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