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 NUM_CHIPSELECTS 8
157 #define DRAM_RANGES 8
163 * Create a contiguous bitmask starting at bit position @lo and ending at
164 * position @hi. For example
166 * GENMASK(21, 39) gives us the 64bit vector 0x000000ffffe00000.
168 #define GENMASK(lo, hi) (((1ULL << ((hi) - (lo) + 1)) - 1) << (lo))
171 * PCI-defined configuration space registers
176 * Function 1 - Address Map
178 #define DRAM_BASE_LO 0x40
179 #define DRAM_LIMIT_LO 0x44
181 #define dram_intlv_en(pvt, i) ((pvt->ranges[i].base.lo >> 8) & 0x7)
182 #define dram_rw(pvt, i) (pvt->ranges[i].base.lo & 0x3)
183 #define dram_intlv_sel(pvt, i) ((pvt->ranges[i].lim.lo >> 8) & 0x7)
184 #define dram_dst_node(pvt, i) (pvt->ranges[i].lim.lo & 0x7)
187 #define dhar_valid(pvt) ((pvt)->dhar & BIT(0))
188 #define dhar_mem_hoist_valid(pvt) ((pvt)->dhar & BIT(1))
189 #define dhar_base(pvt) ((pvt)->dhar & 0xff000000)
190 #define k8_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff00) << 16)
192 /* NOTE: Extra mask bit vs K8 */
193 #define f10_dhar_offset(pvt) (((pvt)->dhar & 0x0000ff80) << 16)
195 #define DCT_CFG_SEL 0x10C
197 #define DRAM_BASE_HI 0x140
198 #define DRAM_LIMIT_HI 0x144
202 * Function 2 - DRAM controller
206 #define DCSB_CS_ENABLE BIT(0)
211 #define csrow_enabled(i, dct, pvt) ((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
216 /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
217 #define DBAM_DIMM(i, reg) ((((reg) >> (4*i))) & 0xF)
219 #define DBAM_MAX_VALUE 11
223 #define REVE_WIDTH_128 BIT(16)
224 #define F10_WIDTH_128 BIT(11)
228 #define DDR3_MODE BIT(8)
230 #define DCT_SEL_LO 0x110
231 #define dct_sel_baseaddr(pvt) ((pvt)->dct_sel_lo & 0xFFFFF800)
232 #define dct_sel_interleave_addr(pvt) (((pvt)->dct_sel_lo >> 6) & 0x3)
233 #define dct_high_range_enabled(pvt) ((pvt)->dct_sel_lo & BIT(0))
234 #define dct_interleave_enabled(pvt) ((pvt)->dct_sel_lo & BIT(2))
236 #define dct_ganging_enabled(pvt) ((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
238 #define dct_data_intlv_enabled(pvt) ((pvt)->dct_sel_lo & BIT(5))
239 #define dct_dram_enabled(pvt) ((pvt)->dct_sel_lo & BIT(8))
240 #define dct_memory_cleared(pvt) ((pvt)->dct_sel_lo & BIT(10))
242 #define DCT_SEL_HI 0x114
245 * Function 3 - Misc Control
249 #define K8_NBCFG 0x44
250 #define K8_NBCFG_CHIPKILL BIT(23)
251 #define K8_NBCFG_ECC_ENABLE BIT(22)
255 /* Family F10h: Normalized Extended Error Codes */
256 #define F10_NBSL_EXT_ERR_RES 0x0
257 #define F10_NBSL_EXT_ERR_ECC 0x8
259 /* Next two are overloaded values */
260 #define F10_NBSL_EXT_ERR_LINK_PROTO 0xB
261 #define F10_NBSL_EXT_ERR_L3_PROTO 0xB
263 #define F10_NBSL_EXT_ERR_NB_ARRAY 0xC
264 #define F10_NBSL_EXT_ERR_DRAM_PARITY 0xD
265 #define F10_NBSL_EXT_ERR_LINK_RETRY 0xE
267 /* Next two are overloaded values */
268 #define F10_NBSL_EXT_ERR_GART_WALK 0xF
269 #define F10_NBSL_EXT_ERR_DEV_WALK 0xF
271 /* 0x10 to 0x1B: Reserved */
272 #define F10_NBSL_EXT_ERR_L3_DATA 0x1C
273 #define F10_NBSL_EXT_ERR_L3_TAG 0x1D
274 #define F10_NBSL_EXT_ERR_L3_LRU 0x1E
276 /* K8: Normalized Extended Error Codes */
277 #define K8_NBSL_EXT_ERR_ECC 0x0
278 #define K8_NBSL_EXT_ERR_CRC 0x1
279 #define K8_NBSL_EXT_ERR_SYNC 0x2
280 #define K8_NBSL_EXT_ERR_MST 0x3
281 #define K8_NBSL_EXT_ERR_TGT 0x4
282 #define K8_NBSL_EXT_ERR_GART 0x5
283 #define K8_NBSL_EXT_ERR_RMW 0x6
284 #define K8_NBSL_EXT_ERR_WDT 0x7
285 #define K8_NBSL_EXT_ERR_CHIPKILL_ECC 0x8
286 #define K8_NBSL_EXT_ERR_DRAM_PARITY 0xD
289 * The following are for BUS type errors AFTER values have been normalized by
292 #define K8_NBSL_PP_SRC 0x0
293 #define K8_NBSL_PP_RES 0x1
294 #define K8_NBSL_PP_OBS 0x2
295 #define K8_NBSL_PP_GENERIC 0x3
297 #define EXTRACT_ERR_CPU_MAP(x) ((x) & 0xF)
299 #define K8_NBEAL 0x50
300 #define K8_NBEAH 0x54
301 #define K8_SCRCTRL 0x58
303 #define F10_NB_CFG_LOW 0x88
305 #define F10_ONLINE_SPARE 0xB0
306 #define F10_ONLINE_SPARE_SWAPDONE0(x) ((x) & BIT(1))
307 #define F10_ONLINE_SPARE_SWAPDONE1(x) ((x) & BIT(3))
308 #define F10_ONLINE_SPARE_BADDRAM_CS0(x) (((x) >> 4) & 0x00000007)
309 #define F10_ONLINE_SPARE_BADDRAM_CS1(x) (((x) >> 8) & 0x00000007)
311 #define F10_NB_ARRAY_ADDR 0xB8
313 #define F10_NB_ARRAY_DRAM_ECC 0x80000000
315 /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
316 #define SET_NB_ARRAY_ADDRESS(section) (((section) & 0x3) << 1)
318 #define F10_NB_ARRAY_DATA 0xBC
320 #define SET_NB_DRAM_INJECTION_WRITE(word, bits) \
321 (BIT(((word) & 0xF) + 20) | \
324 #define SET_NB_DRAM_INJECTION_READ(word, bits) \
325 (BIT(((word) & 0xF) + 20) | \
328 #define K8_NBCAP 0xE8
329 #define K8_NBCAP_CORES (BIT(12)|BIT(13))
330 #define K8_NBCAP_CHIPKILL BIT(4)
331 #define K8_NBCAP_SECDED BIT(3)
332 #define K8_NBCAP_DCT_DUAL BIT(0)
334 #define EXT_NB_MCA_CFG 0x180
337 #define K8_MSR_MCGCTL_NBE BIT(4)
339 #define K8_MSR_MC4CTL 0x0410
340 #define K8_MSR_MC4STAT 0x0411
341 #define K8_MSR_MC4ADDR 0x0412
343 /* AMD sets the first MC device at device ID 0x18. */
344 static inline int get_node_id(struct pci_dev
*pdev
)
346 return PCI_SLOT(pdev
->devfn
) - 0x18;
356 /* Error injection control structure */
357 struct error_injection
{
363 /* low and high part of PCI config space regs */
369 * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
372 struct reg_pair base
;
376 /* A DCT chip selects collection */
378 u32 csbases
[NUM_CHIPSELECTS
];
381 u32 csmasks
[NUM_CHIPSELECTS
];
388 /* pci_device handles which we utilize */
389 struct pci_dev
*F1
, *F2
, *F3
;
391 int mc_node_id
; /* MC index of this MC node */
392 int ext_model
; /* extended model value of this node */
396 u32 dclr0
; /* DRAM Configuration Low DCT0 reg */
397 u32 dclr1
; /* DRAM Configuration Low DCT1 reg */
398 u32 dchr0
; /* DRAM Configuration High DCT0 reg */
399 u32 dchr1
; /* DRAM Configuration High DCT1 reg */
400 u32 nbcap
; /* North Bridge Capabilities */
401 u32 nbcfg
; /* F10 North Bridge Configuration */
402 u32 ext_nbcfg
; /* Extended F10 North Bridge Configuration */
403 u32 dhar
; /* DRAM Hoist reg */
404 u32 dbam0
; /* DRAM Base Address Mapping reg for DCT0 */
405 u32 dbam1
; /* DRAM Base Address Mapping reg for DCT1 */
407 /* one for each DCT */
408 struct chip_select csels
[2];
410 /* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
411 struct dram_range ranges
[DRAM_RANGES
];
413 u64 top_mem
; /* top of memory below 4GB */
414 u64 top_mem2
; /* top of memory above 4GB */
416 u32 dct_sel_lo
; /* DRAM Controller Select Low */
417 u32 dct_sel_hi
; /* DRAM Controller Select High */
418 u32 online_spare
; /* On-Line spare Reg */
420 /* x4 or x8 syndromes in use */
423 /* temp storage for when input is received from sysfs */
424 struct err_regs ctl_error_info
;
426 /* place to store error injection parameters prior to issue */
427 struct error_injection injection
;
429 /* DCT per-family scrubrate setting */
432 /* family name this instance is running on */
433 const char *ctl_name
;
437 static inline u64
get_dram_base(struct amd64_pvt
*pvt
, unsigned i
)
439 u64 addr
= ((u64
)pvt
->ranges
[i
].base
.lo
& 0xffff0000) << 8;
441 if (boot_cpu_data
.x86
== 0xf)
444 return (((u64
)pvt
->ranges
[i
].base
.hi
& 0x000000ff) << 40) | addr
;
447 static inline u64
get_dram_limit(struct amd64_pvt
*pvt
, unsigned i
)
449 u64 lim
= (((u64
)pvt
->ranges
[i
].lim
.lo
& 0xffff0000) << 8) | 0x00ffffff;
451 if (boot_cpu_data
.x86
== 0xf)
454 return (((u64
)pvt
->ranges
[i
].lim
.hi
& 0x000000ff) << 40) | lim
;
458 * per-node ECC settings descriptor
460 struct ecc_settings
{
465 unsigned long nb_mce_enable
:1;
466 unsigned long nb_ecc_prev
:1;
470 extern const char *tt_msgs
[4];
471 extern const char *ll_msgs
[4];
472 extern const char *rrrr_msgs
[16];
473 extern const char *to_msgs
[2];
474 extern const char *pp_msgs
[4];
475 extern const char *ii_msgs
[4];
476 extern const char *htlink_msgs
[8];
478 #ifdef CONFIG_EDAC_DEBUG
479 #define NUM_DBG_ATTRS 5
481 #define NUM_DBG_ATTRS 0
484 #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
485 #define NUM_INJ_ATTRS 5
487 #define NUM_INJ_ATTRS 0
490 extern struct mcidev_sysfs_attribute amd64_dbg_attrs
[NUM_DBG_ATTRS
],
491 amd64_inj_attrs
[NUM_INJ_ATTRS
];
494 * Each of the PCI Device IDs types have their own set of hardware accessor
495 * functions and per device encoding/decoding logic.
498 int (*early_channel_count
) (struct amd64_pvt
*pvt
);
500 u64 (*get_error_address
) (struct mem_ctl_info
*mci
,
501 struct err_regs
*info
);
502 void (*read_dram_ctl_register
) (struct amd64_pvt
*pvt
);
503 void (*map_sysaddr_to_csrow
) (struct mem_ctl_info
*mci
,
504 struct err_regs
*info
, u64 SystemAddr
);
505 int (*dbam_to_cs
) (struct amd64_pvt
*pvt
, int cs_mode
);
506 int (*read_dct_pci_cfg
) (struct amd64_pvt
*pvt
, int offset
,
507 u32
*val
, const char *func
);
510 struct amd64_family_type
{
511 const char *ctl_name
;
516 int __amd64_write_pci_cfg_dword(struct pci_dev
*pdev
, int offset
,
517 u32 val
, const char *func
);
519 #define amd64_read_pci_cfg(pdev, offset, val) \
520 __amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
522 #define amd64_write_pci_cfg(pdev, offset, val) \
523 __amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
525 #define amd64_read_dct_pci_cfg(pvt, offset, val) \
526 pvt->ops->read_dct_pci_cfg(pvt, offset, val, __func__)
529 * For future CPU versions, verify the following as new 'slow' rates appear and
530 * modify the necessary skip values for the supported CPU.
532 #define K8_MIN_SCRUB_RATE_BITS 0x0
533 #define F10_MIN_SCRUB_RATE_BITS 0x5
535 int amd64_get_dram_hole_info(struct mem_ctl_info
*mci
, u64
*hole_base
,
536 u64
*hole_offset
, u64
*hole_size
);