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 #ifdef CONFIG_PPC_STD_MMU_64 94 /* 95 * PAPR says this feature is SLB-Buffer but firmware never 96 * reports that. All SPLPAR support SLB shadow buffer. 97 */ 98 if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) { 99 addr = __pa(paca[cpu].slb_shadow_ptr); 100 ret = register_slb_shadow(hwcpu, addr); 101 if (ret) 102 pr_err("WARNING: SLB shadow buffer registration for " 103 "cpu %d (hw %d) of area %lx failed with %ld\n", 104 cpu, hwcpu, addr, ret); 105 } 106 #endif /* CONFIG_PPC_STD_MMU_64 */ 107 108 /* 109 * Register dispatch trace log, if one has been allocated. 110 */ 111 pp = &paca[cpu]; 112 dtl = pp->dispatch_log; 113 if (dtl) { 114 pp->dtl_ridx = 0; 115 pp->dtl_curr = dtl; 116 lppaca_of(cpu).dtl_idx = 0; 117 118 /* hypervisor reads buffer length from this field */ 119 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES); 120 ret = register_dtl(hwcpu, __pa(dtl)); 121 if (ret) 122 pr_err("WARNING: DTL registration of cpu %d (hw %d) " 123 "failed with %ld\n", smp_processor_id(), 124 hwcpu, ret); 125 lppaca_of(cpu).dtl_enable_mask = 2; 126 } 127 } 128 129 #ifdef CONFIG_PPC_STD_MMU_64 130 131 static long pSeries_lpar_hpte_insert(unsigned long hpte_group, 132 unsigned long vpn, unsigned long pa, 133 unsigned long rflags, unsigned long vflags, 134 int psize, int apsize, int ssize) 135 { 136 unsigned long lpar_rc; 137 unsigned long flags; 138 unsigned long slot; 139 unsigned long hpte_v, hpte_r; 140 141 if (!(vflags & HPTE_V_BOLTED)) 142 pr_devel("hpte_insert(group=%lx, vpn=%016lx, " 143 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n", 144 hpte_group, vpn, pa, rflags, vflags, psize); 145 146 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID; 147 hpte_r = hpte_encode_r(pa, psize, apsize, ssize) | rflags; 148 149 if (!(vflags & HPTE_V_BOLTED)) 150 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r); 151 152 /* Now fill in the actual HPTE */ 153 /* Set CEC cookie to 0 */ 154 /* Zero page = 0 */ 155 /* I-cache Invalidate = 0 */ 156 /* I-cache synchronize = 0 */ 157 /* Exact = 0 */ 158 flags = 0; 159 160 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N)) 161 flags |= H_COALESCE_CAND; 162 163 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot); 164 if (unlikely(lpar_rc == H_PTEG_FULL)) { 165 if (!(vflags & HPTE_V_BOLTED)) 166 pr_devel(" full\n"); 167 return -1; 168 } 169 170 /* 171 * Since we try and ioremap PHBs we don't own, the pte insert 172 * will fail. However we must catch the failure in hash_page 173 * or we will loop forever, so return -2 in this case. 174 */ 175 if (unlikely(lpar_rc != H_SUCCESS)) { 176 if (!(vflags & HPTE_V_BOLTED)) 177 pr_devel(" lpar err %ld\n", lpar_rc); 178 return -2; 179 } 180 if (!(vflags & HPTE_V_BOLTED)) 181 pr_devel(" -> slot: %lu\n", slot & 7); 182 183 /* Because of iSeries, we have to pass down the secondary 184 * bucket bit here as well 185 */ 186 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3); 187 } 188 189 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock); 190 191 static long pSeries_lpar_hpte_remove(unsigned long hpte_group) 192 { 193 unsigned long slot_offset; 194 unsigned long lpar_rc; 195 int i; 196 unsigned long dummy1, dummy2; 197 198 /* pick a random slot to start at */ 199 slot_offset = mftb() & 0x7; 200 201 for (i = 0; i < HPTES_PER_GROUP; i++) { 202 203 /* don't remove a bolted entry */ 204 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset, 205 (0x1UL << 4), &dummy1, &dummy2); 206 if (lpar_rc == H_SUCCESS) 207 return i; 208 209 /* 210 * The test for adjunct partition is performed before the 211 * ANDCOND test. H_RESOURCE may be returned, so we need to 212 * check for that as well. 213 */ 214 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE); 215 216 slot_offset++; 217 slot_offset &= 0x7; 218 } 219 220 return -1; 221 } 222 223 static void pSeries_lpar_hptab_clear(void) 224 { 225 unsigned long size_bytes = 1UL << ppc64_pft_size; 226 unsigned long hpte_count = size_bytes >> 4; 227 struct { 228 unsigned long pteh; 229 unsigned long ptel; 230 } ptes[4]; 231 long lpar_rc; 232 unsigned long i, j; 233 234 /* Read in batches of 4, 235 * invalidate only valid entries not in the VRMA 236 * hpte_count will be a multiple of 4 237 */ 238 for (i = 0; i < hpte_count; i += 4) { 239 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes); 240 if (lpar_rc != H_SUCCESS) 241 continue; 242 for (j = 0; j < 4; j++){ 243 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) == 244 HPTE_V_VRMA_MASK) 245 continue; 246 if (ptes[j].pteh & HPTE_V_VALID) 247 plpar_pte_remove_raw(0, i + j, 0, 248 &(ptes[j].pteh), &(ptes[j].ptel)); 249 } 250 } 251 252 #ifdef __LITTLE_ENDIAN__ 253 /* 254 * Reset exceptions to big endian. 255 * 256 * FIXME this is a hack for kexec, we need to reset the exception 257 * endian before starting the new kernel and this is a convenient place 258 * to do it. 259 * 260 * This is also called on boot when a fadump happens. In that case we 261 * must not change the exception endian mode. 262 */ 263 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active()) { 264 long rc; 265 266 rc = pseries_big_endian_exceptions(); 267 /* 268 * At this point it is unlikely panic() will get anything 269 * out to the user, but at least this will stop us from 270 * continuing on further and creating an even more 271 * difficult to debug situation. 272 * 273 * There is a known problem when kdump'ing, if cpus are offline 274 * the above call will fail. Rather than panicking again, keep 275 * going and hope the kdump kernel is also little endian, which 276 * it usually is. 277 */ 278 if (rc && !kdump_in_progress()) 279 panic("Could not enable big endian exceptions"); 280 } 281 #endif 282 } 283 284 /* 285 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and 286 * the low 3 bits of flags happen to line up. So no transform is needed. 287 * We can probably optimize here and assume the high bits of newpp are 288 * already zero. For now I am paranoid. 289 */ 290 static long pSeries_lpar_hpte_updatepp(unsigned long slot, 291 unsigned long newpp, 292 unsigned long vpn, 293 int psize, int apsize, 294 int ssize, unsigned long inv_flags) 295 { 296 unsigned long lpar_rc; 297 unsigned long flags = (newpp & 7) | H_AVPN; 298 unsigned long want_v; 299 300 want_v = hpte_encode_avpn(vpn, psize, ssize); 301 302 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...", 303 want_v, slot, flags, psize); 304 305 lpar_rc = plpar_pte_protect(flags, slot, want_v); 306 307 if (lpar_rc == H_NOT_FOUND) { 308 pr_devel("not found !\n"); 309 return -1; 310 } 311 312 pr_devel("ok\n"); 313 314 BUG_ON(lpar_rc != H_SUCCESS); 315 316 return 0; 317 } 318 319 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group) 320 { 321 long lpar_rc; 322 unsigned long i, j; 323 struct { 324 unsigned long pteh; 325 unsigned long ptel; 326 } ptes[4]; 327 328 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) { 329 330 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes); 331 if (lpar_rc != H_SUCCESS) 332 continue; 333 334 for (j = 0; j < 4; j++) { 335 if (HPTE_V_COMPARE(ptes[j].pteh, want_v) && 336 (ptes[j].pteh & HPTE_V_VALID)) 337 return i + j; 338 } 339 } 340 341 return -1; 342 } 343 344 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize) 345 { 346 long slot; 347 unsigned long hash; 348 unsigned long want_v; 349 unsigned long hpte_group; 350 351 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize); 352 want_v = hpte_encode_avpn(vpn, psize, ssize); 353 354 /* Bolted entries are always in the primary group */ 355 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP; 356 slot = __pSeries_lpar_hpte_find(want_v, hpte_group); 357 if (slot < 0) 358 return -1; 359 return hpte_group + slot; 360 } 361 362 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp, 363 unsigned long ea, 364 int psize, int ssize) 365 { 366 unsigned long vpn; 367 unsigned long lpar_rc, slot, vsid, flags; 368 369 vsid = get_kernel_vsid(ea, ssize); 370 vpn = hpt_vpn(ea, vsid, ssize); 371 372 slot = pSeries_lpar_hpte_find(vpn, psize, ssize); 373 BUG_ON(slot == -1); 374 375 flags = newpp & 7; 376 lpar_rc = plpar_pte_protect(flags, slot, 0); 377 378 BUG_ON(lpar_rc != H_SUCCESS); 379 } 380 381 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn, 382 int psize, int apsize, 383 int ssize, int local) 384 { 385 unsigned long want_v; 386 unsigned long lpar_rc; 387 unsigned long dummy1, dummy2; 388 389 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n", 390 slot, vpn, psize, local); 391 392 want_v = hpte_encode_avpn(vpn, psize, ssize); 393 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2); 394 if (lpar_rc == H_NOT_FOUND) 395 return; 396 397 BUG_ON(lpar_rc != H_SUCCESS); 398 } 399 400 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 401 /* 402 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need 403 * to make sure that we avoid bouncing the hypervisor tlbie lock. 404 */ 405 #define PPC64_HUGE_HPTE_BATCH 12 406 407 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot, 408 unsigned long *vpn, int count, 409 int psize, int ssize) 410 { 411 unsigned long param[8]; 412 int i = 0, pix = 0, rc; 413 unsigned long flags = 0; 414 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE); 415 416 if (lock_tlbie) 417 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags); 418 419 for (i = 0; i < count; i++) { 420 421 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) { 422 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0, 423 ssize, 0); 424 } else { 425 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i]; 426 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize); 427 pix += 2; 428 if (pix == 8) { 429 rc = plpar_hcall9(H_BULK_REMOVE, param, 430 param[0], param[1], param[2], 431 param[3], param[4], param[5], 432 param[6], param[7]); 433 BUG_ON(rc != H_SUCCESS); 434 pix = 0; 435 } 436 } 437 } 438 if (pix) { 439 param[pix] = HBR_END; 440 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1], 441 param[2], param[3], param[4], param[5], 442 param[6], param[7]); 443 BUG_ON(rc != H_SUCCESS); 444 } 445 446 if (lock_tlbie) 447 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags); 448 } 449 450 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid, 451 unsigned long addr, 452 unsigned char *hpte_slot_array, 453 int psize, int ssize, int local) 454 { 455 int i, index = 0; 456 unsigned long s_addr = addr; 457 unsigned int max_hpte_count, valid; 458 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH]; 459 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH]; 460 unsigned long shift, hidx, vpn = 0, hash, slot; 461 462 shift = mmu_psize_defs[psize].shift; 463 max_hpte_count = 1U << (PMD_SHIFT - shift); 464 465 for (i = 0; i < max_hpte_count; i++) { 466 valid = hpte_valid(hpte_slot_array, i); 467 if (!valid) 468 continue; 469 hidx = hpte_hash_index(hpte_slot_array, i); 470 471 /* get the vpn */ 472 addr = s_addr + (i * (1ul << shift)); 473 vpn = hpt_vpn(addr, vsid, ssize); 474 hash = hpt_hash(vpn, shift, ssize); 475 if (hidx & _PTEIDX_SECONDARY) 476 hash = ~hash; 477 478 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; 479 slot += hidx & _PTEIDX_GROUP_IX; 480 481 slot_array[index] = slot; 482 vpn_array[index] = vpn; 483 if (index == PPC64_HUGE_HPTE_BATCH - 1) { 484 /* 485 * Now do a bluk invalidate 486 */ 487 __pSeries_lpar_hugepage_invalidate(slot_array, 488 vpn_array, 489 PPC64_HUGE_HPTE_BATCH, 490 psize, ssize); 491 index = 0; 492 } else 493 index++; 494 } 495 if (index) 496 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array, 497 index, psize, ssize); 498 } 499 #else 500 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid, 501 unsigned long addr, 502 unsigned char *hpte_slot_array, 503 int psize, int ssize, int local) 504 { 505 WARN(1, "%s called without THP support\n", __func__); 506 } 507 #endif 508 509 static int pSeries_lpar_hpte_removebolted(unsigned long ea, 510 int psize, int ssize) 511 { 512 unsigned long vpn; 513 unsigned long slot, vsid; 514 515 vsid = get_kernel_vsid(ea, ssize); 516 vpn = hpt_vpn(ea, vsid, ssize); 517 518 slot = pSeries_lpar_hpte_find(vpn, psize, ssize); 519 if (slot == -1) 520 return -ENOENT; 521 522 /* 523 * lpar doesn't use the passed actual page size 524 */ 525 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0); 526 return 0; 527 } 528 529 /* 530 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie 531 * lock. 532 */ 533 static void pSeries_lpar_flush_hash_range(unsigned long number, int local) 534 { 535 unsigned long vpn; 536 unsigned long i, pix, rc; 537 unsigned long flags = 0; 538 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch); 539 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE); 540 unsigned long param[9]; 541 unsigned long hash, index, shift, hidx, slot; 542 real_pte_t pte; 543 int psize, ssize; 544 545 if (lock_tlbie) 546 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags); 547 548 psize = batch->psize; 549 ssize = batch->ssize; 550 pix = 0; 551 for (i = 0; i < number; i++) { 552 vpn = batch->vpn[i]; 553 pte = batch->pte[i]; 554 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) { 555 hash = hpt_hash(vpn, shift, ssize); 556 hidx = __rpte_to_hidx(pte, index); 557 if (hidx & _PTEIDX_SECONDARY) 558 hash = ~hash; 559 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP; 560 slot += hidx & _PTEIDX_GROUP_IX; 561 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) { 562 /* 563 * lpar doesn't use the passed actual page size 564 */ 565 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 566 0, ssize, local); 567 } else { 568 param[pix] = HBR_REQUEST | HBR_AVPN | slot; 569 param[pix+1] = hpte_encode_avpn(vpn, psize, 570 ssize); 571 pix += 2; 572 if (pix == 8) { 573 rc = plpar_hcall9(H_BULK_REMOVE, param, 574 param[0], param[1], param[2], 575 param[3], param[4], param[5], 576 param[6], param[7]); 577 BUG_ON(rc != H_SUCCESS); 578 pix = 0; 579 } 580 } 581 } pte_iterate_hashed_end(); 582 } 583 if (pix) { 584 param[pix] = HBR_END; 585 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1], 586 param[2], param[3], param[4], param[5], 587 param[6], param[7]); 588 BUG_ON(rc != H_SUCCESS); 589 } 590 591 if (lock_tlbie) 592 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags); 593 } 594 595 static int __init disable_bulk_remove(char *str) 596 { 597 if (strcmp(str, "off") == 0 && 598 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) { 599 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature"); 600 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE; 601 } 602 return 1; 603 } 604 605 __setup("bulk_remove=", disable_bulk_remove); 606 607 void __init hpte_init_lpar(void) 608 { 609 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate; 610 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp; 611 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp; 612 ppc_md.hpte_insert = pSeries_lpar_hpte_insert; 613 ppc_md.hpte_remove = pSeries_lpar_hpte_remove; 614 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted; 615 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range; 616 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear; 617 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate; 618 } 619 620 #ifdef CONFIG_PPC_SMLPAR 621 #define CMO_FREE_HINT_DEFAULT 1 622 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT; 623 624 static int __init cmo_free_hint(char *str) 625 { 626 char *parm; 627 parm = strstrip(str); 628 629 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) { 630 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n"); 631 cmo_free_hint_flag = 0; 632 return 1; 633 } 634 635 cmo_free_hint_flag = 1; 636 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n"); 637 638 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0) 639 return 1; 640 641 return 0; 642 } 643 644 __setup("cmo_free_hint=", cmo_free_hint); 645 646 static void pSeries_set_page_state(struct page *page, int order, 647 unsigned long state) 648 { 649 int i, j; 650 unsigned long cmo_page_sz, addr; 651 652 cmo_page_sz = cmo_get_page_size(); 653 addr = __pa((unsigned long)page_address(page)); 654 655 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) { 656 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz) 657 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0); 658 } 659 } 660 661 void arch_free_page(struct page *page, int order) 662 { 663 if (radix_enabled()) 664 return; 665 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO)) 666 return; 667 668 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED); 669 } 670 EXPORT_SYMBOL(arch_free_page); 671 672 #endif /* CONFIG_PPC_SMLPAR */ 673 #endif /* CONFIG_PPC_STD_MMU_64 */ 674 675 #ifdef CONFIG_TRACEPOINTS 676 #ifdef HAVE_JUMP_LABEL 677 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT; 678 679 void hcall_tracepoint_regfunc(void) 680 { 681 static_key_slow_inc(&hcall_tracepoint_key); 682 } 683 684 void hcall_tracepoint_unregfunc(void) 685 { 686 static_key_slow_dec(&hcall_tracepoint_key); 687 } 688 #else 689 /* 690 * We optimise our hcall path by placing hcall_tracepoint_refcount 691 * directly in the TOC so we can check if the hcall tracepoints are 692 * enabled via a single load. 693 */ 694 695 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */ 696 extern long hcall_tracepoint_refcount; 697 698 void hcall_tracepoint_regfunc(void) 699 { 700 hcall_tracepoint_refcount++; 701 } 702 703 void hcall_tracepoint_unregfunc(void) 704 { 705 hcall_tracepoint_refcount--; 706 } 707 #endif 708 709 /* 710 * Since the tracing code might execute hcalls we need to guard against 711 * recursion. One example of this are spinlocks calling H_YIELD on 712 * shared processor partitions. 713 */ 714 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth); 715 716 717 void __trace_hcall_entry(unsigned long opcode, unsigned long *args) 718 { 719 unsigned long flags; 720 unsigned int *depth; 721 722 /* 723 * We cannot call tracepoints inside RCU idle regions which 724 * means we must not trace H_CEDE. 725 */ 726 if (opcode == H_CEDE) 727 return; 728 729 local_irq_save(flags); 730 731 depth = this_cpu_ptr(&hcall_trace_depth); 732 733 if (*depth) 734 goto out; 735 736 (*depth)++; 737 preempt_disable(); 738 trace_hcall_entry(opcode, args); 739 (*depth)--; 740 741 out: 742 local_irq_restore(flags); 743 } 744 745 void __trace_hcall_exit(long opcode, unsigned long retval, 746 unsigned long *retbuf) 747 { 748 unsigned long flags; 749 unsigned int *depth; 750 751 if (opcode == H_CEDE) 752 return; 753 754 local_irq_save(flags); 755 756 depth = this_cpu_ptr(&hcall_trace_depth); 757 758 if (*depth) 759 goto out; 760 761 (*depth)++; 762 trace_hcall_exit(opcode, retval, retbuf); 763 preempt_enable(); 764 (*depth)--; 765 766 out: 767 local_irq_restore(flags); 768 } 769 #endif 770 771 /** 772 * h_get_mpp 773 * H_GET_MPP hcall returns info in 7 parms 774 */ 775 int h_get_mpp(struct hvcall_mpp_data *mpp_data) 776 { 777 int rc; 778 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE]; 779 780 rc = plpar_hcall9(H_GET_MPP, retbuf); 781 782 mpp_data->entitled_mem = retbuf[0]; 783 mpp_data->mapped_mem = retbuf[1]; 784 785 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff; 786 mpp_data->pool_num = retbuf[2] & 0xffff; 787 788 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff; 789 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff; 790 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL; 791 792 mpp_data->pool_size = retbuf[4]; 793 mpp_data->loan_request = retbuf[5]; 794 mpp_data->backing_mem = retbuf[6]; 795 796 return rc; 797 } 798 EXPORT_SYMBOL(h_get_mpp); 799 800 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data) 801 { 802 int rc; 803 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 }; 804 805 rc = plpar_hcall9(H_GET_MPP_X, retbuf); 806 807 mpp_x_data->coalesced_bytes = retbuf[0]; 808 mpp_x_data->pool_coalesced_bytes = retbuf[1]; 809 mpp_x_data->pool_purr_cycles = retbuf[2]; 810 mpp_x_data->pool_spurr_cycles = retbuf[3]; 811 812 return rc; 813 } 814