1
2
3
4
5
6
7
8#include <linux/kernel.h>
9#include <asm/cpu_device_id.h>
10#include <asm/intel-family.h>
11#include <asm/mce.h>
12#include "edac_module.h"
13#include "skx_common.h"
14
15#define I10NM_REVISION "v0.0.3"
16#define EDAC_MOD_STR "i10nm_edac"
17
18
19#define i10nm_printk(level, fmt, arg...) \
20 edac_printk(level, "i10nm", fmt, ##arg)
21
22#define I10NM_GET_SCK_BAR(d, reg) \
23 pci_read_config_dword((d)->uracu, 0xd0, &(reg))
24#define I10NM_GET_IMC_BAR(d, i, reg) \
25 pci_read_config_dword((d)->uracu, 0xd8 + (i) * 4, &(reg))
26#define I10NM_GET_DIMMMTR(m, i, j) \
27 (*(u32 *)((m)->mbase + 0x2080c + (i) * 0x4000 + (j) * 4))
28#define I10NM_GET_MCDDRTCFG(m, i, j) \
29 (*(u32 *)((m)->mbase + 0x20970 + (i) * 0x4000 + (j) * 4))
30
31#define I10NM_GET_SCK_MMIO_BASE(reg) (GET_BITFIELD(reg, 0, 28) << 23)
32#define I10NM_GET_IMC_MMIO_OFFSET(reg) (GET_BITFIELD(reg, 0, 10) << 12)
33#define I10NM_GET_IMC_MMIO_SIZE(reg) ((GET_BITFIELD(reg, 13, 23) - \
34 GET_BITFIELD(reg, 0, 10) + 1) << 12)
35
36static struct list_head *i10nm_edac_list;
37
38static struct pci_dev *pci_get_dev_wrapper(int dom, unsigned int bus,
39 unsigned int dev, unsigned int fun)
40{
41 struct pci_dev *pdev;
42
43 pdev = pci_get_domain_bus_and_slot(dom, bus, PCI_DEVFN(dev, fun));
44 if (!pdev) {
45 edac_dbg(2, "No device %02x:%02x.%x\n",
46 bus, dev, fun);
47 return NULL;
48 }
49
50 if (unlikely(pci_enable_device(pdev) < 0)) {
51 edac_dbg(2, "Failed to enable device %02x:%02x.%x\n",
52 bus, dev, fun);
53 return NULL;
54 }
55
56 pci_dev_get(pdev);
57
58 return pdev;
59}
60
61static int i10nm_get_all_munits(void)
62{
63 struct pci_dev *mdev;
64 void __iomem *mbase;
65 unsigned long size;
66 struct skx_dev *d;
67 int i, j = 0;
68 u32 reg, off;
69 u64 base;
70
71 list_for_each_entry(d, i10nm_edac_list, list) {
72 d->util_all = pci_get_dev_wrapper(d->seg, d->bus[1], 29, 1);
73 if (!d->util_all)
74 return -ENODEV;
75
76 d->uracu = pci_get_dev_wrapper(d->seg, d->bus[0], 0, 1);
77 if (!d->uracu)
78 return -ENODEV;
79
80 if (I10NM_GET_SCK_BAR(d, reg)) {
81 i10nm_printk(KERN_ERR, "Failed to socket bar\n");
82 return -ENODEV;
83 }
84
85 base = I10NM_GET_SCK_MMIO_BASE(reg);
86 edac_dbg(2, "socket%d mmio base 0x%llx (reg 0x%x)\n",
87 j++, base, reg);
88
89 for (i = 0; i < I10NM_NUM_IMC; i++) {
90 mdev = pci_get_dev_wrapper(d->seg, d->bus[0],
91 12 + i, 0);
92 if (i == 0 && !mdev) {
93 i10nm_printk(KERN_ERR, "No IMC found\n");
94 return -ENODEV;
95 }
96 if (!mdev)
97 continue;
98
99 d->imc[i].mdev = mdev;
100
101 if (I10NM_GET_IMC_BAR(d, i, reg)) {
102 i10nm_printk(KERN_ERR, "Failed to get mc bar\n");
103 return -ENODEV;
104 }
105
106 off = I10NM_GET_IMC_MMIO_OFFSET(reg);
107 size = I10NM_GET_IMC_MMIO_SIZE(reg);
108 edac_dbg(2, "mc%d mmio base 0x%llx size 0x%lx (reg 0x%x)\n",
109 i, base + off, size, reg);
110
111 mbase = ioremap(base + off, size);
112 if (!mbase) {
113 i10nm_printk(KERN_ERR, "Failed to ioremap 0x%llx\n",
114 base + off);
115 return -ENODEV;
116 }
117
118 d->imc[i].mbase = mbase;
119 }
120 }
121
122 return 0;
123}
124
125static const struct x86_cpu_id i10nm_cpuids[] = {
126 { X86_VENDOR_INTEL, 6, INTEL_FAM6_ATOM_TREMONT_D, 0, 0 },
127 { X86_VENDOR_INTEL, 6, INTEL_FAM6_ICELAKE_X, 0, 0 },
128 { X86_VENDOR_INTEL, 6, INTEL_FAM6_ICELAKE_D, 0, 0 },
129 { }
130};
131MODULE_DEVICE_TABLE(x86cpu, i10nm_cpuids);
132
133static bool i10nm_check_ecc(struct skx_imc *imc, int chan)
134{
135 u32 mcmtr;
136
137 mcmtr = *(u32 *)(imc->mbase + 0x20ef8 + chan * 0x4000);
138 edac_dbg(1, "ch%d mcmtr reg %x\n", chan, mcmtr);
139
140 return !!GET_BITFIELD(mcmtr, 2, 2);
141}
142
143static int i10nm_get_dimm_config(struct mem_ctl_info *mci)
144{
145 struct skx_pvt *pvt = mci->pvt_info;
146 struct skx_imc *imc = pvt->imc;
147 struct dimm_info *dimm;
148 u32 mtr, mcddrtcfg;
149 int i, j, ndimms;
150
151 for (i = 0; i < I10NM_NUM_CHANNELS; i++) {
152 if (!imc->mbase)
153 continue;
154
155 ndimms = 0;
156 for (j = 0; j < I10NM_NUM_DIMMS; j++) {
157 dimm = EDAC_DIMM_PTR(mci->layers, mci->dimms,
158 mci->n_layers, i, j, 0);
159 mtr = I10NM_GET_DIMMMTR(imc, i, j);
160 mcddrtcfg = I10NM_GET_MCDDRTCFG(imc, i, j);
161 edac_dbg(1, "dimmmtr 0x%x mcddrtcfg 0x%x (mc%d ch%d dimm%d)\n",
162 mtr, mcddrtcfg, imc->mc, i, j);
163
164 if (IS_DIMM_PRESENT(mtr))
165 ndimms += skx_get_dimm_info(mtr, 0, dimm,
166 imc, i, j);
167 else if (IS_NVDIMM_PRESENT(mcddrtcfg, j))
168 ndimms += skx_get_nvdimm_info(dimm, imc, i, j,
169 EDAC_MOD_STR);
170 }
171 if (ndimms && !i10nm_check_ecc(imc, i)) {
172 i10nm_printk(KERN_ERR, "ECC is disabled on imc %d channel %d\n",
173 imc->mc, i);
174 return -ENODEV;
175 }
176 }
177
178 return 0;
179}
180
181static struct notifier_block i10nm_mce_dec = {
182 .notifier_call = skx_mce_check_error,
183 .priority = MCE_PRIO_EDAC,
184};
185
186#ifdef CONFIG_EDAC_DEBUG
187
188
189
190
191
192static struct dentry *i10nm_test;
193
194static int debugfs_u64_set(void *data, u64 val)
195{
196 struct mce m;
197
198 pr_warn_once("Fake error to 0x%llx injected via debugfs\n", val);
199
200 memset(&m, 0, sizeof(m));
201
202 m.status = MCI_STATUS_ADDRV + 0x90;
203
204 m.status |= BIT_ULL(MCI_STATUS_CEC_SHIFT);
205 m.addr = val;
206 skx_mce_check_error(NULL, 0, &m);
207
208 return 0;
209}
210DEFINE_SIMPLE_ATTRIBUTE(fops_u64_wo, NULL, debugfs_u64_set, "%llu\n");
211
212static void setup_i10nm_debug(void)
213{
214 i10nm_test = edac_debugfs_create_dir("i10nm_test");
215 if (!i10nm_test)
216 return;
217
218 if (!edac_debugfs_create_file("addr", 0200, i10nm_test,
219 NULL, &fops_u64_wo)) {
220 debugfs_remove(i10nm_test);
221 i10nm_test = NULL;
222 }
223}
224
225static void teardown_i10nm_debug(void)
226{
227 debugfs_remove_recursive(i10nm_test);
228}
229#else
230static inline void setup_i10nm_debug(void) {}
231static inline void teardown_i10nm_debug(void) {}
232#endif
233
234static int __init i10nm_init(void)
235{
236 u8 mc = 0, src_id = 0, node_id = 0;
237 const struct x86_cpu_id *id;
238 const char *owner;
239 struct skx_dev *d;
240 int rc, i, off[3] = {0xd0, 0xc8, 0xcc};
241 u64 tolm, tohm;
242
243 edac_dbg(2, "\n");
244
245 owner = edac_get_owner();
246 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
247 return -EBUSY;
248
249 id = x86_match_cpu(i10nm_cpuids);
250 if (!id)
251 return -ENODEV;
252
253 rc = skx_get_hi_lo(0x09a2, off, &tolm, &tohm);
254 if (rc)
255 return rc;
256
257 rc = skx_get_all_bus_mappings(0x3452, 0xcc, I10NM, &i10nm_edac_list);
258 if (rc < 0)
259 goto fail;
260 if (rc == 0) {
261 i10nm_printk(KERN_ERR, "No memory controllers found\n");
262 return -ENODEV;
263 }
264
265 rc = i10nm_get_all_munits();
266 if (rc < 0)
267 goto fail;
268
269 list_for_each_entry(d, i10nm_edac_list, list) {
270 rc = skx_get_src_id(d, 0xf8, &src_id);
271 if (rc < 0)
272 goto fail;
273
274 rc = skx_get_node_id(d, &node_id);
275 if (rc < 0)
276 goto fail;
277
278 edac_dbg(2, "src_id = %d node_id = %d\n", src_id, node_id);
279 for (i = 0; i < I10NM_NUM_IMC; i++) {
280 if (!d->imc[i].mdev)
281 continue;
282
283 d->imc[i].mc = mc++;
284 d->imc[i].lmc = i;
285 d->imc[i].src_id = src_id;
286 d->imc[i].node_id = node_id;
287
288 rc = skx_register_mci(&d->imc[i], d->imc[i].mdev,
289 "Intel_10nm Socket", EDAC_MOD_STR,
290 i10nm_get_dimm_config);
291 if (rc < 0)
292 goto fail;
293 }
294 }
295
296 rc = skx_adxl_get();
297 if (rc)
298 goto fail;
299
300 opstate_init();
301 mce_register_decode_chain(&i10nm_mce_dec);
302 setup_i10nm_debug();
303
304 i10nm_printk(KERN_INFO, "%s\n", I10NM_REVISION);
305
306 return 0;
307fail:
308 skx_remove();
309 return rc;
310}
311
312static void __exit i10nm_exit(void)
313{
314 edac_dbg(2, "\n");
315 teardown_i10nm_debug();
316 mce_unregister_decode_chain(&i10nm_mce_dec);
317 skx_adxl_put();
318 skx_remove();
319}
320
321module_init(i10nm_init);
322module_exit(i10nm_exit);
323
324MODULE_LICENSE("GPL v2");
325MODULE_DESCRIPTION("MC Driver for Intel 10nm server processors");
326
