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