File Index
Files and directories
Sign in
FILE

skx_common.h

File preview with syntax highlighting.

linux-6.19.14/drivers/edac/skx_common.h C
  1/* SPDX-License-Identifier: GPL-2.0 */
  2/*
  3 * Common codes for both the skx_edac driver and Intel 10nm server EDAC driver.
  4 * Originally split out from the skx_edac driver.
  5 *
  6 * Copyright (c) 2018, Intel Corporation.
  7 */
  8
  9#ifndef _SKX_COMM_EDAC_H
 10#define _SKX_COMM_EDAC_H
 11
 12#include <linux/bits.h>
 13#include <asm/mce.h>
 14
 15#define MSG_SIZE		1024
 16
 17/*
 18 * Debug macros
 19 */
 20#define skx_printk(level, fmt, arg...)			\
 21	edac_printk(level, "skx", fmt, ##arg)
 22
 23#define skx_mc_printk(mci, level, fmt, arg...)		\
 24	edac_mc_chipset_printk(mci, level, "skx", fmt, ##arg)
 25
 26/*
 27 * Get a bit field at register value <v>, from bit <lo> to bit <hi>
 28 */
 29#define GET_BITFIELD(v, lo, hi) \
 30	(((v) & GENMASK_ULL((hi), (lo))) >> (lo))
 31
 32#define SKX_NUM_CHANNELS	3	/* Channels per memory controller */
 33#define SKX_NUM_DIMMS		2	/* Max DIMMS per channel */
 34
 35#define I10NM_NUM_DDR_CHANNELS	2
 36#define I10NM_NUM_DDR_DIMMS	2
 37
 38#define I10NM_NUM_HBM_CHANNELS	2
 39#define I10NM_NUM_HBM_DIMMS	1
 40
 41#define I10NM_NUM_CHANNELS	MAX(I10NM_NUM_DDR_CHANNELS, I10NM_NUM_HBM_CHANNELS)
 42#define I10NM_NUM_DIMMS		MAX(I10NM_NUM_DDR_DIMMS, I10NM_NUM_HBM_DIMMS)
 43
 44#define NUM_CHANNELS	MAX(SKX_NUM_CHANNELS, I10NM_NUM_CHANNELS)
 45#define NUM_DIMMS	MAX(SKX_NUM_DIMMS, I10NM_NUM_DIMMS)
 46
 47#define IS_DIMM_PRESENT(r)		GET_BITFIELD(r, 15, 15)
 48#define IS_NVDIMM_PRESENT(r, i)		GET_BITFIELD(r, i, i)
 49
 50#define MCI_MISC_ECC_MODE(m)	(((m) >> 59) & 15)
 51#define MCI_MISC_ECC_DDRT	8	/* read from DDRT */
 52
 53/*
 54 * According to Intel Architecture spec vol 3B,
 55 * Table 15-10 "IA32_MCi_Status [15:0] Compound Error Code Encoding"
 56 * memory errors should fit one of these masks:
 57 *	000f 0000 1mmm cccc (binary)
 58 *	000f 0010 1mmm cccc (binary)	[RAM used as cache]
 59 * where:
 60 *	f = Correction Report Filtering Bit. If 1, subsequent errors
 61 *	    won't be shown
 62 *	mmm = error type
 63 *	cccc = channel
 64 */
 65#define MCACOD_MEM_ERR_MASK	0xef80
 66/*
 67 * Errors from either the memory of the 1-level memory system or the
 68 * 2nd level memory (the slow "far" memory) of the 2-level memory system.
 69 */
 70#define MCACOD_MEM_CTL_ERR	0x80
 71/*
 72 * Errors from the 1st level memory (the fast "near" memory as cache)
 73 * of the 2-level memory system.
 74 */
 75#define MCACOD_EXT_MEM_ERR	0x280
 76
 77/* Max RRL register sets per {,sub-,pseudo-}channel. */
 78#define NUM_RRL_SET		4
 79/* Max RRL registers per set. */
 80#define NUM_RRL_REG		6
 81/* Max correctable error count registers. */
 82#define NUM_CECNT_REG		8
 83
 84/* Modes of RRL register set. */
 85enum rrl_mode {
 86	/* Last read error from patrol scrub. */
 87	LRE_SCRUB,
 88	/* Last read error from demand. */
 89	LRE_DEMAND,
 90	/* First read error from patrol scrub. */
 91	FRE_SCRUB,
 92	/* First read error from demand. */
 93	FRE_DEMAND,
 94};
 95
 96/* RRL registers per {,sub-,pseudo-}channel. */
 97struct reg_rrl {
 98	/* RRL register parts. */
 99	int set_num, reg_num;
100	enum rrl_mode modes[NUM_RRL_SET];
101	u32 offsets[NUM_RRL_SET][NUM_RRL_REG];
102	/* RRL register widths in byte per set. */
103	u8 widths[NUM_RRL_REG];
104	/* RRL control bits of the first register per set. */
105	u32 v_mask;
106	u32 uc_mask;
107	u32 over_mask;
108	u32 en_patspr_mask;
109	u32 noover_mask;
110	u32 en_mask;
111
112	/* CORRERRCNT register parts. */
113	int cecnt_num;
114	u32 cecnt_offsets[NUM_CECNT_REG];
115	u8 cecnt_widths[NUM_CECNT_REG];
116};
117
118/*
119 * Each cpu socket contains some pci devices that provide global
120 * information, and also some that are local to each of the two
121 * memory controllers on the die.
122 */
123struct skx_dev {
124	/* {skx,i10nm}_edac */
125	u8 bus[4];
126	int seg;
127	struct pci_dev *sad_all;
128	struct pci_dev *util_all;
129	struct pci_dev *uracu;
130	struct pci_dev *pcu_cr3;
131	u32 mcroute;
132
133	/* imh_edac */
134	/* System-view MMIO base physical addresses. */
135	u64 mmio_base_h_north;
136	u64 mmio_base_h_south;
137	int pkg;
138
139	int num_imc;
140	struct list_head list;
141	struct skx_imc {
142		/* i10nm_edac */
143		struct pci_dev *mdev;
144
145		/* imh_edac */
146		struct device *dev;
147
148		struct mem_ctl_info *mci;
149		void __iomem *mbase;
150		int chan_mmio_sz;
151		int num_channels; /* channels per memory controller */
152		int num_dimms; /* dimms per channel */
153		bool hbm_mc;
154		u8 mc;	/* system wide mc# */
155		u8 lmc;	/* socket relative mc# */
156		u8 src_id;
157		/*
158		 * Some server BIOS may hide certain memory controllers, and the
159		 * EDAC driver skips those hidden memory controllers. However, the
160		 * ADXL still decodes memory error address using physical memory
161		 * controller indices. The mapping table is used to convert the
162		 * physical indices (reported by ADXL) to the logical indices
163		 * (used the EDAC driver) of present memory controllers during the
164		 * error handling process.
165		 */
166		u8 mc_mapping;
167		struct skx_channel {
168			struct pci_dev	*cdev;
169			struct pci_dev	*edev;
170			/*
171			 * Two groups of RRL control registers per channel to save default RRL
172			 * settings of two {sub-,pseudo-}channels in Linux RRL control mode.
173			 */
174			u32 rrl_ctl[2][NUM_RRL_SET];
175			struct skx_dimm {
176				u8 close_pg;
177				u8 bank_xor_enable;
178				u8 fine_grain_bank;
179				u8 rowbits;
180				u8 colbits;
181			} dimms[NUM_DIMMS];
182		} chan[NUM_CHANNELS];
183	} imc[];
184};
185
186struct skx_pvt {
187	struct skx_imc	*imc;
188};
189
190enum type {
191	SKX,
192	I10NM,
193	SPR,
194	GNR,
195	DMR,
196};
197
198enum {
199	INDEX_SOCKET,
200	INDEX_MEMCTRL,
201	INDEX_CHANNEL,
202	INDEX_DIMM,
203	INDEX_CS,
204	INDEX_NM_FIRST,
205	INDEX_NM_MEMCTRL = INDEX_NM_FIRST,
206	INDEX_NM_CHANNEL,
207	INDEX_NM_DIMM,
208	INDEX_NM_CS,
209	INDEX_MAX
210};
211
212enum error_source {
213	ERR_SRC_1LM,
214	ERR_SRC_2LM_NM,
215	ERR_SRC_2LM_FM,
216	ERR_SRC_NOT_MEMORY,
217};
218
219#define BIT_NM_MEMCTRL	BIT_ULL(INDEX_NM_MEMCTRL)
220#define BIT_NM_CHANNEL	BIT_ULL(INDEX_NM_CHANNEL)
221#define BIT_NM_DIMM	BIT_ULL(INDEX_NM_DIMM)
222#define BIT_NM_CS	BIT_ULL(INDEX_NM_CS)
223
224struct decoded_addr {
225	struct mce *mce;
226	struct skx_dev *dev;
227	u64	addr;
228	int	socket;
229	int	imc;
230	int	channel;
231	u64	chan_addr;
232	int	sktways;
233	int	chanways;
234	int	dimm;
235	int	cs;
236	int	rank;
237	int	channel_rank;
238	u64	rank_address;
239	int	row;
240	int	column;
241	int	bank_address;
242	int	bank_group;
243	bool	decoded_by_adxl;
244};
245
246struct pci_bdf {
247	u32 bus : 8;
248	u32 dev : 5;
249	u32 fun : 3;
250};
251
252struct res_config {
253	enum type type;
254	/* DDR memory controllers per socket */
255	int ddr_imc_num;
256	/* DDR channels per DDR memory controller */
257	int ddr_chan_num;
258	/* DDR DIMMs per DDR memory channel */
259	int ddr_dimm_num;
260	/* Per DDR channel memory-mapped I/O size */
261	int ddr_chan_mmio_sz;
262	/* HBM memory controllers per socket */
263	int hbm_imc_num;
264	/* HBM channels per HBM memory controller */
265	int hbm_chan_num;
266	/* HBM DIMMs per HBM memory channel */
267	int hbm_dimm_num;
268	/* Per HBM channel memory-mapped I/O size */
269	int hbm_chan_mmio_sz;
270	bool support_ddr5;
271	/* RRL register sets per DDR channel */
272	struct reg_rrl *reg_rrl_ddr;
273	/* RRL register sets per HBM channel */
274	struct reg_rrl *reg_rrl_hbm[2];
275	union {
276		/* {skx,i10nm}_edac */
277		struct {
278			/* Configuration agent device ID */
279			unsigned int decs_did;
280			/* Default bus number configuration register offset */
281			int busno_cfg_offset;
282			struct pci_bdf sad_all_bdf;
283			struct pci_bdf pcu_cr3_bdf;
284			struct pci_bdf util_all_bdf;
285			struct pci_bdf uracu_bdf;
286			struct pci_bdf ddr_mdev_bdf;
287			struct pci_bdf hbm_mdev_bdf;
288			int sad_all_offset;
289		};
290		/* imh_edac */
291		struct {
292			/* MMIO base physical address in local package view */
293			u64 mmio_base_l_north;
294			u64 mmio_base_l_south;
295			u64 ddr_imc_base;
296			u64 ddr_reg_mcmtr_offset;
297			u8  ddr_reg_mcmtr_width;
298			u64 ddr_reg_dimmmtr_offset;
299			u8  ddr_reg_dimmmtr_width;
300			u64 ubox_base;
301			u32 ubox_size;
302			u32 ubox_reg_mmio_base_offset;
303			u8  ubox_reg_mmio_base_width;
304			u32 ubox_reg_socket_id_offset;
305			u8  ubox_reg_socket_id_width;
306			u64 pcu_base;
307			u32 pcu_size;
308			u32 pcu_reg_capid3_offset;
309			u8  pcu_reg_capid3_width;
310			u64 sca_base;
311			u32 sca_size;
312			u32 sca_reg_tolm_offset;
313			u8  sca_reg_tolm_width;
314			u32 sca_reg_tohm_offset;
315			u8  sca_reg_tohm_width;
316			u64 ha_base;
317			u32 ha_size;
318			u32 ha_reg_mode_offset;
319			u8  ha_reg_mode_width;
320		};
321	};
322};
323
324typedef int (*get_dimm_config_f)(struct mem_ctl_info *mci,
325				 struct res_config *cfg);
326typedef bool (*skx_decode_f)(struct decoded_addr *res);
327typedef void (*skx_show_retry_log_f)(struct decoded_addr *res, char *msg, int len, bool scrub_err);
328
329int skx_adxl_get(void);
330void skx_adxl_put(void);
331void skx_set_decode(skx_decode_f decode, skx_show_retry_log_f show_retry_log);
332void skx_set_mem_cfg(bool mem_cfg_2lm);
333void skx_set_res_cfg(struct res_config *cfg);
334void skx_init_mc_mapping(struct skx_dev *d);
335void skx_set_mc_mapping(struct skx_dev *d, u8 pmc, u8 lmc);
336
337int skx_get_src_id(struct skx_dev *d, int off, u8 *id);
338
339int skx_get_all_bus_mappings(struct res_config *cfg, struct list_head **list);
340
341struct list_head *skx_get_edac_list(void);
342
343int skx_get_hi_lo(unsigned int did, int off[], u64 *tolm, u64 *tohm);
344void skx_set_hi_lo(u64 tolm, u64 tohm);
345
346int skx_get_dimm_info(u32 mtr, u32 mcmtr, u32 amap, struct dimm_info *dimm,
347		      struct skx_imc *imc, int chan, int dimmno,
348		      struct res_config *cfg);
349
350int skx_get_nvdimm_info(struct dimm_info *dimm, struct skx_imc *imc,
351			int chan, int dimmno, const char *mod_str);
352
353int skx_register_mci(struct skx_imc *imc, struct device *dev, const char *dev_name,
354		     const char *ctl_name, const char *mod_str,
355		     get_dimm_config_f get_dimm_config,
356		     struct res_config *cfg);
357
358int skx_mce_check_error(struct notifier_block *nb, unsigned long val,
359			void *data);
360
361void skx_remove(void);
362
363#ifdef CONFIG_EDAC_DEBUG
364void skx_setup_debug(const char *name);
365void skx_teardown_debug(void);
366#else
367static inline void skx_setup_debug(const char *name) {}
368static inline void skx_teardown_debug(void) {}
369#endif
370
371#endif /* _SKX_COMM_EDAC_H */
~/ Read-only file preview.