_map_hash_count) | |
1538 | continue; | |
1539 | if (enable) | |
1540 | kernel_map_linear_page(vaddr, lmi); | |
1541 | else | |
1542 | kernel_unmap_linear_page(vaddr, lmi); | |
1543 | } | |
1544 | local_irq_restore(flags); | |
1545 | } | |
1546 | #endif /* CONFIG_DEBUG_PAGEALLOC */ | |
cd3db0c4 BH |
1547 | |
1548 | void setup_initial_memory_limit(phys_addr_t first_memblock_base, | |
1549 | phys_addr_t first_memblock_size) | |
1550 | { | |
1551 | /* We don't currently support the first MEMBLOCK not mapping 0 | |
1552 | * physical on those processors | |
1553 | */ | |
1554 | BUG_ON(first_memblock_base != 0); | |
1555 | ||
1556 | /* On LPAR systems, the first entry is our RMA region, | |
1557 | * non-LPAR 64-bit hash MMU systems don't have a limitation | |
1558 | * on real mode access, but using the first entry works well | |
1559 | * enough. We also clamp it to 1G to avoid some funky things | |
1560 | * such as RTAS bugs etc... | |
1561 | */ | |
1562 | ppc64_rma_size = min_t(u64, first_memblock_size, 0x40000000); | |
1563 | ||
1564 | /* Finally limit subsequent allocations */ | |
1565 | memblock_set_current_limit(ppc64_rma_size); | |
1566 | } |