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_256M	SLB_VSID_SHIFT
43 #define SLB_VSID_SHIFT_1T	24
44 #define SLB_VSID_SSIZE_SHIFT	62
45 #define SLB_VSID_B		ASM_CONST(0xc000000000000000)
46 #define SLB_VSID_B_256M		ASM_CONST(0x0000000000000000)
47 #define SLB_VSID_B_1T		ASM_CONST(0x4000000000000000)
48 #define SLB_VSID_KS		ASM_CONST(0x0000000000000800)
49 #define SLB_VSID_KP		ASM_CONST(0x0000000000000400)
50 #define SLB_VSID_N		ASM_CONST(0x0000000000000200) /* no-execute */
51 #define SLB_VSID_L		ASM_CONST(0x0000000000000100)
52 #define SLB_VSID_C		ASM_CONST(0x0000000000000080) /* class */
53 #define SLB_VSID_LP		ASM_CONST(0x0000000000000030)
54 #define SLB_VSID_LP_00		ASM_CONST(0x0000000000000000)
55 #define SLB_VSID_LP_01		ASM_CONST(0x0000000000000010)
56 #define SLB_VSID_LP_10		ASM_CONST(0x0000000000000020)
57 #define SLB_VSID_LP_11		ASM_CONST(0x0000000000000030)
58 #define SLB_VSID_LLP		(SLB_VSID_L|SLB_VSID_LP)
59 
60 #define SLB_VSID_KERNEL		(SLB_VSID_KP)
61 #define SLB_VSID_USER		(SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
62 
63 #define SLBIE_C			(0x08000000)
64 #define SLBIE_SSIZE_SHIFT	25
65 
66 /*
67  * Hash table
68  */
69 
70 #define HPTES_PER_GROUP 8
71 
72 #define HPTE_V_SSIZE_SHIFT	62
73 #define HPTE_V_AVPN_SHIFT	7
74 #define HPTE_V_COMMON_BITS	ASM_CONST(0x000fffffffffffff)
75 #define HPTE_V_AVPN		ASM_CONST(0x3fffffffffffff80)
76 #define HPTE_V_AVPN_3_0		ASM_CONST(0x000fffffffffff80)
77 #define HPTE_V_AVPN_VAL(x)	(((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
78 #define HPTE_V_COMPARE(x,y)	(!(((x) ^ (y)) & 0xffffffffffffff80UL))
79 #define HPTE_V_BOLTED		ASM_CONST(0x0000000000000010)
80 #define HPTE_V_LOCK		ASM_CONST(0x0000000000000008)
81 #define HPTE_V_LARGE		ASM_CONST(0x0000000000000004)
82 #define HPTE_V_SECONDARY	ASM_CONST(0x0000000000000002)
83 #define HPTE_V_VALID		ASM_CONST(0x0000000000000001)
84 
85 /*
86  * ISA 3.0 has a different HPTE format.
87  */
88 #define HPTE_R_3_0_SSIZE_SHIFT	58
89 #define HPTE_R_3_0_SSIZE_MASK	(3ull << HPTE_R_3_0_SSIZE_SHIFT)
90 #define HPTE_R_PP0		ASM_CONST(0x8000000000000000)
91 #define HPTE_R_TS		ASM_CONST(0x4000000000000000)
92 #define HPTE_R_KEY_HI		ASM_CONST(0x3000000000000000)
93 #define HPTE_R_RPN_SHIFT	12
94 #define HPTE_R_RPN		ASM_CONST(0x0ffffffffffff000)
95 #define HPTE_R_RPN_3_0		ASM_CONST(0x01fffffffffff000)
96 #define HPTE_R_PP		ASM_CONST(0x0000000000000003)
97 #define HPTE_R_PPP		ASM_CONST(0x8000000000000003)
98 #define HPTE_R_N		ASM_CONST(0x0000000000000004)
99 #define HPTE_R_G		ASM_CONST(0x0000000000000008)
100 #define HPTE_R_M		ASM_CONST(0x0000000000000010)
101 #define HPTE_R_I		ASM_CONST(0x0000000000000020)
102 #define HPTE_R_W		ASM_CONST(0x0000000000000040)
103 #define HPTE_R_WIMG		ASM_CONST(0x0000000000000078)
104 #define HPTE_R_C		ASM_CONST(0x0000000000000080)
105 #define HPTE_R_R		ASM_CONST(0x0000000000000100)
106 #define HPTE_R_KEY_LO		ASM_CONST(0x0000000000000e00)
107 
108 #define HPTE_V_1TB_SEG		ASM_CONST(0x4000000000000000)
109 #define HPTE_V_VRMA_MASK	ASM_CONST(0x4001ffffff000000)
110 
111 /* Values for PP (assumes Ks=0, Kp=1) */
112 #define PP_RWXX	0	/* Supervisor read/write, User none */
113 #define PP_RWRX 1	/* Supervisor read/write, User read */
114 #define PP_RWRW 2	/* Supervisor read/write, User read/write */
115 #define PP_RXRX 3	/* Supervisor read,       User read */
116 #define PP_RXXX	(HPTE_R_PP0 | 2)	/* Supervisor read, user none */
117 
118 /* Fields for tlbiel instruction in architecture 2.06 */
119 #define TLBIEL_INVAL_SEL_MASK	0xc00	/* invalidation selector */
120 #define  TLBIEL_INVAL_PAGE	0x000	/* invalidate a single page */
121 #define  TLBIEL_INVAL_SET_LPID	0x800	/* invalidate a set for current LPID */
122 #define  TLBIEL_INVAL_SET	0xc00	/* invalidate a set for all LPIDs */
123 #define TLBIEL_INVAL_SET_MASK	0xfff000	/* set number to inval. */
124 #define TLBIEL_INVAL_SET_SHIFT	12
125 
126 #define POWER7_TLB_SETS		128	/* # sets in POWER7 TLB */
127 #define POWER8_TLB_SETS		512	/* # sets in POWER8 TLB */
128 #define POWER9_TLB_SETS_HASH	256	/* # sets in POWER9 TLB Hash mode */
129 #define POWER9_TLB_SETS_RADIX	128	/* # sets in POWER9 TLB Radix mode */
130 
131 #ifndef __ASSEMBLY__
132 
133 struct mmu_hash_ops {
134 	void            (*hpte_invalidate)(unsigned long slot,
135 					   unsigned long vpn,
136 					   int bpsize, int apsize,
137 					   int ssize, int local);
138 	long		(*hpte_updatepp)(unsigned long slot,
139 					 unsigned long newpp,
140 					 unsigned long vpn,
141 					 int bpsize, int apsize,
142 					 int ssize, unsigned long flags);
143 	void            (*hpte_updateboltedpp)(unsigned long newpp,
144 					       unsigned long ea,
145 					       int psize, int ssize);
146 	long		(*hpte_insert)(unsigned long hpte_group,
147 				       unsigned long vpn,
148 				       unsigned long prpn,
149 				       unsigned long rflags,
150 				       unsigned long vflags,
151 				       int psize, int apsize,
152 				       int ssize);
153 	long		(*hpte_remove)(unsigned long hpte_group);
154 	int             (*hpte_removebolted)(unsigned long ea,
155 					     int psize, int ssize);
156 	void		(*flush_hash_range)(unsigned long number, int local);
157 	void		(*hugepage_invalidate)(unsigned long vsid,
158 					       unsigned long addr,
159 					       unsigned char *hpte_slot_array,
160 					       int psize, int ssize, int local);
161 	int		(*resize_hpt)(unsigned long shift);
162 	/*
163 	 * Special for kexec.
164 	 * To be called in real mode with interrupts disabled. No locks are
165 	 * taken as such, concurrent access on pre POWER5 hardware could result
166 	 * in a deadlock.
167 	 * The linear mapping is destroyed as well.
168 	 */
169 	void		(*hpte_clear_all)(void);
170 };
171 extern struct mmu_hash_ops mmu_hash_ops;
172 
173 struct hash_pte {
174 	__be64 v;
175 	__be64 r;
176 };
177 
178 extern struct hash_pte *htab_address;
179 extern unsigned long htab_size_bytes;
180 extern unsigned long htab_hash_mask;
181 
182 
183 static inline int shift_to_mmu_psize(unsigned int shift)
184 {
185 	int psize;
186 
187 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
188 		if (mmu_psize_defs[psize].shift == shift)
189 			return psize;
190 	return -1;
191 }
192 
193 static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
194 {
195 	if (mmu_psize_defs[mmu_psize].shift)
196 		return mmu_psize_defs[mmu_psize].shift;
197 	BUG();
198 }
199 
200 static inline unsigned long get_sllp_encoding(int psize)
201 {
202 	unsigned long sllp;
203 
204 	sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
205 		((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
206 	return sllp;
207 }
208 
209 #endif /* __ASSEMBLY__ */
210 
211 /*
212  * Segment sizes.
213  * These are the values used by hardware in the B field of
214  * SLB entries and the first dword of MMU hashtable entries.
215  * The B field is 2 bits; the values 2 and 3 are unused and reserved.
216  */
217 #define MMU_SEGSIZE_256M	0
218 #define MMU_SEGSIZE_1T		1
219 
220 /*
221  * encode page number shift.
222  * in order to fit the 78 bit va in a 64 bit variable we shift the va by
223  * 12 bits. This enable us to address upto 76 bit va.
224  * For hpt hash from a va we can ignore the page size bits of va and for
225  * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
226  * we work in all cases including 4k page size.
227  */
228 #define VPN_SHIFT	12
229 
230 /*
231  * HPTE Large Page (LP) details
232  */
233 #define LP_SHIFT	12
234 #define LP_BITS		8
235 #define LP_MASK(i)	((0xFF >> (i)) << LP_SHIFT)
236 
237 #ifndef __ASSEMBLY__
238 
239 static inline int slb_vsid_shift(int ssize)
240 {
241 	if (ssize == MMU_SEGSIZE_256M)
242 		return SLB_VSID_SHIFT;
243 	return SLB_VSID_SHIFT_1T;
244 }
245 
246 static inline int segment_shift(int ssize)
247 {
248 	if (ssize == MMU_SEGSIZE_256M)
249 		return SID_SHIFT;
250 	return SID_SHIFT_1T;
251 }
252 
253 /*
254  * This array is indexed by the LP field of the HPTE second dword.
255  * Since this field may contain some RPN bits, some entries are
256  * replicated so that we get the same value irrespective of RPN.
257  * The top 4 bits are the page size index (MMU_PAGE_*) for the
258  * actual page size, the bottom 4 bits are the base page size.
259  */
260 extern u8 hpte_page_sizes[1 << LP_BITS];
261 
262 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
263 					     bool is_base_size)
264 {
265 	unsigned int i, lp;
266 
267 	if (!(h & HPTE_V_LARGE))
268 		return 1ul << 12;
269 
270 	/* Look at the 8 bit LP value */
271 	lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
272 	i = hpte_page_sizes[lp];
273 	if (!i)
274 		return 0;
275 	if (!is_base_size)
276 		i >>= 4;
277 	return 1ul << mmu_psize_defs[i & 0xf].shift;
278 }
279 
280 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
281 {
282 	return __hpte_page_size(h, l, 0);
283 }
284 
285 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
286 {
287 	return __hpte_page_size(h, l, 1);
288 }
289 
290 /*
291  * The current system page and segment sizes
292  */
293 extern int mmu_kernel_ssize;
294 extern int mmu_highuser_ssize;
295 extern u16 mmu_slb_size;
296 extern unsigned long tce_alloc_start, tce_alloc_end;
297 
298 /*
299  * If the processor supports 64k normal pages but not 64k cache
300  * inhibited pages, we have to be prepared to switch processes
301  * to use 4k pages when they create cache-inhibited mappings.
302  * If this is the case, mmu_ci_restrictions will be set to 1.
303  */
304 extern int mmu_ci_restrictions;
305 
306 /*
307  * This computes the AVPN and B fields of the first dword of a HPTE,
308  * for use when we want to match an existing PTE.  The bottom 7 bits
309  * of the returned value are zero.
310  */
311 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
312 					     int ssize)
313 {
314 	unsigned long v;
315 	/*
316 	 * The AVA field omits the low-order 23 bits of the 78 bits VA.
317 	 * These bits are not needed in the PTE, because the
318 	 * low-order b of these bits are part of the byte offset
319 	 * into the virtual page and, if b < 23, the high-order
320 	 * 23-b of these bits are always used in selecting the
321 	 * PTEGs to be searched
322 	 */
323 	v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
324 	v <<= HPTE_V_AVPN_SHIFT;
325 	v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
326 	return v;
327 }
328 
329 /*
330  * ISA v3.0 defines a new HPTE format, which differs from the old
331  * format in having smaller AVPN and ARPN fields, and the B field
332  * in the second dword instead of the first.
333  */
334 static inline unsigned long hpte_old_to_new_v(unsigned long v)
335 {
336 	/* trim AVPN, drop B */
337 	return v & HPTE_V_COMMON_BITS;
338 }
339 
340 static inline unsigned long hpte_old_to_new_r(unsigned long v, unsigned long r)
341 {
342 	/* move B field from 1st to 2nd dword, trim ARPN */
343 	return (r & ~HPTE_R_3_0_SSIZE_MASK) |
344 		(((v) >> HPTE_V_SSIZE_SHIFT) << HPTE_R_3_0_SSIZE_SHIFT);
345 }
346 
347 static inline unsigned long hpte_new_to_old_v(unsigned long v, unsigned long r)
348 {
349 	/* insert B field */
350 	return (v & HPTE_V_COMMON_BITS) |
351 		((r & HPTE_R_3_0_SSIZE_MASK) <<
352 		 (HPTE_V_SSIZE_SHIFT - HPTE_R_3_0_SSIZE_SHIFT));
353 }
354 
355 static inline unsigned long hpte_new_to_old_r(unsigned long r)
356 {
357 	/* clear out B field */
358 	return r & ~HPTE_R_3_0_SSIZE_MASK;
359 }
360 
361 /*
362  * This function sets the AVPN and L fields of the HPTE  appropriately
363  * using the base page size and actual page size.
364  */
365 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
366 					  int actual_psize, int ssize)
367 {
368 	unsigned long v;
369 	v = hpte_encode_avpn(vpn, base_psize, ssize);
370 	if (actual_psize != MMU_PAGE_4K)
371 		v |= HPTE_V_LARGE;
372 	return v;
373 }
374 
375 /*
376  * This function sets the ARPN, and LP fields of the HPTE appropriately
377  * for the page size. We assume the pa is already "clean" that is properly
378  * aligned for the requested page size
379  */
380 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
381 					  int actual_psize)
382 {
383 	/* A 4K page needs no special encoding */
384 	if (actual_psize == MMU_PAGE_4K)
385 		return pa & HPTE_R_RPN;
386 	else {
387 		unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
388 		unsigned int shift = mmu_psize_defs[actual_psize].shift;
389 		return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
390 	}
391 }
392 
393 /*
394  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
395  */
396 static inline unsigned long hpt_vpn(unsigned long ea,
397 				    unsigned long vsid, int ssize)
398 {
399 	unsigned long mask;
400 	int s_shift = segment_shift(ssize);
401 
402 	mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
403 	return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
404 }
405 
406 /*
407  * This hashes a virtual address
408  */
409 static inline unsigned long hpt_hash(unsigned long vpn,
410 				     unsigned int shift, int ssize)
411 {
412 	unsigned long mask;
413 	unsigned long hash, vsid;
414 
415 	/* VPN_SHIFT can be atmost 12 */
416 	if (ssize == MMU_SEGSIZE_256M) {
417 		mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
418 		hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
419 			((vpn & mask) >> (shift - VPN_SHIFT));
420 	} else {
421 		mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
422 		vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
423 		hash = vsid ^ (vsid << 25) ^
424 			((vpn & mask) >> (shift - VPN_SHIFT)) ;
425 	}
426 	return hash & 0x7fffffffffUL;
427 }
428 
429 #define HPTE_LOCAL_UPDATE	0x1
430 #define HPTE_NOHPTE_UPDATE	0x2
431 
432 extern int __hash_page_4K(unsigned long ea, unsigned long access,
433 			  unsigned long vsid, pte_t *ptep, unsigned long trap,
434 			  unsigned long flags, int ssize, int subpage_prot);
435 extern int __hash_page_64K(unsigned long ea, unsigned long access,
436 			   unsigned long vsid, pte_t *ptep, unsigned long trap,
437 			   unsigned long flags, int ssize);
438 struct mm_struct;
439 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
440 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
441 			unsigned long access, unsigned long trap,
442 			unsigned long flags);
443 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
444 		     unsigned long dsisr);
445 int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
446 		     pte_t *ptep, unsigned long trap, unsigned long flags,
447 		     int ssize, unsigned int shift, unsigned int mmu_psize);
448 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
449 extern int __hash_page_thp(unsigned long ea, unsigned long access,
450 			   unsigned long vsid, pmd_t *pmdp, unsigned long trap,
451 			   unsigned long flags, int ssize, unsigned int psize);
452 #else
453 static inline int __hash_page_thp(unsigned long ea, unsigned long access,
454 				  unsigned long vsid, pmd_t *pmdp,
455 				  unsigned long trap, unsigned long flags,
456 				  int ssize, unsigned int psize)
457 {
458 	BUG();
459 	return -1;
460 }
461 #endif
462 extern void hash_failure_debug(unsigned long ea, unsigned long access,
463 			       unsigned long vsid, unsigned long trap,
464 			       int ssize, int psize, int lpsize,
465 			       unsigned long pte);
466 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
467 			     unsigned long pstart, unsigned long prot,
468 			     int psize, int ssize);
469 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
470 			int psize, int ssize);
471 extern void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
472 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
473 
474 #ifdef CONFIG_PPC_PSERIES
475 void hpte_init_pseries(void);
476 #else
477 static inline void hpte_init_pseries(void) { }
478 #endif
479 
480 extern void hpte_init_native(void);
481 
482 extern void slb_initialize(void);
483 extern void slb_flush_and_rebolt(void);
484 
485 extern void slb_vmalloc_update(void);
486 extern void slb_set_size(u16 size);
487 #endif /* __ASSEMBLY__ */
488 
489 /*
490  * VSID allocation (256MB segment)
491  *
492  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
493  * from mmu context id and effective segment id of the address.
494  *
495  * For user processes max context id is limited to MAX_USER_CONTEXT.
496 
497  * For kernel space, we use context ids 1-4 to map addresses as below:
498  * NOTE: each context only support 64TB now.
499  * 0x00001 -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
500  * 0x00002 -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
501  * 0x00003 -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
502  * 0x00004 -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
503  *
504  * The proto-VSIDs are then scrambled into real VSIDs with the
505  * multiplicative hash:
506  *
507  *	VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
508  *
509  * VSID_MULTIPLIER is prime, so in particular it is
510  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
511  * Because the modulus is 2^n-1 we can compute it efficiently without
512  * a divide or extra multiply (see below). The scramble function gives
513  * robust scattering in the hash table (at least based on some initial
514  * results).
515  *
516  * We use VSID 0 to indicate an invalid VSID. The means we can't use context id
517  * 0, because a context id of 0 and an EA of 0 gives a proto-VSID of 0, which
518  * will produce a VSID of 0.
519  *
520  * We also need to avoid the last segment of the last context, because that
521  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
522  * because of the modulo operation in vsid scramble.
523  */
524 
525 /*
526  * Max Va bits we support as of now is 68 bits. We want 19 bit
527  * context ID.
528  * Restrictions:
529  * GPU has restrictions of not able to access beyond 128TB
530  * (47 bit effective address). We also cannot do more than 20bit PID.
531  * For p4 and p5 which can only do 65 bit VA, we restrict our CONTEXT_BITS
532  * to 16 bits (ie, we can only have 2^16 pids at the same time).
533  */
534 #define VA_BITS			68
535 #define CONTEXT_BITS		19
536 #define ESID_BITS		(VA_BITS - (SID_SHIFT + CONTEXT_BITS))
537 #define ESID_BITS_1T		(VA_BITS - (SID_SHIFT_1T + CONTEXT_BITS))
538 
539 #define ESID_BITS_MASK		((1 << ESID_BITS) - 1)
540 #define ESID_BITS_1T_MASK	((1 << ESID_BITS_1T) - 1)
541 
542 /*
543  * 256MB segment
544  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
545  * available for user + kernel mapping. VSID 0 is reserved as invalid, contexts
546  * 1-4 are used for kernel mapping. Each segment contains 2^28 bytes. Each
547  * context maps 2^49 bytes (512TB).
548  *
549  * We also need to avoid the last segment of the last context, because that
550  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
551  * because of the modulo operation in vsid scramble.
552  */
553 #define MAX_USER_CONTEXT	((ASM_CONST(1) << CONTEXT_BITS) - 2)
554 #define MIN_USER_CONTEXT	(5)
555 
556 /* Would be nice to use KERNEL_REGION_ID here */
557 #define KERNEL_REGION_CONTEXT_OFFSET	(0xc - 1)
558 
559 /*
560  * For platforms that support on 65bit VA we limit the context bits
561  */
562 #define MAX_USER_CONTEXT_65BIT_VA ((ASM_CONST(1) << (65 - (SID_SHIFT + ESID_BITS))) - 2)
563 
564 /*
565  * This should be computed such that protovosid * vsid_mulitplier
566  * doesn't overflow 64 bits. The vsid_mutliplier should also be
567  * co-prime to vsid_modulus. We also need to make sure that number
568  * of bits in multiplied result (dividend) is less than twice the number of
569  * protovsid bits for our modulus optmization to work.
570  *
571  * The below table shows the current values used.
572  * |-------+------------+----------------------+------------+-------------------|
573  * |       | Prime Bits | proto VSID_BITS_65VA | Total Bits | 2* prot VSID_BITS |
574  * |-------+------------+----------------------+------------+-------------------|
575  * | 1T    |         24 |                   25 |         49 |                50 |
576  * |-------+------------+----------------------+------------+-------------------|
577  * | 256MB |         24 |                   37 |         61 |                74 |
578  * |-------+------------+----------------------+------------+-------------------|
579  *
580  * |-------+------------+----------------------+------------+--------------------|
581  * |       | Prime Bits | proto VSID_BITS_68VA | Total Bits | 2* proto VSID_BITS |
582  * |-------+------------+----------------------+------------+--------------------|
583  * | 1T    |         24 |                   28 |         52 |                 56 |
584  * |-------+------------+----------------------+------------+--------------------|
585  * | 256MB |         24 |                   40 |         64 |                 80 |
586  * |-------+------------+----------------------+------------+--------------------|
587  *
588  */
589 #define VSID_MULTIPLIER_256M	ASM_CONST(12538073)	/* 24-bit prime */
590 #define VSID_BITS_256M		(VA_BITS - SID_SHIFT)
591 #define VSID_BITS_65_256M	(65 - SID_SHIFT)
592 /*
593  * Modular multiplicative inverse of VSID_MULTIPLIER under modulo VSID_MODULUS
594  */
595 #define VSID_MULINV_256M	ASM_CONST(665548017062)
596 
597 #define VSID_MULTIPLIER_1T	ASM_CONST(12538073)	/* 24-bit prime */
598 #define VSID_BITS_1T		(VA_BITS - SID_SHIFT_1T)
599 #define VSID_BITS_65_1T		(65 - SID_SHIFT_1T)
600 #define VSID_MULINV_1T		ASM_CONST(209034062)
601 
602 /* 1TB VSID reserved for VRMA */
603 #define VRMA_VSID	0x1ffffffUL
604 #define USER_VSID_RANGE	(1UL << (ESID_BITS + SID_SHIFT))
605 
606 /* 4 bits per slice and we have one slice per 1TB */
607 #define SLICE_ARRAY_SIZE	(H_PGTABLE_RANGE >> 41)
608 #define TASK_SLICE_ARRAY_SZ(x)	((x)->context.addr_limit >> 41)
609 
610 #ifndef __ASSEMBLY__
611 
612 #ifdef CONFIG_PPC_SUBPAGE_PROT
613 /*
614  * For the sub-page protection option, we extend the PGD with one of
615  * these.  Basically we have a 3-level tree, with the top level being
616  * the protptrs array.  To optimize speed and memory consumption when
617  * only addresses < 4GB are being protected, pointers to the first
618  * four pages of sub-page protection words are stored in the low_prot
619  * array.
620  * Each page of sub-page protection words protects 1GB (4 bytes
621  * protects 64k).  For the 3-level tree, each page of pointers then
622  * protects 8TB.
623  */
624 struct subpage_prot_table {
625 	unsigned long maxaddr;	/* only addresses < this are protected */
626 	unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
627 	unsigned int *low_prot[4];
628 };
629 
630 #define SBP_L1_BITS		(PAGE_SHIFT - 2)
631 #define SBP_L2_BITS		(PAGE_SHIFT - 3)
632 #define SBP_L1_COUNT		(1 << SBP_L1_BITS)
633 #define SBP_L2_COUNT		(1 << SBP_L2_BITS)
634 #define SBP_L2_SHIFT		(PAGE_SHIFT + SBP_L1_BITS)
635 #define SBP_L3_SHIFT		(SBP_L2_SHIFT + SBP_L2_BITS)
636 
637 extern void subpage_prot_free(struct mm_struct *mm);
638 extern void subpage_prot_init_new_context(struct mm_struct *mm);
639 #else
640 static inline void subpage_prot_free(struct mm_struct *mm) {}
641 static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
642 #endif /* CONFIG_PPC_SUBPAGE_PROT */
643 
644 #if 0
645 /*
646  * The code below is equivalent to this function for arguments
647  * < 2^VSID_BITS, which is all this should ever be called
648  * with.  However gcc is not clever enough to compute the
649  * modulus (2^n-1) without a second multiply.
650  */
651 #define vsid_scramble(protovsid, size) \
652 	((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
653 
654 /* simplified form avoiding mod operation */
655 #define vsid_scramble(protovsid, size) \
656 	({								 \
657 		unsigned long x;					 \
658 		x = (protovsid) * VSID_MULTIPLIER_##size;		 \
659 		x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
660 		(x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
661 	})
662 
663 #else /* 1 */
664 static inline unsigned long vsid_scramble(unsigned long protovsid,
665 				  unsigned long vsid_multiplier, int vsid_bits)
666 {
667 	unsigned long vsid;
668 	unsigned long vsid_modulus = ((1UL << vsid_bits) - 1);
669 	/*
670 	 * We have same multipler for both 256 and 1T segements now
671 	 */
672 	vsid = protovsid * vsid_multiplier;
673 	vsid = (vsid >> vsid_bits) + (vsid & vsid_modulus);
674 	return (vsid + ((vsid + 1) >> vsid_bits)) & vsid_modulus;
675 }
676 
677 #endif /* 1 */
678 
679 /* Returns the segment size indicator for a user address */
680 static inline int user_segment_size(unsigned long addr)
681 {
682 	/* Use 1T segments if possible for addresses >= 1T */
683 	if (addr >= (1UL << SID_SHIFT_1T))
684 		return mmu_highuser_ssize;
685 	return MMU_SEGSIZE_256M;
686 }
687 
688 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
689 				     int ssize)
690 {
691 	unsigned long va_bits = VA_BITS;
692 	unsigned long vsid_bits;
693 	unsigned long protovsid;
694 
695 	/*
696 	 * Bad address. We return VSID 0 for that
697 	 */
698 	if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
699 		return 0;
700 
701 	if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
702 		va_bits = 65;
703 
704 	if (ssize == MMU_SEGSIZE_256M) {
705 		vsid_bits = va_bits - SID_SHIFT;
706 		protovsid = (context << ESID_BITS) |
707 			((ea >> SID_SHIFT) & ESID_BITS_MASK);
708 		return vsid_scramble(protovsid, VSID_MULTIPLIER_256M, vsid_bits);
709 	}
710 	/* 1T segment */
711 	vsid_bits = va_bits - SID_SHIFT_1T;
712 	protovsid = (context << ESID_BITS_1T) |
713 		((ea >> SID_SHIFT_1T) & ESID_BITS_1T_MASK);
714 	return vsid_scramble(protovsid, VSID_MULTIPLIER_1T, vsid_bits);
715 }
716 
717 /*
718  * This is only valid for addresses >= PAGE_OFFSET
719  */
720 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
721 {
722 	unsigned long context;
723 
724 	if (!is_kernel_addr(ea))
725 		return 0;
726 
727 	/*
728 	 * For kernel space, we use context ids 1-4 to map the address space as
729 	 * below:
730 	 *
731 	 * 0x00001 -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
732 	 * 0x00002 -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
733 	 * 0x00003 -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
734 	 * 0x00004 -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
735 	 *
736 	 * So we can compute the context from the region (top nibble) by
737 	 * subtracting 11, or 0xc - 1.
738 	 */
739 	context = (ea >> 60) - KERNEL_REGION_CONTEXT_OFFSET;
740 
741 	return get_vsid(context, ea, ssize);
742 }
743 
744 unsigned htab_shift_for_mem_size(unsigned long mem_size);
745 
746 #endif /* __ASSEMBLY__ */
747 
748 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */
749