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/page.h>
16 #include <asm/bug.h>
17 #include <asm/asm-const.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/task_size_64.h>
27 #include <asm/cpu_has_feature.h>
28 
29 /*
30  * SLB
31  */
32 
33 #define SLB_NUM_BOLTED		2
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_KEY_BIT0		ASM_CONST(0x2000000000000000)
94 #define HPTE_R_KEY_BIT1		ASM_CONST(0x1000000000000000)
95 #define HPTE_R_RPN_SHIFT	12
96 #define HPTE_R_RPN		ASM_CONST(0x0ffffffffffff000)
97 #define HPTE_R_RPN_3_0		ASM_CONST(0x01fffffffffff000)
98 #define HPTE_R_PP		ASM_CONST(0x0000000000000003)
99 #define HPTE_R_PPP		ASM_CONST(0x8000000000000003)
100 #define HPTE_R_N		ASM_CONST(0x0000000000000004)
101 #define HPTE_R_G		ASM_CONST(0x0000000000000008)
102 #define HPTE_R_M		ASM_CONST(0x0000000000000010)
103 #define HPTE_R_I		ASM_CONST(0x0000000000000020)
104 #define HPTE_R_W		ASM_CONST(0x0000000000000040)
105 #define HPTE_R_WIMG		ASM_CONST(0x0000000000000078)
106 #define HPTE_R_C		ASM_CONST(0x0000000000000080)
107 #define HPTE_R_R		ASM_CONST(0x0000000000000100)
108 #define HPTE_R_KEY_LO		ASM_CONST(0x0000000000000e00)
109 #define HPTE_R_KEY_BIT2		ASM_CONST(0x0000000000000800)
110 #define HPTE_R_KEY_BIT3		ASM_CONST(0x0000000000000400)
111 #define HPTE_R_KEY_BIT4		ASM_CONST(0x0000000000000200)
112 #define HPTE_R_KEY		(HPTE_R_KEY_LO | HPTE_R_KEY_HI)
113 
114 #define HPTE_V_1TB_SEG		ASM_CONST(0x4000000000000000)
115 #define HPTE_V_VRMA_MASK	ASM_CONST(0x4001ffffff000000)
116 
117 /* Values for PP (assumes Ks=0, Kp=1) */
118 #define PP_RWXX	0	/* Supervisor read/write, User none */
119 #define PP_RWRX 1	/* Supervisor read/write, User read */
120 #define PP_RWRW 2	/* Supervisor read/write, User read/write */
121 #define PP_RXRX 3	/* Supervisor read,       User read */
122 #define PP_RXXX	(HPTE_R_PP0 | 2)	/* Supervisor read, user none */
123 
124 /* Fields for tlbiel instruction in architecture 2.06 */
125 #define TLBIEL_INVAL_SEL_MASK	0xc00	/* invalidation selector */
126 #define  TLBIEL_INVAL_PAGE	0x000	/* invalidate a single page */
127 #define  TLBIEL_INVAL_SET_LPID	0x800	/* invalidate a set for current LPID */
128 #define  TLBIEL_INVAL_SET	0xc00	/* invalidate a set for all LPIDs */
129 #define TLBIEL_INVAL_SET_MASK	0xfff000	/* set number to inval. */
130 #define TLBIEL_INVAL_SET_SHIFT	12
131 
132 #define POWER7_TLB_SETS		128	/* # sets in POWER7 TLB */
133 #define POWER8_TLB_SETS		512	/* # sets in POWER8 TLB */
134 #define POWER9_TLB_SETS_HASH	256	/* # sets in POWER9 TLB Hash mode */
135 #define POWER9_TLB_SETS_RADIX	128	/* # sets in POWER9 TLB Radix mode */
136 
137 #ifndef __ASSEMBLY__
138 
139 struct mmu_hash_ops {
140 	void            (*hpte_invalidate)(unsigned long slot,
141 					   unsigned long vpn,
142 					   int bpsize, int apsize,
143 					   int ssize, int local);
144 	long		(*hpte_updatepp)(unsigned long slot,
145 					 unsigned long newpp,
146 					 unsigned long vpn,
147 					 int bpsize, int apsize,
148 					 int ssize, unsigned long flags);
149 	void            (*hpte_updateboltedpp)(unsigned long newpp,
150 					       unsigned long ea,
151 					       int psize, int ssize);
152 	long		(*hpte_insert)(unsigned long hpte_group,
153 				       unsigned long vpn,
154 				       unsigned long prpn,
155 				       unsigned long rflags,
156 				       unsigned long vflags,
157 				       int psize, int apsize,
158 				       int ssize);
159 	long		(*hpte_remove)(unsigned long hpte_group);
160 	int             (*hpte_removebolted)(unsigned long ea,
161 					     int psize, int ssize);
162 	void		(*flush_hash_range)(unsigned long number, int local);
163 	void		(*hugepage_invalidate)(unsigned long vsid,
164 					       unsigned long addr,
165 					       unsigned char *hpte_slot_array,
166 					       int psize, int ssize, int local);
167 	int		(*resize_hpt)(unsigned long shift);
168 	/*
169 	 * Special for kexec.
170 	 * To be called in real mode with interrupts disabled. No locks are
171 	 * taken as such, concurrent access on pre POWER5 hardware could result
172 	 * in a deadlock.
173 	 * The linear mapping is destroyed as well.
174 	 */
175 	void		(*hpte_clear_all)(void);
176 };
177 extern struct mmu_hash_ops mmu_hash_ops;
178 
179 struct hash_pte {
180 	__be64 v;
181 	__be64 r;
182 };
183 
184 extern struct hash_pte *htab_address;
185 extern unsigned long htab_size_bytes;
186 extern unsigned long htab_hash_mask;
187 
188 
189 static inline int shift_to_mmu_psize(unsigned int shift)
190 {
191 	int psize;
192 
193 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
194 		if (mmu_psize_defs[psize].shift == shift)
195 			return psize;
196 	return -1;
197 }
198 
199 static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
200 {
201 	if (mmu_psize_defs[mmu_psize].shift)
202 		return mmu_psize_defs[mmu_psize].shift;
203 	BUG();
204 }
205 
206 static inline unsigned int ap_to_shift(unsigned long ap)
207 {
208 	int psize;
209 
210 	for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
211 		if (mmu_psize_defs[psize].ap == ap)
212 			return mmu_psize_defs[psize].shift;
213 	}
214 
215 	return -1;
216 }
217 
218 static inline unsigned long get_sllp_encoding(int psize)
219 {
220 	unsigned long sllp;
221 
222 	sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
223 		((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
224 	return sllp;
225 }
226 
227 #endif /* __ASSEMBLY__ */
228 
229 /*
230  * Segment sizes.
231  * These are the values used by hardware in the B field of
232  * SLB entries and the first dword of MMU hashtable entries.
233  * The B field is 2 bits; the values 2 and 3 are unused and reserved.
234  */
235 #define MMU_SEGSIZE_256M	0
236 #define MMU_SEGSIZE_1T		1
237 
238 /*
239  * encode page number shift.
240  * in order to fit the 78 bit va in a 64 bit variable we shift the va by
241  * 12 bits. This enable us to address upto 76 bit va.
242  * For hpt hash from a va we can ignore the page size bits of va and for
243  * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
244  * we work in all cases including 4k page size.
245  */
246 #define VPN_SHIFT	12
247 
248 /*
249  * HPTE Large Page (LP) details
250  */
251 #define LP_SHIFT	12
252 #define LP_BITS		8
253 #define LP_MASK(i)	((0xFF >> (i)) << LP_SHIFT)
254 
255 #ifndef __ASSEMBLY__
256 
257 static inline int slb_vsid_shift(int ssize)
258 {
259 	if (ssize == MMU_SEGSIZE_256M)
260 		return SLB_VSID_SHIFT;
261 	return SLB_VSID_SHIFT_1T;
262 }
263 
264 static inline int segment_shift(int ssize)
265 {
266 	if (ssize == MMU_SEGSIZE_256M)
267 		return SID_SHIFT;
268 	return SID_SHIFT_1T;
269 }
270 
271 /*
272  * This array is indexed by the LP field of the HPTE second dword.
273  * Since this field may contain some RPN bits, some entries are
274  * replicated so that we get the same value irrespective of RPN.
275  * The top 4 bits are the page size index (MMU_PAGE_*) for the
276  * actual page size, the bottom 4 bits are the base page size.
277  */
278 extern u8 hpte_page_sizes[1 << LP_BITS];
279 
280 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
281 					     bool is_base_size)
282 {
283 	unsigned int i, lp;
284 
285 	if (!(h & HPTE_V_LARGE))
286 		return 1ul << 12;
287 
288 	/* Look at the 8 bit LP value */
289 	lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
290 	i = hpte_page_sizes[lp];
291 	if (!i)
292 		return 0;
293 	if (!is_base_size)
294 		i >>= 4;
295 	return 1ul << mmu_psize_defs[i & 0xf].shift;
296 }
297 
298 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
299 {
300 	return __hpte_page_size(h, l, 0);
301 }
302 
303 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
304 {
305 	return __hpte_page_size(h, l, 1);
306 }
307 
308 /*
309  * The current system page and segment sizes
310  */
311 extern int mmu_kernel_ssize;
312 extern int mmu_highuser_ssize;
313 extern u16 mmu_slb_size;
314 extern unsigned long tce_alloc_start, tce_alloc_end;
315 
316 /*
317  * If the processor supports 64k normal pages but not 64k cache
318  * inhibited pages, we have to be prepared to switch processes
319  * to use 4k pages when they create cache-inhibited mappings.
320  * If this is the case, mmu_ci_restrictions will be set to 1.
321  */
322 extern int mmu_ci_restrictions;
323 
324 /*
325  * This computes the AVPN and B fields of the first dword of a HPTE,
326  * for use when we want to match an existing PTE.  The bottom 7 bits
327  * of the returned value are zero.
328  */
329 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
330 					     int ssize)
331 {
332 	unsigned long v;
333 	/*
334 	 * The AVA field omits the low-order 23 bits of the 78 bits VA.
335 	 * These bits are not needed in the PTE, because the
336 	 * low-order b of these bits are part of the byte offset
337 	 * into the virtual page and, if b < 23, the high-order
338 	 * 23-b of these bits are always used in selecting the
339 	 * PTEGs to be searched
340 	 */
341 	v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
342 	v <<= HPTE_V_AVPN_SHIFT;
343 	v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
344 	return v;
345 }
346 
347 /*
348  * ISA v3.0 defines a new HPTE format, which differs from the old
349  * format in having smaller AVPN and ARPN fields, and the B field
350  * in the second dword instead of the first.
351  */
352 static inline unsigned long hpte_old_to_new_v(unsigned long v)
353 {
354 	/* trim AVPN, drop B */
355 	return v & HPTE_V_COMMON_BITS;
356 }
357 
358 static inline unsigned long hpte_old_to_new_r(unsigned long v, unsigned long r)
359 {
360 	/* move B field from 1st to 2nd dword, trim ARPN */
361 	return (r & ~HPTE_R_3_0_SSIZE_MASK) |
362 		(((v) >> HPTE_V_SSIZE_SHIFT) << HPTE_R_3_0_SSIZE_SHIFT);
363 }
364 
365 static inline unsigned long hpte_new_to_old_v(unsigned long v, unsigned long r)
366 {
367 	/* insert B field */
368 	return (v & HPTE_V_COMMON_BITS) |
369 		((r & HPTE_R_3_0_SSIZE_MASK) <<
370 		 (HPTE_V_SSIZE_SHIFT - HPTE_R_3_0_SSIZE_SHIFT));
371 }
372 
373 static inline unsigned long hpte_new_to_old_r(unsigned long r)
374 {
375 	/* clear out B field */
376 	return r & ~HPTE_R_3_0_SSIZE_MASK;
377 }
378 
379 static inline unsigned long hpte_get_old_v(struct hash_pte *hptep)
380 {
381 	unsigned long hpte_v;
382 
383 	hpte_v = be64_to_cpu(hptep->v);
384 	if (cpu_has_feature(CPU_FTR_ARCH_300))
385 		hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r));
386 	return hpte_v;
387 }
388 
389 /*
390  * This function sets the AVPN and L fields of the HPTE  appropriately
391  * using the base page size and actual page size.
392  */
393 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
394 					  int actual_psize, int ssize)
395 {
396 	unsigned long v;
397 	v = hpte_encode_avpn(vpn, base_psize, ssize);
398 	if (actual_psize != MMU_PAGE_4K)
399 		v |= HPTE_V_LARGE;
400 	return v;
401 }
402 
403 /*
404  * This function sets the ARPN, and LP fields of the HPTE appropriately
405  * for the page size. We assume the pa is already "clean" that is properly
406  * aligned for the requested page size
407  */
408 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
409 					  int actual_psize)
410 {
411 	/* A 4K page needs no special encoding */
412 	if (actual_psize == MMU_PAGE_4K)
413 		return pa & HPTE_R_RPN;
414 	else {
415 		unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
416 		unsigned int shift = mmu_psize_defs[actual_psize].shift;
417 		return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
418 	}
419 }
420 
421 /*
422  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
423  */
424 static inline unsigned long hpt_vpn(unsigned long ea,
425 				    unsigned long vsid, int ssize)
426 {
427 	unsigned long mask;
428 	int s_shift = segment_shift(ssize);
429 
430 	mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
431 	return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
432 }
433 
434 /*
435  * This hashes a virtual address
436  */
437 static inline unsigned long hpt_hash(unsigned long vpn,
438 				     unsigned int shift, int ssize)
439 {
440 	unsigned long mask;
441 	unsigned long hash, vsid;
442 
443 	/* VPN_SHIFT can be atmost 12 */
444 	if (ssize == MMU_SEGSIZE_256M) {
445 		mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
446 		hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
447 			((vpn & mask) >> (shift - VPN_SHIFT));
448 	} else {
449 		mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
450 		vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
451 		hash = vsid ^ (vsid << 25) ^
452 			((vpn & mask) >> (shift - VPN_SHIFT)) ;
453 	}
454 	return hash & 0x7fffffffffUL;
455 }
456 
457 #define HPTE_LOCAL_UPDATE	0x1
458 #define HPTE_NOHPTE_UPDATE	0x2
459 
460 extern int __hash_page_4K(unsigned long ea, unsigned long access,
461 			  unsigned long vsid, pte_t *ptep, unsigned long trap,
462 			  unsigned long flags, int ssize, int subpage_prot);
463 extern int __hash_page_64K(unsigned long ea, unsigned long access,
464 			   unsigned long vsid, pte_t *ptep, unsigned long trap,
465 			   unsigned long flags, int ssize);
466 struct mm_struct;
467 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
468 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
469 			unsigned long access, unsigned long trap,
470 			unsigned long flags);
471 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
472 		     unsigned long dsisr);
473 int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
474 		     pte_t *ptep, unsigned long trap, unsigned long flags,
475 		     int ssize, unsigned int shift, unsigned int mmu_psize);
476 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
477 extern int __hash_page_thp(unsigned long ea, unsigned long access,
478 			   unsigned long vsid, pmd_t *pmdp, unsigned long trap,
479 			   unsigned long flags, int ssize, unsigned int psize);
480 #else
481 static inline int __hash_page_thp(unsigned long ea, unsigned long access,
482 				  unsigned long vsid, pmd_t *pmdp,
483 				  unsigned long trap, unsigned long flags,
484 				  int ssize, unsigned int psize)
485 {
486 	BUG();
487 	return -1;
488 }
489 #endif
490 extern void hash_failure_debug(unsigned long ea, unsigned long access,
491 			       unsigned long vsid, unsigned long trap,
492 			       int ssize, int psize, int lpsize,
493 			       unsigned long pte);
494 extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
495 			     unsigned long pstart, unsigned long prot,
496 			     int psize, int ssize);
497 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
498 			int psize, int ssize);
499 extern void pseries_add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
500 extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
501 
502 extern void hash__setup_new_exec(void);
503 
504 #ifdef CONFIG_PPC_PSERIES
505 void hpte_init_pseries(void);
506 #else
507 static inline void hpte_init_pseries(void) { }
508 #endif
509 
510 extern void hpte_init_native(void);
511 
512 struct slb_entry {
513 	u64	esid;
514 	u64	vsid;
515 };
516 
517 extern void slb_initialize(void);
518 void slb_flush_and_restore_bolted(void);
519 void slb_flush_all_realmode(void);
520 void __slb_restore_bolted_realmode(void);
521 void slb_restore_bolted_realmode(void);
522 void slb_save_contents(struct slb_entry *slb_ptr);
523 void slb_dump_contents(struct slb_entry *slb_ptr);
524 
525 extern void slb_vmalloc_update(void);
526 extern void slb_set_size(u16 size);
527 #endif /* __ASSEMBLY__ */
528 
529 /*
530  * VSID allocation (256MB segment)
531  *
532  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
533  * from mmu context id and effective segment id of the address.
534  *
535  * For user processes max context id is limited to MAX_USER_CONTEXT.
536  * more details in get_user_context
537  *
538  * For kernel space get_kernel_context
539  *
540  * The proto-VSIDs are then scrambled into real VSIDs with the
541  * multiplicative hash:
542  *
543  *	VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
544  *
545  * VSID_MULTIPLIER is prime, so in particular it is
546  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
547  * Because the modulus is 2^n-1 we can compute it efficiently without
548  * a divide or extra multiply (see below). The scramble function gives
549  * robust scattering in the hash table (at least based on some initial
550  * results).
551  *
552  * We use VSID 0 to indicate an invalid VSID. The means we can't use context id
553  * 0, because a context id of 0 and an EA of 0 gives a proto-VSID of 0, which
554  * will produce a VSID of 0.
555  *
556  * We also need to avoid the last segment of the last context, because that
557  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
558  * because of the modulo operation in vsid scramble.
559  */
560 
561 /*
562  * Max Va bits we support as of now is 68 bits. We want 19 bit
563  * context ID.
564  * Restrictions:
565  * GPU has restrictions of not able to access beyond 128TB
566  * (47 bit effective address). We also cannot do more than 20bit PID.
567  * For p4 and p5 which can only do 65 bit VA, we restrict our CONTEXT_BITS
568  * to 16 bits (ie, we can only have 2^16 pids at the same time).
569  */
570 #define VA_BITS			68
571 #define CONTEXT_BITS		19
572 #define ESID_BITS		(VA_BITS - (SID_SHIFT + CONTEXT_BITS))
573 #define ESID_BITS_1T		(VA_BITS - (SID_SHIFT_1T + CONTEXT_BITS))
574 
575 #define ESID_BITS_MASK		((1 << ESID_BITS) - 1)
576 #define ESID_BITS_1T_MASK	((1 << ESID_BITS_1T) - 1)
577 
578 /*
579  * Now certain config support MAX_PHYSMEM more than 512TB. Hence we will need
580  * to use more than one context for linear mapping the kernel.
581  * For vmalloc and memmap, we use just one context with 512TB. With 64 byte
582  * struct page size, we need ony 32 TB in memmap for 2PB (51 bits (MAX_PHYSMEM_BITS)).
583  */
584 #if (MAX_PHYSMEM_BITS > MAX_EA_BITS_PER_CONTEXT)
585 #define MAX_KERNEL_CTX_CNT	(1UL << (MAX_PHYSMEM_BITS - MAX_EA_BITS_PER_CONTEXT))
586 #else
587 #define MAX_KERNEL_CTX_CNT	1
588 #endif
589 
590 #define MAX_VMALLOC_CTX_CNT	1
591 #define MAX_MEMMAP_CTX_CNT	1
592 
593 /*
594  * 256MB segment
595  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
596  * available for user + kernel mapping. VSID 0 is reserved as invalid, contexts
597  * 1-4 are used for kernel mapping. Each segment contains 2^28 bytes. Each
598  * context maps 2^49 bytes (512TB).
599  *
600  * We also need to avoid the last segment of the last context, because that
601  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
602  * because of the modulo operation in vsid scramble.
603  *
604  * We add one extra context to MIN_USER_CONTEXT so that we can map kernel
605  * context easily. The +1 is to map the unused 0xe region mapping.
606  */
607 #define MAX_USER_CONTEXT	((ASM_CONST(1) << CONTEXT_BITS) - 2)
608 #define MIN_USER_CONTEXT	(MAX_KERNEL_CTX_CNT + MAX_VMALLOC_CTX_CNT + \
609 				 MAX_MEMMAP_CTX_CNT + 2)
610 
611 /*
612  * For platforms that support on 65bit VA we limit the context bits
613  */
614 #define MAX_USER_CONTEXT_65BIT_VA ((ASM_CONST(1) << (65 - (SID_SHIFT + ESID_BITS))) - 2)
615 
616 /*
617  * This should be computed such that protovosid * vsid_mulitplier
618  * doesn't overflow 64 bits. The vsid_mutliplier should also be
619  * co-prime to vsid_modulus. We also need to make sure that number
620  * of bits in multiplied result (dividend) is less than twice the number of
621  * protovsid bits for our modulus optmization to work.
622  *
623  * The below table shows the current values used.
624  * |-------+------------+----------------------+------------+-------------------|
625  * |       | Prime Bits | proto VSID_BITS_65VA | Total Bits | 2* prot VSID_BITS |
626  * |-------+------------+----------------------+------------+-------------------|
627  * | 1T    |         24 |                   25 |         49 |                50 |
628  * |-------+------------+----------------------+------------+-------------------|
629  * | 256MB |         24 |                   37 |         61 |                74 |
630  * |-------+------------+----------------------+------------+-------------------|
631  *
632  * |-------+------------+----------------------+------------+--------------------|
633  * |       | Prime Bits | proto VSID_BITS_68VA | Total Bits | 2* proto VSID_BITS |
634  * |-------+------------+----------------------+------------+--------------------|
635  * | 1T    |         24 |                   28 |         52 |                 56 |
636  * |-------+------------+----------------------+------------+--------------------|
637  * | 256MB |         24 |                   40 |         64 |                 80 |
638  * |-------+------------+----------------------+------------+--------------------|
639  *
640  */
641 #define VSID_MULTIPLIER_256M	ASM_CONST(12538073)	/* 24-bit prime */
642 #define VSID_BITS_256M		(VA_BITS - SID_SHIFT)
643 #define VSID_BITS_65_256M	(65 - SID_SHIFT)
644 /*
645  * Modular multiplicative inverse of VSID_MULTIPLIER under modulo VSID_MODULUS
646  */
647 #define VSID_MULINV_256M	ASM_CONST(665548017062)
648 
649 #define VSID_MULTIPLIER_1T	ASM_CONST(12538073)	/* 24-bit prime */
650 #define VSID_BITS_1T		(VA_BITS - SID_SHIFT_1T)
651 #define VSID_BITS_65_1T		(65 - SID_SHIFT_1T)
652 #define VSID_MULINV_1T		ASM_CONST(209034062)
653 
654 /* 1TB VSID reserved for VRMA */
655 #define VRMA_VSID	0x1ffffffUL
656 #define USER_VSID_RANGE	(1UL << (ESID_BITS + SID_SHIFT))
657 
658 /* 4 bits per slice and we have one slice per 1TB */
659 #define SLICE_ARRAY_SIZE	(H_PGTABLE_RANGE >> 41)
660 #define TASK_SLICE_ARRAY_SZ(x)	((x)->context.slb_addr_limit >> 41)
661 
662 #ifndef __ASSEMBLY__
663 
664 #ifdef CONFIG_PPC_SUBPAGE_PROT
665 /*
666  * For the sub-page protection option, we extend the PGD with one of
667  * these.  Basically we have a 3-level tree, with the top level being
668  * the protptrs array.  To optimize speed and memory consumption when
669  * only addresses < 4GB are being protected, pointers to the first
670  * four pages of sub-page protection words are stored in the low_prot
671  * array.
672  * Each page of sub-page protection words protects 1GB (4 bytes
673  * protects 64k).  For the 3-level tree, each page of pointers then
674  * protects 8TB.
675  */
676 struct subpage_prot_table {
677 	unsigned long maxaddr;	/* only addresses < this are protected */
678 	unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
679 	unsigned int *low_prot[4];
680 };
681 
682 #define SBP_L1_BITS		(PAGE_SHIFT - 2)
683 #define SBP_L2_BITS		(PAGE_SHIFT - 3)
684 #define SBP_L1_COUNT		(1 << SBP_L1_BITS)
685 #define SBP_L2_COUNT		(1 << SBP_L2_BITS)
686 #define SBP_L2_SHIFT		(PAGE_SHIFT + SBP_L1_BITS)
687 #define SBP_L3_SHIFT		(SBP_L2_SHIFT + SBP_L2_BITS)
688 
689 extern void subpage_prot_free(struct mm_struct *mm);
690 extern void subpage_prot_init_new_context(struct mm_struct *mm);
691 #else
692 static inline void subpage_prot_free(struct mm_struct *mm) {}
693 static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
694 #endif /* CONFIG_PPC_SUBPAGE_PROT */
695 
696 #if 0
697 /*
698  * The code below is equivalent to this function for arguments
699  * < 2^VSID_BITS, which is all this should ever be called
700  * with.  However gcc is not clever enough to compute the
701  * modulus (2^n-1) without a second multiply.
702  */
703 #define vsid_scramble(protovsid, size) \
704 	((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
705 
706 /* simplified form avoiding mod operation */
707 #define vsid_scramble(protovsid, size) \
708 	({								 \
709 		unsigned long x;					 \
710 		x = (protovsid) * VSID_MULTIPLIER_##size;		 \
711 		x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
712 		(x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
713 	})
714 
715 #else /* 1 */
716 static inline unsigned long vsid_scramble(unsigned long protovsid,
717 				  unsigned long vsid_multiplier, int vsid_bits)
718 {
719 	unsigned long vsid;
720 	unsigned long vsid_modulus = ((1UL << vsid_bits) - 1);
721 	/*
722 	 * We have same multipler for both 256 and 1T segements now
723 	 */
724 	vsid = protovsid * vsid_multiplier;
725 	vsid = (vsid >> vsid_bits) + (vsid & vsid_modulus);
726 	return (vsid + ((vsid + 1) >> vsid_bits)) & vsid_modulus;
727 }
728 
729 #endif /* 1 */
730 
731 /* Returns the segment size indicator for a user address */
732 static inline int user_segment_size(unsigned long addr)
733 {
734 	/* Use 1T segments if possible for addresses >= 1T */
735 	if (addr >= (1UL << SID_SHIFT_1T))
736 		return mmu_highuser_ssize;
737 	return MMU_SEGSIZE_256M;
738 }
739 
740 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
741 				     int ssize)
742 {
743 	unsigned long va_bits = VA_BITS;
744 	unsigned long vsid_bits;
745 	unsigned long protovsid;
746 
747 	/*
748 	 * Bad address. We return VSID 0 for that
749 	 */
750 	if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
751 		return 0;
752 
753 	if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
754 		va_bits = 65;
755 
756 	if (ssize == MMU_SEGSIZE_256M) {
757 		vsid_bits = va_bits - SID_SHIFT;
758 		protovsid = (context << ESID_BITS) |
759 			((ea >> SID_SHIFT) & ESID_BITS_MASK);
760 		return vsid_scramble(protovsid, VSID_MULTIPLIER_256M, vsid_bits);
761 	}
762 	/* 1T segment */
763 	vsid_bits = va_bits - SID_SHIFT_1T;
764 	protovsid = (context << ESID_BITS_1T) |
765 		((ea >> SID_SHIFT_1T) & ESID_BITS_1T_MASK);
766 	return vsid_scramble(protovsid, VSID_MULTIPLIER_1T, vsid_bits);
767 }
768 
769 /*
770  * For kernel space, we use context ids as below
771  * below. Range is 512TB per context.
772  *
773  * 0x00001 -  [ 0xc000000000000000 - 0xc001ffffffffffff]
774  * 0x00002 -  [ 0xc002000000000000 - 0xc003ffffffffffff]
775  * 0x00003 -  [ 0xc004000000000000 - 0xc005ffffffffffff]
776  * 0x00004 -  [ 0xc006000000000000 - 0xc007ffffffffffff]
777 
778  * 0x00005 -  [ 0xd000000000000000 - 0xd001ffffffffffff ]
779  * 0x00006 -  Not used - Can map 0xe000000000000000 range.
780  * 0x00007 -  [ 0xf000000000000000 - 0xf001ffffffffffff ]
781  *
782  * So we can compute the context from the region (top nibble) by
783  * subtracting 11, or 0xc - 1.
784  */
785 static inline unsigned long get_kernel_context(unsigned long ea)
786 {
787 	unsigned long region_id = REGION_ID(ea);
788 	unsigned long ctx;
789 	/*
790 	 * For linear mapping we do support multiple context
791 	 */
792 	if (region_id == KERNEL_REGION_ID) {
793 		/*
794 		 * We already verified ea to be not beyond the addr limit.
795 		 */
796 		ctx =  1 + ((ea & ~REGION_MASK) >> MAX_EA_BITS_PER_CONTEXT);
797 	} else
798 		ctx = (region_id - 0xc) + MAX_KERNEL_CTX_CNT;
799 	return ctx;
800 }
801 
802 /*
803  * This is only valid for addresses >= PAGE_OFFSET
804  */
805 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
806 {
807 	unsigned long context;
808 
809 	if (!is_kernel_addr(ea))
810 		return 0;
811 
812 	context = get_kernel_context(ea);
813 	return get_vsid(context, ea, ssize);
814 }
815 
816 unsigned htab_shift_for_mem_size(unsigned long mem_size);
817 
818 #endif /* __ASSEMBLY__ */
819 
820 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */
821