xref: /openbmc/linux/drivers/edac/i10nm_base.c (revision b4e18b29)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Driver for Intel(R) 10nm server memory controller.
4  * Copyright (c) 2019, Intel Corporation.
5  *
6  */
7 
8 #include <linux/kernel.h>
9 #include <linux/io.h>
10 #include <asm/cpu_device_id.h>
11 #include <asm/intel-family.h>
12 #include <asm/mce.h>
13 #include "edac_module.h"
14 #include "skx_common.h"
15 
16 #define I10NM_REVISION	"v0.0.4"
17 #define EDAC_MOD_STR	"i10nm_edac"
18 
19 /* Debug macros */
20 #define i10nm_printk(level, fmt, arg...)	\
21 	edac_printk(level, "i10nm", fmt, ##arg)
22 
23 #define I10NM_GET_SCK_BAR(d, reg)	\
24 	pci_read_config_dword((d)->uracu, 0xd0, &(reg))
25 #define I10NM_GET_IMC_BAR(d, i, reg)	\
26 	pci_read_config_dword((d)->uracu, 0xd8 + (i) * 4, &(reg))
27 #define I10NM_GET_DIMMMTR(m, i, j)	\
28 	readl((m)->mbase + 0x2080c + (i) * (m)->chan_mmio_sz + (j) * 4)
29 #define I10NM_GET_MCDDRTCFG(m, i, j)	\
30 	readl((m)->mbase + 0x20970 + (i) * (m)->chan_mmio_sz + (j) * 4)
31 #define I10NM_GET_MCMTR(m, i)		\
32 	readl((m)->mbase + 0x20ef8 + (i) * (m)->chan_mmio_sz)
33 #define I10NM_GET_AMAP(m, i)		\
34 	readl((m)->mbase + 0x20814 + (i) * (m)->chan_mmio_sz)
35 
36 #define I10NM_GET_SCK_MMIO_BASE(reg)	(GET_BITFIELD(reg, 0, 28) << 23)
37 #define I10NM_GET_IMC_MMIO_OFFSET(reg)	(GET_BITFIELD(reg, 0, 10) << 12)
38 #define I10NM_GET_IMC_MMIO_SIZE(reg)	((GET_BITFIELD(reg, 13, 23) - \
39 					 GET_BITFIELD(reg, 0, 10) + 1) << 12)
40 
41 static struct list_head *i10nm_edac_list;
42 
43 static struct pci_dev *pci_get_dev_wrapper(int dom, unsigned int bus,
44 					   unsigned int dev, unsigned int fun)
45 {
46 	struct pci_dev *pdev;
47 
48 	pdev = pci_get_domain_bus_and_slot(dom, bus, PCI_DEVFN(dev, fun));
49 	if (!pdev) {
50 		edac_dbg(2, "No device %02x:%02x.%x\n",
51 			 bus, dev, fun);
52 		return NULL;
53 	}
54 
55 	if (unlikely(pci_enable_device(pdev) < 0)) {
56 		edac_dbg(2, "Failed to enable device %02x:%02x.%x\n",
57 			 bus, dev, fun);
58 		return NULL;
59 	}
60 
61 	pci_dev_get(pdev);
62 
63 	return pdev;
64 }
65 
66 static int i10nm_get_all_munits(void)
67 {
68 	struct pci_dev *mdev;
69 	void __iomem *mbase;
70 	unsigned long size;
71 	struct skx_dev *d;
72 	int i, j = 0;
73 	u32 reg, off;
74 	u64 base;
75 
76 	list_for_each_entry(d, i10nm_edac_list, list) {
77 		d->util_all = pci_get_dev_wrapper(d->seg, d->bus[1], 29, 1);
78 		if (!d->util_all)
79 			return -ENODEV;
80 
81 		d->uracu = pci_get_dev_wrapper(d->seg, d->bus[0], 0, 1);
82 		if (!d->uracu)
83 			return -ENODEV;
84 
85 		if (I10NM_GET_SCK_BAR(d, reg)) {
86 			i10nm_printk(KERN_ERR, "Failed to socket bar\n");
87 			return -ENODEV;
88 		}
89 
90 		base = I10NM_GET_SCK_MMIO_BASE(reg);
91 		edac_dbg(2, "socket%d mmio base 0x%llx (reg 0x%x)\n",
92 			 j++, base, reg);
93 
94 		for (i = 0; i < I10NM_NUM_IMC; i++) {
95 			mdev = pci_get_dev_wrapper(d->seg, d->bus[0],
96 						   12 + i, 0);
97 			if (i == 0 && !mdev) {
98 				i10nm_printk(KERN_ERR, "No IMC found\n");
99 				return -ENODEV;
100 			}
101 			if (!mdev)
102 				continue;
103 
104 			d->imc[i].mdev = mdev;
105 
106 			if (I10NM_GET_IMC_BAR(d, i, reg)) {
107 				i10nm_printk(KERN_ERR, "Failed to get mc bar\n");
108 				return -ENODEV;
109 			}
110 
111 			off  = I10NM_GET_IMC_MMIO_OFFSET(reg);
112 			size = I10NM_GET_IMC_MMIO_SIZE(reg);
113 			edac_dbg(2, "mc%d mmio base 0x%llx size 0x%lx (reg 0x%x)\n",
114 				 i, base + off, size, reg);
115 
116 			mbase = ioremap(base + off, size);
117 			if (!mbase) {
118 				i10nm_printk(KERN_ERR, "Failed to ioremap 0x%llx\n",
119 					     base + off);
120 				return -ENODEV;
121 			}
122 
123 			d->imc[i].mbase = mbase;
124 		}
125 	}
126 
127 	return 0;
128 }
129 
130 static struct res_config i10nm_cfg0 = {
131 	.type			= I10NM,
132 	.decs_did		= 0x3452,
133 	.busno_cfg_offset	= 0xcc,
134 	.ddr_chan_mmio_sz	= 0x4000,
135 };
136 
137 static struct res_config i10nm_cfg1 = {
138 	.type			= I10NM,
139 	.decs_did		= 0x3452,
140 	.busno_cfg_offset	= 0xd0,
141 	.ddr_chan_mmio_sz	= 0x4000,
142 };
143 
144 static struct res_config spr_cfg = {
145 	.type			= SPR,
146 	.decs_did		= 0x3252,
147 	.busno_cfg_offset	= 0xd0,
148 	.ddr_chan_mmio_sz	= 0x8000,
149 	.support_ddr5		= true,
150 };
151 
152 static const struct x86_cpu_id i10nm_cpuids[] = {
153 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ATOM_TREMONT_D,	X86_STEPPINGS(0x0, 0x3), &i10nm_cfg0),
154 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ATOM_TREMONT_D,	X86_STEPPINGS(0x4, 0xf), &i10nm_cfg1),
155 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ICELAKE_X,		X86_STEPPINGS(0x0, 0x3), &i10nm_cfg0),
156 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ICELAKE_X,		X86_STEPPINGS(0x4, 0xf), &i10nm_cfg1),
157 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ICELAKE_D,		X86_STEPPINGS(0x0, 0xf), &i10nm_cfg1),
158 	X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SAPPHIRERAPIDS_X,	X86_STEPPINGS(0x0, 0xf), &spr_cfg),
159 	{}
160 };
161 MODULE_DEVICE_TABLE(x86cpu, i10nm_cpuids);
162 
163 static bool i10nm_check_ecc(struct skx_imc *imc, int chan)
164 {
165 	u32 mcmtr;
166 
167 	mcmtr = I10NM_GET_MCMTR(imc, chan);
168 	edac_dbg(1, "ch%d mcmtr reg %x\n", chan, mcmtr);
169 
170 	return !!GET_BITFIELD(mcmtr, 2, 2);
171 }
172 
173 static int i10nm_get_dimm_config(struct mem_ctl_info *mci,
174 				 struct res_config *cfg)
175 {
176 	struct skx_pvt *pvt = mci->pvt_info;
177 	struct skx_imc *imc = pvt->imc;
178 	u32 mtr, amap, mcddrtcfg;
179 	struct dimm_info *dimm;
180 	int i, j, ndimms;
181 
182 	for (i = 0; i < I10NM_NUM_CHANNELS; i++) {
183 		if (!imc->mbase)
184 			continue;
185 
186 		ndimms = 0;
187 		amap = I10NM_GET_AMAP(imc, i);
188 		for (j = 0; j < I10NM_NUM_DIMMS; j++) {
189 			dimm = edac_get_dimm(mci, i, j, 0);
190 			mtr = I10NM_GET_DIMMMTR(imc, i, j);
191 			mcddrtcfg = I10NM_GET_MCDDRTCFG(imc, i, j);
192 			edac_dbg(1, "dimmmtr 0x%x mcddrtcfg 0x%x (mc%d ch%d dimm%d)\n",
193 				 mtr, mcddrtcfg, imc->mc, i, j);
194 
195 			if (IS_DIMM_PRESENT(mtr))
196 				ndimms += skx_get_dimm_info(mtr, 0, amap, dimm,
197 							    imc, i, j, cfg);
198 			else if (IS_NVDIMM_PRESENT(mcddrtcfg, j))
199 				ndimms += skx_get_nvdimm_info(dimm, imc, i, j,
200 							      EDAC_MOD_STR);
201 		}
202 		if (ndimms && !i10nm_check_ecc(imc, i)) {
203 			i10nm_printk(KERN_ERR, "ECC is disabled on imc %d channel %d\n",
204 				     imc->mc, i);
205 			return -ENODEV;
206 		}
207 	}
208 
209 	return 0;
210 }
211 
212 static struct notifier_block i10nm_mce_dec = {
213 	.notifier_call	= skx_mce_check_error,
214 	.priority	= MCE_PRIO_EDAC,
215 };
216 
217 #ifdef CONFIG_EDAC_DEBUG
218 /*
219  * Debug feature.
220  * Exercise the address decode logic by writing an address to
221  * /sys/kernel/debug/edac/i10nm_test/addr.
222  */
223 static struct dentry *i10nm_test;
224 
225 static int debugfs_u64_set(void *data, u64 val)
226 {
227 	struct mce m;
228 
229 	pr_warn_once("Fake error to 0x%llx injected via debugfs\n", val);
230 
231 	memset(&m, 0, sizeof(m));
232 	/* ADDRV + MemRd + Unknown channel */
233 	m.status = MCI_STATUS_ADDRV + 0x90;
234 	/* One corrected error */
235 	m.status |= BIT_ULL(MCI_STATUS_CEC_SHIFT);
236 	m.addr = val;
237 	skx_mce_check_error(NULL, 0, &m);
238 
239 	return 0;
240 }
241 DEFINE_SIMPLE_ATTRIBUTE(fops_u64_wo, NULL, debugfs_u64_set, "%llu\n");
242 
243 static void setup_i10nm_debug(void)
244 {
245 	i10nm_test = edac_debugfs_create_dir("i10nm_test");
246 	if (!i10nm_test)
247 		return;
248 
249 	if (!edac_debugfs_create_file("addr", 0200, i10nm_test,
250 				      NULL, &fops_u64_wo)) {
251 		debugfs_remove(i10nm_test);
252 		i10nm_test = NULL;
253 	}
254 }
255 
256 static void teardown_i10nm_debug(void)
257 {
258 	debugfs_remove_recursive(i10nm_test);
259 }
260 #else
261 static inline void setup_i10nm_debug(void) {}
262 static inline void teardown_i10nm_debug(void) {}
263 #endif /*CONFIG_EDAC_DEBUG*/
264 
265 static int __init i10nm_init(void)
266 {
267 	u8 mc = 0, src_id = 0, node_id = 0;
268 	const struct x86_cpu_id *id;
269 	struct res_config *cfg;
270 	const char *owner;
271 	struct skx_dev *d;
272 	int rc, i, off[3] = {0xd0, 0xc8, 0xcc};
273 	u64 tolm, tohm;
274 
275 	edac_dbg(2, "\n");
276 
277 	owner = edac_get_owner();
278 	if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
279 		return -EBUSY;
280 
281 	id = x86_match_cpu(i10nm_cpuids);
282 	if (!id)
283 		return -ENODEV;
284 
285 	cfg = (struct res_config *)id->driver_data;
286 
287 	rc = skx_get_hi_lo(0x09a2, off, &tolm, &tohm);
288 	if (rc)
289 		return rc;
290 
291 	rc = skx_get_all_bus_mappings(cfg, &i10nm_edac_list);
292 	if (rc < 0)
293 		goto fail;
294 	if (rc == 0) {
295 		i10nm_printk(KERN_ERR, "No memory controllers found\n");
296 		return -ENODEV;
297 	}
298 
299 	rc = i10nm_get_all_munits();
300 	if (rc < 0)
301 		goto fail;
302 
303 	list_for_each_entry(d, i10nm_edac_list, list) {
304 		rc = skx_get_src_id(d, 0xf8, &src_id);
305 		if (rc < 0)
306 			goto fail;
307 
308 		rc = skx_get_node_id(d, &node_id);
309 		if (rc < 0)
310 			goto fail;
311 
312 		edac_dbg(2, "src_id = %d node_id = %d\n", src_id, node_id);
313 		for (i = 0; i < I10NM_NUM_IMC; i++) {
314 			if (!d->imc[i].mdev)
315 				continue;
316 
317 			d->imc[i].mc  = mc++;
318 			d->imc[i].lmc = i;
319 			d->imc[i].src_id  = src_id;
320 			d->imc[i].node_id = node_id;
321 			d->imc[i].chan_mmio_sz = cfg->ddr_chan_mmio_sz;
322 
323 			rc = skx_register_mci(&d->imc[i], d->imc[i].mdev,
324 					      "Intel_10nm Socket", EDAC_MOD_STR,
325 					      i10nm_get_dimm_config, cfg);
326 			if (rc < 0)
327 				goto fail;
328 		}
329 	}
330 
331 	rc = skx_adxl_get();
332 	if (rc)
333 		goto fail;
334 
335 	opstate_init();
336 	mce_register_decode_chain(&i10nm_mce_dec);
337 	setup_i10nm_debug();
338 
339 	i10nm_printk(KERN_INFO, "%s\n", I10NM_REVISION);
340 
341 	return 0;
342 fail:
343 	skx_remove();
344 	return rc;
345 }
346 
347 static void __exit i10nm_exit(void)
348 {
349 	edac_dbg(2, "\n");
350 	teardown_i10nm_debug();
351 	mce_unregister_decode_chain(&i10nm_mce_dec);
352 	skx_adxl_put();
353 	skx_remove();
354 }
355 
356 module_init(i10nm_init);
357 module_exit(i10nm_exit);
358 
359 MODULE_LICENSE("GPL v2");
360 MODULE_DESCRIPTION("MC Driver for Intel 10nm server processors");
361