diff options
author | David Gibson <david@gibson.dropbear.id.au> | 2007-04-26 21:53:52 -0400 |
---|---|---|
committer | Paul Mackerras <paulus@samba.org> | 2007-04-27 07:14:26 -0400 |
commit | 8d2169e8d6b8a91413df33bc402e0f602ceaabcc (patch) | |
tree | 4de3e19d8cd9a09049e531e4cc1f02b2328b943c | |
parent | 173ba87b9584e4cba41ce9a06916eba80baa1bf4 (diff) |
[POWERPC] Prepare for splitting up mmu.h by MMU type
Currently asm-powerpc/mmu.h has definitions for the 64-bit hash based
MMU. If CONFIG_PPC64 is not set, it instead includes asm-ppc/mmu.h
which contains a particularly horrible mess of #ifdefs giving the
definitions for all the various 32-bit MMUs.
It would be nice to have the low level definitions for each MMU type
neatly in their own separate files. It would also be good to wean
arch/powerpc off dependence on the old asm-ppc/mmu.h.
This patch makes a start on such a cleanup by moving the definitions
for the 64-bit hash MMU to their own file, asm-powerpc/mmu_hash64.h.
Definitions for the other MMUs still all come from asm-ppc/mmu.h,
however each MMU type can now be one-by-one moved over to their own
file, in the process cleaning them up stripping them of cruft no
longer necessary in arch/powerpc.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
-rw-r--r-- | include/asm-powerpc/mmu-hash64.h | 400 | ||||
-rw-r--r-- | include/asm-powerpc/mmu.h | 406 |
2 files changed, 406 insertions, 400 deletions
diff --git a/include/asm-powerpc/mmu-hash64.h b/include/asm-powerpc/mmu-hash64.h new file mode 100644 index 000000000000..6739457d8bc0 --- /dev/null +++ b/include/asm-powerpc/mmu-hash64.h | |||
@@ -0,0 +1,400 @@ | |||
1 | #ifndef _ASM_POWERPC_MMU_HASH64_H_ | ||
2 | #define _ASM_POWERPC_MMU_HASH64_H_ | ||
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> | ||
17 | |||
18 | /* | ||
19 | * Segment table | ||
20 | */ | ||
21 | |||
22 | #define STE_ESID_V 0x80 | ||
23 | #define STE_ESID_KS 0x20 | ||
24 | #define STE_ESID_KP 0x10 | ||
25 | #define STE_ESID_N 0x08 | ||
26 | |||
27 | #define STE_VSID_SHIFT 12 | ||
28 | |||
29 | /* Location of cpu0's segment table */ | ||
30 | #define STAB0_PAGE 0x6 | ||
31 | #define STAB0_OFFSET (STAB0_PAGE << 12) | ||
32 | #define STAB0_PHYS_ADDR (STAB0_OFFSET + PHYSICAL_START) | ||
33 | |||
34 | #ifndef __ASSEMBLY__ | ||
35 | extern char initial_stab[]; | ||
36 | #endif /* ! __ASSEMBLY */ | ||
37 | |||
38 | /* | ||
39 | * SLB | ||
40 | */ | ||
41 | |||
42 | #define SLB_NUM_BOLTED 3 | ||
43 | #define SLB_CACHE_ENTRIES 8 | ||
44 | |||
45 | /* Bits in the SLB ESID word */ | ||
46 | #define SLB_ESID_V ASM_CONST(0x0000000008000000) /* valid */ | ||
47 | |||
48 | /* Bits in the SLB VSID word */ | ||
49 | #define SLB_VSID_SHIFT 12 | ||
50 | #define SLB_VSID_B ASM_CONST(0xc000000000000000) | ||
51 | #define SLB_VSID_B_256M ASM_CONST(0x0000000000000000) | ||
52 | #define SLB_VSID_B_1T ASM_CONST(0x4000000000000000) | ||
53 | #define SLB_VSID_KS ASM_CONST(0x0000000000000800) | ||
54 | #define SLB_VSID_KP ASM_CONST(0x0000000000000400) | ||
55 | #define SLB_VSID_N ASM_CONST(0x0000000000000200) /* no-execute */ | ||
56 | #define SLB_VSID_L ASM_CONST(0x0000000000000100) | ||
57 | #define SLB_VSID_C ASM_CONST(0x0000000000000080) /* class */ | ||
58 | #define SLB_VSID_LP ASM_CONST(0x0000000000000030) | ||
59 | #define SLB_VSID_LP_00 ASM_CONST(0x0000000000000000) | ||
60 | #define SLB_VSID_LP_01 ASM_CONST(0x0000000000000010) | ||
61 | #define SLB_VSID_LP_10 ASM_CONST(0x0000000000000020) | ||
62 | #define SLB_VSID_LP_11 ASM_CONST(0x0000000000000030) | ||
63 | #define SLB_VSID_LLP (SLB_VSID_L|SLB_VSID_LP) | ||
64 | |||
65 | #define SLB_VSID_KERNEL (SLB_VSID_KP) | ||
66 | #define SLB_VSID_USER (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C) | ||
67 | |||
68 | #define SLBIE_C (0x08000000) | ||
69 | |||
70 | /* | ||
71 | * Hash table | ||
72 | */ | ||
73 | |||
74 | #define HPTES_PER_GROUP 8 | ||
75 | |||
76 | #define HPTE_V_AVPN_SHIFT 7 | ||
77 | #define HPTE_V_AVPN ASM_CONST(0xffffffffffffff80) | ||
78 | #define HPTE_V_AVPN_VAL(x) (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT) | ||
79 | #define HPTE_V_COMPARE(x,y) (!(((x) ^ (y)) & HPTE_V_AVPN)) | ||
80 | #define HPTE_V_BOLTED ASM_CONST(0x0000000000000010) | ||
81 | #define HPTE_V_LOCK ASM_CONST(0x0000000000000008) | ||
82 | #define HPTE_V_LARGE ASM_CONST(0x0000000000000004) | ||
83 | #define HPTE_V_SECONDARY ASM_CONST(0x0000000000000002) | ||
84 | #define HPTE_V_VALID ASM_CONST(0x0000000000000001) | ||
85 | |||
86 | #define HPTE_R_PP0 ASM_CONST(0x8000000000000000) | ||
87 | #define HPTE_R_TS ASM_CONST(0x4000000000000000) | ||
88 | #define HPTE_R_RPN_SHIFT 12 | ||
89 | #define HPTE_R_RPN ASM_CONST(0x3ffffffffffff000) | ||
90 | #define HPTE_R_FLAGS ASM_CONST(0x00000000000003ff) | ||
91 | #define HPTE_R_PP ASM_CONST(0x0000000000000003) | ||
92 | #define HPTE_R_N ASM_CONST(0x0000000000000004) | ||
93 | #define HPTE_R_C ASM_CONST(0x0000000000000080) | ||
94 | #define HPTE_R_R ASM_CONST(0x0000000000000100) | ||
95 | |||
96 | /* Values for PP (assumes Ks=0, Kp=1) */ | ||
97 | /* pp0 will always be 0 for linux */ | ||
98 | #define PP_RWXX 0 /* Supervisor read/write, User none */ | ||
99 | #define PP_RWRX 1 /* Supervisor read/write, User read */ | ||
100 | #define PP_RWRW 2 /* Supervisor read/write, User read/write */ | ||
101 | #define PP_RXRX 3 /* Supervisor read, User read */ | ||
102 | |||
103 | #ifndef __ASSEMBLY__ | ||
104 | |||
105 | typedef struct { | ||
106 | unsigned long v; | ||
107 | unsigned long r; | ||
108 | } hpte_t; | ||
109 | |||
110 | extern hpte_t *htab_address; | ||
111 | extern unsigned long htab_size_bytes; | ||
112 | extern unsigned long htab_hash_mask; | ||
113 | |||
114 | /* | ||
115 | * Page size definition | ||
116 | * | ||
117 | * shift : is the "PAGE_SHIFT" value for that page size | ||
118 | * sllp : is a bit mask with the value of SLB L || LP to be or'ed | ||
119 | * directly to a slbmte "vsid" value | ||
120 | * penc : is the HPTE encoding mask for the "LP" field: | ||
121 | * | ||
122 | */ | ||
123 | struct mmu_psize_def | ||
124 | { | ||
125 | unsigned int shift; /* number of bits */ | ||
126 | unsigned int penc; /* HPTE encoding */ | ||
127 | unsigned int tlbiel; /* tlbiel supported for that page size */ | ||
128 | unsigned long avpnm; /* bits to mask out in AVPN in the HPTE */ | ||
129 | unsigned long sllp; /* SLB L||LP (exact mask to use in slbmte) */ | ||
130 | }; | ||
131 | |||
132 | #endif /* __ASSEMBLY__ */ | ||
133 | |||
134 | /* | ||
135 | * The kernel use the constants below to index in the page sizes array. | ||
136 | * The use of fixed constants for this purpose is better for performances | ||
137 | * of the low level hash refill handlers. | ||
138 | * | ||
139 | * A non supported page size has a "shift" field set to 0 | ||
140 | * | ||
141 | * Any new page size being implemented can get a new entry in here. Whether | ||
142 | * the kernel will use it or not is a different matter though. The actual page | ||
143 | * size used by hugetlbfs is not defined here and may be made variable | ||
144 | */ | ||
145 | |||
146 | #define MMU_PAGE_4K 0 /* 4K */ | ||
147 | #define MMU_PAGE_64K 1 /* 64K */ | ||
148 | #define MMU_PAGE_64K_AP 2 /* 64K Admixed (in a 4K segment) */ | ||
149 | #define MMU_PAGE_1M 3 /* 1M */ | ||
150 | #define MMU_PAGE_16M 4 /* 16M */ | ||
151 | #define MMU_PAGE_16G 5 /* 16G */ | ||
152 | #define MMU_PAGE_COUNT 6 | ||
153 | |||
154 | #ifndef __ASSEMBLY__ | ||
155 | |||
156 | /* | ||
157 | * The current system page sizes | ||
158 | */ | ||
159 | extern struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT]; | ||
160 | extern int mmu_linear_psize; | ||
161 | extern int mmu_virtual_psize; | ||
162 | extern int mmu_vmalloc_psize; | ||
163 | extern int mmu_io_psize; | ||
164 | |||
165 | /* | ||
166 | * If the processor supports 64k normal pages but not 64k cache | ||
167 | * inhibited pages, we have to be prepared to switch processes | ||
168 | * to use 4k pages when they create cache-inhibited mappings. | ||
169 | * If this is the case, mmu_ci_restrictions will be set to 1. | ||
170 | */ | ||
171 | extern int mmu_ci_restrictions; | ||
172 | |||
173 | #ifdef CONFIG_HUGETLB_PAGE | ||
174 | /* | ||
175 | * The page size index of the huge pages for use by hugetlbfs | ||
176 | */ | ||
177 | extern int mmu_huge_psize; | ||
178 | |||
179 | #endif /* CONFIG_HUGETLB_PAGE */ | ||
180 | |||
181 | /* | ||
182 | * This function sets the AVPN and L fields of the HPTE appropriately | ||
183 | * for the page size | ||
184 | */ | ||
185 | static inline unsigned long hpte_encode_v(unsigned long va, int psize) | ||
186 | { | ||
187 | unsigned long v = | ||
188 | v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm); | ||
189 | v <<= HPTE_V_AVPN_SHIFT; | ||
190 | if (psize != MMU_PAGE_4K) | ||
191 | v |= HPTE_V_LARGE; | ||
192 | return v; | ||
193 | } | ||
194 | |||
195 | /* | ||
196 | * This function sets the ARPN, and LP fields of the HPTE appropriately | ||
197 | * for the page size. We assume the pa is already "clean" that is properly | ||
198 | * aligned for the requested page size | ||
199 | */ | ||
200 | static inline unsigned long hpte_encode_r(unsigned long pa, int psize) | ||
201 | { | ||
202 | unsigned long r; | ||
203 | |||
204 | /* A 4K page needs no special encoding */ | ||
205 | if (psize == MMU_PAGE_4K) | ||
206 | return pa & HPTE_R_RPN; | ||
207 | else { | ||
208 | unsigned int penc = mmu_psize_defs[psize].penc; | ||
209 | unsigned int shift = mmu_psize_defs[psize].shift; | ||
210 | return (pa & ~((1ul << shift) - 1)) | (penc << 12); | ||
211 | } | ||
212 | return r; | ||
213 | } | ||
214 | |||
215 | /* | ||
216 | * This hashes a virtual address for a 256Mb segment only for now | ||
217 | */ | ||
218 | |||
219 | static inline unsigned long hpt_hash(unsigned long va, unsigned int shift) | ||
220 | { | ||
221 | return ((va >> 28) & 0x7fffffffffUL) ^ ((va & 0x0fffffffUL) >> shift); | ||
222 | } | ||
223 | |||
224 | extern int __hash_page_4K(unsigned long ea, unsigned long access, | ||
225 | unsigned long vsid, pte_t *ptep, unsigned long trap, | ||
226 | unsigned int local); | ||
227 | extern int __hash_page_64K(unsigned long ea, unsigned long access, | ||
228 | unsigned long vsid, pte_t *ptep, unsigned long trap, | ||
229 | unsigned int local); | ||
230 | struct mm_struct; | ||
231 | extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap); | ||
232 | extern int hash_huge_page(struct mm_struct *mm, unsigned long access, | ||
233 | unsigned long ea, unsigned long vsid, int local, | ||
234 | unsigned long trap); | ||
235 | |||
236 | extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend, | ||
237 | unsigned long pstart, unsigned long mode, | ||
238 | int psize); | ||
239 | |||
240 | extern void htab_initialize(void); | ||
241 | extern void htab_initialize_secondary(void); | ||
242 | extern void hpte_init_native(void); | ||
243 | extern void hpte_init_lpar(void); | ||
244 | extern void hpte_init_iSeries(void); | ||
245 | extern void hpte_init_beat(void); | ||
246 | |||
247 | extern void stabs_alloc(void); | ||
248 | extern void slb_initialize(void); | ||
249 | extern void slb_flush_and_rebolt(void); | ||
250 | extern void stab_initialize(unsigned long stab); | ||
251 | |||
252 | #endif /* __ASSEMBLY__ */ | ||
253 | |||
254 | /* | ||
255 | * VSID allocation | ||
256 | * | ||
257 | * We first generate a 36-bit "proto-VSID". For kernel addresses this | ||
258 | * is equal to the ESID, for user addresses it is: | ||
259 | * (context << 15) | (esid & 0x7fff) | ||
260 | * | ||
261 | * The two forms are distinguishable because the top bit is 0 for user | ||
262 | * addresses, whereas the top two bits are 1 for kernel addresses. | ||
263 | * Proto-VSIDs with the top two bits equal to 0b10 are reserved for | ||
264 | * now. | ||
265 | * | ||
266 | * The proto-VSIDs are then scrambled into real VSIDs with the | ||
267 | * multiplicative hash: | ||
268 | * | ||
269 | * VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS | ||
270 | * where VSID_MULTIPLIER = 268435399 = 0xFFFFFC7 | ||
271 | * VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF | ||
272 | * | ||
273 | * This scramble is only well defined for proto-VSIDs below | ||
274 | * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are | ||
275 | * reserved. VSID_MULTIPLIER is prime, so in particular it is | ||
276 | * co-prime to VSID_MODULUS, making this a 1:1 scrambling function. | ||
277 | * Because the modulus is 2^n-1 we can compute it efficiently without | ||
278 | * a divide or extra multiply (see below). | ||
279 | * | ||
280 | * This scheme has several advantages over older methods: | ||
281 | * | ||
282 | * - We have VSIDs allocated for every kernel address | ||
283 | * (i.e. everything above 0xC000000000000000), except the very top | ||
284 | * segment, which simplifies several things. | ||
285 | * | ||
286 | * - We allow for 15 significant bits of ESID and 20 bits of | ||
287 | * context for user addresses. i.e. 8T (43 bits) of address space for | ||
288 | * up to 1M contexts (although the page table structure and context | ||
289 | * allocation will need changes to take advantage of this). | ||
290 | * | ||
291 | * - The scramble function gives robust scattering in the hash | ||
292 | * table (at least based on some initial results). The previous | ||
293 | * method was more susceptible to pathological cases giving excessive | ||
294 | * hash collisions. | ||
295 | */ | ||
296 | /* | ||
297 | * WARNING - If you change these you must make sure the asm | ||
298 | * implementations in slb_allocate (slb_low.S), do_stab_bolted | ||
299 | * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly. | ||
300 | * | ||
301 | * You'll also need to change the precomputed VSID values in head.S | ||
302 | * which are used by the iSeries firmware. | ||
303 | */ | ||
304 | |||
305 | #define VSID_MULTIPLIER ASM_CONST(200730139) /* 28-bit prime */ | ||
306 | #define VSID_BITS 36 | ||
307 | #define VSID_MODULUS ((1UL<<VSID_BITS)-1) | ||
308 | |||
309 | #define CONTEXT_BITS 19 | ||
310 | #define USER_ESID_BITS 16 | ||
311 | |||
312 | #define USER_VSID_RANGE (1UL << (USER_ESID_BITS + SID_SHIFT)) | ||
313 | |||
314 | /* | ||
315 | * This macro generates asm code to compute the VSID scramble | ||
316 | * function. Used in slb_allocate() and do_stab_bolted. The function | ||
317 | * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS | ||
318 | * | ||
319 | * rt = register continaing the proto-VSID and into which the | ||
320 | * VSID will be stored | ||
321 | * rx = scratch register (clobbered) | ||
322 | * | ||
323 | * - rt and rx must be different registers | ||
324 | * - The answer will end up in the low 36 bits of rt. The higher | ||
325 | * bits may contain other garbage, so you may need to mask the | ||
326 | * result. | ||
327 | */ | ||
328 | #define ASM_VSID_SCRAMBLE(rt, rx) \ | ||
329 | lis rx,VSID_MULTIPLIER@h; \ | ||
330 | ori rx,rx,VSID_MULTIPLIER@l; \ | ||
331 | mulld rt,rt,rx; /* rt = rt * MULTIPLIER */ \ | ||
332 | \ | ||
333 | srdi rx,rt,VSID_BITS; \ | ||
334 | clrldi rt,rt,(64-VSID_BITS); \ | ||
335 | add rt,rt,rx; /* add high and low bits */ \ | ||
336 | /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and \ | ||
337 | * 2^36-1+2^28-1. That in particular means that if r3 >= \ | ||
338 | * 2^36-1, then r3+1 has the 2^36 bit set. So, if r3+1 has \ | ||
339 | * the bit clear, r3 already has the answer we want, if it \ | ||
340 | * doesn't, the answer is the low 36 bits of r3+1. So in all \ | ||
341 | * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\ | ||
342 | addi rx,rt,1; \ | ||
343 | srdi rx,rx,VSID_BITS; /* extract 2^36 bit */ \ | ||
344 | add rt,rt,rx | ||
345 | |||
346 | |||
347 | #ifndef __ASSEMBLY__ | ||
348 | |||
349 | typedef unsigned long mm_context_id_t; | ||
350 | |||
351 | typedef struct { | ||
352 | mm_context_id_t id; | ||
353 | u16 user_psize; /* page size index */ | ||
354 | u16 sllp; /* SLB entry page size encoding */ | ||
355 | #ifdef CONFIG_HUGETLB_PAGE | ||
356 | u16 low_htlb_areas, high_htlb_areas; | ||
357 | #endif | ||
358 | unsigned long vdso_base; | ||
359 | } mm_context_t; | ||
360 | |||
361 | |||
362 | static inline unsigned long vsid_scramble(unsigned long protovsid) | ||
363 | { | ||
364 | #if 0 | ||
365 | /* The code below is equivalent to this function for arguments | ||
366 | * < 2^VSID_BITS, which is all this should ever be called | ||
367 | * with. However gcc is not clever enough to compute the | ||
368 | * modulus (2^n-1) without a second multiply. */ | ||
369 | return ((protovsid * VSID_MULTIPLIER) % VSID_MODULUS); | ||
370 | #else /* 1 */ | ||
371 | unsigned long x; | ||
372 | |||
373 | x = protovsid * VSID_MULTIPLIER; | ||
374 | x = (x >> VSID_BITS) + (x & VSID_MODULUS); | ||
375 | return (x + ((x+1) >> VSID_BITS)) & VSID_MODULUS; | ||
376 | #endif /* 1 */ | ||
377 | } | ||
378 | |||
379 | /* This is only valid for addresses >= KERNELBASE */ | ||
380 | static inline unsigned long get_kernel_vsid(unsigned long ea) | ||
381 | { | ||
382 | return vsid_scramble(ea >> SID_SHIFT); | ||
383 | } | ||
384 | |||
385 | /* This is only valid for user addresses (which are below 2^41) */ | ||
386 | static inline unsigned long get_vsid(unsigned long context, unsigned long ea) | ||
387 | { | ||
388 | return vsid_scramble((context << USER_ESID_BITS) | ||
389 | | (ea >> SID_SHIFT)); | ||
390 | } | ||
391 | |||
392 | #define VSID_SCRAMBLE(pvsid) (((pvsid) * VSID_MULTIPLIER) % VSID_MODULUS) | ||
393 | #define KERNEL_VSID(ea) VSID_SCRAMBLE(GET_ESID(ea)) | ||
394 | |||
395 | /* Physical address used by some IO functions */ | ||
396 | typedef unsigned long phys_addr_t; | ||
397 | |||
398 | #endif /* __ASSEMBLY__ */ | ||
399 | |||
400 | #endif /* _ASM_POWERPC_MMU_HASH64_H_ */ | ||
diff --git a/include/asm-powerpc/mmu.h b/include/asm-powerpc/mmu.h index e22fd8811505..06b3e6d336cb 100644 --- a/include/asm-powerpc/mmu.h +++ b/include/asm-powerpc/mmu.h | |||
@@ -2,408 +2,14 @@ | |||
2 | #define _ASM_POWERPC_MMU_H_ | 2 | #define _ASM_POWERPC_MMU_H_ |
3 | #ifdef __KERNEL__ | 3 | #ifdef __KERNEL__ |
4 | 4 | ||
5 | #ifndef CONFIG_PPC64 | 5 | #ifdef CONFIG_PPC64 |
6 | #include <asm-ppc/mmu.h> | 6 | /* 64-bit classic hash table MMU */ |
7 | # include <asm/mmu-hash64.h> | ||
7 | #else | 8 | #else |
8 | 9 | /* 32-bit. FIXME: split up the 32-bit MMU types, and revise for | |
9 | /* | 10 | * arch/powerpc */ |
10 | * PowerPC memory management structures | 11 | # include <asm-ppc/mmu.h> |
11 | * | ||
12 | * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com> | ||
13 | * PPC64 rework. | ||
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 | #include <asm/asm-compat.h> | ||
22 | #include <asm/page.h> | ||
23 | |||
24 | /* | ||
25 | * Segment table | ||
26 | */ | ||
27 | |||
28 | #define STE_ESID_V 0x80 | ||
29 | #define STE_ESID_KS 0x20 | ||
30 | #define STE_ESID_KP 0x10 | ||
31 | #define STE_ESID_N 0x08 | ||
32 | |||
33 | #define STE_VSID_SHIFT 12 | ||
34 | |||
35 | /* Location of cpu0's segment table */ | ||
36 | #define STAB0_PAGE 0x6 | ||
37 | #define STAB0_OFFSET (STAB0_PAGE << 12) | ||
38 | #define STAB0_PHYS_ADDR (STAB0_OFFSET + PHYSICAL_START) | ||
39 | |||
40 | #ifndef __ASSEMBLY__ | ||
41 | extern char initial_stab[]; | ||
42 | #endif /* ! __ASSEMBLY */ | ||
43 | |||
44 | /* | ||
45 | * SLB | ||
46 | */ | ||
47 | |||
48 | #define SLB_NUM_BOLTED 3 | ||
49 | #define SLB_CACHE_ENTRIES 8 | ||
50 | |||
51 | /* Bits in the SLB ESID word */ | ||
52 | #define SLB_ESID_V ASM_CONST(0x0000000008000000) /* valid */ | ||
53 | |||
54 | /* Bits in the SLB VSID word */ | ||
55 | #define SLB_VSID_SHIFT 12 | ||
56 | #define SLB_VSID_B ASM_CONST(0xc000000000000000) | ||
57 | #define SLB_VSID_B_256M ASM_CONST(0x0000000000000000) | ||
58 | #define SLB_VSID_B_1T ASM_CONST(0x4000000000000000) | ||
59 | #define SLB_VSID_KS ASM_CONST(0x0000000000000800) | ||
60 | #define SLB_VSID_KP ASM_CONST(0x0000000000000400) | ||
61 | #define SLB_VSID_N ASM_CONST(0x0000000000000200) /* no-execute */ | ||
62 | #define SLB_VSID_L ASM_CONST(0x0000000000000100) | ||
63 | #define SLB_VSID_C ASM_CONST(0x0000000000000080) /* class */ | ||
64 | #define SLB_VSID_LP ASM_CONST(0x0000000000000030) | ||
65 | #define SLB_VSID_LP_00 ASM_CONST(0x0000000000000000) | ||
66 | #define SLB_VSID_LP_01 ASM_CONST(0x0000000000000010) | ||
67 | #define SLB_VSID_LP_10 ASM_CONST(0x0000000000000020) | ||
68 | #define SLB_VSID_LP_11 ASM_CONST(0x0000000000000030) | ||
69 | #define SLB_VSID_LLP (SLB_VSID_L|SLB_VSID_LP) | ||
70 | |||
71 | #define SLB_VSID_KERNEL (SLB_VSID_KP) | ||
72 | #define SLB_VSID_USER (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C) | ||
73 | |||
74 | #define SLBIE_C (0x08000000) | ||
75 | |||
76 | /* | ||
77 | * Hash table | ||
78 | */ | ||
79 | |||
80 | #define HPTES_PER_GROUP 8 | ||
81 | |||
82 | #define HPTE_V_AVPN_SHIFT 7 | ||
83 | #define HPTE_V_AVPN ASM_CONST(0xffffffffffffff80) | ||
84 | #define HPTE_V_AVPN_VAL(x) (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT) | ||
85 | #define HPTE_V_COMPARE(x,y) (!(((x) ^ (y)) & HPTE_V_AVPN)) | ||
86 | #define HPTE_V_BOLTED ASM_CONST(0x0000000000000010) | ||
87 | #define HPTE_V_LOCK ASM_CONST(0x0000000000000008) | ||
88 | #define HPTE_V_LARGE ASM_CONST(0x0000000000000004) | ||
89 | #define HPTE_V_SECONDARY ASM_CONST(0x0000000000000002) | ||
90 | #define HPTE_V_VALID ASM_CONST(0x0000000000000001) | ||
91 | |||
92 | #define HPTE_R_PP0 ASM_CONST(0x8000000000000000) | ||
93 | #define HPTE_R_TS ASM_CONST(0x4000000000000000) | ||
94 | #define HPTE_R_RPN_SHIFT 12 | ||
95 | #define HPTE_R_RPN ASM_CONST(0x3ffffffffffff000) | ||
96 | #define HPTE_R_FLAGS ASM_CONST(0x00000000000003ff) | ||
97 | #define HPTE_R_PP ASM_CONST(0x0000000000000003) | ||
98 | #define HPTE_R_N ASM_CONST(0x0000000000000004) | ||
99 | #define HPTE_R_C ASM_CONST(0x0000000000000080) | ||
100 | #define HPTE_R_R ASM_CONST(0x0000000000000100) | ||
101 | |||
102 | /* Values for PP (assumes Ks=0, Kp=1) */ | ||
103 | /* pp0 will always be 0 for linux */ | ||
104 | #define PP_RWXX 0 /* Supervisor read/write, User none */ | ||
105 | #define PP_RWRX 1 /* Supervisor read/write, User read */ | ||
106 | #define PP_RWRW 2 /* Supervisor read/write, User read/write */ | ||
107 | #define PP_RXRX 3 /* Supervisor read, User read */ | ||
108 | |||
109 | #ifndef __ASSEMBLY__ | ||
110 | |||
111 | typedef struct { | ||
112 | unsigned long v; | ||
113 | unsigned long r; | ||
114 | } hpte_t; | ||
115 | |||
116 | extern hpte_t *htab_address; | ||
117 | extern unsigned long htab_size_bytes; | ||
118 | extern unsigned long htab_hash_mask; | ||
119 | |||
120 | /* | ||
121 | * Page size definition | ||
122 | * | ||
123 | * shift : is the "PAGE_SHIFT" value for that page size | ||
124 | * sllp : is a bit mask with the value of SLB L || LP to be or'ed | ||
125 | * directly to a slbmte "vsid" value | ||
126 | * penc : is the HPTE encoding mask for the "LP" field: | ||
127 | * | ||
128 | */ | ||
129 | struct mmu_psize_def | ||
130 | { | ||
131 | unsigned int shift; /* number of bits */ | ||
132 | unsigned int penc; /* HPTE encoding */ | ||
133 | unsigned int tlbiel; /* tlbiel supported for that page size */ | ||
134 | unsigned long avpnm; /* bits to mask out in AVPN in the HPTE */ | ||
135 | unsigned long sllp; /* SLB L||LP (exact mask to use in slbmte) */ | ||
136 | }; | ||
137 | |||
138 | #endif /* __ASSEMBLY__ */ | ||
139 | |||
140 | /* | ||
141 | * The kernel use the constants below to index in the page sizes array. | ||
142 | * The use of fixed constants for this purpose is better for performances | ||
143 | * of the low level hash refill handlers. | ||
144 | * | ||
145 | * A non supported page size has a "shift" field set to 0 | ||
146 | * | ||
147 | * Any new page size being implemented can get a new entry in here. Whether | ||
148 | * the kernel will use it or not is a different matter though. The actual page | ||
149 | * size used by hugetlbfs is not defined here and may be made variable | ||
150 | */ | ||
151 | |||
152 | #define MMU_PAGE_4K 0 /* 4K */ | ||
153 | #define MMU_PAGE_64K 1 /* 64K */ | ||
154 | #define MMU_PAGE_64K_AP 2 /* 64K Admixed (in a 4K segment) */ | ||
155 | #define MMU_PAGE_1M 3 /* 1M */ | ||
156 | #define MMU_PAGE_16M 4 /* 16M */ | ||
157 | #define MMU_PAGE_16G 5 /* 16G */ | ||
158 | #define MMU_PAGE_COUNT 6 | ||
159 | |||
160 | #ifndef __ASSEMBLY__ | ||
161 | |||
162 | /* | ||
163 | * The current system page sizes | ||
164 | */ | ||
165 | extern struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT]; | ||
166 | extern int mmu_linear_psize; | ||
167 | extern int mmu_virtual_psize; | ||
168 | extern int mmu_vmalloc_psize; | ||
169 | extern int mmu_io_psize; | ||
170 | |||
171 | /* | ||
172 | * If the processor supports 64k normal pages but not 64k cache | ||
173 | * inhibited pages, we have to be prepared to switch processes | ||
174 | * to use 4k pages when they create cache-inhibited mappings. | ||
175 | * If this is the case, mmu_ci_restrictions will be set to 1. | ||
176 | */ | ||
177 | extern int mmu_ci_restrictions; | ||
178 | |||
179 | #ifdef CONFIG_HUGETLB_PAGE | ||
180 | /* | ||
181 | * The page size index of the huge pages for use by hugetlbfs | ||
182 | */ | ||
183 | extern int mmu_huge_psize; | ||
184 | |||
185 | #endif /* CONFIG_HUGETLB_PAGE */ | ||
186 | |||
187 | /* | ||
188 | * This function sets the AVPN and L fields of the HPTE appropriately | ||
189 | * for the page size | ||
190 | */ | ||
191 | static inline unsigned long hpte_encode_v(unsigned long va, int psize) | ||
192 | { | ||
193 | unsigned long v = | ||
194 | v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm); | ||
195 | v <<= HPTE_V_AVPN_SHIFT; | ||
196 | if (psize != MMU_PAGE_4K) | ||
197 | v |= HPTE_V_LARGE; | ||
198 | return v; | ||
199 | } | ||
200 | |||
201 | /* | ||
202 | * This function sets the ARPN, and LP fields of the HPTE appropriately | ||
203 | * for the page size. We assume the pa is already "clean" that is properly | ||
204 | * aligned for the requested page size | ||
205 | */ | ||
206 | static inline unsigned long hpte_encode_r(unsigned long pa, int psize) | ||
207 | { | ||
208 | unsigned long r; | ||
209 | |||
210 | /* A 4K page needs no special encoding */ | ||
211 | if (psize == MMU_PAGE_4K) | ||
212 | return pa & HPTE_R_RPN; | ||
213 | else { | ||
214 | unsigned int penc = mmu_psize_defs[psize].penc; | ||
215 | unsigned int shift = mmu_psize_defs[psize].shift; | ||
216 | return (pa & ~((1ul << shift) - 1)) | (penc << 12); | ||
217 | } | ||
218 | return r; | ||
219 | } | ||
220 | |||
221 | /* | ||
222 | * This hashes a virtual address for a 256Mb segment only for now | ||
223 | */ | ||
224 | |||
225 | static inline unsigned long hpt_hash(unsigned long va, unsigned int shift) | ||
226 | { | ||
227 | return ((va >> 28) & 0x7fffffffffUL) ^ ((va & 0x0fffffffUL) >> shift); | ||
228 | } | ||
229 | |||
230 | extern int __hash_page_4K(unsigned long ea, unsigned long access, | ||
231 | unsigned long vsid, pte_t *ptep, unsigned long trap, | ||
232 | unsigned int local); | ||
233 | extern int __hash_page_64K(unsigned long ea, unsigned long access, | ||
234 | unsigned long vsid, pte_t *ptep, unsigned long trap, | ||
235 | unsigned int local); | ||
236 | struct mm_struct; | ||
237 | extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap); | ||
238 | extern int hash_huge_page(struct mm_struct *mm, unsigned long access, | ||
239 | unsigned long ea, unsigned long vsid, int local, | ||
240 | unsigned long trap); | ||
241 | |||
242 | extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend, | ||
243 | unsigned long pstart, unsigned long mode, | ||
244 | int psize); | ||
245 | |||
246 | extern void htab_initialize(void); | ||
247 | extern void htab_initialize_secondary(void); | ||
248 | extern void hpte_init_native(void); | ||
249 | extern void hpte_init_lpar(void); | ||
250 | extern void hpte_init_iSeries(void); | ||
251 | extern void hpte_init_beat(void); | ||
252 | |||
253 | extern void stabs_alloc(void); | ||
254 | extern void slb_initialize(void); | ||
255 | extern void slb_flush_and_rebolt(void); | ||
256 | extern void stab_initialize(unsigned long stab); | ||
257 | |||
258 | #endif /* __ASSEMBLY__ */ | ||
259 | |||
260 | /* | ||
261 | * VSID allocation | ||
262 | * | ||
263 | * We first generate a 36-bit "proto-VSID". For kernel addresses this | ||
264 | * is equal to the ESID, for user addresses it is: | ||
265 | * (context << 15) | (esid & 0x7fff) | ||
266 | * | ||
267 | * The two forms are distinguishable because the top bit is 0 for user | ||
268 | * addresses, whereas the top two bits are 1 for kernel addresses. | ||
269 | * Proto-VSIDs with the top two bits equal to 0b10 are reserved for | ||
270 | * now. | ||
271 | * | ||
272 | * The proto-VSIDs are then scrambled into real VSIDs with the | ||
273 | * multiplicative hash: | ||
274 | * | ||
275 | * VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS | ||
276 | * where VSID_MULTIPLIER = 268435399 = 0xFFFFFC7 | ||
277 | * VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF | ||
278 | * | ||
279 | * This scramble is only well defined for proto-VSIDs below | ||
280 | * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are | ||
281 | * reserved. VSID_MULTIPLIER is prime, so in particular it is | ||
282 | * co-prime to VSID_MODULUS, making this a 1:1 scrambling function. | ||
283 | * Because the modulus is 2^n-1 we can compute it efficiently without | ||
284 | * a divide or extra multiply (see below). | ||
285 | * | ||
286 | * This scheme has several advantages over older methods: | ||
287 | * | ||
288 | * - We have VSIDs allocated for every kernel address | ||
289 | * (i.e. everything above 0xC000000000000000), except the very top | ||
290 | * segment, which simplifies several things. | ||
291 | * | ||
292 | * - We allow for 15 significant bits of ESID and 20 bits of | ||
293 | * context for user addresses. i.e. 8T (43 bits) of address space for | ||
294 | * up to 1M contexts (although the page table structure and context | ||
295 | * allocation will need changes to take advantage of this). | ||
296 | * | ||
297 | * - The scramble function gives robust scattering in the hash | ||
298 | * table (at least based on some initial results). The previous | ||
299 | * method was more susceptible to pathological cases giving excessive | ||
300 | * hash collisions. | ||
301 | */ | ||
302 | /* | ||
303 | * WARNING - If you change these you must make sure the asm | ||
304 | * implementations in slb_allocate (slb_low.S), do_stab_bolted | ||
305 | * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly. | ||
306 | * | ||
307 | * You'll also need to change the precomputed VSID values in head.S | ||
308 | * which are used by the iSeries firmware. | ||
309 | */ | ||
310 | |||
311 | #define VSID_MULTIPLIER ASM_CONST(200730139) /* 28-bit prime */ | ||
312 | #define VSID_BITS 36 | ||
313 | #define VSID_MODULUS ((1UL<<VSID_BITS)-1) | ||
314 | |||
315 | #define CONTEXT_BITS 19 | ||
316 | #define USER_ESID_BITS 16 | ||
317 | |||
318 | #define USER_VSID_RANGE (1UL << (USER_ESID_BITS + SID_SHIFT)) | ||
319 | |||
320 | /* | ||
321 | * This macro generates asm code to compute the VSID scramble | ||
322 | * function. Used in slb_allocate() and do_stab_bolted. The function | ||
323 | * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS | ||
324 | * | ||
325 | * rt = register continaing the proto-VSID and into which the | ||
326 | * VSID will be stored | ||
327 | * rx = scratch register (clobbered) | ||
328 | * | ||
329 | * - rt and rx must be different registers | ||
330 | * - The answer will end up in the low 36 bits of rt. The higher | ||
331 | * bits may contain other garbage, so you may need to mask the | ||
332 | * result. | ||
333 | */ | ||
334 | #define ASM_VSID_SCRAMBLE(rt, rx) \ | ||
335 | lis rx,VSID_MULTIPLIER@h; \ | ||
336 | ori rx,rx,VSID_MULTIPLIER@l; \ | ||
337 | mulld rt,rt,rx; /* rt = rt * MULTIPLIER */ \ | ||
338 | \ | ||
339 | srdi rx,rt,VSID_BITS; \ | ||
340 | clrldi rt,rt,(64-VSID_BITS); \ | ||
341 | add rt,rt,rx; /* add high and low bits */ \ | ||
342 | /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and \ | ||
343 | * 2^36-1+2^28-1. That in particular means that if r3 >= \ | ||
344 | * 2^36-1, then r3+1 has the 2^36 bit set. So, if r3+1 has \ | ||
345 | * the bit clear, r3 already has the answer we want, if it \ | ||
346 | * doesn't, the answer is the low 36 bits of r3+1. So in all \ | ||
347 | * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\ | ||
348 | addi rx,rt,1; \ | ||
349 | srdi rx,rx,VSID_BITS; /* extract 2^36 bit */ \ | ||
350 | add rt,rt,rx | ||
351 | |||
352 | |||
353 | #ifndef __ASSEMBLY__ | ||
354 | |||
355 | typedef unsigned long mm_context_id_t; | ||
356 | |||
357 | typedef struct { | ||
358 | mm_context_id_t id; | ||
359 | u16 user_psize; /* page size index */ | ||
360 | u16 sllp; /* SLB entry page size encoding */ | ||
361 | #ifdef CONFIG_HUGETLB_PAGE | ||
362 | u16 low_htlb_areas, high_htlb_areas; | ||
363 | #endif | 12 | #endif |
364 | unsigned long vdso_base; | ||
365 | } mm_context_t; | ||
366 | |||
367 | |||
368 | static inline unsigned long vsid_scramble(unsigned long protovsid) | ||
369 | { | ||
370 | #if 0 | ||
371 | /* The code below is equivalent to this function for arguments | ||
372 | * < 2^VSID_BITS, which is all this should ever be called | ||
373 | * with. However gcc is not clever enough to compute the | ||
374 | * modulus (2^n-1) without a second multiply. */ | ||
375 | return ((protovsid * VSID_MULTIPLIER) % VSID_MODULUS); | ||
376 | #else /* 1 */ | ||
377 | unsigned long x; | ||
378 | |||
379 | x = protovsid * VSID_MULTIPLIER; | ||
380 | x = (x >> VSID_BITS) + (x & VSID_MODULUS); | ||
381 | return (x + ((x+1) >> VSID_BITS)) & VSID_MODULUS; | ||
382 | #endif /* 1 */ | ||
383 | } | ||
384 | |||
385 | /* This is only valid for addresses >= KERNELBASE */ | ||
386 | static inline unsigned long get_kernel_vsid(unsigned long ea) | ||
387 | { | ||
388 | return vsid_scramble(ea >> SID_SHIFT); | ||
389 | } | ||
390 | |||
391 | /* This is only valid for user addresses (which are below 2^41) */ | ||
392 | static inline unsigned long get_vsid(unsigned long context, unsigned long ea) | ||
393 | { | ||
394 | return vsid_scramble((context << USER_ESID_BITS) | ||
395 | | (ea >> SID_SHIFT)); | ||
396 | } | ||
397 | |||
398 | #define VSID_SCRAMBLE(pvsid) (((pvsid) * VSID_MULTIPLIER) % VSID_MODULUS) | ||
399 | #define KERNEL_VSID(ea) VSID_SCRAMBLE(GET_ESID(ea)) | ||
400 | |||
401 | /* Physical address used by some IO functions */ | ||
402 | typedef unsigned long phys_addr_t; | ||
403 | |||
404 | |||
405 | #endif /* __ASSEMBLY */ | ||
406 | 13 | ||
407 | #endif /* CONFIG_PPC64 */ | ||
408 | #endif /* __KERNEL__ */ | 14 | #endif /* __KERNEL__ */ |
409 | #endif /* _ASM_POWERPC_MMU_H_ */ | 15 | #endif /* _ASM_POWERPC_MMU_H_ */ |