1
2
3
4
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 "v0.0.5"
17#define EDAC_MOD_STR "i10nm_edac"
18
19
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_SAD(d, offset, i, reg)\
28 pci_read_config_dword((d)->sad_all, (offset) + (i) * 8, &(reg))
29#define I10NM_GET_HBM_IMC_BAR(d, reg) \
30 pci_read_config_dword((d)->uracu, 0xd4, &(reg))
31#define I10NM_GET_CAPID3_CFG(d, reg) \
32 pci_read_config_dword((d)->pcu_cr3, 0x90, &(reg))
33#define I10NM_GET_DIMMMTR(m, i, j) \
34 readl((m)->mbase + ((m)->hbm_mc ? 0x80c : 0x2080c) + \
35 (i) * (m)->chan_mmio_sz + (j) * 4)
36#define I10NM_GET_MCDDRTCFG(m, i, j) \
37 readl((m)->mbase + ((m)->hbm_mc ? 0x970 : 0x20970) + \
38 (i) * (m)->chan_mmio_sz + (j) * 4)
39#define I10NM_GET_MCMTR(m, i) \
40 readl((m)->mbase + ((m)->hbm_mc ? 0xef8 : 0x20ef8) + \
41 (i) * (m)->chan_mmio_sz)
42#define I10NM_GET_AMAP(m, i) \
43 readl((m)->mbase + ((m)->hbm_mc ? 0x814 : 0x20814) + \
44 (i) * (m)->chan_mmio_sz)
45
46#define I10NM_GET_SCK_MMIO_BASE(reg) (GET_BITFIELD(reg, 0, 28) << 23)
47#define I10NM_GET_IMC_MMIO_OFFSET(reg) (GET_BITFIELD(reg, 0, 10) << 12)
48#define I10NM_GET_IMC_MMIO_SIZE(reg) ((GET_BITFIELD(reg, 13, 23) - \
49 GET_BITFIELD(reg, 0, 10) + 1) << 12)
50#define I10NM_GET_HBM_IMC_MMIO_OFFSET(reg) \
51 ((GET_BITFIELD(reg, 0, 10) << 12) + 0x140000)
52
53#define I10NM_HBM_IMC_MMIO_SIZE 0x9000
54#define I10NM_IS_HBM_PRESENT(reg) GET_BITFIELD(reg, 27, 30)
55#define I10NM_IS_HBM_IMC(reg) GET_BITFIELD(reg, 29, 29)
56
57#define I10NM_MAX_SAD 16
58#define I10NM_SAD_ENABLE(reg) GET_BITFIELD(reg, 0, 0)
59#define I10NM_SAD_NM_CACHEABLE(reg) GET_BITFIELD(reg, 5, 5)
60
61static struct list_head *i10nm_edac_list;
62
63static struct pci_dev *pci_get_dev_wrapper(int dom, unsigned int bus,
64 unsigned int dev, unsigned int fun)
65{
66 struct pci_dev *pdev;
67
68 pdev = pci_get_domain_bus_and_slot(dom, bus, PCI_DEVFN(dev, fun));
69 if (!pdev) {
70 edac_dbg(2, "No device %02x:%02x.%x\n",
71 bus, dev, fun);
72 return NULL;
73 }
74
75 if (unlikely(pci_enable_device(pdev) < 0)) {
76 edac_dbg(2, "Failed to enable device %02x:%02x.%x\n",
77 bus, dev, fun);
78 return NULL;
79 }
80
81 pci_dev_get(pdev);
82
83 return pdev;
84}
85
86static bool i10nm_check_2lm(struct res_config *cfg)
87{
88 struct skx_dev *d;
89 u32 reg;
90 int i;
91
92 list_for_each_entry(d, i10nm_edac_list, list) {
93 d->sad_all = pci_get_dev_wrapper(d->seg, d->bus[1],
94 PCI_SLOT(cfg->sad_all_devfn),
95 PCI_FUNC(cfg->sad_all_devfn));
96 if (!d->sad_all)
97 continue;
98
99 for (i = 0; i < I10NM_MAX_SAD; i++) {
100 I10NM_GET_SAD(d, cfg->sad_all_offset, i, reg);
101 if (I10NM_SAD_ENABLE(reg) && I10NM_SAD_NM_CACHEABLE(reg)) {
102 edac_dbg(2, "2-level memory configuration.\n");
103 return true;
104 }
105 }
106 }
107
108 return false;
109}
110
111static int i10nm_get_ddr_munits(void)
112{
113 struct pci_dev *mdev;
114 void __iomem *mbase;
115 unsigned long size;
116 struct skx_dev *d;
117 int i, j = 0;
118 u32 reg, off;
119 u64 base;
120
121 list_for_each_entry(d, i10nm_edac_list, list) {
122 d->util_all = pci_get_dev_wrapper(d->seg, d->bus[1], 29, 1);
123 if (!d->util_all)
124 return -ENODEV;
125
126 d->uracu = pci_get_dev_wrapper(d->seg, d->bus[0], 0, 1);
127 if (!d->uracu)
128 return -ENODEV;
129
130 if (I10NM_GET_SCK_BAR(d, reg)) {
131 i10nm_printk(KERN_ERR, "Failed to socket bar\n");
132 return -ENODEV;
133 }
134
135 base = I10NM_GET_SCK_MMIO_BASE(reg);
136 edac_dbg(2, "socket%d mmio base 0x%llx (reg 0x%x)\n",
137 j++, base, reg);
138
139 for (i = 0; i < I10NM_NUM_DDR_IMC; i++) {
140 mdev = pci_get_dev_wrapper(d->seg, d->bus[0],
141 12 + i, 0);
142 if (i == 0 && !mdev) {
143 i10nm_printk(KERN_ERR, "No IMC found\n");
144 return -ENODEV;
145 }
146 if (!mdev)
147 continue;
148
149 d->imc[i].mdev = mdev;
150
151 if (I10NM_GET_IMC_BAR(d, i, reg)) {
152 i10nm_printk(KERN_ERR, "Failed to get mc bar\n");
153 return -ENODEV;
154 }
155
156 off = I10NM_GET_IMC_MMIO_OFFSET(reg);
157 size = I10NM_GET_IMC_MMIO_SIZE(reg);
158 edac_dbg(2, "mc%d mmio base 0x%llx size 0x%lx (reg 0x%x)\n",
159 i, base + off, size, reg);
160
161 mbase = ioremap(base + off, size);
162 if (!mbase) {
163 i10nm_printk(KERN_ERR, "Failed to ioremap 0x%llx\n",
164 base + off);
165 return -ENODEV;
166 }
167
168 d->imc[i].mbase = mbase;
169 }
170 }
171
172 return 0;
173}
174
175static bool i10nm_check_hbm_imc(struct skx_dev *d)
176{
177 u32 reg;
178
179 if (I10NM_GET_CAPID3_CFG(d, reg)) {
180 i10nm_printk(KERN_ERR, "Failed to get capid3_cfg\n");
181 return false;
182 }
183
184 return I10NM_IS_HBM_PRESENT(reg) != 0;
185}
186
187static int i10nm_get_hbm_munits(void)
188{
189 struct pci_dev *mdev;
190 void __iomem *mbase;
191 u32 reg, off, mcmtr;
192 struct skx_dev *d;
193 int i, lmc;
194 u64 base;
195
196 list_for_each_entry(d, i10nm_edac_list, list) {
197 d->pcu_cr3 = pci_get_dev_wrapper(d->seg, d->bus[1], 30, 3);
198 if (!d->pcu_cr3)
199 return -ENODEV;
200
201 if (!i10nm_check_hbm_imc(d)) {
202 i10nm_printk(KERN_DEBUG, "No hbm memory\n");
203 return -ENODEV;
204 }
205
206 if (I10NM_GET_SCK_BAR(d, reg)) {
207 i10nm_printk(KERN_ERR, "Failed to get socket bar\n");
208 return -ENODEV;
209 }
210 base = I10NM_GET_SCK_MMIO_BASE(reg);
211
212 if (I10NM_GET_HBM_IMC_BAR(d, reg)) {
213 i10nm_printk(KERN_ERR, "Failed to get hbm mc bar\n");
214 return -ENODEV;
215 }
216 base += I10NM_GET_HBM_IMC_MMIO_OFFSET(reg);
217
218 lmc = I10NM_NUM_DDR_IMC;
219
220 for (i = 0; i < I10NM_NUM_HBM_IMC; i++) {
221 mdev = pci_get_dev_wrapper(d->seg, d->bus[0],
222 12 + i / 4, 1 + i % 4);
223 if (i == 0 && !mdev) {
224 i10nm_printk(KERN_ERR, "No hbm mc found\n");
225 return -ENODEV;
226 }
227 if (!mdev)
228 continue;
229
230 d->imc[lmc].mdev = mdev;
231 off = i * I10NM_HBM_IMC_MMIO_SIZE;
232
233 edac_dbg(2, "hbm mc%d mmio base 0x%llx size 0x%x\n",
234 lmc, base + off, I10NM_HBM_IMC_MMIO_SIZE);
235
236 mbase = ioremap(base + off, I10NM_HBM_IMC_MMIO_SIZE);
237 if (!mbase) {
238 i10nm_printk(KERN_ERR, "Failed to ioremap for hbm mc 0x%llx\n",
239 base + off);
240 return -ENOMEM;
241 }
242
243 d->imc[lmc].mbase = mbase;
244 d->imc[lmc].hbm_mc = true;
245
246 mcmtr = I10NM_GET_MCMTR(&d->imc[lmc], 0);
247 if (!I10NM_IS_HBM_IMC(mcmtr)) {
248 i10nm_printk(KERN_ERR, "This isn't an hbm mc!\n");
249 return -ENODEV;
250 }
251
252 lmc++;
253 }
254 }
255
256 return 0;
257}
258
259static struct res_config i10nm_cfg0 = {
260 .type = I10NM,
261 .decs_did = 0x3452,
262 .busno_cfg_offset = 0xcc,
263 .ddr_chan_mmio_sz = 0x4000,
264 .sad_all_devfn = PCI_DEVFN(29, 0),
265 .sad_all_offset = 0x108,
266};
267
268static struct res_config i10nm_cfg1 = {
269 .type = I10NM,
270 .decs_did = 0x3452,
271 .busno_cfg_offset = 0xd0,
272 .ddr_chan_mmio_sz = 0x4000,
273 .sad_all_devfn = PCI_DEVFN(29, 0),
274 .sad_all_offset = 0x108,
275};
276
277static struct res_config spr_cfg = {
278 .type = SPR,
279 .decs_did = 0x3252,
280 .busno_cfg_offset = 0xd0,
281 .ddr_chan_mmio_sz = 0x8000,
282 .hbm_chan_mmio_sz = 0x4000,
283 .support_ddr5 = true,
284 .sad_all_devfn = PCI_DEVFN(10, 0),
285 .sad_all_offset = 0x300,
286};
287
288static const struct x86_cpu_id i10nm_cpuids[] = {
289 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ATOM_TREMONT_D, X86_STEPPINGS(0x0, 0x3), &i10nm_cfg0),
290 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ATOM_TREMONT_D, X86_STEPPINGS(0x4, 0xf), &i10nm_cfg1),
291 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ICELAKE_X, X86_STEPPINGS(0x0, 0x3), &i10nm_cfg0),
292 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ICELAKE_X, X86_STEPPINGS(0x4, 0xf), &i10nm_cfg1),
293 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(ICELAKE_D, X86_STEPPINGS(0x0, 0xf), &i10nm_cfg1),
294 X86_MATCH_INTEL_FAM6_MODEL_STEPPINGS(SAPPHIRERAPIDS_X, X86_STEPPINGS(0x0, 0xf), &spr_cfg),
295 {}
296};
297MODULE_DEVICE_TABLE(x86cpu, i10nm_cpuids);
298
299static bool i10nm_check_ecc(struct skx_imc *imc, int chan)
300{
301 u32 mcmtr;
302
303 mcmtr = I10NM_GET_MCMTR(imc, chan);
304 edac_dbg(1, "ch%d mcmtr reg %x\n", chan, mcmtr);
305
306 return !!GET_BITFIELD(mcmtr, 2, 2);
307}
308
309static int i10nm_get_dimm_config(struct mem_ctl_info *mci,
310 struct res_config *cfg)
311{
312 struct skx_pvt *pvt = mci->pvt_info;
313 struct skx_imc *imc = pvt->imc;
314 u32 mtr, amap, mcddrtcfg;
315 struct dimm_info *dimm;
316 int i, j, ndimms;
317
318 for (i = 0; i < imc->num_channels; i++) {
319 if (!imc->mbase)
320 continue;
321
322 ndimms = 0;
323 amap = I10NM_GET_AMAP(imc, i);
324 for (j = 0; j < imc->num_dimms; j++) {
325 dimm = edac_get_dimm(mci, i, j, 0);
326 mtr = I10NM_GET_DIMMMTR(imc, i, j);
327 mcddrtcfg = I10NM_GET_MCDDRTCFG(imc, i, j);
328 edac_dbg(1, "dimmmtr 0x%x mcddrtcfg 0x%x (mc%d ch%d dimm%d)\n",
329 mtr, mcddrtcfg, imc->mc, i, j);
330
331 if (IS_DIMM_PRESENT(mtr))
332 ndimms += skx_get_dimm_info(mtr, 0, amap, dimm,
333 imc, i, j, cfg);
334 else if (IS_NVDIMM_PRESENT(mcddrtcfg, j))
335 ndimms += skx_get_nvdimm_info(dimm, imc, i, j,
336 EDAC_MOD_STR);
337 }
338 if (ndimms && !i10nm_check_ecc(imc, i)) {
339 i10nm_printk(KERN_ERR, "ECC is disabled on imc %d channel %d\n",
340 imc->mc, i);
341 return -ENODEV;
342 }
343 }
344
345 return 0;
346}
347
348static struct notifier_block i10nm_mce_dec = {
349 .notifier_call = skx_mce_check_error,
350 .priority = MCE_PRIO_EDAC,
351};
352
353#ifdef CONFIG_EDAC_DEBUG
354
355
356
357
358
359static struct dentry *i10nm_test;
360
361static int debugfs_u64_set(void *data, u64 val)
362{
363 struct mce m;
364
365 pr_warn_once("Fake error to 0x%llx injected via debugfs\n", val);
366
367 memset(&m, 0, sizeof(m));
368
369 m.status = MCI_STATUS_ADDRV + 0x90;
370
371 m.status |= BIT_ULL(MCI_STATUS_CEC_SHIFT);
372 m.addr = val;
373 skx_mce_check_error(NULL, 0, &m);
374
375 return 0;
376}
377DEFINE_SIMPLE_ATTRIBUTE(fops_u64_wo, NULL, debugfs_u64_set, "%llu\n");
378
379static void setup_i10nm_debug(void)
380{
381 i10nm_test = edac_debugfs_create_dir("i10nm_test");
382 if (!i10nm_test)
383 return;
384
385 if (!edac_debugfs_create_file("addr", 0200, i10nm_test,
386 NULL, &fops_u64_wo)) {
387 debugfs_remove(i10nm_test);
388 i10nm_test = NULL;
389 }
390}
391
392static void teardown_i10nm_debug(void)
393{
394 debugfs_remove_recursive(i10nm_test);
395}
396#else
397static inline void setup_i10nm_debug(void) {}
398static inline void teardown_i10nm_debug(void) {}
399#endif
400
401static int __init i10nm_init(void)
402{
403 u8 mc = 0, src_id = 0, node_id = 0;
404 const struct x86_cpu_id *id;
405 struct res_config *cfg;
406 const char *owner;
407 struct skx_dev *d;
408 int rc, i, off[3] = {0xd0, 0xc8, 0xcc};
409 u64 tolm, tohm;
410
411 edac_dbg(2, "\n");
412
413 owner = edac_get_owner();
414 if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR)))
415 return -EBUSY;
416
417 if (cpu_feature_enabled(X86_FEATURE_HYPERVISOR))
418 return -ENODEV;
419
420 id = x86_match_cpu(i10nm_cpuids);
421 if (!id)
422 return -ENODEV;
423
424 cfg = (struct res_config *)id->driver_data;
425
426 rc = skx_get_hi_lo(0x09a2, off, &tolm, &tohm);
427 if (rc)
428 return rc;
429
430 rc = skx_get_all_bus_mappings(cfg, &i10nm_edac_list);
431 if (rc < 0)
432 goto fail;
433 if (rc == 0) {
434 i10nm_printk(KERN_ERR, "No memory controllers found\n");
435 return -ENODEV;
436 }
437
438 skx_set_mem_cfg(i10nm_check_2lm(cfg));
439
440 rc = i10nm_get_ddr_munits();
441
442 if (i10nm_get_hbm_munits() && rc)
443 goto fail;
444
445 list_for_each_entry(d, i10nm_edac_list, list) {
446 rc = skx_get_src_id(d, 0xf8, &src_id);
447 if (rc < 0)
448 goto fail;
449
450 rc = skx_get_node_id(d, &node_id);
451 if (rc < 0)
452 goto fail;
453
454 edac_dbg(2, "src_id = %d node_id = %d\n", src_id, node_id);
455 for (i = 0; i < I10NM_NUM_IMC; i++) {
456 if (!d->imc[i].mdev)
457 continue;
458
459 d->imc[i].mc = mc++;
460 d->imc[i].lmc = i;
461 d->imc[i].src_id = src_id;
462 d->imc[i].node_id = node_id;
463 if (d->imc[i].hbm_mc) {
464 d->imc[i].chan_mmio_sz = cfg->hbm_chan_mmio_sz;
465 d->imc[i].num_channels = I10NM_NUM_HBM_CHANNELS;
466 d->imc[i].num_dimms = I10NM_NUM_HBM_DIMMS;
467 } else {
468 d->imc[i].chan_mmio_sz = cfg->ddr_chan_mmio_sz;
469 d->imc[i].num_channels = I10NM_NUM_DDR_CHANNELS;
470 d->imc[i].num_dimms = I10NM_NUM_DDR_DIMMS;
471 }
472
473 rc = skx_register_mci(&d->imc[i], d->imc[i].mdev,
474 "Intel_10nm Socket", EDAC_MOD_STR,
475 i10nm_get_dimm_config, cfg);
476 if (rc < 0)
477 goto fail;
478 }
479 }
480
481 rc = skx_adxl_get();
482 if (rc)
483 goto fail;
484
485 opstate_init();
486 mce_register_decode_chain(&i10nm_mce_dec);
487 setup_i10nm_debug();
488
489 i10nm_printk(KERN_INFO, "%s\n", I10NM_REVISION);
490
491 return 0;
492fail:
493 skx_remove();
494 return rc;
495}
496
497static void __exit i10nm_exit(void)
498{
499 edac_dbg(2, "\n");
500 teardown_i10nm_debug();
501 mce_unregister_decode_chain(&i10nm_mce_dec);
502 skx_adxl_put();
503 skx_remove();
504}
505
506module_init(i10nm_init);
507module_exit(i10nm_exit);
508
509MODULE_LICENSE("GPL v2");
510MODULE_DESCRIPTION("MC Driver for Intel 10nm server processors");
511