1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
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
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
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
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
|
From 2236ebb926291622792fe547ed0fe3c516f66c44 Mon Sep 17 00:00:00 2001
From: Raju Rangoju <Raju.Rangoju@amd.com>
Date: Fri, 9 Feb 2024 18:20:58 +0530
Subject: [PATCH 4/5] ntb_perf: aggregate and log the throughput periodically
Add necessary support to consolidate the total aggregated throughput.
And, provide a module param to log the performance every 500ms.
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
---
drivers/ntb/test/ntb_perf.c | 129 ++++++++++++++++++++++++++++--------
1 file changed, 100 insertions(+), 29 deletions(-)
diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c
index e6b01254dcce..1d0b95661b0b 100644
--- a/drivers/ntb/test/ntb_perf.c
+++ b/drivers/ntb/test/ntb_perf.c
@@ -127,6 +127,10 @@ static bool use_dma_2p; /* default to 0 */
module_param(use_dma_2p, bool, 0644);
MODULE_PARM_DESC(use_dma_2p, "Use DMA engine also from other node to measure performance");
+static bool use_500ms_log; /* default to 0 */
+module_param(use_500ms_log, bool, 0644);
+MODULE_PARM_DESC(use_500ms_log, "Logs dmachannel performance every 500ms");
+
/*==============================================================================
* Perf driver data definition
@@ -145,6 +149,7 @@ enum perf_cmd {
};
struct perf_ctx;
+struct perf_dma_callback_data;
struct perf_peer {
struct perf_ctx *perf;
@@ -170,6 +175,11 @@ struct perf_peer {
#define to_peer_service(__work) \
container_of(__work, struct perf_peer, service)
+struct perf_dma_callback_data {
+ struct perf_thread *pthr;
+ int didx;
+};
+
struct perf_thread {
struct perf_ctx *perf;
int tidx;
@@ -187,6 +197,8 @@ struct perf_thread {
int status;
struct work_struct work;
dma_addr_t dma_dst_addr[MAX_DMACH_CNT];
+ u64 dma_desc_complete_count[MAX_DMACH_CNT];
+ struct perf_dma_callback_data perf_dma_data;
};
#define to_thread_work(__work) \
container_of(__work, struct perf_thread, work)
@@ -214,6 +226,11 @@ struct perf_ctx {
u64 *data);
struct dentry *dbgfs_dir;
+
+ /* Overall Statistics */
+ ktime_t duration, kstart;
+ u64 perf_speed;
+ u64 total_data;
};
/*
@@ -247,6 +264,10 @@ static struct dentry *perf_dbgfs_topdir;
static struct workqueue_struct *perf_wq __read_mostly;
+static struct workqueue_struct *my_workqueue;
+static struct delayed_work my_work;
+struct perf_peer *timer_peer;
+u64 tmr_chunk_size = 0;
/*==============================================================================
* NTB cross-link commands execution service
*==============================================================================
@@ -784,8 +805,11 @@ static void perf_disable_service(struct perf_ctx *perf)
static void perf_dma_copy_callback(void *data)
{
- struct perf_thread *pthr = data;
+ struct perf_dma_callback_data *perf_dma_data = data;
+ struct perf_thread *pthr = perf_dma_data->pthr;
+ int didx = perf_dma_data->didx;
+ pthr->dma_desc_complete_count[didx] += tmr_chunk_size;
atomic_dec(&pthr->dma_sync);
wake_up(&pthr->dma_wait);
}
@@ -802,15 +826,18 @@ static int perf_copy_chunk(struct perf_thread *pthr,
void __iomem *dst_vaddr;
dma_addr_t dma_dst_addr, dma_src_addr;
int tidx;
+ struct perf_dma_callback_data *perf_dma_data = &pthr->perf_dma_data;
if (!use_dma) {
memcpy_toio(dst, src, len);
goto ret_check_tsync;
}
+ perf_dma_data->pthr = pthr;
+
for (tidx = 0; tidx < perf->dch_cnt; tidx++) {
dma_dev = pthr->dma_chan[tidx]->device->dev;
-
+ perf_dma_data->didx = tidx;
if (!is_dma_copy_aligned(pthr->dma_chan[tidx]->device, offset_in_page(src),
offset_in_page(dst), len))
return -EIO;
@@ -820,6 +847,7 @@ static int perf_copy_chunk(struct perf_thread *pthr,
dma_dst_addr = pthr->dma_dst_addr[tidx] + (dst_vaddr - vbase);
dma_src_addr = pthr->dma_src_addr[tidx] + (dst_vaddr - vbase);
+
do {
tx = dmaengine_prep_dma_memcpy(pthr->dma_chan[tidx], dma_dst_addr,
dma_src_addr, len, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
@@ -833,7 +861,7 @@ static int perf_copy_chunk(struct perf_thread *pthr,
}
tx->callback = perf_dma_copy_callback;
- tx->callback_param = pthr;
+ tx->callback_param = perf_dma_data;
ret = dma_submit_error(dmaengine_submit(tx));
if (ret)
@@ -878,8 +906,12 @@ static int perf_init_test(struct perf_thread *pthr)
pr_info("ntb_perf: outbuf size reduced to 0x%llx \n", chunk_size);
}
- if (!use_dma)
+ if (!use_dma) {
+ pthr->src[0] = dma_alloc_coherent(&perf->ntb->pdev->dev, perf->test_peer->outbuf_size,
+ &pthr->dma_src_addr[0], GFP_KERNEL);
+ get_random_bytes(pthr->src[0], perf->test_peer->outbuf_size);
return 0;
+ }
dma_cap_zero(dma_mask);
dma_cap_set(DMA_MEMCPY, dma_mask);
@@ -1003,16 +1035,6 @@ static int perf_sync_test(struct perf_thread *pthr)
no_dma_ret:
pthr->duration = ktime_sub(ktime_get(), pthr->duration);
-
- dev_dbg(&perf->ntb->dev, "%d: copied %llu bytes\n",
- pthr->tidx, pthr->copied);
-
- dev_dbg(&perf->ntb->dev, "%d: lasted %llu usecs\n",
- pthr->tidx, ktime_to_us(pthr->duration));
-
- dev_dbg(&perf->ntb->dev, "%d: %llu MBytes/s\n", pthr->tidx,
- div64_u64(pthr->copied, ktime_to_us(pthr->duration)));
-
return 0;
}
@@ -1045,6 +1067,7 @@ static void perf_clear_test(struct perf_thread *pthr)
static void perf_thread_work(struct work_struct *work)
{
struct perf_thread *pthr = to_thread_work(work);
+ struct perf_ctx *perf = pthr->perf;
int ret;
/*
@@ -1054,22 +1077,14 @@ static void perf_thread_work(struct work_struct *work)
* synchronization is performed only if test fininshed
* without an error or interruption.
*/
- ret = perf_init_test(pthr);
- if (ret) {
- pthr->status = ret;
- return;
- }
-
ret = perf_run_test(pthr);
- if (ret) {
+ if (ret)
pthr->status = ret;
- goto err_clear_test;
- }
pthr->status = perf_sync_test(pthr);
-err_clear_test:
- perf_clear_test(pthr);
+ atomic_dec(&perf->tsync);
+ wake_up(&perf->twait);
}
static void perf_terminate_test(struct perf_ctx *perf)
@@ -1085,6 +1100,7 @@ static void perf_terminate_test(struct perf_ctx *perf)
}
}
+
static int perf_set_dch_cnt(struct perf_ctx *perf, u8 dch_cnt)
{
if (dch_cnt == 0 || dch_cnt > MAX_THREADS_CNT)
@@ -1115,11 +1131,29 @@ static int perf_set_tcnt(struct perf_ctx *perf, u8 tcnt)
return 0;
}
+static void perf_print_test_complete_cnt(struct work_struct *work)
+{
+ struct perf_ctx *perf = timer_peer->perf;
+ struct perf_thread *pthr;
+ int tidx, didx;
+
+ for (tidx = 0; tidx < perf->tcnt; tidx++) {
+ pthr = &perf->threads[tidx];
+ for (didx = 0; didx < perf->dch_cnt; didx++) {
+ pr_info("%d:%s: perf = %llu MB/s\n",tidx, dma_chan_name(pthr->dma_chan[didx]), div64_u64(pthr->dma_desc_complete_count[didx], 500000));
+ pthr->dma_desc_complete_count[didx] = 0;
+ }
+ }
+ pr_alert("\n");
+ queue_delayed_work(my_workqueue, &my_work, msecs_to_jiffies(500));
+}
+
static int perf_submit_test(struct perf_peer *peer)
{
struct perf_ctx *perf = peer->perf;
struct perf_thread *pthr;
- int tidx, ret;
+ int tidx, didx, ret;
+ my_workqueue = create_singlethread_workqueue("my_workqueue");
ret = wait_for_completion_interruptible(&peer->init_comp);
if (ret < 0)
@@ -1130,6 +1164,16 @@ static int perf_submit_test(struct perf_peer *peer)
perf->test_peer = peer;
atomic_set(&perf->tsync, perf->tcnt);
+ timer_peer = peer;
+ if(use_500ms_log) {
+ INIT_DELAYED_WORK(&my_work, perf_print_test_complete_cnt);
+ queue_delayed_work(my_workqueue, &my_work, msecs_to_jiffies(500));
+ }
+ tmr_chunk_size = 1ULL << chunk_order;
+
+
+ perf->total_data = (1ULL << total_order);
+ perf->total_data *= (perf->tcnt * perf->dch_cnt);
for (tidx = 0; tidx < MAX_THREADS_CNT; tidx++) {
pthr = &perf->threads[tidx];
@@ -1137,17 +1181,42 @@ static int perf_submit_test(struct perf_peer *peer)
pthr->status = -ENODATA;
pthr->copied = 0;
pthr->duration = ktime_set(0, 0);
+ for (didx = 0; didx < perf->dch_cnt; didx++)
+ pthr->dma_desc_complete_count[didx] = 0;
+ if (tidx < perf->tcnt) {
+ ret = perf_init_test(pthr);
+ if (ret) {
+ pthr->status = ret;
+ return ret;
+ }
+ }
+ }
+
+ perf->kstart = ktime_get();
+ for (tidx = 0; tidx < MAX_THREADS_CNT; tidx++) {
+ pthr = &perf->threads[tidx];
if (tidx < perf->tcnt)
(void)queue_work(perf_wq, &pthr->work);
}
-
ret = wait_event_interruptible(perf->twait,
atomic_read(&perf->tsync) <= 0);
+ perf->duration = ktime_sub(ktime_get(), perf->kstart);
+ perf->perf_speed = div64_u64(perf->total_data, ktime_to_us(perf->duration));
+
+ for (tidx = 0; tidx < MAX_THREADS_CNT; tidx++) {
+ pthr = &perf->threads[tidx];
+ if (tidx < perf->tcnt)
+ perf_clear_test(pthr);
+ }
+
if (ret == -ERESTARTSYS) {
perf_terminate_test(perf);
ret = -EINTR;
}
-
+ if(use_500ms_log) {
+ cancel_delayed_work_sync(&my_work);
+ destroy_workqueue(my_workqueue);
+ }
clear_bit_unlock(0, &perf->busy_flag);
return ret;
@@ -1192,6 +1261,8 @@ static int perf_read_stats(struct perf_ctx *perf, char *buf,
}
}
+ (*pos) += scnprintf(buf + *pos, size - *pos, "\n Total throughput : %llu MBytes/s\n", perf->perf_speed);
+
clear_bit_unlock(0, &perf->busy_flag);
return 0;
@@ -1406,7 +1477,7 @@ static ssize_t perf_dbgfs_read_dch_cnt(struct file *filep, char __user *ubuf,
char buf[8];
ssize_t pos;
- pos = scnprintf(buf, sizeof(buf), "%hhu\n", perf->tcnt);
+ pos = scnprintf(buf, sizeof(buf), "%hhu\n", perf->dch_cnt);
return simple_read_from_buffer(ubuf, size, offp, buf, pos);
}
--
2.34.1
|
