File Index
Files and directories
Sign in
FILE

ie31200_edac.c

File preview with syntax highlighting.

linux-6.19.14/drivers/edac/ie31200_edac.c C
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Intel E3-1200
  4 * Copyright (C) 2014 Jason Baron <jbaron@akamai.com>
  5 *
  6 * Support for the E3-1200 processor family. Heavily based on previous
  7 * Intel EDAC drivers.
  8 *
  9 * Since the DRAM controller is on the cpu chip, we can use its PCI device
 10 * id to identify these processors.
 11 *
 12 * PCI DRAM controller device ids (Taken from The PCI ID Repository - https://pci-ids.ucw.cz/)
 13 *
 14 * 0108: Xeon E3-1200 Processor Family DRAM Controller
 15 * 010c: Xeon E3-1200/2nd Generation Core Processor Family DRAM Controller
 16 * 0150: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
 17 * 0158: Xeon E3-1200 v2/Ivy Bridge DRAM Controller
 18 * 015c: Xeon E3-1200 v2/3rd Gen Core processor DRAM Controller
 19 * 0c04: Xeon E3-1200 v3/4th Gen Core Processor DRAM Controller
 20 * 0c08: Xeon E3-1200 v3 Processor DRAM Controller
 21 * 1918: Xeon E3-1200 v5 Skylake Host Bridge/DRAM Registers
 22 * 590f: Xeon E3-1200 v6/7th Gen Core Processor Host Bridge/DRAM Registers
 23 * 5918: Xeon E3-1200 v6/7th Gen Core Processor Host Bridge/DRAM Registers
 24 * 190f: 6th Gen Core Dual-Core Processor Host Bridge/DRAM Registers
 25 * 191f: 6th Gen Core Quad-Core Processor Host Bridge/DRAM Registers
 26 * 3e..: 8th/9th Gen Core Processor Host Bridge/DRAM Registers
 27 *
 28 * Based on Intel specification:
 29 * https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-e3-1200v3-vol-2-datasheet.pdf
 30 * http://www.intel.com/content/www/us/en/processors/xeon/xeon-e3-1200-family-vol-2-datasheet.html
 31 * https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/desktop-6th-gen-core-family-datasheet-vol-2.pdf
 32 * https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/xeon-e3-1200v6-vol-2-datasheet.pdf
 33 * https://www.intel.com/content/www/us/en/processors/core/7th-gen-core-family-mobile-h-processor-lines-datasheet-vol-2.html
 34 * https://www.intel.com/content/www/us/en/products/docs/processors/core/8th-gen-core-family-datasheet-vol-2.html
 35 *
 36 * According to the above datasheet (p.16):
 37 * "
 38 * 6. Software must not access B0/D0/F0 32-bit memory-mapped registers with
 39 * requests that cross a DW boundary.
 40 * "
 41 *
 42 * Thus, we make use of the explicit: lo_hi_readq(), which breaks the readq into
 43 * 2 readl() calls. This restriction may be lifted in subsequent chip releases,
 44 * but lo_hi_readq() ensures that we are safe across all e3-1200 processors.
 45 */
 46
 47#include <linux/bitfield.h>
 48#include <linux/module.h>
 49#include <linux/init.h>
 50#include <linux/pci.h>
 51#include <linux/pci_ids.h>
 52#include <linux/edac.h>
 53
 54#include <linux/io-64-nonatomic-lo-hi.h>
 55#include <asm/mce.h>
 56#include <asm/msr.h>
 57#include "edac_module.h"
 58
 59#define EDAC_MOD_STR "ie31200_edac"
 60
 61#define ie31200_printk(level, fmt, arg...) \
 62	edac_printk(level, "ie31200", fmt, ##arg)
 63
 64#define PCI_DEVICE_ID_INTEL_IE31200_HB_1  0x0108
 65#define PCI_DEVICE_ID_INTEL_IE31200_HB_2  0x010c
 66#define PCI_DEVICE_ID_INTEL_IE31200_HB_3  0x0150
 67#define PCI_DEVICE_ID_INTEL_IE31200_HB_4  0x0158
 68#define PCI_DEVICE_ID_INTEL_IE31200_HB_5  0x015c
 69#define PCI_DEVICE_ID_INTEL_IE31200_HB_6  0x0c04
 70#define PCI_DEVICE_ID_INTEL_IE31200_HB_7  0x0c08
 71#define PCI_DEVICE_ID_INTEL_IE31200_HB_8  0x190F
 72#define PCI_DEVICE_ID_INTEL_IE31200_HB_9  0x1918
 73#define PCI_DEVICE_ID_INTEL_IE31200_HB_10 0x191F
 74#define PCI_DEVICE_ID_INTEL_IE31200_HB_11 0x590f
 75#define PCI_DEVICE_ID_INTEL_IE31200_HB_12 0x5918
 76
 77/* Coffee Lake-S */
 78#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_MASK 0x3e00
 79#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_1    0x3e0f
 80#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_2    0x3e18
 81#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_3    0x3e1f
 82#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_4    0x3e30
 83#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_5    0x3e31
 84#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_6    0x3e32
 85#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_7    0x3e33
 86#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_8    0x3ec2
 87#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_9    0x3ec6
 88#define PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_10   0x3eca
 89
 90/* Raptor Lake-S */
 91#define PCI_DEVICE_ID_INTEL_IE31200_RPL_S_1	0xa703 /* 8P+8E,  e.g. i7-13700 */
 92#define PCI_DEVICE_ID_INTEL_IE31200_RPL_S_2	0x4640 /* 6P+8E,  e.g. i5-13500, i5-13600, i5-14500 */
 93#define PCI_DEVICE_ID_INTEL_IE31200_RPL_S_3	0x4630 /* 4P+0E,  e.g. i3-13100E */
 94#define PCI_DEVICE_ID_INTEL_IE31200_RPL_S_4	0xa700 /* 8P+16E, e.g. i9-13900, i9-14900 */
 95#define PCI_DEVICE_ID_INTEL_IE31200_RPL_S_5	0xa740 /* 8P+12E, e.g. i7-14700 */
 96#define PCI_DEVICE_ID_INTEL_IE31200_RPL_S_6	0xa704 /* 6P+8E,  e.g. i5-14600 */
 97
 98/* Raptor Lake-HX */
 99#define PCI_DEVICE_ID_INTEL_IE31200_RPL_HX_1	0xa702 /* 8P+16E, e.g. i9-13950HX */
100
101/* Alder Lake-S */
102#define PCI_DEVICE_ID_INTEL_IE31200_ADL_S_1	0x4660
103#define PCI_DEVICE_ID_INTEL_IE31200_ADL_S_2	0x4668	/* 8P+4E, e.g. i7-12700K */
104#define PCI_DEVICE_ID_INTEL_IE31200_ADL_S_3	0x4648	/* 6P+4E, e.g. i5-12600K */
105
106/* Bartlett Lake-S */
107#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_1	0x4639
108#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_2	0x463c
109#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_3	0x4642
110#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_4	0x4643
111#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_5	0xa731
112#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_6	0xa732
113#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_7	0xa733
114#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_8	0xa741
115#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_9	0xa744
116#define PCI_DEVICE_ID_INTEL_IE31200_BTL_S_10	0xa745
117
118#define IE31200_RANKS_PER_CHANNEL	8
119#define IE31200_DIMMS_PER_CHANNEL	2
120#define IE31200_CHANNELS		2
121#define IE31200_IMC_NUM			2
122
123/* Intel IE31200 register addresses - device 0 function 0 - DRAM Controller */
124#define IE31200_MCHBAR_LOW		0x48
125#define IE31200_MCHBAR_HIGH		0x4c
126
127/*
128 * Error Status Register (16b)
129 *
130 *  1    Multi-bit DRAM ECC Error Flag (DMERR)
131 *  0    Single-bit DRAM ECC Error Flag (DSERR)
132 */
133#define IE31200_ERRSTS			0xc8
134#define IE31200_ERRSTS_UE		BIT(1)
135#define IE31200_ERRSTS_CE		BIT(0)
136#define IE31200_ERRSTS_BITS		(IE31200_ERRSTS_UE | IE31200_ERRSTS_CE)
137
138#define IE31200_CAPID0			0xe4
139#define IE31200_CAPID0_PDCD		BIT(4)
140#define IE31200_CAPID0_DDPCD		BIT(6)
141#define IE31200_CAPID0_ECC		BIT(1)
142
143static int nr_channels;
144static struct pci_dev *mci_pdev;
145static int ie31200_registered = 1;
146
147struct res_config {
148	enum mem_type mtype;
149	bool cmci;
150	int imc_num;
151	/* Host MMIO configuration register */
152	u64 reg_mchbar_mask;
153	u64 reg_mchbar_window_size;
154	/* ECC error log register */
155	u64 reg_eccerrlog_offset[IE31200_CHANNELS];
156	u64 reg_eccerrlog_ce_mask;
157	u64 reg_eccerrlog_ce_ovfl_mask;
158	u64 reg_eccerrlog_ue_mask;
159	u64 reg_eccerrlog_ue_ovfl_mask;
160	u64 reg_eccerrlog_rank_mask;
161	u64 reg_eccerrlog_syndrome_mask;
162	/* MSR to clear ECC error log register */
163	u32 msr_clear_eccerrlog_offset;
164	/* DIMM characteristics register */
165	u64 reg_mad_dimm_size_granularity;
166	u64 reg_mad_dimm_offset[IE31200_CHANNELS];
167	u32 reg_mad_dimm_size_mask[IE31200_DIMMS_PER_CHANNEL];
168	u32 reg_mad_dimm_rank_mask[IE31200_DIMMS_PER_CHANNEL];
169	u32 reg_mad_dimm_width_mask[IE31200_DIMMS_PER_CHANNEL];
170};
171
172struct ie31200_priv {
173	void __iomem *window;
174	void __iomem *c0errlog;
175	void __iomem *c1errlog;
176	struct res_config *cfg;
177	struct mem_ctl_info *mci;
178	struct pci_dev *pdev;
179	struct device dev;
180};
181
182static struct ie31200_pvt {
183	struct ie31200_priv *priv[IE31200_IMC_NUM];
184} ie31200_pvt;
185
186enum ie31200_chips {
187	IE31200 = 0,
188	IE31200_1 = 1,
189};
190
191struct ie31200_dev_info {
192	const char *ctl_name;
193};
194
195struct ie31200_error_info {
196	u16 errsts;
197	u16 errsts2;
198	u64 eccerrlog[IE31200_CHANNELS];
199	u64 erraddr;
200};
201
202static const struct ie31200_dev_info ie31200_devs[] = {
203	[IE31200] = {
204		.ctl_name = "IE31200"
205	},
206	[IE31200_1] = {
207		.ctl_name = "IE31200_1"
208	},
209};
210
211struct dimm_data {
212	u64 size; /* in bytes */
213	u8  ranks;
214	enum dev_type dtype;
215};
216
217static int how_many_channels(struct pci_dev *pdev)
218{
219	int n_channels;
220	unsigned char capid0_2b; /* 2nd byte of CAPID0 */
221
222	pci_read_config_byte(pdev, IE31200_CAPID0 + 1, &capid0_2b);
223
224	/* check PDCD: Dual Channel Disable */
225	if (capid0_2b & IE31200_CAPID0_PDCD) {
226		edac_dbg(0, "In single channel mode\n");
227		n_channels = 1;
228	} else {
229		edac_dbg(0, "In dual channel mode\n");
230		n_channels = 2;
231	}
232
233	/* check DDPCD - check if both channels are filled */
234	if (capid0_2b & IE31200_CAPID0_DDPCD)
235		edac_dbg(0, "2 DIMMS per channel disabled\n");
236	else
237		edac_dbg(0, "2 DIMMS per channel enabled\n");
238
239	return n_channels;
240}
241
242static bool ecc_capable(struct pci_dev *pdev)
243{
244	unsigned char capid0_4b; /* 4th byte of CAPID0 */
245
246	pci_read_config_byte(pdev, IE31200_CAPID0 + 3, &capid0_4b);
247	if (capid0_4b & IE31200_CAPID0_ECC)
248		return false;
249	return true;
250}
251
252#define mci_to_pci_dev(mci)	(((struct ie31200_priv *)(mci)->pvt_info)->pdev)
253
254static void ie31200_clear_error_info(struct mem_ctl_info *mci)
255{
256	struct ie31200_priv *priv = mci->pvt_info;
257	struct res_config *cfg = priv->cfg;
258
259	/*
260	 * The PCI ERRSTS register is deprecated. Write the MSR to clear
261	 * the ECC error log registers in all memory controllers.
262	 */
263	if (cfg->msr_clear_eccerrlog_offset) {
264		if (wrmsr_safe(cfg->msr_clear_eccerrlog_offset,
265			       cfg->reg_eccerrlog_ce_mask |
266			       cfg->reg_eccerrlog_ce_ovfl_mask |
267			       cfg->reg_eccerrlog_ue_mask |
268			       cfg->reg_eccerrlog_ue_ovfl_mask, 0) < 0)
269			ie31200_printk(KERN_ERR, "Failed to wrmsr.\n");
270
271		return;
272	}
273
274	/*
275	 * Clear any error bits.
276	 * (Yes, we really clear bits by writing 1 to them.)
277	 */
278	pci_write_bits16(mci_to_pci_dev(mci), IE31200_ERRSTS,
279			 IE31200_ERRSTS_BITS, IE31200_ERRSTS_BITS);
280}
281
282static void ie31200_get_and_clear_error_info(struct mem_ctl_info *mci,
283					     struct ie31200_error_info *info)
284{
285	struct pci_dev *pdev = mci_to_pci_dev(mci);
286	struct ie31200_priv *priv = mci->pvt_info;
287
288	/*
289	 * The PCI ERRSTS register is deprecated, directly read the
290	 * MMIO-mapped ECC error log registers.
291	 */
292	if (priv->cfg->msr_clear_eccerrlog_offset) {
293		info->eccerrlog[0] = lo_hi_readq(priv->c0errlog);
294		if (nr_channels == 2)
295			info->eccerrlog[1] = lo_hi_readq(priv->c1errlog);
296
297		ie31200_clear_error_info(mci);
298		return;
299	}
300
301	/*
302	 * This is a mess because there is no atomic way to read all the
303	 * registers at once and the registers can transition from CE being
304	 * overwritten by UE.
305	 */
306	pci_read_config_word(pdev, IE31200_ERRSTS, &info->errsts);
307	if (!(info->errsts & IE31200_ERRSTS_BITS))
308		return;
309
310	info->eccerrlog[0] = lo_hi_readq(priv->c0errlog);
311	if (nr_channels == 2)
312		info->eccerrlog[1] = lo_hi_readq(priv->c1errlog);
313
314	pci_read_config_word(pdev, IE31200_ERRSTS, &info->errsts2);
315
316	/*
317	 * If the error is the same for both reads then the first set
318	 * of reads is valid.  If there is a change then there is a CE
319	 * with no info and the second set of reads is valid and
320	 * should be UE info.
321	 */
322	if ((info->errsts ^ info->errsts2) & IE31200_ERRSTS_BITS) {
323		info->eccerrlog[0] = lo_hi_readq(priv->c0errlog);
324		if (nr_channels == 2)
325			info->eccerrlog[1] =
326				lo_hi_readq(priv->c1errlog);
327	}
328
329	ie31200_clear_error_info(mci);
330}
331
332static void ie31200_process_error_info(struct mem_ctl_info *mci,
333				       struct ie31200_error_info *info)
334{
335	struct ie31200_priv *priv = mci->pvt_info;
336	struct res_config *cfg = priv->cfg;
337	int channel;
338	u64 log;
339
340	if (!cfg->msr_clear_eccerrlog_offset) {
341		if (!(info->errsts & IE31200_ERRSTS_BITS))
342			return;
343
344		if ((info->errsts ^ info->errsts2) & IE31200_ERRSTS_BITS) {
345			edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0,
346					     -1, -1, -1, "UE overwrote CE", "");
347			info->errsts = info->errsts2;
348		}
349	}
350
351	for (channel = 0; channel < nr_channels; channel++) {
352		log = info->eccerrlog[channel];
353		if (log & cfg->reg_eccerrlog_ue_mask) {
354			edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
355					     info->erraddr >> PAGE_SHIFT, 0, 0,
356					     field_get(cfg->reg_eccerrlog_rank_mask, log),
357					     channel, -1,
358					     "ie31200 UE", "");
359		} else if (log & cfg->reg_eccerrlog_ce_mask) {
360			edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
361					     info->erraddr >> PAGE_SHIFT, 0,
362					     field_get(cfg->reg_eccerrlog_syndrome_mask, log),
363					     field_get(cfg->reg_eccerrlog_rank_mask, log),
364					     channel, -1,
365					     "ie31200 CE", "");
366		}
367	}
368}
369
370static void __ie31200_check(struct mem_ctl_info *mci, struct mce *mce)
371{
372	struct ie31200_error_info info;
373
374	info.erraddr = mce ? mce->addr : 0;
375	ie31200_get_and_clear_error_info(mci, &info);
376	ie31200_process_error_info(mci, &info);
377}
378
379static void ie31200_check(struct mem_ctl_info *mci)
380{
381	__ie31200_check(mci, NULL);
382}
383
384static void __iomem *ie31200_map_mchbar(struct pci_dev *pdev, struct res_config *cfg, int mc)
385{
386	union {
387		u64 mchbar;
388		struct {
389			u32 mchbar_low;
390			u32 mchbar_high;
391		};
392	} u;
393	void __iomem *window;
394
395	pci_read_config_dword(pdev, IE31200_MCHBAR_LOW, &u.mchbar_low);
396	pci_read_config_dword(pdev, IE31200_MCHBAR_HIGH, &u.mchbar_high);
397	u.mchbar &= cfg->reg_mchbar_mask;
398	u.mchbar += cfg->reg_mchbar_window_size * mc;
399
400	if (u.mchbar != (resource_size_t)u.mchbar) {
401		ie31200_printk(KERN_ERR, "mmio space beyond accessible range (0x%llx)\n",
402			       (unsigned long long)u.mchbar);
403		return NULL;
404	}
405
406	window = ioremap(u.mchbar, cfg->reg_mchbar_window_size);
407	if (!window)
408		ie31200_printk(KERN_ERR, "Cannot map mmio space at 0x%llx\n",
409			       (unsigned long long)u.mchbar);
410
411	return window;
412}
413
414static void populate_dimm_info(struct dimm_data *dd, u32 addr_decode, int dimm,
415			       struct res_config *cfg)
416{
417	dd->size = field_get(cfg->reg_mad_dimm_size_mask[dimm], addr_decode) * cfg->reg_mad_dimm_size_granularity;
418	dd->ranks = field_get(cfg->reg_mad_dimm_rank_mask[dimm], addr_decode) + 1;
419	dd->dtype = field_get(cfg->reg_mad_dimm_width_mask[dimm], addr_decode) + DEV_X8;
420}
421
422static void ie31200_get_dimm_config(struct mem_ctl_info *mci, void __iomem *window,
423				    struct res_config *cfg, int mc)
424{
425	struct dimm_data dimm_info;
426	struct dimm_info *dimm;
427	unsigned long nr_pages;
428	u32 addr_decode;
429	int i, j, k;
430
431	for (i = 0; i < IE31200_CHANNELS; i++) {
432		addr_decode = readl(window + cfg->reg_mad_dimm_offset[i]);
433		edac_dbg(0, "addr_decode: 0x%x\n", addr_decode);
434
435		for (j = 0; j < IE31200_DIMMS_PER_CHANNEL; j++) {
436			populate_dimm_info(&dimm_info, addr_decode, j, cfg);
437			edac_dbg(0, "mc: %d, channel: %d, dimm: %d, size: %lld MiB, ranks: %d, DRAM chip type: %d\n",
438				 mc, i, j, dimm_info.size >> 20,
439				 dimm_info.ranks,
440				 dimm_info.dtype);
441
442			nr_pages = MiB_TO_PAGES(dimm_info.size >> 20);
443			if (nr_pages == 0)
444				continue;
445
446			nr_pages = nr_pages / dimm_info.ranks;
447			for (k = 0; k < dimm_info.ranks; k++) {
448				dimm = edac_get_dimm(mci, (j * dimm_info.ranks) + k, i, 0);
449				dimm->nr_pages = nr_pages;
450				edac_dbg(0, "set nr pages: 0x%lx\n", nr_pages);
451				dimm->grain = 8; /* just a guess */
452				dimm->mtype = cfg->mtype;
453				dimm->dtype = dimm_info.dtype;
454				dimm->edac_mode = EDAC_UNKNOWN;
455			}
456		}
457	}
458}
459
460static int ie31200_register_mci(struct pci_dev *pdev, struct res_config *cfg, int mc)
461{
462	struct edac_mc_layer layers[2];
463	struct ie31200_priv *priv;
464	struct mem_ctl_info *mci;
465	void __iomem *window;
466	int ret;
467
468	nr_channels = how_many_channels(pdev);
469	layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
470	layers[0].size = IE31200_RANKS_PER_CHANNEL;
471	layers[0].is_virt_csrow = true;
472	layers[1].type = EDAC_MC_LAYER_CHANNEL;
473	layers[1].size = nr_channels;
474	layers[1].is_virt_csrow = false;
475	mci = edac_mc_alloc(mc, ARRAY_SIZE(layers), layers,
476			    sizeof(struct ie31200_priv));
477	if (!mci)
478		return -ENOMEM;
479
480	window = ie31200_map_mchbar(pdev, cfg, mc);
481	if (!window) {
482		ret = -ENODEV;
483		goto fail_free;
484	}
485
486	edac_dbg(3, "MC: init mci\n");
487	mci->mtype_cap = BIT(cfg->mtype);
488	mci->edac_ctl_cap = EDAC_FLAG_SECDED;
489	mci->edac_cap = EDAC_FLAG_SECDED;
490	mci->mod_name = EDAC_MOD_STR;
491	mci->ctl_name = ie31200_devs[mc].ctl_name;
492	mci->dev_name = pci_name(pdev);
493	mci->edac_check = cfg->cmci ? NULL : ie31200_check;
494	mci->ctl_page_to_phys = NULL;
495	priv = mci->pvt_info;
496	priv->window = window;
497	priv->c0errlog = window + cfg->reg_eccerrlog_offset[0];
498	priv->c1errlog = window + cfg->reg_eccerrlog_offset[1];
499	priv->cfg = cfg;
500	priv->mci = mci;
501	priv->pdev = pdev;
502	device_initialize(&priv->dev);
503	/*
504	 * The EDAC core uses mci->pdev (pointer to the structure device)
505	 * as the memory controller ID. The SoCs attach one or more memory
506	 * controllers to a single pci_dev (a single pci_dev->dev can
507	 * correspond to multiple memory controllers).
508	 *
509	 * To make mci->pdev unique, assign pci_dev->dev to mci->pdev
510	 * for the first memory controller and assign a unique priv->dev
511	 * to mci->pdev for each additional memory controller.
512	 */
513	mci->pdev = mc ? &priv->dev : &pdev->dev;
514
515	ie31200_get_dimm_config(mci, window, cfg, mc);
516	ie31200_clear_error_info(mci);
517
518	if (edac_mc_add_mc(mci)) {
519		edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
520		ret = -ENODEV;
521		goto fail_unmap;
522	}
523
524	ie31200_pvt.priv[mc] = priv;
525	return 0;
526fail_unmap:
527	put_device(&priv->dev);
528	iounmap(window);
529fail_free:
530	edac_mc_free(mci);
531	return ret;
532}
533
534static void mce_check(struct mce *mce)
535{
536	struct ie31200_priv *priv;
537	int i;
538
539	for (i = 0; i < IE31200_IMC_NUM; i++) {
540		priv = ie31200_pvt.priv[i];
541		if (!priv)
542			continue;
543
544		__ie31200_check(priv->mci, mce);
545	}
546}
547
548static int mce_handler(struct notifier_block *nb, unsigned long val, void *data)
549{
550	struct mce *mce = (struct mce *)data;
551	char *type;
552
553	if (mce->kflags & MCE_HANDLED_CEC)
554		return NOTIFY_DONE;
555
556	/*
557	 * Ignore unless this is a memory related error.
558	 * Don't check MCI_STATUS_ADDRV since it's not set on some CPUs.
559	 */
560	if ((mce->status & 0xefff) >> 7 != 1)
561		return NOTIFY_DONE;
562
563	type = mce->mcgstatus & MCG_STATUS_MCIP ?  "Exception" : "Event";
564
565	edac_dbg(0, "CPU %d: Machine Check %s: 0x%llx Bank %d: 0x%llx\n",
566		 mce->extcpu, type, mce->mcgstatus,
567		 mce->bank, mce->status);
568	edac_dbg(0, "TSC 0x%llx\n", mce->tsc);
569	edac_dbg(0, "ADDR 0x%llx\n", mce->addr);
570	edac_dbg(0, "MISC 0x%llx\n", mce->misc);
571	edac_dbg(0, "PROCESSOR %u:0x%x TIME %llu SOCKET %u APIC 0x%x\n",
572		 mce->cpuvendor, mce->cpuid, mce->time,
573		 mce->socketid, mce->apicid);
574
575	mce_check(mce);
576	mce->kflags |= MCE_HANDLED_EDAC;
577
578	return NOTIFY_DONE;
579}
580
581static struct notifier_block ie31200_mce_dec = {
582	.notifier_call	= mce_handler,
583	.priority	= MCE_PRIO_EDAC,
584};
585
586static void ie31200_unregister_mcis(void)
587{
588	struct ie31200_priv *priv;
589	struct mem_ctl_info *mci;
590	int i;
591
592	for (i = 0; i < IE31200_IMC_NUM; i++) {
593		priv = ie31200_pvt.priv[i];
594		if (!priv)
595			continue;
596
597		mci = priv->mci;
598		edac_mc_del_mc(mci->pdev);
599		iounmap(priv->window);
600		put_device(&priv->dev);
601		edac_mc_free(mci);
602	}
603}
604
605static int ie31200_probe1(struct pci_dev *pdev, struct res_config *cfg)
606{
607	int i, ret;
608
609	edac_dbg(0, "MC:\n");
610
611	if (!ecc_capable(pdev)) {
612		ie31200_printk(KERN_INFO, "No ECC support\n");
613		return -ENODEV;
614	}
615
616	for (i = 0; i < cfg->imc_num; i++) {
617		ret = ie31200_register_mci(pdev, cfg, i);
618		if (ret)
619			goto fail_register;
620	}
621
622	if (cfg->cmci) {
623		mce_register_decode_chain(&ie31200_mce_dec);
624		edac_op_state = EDAC_OPSTATE_INT;
625	} else {
626		edac_op_state = EDAC_OPSTATE_POLL;
627	}
628
629	/* get this far and it's successful. */
630	edac_dbg(3, "MC: success\n");
631	return 0;
632
633fail_register:
634	ie31200_unregister_mcis();
635	return ret;
636}
637
638static int ie31200_init_one(struct pci_dev *pdev,
639			    const struct pci_device_id *ent)
640{
641	int rc;
642
643	edac_dbg(0, "MC:\n");
644	if (pci_enable_device(pdev) < 0)
645		return -EIO;
646	rc = ie31200_probe1(pdev, (struct res_config *)ent->driver_data);
647	if (rc == 0 && !mci_pdev)
648		mci_pdev = pci_dev_get(pdev);
649
650	return rc;
651}
652
653static void ie31200_remove_one(struct pci_dev *pdev)
654{
655	struct ie31200_priv *priv = ie31200_pvt.priv[0];
656
657	edac_dbg(0, "\n");
658	pci_dev_put(mci_pdev);
659	mci_pdev = NULL;
660	if (priv->cfg->cmci)
661		mce_unregister_decode_chain(&ie31200_mce_dec);
662	ie31200_unregister_mcis();
663}
664
665static struct res_config snb_cfg = {
666	.mtype				= MEM_DDR3,
667	.imc_num			= 1,
668	.reg_mchbar_mask		= GENMASK_ULL(38, 15),
669	.reg_mchbar_window_size		= BIT_ULL(15),
670	.reg_eccerrlog_offset[0]	= 0x40c8,
671	.reg_eccerrlog_offset[1]	= 0x44c8,
672	.reg_eccerrlog_ce_mask		= BIT_ULL(0),
673	.reg_eccerrlog_ue_mask		= BIT_ULL(1),
674	.reg_eccerrlog_rank_mask	= GENMASK_ULL(28, 27),
675	.reg_eccerrlog_syndrome_mask	= GENMASK_ULL(23, 16),
676	.reg_mad_dimm_size_granularity	= BIT_ULL(28),
677	.reg_mad_dimm_offset[0]		= 0x5004,
678	.reg_mad_dimm_offset[1]		= 0x5008,
679	.reg_mad_dimm_size_mask[0]	= GENMASK(7, 0),
680	.reg_mad_dimm_size_mask[1]	= GENMASK(15, 8),
681	.reg_mad_dimm_rank_mask[0]	= BIT(17),
682	.reg_mad_dimm_rank_mask[1]	= BIT(18),
683	.reg_mad_dimm_width_mask[0]	= BIT(19),
684	.reg_mad_dimm_width_mask[1]	= BIT(20),
685};
686
687static struct res_config skl_cfg = {
688	.mtype				= MEM_DDR4,
689	.imc_num			= 1,
690	.reg_mchbar_mask		= GENMASK_ULL(38, 15),
691	.reg_mchbar_window_size		= BIT_ULL(15),
692	.reg_eccerrlog_offset[0]	= 0x4048,
693	.reg_eccerrlog_offset[1]	= 0x4448,
694	.reg_eccerrlog_ce_mask		= BIT_ULL(0),
695	.reg_eccerrlog_ue_mask		= BIT_ULL(1),
696	.reg_eccerrlog_rank_mask	= GENMASK_ULL(28, 27),
697	.reg_eccerrlog_syndrome_mask	= GENMASK_ULL(23, 16),
698	.reg_mad_dimm_size_granularity	= BIT_ULL(30),
699	.reg_mad_dimm_offset[0]		= 0x500c,
700	.reg_mad_dimm_offset[1]		= 0x5010,
701	.reg_mad_dimm_size_mask[0]	= GENMASK(5, 0),
702	.reg_mad_dimm_size_mask[1]	= GENMASK(21, 16),
703	.reg_mad_dimm_rank_mask[0]	= BIT(10),
704	.reg_mad_dimm_rank_mask[1]	= BIT(26),
705	.reg_mad_dimm_width_mask[0]	= GENMASK(9, 8),
706	.reg_mad_dimm_width_mask[1]	= GENMASK(25, 24),
707};
708
709struct res_config rpl_s_cfg = {
710	.mtype				= MEM_DDR5,
711	.cmci				= true,
712	.imc_num			= 2,
713	.reg_mchbar_mask		= GENMASK_ULL(41, 17),
714	.reg_mchbar_window_size		= BIT_ULL(16),
715	.reg_eccerrlog_offset[0]	= 0xe048,
716	.reg_eccerrlog_offset[1]	= 0xe848,
717	.reg_eccerrlog_ce_mask		= BIT_ULL(0),
718	.reg_eccerrlog_ce_ovfl_mask	= BIT_ULL(1),
719	.reg_eccerrlog_ue_mask		= BIT_ULL(2),
720	.reg_eccerrlog_ue_ovfl_mask	= BIT_ULL(3),
721	.reg_eccerrlog_rank_mask	= GENMASK_ULL(28, 27),
722	.reg_eccerrlog_syndrome_mask	= GENMASK_ULL(23, 16),
723	.msr_clear_eccerrlog_offset	= 0x791,
724	.reg_mad_dimm_offset[0]		= 0xd80c,
725	.reg_mad_dimm_offset[1]		= 0xd810,
726	.reg_mad_dimm_size_granularity	= BIT_ULL(29),
727	.reg_mad_dimm_size_mask[0]	= GENMASK(6, 0),
728	.reg_mad_dimm_size_mask[1]	= GENMASK(22, 16),
729	.reg_mad_dimm_rank_mask[0]	= GENMASK(10, 9),
730	.reg_mad_dimm_rank_mask[1]	= GENMASK(27, 26),
731	.reg_mad_dimm_width_mask[0]	= GENMASK(8, 7),
732	.reg_mad_dimm_width_mask[1]	= GENMASK(25, 24),
733};
734
735static const struct pci_device_id ie31200_pci_tbl[] = {
736	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_1), (kernel_ulong_t)&snb_cfg },
737	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_2), (kernel_ulong_t)&snb_cfg },
738	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_3), (kernel_ulong_t)&snb_cfg },
739	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_4), (kernel_ulong_t)&snb_cfg },
740	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_5), (kernel_ulong_t)&snb_cfg },
741	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_6), (kernel_ulong_t)&snb_cfg },
742	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_7), (kernel_ulong_t)&snb_cfg },
743	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_8), (kernel_ulong_t)&skl_cfg },
744	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_9), (kernel_ulong_t)&skl_cfg },
745	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_10), (kernel_ulong_t)&skl_cfg },
746	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_11), (kernel_ulong_t)&skl_cfg },
747	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_12), (kernel_ulong_t)&skl_cfg },
748	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_1), (kernel_ulong_t)&skl_cfg },
749	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_2), (kernel_ulong_t)&skl_cfg },
750	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_3), (kernel_ulong_t)&skl_cfg },
751	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_4), (kernel_ulong_t)&skl_cfg },
752	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_5), (kernel_ulong_t)&skl_cfg },
753	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_6), (kernel_ulong_t)&skl_cfg },
754	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_7), (kernel_ulong_t)&skl_cfg },
755	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_8), (kernel_ulong_t)&skl_cfg },
756	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_9), (kernel_ulong_t)&skl_cfg },
757	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_HB_CFL_10), (kernel_ulong_t)&skl_cfg },
758	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_RPL_S_1), (kernel_ulong_t)&rpl_s_cfg},
759	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_RPL_S_2), (kernel_ulong_t)&rpl_s_cfg},
760	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_RPL_S_3), (kernel_ulong_t)&rpl_s_cfg},
761	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_RPL_S_4), (kernel_ulong_t)&rpl_s_cfg},
762	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_RPL_S_5), (kernel_ulong_t)&rpl_s_cfg},
763	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_RPL_S_6), (kernel_ulong_t)&rpl_s_cfg},
764	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_RPL_HX_1), (kernel_ulong_t)&rpl_s_cfg},
765	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_ADL_S_1), (kernel_ulong_t)&rpl_s_cfg},
766	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_ADL_S_2), (kernel_ulong_t)&rpl_s_cfg},
767	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_ADL_S_3), (kernel_ulong_t)&rpl_s_cfg},
768	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_1), (kernel_ulong_t)&rpl_s_cfg},
769	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_2), (kernel_ulong_t)&rpl_s_cfg},
770	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_3), (kernel_ulong_t)&rpl_s_cfg},
771	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_4), (kernel_ulong_t)&rpl_s_cfg},
772	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_5), (kernel_ulong_t)&rpl_s_cfg},
773	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_6), (kernel_ulong_t)&rpl_s_cfg},
774	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_7), (kernel_ulong_t)&rpl_s_cfg},
775	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_8), (kernel_ulong_t)&rpl_s_cfg},
776	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_9), (kernel_ulong_t)&rpl_s_cfg},
777	{ PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IE31200_BTL_S_10), (kernel_ulong_t)&rpl_s_cfg},
778	{ 0, } /* 0 terminated list. */
779};
780MODULE_DEVICE_TABLE(pci, ie31200_pci_tbl);
781
782static struct pci_driver ie31200_driver = {
783	.name = EDAC_MOD_STR,
784	.probe = ie31200_init_one,
785	.remove = ie31200_remove_one,
786	.id_table = ie31200_pci_tbl,
787};
788
789static int __init ie31200_init(void)
790{
791	int pci_rc, i;
792
793	edac_dbg(3, "MC:\n");
794
795	pci_rc = pci_register_driver(&ie31200_driver);
796	if (pci_rc < 0)
797		return pci_rc;
798
799	if (!mci_pdev) {
800		ie31200_registered = 0;
801		for (i = 0; ie31200_pci_tbl[i].vendor != 0; i++) {
802			mci_pdev = pci_get_device(ie31200_pci_tbl[i].vendor,
803						  ie31200_pci_tbl[i].device,
804						  NULL);
805			if (mci_pdev)
806				break;
807		}
808
809		if (!mci_pdev) {
810			edac_dbg(0, "ie31200 pci_get_device fail\n");
811			pci_rc = -ENODEV;
812			goto fail0;
813		}
814
815		pci_rc = ie31200_init_one(mci_pdev, &ie31200_pci_tbl[i]);
816		if (pci_rc < 0) {
817			edac_dbg(0, "ie31200 init fail\n");
818			pci_rc = -ENODEV;
819			goto fail1;
820		}
821	}
822
823	return 0;
824fail1:
825	pci_dev_put(mci_pdev);
826fail0:
827	pci_unregister_driver(&ie31200_driver);
828
829	return pci_rc;
830}
831
832static void __exit ie31200_exit(void)
833{
834	edac_dbg(3, "MC:\n");
835	pci_unregister_driver(&ie31200_driver);
836	if (!ie31200_registered)
837		ie31200_remove_one(mci_pdev);
838}
839
840module_init(ie31200_init);
841module_exit(ie31200_exit);
842
843MODULE_LICENSE("GPL");
844MODULE_AUTHOR("Jason Baron <jbaron@akamai.com>");
845MODULE_DESCRIPTION("MC support for Intel Processor E31200 memory hub controllers");
~/ Read-only file preview.