12874c5fdSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-or-later 229310e5eSGavin Shan /* 341732bdcSStewart Smith * PowerNV Platform dependent EEH operations 429310e5eSGavin Shan * 529310e5eSGavin Shan * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013. 629310e5eSGavin Shan */ 729310e5eSGavin Shan 829310e5eSGavin Shan #include <linux/atomic.h> 94cf17445SGavin Shan #include <linux/debugfs.h> 1029310e5eSGavin Shan #include <linux/delay.h> 1129310e5eSGavin Shan #include <linux/export.h> 1229310e5eSGavin Shan #include <linux/init.h> 1379231448SAlistair Popple #include <linux/interrupt.h> 1429310e5eSGavin Shan #include <linux/list.h> 1529310e5eSGavin Shan #include <linux/msi.h> 1629310e5eSGavin Shan #include <linux/of.h> 1729310e5eSGavin Shan #include <linux/pci.h> 1829310e5eSGavin Shan #include <linux/proc_fs.h> 1929310e5eSGavin Shan #include <linux/rbtree.h> 2029310e5eSGavin Shan #include <linux/sched.h> 2129310e5eSGavin Shan #include <linux/seq_file.h> 2229310e5eSGavin Shan #include <linux/spinlock.h> 2329310e5eSGavin Shan 2429310e5eSGavin Shan #include <asm/eeh.h> 2529310e5eSGavin Shan #include <asm/eeh_event.h> 2629310e5eSGavin Shan #include <asm/firmware.h> 2729310e5eSGavin Shan #include <asm/io.h> 2829310e5eSGavin Shan #include <asm/iommu.h> 2929310e5eSGavin Shan #include <asm/machdep.h> 3029310e5eSGavin Shan #include <asm/msi_bitmap.h> 3129310e5eSGavin Shan #include <asm/opal.h> 3229310e5eSGavin Shan #include <asm/ppc-pci.h> 339c0e1ecbSGavin Shan #include <asm/pnv-pci.h> 3429310e5eSGavin Shan 3529310e5eSGavin Shan #include "powernv.h" 3629310e5eSGavin Shan #include "pci.h" 3798fd32cdSOliver O'Halloran #include "../../../../drivers/pci/pci.h" 3829310e5eSGavin Shan 3979231448SAlistair Popple static int eeh_event_irq = -EINVAL; 404cf17445SGavin Shan 413b70464aSOliver O'Halloran static void pnv_pcibios_bus_add_device(struct pci_dev *pdev) 42988fc3baSBryant G. Ly { 431ff8f36fSSam Bobroff dev_dbg(&pdev->dev, "EEH: Setting up device\n"); 44e86350f7SOliver O'Halloran eeh_probe_device(pdev); 45988fc3baSBryant G. Ly } 46988fc3baSBryant G. Ly 4701f3bfb7SGavin Shan static int pnv_eeh_init(void) 4829310e5eSGavin Shan { 49dc561fb9SGavin Shan struct pci_controller *hose; 50dc561fb9SGavin Shan struct pnv_phb *phb; 515cb1f8fdSRussell Currey int max_diag_size = PNV_PCI_DIAG_BUF_SIZE; 52dc561fb9SGavin Shan 53e4d54f71SStewart Smith if (!firmware_has_feature(FW_FEATURE_OPAL)) { 54e4d54f71SStewart Smith pr_warn("%s: OPAL is required !\n", 550dae2743SGavin Shan __func__); 5629310e5eSGavin Shan return -EINVAL; 5729310e5eSGavin Shan } 5829310e5eSGavin Shan 5905b1721dSGavin Shan /* Set probe mode */ 6005b1721dSGavin Shan eeh_add_flag(EEH_PROBE_MODE_DEV); 6129310e5eSGavin Shan 62dc561fb9SGavin Shan /* 63dc561fb9SGavin Shan * P7IOC blocks PCI config access to frozen PE, but PHB3 64dc561fb9SGavin Shan * doesn't do that. So we have to selectively enable I/O 65dc561fb9SGavin Shan * prior to collecting error log. 66dc561fb9SGavin Shan */ 67dc561fb9SGavin Shan list_for_each_entry(hose, &hose_list, list_node) { 68dc561fb9SGavin Shan phb = hose->private_data; 69dc561fb9SGavin Shan 70dc561fb9SGavin Shan if (phb->model == PNV_PHB_MODEL_P7IOC) 71dc561fb9SGavin Shan eeh_add_flag(EEH_ENABLE_IO_FOR_LOG); 722aa5cf9eSGavin Shan 735cb1f8fdSRussell Currey if (phb->diag_data_size > max_diag_size) 745cb1f8fdSRussell Currey max_diag_size = phb->diag_data_size; 755cb1f8fdSRussell Currey 762aa5cf9eSGavin Shan /* 772aa5cf9eSGavin Shan * PE#0 should be regarded as valid by EEH core 782aa5cf9eSGavin Shan * if it's not the reserved one. Currently, we 79608fb9c2SGavin Shan * have the reserved PE#255 and PE#127 for PHB3 802aa5cf9eSGavin Shan * and P7IOC separately. So we should regard 81608fb9c2SGavin Shan * PE#0 as valid for PHB3 and P7IOC. 822aa5cf9eSGavin Shan */ 8392b8f137SGavin Shan if (phb->ioda.reserved_pe_idx != 0) 842aa5cf9eSGavin Shan eeh_add_flag(EEH_VALID_PE_ZERO); 852aa5cf9eSGavin Shan 86dc561fb9SGavin Shan break; 87dc561fb9SGavin Shan } 88dc561fb9SGavin Shan 895cb1f8fdSRussell Currey eeh_set_pe_aux_size(max_diag_size); 90988fc3baSBryant G. Ly ppc_md.pcibios_bus_add_device = pnv_pcibios_bus_add_device; 915cb1f8fdSRussell Currey 9229310e5eSGavin Shan return 0; 9329310e5eSGavin Shan } 9429310e5eSGavin Shan 9579231448SAlistair Popple static irqreturn_t pnv_eeh_event(int irq, void *data) 964cf17445SGavin Shan { 974cf17445SGavin Shan /* 9879231448SAlistair Popple * We simply send a special EEH event if EEH has been 9979231448SAlistair Popple * enabled. We don't care about EEH events until we've 10079231448SAlistair Popple * finished processing the outstanding ones. Event processing 10179231448SAlistair Popple * gets unmasked in next_error() if EEH is enabled. 1024cf17445SGavin Shan */ 10379231448SAlistair Popple disable_irq_nosync(irq); 1044cf17445SGavin Shan 1054cf17445SGavin Shan if (eeh_enabled()) 1064cf17445SGavin Shan eeh_send_failure_event(NULL); 1074cf17445SGavin Shan 10879231448SAlistair Popple return IRQ_HANDLED; 1094cf17445SGavin Shan } 1104cf17445SGavin Shan 1114cf17445SGavin Shan #ifdef CONFIG_DEBUG_FS 1124cf17445SGavin Shan static ssize_t pnv_eeh_ei_write(struct file *filp, 1134cf17445SGavin Shan const char __user *user_buf, 1144cf17445SGavin Shan size_t count, loff_t *ppos) 1154cf17445SGavin Shan { 1164cf17445SGavin Shan struct pci_controller *hose = filp->private_data; 1174cf17445SGavin Shan struct eeh_pe *pe; 1184cf17445SGavin Shan int pe_no, type, func; 1194cf17445SGavin Shan unsigned long addr, mask; 1204cf17445SGavin Shan char buf[50]; 1214cf17445SGavin Shan int ret; 1224cf17445SGavin Shan 1234cf17445SGavin Shan if (!eeh_ops || !eeh_ops->err_inject) 1244cf17445SGavin Shan return -ENXIO; 1254cf17445SGavin Shan 1264cf17445SGavin Shan /* Copy over argument buffer */ 1274cf17445SGavin Shan ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count); 1284cf17445SGavin Shan if (!ret) 1294cf17445SGavin Shan return -EFAULT; 1304cf17445SGavin Shan 1314cf17445SGavin Shan /* Retrieve parameters */ 1324cf17445SGavin Shan ret = sscanf(buf, "%x:%x:%x:%lx:%lx", 1334cf17445SGavin Shan &pe_no, &type, &func, &addr, &mask); 1344cf17445SGavin Shan if (ret != 5) 1354cf17445SGavin Shan return -EINVAL; 1364cf17445SGavin Shan 1374cf17445SGavin Shan /* Retrieve PE */ 1388bae6a23SAlexey Kardashevskiy pe = eeh_pe_get(hose, pe_no, 0); 1394cf17445SGavin Shan if (!pe) 1404cf17445SGavin Shan return -ENODEV; 1414cf17445SGavin Shan 1424cf17445SGavin Shan /* Do error injection */ 1434cf17445SGavin Shan ret = eeh_ops->err_inject(pe, type, func, addr, mask); 1444cf17445SGavin Shan return ret < 0 ? ret : count; 1454cf17445SGavin Shan } 1464cf17445SGavin Shan 1474cf17445SGavin Shan static const struct file_operations pnv_eeh_ei_fops = { 1484cf17445SGavin Shan .open = simple_open, 1494cf17445SGavin Shan .llseek = no_llseek, 1504cf17445SGavin Shan .write = pnv_eeh_ei_write, 1514cf17445SGavin Shan }; 1524cf17445SGavin Shan 1534cf17445SGavin Shan static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val) 1544cf17445SGavin Shan { 1554cf17445SGavin Shan struct pci_controller *hose = data; 1564cf17445SGavin Shan struct pnv_phb *phb = hose->private_data; 1574cf17445SGavin Shan 1584cf17445SGavin Shan out_be64(phb->regs + offset, val); 1594cf17445SGavin Shan return 0; 1604cf17445SGavin Shan } 1614cf17445SGavin Shan 1624cf17445SGavin Shan static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val) 1634cf17445SGavin Shan { 1644cf17445SGavin Shan struct pci_controller *hose = data; 1654cf17445SGavin Shan struct pnv_phb *phb = hose->private_data; 1664cf17445SGavin Shan 1674cf17445SGavin Shan *val = in_be64(phb->regs + offset); 1684cf17445SGavin Shan return 0; 1694cf17445SGavin Shan } 1704cf17445SGavin Shan 171ccc9662dSGavin Shan #define PNV_EEH_DBGFS_ENTRY(name, reg) \ 172ccc9662dSGavin Shan static int pnv_eeh_dbgfs_set_##name(void *data, u64 val) \ 173ccc9662dSGavin Shan { \ 174ccc9662dSGavin Shan return pnv_eeh_dbgfs_set(data, reg, val); \ 175ccc9662dSGavin Shan } \ 176ccc9662dSGavin Shan \ 177ccc9662dSGavin Shan static int pnv_eeh_dbgfs_get_##name(void *data, u64 *val) \ 178ccc9662dSGavin Shan { \ 179ccc9662dSGavin Shan return pnv_eeh_dbgfs_get(data, reg, val); \ 180ccc9662dSGavin Shan } \ 181ccc9662dSGavin Shan \ 182ccc9662dSGavin Shan DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_dbgfs_ops_##name, \ 183ccc9662dSGavin Shan pnv_eeh_dbgfs_get_##name, \ 184ccc9662dSGavin Shan pnv_eeh_dbgfs_set_##name, \ 185ccc9662dSGavin Shan "0x%llx\n") 1864cf17445SGavin Shan 187ccc9662dSGavin Shan PNV_EEH_DBGFS_ENTRY(outb, 0xD10); 188ccc9662dSGavin Shan PNV_EEH_DBGFS_ENTRY(inbA, 0xD90); 189ccc9662dSGavin Shan PNV_EEH_DBGFS_ENTRY(inbB, 0xE10); 1904cf17445SGavin Shan 1914cf17445SGavin Shan #endif /* CONFIG_DEBUG_FS */ 1924cf17445SGavin Shan 1933b70464aSOliver O'Halloran static void pnv_eeh_enable_phbs(void) 194b905f8cdSSam Bobroff { 195b905f8cdSSam Bobroff struct pci_controller *hose; 196b905f8cdSSam Bobroff struct pnv_phb *phb; 197b905f8cdSSam Bobroff 198b905f8cdSSam Bobroff list_for_each_entry(hose, &hose_list, list_node) { 199b905f8cdSSam Bobroff phb = hose->private_data; 200b905f8cdSSam Bobroff /* 201b905f8cdSSam Bobroff * If EEH is enabled, we're going to rely on that. 202b905f8cdSSam Bobroff * Otherwise, we restore to conventional mechanism 203b905f8cdSSam Bobroff * to clear frozen PE during PCI config access. 204b905f8cdSSam Bobroff */ 205b905f8cdSSam Bobroff if (eeh_enabled()) 206b905f8cdSSam Bobroff phb->flags |= PNV_PHB_FLAG_EEH; 207b905f8cdSSam Bobroff else 208b905f8cdSSam Bobroff phb->flags &= ~PNV_PHB_FLAG_EEH; 209b905f8cdSSam Bobroff } 210b905f8cdSSam Bobroff } 211b905f8cdSSam Bobroff 21229310e5eSGavin Shan /** 21301f3bfb7SGavin Shan * pnv_eeh_post_init - EEH platform dependent post initialization 21429310e5eSGavin Shan * 21529310e5eSGavin Shan * EEH platform dependent post initialization on powernv. When 21629310e5eSGavin Shan * the function is called, the EEH PEs and devices should have 21729310e5eSGavin Shan * been built. If the I/O cache staff has been built, EEH is 21829310e5eSGavin Shan * ready to supply service. 21929310e5eSGavin Shan */ 220b9fde58dSBenjamin Herrenschmidt int pnv_eeh_post_init(void) 22129310e5eSGavin Shan { 22229310e5eSGavin Shan struct pci_controller *hose; 22329310e5eSGavin Shan struct pnv_phb *phb; 22429310e5eSGavin Shan int ret = 0; 22529310e5eSGavin Shan 226c44e4ccaSSam Bobroff eeh_show_enabled(); 227b9fde58dSBenjamin Herrenschmidt 2284cf17445SGavin Shan /* Register OPAL event notifier */ 22979231448SAlistair Popple eeh_event_irq = opal_event_request(ilog2(OPAL_EVENT_PCI_ERROR)); 23079231448SAlistair Popple if (eeh_event_irq < 0) { 23179231448SAlistair Popple pr_err("%s: Can't register OPAL event interrupt (%d)\n", 23279231448SAlistair Popple __func__, eeh_event_irq); 23379231448SAlistair Popple return eeh_event_irq; 23479231448SAlistair Popple } 23579231448SAlistair Popple 23679231448SAlistair Popple ret = request_irq(eeh_event_irq, pnv_eeh_event, 23779231448SAlistair Popple IRQ_TYPE_LEVEL_HIGH, "opal-eeh", NULL); 23879231448SAlistair Popple if (ret < 0) { 23979231448SAlistair Popple irq_dispose_mapping(eeh_event_irq); 24079231448SAlistair Popple pr_err("%s: Can't request OPAL event interrupt (%d)\n", 24179231448SAlistair Popple __func__, eeh_event_irq); 2424cf17445SGavin Shan return ret; 2434cf17445SGavin Shan } 2444cf17445SGavin Shan 24579231448SAlistair Popple if (!eeh_enabled()) 24679231448SAlistair Popple disable_irq(eeh_event_irq); 24779231448SAlistair Popple 248b905f8cdSSam Bobroff pnv_eeh_enable_phbs(); 249b905f8cdSSam Bobroff 25029310e5eSGavin Shan list_for_each_entry(hose, &hose_list, list_node) { 25129310e5eSGavin Shan phb = hose->private_data; 25229310e5eSGavin Shan 2534cf17445SGavin Shan /* Create debugfs entries */ 2544cf17445SGavin Shan #ifdef CONFIG_DEBUG_FS 2554cf17445SGavin Shan if (phb->has_dbgfs || !phb->dbgfs) 2564cf17445SGavin Shan continue; 2574cf17445SGavin Shan 2584cf17445SGavin Shan phb->has_dbgfs = 1; 2594cf17445SGavin Shan debugfs_create_file("err_injct", 0200, 2604cf17445SGavin Shan phb->dbgfs, hose, 2614cf17445SGavin Shan &pnv_eeh_ei_fops); 2624cf17445SGavin Shan 2634cf17445SGavin Shan debugfs_create_file("err_injct_outbound", 0600, 2644cf17445SGavin Shan phb->dbgfs, hose, 265ccc9662dSGavin Shan &pnv_eeh_dbgfs_ops_outb); 2664cf17445SGavin Shan debugfs_create_file("err_injct_inboundA", 0600, 2674cf17445SGavin Shan phb->dbgfs, hose, 268ccc9662dSGavin Shan &pnv_eeh_dbgfs_ops_inbA); 2694cf17445SGavin Shan debugfs_create_file("err_injct_inboundB", 0600, 2704cf17445SGavin Shan phb->dbgfs, hose, 271ccc9662dSGavin Shan &pnv_eeh_dbgfs_ops_inbB); 2724cf17445SGavin Shan #endif /* CONFIG_DEBUG_FS */ 27329310e5eSGavin Shan } 2744cf17445SGavin Shan 27529310e5eSGavin Shan return ret; 27629310e5eSGavin Shan } 27729310e5eSGavin Shan 2784d6186caSGavin Shan static int pnv_eeh_find_cap(struct pci_dn *pdn, int cap) 279ff57b454SGavin Shan { 2804d6186caSGavin Shan int pos = PCI_CAPABILITY_LIST; 2814d6186caSGavin Shan int cnt = 48; /* Maximal number of capabilities */ 2824d6186caSGavin Shan u32 status, id; 283ff57b454SGavin Shan 284ff57b454SGavin Shan if (!pdn) 285ff57b454SGavin Shan return 0; 286ff57b454SGavin Shan 2874d6186caSGavin Shan /* Check if the device supports capabilities */ 288ff57b454SGavin Shan pnv_pci_cfg_read(pdn, PCI_STATUS, 2, &status); 289ff57b454SGavin Shan if (!(status & PCI_STATUS_CAP_LIST)) 290ff57b454SGavin Shan return 0; 291ff57b454SGavin Shan 292ff57b454SGavin Shan while (cnt--) { 293ff57b454SGavin Shan pnv_pci_cfg_read(pdn, pos, 1, &pos); 294ff57b454SGavin Shan if (pos < 0x40) 295ff57b454SGavin Shan break; 296ff57b454SGavin Shan 297ff57b454SGavin Shan pos &= ~3; 298ff57b454SGavin Shan pnv_pci_cfg_read(pdn, pos + PCI_CAP_LIST_ID, 1, &id); 299ff57b454SGavin Shan if (id == 0xff) 300ff57b454SGavin Shan break; 301ff57b454SGavin Shan 302ff57b454SGavin Shan /* Found */ 303ff57b454SGavin Shan if (id == cap) 304ff57b454SGavin Shan return pos; 305ff57b454SGavin Shan 306ff57b454SGavin Shan /* Next one */ 307ff57b454SGavin Shan pos += PCI_CAP_LIST_NEXT; 308ff57b454SGavin Shan } 309ff57b454SGavin Shan 310ff57b454SGavin Shan return 0; 311ff57b454SGavin Shan } 312ff57b454SGavin Shan 313ff57b454SGavin Shan static int pnv_eeh_find_ecap(struct pci_dn *pdn, int cap) 314ff57b454SGavin Shan { 315ff57b454SGavin Shan struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 316ff57b454SGavin Shan u32 header; 317ff57b454SGavin Shan int pos = 256, ttl = (4096 - 256) / 8; 318ff57b454SGavin Shan 319ff57b454SGavin Shan if (!edev || !edev->pcie_cap) 320ff57b454SGavin Shan return 0; 321ff57b454SGavin Shan if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL) 322ff57b454SGavin Shan return 0; 323ff57b454SGavin Shan else if (!header) 324ff57b454SGavin Shan return 0; 325ff57b454SGavin Shan 326ff57b454SGavin Shan while (ttl-- > 0) { 327ff57b454SGavin Shan if (PCI_EXT_CAP_ID(header) == cap && pos) 328ff57b454SGavin Shan return pos; 329ff57b454SGavin Shan 330ff57b454SGavin Shan pos = PCI_EXT_CAP_NEXT(header); 331ff57b454SGavin Shan if (pos < 256) 332ff57b454SGavin Shan break; 333ff57b454SGavin Shan 334ff57b454SGavin Shan if (pnv_pci_cfg_read(pdn, pos, 4, &header) != PCIBIOS_SUCCESSFUL) 335ff57b454SGavin Shan break; 336ff57b454SGavin Shan } 337ff57b454SGavin Shan 338ff57b454SGavin Shan return 0; 339ff57b454SGavin Shan } 340ff57b454SGavin Shan 341a131bfc6SOliver O'Halloran static struct eeh_pe *pnv_eeh_get_upstream_pe(struct pci_dev *pdev) 342a131bfc6SOliver O'Halloran { 343a131bfc6SOliver O'Halloran struct pci_controller *hose = pdev->bus->sysdata; 344a131bfc6SOliver O'Halloran struct pnv_phb *phb = hose->private_data; 345a131bfc6SOliver O'Halloran struct pci_dev *parent = pdev->bus->self; 346a131bfc6SOliver O'Halloran 347a131bfc6SOliver O'Halloran #ifdef CONFIG_PCI_IOV 348a131bfc6SOliver O'Halloran /* for VFs we use the PF's PE as the upstream PE */ 349a131bfc6SOliver O'Halloran if (pdev->is_virtfn) 350a131bfc6SOliver O'Halloran parent = pdev->physfn; 351a131bfc6SOliver O'Halloran #endif 352a131bfc6SOliver O'Halloran 353a131bfc6SOliver O'Halloran /* otherwise use the PE of our parent bridge */ 354a131bfc6SOliver O'Halloran if (parent) { 355a131bfc6SOliver O'Halloran struct pnv_ioda_pe *ioda_pe = pnv_ioda_get_pe(parent); 356a131bfc6SOliver O'Halloran 357a131bfc6SOliver O'Halloran return eeh_pe_get(phb->hose, ioda_pe->pe_number, 0); 358a131bfc6SOliver O'Halloran } 359a131bfc6SOliver O'Halloran 360a131bfc6SOliver O'Halloran return NULL; 361a131bfc6SOliver O'Halloran } 362a131bfc6SOliver O'Halloran 36329310e5eSGavin Shan /** 364ff57b454SGavin Shan * pnv_eeh_probe - Do probe on PCI device 365e86350f7SOliver O'Halloran * @pdev: pci_dev to probe 36629310e5eSGavin Shan * 367e86350f7SOliver O'Halloran * Create, or find the existing, eeh_dev for this pci_dev. 36829310e5eSGavin Shan */ 369e86350f7SOliver O'Halloran static struct eeh_dev *pnv_eeh_probe(struct pci_dev *pdev) 37029310e5eSGavin Shan { 371e86350f7SOliver O'Halloran struct pci_dn *pdn = pci_get_pdn(pdev); 372ff57b454SGavin Shan struct pci_controller *hose = pdn->phb; 37329310e5eSGavin Shan struct pnv_phb *phb = hose->private_data; 374ff57b454SGavin Shan struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 375a131bfc6SOliver O'Halloran struct eeh_pe *upstream_pe; 376ff57b454SGavin Shan uint32_t pcie_flags; 377dadcd6d6SMike Qiu int ret; 378405b33a7SAlexey Kardashevskiy int config_addr = (pdn->busno << 8) | (pdn->devfn); 37929310e5eSGavin Shan 38029310e5eSGavin Shan /* 38129310e5eSGavin Shan * When probing the root bridge, which doesn't have any 38229310e5eSGavin Shan * subordinate PCI devices. We don't have OF node for 38329310e5eSGavin Shan * the root bridge. So it's not reasonable to continue 38429310e5eSGavin Shan * the probing. 38529310e5eSGavin Shan */ 386ff57b454SGavin Shan if (!edev || edev->pe) 387ff57b454SGavin Shan return NULL; 38829310e5eSGavin Shan 389e86350f7SOliver O'Halloran /* already configured? */ 390e86350f7SOliver O'Halloran if (edev->pdev) { 391e86350f7SOliver O'Halloran pr_debug("%s: found existing edev for %04x:%02x:%02x.%01x\n", 392e86350f7SOliver O'Halloran __func__, hose->global_number, config_addr >> 8, 393e86350f7SOliver O'Halloran PCI_SLOT(config_addr), PCI_FUNC(config_addr)); 394e86350f7SOliver O'Halloran return edev; 395e86350f7SOliver O'Halloran } 396e86350f7SOliver O'Halloran 39729310e5eSGavin Shan /* Skip for PCI-ISA bridge */ 398768a4284SOliver O'Halloran if ((pdev->class >> 8) == PCI_CLASS_BRIDGE_ISA) 399ff57b454SGavin Shan return NULL; 40029310e5eSGavin Shan 4011ff8f36fSSam Bobroff eeh_edev_dbg(edev, "Probing device\n"); 4021ff8f36fSSam Bobroff 40329310e5eSGavin Shan /* Initialize eeh device */ 404ab55d218SGavin Shan edev->mode &= 0xFFFFFF00; 405ff57b454SGavin Shan edev->pcix_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_PCIX); 406ff57b454SGavin Shan edev->pcie_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_EXP); 4079312bc5bSWei Yang edev->af_cap = pnv_eeh_find_cap(pdn, PCI_CAP_ID_AF); 408ff57b454SGavin Shan edev->aer_cap = pnv_eeh_find_ecap(pdn, PCI_EXT_CAP_ID_ERR); 409768a4284SOliver O'Halloran if ((pdev->class >> 8) == PCI_CLASS_BRIDGE_PCI) { 4104b83bd45SGavin Shan edev->mode |= EEH_DEV_BRIDGE; 411ff57b454SGavin Shan if (edev->pcie_cap) { 412ff57b454SGavin Shan pnv_pci_cfg_read(pdn, edev->pcie_cap + PCI_EXP_FLAGS, 413ff57b454SGavin Shan 2, &pcie_flags); 414ff57b454SGavin Shan pcie_flags = (pcie_flags & PCI_EXP_FLAGS_TYPE) >> 4; 415ff57b454SGavin Shan if (pcie_flags == PCI_EXP_TYPE_ROOT_PORT) 4164b83bd45SGavin Shan edev->mode |= EEH_DEV_ROOT_PORT; 417ff57b454SGavin Shan else if (pcie_flags == PCI_EXP_TYPE_DOWNSTREAM) 4184b83bd45SGavin Shan edev->mode |= EEH_DEV_DS_PORT; 419ff57b454SGavin Shan } 4204b83bd45SGavin Shan } 4214b83bd45SGavin Shan 422405b33a7SAlexey Kardashevskiy edev->pe_config_addr = phb->ioda.pe_rmap[config_addr]; 42329310e5eSGavin Shan 424a131bfc6SOliver O'Halloran upstream_pe = pnv_eeh_get_upstream_pe(pdev); 425a131bfc6SOliver O'Halloran 42629310e5eSGavin Shan /* Create PE */ 427a131bfc6SOliver O'Halloran ret = eeh_pe_tree_insert(edev, upstream_pe); 428dadcd6d6SMike Qiu if (ret) { 4291ff8f36fSSam Bobroff eeh_edev_warn(edev, "Failed to add device to PE (code %d)\n", ret); 430ff57b454SGavin Shan return NULL; 431dadcd6d6SMike Qiu } 432dadcd6d6SMike Qiu 433dadcd6d6SMike Qiu /* 434b6541db1SGavin Shan * If the PE contains any one of following adapters, the 435b6541db1SGavin Shan * PCI config space can't be accessed when dumping EEH log. 436b6541db1SGavin Shan * Otherwise, we will run into fenced PHB caused by shortage 437b6541db1SGavin Shan * of outbound credits in the adapter. The PCI config access 438b6541db1SGavin Shan * should be blocked until PE reset. MMIO access is dropped 439b6541db1SGavin Shan * by hardware certainly. In order to drop PCI config requests, 440b6541db1SGavin Shan * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which 441b6541db1SGavin Shan * will be checked in the backend for PE state retrival. If 442b6541db1SGavin Shan * the PE becomes frozen for the first time and the flag has 443b6541db1SGavin Shan * been set for the PE, we will set EEH_PE_CFG_BLOCKED for 444b6541db1SGavin Shan * that PE to block its config space. 445b6541db1SGavin Shan * 446c374ed27SGavin Shan * Broadcom BCM5718 2-ports NICs (14e4:1656) 447b6541db1SGavin Shan * Broadcom Austin 4-ports NICs (14e4:1657) 448353169acSGavin Shan * Broadcom Shiner 4-ports 1G NICs (14e4:168a) 449179ea48bSGavin Shan * Broadcom Shiner 2-ports 10G NICs (14e4:168e) 450b6541db1SGavin Shan */ 451ff57b454SGavin Shan if ((pdn->vendor_id == PCI_VENDOR_ID_BROADCOM && 452c374ed27SGavin Shan pdn->device_id == 0x1656) || 453c374ed27SGavin Shan (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM && 454ff57b454SGavin Shan pdn->device_id == 0x1657) || 455ff57b454SGavin Shan (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM && 456353169acSGavin Shan pdn->device_id == 0x168a) || 457353169acSGavin Shan (pdn->vendor_id == PCI_VENDOR_ID_BROADCOM && 458ff57b454SGavin Shan pdn->device_id == 0x168e)) 459b6541db1SGavin Shan edev->pe->state |= EEH_PE_CFG_RESTRICTED; 460b6541db1SGavin Shan 461b6541db1SGavin Shan /* 462dadcd6d6SMike Qiu * Cache the PE primary bus, which can't be fetched when 463dadcd6d6SMike Qiu * full hotplug is in progress. In that case, all child 464dadcd6d6SMike Qiu * PCI devices of the PE are expected to be removed prior 465dadcd6d6SMike Qiu * to PE reset. 466dadcd6d6SMike Qiu */ 46705ba75f8SGavin Shan if (!(edev->pe->state & EEH_PE_PRI_BUS)) { 468ff57b454SGavin Shan edev->pe->bus = pci_find_bus(hose->global_number, 469ff57b454SGavin Shan pdn->busno); 47005ba75f8SGavin Shan if (edev->pe->bus) 47105ba75f8SGavin Shan edev->pe->state |= EEH_PE_PRI_BUS; 47205ba75f8SGavin Shan } 47329310e5eSGavin Shan 47429310e5eSGavin Shan /* 47529310e5eSGavin Shan * Enable EEH explicitly so that we will do EEH check 47629310e5eSGavin Shan * while accessing I/O stuff 47729310e5eSGavin Shan */ 478b905f8cdSSam Bobroff if (!eeh_has_flag(EEH_ENABLED)) { 479b905f8cdSSam Bobroff enable_irq(eeh_event_irq); 480b905f8cdSSam Bobroff pnv_eeh_enable_phbs(); 48105b1721dSGavin Shan eeh_add_flag(EEH_ENABLED); 482b905f8cdSSam Bobroff } 48329310e5eSGavin Shan 48429310e5eSGavin Shan /* Save memory bars */ 48529310e5eSGavin Shan eeh_save_bars(edev); 48629310e5eSGavin Shan 4871ff8f36fSSam Bobroff eeh_edev_dbg(edev, "EEH enabled on device\n"); 488617082a4SSam Bobroff 489e86350f7SOliver O'Halloran return edev; 49029310e5eSGavin Shan } 49129310e5eSGavin Shan 49229310e5eSGavin Shan /** 49301f3bfb7SGavin Shan * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable 49429310e5eSGavin Shan * @pe: EEH PE 49529310e5eSGavin Shan * @option: operation to be issued 49629310e5eSGavin Shan * 49729310e5eSGavin Shan * The function is used to control the EEH functionality globally. 49829310e5eSGavin Shan * Currently, following options are support according to PAPR: 49929310e5eSGavin Shan * Enable EEH, Disable EEH, Enable MMIO and Enable DMA 50029310e5eSGavin Shan */ 50101f3bfb7SGavin Shan static int pnv_eeh_set_option(struct eeh_pe *pe, int option) 50229310e5eSGavin Shan { 50329310e5eSGavin Shan struct pci_controller *hose = pe->phb; 50429310e5eSGavin Shan struct pnv_phb *phb = hose->private_data; 5057e3e4f8dSGavin Shan bool freeze_pe = false; 506f9433718SGavin Shan int opt; 5077e3e4f8dSGavin Shan s64 rc; 50829310e5eSGavin Shan 5097e3e4f8dSGavin Shan switch (option) { 5107e3e4f8dSGavin Shan case EEH_OPT_DISABLE: 5117e3e4f8dSGavin Shan return -EPERM; 5127e3e4f8dSGavin Shan case EEH_OPT_ENABLE: 5137e3e4f8dSGavin Shan return 0; 5147e3e4f8dSGavin Shan case EEH_OPT_THAW_MMIO: 5157e3e4f8dSGavin Shan opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO; 5167e3e4f8dSGavin Shan break; 5177e3e4f8dSGavin Shan case EEH_OPT_THAW_DMA: 5187e3e4f8dSGavin Shan opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA; 5197e3e4f8dSGavin Shan break; 5207e3e4f8dSGavin Shan case EEH_OPT_FREEZE_PE: 5217e3e4f8dSGavin Shan freeze_pe = true; 5227e3e4f8dSGavin Shan opt = OPAL_EEH_ACTION_SET_FREEZE_ALL; 5237e3e4f8dSGavin Shan break; 5247e3e4f8dSGavin Shan default: 5257e3e4f8dSGavin Shan pr_warn("%s: Invalid option %d\n", __func__, option); 5267e3e4f8dSGavin Shan return -EINVAL; 5277e3e4f8dSGavin Shan } 5287e3e4f8dSGavin Shan 529f9433718SGavin Shan /* Freeze master and slave PEs if PHB supports compound PEs */ 5307e3e4f8dSGavin Shan if (freeze_pe) { 5317e3e4f8dSGavin Shan if (phb->freeze_pe) { 5327e3e4f8dSGavin Shan phb->freeze_pe(phb, pe->addr); 533f9433718SGavin Shan return 0; 5347e3e4f8dSGavin Shan } 53529310e5eSGavin Shan 536f9433718SGavin Shan rc = opal_pci_eeh_freeze_set(phb->opal_id, pe->addr, opt); 537f9433718SGavin Shan if (rc != OPAL_SUCCESS) { 538f9433718SGavin Shan pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n", 539f9433718SGavin Shan __func__, rc, phb->hose->global_number, 540f9433718SGavin Shan pe->addr); 541f9433718SGavin Shan return -EIO; 542f9433718SGavin Shan } 543f9433718SGavin Shan 544f9433718SGavin Shan return 0; 545f9433718SGavin Shan } 546f9433718SGavin Shan 547f9433718SGavin Shan /* Unfreeze master and slave PEs if PHB supports */ 548f9433718SGavin Shan if (phb->unfreeze_pe) 549f9433718SGavin Shan return phb->unfreeze_pe(phb, pe->addr, opt); 550f9433718SGavin Shan 551f9433718SGavin Shan rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe->addr, opt); 552f9433718SGavin Shan if (rc != OPAL_SUCCESS) { 553f9433718SGavin Shan pr_warn("%s: Failure %lld enable %d for PHB#%x-PE#%x\n", 554f9433718SGavin Shan __func__, rc, option, phb->hose->global_number, 555f9433718SGavin Shan pe->addr); 556f9433718SGavin Shan return -EIO; 557f9433718SGavin Shan } 558f9433718SGavin Shan 559f9433718SGavin Shan return 0; 56029310e5eSGavin Shan } 56129310e5eSGavin Shan 56240ae5f69SGavin Shan static void pnv_eeh_get_phb_diag(struct eeh_pe *pe) 56340ae5f69SGavin Shan { 56440ae5f69SGavin Shan struct pnv_phb *phb = pe->phb->private_data; 56540ae5f69SGavin Shan s64 rc; 56640ae5f69SGavin Shan 56740ae5f69SGavin Shan rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data, 5685cb1f8fdSRussell Currey phb->diag_data_size); 56940ae5f69SGavin Shan if (rc != OPAL_SUCCESS) 57040ae5f69SGavin Shan pr_warn("%s: Failure %lld getting PHB#%x diag-data\n", 57140ae5f69SGavin Shan __func__, rc, pe->phb->global_number); 57240ae5f69SGavin Shan } 57340ae5f69SGavin Shan 57440ae5f69SGavin Shan static int pnv_eeh_get_phb_state(struct eeh_pe *pe) 57540ae5f69SGavin Shan { 57640ae5f69SGavin Shan struct pnv_phb *phb = pe->phb->private_data; 577c2057701SAlexey Kardashevskiy u8 fstate = 0; 578c2057701SAlexey Kardashevskiy __be16 pcierr = 0; 57940ae5f69SGavin Shan s64 rc; 58040ae5f69SGavin Shan int result = 0; 58140ae5f69SGavin Shan 58240ae5f69SGavin Shan rc = opal_pci_eeh_freeze_status(phb->opal_id, 58340ae5f69SGavin Shan pe->addr, 58440ae5f69SGavin Shan &fstate, 58540ae5f69SGavin Shan &pcierr, 58640ae5f69SGavin Shan NULL); 58740ae5f69SGavin Shan if (rc != OPAL_SUCCESS) { 58840ae5f69SGavin Shan pr_warn("%s: Failure %lld getting PHB#%x state\n", 58940ae5f69SGavin Shan __func__, rc, phb->hose->global_number); 59040ae5f69SGavin Shan return EEH_STATE_NOT_SUPPORT; 59140ae5f69SGavin Shan } 59240ae5f69SGavin Shan 59340ae5f69SGavin Shan /* 59440ae5f69SGavin Shan * Check PHB state. If the PHB is frozen for the 59540ae5f69SGavin Shan * first time, to dump the PHB diag-data. 59640ae5f69SGavin Shan */ 59740ae5f69SGavin Shan if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) { 59840ae5f69SGavin Shan result = (EEH_STATE_MMIO_ACTIVE | 59940ae5f69SGavin Shan EEH_STATE_DMA_ACTIVE | 60040ae5f69SGavin Shan EEH_STATE_MMIO_ENABLED | 60140ae5f69SGavin Shan EEH_STATE_DMA_ENABLED); 60240ae5f69SGavin Shan } else if (!(pe->state & EEH_PE_ISOLATED)) { 603e762bb89SSam Bobroff eeh_pe_mark_isolated(pe); 60440ae5f69SGavin Shan pnv_eeh_get_phb_diag(pe); 60540ae5f69SGavin Shan 60640ae5f69SGavin Shan if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) 60740ae5f69SGavin Shan pnv_pci_dump_phb_diag_data(pe->phb, pe->data); 60840ae5f69SGavin Shan } 60940ae5f69SGavin Shan 61040ae5f69SGavin Shan return result; 61140ae5f69SGavin Shan } 61240ae5f69SGavin Shan 61340ae5f69SGavin Shan static int pnv_eeh_get_pe_state(struct eeh_pe *pe) 61440ae5f69SGavin Shan { 61540ae5f69SGavin Shan struct pnv_phb *phb = pe->phb->private_data; 616c2057701SAlexey Kardashevskiy u8 fstate = 0; 617c2057701SAlexey Kardashevskiy __be16 pcierr = 0; 61840ae5f69SGavin Shan s64 rc; 61940ae5f69SGavin Shan int result; 62040ae5f69SGavin Shan 62140ae5f69SGavin Shan /* 62240ae5f69SGavin Shan * We don't clobber hardware frozen state until PE 62340ae5f69SGavin Shan * reset is completed. In order to keep EEH core 62440ae5f69SGavin Shan * moving forward, we have to return operational 62540ae5f69SGavin Shan * state during PE reset. 62640ae5f69SGavin Shan */ 62740ae5f69SGavin Shan if (pe->state & EEH_PE_RESET) { 62840ae5f69SGavin Shan result = (EEH_STATE_MMIO_ACTIVE | 62940ae5f69SGavin Shan EEH_STATE_DMA_ACTIVE | 63040ae5f69SGavin Shan EEH_STATE_MMIO_ENABLED | 63140ae5f69SGavin Shan EEH_STATE_DMA_ENABLED); 63240ae5f69SGavin Shan return result; 63340ae5f69SGavin Shan } 63440ae5f69SGavin Shan 63540ae5f69SGavin Shan /* 63640ae5f69SGavin Shan * Fetch PE state from hardware. If the PHB 63740ae5f69SGavin Shan * supports compound PE, let it handle that. 63840ae5f69SGavin Shan */ 63940ae5f69SGavin Shan if (phb->get_pe_state) { 64040ae5f69SGavin Shan fstate = phb->get_pe_state(phb, pe->addr); 64140ae5f69SGavin Shan } else { 64240ae5f69SGavin Shan rc = opal_pci_eeh_freeze_status(phb->opal_id, 64340ae5f69SGavin Shan pe->addr, 64440ae5f69SGavin Shan &fstate, 64540ae5f69SGavin Shan &pcierr, 64640ae5f69SGavin Shan NULL); 64740ae5f69SGavin Shan if (rc != OPAL_SUCCESS) { 64840ae5f69SGavin Shan pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n", 64940ae5f69SGavin Shan __func__, rc, phb->hose->global_number, 65040ae5f69SGavin Shan pe->addr); 65140ae5f69SGavin Shan return EEH_STATE_NOT_SUPPORT; 65240ae5f69SGavin Shan } 65340ae5f69SGavin Shan } 65440ae5f69SGavin Shan 65540ae5f69SGavin Shan /* Figure out state */ 65640ae5f69SGavin Shan switch (fstate) { 65740ae5f69SGavin Shan case OPAL_EEH_STOPPED_NOT_FROZEN: 65840ae5f69SGavin Shan result = (EEH_STATE_MMIO_ACTIVE | 65940ae5f69SGavin Shan EEH_STATE_DMA_ACTIVE | 66040ae5f69SGavin Shan EEH_STATE_MMIO_ENABLED | 66140ae5f69SGavin Shan EEH_STATE_DMA_ENABLED); 66240ae5f69SGavin Shan break; 66340ae5f69SGavin Shan case OPAL_EEH_STOPPED_MMIO_FREEZE: 66440ae5f69SGavin Shan result = (EEH_STATE_DMA_ACTIVE | 66540ae5f69SGavin Shan EEH_STATE_DMA_ENABLED); 66640ae5f69SGavin Shan break; 66740ae5f69SGavin Shan case OPAL_EEH_STOPPED_DMA_FREEZE: 66840ae5f69SGavin Shan result = (EEH_STATE_MMIO_ACTIVE | 66940ae5f69SGavin Shan EEH_STATE_MMIO_ENABLED); 67040ae5f69SGavin Shan break; 67140ae5f69SGavin Shan case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE: 67240ae5f69SGavin Shan result = 0; 67340ae5f69SGavin Shan break; 67440ae5f69SGavin Shan case OPAL_EEH_STOPPED_RESET: 67540ae5f69SGavin Shan result = EEH_STATE_RESET_ACTIVE; 67640ae5f69SGavin Shan break; 67740ae5f69SGavin Shan case OPAL_EEH_STOPPED_TEMP_UNAVAIL: 67840ae5f69SGavin Shan result = EEH_STATE_UNAVAILABLE; 67940ae5f69SGavin Shan break; 68040ae5f69SGavin Shan case OPAL_EEH_STOPPED_PERM_UNAVAIL: 68140ae5f69SGavin Shan result = EEH_STATE_NOT_SUPPORT; 68240ae5f69SGavin Shan break; 68340ae5f69SGavin Shan default: 68440ae5f69SGavin Shan result = EEH_STATE_NOT_SUPPORT; 68540ae5f69SGavin Shan pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n", 68640ae5f69SGavin Shan __func__, phb->hose->global_number, 68740ae5f69SGavin Shan pe->addr, fstate); 68840ae5f69SGavin Shan } 68940ae5f69SGavin Shan 69040ae5f69SGavin Shan /* 69140ae5f69SGavin Shan * If PHB supports compound PE, to freeze all 69240ae5f69SGavin Shan * slave PEs for consistency. 69340ae5f69SGavin Shan * 69440ae5f69SGavin Shan * If the PE is switching to frozen state for the 69540ae5f69SGavin Shan * first time, to dump the PHB diag-data. 69640ae5f69SGavin Shan */ 69740ae5f69SGavin Shan if (!(result & EEH_STATE_NOT_SUPPORT) && 69840ae5f69SGavin Shan !(result & EEH_STATE_UNAVAILABLE) && 69940ae5f69SGavin Shan !(result & EEH_STATE_MMIO_ACTIVE) && 70040ae5f69SGavin Shan !(result & EEH_STATE_DMA_ACTIVE) && 70140ae5f69SGavin Shan !(pe->state & EEH_PE_ISOLATED)) { 70240ae5f69SGavin Shan if (phb->freeze_pe) 70340ae5f69SGavin Shan phb->freeze_pe(phb, pe->addr); 70440ae5f69SGavin Shan 705e762bb89SSam Bobroff eeh_pe_mark_isolated(pe); 70640ae5f69SGavin Shan pnv_eeh_get_phb_diag(pe); 70740ae5f69SGavin Shan 70840ae5f69SGavin Shan if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) 70940ae5f69SGavin Shan pnv_pci_dump_phb_diag_data(pe->phb, pe->data); 71040ae5f69SGavin Shan } 71140ae5f69SGavin Shan 71240ae5f69SGavin Shan return result; 71340ae5f69SGavin Shan } 71440ae5f69SGavin Shan 71529310e5eSGavin Shan /** 71601f3bfb7SGavin Shan * pnv_eeh_get_state - Retrieve PE state 71729310e5eSGavin Shan * @pe: EEH PE 71829310e5eSGavin Shan * @delay: delay while PE state is temporarily unavailable 71929310e5eSGavin Shan * 72029310e5eSGavin Shan * Retrieve the state of the specified PE. For IODA-compitable 72129310e5eSGavin Shan * platform, it should be retrieved from IODA table. Therefore, 72229310e5eSGavin Shan * we prefer passing down to hardware implementation to handle 72329310e5eSGavin Shan * it. 72429310e5eSGavin Shan */ 72501f3bfb7SGavin Shan static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay) 72629310e5eSGavin Shan { 72740ae5f69SGavin Shan int ret; 72829310e5eSGavin Shan 72940ae5f69SGavin Shan if (pe->type & EEH_PE_PHB) 73040ae5f69SGavin Shan ret = pnv_eeh_get_phb_state(pe); 73140ae5f69SGavin Shan else 73240ae5f69SGavin Shan ret = pnv_eeh_get_pe_state(pe); 73340ae5f69SGavin Shan 73440ae5f69SGavin Shan if (!delay) 73540ae5f69SGavin Shan return ret; 73629310e5eSGavin Shan 73729310e5eSGavin Shan /* 73829310e5eSGavin Shan * If the PE state is temporarily unavailable, 73929310e5eSGavin Shan * to inform the EEH core delay for default 74029310e5eSGavin Shan * period (1 second) 74129310e5eSGavin Shan */ 74229310e5eSGavin Shan *delay = 0; 74329310e5eSGavin Shan if (ret & EEH_STATE_UNAVAILABLE) 74429310e5eSGavin Shan *delay = 1000; 74529310e5eSGavin Shan 74629310e5eSGavin Shan return ret; 74729310e5eSGavin Shan } 74829310e5eSGavin Shan 749ebe22531SGavin Shan static s64 pnv_eeh_poll(unsigned long id) 750cadf364dSGavin Shan { 751cadf364dSGavin Shan s64 rc = OPAL_HARDWARE; 752cadf364dSGavin Shan 753cadf364dSGavin Shan while (1) { 754ebe22531SGavin Shan rc = opal_pci_poll(id); 755cadf364dSGavin Shan if (rc <= 0) 756cadf364dSGavin Shan break; 757cadf364dSGavin Shan 758cadf364dSGavin Shan if (system_state < SYSTEM_RUNNING) 759cadf364dSGavin Shan udelay(1000 * rc); 760cadf364dSGavin Shan else 761cadf364dSGavin Shan msleep(rc); 762cadf364dSGavin Shan } 763cadf364dSGavin Shan 764cadf364dSGavin Shan return rc; 765cadf364dSGavin Shan } 766cadf364dSGavin Shan 767cadf364dSGavin Shan int pnv_eeh_phb_reset(struct pci_controller *hose, int option) 768cadf364dSGavin Shan { 769cadf364dSGavin Shan struct pnv_phb *phb = hose->private_data; 770cadf364dSGavin Shan s64 rc = OPAL_HARDWARE; 771cadf364dSGavin Shan 772cadf364dSGavin Shan pr_debug("%s: Reset PHB#%x, option=%d\n", 773cadf364dSGavin Shan __func__, hose->global_number, option); 774cadf364dSGavin Shan 775cadf364dSGavin Shan /* Issue PHB complete reset request */ 776cadf364dSGavin Shan if (option == EEH_RESET_FUNDAMENTAL || 777cadf364dSGavin Shan option == EEH_RESET_HOT) 778cadf364dSGavin Shan rc = opal_pci_reset(phb->opal_id, 779cadf364dSGavin Shan OPAL_RESET_PHB_COMPLETE, 780cadf364dSGavin Shan OPAL_ASSERT_RESET); 781cadf364dSGavin Shan else if (option == EEH_RESET_DEACTIVATE) 782cadf364dSGavin Shan rc = opal_pci_reset(phb->opal_id, 783cadf364dSGavin Shan OPAL_RESET_PHB_COMPLETE, 784cadf364dSGavin Shan OPAL_DEASSERT_RESET); 785cadf364dSGavin Shan if (rc < 0) 786cadf364dSGavin Shan goto out; 787cadf364dSGavin Shan 788cadf364dSGavin Shan /* 789cadf364dSGavin Shan * Poll state of the PHB until the request is done 790cadf364dSGavin Shan * successfully. The PHB reset is usually PHB complete 791cadf364dSGavin Shan * reset followed by hot reset on root bus. So we also 792cadf364dSGavin Shan * need the PCI bus settlement delay. 793cadf364dSGavin Shan */ 794fbce44d0SGavin Shan if (rc > 0) 795ebe22531SGavin Shan rc = pnv_eeh_poll(phb->opal_id); 796cadf364dSGavin Shan if (option == EEH_RESET_DEACTIVATE) { 797cadf364dSGavin Shan if (system_state < SYSTEM_RUNNING) 798cadf364dSGavin Shan udelay(1000 * EEH_PE_RST_SETTLE_TIME); 799cadf364dSGavin Shan else 800cadf364dSGavin Shan msleep(EEH_PE_RST_SETTLE_TIME); 801cadf364dSGavin Shan } 802cadf364dSGavin Shan out: 803cadf364dSGavin Shan if (rc != OPAL_SUCCESS) 804cadf364dSGavin Shan return -EIO; 805cadf364dSGavin Shan 806cadf364dSGavin Shan return 0; 807cadf364dSGavin Shan } 808cadf364dSGavin Shan 809cadf364dSGavin Shan static int pnv_eeh_root_reset(struct pci_controller *hose, int option) 810cadf364dSGavin Shan { 811cadf364dSGavin Shan struct pnv_phb *phb = hose->private_data; 812cadf364dSGavin Shan s64 rc = OPAL_HARDWARE; 813cadf364dSGavin Shan 814cadf364dSGavin Shan pr_debug("%s: Reset PHB#%x, option=%d\n", 815cadf364dSGavin Shan __func__, hose->global_number, option); 816cadf364dSGavin Shan 817cadf364dSGavin Shan /* 818cadf364dSGavin Shan * During the reset deassert time, we needn't care 819cadf364dSGavin Shan * the reset scope because the firmware does nothing 820cadf364dSGavin Shan * for fundamental or hot reset during deassert phase. 821cadf364dSGavin Shan */ 822cadf364dSGavin Shan if (option == EEH_RESET_FUNDAMENTAL) 823cadf364dSGavin Shan rc = opal_pci_reset(phb->opal_id, 824cadf364dSGavin Shan OPAL_RESET_PCI_FUNDAMENTAL, 825cadf364dSGavin Shan OPAL_ASSERT_RESET); 826cadf364dSGavin Shan else if (option == EEH_RESET_HOT) 827cadf364dSGavin Shan rc = opal_pci_reset(phb->opal_id, 828cadf364dSGavin Shan OPAL_RESET_PCI_HOT, 829cadf364dSGavin Shan OPAL_ASSERT_RESET); 830cadf364dSGavin Shan else if (option == EEH_RESET_DEACTIVATE) 831cadf364dSGavin Shan rc = opal_pci_reset(phb->opal_id, 832cadf364dSGavin Shan OPAL_RESET_PCI_HOT, 833cadf364dSGavin Shan OPAL_DEASSERT_RESET); 834cadf364dSGavin Shan if (rc < 0) 835cadf364dSGavin Shan goto out; 836cadf364dSGavin Shan 837cadf364dSGavin Shan /* Poll state of the PHB until the request is done */ 838fbce44d0SGavin Shan if (rc > 0) 839ebe22531SGavin Shan rc = pnv_eeh_poll(phb->opal_id); 840cadf364dSGavin Shan if (option == EEH_RESET_DEACTIVATE) 841cadf364dSGavin Shan msleep(EEH_PE_RST_SETTLE_TIME); 842cadf364dSGavin Shan out: 843cadf364dSGavin Shan if (rc != OPAL_SUCCESS) 844cadf364dSGavin Shan return -EIO; 845cadf364dSGavin Shan 846cadf364dSGavin Shan return 0; 847cadf364dSGavin Shan } 848cadf364dSGavin Shan 8499c0e1ecbSGavin Shan static int __pnv_eeh_bridge_reset(struct pci_dev *dev, int option) 850cadf364dSGavin Shan { 8510bd78587SGavin Shan struct pci_dn *pdn = pci_get_pdn_by_devfn(dev->bus, dev->devfn); 8520bd78587SGavin Shan struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 853cadf364dSGavin Shan int aer = edev ? edev->aer_cap : 0; 854cadf364dSGavin Shan u32 ctrl; 855cadf364dSGavin Shan 85698fd32cdSOliver O'Halloran pr_debug("%s: Secondary Reset PCI bus %04x:%02x with option %d\n", 857cadf364dSGavin Shan __func__, pci_domain_nr(dev->bus), 858cadf364dSGavin Shan dev->bus->number, option); 859cadf364dSGavin Shan 860cadf364dSGavin Shan switch (option) { 861cadf364dSGavin Shan case EEH_RESET_FUNDAMENTAL: 862cadf364dSGavin Shan case EEH_RESET_HOT: 863cadf364dSGavin Shan /* Don't report linkDown event */ 864cadf364dSGavin Shan if (aer) { 86517d2a487SOliver O'Halloran eeh_ops->read_config(edev, aer + PCI_ERR_UNCOR_MASK, 866cadf364dSGavin Shan 4, &ctrl); 867cadf364dSGavin Shan ctrl |= PCI_ERR_UNC_SURPDN; 86817d2a487SOliver O'Halloran eeh_ops->write_config(edev, aer + PCI_ERR_UNCOR_MASK, 869cadf364dSGavin Shan 4, ctrl); 870cadf364dSGavin Shan } 871cadf364dSGavin Shan 87217d2a487SOliver O'Halloran eeh_ops->read_config(edev, PCI_BRIDGE_CONTROL, 2, &ctrl); 873cadf364dSGavin Shan ctrl |= PCI_BRIDGE_CTL_BUS_RESET; 87417d2a487SOliver O'Halloran eeh_ops->write_config(edev, PCI_BRIDGE_CONTROL, 2, ctrl); 875cadf364dSGavin Shan 876cadf364dSGavin Shan msleep(EEH_PE_RST_HOLD_TIME); 877cadf364dSGavin Shan break; 878cadf364dSGavin Shan case EEH_RESET_DEACTIVATE: 87917d2a487SOliver O'Halloran eeh_ops->read_config(edev, PCI_BRIDGE_CONTROL, 2, &ctrl); 880cadf364dSGavin Shan ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET; 88117d2a487SOliver O'Halloran eeh_ops->write_config(edev, PCI_BRIDGE_CONTROL, 2, ctrl); 882cadf364dSGavin Shan 883cadf364dSGavin Shan msleep(EEH_PE_RST_SETTLE_TIME); 884cadf364dSGavin Shan 885cadf364dSGavin Shan /* Continue reporting linkDown event */ 886cadf364dSGavin Shan if (aer) { 88717d2a487SOliver O'Halloran eeh_ops->read_config(edev, aer + PCI_ERR_UNCOR_MASK, 888cadf364dSGavin Shan 4, &ctrl); 889cadf364dSGavin Shan ctrl &= ~PCI_ERR_UNC_SURPDN; 89017d2a487SOliver O'Halloran eeh_ops->write_config(edev, aer + PCI_ERR_UNCOR_MASK, 891cadf364dSGavin Shan 4, ctrl); 892cadf364dSGavin Shan } 893cadf364dSGavin Shan 894cadf364dSGavin Shan break; 895cadf364dSGavin Shan } 896cadf364dSGavin Shan 897cadf364dSGavin Shan return 0; 898cadf364dSGavin Shan } 899cadf364dSGavin Shan 9009c0e1ecbSGavin Shan static int pnv_eeh_bridge_reset(struct pci_dev *pdev, int option) 9019c0e1ecbSGavin Shan { 9029c0e1ecbSGavin Shan struct pci_controller *hose = pci_bus_to_host(pdev->bus); 9039c0e1ecbSGavin Shan struct pnv_phb *phb = hose->private_data; 9049c0e1ecbSGavin Shan struct device_node *dn = pci_device_to_OF_node(pdev); 9059c0e1ecbSGavin Shan uint64_t id = PCI_SLOT_ID(phb->opal_id, 9069c0e1ecbSGavin Shan (pdev->bus->number << 8) | pdev->devfn); 9079c0e1ecbSGavin Shan uint8_t scope; 9089c0e1ecbSGavin Shan int64_t rc; 9099c0e1ecbSGavin Shan 9109c0e1ecbSGavin Shan /* Hot reset to the bus if firmware cannot handle */ 9119c0e1ecbSGavin Shan if (!dn || !of_get_property(dn, "ibm,reset-by-firmware", NULL)) 9129c0e1ecbSGavin Shan return __pnv_eeh_bridge_reset(pdev, option); 9139c0e1ecbSGavin Shan 91498fd32cdSOliver O'Halloran pr_debug("%s: FW reset PCI bus %04x:%02x with option %d\n", 91598fd32cdSOliver O'Halloran __func__, pci_domain_nr(pdev->bus), 91698fd32cdSOliver O'Halloran pdev->bus->number, option); 91798fd32cdSOliver O'Halloran 9189c0e1ecbSGavin Shan switch (option) { 9199c0e1ecbSGavin Shan case EEH_RESET_FUNDAMENTAL: 9209c0e1ecbSGavin Shan scope = OPAL_RESET_PCI_FUNDAMENTAL; 9219c0e1ecbSGavin Shan break; 9229c0e1ecbSGavin Shan case EEH_RESET_HOT: 9239c0e1ecbSGavin Shan scope = OPAL_RESET_PCI_HOT; 9249c0e1ecbSGavin Shan break; 9259c0e1ecbSGavin Shan case EEH_RESET_DEACTIVATE: 9269c0e1ecbSGavin Shan return 0; 9279c0e1ecbSGavin Shan default: 9289c0e1ecbSGavin Shan dev_dbg(&pdev->dev, "%s: Unsupported reset %d\n", 9299c0e1ecbSGavin Shan __func__, option); 9309c0e1ecbSGavin Shan return -EINVAL; 9319c0e1ecbSGavin Shan } 9329c0e1ecbSGavin Shan 9339c0e1ecbSGavin Shan rc = opal_pci_reset(id, scope, OPAL_ASSERT_RESET); 9349c0e1ecbSGavin Shan if (rc <= OPAL_SUCCESS) 9359c0e1ecbSGavin Shan goto out; 9369c0e1ecbSGavin Shan 9379c0e1ecbSGavin Shan rc = pnv_eeh_poll(id); 9389c0e1ecbSGavin Shan out: 9399c0e1ecbSGavin Shan return (rc == OPAL_SUCCESS) ? 0 : -EIO; 9409c0e1ecbSGavin Shan } 9419c0e1ecbSGavin Shan 942cadf364dSGavin Shan void pnv_pci_reset_secondary_bus(struct pci_dev *dev) 943cadf364dSGavin Shan { 944848912e5SMichael Ellerman struct pci_controller *hose; 945848912e5SMichael Ellerman 946848912e5SMichael Ellerman if (pci_is_root_bus(dev->bus)) { 947848912e5SMichael Ellerman hose = pci_bus_to_host(dev->bus); 948848912e5SMichael Ellerman pnv_eeh_root_reset(hose, EEH_RESET_HOT); 949848912e5SMichael Ellerman pnv_eeh_root_reset(hose, EEH_RESET_DEACTIVATE); 950848912e5SMichael Ellerman } else { 951cadf364dSGavin Shan pnv_eeh_bridge_reset(dev, EEH_RESET_HOT); 952cadf364dSGavin Shan pnv_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE); 953cadf364dSGavin Shan } 954848912e5SMichael Ellerman } 955cadf364dSGavin Shan 9569312bc5bSWei Yang static void pnv_eeh_wait_for_pending(struct pci_dn *pdn, const char *type, 9579312bc5bSWei Yang int pos, u16 mask) 9589312bc5bSWei Yang { 95917d2a487SOliver O'Halloran struct eeh_dev *edev = pdn->edev; 9609312bc5bSWei Yang int i, status = 0; 9619312bc5bSWei Yang 9629312bc5bSWei Yang /* Wait for Transaction Pending bit to be cleared */ 9639312bc5bSWei Yang for (i = 0; i < 4; i++) { 96417d2a487SOliver O'Halloran eeh_ops->read_config(edev, pos, 2, &status); 9659312bc5bSWei Yang if (!(status & mask)) 9669312bc5bSWei Yang return; 9679312bc5bSWei Yang 9689312bc5bSWei Yang msleep((1 << i) * 100); 9699312bc5bSWei Yang } 9709312bc5bSWei Yang 9719312bc5bSWei Yang pr_warn("%s: Pending transaction while issuing %sFLR to %04x:%02x:%02x.%01x\n", 9729312bc5bSWei Yang __func__, type, 97369672bd7SAlexey Kardashevskiy pdn->phb->global_number, pdn->busno, 9749312bc5bSWei Yang PCI_SLOT(pdn->devfn), PCI_FUNC(pdn->devfn)); 9759312bc5bSWei Yang } 9769312bc5bSWei Yang 9779312bc5bSWei Yang static int pnv_eeh_do_flr(struct pci_dn *pdn, int option) 9789312bc5bSWei Yang { 9799312bc5bSWei Yang struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 9809312bc5bSWei Yang u32 reg = 0; 9819312bc5bSWei Yang 9829312bc5bSWei Yang if (WARN_ON(!edev->pcie_cap)) 9839312bc5bSWei Yang return -ENOTTY; 9849312bc5bSWei Yang 98517d2a487SOliver O'Halloran eeh_ops->read_config(edev, edev->pcie_cap + PCI_EXP_DEVCAP, 4, ®); 9869312bc5bSWei Yang if (!(reg & PCI_EXP_DEVCAP_FLR)) 9879312bc5bSWei Yang return -ENOTTY; 9889312bc5bSWei Yang 9899312bc5bSWei Yang switch (option) { 9909312bc5bSWei Yang case EEH_RESET_HOT: 9919312bc5bSWei Yang case EEH_RESET_FUNDAMENTAL: 9929312bc5bSWei Yang pnv_eeh_wait_for_pending(pdn, "", 9939312bc5bSWei Yang edev->pcie_cap + PCI_EXP_DEVSTA, 9949312bc5bSWei Yang PCI_EXP_DEVSTA_TRPND); 99517d2a487SOliver O'Halloran eeh_ops->read_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL, 9969312bc5bSWei Yang 4, ®); 9979312bc5bSWei Yang reg |= PCI_EXP_DEVCTL_BCR_FLR; 99817d2a487SOliver O'Halloran eeh_ops->write_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL, 9999312bc5bSWei Yang 4, reg); 10009312bc5bSWei Yang msleep(EEH_PE_RST_HOLD_TIME); 10019312bc5bSWei Yang break; 10029312bc5bSWei Yang case EEH_RESET_DEACTIVATE: 100317d2a487SOliver O'Halloran eeh_ops->read_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL, 10049312bc5bSWei Yang 4, ®); 10059312bc5bSWei Yang reg &= ~PCI_EXP_DEVCTL_BCR_FLR; 100617d2a487SOliver O'Halloran eeh_ops->write_config(edev, edev->pcie_cap + PCI_EXP_DEVCTL, 10079312bc5bSWei Yang 4, reg); 10089312bc5bSWei Yang msleep(EEH_PE_RST_SETTLE_TIME); 10099312bc5bSWei Yang break; 10109312bc5bSWei Yang } 10119312bc5bSWei Yang 10129312bc5bSWei Yang return 0; 10139312bc5bSWei Yang } 10149312bc5bSWei Yang 10159312bc5bSWei Yang static int pnv_eeh_do_af_flr(struct pci_dn *pdn, int option) 10169312bc5bSWei Yang { 10179312bc5bSWei Yang struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 10189312bc5bSWei Yang u32 cap = 0; 10199312bc5bSWei Yang 10209312bc5bSWei Yang if (WARN_ON(!edev->af_cap)) 10219312bc5bSWei Yang return -ENOTTY; 10229312bc5bSWei Yang 102317d2a487SOliver O'Halloran eeh_ops->read_config(edev, edev->af_cap + PCI_AF_CAP, 1, &cap); 10249312bc5bSWei Yang if (!(cap & PCI_AF_CAP_TP) || !(cap & PCI_AF_CAP_FLR)) 10259312bc5bSWei Yang return -ENOTTY; 10269312bc5bSWei Yang 10279312bc5bSWei Yang switch (option) { 10289312bc5bSWei Yang case EEH_RESET_HOT: 10299312bc5bSWei Yang case EEH_RESET_FUNDAMENTAL: 10309312bc5bSWei Yang /* 10319312bc5bSWei Yang * Wait for Transaction Pending bit to clear. A word-aligned 10329312bc5bSWei Yang * test is used, so we use the conrol offset rather than status 10339312bc5bSWei Yang * and shift the test bit to match. 10349312bc5bSWei Yang */ 10359312bc5bSWei Yang pnv_eeh_wait_for_pending(pdn, "AF", 10369312bc5bSWei Yang edev->af_cap + PCI_AF_CTRL, 10379312bc5bSWei Yang PCI_AF_STATUS_TP << 8); 103817d2a487SOliver O'Halloran eeh_ops->write_config(edev, edev->af_cap + PCI_AF_CTRL, 10399312bc5bSWei Yang 1, PCI_AF_CTRL_FLR); 10409312bc5bSWei Yang msleep(EEH_PE_RST_HOLD_TIME); 10419312bc5bSWei Yang break; 10429312bc5bSWei Yang case EEH_RESET_DEACTIVATE: 104317d2a487SOliver O'Halloran eeh_ops->write_config(edev, edev->af_cap + PCI_AF_CTRL, 1, 0); 10449312bc5bSWei Yang msleep(EEH_PE_RST_SETTLE_TIME); 10459312bc5bSWei Yang break; 10469312bc5bSWei Yang } 10479312bc5bSWei Yang 10489312bc5bSWei Yang return 0; 10499312bc5bSWei Yang } 10509312bc5bSWei Yang 10519312bc5bSWei Yang static int pnv_eeh_reset_vf_pe(struct eeh_pe *pe, int option) 10529312bc5bSWei Yang { 10539312bc5bSWei Yang struct eeh_dev *edev; 10549312bc5bSWei Yang struct pci_dn *pdn; 10559312bc5bSWei Yang int ret; 10569312bc5bSWei Yang 10579312bc5bSWei Yang /* The VF PE should have only one child device */ 105880e65b00SSam Bobroff edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, entry); 10599312bc5bSWei Yang pdn = eeh_dev_to_pdn(edev); 10609312bc5bSWei Yang if (!pdn) 10619312bc5bSWei Yang return -ENXIO; 10629312bc5bSWei Yang 10639312bc5bSWei Yang ret = pnv_eeh_do_flr(pdn, option); 10649312bc5bSWei Yang if (!ret) 10659312bc5bSWei Yang return ret; 10669312bc5bSWei Yang 10679312bc5bSWei Yang return pnv_eeh_do_af_flr(pdn, option); 10689312bc5bSWei Yang } 10699312bc5bSWei Yang 107029310e5eSGavin Shan /** 107101f3bfb7SGavin Shan * pnv_eeh_reset - Reset the specified PE 107229310e5eSGavin Shan * @pe: EEH PE 107329310e5eSGavin Shan * @option: reset option 107429310e5eSGavin Shan * 1075cadf364dSGavin Shan * Do reset on the indicated PE. For PCI bus sensitive PE, 1076cadf364dSGavin Shan * we need to reset the parent p2p bridge. The PHB has to 1077cadf364dSGavin Shan * be reinitialized if the p2p bridge is root bridge. For 1078cadf364dSGavin Shan * PCI device sensitive PE, we will try to reset the device 1079cadf364dSGavin Shan * through FLR. For now, we don't have OPAL APIs to do HARD 1080cadf364dSGavin Shan * reset yet, so all reset would be SOFT (HOT) reset. 108129310e5eSGavin Shan */ 108201f3bfb7SGavin Shan static int pnv_eeh_reset(struct eeh_pe *pe, int option) 108329310e5eSGavin Shan { 108429310e5eSGavin Shan struct pci_controller *hose = pe->phb; 10854fad4943SGavin Shan struct pnv_phb *phb; 1086cadf364dSGavin Shan struct pci_bus *bus; 10874fad4943SGavin Shan int64_t rc; 108829310e5eSGavin Shan 1089cadf364dSGavin Shan /* 1090cadf364dSGavin Shan * For PHB reset, we always have complete reset. For those PEs whose 1091cadf364dSGavin Shan * primary bus derived from root complex (root bus) or root port 1092cadf364dSGavin Shan * (usually bus#1), we apply hot or fundamental reset on the root port. 1093cadf364dSGavin Shan * For other PEs, we always have hot reset on the PE primary bus. 1094cadf364dSGavin Shan * 1095cadf364dSGavin Shan * Here, we have different design to pHyp, which always clear the 1096cadf364dSGavin Shan * frozen state during PE reset. However, the good idea here from 1097cadf364dSGavin Shan * benh is to keep frozen state before we get PE reset done completely 1098cadf364dSGavin Shan * (until BAR restore). With the frozen state, HW drops illegal IO 1099cadf364dSGavin Shan * or MMIO access, which can incur recrusive frozen PE during PE 1100cadf364dSGavin Shan * reset. The side effect is that EEH core has to clear the frozen 1101cadf364dSGavin Shan * state explicitly after BAR restore. 1102cadf364dSGavin Shan */ 11034fad4943SGavin Shan if (pe->type & EEH_PE_PHB) 11044fad4943SGavin Shan return pnv_eeh_phb_reset(hose, option); 1105cadf364dSGavin Shan 1106cadf364dSGavin Shan /* 1107cadf364dSGavin Shan * The frozen PE might be caused by PAPR error injection 1108cadf364dSGavin Shan * registers, which are expected to be cleared after hitting 1109cadf364dSGavin Shan * frozen PE as stated in the hardware spec. Unfortunately, 1110cadf364dSGavin Shan * that's not true on P7IOC. So we have to clear it manually 1111cadf364dSGavin Shan * to avoid recursive EEH errors during recovery. 1112cadf364dSGavin Shan */ 1113cadf364dSGavin Shan phb = hose->private_data; 1114cadf364dSGavin Shan if (phb->model == PNV_PHB_MODEL_P7IOC && 1115cadf364dSGavin Shan (option == EEH_RESET_HOT || 1116cadf364dSGavin Shan option == EEH_RESET_FUNDAMENTAL)) { 1117cadf364dSGavin Shan rc = opal_pci_reset(phb->opal_id, 1118cadf364dSGavin Shan OPAL_RESET_PHB_ERROR, 1119cadf364dSGavin Shan OPAL_ASSERT_RESET); 1120cadf364dSGavin Shan if (rc != OPAL_SUCCESS) { 11214fad4943SGavin Shan pr_warn("%s: Failure %lld clearing error injection registers\n", 1122cadf364dSGavin Shan __func__, rc); 1123cadf364dSGavin Shan return -EIO; 1124cadf364dSGavin Shan } 1125cadf364dSGavin Shan } 1126cadf364dSGavin Shan 1127e98ddb77SRussell Currey if (pe->type & EEH_PE_VF) 1128e98ddb77SRussell Currey return pnv_eeh_reset_vf_pe(pe, option); 1129e98ddb77SRussell Currey 1130cadf364dSGavin Shan bus = eeh_pe_bus_get(pe); 113104fec21cSRussell Currey if (!bus) { 11321f52f176SRussell Currey pr_err("%s: Cannot find PCI bus for PHB#%x-PE#%x\n", 113304fec21cSRussell Currey __func__, pe->phb->global_number, pe->addr); 113404fec21cSRussell Currey return -EIO; 113504fec21cSRussell Currey } 113629310e5eSGavin Shan 113798fd32cdSOliver O'Halloran if (pci_is_root_bus(bus)) 11384fad4943SGavin Shan return pnv_eeh_root_reset(hose, option); 11394fad4943SGavin Shan 114098fd32cdSOliver O'Halloran /* 114198fd32cdSOliver O'Halloran * For hot resets try use the generic PCI error recovery reset 114298fd32cdSOliver O'Halloran * functions. These correctly handles the case where the secondary 114398fd32cdSOliver O'Halloran * bus is behind a hotplug slot and it will use the slot provided 114498fd32cdSOliver O'Halloran * reset methods to prevent spurious hotplug events during the reset. 114598fd32cdSOliver O'Halloran * 114698fd32cdSOliver O'Halloran * Fundemental resets need to be handled internally to EEH since the 114798fd32cdSOliver O'Halloran * PCI core doesn't really have a concept of a fundemental reset, 114898fd32cdSOliver O'Halloran * mainly because there's no standard way to generate one. Only a 114998fd32cdSOliver O'Halloran * few devices require an FRESET so it should be fine. 115098fd32cdSOliver O'Halloran */ 115198fd32cdSOliver O'Halloran if (option != EEH_RESET_FUNDAMENTAL) { 115298fd32cdSOliver O'Halloran /* 115398fd32cdSOliver O'Halloran * NB: Skiboot and pnv_eeh_bridge_reset() also no-op the 115498fd32cdSOliver O'Halloran * de-assert step. It's like the OPAL reset API was 115598fd32cdSOliver O'Halloran * poorly designed or something... 115698fd32cdSOliver O'Halloran */ 115798fd32cdSOliver O'Halloran if (option == EEH_RESET_DEACTIVATE) 115898fd32cdSOliver O'Halloran return 0; 115998fd32cdSOliver O'Halloran 116098fd32cdSOliver O'Halloran rc = pci_bus_error_reset(bus->self); 116198fd32cdSOliver O'Halloran if (!rc) 116298fd32cdSOliver O'Halloran return 0; 116398fd32cdSOliver O'Halloran } 116498fd32cdSOliver O'Halloran 116598fd32cdSOliver O'Halloran /* otherwise, use the generic bridge reset. this might call into FW */ 116698fd32cdSOliver O'Halloran if (pci_is_root_bus(bus->parent)) 116798fd32cdSOliver O'Halloran return pnv_eeh_root_reset(hose, option); 11684fad4943SGavin Shan return pnv_eeh_bridge_reset(bus->self, option); 116929310e5eSGavin Shan } 117029310e5eSGavin Shan 117129310e5eSGavin Shan /** 117201f3bfb7SGavin Shan * pnv_eeh_get_log - Retrieve error log 117329310e5eSGavin Shan * @pe: EEH PE 117429310e5eSGavin Shan * @severity: temporary or permanent error log 117529310e5eSGavin Shan * @drv_log: driver log to be combined with retrieved error log 117629310e5eSGavin Shan * @len: length of driver log 117729310e5eSGavin Shan * 117829310e5eSGavin Shan * Retrieve the temporary or permanent error from the PE. 117929310e5eSGavin Shan */ 118001f3bfb7SGavin Shan static int pnv_eeh_get_log(struct eeh_pe *pe, int severity, 118129310e5eSGavin Shan char *drv_log, unsigned long len) 118229310e5eSGavin Shan { 118395edcdeaSGavin Shan if (!eeh_has_flag(EEH_EARLY_DUMP_LOG)) 118495edcdeaSGavin Shan pnv_pci_dump_phb_diag_data(pe->phb, pe->data); 118529310e5eSGavin Shan 118695edcdeaSGavin Shan return 0; 118729310e5eSGavin Shan } 118829310e5eSGavin Shan 118929310e5eSGavin Shan /** 119001f3bfb7SGavin Shan * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE 119129310e5eSGavin Shan * @pe: EEH PE 119229310e5eSGavin Shan * 119329310e5eSGavin Shan * The function will be called to reconfigure the bridges included 119429310e5eSGavin Shan * in the specified PE so that the mulfunctional PE would be recovered 119529310e5eSGavin Shan * again. 119629310e5eSGavin Shan */ 119701f3bfb7SGavin Shan static int pnv_eeh_configure_bridge(struct eeh_pe *pe) 119829310e5eSGavin Shan { 1199bbe170edSGavin Shan return 0; 120029310e5eSGavin Shan } 120129310e5eSGavin Shan 120229310e5eSGavin Shan /** 120301f3bfb7SGavin Shan * pnv_pe_err_inject - Inject specified error to the indicated PE 1204131c123aSGavin Shan * @pe: the indicated PE 1205131c123aSGavin Shan * @type: error type 1206131c123aSGavin Shan * @func: specific error type 1207131c123aSGavin Shan * @addr: address 1208131c123aSGavin Shan * @mask: address mask 1209131c123aSGavin Shan * 1210131c123aSGavin Shan * The routine is called to inject specified error, which is 1211131c123aSGavin Shan * determined by @type and @func, to the indicated PE for 1212131c123aSGavin Shan * testing purpose. 1213131c123aSGavin Shan */ 121401f3bfb7SGavin Shan static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func, 1215131c123aSGavin Shan unsigned long addr, unsigned long mask) 1216131c123aSGavin Shan { 1217131c123aSGavin Shan struct pci_controller *hose = pe->phb; 1218131c123aSGavin Shan struct pnv_phb *phb = hose->private_data; 1219fa646c3cSGavin Shan s64 rc; 1220131c123aSGavin Shan 1221fa646c3cSGavin Shan if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR && 1222fa646c3cSGavin Shan type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) { 1223fa646c3cSGavin Shan pr_warn("%s: Invalid error type %d\n", 1224fa646c3cSGavin Shan __func__, type); 1225fa646c3cSGavin Shan return -ERANGE; 1226fa646c3cSGavin Shan } 1227131c123aSGavin Shan 1228fa646c3cSGavin Shan if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR || 1229fa646c3cSGavin Shan func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) { 1230fa646c3cSGavin Shan pr_warn("%s: Invalid error function %d\n", 1231fa646c3cSGavin Shan __func__, func); 1232fa646c3cSGavin Shan return -ERANGE; 1233fa646c3cSGavin Shan } 1234fa646c3cSGavin Shan 1235fa646c3cSGavin Shan /* Firmware supports error injection ? */ 1236fa646c3cSGavin Shan if (!opal_check_token(OPAL_PCI_ERR_INJECT)) { 1237fa646c3cSGavin Shan pr_warn("%s: Firmware doesn't support error injection\n", 1238fa646c3cSGavin Shan __func__); 1239fa646c3cSGavin Shan return -ENXIO; 1240fa646c3cSGavin Shan } 1241fa646c3cSGavin Shan 1242fa646c3cSGavin Shan /* Do error injection */ 1243fa646c3cSGavin Shan rc = opal_pci_err_inject(phb->opal_id, pe->addr, 1244fa646c3cSGavin Shan type, func, addr, mask); 1245fa646c3cSGavin Shan if (rc != OPAL_SUCCESS) { 1246fa646c3cSGavin Shan pr_warn("%s: Failure %lld injecting error " 1247fa646c3cSGavin Shan "%d-%d to PHB#%x-PE#%x\n", 1248fa646c3cSGavin Shan __func__, rc, type, func, 1249fa646c3cSGavin Shan hose->global_number, pe->addr); 1250fa646c3cSGavin Shan return -EIO; 1251fa646c3cSGavin Shan } 1252fa646c3cSGavin Shan 1253fa646c3cSGavin Shan return 0; 1254131c123aSGavin Shan } 1255131c123aSGavin Shan 12560bd78587SGavin Shan static inline bool pnv_eeh_cfg_blocked(struct pci_dn *pdn) 1257d2cfbcd7SGavin Shan { 12580bd78587SGavin Shan struct eeh_dev *edev = pdn_to_eeh_dev(pdn); 1259d2cfbcd7SGavin Shan 1260d2cfbcd7SGavin Shan if (!edev || !edev->pe) 1261d2cfbcd7SGavin Shan return false; 1262d2cfbcd7SGavin Shan 12639312bc5bSWei Yang /* 12649312bc5bSWei Yang * We will issue FLR or AF FLR to all VFs, which are contained 12659312bc5bSWei Yang * in VF PE. It relies on the EEH PCI config accessors. So we 12669312bc5bSWei Yang * can't block them during the window. 12679312bc5bSWei Yang */ 12689312bc5bSWei Yang if (edev->physfn && (edev->pe->state & EEH_PE_RESET)) 12699312bc5bSWei Yang return false; 12709312bc5bSWei Yang 1271d2cfbcd7SGavin Shan if (edev->pe->state & EEH_PE_CFG_BLOCKED) 1272d2cfbcd7SGavin Shan return true; 1273d2cfbcd7SGavin Shan 1274d2cfbcd7SGavin Shan return false; 1275d2cfbcd7SGavin Shan } 1276d2cfbcd7SGavin Shan 127717d2a487SOliver O'Halloran static int pnv_eeh_read_config(struct eeh_dev *edev, 1278d2cfbcd7SGavin Shan int where, int size, u32 *val) 1279d2cfbcd7SGavin Shan { 128017d2a487SOliver O'Halloran struct pci_dn *pdn = eeh_dev_to_pdn(edev); 128117d2a487SOliver O'Halloran 12823532a741SGavin Shan if (!pdn) 12833532a741SGavin Shan return PCIBIOS_DEVICE_NOT_FOUND; 12843532a741SGavin Shan 12850bd78587SGavin Shan if (pnv_eeh_cfg_blocked(pdn)) { 1286d2cfbcd7SGavin Shan *val = 0xFFFFFFFF; 1287d2cfbcd7SGavin Shan return PCIBIOS_SET_FAILED; 1288d2cfbcd7SGavin Shan } 1289d2cfbcd7SGavin Shan 12903532a741SGavin Shan return pnv_pci_cfg_read(pdn, where, size, val); 1291d2cfbcd7SGavin Shan } 1292d2cfbcd7SGavin Shan 129317d2a487SOliver O'Halloran static int pnv_eeh_write_config(struct eeh_dev *edev, 1294d2cfbcd7SGavin Shan int where, int size, u32 val) 1295d2cfbcd7SGavin Shan { 129617d2a487SOliver O'Halloran struct pci_dn *pdn = eeh_dev_to_pdn(edev); 129717d2a487SOliver O'Halloran 12983532a741SGavin Shan if (!pdn) 12993532a741SGavin Shan return PCIBIOS_DEVICE_NOT_FOUND; 13003532a741SGavin Shan 13010bd78587SGavin Shan if (pnv_eeh_cfg_blocked(pdn)) 1302d2cfbcd7SGavin Shan return PCIBIOS_SET_FAILED; 1303d2cfbcd7SGavin Shan 13043532a741SGavin Shan return pnv_pci_cfg_write(pdn, where, size, val); 1305d2cfbcd7SGavin Shan } 1306d2cfbcd7SGavin Shan 13072a485ad7SGavin Shan static void pnv_eeh_dump_hub_diag_common(struct OpalIoP7IOCErrorData *data) 13082a485ad7SGavin Shan { 13092a485ad7SGavin Shan /* GEM */ 13102a485ad7SGavin Shan if (data->gemXfir || data->gemRfir || 13112a485ad7SGavin Shan data->gemRirqfir || data->gemMask || data->gemRwof) 13122a485ad7SGavin Shan pr_info(" GEM: %016llx %016llx %016llx %016llx %016llx\n", 13132a485ad7SGavin Shan be64_to_cpu(data->gemXfir), 13142a485ad7SGavin Shan be64_to_cpu(data->gemRfir), 13152a485ad7SGavin Shan be64_to_cpu(data->gemRirqfir), 13162a485ad7SGavin Shan be64_to_cpu(data->gemMask), 13172a485ad7SGavin Shan be64_to_cpu(data->gemRwof)); 13182a485ad7SGavin Shan 13192a485ad7SGavin Shan /* LEM */ 13202a485ad7SGavin Shan if (data->lemFir || data->lemErrMask || 13212a485ad7SGavin Shan data->lemAction0 || data->lemAction1 || data->lemWof) 13222a485ad7SGavin Shan pr_info(" LEM: %016llx %016llx %016llx %016llx %016llx\n", 13232a485ad7SGavin Shan be64_to_cpu(data->lemFir), 13242a485ad7SGavin Shan be64_to_cpu(data->lemErrMask), 13252a485ad7SGavin Shan be64_to_cpu(data->lemAction0), 13262a485ad7SGavin Shan be64_to_cpu(data->lemAction1), 13272a485ad7SGavin Shan be64_to_cpu(data->lemWof)); 13282a485ad7SGavin Shan } 13292a485ad7SGavin Shan 13302a485ad7SGavin Shan static void pnv_eeh_get_and_dump_hub_diag(struct pci_controller *hose) 13312a485ad7SGavin Shan { 13322a485ad7SGavin Shan struct pnv_phb *phb = hose->private_data; 13335cb1f8fdSRussell Currey struct OpalIoP7IOCErrorData *data = 13345cb1f8fdSRussell Currey (struct OpalIoP7IOCErrorData*)phb->diag_data; 13352a485ad7SGavin Shan long rc; 13362a485ad7SGavin Shan 13372a485ad7SGavin Shan rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data)); 13382a485ad7SGavin Shan if (rc != OPAL_SUCCESS) { 13392a485ad7SGavin Shan pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n", 13402a485ad7SGavin Shan __func__, phb->hub_id, rc); 13412a485ad7SGavin Shan return; 13422a485ad7SGavin Shan } 13432a485ad7SGavin Shan 1344a7032132SGavin Shan switch (be16_to_cpu(data->type)) { 13452a485ad7SGavin Shan case OPAL_P7IOC_DIAG_TYPE_RGC: 13462a485ad7SGavin Shan pr_info("P7IOC diag-data for RGC\n\n"); 13472a485ad7SGavin Shan pnv_eeh_dump_hub_diag_common(data); 13482a485ad7SGavin Shan if (data->rgc.rgcStatus || data->rgc.rgcLdcp) 13492a485ad7SGavin Shan pr_info(" RGC: %016llx %016llx\n", 13502a485ad7SGavin Shan be64_to_cpu(data->rgc.rgcStatus), 13512a485ad7SGavin Shan be64_to_cpu(data->rgc.rgcLdcp)); 13522a485ad7SGavin Shan break; 13532a485ad7SGavin Shan case OPAL_P7IOC_DIAG_TYPE_BI: 13542a485ad7SGavin Shan pr_info("P7IOC diag-data for BI %s\n\n", 13552a485ad7SGavin Shan data->bi.biDownbound ? "Downbound" : "Upbound"); 13562a485ad7SGavin Shan pnv_eeh_dump_hub_diag_common(data); 13572a485ad7SGavin Shan if (data->bi.biLdcp0 || data->bi.biLdcp1 || 13582a485ad7SGavin Shan data->bi.biLdcp2 || data->bi.biFenceStatus) 13592a485ad7SGavin Shan pr_info(" BI: %016llx %016llx %016llx %016llx\n", 13602a485ad7SGavin Shan be64_to_cpu(data->bi.biLdcp0), 13612a485ad7SGavin Shan be64_to_cpu(data->bi.biLdcp1), 13622a485ad7SGavin Shan be64_to_cpu(data->bi.biLdcp2), 13632a485ad7SGavin Shan be64_to_cpu(data->bi.biFenceStatus)); 13642a485ad7SGavin Shan break; 13652a485ad7SGavin Shan case OPAL_P7IOC_DIAG_TYPE_CI: 13662a485ad7SGavin Shan pr_info("P7IOC diag-data for CI Port %d\n\n", 13672a485ad7SGavin Shan data->ci.ciPort); 13682a485ad7SGavin Shan pnv_eeh_dump_hub_diag_common(data); 13692a485ad7SGavin Shan if (data->ci.ciPortStatus || data->ci.ciPortLdcp) 13702a485ad7SGavin Shan pr_info(" CI: %016llx %016llx\n", 13712a485ad7SGavin Shan be64_to_cpu(data->ci.ciPortStatus), 13722a485ad7SGavin Shan be64_to_cpu(data->ci.ciPortLdcp)); 13732a485ad7SGavin Shan break; 13742a485ad7SGavin Shan case OPAL_P7IOC_DIAG_TYPE_MISC: 13752a485ad7SGavin Shan pr_info("P7IOC diag-data for MISC\n\n"); 13762a485ad7SGavin Shan pnv_eeh_dump_hub_diag_common(data); 13772a485ad7SGavin Shan break; 13782a485ad7SGavin Shan case OPAL_P7IOC_DIAG_TYPE_I2C: 13792a485ad7SGavin Shan pr_info("P7IOC diag-data for I2C\n\n"); 13802a485ad7SGavin Shan pnv_eeh_dump_hub_diag_common(data); 13812a485ad7SGavin Shan break; 13822a485ad7SGavin Shan default: 13832a485ad7SGavin Shan pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n", 13842a485ad7SGavin Shan __func__, phb->hub_id, data->type); 13852a485ad7SGavin Shan } 13862a485ad7SGavin Shan } 13872a485ad7SGavin Shan 13882a485ad7SGavin Shan static int pnv_eeh_get_pe(struct pci_controller *hose, 13892a485ad7SGavin Shan u16 pe_no, struct eeh_pe **pe) 13902a485ad7SGavin Shan { 13912a485ad7SGavin Shan struct pnv_phb *phb = hose->private_data; 13922a485ad7SGavin Shan struct pnv_ioda_pe *pnv_pe; 13932a485ad7SGavin Shan struct eeh_pe *dev_pe; 13942a485ad7SGavin Shan 13952a485ad7SGavin Shan /* 13962a485ad7SGavin Shan * If PHB supports compound PE, to fetch 13972a485ad7SGavin Shan * the master PE because slave PE is invisible 13982a485ad7SGavin Shan * to EEH core. 13992a485ad7SGavin Shan */ 14002a485ad7SGavin Shan pnv_pe = &phb->ioda.pe_array[pe_no]; 14012a485ad7SGavin Shan if (pnv_pe->flags & PNV_IODA_PE_SLAVE) { 14022a485ad7SGavin Shan pnv_pe = pnv_pe->master; 14032a485ad7SGavin Shan WARN_ON(!pnv_pe || 14042a485ad7SGavin Shan !(pnv_pe->flags & PNV_IODA_PE_MASTER)); 14052a485ad7SGavin Shan pe_no = pnv_pe->pe_number; 14062a485ad7SGavin Shan } 14072a485ad7SGavin Shan 14082a485ad7SGavin Shan /* Find the PE according to PE# */ 14098bae6a23SAlexey Kardashevskiy dev_pe = eeh_pe_get(hose, pe_no, 0); 14102a485ad7SGavin Shan if (!dev_pe) 14112a485ad7SGavin Shan return -EEXIST; 14122a485ad7SGavin Shan 14132a485ad7SGavin Shan /* Freeze the (compound) PE */ 14142a485ad7SGavin Shan *pe = dev_pe; 14152a485ad7SGavin Shan if (!(dev_pe->state & EEH_PE_ISOLATED)) 14162a485ad7SGavin Shan phb->freeze_pe(phb, pe_no); 14172a485ad7SGavin Shan 14182a485ad7SGavin Shan /* 14192a485ad7SGavin Shan * At this point, we're sure the (compound) PE should 14202a485ad7SGavin Shan * have been frozen. However, we still need poke until 14212a485ad7SGavin Shan * hitting the frozen PE on top level. 14222a485ad7SGavin Shan */ 14232a485ad7SGavin Shan dev_pe = dev_pe->parent; 14242a485ad7SGavin Shan while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) { 14252a485ad7SGavin Shan int ret; 14262a485ad7SGavin Shan ret = eeh_ops->get_state(dev_pe, NULL); 142734a286a4SSam Bobroff if (ret <= 0 || eeh_state_active(ret)) { 14282a485ad7SGavin Shan dev_pe = dev_pe->parent; 14292a485ad7SGavin Shan continue; 14302a485ad7SGavin Shan } 14312a485ad7SGavin Shan 14322a485ad7SGavin Shan /* Frozen parent PE */ 14332a485ad7SGavin Shan *pe = dev_pe; 14342a485ad7SGavin Shan if (!(dev_pe->state & EEH_PE_ISOLATED)) 14352a485ad7SGavin Shan phb->freeze_pe(phb, dev_pe->addr); 14362a485ad7SGavin Shan 14372a485ad7SGavin Shan /* Next one */ 14382a485ad7SGavin Shan dev_pe = dev_pe->parent; 14392a485ad7SGavin Shan } 14402a485ad7SGavin Shan 14412a485ad7SGavin Shan return 0; 14422a485ad7SGavin Shan } 14432a485ad7SGavin Shan 1444131c123aSGavin Shan /** 144501f3bfb7SGavin Shan * pnv_eeh_next_error - Retrieve next EEH error to handle 144629310e5eSGavin Shan * @pe: Affected PE 144729310e5eSGavin Shan * 14482a485ad7SGavin Shan * The function is expected to be called by EEH core while it gets 14492a485ad7SGavin Shan * special EEH event (without binding PE). The function calls to 14502a485ad7SGavin Shan * OPAL APIs for next error to handle. The informational error is 14512a485ad7SGavin Shan * handled internally by platform. However, the dead IOC, dead PHB, 14522a485ad7SGavin Shan * fenced PHB and frozen PE should be handled by EEH core eventually. 145329310e5eSGavin Shan */ 145401f3bfb7SGavin Shan static int pnv_eeh_next_error(struct eeh_pe **pe) 145529310e5eSGavin Shan { 145629310e5eSGavin Shan struct pci_controller *hose; 14572a485ad7SGavin Shan struct pnv_phb *phb; 14582a485ad7SGavin Shan struct eeh_pe *phb_pe, *parent_pe; 14592a485ad7SGavin Shan __be64 frozen_pe_no; 14602a485ad7SGavin Shan __be16 err_type, severity; 14612a485ad7SGavin Shan long rc; 14622a485ad7SGavin Shan int state, ret = EEH_NEXT_ERR_NONE; 14632a485ad7SGavin Shan 14642a485ad7SGavin Shan /* 146579231448SAlistair Popple * While running here, it's safe to purge the event queue. The 146679231448SAlistair Popple * event should still be masked. 14672a485ad7SGavin Shan */ 14682a485ad7SGavin Shan eeh_remove_event(NULL, false); 146929310e5eSGavin Shan 147029310e5eSGavin Shan list_for_each_entry(hose, &hose_list, list_node) { 14712a485ad7SGavin Shan /* 14722a485ad7SGavin Shan * If the subordinate PCI buses of the PHB has been 14732a485ad7SGavin Shan * removed or is exactly under error recovery, we 14742a485ad7SGavin Shan * needn't take care of it any more. 14752a485ad7SGavin Shan */ 147629310e5eSGavin Shan phb = hose->private_data; 14772a485ad7SGavin Shan phb_pe = eeh_phb_pe_get(hose); 14782a485ad7SGavin Shan if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED)) 14792a485ad7SGavin Shan continue; 14802a485ad7SGavin Shan 14812a485ad7SGavin Shan rc = opal_pci_next_error(phb->opal_id, 14822a485ad7SGavin Shan &frozen_pe_no, &err_type, &severity); 14832a485ad7SGavin Shan if (rc != OPAL_SUCCESS) { 14842a485ad7SGavin Shan pr_devel("%s: Invalid return value on " 14852a485ad7SGavin Shan "PHB#%x (0x%lx) from opal_pci_next_error", 14862a485ad7SGavin Shan __func__, hose->global_number, rc); 14872a485ad7SGavin Shan continue; 14882a485ad7SGavin Shan } 14892a485ad7SGavin Shan 14902a485ad7SGavin Shan /* If the PHB doesn't have error, stop processing */ 14912a485ad7SGavin Shan if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR || 14922a485ad7SGavin Shan be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) { 14932a485ad7SGavin Shan pr_devel("%s: No error found on PHB#%x\n", 14942a485ad7SGavin Shan __func__, hose->global_number); 14952a485ad7SGavin Shan continue; 14962a485ad7SGavin Shan } 14972a485ad7SGavin Shan 14982a485ad7SGavin Shan /* 14992a485ad7SGavin Shan * Processing the error. We're expecting the error with 15002a485ad7SGavin Shan * highest priority reported upon multiple errors on the 15012a485ad7SGavin Shan * specific PHB. 15022a485ad7SGavin Shan */ 15032a485ad7SGavin Shan pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n", 15042a485ad7SGavin Shan __func__, be16_to_cpu(err_type), 15052a485ad7SGavin Shan be16_to_cpu(severity), be64_to_cpu(frozen_pe_no), 15062a485ad7SGavin Shan hose->global_number); 15072a485ad7SGavin Shan switch (be16_to_cpu(err_type)) { 15082a485ad7SGavin Shan case OPAL_EEH_IOC_ERROR: 15092a485ad7SGavin Shan if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) { 15102a485ad7SGavin Shan pr_err("EEH: dead IOC detected\n"); 15112a485ad7SGavin Shan ret = EEH_NEXT_ERR_DEAD_IOC; 15122a485ad7SGavin Shan } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) { 15132a485ad7SGavin Shan pr_info("EEH: IOC informative error " 15142a485ad7SGavin Shan "detected\n"); 15152a485ad7SGavin Shan pnv_eeh_get_and_dump_hub_diag(hose); 15162a485ad7SGavin Shan ret = EEH_NEXT_ERR_NONE; 15172a485ad7SGavin Shan } 15182a485ad7SGavin Shan 15192a485ad7SGavin Shan break; 15202a485ad7SGavin Shan case OPAL_EEH_PHB_ERROR: 15212a485ad7SGavin Shan if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) { 15222a485ad7SGavin Shan *pe = phb_pe; 15232a485ad7SGavin Shan pr_err("EEH: dead PHB#%x detected, " 15242a485ad7SGavin Shan "location: %s\n", 15252a485ad7SGavin Shan hose->global_number, 15262a485ad7SGavin Shan eeh_pe_loc_get(phb_pe)); 15272a485ad7SGavin Shan ret = EEH_NEXT_ERR_DEAD_PHB; 15282a485ad7SGavin Shan } else if (be16_to_cpu(severity) == 15292a485ad7SGavin Shan OPAL_EEH_SEV_PHB_FENCED) { 15302a485ad7SGavin Shan *pe = phb_pe; 15312a485ad7SGavin Shan pr_err("EEH: Fenced PHB#%x detected, " 15322a485ad7SGavin Shan "location: %s\n", 15332a485ad7SGavin Shan hose->global_number, 15342a485ad7SGavin Shan eeh_pe_loc_get(phb_pe)); 15352a485ad7SGavin Shan ret = EEH_NEXT_ERR_FENCED_PHB; 15362a485ad7SGavin Shan } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) { 15372a485ad7SGavin Shan pr_info("EEH: PHB#%x informative error " 15382a485ad7SGavin Shan "detected, location: %s\n", 15392a485ad7SGavin Shan hose->global_number, 15402a485ad7SGavin Shan eeh_pe_loc_get(phb_pe)); 15412a485ad7SGavin Shan pnv_eeh_get_phb_diag(phb_pe); 15422a485ad7SGavin Shan pnv_pci_dump_phb_diag_data(hose, phb_pe->data); 15432a485ad7SGavin Shan ret = EEH_NEXT_ERR_NONE; 15442a485ad7SGavin Shan } 15452a485ad7SGavin Shan 15462a485ad7SGavin Shan break; 15472a485ad7SGavin Shan case OPAL_EEH_PE_ERROR: 15482a485ad7SGavin Shan /* 15492a485ad7SGavin Shan * If we can't find the corresponding PE, we 15502a485ad7SGavin Shan * just try to unfreeze. 15512a485ad7SGavin Shan */ 15522a485ad7SGavin Shan if (pnv_eeh_get_pe(hose, 15532a485ad7SGavin Shan be64_to_cpu(frozen_pe_no), pe)) { 15542a485ad7SGavin Shan pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n", 15550f36db77SGavin Shan hose->global_number, be64_to_cpu(frozen_pe_no)); 15562a485ad7SGavin Shan pr_info("EEH: PHB location: %s\n", 15572a485ad7SGavin Shan eeh_pe_loc_get(phb_pe)); 155879cd9520SGavin Shan 155979cd9520SGavin Shan /* Dump PHB diag-data */ 156079cd9520SGavin Shan rc = opal_pci_get_phb_diag_data2(phb->opal_id, 15615cb1f8fdSRussell Currey phb->diag_data, phb->diag_data_size); 156279cd9520SGavin Shan if (rc == OPAL_SUCCESS) 156379cd9520SGavin Shan pnv_pci_dump_phb_diag_data(hose, 15645cb1f8fdSRussell Currey phb->diag_data); 156579cd9520SGavin Shan 156679cd9520SGavin Shan /* Try best to clear it */ 15672a485ad7SGavin Shan opal_pci_eeh_freeze_clear(phb->opal_id, 1568d63e51b3SGavin Shan be64_to_cpu(frozen_pe_no), 15692a485ad7SGavin Shan OPAL_EEH_ACTION_CLEAR_FREEZE_ALL); 15702a485ad7SGavin Shan ret = EEH_NEXT_ERR_NONE; 15712a485ad7SGavin Shan } else if ((*pe)->state & EEH_PE_ISOLATED || 15722a485ad7SGavin Shan eeh_pe_passed(*pe)) { 15732a485ad7SGavin Shan ret = EEH_NEXT_ERR_NONE; 15742a485ad7SGavin Shan } else { 15752a485ad7SGavin Shan pr_err("EEH: Frozen PE#%x " 15762a485ad7SGavin Shan "on PHB#%x detected\n", 15772a485ad7SGavin Shan (*pe)->addr, 15782a485ad7SGavin Shan (*pe)->phb->global_number); 15792a485ad7SGavin Shan pr_err("EEH: PE location: %s, " 15802a485ad7SGavin Shan "PHB location: %s\n", 15812a485ad7SGavin Shan eeh_pe_loc_get(*pe), 15822a485ad7SGavin Shan eeh_pe_loc_get(phb_pe)); 15832a485ad7SGavin Shan ret = EEH_NEXT_ERR_FROZEN_PE; 15842a485ad7SGavin Shan } 15852a485ad7SGavin Shan 15862a485ad7SGavin Shan break; 15872a485ad7SGavin Shan default: 15882a485ad7SGavin Shan pr_warn("%s: Unexpected error type %d\n", 15892a485ad7SGavin Shan __func__, be16_to_cpu(err_type)); 15902a485ad7SGavin Shan } 15912a485ad7SGavin Shan 15922a485ad7SGavin Shan /* 15932a485ad7SGavin Shan * EEH core will try recover from fenced PHB or 15942a485ad7SGavin Shan * frozen PE. In the time for frozen PE, EEH core 15952a485ad7SGavin Shan * enable IO path for that before collecting logs, 15962a485ad7SGavin Shan * but it ruins the site. So we have to dump the 15972a485ad7SGavin Shan * log in advance here. 15982a485ad7SGavin Shan */ 15992a485ad7SGavin Shan if ((ret == EEH_NEXT_ERR_FROZEN_PE || 16002a485ad7SGavin Shan ret == EEH_NEXT_ERR_FENCED_PHB) && 16012a485ad7SGavin Shan !((*pe)->state & EEH_PE_ISOLATED)) { 1602e762bb89SSam Bobroff eeh_pe_mark_isolated(*pe); 16032a485ad7SGavin Shan pnv_eeh_get_phb_diag(*pe); 16042a485ad7SGavin Shan 16052a485ad7SGavin Shan if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) 16062a485ad7SGavin Shan pnv_pci_dump_phb_diag_data((*pe)->phb, 16072a485ad7SGavin Shan (*pe)->data); 16082a485ad7SGavin Shan } 16092a485ad7SGavin Shan 16102a485ad7SGavin Shan /* 16112a485ad7SGavin Shan * We probably have the frozen parent PE out there and 16122a485ad7SGavin Shan * we need have to handle frozen parent PE firstly. 16132a485ad7SGavin Shan */ 16142a485ad7SGavin Shan if (ret == EEH_NEXT_ERR_FROZEN_PE) { 16152a485ad7SGavin Shan parent_pe = (*pe)->parent; 16162a485ad7SGavin Shan while (parent_pe) { 16172a485ad7SGavin Shan /* Hit the ceiling ? */ 16182a485ad7SGavin Shan if (parent_pe->type & EEH_PE_PHB) 16192a485ad7SGavin Shan break; 16202a485ad7SGavin Shan 16212a485ad7SGavin Shan /* Frozen parent PE ? */ 16222a485ad7SGavin Shan state = eeh_ops->get_state(parent_pe, NULL); 162334a286a4SSam Bobroff if (state > 0 && !eeh_state_active(state)) 16242a485ad7SGavin Shan *pe = parent_pe; 16252a485ad7SGavin Shan 16262a485ad7SGavin Shan /* Next parent level */ 16272a485ad7SGavin Shan parent_pe = parent_pe->parent; 16282a485ad7SGavin Shan } 16292a485ad7SGavin Shan 16302a485ad7SGavin Shan /* We possibly migrate to another PE */ 1631e762bb89SSam Bobroff eeh_pe_mark_isolated(*pe); 16322a485ad7SGavin Shan } 16332a485ad7SGavin Shan 16342a485ad7SGavin Shan /* 16352a485ad7SGavin Shan * If we have no errors on the specific PHB or only 16362a485ad7SGavin Shan * informative error there, we continue poking it. 16372a485ad7SGavin Shan * Otherwise, we need actions to be taken by upper 16382a485ad7SGavin Shan * layer. 16392a485ad7SGavin Shan */ 16402a485ad7SGavin Shan if (ret > EEH_NEXT_ERR_INF) 164129310e5eSGavin Shan break; 164229310e5eSGavin Shan } 164329310e5eSGavin Shan 164479231448SAlistair Popple /* Unmask the event */ 1645b8d65e96SAlistair Popple if (ret == EEH_NEXT_ERR_NONE && eeh_enabled()) 164679231448SAlistair Popple enable_irq(eeh_event_irq); 164779231448SAlistair Popple 16482a485ad7SGavin Shan return ret; 164929310e5eSGavin Shan } 165029310e5eSGavin Shan 16510c2c7652SOliver O'Halloran static int pnv_eeh_restore_config(struct eeh_dev *edev) 16529be3beccSGavin Shan { 16539be3beccSGavin Shan struct pnv_phb *phb; 165464ba3dc7SBryant G. Ly s64 ret = 0; 16559be3beccSGavin Shan 16569be3beccSGavin Shan if (!edev) 16579be3beccSGavin Shan return -EEXIST; 16589be3beccSGavin Shan 165921b43bd5SOliver O'Halloran if (edev->physfn) 166021b43bd5SOliver O'Halloran return 0; 166121b43bd5SOliver O'Halloran 166221b43bd5SOliver O'Halloran phb = edev->controller->private_data; 16639be3beccSGavin Shan ret = opal_pci_reinit(phb->opal_id, 166421b43bd5SOliver O'Halloran OPAL_REINIT_PCI_DEV, edev->bdfn); 16650dc2830eSWei Yang 16669be3beccSGavin Shan if (ret) { 16679be3beccSGavin Shan pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n", 16680c2c7652SOliver O'Halloran __func__, edev->bdfn, ret); 16699be3beccSGavin Shan return -EIO; 16709be3beccSGavin Shan } 16719be3beccSGavin Shan 167264ba3dc7SBryant G. Ly return ret; 16739be3beccSGavin Shan } 16749be3beccSGavin Shan 167501f3bfb7SGavin Shan static struct eeh_ops pnv_eeh_ops = { 167629310e5eSGavin Shan .name = "powernv", 167701f3bfb7SGavin Shan .init = pnv_eeh_init, 1678ff57b454SGavin Shan .probe = pnv_eeh_probe, 167901f3bfb7SGavin Shan .set_option = pnv_eeh_set_option, 168001f3bfb7SGavin Shan .get_state = pnv_eeh_get_state, 168101f3bfb7SGavin Shan .reset = pnv_eeh_reset, 168201f3bfb7SGavin Shan .get_log = pnv_eeh_get_log, 168301f3bfb7SGavin Shan .configure_bridge = pnv_eeh_configure_bridge, 168401f3bfb7SGavin Shan .err_inject = pnv_eeh_err_inject, 168501f3bfb7SGavin Shan .read_config = pnv_eeh_read_config, 168601f3bfb7SGavin Shan .write_config = pnv_eeh_write_config, 168701f3bfb7SGavin Shan .next_error = pnv_eeh_next_error, 168867923cfcSBryant G. Ly .restore_config = pnv_eeh_restore_config, 168967923cfcSBryant G. Ly .notify_resume = NULL 169029310e5eSGavin Shan }; 169129310e5eSGavin Shan 16920dc2830eSWei Yang #ifdef CONFIG_PCI_IOV 16930dc2830eSWei Yang static void pnv_pci_fixup_vf_mps(struct pci_dev *pdev) 16940dc2830eSWei Yang { 16950dc2830eSWei Yang struct pci_dn *pdn = pci_get_pdn(pdev); 16960dc2830eSWei Yang int parent_mps; 16970dc2830eSWei Yang 16980dc2830eSWei Yang if (!pdev->is_virtfn) 16990dc2830eSWei Yang return; 17000dc2830eSWei Yang 17010dc2830eSWei Yang /* Synchronize MPS for VF and PF */ 17020dc2830eSWei Yang parent_mps = pcie_get_mps(pdev->physfn); 17030dc2830eSWei Yang if ((128 << pdev->pcie_mpss) >= parent_mps) 17040dc2830eSWei Yang pcie_set_mps(pdev, parent_mps); 17050dc2830eSWei Yang pdn->mps = pcie_get_mps(pdev); 17060dc2830eSWei Yang } 17070dc2830eSWei Yang DECLARE_PCI_FIXUP_HEADER(PCI_ANY_ID, PCI_ANY_ID, pnv_pci_fixup_vf_mps); 17080dc2830eSWei Yang #endif /* CONFIG_PCI_IOV */ 17090dc2830eSWei Yang 171029310e5eSGavin Shan /** 171129310e5eSGavin Shan * eeh_powernv_init - Register platform dependent EEH operations 171229310e5eSGavin Shan * 171329310e5eSGavin Shan * EEH initialization on powernv platform. This function should be 171429310e5eSGavin Shan * called before any EEH related functions. 171529310e5eSGavin Shan */ 171629310e5eSGavin Shan static int __init eeh_powernv_init(void) 171729310e5eSGavin Shan { 171829310e5eSGavin Shan int ret = -EINVAL; 171929310e5eSGavin Shan 172001f3bfb7SGavin Shan ret = eeh_ops_register(&pnv_eeh_ops); 172129310e5eSGavin Shan if (!ret) 172229310e5eSGavin Shan pr_info("EEH: PowerNV platform initialized\n"); 172329310e5eSGavin Shan else 172429310e5eSGavin Shan pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret); 172529310e5eSGavin Shan 172629310e5eSGavin Shan return ret; 172729310e5eSGavin Shan } 1728b14726c5SMichael Ellerman machine_early_initcall(powernv, eeh_powernv_init); 1729