diff options
Diffstat (limited to 'arch/powerpc/platforms/pseries/dtl.c')
-rw-r--r-- | arch/powerpc/platforms/pseries/dtl.c | 224 |
1 files changed, 169 insertions, 55 deletions
diff --git a/arch/powerpc/platforms/pseries/dtl.c b/arch/powerpc/platforms/pseries/dtl.c index a00addb55945..c371bc06434b 100644 --- a/arch/powerpc/platforms/pseries/dtl.c +++ b/arch/powerpc/platforms/pseries/dtl.c | |||
@@ -23,37 +23,22 @@ | |||
23 | #include <linux/init.h> | 23 | #include <linux/init.h> |
24 | #include <linux/slab.h> | 24 | #include <linux/slab.h> |
25 | #include <linux/debugfs.h> | 25 | #include <linux/debugfs.h> |
26 | #include <linux/spinlock.h> | ||
26 | #include <asm/smp.h> | 27 | #include <asm/smp.h> |
27 | #include <asm/system.h> | 28 | #include <asm/system.h> |
28 | #include <asm/uaccess.h> | 29 | #include <asm/uaccess.h> |
29 | #include <asm/firmware.h> | 30 | #include <asm/firmware.h> |
31 | #include <asm/lppaca.h> | ||
30 | 32 | ||
31 | #include "plpar_wrappers.h" | 33 | #include "plpar_wrappers.h" |
32 | 34 | ||
33 | /* | ||
34 | * Layout of entries in the hypervisor's DTL buffer. Although we don't | ||
35 | * actually access the internals of an entry (we only need to know the size), | ||
36 | * we might as well define it here for reference. | ||
37 | */ | ||
38 | struct dtl_entry { | ||
39 | u8 dispatch_reason; | ||
40 | u8 preempt_reason; | ||
41 | u16 processor_id; | ||
42 | u32 enqueue_to_dispatch_time; | ||
43 | u32 ready_to_enqueue_time; | ||
44 | u32 waiting_to_ready_time; | ||
45 | u64 timebase; | ||
46 | u64 fault_addr; | ||
47 | u64 srr0; | ||
48 | u64 srr1; | ||
49 | }; | ||
50 | |||
51 | struct dtl { | 35 | struct dtl { |
52 | struct dtl_entry *buf; | 36 | struct dtl_entry *buf; |
53 | struct dentry *file; | 37 | struct dentry *file; |
54 | int cpu; | 38 | int cpu; |
55 | int buf_entries; | 39 | int buf_entries; |
56 | u64 last_idx; | 40 | u64 last_idx; |
41 | spinlock_t lock; | ||
57 | }; | 42 | }; |
58 | static DEFINE_PER_CPU(struct dtl, cpu_dtl); | 43 | static DEFINE_PER_CPU(struct dtl, cpu_dtl); |
59 | 44 | ||
@@ -72,25 +57,97 @@ static u8 dtl_event_mask = 0x7; | |||
72 | static int dtl_buf_entries = (16 * 85); | 57 | static int dtl_buf_entries = (16 * 85); |
73 | 58 | ||
74 | 59 | ||
75 | static int dtl_enable(struct dtl *dtl) | 60 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING |
61 | struct dtl_ring { | ||
62 | u64 write_index; | ||
63 | struct dtl_entry *write_ptr; | ||
64 | struct dtl_entry *buf; | ||
65 | struct dtl_entry *buf_end; | ||
66 | u8 saved_dtl_mask; | ||
67 | }; | ||
68 | |||
69 | static DEFINE_PER_CPU(struct dtl_ring, dtl_rings); | ||
70 | |||
71 | static atomic_t dtl_count; | ||
72 | |||
73 | /* | ||
74 | * The cpu accounting code controls the DTL ring buffer, and we get | ||
75 | * given entries as they are processed. | ||
76 | */ | ||
77 | static void consume_dtle(struct dtl_entry *dtle, u64 index) | ||
76 | { | 78 | { |
77 | unsigned long addr; | 79 | struct dtl_ring *dtlr = &__get_cpu_var(dtl_rings); |
78 | int ret, hwcpu; | 80 | struct dtl_entry *wp = dtlr->write_ptr; |
81 | struct lppaca *vpa = local_paca->lppaca_ptr; | ||
79 | 82 | ||
80 | /* only allow one reader */ | 83 | if (!wp) |
81 | if (dtl->buf) | 84 | return; |
82 | return -EBUSY; | ||
83 | 85 | ||
84 | /* we need to store the original allocation size for use during read */ | 86 | *wp = *dtle; |
85 | dtl->buf_entries = dtl_buf_entries; | 87 | barrier(); |
86 | 88 | ||
87 | dtl->buf = kmalloc_node(dtl->buf_entries * sizeof(struct dtl_entry), | 89 | /* check for hypervisor ring buffer overflow, ignore this entry if so */ |
88 | GFP_KERNEL, cpu_to_node(dtl->cpu)); | 90 | if (index + N_DISPATCH_LOG < vpa->dtl_idx) |
89 | if (!dtl->buf) { | 91 | return; |
90 | printk(KERN_WARNING "%s: buffer alloc failed for cpu %d\n", | 92 | |
91 | __func__, dtl->cpu); | 93 | ++wp; |
92 | return -ENOMEM; | 94 | if (wp == dtlr->buf_end) |
93 | } | 95 | wp = dtlr->buf; |
96 | dtlr->write_ptr = wp; | ||
97 | |||
98 | /* incrementing write_index makes the new entry visible */ | ||
99 | smp_wmb(); | ||
100 | ++dtlr->write_index; | ||
101 | } | ||
102 | |||
103 | static int dtl_start(struct dtl *dtl) | ||
104 | { | ||
105 | struct dtl_ring *dtlr = &per_cpu(dtl_rings, dtl->cpu); | ||
106 | |||
107 | dtlr->buf = dtl->buf; | ||
108 | dtlr->buf_end = dtl->buf + dtl->buf_entries; | ||
109 | dtlr->write_index = 0; | ||
110 | |||
111 | /* setting write_ptr enables logging into our buffer */ | ||
112 | smp_wmb(); | ||
113 | dtlr->write_ptr = dtl->buf; | ||
114 | |||
115 | /* enable event logging */ | ||
116 | dtlr->saved_dtl_mask = lppaca_of(dtl->cpu).dtl_enable_mask; | ||
117 | lppaca_of(dtl->cpu).dtl_enable_mask |= dtl_event_mask; | ||
118 | |||
119 | dtl_consumer = consume_dtle; | ||
120 | atomic_inc(&dtl_count); | ||
121 | return 0; | ||
122 | } | ||
123 | |||
124 | static void dtl_stop(struct dtl *dtl) | ||
125 | { | ||
126 | struct dtl_ring *dtlr = &per_cpu(dtl_rings, dtl->cpu); | ||
127 | |||
128 | dtlr->write_ptr = NULL; | ||
129 | smp_wmb(); | ||
130 | |||
131 | dtlr->buf = NULL; | ||
132 | |||
133 | /* restore dtl_enable_mask */ | ||
134 | lppaca_of(dtl->cpu).dtl_enable_mask = dtlr->saved_dtl_mask; | ||
135 | |||
136 | if (atomic_dec_and_test(&dtl_count)) | ||
137 | dtl_consumer = NULL; | ||
138 | } | ||
139 | |||
140 | static u64 dtl_current_index(struct dtl *dtl) | ||
141 | { | ||
142 | return per_cpu(dtl_rings, dtl->cpu).write_index; | ||
143 | } | ||
144 | |||
145 | #else /* CONFIG_VIRT_CPU_ACCOUNTING */ | ||
146 | |||
147 | static int dtl_start(struct dtl *dtl) | ||
148 | { | ||
149 | unsigned long addr; | ||
150 | int ret, hwcpu; | ||
94 | 151 | ||
95 | /* Register our dtl buffer with the hypervisor. The HV expects the | 152 | /* Register our dtl buffer with the hypervisor. The HV expects the |
96 | * buffer size to be passed in the second word of the buffer */ | 153 | * buffer size to be passed in the second word of the buffer */ |
@@ -102,34 +159,82 @@ static int dtl_enable(struct dtl *dtl) | |||
102 | if (ret) { | 159 | if (ret) { |
103 | printk(KERN_WARNING "%s: DTL registration for cpu %d (hw %d) " | 160 | printk(KERN_WARNING "%s: DTL registration for cpu %d (hw %d) " |
104 | "failed with %d\n", __func__, dtl->cpu, hwcpu, ret); | 161 | "failed with %d\n", __func__, dtl->cpu, hwcpu, ret); |
105 | kfree(dtl->buf); | ||
106 | return -EIO; | 162 | return -EIO; |
107 | } | 163 | } |
108 | 164 | ||
109 | /* set our initial buffer indices */ | 165 | /* set our initial buffer indices */ |
110 | dtl->last_idx = lppaca[dtl->cpu].dtl_idx = 0; | 166 | lppaca_of(dtl->cpu).dtl_idx = 0; |
111 | 167 | ||
112 | /* ensure that our updates to the lppaca fields have occurred before | 168 | /* ensure that our updates to the lppaca fields have occurred before |
113 | * we actually enable the logging */ | 169 | * we actually enable the logging */ |
114 | smp_wmb(); | 170 | smp_wmb(); |
115 | 171 | ||
116 | /* enable event logging */ | 172 | /* enable event logging */ |
117 | lppaca[dtl->cpu].dtl_enable_mask = dtl_event_mask; | 173 | lppaca_of(dtl->cpu).dtl_enable_mask = dtl_event_mask; |
118 | 174 | ||
119 | return 0; | 175 | return 0; |
120 | } | 176 | } |
121 | 177 | ||
122 | static void dtl_disable(struct dtl *dtl) | 178 | static void dtl_stop(struct dtl *dtl) |
123 | { | 179 | { |
124 | int hwcpu = get_hard_smp_processor_id(dtl->cpu); | 180 | int hwcpu = get_hard_smp_processor_id(dtl->cpu); |
125 | 181 | ||
126 | lppaca[dtl->cpu].dtl_enable_mask = 0x0; | 182 | lppaca_of(dtl->cpu).dtl_enable_mask = 0x0; |
127 | 183 | ||
128 | unregister_dtl(hwcpu, __pa(dtl->buf)); | 184 | unregister_dtl(hwcpu, __pa(dtl->buf)); |
185 | } | ||
186 | |||
187 | static u64 dtl_current_index(struct dtl *dtl) | ||
188 | { | ||
189 | return lppaca_of(dtl->cpu).dtl_idx; | ||
190 | } | ||
191 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING */ | ||
192 | |||
193 | static int dtl_enable(struct dtl *dtl) | ||
194 | { | ||
195 | long int n_entries; | ||
196 | long int rc; | ||
197 | struct dtl_entry *buf = NULL; | ||
129 | 198 | ||
199 | /* only allow one reader */ | ||
200 | if (dtl->buf) | ||
201 | return -EBUSY; | ||
202 | |||
203 | n_entries = dtl_buf_entries; | ||
204 | buf = kmalloc_node(n_entries * sizeof(struct dtl_entry), | ||
205 | GFP_KERNEL, cpu_to_node(dtl->cpu)); | ||
206 | if (!buf) { | ||
207 | printk(KERN_WARNING "%s: buffer alloc failed for cpu %d\n", | ||
208 | __func__, dtl->cpu); | ||
209 | return -ENOMEM; | ||
210 | } | ||
211 | |||
212 | spin_lock(&dtl->lock); | ||
213 | rc = -EBUSY; | ||
214 | if (!dtl->buf) { | ||
215 | /* store the original allocation size for use during read */ | ||
216 | dtl->buf_entries = n_entries; | ||
217 | dtl->buf = buf; | ||
218 | dtl->last_idx = 0; | ||
219 | rc = dtl_start(dtl); | ||
220 | if (rc) | ||
221 | dtl->buf = NULL; | ||
222 | } | ||
223 | spin_unlock(&dtl->lock); | ||
224 | |||
225 | if (rc) | ||
226 | kfree(buf); | ||
227 | return rc; | ||
228 | } | ||
229 | |||
230 | static void dtl_disable(struct dtl *dtl) | ||
231 | { | ||
232 | spin_lock(&dtl->lock); | ||
233 | dtl_stop(dtl); | ||
130 | kfree(dtl->buf); | 234 | kfree(dtl->buf); |
131 | dtl->buf = NULL; | 235 | dtl->buf = NULL; |
132 | dtl->buf_entries = 0; | 236 | dtl->buf_entries = 0; |
237 | spin_unlock(&dtl->lock); | ||
133 | } | 238 | } |
134 | 239 | ||
135 | /* file interface */ | 240 | /* file interface */ |
@@ -157,8 +262,9 @@ static int dtl_file_release(struct inode *inode, struct file *filp) | |||
157 | static ssize_t dtl_file_read(struct file *filp, char __user *buf, size_t len, | 262 | static ssize_t dtl_file_read(struct file *filp, char __user *buf, size_t len, |
158 | loff_t *pos) | 263 | loff_t *pos) |
159 | { | 264 | { |
160 | int rc, cur_idx, last_idx, n_read, n_req, read_size; | 265 | long int rc, n_read, n_req, read_size; |
161 | struct dtl *dtl; | 266 | struct dtl *dtl; |
267 | u64 cur_idx, last_idx, i; | ||
162 | 268 | ||
163 | if ((len % sizeof(struct dtl_entry)) != 0) | 269 | if ((len % sizeof(struct dtl_entry)) != 0) |
164 | return -EINVAL; | 270 | return -EINVAL; |
@@ -171,41 +277,48 @@ static ssize_t dtl_file_read(struct file *filp, char __user *buf, size_t len, | |||
171 | /* actual number of entries read */ | 277 | /* actual number of entries read */ |
172 | n_read = 0; | 278 | n_read = 0; |
173 | 279 | ||
174 | cur_idx = lppaca[dtl->cpu].dtl_idx; | 280 | spin_lock(&dtl->lock); |
281 | |||
282 | cur_idx = dtl_current_index(dtl); | ||
175 | last_idx = dtl->last_idx; | 283 | last_idx = dtl->last_idx; |
176 | 284 | ||
177 | if (cur_idx - last_idx > dtl->buf_entries) { | 285 | if (last_idx + dtl->buf_entries <= cur_idx) |
178 | pr_debug("%s: hv buffer overflow for cpu %d, samples lost\n", | 286 | last_idx = cur_idx - dtl->buf_entries + 1; |
179 | __func__, dtl->cpu); | 287 | |
180 | } | 288 | if (last_idx + n_req > cur_idx) |
289 | n_req = cur_idx - last_idx; | ||
290 | |||
291 | if (n_req > 0) | ||
292 | dtl->last_idx = last_idx + n_req; | ||
293 | |||
294 | spin_unlock(&dtl->lock); | ||
295 | |||
296 | if (n_req <= 0) | ||
297 | return 0; | ||
181 | 298 | ||
182 | cur_idx %= dtl->buf_entries; | 299 | i = last_idx % dtl->buf_entries; |
183 | last_idx %= dtl->buf_entries; | ||
184 | 300 | ||
185 | /* read the tail of the buffer if we've wrapped */ | 301 | /* read the tail of the buffer if we've wrapped */ |
186 | if (last_idx > cur_idx) { | 302 | if (i + n_req > dtl->buf_entries) { |
187 | read_size = min(n_req, dtl->buf_entries - last_idx); | 303 | read_size = dtl->buf_entries - i; |
188 | 304 | ||
189 | rc = copy_to_user(buf, &dtl->buf[last_idx], | 305 | rc = copy_to_user(buf, &dtl->buf[i], |
190 | read_size * sizeof(struct dtl_entry)); | 306 | read_size * sizeof(struct dtl_entry)); |
191 | if (rc) | 307 | if (rc) |
192 | return -EFAULT; | 308 | return -EFAULT; |
193 | 309 | ||
194 | last_idx = 0; | 310 | i = 0; |
195 | n_req -= read_size; | 311 | n_req -= read_size; |
196 | n_read += read_size; | 312 | n_read += read_size; |
197 | buf += read_size * sizeof(struct dtl_entry); | 313 | buf += read_size * sizeof(struct dtl_entry); |
198 | } | 314 | } |
199 | 315 | ||
200 | /* .. and now the head */ | 316 | /* .. and now the head */ |
201 | read_size = min(n_req, cur_idx - last_idx); | 317 | rc = copy_to_user(buf, &dtl->buf[i], n_req * sizeof(struct dtl_entry)); |
202 | rc = copy_to_user(buf, &dtl->buf[last_idx], | ||
203 | read_size * sizeof(struct dtl_entry)); | ||
204 | if (rc) | 318 | if (rc) |
205 | return -EFAULT; | 319 | return -EFAULT; |
206 | 320 | ||
207 | n_read += read_size; | 321 | n_read += n_req; |
208 | dtl->last_idx += n_read; | ||
209 | 322 | ||
210 | return n_read * sizeof(struct dtl_entry); | 323 | return n_read * sizeof(struct dtl_entry); |
211 | } | 324 | } |
@@ -263,6 +376,7 @@ static int dtl_init(void) | |||
263 | /* set up the per-cpu log structures */ | 376 | /* set up the per-cpu log structures */ |
264 | for_each_possible_cpu(i) { | 377 | for_each_possible_cpu(i) { |
265 | struct dtl *dtl = &per_cpu(cpu_dtl, i); | 378 | struct dtl *dtl = &per_cpu(cpu_dtl, i); |
379 | spin_lock_init(&dtl->lock); | ||
266 | dtl->cpu = i; | 380 | dtl->cpu = i; |
267 | 381 | ||
268 | rc = dtl_setup_file(dtl); | 382 | rc = dtl_setup_file(dtl); |