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_AVPN		ASM_CONST(0x3fffffffffffff80)
74 #define HPTE_V_AVPN_VAL(x)	(((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
75 #define HPTE_V_COMPARE(x,y)	(!(((x) ^ (y)) & 0xffffffffffffff80UL))
76 #define HPTE_V_BOLTED		ASM_CONST(0x0000000000000010)
77 #define HPTE_V_LOCK		ASM_CONST(0x0000000000000008)
78 #define HPTE_V_LARGE		ASM_CONST(0x0000000000000004)
79 #define HPTE_V_SECONDARY	ASM_CONST(0x0000000000000002)
80 #define HPTE_V_VALID		ASM_CONST(0x0000000000000001)
81 
82 /*
83  * ISA 3.0 have a different HPTE format.
84  */
85 #define HPTE_R_3_0_SSIZE_SHIFT	58
86 #define HPTE_R_PP0		ASM_CONST(0x8000000000000000)
87 #define HPTE_R_TS		ASM_CONST(0x4000000000000000)
88 #define HPTE_R_KEY_HI		ASM_CONST(0x3000000000000000)
89 #define HPTE_R_RPN_SHIFT	12
90 #define HPTE_R_RPN		ASM_CONST(0x0ffffffffffff000)
91 #define HPTE_R_PP		ASM_CONST(0x0000000000000003)
92 #define HPTE_R_PPP		ASM_CONST(0x8000000000000003)
93 #define HPTE_R_N		ASM_CONST(0x0000000000000004)
94 #define HPTE_R_G		ASM_CONST(0x0000000000000008)
95 #define HPTE_R_M		ASM_CONST(0x0000000000000010)
96 #define HPTE_R_I		ASM_CONST(0x0000000000000020)
97 #define HPTE_R_W		ASM_CONST(0x0000000000000040)
98 #define HPTE_R_WIMG		ASM_CONST(0x0000000000000078)
99 #define HPTE_R_C		ASM_CONST(0x0000000000000080)
100 #define HPTE_R_R		ASM_CONST(0x0000000000000100)
101 #define HPTE_R_KEY_LO		ASM_CONST(0x0000000000000e00)
102 
103 #define HPTE_V_1TB_SEG		ASM_CONST(0x4000000000000000)
104 #define HPTE_V_VRMA_MASK	ASM_CONST(0x4001ffffff000000)
105 
106 /* Values for PP (assumes Ks=0, Kp=1) */
107 #define PP_RWXX	0	/* Supervisor read/write, User none */
108 #define PP_RWRX 1	/* Supervisor read/write, User read */
109 #define PP_RWRW 2	/* Supervisor read/write, User read/write */
110 #define PP_RXRX 3	/* Supervisor read,       User read */
111 #define PP_RXXX	(HPTE_R_PP0 | 2)	/* Supervisor read, user none */
112 
113 /* Fields for tlbiel instruction in architecture 2.06 */
114 #define TLBIEL_INVAL_SEL_MASK	0xc00	/* invalidation selector */
115 #define  TLBIEL_INVAL_PAGE	0x000	/* invalidate a single page */
116 #define  TLBIEL_INVAL_SET_LPID	0x800	/* invalidate a set for current LPID */
117 #define  TLBIEL_INVAL_SET	0xc00	/* invalidate a set for all LPIDs */
118 #define TLBIEL_INVAL_SET_MASK	0xfff000	/* set number to inval. */
119 #define TLBIEL_INVAL_SET_SHIFT	12
120 
121 #define POWER7_TLB_SETS		128	/* # sets in POWER7 TLB */
122 #define POWER8_TLB_SETS		512	/* # sets in POWER8 TLB */
123 #define POWER9_TLB_SETS_HASH	256	/* # sets in POWER9 TLB Hash mode */
124 #define POWER9_TLB_SETS_RADIX	128	/* # sets in POWER9 TLB Radix mode */
125 
126 #ifndef __ASSEMBLY__
127 
128 struct mmu_hash_ops {
129 	void            (*hpte_invalidate)(unsigned long slot,
130 					   unsigned long vpn,
131 					   int bpsize, int apsize,
132 					   int ssize, int local);
133 	long		(*hpte_updatepp)(unsigned long slot,
134 					 unsigned long newpp,
135 					 unsigned long vpn,
136 					 int bpsize, int apsize,
137 					 int ssize, unsigned long flags);
138 	void            (*hpte_updateboltedpp)(unsigned long newpp,
139 					       unsigned long ea,
140 					       int psize, int ssize);
141 	long		(*hpte_insert)(unsigned long hpte_group,
142 				       unsigned long vpn,
143 				       unsigned long prpn,
144 				       unsigned long rflags,
145 				       unsigned long vflags,
146 				       int psize, int apsize,
147 				       int ssize);
148 	long		(*hpte_remove)(unsigned long hpte_group);
149 	int             (*hpte_removebolted)(unsigned long ea,
150 					     int psize, int ssize);
151 	void		(*flush_hash_range)(unsigned long number, int local);
152 	void		(*hugepage_invalidate)(unsigned long vsid,
153 					       unsigned long addr,
154 					       unsigned char *hpte_slot_array,
155 					       int psize, int ssize, int local);
156 	/*
157 	 * Special for kexec.
158 	 * To be called in real mode with interrupts disabled. No locks are
159 	 * taken as such, concurrent access on pre POWER5 hardware could result
160 	 * in a deadlock.
161 	 * The linear mapping is destroyed as well.
162 	 */
163 	void		(*hpte_clear_all)(void);
164 };
165 extern struct mmu_hash_ops mmu_hash_ops;
166 
167 struct hash_pte {
168 	__be64 v;
169 	__be64 r;
170 };
171 
172 extern struct hash_pte *htab_address;
173 extern unsigned long htab_size_bytes;
174 extern unsigned long htab_hash_mask;
175 
176 
177 static inline int shift_to_mmu_psize(unsigned int shift)
178 {
179 	int psize;
180 
181 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
182 		if (mmu_psize_defs[psize].shift == shift)
183 			return psize;
184 	return -1;
185 }
186 
187 static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
188 {
189 	if (mmu_psize_defs[mmu_psize].shift)
190 		return mmu_psize_defs[mmu_psize].shift;
191 	BUG();
192 }
193 
194 static inline unsigned long get_sllp_encoding(int psize)
195 {
196 	unsigned long sllp;
197 
198 	sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
199 		((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
200 	return sllp;
201 }
202 
203 #endif /* __ASSEMBLY__ */
204 
205 /*
206  * Segment sizes.
207  * These are the values used by hardware in the B field of
208  * SLB entries and the first dword of MMU hashtable entries.
209  * The B field is 2 bits; the values 2 and 3 are unused and reserved.
210  */
211 #define MMU_SEGSIZE_256M	0
212 #define MMU_SEGSIZE_1T		1
213 
214 /*
215  * encode page number shift.
216  * in order to fit the 78 bit va in a 64 bit variable we shift the va by
217  * 12 bits. This enable us to address upto 76 bit va.
218  * For hpt hash from a va we can ignore the page size bits of va and for
219  * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
220  * we work in all cases including 4k page size.
221  */
222 #define VPN_SHIFT	12
223 
224 /*
225  * HPTE Large Page (LP) details
226  */
227 #define LP_SHIFT	12
228 #define LP_BITS		8
229 #define LP_MASK(i)	((0xFF >> (i)) << LP_SHIFT)
230 
231 #ifndef __ASSEMBLY__
232 
233 static inline int slb_vsid_shift(int ssize)
234 {
235 	if (ssize == MMU_SEGSIZE_256M)
236 		return SLB_VSID_SHIFT;
237 	return SLB_VSID_SHIFT_1T;
238 }
239 
240 static inline int segment_shift(int ssize)
241 {
242 	if (ssize == MMU_SEGSIZE_256M)
243 		return SID_SHIFT;
244 	return SID_SHIFT_1T;
245 }
246 
247 /*
248  * This array is indexed by the LP field of the HPTE second dword.
249  * Since this field may contain some RPN bits, some entries are
250  * replicated so that we get the same value irrespective of RPN.
251  * The top 4 bits are the page size index (MMU_PAGE_*) for the
252  * actual page size, the bottom 4 bits are the base page size.
253  */
254 extern u8 hpte_page_sizes[1 << LP_BITS];
255 
256 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
257 					     bool is_base_size)
258 {
259 	unsigned int i, lp;
260 
261 	if (!(h & HPTE_V_LARGE))
262 		return 1ul << 12;
263 
264 	/* Look at the 8 bit LP value */
265 	lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
266 	i = hpte_page_sizes[lp];
267 	if (!i)
268 		return 0;
269 	if (!is_base_size)
270 		i >>= 4;
271 	return 1ul << mmu_psize_defs[i & 0xf].shift;
272 }
273 
274 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
275 {
276 	return __hpte_page_size(h, l, 0);
277 }
278 
279 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
280 {
281 	return __hpte_page_size(h, l, 1);
282 }
283 
284 /*
285  * The current system page and segment sizes
286  */
287 extern int mmu_kernel_ssize;
288 extern int mmu_highuser_ssize;
289 extern u16 mmu_slb_size;
290 extern unsigned long tce_alloc_start, tce_alloc_end;
291 
292 /*
293  * If the processor supports 64k normal pages but not 64k cache
294  * inhibited pages, we have to be prepared to switch processes
295  * to use 4k pages when they create cache-inhibited mappings.
296  * If this is the case, mmu_ci_restrictions will be set to 1.
297  */
298 extern int mmu_ci_restrictions;
299 
300 /*
301  * This computes the AVPN and B fields of the first dword of a HPTE,
302  * for use when we want to match an existing PTE.  The bottom 7 bits
303  * of the returned value are zero.
304  */
305 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
306 					     int ssize)
307 {
308 	unsigned long v;
309 	/*
310 	 * The AVA field omits the low-order 23 bits of the 78 bits VA.
311 	 * These bits are not needed in the PTE, because the
312 	 * low-order b of these bits are part of the byte offset
313 	 * into the virtual page and, if b < 23, the high-order
314 	 * 23-b of these bits are always used in selecting the
315 	 * PTEGs to be searched
316 	 */
317 	v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
318 	v <<= HPTE_V_AVPN_SHIFT;
319 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
320 		v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
321 	return v;
322 }
323 
324 /*
325  * This function sets the AVPN and L fields of the HPTE  appropriately
326  * using the base page size and actual page size.
327  */
328 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
329 					  int actual_psize, int ssize)
330 {
331 	unsigned long v;
332 	v = hpte_encode_avpn(vpn, base_psize, ssize);
333 	if (actual_psize != MMU_PAGE_4K)
334 		v |= HPTE_V_LARGE;
335 	return v;
336 }
337 
338 /*
339  * This function sets the ARPN, and LP fields of the HPTE appropriately
340  * for the page size. We assume the pa is already "clean" that is properly
341  * aligned for the requested page size
342  */
343 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
344 					  int actual_psize, int ssize)
345 {
346 
347 	if (cpu_has_feature(CPU_FTR_ARCH_300))
348 		pa |= ((unsigned long) ssize) << HPTE_R_3_0_SSIZE_SHIFT;
349 
350 	/* A 4K page needs no special encoding */
351 	if (actual_psize == MMU_PAGE_4K)
352 		return pa & HPTE_R_RPN;
353 	else {
354 		unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
355 		unsigned int shift = mmu_psize_defs[actual_psize].shift;
356 		return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
357 	}
358 }
359 
360 /*
361  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
362  */
363 static inline unsigned long hpt_vpn(unsigned long ea,
364 				    unsigned long vsid, int ssize)
365 {
366 	unsigned long mask;
367 	int s_shift = segment_shift(ssize);
368 
369 	mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
370 	return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
371 }
372 
373 /*
374  * This hashes a virtual address
375  */
376 static inline unsigned long hpt_hash(unsigned long vpn,
377 				     unsigned int shift, int ssize)
378 {
379 	int mask;
380 	unsigned long hash, vsid;
381 
382 	/* VPN_SHIFT can be atmost 12 */
383 	if (ssize == MMU_SEGSIZE_256M) {
384 		mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
385 		hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
386 			((vpn & mask) >> (shift - VPN_SHIFT));
387 	} else {
388 		mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
389 		vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
390 		hash = vsid ^ (vsid << 25) ^
391 			((vpn & mask) >> (shift - VPN_SHIFT)) ;
392 	}
393 	return hash & 0x7fffffffffUL;
394 }
395 
396 #define HPTE_LOCAL_UPDATE	0x1
397 #define HPTE_NOHPTE_UPDATE	0x2
398 
399 extern int __hash_page_4K(unsigned long ea, unsigned long access,
400 			  unsigned long vsid, pte_t *ptep, unsigned long trap,
401 			  unsigned long flags, int ssize, int subpage_prot);
402 extern int __hash_page_64K(unsigned long ea, unsigned long access,
403 			   unsigned long vsid, pte_t *ptep, unsigned long trap,
404 			   unsigned long flags, int ssize);
405 struct mm_struct;
406 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
407 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
408 			unsigned long access, unsigned long trap,
409 			unsigned long flags);
410 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
411 		     unsigned long dsisr);
412 int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
413 		     pte_t *ptep, unsigned long trap, unsigned long flags,
414 		     int ssize, unsigned int shift, unsigned int mmu_psize);
415 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
416 extern int __hash_page_thp(unsigned long ea, unsigned long access,
417 			   unsigned long vsid, pmd_t *pmdp, unsigned long trap,
418 			   unsigned long flags, int ssize, unsigned int psize);
419 #else
420 static inline int __hash_page_thp(unsigned long ea, unsigned long access,
421 				  unsigned long vsid, pmd_t *pmdp,
422 				  unsigned long trap, unsigned long flags,
423 				  int ssize, unsigned int psize)
424 {
425 	BUG();
426 	return -1;
427 }
428 #endif
429 extern void hash_failure_debug(unsigned long ea, unsigned long access,
430 			       unsigned long vsid, unsigned long trap,
431 			       int ssize, int psize, int lpsize,
432 			       unsigned long pte);
433 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
434 			     unsigned long pstart, unsigned long prot,
435 			     int psize, int ssize);
436 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
437 			int psize, int ssize);
438 extern void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
439 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
440 
441 #ifdef CONFIG_PPC_PSERIES
442 void hpte_init_pseries(void);
443 #else
444 static inline void hpte_init_pseries(void) { }
445 #endif
446 
447 extern void hpte_init_native(void);
448 
449 extern void slb_initialize(void);
450 extern void slb_flush_and_rebolt(void);
451 
452 extern void slb_vmalloc_update(void);
453 extern void slb_set_size(u16 size);
454 #endif /* __ASSEMBLY__ */
455 
456 /*
457  * VSID allocation (256MB segment)
458  *
459  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
460  * from mmu context id and effective segment id of the address.
461  *
462  * For user processes max context id is limited to ((1ul << 19) - 5)
463  * for kernel space, we use the top 4 context ids to map address as below
464  * NOTE: each context only support 64TB now.
465  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
466  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
467  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
468  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
469  *
470  * The proto-VSIDs are then scrambled into real VSIDs with the
471  * multiplicative hash:
472  *
473  *	VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
474  *
475  * VSID_MULTIPLIER is prime, so in particular it is
476  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
477  * Because the modulus is 2^n-1 we can compute it efficiently without
478  * a divide or extra multiply (see below). The scramble function gives
479  * robust scattering in the hash table (at least based on some initial
480  * results).
481  *
482  * We also consider VSID 0 special. We use VSID 0 for slb entries mapping
483  * bad address. This enables us to consolidate bad address handling in
484  * hash_page.
485  *
486  * We also need to avoid the last segment of the last context, because that
487  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
488  * because of the modulo operation in vsid scramble. But the vmemmap
489  * (which is what uses region 0xf) will never be close to 64TB in size
490  * (it's 56 bytes per page of system memory).
491  */
492 
493 #define CONTEXT_BITS		19
494 #define ESID_BITS		18
495 #define ESID_BITS_1T		6
496 
497 /*
498  * 256MB segment
499  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
500  * available for user + kernel mapping. The top 4 contexts are used for
501  * kernel mapping. Each segment contains 2^28 bytes. Each
502  * context maps 2^46 bytes (64TB) so we can support 2^19-1 contexts
503  * (19 == 37 + 28 - 46).
504  */
505 #define MAX_USER_CONTEXT	((ASM_CONST(1) << CONTEXT_BITS) - 5)
506 
507 /*
508  * This should be computed such that protovosid * vsid_mulitplier
509  * doesn't overflow 64 bits. It should also be co-prime to vsid_modulus
510  */
511 #define VSID_MULTIPLIER_256M	ASM_CONST(12538073)	/* 24-bit prime */
512 #define VSID_BITS_256M		(CONTEXT_BITS + ESID_BITS)
513 #define VSID_MODULUS_256M	((1UL<<VSID_BITS_256M)-1)
514 
515 #define VSID_MULTIPLIER_1T	ASM_CONST(12538073)	/* 24-bit prime */
516 #define VSID_BITS_1T		(CONTEXT_BITS + ESID_BITS_1T)
517 #define VSID_MODULUS_1T		((1UL<<VSID_BITS_1T)-1)
518 
519 
520 #define USER_VSID_RANGE	(1UL << (ESID_BITS + SID_SHIFT))
521 
522 /*
523  * This macro generates asm code to compute the VSID scramble
524  * function.  Used in slb_allocate() and do_stab_bolted.  The function
525  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
526  *
527  *	rt = register containing the proto-VSID and into which the
528  *		VSID will be stored
529  *	rx = scratch register (clobbered)
530  *
531  * 	- rt and rx must be different registers
532  * 	- The answer will end up in the low VSID_BITS bits of rt.  The higher
533  * 	  bits may contain other garbage, so you may need to mask the
534  * 	  result.
535  */
536 #define ASM_VSID_SCRAMBLE(rt, rx, size)					\
537 	lis	rx,VSID_MULTIPLIER_##size@h;				\
538 	ori	rx,rx,VSID_MULTIPLIER_##size@l;				\
539 	mulld	rt,rt,rx;		/* rt = rt * MULTIPLIER */	\
540 									\
541 	srdi	rx,rt,VSID_BITS_##size;					\
542 	clrldi	rt,rt,(64-VSID_BITS_##size);				\
543 	add	rt,rt,rx;		/* add high and low bits */	\
544 	/* NOTE: explanation based on VSID_BITS_##size = 36		\
545 	 * Now, r3 == VSID (mod 2^36-1), and lies between 0 and		\
546 	 * 2^36-1+2^28-1.  That in particular means that if r3 >=	\
547 	 * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has	\
548 	 * the bit clear, r3 already has the answer we want, if it	\
549 	 * doesn't, the answer is the low 36 bits of r3+1.  So in all	\
550 	 * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
551 	addi	rx,rt,1;						\
552 	srdi	rx,rx,VSID_BITS_##size;	/* extract 2^VSID_BITS bit */	\
553 	add	rt,rt,rx
554 
555 /* 4 bits per slice and we have one slice per 1TB */
556 #define SLICE_ARRAY_SIZE  (H_PGTABLE_RANGE >> 41)
557 
558 #ifndef __ASSEMBLY__
559 
560 #ifdef CONFIG_PPC_SUBPAGE_PROT
561 /*
562  * For the sub-page protection option, we extend the PGD with one of
563  * these.  Basically we have a 3-level tree, with the top level being
564  * the protptrs array.  To optimize speed and memory consumption when
565  * only addresses < 4GB are being protected, pointers to the first
566  * four pages of sub-page protection words are stored in the low_prot
567  * array.
568  * Each page of sub-page protection words protects 1GB (4 bytes
569  * protects 64k).  For the 3-level tree, each page of pointers then
570  * protects 8TB.
571  */
572 struct subpage_prot_table {
573 	unsigned long maxaddr;	/* only addresses < this are protected */
574 	unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
575 	unsigned int *low_prot[4];
576 };
577 
578 #define SBP_L1_BITS		(PAGE_SHIFT - 2)
579 #define SBP_L2_BITS		(PAGE_SHIFT - 3)
580 #define SBP_L1_COUNT		(1 << SBP_L1_BITS)
581 #define SBP_L2_COUNT		(1 << SBP_L2_BITS)
582 #define SBP_L2_SHIFT		(PAGE_SHIFT + SBP_L1_BITS)
583 #define SBP_L3_SHIFT		(SBP_L2_SHIFT + SBP_L2_BITS)
584 
585 extern void subpage_prot_free(struct mm_struct *mm);
586 extern void subpage_prot_init_new_context(struct mm_struct *mm);
587 #else
588 static inline void subpage_prot_free(struct mm_struct *mm) {}
589 static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
590 #endif /* CONFIG_PPC_SUBPAGE_PROT */
591 
592 #if 0
593 /*
594  * The code below is equivalent to this function for arguments
595  * < 2^VSID_BITS, which is all this should ever be called
596  * with.  However gcc is not clever enough to compute the
597  * modulus (2^n-1) without a second multiply.
598  */
599 #define vsid_scramble(protovsid, size) \
600 	((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
601 
602 #else /* 1 */
603 #define vsid_scramble(protovsid, size) \
604 	({								 \
605 		unsigned long x;					 \
606 		x = (protovsid) * VSID_MULTIPLIER_##size;		 \
607 		x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
608 		(x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
609 	})
610 #endif /* 1 */
611 
612 /* Returns the segment size indicator for a user address */
613 static inline int user_segment_size(unsigned long addr)
614 {
615 	/* Use 1T segments if possible for addresses >= 1T */
616 	if (addr >= (1UL << SID_SHIFT_1T))
617 		return mmu_highuser_ssize;
618 	return MMU_SEGSIZE_256M;
619 }
620 
621 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
622 				     int ssize)
623 {
624 	/*
625 	 * Bad address. We return VSID 0 for that
626 	 */
627 	if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
628 		return 0;
629 
630 	if (ssize == MMU_SEGSIZE_256M)
631 		return vsid_scramble((context << ESID_BITS)
632 				     | (ea >> SID_SHIFT), 256M);
633 	return vsid_scramble((context << ESID_BITS_1T)
634 			     | (ea >> SID_SHIFT_1T), 1T);
635 }
636 
637 /*
638  * This is only valid for addresses >= PAGE_OFFSET
639  *
640  * For kernel space, we use the top 4 context ids to map address as below
641  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
642  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
643  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
644  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
645  */
646 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
647 {
648 	unsigned long context;
649 
650 	/*
651 	 * kernel take the top 4 context from the available range
652 	 */
653 	context = (MAX_USER_CONTEXT) + ((ea >> 60) - 0xc) + 1;
654 	return get_vsid(context, ea, ssize);
655 }
656 
657 unsigned htab_shift_for_mem_size(unsigned long mem_size);
658 
659 #endif /* __ASSEMBLY__ */
660 
661 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */
662