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 	int		(*resize_hpt)(unsigned long shift);
161 	/*
162 	 * Special for kexec.
163 	 * To be called in real mode with interrupts disabled. No locks are
164 	 * taken as such, concurrent access on pre POWER5 hardware could result
165 	 * in a deadlock.
166 	 * The linear mapping is destroyed as well.
167 	 */
168 	void		(*hpte_clear_all)(void);
169 };
170 extern struct mmu_hash_ops mmu_hash_ops;
171 
172 struct hash_pte {
173 	__be64 v;
174 	__be64 r;
175 };
176 
177 extern struct hash_pte *htab_address;
178 extern unsigned long htab_size_bytes;
179 extern unsigned long htab_hash_mask;
180 
181 
182 static inline int shift_to_mmu_psize(unsigned int shift)
183 {
184 	int psize;
185 
186 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
187 		if (mmu_psize_defs[psize].shift == shift)
188 			return psize;
189 	return -1;
190 }
191 
192 static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
193 {
194 	if (mmu_psize_defs[mmu_psize].shift)
195 		return mmu_psize_defs[mmu_psize].shift;
196 	BUG();
197 }
198 
199 static inline unsigned long get_sllp_encoding(int psize)
200 {
201 	unsigned long sllp;
202 
203 	sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
204 		((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
205 	return sllp;
206 }
207 
208 #endif /* __ASSEMBLY__ */
209 
210 /*
211  * Segment sizes.
212  * These are the values used by hardware in the B field of
213  * SLB entries and the first dword of MMU hashtable entries.
214  * The B field is 2 bits; the values 2 and 3 are unused and reserved.
215  */
216 #define MMU_SEGSIZE_256M	0
217 #define MMU_SEGSIZE_1T		1
218 
219 /*
220  * encode page number shift.
221  * in order to fit the 78 bit va in a 64 bit variable we shift the va by
222  * 12 bits. This enable us to address upto 76 bit va.
223  * For hpt hash from a va we can ignore the page size bits of va and for
224  * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
225  * we work in all cases including 4k page size.
226  */
227 #define VPN_SHIFT	12
228 
229 /*
230  * HPTE Large Page (LP) details
231  */
232 #define LP_SHIFT	12
233 #define LP_BITS		8
234 #define LP_MASK(i)	((0xFF >> (i)) << LP_SHIFT)
235 
236 #ifndef __ASSEMBLY__
237 
238 static inline int slb_vsid_shift(int ssize)
239 {
240 	if (ssize == MMU_SEGSIZE_256M)
241 		return SLB_VSID_SHIFT;
242 	return SLB_VSID_SHIFT_1T;
243 }
244 
245 static inline int segment_shift(int ssize)
246 {
247 	if (ssize == MMU_SEGSIZE_256M)
248 		return SID_SHIFT;
249 	return SID_SHIFT_1T;
250 }
251 
252 /*
253  * This array is indexed by the LP field of the HPTE second dword.
254  * Since this field may contain some RPN bits, some entries are
255  * replicated so that we get the same value irrespective of RPN.
256  * The top 4 bits are the page size index (MMU_PAGE_*) for the
257  * actual page size, the bottom 4 bits are the base page size.
258  */
259 extern u8 hpte_page_sizes[1 << LP_BITS];
260 
261 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
262 					     bool is_base_size)
263 {
264 	unsigned int i, lp;
265 
266 	if (!(h & HPTE_V_LARGE))
267 		return 1ul << 12;
268 
269 	/* Look at the 8 bit LP value */
270 	lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
271 	i = hpte_page_sizes[lp];
272 	if (!i)
273 		return 0;
274 	if (!is_base_size)
275 		i >>= 4;
276 	return 1ul << mmu_psize_defs[i & 0xf].shift;
277 }
278 
279 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
280 {
281 	return __hpte_page_size(h, l, 0);
282 }
283 
284 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
285 {
286 	return __hpte_page_size(h, l, 1);
287 }
288 
289 /*
290  * The current system page and segment sizes
291  */
292 extern int mmu_kernel_ssize;
293 extern int mmu_highuser_ssize;
294 extern u16 mmu_slb_size;
295 extern unsigned long tce_alloc_start, tce_alloc_end;
296 
297 /*
298  * If the processor supports 64k normal pages but not 64k cache
299  * inhibited pages, we have to be prepared to switch processes
300  * to use 4k pages when they create cache-inhibited mappings.
301  * If this is the case, mmu_ci_restrictions will be set to 1.
302  */
303 extern int mmu_ci_restrictions;
304 
305 /*
306  * This computes the AVPN and B fields of the first dword of a HPTE,
307  * for use when we want to match an existing PTE.  The bottom 7 bits
308  * of the returned value are zero.
309  */
310 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
311 					     int ssize)
312 {
313 	unsigned long v;
314 	/*
315 	 * The AVA field omits the low-order 23 bits of the 78 bits VA.
316 	 * These bits are not needed in the PTE, because the
317 	 * low-order b of these bits are part of the byte offset
318 	 * into the virtual page and, if b < 23, the high-order
319 	 * 23-b of these bits are always used in selecting the
320 	 * PTEGs to be searched
321 	 */
322 	v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
323 	v <<= HPTE_V_AVPN_SHIFT;
324 	v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
325 	return v;
326 }
327 
328 /*
329  * ISA v3.0 defines a new HPTE format, which differs from the old
330  * format in having smaller AVPN and ARPN fields, and the B field
331  * in the second dword instead of the first.
332  */
333 static inline unsigned long hpte_old_to_new_v(unsigned long v)
334 {
335 	/* trim AVPN, drop B */
336 	return v & HPTE_V_COMMON_BITS;
337 }
338 
339 static inline unsigned long hpte_old_to_new_r(unsigned long v, unsigned long r)
340 {
341 	/* move B field from 1st to 2nd dword, trim ARPN */
342 	return (r & ~HPTE_R_3_0_SSIZE_MASK) |
343 		(((v) >> HPTE_V_SSIZE_SHIFT) << HPTE_R_3_0_SSIZE_SHIFT);
344 }
345 
346 static inline unsigned long hpte_new_to_old_v(unsigned long v, unsigned long r)
347 {
348 	/* insert B field */
349 	return (v & HPTE_V_COMMON_BITS) |
350 		((r & HPTE_R_3_0_SSIZE_MASK) <<
351 		 (HPTE_V_SSIZE_SHIFT - HPTE_R_3_0_SSIZE_SHIFT));
352 }
353 
354 static inline unsigned long hpte_new_to_old_r(unsigned long r)
355 {
356 	/* clear out B field */
357 	return r & ~HPTE_R_3_0_SSIZE_MASK;
358 }
359 
360 /*
361  * This function sets the AVPN and L fields of the HPTE  appropriately
362  * using the base page size and actual page size.
363  */
364 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
365 					  int actual_psize, int ssize)
366 {
367 	unsigned long v;
368 	v = hpte_encode_avpn(vpn, base_psize, ssize);
369 	if (actual_psize != MMU_PAGE_4K)
370 		v |= HPTE_V_LARGE;
371 	return v;
372 }
373 
374 /*
375  * This function sets the ARPN, and LP fields of the HPTE appropriately
376  * for the page size. We assume the pa is already "clean" that is properly
377  * aligned for the requested page size
378  */
379 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
380 					  int actual_psize)
381 {
382 	/* A 4K page needs no special encoding */
383 	if (actual_psize == MMU_PAGE_4K)
384 		return pa & HPTE_R_RPN;
385 	else {
386 		unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
387 		unsigned int shift = mmu_psize_defs[actual_psize].shift;
388 		return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
389 	}
390 }
391 
392 /*
393  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
394  */
395 static inline unsigned long hpt_vpn(unsigned long ea,
396 				    unsigned long vsid, int ssize)
397 {
398 	unsigned long mask;
399 	int s_shift = segment_shift(ssize);
400 
401 	mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
402 	return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
403 }
404 
405 /*
406  * This hashes a virtual address
407  */
408 static inline unsigned long hpt_hash(unsigned long vpn,
409 				     unsigned int shift, int ssize)
410 {
411 	int mask;
412 	unsigned long hash, vsid;
413 
414 	/* VPN_SHIFT can be atmost 12 */
415 	if (ssize == MMU_SEGSIZE_256M) {
416 		mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
417 		hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
418 			((vpn & mask) >> (shift - VPN_SHIFT));
419 	} else {
420 		mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
421 		vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
422 		hash = vsid ^ (vsid << 25) ^
423 			((vpn & mask) >> (shift - VPN_SHIFT)) ;
424 	}
425 	return hash & 0x7fffffffffUL;
426 }
427 
428 #define HPTE_LOCAL_UPDATE	0x1
429 #define HPTE_NOHPTE_UPDATE	0x2
430 
431 extern int __hash_page_4K(unsigned long ea, unsigned long access,
432 			  unsigned long vsid, pte_t *ptep, unsigned long trap,
433 			  unsigned long flags, int ssize, int subpage_prot);
434 extern int __hash_page_64K(unsigned long ea, unsigned long access,
435 			   unsigned long vsid, pte_t *ptep, unsigned long trap,
436 			   unsigned long flags, int ssize);
437 struct mm_struct;
438 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
439 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
440 			unsigned long access, unsigned long trap,
441 			unsigned long flags);
442 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
443 		     unsigned long dsisr);
444 int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
445 		     pte_t *ptep, unsigned long trap, unsigned long flags,
446 		     int ssize, unsigned int shift, unsigned int mmu_psize);
447 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
448 extern int __hash_page_thp(unsigned long ea, unsigned long access,
449 			   unsigned long vsid, pmd_t *pmdp, unsigned long trap,
450 			   unsigned long flags, int ssize, unsigned int psize);
451 #else
452 static inline int __hash_page_thp(unsigned long ea, unsigned long access,
453 				  unsigned long vsid, pmd_t *pmdp,
454 				  unsigned long trap, unsigned long flags,
455 				  int ssize, unsigned int psize)
456 {
457 	BUG();
458 	return -1;
459 }
460 #endif
461 extern void hash_failure_debug(unsigned long ea, unsigned long access,
462 			       unsigned long vsid, unsigned long trap,
463 			       int ssize, int psize, int lpsize,
464 			       unsigned long pte);
465 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
466 			     unsigned long pstart, unsigned long prot,
467 			     int psize, int ssize);
468 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
469 			int psize, int ssize);
470 extern void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
471 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
472 
473 #ifdef CONFIG_PPC_PSERIES
474 void hpte_init_pseries(void);
475 #else
476 static inline void hpte_init_pseries(void) { }
477 #endif
478 
479 extern void hpte_init_native(void);
480 
481 extern void slb_initialize(void);
482 extern void slb_flush_and_rebolt(void);
483 
484 extern void slb_vmalloc_update(void);
485 extern void slb_set_size(u16 size);
486 #endif /* __ASSEMBLY__ */
487 
488 /*
489  * VSID allocation (256MB segment)
490  *
491  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
492  * from mmu context id and effective segment id of the address.
493  *
494  * For user processes max context id is limited to ((1ul << 19) - 5)
495  * for kernel space, we use the top 4 context ids to map address as below
496  * NOTE: each context only support 64TB now.
497  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
498  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
499  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
500  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
501  *
502  * The proto-VSIDs are then scrambled into real VSIDs with the
503  * multiplicative hash:
504  *
505  *	VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
506  *
507  * VSID_MULTIPLIER is prime, so in particular it is
508  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
509  * Because the modulus is 2^n-1 we can compute it efficiently without
510  * a divide or extra multiply (see below). The scramble function gives
511  * robust scattering in the hash table (at least based on some initial
512  * results).
513  *
514  * We also consider VSID 0 special. We use VSID 0 for slb entries mapping
515  * bad address. This enables us to consolidate bad address handling in
516  * hash_page.
517  *
518  * We also need to avoid the last segment of the last context, because that
519  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
520  * because of the modulo operation in vsid scramble. But the vmemmap
521  * (which is what uses region 0xf) will never be close to 64TB in size
522  * (it's 56 bytes per page of system memory).
523  */
524 
525 #define CONTEXT_BITS		19
526 #define ESID_BITS		18
527 #define ESID_BITS_1T		6
528 
529 #define ESID_BITS_MASK		((1 << ESID_BITS) - 1)
530 #define ESID_BITS_1T_MASK	((1 << ESID_BITS_1T) - 1)
531 
532 /*
533  * 256MB segment
534  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
535  * available for user + kernel mapping. The top 4 contexts are used for
536  * kernel mapping. Each segment contains 2^28 bytes. Each
537  * context maps 2^46 bytes (64TB) so we can support 2^19-1 contexts
538  * (19 == 37 + 28 - 46).
539  */
540 #define MAX_USER_CONTEXT	((ASM_CONST(1) << CONTEXT_BITS) - 5)
541 
542 /*
543  * This should be computed such that protovosid * vsid_mulitplier
544  * doesn't overflow 64 bits. It should also be co-prime to vsid_modulus
545  */
546 #define VSID_MULTIPLIER_256M	ASM_CONST(12538073)	/* 24-bit prime */
547 #define VSID_BITS_256M		(CONTEXT_BITS + ESID_BITS)
548 #define VSID_MODULUS_256M	((1UL<<VSID_BITS_256M)-1)
549 
550 #define VSID_MULTIPLIER_1T	ASM_CONST(12538073)	/* 24-bit prime */
551 #define VSID_BITS_1T		(CONTEXT_BITS + ESID_BITS_1T)
552 #define VSID_MODULUS_1T		((1UL<<VSID_BITS_1T)-1)
553 
554 
555 #define USER_VSID_RANGE	(1UL << (ESID_BITS + SID_SHIFT))
556 
557 /*
558  * This macro generates asm code to compute the VSID scramble
559  * function.  Used in slb_allocate() and do_stab_bolted.  The function
560  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
561  *
562  *	rt = register containing the proto-VSID and into which the
563  *		VSID will be stored
564  *	rx = scratch register (clobbered)
565  *
566  * 	- rt and rx must be different registers
567  * 	- The answer will end up in the low VSID_BITS bits of rt.  The higher
568  * 	  bits may contain other garbage, so you may need to mask the
569  * 	  result.
570  */
571 #define ASM_VSID_SCRAMBLE(rt, rx, size)					\
572 	lis	rx,VSID_MULTIPLIER_##size@h;				\
573 	ori	rx,rx,VSID_MULTIPLIER_##size@l;				\
574 	mulld	rt,rt,rx;		/* rt = rt * MULTIPLIER */	\
575 									\
576 	srdi	rx,rt,VSID_BITS_##size;					\
577 	clrldi	rt,rt,(64-VSID_BITS_##size);				\
578 	add	rt,rt,rx;		/* add high and low bits */	\
579 	/* NOTE: explanation based on VSID_BITS_##size = 36		\
580 	 * Now, r3 == VSID (mod 2^36-1), and lies between 0 and		\
581 	 * 2^36-1+2^28-1.  That in particular means that if r3 >=	\
582 	 * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has	\
583 	 * the bit clear, r3 already has the answer we want, if it	\
584 	 * doesn't, the answer is the low 36 bits of r3+1.  So in all	\
585 	 * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
586 	addi	rx,rt,1;						\
587 	srdi	rx,rx,VSID_BITS_##size;	/* extract 2^VSID_BITS bit */	\
588 	add	rt,rt,rx
589 
590 /* 4 bits per slice and we have one slice per 1TB */
591 #define SLICE_ARRAY_SIZE  (H_PGTABLE_RANGE >> 41)
592 
593 #ifndef __ASSEMBLY__
594 
595 #ifdef CONFIG_PPC_SUBPAGE_PROT
596 /*
597  * For the sub-page protection option, we extend the PGD with one of
598  * these.  Basically we have a 3-level tree, with the top level being
599  * the protptrs array.  To optimize speed and memory consumption when
600  * only addresses < 4GB are being protected, pointers to the first
601  * four pages of sub-page protection words are stored in the low_prot
602  * array.
603  * Each page of sub-page protection words protects 1GB (4 bytes
604  * protects 64k).  For the 3-level tree, each page of pointers then
605  * protects 8TB.
606  */
607 struct subpage_prot_table {
608 	unsigned long maxaddr;	/* only addresses < this are protected */
609 	unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
610 	unsigned int *low_prot[4];
611 };
612 
613 #define SBP_L1_BITS		(PAGE_SHIFT - 2)
614 #define SBP_L2_BITS		(PAGE_SHIFT - 3)
615 #define SBP_L1_COUNT		(1 << SBP_L1_BITS)
616 #define SBP_L2_COUNT		(1 << SBP_L2_BITS)
617 #define SBP_L2_SHIFT		(PAGE_SHIFT + SBP_L1_BITS)
618 #define SBP_L3_SHIFT		(SBP_L2_SHIFT + SBP_L2_BITS)
619 
620 extern void subpage_prot_free(struct mm_struct *mm);
621 extern void subpage_prot_init_new_context(struct mm_struct *mm);
622 #else
623 static inline void subpage_prot_free(struct mm_struct *mm) {}
624 static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
625 #endif /* CONFIG_PPC_SUBPAGE_PROT */
626 
627 #if 0
628 /*
629  * The code below is equivalent to this function for arguments
630  * < 2^VSID_BITS, which is all this should ever be called
631  * with.  However gcc is not clever enough to compute the
632  * modulus (2^n-1) without a second multiply.
633  */
634 #define vsid_scramble(protovsid, size) \
635 	((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
636 
637 #else /* 1 */
638 #define vsid_scramble(protovsid, size) \
639 	({								 \
640 		unsigned long x;					 \
641 		x = (protovsid) * VSID_MULTIPLIER_##size;		 \
642 		x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
643 		(x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
644 	})
645 #endif /* 1 */
646 
647 /* Returns the segment size indicator for a user address */
648 static inline int user_segment_size(unsigned long addr)
649 {
650 	/* Use 1T segments if possible for addresses >= 1T */
651 	if (addr >= (1UL << SID_SHIFT_1T))
652 		return mmu_highuser_ssize;
653 	return MMU_SEGSIZE_256M;
654 }
655 
656 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
657 				     int ssize)
658 {
659 	/*
660 	 * Bad address. We return VSID 0 for that
661 	 */
662 	if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
663 		return 0;
664 
665 	if (ssize == MMU_SEGSIZE_256M)
666 		return vsid_scramble((context << ESID_BITS)
667 				     | ((ea >> SID_SHIFT) & ESID_BITS_MASK), 256M);
668 	return vsid_scramble((context << ESID_BITS_1T)
669 			     | ((ea >> SID_SHIFT_1T) & ESID_BITS_1T_MASK), 1T);
670 }
671 
672 /*
673  * This is only valid for addresses >= PAGE_OFFSET
674  *
675  * For kernel space, we use the top 4 context ids to map address as below
676  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
677  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
678  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
679  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
680  */
681 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
682 {
683 	unsigned long context;
684 
685 	/*
686 	 * kernel take the top 4 context from the available range
687 	 */
688 	context = (MAX_USER_CONTEXT) + ((ea >> 60) - 0xc) + 1;
689 	return get_vsid(context, ea, ssize);
690 }
691 
692 unsigned htab_shift_for_mem_size(unsigned long mem_size);
693 
694 #endif /* __ASSEMBLY__ */
695 
696 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */
697