xref: /openbmc/linux/drivers/edac/amd64_edac.h (revision c4f7ac64)
1 /*
2  * AMD64 class Memory Controller kernel module
3  *
4  * Copyright (c) 2009 SoftwareBitMaker.
5  * Copyright (c) 2009-15 Advanced Micro Devices, Inc.
6  *
7  * This file may be distributed under the terms of the
8  * GNU General Public License.
9  */
10 
11 #include <linux/module.h>
12 #include <linux/ctype.h>
13 #include <linux/init.h>
14 #include <linux/pci.h>
15 #include <linux/pci_ids.h>
16 #include <linux/slab.h>
17 #include <linux/mmzone.h>
18 #include <linux/edac.h>
19 #include <asm/cpu_device_id.h>
20 #include <asm/msr.h>
21 #include "edac_module.h"
22 #include "mce_amd.h"
23 
24 #define amd64_info(fmt, arg...) \
25 	edac_printk(KERN_INFO, "amd64", fmt, ##arg)
26 
27 #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 #define amd64_mc_warn(mci, fmt, arg...) \
34 	edac_mc_chipset_printk(mci, KERN_WARNING, "amd64", fmt, ##arg)
35 
36 #define amd64_mc_err(mci, fmt, arg...) \
37 	edac_mc_chipset_printk(mci, KERN_ERR, "amd64", fmt, ##arg)
38 
39 /*
40  * Throughout the comments in this code, the following terms are used:
41  *
42  *	SysAddr, DramAddr, and InputAddr
43  *
44  *  These terms come directly from the amd64 documentation
45  * (AMD publication #26094).  They are defined as follows:
46  *
47  *     SysAddr:
48  *         This is a physical address generated by a CPU core or a device
49  *         doing DMA.  If generated by a CPU core, a SysAddr is the result of
50  *         a virtual to physical address translation by the CPU core's address
51  *         translation mechanism (MMU).
52  *
53  *     DramAddr:
54  *         A DramAddr is derived from a SysAddr by subtracting an offset that
55  *         depends on which node the SysAddr maps to and whether the SysAddr
56  *         is within a range affected by memory hoisting.  The DRAM Base
57  *         (section 3.4.4.1) and DRAM Limit (section 3.4.4.2) registers
58  *         determine which node a SysAddr maps to.
59  *
60  *         If the DRAM Hole Address Register (DHAR) is enabled and the SysAddr
61  *         is within the range of addresses specified by this register, then
62  *         a value x from the DHAR is subtracted from the SysAddr to produce a
63  *         DramAddr.  Here, x represents the base address for the node that
64  *         the SysAddr maps to plus an offset due to memory hoisting.  See
65  *         section 3.4.8 and the comments in amd64_get_dram_hole_info() and
66  *         sys_addr_to_dram_addr() below for more information.
67  *
68  *         If the SysAddr is not affected by the DHAR then a value y is
69  *         subtracted from the SysAddr to produce a DramAddr.  Here, y is the
70  *         base address for the node that the SysAddr maps to.  See section
71  *         3.4.4 and the comments in sys_addr_to_dram_addr() below for more
72  *         information.
73  *
74  *     InputAddr:
75  *         A DramAddr is translated to an InputAddr before being passed to the
76  *         memory controller for the node that the DramAddr is associated
77  *         with.  The memory controller then maps the InputAddr to a csrow.
78  *         If node interleaving is not in use, then the InputAddr has the same
79  *         value as the DramAddr.  Otherwise, the InputAddr is produced by
80  *         discarding the bits used for node interleaving from the DramAddr.
81  *         See section 3.4.4 for more information.
82  *
83  *         The memory controller for a given node uses its DRAM CS Base and
84  *         DRAM CS Mask registers to map an InputAddr to a csrow.  See
85  *         sections 3.5.4 and 3.5.5 for more information.
86  */
87 
88 #define EDAC_AMD64_VERSION		"3.5.0"
89 #define EDAC_MOD_STR			"amd64_edac"
90 
91 /* Extended Model from CPUID, for CPU Revision numbers */
92 #define K8_REV_D			1
93 #define K8_REV_E			2
94 #define K8_REV_F			4
95 
96 /* Hardware limit on ChipSelect rows per MC and processors per system */
97 #define NUM_CHIPSELECTS			8
98 #define DRAM_RANGES			8
99 #define NUM_CONTROLLERS			8
100 
101 #define ON true
102 #define OFF false
103 
104 /*
105  * PCI-defined configuration space registers
106  */
107 #define PCI_DEVICE_ID_AMD_15H_NB_F1	0x1601
108 #define PCI_DEVICE_ID_AMD_15H_NB_F2	0x1602
109 #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F1 0x141b
110 #define PCI_DEVICE_ID_AMD_15H_M30H_NB_F2 0x141c
111 #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F1 0x1571
112 #define PCI_DEVICE_ID_AMD_15H_M60H_NB_F2 0x1572
113 #define PCI_DEVICE_ID_AMD_16H_NB_F1	0x1531
114 #define PCI_DEVICE_ID_AMD_16H_NB_F2	0x1532
115 #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F1 0x1581
116 #define PCI_DEVICE_ID_AMD_16H_M30H_NB_F2 0x1582
117 #define PCI_DEVICE_ID_AMD_17H_DF_F0	0x1460
118 #define PCI_DEVICE_ID_AMD_17H_DF_F6	0x1466
119 #define PCI_DEVICE_ID_AMD_17H_M10H_DF_F0 0x15e8
120 #define PCI_DEVICE_ID_AMD_17H_M10H_DF_F6 0x15ee
121 #define PCI_DEVICE_ID_AMD_17H_M30H_DF_F0 0x1490
122 #define PCI_DEVICE_ID_AMD_17H_M30H_DF_F6 0x1496
123 #define PCI_DEVICE_ID_AMD_17H_M60H_DF_F0 0x1448
124 #define PCI_DEVICE_ID_AMD_17H_M60H_DF_F6 0x144e
125 #define PCI_DEVICE_ID_AMD_17H_M70H_DF_F0 0x1440
126 #define PCI_DEVICE_ID_AMD_17H_M70H_DF_F6 0x1446
127 #define PCI_DEVICE_ID_AMD_19H_DF_F0	0x1650
128 #define PCI_DEVICE_ID_AMD_19H_DF_F6	0x1656
129 
130 /*
131  * Function 1 - Address Map
132  */
133 #define DRAM_BASE_LO			0x40
134 #define DRAM_LIMIT_LO			0x44
135 
136 /*
137  * F15 M30h D18F1x2[1C:00]
138  */
139 #define DRAM_CONT_BASE			0x200
140 #define DRAM_CONT_LIMIT			0x204
141 
142 /*
143  * F15 M30h D18F1x2[4C:40]
144  */
145 #define DRAM_CONT_HIGH_OFF		0x240
146 
147 #define dram_rw(pvt, i)			((u8)(pvt->ranges[i].base.lo & 0x3))
148 #define dram_intlv_sel(pvt, i)		((u8)((pvt->ranges[i].lim.lo >> 8) & 0x7))
149 #define dram_dst_node(pvt, i)		((u8)(pvt->ranges[i].lim.lo & 0x7))
150 
151 #define DHAR				0xf0
152 #define dhar_mem_hoist_valid(pvt)	((pvt)->dhar & BIT(1))
153 #define dhar_base(pvt)			((pvt)->dhar & 0xff000000)
154 #define k8_dhar_offset(pvt)		(((pvt)->dhar & 0x0000ff00) << 16)
155 
156 					/* NOTE: Extra mask bit vs K8 */
157 #define f10_dhar_offset(pvt)		(((pvt)->dhar & 0x0000ff80) << 16)
158 
159 #define DCT_CFG_SEL			0x10C
160 
161 #define DRAM_LOCAL_NODE_BASE		0x120
162 #define DRAM_LOCAL_NODE_LIM		0x124
163 
164 #define DRAM_BASE_HI			0x140
165 #define DRAM_LIMIT_HI			0x144
166 
167 
168 /*
169  * Function 2 - DRAM controller
170  */
171 #define DCSB0				0x40
172 #define DCSB1				0x140
173 #define DCSB_CS_ENABLE			BIT(0)
174 
175 #define DCSM0				0x60
176 #define DCSM1				0x160
177 
178 #define csrow_enabled(i, dct, pvt)	((pvt)->csels[(dct)].csbases[(i)]     & DCSB_CS_ENABLE)
179 #define csrow_sec_enabled(i, dct, pvt)	((pvt)->csels[(dct)].csbases_sec[(i)] & DCSB_CS_ENABLE)
180 
181 #define DRAM_CONTROL			0x78
182 
183 #define DBAM0				0x80
184 #define DBAM1				0x180
185 
186 /* Extract the DIMM 'type' on the i'th DIMM from the DBAM reg value passed */
187 #define DBAM_DIMM(i, reg)		((((reg) >> (4*(i)))) & 0xF)
188 
189 #define DBAM_MAX_VALUE			11
190 
191 #define DCLR0				0x90
192 #define DCLR1				0x190
193 #define REVE_WIDTH_128			BIT(16)
194 #define WIDTH_128			BIT(11)
195 
196 #define DCHR0				0x94
197 #define DCHR1				0x194
198 #define DDR3_MODE			BIT(8)
199 
200 #define DCT_SEL_LO			0x110
201 #define dct_high_range_enabled(pvt)	((pvt)->dct_sel_lo & BIT(0))
202 #define dct_interleave_enabled(pvt)	((pvt)->dct_sel_lo & BIT(2))
203 
204 #define dct_ganging_enabled(pvt)	((boot_cpu_data.x86 == 0x10) && ((pvt)->dct_sel_lo & BIT(4)))
205 
206 #define dct_data_intlv_enabled(pvt)	((pvt)->dct_sel_lo & BIT(5))
207 #define dct_memory_cleared(pvt)		((pvt)->dct_sel_lo & BIT(10))
208 
209 #define SWAP_INTLV_REG			0x10c
210 
211 #define DCT_SEL_HI			0x114
212 
213 #define F15H_M60H_SCRCTRL		0x1C8
214 #define F17H_SCR_BASE_ADDR		0x48
215 #define F17H_SCR_LIMIT_ADDR		0x4C
216 
217 /*
218  * Function 3 - Misc Control
219  */
220 #define NBCTL				0x40
221 
222 #define NBCFG				0x44
223 #define NBCFG_CHIPKILL			BIT(23)
224 #define NBCFG_ECC_ENABLE		BIT(22)
225 
226 /* F3x48: NBSL */
227 #define F10_NBSL_EXT_ERR_ECC		0x8
228 #define NBSL_PP_OBS			0x2
229 
230 #define SCRCTRL				0x58
231 
232 #define F10_ONLINE_SPARE		0xB0
233 #define online_spare_swap_done(pvt, c)	(((pvt)->online_spare >> (1 + 2 * (c))) & 0x1)
234 #define online_spare_bad_dramcs(pvt, c)	(((pvt)->online_spare >> (4 + 4 * (c))) & 0x7)
235 
236 #define F10_NB_ARRAY_ADDR		0xB8
237 #define F10_NB_ARRAY_DRAM		BIT(31)
238 
239 /* Bits [2:1] are used to select 16-byte section within a 64-byte cacheline  */
240 #define SET_NB_ARRAY_ADDR(section)	(((section) & 0x3) << 1)
241 
242 #define F10_NB_ARRAY_DATA		0xBC
243 #define F10_NB_ARR_ECC_WR_REQ		BIT(17)
244 #define SET_NB_DRAM_INJECTION_WRITE(inj)  \
245 					(BIT(((inj.word) & 0xF) + 20) | \
246 					F10_NB_ARR_ECC_WR_REQ | inj.bit_map)
247 #define SET_NB_DRAM_INJECTION_READ(inj)  \
248 					(BIT(((inj.word) & 0xF) + 20) | \
249 					BIT(16) |  inj.bit_map)
250 
251 
252 #define NBCAP				0xE8
253 #define NBCAP_CHIPKILL			BIT(4)
254 #define NBCAP_SECDED			BIT(3)
255 #define NBCAP_DCT_DUAL			BIT(0)
256 
257 #define EXT_NB_MCA_CFG			0x180
258 
259 /* MSRs */
260 #define MSR_MCGCTL_NBE			BIT(4)
261 
262 /* F17h */
263 
264 /* F0: */
265 #define DF_DHAR				0x104
266 
267 /* UMC CH register offsets */
268 #define UMCCH_BASE_ADDR			0x0
269 #define UMCCH_BASE_ADDR_SEC		0x10
270 #define UMCCH_ADDR_MASK			0x20
271 #define UMCCH_ADDR_MASK_SEC		0x28
272 #define UMCCH_ADDR_CFG			0x30
273 #define UMCCH_DIMM_CFG			0x80
274 #define UMCCH_UMC_CFG			0x100
275 #define UMCCH_SDP_CTRL			0x104
276 #define UMCCH_ECC_CTRL			0x14C
277 #define UMCCH_ECC_BAD_SYMBOL		0xD90
278 #define UMCCH_UMC_CAP			0xDF0
279 #define UMCCH_UMC_CAP_HI		0xDF4
280 
281 /* UMC CH bitfields */
282 #define UMC_ECC_CHIPKILL_CAP		BIT(31)
283 #define UMC_ECC_ENABLED			BIT(30)
284 
285 #define UMC_SDP_INIT			BIT(31)
286 
287 enum amd_families {
288 	K8_CPUS = 0,
289 	F10_CPUS,
290 	F15_CPUS,
291 	F15_M30H_CPUS,
292 	F15_M60H_CPUS,
293 	F16_CPUS,
294 	F16_M30H_CPUS,
295 	F17_CPUS,
296 	F17_M10H_CPUS,
297 	F17_M30H_CPUS,
298 	F17_M60H_CPUS,
299 	F17_M70H_CPUS,
300 	F19_CPUS,
301 	NUM_FAMILIES,
302 };
303 
304 /* Error injection control structure */
305 struct error_injection {
306 	u32	 section;
307 	u32	 word;
308 	u32	 bit_map;
309 };
310 
311 /* low and high part of PCI config space regs */
312 struct reg_pair {
313 	u32 lo, hi;
314 };
315 
316 /*
317  * See F1x[1, 0][7C:40] DRAM Base/Limit Registers
318  */
319 struct dram_range {
320 	struct reg_pair base;
321 	struct reg_pair lim;
322 };
323 
324 /* A DCT chip selects collection */
325 struct chip_select {
326 	u32 csbases[NUM_CHIPSELECTS];
327 	u32 csbases_sec[NUM_CHIPSELECTS];
328 	u8 b_cnt;
329 
330 	u32 csmasks[NUM_CHIPSELECTS];
331 	u32 csmasks_sec[NUM_CHIPSELECTS];
332 	u8 m_cnt;
333 };
334 
335 struct amd64_umc {
336 	u32 dimm_cfg;		/* DIMM Configuration reg */
337 	u32 umc_cfg;		/* Configuration reg */
338 	u32 sdp_ctrl;		/* SDP Control reg */
339 	u32 ecc_ctrl;		/* DRAM ECC Control reg */
340 	u32 umc_cap_hi;		/* Capabilities High reg */
341 };
342 
343 struct amd64_pvt {
344 	struct low_ops *ops;
345 
346 	/* pci_device handles which we utilize */
347 	struct pci_dev *F0, *F1, *F2, *F3, *F6;
348 
349 	u16 mc_node_id;		/* MC index of this MC node */
350 	u8 fam;			/* CPU family */
351 	u8 model;		/* ... model */
352 	u8 stepping;		/* ... stepping */
353 
354 	int ext_model;		/* extended model value of this node */
355 	int channel_count;
356 
357 	/* Raw registers */
358 	u32 dclr0;		/* DRAM Configuration Low DCT0 reg */
359 	u32 dclr1;		/* DRAM Configuration Low DCT1 reg */
360 	u32 dchr0;		/* DRAM Configuration High DCT0 reg */
361 	u32 dchr1;		/* DRAM Configuration High DCT1 reg */
362 	u32 nbcap;		/* North Bridge Capabilities */
363 	u32 nbcfg;		/* F10 North Bridge Configuration */
364 	u32 ext_nbcfg;		/* Extended F10 North Bridge Configuration */
365 	u32 dhar;		/* DRAM Hoist reg */
366 	u32 dbam0;		/* DRAM Base Address Mapping reg for DCT0 */
367 	u32 dbam1;		/* DRAM Base Address Mapping reg for DCT1 */
368 
369 	/* one for each DCT/UMC */
370 	struct chip_select csels[NUM_CONTROLLERS];
371 
372 	/* DRAM base and limit pairs F1x[78,70,68,60,58,50,48,40] */
373 	struct dram_range ranges[DRAM_RANGES];
374 
375 	u64 top_mem;		/* top of memory below 4GB */
376 	u64 top_mem2;		/* top of memory above 4GB */
377 
378 	u32 dct_sel_lo;		/* DRAM Controller Select Low */
379 	u32 dct_sel_hi;		/* DRAM Controller Select High */
380 	u32 online_spare;	/* On-Line spare Reg */
381 
382 	/* x4, x8, or x16 syndromes in use */
383 	u8 ecc_sym_sz;
384 
385 	/* place to store error injection parameters prior to issue */
386 	struct error_injection injection;
387 
388 	/* cache the dram_type */
389 	enum mem_type dram_type;
390 
391 	struct amd64_umc *umc;	/* UMC registers */
392 };
393 
394 enum err_codes {
395 	DECODE_OK	=  0,
396 	ERR_NODE	= -1,
397 	ERR_CSROW	= -2,
398 	ERR_CHANNEL	= -3,
399 	ERR_SYND	= -4,
400 	ERR_NORM_ADDR	= -5,
401 };
402 
403 struct err_info {
404 	int err_code;
405 	struct mem_ctl_info *src_mci;
406 	int csrow;
407 	int channel;
408 	u16 syndrome;
409 	u32 page;
410 	u32 offset;
411 };
412 
413 static inline u32 get_umc_base(u8 channel)
414 {
415 	/* chY: 0xY50000 */
416 	return 0x50000 + (channel << 20);
417 }
418 
419 static inline u64 get_dram_base(struct amd64_pvt *pvt, u8 i)
420 {
421 	u64 addr = ((u64)pvt->ranges[i].base.lo & 0xffff0000) << 8;
422 
423 	if (boot_cpu_data.x86 == 0xf)
424 		return addr;
425 
426 	return (((u64)pvt->ranges[i].base.hi & 0x000000ff) << 40) | addr;
427 }
428 
429 static inline u64 get_dram_limit(struct amd64_pvt *pvt, u8 i)
430 {
431 	u64 lim = (((u64)pvt->ranges[i].lim.lo & 0xffff0000) << 8) | 0x00ffffff;
432 
433 	if (boot_cpu_data.x86 == 0xf)
434 		return lim;
435 
436 	return (((u64)pvt->ranges[i].lim.hi & 0x000000ff) << 40) | lim;
437 }
438 
439 static inline u16 extract_syndrome(u64 status)
440 {
441 	return ((status >> 47) & 0xff) | ((status >> 16) & 0xff00);
442 }
443 
444 static inline u8 dct_sel_interleave_addr(struct amd64_pvt *pvt)
445 {
446 	if (pvt->fam == 0x15 && pvt->model >= 0x30)
447 		return (((pvt->dct_sel_hi >> 9) & 0x1) << 2) |
448 			((pvt->dct_sel_lo >> 6) & 0x3);
449 
450 	return	((pvt)->dct_sel_lo >> 6) & 0x3;
451 }
452 /*
453  * per-node ECC settings descriptor
454  */
455 struct ecc_settings {
456 	u32 old_nbctl;
457 	bool nbctl_valid;
458 
459 	struct flags {
460 		unsigned long nb_mce_enable:1;
461 		unsigned long nb_ecc_prev:1;
462 	} flags;
463 };
464 
465 /*
466  * Each of the PCI Device IDs types have their own set of hardware accessor
467  * functions and per device encoding/decoding logic.
468  */
469 struct low_ops {
470 	int (*early_channel_count)	(struct amd64_pvt *pvt);
471 	void (*map_sysaddr_to_csrow)	(struct mem_ctl_info *mci, u64 sys_addr,
472 					 struct err_info *);
473 	int (*dbam_to_cs)		(struct amd64_pvt *pvt, u8 dct,
474 					 unsigned cs_mode, int cs_mask_nr);
475 };
476 
477 struct amd64_family_type {
478 	const char *ctl_name;
479 	u16 f0_id, f1_id, f2_id, f6_id;
480 	/* Maximum number of memory controllers per die/node. */
481 	u8 max_mcs;
482 	struct low_ops ops;
483 };
484 
485 int __amd64_read_pci_cfg_dword(struct pci_dev *pdev, int offset,
486 			       u32 *val, const char *func);
487 int __amd64_write_pci_cfg_dword(struct pci_dev *pdev, int offset,
488 				u32 val, const char *func);
489 
490 #define amd64_read_pci_cfg(pdev, offset, val)	\
491 	__amd64_read_pci_cfg_dword(pdev, offset, val, __func__)
492 
493 #define amd64_write_pci_cfg(pdev, offset, val)	\
494 	__amd64_write_pci_cfg_dword(pdev, offset, val, __func__)
495 
496 #define to_mci(k) container_of(k, struct mem_ctl_info, dev)
497 
498 /* Injection helpers */
499 static inline void disable_caches(void *dummy)
500 {
501 	write_cr0(read_cr0() | X86_CR0_CD);
502 	wbinvd();
503 }
504 
505 static inline void enable_caches(void *dummy)
506 {
507 	write_cr0(read_cr0() & ~X86_CR0_CD);
508 }
509 
510 static inline u8 dram_intlv_en(struct amd64_pvt *pvt, unsigned int i)
511 {
512 	if (pvt->fam == 0x15 && pvt->model >= 0x30) {
513 		u32 tmp;
514 		amd64_read_pci_cfg(pvt->F1, DRAM_CONT_LIMIT, &tmp);
515 		return (u8) tmp & 0xF;
516 	}
517 	return (u8) (pvt->ranges[i].base.lo >> 8) & 0x7;
518 }
519 
520 static inline u8 dhar_valid(struct amd64_pvt *pvt)
521 {
522 	if (pvt->fam == 0x15 && pvt->model >= 0x30) {
523 		u32 tmp;
524 		amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
525 		return (tmp >> 1) & BIT(0);
526 	}
527 	return (pvt)->dhar & BIT(0);
528 }
529 
530 static inline u32 dct_sel_baseaddr(struct amd64_pvt *pvt)
531 {
532 	if (pvt->fam == 0x15 && pvt->model >= 0x30) {
533 		u32 tmp;
534 		amd64_read_pci_cfg(pvt->F1, DRAM_CONT_BASE, &tmp);
535 		return (tmp >> 11) & 0x1FFF;
536 	}
537 	return (pvt)->dct_sel_lo & 0xFFFFF800;
538 }
539