xref: /openbmc/linux/drivers/infiniband/hw/mlx5/cong.c (revision 89e33ea7)
1 /*
2  * Copyright (c) 2013-2017, Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/debugfs.h>
34 
35 #include "mlx5_ib.h"
36 #include "cmd.h"
37 
38 enum mlx5_ib_cong_node_type {
39 	MLX5_IB_RROCE_ECN_RP = 1,
40 	MLX5_IB_RROCE_ECN_NP = 2,
41 };
42 
43 static const char * const mlx5_ib_dbg_cc_name[] = {
44 	"rp_clamp_tgt_rate",
45 	"rp_clamp_tgt_rate_ati",
46 	"rp_time_reset",
47 	"rp_byte_reset",
48 	"rp_threshold",
49 	"rp_ai_rate",
50 	"rp_hai_rate",
51 	"rp_min_dec_fac",
52 	"rp_min_rate",
53 	"rp_rate_to_set_on_first_cnp",
54 	"rp_dce_tcp_g",
55 	"rp_dce_tcp_rtt",
56 	"rp_rate_reduce_monitor_period",
57 	"rp_initial_alpha_value",
58 	"rp_gd",
59 	"np_cnp_dscp",
60 	"np_cnp_prio_mode",
61 	"np_cnp_prio",
62 };
63 
64 #define MLX5_IB_RP_CLAMP_TGT_RATE_ATTR			BIT(1)
65 #define MLX5_IB_RP_CLAMP_TGT_RATE_ATI_ATTR		BIT(2)
66 #define MLX5_IB_RP_TIME_RESET_ATTR			BIT(3)
67 #define MLX5_IB_RP_BYTE_RESET_ATTR			BIT(4)
68 #define MLX5_IB_RP_THRESHOLD_ATTR			BIT(5)
69 #define MLX5_IB_RP_AI_RATE_ATTR				BIT(7)
70 #define MLX5_IB_RP_HAI_RATE_ATTR			BIT(8)
71 #define MLX5_IB_RP_MIN_DEC_FAC_ATTR			BIT(9)
72 #define MLX5_IB_RP_MIN_RATE_ATTR			BIT(10)
73 #define MLX5_IB_RP_RATE_TO_SET_ON_FIRST_CNP_ATTR	BIT(11)
74 #define MLX5_IB_RP_DCE_TCP_G_ATTR			BIT(12)
75 #define MLX5_IB_RP_DCE_TCP_RTT_ATTR			BIT(13)
76 #define MLX5_IB_RP_RATE_REDUCE_MONITOR_PERIOD_ATTR	BIT(14)
77 #define MLX5_IB_RP_INITIAL_ALPHA_VALUE_ATTR		BIT(15)
78 #define MLX5_IB_RP_GD_ATTR				BIT(16)
79 
80 #define MLX5_IB_NP_CNP_DSCP_ATTR			BIT(3)
81 #define MLX5_IB_NP_CNP_PRIO_MODE_ATTR			BIT(4)
82 
83 static enum mlx5_ib_cong_node_type
84 mlx5_ib_param_to_node(enum mlx5_ib_dbg_cc_types param_offset)
85 {
86 	if (param_offset >= MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE &&
87 	    param_offset <= MLX5_IB_DBG_CC_RP_GD)
88 		return MLX5_IB_RROCE_ECN_RP;
89 	else
90 		return MLX5_IB_RROCE_ECN_NP;
91 }
92 
93 static u32 mlx5_get_cc_param_val(void *field, int offset)
94 {
95 	switch (offset) {
96 	case MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE:
97 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
98 				clamp_tgt_rate);
99 	case MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI:
100 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
101 				clamp_tgt_rate_after_time_inc);
102 	case MLX5_IB_DBG_CC_RP_TIME_RESET:
103 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
104 				rpg_time_reset);
105 	case MLX5_IB_DBG_CC_RP_BYTE_RESET:
106 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
107 				rpg_byte_reset);
108 	case MLX5_IB_DBG_CC_RP_THRESHOLD:
109 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
110 				rpg_threshold);
111 	case MLX5_IB_DBG_CC_RP_AI_RATE:
112 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
113 				rpg_ai_rate);
114 	case MLX5_IB_DBG_CC_RP_HAI_RATE:
115 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
116 				rpg_hai_rate);
117 	case MLX5_IB_DBG_CC_RP_MIN_DEC_FAC:
118 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
119 				rpg_min_dec_fac);
120 	case MLX5_IB_DBG_CC_RP_MIN_RATE:
121 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
122 				rpg_min_rate);
123 	case MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP:
124 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
125 				rate_to_set_on_first_cnp);
126 	case MLX5_IB_DBG_CC_RP_DCE_TCP_G:
127 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
128 				dce_tcp_g);
129 	case MLX5_IB_DBG_CC_RP_DCE_TCP_RTT:
130 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
131 				dce_tcp_rtt);
132 	case MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD:
133 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
134 				rate_reduce_monitor_period);
135 	case MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE:
136 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
137 				initial_alpha_value);
138 	case MLX5_IB_DBG_CC_RP_GD:
139 		return MLX5_GET(cong_control_r_roce_ecn_rp, field,
140 				rpg_gd);
141 	case MLX5_IB_DBG_CC_NP_CNP_DSCP:
142 		return MLX5_GET(cong_control_r_roce_ecn_np, field,
143 				cnp_dscp);
144 	case MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE:
145 		return MLX5_GET(cong_control_r_roce_ecn_np, field,
146 				cnp_prio_mode);
147 	case MLX5_IB_DBG_CC_NP_CNP_PRIO:
148 		return MLX5_GET(cong_control_r_roce_ecn_np, field,
149 				cnp_802p_prio);
150 	default:
151 		return 0;
152 	}
153 }
154 
155 static void mlx5_ib_set_cc_param_mask_val(void *field, int offset,
156 					  u32 var, u32 *attr_mask)
157 {
158 	switch (offset) {
159 	case MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE:
160 		*attr_mask |= MLX5_IB_RP_CLAMP_TGT_RATE_ATTR;
161 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
162 			 clamp_tgt_rate, var);
163 		break;
164 	case MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI:
165 		*attr_mask |= MLX5_IB_RP_CLAMP_TGT_RATE_ATI_ATTR;
166 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
167 			 clamp_tgt_rate_after_time_inc, var);
168 		break;
169 	case MLX5_IB_DBG_CC_RP_TIME_RESET:
170 		*attr_mask |= MLX5_IB_RP_TIME_RESET_ATTR;
171 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
172 			 rpg_time_reset, var);
173 		break;
174 	case MLX5_IB_DBG_CC_RP_BYTE_RESET:
175 		*attr_mask |= MLX5_IB_RP_BYTE_RESET_ATTR;
176 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
177 			 rpg_byte_reset, var);
178 		break;
179 	case MLX5_IB_DBG_CC_RP_THRESHOLD:
180 		*attr_mask |= MLX5_IB_RP_THRESHOLD_ATTR;
181 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
182 			 rpg_threshold, var);
183 		break;
184 	case MLX5_IB_DBG_CC_RP_AI_RATE:
185 		*attr_mask |= MLX5_IB_RP_AI_RATE_ATTR;
186 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
187 			 rpg_ai_rate, var);
188 		break;
189 	case MLX5_IB_DBG_CC_RP_HAI_RATE:
190 		*attr_mask |= MLX5_IB_RP_HAI_RATE_ATTR;
191 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
192 			 rpg_hai_rate, var);
193 		break;
194 	case MLX5_IB_DBG_CC_RP_MIN_DEC_FAC:
195 		*attr_mask |= MLX5_IB_RP_MIN_DEC_FAC_ATTR;
196 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
197 			 rpg_min_dec_fac, var);
198 		break;
199 	case MLX5_IB_DBG_CC_RP_MIN_RATE:
200 		*attr_mask |= MLX5_IB_RP_MIN_RATE_ATTR;
201 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
202 			 rpg_min_rate, var);
203 		break;
204 	case MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP:
205 		*attr_mask |= MLX5_IB_RP_RATE_TO_SET_ON_FIRST_CNP_ATTR;
206 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
207 			 rate_to_set_on_first_cnp, var);
208 		break;
209 	case MLX5_IB_DBG_CC_RP_DCE_TCP_G:
210 		*attr_mask |= MLX5_IB_RP_DCE_TCP_G_ATTR;
211 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
212 			 dce_tcp_g, var);
213 		break;
214 	case MLX5_IB_DBG_CC_RP_DCE_TCP_RTT:
215 		*attr_mask |= MLX5_IB_RP_DCE_TCP_RTT_ATTR;
216 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
217 			 dce_tcp_rtt, var);
218 		break;
219 	case MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD:
220 		*attr_mask |= MLX5_IB_RP_RATE_REDUCE_MONITOR_PERIOD_ATTR;
221 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
222 			 rate_reduce_monitor_period, var);
223 		break;
224 	case MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE:
225 		*attr_mask |= MLX5_IB_RP_INITIAL_ALPHA_VALUE_ATTR;
226 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
227 			 initial_alpha_value, var);
228 		break;
229 	case MLX5_IB_DBG_CC_RP_GD:
230 		*attr_mask |= MLX5_IB_RP_GD_ATTR;
231 		MLX5_SET(cong_control_r_roce_ecn_rp, field,
232 			 rpg_gd, var);
233 		break;
234 	case MLX5_IB_DBG_CC_NP_CNP_DSCP:
235 		*attr_mask |= MLX5_IB_NP_CNP_DSCP_ATTR;
236 		MLX5_SET(cong_control_r_roce_ecn_np, field, cnp_dscp, var);
237 		break;
238 	case MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE:
239 		*attr_mask |= MLX5_IB_NP_CNP_PRIO_MODE_ATTR;
240 		MLX5_SET(cong_control_r_roce_ecn_np, field, cnp_prio_mode, var);
241 		break;
242 	case MLX5_IB_DBG_CC_NP_CNP_PRIO:
243 		*attr_mask |= MLX5_IB_NP_CNP_PRIO_MODE_ATTR;
244 		MLX5_SET(cong_control_r_roce_ecn_np, field, cnp_prio_mode, 0);
245 		MLX5_SET(cong_control_r_roce_ecn_np, field, cnp_802p_prio, var);
246 		break;
247 	}
248 }
249 
250 static int mlx5_ib_get_cc_params(struct mlx5_ib_dev *dev, u8 port_num,
251 				 int offset, u32 *var)
252 {
253 	int outlen = MLX5_ST_SZ_BYTES(query_cong_params_out);
254 	void *out;
255 	void *field;
256 	int err;
257 	enum mlx5_ib_cong_node_type node;
258 	struct mlx5_core_dev *mdev;
259 
260 	/* Takes a 1-based port number */
261 	mdev = mlx5_ib_get_native_port_mdev(dev, port_num + 1, NULL);
262 	if (!mdev)
263 		return -ENODEV;
264 
265 	out = kvzalloc(outlen, GFP_KERNEL);
266 	if (!out) {
267 		err = -ENOMEM;
268 		goto alloc_err;
269 	}
270 
271 	node = mlx5_ib_param_to_node(offset);
272 
273 	err = mlx5_cmd_query_cong_params(mdev, node, out, outlen);
274 	if (err)
275 		goto free;
276 
277 	field = MLX5_ADDR_OF(query_cong_params_out, out, congestion_parameters);
278 	*var = mlx5_get_cc_param_val(field, offset);
279 
280 free:
281 	kvfree(out);
282 alloc_err:
283 	mlx5_ib_put_native_port_mdev(dev, port_num + 1);
284 	return err;
285 }
286 
287 static int mlx5_ib_set_cc_params(struct mlx5_ib_dev *dev, u8 port_num,
288 				 int offset, u32 var)
289 {
290 	int inlen = MLX5_ST_SZ_BYTES(modify_cong_params_in);
291 	void *in;
292 	void *field;
293 	enum mlx5_ib_cong_node_type node;
294 	struct mlx5_core_dev *mdev;
295 	u32 attr_mask = 0;
296 	int err;
297 
298 	/* Takes a 1-based port number */
299 	mdev = mlx5_ib_get_native_port_mdev(dev, port_num + 1, NULL);
300 	if (!mdev)
301 		return -ENODEV;
302 
303 	in = kvzalloc(inlen, GFP_KERNEL);
304 	if (!in) {
305 		err = -ENOMEM;
306 		goto alloc_err;
307 	}
308 
309 	MLX5_SET(modify_cong_params_in, in, opcode,
310 		 MLX5_CMD_OP_MODIFY_CONG_PARAMS);
311 
312 	node = mlx5_ib_param_to_node(offset);
313 	MLX5_SET(modify_cong_params_in, in, cong_protocol, node);
314 
315 	field = MLX5_ADDR_OF(modify_cong_params_in, in, congestion_parameters);
316 	mlx5_ib_set_cc_param_mask_val(field, offset, var, &attr_mask);
317 
318 	field = MLX5_ADDR_OF(modify_cong_params_in, in, field_select);
319 	MLX5_SET(field_select_r_roce_rp, field, field_select_r_roce_rp,
320 		 attr_mask);
321 
322 	err = mlx5_cmd_modify_cong_params(mdev, in, inlen);
323 	kvfree(in);
324 alloc_err:
325 	mlx5_ib_put_native_port_mdev(dev, port_num + 1);
326 	return err;
327 }
328 
329 static ssize_t set_param(struct file *filp, const char __user *buf,
330 			 size_t count, loff_t *pos)
331 {
332 	struct mlx5_ib_dbg_param *param = filp->private_data;
333 	int offset = param->offset;
334 	char lbuf[11] = { };
335 	u32 var;
336 	int ret;
337 
338 	if (count > sizeof(lbuf))
339 		return -EINVAL;
340 
341 	if (copy_from_user(lbuf, buf, count))
342 		return -EFAULT;
343 
344 	lbuf[sizeof(lbuf) - 1] = '\0';
345 
346 	if (kstrtou32(lbuf, 0, &var))
347 		return -EINVAL;
348 
349 	ret = mlx5_ib_set_cc_params(param->dev, param->port_num, offset, var);
350 	return ret ? ret : count;
351 }
352 
353 static ssize_t get_param(struct file *filp, char __user *buf, size_t count,
354 			 loff_t *pos)
355 {
356 	struct mlx5_ib_dbg_param *param = filp->private_data;
357 	int offset = param->offset;
358 	u32 var = 0;
359 	int ret;
360 	char lbuf[11];
361 
362 	ret = mlx5_ib_get_cc_params(param->dev, param->port_num, offset, &var);
363 	if (ret)
364 		return ret;
365 
366 	ret = snprintf(lbuf, sizeof(lbuf), "%d\n", var);
367 	if (ret < 0)
368 		return ret;
369 
370 	return simple_read_from_buffer(buf, count, pos, lbuf, ret);
371 }
372 
373 static const struct file_operations dbg_cc_fops = {
374 	.owner	= THIS_MODULE,
375 	.open	= simple_open,
376 	.write	= set_param,
377 	.read	= get_param,
378 };
379 
380 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num)
381 {
382 	if (!mlx5_debugfs_root ||
383 	    !dev->port[port_num].dbg_cc_params ||
384 	    !dev->port[port_num].dbg_cc_params->root)
385 		return;
386 
387 	debugfs_remove_recursive(dev->port[port_num].dbg_cc_params->root);
388 	kfree(dev->port[port_num].dbg_cc_params);
389 	dev->port[port_num].dbg_cc_params = NULL;
390 }
391 
392 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num)
393 {
394 	struct mlx5_ib_dbg_cc_params *dbg_cc_params;
395 	struct mlx5_core_dev *mdev;
396 	int i;
397 
398 	if (!mlx5_debugfs_root)
399 		return;
400 
401 	/* Takes a 1-based port number */
402 	mdev = mlx5_ib_get_native_port_mdev(dev, port_num + 1, NULL);
403 	if (!mdev)
404 		return;
405 
406 	if (!MLX5_CAP_GEN(mdev, cc_query_allowed) ||
407 	    !MLX5_CAP_GEN(mdev, cc_modify_allowed))
408 		goto put_mdev;
409 
410 	dbg_cc_params = kzalloc(sizeof(*dbg_cc_params), GFP_KERNEL);
411 	if (!dbg_cc_params)
412 		goto err;
413 
414 	dev->port[port_num].dbg_cc_params = dbg_cc_params;
415 
416 	dbg_cc_params->root = debugfs_create_dir("cc_params",
417 						 mdev->priv.dbg_root);
418 
419 	for (i = 0; i < MLX5_IB_DBG_CC_MAX; i++) {
420 		dbg_cc_params->params[i].offset = i;
421 		dbg_cc_params->params[i].dev = dev;
422 		dbg_cc_params->params[i].port_num = port_num;
423 		dbg_cc_params->params[i].dentry =
424 			debugfs_create_file(mlx5_ib_dbg_cc_name[i],
425 					    0600, dbg_cc_params->root,
426 					    &dbg_cc_params->params[i],
427 					    &dbg_cc_fops);
428 	}
429 
430 put_mdev:
431 	mlx5_ib_put_native_port_mdev(dev, port_num + 1);
432 	return;
433 
434 err:
435 	mlx5_ib_warn(dev, "cong debugfs failure\n");
436 	mlx5_ib_cleanup_cong_debugfs(dev, port_num);
437 	mlx5_ib_put_native_port_mdev(dev, port_num + 1);
438 
439 	/*
440 	 * We don't want to fail driver if debugfs failed to initialize,
441 	 * so we are not forwarding error to the user.
442 	 */
443 	return;
444 }
445