lpar.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747
  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 <asm/processor.h>
  28. #include <asm/mmu.h>
  29. #include <asm/page.h>
  30. #include <asm/pgtable.h>
  31. #include <asm/machdep.h>
  32. #include <asm/mmu_context.h>
  33. #include <asm/iommu.h>
  34. #include <asm/tlbflush.h>
  35. #include <asm/tlb.h>
  36. #include <asm/prom.h>
  37. #include <asm/cputable.h>
  38. #include <asm/udbg.h>
  39. #include <asm/smp.h>
  40. #include <asm/trace.h>
  41. #include <asm/firmware.h>
  42. #include "plpar_wrappers.h"
  43. #include "pseries.h"
  44. /* Flag bits for H_BULK_REMOVE */
  45. #define HBR_REQUEST 0x4000000000000000UL
  46. #define HBR_RESPONSE 0x8000000000000000UL
  47. #define HBR_END 0xc000000000000000UL
  48. #define HBR_AVPN 0x0200000000000000UL
  49. #define HBR_ANDCOND 0x0100000000000000UL
  50. /* in hvCall.S */
  51. EXPORT_SYMBOL(plpar_hcall);
  52. EXPORT_SYMBOL(plpar_hcall9);
  53. EXPORT_SYMBOL(plpar_hcall_norets);
  54. extern void pSeries_find_serial_port(void);
  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. if (cpu_has_feature(CPU_FTR_ALTIVEC))
  63. lppaca_of(cpu).vmxregs_in_use = 1;
  64. addr = __pa(&lppaca_of(cpu));
  65. ret = register_vpa(hwcpu, addr);
  66. if (ret) {
  67. pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
  68. "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
  69. return;
  70. }
  71. /*
  72. * PAPR says this feature is SLB-Buffer but firmware never
  73. * reports that. All SPLPAR support SLB shadow buffer.
  74. */
  75. addr = __pa(&slb_shadow[cpu]);
  76. if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
  77. ret = register_slb_shadow(hwcpu, addr);
  78. if (ret)
  79. pr_err("WARNING: SLB shadow buffer registration for "
  80. "cpu %d (hw %d) of area %lx failed with %ld\n",
  81. cpu, hwcpu, addr, ret);
  82. }
  83. /*
  84. * Register dispatch trace log, if one has been allocated.
  85. */
  86. pp = &paca[cpu];
  87. dtl = pp->dispatch_log;
  88. if (dtl) {
  89. pp->dtl_ridx = 0;
  90. pp->dtl_curr = dtl;
  91. lppaca_of(cpu).dtl_idx = 0;
  92. /* hypervisor reads buffer length from this field */
  93. dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
  94. ret = register_dtl(hwcpu, __pa(dtl));
  95. if (ret)
  96. pr_err("WARNING: DTL registration of cpu %d (hw %d) "
  97. "failed with %ld\n", smp_processor_id(),
  98. hwcpu, ret);
  99. lppaca_of(cpu).dtl_enable_mask = 2;
  100. }
  101. }
  102. static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
  103. unsigned long vpn, unsigned long pa,
  104. unsigned long rflags, unsigned long vflags,
  105. int psize, int apsize, int ssize)
  106. {
  107. unsigned long lpar_rc;
  108. unsigned long flags;
  109. unsigned long slot;
  110. unsigned long hpte_v, hpte_r;
  111. if (!(vflags & HPTE_V_BOLTED))
  112. pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
  113. "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
  114. hpte_group, vpn, pa, rflags, vflags, psize);
  115. hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
  116. hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
  117. if (!(vflags & HPTE_V_BOLTED))
  118. pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
  119. /* Now fill in the actual HPTE */
  120. /* Set CEC cookie to 0 */
  121. /* Zero page = 0 */
  122. /* I-cache Invalidate = 0 */
  123. /* I-cache synchronize = 0 */
  124. /* Exact = 0 */
  125. flags = 0;
  126. /* Make pHyp happy */
  127. if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
  128. hpte_r &= ~_PAGE_COHERENT;
  129. if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
  130. flags |= H_COALESCE_CAND;
  131. lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
  132. if (unlikely(lpar_rc == H_PTEG_FULL)) {
  133. if (!(vflags & HPTE_V_BOLTED))
  134. pr_devel(" full\n");
  135. return -1;
  136. }
  137. /*
  138. * Since we try and ioremap PHBs we don't own, the pte insert
  139. * will fail. However we must catch the failure in hash_page
  140. * or we will loop forever, so return -2 in this case.
  141. */
  142. if (unlikely(lpar_rc != H_SUCCESS)) {
  143. if (!(vflags & HPTE_V_BOLTED))
  144. pr_devel(" lpar err %ld\n", lpar_rc);
  145. return -2;
  146. }
  147. if (!(vflags & HPTE_V_BOLTED))
  148. pr_devel(" -> slot: %lu\n", slot & 7);
  149. /* Because of iSeries, we have to pass down the secondary
  150. * bucket bit here as well
  151. */
  152. return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
  153. }
  154. static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
  155. static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
  156. {
  157. unsigned long slot_offset;
  158. unsigned long lpar_rc;
  159. int i;
  160. unsigned long dummy1, dummy2;
  161. /* pick a random slot to start at */
  162. slot_offset = mftb() & 0x7;
  163. for (i = 0; i < HPTES_PER_GROUP; i++) {
  164. /* don't remove a bolted entry */
  165. lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
  166. (0x1UL << 4), &dummy1, &dummy2);
  167. if (lpar_rc == H_SUCCESS)
  168. return i;
  169. /*
  170. * The test for adjunct partition is performed before the
  171. * ANDCOND test. H_RESOURCE may be returned, so we need to
  172. * check for that as well.
  173. */
  174. BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
  175. slot_offset++;
  176. slot_offset &= 0x7;
  177. }
  178. return -1;
  179. }
  180. static void pSeries_lpar_hptab_clear(void)
  181. {
  182. unsigned long size_bytes = 1UL << ppc64_pft_size;
  183. unsigned long hpte_count = size_bytes >> 4;
  184. struct {
  185. unsigned long pteh;
  186. unsigned long ptel;
  187. } ptes[4];
  188. long lpar_rc;
  189. unsigned long i, j;
  190. /* Read in batches of 4,
  191. * invalidate only valid entries not in the VRMA
  192. * hpte_count will be a multiple of 4
  193. */
  194. for (i = 0; i < hpte_count; i += 4) {
  195. lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
  196. if (lpar_rc != H_SUCCESS)
  197. continue;
  198. for (j = 0; j < 4; j++){
  199. if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
  200. HPTE_V_VRMA_MASK)
  201. continue;
  202. if (ptes[j].pteh & HPTE_V_VALID)
  203. plpar_pte_remove_raw(0, i + j, 0,
  204. &(ptes[j].pteh), &(ptes[j].ptel));
  205. }
  206. }
  207. }
  208. /*
  209. * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
  210. * the low 3 bits of flags happen to line up. So no transform is needed.
  211. * We can probably optimize here and assume the high bits of newpp are
  212. * already zero. For now I am paranoid.
  213. */
  214. static long pSeries_lpar_hpte_updatepp(unsigned long slot,
  215. unsigned long newpp,
  216. unsigned long vpn,
  217. int psize, int apsize,
  218. int ssize, int local)
  219. {
  220. unsigned long lpar_rc;
  221. unsigned long flags = (newpp & 7) | H_AVPN;
  222. unsigned long want_v;
  223. want_v = hpte_encode_avpn(vpn, psize, ssize);
  224. pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
  225. want_v, slot, flags, psize);
  226. lpar_rc = plpar_pte_protect(flags, slot, want_v);
  227. if (lpar_rc == H_NOT_FOUND) {
  228. pr_devel("not found !\n");
  229. return -1;
  230. }
  231. pr_devel("ok\n");
  232. BUG_ON(lpar_rc != H_SUCCESS);
  233. return 0;
  234. }
  235. static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
  236. {
  237. unsigned long dword0;
  238. unsigned long lpar_rc;
  239. unsigned long dummy_word1;
  240. unsigned long flags;
  241. /* Read 1 pte at a time */
  242. /* Do not need RPN to logical page translation */
  243. /* No cross CEC PFT access */
  244. flags = 0;
  245. lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
  246. BUG_ON(lpar_rc != H_SUCCESS);
  247. return dword0;
  248. }
  249. static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
  250. {
  251. unsigned long hash;
  252. unsigned long i;
  253. long slot;
  254. unsigned long want_v, hpte_v;
  255. hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
  256. want_v = hpte_encode_avpn(vpn, psize, ssize);
  257. /* Bolted entries are always in the primary group */
  258. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  259. for (i = 0; i < HPTES_PER_GROUP; i++) {
  260. hpte_v = pSeries_lpar_hpte_getword0(slot);
  261. if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
  262. /* HPTE matches */
  263. return slot;
  264. ++slot;
  265. }
  266. return -1;
  267. }
  268. static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
  269. unsigned long ea,
  270. int psize, int ssize)
  271. {
  272. unsigned long vpn;
  273. unsigned long lpar_rc, slot, vsid, flags;
  274. vsid = get_kernel_vsid(ea, ssize);
  275. vpn = hpt_vpn(ea, vsid, ssize);
  276. slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
  277. BUG_ON(slot == -1);
  278. flags = newpp & 7;
  279. lpar_rc = plpar_pte_protect(flags, slot, 0);
  280. BUG_ON(lpar_rc != H_SUCCESS);
  281. }
  282. static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
  283. int psize, int apsize,
  284. int ssize, int local)
  285. {
  286. unsigned long want_v;
  287. unsigned long lpar_rc;
  288. unsigned long dummy1, dummy2;
  289. pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
  290. slot, vpn, psize, local);
  291. want_v = hpte_encode_avpn(vpn, psize, ssize);
  292. lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
  293. if (lpar_rc == H_NOT_FOUND)
  294. return;
  295. BUG_ON(lpar_rc != H_SUCCESS);
  296. }
  297. /*
  298. * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
  299. * to make sure that we avoid bouncing the hypervisor tlbie lock.
  300. */
  301. #define PPC64_HUGE_HPTE_BATCH 12
  302. static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
  303. unsigned long *vpn, int count,
  304. int psize, int ssize)
  305. {
  306. unsigned long param[8];
  307. int i = 0, pix = 0, rc;
  308. unsigned long flags = 0;
  309. int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
  310. if (lock_tlbie)
  311. spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
  312. for (i = 0; i < count; i++) {
  313. if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
  314. pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
  315. ssize, 0);
  316. } else {
  317. param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
  318. param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
  319. pix += 2;
  320. if (pix == 8) {
  321. rc = plpar_hcall9(H_BULK_REMOVE, param,
  322. param[0], param[1], param[2],
  323. param[3], param[4], param[5],
  324. param[6], param[7]);
  325. BUG_ON(rc != H_SUCCESS);
  326. pix = 0;
  327. }
  328. }
  329. }
  330. if (pix) {
  331. param[pix] = HBR_END;
  332. rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
  333. param[2], param[3], param[4], param[5],
  334. param[6], param[7]);
  335. BUG_ON(rc != H_SUCCESS);
  336. }
  337. if (lock_tlbie)
  338. spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
  339. }
  340. static void pSeries_lpar_hugepage_invalidate(struct mm_struct *mm,
  341. unsigned char *hpte_slot_array,
  342. unsigned long addr, int psize)
  343. {
  344. int ssize = 0, i, index = 0;
  345. unsigned long s_addr = addr;
  346. unsigned int max_hpte_count, valid;
  347. unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
  348. unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
  349. unsigned long shift, hidx, vpn = 0, vsid, hash, slot;
  350. shift = mmu_psize_defs[psize].shift;
  351. max_hpte_count = 1U << (PMD_SHIFT - shift);
  352. for (i = 0; i < max_hpte_count; i++) {
  353. valid = hpte_valid(hpte_slot_array, i);
  354. if (!valid)
  355. continue;
  356. hidx = hpte_hash_index(hpte_slot_array, i);
  357. /* get the vpn */
  358. addr = s_addr + (i * (1ul << shift));
  359. if (!is_kernel_addr(addr)) {
  360. ssize = user_segment_size(addr);
  361. vsid = get_vsid(mm->context.id, addr, ssize);
  362. WARN_ON(vsid == 0);
  363. } else {
  364. vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
  365. ssize = mmu_kernel_ssize;
  366. }
  367. vpn = hpt_vpn(addr, vsid, ssize);
  368. hash = hpt_hash(vpn, shift, ssize);
  369. if (hidx & _PTEIDX_SECONDARY)
  370. hash = ~hash;
  371. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  372. slot += hidx & _PTEIDX_GROUP_IX;
  373. slot_array[index] = slot;
  374. vpn_array[index] = vpn;
  375. if (index == PPC64_HUGE_HPTE_BATCH - 1) {
  376. /*
  377. * Now do a bluk invalidate
  378. */
  379. __pSeries_lpar_hugepage_invalidate(slot_array,
  380. vpn_array,
  381. PPC64_HUGE_HPTE_BATCH,
  382. psize, ssize);
  383. index = 0;
  384. } else
  385. index++;
  386. }
  387. if (index)
  388. __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
  389. index, psize, ssize);
  390. }
  391. static void pSeries_lpar_hpte_removebolted(unsigned long ea,
  392. int psize, int ssize)
  393. {
  394. unsigned long vpn;
  395. unsigned long slot, vsid;
  396. vsid = get_kernel_vsid(ea, ssize);
  397. vpn = hpt_vpn(ea, vsid, ssize);
  398. slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
  399. BUG_ON(slot == -1);
  400. /*
  401. * lpar doesn't use the passed actual page size
  402. */
  403. pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
  404. }
  405. /*
  406. * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
  407. * lock.
  408. */
  409. static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
  410. {
  411. unsigned long vpn;
  412. unsigned long i, pix, rc;
  413. unsigned long flags = 0;
  414. struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
  415. int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
  416. unsigned long param[9];
  417. unsigned long hash, index, shift, hidx, slot;
  418. real_pte_t pte;
  419. int psize, ssize;
  420. if (lock_tlbie)
  421. spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
  422. psize = batch->psize;
  423. ssize = batch->ssize;
  424. pix = 0;
  425. for (i = 0; i < number; i++) {
  426. vpn = batch->vpn[i];
  427. pte = batch->pte[i];
  428. pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
  429. hash = hpt_hash(vpn, shift, ssize);
  430. hidx = __rpte_to_hidx(pte, index);
  431. if (hidx & _PTEIDX_SECONDARY)
  432. hash = ~hash;
  433. slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
  434. slot += hidx & _PTEIDX_GROUP_IX;
  435. if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
  436. /*
  437. * lpar doesn't use the passed actual page size
  438. */
  439. pSeries_lpar_hpte_invalidate(slot, vpn, psize,
  440. 0, ssize, local);
  441. } else {
  442. param[pix] = HBR_REQUEST | HBR_AVPN | slot;
  443. param[pix+1] = hpte_encode_avpn(vpn, psize,
  444. ssize);
  445. pix += 2;
  446. if (pix == 8) {
  447. rc = plpar_hcall9(H_BULK_REMOVE, param,
  448. param[0], param[1], param[2],
  449. param[3], param[4], param[5],
  450. param[6], param[7]);
  451. BUG_ON(rc != H_SUCCESS);
  452. pix = 0;
  453. }
  454. }
  455. } pte_iterate_hashed_end();
  456. }
  457. if (pix) {
  458. param[pix] = HBR_END;
  459. rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
  460. param[2], param[3], param[4], param[5],
  461. param[6], param[7]);
  462. BUG_ON(rc != H_SUCCESS);
  463. }
  464. if (lock_tlbie)
  465. spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
  466. }
  467. static int __init disable_bulk_remove(char *str)
  468. {
  469. if (strcmp(str, "off") == 0 &&
  470. firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
  471. printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
  472. powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
  473. }
  474. return 1;
  475. }
  476. __setup("bulk_remove=", disable_bulk_remove);
  477. void __init hpte_init_lpar(void)
  478. {
  479. ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
  480. ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
  481. ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
  482. ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
  483. ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
  484. ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
  485. ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
  486. ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
  487. ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
  488. }
  489. #ifdef CONFIG_PPC_SMLPAR
  490. #define CMO_FREE_HINT_DEFAULT 1
  491. static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
  492. static int __init cmo_free_hint(char *str)
  493. {
  494. char *parm;
  495. parm = strstrip(str);
  496. if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
  497. printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
  498. cmo_free_hint_flag = 0;
  499. return 1;
  500. }
  501. cmo_free_hint_flag = 1;
  502. printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
  503. if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
  504. return 1;
  505. return 0;
  506. }
  507. __setup("cmo_free_hint=", cmo_free_hint);
  508. static void pSeries_set_page_state(struct page *page, int order,
  509. unsigned long state)
  510. {
  511. int i, j;
  512. unsigned long cmo_page_sz, addr;
  513. cmo_page_sz = cmo_get_page_size();
  514. addr = __pa((unsigned long)page_address(page));
  515. for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
  516. for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
  517. plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
  518. }
  519. }
  520. void arch_free_page(struct page *page, int order)
  521. {
  522. if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
  523. return;
  524. pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
  525. }
  526. EXPORT_SYMBOL(arch_free_page);
  527. #endif
  528. #ifdef CONFIG_TRACEPOINTS
  529. /*
  530. * We optimise our hcall path by placing hcall_tracepoint_refcount
  531. * directly in the TOC so we can check if the hcall tracepoints are
  532. * enabled via a single load.
  533. */
  534. /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
  535. extern long hcall_tracepoint_refcount;
  536. /*
  537. * Since the tracing code might execute hcalls we need to guard against
  538. * recursion. One example of this are spinlocks calling H_YIELD on
  539. * shared processor partitions.
  540. */
  541. static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
  542. void hcall_tracepoint_regfunc(void)
  543. {
  544. hcall_tracepoint_refcount++;
  545. }
  546. void hcall_tracepoint_unregfunc(void)
  547. {
  548. hcall_tracepoint_refcount--;
  549. }
  550. void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
  551. {
  552. unsigned long flags;
  553. unsigned int *depth;
  554. /*
  555. * We cannot call tracepoints inside RCU idle regions which
  556. * means we must not trace H_CEDE.
  557. */
  558. if (opcode == H_CEDE)
  559. return;
  560. local_irq_save(flags);
  561. depth = &__get_cpu_var(hcall_trace_depth);
  562. if (*depth)
  563. goto out;
  564. (*depth)++;
  565. preempt_disable();
  566. trace_hcall_entry(opcode, args);
  567. (*depth)--;
  568. out:
  569. local_irq_restore(flags);
  570. }
  571. void __trace_hcall_exit(long opcode, unsigned long retval,
  572. unsigned long *retbuf)
  573. {
  574. unsigned long flags;
  575. unsigned int *depth;
  576. if (opcode == H_CEDE)
  577. return;
  578. local_irq_save(flags);
  579. depth = &__get_cpu_var(hcall_trace_depth);
  580. if (*depth)
  581. goto out;
  582. (*depth)++;
  583. trace_hcall_exit(opcode, retval, retbuf);
  584. preempt_enable();
  585. (*depth)--;
  586. out:
  587. local_irq_restore(flags);
  588. }
  589. #endif
  590. /**
  591. * h_get_mpp
  592. * H_GET_MPP hcall returns info in 7 parms
  593. */
  594. int h_get_mpp(struct hvcall_mpp_data *mpp_data)
  595. {
  596. int rc;
  597. unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
  598. rc = plpar_hcall9(H_GET_MPP, retbuf);
  599. mpp_data->entitled_mem = retbuf[0];
  600. mpp_data->mapped_mem = retbuf[1];
  601. mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
  602. mpp_data->pool_num = retbuf[2] & 0xffff;
  603. mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
  604. mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
  605. mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffff;
  606. mpp_data->pool_size = retbuf[4];
  607. mpp_data->loan_request = retbuf[5];
  608. mpp_data->backing_mem = retbuf[6];
  609. return rc;
  610. }
  611. EXPORT_SYMBOL(h_get_mpp);
  612. int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
  613. {
  614. int rc;
  615. unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
  616. rc = plpar_hcall9(H_GET_MPP_X, retbuf);
  617. mpp_x_data->coalesced_bytes = retbuf[0];
  618. mpp_x_data->pool_coalesced_bytes = retbuf[1];
  619. mpp_x_data->pool_purr_cycles = retbuf[2];
  620. mpp_x_data->pool_spurr_cycles = retbuf[3];
  621. return rc;
  622. }