amd64_edac.h 14.3 KB
Newer Older
D
Doug Thompson 已提交
1 2 3 4
/*
 * AMD64 class Memory Controller kernel module
 *
 * Copyright (c) 2009 SoftwareBitMaker.
5
 * Copyright (c) 2009-15 Advanced Micro Devices, Inc.
D
Doug Thompson 已提交
6 7 8 9 10 11 12 13 14 15 16 17 18
 *
 * This file may be distributed under the terms of the
 * GNU General Public License.
 */

#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>
19
#include <asm/cpu_device_id.h>
20
#include <asm/msr.h>
21
#include "edac_module.h"
B
Borislav Petkov 已提交
22
#include "mce_amd.h"
D
Doug Thompson 已提交
23

24 25 26 27
#define amd64_info(fmt, arg...) \
	edac_printk(KERN_INFO, "amd64", fmt, ##arg)

#define amd64_warn(fmt, arg...) \
28
	edac_printk(KERN_WARNING, "amd64", "Warning: " fmt, ##arg)
29 30

#define amd64_err(fmt, arg...) \
31
	edac_printk(KERN_ERR, "amd64", "Error: " fmt, ##arg)
32 33 34 35 36 37

#define amd64_mc_warn(mci, fmt, arg...) \
	edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)

#define amd64_mc_err(mci, fmt, arg...) \
	edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
D
Doug Thompson 已提交
38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87

/*
 * 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.
 */

88
#define EDAC_AMD64_VERSION		"3.5.0"
D
Doug Thompson 已提交
89 90 91
#define EDAC_MOD_STR			"amd64_edac"

/* Extended Model from CPUID, for CPU Revision numbers */
92 93 94
#define K8_REV_D			1
#define K8_REV_E			2
#define K8_REV_F			4
D
Doug Thompson 已提交
95 96

/* Hardware limit on ChipSelect rows per MC and processors per system */
97 98
#define NUM_CHIPSELECTS			8
#define DRAM_RANGES			8
99
#define NUM_CONTROLLERS			8
D
Doug Thompson 已提交
100

101 102
#define ON true
#define OFF false
D
Doug Thompson 已提交
103 104 105 106

/*
 * PCI-defined configuration space registers
 */
107 108
#define PCI_DEVICE_ID_AMD_15H_NB_F1	0x1601
#define PCI_DEVICE_ID_AMD_15H_NB_F2	0x1602
109 110 111 112
#define PCI_DEVICE_ID_AMD_15H_M30H_NB_F1 0x141b
#define PCI_DEVICE_ID_AMD_15H_M30H_NB_F2 0x141c
#define PCI_DEVICE_ID_AMD_15H_M60H_NB_F1 0x1571
#define PCI_DEVICE_ID_AMD_15H_M60H_NB_F2 0x1572
113 114
#define PCI_DEVICE_ID_AMD_16H_NB_F1	0x1531
#define PCI_DEVICE_ID_AMD_16H_NB_F2	0x1532
115 116
#define PCI_DEVICE_ID_AMD_16H_M30H_NB_F1 0x1581
#define PCI_DEVICE_ID_AMD_16H_M30H_NB_F2 0x1582
117 118
#define PCI_DEVICE_ID_AMD_17H_DF_F0	0x1460
#define PCI_DEVICE_ID_AMD_17H_DF_F6	0x1466
119 120
#define PCI_DEVICE_ID_AMD_17H_M10H_DF_F0 0x15e8
#define PCI_DEVICE_ID_AMD_17H_M10H_DF_F6 0x15ee
121 122
#define PCI_DEVICE_ID_AMD_17H_M30H_DF_F0 0x1490
#define PCI_DEVICE_ID_AMD_17H_M30H_DF_F6 0x1496
D
Doug Thompson 已提交
123 124 125 126

/*
 * Function 1 - Address Map
 */
127 128 129
#define DRAM_BASE_LO			0x40
#define DRAM_LIMIT_LO			0x44

130 131 132 133 134 135 136 137 138 139 140
/*
 * F15 M30h D18F1x2[1C:00]
 */
#define DRAM_CONT_BASE			0x200
#define DRAM_CONT_LIMIT			0x204

/*
 * F15 M30h D18F1x2[4C:40]
 */
#define DRAM_CONT_HIGH_OFF		0x240

141 142 143
#define dram_rw(pvt, i)			((u8)(pvt->ranges[i].base.lo & 0x3))
#define dram_intlv_sel(pvt, i)		((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
#define dram_dst_node(pvt, i)		((u8)(pvt->ranges[i].lim.lo & 0x7))
144

145
#define DHAR				0xf0
146 147 148
#define dhar_mem_hoist_valid(pvt)	((pvt)->dhar & BIT(1))
#define dhar_base(pvt)			((pvt)->dhar & 0xff000000)
#define k8_dhar_offset(pvt)		(((pvt)->dhar & 0x0000ff00) << 16)
D
Doug Thompson 已提交
149 150

					/* NOTE: Extra mask bit vs K8 */
151
#define f10_dhar_offset(pvt)		(((pvt)->dhar & 0x0000ff80) << 16)
D
Doug Thompson 已提交
152

153
#define DCT_CFG_SEL			0x10C
D
Doug Thompson 已提交
154

155
#define DRAM_LOCAL_NODE_BASE		0x120
156 157
#define DRAM_LOCAL_NODE_LIM		0x124

158 159
#define DRAM_BASE_HI			0x140
#define DRAM_LIMIT_HI			0x144
D
Doug Thompson 已提交
160 161 162 163 164


/*
 * Function 2 - DRAM controller
 */
165 166 167
#define DCSB0				0x40
#define DCSB1				0x140
#define DCSB_CS_ENABLE			BIT(0)
D
Doug Thompson 已提交
168

169 170
#define DCSM0				0x60
#define DCSM1				0x160
D
Doug Thompson 已提交
171

172
#define csrow_enabled(i, dct, pvt)	((pvt)->csels[(dct)].csbases[(i)] & DCSB_CS_ENABLE)
D
Doug Thompson 已提交
173

174 175
#define DRAM_CONTROL			0x78

D
Doug Thompson 已提交
176 177 178 179
#define DBAM0				0x80
#define DBAM1				0x180

/* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
B
Borislav Petkov 已提交
180
#define DBAM_DIMM(i, reg)		((((reg) >> (4*(i)))) & 0xF)
D
Doug Thompson 已提交
181 182 183

#define DBAM_MAX_VALUE			11

184 185
#define DCLR0				0x90
#define DCLR1				0x190
D
Doug Thompson 已提交
186
#define REVE_WIDTH_128			BIT(16)
187
#define WIDTH_128			BIT(11)
D
Doug Thompson 已提交
188

189 190
#define DCHR0				0x94
#define DCHR1				0x194
191
#define DDR3_MODE			BIT(8)
D
Doug Thompson 已提交
192

193 194 195
#define DCT_SEL_LO			0x110
#define dct_high_range_enabled(pvt)	((pvt)->dct_sel_lo & BIT(0))
#define dct_interleave_enabled(pvt)	((pvt)->dct_sel_lo & BIT(2))
196

197
#define dct_ganging_enabled(pvt)	((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
198

199 200
#define dct_data_intlv_enabled(pvt)	((pvt)->dct_sel_lo & BIT(5))
#define dct_memory_cleared(pvt)		((pvt)->dct_sel_lo & BIT(10))
D
Doug Thompson 已提交
201

202 203
#define SWAP_INTLV_REG			0x10c

204
#define DCT_SEL_HI			0x114
D
Doug Thompson 已提交
205

206
#define F15H_M60H_SCRCTRL		0x1C8
207 208
#define F17H_SCR_BASE_ADDR		0x48
#define F17H_SCR_LIMIT_ADDR		0x4C
209

D
Doug Thompson 已提交
210 211 212
/*
 * Function 3 - Misc Control
 */
B
Borislav Petkov 已提交
213
#define NBCTL				0x40
D
Doug Thompson 已提交
214

215 216 217
#define NBCFG				0x44
#define NBCFG_CHIPKILL			BIT(23)
#define NBCFG_ECC_ENABLE		BIT(22)
D
Doug Thompson 已提交
218

219
/* F3x48: NBSL */
D
Doug Thompson 已提交
220
#define F10_NBSL_EXT_ERR_ECC		0x8
221
#define NBSL_PP_OBS			0x2
D
Doug Thompson 已提交
222

223
#define SCRCTRL				0x58
D
Doug Thompson 已提交
224 225

#define F10_ONLINE_SPARE		0xB0
226 227
#define online_spare_swap_done(pvt, c)	(((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
#define online_spare_bad_dramcs(pvt, c)	(((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
D
Doug Thompson 已提交
228 229

#define F10_NB_ARRAY_ADDR		0xB8
230
#define F10_NB_ARRAY_DRAM		BIT(31)
D
Doug Thompson 已提交
231 232

/* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline  */
233
#define SET_NB_ARRAY_ADDR(section)	(((section) & 0x3) << 1)
D
Doug Thompson 已提交
234 235

#define F10_NB_ARRAY_DATA		0xBC
236
#define F10_NB_ARR_ECC_WR_REQ		BIT(17)
237 238
#define SET_NB_DRAM_INJECTION_WRITE(inj)  \
					(BIT(((inj.word) & 0xF) + 20) | \
239
					F10_NB_ARR_ECC_WR_REQ | inj.bit_map)
240 241 242 243
#define SET_NB_DRAM_INJECTION_READ(inj)  \
					(BIT(((inj.word) & 0xF) + 20) | \
					BIT(16) |  inj.bit_map)

D
Doug Thompson 已提交
244

245 246 247 248
#define NBCAP				0xE8
#define NBCAP_CHIPKILL			BIT(4)
#define NBCAP_SECDED			BIT(3)
#define NBCAP_DCT_DUAL			BIT(0)
D
Doug Thompson 已提交
249

250 251
#define EXT_NB_MCA_CFG			0x180

252
/* MSRs */
253
#define MSR_MCGCTL_NBE			BIT(4)
D
Doug Thompson 已提交
254

255 256 257 258 259
/* F17h */

/* F0: */
#define DF_DHAR				0x104

260
/* UMC CH register offsets */
261 262
#define UMCCH_BASE_ADDR			0x0
#define UMCCH_ADDR_MASK			0x20
263
#define UMCCH_ADDR_CFG			0x30
264
#define UMCCH_DIMM_CFG			0x80
265
#define UMCCH_UMC_CFG			0x100
266
#define UMCCH_SDP_CTRL			0x104
267
#define UMCCH_ECC_CTRL			0x14C
268 269
#define UMCCH_ECC_BAD_SYMBOL		0xD90
#define UMCCH_UMC_CAP			0xDF0
270 271 272
#define UMCCH_UMC_CAP_HI		0xDF4

/* UMC CH bitfields */
273
#define UMC_ECC_CHIPKILL_CAP		BIT(31)
274
#define UMC_ECC_ENABLED			BIT(30)
275

276 277
#define UMC_SDP_INIT			BIT(31)

278
enum amd_families {
D
Doug Thompson 已提交
279 280
	K8_CPUS = 0,
	F10_CPUS,
281
	F15_CPUS,
282
	F15_M30H_CPUS,
283
	F15_M60H_CPUS,
284
	F16_CPUS,
285
	F16_M30H_CPUS,
286
	F17_CPUS,
287
	F17_M10H_CPUS,
288
	F17_M30H_CPUS,
289
	NUM_FAMILIES,
D
Doug Thompson 已提交
290 291 292 293
};

/* Error injection control structure */
struct error_injection {
294 295 296
	u32	 section;
	u32	 word;
	u32	 bit_map;
D
Doug Thompson 已提交
297 298
};

299 300 301 302 303 304 305 306 307 308 309 310 311
/* low and high part of PCI config space regs */
struct reg_pair {
	u32 lo, hi;
};

/*
 * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
 */
struct dram_range {
	struct reg_pair base;
	struct reg_pair lim;
};

312 313 314 315 316 317 318 319 320
/* A DCT chip selects collection */
struct chip_select {
	u32 csbases[NUM_CHIPSELECTS];
	u8 b_cnt;

	u32 csmasks[NUM_CHIPSELECTS];
	u8 m_cnt;
};

321
struct amd64_umc {
322
	u32 dimm_cfg;		/* DIMM Configuration reg */
323
	u32 umc_cfg;		/* Configuration reg */
324
	u32 sdp_ctrl;		/* SDP Control reg */
325
	u32 ecc_ctrl;		/* DRAM ECC Control reg */
326
	u32 umc_cap_hi;		/* Capabilities High reg */
327 328
};

D
Doug Thompson 已提交
329
struct amd64_pvt {
330 331
	struct low_ops *ops;

D
Doug Thompson 已提交
332
	/* pci_device handles which we utilize */
333
	struct pci_dev *F0, *F1, *F2, *F3, *F6;
D
Doug Thompson 已提交
334

335
	u16 mc_node_id;		/* MC index of this MC node */
336
	u8 fam;			/* CPU family */
337 338 339
	u8 model;		/* ... model */
	u8 stepping;		/* ... stepping */

D
Doug Thompson 已提交
340 341 342 343 344 345 346 347 348 349 350 351 352 353 354
	int ext_model;		/* extended model value of this node */
	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 */

355 356
	/* one for each DCT/UMC */
	struct chip_select csels[NUM_CONTROLLERS];
D
Doug Thompson 已提交
357

358 359
	/* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
	struct dram_range ranges[DRAM_RANGES];
D
Doug Thompson 已提交
360 361 362 363

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

364 365
	u32 dct_sel_lo;		/* DRAM Controller Select Low */
	u32 dct_sel_hi;		/* DRAM Controller Select High */
366
	u32 online_spare;	/* On-Line spare Reg */
D
Doug Thompson 已提交
367

368
	/* x4, x8, or x16 syndromes in use */
369
	u8 ecc_sym_sz;
370

D
Doug Thompson 已提交
371 372
	/* place to store error injection parameters prior to issue */
	struct error_injection injection;
373 374 375

	/* cache the dram_type */
	enum mem_type dram_type;
376 377

	struct amd64_umc *umc;	/* UMC registers */
378 379
};

380 381 382 383 384
enum err_codes {
	DECODE_OK	=  0,
	ERR_NODE	= -1,
	ERR_CSROW	= -2,
	ERR_CHANNEL	= -3,
385 386
	ERR_SYND	= -4,
	ERR_NORM_ADDR	= -5,
387 388 389 390 391 392 393 394 395 396 397 398
};

struct err_info {
	int err_code;
	struct mem_ctl_info *src_mci;
	int csrow;
	int channel;
	u16 syndrome;
	u32 page;
	u32 offset;
};

399 400
static inline u32 get_umc_base(u8 channel)
{
401 402
	/* chY: 0xY50000 */
	return 0x50000 + (channel << 20);
403 404
}

405
static inline u64 get_dram_base(struct amd64_pvt *pvt, u8 i)
406 407 408 409 410 411 412 413 414
{
	u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;

	if (boot_cpu_data.x86 == 0xf)
		return addr;

	return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
}

415
static inline u64 get_dram_limit(struct amd64_pvt *pvt, u8 i)
416 417 418 419 420 421 422 423 424
{
	u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;

	if (boot_cpu_data.x86 == 0xf)
		return lim;

	return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
}

425 426 427 428 429
static inline u16 extract_syndrome(u64 status)
{
	return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
}

430 431 432 433 434 435 436 437
static inline u8 dct_sel_interleave_addr(struct amd64_pvt *pvt)
{
	if (pvt->fam == 0x15 && pvt->model >= 0x30)
		return (((pvt->dct_sel_hi >> 9) & 0x1) << 2) |
			((pvt->dct_sel_lo >> 6) & 0x3);

	return	((pvt)->dct_sel_lo >> 6) & 0x3;
}
438 439 440 441 442 443 444
/*
 * per-node ECC settings descriptor
 */
struct ecc_settings {
	u32 old_nbctl;
	bool nbctl_valid;

D
Doug Thompson 已提交
445
	struct flags {
446 447
		unsigned long nb_mce_enable:1;
		unsigned long nb_ecc_prev:1;
D
Doug Thompson 已提交
448 449 450
	} flags;
};

451
#ifdef CONFIG_EDAC_DEBUG
452
extern const struct attribute_group amd64_edac_dbg_group;
453 454 455
#endif

#ifdef CONFIG_EDAC_AMD64_ERROR_INJECTION
456
extern const struct attribute_group amd64_edac_inj_group;
457 458
#endif

D
Doug Thompson 已提交
459 460 461 462 463
/*
 * Each of the PCI Device IDs types have their own set of hardware accessor
 * functions and per device encoding/decoding logic.
 */
struct low_ops {
464
	int (*early_channel_count)	(struct amd64_pvt *pvt);
465
	void (*map_sysaddr_to_csrow)	(struct mem_ctl_info *mci, u64 sys_addr,
466
					 struct err_info *);
467 468
	int (*dbam_to_cs)		(struct amd64_pvt *pvt, u8 dct,
					 unsigned cs_mode, int cs_mask_nr);
D
Doug Thompson 已提交
469 470 471 472
};

struct amd64_family_type {
	const char *ctl_name;
473
	u16 f0_id, f1_id, f2_id, f6_id;
D
Doug Thompson 已提交
474 475 476
	struct low_ops ops;
};

477 478
int __amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
			       u32 *val, const char *func);
479 480
int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
				u32 val, const char *func);
481

482 483
#define amd64_read_pci_cfg(pdev, offset, val)	\
	__amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
484

485 486
#define amd64_write_pci_cfg(pdev, offset, val)	\
	__amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
487

D
Doug Thompson 已提交
488 489
int amd64_get_dram_hole_info(struct mem_ctl_info *mci, u64 *hole_base,
			     u64 *hole_offset, u64 *hole_size);
490 491

#define to_mci(k) container_of(k, struct mem_ctl_info, dev)
492 493 494 495 496 497 498 499 500 501 502 503

/* Injection helpers */
static inline void disable_caches(void *dummy)
{
	write_cr0(read_cr0() | X86_CR0_CD);
	wbinvd();
}

static inline void enable_caches(void *dummy)
{
	write_cr0(read_cr0() & ~X86_CR0_CD);
}
504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533

static inline u8 dram_intlv_en(struct amd64_pvt *pvt, unsigned int i)
{
	if (pvt->fam == 0x15 && pvt->model >= 0x30) {
		u32 tmp;
		amd64_read_pci_cfg(pvt->F1, DRAM_CONT_LIMIT, &tmp);
		return (u8) tmp & 0xF;
	}
	return (u8) (pvt->ranges[i].base.lo >> 8) & 0x7;
}

static inline u8 dhar_valid(struct amd64_pvt *pvt)
{
	if (pvt->fam == 0x15 && pvt->model >= 0x30) {
		u32 tmp;
		amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
		return (tmp >> 1) & BIT(0);
	}
	return (pvt)->dhar & BIT(0);
}

static inline u32 dct_sel_baseaddr(struct amd64_pvt *pvt)
{
	if (pvt->fam == 0x15 && pvt->model >= 0x30) {
		u32 tmp;
		amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
		return (tmp >> 11) & 0x1FFF;
	}
	return (pvt)->dct_sel_lo & 0xFFFFF800;
}