arch/powerpc/platforms/pseries/msi.c

   1 /*
   2  * Copyright 2006 Jake Moilanen <moilanen@austin.ibm.com>, IBM Corp.
   3  * Copyright 2006-2007 Michael Ellerman, IBM Corp.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License
   7  * as published by the Free Software Foundation; version 2 of the
   8  * License.
   9  *
  10  */
  11
  12 #include <linux/device.h>
  13 #include <linux/irq.h>
  14 #include <linux/msi.h>
  15
  16 #include <asm/rtas.h>
  17 #include <asm/hw_irq.h>
  18 #include <asm/ppc-pci.h>
  19 #include <asm/machdep.h>
  20
  21 #include "pseries.h"
  22
  23 static int query_token, change_token;
  24
  25 #define RTAS_QUERY_FN           0
  26 #define RTAS_CHANGE_FN          1
  27 #define RTAS_RESET_FN           2
  28 #define RTAS_CHANGE_MSI_FN      3
  29 #define RTAS_CHANGE_MSIX_FN     4
  30 #define RTAS_CHANGE_32MSI_FN    5
  31
  32 /* RTAS Helpers */
  33
  34 static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
  35 {
  36         u32 addr, seq_num, rtas_ret[3];
  37         unsigned long buid;
  38         int rc;
  39
  40         addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
  41         buid = pdn->phb->buid;
  42
  43         seq_num = 1;
  44         do {
  45                 if (func == RTAS_CHANGE_MSI_FN || func == RTAS_CHANGE_MSIX_FN ||
  46                     func == RTAS_CHANGE_32MSI_FN)
  47                         rc = rtas_call(change_token, 6, 4, rtas_ret, addr,
  48                                         BUID_HI(buid), BUID_LO(buid),
  49                                         func, num_irqs, seq_num);
  50                 else
  51                         rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
  52                                         BUID_HI(buid), BUID_LO(buid),
  53                                         func, num_irqs, seq_num);
  54
  55                 seq_num = rtas_ret[1];
  56         } while (rtas_busy_delay(rc));
  57
  58         /*
  59          * If the RTAS call succeeded, return the number of irqs allocated.
  60          * If not, make sure we return a negative error code.
  61          */
  62         if (rc == 0)
  63                 rc = rtas_ret[0];
  64         else if (rc > 0)
  65                 rc = -rc;
  66
  67         pr_debug("rtas_msi: ibm,change_msi(func=%d,num=%d), got %d rc = %d\n",
  68                  func, num_irqs, rtas_ret[0], rc);
  69
  70         return rc;
  71 }
  72
  73 static void rtas_disable_msi(struct pci_dev *pdev)
  74 {
  75         struct pci_dn *pdn;
  76
  77         pdn = pci_get_pdn(pdev);
  78         if (!pdn)
  79                 return;
  80
  81         /*
  82          * disabling MSI with the explicit interface also disables MSI-X
  83          */
  84         if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
  85                 /*
  86                  * may have failed because explicit interface is not
  87                  * present
  88                  */
  89                 if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
  90                         pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
  91                 }
  92         }
  93 }
  94
  95 static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
  96 {
  97         u32 addr, rtas_ret[2];
  98         unsigned long buid;
  99         int rc;
 100
 101         addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
 102         buid = pdn->phb->buid;
 103
 104         do {
 105                 rc = rtas_call(query_token, 4, 3, rtas_ret, addr,
 106                                BUID_HI(buid), BUID_LO(buid), offset);
 107         } while (rtas_busy_delay(rc));
 108
 109         if (rc) {
 110                 pr_debug("rtas_msi: error (%d) querying source number\n", rc);
 111                 return rc;
 112         }
 113
 114         return rtas_ret[0];
 115 }
 116
 117 static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
 118 {
 119         struct msi_desc *entry;
 120
 121         for_each_pci_msi_entry(entry, pdev) {
 122                 if (!entry->irq)
 123                         continue;
 124
 125                 irq_set_msi_desc(entry->irq, NULL);
 126                 irq_dispose_mapping(entry->irq);
 127         }
 128
 129         rtas_disable_msi(pdev);
 130 }
 131
 132 static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
 133 {
 134         struct device_node *dn;
 135         const __be32 *p;
 136         u32 req_msi;
 137
 138         dn = pci_device_to_OF_node(pdev);
 139
 140         p = of_get_property(dn, prop_name, NULL);
 141         if (!p) {
 142                 pr_debug("rtas_msi: No %s on %pOF\n", prop_name, dn);
 143                 return -ENOENT;
 144         }
 145
 146         req_msi = be32_to_cpup(p);
 147         if (req_msi < nvec) {
 148                 pr_debug("rtas_msi: %s requests < %d MSIs\n", prop_name, nvec);
 149
 150                 if (req_msi == 0) /* Be paranoid */
 151                         return -ENOSPC;
 152
 153                 return req_msi;
 154         }
 155
 156         return 0;
 157 }
 158
 159 static int check_req_msi(struct pci_dev *pdev, int nvec)
 160 {
 161         return check_req(pdev, nvec, "ibm,req#msi");
 162 }
 163
 164 static int check_req_msix(struct pci_dev *pdev, int nvec)
 165 {
 166         return check_req(pdev, nvec, "ibm,req#msi-x");
 167 }
 168
 169 /* Quota calculation */
 170
 171 static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
 172 {
 173         struct device_node *dn;
 174         const __be32 *p;
 175
 176         dn = of_node_get(pci_device_to_OF_node(dev));
 177         while (dn) {
 178                 p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
 179                 if (p) {
 180                         pr_debug("rtas_msi: found prop on dn %pOF\n",
 181                                 dn);
 182                         *total = be32_to_cpup(p);
 183                         return dn;
 184                 }
 185
 186                 dn = of_get_next_parent(dn);
 187         }
 188
 189         return NULL;
 190 }
 191
 192 static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
 193 {
 194         struct device_node *dn;
 195         struct eeh_dev *edev;
 196
 197         /* Found our PE and assume 8 at that point. */
 198
 199         dn = pci_device_to_OF_node(dev);
 200         if (!dn)
 201                 return NULL;
 202
 203         /* Get the top level device in the PE */
 204         edev = pdn_to_eeh_dev(PCI_DN(dn));
 205         if (edev->pe)
 206                 edev = list_first_entry(&edev->pe->edevs, struct eeh_dev, list);
 207         dn = pci_device_to_OF_node(edev->pdev);
 208         if (!dn)
 209                 return NULL;
 210
 211         /* We actually want the parent */
 212         dn = of_get_parent(dn);
 213         if (!dn)
 214                 return NULL;
 215
 216         /* Hardcode of 8 for old firmwares */
 217         *total = 8;
 218         pr_debug("rtas_msi: using PE dn %pOF\n", dn);
 219
 220         return dn;
 221 }
 222
 223 struct msi_counts {
 224         struct device_node *requestor;
 225         int num_devices;
 226         int request;
 227         int quota;
 228         int spare;
 229         int over_quota;
 230 };
 231
 232 static void *count_non_bridge_devices(struct device_node *dn, void *data)
 233 {
 234         struct msi_counts *counts = data;
 235         const __be32 *p;
 236         u32 class;
 237
 238         pr_debug("rtas_msi: counting %pOF\n", dn);
 239
 240         p = of_get_property(dn, "class-code", NULL);
 241         class = p ? be32_to_cpup(p) : 0;
 242
 243         if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
 244                 counts->num_devices++;
 245
 246         return NULL;
 247 }
 248
 249 static void *count_spare_msis(struct device_node *dn, void *data)
 250 {
 251         struct msi_counts *counts = data;
 252         const __be32 *p;
 253         int req;
 254
 255         if (dn == counts->requestor)
 256                 req = counts->request;
 257         else {
 258                 /* We don't know if a driver will try to use MSI or MSI-X,
 259                  * so we just have to punt and use the larger of the two. */
 260                 req = 0;
 261                 p = of_get_property(dn, "ibm,req#msi", NULL);
 262                 if (p)
 263                         req = be32_to_cpup(p);
 264
 265                 p = of_get_property(dn, "ibm,req#msi-x", NULL);
 266                 if (p)
 267                         req = max(req, (int)be32_to_cpup(p));
 268         }
 269
 270         if (req < counts->quota)
 271                 counts->spare += counts->quota - req;
 272         else if (req > counts->quota)
 273                 counts->over_quota++;
 274
 275         return NULL;
 276 }
 277
 278 static int msi_quota_for_device(struct pci_dev *dev, int request)
 279 {
 280         struct device_node *pe_dn;
 281         struct msi_counts counts;
 282         int total;
 283
 284         pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
 285                   request);
 286
 287         pe_dn = find_pe_total_msi(dev, &total);
 288         if (!pe_dn)
 289                 pe_dn = find_pe_dn(dev, &total);
 290
 291         if (!pe_dn) {
 292                 pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
 293                 goto out;
 294         }
 295
 296         pr_debug("rtas_msi: found PE %pOF\n", pe_dn);
 297
 298         memset(&counts, 0, sizeof(struct msi_counts));
 299
 300         /* Work out how many devices we have below this PE */
 301         pci_traverse_device_nodes(pe_dn, count_non_bridge_devices, &counts);
 302
 303         if (counts.num_devices == 0) {
 304                 pr_err("rtas_msi: found 0 devices under PE for %s\n",
 305                         pci_name(dev));
 306                 goto out;
 307         }
 308
 309         counts.quota = total / counts.num_devices;
 310         if (request <= counts.quota)
 311                 goto out;
 312
 313         /* else, we have some more calculating to do */
 314         counts.requestor = pci_device_to_OF_node(dev);
 315         counts.request = request;
 316         pci_traverse_device_nodes(pe_dn, count_spare_msis, &counts);
 317
 318         /* If the quota isn't an integer multiple of the total, we can
 319          * use the remainder as spare MSIs for anyone that wants them. */
 320         counts.spare += total % counts.num_devices;
 321
 322         /* Divide any spare by the number of over-quota requestors */
 323         if (counts.over_quota)
 324                 counts.quota += counts.spare / counts.over_quota;
 325
 326         /* And finally clamp the request to the possibly adjusted quota */
 327         request = min(counts.quota, request);
 328
 329         pr_debug("rtas_msi: request clamped to quota %d\n", request);
 330 out:
 331         of_node_put(pe_dn);
 332
 333         return request;
 334 }
 335
 336 static int check_msix_entries(struct pci_dev *pdev)
 337 {
 338         struct msi_desc *entry;
 339         int expected;
 340
 341         /* There's no way for us to express to firmware that we want
 342          * a discontiguous, or non-zero based, range of MSI-X entries.
 343          * So we must reject such requests. */
 344
 345         expected = 0;
 346         for_each_pci_msi_entry(entry, pdev) {
 347                 if (entry->msi_attrib.entry_nr != expected) {
 348                         pr_debug("rtas_msi: bad MSI-X entries.\n");
 349                         return -EINVAL;
 350                 }
 351                 expected++;
 352         }
 353
 354         return 0;
 355 }
 356
 357 static void rtas_hack_32bit_msi_gen2(struct pci_dev *pdev)
 358 {
 359         u32 addr_hi, addr_lo;
 360
 361         /*
 362          * We should only get in here for IODA1 configs. This is based on the
 363          * fact that we using RTAS for MSIs, we don't have the 32 bit MSI RTAS
 364          * support, and we are in a PCIe Gen2 slot.
 365          */
 366         dev_info(&pdev->dev,
 367                  "rtas_msi: No 32 bit MSI firmware support, forcing 32 bit MSI\n");
 368         pci_read_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, &addr_hi);
 369         addr_lo = 0xffff0000 | ((addr_hi >> (48 - 32)) << 4);
 370         pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_LO, addr_lo);
 371         pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, 0);
 372 }
 373
 374 static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type)
 375 {
 376         struct pci_dn *pdn;
 377         int hwirq, virq, i, quota, rc;
 378         struct msi_desc *entry;
 379         struct msi_msg msg;
 380         int nvec = nvec_in;
 381         int use_32bit_msi_hack = 0;
 382
 383         if (type == PCI_CAP_ID_MSIX)
 384                 rc = check_req_msix(pdev, nvec);
 385         else
 386                 rc = check_req_msi(pdev, nvec);
 387
 388         if (rc)
 389                 return rc;
 390
 391         quota = msi_quota_for_device(pdev, nvec);
 392
 393         if (quota && quota < nvec)
 394                 return quota;
 395
 396         if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
 397                 return -EINVAL;
 398
 399         /*
 400          * Firmware currently refuse any non power of two allocation
 401          * so we round up if the quota will allow it.
 402          */
 403         if (type == PCI_CAP_ID_MSIX) {
 404                 int m = roundup_pow_of_two(nvec);
 405                 quota = msi_quota_for_device(pdev, m);
 406
 407                 if (quota >= m)
 408                         nvec = m;
 409         }
 410
 411         pdn = pci_get_pdn(pdev);
 412
 413         /*
 414          * Try the new more explicit firmware interface, if that fails fall
 415          * back to the old interface. The old interface is known to never
 416          * return MSI-Xs.
 417          */
 418 again:
 419         if (type == PCI_CAP_ID_MSI) {
 420                 if (pdev->no_64bit_msi) {
 421                         rc = rtas_change_msi(pdn, RTAS_CHANGE_32MSI_FN, nvec);
 422                         if (rc < 0) {
 423                                 /*
 424                                  * We only want to run the 32 bit MSI hack below if
 425                                  * the max bus speed is Gen2 speed
 426                                  */
 427                                 if (pdev->bus->max_bus_speed != PCIE_SPEED_5_0GT)
 428                                         return rc;
 429
 430                                 use_32bit_msi_hack = 1;
 431                         }
 432                 } else
 433                         rc = -1;
 434
 435                 if (rc < 0)
 436                         rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
 437
 438                 if (rc < 0) {
 439                         pr_debug("rtas_msi: trying the old firmware call.\n");
 440                         rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
 441                 }
 442
 443                 if (use_32bit_msi_hack && rc > 0)
 444                         rtas_hack_32bit_msi_gen2(pdev);
 445         } else
 446                 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
 447
 448         if (rc != nvec) {
 449                 if (nvec != nvec_in) {
 450                         nvec = nvec_in;
 451                         goto again;
 452                 }
 453                 pr_debug("rtas_msi: rtas_change_msi() failed\n");
 454                 return rc;
 455         }
 456
 457         i = 0;
 458         for_each_pci_msi_entry(entry, pdev) {
 459                 hwirq = rtas_query_irq_number(pdn, i++);
 460                 if (hwirq < 0) {
 461                         pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
 462                         return hwirq;
 463                 }
 464
 465                 virq = irq_create_mapping(NULL, hwirq);
 466
 467                 if (!virq) {
 468                         pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
 469                         return -ENOSPC;
 470                 }
 471
 472                 dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
 473                 irq_set_msi_desc(virq, entry);
 474
 475                 /* Read config space back so we can restore after reset */
 476                 __pci_read_msi_msg(entry, &msg);
 477                 entry->msg = msg;
 478         }
 479
 480         return 0;
 481 }
 482
 483 static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
 484 {
 485         /* No LSI -> leave MSIs (if any) configured */
 486         if (!pdev->irq) {
 487                 dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
 488                 return;
 489         }
 490
 491         /* No MSI -> MSIs can't have been assigned by fw, leave LSI */
 492         if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
 493                 dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
 494                 return;
 495         }
 496
 497         dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
 498         rtas_disable_msi(pdev);
 499 }
 500
 501 static int rtas_msi_init(void)
 502 {
 503         struct pci_controller *phb;
 504
 505         query_token  = rtas_token("ibm,query-interrupt-source-number");
 506         change_token = rtas_token("ibm,change-msi");
 507
 508         if ((query_token == RTAS_UNKNOWN_SERVICE) ||
 509                         (change_token == RTAS_UNKNOWN_SERVICE)) {
 510                 pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
 511                 return -1;
 512         }
 513
 514         pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
 515
 516         WARN_ON(pseries_pci_controller_ops.setup_msi_irqs);
 517         pseries_pci_controller_ops.setup_msi_irqs = rtas_setup_msi_irqs;
 518         pseries_pci_controller_ops.teardown_msi_irqs = rtas_teardown_msi_irqs;
 519
 520         list_for_each_entry(phb, &hose_list, list_node) {
 521                 WARN_ON(phb->controller_ops.setup_msi_irqs);
 522                 phb->controller_ops.setup_msi_irqs = rtas_setup_msi_irqs;
 523                 phb->controller_ops.teardown_msi_irqs = rtas_teardown_msi_irqs;
 524         }
 525
 526         WARN_ON(ppc_md.pci_irq_fixup);
 527         ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
 528
 529         return 0;
 530 }
 531 machine_arch_initcall(pseries, rtas_msi_init);