diff options
-rw-r--r-- | arch/powerpc/include/asm/lppaca.h | 8 | ||||
-rw-r--r-- | arch/powerpc/kernel/time.c | 6 | ||||
-rw-r--r-- | arch/powerpc/platforms/pseries/dtl.c | 206 |
3 files changed, 179 insertions, 41 deletions
diff --git a/arch/powerpc/include/asm/lppaca.h b/arch/powerpc/include/asm/lppaca.h index cfb85ec8575..7f5e0fefebb 100644 --- a/arch/powerpc/include/asm/lppaca.h +++ b/arch/powerpc/include/asm/lppaca.h | |||
@@ -191,6 +191,14 @@ struct dtl_entry { | |||
191 | #define DISPATCH_LOG_BYTES 4096 /* bytes per cpu */ | 191 | #define DISPATCH_LOG_BYTES 4096 /* bytes per cpu */ |
192 | #define N_DISPATCH_LOG (DISPATCH_LOG_BYTES / sizeof(struct dtl_entry)) | 192 | #define N_DISPATCH_LOG (DISPATCH_LOG_BYTES / sizeof(struct dtl_entry)) |
193 | 193 | ||
194 | /* | ||
195 | * When CONFIG_VIRT_CPU_ACCOUNTING = y, the cpu accounting code controls | ||
196 | * reading from the dispatch trace log. If other code wants to consume | ||
197 | * DTL entries, it can set this pointer to a function that will get | ||
198 | * called once for each DTL entry that gets processed. | ||
199 | */ | ||
200 | extern void (*dtl_consumer)(struct dtl_entry *entry, u64 index); | ||
201 | |||
194 | #endif /* CONFIG_PPC_BOOK3S */ | 202 | #endif /* CONFIG_PPC_BOOK3S */ |
195 | #endif /* __KERNEL__ */ | 203 | #endif /* __KERNEL__ */ |
196 | #endif /* _ASM_POWERPC_LPPACA_H */ | 204 | #endif /* _ASM_POWERPC_LPPACA_H */ |
diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c index fca20643c36..bcb738b9ff8 100644 --- a/arch/powerpc/kernel/time.c +++ b/arch/powerpc/kernel/time.c | |||
@@ -183,6 +183,8 @@ DEFINE_PER_CPU(unsigned long, cputime_scaled_last_delta); | |||
183 | 183 | ||
184 | cputime_t cputime_one_jiffy; | 184 | cputime_t cputime_one_jiffy; |
185 | 185 | ||
186 | void (*dtl_consumer)(struct dtl_entry *, u64); | ||
187 | |||
186 | static void calc_cputime_factors(void) | 188 | static void calc_cputime_factors(void) |
187 | { | 189 | { |
188 | struct div_result res; | 190 | struct div_result res; |
@@ -218,7 +220,7 @@ static u64 read_spurr(u64 tb) | |||
218 | */ | 220 | */ |
219 | static u64 scan_dispatch_log(u64 stop_tb) | 221 | static u64 scan_dispatch_log(u64 stop_tb) |
220 | { | 222 | { |
221 | unsigned long i = local_paca->dtl_ridx; | 223 | u64 i = local_paca->dtl_ridx; |
222 | struct dtl_entry *dtl = local_paca->dtl_curr; | 224 | struct dtl_entry *dtl = local_paca->dtl_curr; |
223 | struct dtl_entry *dtl_end = local_paca->dispatch_log_end; | 225 | struct dtl_entry *dtl_end = local_paca->dispatch_log_end; |
224 | struct lppaca *vpa = local_paca->lppaca_ptr; | 226 | struct lppaca *vpa = local_paca->lppaca_ptr; |
@@ -229,6 +231,8 @@ static u64 scan_dispatch_log(u64 stop_tb) | |||
229 | if (i == vpa->dtl_idx) | 231 | if (i == vpa->dtl_idx) |
230 | return 0; | 232 | return 0; |
231 | while (i < vpa->dtl_idx) { | 233 | while (i < vpa->dtl_idx) { |
234 | if (dtl_consumer) | ||
235 | dtl_consumer(dtl, i); | ||
232 | dtb = dtl->timebase; | 236 | dtb = dtl->timebase; |
233 | tb_delta = dtl->enqueue_to_dispatch_time + | 237 | tb_delta = dtl->enqueue_to_dispatch_time + |
234 | dtl->ready_to_enqueue_time; | 238 | dtl->ready_to_enqueue_time; |
diff --git a/arch/powerpc/platforms/pseries/dtl.c b/arch/powerpc/platforms/pseries/dtl.c index 0357655db49..c371bc06434 100644 --- a/arch/powerpc/platforms/pseries/dtl.c +++ b/arch/powerpc/platforms/pseries/dtl.c | |||
@@ -23,6 +23,7 @@ | |||
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> |
@@ -37,6 +38,7 @@ struct dtl { | |||
37 | int cpu; | 38 | int cpu; |
38 | int buf_entries; | 39 | int buf_entries; |
39 | u64 last_idx; | 40 | u64 last_idx; |
41 | spinlock_t lock; | ||
40 | }; | 42 | }; |
41 | static DEFINE_PER_CPU(struct dtl, cpu_dtl); | 43 | static DEFINE_PER_CPU(struct dtl, cpu_dtl); |
42 | 44 | ||
@@ -55,25 +57,97 @@ static u8 dtl_event_mask = 0x7; | |||
55 | static int dtl_buf_entries = (16 * 85); | 57 | static int dtl_buf_entries = (16 * 85); |
56 | 58 | ||
57 | 59 | ||
58 | 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) | ||
59 | { | 78 | { |
60 | unsigned long addr; | 79 | struct dtl_ring *dtlr = &__get_cpu_var(dtl_rings); |
61 | int ret, hwcpu; | 80 | struct dtl_entry *wp = dtlr->write_ptr; |
81 | struct lppaca *vpa = local_paca->lppaca_ptr; | ||
62 | 82 | ||
63 | /* only allow one reader */ | 83 | if (!wp) |
64 | if (dtl->buf) | 84 | return; |
65 | return -EBUSY; | ||
66 | 85 | ||
67 | /* we need to store the original allocation size for use during read */ | 86 | *wp = *dtle; |
68 | dtl->buf_entries = dtl_buf_entries; | 87 | barrier(); |
69 | 88 | ||
70 | dtl->buf = kmalloc_node(dtl->buf_entries * sizeof(struct dtl_entry), | 89 | /* check for hypervisor ring buffer overflow, ignore this entry if so */ |
71 | GFP_KERNEL, cpu_to_node(dtl->cpu)); | 90 | if (index + N_DISPATCH_LOG < vpa->dtl_idx) |
72 | if (!dtl->buf) { | 91 | return; |
73 | printk(KERN_WARNING "%s: buffer alloc failed for cpu %d\n", | 92 | |
74 | __func__, dtl->cpu); | 93 | ++wp; |
75 | return -ENOMEM; | 94 | if (wp == dtlr->buf_end) |
76 | } | 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; | ||
77 | 151 | ||
78 | /* Register our dtl buffer with the hypervisor. The HV expects the | 152 | /* Register our dtl buffer with the hypervisor. The HV expects the |
79 | * 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 */ |
@@ -85,12 +159,11 @@ static int dtl_enable(struct dtl *dtl) | |||
85 | if (ret) { | 159 | if (ret) { |
86 | printk(KERN_WARNING "%s: DTL registration for cpu %d (hw %d) " | 160 | printk(KERN_WARNING "%s: DTL registration for cpu %d (hw %d) " |
87 | "failed with %d\n", __func__, dtl->cpu, hwcpu, ret); | 161 | "failed with %d\n", __func__, dtl->cpu, hwcpu, ret); |
88 | kfree(dtl->buf); | ||
89 | return -EIO; | 162 | return -EIO; |
90 | } | 163 | } |
91 | 164 | ||
92 | /* set our initial buffer indices */ | 165 | /* set our initial buffer indices */ |
93 | dtl->last_idx = lppaca_of(dtl->cpu).dtl_idx = 0; | 166 | lppaca_of(dtl->cpu).dtl_idx = 0; |
94 | 167 | ||
95 | /* ensure that our updates to the lppaca fields have occurred before | 168 | /* ensure that our updates to the lppaca fields have occurred before |
96 | * we actually enable the logging */ | 169 | * we actually enable the logging */ |
@@ -102,17 +175,66 @@ static int dtl_enable(struct dtl *dtl) | |||
102 | return 0; | 175 | return 0; |
103 | } | 176 | } |
104 | 177 | ||
105 | static void dtl_disable(struct dtl *dtl) | 178 | static void dtl_stop(struct dtl *dtl) |
106 | { | 179 | { |
107 | int hwcpu = get_hard_smp_processor_id(dtl->cpu); | 180 | int hwcpu = get_hard_smp_processor_id(dtl->cpu); |
108 | 181 | ||
109 | lppaca_of(dtl->cpu).dtl_enable_mask = 0x0; | 182 | lppaca_of(dtl->cpu).dtl_enable_mask = 0x0; |
110 | 183 | ||
111 | 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 */ | ||
112 | 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; | ||
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); | ||
113 | kfree(dtl->buf); | 234 | kfree(dtl->buf); |
114 | dtl->buf = NULL; | 235 | dtl->buf = NULL; |
115 | dtl->buf_entries = 0; | 236 | dtl->buf_entries = 0; |
237 | spin_unlock(&dtl->lock); | ||
116 | } | 238 | } |
117 | 239 | ||
118 | /* file interface */ | 240 | /* file interface */ |
@@ -140,8 +262,9 @@ static int dtl_file_release(struct inode *inode, struct file *filp) | |||
140 | 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, |
141 | loff_t *pos) | 263 | loff_t *pos) |
142 | { | 264 | { |
143 | int rc, cur_idx, last_idx, n_read, n_req, read_size; | 265 | long int rc, n_read, n_req, read_size; |
144 | struct dtl *dtl; | 266 | struct dtl *dtl; |
267 | u64 cur_idx, last_idx, i; | ||
145 | 268 | ||
146 | if ((len % sizeof(struct dtl_entry)) != 0) | 269 | if ((len % sizeof(struct dtl_entry)) != 0) |
147 | return -EINVAL; | 270 | return -EINVAL; |
@@ -154,41 +277,48 @@ static ssize_t dtl_file_read(struct file *filp, char __user *buf, size_t len, | |||
154 | /* actual number of entries read */ | 277 | /* actual number of entries read */ |
155 | n_read = 0; | 278 | n_read = 0; |
156 | 279 | ||
157 | cur_idx = lppaca_of(dtl->cpu).dtl_idx; | 280 | spin_lock(&dtl->lock); |
281 | |||
282 | cur_idx = dtl_current_index(dtl); | ||
158 | last_idx = dtl->last_idx; | 283 | last_idx = dtl->last_idx; |
159 | 284 | ||
160 | if (cur_idx - last_idx > dtl->buf_entries) { | 285 | if (last_idx + dtl->buf_entries <= cur_idx) |
161 | pr_debug("%s: hv buffer overflow for cpu %d, samples lost\n", | 286 | last_idx = cur_idx - dtl->buf_entries + 1; |
162 | __func__, dtl->cpu); | 287 | |
163 | } | 288 | if (last_idx + n_req > cur_idx) |
289 | n_req = cur_idx - last_idx; | ||
164 | 290 | ||
165 | cur_idx %= dtl->buf_entries; | 291 | if (n_req > 0) |
166 | last_idx %= dtl->buf_entries; | 292 | dtl->last_idx = last_idx + n_req; |
293 | |||
294 | spin_unlock(&dtl->lock); | ||
295 | |||
296 | if (n_req <= 0) | ||
297 | return 0; | ||
298 | |||
299 | i = last_idx % dtl->buf_entries; | ||
167 | 300 | ||
168 | /* read the tail of the buffer if we've wrapped */ | 301 | /* read the tail of the buffer if we've wrapped */ |
169 | if (last_idx > cur_idx) { | 302 | if (i + n_req > dtl->buf_entries) { |
170 | read_size = min(n_req, dtl->buf_entries - last_idx); | 303 | read_size = dtl->buf_entries - i; |
171 | 304 | ||
172 | rc = copy_to_user(buf, &dtl->buf[last_idx], | 305 | rc = copy_to_user(buf, &dtl->buf[i], |
173 | read_size * sizeof(struct dtl_entry)); | 306 | read_size * sizeof(struct dtl_entry)); |
174 | if (rc) | 307 | if (rc) |
175 | return -EFAULT; | 308 | return -EFAULT; |
176 | 309 | ||
177 | last_idx = 0; | 310 | i = 0; |
178 | n_req -= read_size; | 311 | n_req -= read_size; |
179 | n_read += read_size; | 312 | n_read += read_size; |
180 | buf += read_size * sizeof(struct dtl_entry); | 313 | buf += read_size * sizeof(struct dtl_entry); |
181 | } | 314 | } |
182 | 315 | ||
183 | /* .. and now the head */ | 316 | /* .. and now the head */ |
184 | read_size = min(n_req, cur_idx - last_idx); | 317 | rc = copy_to_user(buf, &dtl->buf[i], n_req * sizeof(struct dtl_entry)); |
185 | rc = copy_to_user(buf, &dtl->buf[last_idx], | ||
186 | read_size * sizeof(struct dtl_entry)); | ||
187 | if (rc) | 318 | if (rc) |
188 | return -EFAULT; | 319 | return -EFAULT; |
189 | 320 | ||
190 | n_read += read_size; | 321 | n_read += n_req; |
191 | dtl->last_idx += n_read; | ||
192 | 322 | ||
193 | return n_read * sizeof(struct dtl_entry); | 323 | return n_read * sizeof(struct dtl_entry); |
194 | } | 324 | } |
@@ -220,11 +350,6 @@ static int dtl_init(void) | |||
220 | struct dentry *event_mask_file, *buf_entries_file; | 350 | struct dentry *event_mask_file, *buf_entries_file; |
221 | int rc, i; | 351 | int rc, i; |
222 | 352 | ||
223 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | ||
224 | /* disable this for now */ | ||
225 | return -ENODEV; | ||
226 | #endif | ||
227 | |||
228 | if (!firmware_has_feature(FW_FEATURE_SPLPAR)) | 353 | if (!firmware_has_feature(FW_FEATURE_SPLPAR)) |
229 | return -ENODEV; | 354 | return -ENODEV; |
230 | 355 | ||
@@ -251,6 +376,7 @@ static int dtl_init(void) | |||
251 | /* set up the per-cpu log structures */ | 376 | /* set up the per-cpu log structures */ |
252 | for_each_possible_cpu(i) { | 377 | for_each_possible_cpu(i) { |
253 | struct dtl *dtl = &per_cpu(cpu_dtl, i); | 378 | struct dtl *dtl = &per_cpu(cpu_dtl, i); |
379 | spin_lock_init(&dtl->lock); | ||
254 | dtl->cpu = i; | 380 | dtl->cpu = i; |
255 | 381 | ||
256 | rc = dtl_setup_file(dtl); | 382 | rc = dtl_setup_file(dtl); |