pdt.c 8.99 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
2
3
4
/*
 *    Page Deallocation Table (PDT) support
 *
5
6
7
8
 *    The Page Deallocation Table (PDT) is maintained by firmware and holds a
 *    list of memory addresses in which memory errors were detected.
 *    The list contains both single-bit (correctable) and double-bit
 *    (uncorrectable) errors.
9
10
11
 *
 *    Copyright 2017 by Helge Deller <deller@gmx.de>
 *
12
13
 *    possible future enhancements:
 *    - add userspace interface via procfs or sysfs to clear PDT
14
15
16
17
 */

#include <linux/memblock.h>
#include <linux/seq_file.h>
18
#include <linux/kthread.h>
19
#include <linux/initrd.h>
20
21
22
23
24
25
26
27
28
29

#include <asm/pdc.h>
#include <asm/pdcpat.h>
#include <asm/sections.h>
#include <asm/pgtable.h>

enum pdt_access_type {
	PDT_NONE,
	PDT_PDC,
	PDT_PAT_NEW,
30
	PDT_PAT_CELL
31
32
33
34
};

static enum pdt_access_type pdt_type;

35
36
37
38
39
/* PDT poll interval: 1 minute if errors, 5 minutes if everything OK. */
#define PDT_POLL_INTERVAL_DEFAULT	(5*60*HZ)
#define PDT_POLL_INTERVAL_SHORT		(1*60*HZ)
static unsigned long pdt_poll_interval = PDT_POLL_INTERVAL_DEFAULT;

40
41
42
43
44
45
46
/* global PDT status information */
static struct pdc_mem_retinfo pdt_status;

#define MAX_PDT_TABLE_SIZE	PAGE_SIZE
#define MAX_PDT_ENTRIES		(MAX_PDT_TABLE_SIZE / sizeof(unsigned long))
static unsigned long pdt_entry[MAX_PDT_ENTRIES] __page_aligned_bss;

47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
/*
 * Constants for the pdt_entry format:
 * A pdt_entry holds the physical address in bits 0-57, bits 58-61 are
 * reserved, bit 62 is the perm bit and bit 63 is the error_type bit.
 * The perm bit indicates whether the error have been verified as a permanent
 * error (value of 1) or has not been verified, and may be transient (value
 * of 0). The error_type bit indicates whether the error is a single bit error
 * (value of 1) or a multiple bit error.
 * On non-PAT machines phys_addr is encoded in bits 0-59 and error_type in bit
 * 63. Those machines don't provide the perm bit.
 */

#define PDT_ADDR_PHYS_MASK	(pdt_type != PDT_PDC ? ~0x3f : ~0x0f)
#define PDT_ADDR_PERM_ERR	(pdt_type != PDT_PDC ? 2UL : 0UL)
#define PDT_ADDR_SINGLE_ERR	1UL
62
63
64
65
66
67
68
69
70
71
72
73
74

/* report PDT entries via /proc/meminfo */
void arch_report_meminfo(struct seq_file *m)
{
	if (pdt_type == PDT_NONE)
		return;

	seq_printf(m, "PDT_max_entries: %7lu\n",
			pdt_status.pdt_size);
	seq_printf(m, "PDT_cur_entries: %7lu\n",
			pdt_status.pdt_entries);
}

75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
static int get_info_pat_new(void)
{
	struct pdc_pat_mem_retinfo pat_rinfo;
	int ret;

	/* newer PAT machines like C8000 report info for all cells */
	if (is_pdc_pat())
		ret = pdc_pat_mem_pdt_info(&pat_rinfo);
	else
		return PDC_BAD_PROC;

	pdt_status.pdt_size = pat_rinfo.max_pdt_entries;
	pdt_status.pdt_entries = pat_rinfo.current_pdt_entries;
	pdt_status.pdt_status = 0;
	pdt_status.first_dbe_loc = pat_rinfo.first_dbe_loc;
	pdt_status.good_mem = pat_rinfo.good_mem;

	return ret;
}

static int get_info_pat_cell(void)
{
	struct pdc_pat_mem_cell_pdt_retinfo cell_rinfo;
	int ret;

	/* older PAT machines like rp5470 report cell info only */
	if (is_pdc_pat())
		ret = pdc_pat_mem_pdt_cell_info(&cell_rinfo, parisc_cell_num);
	else
		return PDC_BAD_PROC;

	pdt_status.pdt_size = cell_rinfo.max_pdt_entries;
	pdt_status.pdt_entries = cell_rinfo.current_pdt_entries;
	pdt_status.pdt_status = 0;
	pdt_status.first_dbe_loc = cell_rinfo.first_dbe_loc;
	pdt_status.good_mem = cell_rinfo.good_mem;

	return ret;
}

static void report_mem_err(unsigned long pde)
{
	struct pdc_pat_mem_phys_mem_location loc;
	unsigned long addr;
	char dimm_txt[32];

	addr = pde & PDT_ADDR_PHYS_MASK;

	/* show DIMM slot description on PAT machines */
	if (is_pdc_pat()) {
		pdc_pat_mem_get_dimm_phys_location(&loc, addr);
		sprintf(dimm_txt, "DIMM slot %02x, ", loc.dimm_slot);
	} else
		dimm_txt[0] = 0;

	pr_warn("PDT: BAD MEMORY at 0x%08lx, %s%s%s-bit error.\n",
		addr, dimm_txt,
		pde & PDT_ADDR_PERM_ERR ? "permanent ":"",
		pde & PDT_ADDR_SINGLE_ERR ? "single":"multi");
}


137
138
139
140
141
142
143
144
145
146
147
148
149
150
/*
 * pdc_pdt_init()
 *
 * Initialize kernel PDT structures, read initial PDT table from firmware,
 * report all current PDT entries and mark bad memory with memblock_reserve()
 * to avoid that the kernel will use broken memory areas.
 *
 */
void __init pdc_pdt_init(void)
{
	int ret, i;
	unsigned long entries;
	struct pdc_mem_read_pdt pdt_read_ret;

151
152
	pdt_type = PDT_PAT_NEW;
	ret = get_info_pat_new();
153

154
155
156
157
158
159
	if (ret != PDC_OK) {
		pdt_type = PDT_PAT_CELL;
		ret = get_info_pat_cell();
	}

	if (ret != PDC_OK) {
160
		pdt_type = PDT_PDC;
161
		/* non-PAT machines provide the standard PDC call */
162
163
164
165
166
167
168
169
170
171
172
		ret = pdc_mem_pdt_info(&pdt_status);
	}

	if (ret != PDC_OK) {
		pdt_type = PDT_NONE;
		pr_info("PDT: Firmware does not provide any page deallocation"
			" information.\n");
		return;
	}

	entries = pdt_status.pdt_entries;
173
174
	if (WARN_ON(entries > MAX_PDT_ENTRIES))
		entries = pdt_status.pdt_entries = MAX_PDT_ENTRIES;
175

176
177
178
179
180
	pr_info("PDT: type %s, size %lu, entries %lu, status %lu, dbe_loc 0x%lx,"
		" good_mem %lu MB\n",
			pdt_type == PDT_PDC ? __stringify(PDT_PDC) :
			pdt_type == PDT_PAT_CELL ? __stringify(PDT_PAT_CELL)
						 : __stringify(PDT_PAT_NEW),
181
182
			pdt_status.pdt_size, pdt_status.pdt_entries,
			pdt_status.pdt_status, pdt_status.first_dbe_loc,
183
			pdt_status.good_mem / 1024 / 1024);
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202

	if (entries == 0) {
		pr_info("PDT: Firmware reports all memory OK.\n");
		return;
	}

	if (pdt_status.first_dbe_loc &&
		pdt_status.first_dbe_loc <= __pa((unsigned long)&_end))
		pr_crit("CRITICAL: Bad memory inside kernel image memory area!\n");

	pr_warn("PDT: Firmware reports %lu entries of faulty memory:\n",
		entries);

	if (pdt_type == PDT_PDC)
		ret = pdc_mem_pdt_read_entries(&pdt_read_ret, pdt_entry);
	else {
#ifdef CONFIG_64BIT
		struct pdc_pat_mem_read_pd_retinfo pat_pret;

203
204
205
206
		if (pdt_type == PDT_PAT_CELL)
			ret = pdc_pat_mem_read_cell_pdt(&pat_pret, pdt_entry,
				MAX_PDT_ENTRIES);
		else
207
208
209
210
211
212
213
214
215
			ret = pdc_pat_mem_read_pd_pdt(&pat_pret, pdt_entry,
				MAX_PDT_TABLE_SIZE, 0);
#else
		ret = PDC_BAD_PROC;
#endif
	}

	if (ret != PDC_OK) {
		pdt_type = PDT_NONE;
216
		pr_warn("PDT: Get PDT entries failed with %d\n", ret);
217
218
219
220
		return;
	}

	for (i = 0; i < pdt_status.pdt_entries; i++) {
221
222
		unsigned long addr;

223
224
		report_mem_err(pdt_entry[i]);

225
226
227
228
229
230
		addr = pdt_entry[i] & PDT_ADDR_PHYS_MASK;
		if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) &&
			addr >= initrd_start && addr < initrd_end)
			pr_crit("CRITICAL: initrd possibly broken "
				"due to bad memory!\n");

231
232
233
234
		/* mark memory page bad */
		memblock_reserve(pdt_entry[i] & PAGE_MASK, PAGE_SIZE);
	}
}
235

236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293

/*
 * This is the PDT kernel thread main loop.
 */

static int pdt_mainloop(void *unused)
{
	struct pdc_mem_read_pdt pdt_read_ret;
	struct pdc_pat_mem_read_pd_retinfo pat_pret __maybe_unused;
	unsigned long old_num_entries;
	unsigned long *bad_mem_ptr;
	int num, ret;

	for (;;) {
		set_current_state(TASK_INTERRUPTIBLE);

		old_num_entries = pdt_status.pdt_entries;

		schedule_timeout(pdt_poll_interval);
		if (kthread_should_stop())
			break;

		/* Do we have new PDT entries? */
		switch (pdt_type) {
		case PDT_PAT_NEW:
			ret = get_info_pat_new();
			break;
		case PDT_PAT_CELL:
			ret = get_info_pat_cell();
			break;
		default:
			ret = pdc_mem_pdt_info(&pdt_status);
			break;
		}

		if (ret != PDC_OK) {
			pr_warn("PDT: unexpected failure %d\n", ret);
			return -EINVAL;
		}

		/* if no new PDT entries, just wait again */
		num = pdt_status.pdt_entries - old_num_entries;
		if (num <= 0)
			continue;

		/* decrease poll interval in case we found memory errors */
		if (pdt_status.pdt_entries &&
			pdt_poll_interval == PDT_POLL_INTERVAL_DEFAULT)
			pdt_poll_interval = PDT_POLL_INTERVAL_SHORT;

		/* limit entries to get */
		if (num > MAX_PDT_ENTRIES) {
			num = MAX_PDT_ENTRIES;
			pdt_status.pdt_entries = old_num_entries + num;
		}

		/* get new entries */
		switch (pdt_type) {
294
#ifdef CONFIG_64BIT
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
		case PDT_PAT_CELL:
			if (pdt_status.pdt_entries > MAX_PDT_ENTRIES) {
				pr_crit("PDT: too many entries.\n");
				return -ENOMEM;
			}
			ret = pdc_pat_mem_read_cell_pdt(&pat_pret, pdt_entry,
				MAX_PDT_ENTRIES);
			bad_mem_ptr = &pdt_entry[old_num_entries];
			break;
		case PDT_PAT_NEW:
			ret = pdc_pat_mem_read_pd_pdt(&pat_pret,
				pdt_entry,
				num * sizeof(unsigned long),
				old_num_entries * sizeof(unsigned long));
			bad_mem_ptr = &pdt_entry[0];
			break;
311
#endif
312
313
314
315
316
317
		default:
			ret = pdc_mem_pdt_read_entries(&pdt_read_ret,
				pdt_entry);
			bad_mem_ptr = &pdt_entry[old_num_entries];
			break;
		}
318

319
320
321
		/* report and mark memory broken */
		while (num--) {
			unsigned long pde = *bad_mem_ptr++;
322

323
324
325
326
327
			report_mem_err(pde);

#ifdef CONFIG_MEMORY_FAILURE
			if ((pde & PDT_ADDR_PERM_ERR) ||
			    ((pde & PDT_ADDR_SINGLE_ERR) == 0))
328
				memory_failure(pde >> PAGE_SHIFT, 0);
329
330
331
332
333
334
335
336
337
338
339
			else
				soft_offline_page(
					pfn_to_page(pde >> PAGE_SHIFT), 0);
#else
			pr_crit("PDT: memory error at 0x%lx ignored.\n"
				"Rebuild kernel with CONFIG_MEMORY_FAILURE=y "
				"for real handling.\n",
				pde & PDT_ADDR_PHYS_MASK);
#endif

		}
340
	}
341
342

	return 0;
343
}
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362


static int __init pdt_initcall(void)
{
	struct task_struct *kpdtd_task;

	if (pdt_type == PDT_NONE)
		return -ENODEV;

	kpdtd_task = kthread_create(pdt_mainloop, NULL, "kpdtd");
	if (IS_ERR(kpdtd_task))
		return PTR_ERR(kpdtd_task);

	wake_up_process(kpdtd_task);

	return 0;
}

late_initcall(pdt_initcall);