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 | ||
15 | #include <asm/asm-compat.h> | |
16 | #include <asm/page.h> | |
891121e6 | 17 | #include <asm/bug.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> |
78f1dbde | 27 | |
8d2169e8 DG |
28 | /* |
29 | * SLB | |
30 | */ | |
31 | ||
32 | #define SLB_NUM_BOLTED 3 | |
33 | #define SLB_CACHE_ENTRIES 8 | |
46db2f86 | 34 | #define SLB_MIN_SIZE 32 |
8d2169e8 DG |
35 | |
36 | /* Bits in the SLB ESID word */ | |
37 | #define SLB_ESID_V ASM_CONST(0x0000000008000000) /* valid */ | |
38 | ||
39 | /* Bits in the SLB VSID word */ | |
40 | #define SLB_VSID_SHIFT 12 | |
1189be65 PM |
41 | #define SLB_VSID_SHIFT_1T 24 |
42 | #define SLB_VSID_SSIZE_SHIFT 62 | |
8d2169e8 DG |
43 | #define SLB_VSID_B ASM_CONST(0xc000000000000000) |
44 | #define SLB_VSID_B_256M ASM_CONST(0x0000000000000000) | |
45 | #define SLB_VSID_B_1T ASM_CONST(0x4000000000000000) | |
46 | #define SLB_VSID_KS ASM_CONST(0x0000000000000800) | |
47 | #define SLB_VSID_KP ASM_CONST(0x0000000000000400) | |
48 | #define SLB_VSID_N ASM_CONST(0x0000000000000200) /* no-execute */ | |
49 | #define SLB_VSID_L ASM_CONST(0x0000000000000100) | |
50 | #define SLB_VSID_C ASM_CONST(0x0000000000000080) /* class */ | |
51 | #define SLB_VSID_LP ASM_CONST(0x0000000000000030) | |
52 | #define SLB_VSID_LP_00 ASM_CONST(0x0000000000000000) | |
53 | #define SLB_VSID_LP_01 ASM_CONST(0x0000000000000010) | |
54 | #define SLB_VSID_LP_10 ASM_CONST(0x0000000000000020) | |
55 | #define SLB_VSID_LP_11 ASM_CONST(0x0000000000000030) | |
56 | #define SLB_VSID_LLP (SLB_VSID_L|SLB_VSID_LP) | |
57 | ||
58 | #define SLB_VSID_KERNEL (SLB_VSID_KP) | |
59 | #define SLB_VSID_USER (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C) | |
60 | ||
61 | #define SLBIE_C (0x08000000) | |
1189be65 | 62 | #define SLBIE_SSIZE_SHIFT 25 |
8d2169e8 DG |
63 | |
64 | /* | |
65 | * Hash table | |
66 | */ | |
67 | ||
68 | #define HPTES_PER_GROUP 8 | |
69 | ||
2454c7e9 | 70 | #define HPTE_V_SSIZE_SHIFT 62 |
8d2169e8 | 71 | #define HPTE_V_AVPN_SHIFT 7 |
2454c7e9 | 72 | #define HPTE_V_AVPN ASM_CONST(0x3fffffffffffff80) |
8d2169e8 | 73 | #define HPTE_V_AVPN_VAL(x) (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT) |
91bbbe22 | 74 | #define HPTE_V_COMPARE(x,y) (!(((x) ^ (y)) & 0xffffffffffffff80UL)) |
8d2169e8 DG |
75 | #define HPTE_V_BOLTED ASM_CONST(0x0000000000000010) |
76 | #define HPTE_V_LOCK ASM_CONST(0x0000000000000008) | |
77 | #define HPTE_V_LARGE ASM_CONST(0x0000000000000004) | |
78 | #define HPTE_V_SECONDARY ASM_CONST(0x0000000000000002) | |
79 | #define HPTE_V_VALID ASM_CONST(0x0000000000000001) | |
80 | ||
50de596d AK |
81 | /* |
82 | * ISA 3.0 have a different HPTE format. | |
83 | */ | |
84 | #define HPTE_R_3_0_SSIZE_SHIFT 58 | |
8d2169e8 DG |
85 | #define HPTE_R_PP0 ASM_CONST(0x8000000000000000) |
86 | #define HPTE_R_TS ASM_CONST(0x4000000000000000) | |
de56a948 | 87 | #define HPTE_R_KEY_HI ASM_CONST(0x3000000000000000) |
8d2169e8 | 88 | #define HPTE_R_RPN_SHIFT 12 |
de56a948 | 89 | #define HPTE_R_RPN ASM_CONST(0x0ffffffffffff000) |
8d2169e8 DG |
90 | #define HPTE_R_PP ASM_CONST(0x0000000000000003) |
91 | #define HPTE_R_N ASM_CONST(0x0000000000000004) | |
de56a948 PM |
92 | #define HPTE_R_G ASM_CONST(0x0000000000000008) |
93 | #define HPTE_R_M ASM_CONST(0x0000000000000010) | |
94 | #define HPTE_R_I ASM_CONST(0x0000000000000020) | |
95 | #define HPTE_R_W ASM_CONST(0x0000000000000040) | |
96 | #define HPTE_R_WIMG ASM_CONST(0x0000000000000078) | |
8d2169e8 DG |
97 | #define HPTE_R_C ASM_CONST(0x0000000000000080) |
98 | #define HPTE_R_R ASM_CONST(0x0000000000000100) | |
de56a948 | 99 | #define HPTE_R_KEY_LO ASM_CONST(0x0000000000000e00) |
8d2169e8 | 100 | |
b7abc5c5 SS |
101 | #define HPTE_V_1TB_SEG ASM_CONST(0x4000000000000000) |
102 | #define HPTE_V_VRMA_MASK ASM_CONST(0x4001ffffff000000) | |
103 | ||
8d2169e8 | 104 | /* Values for PP (assumes Ks=0, Kp=1) */ |
8d2169e8 DG |
105 | #define PP_RWXX 0 /* Supervisor read/write, User none */ |
106 | #define PP_RWRX 1 /* Supervisor read/write, User read */ | |
107 | #define PP_RWRW 2 /* Supervisor read/write, User read/write */ | |
108 | #define PP_RXRX 3 /* Supervisor read, User read */ | |
697d3899 | 109 | #define PP_RXXX (HPTE_R_PP0 | 2) /* Supervisor read, user none */ |
8d2169e8 | 110 | |
b4072df4 PM |
111 | /* Fields for tlbiel instruction in architecture 2.06 */ |
112 | #define TLBIEL_INVAL_SEL_MASK 0xc00 /* invalidation selector */ | |
113 | #define TLBIEL_INVAL_PAGE 0x000 /* invalidate a single page */ | |
114 | #define TLBIEL_INVAL_SET_LPID 0x800 /* invalidate a set for current LPID */ | |
115 | #define TLBIEL_INVAL_SET 0xc00 /* invalidate a set for all LPIDs */ | |
116 | #define TLBIEL_INVAL_SET_MASK 0xfff000 /* set number to inval. */ | |
117 | #define TLBIEL_INVAL_SET_SHIFT 12 | |
118 | ||
119 | #define POWER7_TLB_SETS 128 /* # sets in POWER7 TLB */ | |
45706bb5 | 120 | #define POWER8_TLB_SETS 512 /* # sets in POWER8 TLB */ |
c3ab300e | 121 | #define POWER9_TLB_SETS_HASH 256 /* # sets in POWER9 TLB Hash mode */ |
1a472c9d | 122 | #define POWER9_TLB_SETS_RADIX 128 /* # sets in POWER9 TLB Radix mode */ |
b4072df4 | 123 | |
8d2169e8 DG |
124 | #ifndef __ASSEMBLY__ |
125 | ||
8e561e7e | 126 | struct hash_pte { |
12f04f2b AB |
127 | __be64 v; |
128 | __be64 r; | |
8e561e7e | 129 | }; |
8d2169e8 | 130 | |
8e561e7e | 131 | extern struct hash_pte *htab_address; |
8d2169e8 DG |
132 | extern unsigned long htab_size_bytes; |
133 | extern unsigned long htab_hash_mask; | |
134 | ||
cf9427b8 AK |
135 | |
136 | static inline int shift_to_mmu_psize(unsigned int shift) | |
137 | { | |
138 | int psize; | |
139 | ||
140 | for (psize = 0; psize < MMU_PAGE_COUNT; ++psize) | |
141 | if (mmu_psize_defs[psize].shift == shift) | |
142 | return psize; | |
143 | return -1; | |
144 | } | |
145 | ||
146 | static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize) | |
147 | { | |
148 | if (mmu_psize_defs[mmu_psize].shift) | |
149 | return mmu_psize_defs[mmu_psize].shift; | |
150 | BUG(); | |
151 | } | |
8d2169e8 DG |
152 | |
153 | #endif /* __ASSEMBLY__ */ | |
154 | ||
2454c7e9 PM |
155 | /* |
156 | * Segment sizes. | |
157 | * These are the values used by hardware in the B field of | |
158 | * SLB entries and the first dword of MMU hashtable entries. | |
159 | * The B field is 2 bits; the values 2 and 3 are unused and reserved. | |
160 | */ | |
161 | #define MMU_SEGSIZE_256M 0 | |
162 | #define MMU_SEGSIZE_1T 1 | |
163 | ||
5524a27d AK |
164 | /* |
165 | * encode page number shift. | |
166 | * in order to fit the 78 bit va in a 64 bit variable we shift the va by | |
167 | * 12 bits. This enable us to address upto 76 bit va. | |
168 | * For hpt hash from a va we can ignore the page size bits of va and for | |
169 | * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure | |
170 | * we work in all cases including 4k page size. | |
171 | */ | |
172 | #define VPN_SHIFT 12 | |
1189be65 | 173 | |
b1022fbd AK |
174 | /* |
175 | * HPTE Large Page (LP) details | |
176 | */ | |
177 | #define LP_SHIFT 12 | |
178 | #define LP_BITS 8 | |
179 | #define LP_MASK(i) ((0xFF >> (i)) << LP_SHIFT) | |
180 | ||
8d2169e8 DG |
181 | #ifndef __ASSEMBLY__ |
182 | ||
73d16a6e IM |
183 | static inline int slb_vsid_shift(int ssize) |
184 | { | |
185 | if (ssize == MMU_SEGSIZE_256M) | |
186 | return SLB_VSID_SHIFT; | |
187 | return SLB_VSID_SHIFT_1T; | |
188 | } | |
189 | ||
5524a27d AK |
190 | static inline int segment_shift(int ssize) |
191 | { | |
192 | if (ssize == MMU_SEGSIZE_256M) | |
193 | return SID_SHIFT; | |
194 | return SID_SHIFT_1T; | |
195 | } | |
196 | ||
8d2169e8 | 197 | /* |
1189be65 | 198 | * The current system page and segment sizes |
8d2169e8 | 199 | */ |
1189be65 PM |
200 | extern int mmu_kernel_ssize; |
201 | extern int mmu_highuser_ssize; | |
584f8b71 | 202 | extern u16 mmu_slb_size; |
572fb578 | 203 | extern unsigned long tce_alloc_start, tce_alloc_end; |
8d2169e8 DG |
204 | |
205 | /* | |
206 | * If the processor supports 64k normal pages but not 64k cache | |
207 | * inhibited pages, we have to be prepared to switch processes | |
208 | * to use 4k pages when they create cache-inhibited mappings. | |
209 | * If this is the case, mmu_ci_restrictions will be set to 1. | |
210 | */ | |
211 | extern int mmu_ci_restrictions; | |
212 | ||
5524a27d AK |
213 | /* |
214 | * This computes the AVPN and B fields of the first dword of a HPTE, | |
215 | * for use when we want to match an existing PTE. The bottom 7 bits | |
216 | * of the returned value are zero. | |
217 | */ | |
218 | static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize, | |
219 | int ssize) | |
220 | { | |
221 | unsigned long v; | |
222 | /* | |
223 | * The AVA field omits the low-order 23 bits of the 78 bits VA. | |
224 | * These bits are not needed in the PTE, because the | |
225 | * low-order b of these bits are part of the byte offset | |
226 | * into the virtual page and, if b < 23, the high-order | |
227 | * 23-b of these bits are always used in selecting the | |
228 | * PTEGs to be searched | |
229 | */ | |
230 | v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm); | |
231 | v <<= HPTE_V_AVPN_SHIFT; | |
50de596d AK |
232 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) |
233 | v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT; | |
5524a27d AK |
234 | return v; |
235 | } | |
236 | ||
8d2169e8 DG |
237 | /* |
238 | * This function sets the AVPN and L fields of the HPTE appropriately | |
b1022fbd | 239 | * using the base page size and actual page size. |
8d2169e8 | 240 | */ |
b1022fbd AK |
241 | static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize, |
242 | int actual_psize, int ssize) | |
8d2169e8 | 243 | { |
1189be65 | 244 | unsigned long v; |
b1022fbd AK |
245 | v = hpte_encode_avpn(vpn, base_psize, ssize); |
246 | if (actual_psize != MMU_PAGE_4K) | |
8d2169e8 DG |
247 | v |= HPTE_V_LARGE; |
248 | return v; | |
249 | } | |
250 | ||
251 | /* | |
252 | * This function sets the ARPN, and LP fields of the HPTE appropriately | |
253 | * for the page size. We assume the pa is already "clean" that is properly | |
254 | * aligned for the requested page size | |
255 | */ | |
b1022fbd | 256 | static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize, |
50de596d | 257 | int actual_psize, int ssize) |
8d2169e8 | 258 | { |
50de596d AK |
259 | |
260 | if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
261 | pa |= ((unsigned long) ssize) << HPTE_R_3_0_SSIZE_SHIFT; | |
262 | ||
8d2169e8 | 263 | /* A 4K page needs no special encoding */ |
b1022fbd | 264 | if (actual_psize == MMU_PAGE_4K) |
8d2169e8 DG |
265 | return pa & HPTE_R_RPN; |
266 | else { | |
b1022fbd AK |
267 | unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize]; |
268 | unsigned int shift = mmu_psize_defs[actual_psize].shift; | |
269 | return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT); | |
8d2169e8 | 270 | } |
8d2169e8 DG |
271 | } |
272 | ||
273 | /* | |
5524a27d | 274 | * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size. |
8d2169e8 | 275 | */ |
5524a27d AK |
276 | static inline unsigned long hpt_vpn(unsigned long ea, |
277 | unsigned long vsid, int ssize) | |
1189be65 | 278 | { |
5524a27d AK |
279 | unsigned long mask; |
280 | int s_shift = segment_shift(ssize); | |
281 | ||
282 | mask = (1ul << (s_shift - VPN_SHIFT)) - 1; | |
283 | return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask); | |
1189be65 | 284 | } |
8d2169e8 | 285 | |
1189be65 PM |
286 | /* |
287 | * This hashes a virtual address | |
288 | */ | |
5524a27d AK |
289 | static inline unsigned long hpt_hash(unsigned long vpn, |
290 | unsigned int shift, int ssize) | |
8d2169e8 | 291 | { |
5524a27d | 292 | int mask; |
1189be65 PM |
293 | unsigned long hash, vsid; |
294 | ||
5524a27d | 295 | /* VPN_SHIFT can be atmost 12 */ |
1189be65 | 296 | if (ssize == MMU_SEGSIZE_256M) { |
5524a27d AK |
297 | mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1; |
298 | hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^ | |
299 | ((vpn & mask) >> (shift - VPN_SHIFT)); | |
1189be65 | 300 | } else { |
5524a27d AK |
301 | mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1; |
302 | vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT); | |
303 | hash = vsid ^ (vsid << 25) ^ | |
304 | ((vpn & mask) >> (shift - VPN_SHIFT)) ; | |
1189be65 PM |
305 | } |
306 | return hash & 0x7fffffffffUL; | |
8d2169e8 DG |
307 | } |
308 | ||
aefa5688 AK |
309 | #define HPTE_LOCAL_UPDATE 0x1 |
310 | #define HPTE_NOHPTE_UPDATE 0x2 | |
311 | ||
8d2169e8 DG |
312 | extern int __hash_page_4K(unsigned long ea, unsigned long access, |
313 | unsigned long vsid, pte_t *ptep, unsigned long trap, | |
aefa5688 | 314 | unsigned long flags, int ssize, int subpage_prot); |
8d2169e8 DG |
315 | extern int __hash_page_64K(unsigned long ea, unsigned long access, |
316 | unsigned long vsid, pte_t *ptep, unsigned long trap, | |
aefa5688 | 317 | unsigned long flags, int ssize); |
8d2169e8 | 318 | struct mm_struct; |
0895ecda | 319 | unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap); |
aefa5688 AK |
320 | extern int hash_page_mm(struct mm_struct *mm, unsigned long ea, |
321 | unsigned long access, unsigned long trap, | |
322 | unsigned long flags); | |
323 | extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap, | |
324 | unsigned long dsisr); | |
a4fe3ce7 | 325 | int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid, |
aefa5688 AK |
326 | pte_t *ptep, unsigned long trap, unsigned long flags, |
327 | int ssize, unsigned int shift, unsigned int mmu_psize); | |
6d492ecc AK |
328 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
329 | extern int __hash_page_thp(unsigned long ea, unsigned long access, | |
330 | unsigned long vsid, pmd_t *pmdp, unsigned long trap, | |
aefa5688 | 331 | unsigned long flags, int ssize, unsigned int psize); |
6d492ecc AK |
332 | #else |
333 | static inline int __hash_page_thp(unsigned long ea, unsigned long access, | |
334 | unsigned long vsid, pmd_t *pmdp, | |
aefa5688 | 335 | unsigned long trap, unsigned long flags, |
6d492ecc AK |
336 | int ssize, unsigned int psize) |
337 | { | |
338 | BUG(); | |
ff1e7683 | 339 | return -1; |
6d492ecc AK |
340 | } |
341 | #endif | |
4b8692c0 BH |
342 | extern void hash_failure_debug(unsigned long ea, unsigned long access, |
343 | unsigned long vsid, unsigned long trap, | |
d8139ebf AK |
344 | int ssize, int psize, int lpsize, |
345 | unsigned long pte); | |
8d2169e8 | 346 | extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend, |
bc033b63 | 347 | unsigned long pstart, unsigned long prot, |
1189be65 | 348 | int psize, int ssize); |
f6026df1 AB |
349 | int htab_remove_mapping(unsigned long vstart, unsigned long vend, |
350 | int psize, int ssize); | |
41151e77 | 351 | extern void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages); |
fa28237c | 352 | extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr); |
8d2169e8 | 353 | |
8d2169e8 DG |
354 | extern void hpte_init_native(void); |
355 | extern void hpte_init_lpar(void); | |
8d2169e8 | 356 | extern void hpte_init_beat(void); |
7f2c8577 | 357 | extern void hpte_init_beat_v3(void); |
8d2169e8 | 358 | |
8d2169e8 DG |
359 | extern void slb_initialize(void); |
360 | extern void slb_flush_and_rebolt(void); | |
8d2169e8 | 361 | |
67439b76 | 362 | extern void slb_vmalloc_update(void); |
46db2f86 | 363 | extern void slb_set_size(u16 size); |
8d2169e8 DG |
364 | #endif /* __ASSEMBLY__ */ |
365 | ||
366 | /* | |
f033d659 | 367 | * VSID allocation (256MB segment) |
8d2169e8 | 368 | * |
c60ac569 AK |
369 | * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated |
370 | * from mmu context id and effective segment id of the address. | |
8d2169e8 | 371 | * |
c60ac569 AK |
372 | * For user processes max context id is limited to ((1ul << 19) - 5) |
373 | * for kernel space, we use the top 4 context ids to map address as below | |
374 | * NOTE: each context only support 64TB now. | |
375 | * 0x7fffc - [ 0xc000000000000000 - 0xc0003fffffffffff ] | |
376 | * 0x7fffd - [ 0xd000000000000000 - 0xd0003fffffffffff ] | |
377 | * 0x7fffe - [ 0xe000000000000000 - 0xe0003fffffffffff ] | |
378 | * 0x7ffff - [ 0xf000000000000000 - 0xf0003fffffffffff ] | |
8d2169e8 DG |
379 | * |
380 | * The proto-VSIDs are then scrambled into real VSIDs with the | |
381 | * multiplicative hash: | |
382 | * | |
383 | * VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS | |
8d2169e8 | 384 | * |
f033d659 | 385 | * VSID_MULTIPLIER is prime, so in particular it is |
8d2169e8 DG |
386 | * co-prime to VSID_MODULUS, making this a 1:1 scrambling function. |
387 | * Because the modulus is 2^n-1 we can compute it efficiently without | |
c60ac569 AK |
388 | * a divide or extra multiply (see below). The scramble function gives |
389 | * robust scattering in the hash table (at least based on some initial | |
390 | * results). | |
8d2169e8 | 391 | * |
c60ac569 AK |
392 | * We also consider VSID 0 special. We use VSID 0 for slb entries mapping |
393 | * bad address. This enables us to consolidate bad address handling in | |
394 | * hash_page. | |
8d2169e8 | 395 | * |
c60ac569 AK |
396 | * We also need to avoid the last segment of the last context, because that |
397 | * would give a protovsid of 0x1fffffffff. That will result in a VSID 0 | |
398 | * because of the modulo operation in vsid scramble. But the vmemmap | |
399 | * (which is what uses region 0xf) will never be close to 64TB in size | |
400 | * (it's 56 bytes per page of system memory). | |
8d2169e8 | 401 | */ |
8d2169e8 | 402 | |
e39d1a47 | 403 | #define CONTEXT_BITS 19 |
af81d787 AK |
404 | #define ESID_BITS 18 |
405 | #define ESID_BITS_1T 6 | |
e39d1a47 | 406 | |
c60ac569 AK |
407 | /* |
408 | * 256MB segment | |
af81d787 | 409 | * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments |
c60ac569 AK |
410 | * available for user + kernel mapping. The top 4 contexts are used for |
411 | * kernel mapping. Each segment contains 2^28 bytes. Each | |
412 | * context maps 2^46 bytes (64TB) so we can support 2^19-1 contexts | |
413 | * (19 == 37 + 28 - 46). | |
414 | */ | |
415 | #define MAX_USER_CONTEXT ((ASM_CONST(1) << CONTEXT_BITS) - 5) | |
416 | ||
048ee099 AK |
417 | /* |
418 | * This should be computed such that protovosid * vsid_mulitplier | |
419 | * doesn't overflow 64 bits. It should also be co-prime to vsid_modulus | |
420 | */ | |
421 | #define VSID_MULTIPLIER_256M ASM_CONST(12538073) /* 24-bit prime */ | |
af81d787 | 422 | #define VSID_BITS_256M (CONTEXT_BITS + ESID_BITS) |
1189be65 | 423 | #define VSID_MODULUS_256M ((1UL<<VSID_BITS_256M)-1) |
8d2169e8 | 424 | |
1189be65 | 425 | #define VSID_MULTIPLIER_1T ASM_CONST(12538073) /* 24-bit prime */ |
af81d787 | 426 | #define VSID_BITS_1T (CONTEXT_BITS + ESID_BITS_1T) |
1189be65 PM |
427 | #define VSID_MODULUS_1T ((1UL<<VSID_BITS_1T)-1) |
428 | ||
8d2169e8 | 429 | |
af81d787 | 430 | #define USER_VSID_RANGE (1UL << (ESID_BITS + SID_SHIFT)) |
8d2169e8 DG |
431 | |
432 | /* | |
433 | * This macro generates asm code to compute the VSID scramble | |
434 | * function. Used in slb_allocate() and do_stab_bolted. The function | |
435 | * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS | |
436 | * | |
437 | * rt = register continaing the proto-VSID and into which the | |
438 | * VSID will be stored | |
439 | * rx = scratch register (clobbered) | |
440 | * | |
441 | * - rt and rx must be different registers | |
1189be65 | 442 | * - The answer will end up in the low VSID_BITS bits of rt. The higher |
8d2169e8 DG |
443 | * bits may contain other garbage, so you may need to mask the |
444 | * result. | |
445 | */ | |
1189be65 PM |
446 | #define ASM_VSID_SCRAMBLE(rt, rx, size) \ |
447 | lis rx,VSID_MULTIPLIER_##size@h; \ | |
448 | ori rx,rx,VSID_MULTIPLIER_##size@l; \ | |
8d2169e8 DG |
449 | mulld rt,rt,rx; /* rt = rt * MULTIPLIER */ \ |
450 | \ | |
1189be65 PM |
451 | srdi rx,rt,VSID_BITS_##size; \ |
452 | clrldi rt,rt,(64-VSID_BITS_##size); \ | |
8d2169e8 | 453 | add rt,rt,rx; /* add high and low bits */ \ |
c60ac569 AK |
454 | /* NOTE: explanation based on VSID_BITS_##size = 36 \ |
455 | * Now, r3 == VSID (mod 2^36-1), and lies between 0 and \ | |
8d2169e8 DG |
456 | * 2^36-1+2^28-1. That in particular means that if r3 >= \ |
457 | * 2^36-1, then r3+1 has the 2^36 bit set. So, if r3+1 has \ | |
458 | * the bit clear, r3 already has the answer we want, if it \ | |
459 | * doesn't, the answer is the low 36 bits of r3+1. So in all \ | |
460 | * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\ | |
461 | addi rx,rt,1; \ | |
1189be65 | 462 | srdi rx,rx,VSID_BITS_##size; /* extract 2^VSID_BITS bit */ \ |
8d2169e8 DG |
463 | add rt,rt,rx |
464 | ||
78f1dbde | 465 | /* 4 bits per slice and we have one slice per 1TB */ |
dd1842a2 | 466 | #define SLICE_ARRAY_SIZE (H_PGTABLE_RANGE >> 41) |
8d2169e8 DG |
467 | |
468 | #ifndef __ASSEMBLY__ | |
469 | ||
d28513bc DG |
470 | #ifdef CONFIG_PPC_SUBPAGE_PROT |
471 | /* | |
472 | * For the sub-page protection option, we extend the PGD with one of | |
473 | * these. Basically we have a 3-level tree, with the top level being | |
474 | * the protptrs array. To optimize speed and memory consumption when | |
475 | * only addresses < 4GB are being protected, pointers to the first | |
476 | * four pages of sub-page protection words are stored in the low_prot | |
477 | * array. | |
478 | * Each page of sub-page protection words protects 1GB (4 bytes | |
479 | * protects 64k). For the 3-level tree, each page of pointers then | |
480 | * protects 8TB. | |
481 | */ | |
482 | struct subpage_prot_table { | |
483 | unsigned long maxaddr; /* only addresses < this are protected */ | |
dad6f37c | 484 | unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)]; |
d28513bc DG |
485 | unsigned int *low_prot[4]; |
486 | }; | |
487 | ||
488 | #define SBP_L1_BITS (PAGE_SHIFT - 2) | |
489 | #define SBP_L2_BITS (PAGE_SHIFT - 3) | |
490 | #define SBP_L1_COUNT (1 << SBP_L1_BITS) | |
491 | #define SBP_L2_COUNT (1 << SBP_L2_BITS) | |
492 | #define SBP_L2_SHIFT (PAGE_SHIFT + SBP_L1_BITS) | |
493 | #define SBP_L3_SHIFT (SBP_L2_SHIFT + SBP_L2_BITS) | |
494 | ||
495 | extern void subpage_prot_free(struct mm_struct *mm); | |
496 | extern void subpage_prot_init_new_context(struct mm_struct *mm); | |
497 | #else | |
498 | static inline void subpage_prot_free(struct mm_struct *mm) {} | |
499 | static inline void subpage_prot_init_new_context(struct mm_struct *mm) { } | |
500 | #endif /* CONFIG_PPC_SUBPAGE_PROT */ | |
501 | ||
8d2169e8 | 502 | #if 0 |
1189be65 PM |
503 | /* |
504 | * The code below is equivalent to this function for arguments | |
505 | * < 2^VSID_BITS, which is all this should ever be called | |
506 | * with. However gcc is not clever enough to compute the | |
507 | * modulus (2^n-1) without a second multiply. | |
508 | */ | |
34692708 | 509 | #define vsid_scramble(protovsid, size) \ |
1189be65 | 510 | ((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size)) |
8d2169e8 | 511 | |
1189be65 PM |
512 | #else /* 1 */ |
513 | #define vsid_scramble(protovsid, size) \ | |
514 | ({ \ | |
515 | unsigned long x; \ | |
516 | x = (protovsid) * VSID_MULTIPLIER_##size; \ | |
517 | x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \ | |
518 | (x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \ | |
519 | }) | |
8d2169e8 | 520 | #endif /* 1 */ |
8d2169e8 | 521 | |
1189be65 PM |
522 | /* Returns the segment size indicator for a user address */ |
523 | static inline int user_segment_size(unsigned long addr) | |
8d2169e8 | 524 | { |
1189be65 PM |
525 | /* Use 1T segments if possible for addresses >= 1T */ |
526 | if (addr >= (1UL << SID_SHIFT_1T)) | |
527 | return mmu_highuser_ssize; | |
528 | return MMU_SEGSIZE_256M; | |
8d2169e8 DG |
529 | } |
530 | ||
1189be65 PM |
531 | static inline unsigned long get_vsid(unsigned long context, unsigned long ea, |
532 | int ssize) | |
533 | { | |
c60ac569 AK |
534 | /* |
535 | * Bad address. We return VSID 0 for that | |
536 | */ | |
dd1842a2 | 537 | if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE) |
c60ac569 AK |
538 | return 0; |
539 | ||
1189be65 | 540 | if (ssize == MMU_SEGSIZE_256M) |
af81d787 | 541 | return vsid_scramble((context << ESID_BITS) |
1189be65 | 542 | | (ea >> SID_SHIFT), 256M); |
af81d787 | 543 | return vsid_scramble((context << ESID_BITS_1T) |
1189be65 PM |
544 | | (ea >> SID_SHIFT_1T), 1T); |
545 | } | |
546 | ||
c60ac569 AK |
547 | /* |
548 | * This is only valid for addresses >= PAGE_OFFSET | |
549 | * | |
550 | * For kernel space, we use the top 4 context ids to map address as below | |
551 | * 0x7fffc - [ 0xc000000000000000 - 0xc0003fffffffffff ] | |
552 | * 0x7fffd - [ 0xd000000000000000 - 0xd0003fffffffffff ] | |
553 | * 0x7fffe - [ 0xe000000000000000 - 0xe0003fffffffffff ] | |
554 | * 0x7ffff - [ 0xf000000000000000 - 0xf0003fffffffffff ] | |
555 | */ | |
556 | static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize) | |
557 | { | |
558 | unsigned long context; | |
559 | ||
560 | /* | |
561 | * kernel take the top 4 context from the available range | |
562 | */ | |
563 | context = (MAX_USER_CONTEXT) + ((ea >> 60) - 0xc) + 1; | |
564 | return get_vsid(context, ea, ssize); | |
565 | } | |
5c3c7ede DG |
566 | |
567 | unsigned htab_shift_for_mem_size(unsigned long mem_size); | |
568 | ||
8d2169e8 DG |
569 | #endif /* __ASSEMBLY__ */ |
570 | ||
11a6f6ab | 571 | #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */ |