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  * The current system page and segment sizes
249  */
250 extern int mmu_kernel_ssize;
251 extern int mmu_highuser_ssize;
252 extern u16 mmu_slb_size;
253 extern unsigned long tce_alloc_start, tce_alloc_end;
254 
255 /*
256  * If the processor supports 64k normal pages but not 64k cache
257  * inhibited pages, we have to be prepared to switch processes
258  * to use 4k pages when they create cache-inhibited mappings.
259  * If this is the case, mmu_ci_restrictions will be set to 1.
260  */
261 extern int mmu_ci_restrictions;
262 
263 /*
264  * This computes the AVPN and B fields of the first dword of a HPTE,
265  * for use when we want to match an existing PTE.  The bottom 7 bits
266  * of the returned value are zero.
267  */
268 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
269 					     int ssize)
270 {
271 	unsigned long v;
272 	/*
273 	 * The AVA field omits the low-order 23 bits of the 78 bits VA.
274 	 * These bits are not needed in the PTE, because the
275 	 * low-order b of these bits are part of the byte offset
276 	 * into the virtual page and, if b < 23, the high-order
277 	 * 23-b of these bits are always used in selecting the
278 	 * PTEGs to be searched
279 	 */
280 	v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
281 	v <<= HPTE_V_AVPN_SHIFT;
282 	if (!cpu_has_feature(CPU_FTR_ARCH_300))
283 		v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
284 	return v;
285 }
286 
287 /*
288  * This function sets the AVPN and L fields of the HPTE  appropriately
289  * using the base page size and actual page size.
290  */
291 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
292 					  int actual_psize, int ssize)
293 {
294 	unsigned long v;
295 	v = hpte_encode_avpn(vpn, base_psize, ssize);
296 	if (actual_psize != MMU_PAGE_4K)
297 		v |= HPTE_V_LARGE;
298 	return v;
299 }
300 
301 /*
302  * This function sets the ARPN, and LP fields of the HPTE appropriately
303  * for the page size. We assume the pa is already "clean" that is properly
304  * aligned for the requested page size
305  */
306 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
307 					  int actual_psize, int ssize)
308 {
309 
310 	if (cpu_has_feature(CPU_FTR_ARCH_300))
311 		pa |= ((unsigned long) ssize) << HPTE_R_3_0_SSIZE_SHIFT;
312 
313 	/* A 4K page needs no special encoding */
314 	if (actual_psize == MMU_PAGE_4K)
315 		return pa & HPTE_R_RPN;
316 	else {
317 		unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
318 		unsigned int shift = mmu_psize_defs[actual_psize].shift;
319 		return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
320 	}
321 }
322 
323 /*
324  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
325  */
326 static inline unsigned long hpt_vpn(unsigned long ea,
327 				    unsigned long vsid, int ssize)
328 {
329 	unsigned long mask;
330 	int s_shift = segment_shift(ssize);
331 
332 	mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
333 	return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
334 }
335 
336 /*
337  * This hashes a virtual address
338  */
339 static inline unsigned long hpt_hash(unsigned long vpn,
340 				     unsigned int shift, int ssize)
341 {
342 	int mask;
343 	unsigned long hash, vsid;
344 
345 	/* VPN_SHIFT can be atmost 12 */
346 	if (ssize == MMU_SEGSIZE_256M) {
347 		mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
348 		hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
349 			((vpn & mask) >> (shift - VPN_SHIFT));
350 	} else {
351 		mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
352 		vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
353 		hash = vsid ^ (vsid << 25) ^
354 			((vpn & mask) >> (shift - VPN_SHIFT)) ;
355 	}
356 	return hash & 0x7fffffffffUL;
357 }
358 
359 #define HPTE_LOCAL_UPDATE	0x1
360 #define HPTE_NOHPTE_UPDATE	0x2
361 
362 extern int __hash_page_4K(unsigned long ea, unsigned long access,
363 			  unsigned long vsid, pte_t *ptep, unsigned long trap,
364 			  unsigned long flags, int ssize, int subpage_prot);
365 extern int __hash_page_64K(unsigned long ea, unsigned long access,
366 			   unsigned long vsid, pte_t *ptep, unsigned long trap,
367 			   unsigned long flags, int ssize);
368 struct mm_struct;
369 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
370 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
371 			unsigned long access, unsigned long trap,
372 			unsigned long flags);
373 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
374 		     unsigned long dsisr);
375 int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
376 		     pte_t *ptep, unsigned long trap, unsigned long flags,
377 		     int ssize, unsigned int shift, unsigned int mmu_psize);
378 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
379 extern int __hash_page_thp(unsigned long ea, unsigned long access,
380 			   unsigned long vsid, pmd_t *pmdp, unsigned long trap,
381 			   unsigned long flags, int ssize, unsigned int psize);
382 #else
383 static inline int __hash_page_thp(unsigned long ea, unsigned long access,
384 				  unsigned long vsid, pmd_t *pmdp,
385 				  unsigned long trap, unsigned long flags,
386 				  int ssize, unsigned int psize)
387 {
388 	BUG();
389 	return -1;
390 }
391 #endif
392 extern void hash_failure_debug(unsigned long ea, unsigned long access,
393 			       unsigned long vsid, unsigned long trap,
394 			       int ssize, int psize, int lpsize,
395 			       unsigned long pte);
396 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
397 			     unsigned long pstart, unsigned long prot,
398 			     int psize, int ssize);
399 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
400 			int psize, int ssize);
401 extern void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
402 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
403 
404 #ifdef CONFIG_PPC_PSERIES
405 void hpte_init_pseries(void);
406 #else
407 static inline void hpte_init_pseries(void) { }
408 #endif
409 
410 extern void hpte_init_native(void);
411 
412 extern void slb_initialize(void);
413 extern void slb_flush_and_rebolt(void);
414 
415 extern void slb_vmalloc_update(void);
416 extern void slb_set_size(u16 size);
417 #endif /* __ASSEMBLY__ */
418 
419 /*
420  * VSID allocation (256MB segment)
421  *
422  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
423  * from mmu context id and effective segment id of the address.
424  *
425  * For user processes max context id is limited to ((1ul << 19) - 5)
426  * for kernel space, we use the top 4 context ids to map address as below
427  * NOTE: each context only support 64TB now.
428  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
429  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
430  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
431  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
432  *
433  * The proto-VSIDs are then scrambled into real VSIDs with the
434  * multiplicative hash:
435  *
436  *	VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
437  *
438  * VSID_MULTIPLIER is prime, so in particular it is
439  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
440  * Because the modulus is 2^n-1 we can compute it efficiently without
441  * a divide or extra multiply (see below). The scramble function gives
442  * robust scattering in the hash table (at least based on some initial
443  * results).
444  *
445  * We also consider VSID 0 special. We use VSID 0 for slb entries mapping
446  * bad address. This enables us to consolidate bad address handling in
447  * hash_page.
448  *
449  * We also need to avoid the last segment of the last context, because that
450  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
451  * because of the modulo operation in vsid scramble. But the vmemmap
452  * (which is what uses region 0xf) will never be close to 64TB in size
453  * (it's 56 bytes per page of system memory).
454  */
455 
456 #define CONTEXT_BITS		19
457 #define ESID_BITS		18
458 #define ESID_BITS_1T		6
459 
460 /*
461  * 256MB segment
462  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
463  * available for user + kernel mapping. The top 4 contexts are used for
464  * kernel mapping. Each segment contains 2^28 bytes. Each
465  * context maps 2^46 bytes (64TB) so we can support 2^19-1 contexts
466  * (19 == 37 + 28 - 46).
467  */
468 #define MAX_USER_CONTEXT	((ASM_CONST(1) << CONTEXT_BITS) - 5)
469 
470 /*
471  * This should be computed such that protovosid * vsid_mulitplier
472  * doesn't overflow 64 bits. It should also be co-prime to vsid_modulus
473  */
474 #define VSID_MULTIPLIER_256M	ASM_CONST(12538073)	/* 24-bit prime */
475 #define VSID_BITS_256M		(CONTEXT_BITS + ESID_BITS)
476 #define VSID_MODULUS_256M	((1UL<<VSID_BITS_256M)-1)
477 
478 #define VSID_MULTIPLIER_1T	ASM_CONST(12538073)	/* 24-bit prime */
479 #define VSID_BITS_1T		(CONTEXT_BITS + ESID_BITS_1T)
480 #define VSID_MODULUS_1T		((1UL<<VSID_BITS_1T)-1)
481 
482 
483 #define USER_VSID_RANGE	(1UL << (ESID_BITS + SID_SHIFT))
484 
485 /*
486  * This macro generates asm code to compute the VSID scramble
487  * function.  Used in slb_allocate() and do_stab_bolted.  The function
488  * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
489  *
490  *	rt = register containing the proto-VSID and into which the
491  *		VSID will be stored
492  *	rx = scratch register (clobbered)
493  *
494  * 	- rt and rx must be different registers
495  * 	- The answer will end up in the low VSID_BITS bits of rt.  The higher
496  * 	  bits may contain other garbage, so you may need to mask the
497  * 	  result.
498  */
499 #define ASM_VSID_SCRAMBLE(rt, rx, size)					\
500 	lis	rx,VSID_MULTIPLIER_##size@h;				\
501 	ori	rx,rx,VSID_MULTIPLIER_##size@l;				\
502 	mulld	rt,rt,rx;		/* rt = rt * MULTIPLIER */	\
503 									\
504 	srdi	rx,rt,VSID_BITS_##size;					\
505 	clrldi	rt,rt,(64-VSID_BITS_##size);				\
506 	add	rt,rt,rx;		/* add high and low bits */	\
507 	/* NOTE: explanation based on VSID_BITS_##size = 36		\
508 	 * Now, r3 == VSID (mod 2^36-1), and lies between 0 and		\
509 	 * 2^36-1+2^28-1.  That in particular means that if r3 >=	\
510 	 * 2^36-1, then r3+1 has the 2^36 bit set.  So, if r3+1 has	\
511 	 * the bit clear, r3 already has the answer we want, if it	\
512 	 * doesn't, the answer is the low 36 bits of r3+1.  So in all	\
513 	 * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
514 	addi	rx,rt,1;						\
515 	srdi	rx,rx,VSID_BITS_##size;	/* extract 2^VSID_BITS bit */	\
516 	add	rt,rt,rx
517 
518 /* 4 bits per slice and we have one slice per 1TB */
519 #define SLICE_ARRAY_SIZE  (H_PGTABLE_RANGE >> 41)
520 
521 #ifndef __ASSEMBLY__
522 
523 #ifdef CONFIG_PPC_SUBPAGE_PROT
524 /*
525  * For the sub-page protection option, we extend the PGD with one of
526  * these.  Basically we have a 3-level tree, with the top level being
527  * the protptrs array.  To optimize speed and memory consumption when
528  * only addresses < 4GB are being protected, pointers to the first
529  * four pages of sub-page protection words are stored in the low_prot
530  * array.
531  * Each page of sub-page protection words protects 1GB (4 bytes
532  * protects 64k).  For the 3-level tree, each page of pointers then
533  * protects 8TB.
534  */
535 struct subpage_prot_table {
536 	unsigned long maxaddr;	/* only addresses < this are protected */
537 	unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
538 	unsigned int *low_prot[4];
539 };
540 
541 #define SBP_L1_BITS		(PAGE_SHIFT - 2)
542 #define SBP_L2_BITS		(PAGE_SHIFT - 3)
543 #define SBP_L1_COUNT		(1 << SBP_L1_BITS)
544 #define SBP_L2_COUNT		(1 << SBP_L2_BITS)
545 #define SBP_L2_SHIFT		(PAGE_SHIFT + SBP_L1_BITS)
546 #define SBP_L3_SHIFT		(SBP_L2_SHIFT + SBP_L2_BITS)
547 
548 extern void subpage_prot_free(struct mm_struct *mm);
549 extern void subpage_prot_init_new_context(struct mm_struct *mm);
550 #else
551 static inline void subpage_prot_free(struct mm_struct *mm) {}
552 static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
553 #endif /* CONFIG_PPC_SUBPAGE_PROT */
554 
555 #if 0
556 /*
557  * The code below is equivalent to this function for arguments
558  * < 2^VSID_BITS, which is all this should ever be called
559  * with.  However gcc is not clever enough to compute the
560  * modulus (2^n-1) without a second multiply.
561  */
562 #define vsid_scramble(protovsid, size) \
563 	((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
564 
565 #else /* 1 */
566 #define vsid_scramble(protovsid, size) \
567 	({								 \
568 		unsigned long x;					 \
569 		x = (protovsid) * VSID_MULTIPLIER_##size;		 \
570 		x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
571 		(x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
572 	})
573 #endif /* 1 */
574 
575 /* Returns the segment size indicator for a user address */
576 static inline int user_segment_size(unsigned long addr)
577 {
578 	/* Use 1T segments if possible for addresses >= 1T */
579 	if (addr >= (1UL << SID_SHIFT_1T))
580 		return mmu_highuser_ssize;
581 	return MMU_SEGSIZE_256M;
582 }
583 
584 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
585 				     int ssize)
586 {
587 	/*
588 	 * Bad address. We return VSID 0 for that
589 	 */
590 	if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
591 		return 0;
592 
593 	if (ssize == MMU_SEGSIZE_256M)
594 		return vsid_scramble((context << ESID_BITS)
595 				     | (ea >> SID_SHIFT), 256M);
596 	return vsid_scramble((context << ESID_BITS_1T)
597 			     | (ea >> SID_SHIFT_1T), 1T);
598 }
599 
600 /*
601  * This is only valid for addresses >= PAGE_OFFSET
602  *
603  * For kernel space, we use the top 4 context ids to map address as below
604  * 0x7fffc -  [ 0xc000000000000000 - 0xc0003fffffffffff ]
605  * 0x7fffd -  [ 0xd000000000000000 - 0xd0003fffffffffff ]
606  * 0x7fffe -  [ 0xe000000000000000 - 0xe0003fffffffffff ]
607  * 0x7ffff -  [ 0xf000000000000000 - 0xf0003fffffffffff ]
608  */
609 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
610 {
611 	unsigned long context;
612 
613 	/*
614 	 * kernel take the top 4 context from the available range
615 	 */
616 	context = (MAX_USER_CONTEXT) + ((ea >> 60) - 0xc) + 1;
617 	return get_vsid(context, ea, ssize);
618 }
619 
620 unsigned htab_shift_for_mem_size(unsigned long mem_size);
621 
622 #endif /* __ASSEMBLY__ */
623 
624 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */
625