amd64_edac: Rework printk macros
[pohmelfs.git] / drivers / edac / amd64_edac.h
blobf15e2b257e720d71f180f6ab71d98ecaba3c2e21
1 /*
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
20 * register
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
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
43 * Opteron Processors
44 * AMD publication #: 26094
45 *` Revision: 3.26
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
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
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.
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>
73 #include <asm/msr.h>
74 #include "edac_core.h"
75 #include "mce_amd.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:
106 * SysAddr:
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).
112 * DramAddr:
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
131 * information.
133 * InputAddr:
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 #define EDAC_MAX_NUMNODES 8
152 /* Extended Model from CPUID, for CPU Revision numbers */
153 #define K8_REV_D 1
154 #define K8_REV_E 2
155 #define K8_REV_F 4
157 /* Hardware limit on ChipSelect rows per MC and processors per system */
158 #define MAX_CS_COUNT 8
159 #define DRAM_REG_COUNT 8
161 #define ON true
162 #define OFF false
165 * PCI-defined configuration space registers
170 * Function 1 - Address Map
172 #define K8_DRAM_BASE_LOW 0x40
173 #define K8_DRAM_LIMIT_LOW 0x44
174 #define K8_DHAR 0xf0
176 #define DHAR_VALID BIT(0)
177 #define F10_DRAM_MEM_HOIST_VALID BIT(1)
179 #define DHAR_BASE_MASK 0xff000000
180 #define dhar_base(dhar) (dhar & DHAR_BASE_MASK)
182 #define K8_DHAR_OFFSET_MASK 0x0000ff00
183 #define k8_dhar_offset(dhar) ((dhar & K8_DHAR_OFFSET_MASK) << 16)
185 #define F10_DHAR_OFFSET_MASK 0x0000ff80
186 /* NOTE: Extra mask bit vs K8 */
187 #define f10_dhar_offset(dhar) ((dhar & F10_DHAR_OFFSET_MASK) << 16)
190 /* F10 High BASE/LIMIT registers */
191 #define F10_DRAM_BASE_HIGH 0x140
192 #define F10_DRAM_LIMIT_HIGH 0x144
196 * Function 2 - DRAM controller
198 #define K8_DCSB0 0x40
199 #define F10_DCSB1 0x140
201 #define K8_DCSB_CS_ENABLE BIT(0)
202 #define K8_DCSB_NPT_SPARE BIT(1)
203 #define K8_DCSB_NPT_TESTFAIL BIT(2)
206 * REV E: select [31:21] and [15:9] from DCSB and the shift amount to form
207 * the address
209 #define REV_E_DCSB_BASE_BITS (0xFFE0FE00ULL)
210 #define REV_E_DCS_SHIFT 4
212 #define REV_F_F1Xh_DCSB_BASE_BITS (0x1FF83FE0ULL)
213 #define REV_F_F1Xh_DCS_SHIFT 8
216 * REV F and later: selects [28:19] and [13:5] from DCSB and the shift amount
217 * to form the address
219 #define REV_F_DCSB_BASE_BITS (0x1FF83FE0ULL)
220 #define REV_F_DCS_SHIFT 8
222 /* DRAM CS Mask Registers */
223 #define K8_DCSM0 0x60
224 #define F10_DCSM1 0x160
226 /* REV E: select [29:21] and [15:9] from DCSM */
227 #define REV_E_DCSM_MASK_BITS 0x3FE0FE00
229 /* unused bits [24:20] and [12:0] */
230 #define REV_E_DCS_NOTUSED_BITS 0x01F01FFF
232 /* REV F and later: select [28:19] and [13:5] from DCSM */
233 #define REV_F_F1Xh_DCSM_MASK_BITS 0x1FF83FE0
235 /* unused bits [26:22] and [12:0] */
236 #define REV_F_F1Xh_DCS_NOTUSED_BITS 0x07C01FFF
238 #define DBAM0 0x80
239 #define DBAM1 0x180
241 /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
242 #define DBAM_DIMM(i, reg) ((((reg) >> (4*i))) & 0xF)
244 #define DBAM_MAX_VALUE 11
247 #define F10_DCLR_0 0x90
248 #define F10_DCLR_1 0x190
249 #define REVE_WIDTH_128 BIT(16)
250 #define F10_WIDTH_128 BIT(11)
253 #define F10_DCHR_0 0x94
254 #define F10_DCHR_1 0x194
256 #define F10_DCHR_FOUR_RANK_DIMM BIT(18)
257 #define DDR3_MODE BIT(8)
258 #define F10_DCHR_MblMode BIT(6)
261 #define F10_DCTL_SEL_LOW 0x110
262 #define dct_sel_baseaddr(pvt) ((pvt->dram_ctl_select_low) & 0xFFFFF800)
263 #define dct_sel_interleave_addr(pvt) (((pvt->dram_ctl_select_low) >> 6) & 0x3)
264 #define dct_high_range_enabled(pvt) (pvt->dram_ctl_select_low & BIT(0))
265 #define dct_interleave_enabled(pvt) (pvt->dram_ctl_select_low & BIT(2))
266 #define dct_ganging_enabled(pvt) (pvt->dram_ctl_select_low & BIT(4))
267 #define dct_data_intlv_enabled(pvt) (pvt->dram_ctl_select_low & BIT(5))
268 #define dct_dram_enabled(pvt) (pvt->dram_ctl_select_low & BIT(8))
269 #define dct_memory_cleared(pvt) (pvt->dram_ctl_select_low & BIT(10))
271 #define F10_DCTL_SEL_HIGH 0x114
274 * Function 3 - Misc Control
276 #define K8_NBCTL 0x40
278 /* Correctable ECC error reporting enable */
279 #define K8_NBCTL_CECCEn BIT(0)
281 /* UnCorrectable ECC error reporting enable */
282 #define K8_NBCTL_UECCEn BIT(1)
284 #define K8_NBCFG 0x44
285 #define K8_NBCFG_CHIPKILL BIT(23)
286 #define K8_NBCFG_ECC_ENABLE BIT(22)
288 #define K8_NBSL 0x48
291 /* Family F10h: Normalized Extended Error Codes */
292 #define F10_NBSL_EXT_ERR_RES 0x0
293 #define F10_NBSL_EXT_ERR_ECC 0x8
295 /* Next two are overloaded values */
296 #define F10_NBSL_EXT_ERR_LINK_PROTO 0xB
297 #define F10_NBSL_EXT_ERR_L3_PROTO 0xB
299 #define F10_NBSL_EXT_ERR_NB_ARRAY 0xC
300 #define F10_NBSL_EXT_ERR_DRAM_PARITY 0xD
301 #define F10_NBSL_EXT_ERR_LINK_RETRY 0xE
303 /* Next two are overloaded values */
304 #define F10_NBSL_EXT_ERR_GART_WALK 0xF
305 #define F10_NBSL_EXT_ERR_DEV_WALK 0xF
307 /* 0x10 to 0x1B: Reserved */
308 #define F10_NBSL_EXT_ERR_L3_DATA 0x1C
309 #define F10_NBSL_EXT_ERR_L3_TAG 0x1D
310 #define F10_NBSL_EXT_ERR_L3_LRU 0x1E
312 /* K8: Normalized Extended Error Codes */
313 #define K8_NBSL_EXT_ERR_ECC 0x0
314 #define K8_NBSL_EXT_ERR_CRC 0x1
315 #define K8_NBSL_EXT_ERR_SYNC 0x2
316 #define K8_NBSL_EXT_ERR_MST 0x3
317 #define K8_NBSL_EXT_ERR_TGT 0x4
318 #define K8_NBSL_EXT_ERR_GART 0x5
319 #define K8_NBSL_EXT_ERR_RMW 0x6
320 #define K8_NBSL_EXT_ERR_WDT 0x7
321 #define K8_NBSL_EXT_ERR_CHIPKILL_ECC 0x8
322 #define K8_NBSL_EXT_ERR_DRAM_PARITY 0xD
325 * The following are for BUS type errors AFTER values have been normalized by
326 * shifting right
328 #define K8_NBSL_PP_SRC 0x0
329 #define K8_NBSL_PP_RES 0x1
330 #define K8_NBSL_PP_OBS 0x2
331 #define K8_NBSL_PP_GENERIC 0x3
333 #define EXTRACT_ERR_CPU_MAP(x) ((x) & 0xF)
335 #define K8_NBEAL 0x50
336 #define K8_NBEAH 0x54
337 #define K8_SCRCTRL 0x58
339 #define F10_NB_CFG_LOW 0x88
340 #define F10_NB_CFG_LOW_ENABLE_EXT_CFG BIT(14)
342 #define F10_NB_CFG_HIGH 0x8C
344 #define F10_ONLINE_SPARE 0xB0
345 #define F10_ONLINE_SPARE_SWAPDONE0(x) ((x) & BIT(1))
346 #define F10_ONLINE_SPARE_SWAPDONE1(x) ((x) & BIT(3))
347 #define F10_ONLINE_SPARE_BADDRAM_CS0(x) (((x) >> 4) & 0x00000007)
348 #define F10_ONLINE_SPARE_BADDRAM_CS1(x) (((x) >> 8) & 0x00000007)
350 #define F10_NB_ARRAY_ADDR 0xB8
352 #define F10_NB_ARRAY_DRAM_ECC 0x80000000
354 /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline */
355 #define SET_NB_ARRAY_ADDRESS(section) (((section) & 0x3) << 1)
357 #define F10_NB_ARRAY_DATA 0xBC
359 #define SET_NB_DRAM_INJECTION_WRITE(word, bits) \
360 (BIT(((word) & 0xF) + 20) | \
361 BIT(17) | bits)
363 #define SET_NB_DRAM_INJECTION_READ(word, bits) \
364 (BIT(((word) & 0xF) + 20) | \
365 BIT(16) | bits)
367 #define K8_NBCAP 0xE8
368 #define K8_NBCAP_CORES (BIT(12)|BIT(13))
369 #define K8_NBCAP_CHIPKILL BIT(4)
370 #define K8_NBCAP_SECDED BIT(3)
371 #define K8_NBCAP_DCT_DUAL BIT(0)
373 #define EXT_NB_MCA_CFG 0x180
375 /* MSRs */
376 #define K8_MSR_MCGCTL_NBE BIT(4)
378 #define K8_MSR_MC4CTL 0x0410
379 #define K8_MSR_MC4STAT 0x0411
380 #define K8_MSR_MC4ADDR 0x0412
382 /* AMD sets the first MC device at device ID 0x18. */
383 static inline int get_node_id(struct pci_dev *pdev)
385 return PCI_SLOT(pdev->devfn) - 0x18;
388 enum amd64_chipset_families {
389 K8_CPUS = 0,
390 F10_CPUS,
393 /* Error injection control structure */
394 struct error_injection {
395 u32 section;
396 u32 word;
397 u32 bit_map;
400 struct amd64_pvt {
401 struct low_ops *ops;
403 /* pci_device handles which we utilize */
404 struct pci_dev *F1, *F2, *F3;
406 int mc_node_id; /* MC index of this MC node */
407 int ext_model; /* extended model value of this node */
408 int channel_count;
410 /* Raw registers */
411 u32 dclr0; /* DRAM Configuration Low DCT0 reg */
412 u32 dclr1; /* DRAM Configuration Low DCT1 reg */
413 u32 dchr0; /* DRAM Configuration High DCT0 reg */
414 u32 dchr1; /* DRAM Configuration High DCT1 reg */
415 u32 nbcap; /* North Bridge Capabilities */
416 u32 nbcfg; /* F10 North Bridge Configuration */
417 u32 ext_nbcfg; /* Extended F10 North Bridge Configuration */
418 u32 dhar; /* DRAM Hoist reg */
419 u32 dbam0; /* DRAM Base Address Mapping reg for DCT0 */
420 u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */
422 /* DRAM CS Base Address Registers F2x[1,0][5C:40] */
423 u32 dcsb0[MAX_CS_COUNT];
424 u32 dcsb1[MAX_CS_COUNT];
426 /* DRAM CS Mask Registers F2x[1,0][6C:60] */
427 u32 dcsm0[MAX_CS_COUNT];
428 u32 dcsm1[MAX_CS_COUNT];
431 * Decoded parts of DRAM BASE and LIMIT Registers
432 * F1x[78,70,68,60,58,50,48,40]
434 u64 dram_base[DRAM_REG_COUNT];
435 u64 dram_limit[DRAM_REG_COUNT];
436 u8 dram_IntlvSel[DRAM_REG_COUNT];
437 u8 dram_IntlvEn[DRAM_REG_COUNT];
438 u8 dram_DstNode[DRAM_REG_COUNT];
439 u8 dram_rw_en[DRAM_REG_COUNT];
442 * The following fields are set at (load) run time, after CPU revision
443 * has been determined, since the dct_base and dct_mask registers vary
444 * based on revision
446 u32 dcsb_base; /* DCSB base bits */
447 u32 dcsm_mask; /* DCSM mask bits */
448 u32 cs_count; /* num chip selects (== num DCSB registers) */
449 u32 num_dcsm; /* Number of DCSM registers */
450 u32 dcs_mask_notused; /* DCSM notused mask bits */
451 u32 dcs_shift; /* DCSB and DCSM shift value */
453 u64 top_mem; /* top of memory below 4GB */
454 u64 top_mem2; /* top of memory above 4GB */
456 u32 dram_ctl_select_low; /* DRAM Controller Select Low Reg */
457 u32 dram_ctl_select_high; /* DRAM Controller Select High Reg */
458 u32 online_spare; /* On-Line spare Reg */
460 /* x4 or x8 syndromes in use */
461 u8 syn_type;
463 /* temp storage for when input is received from sysfs */
464 struct err_regs ctl_error_info;
466 /* place to store error injection parameters prior to issue */
467 struct error_injection injection;
469 /* Save old hw registers' values before we modified them */
470 u32 nbctl_mcgctl_saved; /* When true, following 2 are valid */
471 u32 old_nbctl;
473 /* DCT per-family scrubrate setting */
474 u32 min_scrubrate;
476 /* family name this instance is running on */
477 const char *ctl_name;
479 /* misc settings */
480 struct flags {
481 unsigned long cf8_extcfg:1;
482 unsigned long nb_mce_enable:1;
483 unsigned long nb_ecc_prev:1;
484 } flags;
487 struct scrubrate {
488 u32 scrubval; /* bit pattern for scrub rate */
489 u32 bandwidth; /* bandwidth consumed (bytes/sec) */
492 extern struct scrubrate scrubrates[23];
493 extern const char *tt_msgs[4];
494 extern const char *ll_msgs[4];
495 extern const char *rrrr_msgs[16];
496 extern const char *to_msgs[2];
497 extern const char *pp_msgs[4];
498 extern const char *ii_msgs[4];
499 extern const char *htlink_msgs[8];
501 #ifdef CONFIG_EDAC_DEBUG
502 #define NUM_DBG_ATTRS 5
503 #else
504 #define NUM_DBG_ATTRS 0
505 #endif
507 #ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
508 #define NUM_INJ_ATTRS 5
509 #else
510 #define NUM_INJ_ATTRS 0
511 #endif
513 extern struct mcidev_sysfs_attribute amd64_dbg_attrs[NUM_DBG_ATTRS],
514 amd64_inj_attrs[NUM_INJ_ATTRS];
517 * Each of the PCI Device IDs types have their own set of hardware accessor
518 * functions and per device encoding/decoding logic.
520 struct low_ops {
521 int (*early_channel_count) (struct amd64_pvt *pvt);
523 u64 (*get_error_address) (struct mem_ctl_info *mci,
524 struct err_regs *info);
525 void (*read_dram_base_limit) (struct amd64_pvt *pvt, int dram);
526 void (*read_dram_ctl_register) (struct amd64_pvt *pvt);
527 void (*map_sysaddr_to_csrow) (struct mem_ctl_info *mci,
528 struct err_regs *info, u64 SystemAddr);
529 int (*dbam_to_cs) (struct amd64_pvt *pvt, int cs_mode);
532 struct amd64_family_type {
533 const char *ctl_name;
534 u16 f1_id, f3_id;
535 struct low_ops ops;
538 static inline int amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
539 u32 *val, const char *func)
541 int err = 0;
543 err = pci_read_config_dword(pdev, offset, val);
544 if (err)
545 amd64_warn("%s: error reading F%dx%x.\n",
546 func, PCI_FUNC(pdev->devfn), offset);
548 return err;
551 #define amd64_read_pci_cfg(pdev, offset, val) \
552 amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
555 * For future CPU versions, verify the following as new 'slow' rates appear and
556 * modify the necessary skip values for the supported CPU.
558 #define K8_MIN_SCRUB_RATE_BITS 0x0
559 #define F10_MIN_SCRUB_RATE_BITS 0x5
561 int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
562 u64 *hole_offset, u64 *hole_size);