/root/src/xen/xen/arch/x86/cpu/mcheck/amd_nonfatal.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * MCA implementation for AMD CPUs |
3 | | * Copyright (c) 2007 Advanced Micro Devices, Inc. |
4 | | * |
5 | | * This program is free software; you can redistribute it and/or modify |
6 | | * it under the terms of the GNU General Public License as published by |
7 | | * the Free Software Foundation; either version 2 of the License, or |
8 | | * (at your option) any later version. |
9 | | * |
10 | | * This program is distributed in the hope that it will be useful, |
11 | | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
12 | | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
13 | | * GNU General Public License for more details. |
14 | | * |
15 | | * You should have received a copy of the GNU General Public License |
16 | | * along with this program; If not, see <http://www.gnu.org/licenses/>. |
17 | | */ |
18 | | |
19 | | |
20 | | /* K8 common MCA documentation published at |
21 | | * |
22 | | * AMD64 Architecture Programmer's Manual Volume 2: |
23 | | * System Programming |
24 | | * Publication # 24593 Revision: 3.12 |
25 | | * Issue Date: September 2006 |
26 | | * |
27 | | * URL: |
28 | | * http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/24593.pdf |
29 | | */ |
30 | | |
31 | | /* The related documentation for K8 Revisions A - E is: |
32 | | * |
33 | | * BIOS and Kernel Developer's Guide for |
34 | | * AMD Athlon 64 and AMD Opteron Processors |
35 | | * Publication # 26094 Revision: 3.30 |
36 | | * Issue Date: February 2006 |
37 | | * |
38 | | * URL: |
39 | | * http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF |
40 | | */ |
41 | | |
42 | | /* The related documentation for K8 Revisions F - G is: |
43 | | * |
44 | | * BIOS and Kernel Developer's Guide for |
45 | | * AMD NPT Family 0Fh Processors |
46 | | * Publication # 32559 Revision: 3.04 |
47 | | * Issue Date: December 2006 |
48 | | * |
49 | | * URL: |
50 | | * http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/32559.pdf |
51 | | */ |
52 | | |
53 | | #include <xen/init.h> |
54 | | #include <xen/types.h> |
55 | | #include <xen/kernel.h> |
56 | | #include <xen/smp.h> |
57 | | #include <xen/timer.h> |
58 | | #include <xen/event.h> |
59 | | |
60 | | #include <asm/processor.h> |
61 | | #include <asm/system.h> |
62 | | #include <asm/msr.h> |
63 | | |
64 | | #include "mce.h" |
65 | | #include "vmce.h" |
66 | | |
67 | | static struct timer mce_timer; |
68 | | |
69 | | #define MCE_PERIOD MILLISECS(10000) |
70 | 0 | #define MCE_MIN MILLISECS(2000) |
71 | 0 | #define MCE_MAX MILLISECS(30000) |
72 | | |
73 | | static s_time_t period = MCE_PERIOD; |
74 | | static int hw_threshold = 0; |
75 | | static int adjust = 0; |
76 | | static int variable_period = 1; |
77 | | |
78 | | /* The polling service routine: |
79 | | * Collects information of correctable errors and notifies |
80 | | * Dom0 via an event. |
81 | | */ |
82 | | static void mce_amd_checkregs(void *info) |
83 | 0 | { |
84 | 0 | mctelem_cookie_t mctc; |
85 | 0 | struct mca_summary bs; |
86 | 0 |
|
87 | 0 | mctc = mcheck_mca_logout(MCA_POLLER, mca_allbanks, &bs, NULL); |
88 | 0 |
|
89 | 0 | if (bs.errcnt && mctc != NULL) { |
90 | 0 | static uint64_t dumpcount = 0; |
91 | 0 |
|
92 | 0 | /* If Dom0 enabled the VIRQ_MCA event, then notify it. |
93 | 0 | * Otherwise, if dom0 has had plenty of time to register |
94 | 0 | * the virq handler but still hasn't then dump telemetry |
95 | 0 | * to the Xen console. The call count may be incremented |
96 | 0 | * on multiple cpus at once and is indicative only - just |
97 | 0 | * a simple-minded attempt to avoid spamming the console |
98 | 0 | * for corrected errors in early startup. */ |
99 | 0 |
|
100 | 0 | if (dom0_vmce_enabled()) { |
101 | 0 | mctelem_commit(mctc); |
102 | 0 | send_global_virq(VIRQ_MCA); |
103 | 0 | } else if (++dumpcount >= 10) { |
104 | 0 | x86_mcinfo_dump((struct mc_info *)mctelem_dataptr(mctc)); |
105 | 0 | mctelem_dismiss(mctc); |
106 | 0 | } else { |
107 | 0 | mctelem_dismiss(mctc); |
108 | 0 | } |
109 | 0 |
|
110 | 0 | } else if (mctc != NULL) { |
111 | 0 | mctelem_dismiss(mctc); |
112 | 0 | } |
113 | 0 |
|
114 | 0 | /* adjust is global and all cpus may attempt to increment it without |
115 | 0 | * synchronisation, so they race and the final adjust count |
116 | 0 | * (number of cpus seeing any error) is approximate. We can |
117 | 0 | * guarantee that if any cpu observes an error that the |
118 | 0 | * adjust count is at least 1. */ |
119 | 0 | if (bs.errcnt) |
120 | 0 | adjust++; |
121 | 0 | } |
122 | | |
123 | | /* polling service routine invoker: |
124 | | * Adjust poll frequency at runtime. No error means slow polling frequency, |
125 | | * an error means higher polling frequency. |
126 | | * It uses hw threshold register introduced in AMD K8 RevF to detect |
127 | | * multiple correctable errors between two polls. In that case, |
128 | | * increase polling frequency higher than normal. |
129 | | */ |
130 | | static void mce_amd_work_fn(void *data) |
131 | 0 | { |
132 | 0 | on_each_cpu(mce_amd_checkregs, data, 1); |
133 | 0 |
|
134 | 0 | if (adjust > 0) { |
135 | 0 | if (!dom0_vmce_enabled()) { |
136 | 0 | /* Dom0 did not enable VIRQ_MCA, so Xen is reporting. */ |
137 | 0 | printk("MCE: polling routine found correctable error. " |
138 | 0 | " Use mcelog to parse above error output.\n"); |
139 | 0 | } |
140 | 0 | } |
141 | 0 |
|
142 | 0 | if (hw_threshold) { |
143 | 0 | uint64_t value; |
144 | 0 | uint32_t counter; |
145 | 0 |
|
146 | 0 | value = mca_rdmsr(MSR_IA32_MCx_MISC(4)); |
147 | 0 | /* Only the error counter field is of interest |
148 | 0 | * Bit field is described in AMD K8 BKDG chapter 6.4.5.5 |
149 | 0 | */ |
150 | 0 | counter = (value & 0xFFF00000000ULL) >> 32U; |
151 | 0 |
|
152 | 0 | /* HW does not count *all* kinds of correctable errors. |
153 | 0 | * Thus it is possible, that the polling routine finds an |
154 | 0 | * correctable error even if the HW reports nothing. */ |
155 | 0 | if (counter > 0) { |
156 | 0 | /* HW reported correctable errors, |
157 | 0 | * the polling routine did not find... |
158 | 0 | */ |
159 | 0 | if (adjust == 0) { |
160 | 0 | printk("CPU counter reports %"PRIu32 |
161 | 0 | " correctable hardware error%s that %s" |
162 | 0 | " not reported by the status MSRs\n", |
163 | 0 | counter, |
164 | 0 | (counter == 1 ? "" : "s"), |
165 | 0 | (counter == 1 ? "was" : "were")); |
166 | 0 | } |
167 | 0 | /* subtract 1 to not double count the error |
168 | 0 | * from the polling service routine */ |
169 | 0 | adjust += (counter - 1); |
170 | 0 |
|
171 | 0 | /* Restart counter */ |
172 | 0 | /* No interrupt, reset counter value */ |
173 | 0 | value &= ~(0x60FFF00000000ULL); |
174 | 0 | /* Counter enable */ |
175 | 0 | value |= (1ULL << 51); |
176 | 0 | mca_wrmsr(MSR_IA32_MCx_MISC(4), value); |
177 | 0 | } |
178 | 0 | } |
179 | 0 |
|
180 | 0 | if (variable_period && adjust > 0) { |
181 | 0 | /* Increase polling frequency */ |
182 | 0 | adjust++; /* adjust == 1 must have an effect */ |
183 | 0 | period /= adjust; |
184 | 0 | } else if (variable_period) { |
185 | 0 | /* Decrease polling frequency */ |
186 | 0 | period *= 2; |
187 | 0 | } |
188 | 0 | if (variable_period && period > MCE_MAX) { |
189 | 0 | /* limit: Poll at least every 30s */ |
190 | 0 | period = MCE_MAX; |
191 | 0 | } |
192 | 0 | if (variable_period && period < MCE_MIN) { |
193 | 0 | /* limit: Poll every 2s. |
194 | 0 | * When this is reached an uncorrectable error |
195 | 0 | * is expected to happen, if Dom0 does nothing. |
196 | 0 | */ |
197 | 0 | period = MCE_MIN; |
198 | 0 | } |
199 | 0 |
|
200 | 0 | set_timer(&mce_timer, NOW() + period); |
201 | 0 | adjust = 0; |
202 | 0 | } |
203 | | |
204 | | void __init amd_nonfatal_mcheck_init(struct cpuinfo_x86 *c) |
205 | 0 | { |
206 | 0 | if (c->x86_vendor != X86_VENDOR_AMD) |
207 | 0 | return; |
208 | 0 |
|
209 | 0 | /* Assume we are on K8 or newer AMD CPU here */ |
210 | 0 |
|
211 | 0 | /* The threshold bitfields in MSR_IA32_MC4_MISC has |
212 | 0 | * been introduced along with the SVME feature bit. */ |
213 | 0 | if (variable_period && cpu_has(c, X86_FEATURE_SVM)) { |
214 | 0 | uint64_t value; |
215 | 0 |
|
216 | 0 | /* hw threshold registers present */ |
217 | 0 | hw_threshold = 1; |
218 | 0 | rdmsrl(MSR_IA32_MCx_MISC(4), value); |
219 | 0 |
|
220 | 0 | if (value & (1ULL << 61)) { /* Locked bit */ |
221 | 0 | /* Locked by BIOS. Not available for use */ |
222 | 0 | hw_threshold = 0; |
223 | 0 | } |
224 | 0 | if (!(value & (1ULL << 63))) { /* Valid bit */ |
225 | 0 | /* No CtrP present */ |
226 | 0 | hw_threshold = 0; |
227 | 0 | } else { |
228 | 0 | if (!(value & (1ULL << 62))) { /* Counter Bit */ |
229 | 0 | /* No counter field present */ |
230 | 0 | hw_threshold = 0; |
231 | 0 | } |
232 | 0 | } |
233 | 0 |
|
234 | 0 | if (hw_threshold) { |
235 | 0 | /* No interrupt, reset counter value */ |
236 | 0 | value &= ~(0x60FFF00000000ULL); |
237 | 0 | /* Counter enable */ |
238 | 0 | value |= (1ULL << 51); |
239 | 0 | wrmsrl(MSR_IA32_MCx_MISC(4), value); |
240 | 0 | printk(XENLOG_INFO "MCA: Use hw thresholding to adjust polling frequency\n"); |
241 | 0 | } |
242 | 0 | } |
243 | 0 |
|
244 | 0 | init_timer(&mce_timer, mce_amd_work_fn, NULL, 0); |
245 | 0 | set_timer(&mce_timer, NOW() + period); |
246 | 0 | } |