Loading...
1/*
2 * Bus error event handling code for systems equipped with ECC
3 * handling logic, i.e. DECstation/DECsystem 5000/200 (KN02),
4 * 5000/240 (KN03), 5000/260 (KN05) and DECsystem 5900 (KN03),
5 * 5900/260 (KN05) systems.
6 *
7 * Copyright (c) 2003, 2005 Maciej W. Rozycki
8 *
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; either version
12 * 2 of the License, or (at your option) any later version.
13 */
14
15#include <linux/init.h>
16#include <linux/interrupt.h>
17#include <linux/kernel.h>
18#include <linux/sched.h>
19#include <linux/types.h>
20
21#include <asm/addrspace.h>
22#include <asm/bootinfo.h>
23#include <asm/cpu.h>
24#include <asm/irq_regs.h>
25#include <asm/processor.h>
26#include <asm/ptrace.h>
27#include <asm/traps.h>
28
29#include <asm/dec/ecc.h>
30#include <asm/dec/kn02.h>
31#include <asm/dec/kn03.h>
32#include <asm/dec/kn05.h>
33
34static volatile u32 *kn0x_erraddr;
35static volatile u32 *kn0x_chksyn;
36
37static inline void dec_ecc_be_ack(void)
38{
39 *kn0x_erraddr = 0; /* any write clears the IRQ */
40 iob();
41}
42
43static int dec_ecc_be_backend(struct pt_regs *regs, int is_fixup, int invoker)
44{
45 static const char excstr[] = "exception";
46 static const char intstr[] = "interrupt";
47 static const char cpustr[] = "CPU";
48 static const char dmastr[] = "DMA";
49 static const char readstr[] = "read";
50 static const char mreadstr[] = "memory read";
51 static const char writestr[] = "write";
52 static const char mwritstr[] = "partial memory write";
53 static const char timestr[] = "timeout";
54 static const char overstr[] = "overrun";
55 static const char eccstr[] = "ECC error";
56
57 const char *kind, *agent, *cycle, *event;
58 const char *status = "", *xbit = "", *fmt = "";
59 unsigned long address;
60 u16 syn = 0, sngl;
61
62 int i = 0;
63
64 u32 erraddr = *kn0x_erraddr;
65 u32 chksyn = *kn0x_chksyn;
66 int action = MIPS_BE_FATAL;
67
68 /* For non-ECC ack ASAP, so that any subsequent errors get caught. */
69 if ((erraddr & (KN0X_EAR_VALID | KN0X_EAR_ECCERR)) == KN0X_EAR_VALID)
70 dec_ecc_be_ack();
71
72 kind = invoker ? intstr : excstr;
73
74 if (!(erraddr & KN0X_EAR_VALID)) {
75 /* No idea what happened. */
76 printk(KERN_ALERT "Unidentified bus error %s\n", kind);
77 return action;
78 }
79
80 agent = (erraddr & KN0X_EAR_CPU) ? cpustr : dmastr;
81
82 if (erraddr & KN0X_EAR_ECCERR) {
83 /* An ECC error on a CPU or DMA transaction. */
84 cycle = (erraddr & KN0X_EAR_WRITE) ? mwritstr : mreadstr;
85 event = eccstr;
86 } else {
87 /* A CPU timeout or a DMA overrun. */
88 cycle = (erraddr & KN0X_EAR_WRITE) ? writestr : readstr;
89 event = (erraddr & KN0X_EAR_CPU) ? timestr : overstr;
90 }
91
92 address = erraddr & KN0X_EAR_ADDRESS;
93 /* For ECC errors on reads adjust for MT pipelining. */
94 if ((erraddr & (KN0X_EAR_WRITE | KN0X_EAR_ECCERR)) == KN0X_EAR_ECCERR)
95 address = (address & ~0xfffLL) | ((address - 5) & 0xfffLL);
96 address <<= 2;
97
98 /* Only CPU errors are fixable. */
99 if (erraddr & KN0X_EAR_CPU && is_fixup)
100 action = MIPS_BE_FIXUP;
101
102 if (erraddr & KN0X_EAR_ECCERR) {
103 static const u8 data_sbit[32] = {
104 0x4f, 0x4a, 0x52, 0x54, 0x57, 0x58, 0x5b, 0x5d,
105 0x23, 0x25, 0x26, 0x29, 0x2a, 0x2c, 0x31, 0x34,
106 0x0e, 0x0b, 0x13, 0x15, 0x16, 0x19, 0x1a, 0x1c,
107 0x62, 0x64, 0x67, 0x68, 0x6b, 0x6d, 0x70, 0x75,
108 };
109 static const u8 data_mbit[25] = {
110 0x07, 0x0d, 0x1f,
111 0x2f, 0x32, 0x37, 0x38, 0x3b, 0x3d, 0x3e,
112 0x43, 0x45, 0x46, 0x49, 0x4c, 0x51, 0x5e,
113 0x61, 0x6e, 0x73, 0x76, 0x79, 0x7a, 0x7c, 0x7f,
114 };
115 static const char sbestr[] = "corrected single";
116 static const char dbestr[] = "uncorrectable double";
117 static const char mbestr[] = "uncorrectable multiple";
118
119 if (!(address & 0x4))
120 syn = chksyn; /* Low bank. */
121 else
122 syn = chksyn >> 16; /* High bank. */
123
124 if (!(syn & KN0X_ESR_VLDLO)) {
125 /* Ack now, no rewrite will happen. */
126 dec_ecc_be_ack();
127
128 fmt = KERN_ALERT "%s" "invalid\n";
129 } else {
130 sngl = syn & KN0X_ESR_SNGLO;
131 syn &= KN0X_ESR_SYNLO;
132
133 /*
134 * Multibit errors may be tagged incorrectly;
135 * check the syndrome explicitly.
136 */
137 for (i = 0; i < 25; i++)
138 if (syn == data_mbit[i])
139 break;
140
141 if (i < 25) {
142 status = mbestr;
143 } else if (!sngl) {
144 status = dbestr;
145 } else {
146 volatile u32 *ptr =
147 (void *)CKSEG1ADDR(address);
148
149 *ptr = *ptr; /* Rewrite. */
150 iob();
151
152 status = sbestr;
153 action = MIPS_BE_DISCARD;
154 }
155
156 /* Ack now, now we've rewritten (or not). */
157 dec_ecc_be_ack();
158
159 if (syn && syn == (syn & -syn)) {
160 if (syn == 0x01) {
161 fmt = KERN_ALERT "%s"
162 "%#04x -- %s bit error "
163 "at check bit C%s\n";
164 xbit = "X";
165 } else {
166 fmt = KERN_ALERT "%s"
167 "%#04x -- %s bit error "
168 "at check bit C%s%u\n";
169 }
170 i = syn >> 2;
171 } else {
172 for (i = 0; i < 32; i++)
173 if (syn == data_sbit[i])
174 break;
175 if (i < 32)
176 fmt = KERN_ALERT "%s"
177 "%#04x -- %s bit error "
178 "at data bit D%s%u\n";
179 else
180 fmt = KERN_ALERT "%s"
181 "%#04x -- %s bit error\n";
182 }
183 }
184 }
185
186 if (action != MIPS_BE_FIXUP)
187 printk(KERN_ALERT "Bus error %s: %s %s %s at %#010lx\n",
188 kind, agent, cycle, event, address);
189
190 if (action != MIPS_BE_FIXUP && erraddr & KN0X_EAR_ECCERR)
191 printk(fmt, " ECC syndrome ", syn, status, xbit, i);
192
193 return action;
194}
195
196int dec_ecc_be_handler(struct pt_regs *regs, int is_fixup)
197{
198 return dec_ecc_be_backend(regs, is_fixup, 0);
199}
200
201irqreturn_t dec_ecc_be_interrupt(int irq, void *dev_id)
202{
203 struct pt_regs *regs = get_irq_regs();
204
205 int action = dec_ecc_be_backend(regs, 0, 1);
206
207 if (action == MIPS_BE_DISCARD)
208 return IRQ_HANDLED;
209
210 /*
211 * FIXME: Find the affected processes and kill them, otherwise
212 * we must die.
213 *
214 * The interrupt is asynchronously delivered thus EPC and RA
215 * may be irrelevant, but are printed for a reference.
216 */
217 printk(KERN_ALERT "Fatal bus interrupt, epc == %08lx, ra == %08lx\n",
218 regs->cp0_epc, regs->regs[31]);
219 die("Unrecoverable bus error", regs);
220}
221
222
223/*
224 * Initialization differs a bit between KN02 and KN03/KN05, so we
225 * need two variants. Once set up, all systems can be handled the
226 * same way.
227 */
228static inline void dec_kn02_be_init(void)
229{
230 volatile u32 *csr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CSR);
231
232 kn0x_erraddr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_ERRADDR);
233 kn0x_chksyn = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CHKSYN);
234
235 /* Preset write-only bits of the Control Register cache. */
236 cached_kn02_csr = *csr | KN02_CSR_LEDS;
237
238 /* Set normal ECC detection and generation. */
239 cached_kn02_csr &= ~(KN02_CSR_DIAGCHK | KN02_CSR_DIAGGEN);
240 /* Enable ECC correction. */
241 cached_kn02_csr |= KN02_CSR_CORRECT;
242 *csr = cached_kn02_csr;
243 iob();
244}
245
246static inline void dec_kn03_be_init(void)
247{
248 volatile u32 *mcr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_MCR);
249 volatile u32 *mbcs = (void *)CKSEG1ADDR(KN4K_SLOT_BASE + KN4K_MB_CSR);
250
251 kn0x_erraddr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_ERRADDR);
252 kn0x_chksyn = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_CHKSYN);
253
254 /*
255 * Set normal ECC detection and generation, enable ECC correction.
256 * For KN05 we also need to make sure EE (?) is enabled in the MB.
257 * Otherwise DBE/IBE exceptions would be masked but bus error
258 * interrupts would still arrive, resulting in an inevitable crash
259 * if get_dbe() triggers one.
260 */
261 *mcr = (*mcr & ~(KN03_MCR_DIAGCHK | KN03_MCR_DIAGGEN)) |
262 KN03_MCR_CORRECT;
263 if (current_cpu_type() == CPU_R4400SC)
264 *mbcs |= KN4K_MB_CSR_EE;
265 fast_iob();
266}
267
268void __init dec_ecc_be_init(void)
269{
270 if (mips_machtype == MACH_DS5000_200)
271 dec_kn02_be_init();
272 else
273 dec_kn03_be_init();
274
275 /* Clear any leftover errors from the firmware. */
276 dec_ecc_be_ack();
277}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Bus error event handling code for systems equipped with ECC
4 * handling logic, i.e. DECstation/DECsystem 5000/200 (KN02),
5 * 5000/240 (KN03), 5000/260 (KN05) and DECsystem 5900 (KN03),
6 * 5900/260 (KN05) systems.
7 *
8 * Copyright (c) 2003, 2005 Maciej W. Rozycki
9 */
10
11#include <linux/init.h>
12#include <linux/interrupt.h>
13#include <linux/kernel.h>
14#include <linux/sched.h>
15#include <linux/types.h>
16
17#include <asm/addrspace.h>
18#include <asm/bootinfo.h>
19#include <asm/cpu.h>
20#include <asm/cpu-type.h>
21#include <asm/irq_regs.h>
22#include <asm/processor.h>
23#include <asm/ptrace.h>
24#include <asm/traps.h>
25
26#include <asm/dec/ecc.h>
27#include <asm/dec/kn02.h>
28#include <asm/dec/kn03.h>
29#include <asm/dec/kn05.h>
30
31static volatile u32 *kn0x_erraddr;
32static volatile u32 *kn0x_chksyn;
33
34static inline void dec_ecc_be_ack(void)
35{
36 *kn0x_erraddr = 0; /* any write clears the IRQ */
37 iob();
38}
39
40static int dec_ecc_be_backend(struct pt_regs *regs, int is_fixup, int invoker)
41{
42 static const char excstr[] = "exception";
43 static const char intstr[] = "interrupt";
44 static const char cpustr[] = "CPU";
45 static const char dmastr[] = "DMA";
46 static const char readstr[] = "read";
47 static const char mreadstr[] = "memory read";
48 static const char writestr[] = "write";
49 static const char mwritstr[] = "partial memory write";
50 static const char timestr[] = "timeout";
51 static const char overstr[] = "overrun";
52 static const char eccstr[] = "ECC error";
53
54 const char *kind, *agent, *cycle, *event;
55 const char *status = "", *xbit = "", *fmt = "";
56 unsigned long address;
57 u16 syn = 0, sngl;
58
59 int i = 0;
60
61 u32 erraddr = *kn0x_erraddr;
62 u32 chksyn = *kn0x_chksyn;
63 int action = MIPS_BE_FATAL;
64
65 /* For non-ECC ack ASAP, so that any subsequent errors get caught. */
66 if ((erraddr & (KN0X_EAR_VALID | KN0X_EAR_ECCERR)) == KN0X_EAR_VALID)
67 dec_ecc_be_ack();
68
69 kind = invoker ? intstr : excstr;
70
71 if (!(erraddr & KN0X_EAR_VALID)) {
72 /* No idea what happened. */
73 printk(KERN_ALERT "Unidentified bus error %s\n", kind);
74 return action;
75 }
76
77 agent = (erraddr & KN0X_EAR_CPU) ? cpustr : dmastr;
78
79 if (erraddr & KN0X_EAR_ECCERR) {
80 /* An ECC error on a CPU or DMA transaction. */
81 cycle = (erraddr & KN0X_EAR_WRITE) ? mwritstr : mreadstr;
82 event = eccstr;
83 } else {
84 /* A CPU timeout or a DMA overrun. */
85 cycle = (erraddr & KN0X_EAR_WRITE) ? writestr : readstr;
86 event = (erraddr & KN0X_EAR_CPU) ? timestr : overstr;
87 }
88
89 address = erraddr & KN0X_EAR_ADDRESS;
90 /* For ECC errors on reads adjust for MT pipelining. */
91 if ((erraddr & (KN0X_EAR_WRITE | KN0X_EAR_ECCERR)) == KN0X_EAR_ECCERR)
92 address = (address & ~0xfffLL) | ((address - 5) & 0xfffLL);
93 address <<= 2;
94
95 /* Only CPU errors are fixable. */
96 if (erraddr & KN0X_EAR_CPU && is_fixup)
97 action = MIPS_BE_FIXUP;
98
99 if (erraddr & KN0X_EAR_ECCERR) {
100 static const u8 data_sbit[32] = {
101 0x4f, 0x4a, 0x52, 0x54, 0x57, 0x58, 0x5b, 0x5d,
102 0x23, 0x25, 0x26, 0x29, 0x2a, 0x2c, 0x31, 0x34,
103 0x0e, 0x0b, 0x13, 0x15, 0x16, 0x19, 0x1a, 0x1c,
104 0x62, 0x64, 0x67, 0x68, 0x6b, 0x6d, 0x70, 0x75,
105 };
106 static const u8 data_mbit[25] = {
107 0x07, 0x0d, 0x1f,
108 0x2f, 0x32, 0x37, 0x38, 0x3b, 0x3d, 0x3e,
109 0x43, 0x45, 0x46, 0x49, 0x4c, 0x51, 0x5e,
110 0x61, 0x6e, 0x73, 0x76, 0x79, 0x7a, 0x7c, 0x7f,
111 };
112 static const char sbestr[] = "corrected single";
113 static const char dbestr[] = "uncorrectable double";
114 static const char mbestr[] = "uncorrectable multiple";
115
116 if (!(address & 0x4))
117 syn = chksyn; /* Low bank. */
118 else
119 syn = chksyn >> 16; /* High bank. */
120
121 if (!(syn & KN0X_ESR_VLDLO)) {
122 /* Ack now, no rewrite will happen. */
123 dec_ecc_be_ack();
124
125 fmt = KERN_ALERT "%s" "invalid\n";
126 } else {
127 sngl = syn & KN0X_ESR_SNGLO;
128 syn &= KN0X_ESR_SYNLO;
129
130 /*
131 * Multibit errors may be tagged incorrectly;
132 * check the syndrome explicitly.
133 */
134 for (i = 0; i < 25; i++)
135 if (syn == data_mbit[i])
136 break;
137
138 if (i < 25) {
139 status = mbestr;
140 } else if (!sngl) {
141 status = dbestr;
142 } else {
143 volatile u32 *ptr =
144 (void *)CKSEG1ADDR(address);
145
146 *ptr = *ptr; /* Rewrite. */
147 iob();
148
149 status = sbestr;
150 action = MIPS_BE_DISCARD;
151 }
152
153 /* Ack now, now we've rewritten (or not). */
154 dec_ecc_be_ack();
155
156 if (syn && syn == (syn & -syn)) {
157 if (syn == 0x01) {
158 fmt = KERN_ALERT "%s"
159 "%#04x -- %s bit error "
160 "at check bit C%s\n";
161 xbit = "X";
162 } else {
163 fmt = KERN_ALERT "%s"
164 "%#04x -- %s bit error "
165 "at check bit C%s%u\n";
166 }
167 i = syn >> 2;
168 } else {
169 for (i = 0; i < 32; i++)
170 if (syn == data_sbit[i])
171 break;
172 if (i < 32)
173 fmt = KERN_ALERT "%s"
174 "%#04x -- %s bit error "
175 "at data bit D%s%u\n";
176 else
177 fmt = KERN_ALERT "%s"
178 "%#04x -- %s bit error\n";
179 }
180 }
181 }
182
183 if (action != MIPS_BE_FIXUP)
184 printk(KERN_ALERT "Bus error %s: %s %s %s at %#010lx\n",
185 kind, agent, cycle, event, address);
186
187 if (action != MIPS_BE_FIXUP && erraddr & KN0X_EAR_ECCERR)
188 printk(fmt, " ECC syndrome ", syn, status, xbit, i);
189
190 return action;
191}
192
193int dec_ecc_be_handler(struct pt_regs *regs, int is_fixup)
194{
195 return dec_ecc_be_backend(regs, is_fixup, 0);
196}
197
198irqreturn_t dec_ecc_be_interrupt(int irq, void *dev_id)
199{
200 struct pt_regs *regs = get_irq_regs();
201
202 int action = dec_ecc_be_backend(regs, 0, 1);
203
204 if (action == MIPS_BE_DISCARD)
205 return IRQ_HANDLED;
206
207 /*
208 * FIXME: Find the affected processes and kill them, otherwise
209 * we must die.
210 *
211 * The interrupt is asynchronously delivered thus EPC and RA
212 * may be irrelevant, but are printed for a reference.
213 */
214 printk(KERN_ALERT "Fatal bus interrupt, epc == %08lx, ra == %08lx\n",
215 regs->cp0_epc, regs->regs[31]);
216 die("Unrecoverable bus error", regs);
217}
218
219
220/*
221 * Initialization differs a bit between KN02 and KN03/KN05, so we
222 * need two variants. Once set up, all systems can be handled the
223 * same way.
224 */
225static inline void dec_kn02_be_init(void)
226{
227 volatile u32 *csr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CSR);
228
229 kn0x_erraddr = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_ERRADDR);
230 kn0x_chksyn = (void *)CKSEG1ADDR(KN02_SLOT_BASE + KN02_CHKSYN);
231
232 /* Preset write-only bits of the Control Register cache. */
233 cached_kn02_csr = *csr | KN02_CSR_LEDS;
234
235 /* Set normal ECC detection and generation. */
236 cached_kn02_csr &= ~(KN02_CSR_DIAGCHK | KN02_CSR_DIAGGEN);
237 /* Enable ECC correction. */
238 cached_kn02_csr |= KN02_CSR_CORRECT;
239 *csr = cached_kn02_csr;
240 iob();
241}
242
243static inline void dec_kn03_be_init(void)
244{
245 volatile u32 *mcr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_MCR);
246 volatile u32 *mbcs = (void *)CKSEG1ADDR(KN4K_SLOT_BASE + KN4K_MB_CSR);
247
248 kn0x_erraddr = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_ERRADDR);
249 kn0x_chksyn = (void *)CKSEG1ADDR(KN03_SLOT_BASE + IOASIC_CHKSYN);
250
251 /*
252 * Set normal ECC detection and generation, enable ECC correction.
253 * For KN05 we also need to make sure EE (?) is enabled in the MB.
254 * Otherwise DBE/IBE exceptions would be masked but bus error
255 * interrupts would still arrive, resulting in an inevitable crash
256 * if get_dbe() triggers one.
257 */
258 *mcr = (*mcr & ~(KN03_MCR_DIAGCHK | KN03_MCR_DIAGGEN)) |
259 KN03_MCR_CORRECT;
260 if (current_cpu_type() == CPU_R4400SC)
261 *mbcs |= KN4K_MB_CSR_EE;
262 fast_iob();
263}
264
265void __init dec_ecc_be_init(void)
266{
267 if (mips_machtype == MACH_DS5000_200)
268 dec_kn02_be_init();
269 else
270 dec_kn03_be_init();
271
272 /* Clear any leftover errors from the firmware. */
273 dec_ecc_be_ack();
274}