Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * PowerPC memory management structures | |
3 | * | |
4 | * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com> | |
5 | * PPC64 rework. | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version | |
10 | * 2 of the License, or (at your option) any later version. | |
11 | */ | |
12 | ||
13 | #ifndef _PPC64_MMU_H_ | |
14 | #define _PPC64_MMU_H_ | |
15 | ||
16 | #include <linux/config.h> | |
5f7c6907 | 17 | #include <asm/ppc_asm.h> /* for ASM_CONST */ |
1da177e4 | 18 | #include <asm/page.h> |
1da177e4 | 19 | |
1f8d419e DG |
20 | /* |
21 | * Segment table | |
22 | */ | |
1da177e4 LT |
23 | |
24 | #define STE_ESID_V 0x80 | |
25 | #define STE_ESID_KS 0x20 | |
26 | #define STE_ESID_KP 0x10 | |
27 | #define STE_ESID_N 0x08 | |
28 | ||
29 | #define STE_VSID_SHIFT 12 | |
30 | ||
1f8d419e | 31 | /* Location of cpu0's segment table */ |
c59c464a | 32 | #define STAB0_PAGE 0x6 |
637a6ff6 | 33 | #define STAB0_PHYS_ADDR (STAB0_PAGE<<12) |
c59c464a DG |
34 | |
35 | #ifndef __ASSEMBLY__ | |
36 | extern char initial_stab[]; | |
37 | #endif /* ! __ASSEMBLY */ | |
1f8d419e DG |
38 | |
39 | /* | |
40 | * SLB | |
41 | */ | |
1da177e4 | 42 | |
1f8d419e DG |
43 | #define SLB_NUM_BOLTED 3 |
44 | #define SLB_CACHE_ENTRIES 8 | |
45 | ||
46 | /* Bits in the SLB ESID word */ | |
47 | #define SLB_ESID_V ASM_CONST(0x0000000008000000) /* valid */ | |
48 | ||
49 | /* Bits in the SLB VSID word */ | |
50 | #define SLB_VSID_SHIFT 12 | |
3c726f8d BH |
51 | #define SLB_VSID_B ASM_CONST(0xc000000000000000) |
52 | #define SLB_VSID_B_256M ASM_CONST(0x0000000000000000) | |
53 | #define SLB_VSID_B_1T ASM_CONST(0x4000000000000000) | |
1f8d419e DG |
54 | #define SLB_VSID_KS ASM_CONST(0x0000000000000800) |
55 | #define SLB_VSID_KP ASM_CONST(0x0000000000000400) | |
56 | #define SLB_VSID_N ASM_CONST(0x0000000000000200) /* no-execute */ | |
3c726f8d | 57 | #define SLB_VSID_L ASM_CONST(0x0000000000000100) |
1f8d419e | 58 | #define SLB_VSID_C ASM_CONST(0x0000000000000080) /* class */ |
3c726f8d BH |
59 | #define SLB_VSID_LP ASM_CONST(0x0000000000000030) |
60 | #define SLB_VSID_LP_00 ASM_CONST(0x0000000000000000) | |
61 | #define SLB_VSID_LP_01 ASM_CONST(0x0000000000000010) | |
62 | #define SLB_VSID_LP_10 ASM_CONST(0x0000000000000020) | |
63 | #define SLB_VSID_LP_11 ASM_CONST(0x0000000000000030) | |
64 | #define SLB_VSID_LLP (SLB_VSID_L|SLB_VSID_LP) | |
65 | ||
14b34661 DG |
66 | #define SLB_VSID_KERNEL (SLB_VSID_KP) |
67 | #define SLB_VSID_USER (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C) | |
68 | ||
69 | #define SLBIE_C (0x08000000) | |
1f8d419e DG |
70 | |
71 | /* | |
72 | * Hash table | |
73 | */ | |
1da177e4 LT |
74 | |
75 | #define HPTES_PER_GROUP 8 | |
76 | ||
96e28449 DG |
77 | #define HPTE_V_AVPN_SHIFT 7 |
78 | #define HPTE_V_AVPN ASM_CONST(0xffffffffffffff80) | |
79 | #define HPTE_V_AVPN_VAL(x) (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT) | |
3c726f8d | 80 | #define HPTE_V_COMPARE(x,y) (!(((x) ^ (y)) & HPTE_V_AVPN)) |
96e28449 DG |
81 | #define HPTE_V_BOLTED ASM_CONST(0x0000000000000010) |
82 | #define HPTE_V_LOCK ASM_CONST(0x0000000000000008) | |
83 | #define HPTE_V_LARGE ASM_CONST(0x0000000000000004) | |
84 | #define HPTE_V_SECONDARY ASM_CONST(0x0000000000000002) | |
85 | #define HPTE_V_VALID ASM_CONST(0x0000000000000001) | |
86 | ||
87 | #define HPTE_R_PP0 ASM_CONST(0x8000000000000000) | |
88 | #define HPTE_R_TS ASM_CONST(0x4000000000000000) | |
89 | #define HPTE_R_RPN_SHIFT 12 | |
90 | #define HPTE_R_RPN ASM_CONST(0x3ffffffffffff000) | |
91 | #define HPTE_R_FLAGS ASM_CONST(0x00000000000003ff) | |
92 | #define HPTE_R_PP ASM_CONST(0x0000000000000003) | |
3c726f8d | 93 | #define HPTE_R_N ASM_CONST(0x0000000000000004) |
96e28449 | 94 | |
1f8d419e DG |
95 | /* Values for PP (assumes Ks=0, Kp=1) */ |
96 | /* pp0 will always be 0 for linux */ | |
97 | #define PP_RWXX 0 /* Supervisor read/write, User none */ | |
98 | #define PP_RWRX 1 /* Supervisor read/write, User read */ | |
99 | #define PP_RWRW 2 /* Supervisor read/write, User read/write */ | |
100 | #define PP_RXRX 3 /* Supervisor read, User read */ | |
101 | ||
102 | #ifndef __ASSEMBLY__ | |
103 | ||
1da177e4 | 104 | typedef struct { |
96e28449 DG |
105 | unsigned long v; |
106 | unsigned long r; | |
107 | } hpte_t; | |
1da177e4 | 108 | |
96e28449 DG |
109 | extern hpte_t *htab_address; |
110 | extern unsigned long htab_hash_mask; | |
1da177e4 | 111 | |
3c726f8d BH |
112 | /* |
113 | * Page size definition | |
114 | * | |
115 | * shift : is the "PAGE_SHIFT" value for that page size | |
116 | * sllp : is a bit mask with the value of SLB L || LP to be or'ed | |
117 | * directly to a slbmte "vsid" value | |
118 | * penc : is the HPTE encoding mask for the "LP" field: | |
119 | * | |
120 | */ | |
121 | struct mmu_psize_def | |
1da177e4 | 122 | { |
3c726f8d BH |
123 | unsigned int shift; /* number of bits */ |
124 | unsigned int penc; /* HPTE encoding */ | |
125 | unsigned int tlbiel; /* tlbiel supported for that page size */ | |
126 | unsigned long avpnm; /* bits to mask out in AVPN in the HPTE */ | |
127 | unsigned long sllp; /* SLB L||LP (exact mask to use in slbmte) */ | |
128 | }; | |
1da177e4 | 129 | |
3c726f8d | 130 | #endif /* __ASSEMBLY__ */ |
1da177e4 | 131 | |
3c726f8d BH |
132 | /* |
133 | * The kernel use the constants below to index in the page sizes array. | |
134 | * The use of fixed constants for this purpose is better for performances | |
135 | * of the low level hash refill handlers. | |
136 | * | |
137 | * A non supported page size has a "shift" field set to 0 | |
138 | * | |
139 | * Any new page size being implemented can get a new entry in here. Whether | |
140 | * the kernel will use it or not is a different matter though. The actual page | |
141 | * size used by hugetlbfs is not defined here and may be made variable | |
142 | */ | |
1da177e4 | 143 | |
3c726f8d BH |
144 | #define MMU_PAGE_4K 0 /* 4K */ |
145 | #define MMU_PAGE_64K 1 /* 64K */ | |
146 | #define MMU_PAGE_64K_AP 2 /* 64K Admixed (in a 4K segment) */ | |
147 | #define MMU_PAGE_1M 3 /* 1M */ | |
148 | #define MMU_PAGE_16M 4 /* 16M */ | |
149 | #define MMU_PAGE_16G 5 /* 16G */ | |
150 | #define MMU_PAGE_COUNT 6 | |
1da177e4 | 151 | |
3c726f8d | 152 | #ifndef __ASSEMBLY__ |
1da177e4 | 153 | |
3c726f8d BH |
154 | /* |
155 | * The current system page sizes | |
156 | */ | |
157 | extern struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT]; | |
158 | extern int mmu_linear_psize; | |
159 | extern int mmu_virtual_psize; | |
f4c82d51 | 160 | |
3c726f8d BH |
161 | #ifdef CONFIG_HUGETLB_PAGE |
162 | /* | |
163 | * The page size index of the huge pages for use by hugetlbfs | |
164 | */ | |
165 | extern int mmu_huge_psize; | |
f4c82d51 | 166 | |
3c726f8d | 167 | #endif /* CONFIG_HUGETLB_PAGE */ |
f4c82d51 | 168 | |
3c726f8d BH |
169 | /* |
170 | * This function sets the AVPN and L fields of the HPTE appropriately | |
171 | * for the page size | |
172 | */ | |
173 | static inline unsigned long hpte_encode_v(unsigned long va, int psize) | |
174 | { | |
175 | unsigned long v = | |
176 | v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm); | |
177 | v <<= HPTE_V_AVPN_SHIFT; | |
178 | if (psize != MMU_PAGE_4K) | |
179 | v |= HPTE_V_LARGE; | |
180 | return v; | |
181 | } | |
f4c82d51 | 182 | |
3c726f8d BH |
183 | /* |
184 | * This function sets the ARPN, and LP fields of the HPTE appropriately | |
185 | * for the page size. We assume the pa is already "clean" that is properly | |
186 | * aligned for the requested page size | |
187 | */ | |
188 | static inline unsigned long hpte_encode_r(unsigned long pa, int psize) | |
189 | { | |
190 | unsigned long r; | |
f4c82d51 | 191 | |
3c726f8d BH |
192 | /* A 4K page needs no special encoding */ |
193 | if (psize == MMU_PAGE_4K) | |
194 | return pa & HPTE_R_RPN; | |
195 | else { | |
196 | unsigned int penc = mmu_psize_defs[psize].penc; | |
197 | unsigned int shift = mmu_psize_defs[psize].shift; | |
198 | return (pa & ~((1ul << shift) - 1)) | (penc << 12); | |
f4c82d51 | 199 | } |
3c726f8d | 200 | return r; |
f4c82d51 S |
201 | } |
202 | ||
1da177e4 | 203 | /* |
3c726f8d | 204 | * This hashes a virtual address for a 256Mb segment only for now |
1da177e4 | 205 | */ |
3c726f8d BH |
206 | |
207 | static inline unsigned long hpt_hash(unsigned long va, unsigned int shift) | |
208 | { | |
209 | return ((va >> 28) & 0x7fffffffffUL) ^ ((va & 0x0fffffffUL) >> shift); | |
210 | } | |
211 | ||
212 | extern int __hash_page_4K(unsigned long ea, unsigned long access, | |
213 | unsigned long vsid, pte_t *ptep, unsigned long trap, | |
214 | unsigned int local); | |
215 | extern int __hash_page_64K(unsigned long ea, unsigned long access, | |
216 | unsigned long vsid, pte_t *ptep, unsigned long trap, | |
217 | unsigned int local); | |
218 | struct mm_struct; | |
219 | extern int hash_huge_page(struct mm_struct *mm, unsigned long access, | |
220 | unsigned long ea, unsigned long vsid, int local); | |
1da177e4 LT |
221 | |
222 | extern void htab_finish_init(void); | |
3c726f8d BH |
223 | extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend, |
224 | unsigned long pstart, unsigned long mode, | |
225 | int psize); | |
1da177e4 | 226 | |
1f8d419e DG |
227 | extern void hpte_init_native(void); |
228 | extern void hpte_init_lpar(void); | |
229 | extern void hpte_init_iSeries(void); | |
230 | ||
231 | extern long pSeries_lpar_hpte_insert(unsigned long hpte_group, | |
232 | unsigned long va, unsigned long prpn, | |
3c726f8d BH |
233 | unsigned long rflags, |
234 | unsigned long vflags, int psize); | |
235 | ||
236 | extern long native_hpte_insert(unsigned long hpte_group, | |
237 | unsigned long va, unsigned long prpn, | |
238 | unsigned long rflags, | |
239 | unsigned long vflags, int psize); | |
1f8d419e | 240 | |
3c726f8d BH |
241 | extern long iSeries_hpte_insert(unsigned long hpte_group, |
242 | unsigned long va, unsigned long prpn, | |
243 | unsigned long rflags, | |
244 | unsigned long vflags, int psize); | |
4c55130b | 245 | |
533f0817 | 246 | extern void stabs_alloc(void); |
3c726f8d | 247 | extern void slb_initialize(void); |
533f0817 | 248 | |
1da177e4 LT |
249 | #endif /* __ASSEMBLY__ */ |
250 | ||
251 | /* | |
1f8d419e DG |
252 | * VSID allocation |
253 | * | |
254 | * We first generate a 36-bit "proto-VSID". For kernel addresses this | |
255 | * is equal to the ESID, for user addresses it is: | |
256 | * (context << 15) | (esid & 0x7fff) | |
257 | * | |
258 | * The two forms are distinguishable because the top bit is 0 for user | |
259 | * addresses, whereas the top two bits are 1 for kernel addresses. | |
260 | * Proto-VSIDs with the top two bits equal to 0b10 are reserved for | |
261 | * now. | |
262 | * | |
263 | * The proto-VSIDs are then scrambled into real VSIDs with the | |
264 | * multiplicative hash: | |
265 | * | |
266 | * VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS | |
267 | * where VSID_MULTIPLIER = 268435399 = 0xFFFFFC7 | |
268 | * VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF | |
269 | * | |
270 | * This scramble is only well defined for proto-VSIDs below | |
271 | * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are | |
272 | * reserved. VSID_MULTIPLIER is prime, so in particular it is | |
273 | * co-prime to VSID_MODULUS, making this a 1:1 scrambling function. | |
274 | * Because the modulus is 2^n-1 we can compute it efficiently without | |
275 | * a divide or extra multiply (see below). | |
276 | * | |
277 | * This scheme has several advantages over older methods: | |
278 | * | |
279 | * - We have VSIDs allocated for every kernel address | |
280 | * (i.e. everything above 0xC000000000000000), except the very top | |
281 | * segment, which simplifies several things. | |
282 | * | |
283 | * - We allow for 15 significant bits of ESID and 20 bits of | |
284 | * context for user addresses. i.e. 8T (43 bits) of address space for | |
285 | * up to 1M contexts (although the page table structure and context | |
286 | * allocation will need changes to take advantage of this). | |
287 | * | |
288 | * - The scramble function gives robust scattering in the hash | |
289 | * table (at least based on some initial results). The previous | |
290 | * method was more susceptible to pathological cases giving excessive | |
291 | * hash collisions. | |
292 | */ | |
293 | /* | |
294 | * WARNING - If you change these you must make sure the asm | |
295 | * implementations in slb_allocate (slb_low.S), do_stab_bolted | |
296 | * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly. | |
297 | * | |
298 | * You'll also need to change the precomputed VSID values in head.S | |
299 | * which are used by the iSeries firmware. | |
1da177e4 | 300 | */ |
1da177e4 LT |
301 | |
302 | #define VSID_MULTIPLIER ASM_CONST(200730139) /* 28-bit prime */ | |
303 | #define VSID_BITS 36 | |
304 | #define VSID_MODULUS ((1UL<<VSID_BITS)-1) | |
305 | ||
e28f7faf DG |
306 | #define CONTEXT_BITS 19 |
307 | #define USER_ESID_BITS 16 | |
308 | ||
309 | #define USER_VSID_RANGE (1UL << (USER_ESID_BITS + SID_SHIFT)) | |
1da177e4 LT |
310 | |
311 | /* | |
312 | * This macro generates asm code to compute the VSID scramble | |
313 | * function. Used in slb_allocate() and do_stab_bolted. The function | |
314 | * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS | |
315 | * | |
316 | * rt = register continaing the proto-VSID and into which the | |
317 | * VSID will be stored | |
318 | * rx = scratch register (clobbered) | |
319 | * | |
320 | * - rt and rx must be different registers | |
321 | * - The answer will end up in the low 36 bits of rt. The higher | |
322 | * bits may contain other garbage, so you may need to mask the | |
323 | * result. | |
324 | */ | |
325 | #define ASM_VSID_SCRAMBLE(rt, rx) \ | |
326 | lis rx,VSID_MULTIPLIER@h; \ | |
327 | ori rx,rx,VSID_MULTIPLIER@l; \ | |
328 | mulld rt,rt,rx; /* rt = rt * MULTIPLIER */ \ | |
329 | \ | |
330 | srdi rx,rt,VSID_BITS; \ | |
331 | clrldi rt,rt,(64-VSID_BITS); \ | |
332 | add rt,rt,rx; /* add high and low bits */ \ | |
333 | /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and \ | |
334 | * 2^36-1+2^28-1. That in particular means that if r3 >= \ | |
335 | * 2^36-1, then r3+1 has the 2^36 bit set. So, if r3+1 has \ | |
336 | * the bit clear, r3 already has the answer we want, if it \ | |
337 | * doesn't, the answer is the low 36 bits of r3+1. So in all \ | |
338 | * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\ | |
339 | addi rx,rt,1; \ | |
340 | srdi rx,rx,VSID_BITS; /* extract 2^36 bit */ \ | |
341 | add rt,rt,rx | |
342 | ||
1f8d419e DG |
343 | |
344 | #ifndef __ASSEMBLY__ | |
345 | ||
346 | typedef unsigned long mm_context_id_t; | |
347 | ||
348 | typedef struct { | |
349 | mm_context_id_t id; | |
350 | #ifdef CONFIG_HUGETLB_PAGE | |
c594adad | 351 | u16 low_htlb_areas, high_htlb_areas; |
1f8d419e DG |
352 | #endif |
353 | } mm_context_t; | |
354 | ||
355 | ||
356 | static inline unsigned long vsid_scramble(unsigned long protovsid) | |
357 | { | |
358 | #if 0 | |
359 | /* The code below is equivalent to this function for arguments | |
360 | * < 2^VSID_BITS, which is all this should ever be called | |
361 | * with. However gcc is not clever enough to compute the | |
362 | * modulus (2^n-1) without a second multiply. */ | |
363 | return ((protovsid * VSID_MULTIPLIER) % VSID_MODULUS); | |
364 | #else /* 1 */ | |
365 | unsigned long x; | |
366 | ||
367 | x = protovsid * VSID_MULTIPLIER; | |
368 | x = (x >> VSID_BITS) + (x & VSID_MODULUS); | |
369 | return (x + ((x+1) >> VSID_BITS)) & VSID_MODULUS; | |
370 | #endif /* 1 */ | |
371 | } | |
372 | ||
373 | /* This is only valid for addresses >= KERNELBASE */ | |
374 | static inline unsigned long get_kernel_vsid(unsigned long ea) | |
375 | { | |
376 | return vsid_scramble(ea >> SID_SHIFT); | |
377 | } | |
378 | ||
379 | /* This is only valid for user addresses (which are below 2^41) */ | |
380 | static inline unsigned long get_vsid(unsigned long context, unsigned long ea) | |
381 | { | |
382 | return vsid_scramble((context << USER_ESID_BITS) | |
383 | | (ea >> SID_SHIFT)); | |
384 | } | |
385 | ||
488f8499 DG |
386 | #define VSID_SCRAMBLE(pvsid) (((pvsid) * VSID_MULTIPLIER) % VSID_MODULUS) |
387 | #define KERNEL_VSID(ea) VSID_SCRAMBLE(GET_ESID(ea)) | |
388 | ||
1f8d419e DG |
389 | #endif /* __ASSEMBLY */ |
390 | ||
1da177e4 | 391 | #endif /* _PPC64_MMU_H_ */ |