[net-next-2.6.git] / drivers / edac / amd64_edac.h

/*
 * AMD64 class Memory Controller kernel module
 *
 * Copyright (c) 2009 SoftwareBitMaker.
 * Copyright (c) 2009 Advanced Micro Devices, Inc.
 *
 * This file may be distributed under the terms of the
 * GNU General Public License.
 *
 *	Originally Written by Thayne Harbaugh
 *
 *      Changes by Douglas "norsk" Thompson  <dougthompson@xmission.com>:
 *		- K8 CPU Revision D and greater support
 *
 *      Changes by Dave Peterson <dsp@llnl.gov> <dave_peterson@pobox.com>:
 *		- Module largely rewritten, with new (and hopefully correct)
 *		code for dealing with node and chip select interleaving,
 *		various code cleanup, and bug fixes
 *		- Added support for memory hoisting using DRAM hole address
 *		register
 *
 *	Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
 *		-K8 Rev (1207) revision support added, required Revision
 *		specific mini-driver code to support Rev F as well as
 *		prior revisions
 *
 *	Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
 *		-Family 10h revision support added. New PCI Device IDs,
 *		indicating new changes. Actual registers modified
 *		were slight, less than the Rev E to Rev F transition
 *		but changing the PCI Device ID was the proper thing to
 *		do, as it provides for almost automactic family
 *		detection. The mods to Rev F required more family
 *		information detection.
 *
 *	Changes/Fixes by Borislav Petkov <borislav.petkov@amd.com>:
 *		- misc fixes and code cleanups
 *
 * This module is based on the following documents
 * (available from http://www.amd.com/):
 *
 *	Title:	BIOS and Kernel Developer's Guide for AMD Athlon 64 and AMD
 *		Opteron Processors
 *	AMD publication #: 26094
 *`	Revision: 3.26
 *
 *	Title:	BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh
 *		Processors
 *	AMD publication #: 32559
 *	Revision: 3.00
 *	Issue Date: May 2006
 *
 *	Title:	BIOS and Kernel Developer's Guide (BKDG) For AMD Family 10h
 *		Processors
 *	AMD publication #: 31116
 *	Revision: 3.00
 *	Issue Date: September 07, 2007
 *
 * Sections in the first 2 documents are no longer in sync with each other.
 * The Family 10h BKDG was totally re-written from scratch with a new
 * presentation model.
 * Therefore, comments that refer to a Document section might be off.
 */

#include <linux/module.h>
#include <linux/ctype.h>
#include <linux/init.h>
#include <linux/pci.h>
#include <linux/pci_ids.h>
#include <linux/slab.h>
#include <linux/mmzone.h>
#include <linux/edac.h>
#include <asm/msr.h>
#include "edac_core.h"
#include "mce_amd.h"

#define amd64_printk(level, fmt, arg...) \
	edac_printk(level, "amd64", fmt, ##arg)

#define amd64_mc_printk(mci, level, fmt, arg...) \
	edac_mc_chipset_printk(mci, level, "amd64", fmt, ##arg)

/*
 * Throughout the comments in this code, the following terms are used:
 *
 *	SysAddr, DramAddr, and InputAddr
 *
 *  These terms come directly from the amd64 documentation
 * (AMD publication #26094).  They are defined as follows:
 *
 *     SysAddr:
 *         This is a physical address generated by a CPU core or a device
 *         doing DMA.  If generated by a CPU core, a SysAddr is the result of
 *         a virtual to physical address translation by the CPU core's address
 *         translation mechanism (MMU).
 *
 *     DramAddr:
 *         A DramAddr is derived from a SysAddr by subtracting an offset that
 *         depends on which node the SysAddr maps to and whether the SysAddr
 *         is within a range affected by memory hoisting.  The DRAM Base
 *         (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
 *         determine which node a SysAddr maps to.
 *
 *         If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
 *         is within the range of addresses specified by this register, then
 *         a value x from the DHAR is subtracted from the SysAddr to produce a
 *         DramAddr.  Here, x represents the base address for the node that
 *         the SysAddr maps to plus an offset due to memory hoisting.  See
 *         section 3.4.8 and the comments in amd64_get_dram_hole_info() and
 *         sys_addr_to_dram_addr() below for more information.
 *
 *         If the SysAddr is not affected by the DHAR then a value y is
 *         subtracted from the SysAddr to produce a DramAddr.  Here, y is the
 *         base address for the node that the SysAddr maps to.  See section
 *         3.4.4 and the comments in sys_addr_to_dram_addr() below for more
 *         information.
 *
 *     InputAddr:
 *         A DramAddr is translated to an InputAddr before being passed to the
 *         memory controller for the node that the DramAddr is associated
 *         with.  The memory controller then maps the InputAddr to a csrow.
 *         If node interleaving is not in use, then the InputAddr has the same
 *         value as the DramAddr.  Otherwise, the InputAddr is produced by
 *         discarding the bits used for node interleaving from the DramAddr.
 *         See section 3.4.4 for more information.
 *
 *         The memory controller for a given node uses its DRAM CS Base and
 *         DRAM CS Mask registers to map an InputAddr to a csrow.  See
 *         sections 3.5.4 and 3.5.5 for more information.
 */

#define EDAC_AMD64_VERSION		" Ver: 3.3.0 " __DATE__
#define EDAC_MOD_STR			"amd64_edac"

#define EDAC_MAX_NUMNODES		8

/* Extended Model from CPUID, for CPU Revision numbers */
#define K8_REV_D			1
#define K8_REV_E			2
#define K8_REV_F			4

/* Hardware limit on ChipSelect rows per MC and processors per system */
#define MAX_CS_COUNT			8
#define DRAM_REG_COUNT			8

#define ON true
#define OFF false

/*
 * PCI-defined configuration space registers
 */


/*
 * Function 1 - Address Map
 */
#define K8_DRAM_BASE_LOW		0x40
#define K8_DRAM_LIMIT_LOW		0x44
#define K8_DHAR				0xf0

#define DHAR_VALID			BIT(0)
#define F10_DRAM_MEM_HOIST_VALID	BIT(1)

#define DHAR_BASE_MASK			0xff000000
#define dhar_base(dhar)			(dhar & DHAR_BASE_MASK)

#define K8_DHAR_OFFSET_MASK		0x0000ff00
#define k8_dhar_offset(dhar)		((dhar & K8_DHAR_OFFSET_MASK) << 16)

#define F10_DHAR_OFFSET_MASK		0x0000ff80
					/* NOTE: Extra mask bit vs K8 */
#define f10_dhar_offset(dhar)		((dhar & F10_DHAR_OFFSET_MASK) << 16)


/* F10 High BASE/LIMIT registers */
#define F10_DRAM_BASE_HIGH		0x140
#define F10_DRAM_LIMIT_HIGH		0x144


/*
 * Function 2 - DRAM controller
 */
#define K8_DCSB0			0x40
#define F10_DCSB1			0x140

#define K8_DCSB_CS_ENABLE		BIT(0)
#define K8_DCSB_NPT_SPARE		BIT(1)
#define K8_DCSB_NPT_TESTFAIL		BIT(2)

/*
 * REV E: select [31:21] and [15:9] from DCSB and the shift amount to form
 * the address
 */
#define REV_E_DCSB_BASE_BITS		(0xFFE0FE00ULL)
#define REV_E_DCS_SHIFT			4

#define REV_F_F1Xh_DCSB_BASE_BITS	(0x1FF83FE0ULL)
#define REV_F_F1Xh_DCS_SHIFT		8

/*
 * REV F and later: selects [28:19] and [13:5] from DCSB and the shift amount
 * to form the address
 */
#define REV_F_DCSB_BASE_BITS		(0x1FF83FE0ULL)
#define REV_F_DCS_SHIFT			8

/* DRAM CS Mask Registers */
#define K8_DCSM0			0x60
#define F10_DCSM1			0x160

/* REV E: select [29:21] and [15:9] from DCSM */
#define REV_E_DCSM_MASK_BITS		0x3FE0FE00

/* unused bits [24:20] and [12:0] */
#define REV_E_DCS_NOTUSED_BITS		0x01F01FFF

/* REV F and later: select [28:19] and [13:5] from DCSM */
#define REV_F_F1Xh_DCSM_MASK_BITS	0x1FF83FE0

/* unused bits [26:22] and [12:0] */
#define REV_F_F1Xh_DCS_NOTUSED_BITS	0x07C01FFF

#define DBAM0				0x80
#define DBAM1				0x180

/* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
#define DBAM_DIMM(i, reg)		((((reg) >> (4*i))) & 0xF)

#define DBAM_MAX_VALUE			11


#define F10_DCLR_0			0x90
#define F10_DCLR_1			0x190
#define REVE_WIDTH_128			BIT(16)
#define F10_WIDTH_128			BIT(11)


#define F10_DCHR_0			0x94
#define F10_DCHR_1			0x194

#define F10_DCHR_FOUR_RANK_DIMM		BIT(18)
#define DDR3_MODE			BIT(8)
#define F10_DCHR_MblMode		BIT(6)


#define F10_DCTL_SEL_LOW		0x110
#define dct_sel_baseaddr(pvt)		((pvt->dram_ctl_select_low) & 0xFFFFF800)
#define dct_sel_interleave_addr(pvt)	(((pvt->dram_ctl_select_low) >> 6) & 0x3)
#define dct_high_range_enabled(pvt)	(pvt->dram_ctl_select_low & BIT(0))
#define dct_interleave_enabled(pvt)	(pvt->dram_ctl_select_low & BIT(2))
#define dct_ganging_enabled(pvt)	(pvt->dram_ctl_select_low & BIT(4))
#define dct_data_intlv_enabled(pvt)	(pvt->dram_ctl_select_low & BIT(5))
#define dct_dram_enabled(pvt)		(pvt->dram_ctl_select_low & BIT(8))
#define dct_memory_cleared(pvt)		(pvt->dram_ctl_select_low & BIT(10))

#define F10_DCTL_SEL_HIGH		0x114

/*
 * Function 3 - Misc Control
 */
#define K8_NBCTL			0x40

/* Correctable ECC error reporting enable */
#define K8_NBCTL_CECCEn			BIT(0)

/* UnCorrectable ECC error reporting enable */
#define K8_NBCTL_UECCEn			BIT(1)

#define K8_NBCFG			0x44
#define K8_NBCFG_CHIPKILL		BIT(23)
#define K8_NBCFG_ECC_ENABLE		BIT(22)

#define K8_NBSL				0x48


/* Family F10h: Normalized Extended Error Codes */
#define F10_NBSL_EXT_ERR_RES		0x0
#define F10_NBSL_EXT_ERR_ECC		0x8

/* Next two are overloaded values */
#define F10_NBSL_EXT_ERR_LINK_PROTO	0xB
#define F10_NBSL_EXT_ERR_L3_PROTO	0xB

#define F10_NBSL_EXT_ERR_NB_ARRAY	0xC
#define F10_NBSL_EXT_ERR_DRAM_PARITY	0xD
#define F10_NBSL_EXT_ERR_LINK_RETRY	0xE

/* Next two are overloaded values */
#define F10_NBSL_EXT_ERR_GART_WALK	0xF
#define F10_NBSL_EXT_ERR_DEV_WALK	0xF

/* 0x10 to 0x1B: Reserved */
#define F10_NBSL_EXT_ERR_L3_DATA	0x1C
#define F10_NBSL_EXT_ERR_L3_TAG		0x1D
#define F10_NBSL_EXT_ERR_L3_LRU		0x1E

/* K8: Normalized Extended Error Codes */
#define K8_NBSL_EXT_ERR_ECC		0x0
#define K8_NBSL_EXT_ERR_CRC		0x1
#define K8_NBSL_EXT_ERR_SYNC		0x2
#define K8_NBSL_EXT_ERR_MST		0x3
#define K8_NBSL_EXT_ERR_TGT		0x4
#define K8_NBSL_EXT_ERR_GART		0x5
#define K8_NBSL_EXT_ERR_RMW		0x6
#define K8_NBSL_EXT_ERR_WDT		0x7
#define K8_NBSL_EXT_ERR_CHIPKILL_ECC	0x8
#define K8_NBSL_EXT_ERR_DRAM_PARITY	0xD

/*
 * The following are for BUS type errors AFTER values have been normalized by
 * shifting right
 */
#define K8_NBSL_PP_SRC			0x0
#define K8_NBSL_PP_RES			0x1
#define K8_NBSL_PP_OBS			0x2
#define K8_NBSL_PP_GENERIC		0x3

#define EXTRACT_ERR_CPU_MAP(x)		((x) & 0xF)

#define K8_NBEAL			0x50
#define K8_NBEAH			0x54
#define K8_SCRCTRL			0x58

#define F10_NB_CFG_LOW			0x88
#define	F10_NB_CFG_LOW_ENABLE_EXT_CFG	BIT(14)

#define F10_NB_CFG_HIGH			0x8C

#define F10_ONLINE_SPARE		0xB0
#define F10_ONLINE_SPARE_SWAPDONE0(x)	((x) & BIT(1))
#define F10_ONLINE_SPARE_SWAPDONE1(x)	((x) & BIT(3))
#define F10_ONLINE_SPARE_BADDRAM_CS0(x) (((x) >> 4) & 0x00000007)
#define F10_ONLINE_SPARE_BADDRAM_CS1(x) (((x) >> 8) & 0x00000007)

#define F10_NB_ARRAY_ADDR		0xB8

#define F10_NB_ARRAY_DRAM_ECC		0x80000000

/* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline  */
#define SET_NB_ARRAY_ADDRESS(section)	(((section) & 0x3) << 1)

#define F10_NB_ARRAY_DATA		0xBC

#define SET_NB_DRAM_INJECTION_WRITE(word, bits)  \
					(BIT(((word) & 0xF) + 20) | \
					BIT(17) | bits)

#define SET_NB_DRAM_INJECTION_READ(word, bits)  \
					(BIT(((word) & 0xF) + 20) | \
					BIT(16) |  bits)

#define K8_NBCAP			0xE8
#define K8_NBCAP_CORES			(BIT(12)|BIT(13))
#define K8_NBCAP_CHIPKILL		BIT(4)
#define K8_NBCAP_SECDED			BIT(3)
#define K8_NBCAP_DCT_DUAL		BIT(0)

#define EXT_NB_MCA_CFG			0x180

/* MSRs */
#define K8_MSR_MCGCTL_NBE		BIT(4)

#define K8_MSR_MC4CTL			0x0410
#define K8_MSR_MC4STAT			0x0411
#define K8_MSR_MC4ADDR			0x0412

/* AMD sets the first MC device at device ID 0x18. */
static inline int get_node_id(struct pci_dev *pdev)
{
	return PCI_SLOT(pdev->devfn) - 0x18;
}

enum amd64_chipset_families {
	K8_CPUS = 0,
	F10_CPUS,
	F11_CPUS,
};

/* Error injection control structure */
struct error_injection {
	u32	section;
	u32	word;
	u32	bit_map;
};

struct amd64_pvt {
	/* pci_device handles which we utilize */
	struct pci_dev *addr_f1_ctl;
	struct pci_dev *dram_f2_ctl;
	struct pci_dev *misc_f3_ctl;

	int mc_node_id;		/* MC index of this MC node */
	int ext_model;		/* extended model value of this node */

	struct low_ops *ops;	/* pointer to per PCI Device ID func table */

	int channel_count;

	/* Raw registers */
	u32 dclr0;		/* DRAM Configuration Low DCT0 reg */
	u32 dclr1;		/* DRAM Configuration Low DCT1 reg */
	u32 dchr0;		/* DRAM Configuration High DCT0 reg */
	u32 dchr1;		/* DRAM Configuration High DCT1 reg */
	u32 nbcap;		/* North Bridge Capabilities */
	u32 nbcfg;		/* F10 North Bridge Configuration */
	u32 ext_nbcfg;		/* Extended F10 North Bridge Configuration */
	u32 dhar;		/* DRAM Hoist reg */
	u32 dbam0;		/* DRAM Base Address Mapping reg for DCT0 */
	u32 dbam1;		/* DRAM Base Address Mapping reg for DCT1 */

	/* DRAM CS Base Address Registers F2x[1,0][5C:40] */
	u32 dcsb0[MAX_CS_COUNT];
	u32 dcsb1[MAX_CS_COUNT];

	/* DRAM CS Mask Registers F2x[1,0][6C:60] */
	u32 dcsm0[MAX_CS_COUNT];
	u32 dcsm1[MAX_CS_COUNT];

	/*
	 * Decoded parts of DRAM BASE and LIMIT Registers
	 * F1x[78,70,68,60,58,50,48,40]
	 */
	u64 dram_base[DRAM_REG_COUNT];
	u64 dram_limit[DRAM_REG_COUNT];
	u8  dram_IntlvSel[DRAM_REG_COUNT];
	u8  dram_IntlvEn[DRAM_REG_COUNT];
	u8  dram_DstNode[DRAM_REG_COUNT];
	u8  dram_rw_en[DRAM_REG_COUNT];

	/*
	 * The following fields are set at (load) run time, after CPU revision
	 * has been determined, since the dct_base and dct_mask registers vary
	 * based on revision
	 */
	u32 dcsb_base;		/* DCSB base bits */
	u32 dcsm_mask;		/* DCSM mask bits */
	u32 cs_count;		/* num chip selects (== num DCSB registers) */
	u32 num_dcsm;		/* Number of DCSM registers */
	u32 dcs_mask_notused;	/* DCSM notused mask bits */
	u32 dcs_shift;		/* DCSB and DCSM shift value */

	u64 top_mem;		/* top of memory below 4GB */
	u64 top_mem2;		/* top of memory above 4GB */

	u32 dram_ctl_select_low;	/* DRAM Controller Select Low Reg */
	u32 dram_ctl_select_high;	/* DRAM Controller Select High Reg */
	u32 online_spare;               /* On-Line spare Reg */

	/* x4 or x8 syndromes in use */
	u8 syn_type;

	/* temp storage for when input is received from sysfs */
	struct err_regs ctl_error_info;

	/* place to store error injection parameters prior to issue */
	struct error_injection injection;

	/* Save old hw registers' values before we modified them */
	u32 nbctl_mcgctl_saved;		/* When true, following 2 are valid */
	u32 old_nbctl;

	/* MC Type Index value: socket F vs Family 10h */
	u32 mc_type_index;

	/* misc settings */
	struct flags {
		unsigned long cf8_extcfg:1;
		unsigned long nb_mce_enable:1;
		unsigned long nb_ecc_prev:1;
	} flags;
};

struct scrubrate {
       u32 scrubval;           /* bit pattern for scrub rate */
       u32 bandwidth;          /* bandwidth consumed (bytes/sec) */
};

extern struct scrubrate scrubrates[23];
extern const char *tt_msgs[4];
extern const char *ll_msgs[4];
extern const char *rrrr_msgs[16];
extern const char *to_msgs[2];
extern const char *pp_msgs[4];
extern const char *ii_msgs[4];
extern const char *htlink_msgs[8];

#ifdef CONFIG_EDAC_DEBUG
#define NUM_DBG_ATTRS 5
#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.
 */
struct low_ops {
	int (*early_channel_count)	(struct amd64_pvt *pvt);

	u64 (*get_error_address)	(struct mem_ctl_info *mci,
					 struct err_regs *info);
	void (*read_dram_base_limit)	(struct amd64_pvt *pvt, int dram);
	void (*read_dram_ctl_register)	(struct amd64_pvt *pvt);
	void (*map_sysaddr_to_csrow)	(struct mem_ctl_info *mci,
					 struct err_regs *info, u64 SystemAddr);
	int (*dbam_to_cs)		(struct amd64_pvt *pvt, int cs_mode);
};

struct amd64_family_type {
	const char *ctl_name;
	u16 addr_f1_ctl;
	u16 misc_f3_ctl;
	struct low_ops ops;
};

static struct amd64_family_type amd64_family_types[];

static inline const char *get_amd_family_name(int index)
{
	return amd64_family_types[index].ctl_name;
}

static inline struct low_ops *family_ops(int index)
{
	return &amd64_family_types[index].ops;
}

static inline int amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
					   u32 *val, const char *func)
{
	int err = 0;

	err = pci_read_config_dword(pdev, offset, val);
	if (err)
		amd64_printk(KERN_WARNING, "%s: error reading F%dx%x.\n",
			     func, PCI_FUNC(pdev->devfn), offset);

	return err;
}

#define amd64_read_pci_cfg(pdev, offset, val)	\
	amd64_read_pci_cfg_dword(pdev, offset, val, __func__)

/*
 * For future CPU versions, verify the following as new 'slow' rates appear and
 * modify the necessary skip values for the supported CPU.
 */
#define K8_MIN_SCRUB_RATE_BITS	0x0
#define F10_MIN_SCRUB_RATE_BITS	0x5
#define F11_MIN_SCRUB_RATE_BITS	0x6

int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
			     u64 *hole_offset, u64 *hole_size);
Commit	Line	Data
cfe40fdb DT	1	/*
	2	* AMD64 class Memory Controller kernel module
	3	*
	4	* Copyright (c) 2009 SoftwareBitMaker.
	5	* Copyright (c) 2009 Advanced Micro Devices, Inc.
	6	*
	7	* This file may be distributed under the terms of the
	8	* GNU General Public License.
	9	*
	10	* Originally Written by Thayne Harbaugh
	11	*
	12	* Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
	13	* - K8 CPU Revision D and greater support
	14	*
	15	* Changes by Dave Peterson <dsp@llnl.gov> <dave_peterson@pobox.com>:
	16	* - Module largely rewritten, with new (and hopefully correct)
	17	* code for dealing with node and chip select interleaving,
	18	* various code cleanup, and bug fixes
	19	* - Added support for memory hoisting using DRAM hole address
	20	* register
	21	*
	22	* Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
	23	* -K8 Rev (1207) revision support added, required Revision
	24	* specific mini-driver code to support Rev F as well as
	25	* prior revisions
	26	*
	27	* Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
	28	* -Family 10h revision support added. New PCI Device IDs,
	29	* indicating new changes. Actual registers modified
	30	* were slight, less than the Rev E to Rev F transition
	31	* but changing the PCI Device ID was the proper thing to
	32	* do, as it provides for almost automactic family
	33	* detection. The mods to Rev F required more family
	34	* information detection.
	35	*
	36	* Changes/Fixes by Borislav Petkov <borislav.petkov@amd.com>:
	37	* - misc fixes and code cleanups
	38	*
	39	* This module is based on the following documents
	40	* (available from http://www.amd.com/):
	41	*
	42	* Title: BIOS and Kernel Developer's Guide for AMD Athlon 64 and AMD
	43	* Opteron Processors
	44	* AMD publication #: 26094
	45	*` Revision: 3.26
	46	*
	47	* Title: BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh
	48	* Processors
	49	* AMD publication #: 32559
	50	* Revision: 3.00
	51	* Issue Date: May 2006
	52	*
	53	* Title: BIOS and Kernel Developer's Guide (BKDG) For AMD Family 10h
	54	* Processors
	55	* AMD publication #: 31116
	56	* Revision: 3.00
	57	* Issue Date: September 07, 2007
	58	*
	59	* Sections in the first 2 documents are no longer in sync with each other.
	60	* The Family 10h BKDG was totally re-written from scratch with a new
	61	* presentation model.
	62	* Therefore, comments that refer to a Document section might be off.
	63	*/
	64
65	#include <linux/module.h>
66	#include <linux/ctype.h>
67	#include <linux/init.h>
68	#include <linux/pci.h>
69	#include <linux/pci_ids.h>
70	#include <linux/slab.h>
71	#include <linux/mmzone.h>
72	#include <linux/edac.h>
f9431992	73	#include <asm/msr.h>
cfe40fdb	74	#include "edac_core.h"
47ca08a4	75	#include "mce_amd.h"
cfe40fdb DT	76
	77	#define amd64_printk(level, fmt, arg...) \
	78	edac_printk(level, "amd64", fmt, ##arg)
	79
	80	#define amd64_mc_printk(mci, level, fmt, arg...) \
	81	edac_mc_chipset_printk(mci, level, "amd64", fmt, ##arg)
	82
	83	/*
	84	* Throughout the comments in this code, the following terms are used:
	85	*
	86	* SysAddr, DramAddr, and InputAddr
	87	*
	88	* These terms come directly from the amd64 documentation
	89	* (AMD publication #26094). They are defined as follows:
	90	*
	91	* SysAddr:
	92	* This is a physical address generated by a CPU core or a device
	93	* doing DMA. If generated by a CPU core, a SysAddr is the result of
	94	* a virtual to physical address translation by the CPU core's address
	95	* translation mechanism (MMU).
	96	*
	97	* DramAddr:
	98	* A DramAddr is derived from a SysAddr by subtracting an offset that
	99	* depends on which node the SysAddr maps to and whether the SysAddr
	100	* is within a range affected by memory hoisting. The DRAM Base
	101	* (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
	102	* determine which node a SysAddr maps to.
	103	*
	104	* If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
	105	* is within the range of addresses specified by this register, then
	106	* a value x from the DHAR is subtracted from the SysAddr to produce a
	107	* DramAddr. Here, x represents the base address for the node that
	108	* the SysAddr maps to plus an offset due to memory hoisting. See
	109	* section 3.4.8 and the comments in amd64_get_dram_hole_info() and
	110	* sys_addr_to_dram_addr() below for more information.
	111	*
	112	* If the SysAddr is not affected by the DHAR then a value y is
	113	* subtracted from the SysAddr to produce a DramAddr. Here, y is the
	114	* base address for the node that the SysAddr maps to. See section
	115	* 3.4.4 and the comments in sys_addr_to_dram_addr() below for more
	116	* information.
	117	*
	118	* InputAddr:
	119	* A DramAddr is translated to an InputAddr before being passed to the
	120	* memory controller for the node that the DramAddr is associated
	121	* with. The memory controller then maps the InputAddr to a csrow.
	122	* If node interleaving is not in use, then the InputAddr has the same
	123	* value as the DramAddr. Otherwise, the InputAddr is produced by
	124	* discarding the bits used for node interleaving from the DramAddr.
	125	* See section 3.4.4 for more information.
	126	*
	127	* The memory controller for a given node uses its DRAM CS Base and
	128	* DRAM CS Mask registers to map an InputAddr to a csrow. See
	129	* sections 3.5.4 and 3.5.5 for more information.
	130	*/
	131
df5b1606	132	#define EDAC_AMD64_VERSION " Ver: 3.3.0 " __DATE__
cfe40fdb DT	133	#define EDAC_MOD_STR "amd64_edac"
cfe40fdb DT	134
3011b20d BP	135	#define EDAC_MAX_NUMNODES 8
3011b20d BP	136
cfe40fdb	137	/* Extended Model from CPUID, for CPU Revision numbers */
1433eb99 BP	138	#define K8_REV_D 1
	139	#define K8_REV_E 2
	140	#define K8_REV_F 4
cfe40fdb DT	141
cfe40fdb DT	142	/* Hardware limit on ChipSelect rows per MC and processors per system */
9d858bb1	143	#define MAX_CS_COUNT 8
cfe40fdb DT	144	#define DRAM_REG_COUNT 8
cfe40fdb DT	145
f6d6ae96 BP	146	#define ON true
f6d6ae96 BP	147	#define OFF false
cfe40fdb DT	148
	149	/*
	150	* PCI-defined configuration space registers
	151	*/
	152
	153
	154	/*
	155	* Function 1 - Address Map
	156	*/
	157	#define K8_DRAM_BASE_LOW 0x40
	158	#define K8_DRAM_LIMIT_LOW 0x44
	159	#define K8_DHAR 0xf0
	160
	161	#define DHAR_VALID BIT(0)
	162	#define F10_DRAM_MEM_HOIST_VALID BIT(1)
	163
	164	#define DHAR_BASE_MASK 0xff000000
	165	#define dhar_base(dhar) (dhar & DHAR_BASE_MASK)
	166
	167	#define K8_DHAR_OFFSET_MASK 0x0000ff00
	168	#define k8_dhar_offset(dhar) ((dhar & K8_DHAR_OFFSET_MASK) << 16)
	169
	170	#define F10_DHAR_OFFSET_MASK 0x0000ff80
	171	/* NOTE: Extra mask bit vs K8 */
	172	#define f10_dhar_offset(dhar) ((dhar & F10_DHAR_OFFSET_MASK) << 16)
	173
	174
	175	/* F10 High BASE/LIMIT registers */
	176	#define F10_DRAM_BASE_HIGH 0x140
	177	#define F10_DRAM_LIMIT_HIGH 0x144
	178
	179
	180	/*
	181	* Function 2 - DRAM controller
	182	*/
	183	#define K8_DCSB0 0x40
	184	#define F10_DCSB1 0x140
	185
	186	#define K8_DCSB_CS_ENABLE BIT(0)
	187	#define K8_DCSB_NPT_SPARE BIT(1)
	188	#define K8_DCSB_NPT_TESTFAIL BIT(2)
	189
	190	/*
	191	* REV E: select [31:21] and [15:9] from DCSB and the shift amount to form
	192	* the address
	193	*/
	194	#define REV_E_DCSB_BASE_BITS (0xFFE0FE00ULL)
	195	#define REV_E_DCS_SHIFT 4
cfe40fdb DT	196
	197	#define REV_F_F1Xh_DCSB_BASE_BITS (0x1FF83FE0ULL)
	198	#define REV_F_F1Xh_DCS_SHIFT 8
	199
	200	/*
	201	* REV F and later: selects [28:19] and [13:5] from DCSB and the shift amount
	202	* to form the address
	203	*/
	204	#define REV_F_DCSB_BASE_BITS (0x1FF83FE0ULL)
	205	#define REV_F_DCS_SHIFT 8
cfe40fdb DT	206
	207	/* DRAM CS Mask Registers */
	208	#define K8_DCSM0 0x60
	209	#define F10_DCSM1 0x160
	210
	211	/* REV E: select [29:21] and [15:9] from DCSM */
	212	#define REV_E_DCSM_MASK_BITS 0x3FE0FE00
	213
	214	/* unused bits [24:20] and [12:0] */
	215	#define REV_E_DCS_NOTUSED_BITS 0x01F01FFF
	216
	217	/* REV F and later: select [28:19] and [13:5] from DCSM */
	218	#define REV_F_F1Xh_DCSM_MASK_BITS 0x1FF83FE0
	219
	220	/* unused bits [26:22] and [12:0] */
	221	#define REV_F_F1Xh_DCS_NOTUSED_BITS 0x07C01FFF
	222
	223	#define DBAM0 0x80
	224	#define DBAM1 0x180
	225
	226	/* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
	227	#define DBAM_DIMM(i, reg) ((((reg) >> (4*i))) & 0xF)
	228
	229	#define DBAM_MAX_VALUE 11
	230
	231
	232	#define F10_DCLR_0 0x90
	233	#define F10_DCLR_1 0x190
	234	#define REVE_WIDTH_128 BIT(16)
	235	#define F10_WIDTH_128 BIT(11)
	236
	237
	238	#define F10_DCHR_0 0x94
	239	#define F10_DCHR_1 0x194
	240
	241	#define F10_DCHR_FOUR_RANK_DIMM BIT(18)
1433eb99	242	#define DDR3_MODE BIT(8)
cfe40fdb DT	243	#define F10_DCHR_MblMode BIT(6)
	244
	245
	246	#define F10_DCTL_SEL_LOW 0x110
69542650 BP	247	#define dct_sel_baseaddr(pvt) ((pvt->dram_ctl_select_low) & 0xFFFFF800)
	248	#define dct_sel_interleave_addr(pvt) (((pvt->dram_ctl_select_low) >> 6) & 0x3)
	249	#define dct_high_range_enabled(pvt) (pvt->dram_ctl_select_low & BIT(0))
	250	#define dct_interleave_enabled(pvt) (pvt->dram_ctl_select_low & BIT(2))
	251	#define dct_ganging_enabled(pvt) (pvt->dram_ctl_select_low & BIT(4))
	252	#define dct_data_intlv_enabled(pvt) (pvt->dram_ctl_select_low & BIT(5))
	253	#define dct_dram_enabled(pvt) (pvt->dram_ctl_select_low & BIT(8))
	254	#define dct_memory_cleared(pvt) (pvt->dram_ctl_select_low & BIT(10))
cfe40fdb DT	255
	256	#define F10_DCTL_SEL_HIGH 0x114
	257
cfe40fdb DT	258	/*
	259	* Function 3 - Misc Control
	260	*/
	261	#define K8_NBCTL 0x40
	262
	263	/* Correctable ECC error reporting enable */
	264	#define K8_NBCTL_CECCEn BIT(0)
	265
	266	/* UnCorrectable ECC error reporting enable */
	267	#define K8_NBCTL_UECCEn BIT(1)
	268
	269	#define K8_NBCFG 0x44
	270	#define K8_NBCFG_CHIPKILL BIT(23)
	271	#define K8_NBCFG_ECC_ENABLE BIT(22)
	272
	273	#define K8_NBSL 0x48
	274
	275
cfe40fdb DT	276	/* Family F10h: Normalized Extended Error Codes */
cfe40fdb DT	277	#define F10_NBSL_EXT_ERR_RES 0x0
cfe40fdb	278	#define F10_NBSL_EXT_ERR_ECC 0x8
cfe40fdb DT	279
	280	/* Next two are overloaded values */
	281	#define F10_NBSL_EXT_ERR_LINK_PROTO 0xB
	282	#define F10_NBSL_EXT_ERR_L3_PROTO 0xB
	283
	284	#define F10_NBSL_EXT_ERR_NB_ARRAY 0xC
	285	#define F10_NBSL_EXT_ERR_DRAM_PARITY 0xD
	286	#define F10_NBSL_EXT_ERR_LINK_RETRY 0xE
	287
	288	/* Next two are overloaded values */
	289	#define F10_NBSL_EXT_ERR_GART_WALK 0xF
	290	#define F10_NBSL_EXT_ERR_DEV_WALK 0xF
	291
	292	/* 0x10 to 0x1B: Reserved */
	293	#define F10_NBSL_EXT_ERR_L3_DATA 0x1C
	294	#define F10_NBSL_EXT_ERR_L3_TAG 0x1D
	295	#define F10_NBSL_EXT_ERR_L3_LRU 0x1E
	296
	297	/* K8: Normalized Extended Error Codes */
	298	#define K8_NBSL_EXT_ERR_ECC 0x0
	299	#define K8_NBSL_EXT_ERR_CRC 0x1
	300	#define K8_NBSL_EXT_ERR_SYNC 0x2
	301	#define K8_NBSL_EXT_ERR_MST 0x3
	302	#define K8_NBSL_EXT_ERR_TGT 0x4
	303	#define K8_NBSL_EXT_ERR_GART 0x5
	304	#define K8_NBSL_EXT_ERR_RMW 0x6
	305	#define K8_NBSL_EXT_ERR_WDT 0x7
	306	#define K8_NBSL_EXT_ERR_CHIPKILL_ECC 0x8
	307	#define K8_NBSL_EXT_ERR_DRAM_PARITY 0xD
	308
cfe40fdb DT	309	/*
	310	* The following are for BUS type errors AFTER values have been normalized by
	311	* shifting right
	312	*/
	313	#define K8_NBSL_PP_SRC 0x0
	314	#define K8_NBSL_PP_RES 0x1
	315	#define K8_NBSL_PP_OBS 0x2
	316	#define K8_NBSL_PP_GENERIC 0x3
	317
cfe40fdb	318	#define EXTRACT_ERR_CPU_MAP(x) ((x) & 0xF)
cfe40fdb	319
cfe40fdb DT	320	#define K8_NBEAL 0x50
	321	#define K8_NBEAH 0x54
	322	#define K8_SCRCTRL 0x58
	323
	324	#define F10_NB_CFG_LOW 0x88
	325	#define F10_NB_CFG_LOW_ENABLE_EXT_CFG BIT(14)
	326
	327	#define F10_NB_CFG_HIGH 0x8C
	328
	329	#define F10_ONLINE_SPARE 0xB0
	330	#define F10_ONLINE_SPARE_SWAPDONE0(x) ((x) & BIT(1))
	331	#define F10_ONLINE_SPARE_SWAPDONE1(x) ((x) & BIT(3))
	332	#define F10_ONLINE_SPARE_BADDRAM_CS0(x) (((x) >> 4) & 0x00000007)
	333	#define F10_ONLINE_SPARE_BADDRAM_CS1(x) (((x) >> 8) & 0x00000007)
	334
	335	#define F10_NB_ARRAY_ADDR 0xB8
	336
	337	#define F10_NB_ARRAY_DRAM_ECC 0x80000000
	338
	339	/* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
	340	#define SET_NB_ARRAY_ADDRESS(section) (((section) & 0x3) << 1)
	341
	342	#define F10_NB_ARRAY_DATA 0xBC
	343
	344	#define SET_NB_DRAM_INJECTION_WRITE(word, bits) \
	345	(BIT(((word) & 0xF) + 20) \| \
94baaee4	346	BIT(17) \| bits)
cfe40fdb DT	347
	348	#define SET_NB_DRAM_INJECTION_READ(word, bits) \
	349	(BIT(((word) & 0xF) + 20) \| \
94baaee4	350	BIT(16) \| bits)
cfe40fdb DT	351
	352	#define K8_NBCAP 0xE8
	353	#define K8_NBCAP_CORES (BIT(12)\|BIT(13))
	354	#define K8_NBCAP_CHIPKILL BIT(4)
	355	#define K8_NBCAP_SECDED BIT(3)
cfe40fdb DT	356	#define K8_NBCAP_DCT_DUAL BIT(0)
cfe40fdb DT	357
ad6a32e9 BP	358	#define EXT_NB_MCA_CFG 0x180
ad6a32e9 BP	359
f6d6ae96	360	/* MSRs */
cfe40fdb DT	361	#define K8_MSR_MCGCTL_NBE BIT(4)
	362
	363	#define K8_MSR_MC4CTL 0x0410
	364	#define K8_MSR_MC4STAT 0x0411
	365	#define K8_MSR_MC4ADDR 0x0412
	366
	367	/* AMD sets the first MC device at device ID 0x18. */
37da0450	368	static inline int get_node_id(struct pci_dev *pdev)
cfe40fdb DT	369	{
	370	return PCI_SLOT(pdev->devfn) - 0x18;
	371	}
	372
	373	enum amd64_chipset_families {
	374	K8_CPUS = 0,
	375	F10_CPUS,
	376	F11_CPUS,
	377	};
	378
cfe40fdb DT	379	/* Error injection control structure */
	380	struct error_injection {
	381	u32 section;
	382	u32 word;
	383	u32 bit_map;
	384	};
	385
	386	struct amd64_pvt {
	387	/* pci_device handles which we utilize */
	388	struct pci_dev *addr_f1_ctl;
	389	struct pci_dev *dram_f2_ctl;
	390	struct pci_dev *misc_f3_ctl;
	391
	392	int mc_node_id; /* MC index of this MC node */
	393	int ext_model; /* extended model value of this node */
	394
	395	struct low_ops ops; / pointer to per PCI Device ID func table */
	396
	397	int channel_count;
	398
	399	/* Raw registers */
	400	u32 dclr0; /* DRAM Configuration Low DCT0 reg */
	401	u32 dclr1; /* DRAM Configuration Low DCT1 reg */
	402	u32 dchr0; /* DRAM Configuration High DCT0 reg */
	403	u32 dchr1; /* DRAM Configuration High DCT1 reg */
	404	u32 nbcap; /* North Bridge Capabilities */
	405	u32 nbcfg; /* F10 North Bridge Configuration */
	406	u32 ext_nbcfg; /* Extended F10 North Bridge Configuration */
	407	u32 dhar; /* DRAM Hoist reg */
	408	u32 dbam0; /* DRAM Base Address Mapping reg for DCT0 */
	409	u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */
	410
	411	/* DRAM CS Base Address Registers F2x[1,0][5C:40] */
9d858bb1 BP	412	u32 dcsb0[MAX_CS_COUNT];
9d858bb1 BP	413	u32 dcsb1[MAX_CS_COUNT];
cfe40fdb DT	414
cfe40fdb DT	415	/* DRAM CS Mask Registers F2x[1,0][6C:60] */
9d858bb1 BP	416	u32 dcsm0[MAX_CS_COUNT];
9d858bb1 BP	417	u32 dcsm1[MAX_CS_COUNT];
cfe40fdb DT	418
	419	/*
	420	* Decoded parts of DRAM BASE and LIMIT Registers
	421	* F1x[78,70,68,60,58,50,48,40]
	422	*/
	423	u64 dram_base[DRAM_REG_COUNT];
	424	u64 dram_limit[DRAM_REG_COUNT];
	425	u8 dram_IntlvSel[DRAM_REG_COUNT];
	426	u8 dram_IntlvEn[DRAM_REG_COUNT];
	427	u8 dram_DstNode[DRAM_REG_COUNT];
	428	u8 dram_rw_en[DRAM_REG_COUNT];
	429
	430	/*
	431	* The following fields are set at (load) run time, after CPU revision
	432	* has been determined, since the dct_base and dct_mask registers vary
	433	* based on revision
	434	*/
	435	u32 dcsb_base; /* DCSB base bits */
	436	u32 dcsm_mask; /* DCSM mask bits */
9d858bb1	437	u32 cs_count; /* num chip selects (== num DCSB registers) */
cfe40fdb DT	438	u32 num_dcsm; /* Number of DCSM registers */
	439	u32 dcs_mask_notused; /* DCSM notused mask bits */
	440	u32 dcs_shift; /* DCSB and DCSM shift value */
	441
	442	u64 top_mem; /* top of memory below 4GB */
	443	u64 top_mem2; /* top of memory above 4GB */
	444
	445	u32 dram_ctl_select_low; /* DRAM Controller Select Low Reg */
	446	u32 dram_ctl_select_high; /* DRAM Controller Select High Reg */
	447	u32 online_spare; /* On-Line spare Reg */
	448
ad6a32e9 BP	449	/* x4 or x8 syndromes in use */
	450	u8 syn_type;
	451
cfe40fdb	452	/* temp storage for when input is received from sysfs */
ef44cc4c	453	struct err_regs ctl_error_info;
cfe40fdb DT	454
	455	/* place to store error injection parameters prior to issue */
	456	struct error_injection injection;
	457
	458	/* Save old hw registers' values before we modified them */
	459	u32 nbctl_mcgctl_saved; /* When true, following 2 are valid */
	460	u32 old_nbctl;
cfe40fdb DT	461
	462	/* MC Type Index value: socket F vs Family 10h */
	463	u32 mc_type_index;
	464
	465	/* misc settings */
	466	struct flags {
	467	unsigned long cf8_extcfg:1;
d95cf4de BP	468	unsigned long nb_mce_enable:1;
d95cf4de BP	469	unsigned long nb_ecc_prev:1;
cfe40fdb DT	470	} flags;
	471	};
	472
	473	struct scrubrate {
	474	u32 scrubval; /* bit pattern for scrub rate */
	475	u32 bandwidth; /* bandwidth consumed (bytes/sec) */
	476	};
	477
	478	extern struct scrubrate scrubrates[23];
cfe40fdb DT	479	extern const char *tt_msgs[4];
	480	extern const char *ll_msgs[4];
	481	extern const char *rrrr_msgs[16];
	482	extern const char *to_msgs[2];
	483	extern const char *pp_msgs[4];
	484	extern const char *ii_msgs[4];
cfe40fdb DT	485	extern const char *htlink_msgs[8];
cfe40fdb DT	486
7d6034d3	487	#ifdef CONFIG_EDAC_DEBUG
9cdeb404	488	#define NUM_DBG_ATTRS 5
7d6034d3 DT	489	#else
	490	#define NUM_DBG_ATTRS 0
	491	#endif
	492
	493	#ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
	494	#define NUM_INJ_ATTRS 5
	495	#else
	496	#define NUM_INJ_ATTRS 0
	497	#endif
	498
	499	extern struct mcidev_sysfs_attribute amd64_dbg_attrs[NUM_DBG_ATTRS],
	500	amd64_inj_attrs[NUM_INJ_ATTRS];
	501
cfe40fdb DT	502	/*
	503	* Each of the PCI Device IDs types have their own set of hardware accessor
	504	* functions and per device encoding/decoding logic.
	505	*/
	506	struct low_ops {
1433eb99 BP	507	int (early_channel_count) (struct amd64_pvt pvt);
	508
	509	u64 (get_error_address) (struct mem_ctl_info mci,
	510	struct err_regs *info);
	511	void (read_dram_base_limit) (struct amd64_pvt pvt, int dram);
	512	void (read_dram_ctl_register) (struct amd64_pvt pvt);
	513	void (map_sysaddr_to_csrow) (struct mem_ctl_info mci,
	514	struct err_regs *info, u64 SystemAddr);
	515	int (dbam_to_cs) (struct amd64_pvt pvt, int cs_mode);
cfe40fdb DT	516	};
	517
	518	struct amd64_family_type {
	519	const char *ctl_name;
	520	u16 addr_f1_ctl;
	521	u16 misc_f3_ctl;
	522	struct low_ops ops;
	523	};
	524
	525	static struct amd64_family_type amd64_family_types[];
	526
	527	static inline const char *get_amd_family_name(int index)
	528	{
	529	return amd64_family_types[index].ctl_name;
	530	}
	531
	532	static inline struct low_ops *family_ops(int index)
	533	{
	534	return &amd64_family_types[index].ops;
	535	}
	536
6ba5dcdc BP	537	static inline int amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
	538	u32 val, const char func)
	539	{
	540	int err = 0;
	541
	542	err = pci_read_config_dword(pdev, offset, val);
	543	if (err)
	544	amd64_printk(KERN_WARNING, "%s: error reading F%dx%x.\n",
	545	func, PCI_FUNC(pdev->devfn), offset);
	546
	547	return err;
	548	}
	549
	550	#define amd64_read_pci_cfg(pdev, offset, val) \
	551	amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
	552
cfe40fdb DT	553	/*
	554	* For future CPU versions, verify the following as new 'slow' rates appear and
	555	* modify the necessary skip values for the supported CPU.
	556	*/
	557	#define K8_MIN_SCRUB_RATE_BITS 0x0
	558	#define F10_MIN_SCRUB_RATE_BITS 0x5
	559	#define F11_MIN_SCRUB_RATE_BITS 0x6
	560
cfe40fdb DT	561	int amd64_get_dram_hole_info(struct mem_ctl_info mci, u64 hole_base,
cfe40fdb DT	562	u64 hole_offset, u64 hole_size);