xref: /linux/drivers/edac/aspeed_edac.c (revision 786262be6048deab760f68c8acc2c85607165894)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * Copyright 2018, 2019 Cisco Systems
4  */
5 
6 #include <linux/edac.h>
7 #include <linux/module.h>
8 #include <linux/init.h>
9 #include <linux/interrupt.h>
10 #include <linux/platform_device.h>
11 #include <linux/stop_machine.h>
12 #include <linux/io.h>
13 #include <linux/of_address.h>
14 #include <linux/regmap.h>
15 #include "edac_module.h"
16 
17 
18 #define DRV_NAME "aspeed-edac"
19 
20 
21 #define ASPEED_MCR_PROT        0x00 /* protection key register */
22 #define ASPEED_MCR_CONF        0x04 /* configuration register */
23 #define ASPEED_MCR_INTR_CTRL   0x50 /* interrupt control/status register */
24 #define ASPEED_MCR_ADDR_UNREC  0x58 /* address of first un-recoverable error */
25 #define ASPEED_MCR_ADDR_REC    0x5c /* address of last recoverable error */
26 #define ASPEED_MCR_LAST        ASPEED_MCR_ADDR_REC
27 
28 
29 #define ASPEED_MCR_PROT_PASSWD	            0xfc600309
30 #define ASPEED_MCR_CONF_DRAM_TYPE               BIT(4)
31 #define ASPEED_MCR_CONF_ECC                     BIT(7)
32 #define ASPEED_MCR_INTR_CTRL_CLEAR             BIT(31)
33 #define ASPEED_MCR_INTR_CTRL_CNT_REC   GENMASK(23, 16)
34 #define ASPEED_MCR_INTR_CTRL_CNT_UNREC GENMASK(15, 12)
35 #define ASPEED_MCR_INTR_CTRL_ENABLE  (BIT(0) | BIT(1))
36 
37 
38 static struct regmap *aspeed_regmap;
39 
40 
regmap_reg_write(void * context,unsigned int reg,unsigned int val)41 static int regmap_reg_write(void *context, unsigned int reg, unsigned int val)
42 {
43 	void __iomem *regs = (void __iomem *)context;
44 
45 	/* enable write to MCR register set */
46 	writel(ASPEED_MCR_PROT_PASSWD, regs + ASPEED_MCR_PROT);
47 
48 	writel(val, regs + reg);
49 
50 	/* disable write to MCR register set */
51 	writel(~ASPEED_MCR_PROT_PASSWD, regs + ASPEED_MCR_PROT);
52 
53 	return 0;
54 }
55 
56 
regmap_reg_read(void * context,unsigned int reg,unsigned int * val)57 static int regmap_reg_read(void *context, unsigned int reg, unsigned int *val)
58 {
59 	void __iomem *regs = (void __iomem *)context;
60 
61 	*val = readl(regs + reg);
62 
63 	return 0;
64 }
65 
regmap_is_volatile(struct device * dev,unsigned int reg)66 static bool regmap_is_volatile(struct device *dev, unsigned int reg)
67 {
68 	switch (reg) {
69 	case ASPEED_MCR_PROT:
70 	case ASPEED_MCR_INTR_CTRL:
71 	case ASPEED_MCR_ADDR_UNREC:
72 	case ASPEED_MCR_ADDR_REC:
73 		return true;
74 	default:
75 		return false;
76 	}
77 }
78 
79 
80 static const struct regmap_config aspeed_regmap_config = {
81 	.reg_bits = 32,
82 	.val_bits = 32,
83 	.reg_stride = 4,
84 	.max_register = ASPEED_MCR_LAST,
85 	.reg_write = regmap_reg_write,
86 	.reg_read = regmap_reg_read,
87 	.volatile_reg = regmap_is_volatile,
88 	.fast_io = true,
89 };
90 
91 
count_rec(struct mem_ctl_info * mci,u8 rec_cnt,u32 rec_addr)92 static void count_rec(struct mem_ctl_info *mci, u8 rec_cnt, u32 rec_addr)
93 {
94 	struct csrow_info *csrow = mci->csrows[0];
95 	u32 page, offset, syndrome;
96 
97 	if (!rec_cnt)
98 		return;
99 
100 	/* report first few errors (if there are) */
101 	/* note: no addresses are recorded */
102 	if (rec_cnt > 1) {
103 		/* page, offset and syndrome are not available */
104 		page = 0;
105 		offset = 0;
106 		syndrome = 0;
107 		edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, rec_cnt-1,
108 				     page, offset, syndrome, 0, 0, -1,
109 				     "address(es) not available", "");
110 	}
111 
112 	/* report last error */
113 	/* note: rec_addr is the last recoverable error addr */
114 	page = rec_addr >> PAGE_SHIFT;
115 	offset = rec_addr & ~PAGE_MASK;
116 	/* syndrome is not available */
117 	syndrome = 0;
118 	edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1,
119 			     csrow->first_page + page, offset, syndrome,
120 			     0, 0, -1, "", "");
121 }
122 
123 
count_un_rec(struct mem_ctl_info * mci,u8 un_rec_cnt,u32 un_rec_addr)124 static void count_un_rec(struct mem_ctl_info *mci, u8 un_rec_cnt,
125 			 u32 un_rec_addr)
126 {
127 	struct csrow_info *csrow = mci->csrows[0];
128 	u32 page, offset, syndrome;
129 
130 	if (!un_rec_cnt)
131 		return;
132 
133 	/* report 1. error */
134 	/* note: un_rec_addr is the first unrecoverable error addr */
135 	page = un_rec_addr >> PAGE_SHIFT;
136 	offset = un_rec_addr & ~PAGE_MASK;
137 	/* syndrome is not available */
138 	syndrome = 0;
139 	edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1,
140 			     csrow->first_page + page, offset, syndrome,
141 			     0, 0, -1, "", "");
142 
143 	/* report further errors (if there are) */
144 	/* note: no addresses are recorded */
145 	if (un_rec_cnt > 1) {
146 		/* page, offset and syndrome are not available */
147 		page = 0;
148 		offset = 0;
149 		syndrome = 0;
150 		edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, un_rec_cnt-1,
151 				     page, offset, syndrome, 0, 0, -1,
152 				     "address(es) not available", "");
153 	}
154 }
155 
156 
mcr_isr(int irq,void * arg)157 static irqreturn_t mcr_isr(int irq, void *arg)
158 {
159 	struct mem_ctl_info *mci = arg;
160 	u32 rec_addr, un_rec_addr;
161 	u32 reg50, reg5c, reg58;
162 	u8  rec_cnt, un_rec_cnt;
163 
164 	regmap_read(aspeed_regmap, ASPEED_MCR_INTR_CTRL, &reg50);
165 	dev_dbg(mci->pdev, "received edac interrupt w/ mcr register 50: 0x%x\n",
166 		reg50);
167 
168 	/* collect data about recoverable and unrecoverable errors */
169 	rec_cnt = (reg50 & ASPEED_MCR_INTR_CTRL_CNT_REC) >> 16;
170 	un_rec_cnt = (reg50 & ASPEED_MCR_INTR_CTRL_CNT_UNREC) >> 12;
171 
172 	dev_dbg(mci->pdev, "%d recoverable interrupts and %d unrecoverable interrupts\n",
173 		rec_cnt, un_rec_cnt);
174 
175 	regmap_read(aspeed_regmap, ASPEED_MCR_ADDR_UNREC, &reg58);
176 	un_rec_addr = reg58;
177 
178 	regmap_read(aspeed_regmap, ASPEED_MCR_ADDR_REC, &reg5c);
179 	rec_addr = reg5c;
180 
181 	/* clear interrupt flags and error counters: */
182 	regmap_update_bits(aspeed_regmap, ASPEED_MCR_INTR_CTRL,
183 			   ASPEED_MCR_INTR_CTRL_CLEAR,
184 			   ASPEED_MCR_INTR_CTRL_CLEAR);
185 
186 	regmap_update_bits(aspeed_regmap, ASPEED_MCR_INTR_CTRL,
187 			   ASPEED_MCR_INTR_CTRL_CLEAR, 0);
188 
189 	/* process recoverable and unrecoverable errors */
190 	count_rec(mci, rec_cnt, rec_addr);
191 	count_un_rec(mci, un_rec_cnt, un_rec_addr);
192 
193 	if (!rec_cnt && !un_rec_cnt)
194 		dev_dbg(mci->pdev, "received edac interrupt, but did not find any ECC counters\n");
195 
196 	regmap_read(aspeed_regmap, ASPEED_MCR_INTR_CTRL, &reg50);
197 	dev_dbg(mci->pdev, "edac interrupt handled. mcr reg 50 is now: 0x%x\n",
198 		reg50);
199 
200 	return IRQ_HANDLED;
201 }
202 
203 
config_irq(void * ctx,struct platform_device * pdev)204 static int config_irq(void *ctx, struct platform_device *pdev)
205 {
206 	int irq;
207 	int rc;
208 
209 	/* register interrupt handler */
210 	irq = platform_get_irq(pdev, 0);
211 	dev_dbg(&pdev->dev, "got irq %d\n", irq);
212 	if (irq < 0)
213 		return irq;
214 
215 	rc = devm_request_irq(&pdev->dev, irq, mcr_isr, IRQF_TRIGGER_HIGH,
216 			      DRV_NAME, ctx);
217 	if (rc)
218 		return rc;
219 
220 	/* enable interrupts */
221 	regmap_update_bits(aspeed_regmap, ASPEED_MCR_INTR_CTRL,
222 			   ASPEED_MCR_INTR_CTRL_ENABLE,
223 			   ASPEED_MCR_INTR_CTRL_ENABLE);
224 
225 	return 0;
226 }
227 
228 
init_csrows(struct mem_ctl_info * mci)229 static int init_csrows(struct mem_ctl_info *mci)
230 {
231 	struct csrow_info *csrow = mci->csrows[0];
232 	u32 nr_pages, dram_type;
233 	struct dimm_info *dimm;
234 	struct device_node *np;
235 	struct resource r;
236 	u32 reg04;
237 	int rc;
238 
239 	/* retrieve info about physical memory from device tree */
240 	np = of_find_node_by_name(NULL, "memory");
241 	if (!np) {
242 		dev_err(mci->pdev, "dt: missing /memory node\n");
243 		return -ENODEV;
244 	}
245 
246 	rc = of_address_to_resource(np, 0, &r);
247 
248 	of_node_put(np);
249 
250 	if (rc) {
251 		dev_err(mci->pdev, "dt: failed requesting resource for /memory node\n");
252 		return rc;
253 	}
254 
255 	dev_dbg(mci->pdev, "dt: /memory node resources: first page %pR, PAGE_SHIFT macro=0x%x\n",
256 		&r, PAGE_SHIFT);
257 
258 	csrow->first_page = r.start >> PAGE_SHIFT;
259 	nr_pages = resource_size(&r) >> PAGE_SHIFT;
260 	csrow->last_page = csrow->first_page + nr_pages - 1;
261 
262 	regmap_read(aspeed_regmap, ASPEED_MCR_CONF, &reg04);
263 	dram_type = (reg04 & ASPEED_MCR_CONF_DRAM_TYPE) ? MEM_DDR4 : MEM_DDR3;
264 
265 	dimm = csrow->channels[0]->dimm;
266 	dimm->mtype = dram_type;
267 	dimm->edac_mode = EDAC_SECDED;
268 	dimm->nr_pages = nr_pages / csrow->nr_channels;
269 
270 	dev_dbg(mci->pdev, "initialized dimm with first_page=0x%lx and nr_pages=0x%x\n",
271 		csrow->first_page, nr_pages);
272 
273 	return 0;
274 }
275 
276 
aspeed_probe(struct platform_device * pdev)277 static int aspeed_probe(struct platform_device *pdev)
278 {
279 	struct device *dev = &pdev->dev;
280 	struct edac_mc_layer layers[2];
281 	struct mem_ctl_info *mci;
282 	void __iomem *regs;
283 	u32 reg04;
284 	int rc;
285 
286 	regs = devm_platform_ioremap_resource(pdev, 0);
287 	if (IS_ERR(regs))
288 		return PTR_ERR(regs);
289 
290 	aspeed_regmap = devm_regmap_init(dev, NULL, (__force void *)regs,
291 					 &aspeed_regmap_config);
292 	if (IS_ERR(aspeed_regmap))
293 		return PTR_ERR(aspeed_regmap);
294 
295 	/* bail out if ECC mode is not configured */
296 	regmap_read(aspeed_regmap, ASPEED_MCR_CONF, &reg04);
297 	if (!(reg04 & ASPEED_MCR_CONF_ECC)) {
298 		dev_err(&pdev->dev, "ECC mode is not configured in u-boot\n");
299 		return -EPERM;
300 	}
301 
302 	edac_op_state = EDAC_OPSTATE_INT;
303 
304 	/* allocate & init EDAC MC data structure */
305 	layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
306 	layers[0].size = 1;
307 	layers[0].is_virt_csrow = true;
308 	layers[1].type = EDAC_MC_LAYER_CHANNEL;
309 	layers[1].size = 1;
310 	layers[1].is_virt_csrow = false;
311 
312 	mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 0);
313 	if (!mci)
314 		return -ENOMEM;
315 
316 	mci->pdev = &pdev->dev;
317 	mci->mtype_cap = MEM_FLAG_DDR3 | MEM_FLAG_DDR4;
318 	mci->edac_ctl_cap = EDAC_FLAG_SECDED;
319 	mci->edac_cap = EDAC_FLAG_SECDED;
320 	mci->scrub_cap = SCRUB_FLAG_HW_SRC;
321 	mci->scrub_mode = SCRUB_HW_SRC;
322 	mci->mod_name = DRV_NAME;
323 	mci->ctl_name = "MIC";
324 	mci->dev_name = dev_name(&pdev->dev);
325 
326 	rc = init_csrows(mci);
327 	if (rc) {
328 		dev_err(&pdev->dev, "failed to init csrows\n");
329 		goto probe_exit02;
330 	}
331 
332 	platform_set_drvdata(pdev, mci);
333 
334 	/* register with edac core */
335 	rc = edac_mc_add_mc(mci);
336 	if (rc) {
337 		dev_err(&pdev->dev, "failed to register with EDAC core\n");
338 		goto probe_exit02;
339 	}
340 
341 	/* register interrupt handler and enable interrupts */
342 	rc = config_irq(mci, pdev);
343 	if (rc) {
344 		dev_err(&pdev->dev, "failed setting up irq\n");
345 		goto probe_exit01;
346 	}
347 
348 	return 0;
349 
350 probe_exit01:
351 	edac_mc_del_mc(&pdev->dev);
352 probe_exit02:
353 	edac_mc_free(mci);
354 	return rc;
355 }
356 
357 
aspeed_remove(struct platform_device * pdev)358 static void aspeed_remove(struct platform_device *pdev)
359 {
360 	struct mem_ctl_info *mci;
361 
362 	/* disable interrupts */
363 	regmap_update_bits(aspeed_regmap, ASPEED_MCR_INTR_CTRL,
364 			   ASPEED_MCR_INTR_CTRL_ENABLE, 0);
365 
366 	/* free resources */
367 	mci = edac_mc_del_mc(&pdev->dev);
368 	if (mci)
369 		edac_mc_free(mci);
370 }
371 
372 
373 static const struct of_device_id aspeed_of_match[] = {
374 	{ .compatible = "aspeed,ast2400-sdram-edac" },
375 	{ .compatible = "aspeed,ast2500-sdram-edac" },
376 	{ .compatible = "aspeed,ast2600-sdram-edac" },
377 	{},
378 };
379 
380 MODULE_DEVICE_TABLE(of, aspeed_of_match);
381 
382 static struct platform_driver aspeed_driver = {
383 	.driver		= {
384 		.name	= DRV_NAME,
385 		.of_match_table = aspeed_of_match
386 	},
387 	.probe		= aspeed_probe,
388 	.remove		= aspeed_remove
389 };
390 module_platform_driver(aspeed_driver);
391 
392 MODULE_LICENSE("GPL");
393 MODULE_AUTHOR("Stefan Schaeckeler <sschaeck@cisco.com>");
394 MODULE_DESCRIPTION("Aspeed BMC SoC EDAC driver");
395 MODULE_VERSION("1.0");
396