1 #ifndef _ASM_POWERPC_MMU_HASH64_H_
2 #define _ASM_POWERPC_MMU_HASH64_H_
4 * PowerPC64 memory management structures
6 * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com>
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.
15 #include <asm/asm-compat.h>
22 #define STE_ESID_V 0x80
23 #define STE_ESID_KS 0x20
24 #define STE_ESID_KP 0x10
25 #define STE_ESID_N 0x08
27 #define STE_VSID_SHIFT 12
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)
35 extern char initial_stab[];
36 #endif /* ! __ASSEMBLY */
42 #define SLB_NUM_BOLTED 3
43 #define SLB_CACHE_ENTRIES 8
45 /* Bits in the SLB ESID word */
46 #define SLB_ESID_V ASM_CONST(0x0000000008000000) /* valid */
48 /* Bits in the SLB VSID word */
49 #define SLB_VSID_SHIFT 12
50 #define SLB_VSID_SHIFT_1T 24
51 #define SLB_VSID_SSIZE_SHIFT 62
52 #define SLB_VSID_B ASM_CONST(0xc000000000000000)
53 #define SLB_VSID_B_256M ASM_CONST(0x0000000000000000)
54 #define SLB_VSID_B_1T ASM_CONST(0x4000000000000000)
55 #define SLB_VSID_KS ASM_CONST(0x0000000000000800)
56 #define SLB_VSID_KP ASM_CONST(0x0000000000000400)
57 #define SLB_VSID_N ASM_CONST(0x0000000000000200) /* no-execute */
58 #define SLB_VSID_L ASM_CONST(0x0000000000000100)
59 #define SLB_VSID_C ASM_CONST(0x0000000000000080) /* class */
60 #define SLB_VSID_LP ASM_CONST(0x0000000000000030)
61 #define SLB_VSID_LP_00 ASM_CONST(0x0000000000000000)
62 #define SLB_VSID_LP_01 ASM_CONST(0x0000000000000010)
63 #define SLB_VSID_LP_10 ASM_CONST(0x0000000000000020)
64 #define SLB_VSID_LP_11 ASM_CONST(0x0000000000000030)
65 #define SLB_VSID_LLP (SLB_VSID_L|SLB_VSID_LP)
67 #define SLB_VSID_KERNEL (SLB_VSID_KP)
68 #define SLB_VSID_USER (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
70 #define SLBIE_C (0x08000000)
71 #define SLBIE_SSIZE_SHIFT 25
77 #define HPTES_PER_GROUP 8
79 #define HPTE_V_SSIZE_SHIFT 62
80 #define HPTE_V_AVPN_SHIFT 7
81 #define HPTE_V_AVPN ASM_CONST(0x3fffffffffffff80)
82 #define HPTE_V_AVPN_VAL(x) (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
83 #define HPTE_V_COMPARE(x,y) (!(((x) ^ (y)) & 0xffffffffffffff80UL))
84 #define HPTE_V_BOLTED ASM_CONST(0x0000000000000010)
85 #define HPTE_V_LOCK ASM_CONST(0x0000000000000008)
86 #define HPTE_V_LARGE ASM_CONST(0x0000000000000004)
87 #define HPTE_V_SECONDARY ASM_CONST(0x0000000000000002)
88 #define HPTE_V_VALID ASM_CONST(0x0000000000000001)
90 #define HPTE_R_PP0 ASM_CONST(0x8000000000000000)
91 #define HPTE_R_TS ASM_CONST(0x4000000000000000)
92 #define HPTE_R_RPN_SHIFT 12
93 #define HPTE_R_RPN ASM_CONST(0x3ffffffffffff000)
94 #define HPTE_R_FLAGS ASM_CONST(0x00000000000003ff)
95 #define HPTE_R_PP ASM_CONST(0x0000000000000003)
96 #define HPTE_R_N ASM_CONST(0x0000000000000004)
97 #define HPTE_R_C ASM_CONST(0x0000000000000080)
98 #define HPTE_R_R ASM_CONST(0x0000000000000100)
100 #define HPTE_V_1TB_SEG ASM_CONST(0x4000000000000000)
101 #define HPTE_V_VRMA_MASK ASM_CONST(0x4001ffffff000000)
103 /* Values for PP (assumes Ks=0, Kp=1) */
104 /* pp0 will always be 0 for linux */
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 */
117 extern struct hash_pte *htab_address;
118 extern unsigned long htab_size_bytes;
119 extern unsigned long htab_hash_mask;
122 * Page size definition
124 * shift : is the "PAGE_SHIFT" value for that page size
125 * sllp : is a bit mask with the value of SLB L || LP to be or'ed
126 * directly to a slbmte "vsid" value
127 * penc : is the HPTE encoding mask for the "LP" field:
132 unsigned int shift; /* number of bits */
133 unsigned int penc; /* HPTE encoding */
134 unsigned int tlbiel; /* tlbiel supported for that page size */
135 unsigned long avpnm; /* bits to mask out in AVPN in the HPTE */
136 unsigned long sllp; /* SLB L||LP (exact mask to use in slbmte) */
139 #endif /* __ASSEMBLY__ */
142 * The kernel use the constants below to index in the page sizes array.
143 * The use of fixed constants for this purpose is better for performances
144 * of the low level hash refill handlers.
146 * A non supported page size has a "shift" field set to 0
148 * Any new page size being implemented can get a new entry in here. Whether
149 * the kernel will use it or not is a different matter though. The actual page
150 * size used by hugetlbfs is not defined here and may be made variable
153 #define MMU_PAGE_4K 0 /* 4K */
154 #define MMU_PAGE_64K 1 /* 64K */
155 #define MMU_PAGE_64K_AP 2 /* 64K Admixed (in a 4K segment) */
156 #define MMU_PAGE_1M 3 /* 1M */
157 #define MMU_PAGE_16M 4 /* 16M */
158 #define MMU_PAGE_16G 5 /* 16G */
159 #define MMU_PAGE_COUNT 6
163 * These are the values used by hardware in the B field of
164 * SLB entries and the first dword of MMU hashtable entries.
165 * The B field is 2 bits; the values 2 and 3 are unused and reserved.
167 #define MMU_SEGSIZE_256M 0
168 #define MMU_SEGSIZE_1T 1
174 * The current system page and segment sizes
176 extern struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
177 extern int mmu_linear_psize;
178 extern int mmu_virtual_psize;
179 extern int mmu_vmalloc_psize;
180 extern int mmu_vmemmap_psize;
181 extern int mmu_io_psize;
182 extern int mmu_kernel_ssize;
183 extern int mmu_highuser_ssize;
184 extern u16 mmu_slb_size;
185 extern unsigned long tce_alloc_start, tce_alloc_end;
188 * If the processor supports 64k normal pages but not 64k cache
189 * inhibited pages, we have to be prepared to switch processes
190 * to use 4k pages when they create cache-inhibited mappings.
191 * If this is the case, mmu_ci_restrictions will be set to 1.
193 extern int mmu_ci_restrictions;
195 #ifdef CONFIG_HUGETLB_PAGE
197 * The page size indexes of the huge pages for use by hugetlbfs
199 extern unsigned int mmu_huge_psizes[MMU_PAGE_COUNT];
201 #endif /* CONFIG_HUGETLB_PAGE */
204 * This function sets the AVPN and L fields of the HPTE appropriately
207 static inline unsigned long hpte_encode_v(unsigned long va, int psize,
211 v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm);
212 v <<= HPTE_V_AVPN_SHIFT;
213 if (psize != MMU_PAGE_4K)
215 v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
220 * This function sets the ARPN, and LP fields of the HPTE appropriately
221 * for the page size. We assume the pa is already "clean" that is properly
222 * aligned for the requested page size
224 static inline unsigned long hpte_encode_r(unsigned long pa, int psize)
228 /* A 4K page needs no special encoding */
229 if (psize == MMU_PAGE_4K)
230 return pa & HPTE_R_RPN;
232 unsigned int penc = mmu_psize_defs[psize].penc;
233 unsigned int shift = mmu_psize_defs[psize].shift;
234 return (pa & ~((1ul << shift) - 1)) | (penc << 12);
240 * Build a VA given VSID, EA and segment size
242 static inline unsigned long hpt_va(unsigned long ea, unsigned long vsid,
245 if (ssize == MMU_SEGSIZE_256M)
246 return (vsid << 28) | (ea & 0xfffffffUL);
247 return (vsid << 40) | (ea & 0xffffffffffUL);
251 * This hashes a virtual address
254 static inline unsigned long hpt_hash(unsigned long va, unsigned int shift,
257 unsigned long hash, vsid;
259 if (ssize == MMU_SEGSIZE_256M) {
260 hash = (va >> 28) ^ ((va & 0x0fffffffUL) >> shift);
263 hash = vsid ^ (vsid << 25) ^ ((va & 0xffffffffffUL) >> shift);
265 return hash & 0x7fffffffffUL;
268 extern int __hash_page_4K(unsigned long ea, unsigned long access,
269 unsigned long vsid, pte_t *ptep, unsigned long trap,
270 unsigned int local, int ssize, int subpage_prot);
271 extern int __hash_page_64K(unsigned long ea, unsigned long access,
272 unsigned long vsid, pte_t *ptep, unsigned long trap,
273 unsigned int local, int ssize);
275 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap);
276 extern int hash_huge_page(struct mm_struct *mm, unsigned long access,
277 unsigned long ea, unsigned long vsid, int local,
280 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
281 unsigned long pstart, unsigned long mode,
282 int psize, int ssize);
283 extern void set_huge_psize(int psize);
284 extern void add_gpage(unsigned long addr, unsigned long page_size,
285 unsigned long number_of_pages);
286 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
288 extern void htab_initialize(void);
289 extern void htab_initialize_secondary(void);
290 extern void hpte_init_native(void);
291 extern void hpte_init_lpar(void);
292 extern void hpte_init_iSeries(void);
293 extern void hpte_init_beat(void);
294 extern void hpte_init_beat_v3(void);
296 extern void stabs_alloc(void);
297 extern void slb_initialize(void);
298 extern void slb_flush_and_rebolt(void);
299 extern void stab_initialize(unsigned long stab);
301 extern void slb_vmalloc_update(void);
302 #endif /* __ASSEMBLY__ */
307 * We first generate a 36-bit "proto-VSID". For kernel addresses this
308 * is equal to the ESID, for user addresses it is:
309 * (context << 15) | (esid & 0x7fff)
311 * The two forms are distinguishable because the top bit is 0 for user
312 * addresses, whereas the top two bits are 1 for kernel addresses.
313 * Proto-VSIDs with the top two bits equal to 0b10 are reserved for
316 * The proto-VSIDs are then scrambled into real VSIDs with the
317 * multiplicative hash:
319 * VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
320 * where VSID_MULTIPLIER = 268435399 = 0xFFFFFC7
321 * VSID_MODULUS = 2^36-1 = 0xFFFFFFFFF
323 * This scramble is only well defined for proto-VSIDs below
324 * 0xFFFFFFFFF, so both proto-VSID and actual VSID 0xFFFFFFFFF are
325 * reserved. VSID_MULTIPLIER is prime, so in particular it is
326 * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
327 * Because the modulus is 2^n-1 we can compute it efficiently without
328 * a divide or extra multiply (see below).
330 * This scheme has several advantages over older methods:
332 * - We have VSIDs allocated for every kernel address
333 * (i.e. everything above 0xC000000000000000), except the very top
334 * segment, which simplifies several things.
336 * - We allow for 15 significant bits of ESID and 20 bits of
337 * context for user addresses. i.e. 8T (43 bits) of address space for
338 * up to 1M contexts (although the page table structure and context
339 * allocation will need changes to take advantage of this).
341 * - The scramble function gives robust scattering in the hash
342 * table (at least based on some initial results). The previous
343 * method was more susceptible to pathological cases giving excessive
347 * WARNING - If you change these you must make sure the asm
348 * implementations in slb_allocate (slb_low.S), do_stab_bolted
349 * (head.S) and ASM_VSID_SCRAMBLE (below) are changed accordingly.
351 * You'll also need to change the precomputed VSID values in head.S
352 * which are used by the iSeries firmware.
355 #define VSID_MULTIPLIER_256M ASM_CONST(200730139) /* 28-bit prime */
356 #define VSID_BITS_256M 36
357 #define VSID_MODULUS_256M ((1UL<<VSID_BITS_256M)-1)
359 #define VSID_MULTIPLIER_1T ASM_CONST(12538073) /* 24-bit prime */
360 #define VSID_BITS_1T 24
361 #define VSID_MODULUS_1T ((1UL<<VSID_BITS_1T)-1)
363 #define CONTEXT_BITS 19
364 #define USER_ESID_BITS 16
365 #define USER_ESID_BITS_1T 4
367 #define USER_VSID_RANGE (1UL << (USER_ESID_BITS + SID_SHIFT))
370 * This macro generates asm code to compute the VSID scramble
371 * function. Used in slb_allocate() and do_stab_bolted. The function
372 * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
374 * rt = register continaing the proto-VSID and into which the
375 * VSID will be stored
376 * rx = scratch register (clobbered)
378 * - rt and rx must be different registers
379 * - The answer will end up in the low VSID_BITS bits of rt. The higher
380 * bits may contain other garbage, so you may need to mask the
383 #define ASM_VSID_SCRAMBLE(rt, rx, size) \
384 lis rx,VSID_MULTIPLIER_##size@h; \
385 ori rx,rx,VSID_MULTIPLIER_##size@l; \
386 mulld rt,rt,rx; /* rt = rt * MULTIPLIER */ \
388 srdi rx,rt,VSID_BITS_##size; \
389 clrldi rt,rt,(64-VSID_BITS_##size); \
390 add rt,rt,rx; /* add high and low bits */ \
391 /* Now, r3 == VSID (mod 2^36-1), and lies between 0 and \
392 * 2^36-1+2^28-1. That in particular means that if r3 >= \
393 * 2^36-1, then r3+1 has the 2^36 bit set. So, if r3+1 has \
394 * the bit clear, r3 already has the answer we want, if it \
395 * doesn't, the answer is the low 36 bits of r3+1. So in all \
396 * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
398 srdi rx,rx,VSID_BITS_##size; /* extract 2^VSID_BITS bit */ \
404 typedef unsigned long mm_context_id_t;
408 u16 user_psize; /* page size index */
410 #ifdef CONFIG_PPC_MM_SLICES
411 u64 low_slices_psize; /* SLB page size encodings */
412 u64 high_slices_psize; /* 4 bits per slice for now */
414 u16 sllp; /* SLB page size encoding */
416 unsigned long vdso_base;
422 * The code below is equivalent to this function for arguments
423 * < 2^VSID_BITS, which is all this should ever be called
424 * with. However gcc is not clever enough to compute the
425 * modulus (2^n-1) without a second multiply.
427 #define vsid_scrample(protovsid, size) \
428 ((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
431 #define vsid_scramble(protovsid, size) \
434 x = (protovsid) * VSID_MULTIPLIER_##size; \
435 x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
436 (x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
440 /* This is only valid for addresses >= KERNELBASE */
441 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
443 if (ssize == MMU_SEGSIZE_256M)
444 return vsid_scramble(ea >> SID_SHIFT, 256M);
445 return vsid_scramble(ea >> SID_SHIFT_1T, 1T);
448 /* Returns the segment size indicator for a user address */
449 static inline int user_segment_size(unsigned long addr)
451 /* Use 1T segments if possible for addresses >= 1T */
452 if (addr >= (1UL << SID_SHIFT_1T))
453 return mmu_highuser_ssize;
454 return MMU_SEGSIZE_256M;
457 /* This is only valid for user addresses (which are below 2^44) */
458 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
461 if (ssize == MMU_SEGSIZE_256M)
462 return vsid_scramble((context << USER_ESID_BITS)
463 | (ea >> SID_SHIFT), 256M);
464 return vsid_scramble((context << USER_ESID_BITS_1T)
465 | (ea >> SID_SHIFT_1T), 1T);
469 * This is only used on legacy iSeries in lparmap.c,
470 * hence the 256MB segment assumption.
472 #define VSID_SCRAMBLE(pvsid) (((pvsid) * VSID_MULTIPLIER_256M) % \
474 #define KERNEL_VSID(ea) VSID_SCRAMBLE(GET_ESID(ea))
476 #endif /* __ASSEMBLY__ */
478 #endif /* _ASM_POWERPC_MMU_HASH64_H_ */