]>
Commit | Line | Data |
---|---|---|
11a6f6ab AK |
1 | #ifndef _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ |
2 | #define _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ | |
8d2169e8 DG |
3 | /* |
4 | * PowerPC64 memory management structures | |
5 | * | |
6 | * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com> | |
7 | * PPC64 rework. | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU General Public License | |
11 | * as published by the Free Software Foundation; either version | |
12 | * 2 of the License, or (at your option) any later version. | |
13 | */ | |
14 | ||
8d2169e8 | 15 | #include <asm/page.h> |
891121e6 | 16 | #include <asm/bug.h> |
ec0c464c | 17 | #include <asm/asm-const.h> |
8d2169e8 | 18 | |
78f1dbde AK |
19 | /* |
20 | * This is necessary to get the definition of PGTABLE_RANGE which we | |
21 | * need for various slices related matters. Note that this isn't the | |
22 | * complete pgtable.h but only a portion of it. | |
23 | */ | |
3dfcb315 | 24 | #include <asm/book3s/64/pgtable.h> |
cf9427b8 | 25 | #include <asm/bug.h> |
dad6f37c | 26 | #include <asm/processor.h> |
b92a226e | 27 | #include <asm/cpu_has_feature.h> |
78f1dbde | 28 | |
8d2169e8 DG |
29 | /* |
30 | * SLB | |
31 | */ | |
32 | ||
85376e2a | 33 | #define SLB_NUM_BOLTED 2 |
8d2169e8 | 34 | #define SLB_CACHE_ENTRIES 8 |
46db2f86 | 35 | #define SLB_MIN_SIZE 32 |
8d2169e8 DG |
36 | |
37 | /* Bits in the SLB ESID word */ | |
38 | #define SLB_ESID_V ASM_CONST(0x0000000008000000) /* valid */ | |
39 | ||
40 | /* Bits in the SLB VSID word */ | |
41 | #define SLB_VSID_SHIFT 12 | |
e6f81a92 | 42 | #define SLB_VSID_SHIFT_256M SLB_VSID_SHIFT |
1189be65 PM |
43 | #define SLB_VSID_SHIFT_1T 24 |
44 | #define SLB_VSID_SSIZE_SHIFT 62 | |
8d2169e8 DG |
45 | #define SLB_VSID_B ASM_CONST(0xc000000000000000) |
46 | #define SLB_VSID_B_256M ASM_CONST(0x0000000000000000) | |
47 | #define SLB_VSID_B_1T ASM_CONST(0x4000000000000000) | |
48 | #define SLB_VSID_KS ASM_CONST(0x0000000000000800) | |
49 | #define SLB_VSID_KP ASM_CONST(0x0000000000000400) | |
50 | #define SLB_VSID_N ASM_CONST(0x0000000000000200) /* no-execute */ | |
51 | #define SLB_VSID_L ASM_CONST(0x0000000000000100) | |
52 | #define SLB_VSID_C ASM_CONST(0x0000000000000080) /* class */ | |
53 | #define SLB_VSID_LP ASM_CONST(0x0000000000000030) | |
54 | #define SLB_VSID_LP_00 ASM_CONST(0x0000000000000000) | |
55 | #define SLB_VSID_LP_01 ASM_CONST(0x0000000000000010) | |
56 | #define SLB_VSID_LP_10 ASM_CONST(0x0000000000000020) | |
57 | #define SLB_VSID_LP_11 ASM_CONST(0x0000000000000030) | |
58 | #define SLB_VSID_LLP (SLB_VSID_L|SLB_VSID_LP) | |
59 | ||
60 | #define SLB_VSID_KERNEL (SLB_VSID_KP) | |
61 | #define SLB_VSID_USER (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C) | |
62 | ||
63 | #define SLBIE_C (0x08000000) | |
1189be65 | 64 | #define SLBIE_SSIZE_SHIFT 25 |
8d2169e8 DG |
65 | |
66 | /* | |
67 | * Hash table | |
68 | */ | |
69 | ||
70 | #define HPTES_PER_GROUP 8 | |
71 | ||
2454c7e9 | 72 | #define HPTE_V_SSIZE_SHIFT 62 |
8d2169e8 | 73 | #define HPTE_V_AVPN_SHIFT 7 |
6b243fcf | 74 | #define HPTE_V_COMMON_BITS ASM_CONST(0x000fffffffffffff) |
2454c7e9 | 75 | #define HPTE_V_AVPN ASM_CONST(0x3fffffffffffff80) |
6b243fcf | 76 | #define HPTE_V_AVPN_3_0 ASM_CONST(0x000fffffffffff80) |
8d2169e8 | 77 | #define HPTE_V_AVPN_VAL(x) (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT) |
91bbbe22 | 78 | #define HPTE_V_COMPARE(x,y) (!(((x) ^ (y)) & 0xffffffffffffff80UL)) |
8d2169e8 DG |
79 | #define HPTE_V_BOLTED ASM_CONST(0x0000000000000010) |
80 | #define HPTE_V_LOCK ASM_CONST(0x0000000000000008) | |
81 | #define HPTE_V_LARGE ASM_CONST(0x0000000000000004) | |
82 | #define HPTE_V_SECONDARY ASM_CONST(0x0000000000000002) | |
83 | #define HPTE_V_VALID ASM_CONST(0x0000000000000001) | |
84 | ||
50de596d | 85 | /* |
6b243fcf | 86 | * ISA 3.0 has a different HPTE format. |
50de596d AK |
87 | */ |
88 | #define HPTE_R_3_0_SSIZE_SHIFT 58 | |
6b243fcf | 89 | #define HPTE_R_3_0_SSIZE_MASK (3ull << HPTE_R_3_0_SSIZE_SHIFT) |
8d2169e8 DG |
90 | #define HPTE_R_PP0 ASM_CONST(0x8000000000000000) |
91 | #define HPTE_R_TS ASM_CONST(0x4000000000000000) | |
de56a948 | 92 | #define HPTE_R_KEY_HI ASM_CONST(0x3000000000000000) |
a6590ca5 RP |
93 | #define HPTE_R_KEY_BIT0 ASM_CONST(0x2000000000000000) |
94 | #define HPTE_R_KEY_BIT1 ASM_CONST(0x1000000000000000) | |
8d2169e8 | 95 | #define HPTE_R_RPN_SHIFT 12 |
de56a948 | 96 | #define HPTE_R_RPN ASM_CONST(0x0ffffffffffff000) |
6b243fcf | 97 | #define HPTE_R_RPN_3_0 ASM_CONST(0x01fffffffffff000) |
8d2169e8 | 98 | #define HPTE_R_PP ASM_CONST(0x0000000000000003) |
8550e2fa | 99 | #define HPTE_R_PPP ASM_CONST(0x8000000000000003) |
8d2169e8 | 100 | #define HPTE_R_N ASM_CONST(0x0000000000000004) |
de56a948 PM |
101 | #define HPTE_R_G ASM_CONST(0x0000000000000008) |
102 | #define HPTE_R_M ASM_CONST(0x0000000000000010) | |
103 | #define HPTE_R_I ASM_CONST(0x0000000000000020) | |
104 | #define HPTE_R_W ASM_CONST(0x0000000000000040) | |
105 | #define HPTE_R_WIMG ASM_CONST(0x0000000000000078) | |
8d2169e8 DG |
106 | #define HPTE_R_C ASM_CONST(0x0000000000000080) |
107 | #define HPTE_R_R ASM_CONST(0x0000000000000100) | |
de56a948 | 108 | #define HPTE_R_KEY_LO ASM_CONST(0x0000000000000e00) |
a6590ca5 RP |
109 | #define HPTE_R_KEY_BIT2 ASM_CONST(0x0000000000000800) |
110 | #define HPTE_R_KEY_BIT3 ASM_CONST(0x0000000000000400) | |
111 | #define HPTE_R_KEY_BIT4 ASM_CONST(0x0000000000000200) | |
d182b8fd | 112 | #define HPTE_R_KEY (HPTE_R_KEY_LO | HPTE_R_KEY_HI) |
8d2169e8 | 113 | |
b7abc5c5 SS |
114 | #define HPTE_V_1TB_SEG ASM_CONST(0x4000000000000000) |
115 | #define HPTE_V_VRMA_MASK ASM_CONST(0x4001ffffff000000) | |
116 | ||
8d2169e8 | 117 | /* Values for PP (assumes Ks=0, Kp=1) */ |
8d2169e8 DG |
118 | #define PP_RWXX 0 /* Supervisor read/write, User none */ |
119 | #define PP_RWRX 1 /* Supervisor read/write, User read */ | |
120 | #define PP_RWRW 2 /* Supervisor read/write, User read/write */ | |
121 | #define PP_RXRX 3 /* Supervisor read, User read */ | |
697d3899 | 122 | #define PP_RXXX (HPTE_R_PP0 | 2) /* Supervisor read, user none */ |
8d2169e8 | 123 | |
b4072df4 PM |
124 | /* Fields for tlbiel instruction in architecture 2.06 */ |
125 | #define TLBIEL_INVAL_SEL_MASK 0xc00 /* invalidation selector */ | |
126 | #define TLBIEL_INVAL_PAGE 0x000 /* invalidate a single page */ | |
127 | #define TLBIEL_INVAL_SET_LPID 0x800 /* invalidate a set for current LPID */ | |
128 | #define TLBIEL_INVAL_SET 0xc00 /* invalidate a set for all LPIDs */ | |
129 | #define TLBIEL_INVAL_SET_MASK 0xfff000 /* set number to inval. */ | |
130 | #define TLBIEL_INVAL_SET_SHIFT 12 | |
131 | ||
132 | #define POWER7_TLB_SETS 128 /* # sets in POWER7 TLB */ | |
45706bb5 | 133 | #define POWER8_TLB_SETS 512 /* # sets in POWER8 TLB */ |
c3ab300e | 134 | #define POWER9_TLB_SETS_HASH 256 /* # sets in POWER9 TLB Hash mode */ |
1a472c9d | 135 | #define POWER9_TLB_SETS_RADIX 128 /* # sets in POWER9 TLB Radix mode */ |
b4072df4 | 136 | |
8d2169e8 DG |
137 | #ifndef __ASSEMBLY__ |
138 | ||
7025776e BH |
139 | struct mmu_hash_ops { |
140 | void (*hpte_invalidate)(unsigned long slot, | |
141 | unsigned long vpn, | |
142 | int bpsize, int apsize, | |
143 | int ssize, int local); | |
144 | long (*hpte_updatepp)(unsigned long slot, | |
145 | unsigned long newpp, | |
146 | unsigned long vpn, | |
147 | int bpsize, int apsize, | |
148 | int ssize, unsigned long flags); | |
149 | void (*hpte_updateboltedpp)(unsigned long newpp, | |
150 | unsigned long ea, | |
151 | int psize, int ssize); | |
152 | long (*hpte_insert)(unsigned long hpte_group, | |
153 | unsigned long vpn, | |
154 | unsigned long prpn, | |
155 | unsigned long rflags, | |
156 | unsigned long vflags, | |
157 | int psize, int apsize, | |
158 | int ssize); | |
159 | long (*hpte_remove)(unsigned long hpte_group); | |
160 | int (*hpte_removebolted)(unsigned long ea, | |
161 | int psize, int ssize); | |
162 | void (*flush_hash_range)(unsigned long number, int local); | |
163 | void (*hugepage_invalidate)(unsigned long vsid, | |
164 | unsigned long addr, | |
165 | unsigned char *hpte_slot_array, | |
166 | int psize, int ssize, int local); | |
dbcf929c | 167 | int (*resize_hpt)(unsigned long shift); |
7025776e BH |
168 | /* |
169 | * Special for kexec. | |
170 | * To be called in real mode with interrupts disabled. No locks are | |
171 | * taken as such, concurrent access on pre POWER5 hardware could result | |
172 | * in a deadlock. | |
173 | * The linear mapping is destroyed as well. | |
174 | */ | |
175 | void (*hpte_clear_all)(void); | |
176 | }; | |
177 | extern struct mmu_hash_ops mmu_hash_ops; | |
178 | ||
8e561e7e | 179 | struct hash_pte { |
12f04f2b AB |
180 | __be64 v; |
181 | __be64 r; | |
8e561e7e | 182 | }; |
8d2169e8 | 183 | |
8e561e7e | 184 | extern struct hash_pte *htab_address; |
8d2169e8 DG |
185 | extern unsigned long htab_size_bytes; |
186 | extern unsigned long htab_hash_mask; | |
187 | ||
cf9427b8 AK |
188 | |
189 | static inline int shift_to_mmu_psize(unsigned int shift) | |
190 | { | |
191 | int psize; | |
192 | ||
193 | for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) | |
194 | if (mmu_psize_defs[psize].shift == shift) | |
195 | return psize; | |
196 | return -1; | |
197 | } | |
198 | ||
199 | static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize) | |
200 | { | |
201 | if (mmu_psize_defs[mmu_psize].shift) | |
202 | return mmu_psize_defs[mmu_psize].shift; | |
203 | BUG(); | |
204 | } | |
8d2169e8 | 205 | |
138ee7ee AK |
206 | static inline unsigned long get_sllp_encoding(int psize) |
207 | { | |
208 | unsigned long sllp; | |
209 | ||
210 | sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) | | |
211 | ((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4); | |
212 | return sllp; | |
213 | } | |
214 | ||
8d2169e8 DG |
215 | #endif /* __ASSEMBLY__ */ |
216 | ||
2454c7e9 PM |
217 | /* |
218 | * Segment sizes. | |
219 | * These are the values used by hardware in the B field of | |
220 | * SLB entries and the first dword of MMU hashtable entries. | |
221 | * The B field is 2 bits; the values 2 and 3 are unused and reserved. | |
222 | */ | |
223 | #define MMU_SEGSIZE_256M 0 | |
224 | #define MMU_SEGSIZE_1T 1 | |
225 | ||
5524a27d AK |
226 | /* |
227 | * encode page number shift. | |
228 | * in order to fit the 78 bit va in a 64 bit variable we shift the va by | |
229 | * 12 bits. This enable us to address upto 76 bit va. | |
230 | * For hpt hash from a va we can ignore the page size bits of va and for | |
231 | * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure | |
232 | * we work in all cases including 4k page size. | |
233 | */ | |
234 | #define VPN_SHIFT 12 | |
1189be65 | 235 | |
b1022fbd AK |
236 | /* |
237 | * HPTE Large Page (LP) details | |
238 | */ | |
239 | #define LP_SHIFT 12 | |
240 | #define LP_BITS 8 | |
241 | #define LP_MASK(i) ((0xFF >> (i)) << LP_SHIFT) | |
242 | ||
8d2169e8 DG |
243 | #ifndef __ASSEMBLY__ |
244 | ||
73d16a6e IM |
245 | static inline int slb_vsid_shift(int ssize) |
246 | { | |
247 | if (ssize == MMU_SEGSIZE_256M) | |
248 | return SLB_VSID_SHIFT; | |
249 | return SLB_VSID_SHIFT_1T; | |
250 | } | |
251 | ||
5524a27d AK |
252 | static inline int segment_shift(int ssize) |
253 | { | |
254 | if (ssize == MMU_SEGSIZE_256M) | |
255 | return SID_SHIFT; | |
256 | return SID_SHIFT_1T; | |
257 | } | |
258 | ||
0eeede0c PM |
259 | /* |
260 | * This array is indexed by the LP field of the HPTE second dword. | |
261 | * Since this field may contain some RPN bits, some entries are | |
262 | * replicated so that we get the same value irrespective of RPN. | |
263 | * The top 4 bits are the page size index (MMU_PAGE_*) for the | |
264 | * actual page size, the bottom 4 bits are the base page size. | |
265 | */ | |
266 | extern u8 hpte_page_sizes[1 << LP_BITS]; | |
267 | ||
268 | static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l, | |
269 | bool is_base_size) | |
270 | { | |
271 | unsigned int i, lp; | |
272 | ||
273 | if (!(h & HPTE_V_LARGE)) | |
274 | return 1ul << 12; | |
275 | ||
276 | /* Look at the 8 bit LP value */ | |
277 | lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1); | |
278 | i = hpte_page_sizes[lp]; | |
279 | if (!i) | |
280 | return 0; | |
281 | if (!is_base_size) | |
282 | i >>= 4; | |
283 | return 1ul << mmu_psize_defs[i & 0xf].shift; | |
284 | } | |
285 | ||
286 | static inline unsigned long hpte_page_size(unsigned long h, unsigned long l) | |
287 | { | |
288 | return __hpte_page_size(h, l, 0); | |
289 | } | |
290 | ||
291 | static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l) | |
292 | { | |
293 | return __hpte_page_size(h, l, 1); | |
294 | } | |
295 | ||
8d2169e8 | 296 | /* |
1189be65 | 297 | * The current system page and segment sizes |
8d2169e8 | 298 | */ |
1189be65 PM |
299 | extern int mmu_kernel_ssize; |
300 | extern int mmu_highuser_ssize; | |
584f8b71 | 301 | extern u16 mmu_slb_size; |
572fb578 | 302 | extern unsigned long tce_alloc_start, tce_alloc_end; |
8d2169e8 DG |
303 | |
304 | /* | |
305 | * If the processor supports 64k normal pages but not 64k cache | |
306 | * inhibited pages, we have to be prepared to switch processes | |
307 | * to use 4k pages when they create cache-inhibited mappings. | |
308 | * If this is the case, mmu_ci_restrictions will be set to 1. | |
309 | */ | |
310 | extern int mmu_ci_restrictions; | |
311 | ||
5524a27d AK |
312 | /* |
313 | * This computes the AVPN and B fields of the first dword of a HPTE, | |
314 | * for use when we want to match an existing PTE. The bottom 7 bits | |
315 | * of the returned value are zero. | |
316 | */ | |
317 | static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize, | |
318 | int ssize) | |
319 | { | |
320 | unsigned long v; | |
321 | /* | |
322 | * The AVA field omits the low-order 23 bits of the 78 bits VA. | |
323 | * These bits are not needed in the PTE, because the | |
324 | * low-order b of these bits are part of the byte offset | |
325 | * into the virtual page and, if b < 23, the high-order | |
326 | * 23-b of these bits are always used in selecting the | |
327 | * PTEGs to be searched | |
328 | */ | |
329 | v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm); | |
330 | v <<= HPTE_V_AVPN_SHIFT; | |
6b243fcf | 331 | v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT; |
5524a27d AK |
332 | return v; |
333 | } | |
334 | ||
6b243fcf PM |
335 | /* |
336 | * ISA v3.0 defines a new HPTE format, which differs from the old | |
337 | * format in having smaller AVPN and ARPN fields, and the B field | |
338 | * in the second dword instead of the first. | |
339 | */ | |
340 | static inline unsigned long hpte_old_to_new_v(unsigned long v) | |
341 | { | |
342 | /* trim AVPN, drop B */ | |
343 | return v & HPTE_V_COMMON_BITS; | |
344 | } | |
345 | ||
346 | static inline unsigned long hpte_old_to_new_r(unsigned long v, unsigned long r) | |
347 | { | |
348 | /* move B field from 1st to 2nd dword, trim ARPN */ | |
349 | return (r & ~HPTE_R_3_0_SSIZE_MASK) | | |
350 | (((v) >> HPTE_V_SSIZE_SHIFT) << HPTE_R_3_0_SSIZE_SHIFT); | |
351 | } | |
352 | ||
353 | static inline unsigned long hpte_new_to_old_v(unsigned long v, unsigned long r) | |
354 | { | |
355 | /* insert B field */ | |
356 | return (v & HPTE_V_COMMON_BITS) | | |
357 | ((r & HPTE_R_3_0_SSIZE_MASK) << | |
358 | (HPTE_V_SSIZE_SHIFT - HPTE_R_3_0_SSIZE_SHIFT)); | |
359 | } | |
360 | ||
361 | static inline unsigned long hpte_new_to_old_r(unsigned long r) | |
362 | { | |
363 | /* clear out B field */ | |
364 | return r & ~HPTE_R_3_0_SSIZE_MASK; | |
365 | } | |
366 | ||
a833280b AK |
367 | static inline unsigned long hpte_get_old_v(struct hash_pte *hptep) |
368 | { | |
369 | unsigned long hpte_v; | |
370 | ||
371 | hpte_v = be64_to_cpu(hptep->v); | |
372 | if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
373 | hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r)); | |
374 | return hpte_v; | |
375 | } | |
376 | ||
8d2169e8 DG |
377 | /* |
378 | * This function sets the AVPN and L fields of the HPTE appropriately | |
b1022fbd | 379 | * using the base page size and actual page size. |
8d2169e8 | 380 | */ |
b1022fbd AK |
381 | static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize, |
382 | int actual_psize, int ssize) | |
8d2169e8 | 383 | { |
1189be65 | 384 | unsigned long v; |
b1022fbd AK |
385 | v = hpte_encode_avpn(vpn, base_psize, ssize); |
386 | if (actual_psize != MMU_PAGE_4K) | |
8d2169e8 DG |
387 | v |= HPTE_V_LARGE; |
388 | return v; | |
389 | } | |
390 | ||
391 | /* | |
392 | * This function sets the ARPN, and LP fields of the HPTE appropriately | |
393 | * for the page size. We assume the pa is already "clean" that is properly | |
394 | * aligned for the requested page size | |
395 | */ | |
b1022fbd | 396 | static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize, |
6b243fcf | 397 | int actual_psize) |
8d2169e8 | 398 | { |
8d2169e8 | 399 | /* A 4K page needs no special encoding */ |
b1022fbd | 400 | if (actual_psize == MMU_PAGE_4K) |
8d2169e8 DG |
401 | return pa & HPTE_R_RPN; |
402 | else { | |
b1022fbd AK |
403 | unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize]; |
404 | unsigned int shift = mmu_psize_defs[actual_psize].shift; | |
405 | return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT); | |
8d2169e8 | 406 | } |
8d2169e8 DG |
407 | } |
408 | ||
409 | /* | |
5524a27d | 410 | * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size. |
8d2169e8 | 411 | */ |
5524a27d AK |
412 | static inline unsigned long hpt_vpn(unsigned long ea, |
413 | unsigned long vsid, int ssize) | |
1189be65 | 414 | { |
5524a27d AK |
415 | unsigned long mask; |
416 | int s_shift = segment_shift(ssize); | |
417 | ||
418 | mask = (1ul << (s_shift - VPN_SHIFT)) - 1; | |
419 | return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask); | |
1189be65 | 420 | } |
8d2169e8 | 421 | |
1189be65 PM |
422 | /* |
423 | * This hashes a virtual address | |
424 | */ | |
5524a27d AK |
425 | static inline unsigned long hpt_hash(unsigned long vpn, |
426 | unsigned int shift, int ssize) | |
8d2169e8 | 427 | { |
59248aec | 428 | unsigned long mask; |
1189be65 PM |
429 | unsigned long hash, vsid; |
430 | ||
5524a27d | 431 | /* VPN_SHIFT can be atmost 12 */ |
1189be65 | 432 | if (ssize == MMU_SEGSIZE_256M) { |
5524a27d AK |
433 | mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1; |
434 | hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^ | |
435 | ((vpn & mask) >> (shift - VPN_SHIFT)); | |
1189be65 | 436 | } else { |
5524a27d AK |
437 | mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1; |
438 | vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT); | |
439 | hash = vsid ^ (vsid << 25) ^ | |
440 | ((vpn & mask) >> (shift - VPN_SHIFT)) ; | |
1189be65 PM |
441 | } |
442 | return hash & 0x7fffffffffUL; | |
8d2169e8 DG |
443 | } |
444 | ||
aefa5688 AK |
445 | #define HPTE_LOCAL_UPDATE 0x1 |
446 | #define HPTE_NOHPTE_UPDATE 0x2 | |
447 | ||
8d2169e8 DG |
448 | extern int __hash_page_4K(unsigned long ea, unsigned long access, |
449 | unsigned long vsid, pte_t *ptep, unsigned long trap, | |
aefa5688 | 450 | unsigned long flags, int ssize, int subpage_prot); |
8d2169e8 DG |
451 | extern int __hash_page_64K(unsigned long ea, unsigned long access, |
452 | unsigned long vsid, pte_t *ptep, unsigned long trap, | |
aefa5688 | 453 | unsigned long flags, int ssize); |
8d2169e8 | 454 | struct mm_struct; |
0895ecda | 455 | unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap); |
aefa5688 AK |
456 | extern int hash_page_mm(struct mm_struct *mm, unsigned long ea, |
457 | unsigned long access, unsigned long trap, | |
458 | unsigned long flags); | |
459 | extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap, | |
460 | unsigned long dsisr); | |
a4fe3ce7 | 461 | int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid, |
aefa5688 AK |
462 | pte_t *ptep, unsigned long trap, unsigned long flags, |
463 | int ssize, unsigned int shift, unsigned int mmu_psize); | |
6d492ecc AK |
464 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
465 | extern int __hash_page_thp(unsigned long ea, unsigned long access, | |
466 | unsigned long vsid, pmd_t *pmdp, unsigned long trap, | |
aefa5688 | 467 | unsigned long flags, int ssize, unsigned int psize); |
6d492ecc AK |
468 | #else |
469 | static inline int __hash_page_thp(unsigned long ea, unsigned long access, | |
470 | unsigned long vsid, pmd_t *pmdp, | |
aefa5688 | 471 | unsigned long trap, unsigned long flags, |
6d492ecc AK |
472 | int ssize, unsigned int psize) |
473 | { | |
474 | BUG(); | |
ff1e7683 | 475 | return -1; |
6d492ecc AK |
476 | } |
477 | #endif | |
4b8692c0 BH |
478 | extern void hash_failure_debug(unsigned long ea, unsigned long access, |
479 | unsigned long vsid, unsigned long trap, | |
d8139ebf AK |
480 | int ssize, int psize, int lpsize, |
481 | unsigned long pte); | |
8d2169e8 | 482 | extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend, |
bc033b63 | 483 | unsigned long pstart, unsigned long prot, |
1189be65 | 484 | int psize, int ssize); |
f6026df1 AB |
485 | int htab_remove_mapping(unsigned long vstart, unsigned long vend, |
486 | int psize, int ssize); | |
79cc38de | 487 | extern void pseries_add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages); |
fa28237c | 488 | extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr); |
8d2169e8 | 489 | |
6364e84e ME |
490 | #ifdef CONFIG_PPC_PSERIES |
491 | void hpte_init_pseries(void); | |
492 | #else | |
493 | static inline void hpte_init_pseries(void) { } | |
494 | #endif | |
495 | ||
8d2169e8 | 496 | extern void hpte_init_native(void); |
8d2169e8 | 497 | |
c6d15258 MS |
498 | struct slb_entry { |
499 | u64 esid; | |
500 | u64 vsid; | |
501 | }; | |
502 | ||
8d2169e8 DG |
503 | extern void slb_initialize(void); |
504 | extern void slb_flush_and_rebolt(void); | |
e7e81847 NP |
505 | void slb_flush_all_realmode(void); |
506 | void __slb_restore_bolted_realmode(void); | |
507 | void slb_restore_bolted_realmode(void); | |
c6d15258 MS |
508 | void slb_save_contents(struct slb_entry *slb_ptr); |
509 | void slb_dump_contents(struct slb_entry *slb_ptr); | |
8d2169e8 | 510 | |
67439b76 | 511 | extern void slb_vmalloc_update(void); |
46db2f86 | 512 | extern void slb_set_size(u16 size); |
8d2169e8 DG |
513 | #endif /* __ASSEMBLY__ */ |
514 | ||
515 | /* | |
f033d659 | 516 | * VSID allocation (256MB segment) |
8d2169e8 | 517 | * |
c60ac569 AK |
518 | * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated |
519 | * from mmu context id and effective segment id of the address. | |
8d2169e8 | 520 | * |
941711a3 AK |
521 | * For user processes max context id is limited to MAX_USER_CONTEXT. |
522 | ||
add2e1e5 | 523 | * For kernel space, we use context ids 1-4 to map addresses as below: |
c60ac569 | 524 | * NOTE: each context only support 64TB now. |
941711a3 AK |
525 | * 0x00001 - [ 0xc000000000000000 - 0xc0003fffffffffff ] |
526 | * 0x00002 - [ 0xd000000000000000 - 0xd0003fffffffffff ] | |
527 | * 0x00003 - [ 0xe000000000000000 - 0xe0003fffffffffff ] | |
528 | * 0x00004 - [ 0xf000000000000000 - 0xf0003fffffffffff ] | |
8d2169e8 DG |
529 | * |
530 | * The proto-VSIDs are then scrambled into real VSIDs with the | |
531 | * multiplicative hash: | |
532 | * | |
533 | * VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS | |
8d2169e8 | 534 | * |
f033d659 | 535 | * VSID_MULTIPLIER is prime, so in particular it is |
8d2169e8 DG |
536 | * co-prime to VSID_MODULUS, making this a 1:1 scrambling function. |
537 | * Because the modulus is 2^n-1 we can compute it efficiently without | |
c60ac569 AK |
538 | * a divide or extra multiply (see below). The scramble function gives |
539 | * robust scattering in the hash table (at least based on some initial | |
540 | * results). | |
8d2169e8 | 541 | * |
941711a3 AK |
542 | * We use VSID 0 to indicate an invalid VSID. The means we can't use context id |
543 | * 0, because a context id of 0 and an EA of 0 gives a proto-VSID of 0, which | |
544 | * will produce a VSID of 0. | |
8d2169e8 | 545 | * |
c60ac569 AK |
546 | * We also need to avoid the last segment of the last context, because that |
547 | * would give a protovsid of 0x1fffffffff. That will result in a VSID 0 | |
941711a3 | 548 | * because of the modulo operation in vsid scramble. |
8d2169e8 | 549 | */ |
8d2169e8 | 550 | |
e6f81a92 AK |
551 | /* |
552 | * Max Va bits we support as of now is 68 bits. We want 19 bit | |
553 | * context ID. | |
554 | * Restrictions: | |
555 | * GPU has restrictions of not able to access beyond 128TB | |
556 | * (47 bit effective address). We also cannot do more than 20bit PID. | |
557 | * For p4 and p5 which can only do 65 bit VA, we restrict our CONTEXT_BITS | |
558 | * to 16 bits (ie, we can only have 2^16 pids at the same time). | |
559 | */ | |
560 | #define VA_BITS 68 | |
e39d1a47 | 561 | #define CONTEXT_BITS 19 |
e6f81a92 AK |
562 | #define ESID_BITS (VA_BITS - (SID_SHIFT + CONTEXT_BITS)) |
563 | #define ESID_BITS_1T (VA_BITS - (SID_SHIFT_1T + CONTEXT_BITS)) | |
e39d1a47 | 564 | |
79270e0a AK |
565 | #define ESID_BITS_MASK ((1 << ESID_BITS) - 1) |
566 | #define ESID_BITS_1T_MASK ((1 << ESID_BITS_1T) - 1) | |
567 | ||
c60ac569 AK |
568 | /* |
569 | * 256MB segment | |
af81d787 | 570 | * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments |
941711a3 AK |
571 | * available for user + kernel mapping. VSID 0 is reserved as invalid, contexts |
572 | * 1-4 are used for kernel mapping. Each segment contains 2^28 bytes. Each | |
e6f81a92 | 573 | * context maps 2^49 bytes (512TB). |
941711a3 AK |
574 | * |
575 | * We also need to avoid the last segment of the last context, because that | |
576 | * would give a protovsid of 0x1fffffffff. That will result in a VSID 0 | |
577 | * because of the modulo operation in vsid scramble. | |
c60ac569 | 578 | */ |
941711a3 AK |
579 | #define MAX_USER_CONTEXT ((ASM_CONST(1) << CONTEXT_BITS) - 2) |
580 | #define MIN_USER_CONTEXT (5) | |
581 | ||
582 | /* Would be nice to use KERNEL_REGION_ID here */ | |
583 | #define KERNEL_REGION_CONTEXT_OFFSET (0xc - 1) | |
c60ac569 | 584 | |
e6f81a92 AK |
585 | /* |
586 | * For platforms that support on 65bit VA we limit the context bits | |
587 | */ | |
588 | #define MAX_USER_CONTEXT_65BIT_VA ((ASM_CONST(1) << (65 - (SID_SHIFT + ESID_BITS))) - 2) | |
589 | ||
048ee099 AK |
590 | /* |
591 | * This should be computed such that protovosid * vsid_mulitplier | |
e6f81a92 AK |
592 | * doesn't overflow 64 bits. The vsid_mutliplier should also be |
593 | * co-prime to vsid_modulus. We also need to make sure that number | |
594 | * of bits in multiplied result (dividend) is less than twice the number of | |
595 | * protovsid bits for our modulus optmization to work. | |
596 | * | |
597 | * The below table shows the current values used. | |
598 | * |-------+------------+----------------------+------------+-------------------| | |
599 | * | | Prime Bits | proto VSID_BITS_65VA | Total Bits | 2* prot VSID_BITS | | |
600 | * |-------+------------+----------------------+------------+-------------------| | |
601 | * | 1T | 24 | 25 | 49 | 50 | | |
602 | * |-------+------------+----------------------+------------+-------------------| | |
603 | * | 256MB | 24 | 37 | 61 | 74 | | |
604 | * |-------+------------+----------------------+------------+-------------------| | |
605 | * | |
606 | * |-------+------------+----------------------+------------+--------------------| | |
607 | * | | Prime Bits | proto VSID_BITS_68VA | Total Bits | 2* proto VSID_BITS | | |
608 | * |-------+------------+----------------------+------------+--------------------| | |
609 | * | 1T | 24 | 28 | 52 | 56 | | |
610 | * |-------+------------+----------------------+------------+--------------------| | |
611 | * | 256MB | 24 | 40 | 64 | 80 | | |
612 | * |-------+------------+----------------------+------------+--------------------| | |
613 | * | |
048ee099 AK |
614 | */ |
615 | #define VSID_MULTIPLIER_256M ASM_CONST(12538073) /* 24-bit prime */ | |
e6f81a92 AK |
616 | #define VSID_BITS_256M (VA_BITS - SID_SHIFT) |
617 | #define VSID_BITS_65_256M (65 - SID_SHIFT) | |
82228e36 AK |
618 | /* |
619 | * Modular multiplicative inverse of VSID_MULTIPLIER under modulo VSID_MODULUS | |
620 | */ | |
621 | #define VSID_MULINV_256M ASM_CONST(665548017062) | |
8d2169e8 | 622 | |
1189be65 | 623 | #define VSID_MULTIPLIER_1T ASM_CONST(12538073) /* 24-bit prime */ |
e6f81a92 AK |
624 | #define VSID_BITS_1T (VA_BITS - SID_SHIFT_1T) |
625 | #define VSID_BITS_65_1T (65 - SID_SHIFT_1T) | |
82228e36 | 626 | #define VSID_MULINV_1T ASM_CONST(209034062) |
8d2169e8 | 627 | |
82228e36 AK |
628 | /* 1TB VSID reserved for VRMA */ |
629 | #define VRMA_VSID 0x1ffffffUL | |
af81d787 | 630 | #define USER_VSID_RANGE (1UL << (ESID_BITS + SID_SHIFT)) |
8d2169e8 | 631 | |
78f1dbde | 632 | /* 4 bits per slice and we have one slice per 1TB */ |
957b778a | 633 | #define SLICE_ARRAY_SIZE (H_PGTABLE_RANGE >> 41) |
4722476b | 634 | #define TASK_SLICE_ARRAY_SZ(x) ((x)->context.slb_addr_limit >> 41) |
8d2169e8 DG |
635 | |
636 | #ifndef __ASSEMBLY__ | |
637 | ||
d28513bc DG |
638 | #ifdef CONFIG_PPC_SUBPAGE_PROT |
639 | /* | |
640 | * For the sub-page protection option, we extend the PGD with one of | |
641 | * these. Basically we have a 3-level tree, with the top level being | |
642 | * the protptrs array. To optimize speed and memory consumption when | |
643 | * only addresses < 4GB are being protected, pointers to the first | |
644 | * four pages of sub-page protection words are stored in the low_prot | |
645 | * array. | |
646 | * Each page of sub-page protection words protects 1GB (4 bytes | |
647 | * protects 64k). For the 3-level tree, each page of pointers then | |
648 | * protects 8TB. | |
649 | */ | |
650 | struct subpage_prot_table { | |
651 | unsigned long maxaddr; /* only addresses < this are protected */ | |
dad6f37c | 652 | unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)]; |
d28513bc DG |
653 | unsigned int *low_prot[4]; |
654 | }; | |
655 | ||
656 | #define SBP_L1_BITS (PAGE_SHIFT - 2) | |
657 | #define SBP_L2_BITS (PAGE_SHIFT - 3) | |
658 | #define SBP_L1_COUNT (1 << SBP_L1_BITS) | |
659 | #define SBP_L2_COUNT (1 << SBP_L2_BITS) | |
660 | #define SBP_L2_SHIFT (PAGE_SHIFT + SBP_L1_BITS) | |
661 | #define SBP_L3_SHIFT (SBP_L2_SHIFT + SBP_L2_BITS) | |
662 | ||
663 | extern void subpage_prot_free(struct mm_struct *mm); | |
664 | extern void subpage_prot_init_new_context(struct mm_struct *mm); | |
665 | #else | |
666 | static inline void subpage_prot_free(struct mm_struct *mm) {} | |
667 | static inline void subpage_prot_init_new_context(struct mm_struct *mm) { } | |
668 | #endif /* CONFIG_PPC_SUBPAGE_PROT */ | |
669 | ||
8d2169e8 | 670 | #if 0 |
1189be65 PM |
671 | /* |
672 | * The code below is equivalent to this function for arguments | |
673 | * < 2^VSID_BITS, which is all this should ever be called | |
674 | * with. However gcc is not clever enough to compute the | |
675 | * modulus (2^n-1) without a second multiply. | |
676 | */ | |
34692708 | 677 | #define vsid_scramble(protovsid, size) \ |
1189be65 | 678 | ((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size)) |
8d2169e8 | 679 | |
e6f81a92 | 680 | /* simplified form avoiding mod operation */ |
1189be65 PM |
681 | #define vsid_scramble(protovsid, size) \ |
682 | ({ \ | |
683 | unsigned long x; \ | |
684 | x = (protovsid) * VSID_MULTIPLIER_##size; \ | |
685 | x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \ | |
686 | (x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \ | |
687 | }) | |
e6f81a92 AK |
688 | |
689 | #else /* 1 */ | |
690 | static inline unsigned long vsid_scramble(unsigned long protovsid, | |
691 | unsigned long vsid_multiplier, int vsid_bits) | |
692 | { | |
693 | unsigned long vsid; | |
694 | unsigned long vsid_modulus = ((1UL << vsid_bits) - 1); | |
695 | /* | |
696 | * We have same multipler for both 256 and 1T segements now | |
697 | */ | |
698 | vsid = protovsid * vsid_multiplier; | |
699 | vsid = (vsid >> vsid_bits) + (vsid & vsid_modulus); | |
700 | return (vsid + ((vsid + 1) >> vsid_bits)) & vsid_modulus; | |
701 | } | |
702 | ||
8d2169e8 | 703 | #endif /* 1 */ |
8d2169e8 | 704 | |
1189be65 PM |
705 | /* Returns the segment size indicator for a user address */ |
706 | static inline int user_segment_size(unsigned long addr) | |
8d2169e8 | 707 | { |
1189be65 PM |
708 | /* Use 1T segments if possible for addresses >= 1T */ |
709 | if (addr >= (1UL << SID_SHIFT_1T)) | |
710 | return mmu_highuser_ssize; | |
711 | return MMU_SEGSIZE_256M; | |
8d2169e8 DG |
712 | } |
713 | ||
1189be65 PM |
714 | static inline unsigned long get_vsid(unsigned long context, unsigned long ea, |
715 | int ssize) | |
716 | { | |
e6f81a92 AK |
717 | unsigned long va_bits = VA_BITS; |
718 | unsigned long vsid_bits; | |
719 | unsigned long protovsid; | |
720 | ||
c60ac569 AK |
721 | /* |
722 | * Bad address. We return VSID 0 for that | |
723 | */ | |
dd1842a2 | 724 | if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE) |
c60ac569 AK |
725 | return 0; |
726 | ||
e6f81a92 AK |
727 | if (!mmu_has_feature(MMU_FTR_68_BIT_VA)) |
728 | va_bits = 65; | |
729 | ||
730 | if (ssize == MMU_SEGSIZE_256M) { | |
731 | vsid_bits = va_bits - SID_SHIFT; | |
732 | protovsid = (context << ESID_BITS) | | |
733 | ((ea >> SID_SHIFT) & ESID_BITS_MASK); | |
734 | return vsid_scramble(protovsid, VSID_MULTIPLIER_256M, vsid_bits); | |
735 | } | |
736 | /* 1T segment */ | |
737 | vsid_bits = va_bits - SID_SHIFT_1T; | |
738 | protovsid = (context << ESID_BITS_1T) | | |
739 | ((ea >> SID_SHIFT_1T) & ESID_BITS_1T_MASK); | |
740 | return vsid_scramble(protovsid, VSID_MULTIPLIER_1T, vsid_bits); | |
1189be65 PM |
741 | } |
742 | ||
c60ac569 AK |
743 | /* |
744 | * This is only valid for addresses >= PAGE_OFFSET | |
c60ac569 AK |
745 | */ |
746 | static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize) | |
747 | { | |
748 | unsigned long context; | |
749 | ||
85beb1c4 ME |
750 | if (!is_kernel_addr(ea)) |
751 | return 0; | |
752 | ||
c60ac569 | 753 | /* |
941711a3 AK |
754 | * For kernel space, we use context ids 1-4 to map the address space as |
755 | * below: | |
756 | * | |
757 | * 0x00001 - [ 0xc000000000000000 - 0xc0003fffffffffff ] | |
758 | * 0x00002 - [ 0xd000000000000000 - 0xd0003fffffffffff ] | |
759 | * 0x00003 - [ 0xe000000000000000 - 0xe0003fffffffffff ] | |
760 | * 0x00004 - [ 0xf000000000000000 - 0xf0003fffffffffff ] | |
761 | * | |
762 | * So we can compute the context from the region (top nibble) by | |
763 | * subtracting 11, or 0xc - 1. | |
c60ac569 | 764 | */ |
941711a3 AK |
765 | context = (ea >> 60) - KERNEL_REGION_CONTEXT_OFFSET; |
766 | ||
c60ac569 AK |
767 | return get_vsid(context, ea, ssize); |
768 | } | |
5c3c7ede DG |
769 | |
770 | unsigned htab_shift_for_mem_size(unsigned long mem_size); | |
771 | ||
8d2169e8 DG |
772 | #endif /* __ASSEMBLY__ */ |
773 | ||
11a6f6ab | 774 | #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */ |