arch/powerpc/platforms/pseries/lpar.c

   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 <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/abs_addr.h>
  34 #include <asm/mmu_context.h>
  35 #include <asm/iommu.h>
  36 #include <asm/tlbflush.h>
  37 #include <asm/tlb.h>
  38 #include <asm/prom.h>
  39 #include <asm/cputable.h>
  40 #include <asm/udbg.h>
  41 #include <asm/smp.h>
  42 #include <asm/trace.h>
  43
  44 #include "plpar_wrappers.h"
  45 #include "pseries.h"
  46
  47
  48 /* in hvCall.S */
  49 EXPORT_SYMBOL(plpar_hcall);
  50 EXPORT_SYMBOL(plpar_hcall9);
  51 EXPORT_SYMBOL(plpar_hcall_norets);
  52
  53 extern void pSeries_find_serial_port(void);
  54
  55
  56 static int vtermno;     /* virtual terminal# for udbg  */
  57
  58 #define __ALIGNED__ __attribute__((__aligned__(sizeof(long))))
  59 static void udbg_hvsi_putc(char c)
  60 {
  61         /* packet's seqno isn't used anyways */
  62         uint8_t packet[] __ALIGNED__ = { 0xff, 5, 0, 0, c };
  63         int rc;
  64
  65         if (c == '\n')
  66                 udbg_hvsi_putc('\r');
  67
  68         do {
  69                 rc = plpar_put_term_char(vtermno, sizeof(packet), packet);
  70         } while (rc == H_BUSY);
  71 }
  72
  73 static long hvsi_udbg_buf_len;
  74 static uint8_t hvsi_udbg_buf[256];
  75
  76 static int udbg_hvsi_getc_poll(void)
  77 {
  78         unsigned char ch;
  79         int rc, i;
  80
  81         if (hvsi_udbg_buf_len == 0) {
  82                 rc = plpar_get_term_char(vtermno, &hvsi_udbg_buf_len, hvsi_udbg_buf);
  83                 if (rc != H_SUCCESS || hvsi_udbg_buf[0] != 0xff) {
  84                         /* bad read or non-data packet */
  85                         hvsi_udbg_buf_len = 0;
  86                 } else {
  87                         /* remove the packet header */
  88                         for (i = 4; i < hvsi_udbg_buf_len; i++)
  89                                 hvsi_udbg_buf[i-4] = hvsi_udbg_buf[i];
  90                         hvsi_udbg_buf_len -= 4;
  91                 }
  92         }
  93
  94         if (hvsi_udbg_buf_len <= 0 || hvsi_udbg_buf_len > 256) {
  95                 /* no data ready */
  96                 hvsi_udbg_buf_len = 0;
  97                 return -1;
  98         }
  99
 100         ch = hvsi_udbg_buf[0];
 101         /* shift remaining data down */
 102         for (i = 1; i < hvsi_udbg_buf_len; i++) {
 103                 hvsi_udbg_buf[i-1] = hvsi_udbg_buf[i];
 104         }
 105         hvsi_udbg_buf_len--;
 106
 107         return ch;
 108 }
 109
 110 static int udbg_hvsi_getc(void)
 111 {
 112         int ch;
 113         for (;;) {
 114                 ch = udbg_hvsi_getc_poll();
 115                 if (ch == -1) {
 116                         /* This shouldn't be needed...but... */
 117                         volatile unsigned long delay;
 118                         for (delay=0; delay < 2000000; delay++)
 119                                 ;
 120                 } else {
 121                         return ch;
 122                 }
 123         }
 124 }
 125
 126 static void udbg_putcLP(char c)
 127 {
 128         char buf[16];
 129         unsigned long rc;
 130
 131         if (c == '\n')
 132                 udbg_putcLP('\r');
 133
 134         buf[0] = c;
 135         do {
 136                 rc = plpar_put_term_char(vtermno, 1, buf);
 137         } while(rc == H_BUSY);
 138 }
 139
 140 /* Buffered chars getc */
 141 static long inbuflen;
 142 static long inbuf[2];   /* must be 2 longs */
 143
 144 static int udbg_getc_pollLP(void)
 145 {
 146         /* The interface is tricky because it may return up to 16 chars.
 147          * We save them statically for future calls to udbg_getc().
 148          */
 149         char ch, *buf = (char *)inbuf;
 150         int i;
 151         long rc;
 152         if (inbuflen == 0) {
 153                 /* get some more chars. */
 154                 inbuflen = 0;
 155                 rc = plpar_get_term_char(vtermno, &inbuflen, buf);
 156                 if (rc != H_SUCCESS)
 157                         inbuflen = 0;   /* otherwise inbuflen is garbage */
 158         }
 159         if (inbuflen <= 0 || inbuflen > 16) {
 160                 /* Catch error case as well as other oddities (corruption) */
 161                 inbuflen = 0;
 162                 return -1;
 163         }
 164         ch = buf[0];
 165         for (i = 1; i < inbuflen; i++)  /* shuffle them down. */
 166                 buf[i-1] = buf[i];
 167         inbuflen--;
 168         return ch;
 169 }
 170
 171 static int udbg_getcLP(void)
 172 {
 173         int ch;
 174         for (;;) {
 175                 ch = udbg_getc_pollLP();
 176                 if (ch == -1) {
 177                         /* This shouldn't be needed...but... */
 178                         volatile unsigned long delay;
 179                         for (delay=0; delay < 2000000; delay++)
 180                                 ;
 181                 } else {
 182                         return ch;
 183                 }
 184         }
 185 }
 186
 187 /* call this from early_init() for a working debug console on
 188  * vterm capable LPAR machines
 189  */
 190 void __init udbg_init_debug_lpar(void)
 191 {
 192         vtermno = 0;
 193         udbg_putc = udbg_putcLP;
 194         udbg_getc = udbg_getcLP;
 195         udbg_getc_poll = udbg_getc_pollLP;
 196
 197         register_early_udbg_console();
 198 }
 199
 200 /* returns 0 if couldn't find or use /chosen/stdout as console */
 201 void __init find_udbg_vterm(void)
 202 {
 203         struct device_node *stdout_node;
 204         const u32 *termno;
 205         const char *name;
 206
 207         /* find the boot console from /chosen/stdout */
 208         if (!of_chosen)
 209                 return;
 210         name = of_get_property(of_chosen, "linux,stdout-path", NULL);
 211         if (name == NULL)
 212                 return;
 213         stdout_node = of_find_node_by_path(name);
 214         if (!stdout_node)
 215                 return;
 216         name = of_get_property(stdout_node, "name", NULL);
 217         if (!name) {
 218                 printk(KERN_WARNING "stdout node missing 'name' property!\n");
 219                 goto out;
 220         }
 221
 222         /* Check if it's a virtual terminal */
 223         if (strncmp(name, "vty", 3) != 0)
 224                 goto out;
 225         termno = of_get_property(stdout_node, "reg", NULL);
 226         if (termno == NULL)
 227                 goto out;
 228         vtermno = termno[0];
 229
 230         if (of_device_is_compatible(stdout_node, "hvterm1")) {
 231                 udbg_putc = udbg_putcLP;
 232                 udbg_getc = udbg_getcLP;
 233                 udbg_getc_poll = udbg_getc_pollLP;
 234                 add_preferred_console("hvc", termno[0] & 0xff, NULL);
 235         } else if (of_device_is_compatible(stdout_node, "hvterm-protocol")) {
 236                 vtermno = termno[0];
 237                 udbg_putc = udbg_hvsi_putc;
 238                 udbg_getc = udbg_hvsi_getc;
 239                 udbg_getc_poll = udbg_hvsi_getc_poll;
 240                 add_preferred_console("hvsi", termno[0] & 0xff, NULL);
 241         }
 242 out:
 243         of_node_put(stdout_node);
 244 }
 245
 246 void vpa_init(int cpu)
 247 {
 248         int hwcpu = get_hard_smp_processor_id(cpu);
 249         unsigned long addr;
 250         long ret;
 251         struct paca_struct *pp;
 252         struct dtl_entry *dtl;
 253
 254         if (cpu_has_feature(CPU_FTR_ALTIVEC))
 255                 lppaca_of(cpu).vmxregs_in_use = 1;
 256
 257         addr = __pa(&lppaca_of(cpu));
 258         ret = register_vpa(hwcpu, addr);
 259
 260         if (ret) {
 261                 printk(KERN_ERR "WARNING: vpa_init: VPA registration for "
 262                                 "cpu %d (hw %d) of area %lx returns %ld\n",
 263                                 cpu, hwcpu, addr, ret);
 264                 return;
 265         }
 266         /*
 267          * PAPR says this feature is SLB-Buffer but firmware never
 268          * reports that.  All SPLPAR support SLB shadow buffer.
 269          */
 270         addr = __pa(&slb_shadow[cpu]);
 271         if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
 272                 ret = register_slb_shadow(hwcpu, addr);
 273                 if (ret)
 274                         printk(KERN_ERR
 275                                "WARNING: vpa_init: SLB shadow buffer "
 276                                "registration for cpu %d (hw %d) of area %lx "
 277                                "returns %ld\n", cpu, hwcpu, addr, ret);
 278         }
 279
 280         /*
 281          * Register dispatch trace log, if one has been allocated.
 282          */
 283         pp = &paca[cpu];
 284         dtl = pp->dispatch_log;
 285         if (dtl) {
 286                 pp->dtl_ridx = 0;
 287                 pp->dtl_curr = dtl;
 288                 lppaca_of(cpu).dtl_idx = 0;
 289
 290                 /* hypervisor reads buffer length from this field */
 291                 dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
 292                 ret = register_dtl(hwcpu, __pa(dtl));
 293                 if (ret)
 294                         pr_warn("DTL registration failed for cpu %d (%ld)\n",
 295                                 cpu, ret);
 296                 lppaca_of(cpu).dtl_enable_mask = 2;
 297         }
 298 }
 299
 300 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
 301                               unsigned long va, unsigned long pa,
 302                               unsigned long rflags, unsigned long vflags,
 303                               int psize, int ssize)
 304 {
 305         unsigned long lpar_rc;
 306         unsigned long flags;
 307         unsigned long slot;
 308         unsigned long hpte_v, hpte_r;
 309
 310         if (!(vflags & HPTE_V_BOLTED))
 311                 pr_devel("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
 312                          "rflags=%lx, vflags=%lx, psize=%d)\n",
 313                          hpte_group, va, pa, rflags, vflags, psize);
 314
 315         hpte_v = hpte_encode_v(va, psize, ssize) | vflags | HPTE_V_VALID;
 316         hpte_r = hpte_encode_r(pa, psize) | rflags;
 317
 318         if (!(vflags & HPTE_V_BOLTED))
 319                 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
 320
 321         /* Now fill in the actual HPTE */
 322         /* Set CEC cookie to 0         */
 323         /* Zero page = 0               */
 324         /* I-cache Invalidate = 0      */
 325         /* I-cache synchronize = 0     */
 326         /* Exact = 0                   */
 327         flags = 0;
 328
 329         /* Make pHyp happy */
 330         if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
 331                 hpte_r &= ~_PAGE_COHERENT;
 332         if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
 333                 flags |= H_COALESCE_CAND;
 334
 335         lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
 336         if (unlikely(lpar_rc == H_PTEG_FULL)) {
 337                 if (!(vflags & HPTE_V_BOLTED))
 338                         pr_devel(" full\n");
 339                 return -1;
 340         }
 341
 342         /*
 343          * Since we try and ioremap PHBs we don't own, the pte insert
 344          * will fail. However we must catch the failure in hash_page
 345          * or we will loop forever, so return -2 in this case.
 346          */
 347         if (unlikely(lpar_rc != H_SUCCESS)) {
 348                 if (!(vflags & HPTE_V_BOLTED))
 349                         pr_devel(" lpar err %lu\n", lpar_rc);
 350                 return -2;
 351         }
 352         if (!(vflags & HPTE_V_BOLTED))
 353                 pr_devel(" -> slot: %lu\n", slot & 7);
 354
 355         /* Because of iSeries, we have to pass down the secondary
 356          * bucket bit here as well
 357          */
 358         return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
 359 }
 360
 361 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
 362
 363 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
 364 {
 365         unsigned long slot_offset;
 366         unsigned long lpar_rc;
 367         int i;
 368         unsigned long dummy1, dummy2;
 369
 370         /* pick a random slot to start at */
 371         slot_offset = mftb() & 0x7;
 372
 373         for (i = 0; i < HPTES_PER_GROUP; i++) {
 374
 375                 /* don't remove a bolted entry */
 376                 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
 377                                            (0x1UL << 4), &dummy1, &dummy2);
 378                 if (lpar_rc == H_SUCCESS)
 379                         return i;
 380                 BUG_ON(lpar_rc != H_NOT_FOUND);
 381
 382                 slot_offset++;
 383                 slot_offset &= 0x7;
 384         }
 385
 386         return -1;
 387 }
 388
 389 static void pSeries_lpar_hptab_clear(void)
 390 {
 391         unsigned long size_bytes = 1UL << ppc64_pft_size;
 392         unsigned long hpte_count = size_bytes >> 4;
 393         struct {
 394                 unsigned long pteh;
 395                 unsigned long ptel;
 396         } ptes[4];
 397         long lpar_rc;
 398         int i, j;
 399
 400         /* Read in batches of 4,
 401          * invalidate only valid entries not in the VRMA
 402          * hpte_count will be a multiple of 4
 403          */
 404         for (i = 0; i < hpte_count; i += 4) {
 405                 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
 406                 if (lpar_rc != H_SUCCESS)
 407                         continue;
 408                 for (j = 0; j < 4; j++){
 409                         if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
 410                                 HPTE_V_VRMA_MASK)
 411                                 continue;
 412                         if (ptes[j].pteh & HPTE_V_VALID)
 413                                 plpar_pte_remove_raw(0, i + j, 0,
 414                                         &(ptes[j].pteh), &(ptes[j].ptel));
 415                 }
 416         }
 417 }
 418
 419 /*
 420  * This computes the AVPN and B fields of the first dword of a HPTE,
 421  * for use when we want to match an existing PTE.  The bottom 7 bits
 422  * of the returned value are zero.
 423  */
 424 static inline unsigned long hpte_encode_avpn(unsigned long va, int psize,
 425                                              int ssize)
 426 {
 427         unsigned long v;
 428
 429         v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm);
 430         v <<= HPTE_V_AVPN_SHIFT;
 431         v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
 432         return v;
 433 }
 434
 435 /*
 436  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
 437  * the low 3 bits of flags happen to line up.  So no transform is needed.
 438  * We can probably optimize here and assume the high bits of newpp are
 439  * already zero.  For now I am paranoid.
 440  */
 441 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 442                                        unsigned long newpp,
 443                                        unsigned long va,
 444                                        int psize, int ssize, int local)
 445 {
 446         unsigned long lpar_rc;
 447         unsigned long flags = (newpp & 7) | H_AVPN;
 448         unsigned long want_v;
 449
 450         want_v = hpte_encode_avpn(va, psize, ssize);
 451
 452         pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
 453                  want_v, slot, flags, psize);
 454
 455         lpar_rc = plpar_pte_protect(flags, slot, want_v);
 456
 457         if (lpar_rc == H_NOT_FOUND) {
 458                 pr_devel("not found !\n");
 459                 return -1;
 460         }
 461
 462         pr_devel("ok\n");
 463
 464         BUG_ON(lpar_rc != H_SUCCESS);
 465
 466         return 0;
 467 }
 468
 469 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
 470 {
 471         unsigned long dword0;
 472         unsigned long lpar_rc;
 473         unsigned long dummy_word1;
 474         unsigned long flags;
 475
 476         /* Read 1 pte at a time                        */
 477         /* Do not need RPN to logical page translation */
 478         /* No cross CEC PFT access                     */
 479         flags = 0;
 480
 481         lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
 482
 483         BUG_ON(lpar_rc != H_SUCCESS);
 484
 485         return dword0;
 486 }
 487
 488 static long pSeries_lpar_hpte_find(unsigned long va, int psize, int ssize)
 489 {
 490         unsigned long hash;
 491         unsigned long i;
 492         long slot;
 493         unsigned long want_v, hpte_v;
 494
 495         hash = hpt_hash(va, mmu_psize_defs[psize].shift, ssize);
 496         want_v = hpte_encode_avpn(va, psize, ssize);
 497
 498         /* Bolted entries are always in the primary group */
 499         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 500         for (i = 0; i < HPTES_PER_GROUP; i++) {
 501                 hpte_v = pSeries_lpar_hpte_getword0(slot);
 502
 503                 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
 504                         /* HPTE matches */
 505                         return slot;
 506                 ++slot;
 507         }
 508
 509         return -1;
 510 }
 511
 512 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
 513                                              unsigned long ea,
 514                                              int psize, int ssize)
 515 {
 516         unsigned long lpar_rc, slot, vsid, va, flags;
 517
 518         vsid = get_kernel_vsid(ea, ssize);
 519         va = hpt_va(ea, vsid, ssize);
 520
 521         slot = pSeries_lpar_hpte_find(va, psize, ssize);
 522         BUG_ON(slot == -1);
 523
 524         flags = newpp & 7;
 525         lpar_rc = plpar_pte_protect(flags, slot, 0);
 526
 527         BUG_ON(lpar_rc != H_SUCCESS);
 528 }
 529
 530 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long va,
 531                                          int psize, int ssize, int local)
 532 {
 533         unsigned long want_v;
 534         unsigned long lpar_rc;
 535         unsigned long dummy1, dummy2;
 536
 537         pr_devel("    inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
 538                  slot, va, psize, local);
 539
 540         want_v = hpte_encode_avpn(va, psize, ssize);
 541         lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
 542         if (lpar_rc == H_NOT_FOUND)
 543                 return;
 544
 545         BUG_ON(lpar_rc != H_SUCCESS);
 546 }
 547
 548 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
 549                                            int psize, int ssize)
 550 {
 551         unsigned long slot, vsid, va;
 552
 553         vsid = get_kernel_vsid(ea, ssize);
 554         va = hpt_va(ea, vsid, ssize);
 555
 556         slot = pSeries_lpar_hpte_find(va, psize, ssize);
 557         BUG_ON(slot == -1);
 558
 559         pSeries_lpar_hpte_invalidate(slot, va, psize, ssize, 0);
 560 }
 561
 562 /* Flag bits for H_BULK_REMOVE */
 563 #define HBR_REQUEST     0x4000000000000000UL
 564 #define HBR_RESPONSE    0x8000000000000000UL
 565 #define HBR_END         0xc000000000000000UL
 566 #define HBR_AVPN        0x0200000000000000UL
 567 #define HBR_ANDCOND     0x0100000000000000UL
 568
 569 /*
 570  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
 571  * lock.
 572  */
 573 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
 574 {
 575         unsigned long i, pix, rc;
 576         unsigned long flags = 0;
 577         struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
 578         int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
 579         unsigned long param[9];
 580         unsigned long va;
 581         unsigned long hash, index, shift, hidx, slot;
 582         real_pte_t pte;
 583         int psize, ssize;
 584
 585         if (lock_tlbie)
 586                 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
 587
 588         psize = batch->psize;
 589         ssize = batch->ssize;
 590         pix = 0;
 591         for (i = 0; i < number; i++) {
 592                 va = batch->vaddr[i];
 593                 pte = batch->pte[i];
 594                 pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
 595                         hash = hpt_hash(va, shift, ssize);
 596                         hidx = __rpte_to_hidx(pte, index);
 597                         if (hidx & _PTEIDX_SECONDARY)
 598                                 hash = ~hash;
 599                         slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
 600                         slot += hidx & _PTEIDX_GROUP_IX;
 601                         if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 602                                 pSeries_lpar_hpte_invalidate(slot, va, psize,
 603                                                              ssize, local);
 604                         } else {
 605                                 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
 606                                 param[pix+1] = hpte_encode_avpn(va, psize,
 607                                                                 ssize);
 608                                 pix += 2;
 609                                 if (pix == 8) {
 610                                         rc = plpar_hcall9(H_BULK_REMOVE, param,
 611                                                 param[0], param[1], param[2],
 612                                                 param[3], param[4], param[5],
 613                                                 param[6], param[7]);
 614                                         BUG_ON(rc != H_SUCCESS);
 615                                         pix = 0;
 616                                 }
 617                         }
 618                 } pte_iterate_hashed_end();
 619         }
 620         if (pix) {
 621                 param[pix] = HBR_END;
 622                 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
 623                                   param[2], param[3], param[4], param[5],
 624                                   param[6], param[7]);
 625                 BUG_ON(rc != H_SUCCESS);
 626         }
 627
 628         if (lock_tlbie)
 629                 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
 630 }
 631
 632 static int __init disable_bulk_remove(char *str)
 633 {
 634         if (strcmp(str, "off") == 0 &&
 635             firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
 636                         printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
 637                         powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
 638         }
 639         return 1;
 640 }
 641
 642 __setup("bulk_remove=", disable_bulk_remove);
 643
 644 void __init hpte_init_lpar(void)
 645 {
 646         ppc_md.hpte_invalidate  = pSeries_lpar_hpte_invalidate;
 647         ppc_md.hpte_updatepp    = pSeries_lpar_hpte_updatepp;
 648         ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
 649         ppc_md.hpte_insert      = pSeries_lpar_hpte_insert;
 650         ppc_md.hpte_remove      = pSeries_lpar_hpte_remove;
 651         ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
 652         ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
 653         ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
 654 }
 655
 656 #ifdef CONFIG_PPC_SMLPAR
 657 #define CMO_FREE_HINT_DEFAULT 1
 658 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
 659
 660 static int __init cmo_free_hint(char *str)
 661 {
 662         char *parm;
 663         parm = strstrip(str);
 664
 665         if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
 666                 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
 667                 cmo_free_hint_flag = 0;
 668                 return 1;
 669         }
 670
 671         cmo_free_hint_flag = 1;
 672         printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
 673
 674         if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
 675                 return 1;
 676
 677         return 0;
 678 }
 679
 680 __setup("cmo_free_hint=", cmo_free_hint);
 681
 682 static void pSeries_set_page_state(struct page *page, int order,
 683                                    unsigned long state)
 684 {
 685         int i, j;
 686         unsigned long cmo_page_sz, addr;
 687
 688         cmo_page_sz = cmo_get_page_size();
 689         addr = __pa((unsigned long)page_address(page));
 690
 691         for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
 692                 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
 693                         plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
 694         }
 695 }
 696
 697 void arch_free_page(struct page *page, int order)
 698 {
 699         if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
 700                 return;
 701
 702         pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
 703 }
 704 EXPORT_SYMBOL(arch_free_page);
 705
 706 #endif
 707
 708 #ifdef CONFIG_TRACEPOINTS
 709 /*
 710  * We optimise our hcall path by placing hcall_tracepoint_refcount
 711  * directly in the TOC so we can check if the hcall tracepoints are
 712  * enabled via a single load.
 713  */
 714
 715 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
 716 extern long hcall_tracepoint_refcount;
 717
 718 /*
 719  * Since the tracing code might execute hcalls we need to guard against
 720  * recursion. One example of this are spinlocks calling H_YIELD on
 721  * shared processor partitions.
 722  */
 723 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
 724
 725 void hcall_tracepoint_regfunc(void)
 726 {
 727         hcall_tracepoint_refcount++;
 728 }
 729
 730 void hcall_tracepoint_unregfunc(void)
 731 {
 732         hcall_tracepoint_refcount--;
 733 }
 734
 735 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
 736 {
 737         unsigned long flags;
 738         unsigned int *depth;
 739
 740         local_irq_save(flags);
 741
 742         depth = &__get_cpu_var(hcall_trace_depth);
 743
 744         if (*depth)
 745                 goto out;
 746
 747         (*depth)++;
 748         trace_hcall_entry(opcode, args);
 749         (*depth)--;
 750
 751 out:
 752         local_irq_restore(flags);
 753 }
 754
 755 void __trace_hcall_exit(long opcode, unsigned long retval,
 756                         unsigned long *retbuf)
 757 {
 758         unsigned long flags;
 759         unsigned int *depth;
 760
 761         local_irq_save(flags);
 762
 763         depth = &__get_cpu_var(hcall_trace_depth);
 764
 765         if (*depth)
 766                 goto out;
 767
 768         (*depth)++;
 769         trace_hcall_exit(opcode, retval, retbuf);
 770         (*depth)--;
 771
 772 out:
 773         local_irq_restore(flags);
 774 }
 775 #endif
 776
 777 /**
 778  * h_get_mpp
 779  * H_GET_MPP hcall returns info in 7 parms
 780  */
 781 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
 782 {
 783         int rc;
 784         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
 785
 786         rc = plpar_hcall9(H_GET_MPP, retbuf);
 787
 788         mpp_data->entitled_mem = retbuf[0];
 789         mpp_data->mapped_mem = retbuf[1];
 790
 791         mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
 792         mpp_data->pool_num = retbuf[2] & 0xffff;
 793
 794         mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
 795         mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
 796         mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffff;
 797
 798         mpp_data->pool_size = retbuf[4];
 799         mpp_data->loan_request = retbuf[5];
 800         mpp_data->backing_mem = retbuf[6];
 801
 802         return rc;
 803 }
 804 EXPORT_SYMBOL(h_get_mpp);
 805
 806 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
 807 {
 808         int rc;
 809         unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
 810
 811         rc = plpar_hcall9(H_GET_MPP_X, retbuf);
 812
 813         mpp_x_data->coalesced_bytes = retbuf[0];
 814         mpp_x_data->pool_coalesced_bytes = retbuf[1];
 815         mpp_x_data->pool_purr_cycles = retbuf[2];
 816         mpp_x_data->pool_spurr_cycles = retbuf[3];
 817
 818         return rc;
 819 }