1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 #ifndef _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_
3 #define _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_
4 /*
5  * PowerPC64 memory management structures
6  *
7  * Dave Engebretsen & Mike Corrigan <{engebret|mikejc}@us.ibm.com>
8  *   PPC64 rework.
9  */
10 
11 #include <asm/page.h>
12 #include <asm/bug.h>
13 #include <asm/asm-const.h>
14 
15 /*
16  * This is necessary to get the definition of PGTABLE_RANGE which we
17  * need for various slices related matters. Note that this isn't the
18  * complete pgtable.h but only a portion of it.
19  */
20 #include <asm/book3s/64/pgtable.h>
21 #include <asm/task_size_64.h>
22 #include <asm/cpu_has_feature.h>
23 
24 /*
25  * SLB
26  */
27 
28 #define SLB_NUM_BOLTED		2
29 #define SLB_CACHE_ENTRIES	8
30 #define SLB_MIN_SIZE		32
31 
32 /* Bits in the SLB ESID word */
33 #define SLB_ESID_V		ASM_CONST(0x0000000008000000) /* valid */
34 
35 /* Bits in the SLB VSID word */
36 #define SLB_VSID_SHIFT		12
37 #define SLB_VSID_SHIFT_256M	SLB_VSID_SHIFT
38 #define SLB_VSID_SHIFT_1T	24
39 #define SLB_VSID_SSIZE_SHIFT	62
40 #define SLB_VSID_B		ASM_CONST(0xc000000000000000)
41 #define SLB_VSID_B_256M		ASM_CONST(0x0000000000000000)
42 #define SLB_VSID_B_1T		ASM_CONST(0x4000000000000000)
43 #define SLB_VSID_KS		ASM_CONST(0x0000000000000800)
44 #define SLB_VSID_KP		ASM_CONST(0x0000000000000400)
45 #define SLB_VSID_N		ASM_CONST(0x0000000000000200) /* no-execute */
46 #define SLB_VSID_L		ASM_CONST(0x0000000000000100)
47 #define SLB_VSID_C		ASM_CONST(0x0000000000000080) /* class */
48 #define SLB_VSID_LP		ASM_CONST(0x0000000000000030)
49 #define SLB_VSID_LP_00		ASM_CONST(0x0000000000000000)
50 #define SLB_VSID_LP_01		ASM_CONST(0x0000000000000010)
51 #define SLB_VSID_LP_10		ASM_CONST(0x0000000000000020)
52 #define SLB_VSID_LP_11		ASM_CONST(0x0000000000000030)
53 #define SLB_VSID_LLP		(SLB_VSID_L|SLB_VSID_LP)
54 
55 #define SLB_VSID_KERNEL		(SLB_VSID_KP)
56 #define SLB_VSID_USER		(SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
57 
58 #define SLBIE_C			(0x08000000)
59 #define SLBIE_SSIZE_SHIFT	25
60 
61 /*
62  * Hash table
63  */
64 
65 #define HPTES_PER_GROUP 8
66 
67 #define HPTE_V_SSIZE_SHIFT	62
68 #define HPTE_V_AVPN_SHIFT	7
69 #define HPTE_V_COMMON_BITS	ASM_CONST(0x000fffffffffffff)
70 #define HPTE_V_AVPN		ASM_CONST(0x3fffffffffffff80)
71 #define HPTE_V_AVPN_3_0		ASM_CONST(0x000fffffffffff80)
72 #define HPTE_V_AVPN_VAL(x)	(((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
73 #define HPTE_V_COMPARE(x,y)	(!(((x) ^ (y)) & 0xffffffffffffff80UL))
74 #define HPTE_V_BOLTED		ASM_CONST(0x0000000000000010)
75 #define HPTE_V_LOCK		ASM_CONST(0x0000000000000008)
76 #define HPTE_V_LARGE		ASM_CONST(0x0000000000000004)
77 #define HPTE_V_SECONDARY	ASM_CONST(0x0000000000000002)
78 #define HPTE_V_VALID		ASM_CONST(0x0000000000000001)
79 
80 /*
81  * ISA 3.0 has a different HPTE format.
82  */
83 #define HPTE_R_3_0_SSIZE_SHIFT	58
84 #define HPTE_R_3_0_SSIZE_MASK	(3ull << HPTE_R_3_0_SSIZE_SHIFT)
85 #define HPTE_R_PP0		ASM_CONST(0x8000000000000000)
86 #define HPTE_R_TS		ASM_CONST(0x4000000000000000)
87 #define HPTE_R_KEY_HI		ASM_CONST(0x3000000000000000)
88 #define HPTE_R_KEY_BIT4		ASM_CONST(0x2000000000000000)
89 #define HPTE_R_KEY_BIT3		ASM_CONST(0x1000000000000000)
90 #define HPTE_R_RPN_SHIFT	12
91 #define HPTE_R_RPN		ASM_CONST(0x0ffffffffffff000)
92 #define HPTE_R_RPN_3_0		ASM_CONST(0x01fffffffffff000)
93 #define HPTE_R_PP		ASM_CONST(0x0000000000000003)
94 #define HPTE_R_PPP		ASM_CONST(0x8000000000000003)
95 #define HPTE_R_N		ASM_CONST(0x0000000000000004)
96 #define HPTE_R_G		ASM_CONST(0x0000000000000008)
97 #define HPTE_R_M		ASM_CONST(0x0000000000000010)
98 #define HPTE_R_I		ASM_CONST(0x0000000000000020)
99 #define HPTE_R_W		ASM_CONST(0x0000000000000040)
100 #define HPTE_R_WIMG		ASM_CONST(0x0000000000000078)
101 #define HPTE_R_C		ASM_CONST(0x0000000000000080)
102 #define HPTE_R_R		ASM_CONST(0x0000000000000100)
103 #define HPTE_R_KEY_LO		ASM_CONST(0x0000000000000e00)
104 #define HPTE_R_KEY_BIT2		ASM_CONST(0x0000000000000800)
105 #define HPTE_R_KEY_BIT1		ASM_CONST(0x0000000000000400)
106 #define HPTE_R_KEY_BIT0		ASM_CONST(0x0000000000000200)
107 #define HPTE_R_KEY		(HPTE_R_KEY_LO | HPTE_R_KEY_HI)
108 
109 #define HPTE_V_1TB_SEG		ASM_CONST(0x4000000000000000)
110 #define HPTE_V_VRMA_MASK	ASM_CONST(0x4001ffffff000000)
111 
112 /* Values for PP (assumes Ks=0, Kp=1) */
113 #define PP_RWXX	0	/* Supervisor read/write, User none */
114 #define PP_RWRX 1	/* Supervisor read/write, User read */
115 #define PP_RWRW 2	/* Supervisor read/write, User read/write */
116 #define PP_RXRX 3	/* Supervisor read,       User read */
117 #define PP_RXXX	(HPTE_R_PP0 | 2)	/* Supervisor read, user none */
118 
119 /* Fields for tlbiel instruction in architecture 2.06 */
120 #define TLBIEL_INVAL_SEL_MASK	0xc00	/* invalidation selector */
121 #define  TLBIEL_INVAL_PAGE	0x000	/* invalidate a single page */
122 #define  TLBIEL_INVAL_SET_LPID	0x800	/* invalidate a set for current LPID */
123 #define  TLBIEL_INVAL_SET	0xc00	/* invalidate a set for all LPIDs */
124 #define TLBIEL_INVAL_SET_MASK	0xfff000	/* set number to inval. */
125 #define TLBIEL_INVAL_SET_SHIFT	12
126 
127 #define POWER7_TLB_SETS		128	/* # sets in POWER7 TLB */
128 #define POWER8_TLB_SETS		512	/* # sets in POWER8 TLB */
129 #define POWER9_TLB_SETS_HASH	256	/* # sets in POWER9 TLB Hash mode */
130 #define POWER9_TLB_SETS_RADIX	128	/* # sets in POWER9 TLB Radix mode */
131 
132 #ifndef __ASSEMBLY__
133 
134 struct mmu_hash_ops {
135 	void            (*hpte_invalidate)(unsigned long slot,
136 					   unsigned long vpn,
137 					   int bpsize, int apsize,
138 					   int ssize, int local);
139 	long		(*hpte_updatepp)(unsigned long slot,
140 					 unsigned long newpp,
141 					 unsigned long vpn,
142 					 int bpsize, int apsize,
143 					 int ssize, unsigned long flags);
144 	void            (*hpte_updateboltedpp)(unsigned long newpp,
145 					       unsigned long ea,
146 					       int psize, int ssize);
147 	long		(*hpte_insert)(unsigned long hpte_group,
148 				       unsigned long vpn,
149 				       unsigned long prpn,
150 				       unsigned long rflags,
151 				       unsigned long vflags,
152 				       int psize, int apsize,
153 				       int ssize);
154 	long		(*hpte_remove)(unsigned long hpte_group);
155 	int             (*hpte_removebolted)(unsigned long ea,
156 					     int psize, int ssize);
157 	void		(*flush_hash_range)(unsigned long number, int local);
158 	void		(*hugepage_invalidate)(unsigned long vsid,
159 					       unsigned long addr,
160 					       unsigned char *hpte_slot_array,
161 					       int psize, int ssize, int local);
162 	int		(*resize_hpt)(unsigned long shift);
163 	/*
164 	 * Special for kexec.
165 	 * To be called in real mode with interrupts disabled. No locks are
166 	 * taken as such, concurrent access on pre POWER5 hardware could result
167 	 * in a deadlock.
168 	 * The linear mapping is destroyed as well.
169 	 */
170 	void		(*hpte_clear_all)(void);
171 };
172 extern struct mmu_hash_ops mmu_hash_ops;
173 
174 struct hash_pte {
175 	__be64 v;
176 	__be64 r;
177 };
178 
179 extern struct hash_pte *htab_address;
180 extern unsigned long htab_size_bytes;
181 extern unsigned long htab_hash_mask;
182 
183 
184 static inline int shift_to_mmu_psize(unsigned int shift)
185 {
186 	int psize;
187 
188 	for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
189 		if (mmu_psize_defs[psize].shift == shift)
190 			return psize;
191 	return -1;
192 }
193 
194 static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
195 {
196 	if (mmu_psize_defs[mmu_psize].shift)
197 		return mmu_psize_defs[mmu_psize].shift;
198 	BUG();
199 }
200 
201 static inline unsigned int ap_to_shift(unsigned long ap)
202 {
203 	int psize;
204 
205 	for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
206 		if (mmu_psize_defs[psize].ap == ap)
207 			return mmu_psize_defs[psize].shift;
208 	}
209 
210 	return -1;
211 }
212 
213 static inline unsigned long get_sllp_encoding(int psize)
214 {
215 	unsigned long sllp;
216 
217 	sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
218 		((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
219 	return sllp;
220 }
221 
222 #endif /* __ASSEMBLY__ */
223 
224 /*
225  * Segment sizes.
226  * These are the values used by hardware in the B field of
227  * SLB entries and the first dword of MMU hashtable entries.
228  * The B field is 2 bits; the values 2 and 3 are unused and reserved.
229  */
230 #define MMU_SEGSIZE_256M	0
231 #define MMU_SEGSIZE_1T		1
232 
233 /*
234  * encode page number shift.
235  * in order to fit the 78 bit va in a 64 bit variable we shift the va by
236  * 12 bits. This enable us to address upto 76 bit va.
237  * For hpt hash from a va we can ignore the page size bits of va and for
238  * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
239  * we work in all cases including 4k page size.
240  */
241 #define VPN_SHIFT	12
242 
243 /*
244  * HPTE Large Page (LP) details
245  */
246 #define LP_SHIFT	12
247 #define LP_BITS		8
248 #define LP_MASK(i)	((0xFF >> (i)) << LP_SHIFT)
249 
250 #ifndef __ASSEMBLY__
251 
252 static inline int slb_vsid_shift(int ssize)
253 {
254 	if (ssize == MMU_SEGSIZE_256M)
255 		return SLB_VSID_SHIFT;
256 	return SLB_VSID_SHIFT_1T;
257 }
258 
259 static inline int segment_shift(int ssize)
260 {
261 	if (ssize == MMU_SEGSIZE_256M)
262 		return SID_SHIFT;
263 	return SID_SHIFT_1T;
264 }
265 
266 /*
267  * This array is indexed by the LP field of the HPTE second dword.
268  * Since this field may contain some RPN bits, some entries are
269  * replicated so that we get the same value irrespective of RPN.
270  * The top 4 bits are the page size index (MMU_PAGE_*) for the
271  * actual page size, the bottom 4 bits are the base page size.
272  */
273 extern u8 hpte_page_sizes[1 << LP_BITS];
274 
275 static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
276 					     bool is_base_size)
277 {
278 	unsigned int i, lp;
279 
280 	if (!(h & HPTE_V_LARGE))
281 		return 1ul << 12;
282 
283 	/* Look at the 8 bit LP value */
284 	lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
285 	i = hpte_page_sizes[lp];
286 	if (!i)
287 		return 0;
288 	if (!is_base_size)
289 		i >>= 4;
290 	return 1ul << mmu_psize_defs[i & 0xf].shift;
291 }
292 
293 static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
294 {
295 	return __hpte_page_size(h, l, 0);
296 }
297 
298 static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
299 {
300 	return __hpte_page_size(h, l, 1);
301 }
302 
303 /*
304  * The current system page and segment sizes
305  */
306 extern int mmu_kernel_ssize;
307 extern int mmu_highuser_ssize;
308 extern u16 mmu_slb_size;
309 extern unsigned long tce_alloc_start, tce_alloc_end;
310 
311 /*
312  * If the processor supports 64k normal pages but not 64k cache
313  * inhibited pages, we have to be prepared to switch processes
314  * to use 4k pages when they create cache-inhibited mappings.
315  * If this is the case, mmu_ci_restrictions will be set to 1.
316  */
317 extern int mmu_ci_restrictions;
318 
319 /*
320  * This computes the AVPN and B fields of the first dword of a HPTE,
321  * for use when we want to match an existing PTE.  The bottom 7 bits
322  * of the returned value are zero.
323  */
324 static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
325 					     int ssize)
326 {
327 	unsigned long v;
328 	/*
329 	 * The AVA field omits the low-order 23 bits of the 78 bits VA.
330 	 * These bits are not needed in the PTE, because the
331 	 * low-order b of these bits are part of the byte offset
332 	 * into the virtual page and, if b < 23, the high-order
333 	 * 23-b of these bits are always used in selecting the
334 	 * PTEGs to be searched
335 	 */
336 	v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
337 	v <<= HPTE_V_AVPN_SHIFT;
338 	v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
339 	return v;
340 }
341 
342 /*
343  * ISA v3.0 defines a new HPTE format, which differs from the old
344  * format in having smaller AVPN and ARPN fields, and the B field
345  * in the second dword instead of the first.
346  */
347 static inline unsigned long hpte_old_to_new_v(unsigned long v)
348 {
349 	/* trim AVPN, drop B */
350 	return v & HPTE_V_COMMON_BITS;
351 }
352 
353 static inline unsigned long hpte_old_to_new_r(unsigned long v, unsigned long r)
354 {
355 	/* move B field from 1st to 2nd dword, trim ARPN */
356 	return (r & ~HPTE_R_3_0_SSIZE_MASK) |
357 		(((v) >> HPTE_V_SSIZE_SHIFT) << HPTE_R_3_0_SSIZE_SHIFT);
358 }
359 
360 static inline unsigned long hpte_new_to_old_v(unsigned long v, unsigned long r)
361 {
362 	/* insert B field */
363 	return (v & HPTE_V_COMMON_BITS) |
364 		((r & HPTE_R_3_0_SSIZE_MASK) <<
365 		 (HPTE_V_SSIZE_SHIFT - HPTE_R_3_0_SSIZE_SHIFT));
366 }
367 
368 static inline unsigned long hpte_new_to_old_r(unsigned long r)
369 {
370 	/* clear out B field */
371 	return r & ~HPTE_R_3_0_SSIZE_MASK;
372 }
373 
374 static inline unsigned long hpte_get_old_v(struct hash_pte *hptep)
375 {
376 	unsigned long hpte_v;
377 
378 	hpte_v = be64_to_cpu(hptep->v);
379 	if (cpu_has_feature(CPU_FTR_ARCH_300))
380 		hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r));
381 	return hpte_v;
382 }
383 
384 /*
385  * This function sets the AVPN and L fields of the HPTE  appropriately
386  * using the base page size and actual page size.
387  */
388 static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
389 					  int actual_psize, int ssize)
390 {
391 	unsigned long v;
392 	v = hpte_encode_avpn(vpn, base_psize, ssize);
393 	if (actual_psize != MMU_PAGE_4K)
394 		v |= HPTE_V_LARGE;
395 	return v;
396 }
397 
398 /*
399  * This function sets the ARPN, and LP fields of the HPTE appropriately
400  * for the page size. We assume the pa is already "clean" that is properly
401  * aligned for the requested page size
402  */
403 static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
404 					  int actual_psize)
405 {
406 	/* A 4K page needs no special encoding */
407 	if (actual_psize == MMU_PAGE_4K)
408 		return pa & HPTE_R_RPN;
409 	else {
410 		unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
411 		unsigned int shift = mmu_psize_defs[actual_psize].shift;
412 		return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
413 	}
414 }
415 
416 /*
417  * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
418  */
419 static inline unsigned long hpt_vpn(unsigned long ea,
420 				    unsigned long vsid, int ssize)
421 {
422 	unsigned long mask;
423 	int s_shift = segment_shift(ssize);
424 
425 	mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
426 	return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
427 }
428 
429 /*
430  * This hashes a virtual address
431  */
432 static inline unsigned long hpt_hash(unsigned long vpn,
433 				     unsigned int shift, int ssize)
434 {
435 	unsigned long mask;
436 	unsigned long hash, vsid;
437 
438 	/* VPN_SHIFT can be atmost 12 */
439 	if (ssize == MMU_SEGSIZE_256M) {
440 		mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
441 		hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
442 			((vpn & mask) >> (shift - VPN_SHIFT));
443 	} else {
444 		mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
445 		vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
446 		hash = vsid ^ (vsid << 25) ^
447 			((vpn & mask) >> (shift - VPN_SHIFT)) ;
448 	}
449 	return hash & 0x7fffffffffUL;
450 }
451 
452 #define HPTE_LOCAL_UPDATE	0x1
453 #define HPTE_NOHPTE_UPDATE	0x2
454 #define HPTE_USE_KERNEL_KEY	0x4
455 
456 long hpte_insert_repeating(unsigned long hash, unsigned long vpn, unsigned long pa,
457 			   unsigned long rlags, unsigned long vflags, int psize, int ssize);
458 extern int __hash_page_4K(unsigned long ea, unsigned long access,
459 			  unsigned long vsid, pte_t *ptep, unsigned long trap,
460 			  unsigned long flags, int ssize, int subpage_prot);
461 extern int __hash_page_64K(unsigned long ea, unsigned long access,
462 			   unsigned long vsid, pte_t *ptep, unsigned long trap,
463 			   unsigned long flags, int ssize);
464 struct mm_struct;
465 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
466 extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
467 			unsigned long access, unsigned long trap,
468 			unsigned long flags);
469 extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
470 		     unsigned long dsisr);
471 void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc);
472 int __hash_page(unsigned long trap, unsigned long ea, unsigned long dsisr, unsigned long msr);
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 void preload_new_slb_context(unsigned long start, unsigned long sp);
527 
528 #ifdef CONFIG_PPC_64S_HASH_MMU
529 void slb_set_size(u16 size);
530 #else
531 static inline void slb_set_size(u16 size) { }
532 #endif
533 
534 #endif /* __ASSEMBLY__ */
535 
536 /*
537  * VSID allocation (256MB segment)
538  *
539  * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
540  * from mmu context id and effective segment id of the address.
541  *
542  * For user processes max context id is limited to MAX_USER_CONTEXT.
543  * more details in get_user_context
544  *
545  * For kernel space get_kernel_context
546  *
547  * The proto-VSIDs are then scrambled into real VSIDs with the
548  * multiplicative hash:
549  *
550  *	VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
551  *
552  * VSID_MULTIPLIER is prime, so in particular it is
553  * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
554  * Because the modulus is 2^n-1 we can compute it efficiently without
555  * a divide or extra multiply (see below). The scramble function gives
556  * robust scattering in the hash table (at least based on some initial
557  * results).
558  *
559  * We use VSID 0 to indicate an invalid VSID. The means we can't use context id
560  * 0, because a context id of 0 and an EA of 0 gives a proto-VSID of 0, which
561  * will produce a VSID of 0.
562  *
563  * We also need to avoid the last segment of the last context, because that
564  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
565  * because of the modulo operation in vsid scramble.
566  */
567 
568 /*
569  * Max Va bits we support as of now is 68 bits. We want 19 bit
570  * context ID.
571  * Restrictions:
572  * GPU has restrictions of not able to access beyond 128TB
573  * (47 bit effective address). We also cannot do more than 20bit PID.
574  * For p4 and p5 which can only do 65 bit VA, we restrict our CONTEXT_BITS
575  * to 16 bits (ie, we can only have 2^16 pids at the same time).
576  */
577 #define VA_BITS			68
578 #define CONTEXT_BITS		19
579 #define ESID_BITS		(VA_BITS - (SID_SHIFT + CONTEXT_BITS))
580 #define ESID_BITS_1T		(VA_BITS - (SID_SHIFT_1T + CONTEXT_BITS))
581 
582 #define ESID_BITS_MASK		((1 << ESID_BITS) - 1)
583 #define ESID_BITS_1T_MASK	((1 << ESID_BITS_1T) - 1)
584 
585 /*
586  * Now certain config support MAX_PHYSMEM more than 512TB. Hence we will need
587  * to use more than one context for linear mapping the kernel.
588  * For vmalloc and memmap, we use just one context with 512TB. With 64 byte
589  * struct page size, we need ony 32 TB in memmap for 2PB (51 bits (MAX_PHYSMEM_BITS)).
590  */
591 #if (H_MAX_PHYSMEM_BITS > MAX_EA_BITS_PER_CONTEXT)
592 #define MAX_KERNEL_CTX_CNT	(1UL << (H_MAX_PHYSMEM_BITS - MAX_EA_BITS_PER_CONTEXT))
593 #else
594 #define MAX_KERNEL_CTX_CNT	1
595 #endif
596 
597 #define MAX_VMALLOC_CTX_CNT	1
598 #define MAX_IO_CTX_CNT		1
599 #define MAX_VMEMMAP_CTX_CNT	1
600 
601 /*
602  * 256MB segment
603  * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
604  * available for user + kernel mapping. VSID 0 is reserved as invalid, contexts
605  * 1-4 are used for kernel mapping. Each segment contains 2^28 bytes. Each
606  * context maps 2^49 bytes (512TB).
607  *
608  * We also need to avoid the last segment of the last context, because that
609  * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
610  * because of the modulo operation in vsid scramble.
611  *
612  */
613 #define MAX_USER_CONTEXT	((ASM_CONST(1) << CONTEXT_BITS) - 2)
614 
615 // The + 2 accounts for INVALID_REGION and 1 more to avoid overlap with kernel
616 #define MIN_USER_CONTEXT	(MAX_KERNEL_CTX_CNT + MAX_VMALLOC_CTX_CNT + \
617 				 MAX_IO_CTX_CNT + MAX_VMEMMAP_CTX_CNT + 2)
618 
619 /*
620  * For platforms that support on 65bit VA we limit the context bits
621  */
622 #define MAX_USER_CONTEXT_65BIT_VA ((ASM_CONST(1) << (65 - (SID_SHIFT + ESID_BITS))) - 2)
623 
624 /*
625  * This should be computed such that protovosid * vsid_mulitplier
626  * doesn't overflow 64 bits. The vsid_mutliplier should also be
627  * co-prime to vsid_modulus. We also need to make sure that number
628  * of bits in multiplied result (dividend) is less than twice the number of
629  * protovsid bits for our modulus optmization to work.
630  *
631  * The below table shows the current values used.
632  * |-------+------------+----------------------+------------+-------------------|
633  * |       | Prime Bits | proto VSID_BITS_65VA | Total Bits | 2* prot VSID_BITS |
634  * |-------+------------+----------------------+------------+-------------------|
635  * | 1T    |         24 |                   25 |         49 |                50 |
636  * |-------+------------+----------------------+------------+-------------------|
637  * | 256MB |         24 |                   37 |         61 |                74 |
638  * |-------+------------+----------------------+------------+-------------------|
639  *
640  * |-------+------------+----------------------+------------+--------------------|
641  * |       | Prime Bits | proto VSID_BITS_68VA | Total Bits | 2* proto VSID_BITS |
642  * |-------+------------+----------------------+------------+--------------------|
643  * | 1T    |         24 |                   28 |         52 |                 56 |
644  * |-------+------------+----------------------+------------+--------------------|
645  * | 256MB |         24 |                   40 |         64 |                 80 |
646  * |-------+------------+----------------------+------------+--------------------|
647  *
648  */
649 #define VSID_MULTIPLIER_256M	ASM_CONST(12538073)	/* 24-bit prime */
650 #define VSID_BITS_256M		(VA_BITS - SID_SHIFT)
651 #define VSID_BITS_65_256M	(65 - SID_SHIFT)
652 /*
653  * Modular multiplicative inverse of VSID_MULTIPLIER under modulo VSID_MODULUS
654  */
655 #define VSID_MULINV_256M	ASM_CONST(665548017062)
656 
657 #define VSID_MULTIPLIER_1T	ASM_CONST(12538073)	/* 24-bit prime */
658 #define VSID_BITS_1T		(VA_BITS - SID_SHIFT_1T)
659 #define VSID_BITS_65_1T		(65 - SID_SHIFT_1T)
660 #define VSID_MULINV_1T		ASM_CONST(209034062)
661 
662 /* 1TB VSID reserved for VRMA */
663 #define VRMA_VSID	0x1ffffffUL
664 #define USER_VSID_RANGE	(1UL << (ESID_BITS + SID_SHIFT))
665 
666 /* 4 bits per slice and we have one slice per 1TB */
667 #define SLICE_ARRAY_SIZE	(H_PGTABLE_RANGE >> 41)
668 #define LOW_SLICE_ARRAY_SZ	(BITS_PER_LONG / BITS_PER_BYTE)
669 #define TASK_SLICE_ARRAY_SZ(x)	((x)->hash_context->slb_addr_limit >> 41)
670 #ifndef __ASSEMBLY__
671 
672 #ifdef CONFIG_PPC_SUBPAGE_PROT
673 /*
674  * For the sub-page protection option, we extend the PGD with one of
675  * these.  Basically we have a 3-level tree, with the top level being
676  * the protptrs array.  To optimize speed and memory consumption when
677  * only addresses < 4GB are being protected, pointers to the first
678  * four pages of sub-page protection words are stored in the low_prot
679  * array.
680  * Each page of sub-page protection words protects 1GB (4 bytes
681  * protects 64k).  For the 3-level tree, each page of pointers then
682  * protects 8TB.
683  */
684 struct subpage_prot_table {
685 	unsigned long maxaddr;	/* only addresses < this are protected */
686 	unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
687 	unsigned int *low_prot[4];
688 };
689 
690 #define SBP_L1_BITS		(PAGE_SHIFT - 2)
691 #define SBP_L2_BITS		(PAGE_SHIFT - 3)
692 #define SBP_L1_COUNT		(1 << SBP_L1_BITS)
693 #define SBP_L2_COUNT		(1 << SBP_L2_BITS)
694 #define SBP_L2_SHIFT		(PAGE_SHIFT + SBP_L1_BITS)
695 #define SBP_L3_SHIFT		(SBP_L2_SHIFT + SBP_L2_BITS)
696 
697 extern void subpage_prot_free(struct mm_struct *mm);
698 #else
699 static inline void subpage_prot_free(struct mm_struct *mm) {}
700 #endif /* CONFIG_PPC_SUBPAGE_PROT */
701 
702 /*
703  * One bit per slice. We have lower slices which cover 256MB segments
704  * upto 4G range. That gets us 16 low slices. For the rest we track slices
705  * in 1TB size.
706  */
707 struct slice_mask {
708 	u64 low_slices;
709 	DECLARE_BITMAP(high_slices, SLICE_NUM_HIGH);
710 };
711 
712 struct hash_mm_context {
713 	u16 user_psize; /* page size index */
714 
715 	/* SLB page size encodings*/
716 	unsigned char low_slices_psize[LOW_SLICE_ARRAY_SZ];
717 	unsigned char high_slices_psize[SLICE_ARRAY_SIZE];
718 	unsigned long slb_addr_limit;
719 #ifdef CONFIG_PPC_64K_PAGES
720 	struct slice_mask mask_64k;
721 #endif
722 	struct slice_mask mask_4k;
723 #ifdef CONFIG_HUGETLB_PAGE
724 	struct slice_mask mask_16m;
725 	struct slice_mask mask_16g;
726 #endif
727 
728 #ifdef CONFIG_PPC_SUBPAGE_PROT
729 	struct subpage_prot_table *spt;
730 #endif /* CONFIG_PPC_SUBPAGE_PROT */
731 };
732 
733 #if 0
734 /*
735  * The code below is equivalent to this function for arguments
736  * < 2^VSID_BITS, which is all this should ever be called
737  * with.  However gcc is not clever enough to compute the
738  * modulus (2^n-1) without a second multiply.
739  */
740 #define vsid_scramble(protovsid, size) \
741 	((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
742 
743 /* simplified form avoiding mod operation */
744 #define vsid_scramble(protovsid, size) \
745 	({								 \
746 		unsigned long x;					 \
747 		x = (protovsid) * VSID_MULTIPLIER_##size;		 \
748 		x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
749 		(x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
750 	})
751 
752 #else /* 1 */
753 static inline unsigned long vsid_scramble(unsigned long protovsid,
754 				  unsigned long vsid_multiplier, int vsid_bits)
755 {
756 	unsigned long vsid;
757 	unsigned long vsid_modulus = ((1UL << vsid_bits) - 1);
758 	/*
759 	 * We have same multipler for both 256 and 1T segements now
760 	 */
761 	vsid = protovsid * vsid_multiplier;
762 	vsid = (vsid >> vsid_bits) + (vsid & vsid_modulus);
763 	return (vsid + ((vsid + 1) >> vsid_bits)) & vsid_modulus;
764 }
765 
766 #endif /* 1 */
767 
768 /* Returns the segment size indicator for a user address */
769 static inline int user_segment_size(unsigned long addr)
770 {
771 	/* Use 1T segments if possible for addresses >= 1T */
772 	if (addr >= (1UL << SID_SHIFT_1T))
773 		return mmu_highuser_ssize;
774 	return MMU_SEGSIZE_256M;
775 }
776 
777 static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
778 				     int ssize)
779 {
780 	unsigned long va_bits = VA_BITS;
781 	unsigned long vsid_bits;
782 	unsigned long protovsid;
783 
784 	/*
785 	 * Bad address. We return VSID 0 for that
786 	 */
787 	if ((ea & EA_MASK)  >= H_PGTABLE_RANGE)
788 		return 0;
789 
790 	if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
791 		va_bits = 65;
792 
793 	if (ssize == MMU_SEGSIZE_256M) {
794 		vsid_bits = va_bits - SID_SHIFT;
795 		protovsid = (context << ESID_BITS) |
796 			((ea >> SID_SHIFT) & ESID_BITS_MASK);
797 		return vsid_scramble(protovsid, VSID_MULTIPLIER_256M, vsid_bits);
798 	}
799 	/* 1T segment */
800 	vsid_bits = va_bits - SID_SHIFT_1T;
801 	protovsid = (context << ESID_BITS_1T) |
802 		((ea >> SID_SHIFT_1T) & ESID_BITS_1T_MASK);
803 	return vsid_scramble(protovsid, VSID_MULTIPLIER_1T, vsid_bits);
804 }
805 
806 /*
807  * For kernel space, we use context ids as
808  * below. Range is 512TB per context.
809  *
810  * 0x00001 -  [ 0xc000000000000000 - 0xc001ffffffffffff]
811  * 0x00002 -  [ 0xc002000000000000 - 0xc003ffffffffffff]
812  * 0x00003 -  [ 0xc004000000000000 - 0xc005ffffffffffff]
813  * 0x00004 -  [ 0xc006000000000000 - 0xc007ffffffffffff]
814  *
815  * vmap, IO, vmemap
816  *
817  * 0x00005 -  [ 0xc008000000000000 - 0xc009ffffffffffff]
818  * 0x00006 -  [ 0xc00a000000000000 - 0xc00bffffffffffff]
819  * 0x00007 -  [ 0xc00c000000000000 - 0xc00dffffffffffff]
820  *
821  */
822 static inline unsigned long get_kernel_context(unsigned long ea)
823 {
824 	unsigned long region_id = get_region_id(ea);
825 	unsigned long ctx;
826 	/*
827 	 * Depending on Kernel config, kernel region can have one context
828 	 * or more.
829 	 */
830 	if (region_id == LINEAR_MAP_REGION_ID) {
831 		/*
832 		 * We already verified ea to be not beyond the addr limit.
833 		 */
834 		ctx =  1 + ((ea & EA_MASK) >> MAX_EA_BITS_PER_CONTEXT);
835 	} else
836 		ctx = region_id + MAX_KERNEL_CTX_CNT - 1;
837 	return ctx;
838 }
839 
840 /*
841  * This is only valid for addresses >= PAGE_OFFSET
842  */
843 static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
844 {
845 	unsigned long context;
846 
847 	if (!is_kernel_addr(ea))
848 		return 0;
849 
850 	context = get_kernel_context(ea);
851 	return get_vsid(context, ea, ssize);
852 }
853 
854 unsigned htab_shift_for_mem_size(unsigned long mem_size);
855 
856 enum slb_index {
857 	LINEAR_INDEX	= 0, /* Kernel linear map  (0xc000000000000000) */
858 	KSTACK_INDEX	= 1, /* Kernel stack map */
859 };
860 
861 #define slb_esid_mask(ssize)	\
862 	(((ssize) == MMU_SEGSIZE_256M) ? ESID_MASK : ESID_MASK_1T)
863 
864 static inline unsigned long mk_esid_data(unsigned long ea, int ssize,
865 					 enum slb_index index)
866 {
867 	return (ea & slb_esid_mask(ssize)) | SLB_ESID_V | index;
868 }
869 
870 static inline unsigned long __mk_vsid_data(unsigned long vsid, int ssize,
871 					   unsigned long flags)
872 {
873 	return (vsid << slb_vsid_shift(ssize)) | flags |
874 		((unsigned long)ssize << SLB_VSID_SSIZE_SHIFT);
875 }
876 
877 static inline unsigned long mk_vsid_data(unsigned long ea, int ssize,
878 					 unsigned long flags)
879 {
880 	return __mk_vsid_data(get_kernel_vsid(ea, ssize), ssize, flags);
881 }
882 
883 #endif /* __ASSEMBLY__ */
884 #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */
885