diff options
Diffstat (limited to 'drivers/infiniband/hw/qib/qib_pcie.c')
| -rw-r--r-- | drivers/infiniband/hw/qib/qib_pcie.c | 613 |
1 files changed, 613 insertions, 0 deletions
diff --git a/drivers/infiniband/hw/qib/qib_pcie.c b/drivers/infiniband/hw/qib/qib_pcie.c new file mode 100644 index 000000000..3dc6ce033 --- /dev/null +++ b/drivers/infiniband/hw/qib/qib_pcie.c @@ -0,0 +1,613 @@ +/* + * Copyright (c) 2010 - 2017 Intel Corporation. All rights reserved. + * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved. + * + * This software is available to you under a choice of one of two + * licenses. You may choose to be licensed under the terms of the GNU + * General Public License (GPL) Version 2, available from the file + * COPYING in the main directory of this source tree, or the + * OpenIB.org BSD license below: + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS + * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN + * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN + * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ + +#include <linux/pci.h> +#include <linux/io.h> +#include <linux/delay.h> +#include <linux/vmalloc.h> +#include <linux/aer.h> +#include <linux/module.h> + +#include "qib.h" + +/* + * This file contains PCIe utility routines that are common to the + * various QLogic InfiniPath adapters + */ + +/* + * Code to adjust PCIe capabilities. + * To minimize the change footprint, we call it + * from qib_pcie_params, which every chip-specific + * file calls, even though this violates some + * expectations of harmlessness. + */ +static void qib_tune_pcie_caps(struct qib_devdata *); +static void qib_tune_pcie_coalesce(struct qib_devdata *); + +/* + * Do all the common PCIe setup and initialization. + * devdata is not yet allocated, and is not allocated until after this + * routine returns success. Therefore qib_dev_err() can't be used for error + * printing. + */ +int qib_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent) +{ + int ret; + + ret = pci_enable_device(pdev); + if (ret) { + /* + * This can happen (in theory) iff: + * We did a chip reset, and then failed to reprogram the + * BAR, or the chip reset due to an internal error. We then + * unloaded the driver and reloaded it. + * + * Both reset cases set the BAR back to initial state. For + * the latter case, the AER sticky error bit at offset 0x718 + * should be set, but the Linux kernel doesn't yet know + * about that, it appears. If the original BAR was retained + * in the kernel data structures, this may be OK. + */ + qib_early_err(&pdev->dev, "pci enable failed: error %d\n", + -ret); + goto done; + } + + ret = pci_request_regions(pdev, QIB_DRV_NAME); + if (ret) { + qib_devinfo(pdev, "pci_request_regions fails: err %d\n", -ret); + goto bail; + } + + ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); + if (ret) { + /* + * If the 64 bit setup fails, try 32 bit. Some systems + * do not setup 64 bit maps on systems with 2GB or less + * memory installed. + */ + ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); + if (ret) { + qib_devinfo(pdev, "Unable to set DMA mask: %d\n", ret); + goto bail; + } + ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); + } else + ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); + if (ret) { + qib_early_err(&pdev->dev, + "Unable to set DMA consistent mask: %d\n", ret); + goto bail; + } + + pci_set_master(pdev); + ret = pci_enable_pcie_error_reporting(pdev); + if (ret) { + qib_early_err(&pdev->dev, + "Unable to enable pcie error reporting: %d\n", + ret); + ret = 0; + } + goto done; + +bail: + pci_disable_device(pdev); + pci_release_regions(pdev); +done: + return ret; +} + +/* + * Do remaining PCIe setup, once dd is allocated, and save away + * fields required to re-initialize after a chip reset, or for + * various other purposes + */ +int qib_pcie_ddinit(struct qib_devdata *dd, struct pci_dev *pdev, + const struct pci_device_id *ent) +{ + unsigned long len; + resource_size_t addr; + + dd->pcidev = pdev; + pci_set_drvdata(pdev, dd); + + addr = pci_resource_start(pdev, 0); + len = pci_resource_len(pdev, 0); + + dd->kregbase = ioremap(addr, len); + if (!dd->kregbase) + return -ENOMEM; + + dd->kregend = (u64 __iomem *)((void __iomem *) dd->kregbase + len); + dd->physaddr = addr; /* used for io_remap, etc. */ + + /* + * Save BARs to rewrite after device reset. Save all 64 bits of + * BAR, just in case. + */ + dd->pcibar0 = addr; + dd->pcibar1 = addr >> 32; + dd->deviceid = ent->device; /* save for later use */ + dd->vendorid = ent->vendor; + + return 0; +} + +/* + * Do PCIe cleanup, after chip-specific cleanup, etc. Just prior + * to releasing the dd memory. + * void because none of the core pcie cleanup returns are void + */ +void qib_pcie_ddcleanup(struct qib_devdata *dd) +{ + u64 __iomem *base = (void __iomem *) dd->kregbase; + + dd->kregbase = NULL; + iounmap(base); + if (dd->piobase) + iounmap(dd->piobase); + if (dd->userbase) + iounmap(dd->userbase); + if (dd->piovl15base) + iounmap(dd->piovl15base); + + pci_disable_device(dd->pcidev); + pci_release_regions(dd->pcidev); + + pci_set_drvdata(dd->pcidev, NULL); +} + +/** + * We save the msi lo and hi values, so we can restore them after + * chip reset (the kernel PCI infrastructure doesn't yet handle that + * correctly. + */ +static void qib_cache_msi_info(struct qib_devdata *dd, int pos) +{ + struct pci_dev *pdev = dd->pcidev; + u16 control; + + pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO, &dd->msi_lo); + pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI, &dd->msi_hi); + pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &control); + + /* now save the data (vector) info */ + pci_read_config_word(pdev, + pos + ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8), + &dd->msi_data); +} + +int qib_pcie_params(struct qib_devdata *dd, u32 minw, u32 *nent) +{ + u16 linkstat, speed; + int nvec; + int maxvec; + unsigned int flags = PCI_IRQ_MSIX | PCI_IRQ_MSI; + + if (!pci_is_pcie(dd->pcidev)) { + qib_dev_err(dd, "Can't find PCI Express capability!\n"); + /* set up something... */ + dd->lbus_width = 1; + dd->lbus_speed = 2500; /* Gen1, 2.5GHz */ + nvec = -1; + goto bail; + } + + if (dd->flags & QIB_HAS_INTX) + flags |= PCI_IRQ_LEGACY; + maxvec = (nent && *nent) ? *nent : 1; + nvec = pci_alloc_irq_vectors(dd->pcidev, 1, maxvec, flags); + if (nvec < 0) + goto bail; + + /* + * If nent exists, make sure to record how many vectors were allocated. + * If msix_enabled is false, return 0 so the fallback code works + * correctly. + */ + if (nent) + *nent = !dd->pcidev->msix_enabled ? 0 : nvec; + + if (dd->pcidev->msi_enabled) + qib_cache_msi_info(dd, dd->pcidev->msi_cap); + + pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKSTA, &linkstat); + /* + * speed is bits 0-3, linkwidth is bits 4-8 + * no defines for them in headers + */ + speed = linkstat & 0xf; + linkstat >>= 4; + linkstat &= 0x1f; + dd->lbus_width = linkstat; + + switch (speed) { + case 1: + dd->lbus_speed = 2500; /* Gen1, 2.5GHz */ + break; + case 2: + dd->lbus_speed = 5000; /* Gen1, 5GHz */ + break; + default: /* not defined, assume gen1 */ + dd->lbus_speed = 2500; + break; + } + + /* + * Check against expected pcie width and complain if "wrong" + * on first initialization, not afterwards (i.e., reset). + */ + if (minw && linkstat < minw) + qib_dev_err(dd, + "PCIe width %u (x%u HCA), performance reduced\n", + linkstat, minw); + + qib_tune_pcie_caps(dd); + + qib_tune_pcie_coalesce(dd); + +bail: + /* fill in string, even on errors */ + snprintf(dd->lbus_info, sizeof(dd->lbus_info), + "PCIe,%uMHz,x%u\n", dd->lbus_speed, dd->lbus_width); + return nvec < 0 ? nvec : 0; +} + +/** + * qib_free_irq - Cleanup INTx and MSI interrupts + * @dd: valid pointer to qib dev data + * + * Since cleanup for INTx and MSI interrupts is trivial, have a common + * routine. + * + */ +void qib_free_irq(struct qib_devdata *dd) +{ + pci_free_irq(dd->pcidev, 0, dd); + pci_free_irq_vectors(dd->pcidev); +} + +/* + * Setup pcie interrupt stuff again after a reset. I'd like to just call + * pci_enable_msi() again for msi, but when I do that, + * the MSI enable bit doesn't get set in the command word, and + * we switch to to a different interrupt vector, which is confusing, + * so I instead just do it all inline. Perhaps somehow can tie this + * into the PCIe hotplug support at some point + */ +int qib_reinit_intr(struct qib_devdata *dd) +{ + int pos; + u16 control; + int ret = 0; + + /* If we aren't using MSI, don't restore it */ + if (!dd->msi_lo) + goto bail; + + pos = dd->pcidev->msi_cap; + if (!pos) { + qib_dev_err(dd, + "Can't find MSI capability, can't restore MSI settings\n"); + ret = 0; + /* nothing special for MSIx, just MSI */ + goto bail; + } + pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO, + dd->msi_lo); + pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI, + dd->msi_hi); + pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control); + if (!(control & PCI_MSI_FLAGS_ENABLE)) { + control |= PCI_MSI_FLAGS_ENABLE; + pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, + control); + } + /* now rewrite the data (vector) info */ + pci_write_config_word(dd->pcidev, pos + + ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8), + dd->msi_data); + ret = 1; +bail: + qib_free_irq(dd); + + if (!ret && (dd->flags & QIB_HAS_INTX)) + ret = 1; + + /* and now set the pci master bit again */ + pci_set_master(dd->pcidev); + + return ret; +} + +/* + * These two routines are helper routines for the device reset code + * to move all the pcie code out of the chip-specific driver code. + */ +void qib_pcie_getcmd(struct qib_devdata *dd, u16 *cmd, u8 *iline, u8 *cline) +{ + pci_read_config_word(dd->pcidev, PCI_COMMAND, cmd); + pci_read_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline); + pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline); +} + +void qib_pcie_reenable(struct qib_devdata *dd, u16 cmd, u8 iline, u8 cline) +{ + int r; + + r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0, + dd->pcibar0); + if (r) + qib_dev_err(dd, "rewrite of BAR0 failed: %d\n", r); + r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1, + dd->pcibar1); + if (r) + qib_dev_err(dd, "rewrite of BAR1 failed: %d\n", r); + /* now re-enable memory access, and restore cosmetic settings */ + pci_write_config_word(dd->pcidev, PCI_COMMAND, cmd); + pci_write_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline); + pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline); + r = pci_enable_device(dd->pcidev); + if (r) + qib_dev_err(dd, + "pci_enable_device failed after reset: %d\n", r); +} + + +static int qib_pcie_coalesce; +module_param_named(pcie_coalesce, qib_pcie_coalesce, int, S_IRUGO); +MODULE_PARM_DESC(pcie_coalesce, "tune PCIe coalescing on some Intel chipsets"); + +/* + * Enable PCIe completion and data coalescing, on Intel 5x00 and 7300 + * chipsets. This is known to be unsafe for some revisions of some + * of these chipsets, with some BIOS settings, and enabling it on those + * systems may result in the system crashing, and/or data corruption. + */ +static void qib_tune_pcie_coalesce(struct qib_devdata *dd) +{ + struct pci_dev *parent; + u16 devid; + u32 mask, bits, val; + + if (!qib_pcie_coalesce) + return; + + /* Find out supported and configured values for parent (root) */ + parent = dd->pcidev->bus->self; + if (parent->bus->parent) { + qib_devinfo(dd->pcidev, "Parent not root\n"); + return; + } + if (!pci_is_pcie(parent)) + return; + if (parent->vendor != 0x8086) + return; + + /* + * - bit 12: Max_rdcmp_Imt_EN: need to set to 1 + * - bit 11: COALESCE_FORCE: need to set to 0 + * - bit 10: COALESCE_EN: need to set to 1 + * (but limitations on some on some chipsets) + * + * On the Intel 5000, 5100, and 7300 chipsets, there is + * also: - bit 25:24: COALESCE_MODE, need to set to 0 + */ + devid = parent->device; + if (devid >= 0x25e2 && devid <= 0x25fa) { + /* 5000 P/V/X/Z */ + if (parent->revision <= 0xb2) + bits = 1U << 10; + else + bits = 7U << 10; + mask = (3U << 24) | (7U << 10); + } else if (devid >= 0x65e2 && devid <= 0x65fa) { + /* 5100 */ + bits = 1U << 10; + mask = (3U << 24) | (7U << 10); + } else if (devid >= 0x4021 && devid <= 0x402e) { + /* 5400 */ + bits = 7U << 10; + mask = 7U << 10; + } else if (devid >= 0x3604 && devid <= 0x360a) { + /* 7300 */ + bits = 7U << 10; + mask = (3U << 24) | (7U << 10); + } else { + /* not one of the chipsets that we know about */ + return; + } + pci_read_config_dword(parent, 0x48, &val); + val &= ~mask; + val |= bits; + pci_write_config_dword(parent, 0x48, val); +} + +/* + * BIOS may not set PCIe bus-utilization parameters for best performance. + * Check and optionally adjust them to maximize our throughput. + */ +static int qib_pcie_caps; +module_param_named(pcie_caps, qib_pcie_caps, int, S_IRUGO); +MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)"); + +static void qib_tune_pcie_caps(struct qib_devdata *dd) +{ + struct pci_dev *parent; + u16 rc_mpss, rc_mps, ep_mpss, ep_mps; + u16 rc_mrrs, ep_mrrs, max_mrrs; + + /* Find out supported and configured values for parent (root) */ + parent = dd->pcidev->bus->self; + if (!pci_is_root_bus(parent->bus)) { + qib_devinfo(dd->pcidev, "Parent not root\n"); + return; + } + + if (!pci_is_pcie(parent) || !pci_is_pcie(dd->pcidev)) + return; + + rc_mpss = parent->pcie_mpss; + rc_mps = ffs(pcie_get_mps(parent)) - 8; + /* Find out supported and configured values for endpoint (us) */ + ep_mpss = dd->pcidev->pcie_mpss; + ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8; + + /* Find max payload supported by root, endpoint */ + if (rc_mpss > ep_mpss) + rc_mpss = ep_mpss; + + /* If Supported greater than limit in module param, limit it */ + if (rc_mpss > (qib_pcie_caps & 7)) + rc_mpss = qib_pcie_caps & 7; + /* If less than (allowed, supported), bump root payload */ + if (rc_mpss > rc_mps) { + rc_mps = rc_mpss; + pcie_set_mps(parent, 128 << rc_mps); + } + /* If less than (allowed, supported), bump endpoint payload */ + if (rc_mpss > ep_mps) { + ep_mps = rc_mpss; + pcie_set_mps(dd->pcidev, 128 << ep_mps); + } + + /* + * Now the Read Request size. + * No field for max supported, but PCIe spec limits it to 4096, + * which is code '5' (log2(4096) - 7) + */ + max_mrrs = 5; + if (max_mrrs > ((qib_pcie_caps >> 4) & 7)) + max_mrrs = (qib_pcie_caps >> 4) & 7; + + max_mrrs = 128 << max_mrrs; + rc_mrrs = pcie_get_readrq(parent); + ep_mrrs = pcie_get_readrq(dd->pcidev); + + if (max_mrrs > rc_mrrs) { + rc_mrrs = max_mrrs; + pcie_set_readrq(parent, rc_mrrs); + } + if (max_mrrs > ep_mrrs) { + ep_mrrs = max_mrrs; + pcie_set_readrq(dd->pcidev, ep_mrrs); + } +} +/* End of PCIe capability tuning */ + +/* + * From here through qib_pci_err_handler definition is invoked via + * PCI error infrastructure, registered via pci + */ +static pci_ers_result_t +qib_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state) +{ + struct qib_devdata *dd = pci_get_drvdata(pdev); + pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED; + + switch (state) { + case pci_channel_io_normal: + qib_devinfo(pdev, "State Normal, ignoring\n"); + break; + + case pci_channel_io_frozen: + qib_devinfo(pdev, "State Frozen, requesting reset\n"); + pci_disable_device(pdev); + ret = PCI_ERS_RESULT_NEED_RESET; + break; + + case pci_channel_io_perm_failure: + qib_devinfo(pdev, "State Permanent Failure, disabling\n"); + if (dd) { + /* no more register accesses! */ + dd->flags &= ~QIB_PRESENT; + qib_disable_after_error(dd); + } + /* else early, or other problem */ + ret = PCI_ERS_RESULT_DISCONNECT; + break; + + default: /* shouldn't happen */ + qib_devinfo(pdev, "QIB PCI errors detected (state %d)\n", + state); + break; + } + return ret; +} + +static pci_ers_result_t +qib_pci_mmio_enabled(struct pci_dev *pdev) +{ + u64 words = 0U; + struct qib_devdata *dd = pci_get_drvdata(pdev); + pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED; + + if (dd && dd->pport) { + words = dd->f_portcntr(dd->pport, QIBPORTCNTR_WORDRCV); + if (words == ~0ULL) + ret = PCI_ERS_RESULT_NEED_RESET; + } + qib_devinfo(pdev, + "QIB mmio_enabled function called, read wordscntr %Lx, returning %d\n", + words, ret); + return ret; +} + +static pci_ers_result_t +qib_pci_slot_reset(struct pci_dev *pdev) +{ + qib_devinfo(pdev, "QIB slot_reset function called, ignored\n"); + return PCI_ERS_RESULT_CAN_RECOVER; +} + +static void +qib_pci_resume(struct pci_dev *pdev) +{ + struct qib_devdata *dd = pci_get_drvdata(pdev); + + qib_devinfo(pdev, "QIB resume function called\n"); + /* + * Running jobs will fail, since it's asynchronous + * unlike sysfs-requested reset. Better than + * doing nothing. + */ + qib_init(dd, 1); /* same as re-init after reset */ +} + +const struct pci_error_handlers qib_pci_err_handler = { + .error_detected = qib_pci_error_detected, + .mmio_enabled = qib_pci_mmio_enabled, + .slot_reset = qib_pci_slot_reset, + .resume = qib_pci_resume, +}; |