hw/ppc/spapr_iommu.c

   1 /*
   2  * QEMU sPAPR IOMMU (TCE) code
   3  *
   4  * Copyright (c) 2010 David Gibson, IBM Corporation <dwg@au1.ibm.com>
   5  *
   6  * This library is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This library is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with this library; if not, see <http://www.gnu.org/licenses/>.
  18  */
  19 #include "hw/hw.h"
  20 #include "sysemu/kvm.h"
  21 #include "hw/qdev.h"
  22 #include "kvm_ppc.h"
  23 #include "sysemu/dma.h"
  24 #include "exec/address-spaces.h"
  25 #include "trace.h"
  26
  27 #include "hw/ppc/spapr.h"
  28 #include "hw/ppc/spapr_vio.h"
  29
  30 #include <libfdt.h>
  31
  32 enum sPAPRTCEAccess {
  33     SPAPR_TCE_FAULT = 0,
  34     SPAPR_TCE_RO = 1,
  35     SPAPR_TCE_WO = 2,
  36     SPAPR_TCE_RW = 3,
  37 };
  38
  39 #define IOMMU_PAGE_SIZE(shift)      (1ULL << (shift))
  40 #define IOMMU_PAGE_MASK(shift)      (~(IOMMU_PAGE_SIZE(shift) - 1))
  41
  42 static QLIST_HEAD(spapr_tce_tables, sPAPRTCETable) spapr_tce_tables;
  43
  44 sPAPRTCETable *spapr_tce_find_by_liobn(target_ulong liobn)
  45 {
  46     sPAPRTCETable *tcet;
  47
  48     if (liobn & 0xFFFFFFFF00000000ULL) {
  49         hcall_dprintf("Request for out-of-bounds LIOBN 0x" TARGET_FMT_lx "\n",
  50                       liobn);
  51         return NULL;
  52     }
  53
  54     QLIST_FOREACH(tcet, &spapr_tce_tables, list) {
  55         if (tcet->liobn == (uint32_t)liobn) {
  56             return tcet;
  57         }
  58     }
  59
  60     return NULL;
  61 }
  62
  63 static IOMMUAccessFlags spapr_tce_iommu_access_flags(uint64_t tce)
  64 {
  65     switch (tce & SPAPR_TCE_RW) {
  66     case SPAPR_TCE_FAULT:
  67         return IOMMU_NONE;
  68     case SPAPR_TCE_RO:
  69         return IOMMU_RO;
  70     case SPAPR_TCE_WO:
  71         return IOMMU_WO;
  72     default: /* SPAPR_TCE_RW */
  73         return IOMMU_RW;
  74     }
  75 }
  76
  77 /* Called from RCU critical section */
  78 static IOMMUTLBEntry spapr_tce_translate_iommu(MemoryRegion *iommu, hwaddr addr,
  79                                                bool is_write)
  80 {
  81     sPAPRTCETable *tcet = container_of(iommu, sPAPRTCETable, iommu);
  82     uint64_t tce;
  83     IOMMUTLBEntry ret = {
  84         .target_as = &address_space_memory,
  85         .iova = 0,
  86         .translated_addr = 0,
  87         .addr_mask = ~(hwaddr)0,
  88         .perm = IOMMU_NONE,
  89     };
  90
  91     if ((addr >> tcet->page_shift) < tcet->nb_table) {
  92         /* Check if we are in bound */
  93         hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
  94
  95         tce = tcet->table[addr >> tcet->page_shift];
  96         ret.iova = addr & page_mask;
  97         ret.translated_addr = tce & page_mask;
  98         ret.addr_mask = ~page_mask;
  99         ret.perm = spapr_tce_iommu_access_flags(tce);
 100     }
 101     trace_spapr_iommu_xlate(tcet->liobn, addr, ret.iova, ret.perm,
 102                             ret.addr_mask);
 103
 104     return ret;
 105 }
 106
 107 static int spapr_tce_table_post_load(void *opaque, int version_id)
 108 {
 109     sPAPRTCETable *tcet = SPAPR_TCE_TABLE(opaque);
 110
 111     if (tcet->vdev) {
 112         spapr_vio_set_bypass(tcet->vdev, tcet->bypass);
 113     }
 114
 115     return 0;
 116 }
 117
 118 static const VMStateDescription vmstate_spapr_tce_table = {
 119     .name = "spapr_iommu",
 120     .version_id = 2,
 121     .minimum_version_id = 2,
 122     .post_load = spapr_tce_table_post_load,
 123     .fields      = (VMStateField []) {
 124         /* Sanity check */
 125         VMSTATE_UINT32_EQUAL(liobn, sPAPRTCETable),
 126         VMSTATE_UINT32_EQUAL(nb_table, sPAPRTCETable),
 127
 128         /* IOMMU state */
 129         VMSTATE_BOOL(bypass, sPAPRTCETable),
 130         VMSTATE_VARRAY_UINT32(table, sPAPRTCETable, nb_table, 0, vmstate_info_uint64, uint64_t),
 131
 132         VMSTATE_END_OF_LIST()
 133     },
 134 };
 135
 136 static MemoryRegionIOMMUOps spapr_iommu_ops = {
 137     .translate = spapr_tce_translate_iommu,
 138 };
 139
 140 static int spapr_tce_table_realize(DeviceState *dev)
 141 {
 142     sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
 143     uint64_t window_size = (uint64_t)tcet->nb_table << tcet->page_shift;
 144
 145     if (kvm_enabled() && !(window_size >> 32)) {
 146         tcet->table = kvmppc_create_spapr_tce(tcet->liobn,
 147                                               window_size,
 148                                               &tcet->fd,
 149                                               tcet->need_vfio);
 150     }
 151
 152     if (!tcet->table) {
 153         size_t table_size = tcet->nb_table * sizeof(uint64_t);
 154         tcet->table = g_malloc0(table_size);
 155     }
 156
 157     trace_spapr_iommu_new_table(tcet->liobn, tcet, tcet->table, tcet->fd);
 158
 159     memory_region_init_iommu(&tcet->iommu, OBJECT(dev), &spapr_iommu_ops,
 160                              "iommu-spapr",
 161                              (uint64_t)tcet->nb_table << tcet->page_shift);
 162
 163     QLIST_INSERT_HEAD(&spapr_tce_tables, tcet, list);
 164
 165     vmstate_register(DEVICE(tcet), tcet->liobn, &vmstate_spapr_tce_table,
 166                      tcet);
 167
 168     return 0;
 169 }
 170
 171 void spapr_tce_set_need_vfio(sPAPRTCETable *tcet, bool need_vfio)
 172 {
 173     size_t table_size = tcet->nb_table * sizeof(uint64_t);
 174     void *newtable;
 175
 176     if (need_vfio == tcet->need_vfio) {
 177         /* Nothing to do */
 178         return;
 179     }
 180
 181     if (!need_vfio) {
 182         /* FIXME: We don't support transition back to KVM accelerated
 183          * TCEs yet */
 184         return;
 185     }
 186
 187     tcet->need_vfio = true;
 188
 189     if (tcet->fd < 0) {
 190         /* Table is already in userspace, nothing to be do */
 191         return;
 192     }
 193
 194     newtable = g_malloc(table_size);
 195     memcpy(newtable, tcet->table, table_size);
 196
 197     kvmppc_remove_spapr_tce(tcet->table, tcet->fd, tcet->nb_table);
 198
 199     tcet->fd = -1;
 200     tcet->table = newtable;
 201 }
 202
 203 sPAPRTCETable *spapr_tce_new_table(DeviceState *owner, uint32_t liobn,
 204                                    uint64_t bus_offset,
 205                                    uint32_t page_shift,
 206                                    uint32_t nb_table,
 207                                    bool need_vfio)
 208 {
 209     sPAPRTCETable *tcet;
 210     char tmp[64];
 211
 212     if (spapr_tce_find_by_liobn(liobn)) {
 213         fprintf(stderr, "Attempted to create TCE table with duplicate"
 214                 " LIOBN 0x%x\n", liobn);
 215         return NULL;
 216     }
 217
 218     if (!nb_table) {
 219         return NULL;
 220     }
 221
 222     tcet = SPAPR_TCE_TABLE(object_new(TYPE_SPAPR_TCE_TABLE));
 223     tcet->liobn = liobn;
 224     tcet->bus_offset = bus_offset;
 225     tcet->page_shift = page_shift;
 226     tcet->nb_table = nb_table;
 227     tcet->need_vfio = need_vfio;
 228
 229     snprintf(tmp, sizeof(tmp), "tce-table-%x", liobn);
 230     object_property_add_child(OBJECT(owner), tmp, OBJECT(tcet), NULL);
 231
 232     object_property_set_bool(OBJECT(tcet), true, "realized", NULL);
 233
 234     return tcet;
 235 }
 236
 237 static void spapr_tce_table_unrealize(DeviceState *dev, Error **errp)
 238 {
 239     sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
 240
 241     QLIST_REMOVE(tcet, list);
 242
 243     if (!kvm_enabled() ||
 244         (kvmppc_remove_spapr_tce(tcet->table, tcet->fd,
 245                                  tcet->nb_table) != 0)) {
 246         g_free(tcet->table);
 247     }
 248 }
 249
 250 MemoryRegion *spapr_tce_get_iommu(sPAPRTCETable *tcet)
 251 {
 252     return &tcet->iommu;
 253 }
 254
 255 static void spapr_tce_reset(DeviceState *dev)
 256 {
 257     sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
 258     size_t table_size = tcet->nb_table * sizeof(uint64_t);
 259
 260     memset(tcet->table, 0, table_size);
 261 }
 262
 263 static target_ulong put_tce_emu(sPAPRTCETable *tcet, target_ulong ioba,
 264                                 target_ulong tce)
 265 {
 266     IOMMUTLBEntry entry;
 267     hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
 268     unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift;
 269
 270     if (index >= tcet->nb_table) {
 271         hcall_dprintf("spapr_vio_put_tce on out-of-bounds IOBA 0x"
 272                       TARGET_FMT_lx "\n", ioba);
 273         return H_PARAMETER;
 274     }
 275
 276     tcet->table[index] = tce;
 277
 278     entry.target_as = &address_space_memory,
 279     entry.iova = ioba & page_mask;
 280     entry.translated_addr = tce & page_mask;
 281     entry.addr_mask = ~page_mask;
 282     entry.perm = spapr_tce_iommu_access_flags(tce);
 283     memory_region_notify_iommu(&tcet->iommu, entry);
 284
 285     return H_SUCCESS;
 286 }
 287
 288 static target_ulong h_put_tce_indirect(PowerPCCPU *cpu,
 289                                        sPAPRMachineState *spapr,
 290                                        target_ulong opcode, target_ulong *args)
 291 {
 292     int i;
 293     target_ulong liobn = args[0];
 294     target_ulong ioba = args[1];
 295     target_ulong ioba1 = ioba;
 296     target_ulong tce_list = args[2];
 297     target_ulong npages = args[3];
 298     target_ulong ret = H_PARAMETER, tce = 0;
 299     sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
 300     CPUState *cs = CPU(cpu);
 301     hwaddr page_mask, page_size;
 302
 303     if (!tcet) {
 304         return H_PARAMETER;
 305     }
 306
 307     if ((npages > 512) || (tce_list & SPAPR_TCE_PAGE_MASK)) {
 308         return H_PARAMETER;
 309     }
 310
 311     page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
 312     page_size = IOMMU_PAGE_SIZE(tcet->page_shift);
 313     ioba &= page_mask;
 314
 315     for (i = 0; i < npages; ++i, ioba += page_size) {
 316         tce = ldq_be_phys(cs->as, tce_list + i * sizeof(target_ulong));
 317
 318         ret = put_tce_emu(tcet, ioba, tce);
 319         if (ret) {
 320             break;
 321         }
 322     }
 323
 324     /* Trace last successful or the first problematic entry */
 325     i = i ? (i - 1) : 0;
 326     if (SPAPR_IS_PCI_LIOBN(liobn)) {
 327         trace_spapr_iommu_pci_indirect(liobn, ioba1, tce_list, i, tce, ret);
 328     } else {
 329         trace_spapr_iommu_indirect(liobn, ioba1, tce_list, i, tce, ret);
 330     }
 331     return ret;
 332 }
 333
 334 static target_ulong h_stuff_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 335                               target_ulong opcode, target_ulong *args)
 336 {
 337     int i;
 338     target_ulong liobn = args[0];
 339     target_ulong ioba = args[1];
 340     target_ulong tce_value = args[2];
 341     target_ulong npages = args[3];
 342     target_ulong ret = H_PARAMETER;
 343     sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
 344     hwaddr page_mask, page_size;
 345
 346     if (!tcet) {
 347         return H_PARAMETER;
 348     }
 349
 350     if (npages > tcet->nb_table) {
 351         return H_PARAMETER;
 352     }
 353
 354     page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
 355     page_size = IOMMU_PAGE_SIZE(tcet->page_shift);
 356     ioba &= page_mask;
 357
 358     for (i = 0; i < npages; ++i, ioba += page_size) {
 359         ret = put_tce_emu(tcet, ioba, tce_value);
 360         if (ret) {
 361             break;
 362         }
 363     }
 364     if (SPAPR_IS_PCI_LIOBN(liobn)) {
 365         trace_spapr_iommu_pci_stuff(liobn, ioba, tce_value, npages, ret);
 366     } else {
 367         trace_spapr_iommu_stuff(liobn, ioba, tce_value, npages, ret);
 368     }
 369
 370     return ret;
 371 }
 372
 373 static target_ulong h_put_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 374                               target_ulong opcode, target_ulong *args)
 375 {
 376     target_ulong liobn = args[0];
 377     target_ulong ioba = args[1];
 378     target_ulong tce = args[2];
 379     target_ulong ret = H_PARAMETER;
 380     sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
 381
 382     if (tcet) {
 383         hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
 384
 385         ioba &= page_mask;
 386
 387         ret = put_tce_emu(tcet, ioba, tce);
 388     }
 389     if (SPAPR_IS_PCI_LIOBN(liobn)) {
 390         trace_spapr_iommu_pci_put(liobn, ioba, tce, ret);
 391     } else {
 392         trace_spapr_iommu_put(liobn, ioba, tce, ret);
 393     }
 394
 395     return ret;
 396 }
 397
 398 static target_ulong get_tce_emu(sPAPRTCETable *tcet, target_ulong ioba,
 399                                 target_ulong *tce)
 400 {
 401     unsigned long index = (ioba - tcet->bus_offset) >> tcet->page_shift;
 402
 403     if (index >= tcet->nb_table) {
 404         hcall_dprintf("spapr_iommu_get_tce on out-of-bounds IOBA 0x"
 405                       TARGET_FMT_lx "\n", ioba);
 406         return H_PARAMETER;
 407     }
 408
 409     *tce = tcet->table[index];
 410
 411     return H_SUCCESS;
 412 }
 413
 414 static target_ulong h_get_tce(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 415                               target_ulong opcode, target_ulong *args)
 416 {
 417     target_ulong liobn = args[0];
 418     target_ulong ioba = args[1];
 419     target_ulong tce = 0;
 420     target_ulong ret = H_PARAMETER;
 421     sPAPRTCETable *tcet = spapr_tce_find_by_liobn(liobn);
 422
 423     if (tcet) {
 424         hwaddr page_mask = IOMMU_PAGE_MASK(tcet->page_shift);
 425
 426         ioba &= page_mask;
 427
 428         ret = get_tce_emu(tcet, ioba, &tce);
 429         if (!ret) {
 430             args[0] = tce;
 431         }
 432     }
 433     if (SPAPR_IS_PCI_LIOBN(liobn)) {
 434         trace_spapr_iommu_pci_get(liobn, ioba, ret, tce);
 435     } else {
 436         trace_spapr_iommu_get(liobn, ioba, ret, tce);
 437     }
 438
 439     return ret;
 440 }
 441
 442 int spapr_dma_dt(void *fdt, int node_off, const char *propname,
 443                  uint32_t liobn, uint64_t window, uint32_t size)
 444 {
 445     uint32_t dma_prop[5];
 446     int ret;
 447
 448     dma_prop[0] = cpu_to_be32(liobn);
 449     dma_prop[1] = cpu_to_be32(window >> 32);
 450     dma_prop[2] = cpu_to_be32(window & 0xFFFFFFFF);
 451     dma_prop[3] = 0; /* window size is 32 bits */
 452     dma_prop[4] = cpu_to_be32(size);
 453
 454     ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-address-cells", 2);
 455     if (ret < 0) {
 456         return ret;
 457     }
 458
 459     ret = fdt_setprop_cell(fdt, node_off, "ibm,#dma-size-cells", 2);
 460     if (ret < 0) {
 461         return ret;
 462     }
 463
 464     ret = fdt_setprop(fdt, node_off, propname, dma_prop, sizeof(dma_prop));
 465     if (ret < 0) {
 466         return ret;
 467     }
 468
 469     return 0;
 470 }
 471
 472 int spapr_tcet_dma_dt(void *fdt, int node_off, const char *propname,
 473                       sPAPRTCETable *tcet)
 474 {
 475     if (!tcet) {
 476         return 0;
 477     }
 478
 479     return spapr_dma_dt(fdt, node_off, propname,
 480                         tcet->liobn, 0, tcet->nb_table << tcet->page_shift);
 481 }
 482
 483 static void spapr_tce_table_class_init(ObjectClass *klass, void *data)
 484 {
 485     DeviceClass *dc = DEVICE_CLASS(klass);
 486     dc->init = spapr_tce_table_realize;
 487     dc->reset = spapr_tce_reset;
 488     dc->unrealize = spapr_tce_table_unrealize;
 489
 490     QLIST_INIT(&spapr_tce_tables);
 491
 492     /* hcall-tce */
 493     spapr_register_hypercall(H_PUT_TCE, h_put_tce);
 494     spapr_register_hypercall(H_GET_TCE, h_get_tce);
 495     spapr_register_hypercall(H_PUT_TCE_INDIRECT, h_put_tce_indirect);
 496     spapr_register_hypercall(H_STUFF_TCE, h_stuff_tce);
 497 }
 498
 499 static TypeInfo spapr_tce_table_info = {
 500     .name = TYPE_SPAPR_TCE_TABLE,
 501     .parent = TYPE_DEVICE,
 502     .instance_size = sizeof(sPAPRTCETable),
 503     .class_init = spapr_tce_table_class_init,
 504 };
 505
 506 static void register_types(void)
 507 {
 508     type_register_static(&spapr_tce_table_info);
 509 }
 510
 511 type_init(register_types);