amd64_edac: add module registration routines
authorDoug Thompson <dougthompson@xmission.com>
Mon, 27 Apr 2009 18:01:01 +0000 (20:01 +0200)
committerBorislav Petkov <borislav.petkov@amd.com>
Wed, 10 Jun 2009 10:19:28 +0000 (12:19 +0200)
Also, link into Kbuild by adding Kconfig and Makefile entries.

Borislav:
- Kconfig/Makefile splitting
- use zero-sized arrays for the sysfs attrs if not enabled
- rename sysfs attrs to more conform values
- shorten CONFIG_ names
- make multiple structure members assignment vertically aligned
- fix/cleanup comments
- fix function return value patterns
- fix err labels
- fix a memleak bug caught by Ingo
- remove the NUMA dependency and use num_k8_northbrides for initializing
  a driver instance per NB.
- do not copy the pvt contents into the mci struct in
  amd64_init_2nd_stage() and save it in the mci->pvt_info void ptr
  instead.
- cleanup debug calls
- simplify amd64_setup_pci_device()

Reviewed-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
drivers/edac/Kconfig
drivers/edac/Makefile
drivers/edac/amd64_edac.c
drivers/edac/amd64_edac.h

index 956982f..3dc650a 100644 (file)
@@ -58,6 +58,32 @@ config EDAC_MM_EDAC
          occurred so that a particular failing memory module can be
          replaced.  If unsure, select 'Y'.
 
+config EDAC_AMD64
+       tristate "AMD64 (Opteron, Athlon64) K8, F10h, F11h"
+       depends on EDAC_MM_EDAC && X86 && PCI
+       default m
+       help
+       Support for error detection and correction on the AMD 64
+       Families of Memory Controllers (K8, F10h and F11h)
+
+config EDAC_AMD64_ERROR_INJECTION
+       bool "Sysfs Error Injection facilities"
+       depends on EDAC_AMD64
+       help
+         Recent Opterons (Family 10h and later) provide for Memory Error
+         Injection into the ECC detection circuits. The amd64_edac module
+         allows the operator/user to inject Uncorrectable and Correctable
+         errors into DRAM.
+
+         When enabled, in each of the respective memory controller directories
+         (/sys/devices/system/edac/mc/mcX), there are 3 input files:
+
+         - inject_section (0..3, 16-byte section of 64-byte cacheline),
+         - inject_word (0..8, 16-bit word of 16-byte section),
+         - inject_ecc_vector (hex ecc vector: select bits of inject word)
+
+         In addition, there are two control files, inject_read and inject_write,
+         which trigger the DRAM ECC Read and Write respectively.
 
 config EDAC_AMD76X
        tristate "AMD 76x (760, 762, 768)"
index 5907681..633dc56 100644 (file)
@@ -30,6 +30,13 @@ obj-$(CONFIG_EDAC_I3000)             += i3000_edac.o
 obj-$(CONFIG_EDAC_X38)                 += x38_edac.o
 obj-$(CONFIG_EDAC_I82860)              += i82860_edac.o
 obj-$(CONFIG_EDAC_R82600)              += r82600_edac.o
+
+amd64_edac_mod-y :=  amd64_edac_err_types.o amd64_edac.o
+amd64_edac_mod-$(CONFIG_EDAC_DEBUG) += amd64_edac_dbg.o
+amd64_edac_mod-$(CONFIG_EDAC_AMD64_ERROR_INJECTION) += amd64_edac_inj.o
+
+obj-$(CONFIG_EDAC_AMD64)               += amd64_edac_mod.o
+
 obj-$(CONFIG_EDAC_PASEMI)              += pasemi_edac.o
 obj-$(CONFIG_EDAC_MPC85XX)             += mpc85xx_edac.o
 obj-$(CONFIG_EDAC_MV64X60)             += mv64x60_edac.o
index 3b6c421..c36bf40 100644 (file)
@@ -1,4 +1,5 @@
 #include "amd64_edac.h"
+#include <asm/k8.h>
 
 static struct edac_pci_ctl_info *amd64_ctl_pci;
 
@@ -2978,3 +2979,376 @@ static int amd64_check_ecc_enabled(struct amd64_pvt *pvt)
        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;
+       mci->edac_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_mc_node_id_from_pdev(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_mc_node_id_from_pdev(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_exit;
+       }
+
+       amd64_setup_pci_device();
+
+       return 0;
+
+err_exit:
+       debugf0("'finish_setup' stage failed\n");
+       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");
index e7aa760..a159957 100644 (file)
@@ -577,6 +577,21 @@ extern const char *ii_msgs[4];
 extern const char *ext_msgs[32];
 extern const char *htlink_msgs[8];
 
+#ifdef CONFIG_EDAC_DEBUG
+#define NUM_DBG_ATTRS 9
+#else
+#define NUM_DBG_ATTRS 0
+#endif
+
+#ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
+#define NUM_INJ_ATTRS 5
+#else
+#define NUM_INJ_ATTRS 0
+#endif
+
+extern struct mcidev_sysfs_attribute amd64_dbg_attrs[NUM_DBG_ATTRS],
+                                    amd64_inj_attrs[NUM_INJ_ATTRS];
+
 /*
  * Each of the PCI Device IDs types have their own set of hardware accessor
  * functions and per device encoding/decoding logic.