diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2011-11-02 16:55:15 -0700 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2011-11-02 16:55:15 -0700 |
commit | 6681ba7ec480bc839584fd0817991d248b4b9e44 (patch) | |
tree | 994fb1de40d58ce8dac821cf1fec727e2f902f47 /drivers/edac | |
parent | 06ef93e1b8405acac6ec900564e3ad1a8e3a72b2 (diff) | |
parent | 4d096ca7e65584dd5845e64c6400f920e694f672 (diff) | |
download | op-kernel-dev-6681ba7ec480bc839584fd0817991d248b4b9e44.zip op-kernel-dev-6681ba7ec480bc839584fd0817991d248b4b9e44.tar.gz |
Merge branch 'linux_next' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-edac
* 'linux_next' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-edac: (21 commits)
MAINTAINERS: add an entry for Edac Sandy Bridge driver
edac: tag sb_edac as EXPERIMENTAL, as it requires more testing
EDAC: Fix incorrect edac mode reporting in sb_edac
edac: sb_edac: Add it to the building system
edac: Add an experimental new driver to support Sandy Bridge CPU's
i7300_edac: Fix error cleanup logic
i7core_edac: Initialize memory name with cpu, channel, bank
i7core_edac: Fix compilation on 32 bits arch
i7core_edac: scrubbing fixups
EDAC: Correct Kconfig dependencies
i7core_edac: return -ENODEV if no MC is found
i7core_edac: use edac's own way to print errors
MAINTAINERS: remove dropped edac_mce.* from the file
i7core_edac: Drop the edac_mce facility
x86, MCE: Use notifier chain only for MCE decoding
EDAC i7core: Use mce socketid for better compatibility
i7core_edac: Don't enable memory scrubbing for Xeon 35xx
i7core_edac: Add scrubbing support
edac: Move edac main structs to include/linux/edac.h
i7core_edac: Fix oops when trying to inject errors
...
Diffstat (limited to 'drivers/edac')
-rw-r--r-- | drivers/edac/Kconfig | 16 | ||||
-rw-r--r-- | drivers/edac/Makefile | 2 | ||||
-rw-r--r-- | drivers/edac/edac_core.h | 350 | ||||
-rw-r--r-- | drivers/edac/edac_mce.c | 61 | ||||
-rw-r--r-- | drivers/edac/i7300_edac.c | 51 | ||||
-rw-r--r-- | drivers/edac/i7core_edac.c | 415 | ||||
-rw-r--r-- | drivers/edac/sb_edac.c | 1893 |
7 files changed, 2308 insertions, 480 deletions
diff --git a/drivers/edac/Kconfig b/drivers/edac/Kconfig index af1a17d..203361e 100644 --- a/drivers/edac/Kconfig +++ b/drivers/edac/Kconfig @@ -41,7 +41,7 @@ config EDAC_DEBUG config EDAC_DECODE_MCE tristate "Decode MCEs in human-readable form (only on AMD for now)" - depends on CPU_SUP_AMD && X86_MCE + depends on CPU_SUP_AMD && X86_MCE_AMD default y ---help--- Enable this option if you want to decode Machine Check Exceptions @@ -71,9 +71,6 @@ config EDAC_MM_EDAC occurred so that a particular failing memory module can be replaced. If unsure, select 'Y'. -config EDAC_MCE - bool - config EDAC_AMD64 tristate "AMD64 (Opteron, Athlon64) K8, F10h" depends on EDAC_MM_EDAC && AMD_NB && X86_64 && EDAC_DECODE_MCE @@ -173,8 +170,7 @@ config EDAC_I5400 config EDAC_I7CORE tristate "Intel i7 Core (Nehalem) processors" - depends on EDAC_MM_EDAC && PCI && X86 - select EDAC_MCE + depends on EDAC_MM_EDAC && PCI && X86 && X86_MCE_INTEL help Support for error detection and correction the Intel i7 Core (Nehalem) Integrated Memory Controller that exists on @@ -216,6 +212,14 @@ config EDAC_I7300 Support for error detection and correction the Intel Clarksboro MCH (Intel 7300 chipset). +config EDAC_SBRIDGE + tristate "Intel Sandy-Bridge Integrated MC" + depends on EDAC_MM_EDAC && PCI && X86 && X86_MCE_INTEL + depends on EXPERIMENTAL + help + Support for error detection and correction the Intel + Sandy Bridge Integrated Memory Controller. + config EDAC_MPC85XX tristate "Freescale MPC83xx / MPC85xx" depends on EDAC_MM_EDAC && FSL_SOC && (PPC_83xx || PPC_85xx) diff --git a/drivers/edac/Makefile b/drivers/edac/Makefile index 3e23913..196a63d 100644 --- a/drivers/edac/Makefile +++ b/drivers/edac/Makefile @@ -8,7 +8,6 @@ obj-$(CONFIG_EDAC) := edac_stub.o obj-$(CONFIG_EDAC_MM_EDAC) += edac_core.o -obj-$(CONFIG_EDAC_MCE) += edac_mce.o edac_core-y := edac_mc.o edac_device.o edac_mc_sysfs.o edac_pci_sysfs.o edac_core-y += edac_module.o edac_device_sysfs.o @@ -29,6 +28,7 @@ obj-$(CONFIG_EDAC_I5100) += i5100_edac.o obj-$(CONFIG_EDAC_I5400) += i5400_edac.o obj-$(CONFIG_EDAC_I7300) += i7300_edac.o obj-$(CONFIG_EDAC_I7CORE) += i7core_edac.o +obj-$(CONFIG_EDAC_SBRIDGE) += sb_edac.o obj-$(CONFIG_EDAC_E7XXX) += e7xxx_edac.o obj-$(CONFIG_EDAC_E752X) += e752x_edac.o obj-$(CONFIG_EDAC_I82443BXGX) += i82443bxgx_edac.o diff --git a/drivers/edac/edac_core.h b/drivers/edac/edac_core.h index 55b8278..fe90cd4 100644 --- a/drivers/edac/edac_core.h +++ b/drivers/edac/edac_core.h @@ -34,11 +34,10 @@ #include <linux/platform_device.h> #include <linux/sysdev.h> #include <linux/workqueue.h> +#include <linux/edac.h> -#define EDAC_MC_LABEL_LEN 31 #define EDAC_DEVICE_NAME_LEN 31 #define EDAC_ATTRIB_VALUE_LEN 15 -#define MC_PROC_NAME_MAX_LEN 7 #if PAGE_SHIFT < 20 #define PAGES_TO_MiB(pages) ((pages) >> (20 - PAGE_SHIFT)) @@ -101,353 +100,6 @@ extern int edac_debug_level; #define edac_dev_name(dev) (dev)->dev_name -/* memory devices */ -enum dev_type { - DEV_UNKNOWN = 0, - DEV_X1, - DEV_X2, - DEV_X4, - DEV_X8, - DEV_X16, - DEV_X32, /* Do these parts exist? */ - DEV_X64 /* Do these parts exist? */ -}; - -#define DEV_FLAG_UNKNOWN BIT(DEV_UNKNOWN) -#define DEV_FLAG_X1 BIT(DEV_X1) -#define DEV_FLAG_X2 BIT(DEV_X2) -#define DEV_FLAG_X4 BIT(DEV_X4) -#define DEV_FLAG_X8 BIT(DEV_X8) -#define DEV_FLAG_X16 BIT(DEV_X16) -#define DEV_FLAG_X32 BIT(DEV_X32) -#define DEV_FLAG_X64 BIT(DEV_X64) - -/* memory types */ -enum mem_type { - MEM_EMPTY = 0, /* Empty csrow */ - MEM_RESERVED, /* Reserved csrow type */ - MEM_UNKNOWN, /* Unknown csrow type */ - MEM_FPM, /* Fast page mode */ - MEM_EDO, /* Extended data out */ - MEM_BEDO, /* Burst Extended data out */ - MEM_SDR, /* Single data rate SDRAM */ - MEM_RDR, /* Registered single data rate SDRAM */ - MEM_DDR, /* Double data rate SDRAM */ - MEM_RDDR, /* Registered Double data rate SDRAM */ - MEM_RMBS, /* Rambus DRAM */ - MEM_DDR2, /* DDR2 RAM */ - MEM_FB_DDR2, /* fully buffered DDR2 */ - MEM_RDDR2, /* Registered DDR2 RAM */ - MEM_XDR, /* Rambus XDR */ - MEM_DDR3, /* DDR3 RAM */ - MEM_RDDR3, /* Registered DDR3 RAM */ -}; - -#define MEM_FLAG_EMPTY BIT(MEM_EMPTY) -#define MEM_FLAG_RESERVED BIT(MEM_RESERVED) -#define MEM_FLAG_UNKNOWN BIT(MEM_UNKNOWN) -#define MEM_FLAG_FPM BIT(MEM_FPM) -#define MEM_FLAG_EDO BIT(MEM_EDO) -#define MEM_FLAG_BEDO BIT(MEM_BEDO) -#define MEM_FLAG_SDR BIT(MEM_SDR) -#define MEM_FLAG_RDR BIT(MEM_RDR) -#define MEM_FLAG_DDR BIT(MEM_DDR) -#define MEM_FLAG_RDDR BIT(MEM_RDDR) -#define MEM_FLAG_RMBS BIT(MEM_RMBS) -#define MEM_FLAG_DDR2 BIT(MEM_DDR2) -#define MEM_FLAG_FB_DDR2 BIT(MEM_FB_DDR2) -#define MEM_FLAG_RDDR2 BIT(MEM_RDDR2) -#define MEM_FLAG_XDR BIT(MEM_XDR) -#define MEM_FLAG_DDR3 BIT(MEM_DDR3) -#define MEM_FLAG_RDDR3 BIT(MEM_RDDR3) - -/* chipset Error Detection and Correction capabilities and mode */ -enum edac_type { - EDAC_UNKNOWN = 0, /* Unknown if ECC is available */ - EDAC_NONE, /* Doesn't support ECC */ - EDAC_RESERVED, /* Reserved ECC type */ - EDAC_PARITY, /* Detects parity errors */ - EDAC_EC, /* Error Checking - no correction */ - EDAC_SECDED, /* Single bit error correction, Double detection */ - EDAC_S2ECD2ED, /* Chipkill x2 devices - do these exist? */ - EDAC_S4ECD4ED, /* Chipkill x4 devices */ - EDAC_S8ECD8ED, /* Chipkill x8 devices */ - EDAC_S16ECD16ED, /* Chipkill x16 devices */ -}; - -#define EDAC_FLAG_UNKNOWN BIT(EDAC_UNKNOWN) -#define EDAC_FLAG_NONE BIT(EDAC_NONE) -#define EDAC_FLAG_PARITY BIT(EDAC_PARITY) -#define EDAC_FLAG_EC BIT(EDAC_EC) -#define EDAC_FLAG_SECDED BIT(EDAC_SECDED) -#define EDAC_FLAG_S2ECD2ED BIT(EDAC_S2ECD2ED) -#define EDAC_FLAG_S4ECD4ED BIT(EDAC_S4ECD4ED) -#define EDAC_FLAG_S8ECD8ED BIT(EDAC_S8ECD8ED) -#define EDAC_FLAG_S16ECD16ED BIT(EDAC_S16ECD16ED) - -/* scrubbing capabilities */ -enum scrub_type { - SCRUB_UNKNOWN = 0, /* Unknown if scrubber is available */ - SCRUB_NONE, /* No scrubber */ - SCRUB_SW_PROG, /* SW progressive (sequential) scrubbing */ - SCRUB_SW_SRC, /* Software scrub only errors */ - SCRUB_SW_PROG_SRC, /* Progressive software scrub from an error */ - SCRUB_SW_TUNABLE, /* Software scrub frequency is tunable */ - SCRUB_HW_PROG, /* HW progressive (sequential) scrubbing */ - SCRUB_HW_SRC, /* Hardware scrub only errors */ - SCRUB_HW_PROG_SRC, /* Progressive hardware scrub from an error */ - SCRUB_HW_TUNABLE /* Hardware scrub frequency is tunable */ -}; - -#define SCRUB_FLAG_SW_PROG BIT(SCRUB_SW_PROG) -#define SCRUB_FLAG_SW_SRC BIT(SCRUB_SW_SRC) -#define SCRUB_FLAG_SW_PROG_SRC BIT(SCRUB_SW_PROG_SRC) -#define SCRUB_FLAG_SW_TUN BIT(SCRUB_SW_SCRUB_TUNABLE) -#define SCRUB_FLAG_HW_PROG BIT(SCRUB_HW_PROG) -#define SCRUB_FLAG_HW_SRC BIT(SCRUB_HW_SRC) -#define SCRUB_FLAG_HW_PROG_SRC BIT(SCRUB_HW_PROG_SRC) -#define SCRUB_FLAG_HW_TUN BIT(SCRUB_HW_TUNABLE) - -/* FIXME - should have notify capabilities: NMI, LOG, PROC, etc */ - -/* EDAC internal operation states */ -#define OP_ALLOC 0x100 -#define OP_RUNNING_POLL 0x201 -#define OP_RUNNING_INTERRUPT 0x202 -#define OP_RUNNING_POLL_INTR 0x203 -#define OP_OFFLINE 0x300 - -/* - * There are several things to be aware of that aren't at all obvious: - * - * - * SOCKETS, SOCKET SETS, BANKS, ROWS, CHIP-SELECT ROWS, CHANNELS, etc.. - * - * These are some of the many terms that are thrown about that don't always - * mean what people think they mean (Inconceivable!). In the interest of - * creating a common ground for discussion, terms and their definitions - * will be established. - * - * Memory devices: The individual chip on a memory stick. These devices - * commonly output 4 and 8 bits each. Grouping several - * of these in parallel provides 64 bits which is common - * for a memory stick. - * - * Memory Stick: A printed circuit board that aggregates multiple - * memory devices in parallel. This is the atomic - * memory component that is purchaseable by Joe consumer - * and loaded into a memory socket. - * - * Socket: A physical connector on the motherboard that accepts - * a single memory stick. - * - * Channel: Set of memory devices on a memory stick that must be - * grouped in parallel with one or more additional - * channels from other memory sticks. This parallel - * grouping of the output from multiple channels are - * necessary for the smallest granularity of memory access. - * Some memory controllers are capable of single channel - - * which means that memory sticks can be loaded - * individually. Other memory controllers are only - * capable of dual channel - which means that memory - * sticks must be loaded as pairs (see "socket set"). - * - * Chip-select row: All of the memory devices that are selected together. - * for a single, minimum grain of memory access. - * This selects all of the parallel memory devices across - * all of the parallel channels. Common chip-select rows - * for single channel are 64 bits, for dual channel 128 - * bits. - * - * Single-Ranked stick: A Single-ranked stick has 1 chip-select row of memory. - * Motherboards commonly drive two chip-select pins to - * a memory stick. A single-ranked stick, will occupy - * only one of those rows. The other will be unused. - * - * Double-Ranked stick: A double-ranked stick has two chip-select rows which - * access different sets of memory devices. The two - * rows cannot be accessed concurrently. - * - * Double-sided stick: DEPRECATED TERM, see Double-Ranked stick. - * A double-sided stick has two chip-select rows which - * access different sets of memory devices. The two - * rows cannot be accessed concurrently. "Double-sided" - * is irrespective of the memory devices being mounted - * on both sides of the memory stick. - * - * Socket set: All of the memory sticks that are required for - * a single memory access or all of the memory sticks - * spanned by a chip-select row. A single socket set - * has two chip-select rows and if double-sided sticks - * are used these will occupy those chip-select rows. - * - * Bank: This term is avoided because it is unclear when - * needing to distinguish between chip-select rows and - * socket sets. - * - * Controller pages: - * - * Physical pages: - * - * Virtual pages: - * - * - * STRUCTURE ORGANIZATION AND CHOICES - * - * - * - * PS - I enjoyed writing all that about as much as you enjoyed reading it. - */ - -struct channel_info { - int chan_idx; /* channel index */ - u32 ce_count; /* Correctable Errors for this CHANNEL */ - char label[EDAC_MC_LABEL_LEN + 1]; /* DIMM label on motherboard */ - struct csrow_info *csrow; /* the parent */ -}; - -struct csrow_info { - unsigned long first_page; /* first page number in dimm */ - unsigned long last_page; /* last page number in dimm */ - unsigned long page_mask; /* used for interleaving - - * 0UL for non intlv - */ - u32 nr_pages; /* number of pages in csrow */ - u32 grain; /* granularity of reported error in bytes */ - int csrow_idx; /* the chip-select row */ - enum dev_type dtype; /* memory device type */ - u32 ue_count; /* Uncorrectable Errors for this csrow */ - u32 ce_count; /* Correctable Errors for this csrow */ - enum mem_type mtype; /* memory csrow type */ - enum edac_type edac_mode; /* EDAC mode for this csrow */ - struct mem_ctl_info *mci; /* the parent */ - - struct kobject kobj; /* sysfs kobject for this csrow */ - - /* channel information for this csrow */ - u32 nr_channels; - struct channel_info *channels; -}; - -struct mcidev_sysfs_group { - const char *name; /* group name */ - const struct mcidev_sysfs_attribute *mcidev_attr; /* group attributes */ -}; - -struct mcidev_sysfs_group_kobj { - struct list_head list; /* list for all instances within a mc */ - - struct kobject kobj; /* kobj for the group */ - - const struct mcidev_sysfs_group *grp; /* group description table */ - struct mem_ctl_info *mci; /* the parent */ -}; - -/* mcidev_sysfs_attribute structure - * used for driver sysfs attributes and in mem_ctl_info - * sysfs top level entries - */ -struct mcidev_sysfs_attribute { - /* It should use either attr or grp */ - struct attribute attr; - const struct mcidev_sysfs_group *grp; /* Points to a group of attributes */ - - /* Ops for show/store values at the attribute - not used on group */ - ssize_t (*show)(struct mem_ctl_info *,char *); - ssize_t (*store)(struct mem_ctl_info *, const char *,size_t); -}; - -/* MEMORY controller information structure - */ -struct mem_ctl_info { - struct list_head link; /* for global list of mem_ctl_info structs */ - - struct module *owner; /* Module owner of this control struct */ - - unsigned long mtype_cap; /* memory types supported by mc */ - unsigned long edac_ctl_cap; /* Mem controller EDAC capabilities */ - unsigned long edac_cap; /* configuration capabilities - this is - * closely related to edac_ctl_cap. The - * difference is that the controller may be - * capable of s4ecd4ed which would be listed - * in edac_ctl_cap, but if channels aren't - * capable of s4ecd4ed then the edac_cap would - * not have that capability. - */ - unsigned long scrub_cap; /* chipset scrub capabilities */ - enum scrub_type scrub_mode; /* current scrub mode */ - - /* Translates sdram memory scrub rate given in bytes/sec to the - internal representation and configures whatever else needs - to be configured. - */ - int (*set_sdram_scrub_rate) (struct mem_ctl_info * mci, u32 bw); - - /* Get the current sdram memory scrub rate from the internal - representation and converts it to the closest matching - bandwidth in bytes/sec. - */ - int (*get_sdram_scrub_rate) (struct mem_ctl_info * mci); - - - /* pointer to edac checking routine */ - void (*edac_check) (struct mem_ctl_info * mci); - - /* - * Remaps memory pages: controller pages to physical pages. - * For most MC's, this will be NULL. - */ - /* FIXME - why not send the phys page to begin with? */ - unsigned long (*ctl_page_to_phys) (struct mem_ctl_info * mci, - unsigned long page); - int mc_idx; - int nr_csrows; - struct csrow_info *csrows; - /* - * FIXME - what about controllers on other busses? - IDs must be - * unique. dev pointer should be sufficiently unique, but - * BUS:SLOT.FUNC numbers may not be unique. - */ - struct device *dev; - const char *mod_name; - const char *mod_ver; - const char *ctl_name; - const char *dev_name; - char proc_name[MC_PROC_NAME_MAX_LEN + 1]; - void *pvt_info; - u32 ue_noinfo_count; /* Uncorrectable Errors w/o info */ - u32 ce_noinfo_count; /* Correctable Errors w/o info */ - u32 ue_count; /* Total Uncorrectable Errors for this MC */ - u32 ce_count; /* Total Correctable Errors for this MC */ - unsigned long start_time; /* mci load start time (in jiffies) */ - - struct completion complete; - - /* edac sysfs device control */ - struct kobject edac_mci_kobj; - - /* list for all grp instances within a mc */ - struct list_head grp_kobj_list; - - /* Additional top controller level attributes, but specified - * by the low level driver. - * - * Set by the low level driver to provide attributes at the - * controller level, same level as 'ue_count' and 'ce_count' above. - * An array of structures, NULL terminated - * - * If attributes are desired, then set to array of attributes - * If no attributes are desired, leave NULL - */ - const struct mcidev_sysfs_attribute *mc_driver_sysfs_attributes; - - /* work struct for this MC */ - struct delayed_work work; - - /* the internal state of this controller instance */ - int op_state; -}; - /* * The following are the structures to provide for a generic * or abstract 'edac_device'. This set of structures and the diff --git a/drivers/edac/edac_mce.c b/drivers/edac/edac_mce.c deleted file mode 100644 index 9ccdc5b..0000000 --- a/drivers/edac/edac_mce.c +++ /dev/null @@ -1,61 +0,0 @@ -/* Provides edac interface to mcelog events - * - * This file may be distributed under the terms of the - * GNU General Public License version 2. - * - * Copyright (c) 2009 by: - * Mauro Carvalho Chehab <mchehab@redhat.com> - * - * Red Hat Inc. http://www.redhat.com - */ - -#include <linux/module.h> -#include <linux/edac_mce.h> -#include <asm/mce.h> - -int edac_mce_enabled; -EXPORT_SYMBOL_GPL(edac_mce_enabled); - - -/* - * Extension interface - */ - -static LIST_HEAD(edac_mce_list); -static DEFINE_MUTEX(edac_mce_lock); - -int edac_mce_register(struct edac_mce *edac_mce) -{ - mutex_lock(&edac_mce_lock); - list_add_tail(&edac_mce->list, &edac_mce_list); - mutex_unlock(&edac_mce_lock); - return 0; -} -EXPORT_SYMBOL(edac_mce_register); - -void edac_mce_unregister(struct edac_mce *edac_mce) -{ - mutex_lock(&edac_mce_lock); - list_del(&edac_mce->list); - mutex_unlock(&edac_mce_lock); -} -EXPORT_SYMBOL(edac_mce_unregister); - -int edac_mce_parse(struct mce *mce) -{ - struct edac_mce *edac_mce; - - list_for_each_entry(edac_mce, &edac_mce_list, list) { - if (edac_mce->check_error(edac_mce->priv, mce)) - return 1; - } - - /* Nobody queued the error */ - return 0; -} -EXPORT_SYMBOL_GPL(edac_mce_parse); - -MODULE_LICENSE("GPL"); -MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>"); -MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)"); -MODULE_DESCRIPTION("EDAC Driver for mcelog captured errors"); diff --git a/drivers/edac/i7300_edac.c b/drivers/edac/i7300_edac.c index a76fe83..6104dba 100644 --- a/drivers/edac/i7300_edac.c +++ b/drivers/edac/i7300_edac.c @@ -372,7 +372,7 @@ static const char *get_err_from_table(const char *table[], int size, int pos) static void i7300_process_error_global(struct mem_ctl_info *mci) { struct i7300_pvt *pvt; - u32 errnum, value; + u32 errnum, error_reg; unsigned long errors; const char *specific; bool is_fatal; @@ -381,9 +381,9 @@ static void i7300_process_error_global(struct mem_ctl_info *mci) /* read in the 1st FATAL error register */ pci_read_config_dword(pvt->pci_dev_16_2_fsb_err_regs, - FERR_GLOBAL_HI, &value); - if (unlikely(value)) { - errors = value; + FERR_GLOBAL_HI, &error_reg); + if (unlikely(error_reg)) { + errors = error_reg; errnum = find_first_bit(&errors, ARRAY_SIZE(ferr_global_hi_name)); specific = GET_ERR_FROM_TABLE(ferr_global_hi_name, errnum); @@ -391,15 +391,15 @@ static void i7300_process_error_global(struct mem_ctl_info *mci) /* Clear the error bit */ pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs, - FERR_GLOBAL_HI, value); + FERR_GLOBAL_HI, error_reg); goto error_global; } pci_read_config_dword(pvt->pci_dev_16_2_fsb_err_regs, - FERR_GLOBAL_LO, &value); - if (unlikely(value)) { - errors = value; + FERR_GLOBAL_LO, &error_reg); + if (unlikely(error_reg)) { + errors = error_reg; errnum = find_first_bit(&errors, ARRAY_SIZE(ferr_global_lo_name)); specific = GET_ERR_FROM_TABLE(ferr_global_lo_name, errnum); @@ -407,7 +407,7 @@ static void i7300_process_error_global(struct mem_ctl_info *mci) /* Clear the error bit */ pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs, - FERR_GLOBAL_LO, value); + FERR_GLOBAL_LO, error_reg); goto error_global; } @@ -427,7 +427,7 @@ error_global: static void i7300_process_fbd_error(struct mem_ctl_info *mci) { struct i7300_pvt *pvt; - u32 errnum, value; + u32 errnum, value, error_reg; u16 val16; unsigned branch, channel, bank, rank, cas, ras; u32 syndrome; @@ -440,14 +440,14 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci) /* read in the 1st FATAL error register */ pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, - FERR_FAT_FBD, &value); - if (unlikely(value & FERR_FAT_FBD_ERR_MASK)) { - errors = value & FERR_FAT_FBD_ERR_MASK ; + FERR_FAT_FBD, &error_reg); + if (unlikely(error_reg & FERR_FAT_FBD_ERR_MASK)) { + errors = error_reg & FERR_FAT_FBD_ERR_MASK ; errnum = find_first_bit(&errors, ARRAY_SIZE(ferr_fat_fbd_name)); specific = GET_ERR_FROM_TABLE(ferr_fat_fbd_name, errnum); + branch = (GET_FBD_FAT_IDX(error_reg) == 2) ? 1 : 0; - branch = (GET_FBD_FAT_IDX(value) == 2) ? 1 : 0; pci_read_config_word(pvt->pci_dev_16_1_fsb_addr_map, NRECMEMA, &val16); bank = NRECMEMA_BANK(val16); @@ -455,11 +455,14 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci) pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, NRECMEMB, &value); - is_wr = NRECMEMB_IS_WR(value); cas = NRECMEMB_CAS(value); ras = NRECMEMB_RAS(value); + /* Clean the error register */ + pci_write_config_dword(pvt->pci_dev_16_1_fsb_addr_map, + FERR_FAT_FBD, error_reg); + snprintf(pvt->tmp_prt_buffer, PAGE_SIZE, "FATAL (Branch=%d DRAM-Bank=%d %s " "RAS=%d CAS=%d Err=0x%lx (%s))", @@ -476,21 +479,17 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci) /* read in the 1st NON-FATAL error register */ pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, - FERR_NF_FBD, &value); - if (unlikely(value & FERR_NF_FBD_ERR_MASK)) { - errors = value & FERR_NF_FBD_ERR_MASK; + FERR_NF_FBD, &error_reg); + if (unlikely(error_reg & FERR_NF_FBD_ERR_MASK)) { + errors = error_reg & FERR_NF_FBD_ERR_MASK; errnum = find_first_bit(&errors, ARRAY_SIZE(ferr_nf_fbd_name)); specific = GET_ERR_FROM_TABLE(ferr_nf_fbd_name, errnum); - - /* Clear the error bit */ - pci_write_config_dword(pvt->pci_dev_16_2_fsb_err_regs, - FERR_GLOBAL_LO, value); + branch = (GET_FBD_FAT_IDX(error_reg) == 2) ? 1 : 0; pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, REDMEMA, &syndrome); - branch = (GET_FBD_FAT_IDX(value) == 2) ? 1 : 0; pci_read_config_word(pvt->pci_dev_16_1_fsb_addr_map, RECMEMA, &val16); bank = RECMEMA_BANK(val16); @@ -498,18 +497,20 @@ static void i7300_process_fbd_error(struct mem_ctl_info *mci) pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, RECMEMB, &value); - is_wr = RECMEMB_IS_WR(value); cas = RECMEMB_CAS(value); ras = RECMEMB_RAS(value); pci_read_config_dword(pvt->pci_dev_16_1_fsb_addr_map, REDMEMB, &value); - channel = (branch << 1); if (IS_SECOND_CH(value)) channel++; + /* Clear the error bit */ + pci_write_config_dword(pvt->pci_dev_16_1_fsb_addr_map, + FERR_NF_FBD, error_reg); + /* Form out message */ snprintf(pvt->tmp_prt_buffer, PAGE_SIZE, "Corrected error (Branch=%d, Channel %d), " diff --git a/drivers/edac/i7core_edac.c b/drivers/edac/i7core_edac.c index f6cf448..70ad892 100644 --- a/drivers/edac/i7core_edac.c +++ b/drivers/edac/i7core_edac.c @@ -31,11 +31,13 @@ #include <linux/pci_ids.h> #include <linux/slab.h> #include <linux/delay.h> +#include <linux/dmi.h> #include <linux/edac.h> #include <linux/mmzone.h> -#include <linux/edac_mce.h> #include <linux/smp.h> +#include <asm/mce.h> #include <asm/processor.h> +#include <asm/div64.h> #include "edac_core.h" @@ -78,6 +80,8 @@ MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices"); /* OFFSETS for Device 0 Function 0 */ #define MC_CFG_CONTROL 0x90 + #define MC_CFG_UNLOCK 0x02 + #define MC_CFG_LOCK 0x00 /* OFFSETS for Device 3 Function 0 */ @@ -98,6 +102,15 @@ MODULE_PARM_DESC(use_pci_fixup, "Enable PCI fixup to seek for hidden devices"); #define DIMM0_COR_ERR(r) ((r) & 0x7fff) /* OFFSETS for Device 3 Function 2, as inicated on Xeon 5500 datasheet */ +#define MC_SSRCONTROL 0x48 + #define SSR_MODE_DISABLE 0x00 + #define SSR_MODE_ENABLE 0x01 + #define SSR_MODE_MASK 0x03 + +#define MC_SCRUB_CONTROL 0x4c + #define STARTSCRUB (1 << 24) + #define SCRUBINTERVAL_MASK 0xffffff + #define MC_COR_ECC_CNT_0 0x80 #define MC_COR_ECC_CNT_1 0x84 #define MC_COR_ECC_CNT_2 0x88 @@ -253,10 +266,7 @@ struct i7core_pvt { unsigned long rdimm_ce_count[NUM_CHANS][MAX_DIMMS]; int rdimm_last_ce_count[NUM_CHANS][MAX_DIMMS]; - unsigned int is_registered; - - /* mcelog glue */ - struct edac_mce edac_mce; + bool is_registered, enable_scrub; /* Fifo double buffers */ struct mce mce_entry[MCE_LOG_LEN]; @@ -268,6 +278,9 @@ struct i7core_pvt { /* Count indicator to show errors not got */ unsigned mce_overrun; + /* DCLK Frequency used for computing scrub rate */ + int dclk_freq; + /* Struct to control EDAC polling */ struct edac_pci_ctl_info *i7core_pci; }; @@ -281,8 +294,7 @@ static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = { /* Memory controller */ { PCI_DESCR(3, 0, PCI_DEVICE_ID_INTEL_I7_MCR) }, { PCI_DESCR(3, 1, PCI_DEVICE_ID_INTEL_I7_MC_TAD) }, - - /* Exists only for RDIMM */ + /* Exists only for RDIMM */ { PCI_DESCR(3, 2, PCI_DEVICE_ID_INTEL_I7_MC_RAS), .optional = 1 }, { PCI_DESCR(3, 4, PCI_DEVICE_ID_INTEL_I7_MC_TEST) }, @@ -303,6 +315,16 @@ static const struct pci_id_descr pci_dev_descr_i7core_nehalem[] = { { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_I7_MC_CH2_ADDR) }, { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_I7_MC_CH2_RANK) }, { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_I7_MC_CH2_TC) }, + + /* Generic Non-core registers */ + /* + * This is the PCI device on i7core and on Xeon 35xx (8086:2c41) + * On Xeon 55xx, however, it has a different id (8086:2c40). So, + * the probing code needs to test for the other address in case of + * failure of this one + */ + { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_I7_NONCORE) }, + }; static const struct pci_id_descr pci_dev_descr_lynnfield[] = { @@ -319,6 +341,12 @@ static const struct pci_id_descr pci_dev_descr_lynnfield[] = { { PCI_DESCR( 5, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_ADDR) }, { PCI_DESCR( 5, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_RANK) }, { PCI_DESCR( 5, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH1_TC) }, + + /* + * This is the PCI device has an alternate address on some + * processors like Core i7 860 + */ + { PCI_DESCR( 0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE) }, }; static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = { @@ -346,6 +374,10 @@ static const struct pci_id_descr pci_dev_descr_i7core_westmere[] = { { PCI_DESCR(6, 1, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_ADDR_REV2) }, { PCI_DESCR(6, 2, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_RANK_REV2) }, { PCI_DESCR(6, 3, PCI_DEVICE_ID_INTEL_LYNNFIELD_MC_CH2_TC_REV2) }, + + /* Generic Non-core registers */ + { PCI_DESCR(0, 0, PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2) }, + }; #define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) } @@ -714,6 +746,10 @@ static int get_dimm_config(const struct mem_ctl_info *mci) csr->edac_mode = mode; csr->mtype = mtype; + snprintf(csr->channels[0].label, + sizeof(csr->channels[0].label), + "CPU#%uChannel#%u_DIMM#%u", + pvt->i7core_dev->socket, i, j); csrow++; } @@ -731,7 +767,7 @@ static int get_dimm_config(const struct mem_ctl_info *mci) debugf1("\t\t%#x\t%#x\t%#x\n", (value[j] >> 27) & 0x1, (value[j] >> 24) & 0x7, - (value[j] && ((1 << 24) - 1))); + (value[j] & ((1 << 24) - 1))); } return 0; @@ -1324,6 +1360,20 @@ static int i7core_get_onedevice(struct pci_dev **prev, pdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_descr->dev_id, *prev); + /* + * On Xeon 55xx, the Intel Quckpath Arch Generic Non-core regs + * is at addr 8086:2c40, instead of 8086:2c41. So, we need + * to probe for the alternate address in case of failure + */ + if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_I7_NONCORE && !pdev) + pdev = pci_get_device(PCI_VENDOR_ID_INTEL, + PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT, *prev); + + if (dev_descr->dev_id == PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE && !pdev) + pdev = pci_get_device(PCI_VENDOR_ID_INTEL, + PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT, + *prev); + if (!pdev) { if (*prev) { *prev = pdev; @@ -1444,8 +1494,10 @@ static int mci_bind_devs(struct mem_ctl_info *mci, struct i7core_pvt *pvt = mci->pvt_info; struct pci_dev *pdev; int i, func, slot; + char *family; - pvt->is_registered = 0; + pvt->is_registered = false; + pvt->enable_scrub = false; for (i = 0; i < i7core_dev->n_devs; i++) { pdev = i7core_dev->pdev[i]; if (!pdev) @@ -1461,9 +1513,37 @@ static int mci_bind_devs(struct mem_ctl_info *mci, if (unlikely(func > MAX_CHAN_FUNC)) goto error; pvt->pci_ch[slot - 4][func] = pdev; - } else if (!slot && !func) + } else if (!slot && !func) { pvt->pci_noncore = pdev; - else + + /* Detect the processor family */ + switch (pdev->device) { + case PCI_DEVICE_ID_INTEL_I7_NONCORE: + family = "Xeon 35xx/ i7core"; + pvt->enable_scrub = false; + break; + case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_ALT: + family = "i7-800/i5-700"; + pvt->enable_scrub = false; + break; + case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE: + family = "Xeon 34xx"; + pvt->enable_scrub = false; + break; + case PCI_DEVICE_ID_INTEL_I7_NONCORE_ALT: + family = "Xeon 55xx"; + pvt->enable_scrub = true; + break; + case PCI_DEVICE_ID_INTEL_LYNNFIELD_NONCORE_REV2: + family = "Xeon 56xx / i7-900"; + pvt->enable_scrub = true; + break; + default: + family = "unknown"; + pvt->enable_scrub = false; + } + debugf0("Detected a processor type %s\n", family); + } else goto error; debugf0("Associated fn %d.%d, dev = %p, socket %d\n", @@ -1472,7 +1552,7 @@ static int mci_bind_devs(struct mem_ctl_info *mci, if (PCI_SLOT(pdev->devfn) == 3 && PCI_FUNC(pdev->devfn) == 2) - pvt->is_registered = 1; + pvt->is_registered = true; } return 0; @@ -1826,33 +1906,43 @@ check_ce_error: * WARNING: As this routine should be called at NMI time, extra care should * be taken to avoid deadlocks, and to be as fast as possible. */ -static int i7core_mce_check_error(void *priv, struct mce *mce) +static int i7core_mce_check_error(struct notifier_block *nb, unsigned long val, + void *data) { - struct mem_ctl_info *mci = priv; - struct i7core_pvt *pvt = mci->pvt_info; + struct mce *mce = (struct mce *)data; + struct i7core_dev *i7_dev; + struct mem_ctl_info *mci; + struct i7core_pvt *pvt; + + i7_dev = get_i7core_dev(mce->socketid); + if (!i7_dev) + return NOTIFY_BAD; + + mci = i7_dev->mci; + pvt = mci->pvt_info; /* * Just let mcelog handle it if the error is * outside the memory controller */ if (((mce->status & 0xffff) >> 7) != 1) - return 0; + return NOTIFY_DONE; /* Bank 8 registers are the only ones that we know how to handle */ if (mce->bank != 8) - return 0; + return NOTIFY_DONE; #ifdef CONFIG_SMP /* Only handle if it is the right mc controller */ - if (cpu_data(mce->cpu).phys_proc_id != pvt->i7core_dev->socket) - return 0; + if (mce->socketid != pvt->i7core_dev->socket) + return NOTIFY_DONE; #endif smp_rmb(); if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) { smp_wmb(); pvt->mce_overrun++; - return 0; + return NOTIFY_DONE; } /* Copy memory error at the ringbuffer */ @@ -1865,7 +1955,240 @@ static int i7core_mce_check_error(void *priv, struct mce *mce) i7core_check_error(mci); /* Advise mcelog that the errors were handled */ - return 1; + return NOTIFY_STOP; +} + +static struct notifier_block i7_mce_dec = { + .notifier_call = i7core_mce_check_error, +}; + +struct memdev_dmi_entry { + u8 type; + u8 length; + u16 handle; + u16 phys_mem_array_handle; + u16 mem_err_info_handle; + u16 total_width; + u16 data_width; + u16 size; + u8 form; + u8 device_set; + u8 device_locator; + u8 bank_locator; + u8 memory_type; + u16 type_detail; + u16 speed; + u8 manufacturer; + u8 serial_number; + u8 asset_tag; + u8 part_number; + u8 attributes; + u32 extended_size; + u16 conf_mem_clk_speed; +} __attribute__((__packed__)); + + +/* + * Decode the DRAM Clock Frequency, be paranoid, make sure that all + * memory devices show the same speed, and if they don't then consider + * all speeds to be invalid. + */ +static void decode_dclk(const struct dmi_header *dh, void *_dclk_freq) +{ + int *dclk_freq = _dclk_freq; + u16 dmi_mem_clk_speed; + + if (*dclk_freq == -1) + return; + + if (dh->type == DMI_ENTRY_MEM_DEVICE) { + struct memdev_dmi_entry *memdev_dmi_entry = + (struct memdev_dmi_entry *)dh; + unsigned long conf_mem_clk_speed_offset = + (unsigned long)&memdev_dmi_entry->conf_mem_clk_speed - + (unsigned long)&memdev_dmi_entry->type; + unsigned long speed_offset = + (unsigned long)&memdev_dmi_entry->speed - + (unsigned long)&memdev_dmi_entry->type; + + /* Check that a DIMM is present */ + if (memdev_dmi_entry->size == 0) + return; + + /* + * Pick the configured speed if it's available, otherwise + * pick the DIMM speed, or we don't have a speed. + */ + if (memdev_dmi_entry->length > conf_mem_clk_speed_offset) { + dmi_mem_clk_speed = + memdev_dmi_entry->conf_mem_clk_speed; + } else if (memdev_dmi_entry->length > speed_offset) { + dmi_mem_clk_speed = memdev_dmi_entry->speed; + } else { + *dclk_freq = -1; + return; + } + + if (*dclk_freq == 0) { + /* First pass, speed was 0 */ + if (dmi_mem_clk_speed > 0) { + /* Set speed if a valid speed is read */ + *dclk_freq = dmi_mem_clk_speed; + } else { + /* Otherwise we don't have a valid speed */ + *dclk_freq = -1; + } + } else if (*dclk_freq > 0 && + *dclk_freq != dmi_mem_clk_speed) { + /* + * If we have a speed, check that all DIMMS are the same + * speed, otherwise set the speed as invalid. + */ + *dclk_freq = -1; + } + } +} + +/* + * The default DCLK frequency is used as a fallback if we + * fail to find anything reliable in the DMI. The value + * is taken straight from the datasheet. + */ +#define DEFAULT_DCLK_FREQ 800 + +static int get_dclk_freq(void) +{ + int dclk_freq = 0; + + dmi_walk(decode_dclk, (void *)&dclk_freq); + + if (dclk_freq < 1) + return DEFAULT_DCLK_FREQ; + + return dclk_freq; +} + +/* + * set_sdram_scrub_rate This routine sets byte/sec bandwidth scrub rate + * to hardware according to SCRUBINTERVAL formula + * found in datasheet. + */ +static int set_sdram_scrub_rate(struct mem_ctl_info *mci, u32 new_bw) +{ + struct i7core_pvt *pvt = mci->pvt_info; + struct pci_dev *pdev; + u32 dw_scrub; + u32 dw_ssr; + + /* Get data from the MC register, function 2 */ + pdev = pvt->pci_mcr[2]; + if (!pdev) + return -ENODEV; + + pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &dw_scrub); + + if (new_bw == 0) { + /* Prepare to disable petrol scrub */ + dw_scrub &= ~STARTSCRUB; + /* Stop the patrol scrub engine */ + write_and_test(pdev, MC_SCRUB_CONTROL, + dw_scrub & ~SCRUBINTERVAL_MASK); + + /* Get current status of scrub rate and set bit to disable */ + pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr); + dw_ssr &= ~SSR_MODE_MASK; + dw_ssr |= SSR_MODE_DISABLE; + } else { + const int cache_line_size = 64; + const u32 freq_dclk_mhz = pvt->dclk_freq; + unsigned long long scrub_interval; + /* + * Translate the desired scrub rate to a register value and + * program the corresponding register value. + */ + scrub_interval = (unsigned long long)freq_dclk_mhz * + cache_line_size * 1000000; + do_div(scrub_interval, new_bw); + + if (!scrub_interval || scrub_interval > SCRUBINTERVAL_MASK) + return -EINVAL; + + dw_scrub = SCRUBINTERVAL_MASK & scrub_interval; + + /* Start the patrol scrub engine */ + pci_write_config_dword(pdev, MC_SCRUB_CONTROL, + STARTSCRUB | dw_scrub); + + /* Get current status of scrub rate and set bit to enable */ + pci_read_config_dword(pdev, MC_SSRCONTROL, &dw_ssr); + dw_ssr &= ~SSR_MODE_MASK; + dw_ssr |= SSR_MODE_ENABLE; + } + /* Disable or enable scrubbing */ + pci_write_config_dword(pdev, MC_SSRCONTROL, dw_ssr); + + return new_bw; +} + +/* + * get_sdram_scrub_rate This routine convert current scrub rate value + * into byte/sec bandwidth accourding to + * SCRUBINTERVAL formula found in datasheet. + */ +static int get_sdram_scrub_rate(struct mem_ctl_info *mci) +{ + struct i7core_pvt *pvt = mci->pvt_info; + struct pci_dev *pdev; + const u32 cache_line_size = 64; + const u32 freq_dclk_mhz = pvt->dclk_freq; + unsigned long long scrub_rate; + u32 scrubval; + + /* Get data from the MC register, function 2 */ + pdev = pvt->pci_mcr[2]; + if (!pdev) + return -ENODEV; + + /* Get current scrub control data */ + pci_read_config_dword(pdev, MC_SCRUB_CONTROL, &scrubval); + + /* Mask highest 8-bits to 0 */ + scrubval &= SCRUBINTERVAL_MASK; + if (!scrubval) + return 0; + + /* Calculate scrub rate value into byte/sec bandwidth */ + scrub_rate = (unsigned long long)freq_dclk_mhz * + 1000000 * cache_line_size; + do_div(scrub_rate, scrubval); + return (int)scrub_rate; +} + +static void enable_sdram_scrub_setting(struct mem_ctl_info *mci) +{ + struct i7core_pvt *pvt = mci->pvt_info; + u32 pci_lock; + + /* Unlock writes to pci registers */ + pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock); + pci_lock &= ~0x3; + pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, + pci_lock | MC_CFG_UNLOCK); + + mci->set_sdram_scrub_rate = set_sdram_scrub_rate; + mci->get_sdram_scrub_rate = get_sdram_scrub_rate; +} + +static void disable_sdram_scrub_setting(struct mem_ctl_info *mci) +{ + struct i7core_pvt *pvt = mci->pvt_info; + u32 pci_lock; + + /* Lock writes to pci registers */ + pci_read_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, &pci_lock); + pci_lock &= ~0x3; + pci_write_config_dword(pvt->pci_noncore, MC_CFG_CONTROL, + pci_lock | MC_CFG_LOCK); } static void i7core_pci_ctl_create(struct i7core_pvt *pvt) @@ -1874,7 +2197,8 @@ static void i7core_pci_ctl_create(struct i7core_pvt *pvt) &pvt->i7core_dev->pdev[0]->dev, EDAC_MOD_STR); if (unlikely(!pvt->i7core_pci)) - pr_warn("Unable to setup PCI error report via EDAC\n"); + i7core_printk(KERN_WARNING, + "Unable to setup PCI error report via EDAC\n"); } static void i7core_pci_ctl_release(struct i7core_pvt *pvt) @@ -1906,8 +2230,11 @@ static void i7core_unregister_mci(struct i7core_dev *i7core_dev) debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n", __func__, mci, &i7core_dev->pdev[0]->dev); - /* Disable MCE NMI handler */ - edac_mce_unregister(&pvt->edac_mce); + /* Disable scrubrate setting */ + if (pvt->enable_scrub) + disable_sdram_scrub_setting(mci); + + atomic_notifier_chain_unregister(&x86_mce_decoder_chain, &i7_mce_dec); /* Disable EDAC polling */ i7core_pci_ctl_release(pvt); @@ -1979,6 +2306,10 @@ static int i7core_register_mci(struct i7core_dev *i7core_dev) /* Set the function pointer to an actual operation function */ mci->edac_check = i7core_check_error; + /* Enable scrubrate setting */ + if (pvt->enable_scrub) + enable_sdram_scrub_setting(mci); + /* add this new MC control structure to EDAC's list of MCs */ if (unlikely(edac_mc_add_mc(mci))) { debugf0("MC: " __FILE__ @@ -2002,21 +2333,13 @@ static int i7core_register_mci(struct i7core_dev *i7core_dev) /* allocating generic PCI control info */ i7core_pci_ctl_create(pvt); - /* Registers on edac_mce in order to receive memory errors */ - pvt->edac_mce.priv = mci; - pvt->edac_mce.check_error = i7core_mce_check_error; - rc = edac_mce_register(&pvt->edac_mce); - if (unlikely(rc < 0)) { - debugf0("MC: " __FILE__ - ": %s(): failed edac_mce_register()\n", __func__); - goto fail1; - } + /* DCLK for scrub rate setting */ + pvt->dclk_freq = get_dclk_freq(); + + atomic_notifier_chain_register(&x86_mce_decoder_chain, &i7_mce_dec); return 0; -fail1: - i7core_pci_ctl_release(pvt); - edac_mc_del_mc(mci->dev); fail0: kfree(mci->ctl_name); edac_mc_free(mci); @@ -2035,7 +2358,7 @@ fail0: static int __devinit i7core_probe(struct pci_dev *pdev, const struct pci_device_id *id) { - int rc; + int rc, count = 0; struct i7core_dev *i7core_dev; /* get the pci devices we want to reserve for our use */ @@ -2055,12 +2378,28 @@ static int __devinit i7core_probe(struct pci_dev *pdev, goto fail0; list_for_each_entry(i7core_dev, &i7core_edac_list, list) { + count++; rc = i7core_register_mci(i7core_dev); if (unlikely(rc < 0)) goto fail1; } - i7core_printk(KERN_INFO, "Driver loaded.\n"); + /* + * Nehalem-EX uses a different memory controller. However, as the + * memory controller is not visible on some Nehalem/Nehalem-EP, we + * need to indirectly probe via a X58 PCI device. The same devices + * are found on (some) Nehalem-EX. So, on those machines, the + * probe routine needs to return -ENODEV, as the actual Memory + * Controller registers won't be detected. + */ + if (!count) { + rc = -ENODEV; + goto fail1; + } + + i7core_printk(KERN_INFO, + "Driver loaded, %d memory controller(s) found.\n", + count); mutex_unlock(&i7core_edac_lock); return 0; diff --git a/drivers/edac/sb_edac.c b/drivers/edac/sb_edac.c new file mode 100644 index 0000000..7a402bf --- /dev/null +++ b/drivers/edac/sb_edac.c @@ -0,0 +1,1893 @@ +/* Intel Sandy Bridge -EN/-EP/-EX Memory Controller kernel module + * + * This driver supports the memory controllers found on the Intel + * processor family Sandy Bridge. + * + * This file may be distributed under the terms of the + * GNU General Public License version 2 only. + * + * Copyright (c) 2011 by: + * Mauro Carvalho Chehab <mchehab@redhat.com> + */ + +#include <linux/module.h> +#include <linux/init.h> +#include <linux/pci.h> +#include <linux/pci_ids.h> +#include <linux/slab.h> +#include <linux/delay.h> +#include <linux/edac.h> +#include <linux/mmzone.h> +#include <linux/smp.h> +#include <linux/bitmap.h> +#include <asm/processor.h> +#include <asm/mce.h> + +#include "edac_core.h" + +/* Static vars */ +static LIST_HEAD(sbridge_edac_list); +static DEFINE_MUTEX(sbridge_edac_lock); +static int probed; + +/* + * Alter this version for the module when modifications are made + */ +#define SBRIDGE_REVISION " Ver: 1.0.0 " +#define EDAC_MOD_STR "sbridge_edac" + +/* + * Debug macros + */ +#define sbridge_printk(level, fmt, arg...) \ + edac_printk(level, "sbridge", fmt, ##arg) + +#define sbridge_mc_printk(mci, level, fmt, arg...) \ + edac_mc_chipset_printk(mci, level, "sbridge", fmt, ##arg) + +/* + * Get a bit field at register value <v>, from bit <lo> to bit <hi> + */ +#define GET_BITFIELD(v, lo, hi) \ + (((v) & ((1ULL << ((hi) - (lo) + 1)) - 1) << (lo)) >> (lo)) + +/* + * sbridge Memory Controller Registers + */ + +/* + * FIXME: For now, let's order by device function, as it makes + * easier for driver's development proccess. This table should be + * moved to pci_id.h when submitted upstream + */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0 0x3cf4 /* 12.6 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1 0x3cf6 /* 12.7 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_BR 0x3cf5 /* 13.6 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0 0x3ca0 /* 14.0 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA 0x3ca8 /* 15.0 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS 0x3c71 /* 15.1 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0 0x3caa /* 15.2 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1 0x3cab /* 15.3 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2 0x3cac /* 15.4 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3 0x3cad /* 15.5 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO 0x3cb8 /* 17.0 */ + + /* + * Currently, unused, but will be needed in the future + * implementations, as they hold the error counters + */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR0 0x3c72 /* 16.2 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR1 0x3c73 /* 16.3 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR2 0x3c76 /* 16.6 */ +#define PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_ERR3 0x3c77 /* 16.7 */ + +/* Devices 12 Function 6, Offsets 0x80 to 0xcc */ +static const u32 dram_rule[] = { + 0x80, 0x88, 0x90, 0x98, 0xa0, + 0xa8, 0xb0, 0xb8, 0xc0, 0xc8, +}; +#define MAX_SAD ARRAY_SIZE(dram_rule) + +#define SAD_LIMIT(reg) ((GET_BITFIELD(reg, 6, 25) << 26) | 0x3ffffff) +#define DRAM_ATTR(reg) GET_BITFIELD(reg, 2, 3) +#define INTERLEAVE_MODE(reg) GET_BITFIELD(reg, 1, 1) +#define DRAM_RULE_ENABLE(reg) GET_BITFIELD(reg, 0, 0) + +static char *get_dram_attr(u32 reg) +{ + switch(DRAM_ATTR(reg)) { + case 0: + return "DRAM"; + case 1: + return "MMCFG"; + case 2: + return "NXM"; + default: + return "unknown"; + } +} + +static const u32 interleave_list[] = { + 0x84, 0x8c, 0x94, 0x9c, 0xa4, + 0xac, 0xb4, 0xbc, 0xc4, 0xcc, +}; +#define MAX_INTERLEAVE ARRAY_SIZE(interleave_list) + +#define SAD_PKG0(reg) GET_BITFIELD(reg, 0, 2) +#define SAD_PKG1(reg) GET_BITFIELD(reg, 3, 5) +#define SAD_PKG2(reg) GET_BITFIELD(reg, 8, 10) +#define SAD_PKG3(reg) GET_BITFIELD(reg, 11, 13) +#define SAD_PKG4(reg) GET_BITFIELD(reg, 16, 18) +#define SAD_PKG5(reg) GET_BITFIELD(reg, 19, 21) +#define SAD_PKG6(reg) GET_BITFIELD(reg, 24, 26) +#define SAD_PKG7(reg) GET_BITFIELD(reg, 27, 29) + +static inline int sad_pkg(u32 reg, int interleave) +{ + switch (interleave) { + case 0: + return SAD_PKG0(reg); + case 1: + return SAD_PKG1(reg); + case 2: + return SAD_PKG2(reg); + case 3: + return SAD_PKG3(reg); + case 4: + return SAD_PKG4(reg); + case 5: + return SAD_PKG5(reg); + case 6: + return SAD_PKG6(reg); + case 7: + return SAD_PKG7(reg); + default: + return -EINVAL; + } +} + +/* Devices 12 Function 7 */ + +#define TOLM 0x80 +#define TOHM 0x84 + +#define GET_TOLM(reg) ((GET_BITFIELD(reg, 0, 3) << 28) | 0x3ffffff) +#define GET_TOHM(reg) ((GET_BITFIELD(reg, 0, 20) << 25) | 0x3ffffff) + +/* Device 13 Function 6 */ + +#define SAD_TARGET 0xf0 + +#define SOURCE_ID(reg) GET_BITFIELD(reg, 9, 11) + +#define SAD_CONTROL 0xf4 + +#define NODE_ID(reg) GET_BITFIELD(reg, 0, 2) + +/* Device 14 function 0 */ + +static const u32 tad_dram_rule[] = { + 0x40, 0x44, 0x48, 0x4c, + 0x50, 0x54, 0x58, 0x5c, + 0x60, 0x64, 0x68, 0x6c, +}; +#define MAX_TAD ARRAY_SIZE(tad_dram_rule) + +#define TAD_LIMIT(reg) ((GET_BITFIELD(reg, 12, 31) << 26) | 0x3ffffff) +#define TAD_SOCK(reg) GET_BITFIELD(reg, 10, 11) +#define TAD_CH(reg) GET_BITFIELD(reg, 8, 9) +#define TAD_TGT3(reg) GET_BITFIELD(reg, 6, 7) +#define TAD_TGT2(reg) GET_BITFIELD(reg, 4, 5) +#define TAD_TGT1(reg) GET_BITFIELD(reg, 2, 3) +#define TAD_TGT0(reg) GET_BITFIELD(reg, 0, 1) + +/* Device 15, function 0 */ + +#define MCMTR 0x7c + +#define IS_ECC_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 2, 2) +#define IS_LOCKSTEP_ENABLED(mcmtr) GET_BITFIELD(mcmtr, 1, 1) +#define IS_CLOSE_PG(mcmtr) GET_BITFIELD(mcmtr, 0, 0) + +/* Device 15, function 1 */ + +#define RASENABLES 0xac +#define IS_MIRROR_ENABLED(reg) GET_BITFIELD(reg, 0, 0) + +/* Device 15, functions 2-5 */ + +static const int mtr_regs[] = { + 0x80, 0x84, 0x88, +}; + +#define RANK_DISABLE(mtr) GET_BITFIELD(mtr, 16, 19) +#define IS_DIMM_PRESENT(mtr) GET_BITFIELD(mtr, 14, 14) +#define RANK_CNT_BITS(mtr) GET_BITFIELD(mtr, 12, 13) +#define RANK_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 2, 4) +#define COL_WIDTH_BITS(mtr) GET_BITFIELD(mtr, 0, 1) + +static const u32 tad_ch_nilv_offset[] = { + 0x90, 0x94, 0x98, 0x9c, + 0xa0, 0xa4, 0xa8, 0xac, + 0xb0, 0xb4, 0xb8, 0xbc, +}; +#define CHN_IDX_OFFSET(reg) GET_BITFIELD(reg, 28, 29) +#define TAD_OFFSET(reg) (GET_BITFIELD(reg, 6, 25) << 26) + +static const u32 rir_way_limit[] = { + 0x108, 0x10c, 0x110, 0x114, 0x118, +}; +#define MAX_RIR_RANGES ARRAY_SIZE(rir_way_limit) + +#define IS_RIR_VALID(reg) GET_BITFIELD(reg, 31, 31) +#define RIR_WAY(reg) GET_BITFIELD(reg, 28, 29) +#define RIR_LIMIT(reg) ((GET_BITFIELD(reg, 1, 10) << 29)| 0x1fffffff) + +#define MAX_RIR_WAY 8 + +static const u32 rir_offset[MAX_RIR_RANGES][MAX_RIR_WAY] = { + { 0x120, 0x124, 0x128, 0x12c, 0x130, 0x134, 0x138, 0x13c }, + { 0x140, 0x144, 0x148, 0x14c, 0x150, 0x154, 0x158, 0x15c }, + { 0x160, 0x164, 0x168, 0x16c, 0x170, 0x174, 0x178, 0x17c }, + { 0x180, 0x184, 0x188, 0x18c, 0x190, 0x194, 0x198, 0x19c }, + { 0x1a0, 0x1a4, 0x1a8, 0x1ac, 0x1b0, 0x1b4, 0x1b8, 0x1bc }, +}; + +#define RIR_RNK_TGT(reg) GET_BITFIELD(reg, 16, 19) +#define RIR_OFFSET(reg) GET_BITFIELD(reg, 2, 14) + +/* Device 16, functions 2-7 */ + +/* + * FIXME: Implement the error count reads directly + */ + +static const u32 correrrcnt[] = { + 0x104, 0x108, 0x10c, 0x110, +}; + +#define RANK_ODD_OV(reg) GET_BITFIELD(reg, 31, 31) +#define RANK_ODD_ERR_CNT(reg) GET_BITFIELD(reg, 16, 30) +#define RANK_EVEN_OV(reg) GET_BITFIELD(reg, 15, 15) +#define RANK_EVEN_ERR_CNT(reg) GET_BITFIELD(reg, 0, 14) + +static const u32 correrrthrsld[] = { + 0x11c, 0x120, 0x124, 0x128, +}; + +#define RANK_ODD_ERR_THRSLD(reg) GET_BITFIELD(reg, 16, 30) +#define RANK_EVEN_ERR_THRSLD(reg) GET_BITFIELD(reg, 0, 14) + + +/* Device 17, function 0 */ + +#define RANK_CFG_A 0x0328 + +#define IS_RDIMM_ENABLED(reg) GET_BITFIELD(reg, 11, 11) + +/* + * sbridge structs + */ + +#define NUM_CHANNELS 4 +#define MAX_DIMMS 3 /* Max DIMMS per channel */ + +struct sbridge_info { + u32 mcmtr; +}; + +struct sbridge_channel { + u32 ranks; + u32 dimms; +}; + +struct pci_id_descr { + int dev; + int func; + int dev_id; + int optional; +}; + +struct pci_id_table { + const struct pci_id_descr *descr; + int n_devs; +}; + +struct sbridge_dev { + struct list_head list; + u8 bus, mc; + u8 node_id, source_id; + struct pci_dev **pdev; + int n_devs; + struct mem_ctl_info *mci; +}; + +struct sbridge_pvt { + struct pci_dev *pci_ta, *pci_ddrio, *pci_ras; + struct pci_dev *pci_sad0, *pci_sad1, *pci_ha0; + struct pci_dev *pci_br; + struct pci_dev *pci_tad[NUM_CHANNELS]; + + struct sbridge_dev *sbridge_dev; + + struct sbridge_info info; + struct sbridge_channel channel[NUM_CHANNELS]; + + int csrow_map[NUM_CHANNELS][MAX_DIMMS]; + + /* Memory type detection */ + bool is_mirrored, is_lockstep, is_close_pg; + + /* Fifo double buffers */ + struct mce mce_entry[MCE_LOG_LEN]; + struct mce mce_outentry[MCE_LOG_LEN]; + + /* Fifo in/out counters */ + unsigned mce_in, mce_out; + + /* Count indicator to show errors not got */ + unsigned mce_overrun; + + /* Memory description */ + u64 tolm, tohm; +}; + +#define PCI_DESCR(device, function, device_id) \ + .dev = (device), \ + .func = (function), \ + .dev_id = (device_id) + +static const struct pci_id_descr pci_dev_descr_sbridge[] = { + /* Processor Home Agent */ + { PCI_DESCR(14, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_HA0) }, + + /* Memory controller */ + { PCI_DESCR(15, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA) }, + { PCI_DESCR(15, 1, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_RAS) }, + { PCI_DESCR(15, 2, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD0) }, + { PCI_DESCR(15, 3, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD1) }, + { PCI_DESCR(15, 4, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD2) }, + { PCI_DESCR(15, 5, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TAD3) }, + { PCI_DESCR(17, 0, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_DDRIO) }, + + /* System Address Decoder */ + { PCI_DESCR(12, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD0) }, + { PCI_DESCR(12, 7, PCI_DEVICE_ID_INTEL_SBRIDGE_SAD1) }, + + /* Broadcast Registers */ + { PCI_DESCR(13, 6, PCI_DEVICE_ID_INTEL_SBRIDGE_BR) }, +}; + +#define PCI_ID_TABLE_ENTRY(A) { .descr=A, .n_devs = ARRAY_SIZE(A) } +static const struct pci_id_table pci_dev_descr_sbridge_table[] = { + PCI_ID_TABLE_ENTRY(pci_dev_descr_sbridge), + {0,} /* 0 terminated list. */ +}; + +/* + * pci_device_id table for which devices we are looking for + */ +static const struct pci_device_id sbridge_pci_tbl[] __devinitdata = { + {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SBRIDGE_IMC_TA)}, + {0,} /* 0 terminated list. */ +}; + + +/**************************************************************************** + Anciliary status routines + ****************************************************************************/ + +static inline int numrank(u32 mtr) +{ + int ranks = (1 << RANK_CNT_BITS(mtr)); + + if (ranks > 4) { + debugf0("Invalid number of ranks: %d (max = 4) raw value = %x (%04x)", + ranks, (unsigned int)RANK_CNT_BITS(mtr), mtr); + return -EINVAL; + } + + return ranks; +} + +static inline int numrow(u32 mtr) +{ + int rows = (RANK_WIDTH_BITS(mtr) + 12); + + if (rows < 13 || rows > 18) { + debugf0("Invalid number of rows: %d (should be between 14 and 17) raw value = %x (%04x)", + rows, (unsigned int)RANK_WIDTH_BITS(mtr), mtr); + return -EINVAL; + } + + return 1 << rows; +} + +static inline int numcol(u32 mtr) +{ + int cols = (COL_WIDTH_BITS(mtr) + 10); + + if (cols > 12) { + debugf0("Invalid number of cols: %d (max = 4) raw value = %x (%04x)", + cols, (unsigned int)COL_WIDTH_BITS(mtr), mtr); + return -EINVAL; + } + + return 1 << cols; +} + +static struct sbridge_dev *get_sbridge_dev(u8 bus) +{ + struct sbridge_dev *sbridge_dev; + + list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { + if (sbridge_dev->bus == bus) + return sbridge_dev; + } + + return NULL; +} + +static struct sbridge_dev *alloc_sbridge_dev(u8 bus, + const struct pci_id_table *table) +{ + struct sbridge_dev *sbridge_dev; + + sbridge_dev = kzalloc(sizeof(*sbridge_dev), GFP_KERNEL); + if (!sbridge_dev) + return NULL; + + sbridge_dev->pdev = kzalloc(sizeof(*sbridge_dev->pdev) * table->n_devs, + GFP_KERNEL); + if (!sbridge_dev->pdev) { + kfree(sbridge_dev); + return NULL; + } + + sbridge_dev->bus = bus; + sbridge_dev->n_devs = table->n_devs; + list_add_tail(&sbridge_dev->list, &sbridge_edac_list); + + return sbridge_dev; +} + +static void free_sbridge_dev(struct sbridge_dev *sbridge_dev) +{ + list_del(&sbridge_dev->list); + kfree(sbridge_dev->pdev); + kfree(sbridge_dev); +} + +/**************************************************************************** + Memory check routines + ****************************************************************************/ +static struct pci_dev *get_pdev_slot_func(u8 bus, unsigned slot, + unsigned func) +{ + struct sbridge_dev *sbridge_dev = get_sbridge_dev(bus); + int i; + + if (!sbridge_dev) + return NULL; + + for (i = 0; i < sbridge_dev->n_devs; i++) { + if (!sbridge_dev->pdev[i]) + continue; + + if (PCI_SLOT(sbridge_dev->pdev[i]->devfn) == slot && + PCI_FUNC(sbridge_dev->pdev[i]->devfn) == func) { + debugf1("Associated %02x.%02x.%d with %p\n", + bus, slot, func, sbridge_dev->pdev[i]); + return sbridge_dev->pdev[i]; + } + } + + return NULL; +} + +/** + * sbridge_get_active_channels() - gets the number of channels and csrows + * bus: Device bus + * @channels: Number of channels that will be returned + * @csrows: Number of csrows found + * + * Since EDAC core needs to know in advance the number of available channels + * and csrows, in order to allocate memory for csrows/channels, it is needed + * to run two similar steps. At the first step, implemented on this function, + * it checks the number of csrows/channels present at one socket, identified + * by the associated PCI bus. + * this is used in order to properly allocate the size of mci components. + * Note: one csrow is one dimm. + */ +static int sbridge_get_active_channels(const u8 bus, unsigned *channels, + unsigned *csrows) +{ + struct pci_dev *pdev = NULL; + int i, j; + u32 mcmtr; + + *channels = 0; + *csrows = 0; + + pdev = get_pdev_slot_func(bus, 15, 0); + if (!pdev) { + sbridge_printk(KERN_ERR, "Couldn't find PCI device " + "%2x.%02d.%d!!!\n", + bus, 15, 0); + return -ENODEV; + } + + pci_read_config_dword(pdev, MCMTR, &mcmtr); + if (!IS_ECC_ENABLED(mcmtr)) { + sbridge_printk(KERN_ERR, "ECC is disabled. Aborting\n"); + return -ENODEV; + } + + for (i = 0; i < NUM_CHANNELS; i++) { + u32 mtr; + + /* Device 15 functions 2 - 5 */ + pdev = get_pdev_slot_func(bus, 15, 2 + i); + if (!pdev) { + sbridge_printk(KERN_ERR, "Couldn't find PCI device " + "%2x.%02d.%d!!!\n", + bus, 15, 2 + i); + return -ENODEV; + } + (*channels)++; + + for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) { + pci_read_config_dword(pdev, mtr_regs[j], &mtr); + debugf1("Bus#%02x channel #%d MTR%d = %x\n", bus, i, j, mtr); + if (IS_DIMM_PRESENT(mtr)) + (*csrows)++; + } + } + + debugf0("Number of active channels: %d, number of active dimms: %d\n", + *channels, *csrows); + + return 0; +} + +static int get_dimm_config(const struct mem_ctl_info *mci) +{ + struct sbridge_pvt *pvt = mci->pvt_info; + struct csrow_info *csr; + int i, j, banks, ranks, rows, cols, size, npages; + int csrow = 0; + unsigned long last_page = 0; + u32 reg; + enum edac_type mode; + enum mem_type mtype; + + pci_read_config_dword(pvt->pci_br, SAD_TARGET, ®); + pvt->sbridge_dev->source_id = SOURCE_ID(reg); + + pci_read_config_dword(pvt->pci_br, SAD_CONTROL, ®); + pvt->sbridge_dev->node_id = NODE_ID(reg); + debugf0("mc#%d: Node ID: %d, source ID: %d\n", + pvt->sbridge_dev->mc, + pvt->sbridge_dev->node_id, + pvt->sbridge_dev->source_id); + + pci_read_config_dword(pvt->pci_ras, RASENABLES, ®); + if (IS_MIRROR_ENABLED(reg)) { + debugf0("Memory mirror is enabled\n"); + pvt->is_mirrored = true; + } else { + debugf0("Memory mirror is disabled\n"); + pvt->is_mirrored = false; + } + + pci_read_config_dword(pvt->pci_ta, MCMTR, &pvt->info.mcmtr); + if (IS_LOCKSTEP_ENABLED(pvt->info.mcmtr)) { + debugf0("Lockstep is enabled\n"); + mode = EDAC_S8ECD8ED; + pvt->is_lockstep = true; + } else { + debugf0("Lockstep is disabled\n"); + mode = EDAC_S4ECD4ED; + pvt->is_lockstep = false; + } + if (IS_CLOSE_PG(pvt->info.mcmtr)) { + debugf0("address map is on closed page mode\n"); + pvt->is_close_pg = true; + } else { + debugf0("address map is on open page mode\n"); + pvt->is_close_pg = false; + } + + pci_read_config_dword(pvt->pci_ta, RANK_CFG_A, ®); + if (IS_RDIMM_ENABLED(reg)) { + /* FIXME: Can also be LRDIMM */ + debugf0("Memory is registered\n"); + mtype = MEM_RDDR3; + } else { + debugf0("Memory is unregistered\n"); + mtype = MEM_DDR3; + } + + /* On all supported DDR3 DIMM types, there are 8 banks available */ + banks = 8; + + for (i = 0; i < NUM_CHANNELS; i++) { + u32 mtr; + + for (j = 0; j < ARRAY_SIZE(mtr_regs); j++) { + pci_read_config_dword(pvt->pci_tad[i], + mtr_regs[j], &mtr); + debugf4("Channel #%d MTR%d = %x\n", i, j, mtr); + if (IS_DIMM_PRESENT(mtr)) { + pvt->channel[i].dimms++; + + ranks = numrank(mtr); + rows = numrow(mtr); + cols = numcol(mtr); + + /* DDR3 has 8 I/O banks */ + size = (rows * cols * banks * ranks) >> (20 - 3); + npages = MiB_TO_PAGES(size); + + debugf0("mc#%d: channel %d, dimm %d, %d Mb (%d pages) bank: %d, rank: %d, row: %#x, col: %#x\n", + pvt->sbridge_dev->mc, i, j, + size, npages, + banks, ranks, rows, cols); + csr = &mci->csrows[csrow]; + + csr->first_page = last_page; + csr->last_page = last_page + npages - 1; + csr->page_mask = 0UL; /* Unused */ + csr->nr_pages = npages; + csr->grain = 32; + csr->csrow_idx = csrow; + csr->dtype = (banks == 8) ? DEV_X8 : DEV_X4; + csr->ce_count = 0; + csr->ue_count = 0; + csr->mtype = mtype; + csr->edac_mode = mode; + csr->nr_channels = 1; + csr->channels[0].chan_idx = i; + csr->channels[0].ce_count = 0; + pvt->csrow_map[i][j] = csrow; + snprintf(csr->channels[0].label, + sizeof(csr->channels[0].label), + "CPU_SrcID#%u_Channel#%u_DIMM#%u", + pvt->sbridge_dev->source_id, i, j); + last_page += npages; + csrow++; + } + } + } + + return 0; +} + +static void get_memory_layout(const struct mem_ctl_info *mci) +{ + struct sbridge_pvt *pvt = mci->pvt_info; + int i, j, k, n_sads, n_tads, sad_interl; + u32 reg; + u64 limit, prv = 0; + u64 tmp_mb; + u32 rir_way; + + /* + * Step 1) Get TOLM/TOHM ranges + */ + + /* Address range is 32:28 */ + pci_read_config_dword(pvt->pci_sad1, TOLM, + ®); + pvt->tolm = GET_TOLM(reg); + tmp_mb = (1 + pvt->tolm) >> 20; + + debugf0("TOLM: %Lu.%03Lu GB (0x%016Lx)\n", + tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tolm); + + /* Address range is already 45:25 */ + pci_read_config_dword(pvt->pci_sad1, TOHM, + ®); + pvt->tohm = GET_TOHM(reg); + tmp_mb = (1 + pvt->tohm) >> 20; + + debugf0("TOHM: %Lu.%03Lu GB (0x%016Lx)", + tmp_mb / 1000, tmp_mb % 1000, (u64)pvt->tohm); + + /* + * Step 2) Get SAD range and SAD Interleave list + * TAD registers contain the interleave wayness. However, it + * seems simpler to just discover it indirectly, with the + * algorithm bellow. + */ + prv = 0; + for (n_sads = 0; n_sads < MAX_SAD; n_sads++) { + /* SAD_LIMIT Address range is 45:26 */ + pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads], + ®); + limit = SAD_LIMIT(reg); + + if (!DRAM_RULE_ENABLE(reg)) + continue; + + if (limit <= prv) + break; + + tmp_mb = (limit + 1) >> 20; + debugf0("SAD#%d %s up to %Lu.%03Lu GB (0x%016Lx) %s reg=0x%08x\n", + n_sads, + get_dram_attr(reg), + tmp_mb / 1000, tmp_mb % 1000, + ((u64)tmp_mb) << 20L, + INTERLEAVE_MODE(reg) ? "Interleave: 8:6" : "Interleave: [8:6]XOR[18:16]", + reg); + prv = limit; + + pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads], + ®); + sad_interl = sad_pkg(reg, 0); + for (j = 0; j < 8; j++) { + if (j > 0 && sad_interl == sad_pkg(reg, j)) + break; + + debugf0("SAD#%d, interleave #%d: %d\n", + n_sads, j, sad_pkg(reg, j)); + } + } + + /* + * Step 3) Get TAD range + */ + prv = 0; + for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { + pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads], + ®); + limit = TAD_LIMIT(reg); + if (limit <= prv) + break; + tmp_mb = (limit + 1) >> 20; + + debugf0("TAD#%d: up to %Lu.%03Lu GB (0x%016Lx), socket interleave %d, memory interleave %d, TGT: %d, %d, %d, %d, reg=0x%08x\n", + n_tads, tmp_mb / 1000, tmp_mb % 1000, + ((u64)tmp_mb) << 20L, + (u32)TAD_SOCK(reg), + (u32)TAD_CH(reg), + (u32)TAD_TGT0(reg), + (u32)TAD_TGT1(reg), + (u32)TAD_TGT2(reg), + (u32)TAD_TGT3(reg), + reg); + prv = tmp_mb; + } + + /* + * Step 4) Get TAD offsets, per each channel + */ + for (i = 0; i < NUM_CHANNELS; i++) { + if (!pvt->channel[i].dimms) + continue; + for (j = 0; j < n_tads; j++) { + pci_read_config_dword(pvt->pci_tad[i], + tad_ch_nilv_offset[j], + ®); + tmp_mb = TAD_OFFSET(reg) >> 20; + debugf0("TAD CH#%d, offset #%d: %Lu.%03Lu GB (0x%016Lx), reg=0x%08x\n", + i, j, + tmp_mb / 1000, tmp_mb % 1000, + ((u64)tmp_mb) << 20L, + reg); + } + } + + /* + * Step 6) Get RIR Wayness/Limit, per each channel + */ + for (i = 0; i < NUM_CHANNELS; i++) { + if (!pvt->channel[i].dimms) + continue; + for (j = 0; j < MAX_RIR_RANGES; j++) { + pci_read_config_dword(pvt->pci_tad[i], + rir_way_limit[j], + ®); + + if (!IS_RIR_VALID(reg)) + continue; + + tmp_mb = RIR_LIMIT(reg) >> 20; + rir_way = 1 << RIR_WAY(reg); + debugf0("CH#%d RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d, reg=0x%08x\n", + i, j, + tmp_mb / 1000, tmp_mb % 1000, + ((u64)tmp_mb) << 20L, + rir_way, + reg); + + for (k = 0; k < rir_way; k++) { + pci_read_config_dword(pvt->pci_tad[i], + rir_offset[j][k], + ®); + tmp_mb = RIR_OFFSET(reg) << 6; + + debugf0("CH#%d RIR#%d INTL#%d, offset %Lu.%03Lu GB (0x%016Lx), tgt: %d, reg=0x%08x\n", + i, j, k, + tmp_mb / 1000, tmp_mb % 1000, + ((u64)tmp_mb) << 20L, + (u32)RIR_RNK_TGT(reg), + reg); + } + } + } +} + +struct mem_ctl_info *get_mci_for_node_id(u8 node_id) +{ + struct sbridge_dev *sbridge_dev; + + list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { + if (sbridge_dev->node_id == node_id) + return sbridge_dev->mci; + } + return NULL; +} + +static int get_memory_error_data(struct mem_ctl_info *mci, + u64 addr, + u8 *socket, + long *channel_mask, + u8 *rank, + char *area_type) +{ + struct mem_ctl_info *new_mci; + struct sbridge_pvt *pvt = mci->pvt_info; + char msg[256]; + int n_rir, n_sads, n_tads, sad_way, sck_xch; + int sad_interl, idx, base_ch; + int interleave_mode; + unsigned sad_interleave[MAX_INTERLEAVE]; + u32 reg; + u8 ch_way,sck_way; + u32 tad_offset; + u32 rir_way; + u64 ch_addr, offset, limit, prv = 0; + + + /* + * Step 0) Check if the address is at special memory ranges + * The check bellow is probably enough to fill all cases where + * the error is not inside a memory, except for the legacy + * range (e. g. VGA addresses). It is unlikely, however, that the + * memory controller would generate an error on that range. + */ + if ((addr > (u64) pvt->tolm) && (addr < (1L << 32))) { + sprintf(msg, "Error at TOLM area, on addr 0x%08Lx", addr); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + if (addr >= (u64)pvt->tohm) { + sprintf(msg, "Error at MMIOH area, on addr 0x%016Lx", addr); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + + /* + * Step 1) Get socket + */ + for (n_sads = 0; n_sads < MAX_SAD; n_sads++) { + pci_read_config_dword(pvt->pci_sad0, dram_rule[n_sads], + ®); + + if (!DRAM_RULE_ENABLE(reg)) + continue; + + limit = SAD_LIMIT(reg); + if (limit <= prv) { + sprintf(msg, "Can't discover the memory socket"); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + if (addr <= limit) + break; + prv = limit; + } + if (n_sads == MAX_SAD) { + sprintf(msg, "Can't discover the memory socket"); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + area_type = get_dram_attr(reg); + interleave_mode = INTERLEAVE_MODE(reg); + + pci_read_config_dword(pvt->pci_sad0, interleave_list[n_sads], + ®); + sad_interl = sad_pkg(reg, 0); + for (sad_way = 0; sad_way < 8; sad_way++) { + if (sad_way > 0 && sad_interl == sad_pkg(reg, sad_way)) + break; + sad_interleave[sad_way] = sad_pkg(reg, sad_way); + debugf0("SAD interleave #%d: %d\n", + sad_way, sad_interleave[sad_way]); + } + debugf0("mc#%d: Error detected on SAD#%d: address 0x%016Lx < 0x%016Lx, Interleave [%d:6]%s\n", + pvt->sbridge_dev->mc, + n_sads, + addr, + limit, + sad_way + 7, + INTERLEAVE_MODE(reg) ? "" : "XOR[18:16]"); + if (interleave_mode) + idx = ((addr >> 6) ^ (addr >> 16)) & 7; + else + idx = (addr >> 6) & 7; + switch (sad_way) { + case 1: + idx = 0; + break; + case 2: + idx = idx & 1; + break; + case 4: + idx = idx & 3; + break; + case 8: + break; + default: + sprintf(msg, "Can't discover socket interleave"); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + *socket = sad_interleave[idx]; + debugf0("SAD interleave index: %d (wayness %d) = CPU socket %d\n", + idx, sad_way, *socket); + + /* + * Move to the proper node structure, in order to access the + * right PCI registers + */ + new_mci = get_mci_for_node_id(*socket); + if (!new_mci) { + sprintf(msg, "Struct for socket #%u wasn't initialized", + *socket); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + mci = new_mci; + pvt = mci->pvt_info; + + /* + * Step 2) Get memory channel + */ + prv = 0; + for (n_tads = 0; n_tads < MAX_TAD; n_tads++) { + pci_read_config_dword(pvt->pci_ha0, tad_dram_rule[n_tads], + ®); + limit = TAD_LIMIT(reg); + if (limit <= prv) { + sprintf(msg, "Can't discover the memory channel"); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + if (addr <= limit) + break; + prv = limit; + } + ch_way = TAD_CH(reg) + 1; + sck_way = TAD_SOCK(reg) + 1; + /* + * FIXME: Is it right to always use channel 0 for offsets? + */ + pci_read_config_dword(pvt->pci_tad[0], + tad_ch_nilv_offset[n_tads], + &tad_offset); + + if (ch_way == 3) + idx = addr >> 6; + else + idx = addr >> (6 + sck_way); + idx = idx % ch_way; + + /* + * FIXME: Shouldn't we use CHN_IDX_OFFSET() here, when ch_way == 3 ??? + */ + switch (idx) { + case 0: + base_ch = TAD_TGT0(reg); + break; + case 1: + base_ch = TAD_TGT1(reg); + break; + case 2: + base_ch = TAD_TGT2(reg); + break; + case 3: + base_ch = TAD_TGT3(reg); + break; + default: + sprintf(msg, "Can't discover the TAD target"); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + *channel_mask = 1 << base_ch; + + if (pvt->is_mirrored) { + *channel_mask |= 1 << ((base_ch + 2) % 4); + switch(ch_way) { + case 2: + case 4: + sck_xch = 1 << sck_way * (ch_way >> 1); + break; + default: + sprintf(msg, "Invalid mirror set. Can't decode addr"); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + } else + sck_xch = (1 << sck_way) * ch_way; + + if (pvt->is_lockstep) + *channel_mask |= 1 << ((base_ch + 1) % 4); + + offset = TAD_OFFSET(tad_offset); + + debugf0("TAD#%d: address 0x%016Lx < 0x%016Lx, socket interleave %d, channel interleave %d (offset 0x%08Lx), index %d, base ch: %d, ch mask: 0x%02lx\n", + n_tads, + addr, + limit, + (u32)TAD_SOCK(reg), + ch_way, + offset, + idx, + base_ch, + *channel_mask); + + /* Calculate channel address */ + /* Remove the TAD offset */ + + if (offset > addr) { + sprintf(msg, "Can't calculate ch addr: TAD offset 0x%08Lx is too high for addr 0x%08Lx!", + offset, addr); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + addr -= offset; + /* Store the low bits [0:6] of the addr */ + ch_addr = addr & 0x7f; + /* Remove socket wayness and remove 6 bits */ + addr >>= 6; + addr /= sck_xch; +#if 0 + /* Divide by channel way */ + addr = addr / ch_way; +#endif + /* Recover the last 6 bits */ + ch_addr |= addr << 6; + + /* + * Step 3) Decode rank + */ + for (n_rir = 0; n_rir < MAX_RIR_RANGES; n_rir++) { + pci_read_config_dword(pvt->pci_tad[base_ch], + rir_way_limit[n_rir], + ®); + + if (!IS_RIR_VALID(reg)) + continue; + + limit = RIR_LIMIT(reg); + + debugf0("RIR#%d, limit: %Lu.%03Lu GB (0x%016Lx), way: %d\n", + n_rir, + (limit >> 20) / 1000, (limit >> 20) % 1000, + limit, + 1 << RIR_WAY(reg)); + if (ch_addr <= limit) + break; + } + if (n_rir == MAX_RIR_RANGES) { + sprintf(msg, "Can't discover the memory rank for ch addr 0x%08Lx", + ch_addr); + edac_mc_handle_ce_no_info(mci, msg); + return -EINVAL; + } + rir_way = RIR_WAY(reg); + if (pvt->is_close_pg) + idx = (ch_addr >> 6); + else + idx = (ch_addr >> 13); /* FIXME: Datasheet says to shift by 15 */ + idx %= 1 << rir_way; + + pci_read_config_dword(pvt->pci_tad[base_ch], + rir_offset[n_rir][idx], + ®); + *rank = RIR_RNK_TGT(reg); + + debugf0("RIR#%d: channel address 0x%08Lx < 0x%08Lx, RIR interleave %d, index %d\n", + n_rir, + ch_addr, + limit, + rir_way, + idx); + + return 0; +} + +/**************************************************************************** + Device initialization routines: put/get, init/exit + ****************************************************************************/ + +/* + * sbridge_put_all_devices 'put' all the devices that we have + * reserved via 'get' + */ +static void sbridge_put_devices(struct sbridge_dev *sbridge_dev) +{ + int i; + + debugf0(__FILE__ ": %s()\n", __func__); + for (i = 0; i < sbridge_dev->n_devs; i++) { + struct pci_dev *pdev = sbridge_dev->pdev[i]; + if (!pdev) + continue; + debugf0("Removing dev %02x:%02x.%d\n", + pdev->bus->number, + PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); + pci_dev_put(pdev); + } +} + +static void sbridge_put_all_devices(void) +{ + struct sbridge_dev *sbridge_dev, *tmp; + + list_for_each_entry_safe(sbridge_dev, tmp, &sbridge_edac_list, list) { + sbridge_put_devices(sbridge_dev); + free_sbridge_dev(sbridge_dev); + } +} + +/* + * sbridge_get_all_devices Find and perform 'get' operation on the MCH's + * device/functions we want to reference for this driver + * + * Need to 'get' device 16 func 1 and func 2 + */ +static int sbridge_get_onedevice(struct pci_dev **prev, + u8 *num_mc, + const struct pci_id_table *table, + const unsigned devno) +{ + struct sbridge_dev *sbridge_dev; + const struct pci_id_descr *dev_descr = &table->descr[devno]; + + struct pci_dev *pdev = NULL; + u8 bus = 0; + + sbridge_printk(KERN_INFO, + "Seeking for: dev %02x.%d PCI ID %04x:%04x\n", + dev_descr->dev, dev_descr->func, + PCI_VENDOR_ID_INTEL, dev_descr->dev_id); + + pdev = pci_get_device(PCI_VENDOR_ID_INTEL, + dev_descr->dev_id, *prev); + + if (!pdev) { + if (*prev) { + *prev = pdev; + return 0; + } + + if (dev_descr->optional) + return 0; + + if (devno == 0) + return -ENODEV; + + sbridge_printk(KERN_INFO, + "Device not found: dev %02x.%d PCI ID %04x:%04x\n", + dev_descr->dev, dev_descr->func, + PCI_VENDOR_ID_INTEL, dev_descr->dev_id); + + /* End of list, leave */ + return -ENODEV; + } + bus = pdev->bus->number; + + sbridge_dev = get_sbridge_dev(bus); + if (!sbridge_dev) { + sbridge_dev = alloc_sbridge_dev(bus, table); + if (!sbridge_dev) { + pci_dev_put(pdev); + return -ENOMEM; + } + (*num_mc)++; + } + + if (sbridge_dev->pdev[devno]) { + sbridge_printk(KERN_ERR, + "Duplicated device for " + "dev %02x:%d.%d PCI ID %04x:%04x\n", + bus, dev_descr->dev, dev_descr->func, + PCI_VENDOR_ID_INTEL, dev_descr->dev_id); + pci_dev_put(pdev); + return -ENODEV; + } + + sbridge_dev->pdev[devno] = pdev; + + /* Sanity check */ + if (unlikely(PCI_SLOT(pdev->devfn) != dev_descr->dev || + PCI_FUNC(pdev->devfn) != dev_descr->func)) { + sbridge_printk(KERN_ERR, + "Device PCI ID %04x:%04x " + "has dev %02x:%d.%d instead of dev %02x:%02x.%d\n", + PCI_VENDOR_ID_INTEL, dev_descr->dev_id, + bus, PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), + bus, dev_descr->dev, dev_descr->func); + return -ENODEV; + } + + /* Be sure that the device is enabled */ + if (unlikely(pci_enable_device(pdev) < 0)) { + sbridge_printk(KERN_ERR, + "Couldn't enable " + "dev %02x:%d.%d PCI ID %04x:%04x\n", + bus, dev_descr->dev, dev_descr->func, + PCI_VENDOR_ID_INTEL, dev_descr->dev_id); + return -ENODEV; + } + + debugf0("Detected dev %02x:%d.%d PCI ID %04x:%04x\n", + bus, dev_descr->dev, + dev_descr->func, + PCI_VENDOR_ID_INTEL, dev_descr->dev_id); + + /* + * As stated on drivers/pci/search.c, the reference count for + * @from is always decremented if it is not %NULL. So, as we need + * to get all devices up to null, we need to do a get for the device + */ + pci_dev_get(pdev); + + *prev = pdev; + + return 0; +} + +static int sbridge_get_all_devices(u8 *num_mc) +{ + int i, rc; + struct pci_dev *pdev = NULL; + const struct pci_id_table *table = pci_dev_descr_sbridge_table; + + while (table && table->descr) { + for (i = 0; i < table->n_devs; i++) { + pdev = NULL; + do { + rc = sbridge_get_onedevice(&pdev, num_mc, + table, i); + if (rc < 0) { + if (i == 0) { + i = table->n_devs; + break; + } + sbridge_put_all_devices(); + return -ENODEV; + } + } while (pdev); + } + table++; + } + + return 0; +} + +static int mci_bind_devs(struct mem_ctl_info *mci, + struct sbridge_dev *sbridge_dev) +{ + struct sbridge_pvt *pvt = mci->pvt_info; + struct pci_dev *pdev; + int i, func, slot; + + for (i = 0; i < sbridge_dev->n_devs; i++) { + pdev = sbridge_dev->pdev[i]; + if (!pdev) + continue; + slot = PCI_SLOT(pdev->devfn); + func = PCI_FUNC(pdev->devfn); + switch (slot) { + case 12: + switch (func) { + case 6: + pvt->pci_sad0 = pdev; + break; + case 7: + pvt->pci_sad1 = pdev; + break; + default: + goto error; + } + break; + case 13: + switch (func) { + case 6: + pvt->pci_br = pdev; + break; + default: + goto error; + } + break; + case 14: + switch (func) { + case 0: + pvt->pci_ha0 = pdev; + break; + default: + goto error; + } + break; + case 15: + switch (func) { + case 0: + pvt->pci_ta = pdev; + break; + case 1: + pvt->pci_ras = pdev; + break; + case 2: + case 3: + case 4: + case 5: + pvt->pci_tad[func - 2] = pdev; + break; + default: + goto error; + } + break; + case 17: + switch (func) { + case 0: + pvt->pci_ddrio = pdev; + break; + default: + goto error; + } + break; + default: + goto error; + } + + debugf0("Associated PCI %02x.%02d.%d with dev = %p\n", + sbridge_dev->bus, + PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn), + pdev); + } + + /* Check if everything were registered */ + if (!pvt->pci_sad0 || !pvt->pci_sad1 || !pvt->pci_ha0 || + !pvt-> pci_tad || !pvt->pci_ras || !pvt->pci_ta || + !pvt->pci_ddrio) + goto enodev; + + for (i = 0; i < NUM_CHANNELS; i++) { + if (!pvt->pci_tad[i]) + goto enodev; + } + return 0; + +enodev: + sbridge_printk(KERN_ERR, "Some needed devices are missing\n"); + return -ENODEV; + +error: + sbridge_printk(KERN_ERR, "Device %d, function %d " + "is out of the expected range\n", + slot, func); + return -EINVAL; +} + +/**************************************************************************** + Error check routines + ****************************************************************************/ + +/* + * While Sandy Bridge has error count registers, SMI BIOS read values from + * and resets the counters. So, they are not reliable for the OS to read + * from them. So, we have no option but to just trust on whatever MCE is + * telling us about the errors. + */ +static void sbridge_mce_output_error(struct mem_ctl_info *mci, + const struct mce *m) +{ + struct mem_ctl_info *new_mci; + struct sbridge_pvt *pvt = mci->pvt_info; + char *type, *optype, *msg, *recoverable_msg; + bool ripv = GET_BITFIELD(m->mcgstatus, 0, 0); + bool overflow = GET_BITFIELD(m->status, 62, 62); + bool uncorrected_error = GET_BITFIELD(m->status, 61, 61); + bool recoverable = GET_BITFIELD(m->status, 56, 56); + u32 core_err_cnt = GET_BITFIELD(m->status, 38, 52); + u32 mscod = GET_BITFIELD(m->status, 16, 31); + u32 errcode = GET_BITFIELD(m->status, 0, 15); + u32 channel = GET_BITFIELD(m->status, 0, 3); + u32 optypenum = GET_BITFIELD(m->status, 4, 6); + long channel_mask, first_channel; + u8 rank, socket; + int csrow, rc, dimm; + char *area_type = "Unknown"; + + if (ripv) + type = "NON_FATAL"; + else + type = "FATAL"; + + /* + * According with Table 15-9 of the Intel Archictecture spec vol 3A, + * memory errors should fit in this mask: + * 000f 0000 1mmm cccc (binary) + * where: + * f = Correction Report Filtering Bit. If 1, subsequent errors + * won't be shown + * mmm = error type + * cccc = channel + * If the mask doesn't match, report an error to the parsing logic + */ + if (! ((errcode & 0xef80) == 0x80)) { + optype = "Can't parse: it is not a mem"; + } else { + switch (optypenum) { + case 0: + optype = "generic undef request"; + break; + case 1: + optype = "memory read"; + break; + case 2: + optype = "memory write"; + break; + case 3: + optype = "addr/cmd"; + break; + case 4: + optype = "memory scrubbing"; + break; + default: + optype = "reserved"; + break; + } + } + + rc = get_memory_error_data(mci, m->addr, &socket, + &channel_mask, &rank, area_type); + if (rc < 0) + return; + new_mci = get_mci_for_node_id(socket); + if (!new_mci) { + edac_mc_handle_ce_no_info(mci, "Error: socket got corrupted!"); + return; + } + mci = new_mci; + pvt = mci->pvt_info; + + first_channel = find_first_bit(&channel_mask, NUM_CHANNELS); + + if (rank < 4) + dimm = 0; + else if (rank < 8) + dimm = 1; + else + dimm = 2; + + csrow = pvt->csrow_map[first_channel][dimm]; + + if (uncorrected_error && recoverable) + recoverable_msg = " recoverable"; + else + recoverable_msg = ""; + + /* + * FIXME: What should we do with "channel" information on mcelog? + * Probably, we can just discard it, as the channel information + * comes from the get_memory_error_data() address decoding + */ + msg = kasprintf(GFP_ATOMIC, + "%d %s error(s): %s on %s area %s%s: cpu=%d Err=%04x:%04x (ch=%d), " + "addr = 0x%08llx => socket=%d, Channel=%ld(mask=%ld), rank=%d\n", + core_err_cnt, + area_type, + optype, + type, + recoverable_msg, + overflow ? "OVERFLOW" : "", + m->cpu, + mscod, errcode, + channel, /* 1111b means not specified */ + (long long) m->addr, + socket, + first_channel, /* This is the real channel on SB */ + channel_mask, + rank); + + debugf0("%s", msg); + + /* Call the helper to output message */ + if (uncorrected_error) + edac_mc_handle_fbd_ue(mci, csrow, 0, 0, msg); + else + edac_mc_handle_fbd_ce(mci, csrow, 0, msg); + + kfree(msg); +} + +/* + * sbridge_check_error Retrieve and process errors reported by the + * hardware. Called by the Core module. + */ +static void sbridge_check_error(struct mem_ctl_info *mci) +{ + struct sbridge_pvt *pvt = mci->pvt_info; + int i; + unsigned count = 0; + struct mce *m; + + /* + * MCE first step: Copy all mce errors into a temporary buffer + * We use a double buffering here, to reduce the risk of + * loosing an error. + */ + smp_rmb(); + count = (pvt->mce_out + MCE_LOG_LEN - pvt->mce_in) + % MCE_LOG_LEN; + if (!count) + return; + + m = pvt->mce_outentry; + if (pvt->mce_in + count > MCE_LOG_LEN) { + unsigned l = MCE_LOG_LEN - pvt->mce_in; + + memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * l); + smp_wmb(); + pvt->mce_in = 0; + count -= l; + m += l; + } + memcpy(m, &pvt->mce_entry[pvt->mce_in], sizeof(*m) * count); + smp_wmb(); + pvt->mce_in += count; + + smp_rmb(); + if (pvt->mce_overrun) { + sbridge_printk(KERN_ERR, "Lost %d memory errors\n", + pvt->mce_overrun); + smp_wmb(); + pvt->mce_overrun = 0; + } + + /* + * MCE second step: parse errors and display + */ + for (i = 0; i < count; i++) + sbridge_mce_output_error(mci, &pvt->mce_outentry[i]); +} + +/* + * sbridge_mce_check_error Replicates mcelog routine to get errors + * This routine simply queues mcelog errors, and + * return. The error itself should be handled later + * by sbridge_check_error. + * WARNING: As this routine should be called at NMI time, extra care should + * be taken to avoid deadlocks, and to be as fast as possible. + */ +static int sbridge_mce_check_error(struct notifier_block *nb, unsigned long val, + void *data) +{ + struct mce *mce = (struct mce *)data; + struct mem_ctl_info *mci; + struct sbridge_pvt *pvt; + + mci = get_mci_for_node_id(mce->socketid); + if (!mci) + return NOTIFY_BAD; + pvt = mci->pvt_info; + + /* + * Just let mcelog handle it if the error is + * outside the memory controller. A memory error + * is indicated by bit 7 = 1 and bits = 8-11,13-15 = 0. + * bit 12 has an special meaning. + */ + if ((mce->status & 0xefff) >> 7 != 1) + return NOTIFY_DONE; + + printk("sbridge: HANDLING MCE MEMORY ERROR\n"); + + printk("CPU %d: Machine Check Exception: %Lx Bank %d: %016Lx\n", + mce->extcpu, mce->mcgstatus, mce->bank, mce->status); + printk("TSC %llx ", mce->tsc); + printk("ADDR %llx ", mce->addr); + printk("MISC %llx ", mce->misc); + + printk("PROCESSOR %u:%x TIME %llu SOCKET %u APIC %x\n", + mce->cpuvendor, mce->cpuid, mce->time, + mce->socketid, mce->apicid); + +#ifdef CONFIG_SMP + /* Only handle if it is the right mc controller */ + if (cpu_data(mce->cpu).phys_proc_id != pvt->sbridge_dev->mc) + return NOTIFY_DONE; +#endif + + smp_rmb(); + if ((pvt->mce_out + 1) % MCE_LOG_LEN == pvt->mce_in) { + smp_wmb(); + pvt->mce_overrun++; + return NOTIFY_DONE; + } + + /* Copy memory error at the ringbuffer */ + memcpy(&pvt->mce_entry[pvt->mce_out], mce, sizeof(*mce)); + smp_wmb(); + pvt->mce_out = (pvt->mce_out + 1) % MCE_LOG_LEN; + + /* Handle fatal errors immediately */ + if (mce->mcgstatus & 1) + sbridge_check_error(mci); + + /* Advice mcelog that the error were handled */ + return NOTIFY_STOP; +} + +static struct notifier_block sbridge_mce_dec = { + .notifier_call = sbridge_mce_check_error, +}; + +/**************************************************************************** + EDAC register/unregister logic + ****************************************************************************/ + +static void sbridge_unregister_mci(struct sbridge_dev *sbridge_dev) +{ + struct mem_ctl_info *mci = sbridge_dev->mci; + struct sbridge_pvt *pvt; + + if (unlikely(!mci || !mci->pvt_info)) { + debugf0("MC: " __FILE__ ": %s(): dev = %p\n", + __func__, &sbridge_dev->pdev[0]->dev); + + sbridge_printk(KERN_ERR, "Couldn't find mci handler\n"); + return; + } + + pvt = mci->pvt_info; + + debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n", + __func__, mci, &sbridge_dev->pdev[0]->dev); + + atomic_notifier_chain_unregister(&x86_mce_decoder_chain, + &sbridge_mce_dec); + + /* Remove MC sysfs nodes */ + edac_mc_del_mc(mci->dev); + + debugf1("%s: free mci struct\n", mci->ctl_name); + kfree(mci->ctl_name); + edac_mc_free(mci); + sbridge_dev->mci = NULL; +} + +static int sbridge_register_mci(struct sbridge_dev *sbridge_dev) +{ + struct mem_ctl_info *mci; + struct sbridge_pvt *pvt; + int rc, channels, csrows; + + /* Check the number of active and not disabled channels */ + rc = sbridge_get_active_channels(sbridge_dev->bus, &channels, &csrows); + if (unlikely(rc < 0)) + return rc; + + /* allocate a new MC control structure */ + mci = edac_mc_alloc(sizeof(*pvt), csrows, channels, sbridge_dev->mc); + if (unlikely(!mci)) + return -ENOMEM; + + debugf0("MC: " __FILE__ ": %s(): mci = %p, dev = %p\n", + __func__, mci, &sbridge_dev->pdev[0]->dev); + + pvt = mci->pvt_info; + memset(pvt, 0, sizeof(*pvt)); + + /* Associate sbridge_dev and mci for future usage */ + pvt->sbridge_dev = sbridge_dev; + sbridge_dev->mci = mci; + + mci->mtype_cap = MEM_FLAG_DDR3; + mci->edac_ctl_cap = EDAC_FLAG_NONE; + mci->edac_cap = EDAC_FLAG_NONE; + mci->mod_name = "sbridge_edac.c"; + mci->mod_ver = SBRIDGE_REVISION; + mci->ctl_name = kasprintf(GFP_KERNEL, "Sandy Bridge Socket#%d", mci->mc_idx); + mci->dev_name = pci_name(sbridge_dev->pdev[0]); + mci->ctl_page_to_phys = NULL; + + /* Set the function pointer to an actual operation function */ + mci->edac_check = sbridge_check_error; + + /* Store pci devices at mci for faster access */ + rc = mci_bind_devs(mci, sbridge_dev); + if (unlikely(rc < 0)) + goto fail0; + + /* Get dimm basic config and the memory layout */ + get_dimm_config(mci); + get_memory_layout(mci); + + /* record ptr to the generic device */ + mci->dev = &sbridge_dev->pdev[0]->dev; + + /* add this new MC control structure to EDAC's list of MCs */ + if (unlikely(edac_mc_add_mc(mci))) { + debugf0("MC: " __FILE__ + ": %s(): failed edac_mc_add_mc()\n", __func__); + rc = -EINVAL; + goto fail0; + } + + atomic_notifier_chain_register(&x86_mce_decoder_chain, + &sbridge_mce_dec); + return 0; + +fail0: + kfree(mci->ctl_name); + edac_mc_free(mci); + sbridge_dev->mci = NULL; + return rc; +} + +/* + * sbridge_probe Probe for ONE instance of device to see if it is + * present. + * return: + * 0 for FOUND a device + * < 0 for error code + */ + +static int __devinit sbridge_probe(struct pci_dev *pdev, + const struct pci_device_id *id) +{ + int rc; + u8 mc, num_mc = 0; + struct sbridge_dev *sbridge_dev; + + /* get the pci devices we want to reserve for our use */ + mutex_lock(&sbridge_edac_lock); + + /* + * All memory controllers are allocated at the first pass. + */ + if (unlikely(probed >= 1)) { + mutex_unlock(&sbridge_edac_lock); + return -ENODEV; + } + probed++; + + rc = sbridge_get_all_devices(&num_mc); + if (unlikely(rc < 0)) + goto fail0; + mc = 0; + + list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) { + debugf0("Registering MC#%d (%d of %d)\n", mc, mc + 1, num_mc); + sbridge_dev->mc = mc++; + rc = sbridge_register_mci(sbridge_dev); + if (unlikely(rc < 0)) + goto fail1; + } + + sbridge_printk(KERN_INFO, "Driver loaded.\n"); + + mutex_unlock(&sbridge_edac_lock); + return 0; + +fail1: + list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) + sbridge_unregister_mci(sbridge_dev); + + sbridge_put_all_devices(); +fail0: + mutex_unlock(&sbridge_edac_lock); + return rc; +} + +/* + * sbridge_remove destructor for one instance of device + * + */ +static void __devexit sbridge_remove(struct pci_dev *pdev) +{ + struct sbridge_dev *sbridge_dev; + + debugf0(__FILE__ ": %s()\n", __func__); + + /* + * we have a trouble here: pdev value for removal will be wrong, since + * it will point to the X58 register used to detect that the machine + * is a Nehalem or upper design. However, due to the way several PCI + * devices are grouped together to provide MC functionality, we need + * to use a different method for releasing the devices + */ + + mutex_lock(&sbridge_edac_lock); + + if (unlikely(!probed)) { + mutex_unlock(&sbridge_edac_lock); + return; + } + + list_for_each_entry(sbridge_dev, &sbridge_edac_list, list) + sbridge_unregister_mci(sbridge_dev); + + /* Release PCI resources */ + sbridge_put_all_devices(); + + probed--; + + mutex_unlock(&sbridge_edac_lock); +} + +MODULE_DEVICE_TABLE(pci, sbridge_pci_tbl); + +/* + * sbridge_driver pci_driver structure for this module + * + */ +static struct pci_driver sbridge_driver = { + .name = "sbridge_edac", + .probe = sbridge_probe, + .remove = __devexit_p(sbridge_remove), + .id_table = sbridge_pci_tbl, +}; + +/* + * sbridge_init Module entry function + * Try to initialize this module for its devices + */ +static int __init sbridge_init(void) +{ + int pci_rc; + + debugf2("MC: " __FILE__ ": %s()\n", __func__); + + /* Ensure that the OPSTATE is set correctly for POLL or NMI */ + opstate_init(); + + pci_rc = pci_register_driver(&sbridge_driver); + + if (pci_rc >= 0) + return 0; + + sbridge_printk(KERN_ERR, "Failed to register device with error %d.\n", + pci_rc); + + return pci_rc; +} + +/* + * sbridge_exit() Module exit function + * Unregister the driver + */ +static void __exit sbridge_exit(void) +{ + debugf2("MC: " __FILE__ ": %s()\n", __func__); + pci_unregister_driver(&sbridge_driver); +} + +module_init(sbridge_init); +module_exit(sbridge_exit); + +module_param(edac_op_state, int, 0444); +MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); + +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("Mauro Carvalho Chehab <mchehab@redhat.com>"); +MODULE_AUTHOR("Red Hat Inc. (http://www.redhat.com)"); +MODULE_DESCRIPTION("MC Driver for Intel Sandy Bridge memory controllers - " + SBRIDGE_REVISION); |