/*- * SPDX-License-Identifier: BSD-2-Clause * * Authors: Joe Kloss; Ravi Pokala (rpokala@freebsd.org) * * Copyright (c) 2017-2018 Panasas * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. 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. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "imcsmb_reg.h" #include "imcsmb_var.h" /* (Sandy,Ivy)bridge-Xeon and (Has,Broad)well-Xeon CPUs contain one or two * "Integrated Memory Controllers" (iMCs), and each iMC contains two separate * SMBus controllers. These are used for reading SPD data from the DIMMs, and * for reading the "Thermal Sensor on DIMM" (TSODs). The iMC SMBus controllers * are very simple devices, and have limited functionality compared to * full-fledged SMBus controllers, like the one in Intel ICHs and PCHs. * * The publicly available documentation for the iMC SMBus controllers can be * found in the CPU datasheets for (Sandy,Ivy)bridge-Xeon and * (Has,broad)well-Xeon, respectively: * * https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/ * Sandybridge xeon-e5-1600-2600-vol-2-datasheet.pdf * Ivybridge xeon-e5-v2-datasheet-vol-2.pdf * Haswell xeon-e5-v3-datasheet-vol-2.pdf * Broadwell xeon-e5-v4-datasheet-vol-2.pdf * * Another useful resource is the Linux driver. It is not in the main tree. * * https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg840043.html * * The iMC SMBus controllers do not support interrupts (thus, they must be * polled for IO completion). All of the iMC registers are in PCI configuration * space; there is no support for PIO or MMIO. As a result, this driver does * not need to perform and newbus resource manipulation. * * Because there are multiple SMBus controllers sharing the same PCI device, * this driver is actually *two* drivers: * * - "imcsmb" is an smbus(4)-compliant SMBus controller driver * * - "imcsmb_pci" recognizes the PCI device and assigns the appropriate set of * PCI config registers to a specific "imcsmb" instance. */ /* Depending on the motherboard and firmware, the TSODs might be polled by * firmware. Therefore, when this driver accesses these SMBus controllers, the * firmware polling must be disabled as part of requesting the bus, and * re-enabled when releasing the bus. Unfortunately, the details of how to do * this are vendor-specific. Contact your motherboard vendor to get the * information you need to do proper implementations. * * For NVDIMMs which conform to the ACPI "NFIT" standard, the ACPI firmware * manages the NVDIMM; for those which pre-date the standard, the operating * system interacts with the NVDIMM controller using a vendor-proprietary API * over the SMBus. In that case, the NVDIMM driver would be an SMBus slave * device driver, and would interface with the hardware via an SMBus controller * driver such as this one. */ /* PCIe device IDs for (Sandy,Ivy)bridge)-Xeon and (Has,Broad)well-Xeon */ #define PCI_VENDOR_INTEL 0x8086 #define IMCSMB_PCI_DEV_ID_IMC0_SBX 0x3ca8 #define IMCSMB_PCI_DEV_ID_IMC0_IBX 0x0ea8 #define IMCSMB_PCI_DEV_ID_IMC0_HSX 0x2fa8 #define IMCSMB_PCI_DEV_ID_IMC0_BDX 0x6fa8 /* (Sandy,Ivy)bridge-Xeon only have a single memory controller per socket */ #define IMCSMB_PCI_DEV_ID_IMC1_HSX 0x2f68 #define IMCSMB_PCI_DEV_ID_IMC1_BDX 0x6f68 /* There are two SMBus controllers in each device. These define the registers * for each of these devices. */ static struct imcsmb_reg_set imcsmb_regs[] = { { .smb_stat = IMCSMB_REG_STATUS0, .smb_cmd = IMCSMB_REG_COMMAND0, .smb_cntl = IMCSMB_REG_CONTROL0 }, { .smb_stat = IMCSMB_REG_STATUS1, .smb_cmd = IMCSMB_REG_COMMAND1, .smb_cntl = IMCSMB_REG_CONTROL1 }, }; static struct imcsmb_pci_device { uint16_t id; char *name; } imcsmb_pci_devices[] = { {IMCSMB_PCI_DEV_ID_IMC0_SBX, "Intel Sandybridge Xeon iMC 0 SMBus controllers" }, {IMCSMB_PCI_DEV_ID_IMC0_IBX, "Intel Ivybridge Xeon iMC 0 SMBus controllers" }, {IMCSMB_PCI_DEV_ID_IMC0_HSX, "Intel Haswell Xeon iMC 0 SMBus controllers" }, {IMCSMB_PCI_DEV_ID_IMC1_HSX, "Intel Haswell Xeon iMC 1 SMBus controllers" }, {IMCSMB_PCI_DEV_ID_IMC0_BDX, "Intel Broadwell Xeon iMC 0 SMBus controllers" }, {IMCSMB_PCI_DEV_ID_IMC1_BDX, "Intel Broadwell Xeon iMC 1 SMBus controllers" }, {0, NULL}, }; /* Device methods. */ static int imcsmb_pci_attach(device_t dev); static int imcsmb_pci_detach(device_t dev); static int imcsmb_pci_probe(device_t dev); /** * device_attach() method. Set up the PCI device's softc, then explicitly create * children for the actual imcsmbX controllers. Set up the child's ivars to * point to the proper set of the PCI device's config registers. * * @author Joe Kloss, rpokala * * @param[in,out] dev * Device being attached. */ static int imcsmb_pci_attach(device_t dev) { struct imcsmb_pci_softc *sc; device_t child; int rc; int unit; /* Initialize private state */ sc = device_get_softc(dev); sc->dev = dev; sc->semaphore = 0; /* Create the imcsmbX children */ for (unit = 0; unit < 2; unit++) { child = device_add_child(dev, "imcsmb", -1); if (child == NULL) { /* Nothing has been allocated, so there's no cleanup. */ device_printf(dev, "Child imcsmb not added\n"); rc = ENXIO; goto out; } /* Set the child's ivars to point to the appropriate set of * the PCI device's registers. */ device_set_ivars(child, &imcsmb_regs[unit]); } /* Attach the imcsmbX children. */ if ((rc = bus_generic_attach(dev)) != 0) { device_printf(dev, "failed to attach children: %d\n", rc); goto out; } out: return (rc); } /** * device_detach() method. attach() didn't do any allocations, so all that's * needed here is to free up any downstream drivers and children. * * @author Joe Kloss * * @param[in] dev * Device being detached. */ static int imcsmb_pci_detach(device_t dev) { int rc; /* Detach any attached drivers */ rc = bus_generic_detach(dev); if (rc == 0) { /* Remove all children */ rc = device_delete_children(dev); } return (rc); } /** * device_probe() method. Look for the right PCI vendor/device IDs. * * @author Joe Kloss, rpokala * * @param[in,out] dev * Device being probed. */ static int imcsmb_pci_probe(device_t dev) { struct imcsmb_pci_device *pci_device; int rc; uint16_t pci_dev_id; rc = ENXIO; if (pci_get_vendor(dev) != PCI_VENDOR_INTEL) { goto out; } pci_dev_id = pci_get_device(dev); for (pci_device = imcsmb_pci_devices; pci_device->name != NULL; pci_device++) { if (pci_dev_id == pci_device->id) { device_set_desc(dev, pci_device->name); rc = BUS_PROBE_DEFAULT; goto out; } } out: return (rc); } /** * Invoked via smbus_callback() -> imcsmb_callback(); clear the semaphore, and * re-enable motherboard-specific DIMM temperature monitoring if needed. This * gets called after the transaction completes. * * @author Joe Kloss * * @param[in,out] dev * The device whose busses to release. */ void imcsmb_pci_release_bus(device_t dev) { struct imcsmb_pci_softc *sc; sc = device_get_softc(dev); /* * IF NEEDED, INSERT MOTHERBOARD-SPECIFIC CODE TO RE-ENABLE DIMM * TEMPERATURE MONITORING HERE. */ atomic_store_rel_int(&sc->semaphore, 0); } /** * Invoked via smbus_callback() -> imcsmb_callback(); set the semaphore, and * disable motherboard-specific DIMM temperature monitoring if needed. This gets * called before the transaction starts. * * @author Joe Kloss * * @param[in,out] dev * The device whose busses to request. */ int imcsmb_pci_request_bus(device_t dev) { struct imcsmb_pci_softc *sc; int rc; sc = device_get_softc(dev); rc = 0; /* We don't want to block. Use a simple test-and-set semaphore to * protect the bus. */ if (atomic_cmpset_acq_int(&sc->semaphore, 0, 1) == 0) { rc = EWOULDBLOCK; } /* * IF NEEDED, INSERT MOTHERBOARD-SPECIFIC CODE TO DISABLE DIMM * TEMPERATURE MONITORING HERE. */ return (rc); } /* Our device class */ static devclass_t imcsmb_pci_devclass; /* Device methods */ static device_method_t imcsmb_pci_methods[] = { /* Device interface */ DEVMETHOD(device_attach, imcsmb_pci_attach), DEVMETHOD(device_detach, imcsmb_pci_detach), DEVMETHOD(device_probe, imcsmb_pci_probe), DEVMETHOD_END }; static driver_t imcsmb_pci_driver = { .name = "imcsmb_pci", .methods = imcsmb_pci_methods, .size = sizeof(struct imcsmb_pci_softc), }; DRIVER_MODULE(imcsmb_pci, pci, imcsmb_pci_driver, imcsmb_pci_devclass, 0, 0); MODULE_DEPEND(imcsmb_pci, pci, 1, 1, 1); MODULE_VERSION(imcsmb_pci, 1); /* vi: set ts=8 sw=4 sts=8 noet: */