#include "amd64_edac.h"
+#include <asm/k8.h>
static struct edac_pci_ctl_info *amd64_ctl_pci;
static enum edac_type amd64_determine_edac_cap(struct amd64_pvt *pvt)
{
int bit;
- enum dev_type edac_cap = EDAC_NONE;
+ enum dev_type edac_cap = EDAC_FLAG_NONE;
bit = (boot_cpu_data.x86 > 0xf || pvt->ext_model >= OPTERON_CPU_REV_F)
? 19
: 17;
- if (pvt->dclr0 >> BIT(bit))
+ if (pvt->dclr0 & BIT(bit))
edac_cap = EDAC_FLAG_SECDED;
return edac_cap;
if (channels == 0)
channels = 1;
- debugf0("DIMM count= %d\n", channels);
+ debugf0("MCT channel count: %d\n", channels);
return channels;
}
EXPORT_SYMBOL_GPL(amd64_process_error_info);
+/*
+ * The main polling 'check' function, called FROM the edac core to perform the
+ * error checking and if an error is encountered, error processing.
+ */
+static void amd64_check(struct mem_ctl_info *mci)
+{
+ struct amd64_error_info_regs info;
+
+ if (amd64_get_error_info(mci, &info))
+ amd64_process_error_info(mci, &info, 1);
+}
+
+/*
+ * Input:
+ * 1) struct amd64_pvt which contains pvt->dram_f2_ctl pointer
+ * 2) AMD Family index value
+ *
+ * Ouput:
+ * Upon return of 0, the following filled in:
+ *
+ * struct pvt->addr_f1_ctl
+ * struct pvt->misc_f3_ctl
+ *
+ * Filled in with related device funcitions of 'dram_f2_ctl'
+ * These devices are "reserved" via the pci_get_device()
+ *
+ * Upon return of 1 (error status):
+ *
+ * Nothing reserved
+ */
+static int amd64_reserve_mc_sibling_devices(struct amd64_pvt *pvt, int mc_idx)
+{
+ const struct amd64_family_type *amd64_dev = &amd64_family_types[mc_idx];
+
+ /* Reserve the ADDRESS MAP Device */
+ pvt->addr_f1_ctl = pci_get_related_function(pvt->dram_f2_ctl->vendor,
+ amd64_dev->addr_f1_ctl,
+ pvt->dram_f2_ctl);
+
+ if (!pvt->addr_f1_ctl) {
+ amd64_printk(KERN_ERR, "error address map device not found: "
+ "vendor %x device 0x%x (broken BIOS?)\n",
+ PCI_VENDOR_ID_AMD, amd64_dev->addr_f1_ctl);
+ return 1;
+ }
+
+ /* Reserve the MISC Device */
+ pvt->misc_f3_ctl = pci_get_related_function(pvt->dram_f2_ctl->vendor,
+ amd64_dev->misc_f3_ctl,
+ pvt->dram_f2_ctl);
+
+ if (!pvt->misc_f3_ctl) {
+ pci_dev_put(pvt->addr_f1_ctl);
+ pvt->addr_f1_ctl = NULL;
+
+ amd64_printk(KERN_ERR, "error miscellaneous device not found: "
+ "vendor %x device 0x%x (broken BIOS?)\n",
+ PCI_VENDOR_ID_AMD, amd64_dev->misc_f3_ctl);
+ return 1;
+ }
+
+ debugf1(" Addr Map device PCI Bus ID:\t%s\n",
+ pci_name(pvt->addr_f1_ctl));
+ debugf1(" DRAM MEM-CTL PCI Bus ID:\t%s\n",
+ pci_name(pvt->dram_f2_ctl));
+ debugf1(" Misc device PCI Bus ID:\t%s\n",
+ pci_name(pvt->misc_f3_ctl));
+
+ return 0;
+}
+
+static void amd64_free_mc_sibling_devices(struct amd64_pvt *pvt)
+{
+ pci_dev_put(pvt->addr_f1_ctl);
+ pci_dev_put(pvt->misc_f3_ctl);
+}
+
+/*
+ * Retrieve the hardware registers of the memory controller (this includes the
+ * 'Address Map' and 'Misc' device regs)
+ */
+static void amd64_read_mc_registers(struct amd64_pvt *pvt)
+{
+ u64 msr_val;
+ int dram, err = 0;
+
+ /*
+ * Retrieve TOP_MEM and TOP_MEM2; no masking off of reserved bits since
+ * those are Read-As-Zero
+ */
+ rdmsrl(MSR_K8_TOP_MEM1, msr_val);
+ pvt->top_mem = msr_val >> 23;
+ debugf0(" TOP_MEM=0x%08llx\n", pvt->top_mem);
+
+ /* check first whether TOP_MEM2 is enabled */
+ rdmsrl(MSR_K8_SYSCFG, msr_val);
+ if (msr_val & (1U << 21)) {
+ rdmsrl(MSR_K8_TOP_MEM2, msr_val);
+ pvt->top_mem2 = msr_val >> 23;
+ debugf0(" TOP_MEM2=0x%08llx\n", pvt->top_mem2);
+ } else
+ debugf0(" TOP_MEM2 disabled.\n");
+
+ amd64_cpu_display_info(pvt);
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCAP, &pvt->nbcap);
+ if (err)
+ goto err_reg;
+
+ if (pvt->ops->read_dram_ctl_register)
+ pvt->ops->read_dram_ctl_register(pvt);
+
+ for (dram = 0; dram < DRAM_REG_COUNT; dram++) {
+ /*
+ * Call CPU specific READ function to get the DRAM Base and
+ * Limit values from the DCT.
+ */
+ pvt->ops->read_dram_base_limit(pvt, dram);
+
+ /*
+ * Only print out debug info on rows with both R and W Enabled.
+ * Normal processing, compiler should optimize this whole 'if'
+ * debug output block away.
+ */
+ if (pvt->dram_rw_en[dram] != 0) {
+ debugf1(" DRAM_BASE[%d]: 0x%8.08x-%8.08x "
+ "DRAM_LIMIT: 0x%8.08x-%8.08x\n",
+ dram,
+ (u32)(pvt->dram_base[dram] >> 32),
+ (u32)(pvt->dram_base[dram] & 0xFFFFFFFF),
+ (u32)(pvt->dram_limit[dram] >> 32),
+ (u32)(pvt->dram_limit[dram] & 0xFFFFFFFF));
+ debugf1(" IntlvEn=%s %s %s "
+ "IntlvSel=%d DstNode=%d\n",
+ pvt->dram_IntlvEn[dram] ?
+ "Enabled" : "Disabled",
+ (pvt->dram_rw_en[dram] & 0x2) ? "W" : "!W",
+ (pvt->dram_rw_en[dram] & 0x1) ? "R" : "!R",
+ pvt->dram_IntlvSel[dram],
+ pvt->dram_DstNode[dram]);
+ }
+ }
+
+ amd64_read_dct_base_mask(pvt);
+
+ err = pci_read_config_dword(pvt->addr_f1_ctl, K8_DHAR, &pvt->dhar);
+ if (err)
+ goto err_reg;
+
+ amd64_read_dbam_reg(pvt);
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl,
+ F10_ONLINE_SPARE, &pvt->online_spare);
+ if (err)
+ goto err_reg;
+
+ err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_0, &pvt->dclr0);
+ if (err)
+ goto err_reg;
+
+ err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCHR_0, &pvt->dchr0);
+ if (err)
+ goto err_reg;
+
+ if (!dct_ganging_enabled(pvt)) {
+ err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCLR_1,
+ &pvt->dclr1);
+ if (err)
+ goto err_reg;
+
+ err = pci_read_config_dword(pvt->dram_f2_ctl, F10_DCHR_1,
+ &pvt->dchr1);
+ if (err)
+ goto err_reg;
+ }
+
+ amd64_dump_misc_regs(pvt);
+
+err_reg:
+ debugf0("Reading an MC register failed\n");
+
+}
+
+/*
+ * NOTE: CPU Revision Dependent code
+ *
+ * Input:
+ * @csrow_nr ChipSelect Row Number (0..CHIPSELECT_COUNT-1)
+ * k8 private pointer to -->
+ * DRAM Bank Address mapping register
+ * node_id
+ * DCL register where dual_channel_active is
+ *
+ * The DBAM register consists of 4 sets of 4 bits each definitions:
+ *
+ * Bits: CSROWs
+ * 0-3 CSROWs 0 and 1
+ * 4-7 CSROWs 2 and 3
+ * 8-11 CSROWs 4 and 5
+ * 12-15 CSROWs 6 and 7
+ *
+ * Values range from: 0 to 15
+ * The meaning of the values depends on CPU revision and dual-channel state,
+ * see relevant BKDG more info.
+ *
+ * The memory controller provides for total of only 8 CSROWs in its current
+ * architecture. Each "pair" of CSROWs normally represents just one DIMM in
+ * single channel or two (2) DIMMs in dual channel mode.
+ *
+ * The following code logic collapses the various tables for CSROW based on CPU
+ * revision.
+ *
+ * Returns:
+ * The number of PAGE_SIZE pages on the specified CSROW number it
+ * encompasses
+ *
+ */
+static u32 amd64_csrow_nr_pages(int csrow_nr, struct amd64_pvt *pvt)
+{
+ u32 dram_map, nr_pages;
+
+ /*
+ * The math on this doesn't look right on the surface because x/2*4 can
+ * be simplified to x*2 but this expression makes use of the fact that
+ * it is integral math where 1/2=0. This intermediate value becomes the
+ * number of bits to shift the DBAM register to extract the proper CSROW
+ * field.
+ */
+ dram_map = (pvt->dbam0 >> ((csrow_nr / 2) * 4)) & 0xF;
+
+ nr_pages = pvt->ops->dbam_map_to_pages(pvt, dram_map);
+
+ /*
+ * If dual channel then double the memory size of single channel.
+ * Channel count is 1 or 2
+ */
+ nr_pages <<= (pvt->channel_count - 1);
+
+ debugf0(" (csrow=%d) DBAM map index= %d\n", csrow_nr, dram_map);
+ debugf0(" nr_pages= %u channel-count = %d\n",
+ nr_pages, pvt->channel_count);
+
+ return nr_pages;
+}
+
+/*
+ * Initialize the array of csrow attribute instances, based on the values
+ * from pci config hardware registers.
+ */
+static int amd64_init_csrows(struct mem_ctl_info *mci)
+{
+ struct csrow_info *csrow;
+ struct amd64_pvt *pvt;
+ u64 input_addr_min, input_addr_max, sys_addr;
+ int i, err = 0, empty = 1;
+
+ pvt = mci->pvt_info;
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &pvt->nbcfg);
+ if (err)
+ debugf0("Reading K8_NBCFG failed\n");
+
+ debugf0("NBCFG= 0x%x CHIPKILL= %s DRAM ECC= %s\n", pvt->nbcfg,
+ (pvt->nbcfg & K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled",
+ (pvt->nbcfg & K8_NBCFG_ECC_ENABLE) ? "Enabled" : "Disabled"
+ );
+
+ for (i = 0; i < CHIPSELECT_COUNT; i++) {
+ csrow = &mci->csrows[i];
+
+ if ((pvt->dcsb0[i] & K8_DCSB_CS_ENABLE) == 0) {
+ debugf1("----CSROW %d EMPTY for node %d\n", i,
+ pvt->mc_node_id);
+ continue;
+ }
+
+ debugf1("----CSROW %d VALID for MC node %d\n",
+ i, pvt->mc_node_id);
+
+ empty = 0;
+ csrow->nr_pages = amd64_csrow_nr_pages(i, pvt);
+ find_csrow_limits(mci, i, &input_addr_min, &input_addr_max);
+ sys_addr = input_addr_to_sys_addr(mci, input_addr_min);
+ csrow->first_page = (u32) (sys_addr >> PAGE_SHIFT);
+ sys_addr = input_addr_to_sys_addr(mci, input_addr_max);
+ csrow->last_page = (u32) (sys_addr >> PAGE_SHIFT);
+ csrow->page_mask = ~mask_from_dct_mask(pvt, i);
+ /* 8 bytes of resolution */
+
+ csrow->mtype = amd64_determine_memory_type(pvt);
+
+ debugf1(" for MC node %d csrow %d:\n", pvt->mc_node_id, i);
+ debugf1(" input_addr_min: 0x%lx input_addr_max: 0x%lx\n",
+ (unsigned long)input_addr_min,
+ (unsigned long)input_addr_max);
+ debugf1(" sys_addr: 0x%lx page_mask: 0x%lx\n",
+ (unsigned long)sys_addr, csrow->page_mask);
+ debugf1(" nr_pages: %u first_page: 0x%lx "
+ "last_page: 0x%lx\n",
+ (unsigned)csrow->nr_pages,
+ csrow->first_page, csrow->last_page);
+
+ /*
+ * determine whether CHIPKILL or JUST ECC or NO ECC is operating
+ */
+ if (pvt->nbcfg & K8_NBCFG_ECC_ENABLE)
+ csrow->edac_mode =
+ (pvt->nbcfg & K8_NBCFG_CHIPKILL) ?
+ EDAC_S4ECD4ED : EDAC_SECDED;
+ else
+ csrow->edac_mode = EDAC_NONE;
+ }
+
+ return empty;
+}
+
+/*
+ * Only if 'ecc_enable_override' is set AND BIOS had ECC disabled, do "we"
+ * enable it.
+ */
+static void amd64_enable_ecc_error_reporting(struct mem_ctl_info *mci)
+{
+ struct amd64_pvt *pvt = mci->pvt_info;
+ const cpumask_t *cpumask = cpumask_of_node(pvt->mc_node_id);
+ int cpu, idx = 0, err = 0;
+ struct msr msrs[cpumask_weight(cpumask)];
+ u32 value;
+ u32 mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn;
+
+ if (!ecc_enable_override)
+ return;
+
+ memset(msrs, 0, sizeof(msrs));
+
+ amd64_printk(KERN_WARNING,
+ "'ecc_enable_override' parameter is active, "
+ "Enabling AMD ECC hardware now: CAUTION\n");
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCTL, &value);
+ if (err)
+ debugf0("Reading K8_NBCTL failed\n");
+
+ /* turn on UECCn and CECCEn bits */
+ pvt->old_nbctl = value & mask;
+ pvt->nbctl_mcgctl_saved = 1;
+
+ value |= mask;
+ pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCTL, value);
+
+ rdmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs);
+
+ for_each_cpu(cpu, cpumask) {
+ if (msrs[idx].l & K8_MSR_MCGCTL_NBE)
+ set_bit(idx, &pvt->old_mcgctl);
+
+ msrs[idx].l |= K8_MSR_MCGCTL_NBE;
+ idx++;
+ }
+ wrmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs);
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value);
+ if (err)
+ debugf0("Reading K8_NBCFG failed\n");
+
+ debugf0("NBCFG(1)= 0x%x CHIPKILL= %s ECC_ENABLE= %s\n", value,
+ (value & K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled",
+ (value & K8_NBCFG_ECC_ENABLE) ? "Enabled" : "Disabled");
+
+ if (!(value & K8_NBCFG_ECC_ENABLE)) {
+ amd64_printk(KERN_WARNING,
+ "This node reports that DRAM ECC is "
+ "currently Disabled; ENABLING now\n");
+
+ /* Attempt to turn on DRAM ECC Enable */
+ value |= K8_NBCFG_ECC_ENABLE;
+ pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCFG, value);
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value);
+ if (err)
+ debugf0("Reading K8_NBCFG failed\n");
+
+ if (!(value & K8_NBCFG_ECC_ENABLE)) {
+ amd64_printk(KERN_WARNING,
+ "Hardware rejects Enabling DRAM ECC checking\n"
+ "Check memory DIMM configuration\n");
+ } else {
+ amd64_printk(KERN_DEBUG,
+ "Hardware accepted DRAM ECC Enable\n");
+ }
+ }
+ debugf0("NBCFG(2)= 0x%x CHIPKILL= %s ECC_ENABLE= %s\n", value,
+ (value & K8_NBCFG_CHIPKILL) ? "Enabled" : "Disabled",
+ (value & K8_NBCFG_ECC_ENABLE) ? "Enabled" : "Disabled");
+
+ pvt->ctl_error_info.nbcfg = value;
+}
+
+static void amd64_restore_ecc_error_reporting(struct amd64_pvt *pvt)
+{
+ const cpumask_t *cpumask = cpumask_of_node(pvt->mc_node_id);
+ int cpu, idx = 0, err = 0;
+ struct msr msrs[cpumask_weight(cpumask)];
+ u32 value;
+ u32 mask = K8_NBCTL_CECCEn | K8_NBCTL_UECCEn;
+
+ if (!pvt->nbctl_mcgctl_saved)
+ return;
+
+ memset(msrs, 0, sizeof(msrs));
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCTL, &value);
+ if (err)
+ debugf0("Reading K8_NBCTL failed\n");
+ value &= ~mask;
+ value |= pvt->old_nbctl;
+
+ /* restore the NB Enable MCGCTL bit */
+ pci_write_config_dword(pvt->misc_f3_ctl, K8_NBCTL, value);
+
+ rdmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs);
+
+ for_each_cpu(cpu, cpumask) {
+ msrs[idx].l &= ~K8_MSR_MCGCTL_NBE;
+ msrs[idx].l |=
+ test_bit(idx, &pvt->old_mcgctl) << K8_MSR_MCGCTL_NBE;
+ idx++;
+ }
+
+ wrmsr_on_cpus(cpumask, K8_MSR_MCGCTL, msrs);
+}
+
+static void check_mcg_ctl(void *ret)
+{
+ u64 msr_val = 0;
+ u8 nbe;
+
+ rdmsrl(MSR_IA32_MCG_CTL, msr_val);
+ nbe = msr_val & K8_MSR_MCGCTL_NBE;
+
+ debugf0("core: %u, MCG_CTL: 0x%llx, NB MSR is %s\n",
+ raw_smp_processor_id(), msr_val,
+ (nbe ? "enabled" : "disabled"));
+
+ if (!nbe)
+ *(int *)ret = 0;
+}
+
+/* check MCG_CTL on all the cpus on this node */
+static int amd64_mcg_ctl_enabled_on_cpus(const cpumask_t *mask)
+{
+ int ret = 1;
+ preempt_disable();
+ smp_call_function_many(mask, check_mcg_ctl, &ret, 1);
+ preempt_enable();
+
+ return ret;
+}
+
+/*
+ * EDAC requires that the BIOS have ECC enabled before taking over the
+ * processing of ECC errors. This is because the BIOS can properly initialize
+ * the memory system completely. A command line option allows to force-enable
+ * hardware ECC later in amd64_enable_ecc_error_reporting().
+ */
+static int amd64_check_ecc_enabled(struct amd64_pvt *pvt)
+{
+ u32 value;
+ int err = 0, ret = 0;
+ u8 ecc_enabled = 0;
+
+ err = pci_read_config_dword(pvt->misc_f3_ctl, K8_NBCFG, &value);
+ if (err)
+ debugf0("Reading K8_NBCTL failed\n");
+
+ ecc_enabled = !!(value & K8_NBCFG_ECC_ENABLE);
+
+ ret = amd64_mcg_ctl_enabled_on_cpus(cpumask_of_node(pvt->mc_node_id));
+
+ debugf0("K8_NBCFG=0x%x, DRAM ECC is %s\n", value,
+ (value & K8_NBCFG_ECC_ENABLE ? "enabled" : "disabled"));
+
+ if (!ecc_enabled || !ret) {
+ if (!ecc_enabled) {
+ amd64_printk(KERN_WARNING, "This node reports that "
+ "Memory ECC is currently "
+ "disabled.\n");
+
+ amd64_printk(KERN_WARNING, "bit 0x%lx in register "
+ "F3x%x of the MISC_CONTROL device (%s) "
+ "should be enabled\n", K8_NBCFG_ECC_ENABLE,
+ K8_NBCFG, pci_name(pvt->misc_f3_ctl));
+ }
+ if (!ret) {
+ amd64_printk(KERN_WARNING, "bit 0x%016lx in MSR 0x%08x "
+ "of node %d should be enabled\n",
+ K8_MSR_MCGCTL_NBE, MSR_IA32_MCG_CTL,
+ pvt->mc_node_id);
+ }
+ if (!ecc_enable_override) {
+ amd64_printk(KERN_WARNING, "WARNING: ECC is NOT "
+ "currently enabled by the BIOS. Module "
+ "will NOT be loaded.\n"
+ " Either Enable ECC in the BIOS, "
+ "or use the 'ecc_enable_override' "
+ "parameter.\n"
+ " Might be a BIOS bug, if BIOS says "
+ "ECC is enabled\n"
+ " Use of the override can cause "
+ "unknown side effects.\n");
+ ret = -ENODEV;
+ } else
+ /*
+ * enable further driver loading if ECC enable is
+ * overridden.
+ */
+ ret = 0;
+ } else {
+ amd64_printk(KERN_INFO,
+ "ECC is enabled by BIOS, Proceeding "
+ "with EDAC module initialization\n");
+
+ /* CLEAR the override, since BIOS controlled it */
+ ecc_enable_override = 0;
+ }
+
+ return ret;
+}
+
+struct mcidev_sysfs_attribute sysfs_attrs[ARRAY_SIZE(amd64_dbg_attrs) +
+ ARRAY_SIZE(amd64_inj_attrs) +
+ 1];
+
+struct mcidev_sysfs_attribute terminator = { .attr = { .name = NULL } };
+
+static void amd64_set_mc_sysfs_attributes(struct mem_ctl_info *mci)
+{
+ unsigned int i = 0, j = 0;
+
+ for (; i < ARRAY_SIZE(amd64_dbg_attrs); i++)
+ sysfs_attrs[i] = amd64_dbg_attrs[i];
+
+ for (j = 0; j < ARRAY_SIZE(amd64_inj_attrs); j++, i++)
+ sysfs_attrs[i] = amd64_inj_attrs[j];
+
+ sysfs_attrs[i] = terminator;
+
+ mci->mc_driver_sysfs_attributes = sysfs_attrs;
+}
+
+static void amd64_setup_mci_misc_attributes(struct mem_ctl_info *mci)
+{
+ struct amd64_pvt *pvt = mci->pvt_info;
+
+ mci->mtype_cap = MEM_FLAG_DDR2 | MEM_FLAG_RDDR2;
+ mci->edac_ctl_cap = EDAC_FLAG_NONE;
+
+ if (pvt->nbcap & K8_NBCAP_SECDED)
+ mci->edac_ctl_cap |= EDAC_FLAG_SECDED;
+
+ if (pvt->nbcap & K8_NBCAP_CHIPKILL)
+ mci->edac_ctl_cap |= EDAC_FLAG_S4ECD4ED;
+
+ mci->edac_cap = amd64_determine_edac_cap(pvt);
+ mci->mod_name = EDAC_MOD_STR;
+ mci->mod_ver = EDAC_AMD64_VERSION;
+ mci->ctl_name = get_amd_family_name(pvt->mc_type_index);
+ mci->dev_name = pci_name(pvt->dram_f2_ctl);
+ mci->ctl_page_to_phys = NULL;
+
+ /* IMPORTANT: Set the polling 'check' function in this module */
+ mci->edac_check = amd64_check;
+
+ /* memory scrubber interface */
+ mci->set_sdram_scrub_rate = amd64_set_scrub_rate;
+ mci->get_sdram_scrub_rate = amd64_get_scrub_rate;
+}
+
+/*
+ * Init stuff for this DRAM Controller device.
+ *
+ * Due to a hardware feature on Fam10h CPUs, the Enable Extended Configuration
+ * Space feature MUST be enabled on ALL Processors prior to actually reading
+ * from the ECS registers. Since the loading of the module can occur on any
+ * 'core', and cores don't 'see' all the other processors ECS data when the
+ * others are NOT enabled. Our solution is to first enable ECS access in this
+ * routine on all processors, gather some data in a amd64_pvt structure and
+ * later come back in a finish-setup function to perform that final
+ * initialization. See also amd64_init_2nd_stage() for that.
+ */
+static int amd64_probe_one_instance(struct pci_dev *dram_f2_ctl,
+ int mc_type_index)
+{
+ struct amd64_pvt *pvt = NULL;
+ int err = 0, ret;
+
+ ret = -ENOMEM;
+ pvt = kzalloc(sizeof(struct amd64_pvt), GFP_KERNEL);
+ if (!pvt)
+ goto err_exit;
+
+ pvt->mc_node_id = get_node_id(dram_f2_ctl);
+
+ pvt->dram_f2_ctl = dram_f2_ctl;
+ pvt->ext_model = boot_cpu_data.x86_model >> 4;
+ pvt->mc_type_index = mc_type_index;
+ pvt->ops = family_ops(mc_type_index);
+ pvt->old_mcgctl = 0;
+
+ /*
+ * We have the dram_f2_ctl device as an argument, now go reserve its
+ * sibling devices from the PCI system.
+ */
+ ret = -ENODEV;
+ err = amd64_reserve_mc_sibling_devices(pvt, mc_type_index);
+ if (err)
+ goto err_free;
+
+ ret = -EINVAL;
+ err = amd64_check_ecc_enabled(pvt);
+ if (err)
+ goto err_put;
+
+ /*
+ * Key operation here: setup of HW prior to performing ops on it. Some
+ * setup is required to access ECS data. After this is performed, the
+ * 'teardown' function must be called upon error and normal exit paths.
+ */
+ if (boot_cpu_data.x86 >= 0x10)
+ amd64_setup(pvt);
+
+ /*
+ * Save the pointer to the private data for use in 2nd initialization
+ * stage
+ */
+ pvt_lookup[pvt->mc_node_id] = pvt;
+
+ return 0;
+
+err_put:
+ amd64_free_mc_sibling_devices(pvt);
+
+err_free:
+ kfree(pvt);
+
+err_exit:
+ return ret;
+}
+
+/*
+ * This is the finishing stage of the init code. Needs to be performed after all
+ * MCs' hardware have been prepped for accessing extended config space.
+ */
+static int amd64_init_2nd_stage(struct amd64_pvt *pvt)
+{
+ int node_id = pvt->mc_node_id;
+ struct mem_ctl_info *mci;
+ int ret, err = 0;
+
+ amd64_read_mc_registers(pvt);
+
+ ret = -ENODEV;
+ if (pvt->ops->probe_valid_hardware) {
+ err = pvt->ops->probe_valid_hardware(pvt);
+ if (err)
+ goto err_exit;
+ }
+
+ /*
+ * We need to determine how many memory channels there are. Then use
+ * that information for calculating the size of the dynamic instance
+ * tables in the 'mci' structure
+ */
+ pvt->channel_count = pvt->ops->early_channel_count(pvt);
+ if (pvt->channel_count < 0)
+ goto err_exit;
+
+ ret = -ENOMEM;
+ mci = edac_mc_alloc(0, CHIPSELECT_COUNT, pvt->channel_count, node_id);
+ if (!mci)
+ goto err_exit;
+
+ mci->pvt_info = pvt;
+
+ mci->dev = &pvt->dram_f2_ctl->dev;
+ amd64_setup_mci_misc_attributes(mci);
+
+ if (amd64_init_csrows(mci))
+ mci->edac_cap = EDAC_FLAG_NONE;
+
+ amd64_enable_ecc_error_reporting(mci);
+ amd64_set_mc_sysfs_attributes(mci);
+
+ ret = -ENODEV;
+ if (edac_mc_add_mc(mci)) {
+ debugf1("failed edac_mc_add_mc()\n");
+ goto err_add_mc;
+ }
+
+ mci_lookup[node_id] = mci;
+ pvt_lookup[node_id] = NULL;
+ return 0;
+
+err_add_mc:
+ edac_mc_free(mci);
+
+err_exit:
+ debugf0("failure to init 2nd stage: ret=%d\n", ret);
+
+ amd64_restore_ecc_error_reporting(pvt);
+
+ if (boot_cpu_data.x86 > 0xf)
+ amd64_teardown(pvt);
+
+ amd64_free_mc_sibling_devices(pvt);
+
+ kfree(pvt_lookup[pvt->mc_node_id]);
+ pvt_lookup[node_id] = NULL;
+
+ return ret;
+}
+
+
+static int __devinit amd64_init_one_instance(struct pci_dev *pdev,
+ const struct pci_device_id *mc_type)
+{
+ int ret = 0;
+
+ debugf0("(MC node=%d,mc_type='%s')\n", get_node_id(pdev),
+ get_amd_family_name(mc_type->driver_data));
+
+ ret = pci_enable_device(pdev);
+ if (ret < 0)
+ ret = -EIO;
+ else
+ ret = amd64_probe_one_instance(pdev, mc_type->driver_data);
+
+ if (ret < 0)
+ debugf0("ret=%d\n", ret);
+
+ return ret;
+}
+
+static void __devexit amd64_remove_one_instance(struct pci_dev *pdev)
+{
+ struct mem_ctl_info *mci;
+ struct amd64_pvt *pvt;
+
+ /* Remove from EDAC CORE tracking list */
+ mci = edac_mc_del_mc(&pdev->dev);
+ if (!mci)
+ return;
+
+ pvt = mci->pvt_info;
+
+ amd64_restore_ecc_error_reporting(pvt);
+
+ if (boot_cpu_data.x86 > 0xf)
+ amd64_teardown(pvt);
+
+ amd64_free_mc_sibling_devices(pvt);
+
+ kfree(pvt);
+ mci->pvt_info = NULL;
+
+ mci_lookup[pvt->mc_node_id] = NULL;
+
+ /* Free the EDAC CORE resources */
+ edac_mc_free(mci);
+}
+
+/*
+ * This table is part of the interface for loading drivers for PCI devices. The
+ * PCI core identifies what devices are on a system during boot, and then
+ * inquiry this table to see if this driver is for a given device found.
+ */
+static const struct pci_device_id amd64_pci_table[] __devinitdata = {
+ {
+ .vendor = PCI_VENDOR_ID_AMD,
+ .device = PCI_DEVICE_ID_AMD_K8_NB_MEMCTL,
+ .subvendor = PCI_ANY_ID,
+ .subdevice = PCI_ANY_ID,
+ .class = 0,
+ .class_mask = 0,
+ .driver_data = K8_CPUS
+ },
+ {
+ .vendor = PCI_VENDOR_ID_AMD,
+ .device = PCI_DEVICE_ID_AMD_10H_NB_DRAM,
+ .subvendor = PCI_ANY_ID,
+ .subdevice = PCI_ANY_ID,
+ .class = 0,
+ .class_mask = 0,
+ .driver_data = F10_CPUS
+ },
+ {
+ .vendor = PCI_VENDOR_ID_AMD,
+ .device = PCI_DEVICE_ID_AMD_11H_NB_DRAM,
+ .subvendor = PCI_ANY_ID,
+ .subdevice = PCI_ANY_ID,
+ .class = 0,
+ .class_mask = 0,
+ .driver_data = F11_CPUS
+ },
+ {0, }
+};
+MODULE_DEVICE_TABLE(pci, amd64_pci_table);
+
+static struct pci_driver amd64_pci_driver = {
+ .name = EDAC_MOD_STR,
+ .probe = amd64_init_one_instance,
+ .remove = __devexit_p(amd64_remove_one_instance),
+ .id_table = amd64_pci_table,
+};
+
+static void amd64_setup_pci_device(void)
+{
+ struct mem_ctl_info *mci;
+ struct amd64_pvt *pvt;
+
+ if (amd64_ctl_pci)
+ return;
+
+ mci = mci_lookup[0];
+ if (mci) {
+
+ pvt = mci->pvt_info;
+ amd64_ctl_pci =
+ edac_pci_create_generic_ctl(&pvt->dram_f2_ctl->dev,
+ EDAC_MOD_STR);
+
+ if (!amd64_ctl_pci) {
+ pr_warning("%s(): Unable to create PCI control\n",
+ __func__);
+
+ pr_warning("%s(): PCI error report via EDAC not set\n",
+ __func__);
+ }
+ }
+}
+
+static int __init amd64_edac_init(void)
+{
+ int nb, err = -ENODEV;
+
+ edac_printk(KERN_INFO, EDAC_MOD_STR, EDAC_AMD64_VERSION "\n");
+
+ opstate_init();
+
+ if (cache_k8_northbridges() < 0)
+ goto err_exit;
+
+ err = pci_register_driver(&amd64_pci_driver);
+ if (err)
+ return err;
+
+ /*
+ * At this point, the array 'pvt_lookup[]' contains pointers to alloc'd
+ * amd64_pvt structs. These will be used in the 2nd stage init function
+ * to finish initialization of the MC instances.
+ */
+ for (nb = 0; nb < num_k8_northbridges; nb++) {
+ if (!pvt_lookup[nb])
+ continue;
+
+ err = amd64_init_2nd_stage(pvt_lookup[nb]);
+ if (err)
+ goto err_2nd_stage;
+ }
+
+ amd64_setup_pci_device();
+
+ return 0;
+
+err_2nd_stage:
+ debugf0("2nd stage failed\n");
+
+err_exit:
+ pci_unregister_driver(&amd64_pci_driver);
+
+ return err;
+}
+
+static void __exit amd64_edac_exit(void)
+{
+ if (amd64_ctl_pci)
+ edac_pci_release_generic_ctl(amd64_ctl_pci);
+
+ pci_unregister_driver(&amd64_pci_driver);
+}
+
+module_init(amd64_edac_init);
+module_exit(amd64_edac_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("SoftwareBitMaker: Doug Thompson, "
+ "Dave Peterson, Thayne Harbaugh");
+MODULE_DESCRIPTION("MC support for AMD64 memory controllers - "
+ EDAC_AMD64_VERSION);
+module_param(edac_op_state, int, 0444);
+MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI");