mmu-hash.h 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661
  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. #include <asm/asm-compat.h>
  15. #include <asm/page.h>
  16. #include <asm/bug.h>
  17. /*
  18. * This is necessary to get the definition of PGTABLE_RANGE which we
  19. * need for various slices related matters. Note that this isn't the
  20. * complete pgtable.h but only a portion of it.
  21. */
  22. #include <asm/book3s/64/pgtable.h>
  23. #include <asm/bug.h>
  24. #include <asm/processor.h>
  25. #include <asm/cpu_has_feature.h>
  26. /*
  27. * SLB
  28. */
  29. #define SLB_NUM_BOLTED 3
  30. #define SLB_CACHE_ENTRIES 8
  31. #define SLB_MIN_SIZE 32
  32. /* Bits in the SLB ESID word */
  33. #define SLB_ESID_V ASM_CONST(0x0000000008000000) /* valid */
  34. /* Bits in the SLB VSID word */
  35. #define SLB_VSID_SHIFT 12
  36. #define SLB_VSID_SHIFT_1T 24
  37. #define SLB_VSID_SSIZE_SHIFT 62
  38. #define SLB_VSID_B ASM_CONST(0xc000000000000000)
  39. #define SLB_VSID_B_256M ASM_CONST(0x0000000000000000)
  40. #define SLB_VSID_B_1T ASM_CONST(0x4000000000000000)
  41. #define SLB_VSID_KS ASM_CONST(0x0000000000000800)
  42. #define SLB_VSID_KP ASM_CONST(0x0000000000000400)
  43. #define SLB_VSID_N ASM_CONST(0x0000000000000200) /* no-execute */
  44. #define SLB_VSID_L ASM_CONST(0x0000000000000100)
  45. #define SLB_VSID_C ASM_CONST(0x0000000000000080) /* class */
  46. #define SLB_VSID_LP ASM_CONST(0x0000000000000030)
  47. #define SLB_VSID_LP_00 ASM_CONST(0x0000000000000000)
  48. #define SLB_VSID_LP_01 ASM_CONST(0x0000000000000010)
  49. #define SLB_VSID_LP_10 ASM_CONST(0x0000000000000020)
  50. #define SLB_VSID_LP_11 ASM_CONST(0x0000000000000030)
  51. #define SLB_VSID_LLP (SLB_VSID_L|SLB_VSID_LP)
  52. #define SLB_VSID_KERNEL (SLB_VSID_KP)
  53. #define SLB_VSID_USER (SLB_VSID_KP|SLB_VSID_KS|SLB_VSID_C)
  54. #define SLBIE_C (0x08000000)
  55. #define SLBIE_SSIZE_SHIFT 25
  56. /*
  57. * Hash table
  58. */
  59. #define HPTES_PER_GROUP 8
  60. #define HPTE_V_SSIZE_SHIFT 62
  61. #define HPTE_V_AVPN_SHIFT 7
  62. #define HPTE_V_AVPN ASM_CONST(0x3fffffffffffff80)
  63. #define HPTE_V_AVPN_VAL(x) (((x) & HPTE_V_AVPN) >> HPTE_V_AVPN_SHIFT)
  64. #define HPTE_V_COMPARE(x,y) (!(((x) ^ (y)) & 0xffffffffffffff80UL))
  65. #define HPTE_V_BOLTED ASM_CONST(0x0000000000000010)
  66. #define HPTE_V_LOCK ASM_CONST(0x0000000000000008)
  67. #define HPTE_V_LARGE ASM_CONST(0x0000000000000004)
  68. #define HPTE_V_SECONDARY ASM_CONST(0x0000000000000002)
  69. #define HPTE_V_VALID ASM_CONST(0x0000000000000001)
  70. /*
  71. * ISA 3.0 have a different HPTE format.
  72. */
  73. #define HPTE_R_3_0_SSIZE_SHIFT 58
  74. #define HPTE_R_PP0 ASM_CONST(0x8000000000000000)
  75. #define HPTE_R_TS ASM_CONST(0x4000000000000000)
  76. #define HPTE_R_KEY_HI ASM_CONST(0x3000000000000000)
  77. #define HPTE_R_RPN_SHIFT 12
  78. #define HPTE_R_RPN ASM_CONST(0x0ffffffffffff000)
  79. #define HPTE_R_PP ASM_CONST(0x0000000000000003)
  80. #define HPTE_R_PPP ASM_CONST(0x8000000000000003)
  81. #define HPTE_R_N ASM_CONST(0x0000000000000004)
  82. #define HPTE_R_G ASM_CONST(0x0000000000000008)
  83. #define HPTE_R_M ASM_CONST(0x0000000000000010)
  84. #define HPTE_R_I ASM_CONST(0x0000000000000020)
  85. #define HPTE_R_W ASM_CONST(0x0000000000000040)
  86. #define HPTE_R_WIMG ASM_CONST(0x0000000000000078)
  87. #define HPTE_R_C ASM_CONST(0x0000000000000080)
  88. #define HPTE_R_R ASM_CONST(0x0000000000000100)
  89. #define HPTE_R_KEY_LO ASM_CONST(0x0000000000000e00)
  90. #define HPTE_V_1TB_SEG ASM_CONST(0x4000000000000000)
  91. #define HPTE_V_VRMA_MASK ASM_CONST(0x4001ffffff000000)
  92. /* Values for PP (assumes Ks=0, Kp=1) */
  93. #define PP_RWXX 0 /* Supervisor read/write, User none */
  94. #define PP_RWRX 1 /* Supervisor read/write, User read */
  95. #define PP_RWRW 2 /* Supervisor read/write, User read/write */
  96. #define PP_RXRX 3 /* Supervisor read, User read */
  97. #define PP_RXXX (HPTE_R_PP0 | 2) /* Supervisor read, user none */
  98. /* Fields for tlbiel instruction in architecture 2.06 */
  99. #define TLBIEL_INVAL_SEL_MASK 0xc00 /* invalidation selector */
  100. #define TLBIEL_INVAL_PAGE 0x000 /* invalidate a single page */
  101. #define TLBIEL_INVAL_SET_LPID 0x800 /* invalidate a set for current LPID */
  102. #define TLBIEL_INVAL_SET 0xc00 /* invalidate a set for all LPIDs */
  103. #define TLBIEL_INVAL_SET_MASK 0xfff000 /* set number to inval. */
  104. #define TLBIEL_INVAL_SET_SHIFT 12
  105. #define POWER7_TLB_SETS 128 /* # sets in POWER7 TLB */
  106. #define POWER8_TLB_SETS 512 /* # sets in POWER8 TLB */
  107. #define POWER9_TLB_SETS_HASH 256 /* # sets in POWER9 TLB Hash mode */
  108. #define POWER9_TLB_SETS_RADIX 128 /* # sets in POWER9 TLB Radix mode */
  109. #ifndef __ASSEMBLY__
  110. struct mmu_hash_ops {
  111. void (*hpte_invalidate)(unsigned long slot,
  112. unsigned long vpn,
  113. int bpsize, int apsize,
  114. int ssize, int local);
  115. long (*hpte_updatepp)(unsigned long slot,
  116. unsigned long newpp,
  117. unsigned long vpn,
  118. int bpsize, int apsize,
  119. int ssize, unsigned long flags);
  120. void (*hpte_updateboltedpp)(unsigned long newpp,
  121. unsigned long ea,
  122. int psize, int ssize);
  123. long (*hpte_insert)(unsigned long hpte_group,
  124. unsigned long vpn,
  125. unsigned long prpn,
  126. unsigned long rflags,
  127. unsigned long vflags,
  128. int psize, int apsize,
  129. int ssize);
  130. long (*hpte_remove)(unsigned long hpte_group);
  131. int (*hpte_removebolted)(unsigned long ea,
  132. int psize, int ssize);
  133. void (*flush_hash_range)(unsigned long number, int local);
  134. void (*hugepage_invalidate)(unsigned long vsid,
  135. unsigned long addr,
  136. unsigned char *hpte_slot_array,
  137. int psize, int ssize, int local);
  138. /*
  139. * Special for kexec.
  140. * To be called in real mode with interrupts disabled. No locks are
  141. * taken as such, concurrent access on pre POWER5 hardware could result
  142. * in a deadlock.
  143. * The linear mapping is destroyed as well.
  144. */
  145. void (*hpte_clear_all)(void);
  146. };
  147. extern struct mmu_hash_ops mmu_hash_ops;
  148. struct hash_pte {
  149. __be64 v;
  150. __be64 r;
  151. };
  152. extern struct hash_pte *htab_address;
  153. extern unsigned long htab_size_bytes;
  154. extern unsigned long htab_hash_mask;
  155. static inline int shift_to_mmu_psize(unsigned int shift)
  156. {
  157. int psize;
  158. for (psize = 0; psize < MMU_PAGE_COUNT; ++psize)
  159. if (mmu_psize_defs[psize].shift == shift)
  160. return psize;
  161. return -1;
  162. }
  163. static inline unsigned int mmu_psize_to_shift(unsigned int mmu_psize)
  164. {
  165. if (mmu_psize_defs[mmu_psize].shift)
  166. return mmu_psize_defs[mmu_psize].shift;
  167. BUG();
  168. }
  169. static inline unsigned long get_sllp_encoding(int psize)
  170. {
  171. unsigned long sllp;
  172. sllp = ((mmu_psize_defs[psize].sllp & SLB_VSID_L) >> 6) |
  173. ((mmu_psize_defs[psize].sllp & SLB_VSID_LP) >> 4);
  174. return sllp;
  175. }
  176. #endif /* __ASSEMBLY__ */
  177. /*
  178. * Segment sizes.
  179. * These are the values used by hardware in the B field of
  180. * SLB entries and the first dword of MMU hashtable entries.
  181. * The B field is 2 bits; the values 2 and 3 are unused and reserved.
  182. */
  183. #define MMU_SEGSIZE_256M 0
  184. #define MMU_SEGSIZE_1T 1
  185. /*
  186. * encode page number shift.
  187. * in order to fit the 78 bit va in a 64 bit variable we shift the va by
  188. * 12 bits. This enable us to address upto 76 bit va.
  189. * For hpt hash from a va we can ignore the page size bits of va and for
  190. * hpte encoding we ignore up to 23 bits of va. So ignoring lower 12 bits ensure
  191. * we work in all cases including 4k page size.
  192. */
  193. #define VPN_SHIFT 12
  194. /*
  195. * HPTE Large Page (LP) details
  196. */
  197. #define LP_SHIFT 12
  198. #define LP_BITS 8
  199. #define LP_MASK(i) ((0xFF >> (i)) << LP_SHIFT)
  200. #ifndef __ASSEMBLY__
  201. static inline int slb_vsid_shift(int ssize)
  202. {
  203. if (ssize == MMU_SEGSIZE_256M)
  204. return SLB_VSID_SHIFT;
  205. return SLB_VSID_SHIFT_1T;
  206. }
  207. static inline int segment_shift(int ssize)
  208. {
  209. if (ssize == MMU_SEGSIZE_256M)
  210. return SID_SHIFT;
  211. return SID_SHIFT_1T;
  212. }
  213. /*
  214. * This array is indexed by the LP field of the HPTE second dword.
  215. * Since this field may contain some RPN bits, some entries are
  216. * replicated so that we get the same value irrespective of RPN.
  217. * The top 4 bits are the page size index (MMU_PAGE_*) for the
  218. * actual page size, the bottom 4 bits are the base page size.
  219. */
  220. extern u8 hpte_page_sizes[1 << LP_BITS];
  221. static inline unsigned long __hpte_page_size(unsigned long h, unsigned long l,
  222. bool is_base_size)
  223. {
  224. unsigned int i, lp;
  225. if (!(h & HPTE_V_LARGE))
  226. return 1ul << 12;
  227. /* Look at the 8 bit LP value */
  228. lp = (l >> LP_SHIFT) & ((1 << LP_BITS) - 1);
  229. i = hpte_page_sizes[lp];
  230. if (!i)
  231. return 0;
  232. if (!is_base_size)
  233. i >>= 4;
  234. return 1ul << mmu_psize_defs[i & 0xf].shift;
  235. }
  236. static inline unsigned long hpte_page_size(unsigned long h, unsigned long l)
  237. {
  238. return __hpte_page_size(h, l, 0);
  239. }
  240. static inline unsigned long hpte_base_page_size(unsigned long h, unsigned long l)
  241. {
  242. return __hpte_page_size(h, l, 1);
  243. }
  244. /*
  245. * The current system page and segment sizes
  246. */
  247. extern int mmu_kernel_ssize;
  248. extern int mmu_highuser_ssize;
  249. extern u16 mmu_slb_size;
  250. extern unsigned long tce_alloc_start, tce_alloc_end;
  251. /*
  252. * If the processor supports 64k normal pages but not 64k cache
  253. * inhibited pages, we have to be prepared to switch processes
  254. * to use 4k pages when they create cache-inhibited mappings.
  255. * If this is the case, mmu_ci_restrictions will be set to 1.
  256. */
  257. extern int mmu_ci_restrictions;
  258. /*
  259. * This computes the AVPN and B fields of the first dword of a HPTE,
  260. * for use when we want to match an existing PTE. The bottom 7 bits
  261. * of the returned value are zero.
  262. */
  263. static inline unsigned long hpte_encode_avpn(unsigned long vpn, int psize,
  264. int ssize)
  265. {
  266. unsigned long v;
  267. /*
  268. * The AVA field omits the low-order 23 bits of the 78 bits VA.
  269. * These bits are not needed in the PTE, because the
  270. * low-order b of these bits are part of the byte offset
  271. * into the virtual page and, if b < 23, the high-order
  272. * 23-b of these bits are always used in selecting the
  273. * PTEGs to be searched
  274. */
  275. v = (vpn >> (23 - VPN_SHIFT)) & ~(mmu_psize_defs[psize].avpnm);
  276. v <<= HPTE_V_AVPN_SHIFT;
  277. if (!cpu_has_feature(CPU_FTR_ARCH_300))
  278. v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
  279. return v;
  280. }
  281. /*
  282. * This function sets the AVPN and L fields of the HPTE appropriately
  283. * using the base page size and actual page size.
  284. */
  285. static inline unsigned long hpte_encode_v(unsigned long vpn, int base_psize,
  286. int actual_psize, int ssize)
  287. {
  288. unsigned long v;
  289. v = hpte_encode_avpn(vpn, base_psize, ssize);
  290. if (actual_psize != MMU_PAGE_4K)
  291. v |= HPTE_V_LARGE;
  292. return v;
  293. }
  294. /*
  295. * This function sets the ARPN, and LP fields of the HPTE appropriately
  296. * for the page size. We assume the pa is already "clean" that is properly
  297. * aligned for the requested page size
  298. */
  299. static inline unsigned long hpte_encode_r(unsigned long pa, int base_psize,
  300. int actual_psize, int ssize)
  301. {
  302. if (cpu_has_feature(CPU_FTR_ARCH_300))
  303. pa |= ((unsigned long) ssize) << HPTE_R_3_0_SSIZE_SHIFT;
  304. /* A 4K page needs no special encoding */
  305. if (actual_psize == MMU_PAGE_4K)
  306. return pa & HPTE_R_RPN;
  307. else {
  308. unsigned int penc = mmu_psize_defs[base_psize].penc[actual_psize];
  309. unsigned int shift = mmu_psize_defs[actual_psize].shift;
  310. return (pa & ~((1ul << shift) - 1)) | (penc << LP_SHIFT);
  311. }
  312. }
  313. /*
  314. * Build a VPN_SHIFT bit shifted va given VSID, EA and segment size.
  315. */
  316. static inline unsigned long hpt_vpn(unsigned long ea,
  317. unsigned long vsid, int ssize)
  318. {
  319. unsigned long mask;
  320. int s_shift = segment_shift(ssize);
  321. mask = (1ul << (s_shift - VPN_SHIFT)) - 1;
  322. return (vsid << (s_shift - VPN_SHIFT)) | ((ea >> VPN_SHIFT) & mask);
  323. }
  324. /*
  325. * This hashes a virtual address
  326. */
  327. static inline unsigned long hpt_hash(unsigned long vpn,
  328. unsigned int shift, int ssize)
  329. {
  330. int mask;
  331. unsigned long hash, vsid;
  332. /* VPN_SHIFT can be atmost 12 */
  333. if (ssize == MMU_SEGSIZE_256M) {
  334. mask = (1ul << (SID_SHIFT - VPN_SHIFT)) - 1;
  335. hash = (vpn >> (SID_SHIFT - VPN_SHIFT)) ^
  336. ((vpn & mask) >> (shift - VPN_SHIFT));
  337. } else {
  338. mask = (1ul << (SID_SHIFT_1T - VPN_SHIFT)) - 1;
  339. vsid = vpn >> (SID_SHIFT_1T - VPN_SHIFT);
  340. hash = vsid ^ (vsid << 25) ^
  341. ((vpn & mask) >> (shift - VPN_SHIFT)) ;
  342. }
  343. return hash & 0x7fffffffffUL;
  344. }
  345. #define HPTE_LOCAL_UPDATE 0x1
  346. #define HPTE_NOHPTE_UPDATE 0x2
  347. extern int __hash_page_4K(unsigned long ea, unsigned long access,
  348. unsigned long vsid, pte_t *ptep, unsigned long trap,
  349. unsigned long flags, int ssize, int subpage_prot);
  350. extern int __hash_page_64K(unsigned long ea, unsigned long access,
  351. unsigned long vsid, pte_t *ptep, unsigned long trap,
  352. unsigned long flags, int ssize);
  353. struct mm_struct;
  354. unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap);
  355. extern int hash_page_mm(struct mm_struct *mm, unsigned long ea,
  356. unsigned long access, unsigned long trap,
  357. unsigned long flags);
  358. extern int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
  359. unsigned long dsisr);
  360. int __hash_page_huge(unsigned long ea, unsigned long access, unsigned long vsid,
  361. pte_t *ptep, unsigned long trap, unsigned long flags,
  362. int ssize, unsigned int shift, unsigned int mmu_psize);
  363. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  364. extern int __hash_page_thp(unsigned long ea, unsigned long access,
  365. unsigned long vsid, pmd_t *pmdp, unsigned long trap,
  366. unsigned long flags, int ssize, unsigned int psize);
  367. #else
  368. static inline int __hash_page_thp(unsigned long ea, unsigned long access,
  369. unsigned long vsid, pmd_t *pmdp,
  370. unsigned long trap, unsigned long flags,
  371. int ssize, unsigned int psize)
  372. {
  373. BUG();
  374. return -1;
  375. }
  376. #endif
  377. extern void hash_failure_debug(unsigned long ea, unsigned long access,
  378. unsigned long vsid, unsigned long trap,
  379. int ssize, int psize, int lpsize,
  380. unsigned long pte);
  381. extern int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
  382. unsigned long pstart, unsigned long prot,
  383. int psize, int ssize);
  384. int htab_remove_mapping(unsigned long vstart, unsigned long vend,
  385. int psize, int ssize);
  386. extern void add_gpage(u64 addr, u64 page_size, unsigned long number_of_pages);
  387. extern void demote_segment_4k(struct mm_struct *mm, unsigned long addr);
  388. #ifdef CONFIG_PPC_PSERIES
  389. void hpte_init_pseries(void);
  390. #else
  391. static inline void hpte_init_pseries(void) { }
  392. #endif
  393. extern void hpte_init_native(void);
  394. extern void slb_initialize(void);
  395. extern void slb_flush_and_rebolt(void);
  396. extern void slb_vmalloc_update(void);
  397. extern void slb_set_size(u16 size);
  398. #endif /* __ASSEMBLY__ */
  399. /*
  400. * VSID allocation (256MB segment)
  401. *
  402. * We first generate a 37-bit "proto-VSID". Proto-VSIDs are generated
  403. * from mmu context id and effective segment id of the address.
  404. *
  405. * For user processes max context id is limited to ((1ul << 19) - 5)
  406. * for kernel space, we use the top 4 context ids to map address as below
  407. * NOTE: each context only support 64TB now.
  408. * 0x7fffc - [ 0xc000000000000000 - 0xc0003fffffffffff ]
  409. * 0x7fffd - [ 0xd000000000000000 - 0xd0003fffffffffff ]
  410. * 0x7fffe - [ 0xe000000000000000 - 0xe0003fffffffffff ]
  411. * 0x7ffff - [ 0xf000000000000000 - 0xf0003fffffffffff ]
  412. *
  413. * The proto-VSIDs are then scrambled into real VSIDs with the
  414. * multiplicative hash:
  415. *
  416. * VSID = (proto-VSID * VSID_MULTIPLIER) % VSID_MODULUS
  417. *
  418. * VSID_MULTIPLIER is prime, so in particular it is
  419. * co-prime to VSID_MODULUS, making this a 1:1 scrambling function.
  420. * Because the modulus is 2^n-1 we can compute it efficiently without
  421. * a divide or extra multiply (see below). The scramble function gives
  422. * robust scattering in the hash table (at least based on some initial
  423. * results).
  424. *
  425. * We also consider VSID 0 special. We use VSID 0 for slb entries mapping
  426. * bad address. This enables us to consolidate bad address handling in
  427. * hash_page.
  428. *
  429. * We also need to avoid the last segment of the last context, because that
  430. * would give a protovsid of 0x1fffffffff. That will result in a VSID 0
  431. * because of the modulo operation in vsid scramble. But the vmemmap
  432. * (which is what uses region 0xf) will never be close to 64TB in size
  433. * (it's 56 bytes per page of system memory).
  434. */
  435. #define CONTEXT_BITS 19
  436. #define ESID_BITS 18
  437. #define ESID_BITS_1T 6
  438. /*
  439. * 256MB segment
  440. * The proto-VSID space has 2^(CONTEX_BITS + ESID_BITS) - 1 segments
  441. * available for user + kernel mapping. The top 4 contexts are used for
  442. * kernel mapping. Each segment contains 2^28 bytes. Each
  443. * context maps 2^46 bytes (64TB) so we can support 2^19-1 contexts
  444. * (19 == 37 + 28 - 46).
  445. */
  446. #define MAX_USER_CONTEXT ((ASM_CONST(1) << CONTEXT_BITS) - 5)
  447. /*
  448. * This should be computed such that protovosid * vsid_mulitplier
  449. * doesn't overflow 64 bits. It should also be co-prime to vsid_modulus
  450. */
  451. #define VSID_MULTIPLIER_256M ASM_CONST(12538073) /* 24-bit prime */
  452. #define VSID_BITS_256M (CONTEXT_BITS + ESID_BITS)
  453. #define VSID_MODULUS_256M ((1UL<<VSID_BITS_256M)-1)
  454. #define VSID_MULTIPLIER_1T ASM_CONST(12538073) /* 24-bit prime */
  455. #define VSID_BITS_1T (CONTEXT_BITS + ESID_BITS_1T)
  456. #define VSID_MODULUS_1T ((1UL<<VSID_BITS_1T)-1)
  457. #define USER_VSID_RANGE (1UL << (ESID_BITS + SID_SHIFT))
  458. /*
  459. * This macro generates asm code to compute the VSID scramble
  460. * function. Used in slb_allocate() and do_stab_bolted. The function
  461. * computed is: (protovsid*VSID_MULTIPLIER) % VSID_MODULUS
  462. *
  463. * rt = register containing the proto-VSID and into which the
  464. * VSID will be stored
  465. * rx = scratch register (clobbered)
  466. *
  467. * - rt and rx must be different registers
  468. * - The answer will end up in the low VSID_BITS bits of rt. The higher
  469. * bits may contain other garbage, so you may need to mask the
  470. * result.
  471. */
  472. #define ASM_VSID_SCRAMBLE(rt, rx, size) \
  473. lis rx,VSID_MULTIPLIER_##size@h; \
  474. ori rx,rx,VSID_MULTIPLIER_##size@l; \
  475. mulld rt,rt,rx; /* rt = rt * MULTIPLIER */ \
  476. \
  477. srdi rx,rt,VSID_BITS_##size; \
  478. clrldi rt,rt,(64-VSID_BITS_##size); \
  479. add rt,rt,rx; /* add high and low bits */ \
  480. /* NOTE: explanation based on VSID_BITS_##size = 36 \
  481. * Now, r3 == VSID (mod 2^36-1), and lies between 0 and \
  482. * 2^36-1+2^28-1. That in particular means that if r3 >= \
  483. * 2^36-1, then r3+1 has the 2^36 bit set. So, if r3+1 has \
  484. * the bit clear, r3 already has the answer we want, if it \
  485. * doesn't, the answer is the low 36 bits of r3+1. So in all \
  486. * cases the answer is the low 36 bits of (r3 + ((r3+1) >> 36))*/\
  487. addi rx,rt,1; \
  488. srdi rx,rx,VSID_BITS_##size; /* extract 2^VSID_BITS bit */ \
  489. add rt,rt,rx
  490. /* 4 bits per slice and we have one slice per 1TB */
  491. #define SLICE_ARRAY_SIZE (H_PGTABLE_RANGE >> 41)
  492. #ifndef __ASSEMBLY__
  493. #ifdef CONFIG_PPC_SUBPAGE_PROT
  494. /*
  495. * For the sub-page protection option, we extend the PGD with one of
  496. * these. Basically we have a 3-level tree, with the top level being
  497. * the protptrs array. To optimize speed and memory consumption when
  498. * only addresses < 4GB are being protected, pointers to the first
  499. * four pages of sub-page protection words are stored in the low_prot
  500. * array.
  501. * Each page of sub-page protection words protects 1GB (4 bytes
  502. * protects 64k). For the 3-level tree, each page of pointers then
  503. * protects 8TB.
  504. */
  505. struct subpage_prot_table {
  506. unsigned long maxaddr; /* only addresses < this are protected */
  507. unsigned int **protptrs[(TASK_SIZE_USER64 >> 43)];
  508. unsigned int *low_prot[4];
  509. };
  510. #define SBP_L1_BITS (PAGE_SHIFT - 2)
  511. #define SBP_L2_BITS (PAGE_SHIFT - 3)
  512. #define SBP_L1_COUNT (1 << SBP_L1_BITS)
  513. #define SBP_L2_COUNT (1 << SBP_L2_BITS)
  514. #define SBP_L2_SHIFT (PAGE_SHIFT + SBP_L1_BITS)
  515. #define SBP_L3_SHIFT (SBP_L2_SHIFT + SBP_L2_BITS)
  516. extern void subpage_prot_free(struct mm_struct *mm);
  517. extern void subpage_prot_init_new_context(struct mm_struct *mm);
  518. #else
  519. static inline void subpage_prot_free(struct mm_struct *mm) {}
  520. static inline void subpage_prot_init_new_context(struct mm_struct *mm) { }
  521. #endif /* CONFIG_PPC_SUBPAGE_PROT */
  522. #if 0
  523. /*
  524. * The code below is equivalent to this function for arguments
  525. * < 2^VSID_BITS, which is all this should ever be called
  526. * with. However gcc is not clever enough to compute the
  527. * modulus (2^n-1) without a second multiply.
  528. */
  529. #define vsid_scramble(protovsid, size) \
  530. ((((protovsid) * VSID_MULTIPLIER_##size) % VSID_MODULUS_##size))
  531. #else /* 1 */
  532. #define vsid_scramble(protovsid, size) \
  533. ({ \
  534. unsigned long x; \
  535. x = (protovsid) * VSID_MULTIPLIER_##size; \
  536. x = (x >> VSID_BITS_##size) + (x & VSID_MODULUS_##size); \
  537. (x + ((x+1) >> VSID_BITS_##size)) & VSID_MODULUS_##size; \
  538. })
  539. #endif /* 1 */
  540. /* Returns the segment size indicator for a user address */
  541. static inline int user_segment_size(unsigned long addr)
  542. {
  543. /* Use 1T segments if possible for addresses >= 1T */
  544. if (addr >= (1UL << SID_SHIFT_1T))
  545. return mmu_highuser_ssize;
  546. return MMU_SEGSIZE_256M;
  547. }
  548. static inline unsigned long get_vsid(unsigned long context, unsigned long ea,
  549. int ssize)
  550. {
  551. /*
  552. * Bad address. We return VSID 0 for that
  553. */
  554. if ((ea & ~REGION_MASK) >= H_PGTABLE_RANGE)
  555. return 0;
  556. if (ssize == MMU_SEGSIZE_256M)
  557. return vsid_scramble((context << ESID_BITS)
  558. | (ea >> SID_SHIFT), 256M);
  559. return vsid_scramble((context << ESID_BITS_1T)
  560. | (ea >> SID_SHIFT_1T), 1T);
  561. }
  562. /*
  563. * This is only valid for addresses >= PAGE_OFFSET
  564. *
  565. * For kernel space, we use the top 4 context ids to map address as below
  566. * 0x7fffc - [ 0xc000000000000000 - 0xc0003fffffffffff ]
  567. * 0x7fffd - [ 0xd000000000000000 - 0xd0003fffffffffff ]
  568. * 0x7fffe - [ 0xe000000000000000 - 0xe0003fffffffffff ]
  569. * 0x7ffff - [ 0xf000000000000000 - 0xf0003fffffffffff ]
  570. */
  571. static inline unsigned long get_kernel_vsid(unsigned long ea, int ssize)
  572. {
  573. unsigned long context;
  574. /*
  575. * kernel take the top 4 context from the available range
  576. */
  577. context = (MAX_USER_CONTEXT) + ((ea >> 60) - 0xc) + 1;
  578. return get_vsid(context, ea, ssize);
  579. }
  580. unsigned htab_shift_for_mem_size(unsigned long mem_size);
  581. #endif /* __ASSEMBLY__ */
  582. #endif /* _ASM_POWERPC_BOOK3S_64_MMU_HASH_H_ */