hash_native_64.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735
  1. /*
  2. * native hashtable management.
  3. *
  4. * SMP scalability work:
  5. * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
  6. *
  7. * This program is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU General Public License
  9. * as published by the Free Software Foundation; either version
  10. * 2 of the License, or (at your option) any later version.
  11. */
  12. #undef DEBUG_LOW
  13. #include <linux/spinlock.h>
  14. #include <linux/bitops.h>
  15. #include <linux/of.h>
  16. #include <linux/threads.h>
  17. #include <linux/smp.h>
  18. #include <asm/machdep.h>
  19. #include <asm/mmu.h>
  20. #include <asm/mmu_context.h>
  21. #include <asm/pgtable.h>
  22. #include <asm/tlbflush.h>
  23. #include <asm/tlb.h>
  24. #include <asm/cputable.h>
  25. #include <asm/udbg.h>
  26. #include <asm/kexec.h>
  27. #include <asm/ppc-opcode.h>
  28. #include <misc/cxl-base.h>
  29. #ifdef DEBUG_LOW
  30. #define DBG_LOW(fmt...) udbg_printf(fmt)
  31. #else
  32. #define DBG_LOW(fmt...)
  33. #endif
  34. #ifdef __BIG_ENDIAN__
  35. #define HPTE_LOCK_BIT 3
  36. #else
  37. #define HPTE_LOCK_BIT (56+3)
  38. #endif
  39. DEFINE_RAW_SPINLOCK(native_tlbie_lock);
  40. static inline void __tlbie(unsigned long vpn, int psize, int apsize, int ssize)
  41. {
  42. unsigned long va;
  43. unsigned int penc;
  44. unsigned long sllp;
  45. /*
  46. * We need 14 to 65 bits of va for a tlibe of 4K page
  47. * With vpn we ignore the lower VPN_SHIFT bits already.
  48. * And top two bits are already ignored because we can
  49. * only accomodate 76 bits in a 64 bit vpn with a VPN_SHIFT
  50. * of 12.
  51. */
  52. va = vpn << VPN_SHIFT;
  53. /*
  54. * clear top 16 bits of 64bit va, non SLS segment
  55. * Older versions of the architecture (2.02 and earler) require the
  56. * masking of the top 16 bits.
  57. */
  58. if (mmu_has_feature(MMU_FTR_TLBIE_CROP_VA))
  59. va &= ~(0xffffULL << 48);
  60. switch (psize) {
  61. case MMU_PAGE_4K:
  62. /* clear out bits after (52) [0....52.....63] */
  63. va &= ~((1ul << (64 - 52)) - 1);
  64. va |= ssize << 8;
  65. sllp = get_sllp_encoding(apsize);
  66. va |= sllp << 5;
  67. asm volatile(ASM_FTR_IFCLR("tlbie %0,0", PPC_TLBIE(%1,%0), %2)
  68. : : "r" (va), "r"(0), "i" (CPU_FTR_ARCH_206)
  69. : "memory");
  70. break;
  71. default:
  72. /* We need 14 to 14 + i bits of va */
  73. penc = mmu_psize_defs[psize].penc[apsize];
  74. va &= ~((1ul << mmu_psize_defs[apsize].shift) - 1);
  75. va |= penc << 12;
  76. va |= ssize << 8;
  77. /*
  78. * AVAL bits:
  79. * We don't need all the bits, but rest of the bits
  80. * must be ignored by the processor.
  81. * vpn cover upto 65 bits of va. (0...65) and we need
  82. * 58..64 bits of va.
  83. */
  84. va |= (vpn & 0xfe); /* AVAL */
  85. va |= 1; /* L */
  86. asm volatile(ASM_FTR_IFCLR("tlbie %0,1", PPC_TLBIE(%1,%0), %2)
  87. : : "r" (va), "r"(0), "i" (CPU_FTR_ARCH_206)
  88. : "memory");
  89. break;
  90. }
  91. }
  92. static inline void __tlbiel(unsigned long vpn, int psize, int apsize, int ssize)
  93. {
  94. unsigned long va;
  95. unsigned int penc;
  96. unsigned long sllp;
  97. /* VPN_SHIFT can be atmost 12 */
  98. va = vpn << VPN_SHIFT;
  99. /*
  100. * clear top 16 bits of 64 bit va, non SLS segment
  101. * Older versions of the architecture (2.02 and earler) require the
  102. * masking of the top 16 bits.
  103. */
  104. if (mmu_has_feature(MMU_FTR_TLBIE_CROP_VA))
  105. va &= ~(0xffffULL << 48);
  106. switch (psize) {
  107. case MMU_PAGE_4K:
  108. /* clear out bits after(52) [0....52.....63] */
  109. va &= ~((1ul << (64 - 52)) - 1);
  110. va |= ssize << 8;
  111. sllp = get_sllp_encoding(apsize);
  112. va |= sllp << 5;
  113. asm volatile(".long 0x7c000224 | (%0 << 11) | (0 << 21)"
  114. : : "r"(va) : "memory");
  115. break;
  116. default:
  117. /* We need 14 to 14 + i bits of va */
  118. penc = mmu_psize_defs[psize].penc[apsize];
  119. va &= ~((1ul << mmu_psize_defs[apsize].shift) - 1);
  120. va |= penc << 12;
  121. va |= ssize << 8;
  122. /*
  123. * AVAL bits:
  124. * We don't need all the bits, but rest of the bits
  125. * must be ignored by the processor.
  126. * vpn cover upto 65 bits of va. (0...65) and we need
  127. * 58..64 bits of va.
  128. */
  129. va |= (vpn & 0xfe);
  130. va |= 1; /* L */
  131. asm volatile(".long 0x7c000224 | (%0 << 11) | (1 << 21)"
  132. : : "r"(va) : "memory");
  133. break;
  134. }
  135. }
  136. static inline void tlbie(unsigned long vpn, int psize, int apsize,
  137. int ssize, int local)
  138. {
  139. unsigned int use_local;
  140. int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
  141. use_local = local && mmu_has_feature(MMU_FTR_TLBIEL) && !cxl_ctx_in_use();
  142. if (use_local)
  143. use_local = mmu_psize_defs[psize].tlbiel;
  144. if (lock_tlbie && !use_local)
  145. raw_spin_lock(&native_tlbie_lock);
  146. asm volatile("ptesync": : :"memory");
  147. if (use_local) {
  148. __tlbiel(vpn, psize, apsize, ssize);
  149. asm volatile("ptesync": : :"memory");
  150. } else {
  151. __tlbie(vpn, psize, apsize, ssize);
  152. asm volatile("eieio; tlbsync; ptesync": : :"memory");
  153. }
  154. if (lock_tlbie && !use_local)
  155. raw_spin_unlock(&native_tlbie_lock);
  156. }
  157. static inline void native_lock_hpte(struct hash_pte *hptep)
  158. {
  159. unsigned long *word = (unsigned long *)&hptep->v;
  160. while (1) {
  161. if (!test_and_set_bit_lock(HPTE_LOCK_BIT, word))
  162. break;
  163. while(test_bit(HPTE_LOCK_BIT, word))
  164. cpu_relax();
  165. }
  166. }
  167. static inline void native_unlock_hpte(struct hash_pte *hptep)
  168. {
  169. unsigned long *word = (unsigned long *)&hptep->v;
  170. clear_bit_unlock(HPTE_LOCK_BIT, word);
  171. }
  172. static long native_hpte_insert(unsigned long hpte_group, unsigned long vpn,
  173. unsigned long pa, unsigned long rflags,
  174. unsigned long vflags, int psize, int apsize, int ssize)
  175. {
  176. struct hash_pte *hptep = htab_address + hpte_group;
  177. unsigned long hpte_v, hpte_r;
  178. int i;
  179. if (!(vflags & HPTE_V_BOLTED)) {
  180. DBG_LOW(" insert(group=%lx, vpn=%016lx, pa=%016lx,"
  181. " rflags=%lx, vflags=%lx, psize=%d)\n",
  182. hpte_group, vpn, pa, rflags, vflags, psize);
  183. }
  184. for (i = 0; i < HPTES_PER_GROUP; i++) {
  185. if (! (be64_to_cpu(hptep->v) & HPTE_V_VALID)) {
  186. /* retry with lock held */
  187. native_lock_hpte(hptep);
  188. if (! (be64_to_cpu(hptep->v) & HPTE_V_VALID))
  189. break;
  190. native_unlock_hpte(hptep);
  191. }
  192. hptep++;
  193. }
  194. if (i == HPTES_PER_GROUP)
  195. return -1;
  196. hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
  197. hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
  198. if (!(vflags & HPTE_V_BOLTED)) {
  199. DBG_LOW(" i=%x hpte_v=%016lx, hpte_r=%016lx\n",
  200. i, hpte_v, hpte_r);
  201. }
  202. if (cpu_has_feature(CPU_FTR_ARCH_300)) {
  203. hpte_r = hpte_old_to_new_r(hpte_v, hpte_r);
  204. hpte_v = hpte_old_to_new_v(hpte_v);
  205. }
  206. hptep->r = cpu_to_be64(hpte_r);
  207. /* Guarantee the second dword is visible before the valid bit */
  208. eieio();
  209. /*
  210. * Now set the first dword including the valid bit
  211. * NOTE: this also unlocks the hpte
  212. */
  213. hptep->v = cpu_to_be64(hpte_v);
  214. __asm__ __volatile__ ("ptesync" : : : "memory");
  215. return i | (!!(vflags & HPTE_V_SECONDARY) << 3);
  216. }
  217. static long native_hpte_remove(unsigned long hpte_group)
  218. {
  219. struct hash_pte *hptep;
  220. int i;
  221. int slot_offset;
  222. unsigned long hpte_v;
  223. DBG_LOW(" remove(group=%lx)\n", hpte_group);
  224. /* pick a random entry to start at */
  225. slot_offset = mftb() & 0x7;
  226. for (i = 0; i < HPTES_PER_GROUP; i++) {
  227. hptep = htab_address + hpte_group + slot_offset;
  228. hpte_v = be64_to_cpu(hptep->v);
  229. if ((hpte_v & HPTE_V_VALID) && !(hpte_v & HPTE_V_BOLTED)) {
  230. /* retry with lock held */
  231. native_lock_hpte(hptep);
  232. hpte_v = be64_to_cpu(hptep->v);
  233. if ((hpte_v & HPTE_V_VALID)
  234. && !(hpte_v & HPTE_V_BOLTED))
  235. break;
  236. native_unlock_hpte(hptep);
  237. }
  238. slot_offset++;
  239. slot_offset &= 0x7;
  240. }
  241. if (i == HPTES_PER_GROUP)
  242. return -1;
  243. /* Invalidate the hpte. NOTE: this also unlocks it */
  244. hptep->v = 0;
  245. return i;
  246. }
  247. static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
  248. unsigned long vpn, int bpsize,
  249. int apsize, int ssize, unsigned long flags)
  250. {
  251. struct hash_pte *hptep = htab_address + slot;
  252. unsigned long hpte_v, want_v;
  253. int ret = 0, local = 0;
  254. want_v = hpte_encode_avpn(vpn, bpsize, ssize);
  255. DBG_LOW(" update(vpn=%016lx, avpnv=%016lx, group=%lx, newpp=%lx)",
  256. vpn, want_v & HPTE_V_AVPN, slot, newpp);
  257. hpte_v = be64_to_cpu(hptep->v);
  258. if (cpu_has_feature(CPU_FTR_ARCH_300))
  259. hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r));
  260. /*
  261. * We need to invalidate the TLB always because hpte_remove doesn't do
  262. * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
  263. * random entry from it. When we do that we don't invalidate the TLB
  264. * (hpte_remove) because we assume the old translation is still
  265. * technically "valid".
  266. */
  267. if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID)) {
  268. DBG_LOW(" -> miss\n");
  269. ret = -1;
  270. } else {
  271. native_lock_hpte(hptep);
  272. /* recheck with locks held */
  273. hpte_v = be64_to_cpu(hptep->v);
  274. if (cpu_has_feature(CPU_FTR_ARCH_300))
  275. hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r));
  276. if (unlikely(!HPTE_V_COMPARE(hpte_v, want_v) ||
  277. !(hpte_v & HPTE_V_VALID))) {
  278. ret = -1;
  279. } else {
  280. DBG_LOW(" -> hit\n");
  281. /* Update the HPTE */
  282. hptep->r = cpu_to_be64((be64_to_cpu(hptep->r) &
  283. ~(HPTE_R_PPP | HPTE_R_N)) |
  284. (newpp & (HPTE_R_PPP | HPTE_R_N |
  285. HPTE_R_C)));
  286. }
  287. native_unlock_hpte(hptep);
  288. }
  289. if (flags & HPTE_LOCAL_UPDATE)
  290. local = 1;
  291. /*
  292. * Ensure it is out of the tlb too if it is not a nohpte fault
  293. */
  294. if (!(flags & HPTE_NOHPTE_UPDATE))
  295. tlbie(vpn, bpsize, apsize, ssize, local);
  296. return ret;
  297. }
  298. static long native_hpte_find(unsigned long vpn, int psize, int ssize)
  299. {
  300. struct hash_pte *hptep;
  301. unsigned long hash;
  302. unsigned long i;
  303. long slot;
  304. unsigned long want_v, hpte_v;
  305. hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
  306. want_v = hpte_encode_avpn(vpn, psize, ssize);
  307. /* Bolted mappings are only ever in the primary group */
  308. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  309. for (i = 0; i < HPTES_PER_GROUP; i++) {
  310. hptep = htab_address + slot;
  311. hpte_v = be64_to_cpu(hptep->v);
  312. if (cpu_has_feature(CPU_FTR_ARCH_300))
  313. hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r));
  314. if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
  315. /* HPTE matches */
  316. return slot;
  317. ++slot;
  318. }
  319. return -1;
  320. }
  321. /*
  322. * Update the page protection bits. Intended to be used to create
  323. * guard pages for kernel data structures on pages which are bolted
  324. * in the HPT. Assumes pages being operated on will not be stolen.
  325. *
  326. * No need to lock here because we should be the only user.
  327. */
  328. static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
  329. int psize, int ssize)
  330. {
  331. unsigned long vpn;
  332. unsigned long vsid;
  333. long slot;
  334. struct hash_pte *hptep;
  335. vsid = get_kernel_vsid(ea, ssize);
  336. vpn = hpt_vpn(ea, vsid, ssize);
  337. slot = native_hpte_find(vpn, psize, ssize);
  338. if (slot == -1)
  339. panic("could not find page to bolt\n");
  340. hptep = htab_address + slot;
  341. /* Update the HPTE */
  342. hptep->r = cpu_to_be64((be64_to_cpu(hptep->r) &
  343. ~(HPTE_R_PPP | HPTE_R_N)) |
  344. (newpp & (HPTE_R_PPP | HPTE_R_N)));
  345. /*
  346. * Ensure it is out of the tlb too. Bolted entries base and
  347. * actual page size will be same.
  348. */
  349. tlbie(vpn, psize, psize, ssize, 0);
  350. }
  351. static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
  352. int bpsize, int apsize, int ssize, int local)
  353. {
  354. struct hash_pte *hptep = htab_address + slot;
  355. unsigned long hpte_v;
  356. unsigned long want_v;
  357. unsigned long flags;
  358. local_irq_save(flags);
  359. DBG_LOW(" invalidate(vpn=%016lx, hash: %lx)\n", vpn, slot);
  360. want_v = hpte_encode_avpn(vpn, bpsize, ssize);
  361. native_lock_hpte(hptep);
  362. hpte_v = be64_to_cpu(hptep->v);
  363. if (cpu_has_feature(CPU_FTR_ARCH_300))
  364. hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r));
  365. /*
  366. * We need to invalidate the TLB always because hpte_remove doesn't do
  367. * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
  368. * random entry from it. When we do that we don't invalidate the TLB
  369. * (hpte_remove) because we assume the old translation is still
  370. * technically "valid".
  371. */
  372. if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
  373. native_unlock_hpte(hptep);
  374. else
  375. /* Invalidate the hpte. NOTE: this also unlocks it */
  376. hptep->v = 0;
  377. /* Invalidate the TLB */
  378. tlbie(vpn, bpsize, apsize, ssize, local);
  379. local_irq_restore(flags);
  380. }
  381. #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  382. static void native_hugepage_invalidate(unsigned long vsid,
  383. unsigned long addr,
  384. unsigned char *hpte_slot_array,
  385. int psize, int ssize, int local)
  386. {
  387. int i;
  388. struct hash_pte *hptep;
  389. int actual_psize = MMU_PAGE_16M;
  390. unsigned int max_hpte_count, valid;
  391. unsigned long flags, s_addr = addr;
  392. unsigned long hpte_v, want_v, shift;
  393. unsigned long hidx, vpn = 0, hash, slot;
  394. shift = mmu_psize_defs[psize].shift;
  395. max_hpte_count = 1U << (PMD_SHIFT - shift);
  396. local_irq_save(flags);
  397. for (i = 0; i < max_hpte_count; i++) {
  398. valid = hpte_valid(hpte_slot_array, i);
  399. if (!valid)
  400. continue;
  401. hidx = hpte_hash_index(hpte_slot_array, i);
  402. /* get the vpn */
  403. addr = s_addr + (i * (1ul << shift));
  404. vpn = hpt_vpn(addr, vsid, ssize);
  405. hash = hpt_hash(vpn, shift, ssize);
  406. if (hidx & _PTEIDX_SECONDARY)
  407. hash = ~hash;
  408. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  409. slot += hidx & _PTEIDX_GROUP_IX;
  410. hptep = htab_address + slot;
  411. want_v = hpte_encode_avpn(vpn, psize, ssize);
  412. native_lock_hpte(hptep);
  413. hpte_v = be64_to_cpu(hptep->v);
  414. if (cpu_has_feature(CPU_FTR_ARCH_300))
  415. hpte_v = hpte_new_to_old_v(hpte_v, be64_to_cpu(hptep->r));
  416. /* Even if we miss, we need to invalidate the TLB */
  417. if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
  418. native_unlock_hpte(hptep);
  419. else
  420. /* Invalidate the hpte. NOTE: this also unlocks it */
  421. hptep->v = 0;
  422. /*
  423. * We need to do tlb invalidate for all the address, tlbie
  424. * instruction compares entry_VA in tlb with the VA specified
  425. * here
  426. */
  427. tlbie(vpn, psize, actual_psize, ssize, local);
  428. }
  429. local_irq_restore(flags);
  430. }
  431. #else
  432. static void native_hugepage_invalidate(unsigned long vsid,
  433. unsigned long addr,
  434. unsigned char *hpte_slot_array,
  435. int psize, int ssize, int local)
  436. {
  437. WARN(1, "%s called without THP support\n", __func__);
  438. }
  439. #endif
  440. static void hpte_decode(struct hash_pte *hpte, unsigned long slot,
  441. int *psize, int *apsize, int *ssize, unsigned long *vpn)
  442. {
  443. unsigned long avpn, pteg, vpi;
  444. unsigned long hpte_v = be64_to_cpu(hpte->v);
  445. unsigned long hpte_r = be64_to_cpu(hpte->r);
  446. unsigned long vsid, seg_off;
  447. int size, a_size, shift;
  448. /* Look at the 8 bit LP value */
  449. unsigned int lp = (hpte_r >> LP_SHIFT) & ((1 << LP_BITS) - 1);
  450. if (cpu_has_feature(CPU_FTR_ARCH_300)) {
  451. hpte_v = hpte_new_to_old_v(hpte_v, hpte_r);
  452. hpte_r = hpte_new_to_old_r(hpte_r);
  453. }
  454. if (!(hpte_v & HPTE_V_LARGE)) {
  455. size = MMU_PAGE_4K;
  456. a_size = MMU_PAGE_4K;
  457. } else {
  458. size = hpte_page_sizes[lp] & 0xf;
  459. a_size = hpte_page_sizes[lp] >> 4;
  460. }
  461. /* This works for all page sizes, and for 256M and 1T segments */
  462. *ssize = hpte_v >> HPTE_V_SSIZE_SHIFT;
  463. shift = mmu_psize_defs[size].shift;
  464. avpn = (HPTE_V_AVPN_VAL(hpte_v) & ~mmu_psize_defs[size].avpnm);
  465. pteg = slot / HPTES_PER_GROUP;
  466. if (hpte_v & HPTE_V_SECONDARY)
  467. pteg = ~pteg;
  468. switch (*ssize) {
  469. case MMU_SEGSIZE_256M:
  470. /* We only have 28 - 23 bits of seg_off in avpn */
  471. seg_off = (avpn & 0x1f) << 23;
  472. vsid = avpn >> 5;
  473. /* We can find more bits from the pteg value */
  474. if (shift < 23) {
  475. vpi = (vsid ^ pteg) & htab_hash_mask;
  476. seg_off |= vpi << shift;
  477. }
  478. *vpn = vsid << (SID_SHIFT - VPN_SHIFT) | seg_off >> VPN_SHIFT;
  479. break;
  480. case MMU_SEGSIZE_1T:
  481. /* We only have 40 - 23 bits of seg_off in avpn */
  482. seg_off = (avpn & 0x1ffff) << 23;
  483. vsid = avpn >> 17;
  484. if (shift < 23) {
  485. vpi = (vsid ^ (vsid << 25) ^ pteg) & htab_hash_mask;
  486. seg_off |= vpi << shift;
  487. }
  488. *vpn = vsid << (SID_SHIFT_1T - VPN_SHIFT) | seg_off >> VPN_SHIFT;
  489. break;
  490. default:
  491. *vpn = size = 0;
  492. }
  493. *psize = size;
  494. *apsize = a_size;
  495. }
  496. /*
  497. * clear all mappings on kexec. All cpus are in real mode (or they will
  498. * be when they isi), and we are the only one left. We rely on our kernel
  499. * mapping being 0xC0's and the hardware ignoring those two real bits.
  500. *
  501. * This must be called with interrupts disabled.
  502. *
  503. * Taking the native_tlbie_lock is unsafe here due to the possibility of
  504. * lockdep being on. On pre POWER5 hardware, not taking the lock could
  505. * cause deadlock. POWER5 and newer not taking the lock is fine. This only
  506. * gets called during boot before secondary CPUs have come up and during
  507. * crashdump and all bets are off anyway.
  508. *
  509. * TODO: add batching support when enabled. remember, no dynamic memory here,
  510. * although there is the control page available...
  511. */
  512. static void native_hpte_clear(void)
  513. {
  514. unsigned long vpn = 0;
  515. unsigned long slot, slots;
  516. struct hash_pte *hptep = htab_address;
  517. unsigned long hpte_v;
  518. unsigned long pteg_count;
  519. int psize, apsize, ssize;
  520. pteg_count = htab_hash_mask + 1;
  521. slots = pteg_count * HPTES_PER_GROUP;
  522. for (slot = 0; slot < slots; slot++, hptep++) {
  523. /*
  524. * we could lock the pte here, but we are the only cpu
  525. * running, right? and for crash dump, we probably
  526. * don't want to wait for a maybe bad cpu.
  527. */
  528. hpte_v = be64_to_cpu(hptep->v);
  529. /*
  530. * Call __tlbie() here rather than tlbie() since we can't take the
  531. * native_tlbie_lock.
  532. */
  533. if (hpte_v & HPTE_V_VALID) {
  534. hpte_decode(hptep, slot, &psize, &apsize, &ssize, &vpn);
  535. hptep->v = 0;
  536. __tlbie(vpn, psize, apsize, ssize);
  537. }
  538. }
  539. asm volatile("eieio; tlbsync; ptesync":::"memory");
  540. }
  541. /*
  542. * Batched hash table flush, we batch the tlbie's to avoid taking/releasing
  543. * the lock all the time
  544. */
  545. static void native_flush_hash_range(unsigned long number, int local)
  546. {
  547. unsigned long vpn;
  548. unsigned long hash, index, hidx, shift, slot;
  549. struct hash_pte *hptep;
  550. unsigned long hpte_v;
  551. unsigned long want_v;
  552. unsigned long flags;
  553. real_pte_t pte;
  554. struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
  555. unsigned long psize = batch->psize;
  556. int ssize = batch->ssize;
  557. int i;
  558. unsigned int use_local;
  559. use_local = local && mmu_has_feature(MMU_FTR_TLBIEL) &&
  560. mmu_psize_defs[psize].tlbiel && !cxl_ctx_in_use();
  561. local_irq_save(flags);
  562. for (i = 0; i < number; i++) {
  563. vpn = batch->vpn[i];
  564. pte = batch->pte[i];
  565. pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
  566. hash = hpt_hash(vpn, shift, ssize);
  567. hidx = __rpte_to_hidx(pte, index);
  568. if (hidx & _PTEIDX_SECONDARY)
  569. hash = ~hash;
  570. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  571. slot += hidx & _PTEIDX_GROUP_IX;
  572. hptep = htab_address + slot;
  573. want_v = hpte_encode_avpn(vpn, psize, ssize);
  574. native_lock_hpte(hptep);
  575. hpte_v = be64_to_cpu(hptep->v);
  576. if (cpu_has_feature(CPU_FTR_ARCH_300))
  577. hpte_v = hpte_new_to_old_v(hpte_v,
  578. be64_to_cpu(hptep->r));
  579. if (!HPTE_V_COMPARE(hpte_v, want_v) ||
  580. !(hpte_v & HPTE_V_VALID))
  581. native_unlock_hpte(hptep);
  582. else
  583. hptep->v = 0;
  584. } pte_iterate_hashed_end();
  585. }
  586. if (use_local) {
  587. asm volatile("ptesync":::"memory");
  588. for (i = 0; i < number; i++) {
  589. vpn = batch->vpn[i];
  590. pte = batch->pte[i];
  591. pte_iterate_hashed_subpages(pte, psize,
  592. vpn, index, shift) {
  593. __tlbiel(vpn, psize, psize, ssize);
  594. } pte_iterate_hashed_end();
  595. }
  596. asm volatile("ptesync":::"memory");
  597. } else {
  598. int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
  599. if (lock_tlbie)
  600. raw_spin_lock(&native_tlbie_lock);
  601. asm volatile("ptesync":::"memory");
  602. for (i = 0; i < number; i++) {
  603. vpn = batch->vpn[i];
  604. pte = batch->pte[i];
  605. pte_iterate_hashed_subpages(pte, psize,
  606. vpn, index, shift) {
  607. __tlbie(vpn, psize, psize, ssize);
  608. } pte_iterate_hashed_end();
  609. }
  610. asm volatile("eieio; tlbsync; ptesync":::"memory");
  611. if (lock_tlbie)
  612. raw_spin_unlock(&native_tlbie_lock);
  613. }
  614. local_irq_restore(flags);
  615. }
  616. static int native_register_proc_table(unsigned long base, unsigned long page_size,
  617. unsigned long table_size)
  618. {
  619. unsigned long patb1 = base << 25; /* VSID */
  620. patb1 |= (page_size << 5); /* sllp */
  621. patb1 |= table_size;
  622. partition_tb->patb1 = cpu_to_be64(patb1);
  623. return 0;
  624. }
  625. void __init hpte_init_native(void)
  626. {
  627. mmu_hash_ops.hpte_invalidate = native_hpte_invalidate;
  628. mmu_hash_ops.hpte_updatepp = native_hpte_updatepp;
  629. mmu_hash_ops.hpte_updateboltedpp = native_hpte_updateboltedpp;
  630. mmu_hash_ops.hpte_insert = native_hpte_insert;
  631. mmu_hash_ops.hpte_remove = native_hpte_remove;
  632. mmu_hash_ops.hpte_clear_all = native_hpte_clear;
  633. mmu_hash_ops.flush_hash_range = native_flush_hash_range;
  634. mmu_hash_ops.hugepage_invalidate = native_hugepage_invalidate;
  635. if (cpu_has_feature(CPU_FTR_ARCH_300))
  636. register_process_table = native_register_proc_table;
  637. }