1 #ifndef _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_
2 #define _ASM_POWERPC_BOOK3S_64_MMU_HASH_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 #include <asm/bug.h>
18 
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  */
24 #include <asm/book3s/64/pgtable.h>
25 #include <asm/bug.h>
26 #include <asm/processor.h>
27 #include <asm/cpu_has_feature.h>
28 
29 /*
30  * SLB
31  */
32 
33 #define SLB_NUM_BOLTED		3
34 #define SLB_CACHE_ENTRIES	8
35 #define SLB_MIN_SIZE		32
36 
37 /* Bits in the SLB ESID word */
38 #define SLB_ESID_V		ASM_CONST(0x0000000008000000) /* valid */
39 
40 /* Bits in the SLB VSID word */
41 #define SLB_VSID_SHIFT		12
42 #define SLB_VSID_SHIFT_1T	24
43 #define SLB_VSID_SSIZE_SHIFT	62
44 #define SLB_VSID_B		ASM_CONST(0xc000000000000000)
45 #define SLB_VSID_B_256M		ASM_CONST(0x0000000000000000)
46 #define SLB_VSID_B_1T		ASM_CONST(0x4000000000000000)
47 #define SLB_VSID_KS		ASM_CONST(0x0000000000000800)
48 #define SLB_VSID_KP		ASM_CONST(0x0000000000000400)
49 #define SLB_VSID_N		ASM_CONST(0x0000000000000200) /* no-execute */
50 #define SLB_VSID_L		ASM_CONST(0x0000000000000100)
51 #define SLB_VSID_C		ASM_CONST(0x0000000000000080) /* class */
52 #define SLB_VSID_LP		ASM_CONST(0x0000000000000030)
53 #define SLB_VSID_LP_00		ASM_CONST(0x0000000000000000)
54 #define SLB_VSID_LP_01		ASM_CONST(0x0000000000000010)
55 #define SLB_VSID_LP_10		ASM_CONST(0x0000000000000020)
56 #define SLB_VSID_LP_11		ASM_CONST(0x0000000000000030)
57 #define SLB_VSID_LLP		(SLB_VSID_L|SLB_VSID_LP)
58 
59 #define SLB_VSID_KERNEL		(SLB_VSID_KP)
60 #define SLB_VSID_USER		(SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
61 
62 #define SLBIE_C			(0x08000000)
63 #define SLBIE_SSIZE_SHIFT	25
64 
65 /*
66  * Hash table
67  */
68 
69 #define HPTES_PER_GROUP 8
70 
71 #define HPTE_V_SSIZE_SHIFT	62
72 #define HPTE_V_AVPN_SHIFT	7
73 #define HPTE_V_COMMON_BITS	ASM_CONST(0x000fffffffffffff)
74 #define HPTE_V_AVPN		ASM_CONST(0x3fffffffffffff80)
75 #define HPTE_V_AVPN_3_0		ASM_CONST(0x000fffffffffff80)
76 #define HPTE_V_AVPN_VAL(x)	(((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
77 #define HPTE_V_COMPARE(x,y)	(!(((x) ^ (y)) & 0xffffffffffffff80UL))
78 #define HPTE_V_BOLTED		ASM_CONST(0x0000000000000010)
79 #define HPTE_V_LOCK		ASM_CONST(0x0000000000000008)
80 #define HPTE_V_LARGE		ASM_CONST(0x0000000000000004)
81 #define HPTE_V_SECONDARY	ASM_CONST(0x0000000000000002)
82 #define HPTE_V_VALID		ASM_CONST(0x0000000000000001)
83 
84 /*
85  * ISA 3.0 has a different HPTE format.
86  */
87 #define HPTE_R_3_0_SSIZE_SHIFT	58
88 #define HPTE_R_3_0_SSIZE_MASK	(3ull << HPTE_R_3_0_SSIZE_SHIFT)
89 #define HPTE_R_PP0		ASM_CONST(0x8000000000000000)
90 #define HPTE_R_TS		ASM_CONST(0x4000000000000000)
91 #define HPTE_R_KEY_HI		ASM_CONST(0x3000000000000000)
92 #define HPTE_R_RPN_SHIFT	12
93 #define HPTE_R_RPN		ASM_CONST(0x0ffffffffffff000)
94 #define HPTE_R_RPN_3_0		ASM_CONST(0x01fffffffffff000)
95 #define HPTE_R_PP		ASM_CONST(0x0000000000000003)
96 #define HPTE_R_PPP		ASM_CONST(0x8000000000000003)
97 #define HPTE_R_N		ASM_CONST(0x0000000000000004)
98 #define HPTE_R_G		ASM_CONST(0x0000000000000008)
99 #define HPTE_R_M		ASM_CONST(0x0000000000000010)
100 #define HPTE_R_I		ASM_CONST(0x0000000000000020)
101 #define HPTE_R_W		ASM_CONST(0x0000000000000040)
102 #define HPTE_R_WIMG		ASM_CONST(0x0000000000000078)
103 #define HPTE_R_C		ASM_CONST(0x0000000000000080)
104 #define HPTE_R_R		ASM_CONST(0x0000000000000100)
105 #define HPTE_R_KEY_LO		ASM_CONST(0x0000000000000e00)
106 
107 #define HPTE_V_1TB_SEG		ASM_CONST(0x4000000000000000)
108 #define HPTE_V_VRMA_MASK	ASM_CONST(0x4001ffffff000000)
109 
110 /* Values for PP (assumes Ks=0, Kp=1) */
111 #define PP_RWXX	0	/* Supervisor read/write, User none */
112 #define PP_RWRX 1	/* Supervisor read/write, User read */
113 #define PP_RWRW 2	/* Supervisor read/write, User read/write */
114 #define PP_RXRX 3	/* Supervisor read,       User read */
115 #define PP_RXXX	(HPTE_R_PP0 | 2)	/* Supervisor read, user none */
116 
117 /* Fields for tlbiel instruction in architecture 2.06 */
118 #define TLBIEL_INVAL_SEL_MASK	0xc00	/* invalidation selector */
119 #define  TLBIEL_INVAL_PAGE	0x000	/* invalidate a single page */
120 #define  TLBIEL_INVAL_SET_LPID	0x800	/* invalidate a set for current LPID */
121 #define  TLBIEL_INVAL_SET	0xc00	/* invalidate a set for all LPIDs */
122 #define TLBIEL_INVAL_SET_MASK	0xfff000	/* set number to inval. */
123 #define TLBIEL_INVAL_SET_SHIFT	12
124 
125 #define POWER7_TLB_SETS		128	/* # sets in POWER7 TLB */
126 #define POWER8_TLB_SETS		512	/* # sets in POWER8 TLB */
127 #define POWER9_TLB_SETS_HASH	256	/* # sets in POWER9 TLB Hash mode */
128 #define POWER9_TLB_SETS_RADIX	128	/* # sets in POWER9 TLB Radix mode */
129 
130 #ifndef __ASSEMBLY__
131 
132 struct mmu_hash_ops {
133 	void            (*hpte_invalidate)(unsigned long slot,
134 					   unsigned long vpn,
135 					   int bpsize, int apsize,
136 					   int ssize, int local);
137 	long		(*hpte_updatepp)(unsigned long slot,
138 					 unsigned long newpp,
139 					 unsigned long vpn,
140 					 int bpsize, int apsize,
141 					 int ssize, unsigned long flags);
142 	void            (*hpte_updateboltedpp)(unsigned long newpp,
143 					       unsigned long ea,
144 					       int psize, int ssize);
145 	long		(*hpte_insert)(unsigned long hpte_group,
146 				       unsigned long vpn,
147 				       unsigned long prpn,
148 				       unsigned long rflags,
149 				       unsigned long vflags,
150 				       int psize, int apsize,
151 				       int ssize);
152 	long		(*hpte_remove)(unsigned long hpte_group);
153 	int             (*hpte_removebolted)(unsigned long ea,
154 					     int psize, int ssize);
155 	void		(*flush_hash_range)(unsigned long number, int local);
156 	void		(*hugepage_invalidate)(unsigned long vsid,
157 					       unsigned long addr,
158 					       unsigned char *hpte_slot_array,
159 					       int psize, int ssize, int local);
160 	/*
161 	 * Special for kexec.
162 	 * To be called in real mode with interrupts disabled. No locks are
163 	 * taken as such, concurrent access on pre POWER5 hardware could result
164 	 * in a deadlock.
165 	 * The linear mapping is destroyed as well.
166 	 */
167 	void		(*hpte_clear_all)(void);
168 };
169 extern struct mmu_hash_ops mmu_hash_ops;
170 
171 struct hash_pte {
172 	__be64 v;
173 	__be64 r;
174 };
175 
176 extern struct hash_pte *htab_address;
177 extern unsigned long htab_size_bytes;
178 extern unsigned long htab_hash_mask;
179 
180 
181 static inline int shift_to_mmu_psize(unsigned int shift)
182 {
183 	int psize;
184 
185 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
186 		if (mmu_psize_defs[psize].shift == shift)
187 			return psize;
188 	return -1;
189 }
190 
191 static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
192 {
193 	if (mmu_psize_defs[mmu_psize].shift)
194 		return mmu_psize_defs[mmu_psize].shift;
195 	BUG();
196 }
197 
198 static inline unsigned long get_sllp_encoding(int psize)
199 {
200 	unsigned long sllp;
201 
202 	sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
203 		((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
204 	return sllp;
205 }
206 
207 #endif /* __ASSEMBLY__ */
208 
209 /*
210  * Segment sizes.
211  * These are the values used by hardware in the B field of
212  * SLB entries and the first dword of MMU hashtable entries.
213  * The B field is 2 bits; the values 2 and 3 are unused and reserved.
214  */
215 #define MMU_SEGSIZE_256M	0
216 #define MMU_SEGSIZE_1T		1
217 
218 /*
219  * encode page number shift.
220  * in order to fit the 78 bit va in a 64 bit variable we shift the va by
221  * 12 bits. This enable us to address upto 76 bit va.
222  * For hpt hash from a va we can ignore the page size bits of va and for
223  * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
224  * we work in all cases including 4k page size.
225  */
226 #define VPN_SHIFT	12
227 
228 /*
229  * HPTE Large Page (LP) details
230  */
231 #define LP_SHIFT	12
232 #define LP_BITS		8
233 #define LP_MASK(i)	((0xFF >> (i)) << LP_SHIFT)
234 
235 #ifndef __ASSEMBLY__
236 
237 static inline int slb_vsid_shift(int ssize)
238 {
239 	if (ssize == MMU_SEGSIZE_256M)
240 		return SLB_VSID_SHIFT;
241 	return SLB_VSID_SHIFT_1T;
242 }
243 
244 static inline int segment_shift(int ssize)
245 {
246 	if (ssize == MMU_SEGSIZE_256M)
247 		return SID_SHIFT;
248 	return SID_SHIFT_1T;
249 }
250 
251 /*
252  * This array is indexed by the LP field of the HPTE second dword.
253  * Since this field may contain some RPN bits, some entries are
254  * replicated so that we get the same value irrespective of RPN.
255  * The top 4 bits are the page size index (MMU_PAGE_*) for the
256  * actual page size, the bottom 4 bits are the base page size.
257  */
258 extern u8 hpte_page_sizes[1 << LP_BITS];
259 
260 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
261 					     bool is_base_size)
262 {
263 	unsigned int i, lp;
264 
265 	if (!(h & HPTE_V_LARGE))
266 		return 1ul << 12;
267 
268 	/* Look at the 8 bit LP value */
269 	lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
270 	i = hpte_page_sizes[lp];
271 	if (!i)
272 		return 0;
273 	if (!is_base_size)
274 		i >>= 4;
275 	return 1ul << mmu_psize_defs[i & 0xf].shift;
276 }
277 
278 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
279 {
280 	return __hpte_page_size(h, l, 0);
281 }
282 
283 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
284 {
285 	return __hpte_page_size(h, l, 1);
286 }
287 
288 /*
289  * The current system page and segment sizes
290  */
291 extern int mmu_kernel_ssize;
292 extern int mmu_highuser_ssize;
293 extern u16 mmu_slb_size;
294 extern unsigned long tce_alloc_start, tce_alloc_end;
295 
296 /*
297  * If the processor supports 64k normal pages but not 64k cache
298  * inhibited pages, we have to be prepared to switch processes
299  * to use 4k pages when they create cache-inhibited mappings.
300  * If this is the case, mmu_ci_restrictions will be set to 1.
301  */
302 extern int mmu_ci_restrictions;
303 
304 /*
305  * This computes the AVPN and B fields of the first dword of a HPTE,
306  * for use when we want to match an existing PTE.  The bottom 7 bits
307  * of the returned value are zero.
308  */
309 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
310 					     int ssize)
311 {
312 	unsigned long v;
313 	/*
314 	 * The AVA field omits the low-order 23 bits of the 78 bits VA.
315 	 * These bits are not needed in the PTE, because the
316 	 * low-order b of these bits are part of the byte offset
317 	 * into the virtual page and, if b < 23, the high-order
318 	 * 23-b of these bits are always used in selecting the
319 	 * PTEGs to be searched
320 	 */
321 	v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
322 	v <<= HPTE_V_AVPN_SHIFT;
323 	v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
324 	return v;
325 }
326 
327 /*
328  * ISA v3.0 defines a new HPTE format, which differs from the old
329  * format in having smaller AVPN and ARPN fields, and the B field
330  * in the second dword instead of the first.
331  */
332 static inline unsigned long hpte_old_to_new_v(unsigned long v)
333 {
334 	/* trim AVPN, drop B */
335 	return v & HPTE_V_COMMON_BITS;
336 }
337 
338 static inline unsigned long hpte_old_to_new_r(unsigned long v, unsigned long r)
339 {
340 	/* move B field from 1st to 2nd dword, trim ARPN */
341 	return (r & ~HPTE_R_3_0_SSIZE_MASK) |
342 		(((v) >> HPTE_V_SSIZE_SHIFT) << HPTE_R_3_0_SSIZE_SHIFT);
343 }
344 
345 static inline unsigned long hpte_new_to_old_v(unsigned long v, unsigned long r)
346 {
347 	/* insert B field */
348 	return (v & HPTE_V_COMMON_BITS) |
349 		((r & HPTE_R_3_0_SSIZE_MASK) <<
350 		 (HPTE_V_SSIZE_SHIFT - HPTE_R_3_0_SSIZE_SHIFT));
351 }
352 
353 static inline unsigned long hpte_new_to_old_r(unsigned long r)
354 {
355 	/* clear out B field */
356 	return r & ~HPTE_R_3_0_SSIZE_MASK;
357 }
358 
359 /*
360  * This function sets the AVPN and L fields of the HPTE  appropriately
361  * using the base page size and actual page size.
362  */
363 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
364 					  int actual_psize, int ssize)
365 {
366 	unsigned long v;
367 	v = hpte_encode_avpn(vpn, base_psize, ssize);
368 	if (actual_psize != MMU_PAGE_4K)
369 		v |= HPTE_V_LARGE;
370 	return v;
371 }
372 
373 /*
374  * This function sets the ARPN, and LP fields of the HPTE appropriately
375  * for the page size. We assume the pa is already "clean" that is properly
376  * aligned for the requested page size
377  */
378 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
379 					  int actual_psize)
380 {
381 	/* A 4K page needs no special encoding */
382 	if (actual_psize == MMU_PAGE_4K)
383 		return pa & HPTE_R_RPN;
384 	else {
385 		unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
386 		unsigned int shift = mmu_psize_defs[actual_psize].shift;
387 		return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
388 	}
389 }
390 
391 /*
392  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
393  */
394 static inline unsigned long hpt_vpn(unsigned long ea,
395 				    unsigned long vsid, int ssize)
396 {
397 	unsigned long mask;
398 	int s_shift = segment_shift(ssize);
399 
400 	mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
401 	return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
402 }
403 
404 /*
405  * This hashes a virtual address
406  */
407 static inline unsigned long hpt_hash(unsigned long vpn,
408 				     unsigned int shift, int ssize)
409 {
410 	int mask;
411 	unsigned long hash, vsid;
412 
413 	/* VPN_SHIFT can be atmost 12 */
414 	if (ssize == MMU_SEGSIZE_256M) {
415 		mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
416 		hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
417 			((vpn & mask) >> (shift - VPN_SHIFT));
418 	} else {
419 		mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
420 		vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
421 		hash = vsid ^ (vsid << 25) ^
422 			((vpn & mask) >> (shift - VPN_SHIFT)) ;
423 	}
424 	return hash & 0x7fffffffffUL;
425 }
426 
427 #define HPTE_LOCAL_UPDATE	0x1
428 #define HPTE_NOHPTE_UPDATE	0x2
429 
430 extern int __hash_page_4K(unsigned long ea, unsigned long access,
431 			  unsigned long vsid, pte_t *ptep, unsigned long trap,
432 			  unsigned long flags, int ssize, int subpage_prot);
433 extern int __hash_page_64K(unsigned long ea, unsigned long access,
434 			   unsigned long vsid, pte_t *ptep, unsigned long trap,
435 			   unsigned long flags, int ssize);
436 struct mm_struct;
437 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
438 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
439 			unsigned long access, unsigned long trap,
440 			unsigned long flags);
441 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
442 		     unsigned long dsisr);
443 int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
444 		     pte_t *ptep, unsigned long trap, unsigned long flags,
445 		     int ssize, unsigned int shift, unsigned int mmu_psize);
446 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
447 extern int __hash_page_thp(unsigned long ea, unsigned long access,
448 			   unsigned long vsid, pmd_t *pmdp, unsigned long trap,
449 			   unsigned long flags, int ssize, unsigned int psize);
450 #else
451 static inline int __hash_page_thp(unsigned long ea, unsigned long access,
452 				  unsigned long vsid, pmd_t *pmdp,
453 				  unsigned long trap, unsigned long flags,
454 				  int ssize, unsigned int psize)
455 {
456 	BUG();
457 	return -1;
458 }
459 #endif
460 extern void hash_failure_debug(unsigned long ea, unsigned long access,
461 			       unsigned long vsid, unsigned long trap,
462 			       int ssize, int psize, int lpsize,
463 			       unsigned long pte);
464 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
465 			     unsigned long pstart, unsigned long prot,
466 			     int psize, int ssize);
467 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
468 			int psize, int ssize);
469 extern void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
470 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
471 
472 #ifdef CONFIG_PPC_PSERIES
473 void hpte_init_pseries(void);
474 #else
475 static inline void hpte_init_pseries(void) { }
476 #endif
477 
478 extern void hpte_init_native(void);
479 
480 extern void slb_initialize(void);
481 extern void slb_flush_and_rebolt(void);
482 
483 extern void slb_vmalloc_update(void);
484 extern void slb_set_size(u16 size);
485 #endif /* __ASSEMBLY__ */
486 
487 /*
488  * VSID allocation (256MB segment)
489  *
490  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
491  * from mmu context id and effective segment id of the address.
492  *
493  * For user processes max context id is limited to ((1ul << 19) - 5)
494  * for kernel space, we use the top 4 context ids to map address as below
495  * NOTE: each context only support 64TB now.
496  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
497  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
498  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
499  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
500  *
501  * The proto-VSIDs are then scrambled into real VSIDs with the
502  * multiplicative hash:
503  *
504  *	VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
505  *
506  * VSID_MULTIPLIER is prime, so in particular it is
507  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
508  * Because the modulus is 2^n-1 we can compute it efficiently without
509  * a divide or extra multiply (see below). The scramble function gives
510  * robust scattering in the hash table (at least based on some initial
511  * results).
512  *
513  * We also consider VSID 0 special. We use VSID 0 for slb entries mapping
514  * bad address. This enables us to consolidate bad address handling in
515  * hash_page.
516  *
517  * We also need to avoid the last segment of the last context, because that
518  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
519  * because of the modulo operation in vsid scramble. But the vmemmap
520  * (which is what uses region 0xf) will never be close to 64TB in size
521  * (it's 56 bytes per page of system memory).
522  */
523 
524 #define CONTEXT_BITS		19
525 #define ESID_BITS		18
526 #define ESID_BITS_1T		6
527 
528 /*
529  * 256MB segment
530  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
531  * available for user + kernel mapping. The top 4 contexts are used for
532  * kernel mapping. Each segment contains 2^28 bytes. Each
533  * context maps 2^46 bytes (64TB) so we can support 2^19-1 contexts
534  * (19 == 37 + 28 - 46).
535  */
536 #define MAX_USER_CONTEXT	((ASM_CONST(1) << CONTEXT_BITS) - 5)
537 
538 /*
539  * This should be computed such that protovosid * vsid_mulitplier
540  * doesn't overflow 64 bits. It should also be co-prime to vsid_modulus
541  */
542 #define VSID_MULTIPLIER_256M	ASM_CONST(12538073)	/* 24-bit prime */
543 #define VSID_BITS_256M		(CONTEXT_BITS + ESID_BITS)
544 #define VSID_MODULUS_256M	((1UL<<VSID_BITS_256M)-1)
545 
546 #define VSID_MULTIPLIER_1T	ASM_CONST(12538073)	/* 24-bit prime */
547 #define VSID_BITS_1T		(CONTEXT_BITS + ESID_BITS_1T)
548 #define VSID_MODULUS_1T		((1UL<<VSID_BITS_1T)-1)
549 
550 
551 #define USER_VSID_RANGE	(1UL << (ESID_BITS + SID_SHIFT))
552 
553 /*
554  * This macro generates asm code to compute the VSID scramble
555  * function.  Used in slb_allocate() and do_stab_bolted.  The function
556  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
557  *
558  *	rt = register containing the proto-VSID and into which the
559  *		VSID will be stored
560  *	rx = scratch register (clobbered)
561  *
562  * 	- rt and rx must be different registers
563  * 	- The answer will end up in the low VSID_BITS bits of rt.  The higher
564  * 	  bits may contain other garbage, so you may need to mask the
565  * 	  result.
566  */
567 #define ASM_VSID_SCRAMBLE(rt, rx, size)					\
568 	lis	rx,VSID_MULTIPLIER_##size@h;				\
569 	ori	rx,rx,VSID_MULTIPLIER_##size@l;				\
570 	mulld	rt,rt,rx;		/* rt = rt * MULTIPLIER */	\
571 									\
572 	srdi	rx,rt,VSID_BITS_##size;					\
573 	clrldi	rt,rt,(64-VSID_BITS_##size);				\
574 	add	rt,rt,rx;		/* add high and low bits */	\
575 	/* NOTE: explanation based on VSID_BITS_##size = 36		\
576 	 * Now, r3 == VSID (mod 2^36-1), and lies between 0 and		\
577 	 * 2^36-1+2^28-1.  That in particular means that if r3 >=	\
578 	 * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has	\
579 	 * the bit clear, r3 already has the answer we want, if it	\
580 	 * doesn't, the answer is the low 36 bits of r3+1.  So in all	\
581 	 * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
582 	addi	rx,rt,1;						\
583 	srdi	rx,rx,VSID_BITS_##size;	/* extract 2^VSID_BITS bit */	\
584 	add	rt,rt,rx
585 
586 /* 4 bits per slice and we have one slice per 1TB */
587 #define SLICE_ARRAY_SIZE  (H_PGTABLE_RANGE >> 41)
588 
589 #ifndef __ASSEMBLY__
590 
591 #ifdef CONFIG_PPC_SUBPAGE_PROT
592 /*
593  * For the sub-page protection option, we extend the PGD with one of
594  * these.  Basically we have a 3-level tree, with the top level being
595  * the protptrs array.  To optimize speed and memory consumption when
596  * only addresses < 4GB are being protected, pointers to the first
597  * four pages of sub-page protection words are stored in the low_prot
598  * array.
599  * Each page of sub-page protection words protects 1GB (4 bytes
600  * protects 64k).  For the 3-level tree, each page of pointers then
601  * protects 8TB.
602  */
603 struct subpage_prot_table {
604 	unsigned long maxaddr;	/* only addresses < this are protected */
605 	unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
606 	unsigned int *low_prot[4];
607 };
608 
609 #define SBP_L1_BITS		(PAGE_SHIFT - 2)
610 #define SBP_L2_BITS		(PAGE_SHIFT - 3)
611 #define SBP_L1_COUNT		(1 << SBP_L1_BITS)
612 #define SBP_L2_COUNT		(1 << SBP_L2_BITS)
613 #define SBP_L2_SHIFT		(PAGE_SHIFT + SBP_L1_BITS)
614 #define SBP_L3_SHIFT		(SBP_L2_SHIFT + SBP_L2_BITS)
615 
616 extern void subpage_prot_free(struct mm_struct *mm);
617 extern void subpage_prot_init_new_context(struct mm_struct *mm);
618 #else
619 static inline void subpage_prot_free(struct mm_struct *mm) {}
620 static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
621 #endif /* CONFIG_PPC_SUBPAGE_PROT */
622 
623 #if 0
624 /*
625  * The code below is equivalent to this function for arguments
626  * < 2^VSID_BITS, which is all this should ever be called
627  * with.  However gcc is not clever enough to compute the
628  * modulus (2^n-1) without a second multiply.
629  */
630 #define vsid_scramble(protovsid, size) \
631 	((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
632 
633 #else /* 1 */
634 #define vsid_scramble(protovsid, size) \
635 	({								 \
636 		unsigned long x;					 \
637 		x = (protovsid) * VSID_MULTIPLIER_##size;		 \
638 		x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
639 		(x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
640 	})
641 #endif /* 1 */
642 
643 /* Returns the segment size indicator for a user address */
644 static inline int user_segment_size(unsigned long addr)
645 {
646 	/* Use 1T segments if possible for addresses >= 1T */
647 	if (addr >= (1UL << SID_SHIFT_1T))
648 		return mmu_highuser_ssize;
649 	return MMU_SEGSIZE_256M;
650 }
651 
652 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
653 				     int ssize)
654 {
655 	/*
656 	 * Bad address. We return VSID 0 for that
657 	 */
658 	if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
659 		return 0;
660 
661 	if (ssize == MMU_SEGSIZE_256M)
662 		return vsid_scramble((context << ESID_BITS)
663 				     | (ea >> SID_SHIFT), 256M);
664 	return vsid_scramble((context << ESID_BITS_1T)
665 			     | (ea >> SID_SHIFT_1T), 1T);
666 }
667 
668 /*
669  * This is only valid for addresses >= PAGE_OFFSET
670  *
671  * For kernel space, we use the top 4 context ids to map address as below
672  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
673  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
674  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
675  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
676  */
677 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
678 {
679 	unsigned long context;
680 
681 	/*
682 	 * kernel take the top 4 context from the available range
683 	 */
684 	context = (MAX_USER_CONTEXT) + ((ea >> 60) - 0xc) + 1;
685 	return get_vsid(context, ea, ssize);
686 }
687 
688 unsigned htab_shift_for_mem_size(unsigned long mem_size);
689 
690 #endif /* __ASSEMBLY__ */
691 
692 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */
693