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 22 /* Enables debugging of low-level hash table routines - careful! */ 23 #undef DEBUG 24 25 #include <linux/kernel.h> 26 #include <linux/dma-mapping.h> 27 #include <linux/console.h> 28 #include <linux/export.h> 29 #include <linux/jump_label.h> 30 #include <asm/processor.h> 31 #include <asm/mmu.h> 32 #include <asm/page.h> 33 #include <asm/pgtable.h> 34 #include <asm/machdep.h> 35 #include <asm/mmu_context.h> 36 #include <asm/iommu.h> 37 #include <asm/tlbflush.h> 38 #include <asm/tlb.h> 39 #include <asm/prom.h> 40 #include <asm/cputable.h> 41 #include <asm/udbg.h> 42 #include <asm/smp.h> 43 #include <asm/trace.h> 44 #include <asm/firmware.h> 45 #include <asm/plpar_wrappers.h> 46 #include <asm/kexec.h> 47 #include <asm/fadump.h> 48 49 #include "pseries.h" 50 51 /* Flag bits for H_BULK_REMOVE */ 52 #define HBR_REQUEST 0x4000000000000000UL 53 #define HBR_RESPONSE 0x8000000000000000UL 54 #define HBR_END 0xc000000000000000UL 55 #define HBR_AVPN 0x0200000000000000UL 56 #define HBR_ANDCOND 0x0100000000000000UL 57 58 59 /* in hvCall.S */ 60 EXPORT_SYMBOL(plpar_hcall); 61 EXPORT_SYMBOL(plpar_hcall9); 62 EXPORT_SYMBOL(plpar_hcall_norets); 63 64 void vpa_init(int cpu) 65 { 66 int hwcpu = get_hard_smp_processor_id(cpu); 67 unsigned long addr; 68 long ret; 69 struct paca_struct *pp; 70 struct dtl_entry *dtl; 71 72 /* 73 * The spec says it "may be problematic" if CPU x registers the VPA of 74 * CPU y. We should never do that, but wail if we ever do. 75 */ 76 WARN_ON(cpu != smp_processor_id()); 77 78 if (cpu_has_feature(CPU_FTR_ALTIVEC)) 79 lppaca_of(cpu).vmxregs_in_use = 1; 80 81 if (cpu_has_feature(CPU_FTR_ARCH_207S)) 82 lppaca_of(cpu).ebb_regs_in_use = 1; 83 84 addr = __pa(&lppaca_of(cpu)); 85 ret = register_vpa(hwcpu, addr); 86 87 if (ret) { 88 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area " 89 "%lx failed with %ld\n", cpu, hwcpu, addr, ret); 90 return; 91 } 92 /* 93 * PAPR says this feature is SLB-Buffer but firmware never 94 * reports that. All SPLPAR support SLB shadow buffer. 95 */ 96 addr = __pa(paca[cpu].slb_shadow_ptr); 97 if (firmware_has_feature(FW_FEATURE_SPLPAR)) { 98 ret = register_slb_shadow(hwcpu, addr); 99 if (ret) 100 pr_err("WARNING: SLB shadow buffer registration for " 101 "cpu %d (hw %d) of area %lx failed with %ld\n", 102 cpu, hwcpu, addr, ret); 103 } 104 105 /* 106 * Register dispatch trace log, if one has been allocated. 107 */ 108 pp = &paca[cpu]; 109 dtl = pp->dispatch_log; 110 if (dtl) { 111 pp->dtl_ridx = 0; 112 pp->dtl_curr = dtl; 113 lppaca_of(cpu).dtl_idx = 0; 114 115 /* hypervisor reads buffer length from this field */ 116 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES); 117 ret = register_dtl(hwcpu, __pa(dtl)); 118 if (ret) 119 pr_err("WARNING: DTL registration of cpu %d (hw %d) " 120 "failed with %ld\n", smp_processor_id(), 121 hwcpu, ret); 122 lppaca_of(cpu).dtl_enable_mask = 2; 123 } 124 } 125 126 static long pSeries_lpar_hpte_insert(unsigned long hpte_group, 127 unsigned long vpn, unsigned long pa, 128 unsigned long rflags, unsigned long vflags, 129 int psize, int apsize, int ssize) 130 { 131 unsigned long lpar_rc; 132 unsigned long flags; 133 unsigned long slot; 134 unsigned long hpte_v, hpte_r; 135 136 if (!(vflags & HPTE_V_BOLTED)) 137 pr_devel("hpte_insert(group=%lx, vpn=%016lx, " 138 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n", 139 hpte_group, vpn, pa, rflags, vflags, psize); 140 141 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID; 142 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags; 143 144 if (!(vflags & HPTE_V_BOLTED)) 145 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r); 146 147 /* Now fill in the actual HPTE */ 148 /* Set CEC cookie to 0 */ 149 /* Zero page = 0 */ 150 /* I-cache Invalidate = 0 */ 151 /* I-cache synchronize = 0 */ 152 /* Exact = 0 */ 153 flags = 0; 154 155 /* Make pHyp happy */ 156 if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU)) 157 hpte_r &= ~HPTE_R_M; 158 159 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N)) 160 flags |= H_COALESCE_CAND; 161 162 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot); 163 if (unlikely(lpar_rc == H_PTEG_FULL)) { 164 if (!(vflags & HPTE_V_BOLTED)) 165 pr_devel(" full\n"); 166 return -1; 167 } 168 169 /* 170 * Since we try and ioremap PHBs we don't own, the pte insert 171 * will fail. However we must catch the failure in hash_page 172 * or we will loop forever, so return -2 in this case. 173 */ 174 if (unlikely(lpar_rc != H_SUCCESS)) { 175 if (!(vflags & HPTE_V_BOLTED)) 176 pr_devel(" lpar err %ld\n", lpar_rc); 177 return -2; 178 } 179 if (!(vflags & HPTE_V_BOLTED)) 180 pr_devel(" -> slot: %lu\n", slot & 7); 181 182 /* Because of iSeries, we have to pass down the secondary 183 * bucket bit here as well 184 */ 185 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3); 186 } 187 188 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock); 189 190 static long pSeries_lpar_hpte_remove(unsigned long hpte_group) 191 { 192 unsigned long slot_offset; 193 unsigned long lpar_rc; 194 int i; 195 unsigned long dummy1, dummy2; 196 197 /* pick a random slot to start at */ 198 slot_offset = mftb() & 0x7; 199 200 for (i = 0; i < HPTES_PER_GROUP; i++) { 201 202 /* don't remove a bolted entry */ 203 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset, 204 (0x1UL << 4), &dummy1, &dummy2); 205 if (lpar_rc == H_SUCCESS) 206 return i; 207 208 /* 209 * The test for adjunct partition is performed before the 210 * ANDCOND test. H_RESOURCE may be returned, so we need to 211 * check for that as well. 212 */ 213 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE); 214 215 slot_offset++; 216 slot_offset &= 0x7; 217 } 218 219 return -1; 220 } 221 222 static void pSeries_lpar_hptab_clear(void) 223 { 224 unsigned long size_bytes = 1UL << ppc64_pft_size; 225 unsigned long hpte_count = size_bytes >> 4; 226 struct { 227 unsigned long pteh; 228 unsigned long ptel; 229 } ptes[4]; 230 long lpar_rc; 231 unsigned long i, j; 232 233 /* Read in batches of 4, 234 * invalidate only valid entries not in the VRMA 235 * hpte_count will be a multiple of 4 236 */ 237 for (i = 0; i < hpte_count; i += 4) { 238 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes); 239 if (lpar_rc != H_SUCCESS) 240 continue; 241 for (j = 0; j < 4; j++){ 242 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) == 243 HPTE_V_VRMA_MASK) 244 continue; 245 if (ptes[j].pteh & HPTE_V_VALID) 246 plpar_pte_remove_raw(0, i + j, 0, 247 &(ptes[j].pteh), &(ptes[j].ptel)); 248 } 249 } 250 251 #ifdef __LITTLE_ENDIAN__ 252 /* 253 * Reset exceptions to big endian. 254 * 255 * FIXME this is a hack for kexec, we need to reset the exception 256 * endian before starting the new kernel and this is a convenient place 257 * to do it. 258 * 259 * This is also called on boot when a fadump happens. In that case we 260 * must not change the exception endian mode. 261 */ 262 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active()) { 263 long rc; 264 265 rc = pseries_big_endian_exceptions(); 266 /* 267 * At this point it is unlikely panic() will get anything 268 * out to the user, but at least this will stop us from 269 * continuing on further and creating an even more 270 * difficult to debug situation. 271 * 272 * There is a known problem when kdump'ing, if cpus are offline 273 * the above call will fail. Rather than panicking again, keep 274 * going and hope the kdump kernel is also little endian, which 275 * it usually is. 276 */ 277 if (rc && !kdump_in_progress()) 278 panic("Could not enable big endian exceptions"); 279 } 280 #endif 281 } 282 283 /* 284 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and 285 * the low 3 bits of flags happen to line up. So no transform is needed. 286 * We can probably optimize here and assume the high bits of newpp are 287 * already zero. For now I am paranoid. 288 */ 289 static long pSeries_lpar_hpte_updatepp(unsigned long slot, 290 unsigned long newpp, 291 unsigned long vpn, 292 int psize, int apsize, 293 int ssize, unsigned long inv_flags) 294 { 295 unsigned long lpar_rc; 296 unsigned long flags = (newpp & 7) | H_AVPN; 297 unsigned long want_v; 298 299 want_v = hpte_encode_avpn(vpn, psize, ssize); 300 301 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...", 302 want_v, slot, flags, psize); 303 304 lpar_rc = plpar_pte_protect(flags, slot, want_v); 305 306 if (lpar_rc == H_NOT_FOUND) { 307 pr_devel("not found !\n"); 308 return -1; 309 } 310 311 pr_devel("ok\n"); 312 313 BUG_ON(lpar_rc != H_SUCCESS); 314 315 return 0; 316 } 317 318 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group) 319 { 320 long lpar_rc; 321 unsigned long i, j; 322 struct { 323 unsigned long pteh; 324 unsigned long ptel; 325 } ptes[4]; 326 327 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) { 328 329 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes); 330 if (lpar_rc != H_SUCCESS) 331 continue; 332 333 for (j = 0; j < 4; j++) { 334 if (HPTE_V_COMPARE(ptes[j].pteh, want_v) && 335 (ptes[j].pteh & HPTE_V_VALID)) 336 return i + j; 337 } 338 } 339 340 return -1; 341 } 342 343 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize) 344 { 345 long slot; 346 unsigned long hash; 347 unsigned long want_v; 348 unsigned long hpte_group; 349 350 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize); 351 want_v = hpte_encode_avpn(vpn, psize, ssize); 352 353 /* Bolted entries are always in the primary group */ 354 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP; 355 slot = __pSeries_lpar_hpte_find(want_v, hpte_group); 356 if (slot < 0) 357 return -1; 358 return hpte_group + slot; 359 } 360 361 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp, 362 unsigned long ea, 363 int psize, int ssize) 364 { 365 unsigned long vpn; 366 unsigned long lpar_rc, slot, vsid, flags; 367 368 vsid = get_kernel_vsid(ea, ssize); 369 vpn = hpt_vpn(ea, vsid, ssize); 370 371 slot = pSeries_lpar_hpte_find(vpn, psize, ssize); 372 BUG_ON(slot == -1); 373 374 flags = newpp & 7; 375 lpar_rc = plpar_pte_protect(flags, slot, 0); 376 377 BUG_ON(lpar_rc != H_SUCCESS); 378 } 379 380 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn, 381 int psize, int apsize, 382 int ssize, int local) 383 { 384 unsigned long want_v; 385 unsigned long lpar_rc; 386 unsigned long dummy1, dummy2; 387 388 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n", 389 slot, vpn, psize, local); 390 391 want_v = hpte_encode_avpn(vpn, psize, ssize); 392 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2); 393 if (lpar_rc == H_NOT_FOUND) 394 return; 395 396 BUG_ON(lpar_rc != H_SUCCESS); 397 } 398 399 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 400 /* 401 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need 402 * to make sure that we avoid bouncing the hypervisor tlbie lock. 403 */ 404 #define PPC64_HUGE_HPTE_BATCH 12 405 406 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot, 407 unsigned long *vpn, int count, 408 int psize, int ssize) 409 { 410 unsigned long param[8]; 411 int i = 0, pix = 0, rc; 412 unsigned long flags = 0; 413 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE); 414 415 if (lock_tlbie) 416 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags); 417 418 for (i = 0; i < count; i++) { 419 420 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) { 421 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0, 422 ssize, 0); 423 } else { 424 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i]; 425 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize); 426 pix += 2; 427 if (pix == 8) { 428 rc = plpar_hcall9(H_BULK_REMOVE, param, 429 param[0], param[1], param[2], 430 param[3], param[4], param[5], 431 param[6], param[7]); 432 BUG_ON(rc != H_SUCCESS); 433 pix = 0; 434 } 435 } 436 } 437 if (pix) { 438 param[pix] = HBR_END; 439 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1], 440 param[2], param[3], param[4], param[5], 441 param[6], param[7]); 442 BUG_ON(rc != H_SUCCESS); 443 } 444 445 if (lock_tlbie) 446 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags); 447 } 448 449 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid, 450 unsigned long addr, 451 unsigned char *hpte_slot_array, 452 int psize, int ssize, int local) 453 { 454 int i, index = 0; 455 unsigned long s_addr = addr; 456 unsigned int max_hpte_count, valid; 457 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH]; 458 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH]; 459 unsigned long shift, hidx, vpn = 0, hash, slot; 460 461 shift = mmu_psize_defs[psize].shift; 462 max_hpte_count = 1U << (PMD_SHIFT - shift); 463 464 for (i = 0; i < max_hpte_count; i++) { 465 valid = hpte_valid(hpte_slot_array, i); 466 if (!valid) 467 continue; 468 hidx = hpte_hash_index(hpte_slot_array, i); 469 470 /* get the vpn */ 471 addr = s_addr + (i * (1ul << shift)); 472 vpn = hpt_vpn(addr, vsid, ssize); 473 hash = hpt_hash(vpn, shift, ssize); 474 if (hidx & _PTEIDX_SECONDARY) 475 hash = ~hash; 476 477 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; 478 slot += hidx & _PTEIDX_GROUP_IX; 479 480 slot_array[index] = slot; 481 vpn_array[index] = vpn; 482 if (index == PPC64_HUGE_HPTE_BATCH - 1) { 483 /* 484 * Now do a bluk invalidate 485 */ 486 __pSeries_lpar_hugepage_invalidate(slot_array, 487 vpn_array, 488 PPC64_HUGE_HPTE_BATCH, 489 psize, ssize); 490 index = 0; 491 } else 492 index++; 493 } 494 if (index) 495 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array, 496 index, psize, ssize); 497 } 498 #else 499 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid, 500 unsigned long addr, 501 unsigned char *hpte_slot_array, 502 int psize, int ssize, int local) 503 { 504 WARN(1, "%s called without THP support\n", __func__); 505 } 506 #endif 507 508 static int pSeries_lpar_hpte_removebolted(unsigned long ea, 509 int psize, int ssize) 510 { 511 unsigned long vpn; 512 unsigned long slot, vsid; 513 514 vsid = get_kernel_vsid(ea, ssize); 515 vpn = hpt_vpn(ea, vsid, ssize); 516 517 slot = pSeries_lpar_hpte_find(vpn, psize, ssize); 518 if (slot == -1) 519 return -ENOENT; 520 521 /* 522 * lpar doesn't use the passed actual page size 523 */ 524 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0); 525 return 0; 526 } 527 528 /* 529 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie 530 * lock. 531 */ 532 static void pSeries_lpar_flush_hash_range(unsigned long number, int local) 533 { 534 unsigned long vpn; 535 unsigned long i, pix, rc; 536 unsigned long flags = 0; 537 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch); 538 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE); 539 unsigned long param[9]; 540 unsigned long hash, index, shift, hidx, slot; 541 real_pte_t pte; 542 int psize, ssize; 543 544 if (lock_tlbie) 545 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags); 546 547 psize = batch->psize; 548 ssize = batch->ssize; 549 pix = 0; 550 for (i = 0; i < number; i++) { 551 vpn = batch->vpn[i]; 552 pte = batch->pte[i]; 553 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) { 554 hash = hpt_hash(vpn, shift, ssize); 555 hidx = __rpte_to_hidx(pte, index); 556 if (hidx & _PTEIDX_SECONDARY) 557 hash = ~hash; 558 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; 559 slot += hidx & _PTEIDX_GROUP_IX; 560 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) { 561 /* 562 * lpar doesn't use the passed actual page size 563 */ 564 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 565 0, ssize, local); 566 } else { 567 param[pix] = HBR_REQUEST | HBR_AVPN | slot; 568 param[pix+1] = hpte_encode_avpn(vpn, psize, 569 ssize); 570 pix += 2; 571 if (pix == 8) { 572 rc = plpar_hcall9(H_BULK_REMOVE, param, 573 param[0], param[1], param[2], 574 param[3], param[4], param[5], 575 param[6], param[7]); 576 BUG_ON(rc != H_SUCCESS); 577 pix = 0; 578 } 579 } 580 } pte_iterate_hashed_end(); 581 } 582 if (pix) { 583 param[pix] = HBR_END; 584 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1], 585 param[2], param[3], param[4], param[5], 586 param[6], param[7]); 587 BUG_ON(rc != H_SUCCESS); 588 } 589 590 if (lock_tlbie) 591 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags); 592 } 593 594 static int __init disable_bulk_remove(char *str) 595 { 596 if (strcmp(str, "off") == 0 && 597 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) { 598 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature"); 599 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE; 600 } 601 return 1; 602 } 603 604 __setup("bulk_remove=", disable_bulk_remove); 605 606 void __init hpte_init_lpar(void) 607 { 608 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate; 609 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp; 610 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp; 611 ppc_md.hpte_insert = pSeries_lpar_hpte_insert; 612 ppc_md.hpte_remove = pSeries_lpar_hpte_remove; 613 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted; 614 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range; 615 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear; 616 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate; 617 } 618 619 #ifdef CONFIG_PPC_SMLPAR 620 #define CMO_FREE_HINT_DEFAULT 1 621 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT; 622 623 static int __init cmo_free_hint(char *str) 624 { 625 char *parm; 626 parm = strstrip(str); 627 628 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) { 629 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n"); 630 cmo_free_hint_flag = 0; 631 return 1; 632 } 633 634 cmo_free_hint_flag = 1; 635 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n"); 636 637 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0) 638 return 1; 639 640 return 0; 641 } 642 643 __setup("cmo_free_hint=", cmo_free_hint); 644 645 static void pSeries_set_page_state(struct page *page, int order, 646 unsigned long state) 647 { 648 int i, j; 649 unsigned long cmo_page_sz, addr; 650 651 cmo_page_sz = cmo_get_page_size(); 652 addr = __pa((unsigned long)page_address(page)); 653 654 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) { 655 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz) 656 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0); 657 } 658 } 659 660 void arch_free_page(struct page *page, int order) 661 { 662 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO)) 663 return; 664 665 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED); 666 } 667 EXPORT_SYMBOL(arch_free_page); 668 669 #endif 670 671 #ifdef CONFIG_TRACEPOINTS 672 #ifdef HAVE_JUMP_LABEL 673 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT; 674 675 void hcall_tracepoint_regfunc(void) 676 { 677 static_key_slow_inc(&hcall_tracepoint_key); 678 } 679 680 void hcall_tracepoint_unregfunc(void) 681 { 682 static_key_slow_dec(&hcall_tracepoint_key); 683 } 684 #else 685 /* 686 * We optimise our hcall path by placing hcall_tracepoint_refcount 687 * directly in the TOC so we can check if the hcall tracepoints are 688 * enabled via a single load. 689 */ 690 691 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */ 692 extern long hcall_tracepoint_refcount; 693 694 void hcall_tracepoint_regfunc(void) 695 { 696 hcall_tracepoint_refcount++; 697 } 698 699 void hcall_tracepoint_unregfunc(void) 700 { 701 hcall_tracepoint_refcount--; 702 } 703 #endif 704 705 /* 706 * Since the tracing code might execute hcalls we need to guard against 707 * recursion. One example of this are spinlocks calling H_YIELD on 708 * shared processor partitions. 709 */ 710 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth); 711 712 713 void __trace_hcall_entry(unsigned long opcode, unsigned long *args) 714 { 715 unsigned long flags; 716 unsigned int *depth; 717 718 /* 719 * We cannot call tracepoints inside RCU idle regions which 720 * means we must not trace H_CEDE. 721 */ 722 if (opcode == H_CEDE) 723 return; 724 725 local_irq_save(flags); 726 727 depth = this_cpu_ptr(&hcall_trace_depth); 728 729 if (*depth) 730 goto out; 731 732 (*depth)++; 733 preempt_disable(); 734 trace_hcall_entry(opcode, args); 735 (*depth)--; 736 737 out: 738 local_irq_restore(flags); 739 } 740 741 void __trace_hcall_exit(long opcode, unsigned long retval, 742 unsigned long *retbuf) 743 { 744 unsigned long flags; 745 unsigned int *depth; 746 747 if (opcode == H_CEDE) 748 return; 749 750 local_irq_save(flags); 751 752 depth = this_cpu_ptr(&hcall_trace_depth); 753 754 if (*depth) 755 goto out; 756 757 (*depth)++; 758 trace_hcall_exit(opcode, retval, retbuf); 759 preempt_enable(); 760 (*depth)--; 761 762 out: 763 local_irq_restore(flags); 764 } 765 #endif 766 767 /** 768 * h_get_mpp 769 * H_GET_MPP hcall returns info in 7 parms 770 */ 771 int h_get_mpp(struct hvcall_mpp_data *mpp_data) 772 { 773 int rc; 774 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE]; 775 776 rc = plpar_hcall9(H_GET_MPP, retbuf); 777 778 mpp_data->entitled_mem = retbuf[0]; 779 mpp_data->mapped_mem = retbuf[1]; 780 781 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff; 782 mpp_data->pool_num = retbuf[2] & 0xffff; 783 784 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff; 785 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff; 786 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL; 787 788 mpp_data->pool_size = retbuf[4]; 789 mpp_data->loan_request = retbuf[5]; 790 mpp_data->backing_mem = retbuf[6]; 791 792 return rc; 793 } 794 EXPORT_SYMBOL(h_get_mpp); 795 796 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data) 797 { 798 int rc; 799 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 }; 800 801 rc = plpar_hcall9(H_GET_MPP_X, retbuf); 802 803 mpp_x_data->coalesced_bytes = retbuf[0]; 804 mpp_x_data->pool_coalesced_bytes = retbuf[1]; 805 mpp_x_data->pool_purr_cycles = retbuf[2]; 806 mpp_x_data->pool_spurr_cycles = retbuf[3]; 807 808 return rc; 809 } 810