lpar.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780
  1. /*
  2. * pSeries_lpar.c
  3. * Copyright (C) 2001 Todd Inglett, IBM Corporation
  4. *
  5. * pSeries LPAR support.
  6. *
  7. * This program is free software; you can redistribute it and/or modify
  8. * it under the terms of the GNU General Public License as published by
  9. * the Free Software Foundation; either version 2 of the License, or
  10. * (at your option) any later version.
  11. *
  12. * This program is distributed in the hope that it will be useful,
  13. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  15. * GNU General Public License for more details.
  16. *
  17. * You should have received a copy of the GNU General Public License
  18. * along with this program; if not, write to the Free Software
  19. * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  20. */
  21. /* Enables debugging of low-level hash table routines - careful! */
  22. #undef DEBUG
  23. #include <linux/kernel.h>
  24. #include <linux/dma-mapping.h>
  25. #include <linux/console.h>
  26. #include <linux/export.h>
  27. #include <linux/static_key.h>
  28. #include <asm/processor.h>
  29. #include <asm/mmu.h>
  30. #include <asm/page.h>
  31. #include <asm/pgtable.h>
  32. #include <asm/machdep.h>
  33. #include <asm/mmu_context.h>
  34. #include <asm/iommu.h>
  35. #include <asm/tlbflush.h>
  36. #include <asm/tlb.h>
  37. #include <asm/prom.h>
  38. #include <asm/cputable.h>
  39. #include <asm/udbg.h>
  40. #include <asm/smp.h>
  41. #include <asm/trace.h>
  42. #include <asm/firmware.h>
  43. #include <asm/plpar_wrappers.h>
  44. #include "pseries.h"
  45. /* Flag bits for H_BULK_REMOVE */
  46. #define HBR_REQUEST 0x4000000000000000UL
  47. #define HBR_RESPONSE 0x8000000000000000UL
  48. #define HBR_END 0xc000000000000000UL
  49. #define HBR_AVPN 0x0200000000000000UL
  50. #define HBR_ANDCOND 0x0100000000000000UL
  51. /* in hvCall.S */
  52. EXPORT_SYMBOL(plpar_hcall);
  53. EXPORT_SYMBOL(plpar_hcall9);
  54. EXPORT_SYMBOL(plpar_hcall_norets);
  55. void vpa_init(int cpu)
  56. {
  57. int hwcpu = get_hard_smp_processor_id(cpu);
  58. unsigned long addr;
  59. long ret;
  60. struct paca_struct *pp;
  61. struct dtl_entry *dtl;
  62. /*
  63. * The spec says it "may be problematic" if CPU x registers the VPA of
  64. * CPU y. We should never do that, but wail if we ever do.
  65. */
  66. WARN_ON(cpu != smp_processor_id());
  67. if (cpu_has_feature(CPU_FTR_ALTIVEC))
  68. lppaca_of(cpu).vmxregs_in_use = 1;
  69. if (cpu_has_feature(CPU_FTR_ARCH_207S))
  70. lppaca_of(cpu).ebb_regs_in_use = 1;
  71. addr = __pa(&lppaca_of(cpu));
  72. ret = register_vpa(hwcpu, addr);
  73. if (ret) {
  74. pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
  75. "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
  76. return;
  77. }
  78. /*
  79. * PAPR says this feature is SLB-Buffer but firmware never
  80. * reports that. All SPLPAR support SLB shadow buffer.
  81. */
  82. addr = __pa(paca[cpu].slb_shadow_ptr);
  83. if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
  84. ret = register_slb_shadow(hwcpu, addr);
  85. if (ret)
  86. pr_err("WARNING: SLB shadow buffer registration for "
  87. "cpu %d (hw %d) of area %lx failed with %ld\n",
  88. cpu, hwcpu, addr, ret);
  89. }
  90. /*
  91. * Register dispatch trace log, if one has been allocated.
  92. */
  93. pp = &paca[cpu];
  94. dtl = pp->dispatch_log;
  95. if (dtl) {
  96. pp->dtl_ridx = 0;
  97. pp->dtl_curr = dtl;
  98. lppaca_of(cpu).dtl_idx = 0;
  99. /* hypervisor reads buffer length from this field */
  100. dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
  101. ret = register_dtl(hwcpu, __pa(dtl));
  102. if (ret)
  103. pr_err("WARNING: DTL registration of cpu %d (hw %d) "
  104. "failed with %ld\n", smp_processor_id(),
  105. hwcpu, ret);
  106. lppaca_of(cpu).dtl_enable_mask = 2;
  107. }
  108. }
  109. static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
  110. unsigned long vpn, unsigned long pa,
  111. unsigned long rflags, unsigned long vflags,
  112. int psize, int apsize, int ssize)
  113. {
  114. unsigned long lpar_rc;
  115. unsigned long flags;
  116. unsigned long slot;
  117. unsigned long hpte_v, hpte_r;
  118. if (!(vflags & HPTE_V_BOLTED))
  119. pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
  120. "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
  121. hpte_group, vpn, pa, rflags, vflags, psize);
  122. hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
  123. hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
  124. if (!(vflags & HPTE_V_BOLTED))
  125. pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
  126. /* Now fill in the actual HPTE */
  127. /* Set CEC cookie to 0 */
  128. /* Zero page = 0 */
  129. /* I-cache Invalidate = 0 */
  130. /* I-cache synchronize = 0 */
  131. /* Exact = 0 */
  132. flags = 0;
  133. /* Make pHyp happy */
  134. if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU))
  135. hpte_r &= ~HPTE_R_M;
  136. if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
  137. flags |= H_COALESCE_CAND;
  138. lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
  139. if (unlikely(lpar_rc == H_PTEG_FULL)) {
  140. if (!(vflags & HPTE_V_BOLTED))
  141. pr_devel(" full\n");
  142. return -1;
  143. }
  144. /*
  145. * Since we try and ioremap PHBs we don't own, the pte insert
  146. * will fail. However we must catch the failure in hash_page
  147. * or we will loop forever, so return -2 in this case.
  148. */
  149. if (unlikely(lpar_rc != H_SUCCESS)) {
  150. if (!(vflags & HPTE_V_BOLTED))
  151. pr_devel(" lpar err %ld\n", lpar_rc);
  152. return -2;
  153. }
  154. if (!(vflags & HPTE_V_BOLTED))
  155. pr_devel(" -> slot: %lu\n", slot & 7);
  156. /* Because of iSeries, we have to pass down the secondary
  157. * bucket bit here as well
  158. */
  159. return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
  160. }
  161. static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
  162. static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
  163. {
  164. unsigned long slot_offset;
  165. unsigned long lpar_rc;
  166. int i;
  167. unsigned long dummy1, dummy2;
  168. /* pick a random slot to start at */
  169. slot_offset = mftb() & 0x7;
  170. for (i = 0; i < HPTES_PER_GROUP; i++) {
  171. /* don't remove a bolted entry */
  172. lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
  173. (0x1UL << 4), &dummy1, &dummy2);
  174. if (lpar_rc == H_SUCCESS)
  175. return i;
  176. /*
  177. * The test for adjunct partition is performed before the
  178. * ANDCOND test. H_RESOURCE may be returned, so we need to
  179. * check for that as well.
  180. */
  181. BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
  182. slot_offset++;
  183. slot_offset &= 0x7;
  184. }
  185. return -1;
  186. }
  187. static void pSeries_lpar_hptab_clear(void)
  188. {
  189. unsigned long size_bytes = 1UL << ppc64_pft_size;
  190. unsigned long hpte_count = size_bytes >> 4;
  191. struct {
  192. unsigned long pteh;
  193. unsigned long ptel;
  194. } ptes[4];
  195. long lpar_rc;
  196. unsigned long i, j;
  197. /* Read in batches of 4,
  198. * invalidate only valid entries not in the VRMA
  199. * hpte_count will be a multiple of 4
  200. */
  201. for (i = 0; i < hpte_count; i += 4) {
  202. lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
  203. if (lpar_rc != H_SUCCESS)
  204. continue;
  205. for (j = 0; j < 4; j++){
  206. if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
  207. HPTE_V_VRMA_MASK)
  208. continue;
  209. if (ptes[j].pteh & HPTE_V_VALID)
  210. plpar_pte_remove_raw(0, i + j, 0,
  211. &(ptes[j].pteh), &(ptes[j].ptel));
  212. }
  213. }
  214. #ifdef __LITTLE_ENDIAN__
  215. /* Reset exceptions to big endian */
  216. if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
  217. long rc;
  218. rc = pseries_big_endian_exceptions();
  219. /*
  220. * At this point it is unlikely panic() will get anything
  221. * out to the user, but at least this will stop us from
  222. * continuing on further and creating an even more
  223. * difficult to debug situation.
  224. */
  225. if (rc)
  226. panic("Could not enable big endian exceptions");
  227. }
  228. #endif
  229. }
  230. /*
  231. * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
  232. * the low 3 bits of flags happen to line up. So no transform is needed.
  233. * We can probably optimize here and assume the high bits of newpp are
  234. * already zero. For now I am paranoid.
  235. */
  236. static long pSeries_lpar_hpte_updatepp(unsigned long slot,
  237. unsigned long newpp,
  238. unsigned long vpn,
  239. int psize, int apsize,
  240. int ssize, int local)
  241. {
  242. unsigned long lpar_rc;
  243. unsigned long flags = (newpp & 7) | H_AVPN;
  244. unsigned long want_v;
  245. want_v = hpte_encode_avpn(vpn, psize, ssize);
  246. pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
  247. want_v, slot, flags, psize);
  248. lpar_rc = plpar_pte_protect(flags, slot, want_v);
  249. if (lpar_rc == H_NOT_FOUND) {
  250. pr_devel("not found !\n");
  251. return -1;
  252. }
  253. pr_devel("ok\n");
  254. BUG_ON(lpar_rc != H_SUCCESS);
  255. return 0;
  256. }
  257. static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
  258. {
  259. unsigned long dword0;
  260. unsigned long lpar_rc;
  261. unsigned long dummy_word1;
  262. unsigned long flags;
  263. /* Read 1 pte at a time */
  264. /* Do not need RPN to logical page translation */
  265. /* No cross CEC PFT access */
  266. flags = 0;
  267. lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
  268. BUG_ON(lpar_rc != H_SUCCESS);
  269. return dword0;
  270. }
  271. static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
  272. {
  273. unsigned long hash;
  274. unsigned long i;
  275. long slot;
  276. unsigned long want_v, hpte_v;
  277. hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
  278. want_v = hpte_encode_avpn(vpn, psize, ssize);
  279. /* Bolted entries are always in the primary group */
  280. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  281. for (i = 0; i < HPTES_PER_GROUP; i++) {
  282. hpte_v = pSeries_lpar_hpte_getword0(slot);
  283. if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
  284. /* HPTE matches */
  285. return slot;
  286. ++slot;
  287. }
  288. return -1;
  289. }
  290. static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
  291. unsigned long ea,
  292. int psize, int ssize)
  293. {
  294. unsigned long vpn;
  295. unsigned long lpar_rc, slot, vsid, flags;
  296. vsid = get_kernel_vsid(ea, ssize);
  297. vpn = hpt_vpn(ea, vsid, ssize);
  298. slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
  299. BUG_ON(slot == -1);
  300. flags = newpp & 7;
  301. lpar_rc = plpar_pte_protect(flags, slot, 0);
  302. BUG_ON(lpar_rc != H_SUCCESS);
  303. }
  304. static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
  305. int psize, int apsize,
  306. int ssize, int local)
  307. {
  308. unsigned long want_v;
  309. unsigned long lpar_rc;
  310. unsigned long dummy1, dummy2;
  311. pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
  312. slot, vpn, psize, local);
  313. want_v = hpte_encode_avpn(vpn, psize, ssize);
  314. lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
  315. if (lpar_rc == H_NOT_FOUND)
  316. return;
  317. BUG_ON(lpar_rc != H_SUCCESS);
  318. }
  319. /*
  320. * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
  321. * to make sure that we avoid bouncing the hypervisor tlbie lock.
  322. */
  323. #define PPC64_HUGE_HPTE_BATCH 12
  324. static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
  325. unsigned long *vpn, int count,
  326. int psize, int ssize)
  327. {
  328. unsigned long param[8];
  329. int i = 0, pix = 0, rc;
  330. unsigned long flags = 0;
  331. int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
  332. if (lock_tlbie)
  333. spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
  334. for (i = 0; i < count; i++) {
  335. if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
  336. pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
  337. ssize, 0);
  338. } else {
  339. param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
  340. param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
  341. pix += 2;
  342. if (pix == 8) {
  343. rc = plpar_hcall9(H_BULK_REMOVE, param,
  344. param[0], param[1], param[2],
  345. param[3], param[4], param[5],
  346. param[6], param[7]);
  347. BUG_ON(rc != H_SUCCESS);
  348. pix = 0;
  349. }
  350. }
  351. }
  352. if (pix) {
  353. param[pix] = HBR_END;
  354. rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
  355. param[2], param[3], param[4], param[5],
  356. param[6], param[7]);
  357. BUG_ON(rc != H_SUCCESS);
  358. }
  359. if (lock_tlbie)
  360. spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
  361. }
  362. static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
  363. unsigned long addr,
  364. unsigned char *hpte_slot_array,
  365. int psize, int ssize)
  366. {
  367. int i, index = 0;
  368. unsigned long s_addr = addr;
  369. unsigned int max_hpte_count, valid;
  370. unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
  371. unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
  372. unsigned long shift, hidx, vpn = 0, hash, slot;
  373. shift = mmu_psize_defs[psize].shift;
  374. max_hpte_count = 1U << (PMD_SHIFT - shift);
  375. for (i = 0; i < max_hpte_count; i++) {
  376. valid = hpte_valid(hpte_slot_array, i);
  377. if (!valid)
  378. continue;
  379. hidx = hpte_hash_index(hpte_slot_array, i);
  380. /* get the vpn */
  381. addr = s_addr + (i * (1ul << shift));
  382. vpn = hpt_vpn(addr, vsid, ssize);
  383. hash = hpt_hash(vpn, shift, ssize);
  384. if (hidx & _PTEIDX_SECONDARY)
  385. hash = ~hash;
  386. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  387. slot += hidx & _PTEIDX_GROUP_IX;
  388. slot_array[index] = slot;
  389. vpn_array[index] = vpn;
  390. if (index == PPC64_HUGE_HPTE_BATCH - 1) {
  391. /*
  392. * Now do a bluk invalidate
  393. */
  394. __pSeries_lpar_hugepage_invalidate(slot_array,
  395. vpn_array,
  396. PPC64_HUGE_HPTE_BATCH,
  397. psize, ssize);
  398. index = 0;
  399. } else
  400. index++;
  401. }
  402. if (index)
  403. __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
  404. index, psize, ssize);
  405. }
  406. static void pSeries_lpar_hpte_removebolted(unsigned long ea,
  407. int psize, int ssize)
  408. {
  409. unsigned long vpn;
  410. unsigned long slot, vsid;
  411. vsid = get_kernel_vsid(ea, ssize);
  412. vpn = hpt_vpn(ea, vsid, ssize);
  413. slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
  414. BUG_ON(slot == -1);
  415. /*
  416. * lpar doesn't use the passed actual page size
  417. */
  418. pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
  419. }
  420. /*
  421. * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
  422. * lock.
  423. */
  424. static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
  425. {
  426. unsigned long vpn;
  427. unsigned long i, pix, rc;
  428. unsigned long flags = 0;
  429. struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
  430. int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
  431. unsigned long param[9];
  432. unsigned long hash, index, shift, hidx, slot;
  433. real_pte_t pte;
  434. int psize, ssize;
  435. if (lock_tlbie)
  436. spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
  437. psize = batch->psize;
  438. ssize = batch->ssize;
  439. pix = 0;
  440. for (i = 0; i < number; i++) {
  441. vpn = batch->vpn[i];
  442. pte = batch->pte[i];
  443. pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
  444. hash = hpt_hash(vpn, shift, ssize);
  445. hidx = __rpte_to_hidx(pte, index);
  446. if (hidx & _PTEIDX_SECONDARY)
  447. hash = ~hash;
  448. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  449. slot += hidx & _PTEIDX_GROUP_IX;
  450. if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
  451. /*
  452. * lpar doesn't use the passed actual page size
  453. */
  454. pSeries_lpar_hpte_invalidate(slot, vpn, psize,
  455. 0, ssize, local);
  456. } else {
  457. param[pix] = HBR_REQUEST | HBR_AVPN | slot;
  458. param[pix+1] = hpte_encode_avpn(vpn, psize,
  459. ssize);
  460. pix += 2;
  461. if (pix == 8) {
  462. rc = plpar_hcall9(H_BULK_REMOVE, param,
  463. param[0], param[1], param[2],
  464. param[3], param[4], param[5],
  465. param[6], param[7]);
  466. BUG_ON(rc != H_SUCCESS);
  467. pix = 0;
  468. }
  469. }
  470. } pte_iterate_hashed_end();
  471. }
  472. if (pix) {
  473. param[pix] = HBR_END;
  474. rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
  475. param[2], param[3], param[4], param[5],
  476. param[6], param[7]);
  477. BUG_ON(rc != H_SUCCESS);
  478. }
  479. if (lock_tlbie)
  480. spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
  481. }
  482. static int __init disable_bulk_remove(char *str)
  483. {
  484. if (strcmp(str, "off") == 0 &&
  485. firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
  486. printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
  487. powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
  488. }
  489. return 1;
  490. }
  491. __setup("bulk_remove=", disable_bulk_remove);
  492. void __init hpte_init_lpar(void)
  493. {
  494. ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
  495. ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
  496. ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
  497. ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
  498. ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
  499. ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
  500. ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
  501. ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
  502. ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
  503. }
  504. #ifdef CONFIG_PPC_SMLPAR
  505. #define CMO_FREE_HINT_DEFAULT 1
  506. static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
  507. static int __init cmo_free_hint(char *str)
  508. {
  509. char *parm;
  510. parm = strstrip(str);
  511. if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
  512. printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
  513. cmo_free_hint_flag = 0;
  514. return 1;
  515. }
  516. cmo_free_hint_flag = 1;
  517. printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
  518. if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
  519. return 1;
  520. return 0;
  521. }
  522. __setup("cmo_free_hint=", cmo_free_hint);
  523. static void pSeries_set_page_state(struct page *page, int order,
  524. unsigned long state)
  525. {
  526. int i, j;
  527. unsigned long cmo_page_sz, addr;
  528. cmo_page_sz = cmo_get_page_size();
  529. addr = __pa((unsigned long)page_address(page));
  530. for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
  531. for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
  532. plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
  533. }
  534. }
  535. void arch_free_page(struct page *page, int order)
  536. {
  537. if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
  538. return;
  539. pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
  540. }
  541. EXPORT_SYMBOL(arch_free_page);
  542. #endif
  543. #ifdef CONFIG_TRACEPOINTS
  544. #ifdef HAVE_JUMP_LABEL
  545. struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
  546. void hcall_tracepoint_regfunc(void)
  547. {
  548. static_key_slow_inc(&hcall_tracepoint_key);
  549. }
  550. void hcall_tracepoint_unregfunc(void)
  551. {
  552. static_key_slow_dec(&hcall_tracepoint_key);
  553. }
  554. #else
  555. /*
  556. * We optimise our hcall path by placing hcall_tracepoint_refcount
  557. * directly in the TOC so we can check if the hcall tracepoints are
  558. * enabled via a single load.
  559. */
  560. /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
  561. extern long hcall_tracepoint_refcount;
  562. void hcall_tracepoint_regfunc(void)
  563. {
  564. hcall_tracepoint_refcount++;
  565. }
  566. void hcall_tracepoint_unregfunc(void)
  567. {
  568. hcall_tracepoint_refcount--;
  569. }
  570. #endif
  571. /*
  572. * Since the tracing code might execute hcalls we need to guard against
  573. * recursion. One example of this are spinlocks calling H_YIELD on
  574. * shared processor partitions.
  575. */
  576. static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
  577. void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
  578. {
  579. unsigned long flags;
  580. unsigned int *depth;
  581. /*
  582. * We cannot call tracepoints inside RCU idle regions which
  583. * means we must not trace H_CEDE.
  584. */
  585. if (opcode == H_CEDE)
  586. return;
  587. local_irq_save(flags);
  588. depth = &__get_cpu_var(hcall_trace_depth);
  589. if (*depth)
  590. goto out;
  591. (*depth)++;
  592. preempt_disable();
  593. trace_hcall_entry(opcode, args);
  594. (*depth)--;
  595. out:
  596. local_irq_restore(flags);
  597. }
  598. void __trace_hcall_exit(long opcode, unsigned long retval,
  599. unsigned long *retbuf)
  600. {
  601. unsigned long flags;
  602. unsigned int *depth;
  603. if (opcode == H_CEDE)
  604. return;
  605. local_irq_save(flags);
  606. depth = &__get_cpu_var(hcall_trace_depth);
  607. if (*depth)
  608. goto out;
  609. (*depth)++;
  610. trace_hcall_exit(opcode, retval, retbuf);
  611. preempt_enable();
  612. (*depth)--;
  613. out:
  614. local_irq_restore(flags);
  615. }
  616. #endif
  617. /**
  618. * h_get_mpp
  619. * H_GET_MPP hcall returns info in 7 parms
  620. */
  621. int h_get_mpp(struct hvcall_mpp_data *mpp_data)
  622. {
  623. int rc;
  624. unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
  625. rc = plpar_hcall9(H_GET_MPP, retbuf);
  626. mpp_data->entitled_mem = retbuf[0];
  627. mpp_data->mapped_mem = retbuf[1];
  628. mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
  629. mpp_data->pool_num = retbuf[2] & 0xffff;
  630. mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
  631. mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
  632. mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
  633. mpp_data->pool_size = retbuf[4];
  634. mpp_data->loan_request = retbuf[5];
  635. mpp_data->backing_mem = retbuf[6];
  636. return rc;
  637. }
  638. EXPORT_SYMBOL(h_get_mpp);
  639. int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
  640. {
  641. int rc;
  642. unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
  643. rc = plpar_hcall9(H_GET_MPP_X, retbuf);
  644. mpp_x_data->coalesced_bytes = retbuf[0];
  645. mpp_x_data->pool_coalesced_bytes = retbuf[1];
  646. mpp_x_data->pool_purr_cycles = retbuf[2];
  647. mpp_x_data->pool_spurr_cycles = retbuf[3];
  648. return rc;
  649. }