2 * AMD64 class Memory Controller kernel module
4 * Copyright (c) 2009 SoftwareBitMaker.
5 * Copyright (c) 2009 Advanced Micro Devices, Inc.
7 * This file may be distributed under the terms of the
8 * GNU General Public License.
10 * Originally Written by Thayne Harbaugh
12 * Changes by Douglas "norsk" Thompson <dougthompson@xmission.com>:
13 * - K8 CPU Revision D and greater support
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
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
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.
36 * Changes/Fixes by Borislav Petkov <borislav.petkov@amd.com>:
37 * - misc fixes and code cleanups
39 * This module is based on the following documents
40 * (available from http://www.amd.com/):
42 * Title: BIOS and Kernel Developer's Guide for AMD Athlon 64 and AMD
44 * AMD publication #: 26094
47 * Title: BIOS and Kernel Developer's Guide for AMD NPT Family 0Fh
49 * AMD publication #: 32559
51 * Issue Date: May 2006
53 * Title: BIOS and Kernel Developer's Guide (BKDG) For AMD Family 10h
55 * AMD publication #: 31116
57 * Issue Date: September 07, 2007
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
62 * Therefore, comments that refer to a Document section might be off.
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>
74 #include "edac_core.h"
77 #define amd64_debug(fmt, arg...) \
78 edac_printk(KERN_DEBUG, "amd64", fmt, ##arg)
80 #define amd64_info(fmt, arg...) \
81 edac_printk(KERN_INFO, "amd64", fmt, ##arg)
83 #define amd64_notice(fmt, arg...) \
84 edac_printk(KERN_NOTICE, "amd64", fmt, ##arg)
86 #define amd64_warn(fmt, arg...) \
87 edac_printk(KERN_WARNING, "amd64", fmt, ##arg)
89 #define amd64_err(fmt, arg...) \
90 edac_printk(KERN_ERR, "amd64", fmt, ##arg)
92 #define amd64_mc_warn(mci, fmt, arg...) \
93 edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)
95 #define amd64_mc_err(mci, fmt, arg...) \
96 edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
99 * Throughout the comments in this code, the following terms are used:
101 * SysAddr, DramAddr, and InputAddr
103 * These terms come directly from the amd64 documentation
104 * (AMD publication #26094). They are defined as follows:
107 * This is a physical address generated by a CPU core or a device
108 * doing DMA. If generated by a CPU core, a SysAddr is the result of
109 * a virtual to physical address translation by the CPU core's address
110 * translation mechanism (MMU).
113 * A DramAddr is derived from a SysAddr by subtracting an offset that
114 * depends on which node the SysAddr maps to and whether the SysAddr
115 * is within a range affected by memory hoisting. The DRAM Base
116 * (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
117 * determine which node a SysAddr maps to.
119 * If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
120 * is within the range of addresses specified by this register, then
121 * a value x from the DHAR is subtracted from the SysAddr to produce a
122 * DramAddr. Here, x represents the base address for the node that
123 * the SysAddr maps to plus an offset due to memory hoisting. See
124 * section 3.4.8 and the comments in amd64_get_dram_hole_info() and
125 * sys_addr_to_dram_addr() below for more information.
127 * If the SysAddr is not affected by the DHAR then a value y is
128 * subtracted from the SysAddr to produce a DramAddr. Here, y is the
129 * base address for the node that the SysAddr maps to. See section
130 * 3.4.4 and the comments in sys_addr_to_dram_addr() below for more
134 * A DramAddr is translated to an InputAddr before being passed to the
135 * memory controller for the node that the DramAddr is associated
136 * with. The memory controller then maps the InputAddr to a csrow.
137 * If node interleaving is not in use, then the InputAddr has the same
138 * value as the DramAddr. Otherwise, the InputAddr is produced by
139 * discarding the bits used for node interleaving from the DramAddr.
140 * See section 3.4.4 for more information.
142 * The memory controller for a given node uses its DRAM CS Base and
143 * DRAM CS Mask registers to map an InputAddr to a csrow. See
144 * sections 3.5.4 and 3.5.5 for more information.
147 #define EDAC_AMD64_VERSION "v3.3.0"
148 #define EDAC_MOD_STR "amd64_edac"
150 /* Extended Model from CPUID, for CPU Revision numbers */
155 /* Hardware limit on ChipSelect rows per MC and processors per system */
156 #define MAX_CS_COUNT 8
157 #define DRAM_REG_COUNT 8
163 * PCI-defined configuration space registers
168 * Function 1 - Address Map
170 #define K8_DRAM_BASE_LOW 0x40
171 #define K8_DRAM_LIMIT_LOW 0x44
174 #define DHAR_VALID BIT(0)
175 #define F10_DRAM_MEM_HOIST_VALID BIT(1)
177 #define DHAR_BASE_MASK 0xff000000
178 #define dhar_base(dhar) (dhar & DHAR_BASE_MASK)
180 #define K8_DHAR_OFFSET_MASK 0x0000ff00
181 #define k8_dhar_offset(dhar) ((dhar & K8_DHAR_OFFSET_MASK) << 16)
183 #define F10_DHAR_OFFSET_MASK 0x0000ff80
184 /* NOTE: Extra mask bit vs K8 */
185 #define f10_dhar_offset(dhar) ((dhar & F10_DHAR_OFFSET_MASK) << 16)
188 /* F10 High BASE/LIMIT registers */
189 #define F10_DRAM_BASE_HIGH 0x140
190 #define F10_DRAM_LIMIT_HIGH 0x144
194 * Function 2 - DRAM controller
196 #define K8_DCSB0 0x40
197 #define F10_DCSB1 0x140
199 #define K8_DCSB_CS_ENABLE BIT(0)
200 #define K8_DCSB_NPT_SPARE BIT(1)
201 #define K8_DCSB_NPT_TESTFAIL BIT(2)
204 * REV E: select [31:21] and [15:9] from DCSB and the shift amount to form
207 #define REV_E_DCSB_BASE_BITS (0xFFE0FE00ULL)
208 #define REV_E_DCS_SHIFT 4
210 #define REV_F_F1Xh_DCSB_BASE_BITS (0x1FF83FE0ULL)
211 #define REV_F_F1Xh_DCS_SHIFT 8
214 * REV F and later: selects [28:19] and [13:5] from DCSB and the shift amount
215 * to form the address
217 #define REV_F_DCSB_BASE_BITS (0x1FF83FE0ULL)
218 #define REV_F_DCS_SHIFT 8
220 /* DRAM CS Mask Registers */
221 #define K8_DCSM0 0x60
222 #define F10_DCSM1 0x160
224 /* REV E: select [29:21] and [15:9] from DCSM */
225 #define REV_E_DCSM_MASK_BITS 0x3FE0FE00
227 /* unused bits [24:20] and [12:0] */
228 #define REV_E_DCS_NOTUSED_BITS 0x01F01FFF
230 /* REV F and later: select [28:19] and [13:5] from DCSM */
231 #define REV_F_F1Xh_DCSM_MASK_BITS 0x1FF83FE0
233 /* unused bits [26:22] and [12:0] */
234 #define REV_F_F1Xh_DCS_NOTUSED_BITS 0x07C01FFF
239 /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
240 #define DBAM_DIMM(i, reg) ((((reg) >> (4*i))) & 0xF)
242 #define DBAM_MAX_VALUE 11
245 #define F10_DCLR_0 0x90
246 #define F10_DCLR_1 0x190
247 #define REVE_WIDTH_128 BIT(16)
248 #define F10_WIDTH_128 BIT(11)
251 #define F10_DCHR_0 0x94
252 #define F10_DCHR_1 0x194
254 #define F10_DCHR_FOUR_RANK_DIMM BIT(18)
255 #define DDR3_MODE BIT(8)
256 #define F10_DCHR_MblMode BIT(6)
259 #define F10_DCTL_SEL_LOW 0x110
260 #define dct_sel_baseaddr(pvt) ((pvt->dram_ctl_select_low) & 0xFFFFF800)
261 #define dct_sel_interleave_addr(pvt) (((pvt->dram_ctl_select_low) >> 6) & 0x3)
262 #define dct_high_range_enabled(pvt) (pvt->dram_ctl_select_low & BIT(0))
263 #define dct_interleave_enabled(pvt) (pvt->dram_ctl_select_low & BIT(2))
264 #define dct_ganging_enabled(pvt) (pvt->dram_ctl_select_low & BIT(4))
265 #define dct_data_intlv_enabled(pvt) (pvt->dram_ctl_select_low & BIT(5))
266 #define dct_dram_enabled(pvt) (pvt->dram_ctl_select_low & BIT(8))
267 #define dct_memory_cleared(pvt) (pvt->dram_ctl_select_low & BIT(10))
269 #define F10_DCTL_SEL_HIGH 0x114
272 * Function 3 - Misc Control
274 #define K8_NBCTL 0x40
276 /* Correctable ECC error reporting enable */
277 #define K8_NBCTL_CECCEn BIT(0)
279 /* UnCorrectable ECC error reporting enable */
280 #define K8_NBCTL_UECCEn BIT(1)
282 #define K8_NBCFG 0x44
283 #define K8_NBCFG_CHIPKILL BIT(23)
284 #define K8_NBCFG_ECC_ENABLE BIT(22)
289 /* Family F10h: Normalized Extended Error Codes */
290 #define F10_NBSL_EXT_ERR_RES 0x0
291 #define F10_NBSL_EXT_ERR_ECC 0x8
293 /* Next two are overloaded values */
294 #define F10_NBSL_EXT_ERR_LINK_PROTO 0xB
295 #define F10_NBSL_EXT_ERR_L3_PROTO 0xB
297 #define F10_NBSL_EXT_ERR_NB_ARRAY 0xC
298 #define F10_NBSL_EXT_ERR_DRAM_PARITY 0xD
299 #define F10_NBSL_EXT_ERR_LINK_RETRY 0xE
301 /* Next two are overloaded values */
302 #define F10_NBSL_EXT_ERR_GART_WALK 0xF
303 #define F10_NBSL_EXT_ERR_DEV_WALK 0xF
305 /* 0x10 to 0x1B: Reserved */
306 #define F10_NBSL_EXT_ERR_L3_DATA 0x1C
307 #define F10_NBSL_EXT_ERR_L3_TAG 0x1D
308 #define F10_NBSL_EXT_ERR_L3_LRU 0x1E
310 /* K8: Normalized Extended Error Codes */
311 #define K8_NBSL_EXT_ERR_ECC 0x0
312 #define K8_NBSL_EXT_ERR_CRC 0x1
313 #define K8_NBSL_EXT_ERR_SYNC 0x2
314 #define K8_NBSL_EXT_ERR_MST 0x3
315 #define K8_NBSL_EXT_ERR_TGT 0x4
316 #define K8_NBSL_EXT_ERR_GART 0x5
317 #define K8_NBSL_EXT_ERR_RMW 0x6
318 #define K8_NBSL_EXT_ERR_WDT 0x7
319 #define K8_NBSL_EXT_ERR_CHIPKILL_ECC 0x8
320 #define K8_NBSL_EXT_ERR_DRAM_PARITY 0xD
323 * The following are for BUS type errors AFTER values have been normalized by
326 #define K8_NBSL_PP_SRC 0x0
327 #define K8_NBSL_PP_RES 0x1
328 #define K8_NBSL_PP_OBS 0x2
329 #define K8_NBSL_PP_GENERIC 0x3
331 #define EXTRACT_ERR_CPU_MAP(x) ((x) & 0xF)
333 #define K8_NBEAL 0x50
334 #define K8_NBEAH 0x54
335 #define K8_SCRCTRL 0x58
337 #define F10_NB_CFG_LOW 0x88
339 #define F10_ONLINE_SPARE 0xB0
340 #define F10_ONLINE_SPARE_SWAPDONE0(x) ((x) & BIT(1))
341 #define F10_ONLINE_SPARE_SWAPDONE1(x) ((x) & BIT(3))
342 #define F10_ONLINE_SPARE_BADDRAM_CS0(x) (((x) >> 4) & 0x00000007)
343 #define F10_ONLINE_SPARE_BADDRAM_CS1(x) (((x) >> 8) & 0x00000007)
345 #define F10_NB_ARRAY_ADDR 0xB8
347 #define F10_NB_ARRAY_DRAM_ECC 0x80000000
349 /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
350 #define SET_NB_ARRAY_ADDRESS(section) (((section) & 0x3) << 1)
352 #define F10_NB_ARRAY_DATA 0xBC
354 #define SET_NB_DRAM_INJECTION_WRITE(word, bits) \
355 (BIT(((word) & 0xF) + 20) | \
358 #define SET_NB_DRAM_INJECTION_READ(word, bits) \
359 (BIT(((word) & 0xF) + 20) | \
362 #define K8_NBCAP 0xE8
363 #define K8_NBCAP_CORES (BIT(12)|BIT(13))
364 #define K8_NBCAP_CHIPKILL BIT(4)
365 #define K8_NBCAP_SECDED BIT(3)
366 #define K8_NBCAP_DCT_DUAL BIT(0)
368 #define EXT_NB_MCA_CFG 0x180
371 #define K8_MSR_MCGCTL_NBE BIT(4)
373 #define K8_MSR_MC4CTL 0x0410
374 #define K8_MSR_MC4STAT 0x0411
375 #define K8_MSR_MC4ADDR 0x0412
377 /* AMD sets the first MC device at device ID 0x18. */
378 static inline int get_node_id(struct pci_dev
*pdev
)
380 return PCI_SLOT(pdev
->devfn
) - 0x18;
383 enum amd64_chipset_families
{
388 /* Error injection control structure */
389 struct error_injection
{
398 /* pci_device handles which we utilize */
399 struct pci_dev
*F1
, *F2
, *F3
;
401 int mc_node_id
; /* MC index of this MC node */
402 int ext_model
; /* extended model value of this node */
406 u32 dclr0
; /* DRAM Configuration Low DCT0 reg */
407 u32 dclr1
; /* DRAM Configuration Low DCT1 reg */
408 u32 dchr0
; /* DRAM Configuration High DCT0 reg */
409 u32 dchr1
; /* DRAM Configuration High DCT1 reg */
410 u32 nbcap
; /* North Bridge Capabilities */
411 u32 nbcfg
; /* F10 North Bridge Configuration */
412 u32 ext_nbcfg
; /* Extended F10 North Bridge Configuration */
413 u32 dhar
; /* DRAM Hoist reg */
414 u32 dbam0
; /* DRAM Base Address Mapping reg for DCT0 */
415 u32 dbam1
; /* DRAM Base Address Mapping reg for DCT1 */
417 /* DRAM CS Base Address Registers F2x[1,0][5C:40] */
418 u32 dcsb0
[MAX_CS_COUNT
];
419 u32 dcsb1
[MAX_CS_COUNT
];
421 /* DRAM CS Mask Registers F2x[1,0][6C:60] */
422 u32 dcsm0
[MAX_CS_COUNT
];
423 u32 dcsm1
[MAX_CS_COUNT
];
426 * Decoded parts of DRAM BASE and LIMIT Registers
427 * F1x[78,70,68,60,58,50,48,40]
429 u64 dram_base
[DRAM_REG_COUNT
];
430 u64 dram_limit
[DRAM_REG_COUNT
];
431 u8 dram_IntlvSel
[DRAM_REG_COUNT
];
432 u8 dram_IntlvEn
[DRAM_REG_COUNT
];
433 u8 dram_DstNode
[DRAM_REG_COUNT
];
434 u8 dram_rw_en
[DRAM_REG_COUNT
];
437 * The following fields are set at (load) run time, after CPU revision
438 * has been determined, since the dct_base and dct_mask registers vary
441 u32 dcsb_base
; /* DCSB base bits */
442 u32 dcsm_mask
; /* DCSM mask bits */
443 u32 cs_count
; /* num chip selects (== num DCSB registers) */
444 u32 num_dcsm
; /* Number of DCSM registers */
445 u32 dcs_mask_notused
; /* DCSM notused mask bits */
446 u32 dcs_shift
; /* DCSB and DCSM shift value */
448 u64 top_mem
; /* top of memory below 4GB */
449 u64 top_mem2
; /* top of memory above 4GB */
451 u32 dram_ctl_select_low
; /* DRAM Controller Select Low Reg */
452 u32 dram_ctl_select_high
; /* DRAM Controller Select High Reg */
453 u32 online_spare
; /* On-Line spare Reg */
455 /* x4 or x8 syndromes in use */
458 /* temp storage for when input is received from sysfs */
459 struct err_regs ctl_error_info
;
461 /* place to store error injection parameters prior to issue */
462 struct error_injection injection
;
464 /* DCT per-family scrubrate setting */
467 /* family name this instance is running on */
468 const char *ctl_name
;
473 * per-node ECC settings descriptor
475 struct ecc_settings
{
480 unsigned long nb_mce_enable
:1;
481 unsigned long nb_ecc_prev
:1;
485 extern const char *tt_msgs
[4];
486 extern const char *ll_msgs
[4];
487 extern const char *rrrr_msgs
[16];
488 extern const char *to_msgs
[2];
489 extern const char *pp_msgs
[4];
490 extern const char *ii_msgs
[4];
491 extern const char *htlink_msgs
[8];
493 #ifdef CONFIG_EDAC_DEBUG
494 #define NUM_DBG_ATTRS 5
496 #define NUM_DBG_ATTRS 0
499 #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
500 #define NUM_INJ_ATTRS 5
502 #define NUM_INJ_ATTRS 0
505 extern struct mcidev_sysfs_attribute amd64_dbg_attrs
[NUM_DBG_ATTRS
],
506 amd64_inj_attrs
[NUM_INJ_ATTRS
];
509 * Each of the PCI Device IDs types have their own set of hardware accessor
510 * functions and per device encoding/decoding logic.
513 int (*early_channel_count
) (struct amd64_pvt
*pvt
);
515 u64 (*get_error_address
) (struct mem_ctl_info
*mci
,
516 struct err_regs
*info
);
517 void (*read_dram_base_limit
) (struct amd64_pvt
*pvt
, int dram
);
518 void (*read_dram_ctl_register
) (struct amd64_pvt
*pvt
);
519 void (*map_sysaddr_to_csrow
) (struct mem_ctl_info
*mci
,
520 struct err_regs
*info
, u64 SystemAddr
);
521 int (*dbam_to_cs
) (struct amd64_pvt
*pvt
, int cs_mode
);
524 struct amd64_family_type
{
525 const char *ctl_name
;
530 static inline int amd64_read_pci_cfg_dword(struct pci_dev
*pdev
, int offset
,
531 u32
*val
, const char *func
)
535 err
= pci_read_config_dword(pdev
, offset
, val
);
537 amd64_warn("%s: error reading F%dx%x.\n",
538 func
, PCI_FUNC(pdev
->devfn
), offset
);
543 #define amd64_read_pci_cfg(pdev, offset, val) \
544 amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
547 * For future CPU versions, verify the following as new 'slow' rates appear and
548 * modify the necessary skip values for the supported CPU.
550 #define K8_MIN_SCRUB_RATE_BITS 0x0
551 #define F10_MIN_SCRUB_RATE_BITS 0x5
553 int amd64_get_dram_hole_info(struct mem_ctl_info
*mci
, u64
*hole_base
,
554 u64
*hole_offset
, u64
*hole_size
);