1 /* 2 * The file intends to implement the platform dependent EEH operations on 3 * powernv platform. Actually, the powernv was created in order to fully 4 * hypervisor support. 5 * 6 * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013. 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation; either version 2 of the License, or 11 * (at your option) any later version. 12 */ 13 14 #include <linux/atomic.h> 15 #include <linux/delay.h> 16 #include <linux/export.h> 17 #include <linux/init.h> 18 #include <linux/list.h> 19 #include <linux/msi.h> 20 #include <linux/of.h> 21 #include <linux/pci.h> 22 #include <linux/proc_fs.h> 23 #include <linux/rbtree.h> 24 #include <linux/sched.h> 25 #include <linux/seq_file.h> 26 #include <linux/spinlock.h> 27 28 #include <asm/eeh.h> 29 #include <asm/eeh_event.h> 30 #include <asm/firmware.h> 31 #include <asm/io.h> 32 #include <asm/iommu.h> 33 #include <asm/machdep.h> 34 #include <asm/msi_bitmap.h> 35 #include <asm/opal.h> 36 #include <asm/ppc-pci.h> 37 38 #include "powernv.h" 39 #include "pci.h" 40 41 /** 42 * powernv_eeh_init - EEH platform dependent initialization 43 * 44 * EEH platform dependent initialization on powernv 45 */ 46 static int powernv_eeh_init(void) 47 { 48 struct pci_controller *hose; 49 struct pnv_phb *phb; 50 51 /* We require OPALv3 */ 52 if (!firmware_has_feature(FW_FEATURE_OPALv3)) { 53 pr_warn("%s: OPALv3 is required !\n", 54 __func__); 55 return -EINVAL; 56 } 57 58 /* Set probe mode */ 59 eeh_add_flag(EEH_PROBE_MODE_DEV); 60 61 /* 62 * P7IOC blocks PCI config access to frozen PE, but PHB3 63 * doesn't do that. So we have to selectively enable I/O 64 * prior to collecting error log. 65 */ 66 list_for_each_entry(hose, &hose_list, list_node) { 67 phb = hose->private_data; 68 69 if (phb->model == PNV_PHB_MODEL_P7IOC) 70 eeh_add_flag(EEH_ENABLE_IO_FOR_LOG); 71 break; 72 } 73 74 return 0; 75 } 76 77 /** 78 * powernv_eeh_post_init - EEH platform dependent post initialization 79 * 80 * EEH platform dependent post initialization on powernv. When 81 * the function is called, the EEH PEs and devices should have 82 * been built. If the I/O cache staff has been built, EEH is 83 * ready to supply service. 84 */ 85 static int powernv_eeh_post_init(void) 86 { 87 struct pci_controller *hose; 88 struct pnv_phb *phb; 89 int ret = 0; 90 91 list_for_each_entry(hose, &hose_list, list_node) { 92 phb = hose->private_data; 93 94 if (phb->eeh_ops && phb->eeh_ops->post_init) { 95 ret = phb->eeh_ops->post_init(hose); 96 if (ret) 97 break; 98 } 99 } 100 101 return ret; 102 } 103 104 /** 105 * powernv_eeh_dev_probe - Do probe on PCI device 106 * @dev: PCI device 107 * @flag: unused 108 * 109 * When EEH module is installed during system boot, all PCI devices 110 * are checked one by one to see if it supports EEH. The function 111 * is introduced for the purpose. By default, EEH has been enabled 112 * on all PCI devices. That's to say, we only need do necessary 113 * initialization on the corresponding eeh device and create PE 114 * accordingly. 115 * 116 * It's notable that's unsafe to retrieve the EEH device through 117 * the corresponding PCI device. During the PCI device hotplug, which 118 * was possiblly triggered by EEH core, the binding between EEH device 119 * and the PCI device isn't built yet. 120 */ 121 static int powernv_eeh_dev_probe(struct pci_dev *dev, void *flag) 122 { 123 struct pci_controller *hose = pci_bus_to_host(dev->bus); 124 struct pnv_phb *phb = hose->private_data; 125 struct device_node *dn = pci_device_to_OF_node(dev); 126 struct eeh_dev *edev = of_node_to_eeh_dev(dn); 127 int ret; 128 129 /* 130 * When probing the root bridge, which doesn't have any 131 * subordinate PCI devices. We don't have OF node for 132 * the root bridge. So it's not reasonable to continue 133 * the probing. 134 */ 135 if (!dn || !edev || edev->pe) 136 return 0; 137 138 /* Skip for PCI-ISA bridge */ 139 if ((dev->class >> 8) == PCI_CLASS_BRIDGE_ISA) 140 return 0; 141 142 /* Initialize eeh device */ 143 edev->class_code = dev->class; 144 edev->mode &= 0xFFFFFF00; 145 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) 146 edev->mode |= EEH_DEV_BRIDGE; 147 edev->pcix_cap = pci_find_capability(dev, PCI_CAP_ID_PCIX); 148 if (pci_is_pcie(dev)) { 149 edev->pcie_cap = pci_pcie_cap(dev); 150 151 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) 152 edev->mode |= EEH_DEV_ROOT_PORT; 153 else if (pci_pcie_type(dev) == PCI_EXP_TYPE_DOWNSTREAM) 154 edev->mode |= EEH_DEV_DS_PORT; 155 156 edev->aer_cap = pci_find_ext_capability(dev, 157 PCI_EXT_CAP_ID_ERR); 158 } 159 160 edev->config_addr = ((dev->bus->number << 8) | dev->devfn); 161 edev->pe_config_addr = phb->bdfn_to_pe(phb, dev->bus, dev->devfn & 0xff); 162 163 /* Create PE */ 164 ret = eeh_add_to_parent_pe(edev); 165 if (ret) { 166 pr_warn("%s: Can't add PCI dev %s to parent PE (%d)\n", 167 __func__, pci_name(dev), ret); 168 return ret; 169 } 170 171 /* 172 * Cache the PE primary bus, which can't be fetched when 173 * full hotplug is in progress. In that case, all child 174 * PCI devices of the PE are expected to be removed prior 175 * to PE reset. 176 */ 177 if (!edev->pe->bus) 178 edev->pe->bus = dev->bus; 179 180 /* 181 * Enable EEH explicitly so that we will do EEH check 182 * while accessing I/O stuff 183 */ 184 eeh_add_flag(EEH_ENABLED); 185 186 /* Save memory bars */ 187 eeh_save_bars(edev); 188 189 return 0; 190 } 191 192 /** 193 * powernv_eeh_set_option - Initialize EEH or MMIO/DMA reenable 194 * @pe: EEH PE 195 * @option: operation to be issued 196 * 197 * The function is used to control the EEH functionality globally. 198 * Currently, following options are support according to PAPR: 199 * Enable EEH, Disable EEH, Enable MMIO and Enable DMA 200 */ 201 static int powernv_eeh_set_option(struct eeh_pe *pe, int option) 202 { 203 struct pci_controller *hose = pe->phb; 204 struct pnv_phb *phb = hose->private_data; 205 int ret = -EEXIST; 206 207 /* 208 * What we need do is pass it down for hardware 209 * implementation to handle it. 210 */ 211 if (phb->eeh_ops && phb->eeh_ops->set_option) 212 ret = phb->eeh_ops->set_option(pe, option); 213 214 return ret; 215 } 216 217 /** 218 * powernv_eeh_get_pe_addr - Retrieve PE address 219 * @pe: EEH PE 220 * 221 * Retrieve the PE address according to the given tranditional 222 * PCI BDF (Bus/Device/Function) address. 223 */ 224 static int powernv_eeh_get_pe_addr(struct eeh_pe *pe) 225 { 226 return pe->addr; 227 } 228 229 /** 230 * powernv_eeh_get_state - Retrieve PE state 231 * @pe: EEH PE 232 * @delay: delay while PE state is temporarily unavailable 233 * 234 * Retrieve the state of the specified PE. For IODA-compitable 235 * platform, it should be retrieved from IODA table. Therefore, 236 * we prefer passing down to hardware implementation to handle 237 * it. 238 */ 239 static int powernv_eeh_get_state(struct eeh_pe *pe, int *delay) 240 { 241 struct pci_controller *hose = pe->phb; 242 struct pnv_phb *phb = hose->private_data; 243 int ret = EEH_STATE_NOT_SUPPORT; 244 245 if (phb->eeh_ops && phb->eeh_ops->get_state) { 246 ret = phb->eeh_ops->get_state(pe); 247 248 /* 249 * If the PE state is temporarily unavailable, 250 * to inform the EEH core delay for default 251 * period (1 second) 252 */ 253 if (delay) { 254 *delay = 0; 255 if (ret & EEH_STATE_UNAVAILABLE) 256 *delay = 1000; 257 } 258 } 259 260 return ret; 261 } 262 263 /** 264 * powernv_eeh_reset - Reset the specified PE 265 * @pe: EEH PE 266 * @option: reset option 267 * 268 * Reset the specified PE 269 */ 270 static int powernv_eeh_reset(struct eeh_pe *pe, int option) 271 { 272 struct pci_controller *hose = pe->phb; 273 struct pnv_phb *phb = hose->private_data; 274 int ret = -EEXIST; 275 276 if (phb->eeh_ops && phb->eeh_ops->reset) 277 ret = phb->eeh_ops->reset(pe, option); 278 279 return ret; 280 } 281 282 /** 283 * powernv_eeh_wait_state - Wait for PE state 284 * @pe: EEH PE 285 * @max_wait: maximal period in microsecond 286 * 287 * Wait for the state of associated PE. It might take some time 288 * to retrieve the PE's state. 289 */ 290 static int powernv_eeh_wait_state(struct eeh_pe *pe, int max_wait) 291 { 292 int ret; 293 int mwait; 294 295 while (1) { 296 ret = powernv_eeh_get_state(pe, &mwait); 297 298 /* 299 * If the PE's state is temporarily unavailable, 300 * we have to wait for the specified time. Otherwise, 301 * the PE's state will be returned immediately. 302 */ 303 if (ret != EEH_STATE_UNAVAILABLE) 304 return ret; 305 306 max_wait -= mwait; 307 if (max_wait <= 0) { 308 pr_warn("%s: Timeout getting PE#%x's state (%d)\n", 309 __func__, pe->addr, max_wait); 310 return EEH_STATE_NOT_SUPPORT; 311 } 312 313 msleep(mwait); 314 } 315 316 return EEH_STATE_NOT_SUPPORT; 317 } 318 319 /** 320 * powernv_eeh_get_log - Retrieve error log 321 * @pe: EEH PE 322 * @severity: temporary or permanent error log 323 * @drv_log: driver log to be combined with retrieved error log 324 * @len: length of driver log 325 * 326 * Retrieve the temporary or permanent error from the PE. 327 */ 328 static int powernv_eeh_get_log(struct eeh_pe *pe, int severity, 329 char *drv_log, unsigned long len) 330 { 331 struct pci_controller *hose = pe->phb; 332 struct pnv_phb *phb = hose->private_data; 333 int ret = -EEXIST; 334 335 if (phb->eeh_ops && phb->eeh_ops->get_log) 336 ret = phb->eeh_ops->get_log(pe, severity, drv_log, len); 337 338 return ret; 339 } 340 341 /** 342 * powernv_eeh_configure_bridge - Configure PCI bridges in the indicated PE 343 * @pe: EEH PE 344 * 345 * The function will be called to reconfigure the bridges included 346 * in the specified PE so that the mulfunctional PE would be recovered 347 * again. 348 */ 349 static int powernv_eeh_configure_bridge(struct eeh_pe *pe) 350 { 351 struct pci_controller *hose = pe->phb; 352 struct pnv_phb *phb = hose->private_data; 353 int ret = 0; 354 355 if (phb->eeh_ops && phb->eeh_ops->configure_bridge) 356 ret = phb->eeh_ops->configure_bridge(pe); 357 358 return ret; 359 } 360 361 /** 362 * powernv_pe_err_inject - Inject specified error to the indicated PE 363 * @pe: the indicated PE 364 * @type: error type 365 * @func: specific error type 366 * @addr: address 367 * @mask: address mask 368 * 369 * The routine is called to inject specified error, which is 370 * determined by @type and @func, to the indicated PE for 371 * testing purpose. 372 */ 373 static int powernv_eeh_err_inject(struct eeh_pe *pe, int type, int func, 374 unsigned long addr, unsigned long mask) 375 { 376 struct pci_controller *hose = pe->phb; 377 struct pnv_phb *phb = hose->private_data; 378 int ret = -EEXIST; 379 380 if (phb->eeh_ops && phb->eeh_ops->err_inject) 381 ret = phb->eeh_ops->err_inject(pe, type, func, addr, mask); 382 383 return ret; 384 } 385 386 /** 387 * powernv_eeh_next_error - Retrieve next EEH error to handle 388 * @pe: Affected PE 389 * 390 * Using OPAL API, to retrieve next EEH error for EEH core to handle 391 */ 392 static int powernv_eeh_next_error(struct eeh_pe **pe) 393 { 394 struct pci_controller *hose; 395 struct pnv_phb *phb = NULL; 396 397 list_for_each_entry(hose, &hose_list, list_node) { 398 phb = hose->private_data; 399 break; 400 } 401 402 if (phb && phb->eeh_ops->next_error) 403 return phb->eeh_ops->next_error(pe); 404 405 return -EEXIST; 406 } 407 408 static int powernv_eeh_restore_config(struct device_node *dn) 409 { 410 struct eeh_dev *edev = of_node_to_eeh_dev(dn); 411 struct pnv_phb *phb; 412 s64 ret; 413 414 if (!edev) 415 return -EEXIST; 416 417 phb = edev->phb->private_data; 418 ret = opal_pci_reinit(phb->opal_id, 419 OPAL_REINIT_PCI_DEV, edev->config_addr); 420 if (ret) { 421 pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n", 422 __func__, edev->config_addr, ret); 423 return -EIO; 424 } 425 426 return 0; 427 } 428 429 static struct eeh_ops powernv_eeh_ops = { 430 .name = "powernv", 431 .init = powernv_eeh_init, 432 .post_init = powernv_eeh_post_init, 433 .of_probe = NULL, 434 .dev_probe = powernv_eeh_dev_probe, 435 .set_option = powernv_eeh_set_option, 436 .get_pe_addr = powernv_eeh_get_pe_addr, 437 .get_state = powernv_eeh_get_state, 438 .reset = powernv_eeh_reset, 439 .wait_state = powernv_eeh_wait_state, 440 .get_log = powernv_eeh_get_log, 441 .configure_bridge = powernv_eeh_configure_bridge, 442 .err_inject = powernv_eeh_err_inject, 443 .read_config = pnv_pci_cfg_read, 444 .write_config = pnv_pci_cfg_write, 445 .next_error = powernv_eeh_next_error, 446 .restore_config = powernv_eeh_restore_config 447 }; 448 449 /** 450 * eeh_powernv_init - Register platform dependent EEH operations 451 * 452 * EEH initialization on powernv platform. This function should be 453 * called before any EEH related functions. 454 */ 455 static int __init eeh_powernv_init(void) 456 { 457 int ret = -EINVAL; 458 459 eeh_set_pe_aux_size(PNV_PCI_DIAG_BUF_SIZE); 460 ret = eeh_ops_register(&powernv_eeh_ops); 461 if (!ret) 462 pr_info("EEH: PowerNV platform initialized\n"); 463 else 464 pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret); 465 466 return ret; 467 } 468 machine_early_initcall(powernv, eeh_powernv_init); 469