xref: /linux/drivers/net/ethernet/mellanox/mlx4/en_main.c (revision 34dc1baba215b826e454b8d19e4f24adbeb7d00d)
1 /*
2  * Copyright (c) 2007 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 
34 #include <linux/cpumask.h>
35 #include <linux/module.h>
36 #include <linux/delay.h>
37 #include <linux/netdevice.h>
38 #include <linux/slab.h>
39 
40 #include <linux/mlx4/driver.h>
41 #include <linux/mlx4/device.h>
42 #include <linux/mlx4/cmd.h>
43 
44 #include "mlx4_en.h"
45 
46 MODULE_AUTHOR("Liran Liss, Yevgeny Petrilin");
47 MODULE_DESCRIPTION("Mellanox ConnectX HCA Ethernet driver");
48 MODULE_LICENSE("Dual BSD/GPL");
49 MODULE_VERSION(DRV_VERSION);
50 
51 static const char mlx4_en_version[] =
52 	DRV_NAME ": Mellanox ConnectX HCA Ethernet driver v"
53 	DRV_VERSION "\n";
54 
55 #define MLX4_EN_PARM_INT(X, def_val, desc) \
56 	static unsigned int X = def_val;\
57 	module_param(X , uint, 0444); \
58 	MODULE_PARM_DESC(X, desc);
59 
60 
61 /*
62  * Device scope module parameters
63  */
64 
65 /* Enable RSS UDP traffic */
66 MLX4_EN_PARM_INT(udp_rss, 1,
67 		 "Enable RSS for incoming UDP traffic or disabled (0)");
68 
69 /* Priority pausing */
70 MLX4_EN_PARM_INT(pfctx, 0, "Priority based Flow Control policy on TX[7:0]."
71 			   " Per priority bit mask");
72 MLX4_EN_PARM_INT(pfcrx, 0, "Priority based Flow Control policy on RX[7:0]."
73 			   " Per priority bit mask");
74 
75 MLX4_EN_PARM_INT(inline_thold, MAX_INLINE,
76 		 "Threshold for using inline data (range: 17-104, default: 104)");
77 
78 #define MAX_PFC_TX     0xff
79 #define MAX_PFC_RX     0xff
80 
81 void en_print(const char *level, const struct mlx4_en_priv *priv,
82 	      const char *format, ...)
83 {
84 	va_list args;
85 	struct va_format vaf;
86 
87 	va_start(args, format);
88 
89 	vaf.fmt = format;
90 	vaf.va = &args;
91 	if (priv->registered)
92 		printk("%s%s: %s: %pV",
93 		       level, DRV_NAME, priv->dev->name, &vaf);
94 	else
95 		printk("%s%s: %s: Port %d: %pV",
96 		       level, DRV_NAME, dev_name(&priv->mdev->pdev->dev),
97 		       priv->port, &vaf);
98 	va_end(args);
99 }
100 
101 void mlx4_en_update_loopback_state(struct net_device *dev,
102 				   netdev_features_t features)
103 {
104 	struct mlx4_en_priv *priv = netdev_priv(dev);
105 
106 	if (features & NETIF_F_LOOPBACK)
107 		priv->ctrl_flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
108 	else
109 		priv->ctrl_flags &= cpu_to_be32(~MLX4_WQE_CTRL_FORCE_LOOPBACK);
110 
111 	priv->flags &= ~(MLX4_EN_FLAG_RX_FILTER_NEEDED|
112 			MLX4_EN_FLAG_ENABLE_HW_LOOPBACK);
113 
114 	/* Drop the packet if SRIOV is not enabled
115 	 * and not performing the selftest or flb disabled
116 	 */
117 	if (mlx4_is_mfunc(priv->mdev->dev) &&
118 	    !(features & NETIF_F_LOOPBACK) && !priv->validate_loopback)
119 		priv->flags |= MLX4_EN_FLAG_RX_FILTER_NEEDED;
120 
121 	/* Set dmac in Tx WQE if we are in SRIOV mode or if loopback selftest
122 	 * is requested
123 	 */
124 	if (mlx4_is_mfunc(priv->mdev->dev) || priv->validate_loopback)
125 		priv->flags |= MLX4_EN_FLAG_ENABLE_HW_LOOPBACK;
126 
127 	mutex_lock(&priv->mdev->state_lock);
128 	if ((priv->mdev->dev->caps.flags2 &
129 	     MLX4_DEV_CAP_FLAG2_UPDATE_QP_SRC_CHECK_LB) &&
130 	    priv->rss_map.indir_qp && priv->rss_map.indir_qp->qpn) {
131 		int i;
132 		int err = 0;
133 		int loopback = !!(features & NETIF_F_LOOPBACK);
134 
135 		for (i = 0; i < priv->rx_ring_num; i++) {
136 			int ret;
137 
138 			ret = mlx4_en_change_mcast_lb(priv,
139 						      &priv->rss_map.qps[i],
140 						      loopback);
141 			if (!err)
142 				err = ret;
143 		}
144 		if (err)
145 			mlx4_warn(priv->mdev, "failed to change mcast loopback\n");
146 	}
147 	mutex_unlock(&priv->mdev->state_lock);
148 }
149 
150 static void mlx4_en_get_profile(struct mlx4_en_dev *mdev)
151 {
152 	struct mlx4_en_profile *params = &mdev->profile;
153 	int i;
154 
155 	params->udp_rss = udp_rss;
156 	params->max_num_tx_rings_p_up = mlx4_low_memory_profile() ?
157 		MLX4_EN_MIN_TX_RING_P_UP :
158 		min_t(int, num_online_cpus(), MLX4_EN_MAX_TX_RING_P_UP);
159 
160 	if (params->udp_rss && !(mdev->dev->caps.flags
161 					& MLX4_DEV_CAP_FLAG_UDP_RSS)) {
162 		mlx4_warn(mdev, "UDP RSS is not supported on this device\n");
163 		params->udp_rss = 0;
164 	}
165 	for (i = 1; i <= MLX4_MAX_PORTS; i++) {
166 		params->prof[i].rx_pause = !(pfcrx || pfctx);
167 		params->prof[i].rx_ppp = pfcrx;
168 		params->prof[i].tx_pause = !(pfcrx || pfctx);
169 		params->prof[i].tx_ppp = pfctx;
170 		if (mlx4_low_memory_profile()) {
171 			params->prof[i].tx_ring_size = MLX4_EN_MIN_TX_SIZE;
172 			params->prof[i].rx_ring_size = MLX4_EN_MIN_RX_SIZE;
173 		} else {
174 			params->prof[i].tx_ring_size = MLX4_EN_DEF_TX_RING_SIZE;
175 			params->prof[i].rx_ring_size = MLX4_EN_DEF_RX_RING_SIZE;
176 		}
177 		params->prof[i].num_up = MLX4_EN_NUM_UP_LOW;
178 		params->prof[i].num_tx_rings_p_up = params->max_num_tx_rings_p_up;
179 		params->prof[i].tx_ring_num[TX] = params->max_num_tx_rings_p_up *
180 			params->prof[i].num_up;
181 		params->prof[i].rss_rings = 0;
182 		params->prof[i].inline_thold = inline_thold;
183 	}
184 }
185 
186 static int mlx4_en_event(struct notifier_block *this, unsigned long event,
187 			 void *param)
188 {
189 	struct mlx4_en_dev *mdev =
190 		container_of(this, struct mlx4_en_dev, mlx_nb);
191 	struct mlx4_dev *dev = mdev->dev;
192 	struct mlx4_en_priv *priv;
193 	int port;
194 
195 	switch (event) {
196 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
197 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
198 	case MLX4_DEV_EVENT_SLAVE_INIT:
199 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
200 		break;
201 	default:
202 		port = *(int *)param;
203 		break;
204 	}
205 
206 	switch (event) {
207 	case MLX4_DEV_EVENT_PORT_UP:
208 	case MLX4_DEV_EVENT_PORT_DOWN:
209 		if (!mdev->pndev[port])
210 			return NOTIFY_DONE;
211 		priv = netdev_priv(mdev->pndev[port]);
212 		/* To prevent races, we poll the link state in a separate
213 		  task rather than changing it here */
214 		priv->link_state = event;
215 		queue_work(mdev->workqueue, &priv->linkstate_task);
216 		break;
217 
218 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
219 		mlx4_err(mdev, "Internal error detected, restarting device\n");
220 		break;
221 
222 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
223 	case MLX4_DEV_EVENT_SLAVE_INIT:
224 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
225 		break;
226 	default:
227 		if (port < 1 || port > dev->caps.num_ports ||
228 		    !mdev->pndev[port])
229 			return NOTIFY_DONE;
230 		mlx4_warn(mdev, "Unhandled event %d for port %d\n", (int)event,
231 			  port);
232 	}
233 
234 	return NOTIFY_DONE;
235 }
236 
237 static void mlx4_en_remove(struct auxiliary_device *adev)
238 {
239 	struct mlx4_adev *madev = container_of(adev, struct mlx4_adev, adev);
240 	struct mlx4_dev *dev = madev->mdev;
241 	struct mlx4_en_dev *mdev = auxiliary_get_drvdata(adev);
242 	int i;
243 
244 	mlx4_unregister_event_notifier(dev, &mdev->mlx_nb);
245 
246 	mutex_lock(&mdev->state_lock);
247 	mdev->device_up = false;
248 	mutex_unlock(&mdev->state_lock);
249 
250 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
251 		if (mdev->pndev[i])
252 			mlx4_en_destroy_netdev(mdev->pndev[i]);
253 
254 	destroy_workqueue(mdev->workqueue);
255 	(void) mlx4_mr_free(dev, &mdev->mr);
256 	iounmap(mdev->uar_map);
257 	mlx4_uar_free(dev, &mdev->priv_uar);
258 	mlx4_pd_free(dev, mdev->priv_pdn);
259 	if (mdev->netdev_nb.notifier_call)
260 		unregister_netdevice_notifier(&mdev->netdev_nb);
261 	kfree(mdev);
262 }
263 
264 static int mlx4_en_probe(struct auxiliary_device *adev,
265 			 const struct auxiliary_device_id *id)
266 {
267 	struct mlx4_adev *madev = container_of(adev, struct mlx4_adev, adev);
268 	struct mlx4_dev *dev = madev->mdev;
269 	struct mlx4_en_dev *mdev;
270 	int err, i;
271 
272 	printk_once(KERN_INFO "%s", mlx4_en_version);
273 
274 	mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
275 	if (!mdev) {
276 		err = -ENOMEM;
277 		goto err_free_res;
278 	}
279 
280 	err = mlx4_pd_alloc(dev, &mdev->priv_pdn);
281 	if (err)
282 		goto err_free_dev;
283 
284 	err = mlx4_uar_alloc(dev, &mdev->priv_uar);
285 	if (err)
286 		goto err_pd;
287 
288 	mdev->uar_map = ioremap((phys_addr_t) mdev->priv_uar.pfn << PAGE_SHIFT,
289 				PAGE_SIZE);
290 	if (!mdev->uar_map) {
291 		err = -ENOMEM;
292 		goto err_uar;
293 	}
294 	spin_lock_init(&mdev->uar_lock);
295 
296 	mdev->dev = dev;
297 	mdev->dma_device = &dev->persist->pdev->dev;
298 	mdev->pdev = dev->persist->pdev;
299 	mdev->device_up = false;
300 
301 	mdev->LSO_support = !!(dev->caps.flags & (1 << 15));
302 	if (!mdev->LSO_support)
303 		mlx4_warn(mdev, "LSO not supported, please upgrade to later FW version to enable LSO\n");
304 
305 	err = mlx4_mr_alloc(mdev->dev, mdev->priv_pdn, 0, ~0ull,
306 			    MLX4_PERM_LOCAL_WRITE | MLX4_PERM_LOCAL_READ, 0, 0,
307 			    &mdev->mr);
308 	if (err) {
309 		mlx4_err(mdev, "Failed allocating memory region\n");
310 		goto err_map;
311 	}
312 	err = mlx4_mr_enable(mdev->dev, &mdev->mr);
313 	if (err) {
314 		mlx4_err(mdev, "Failed enabling memory region\n");
315 		goto err_mr;
316 	}
317 
318 	/* Build device profile according to supplied module parameters */
319 	mlx4_en_get_profile(mdev);
320 
321 	/* Configure which ports to start according to module parameters */
322 	mdev->port_cnt = 0;
323 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
324 		mdev->port_cnt++;
325 
326 	/* Set default number of RX rings*/
327 	mlx4_en_set_num_rx_rings(mdev);
328 
329 	/* Create our own workqueue for reset/multicast tasks
330 	 * Note: we cannot use the shared workqueue because of deadlocks caused
331 	 *       by the rtnl lock */
332 	mdev->workqueue = create_singlethread_workqueue("mlx4_en");
333 	if (!mdev->workqueue) {
334 		err = -ENOMEM;
335 		goto err_mr;
336 	}
337 
338 	/* At this stage all non-port specific tasks are complete:
339 	 * mark the card state as up */
340 	mutex_init(&mdev->state_lock);
341 	mdev->device_up = true;
342 
343 	/* register mlx4 core notifier */
344 	mdev->mlx_nb.notifier_call = mlx4_en_event;
345 	err = mlx4_register_event_notifier(dev, &mdev->mlx_nb);
346 	WARN(err, "failed to register mlx4 event notifier (%d)", err);
347 
348 	/* Setup ports */
349 
350 	/* Create a netdev for each port */
351 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
352 		mlx4_info(mdev, "Activating port:%d\n", i);
353 		if (mlx4_en_init_netdev(mdev, i, &mdev->profile.prof[i]))
354 			mdev->pndev[i] = NULL;
355 	}
356 
357 	/* register netdev notifier */
358 	mdev->netdev_nb.notifier_call = mlx4_en_netdev_event;
359 	if (register_netdevice_notifier(&mdev->netdev_nb)) {
360 		mdev->netdev_nb.notifier_call = NULL;
361 		mlx4_err(mdev, "Failed to create netdev notifier\n");
362 	}
363 
364 	auxiliary_set_drvdata(adev, mdev);
365 	return 0;
366 
367 err_mr:
368 	(void) mlx4_mr_free(dev, &mdev->mr);
369 err_map:
370 	if (mdev->uar_map)
371 		iounmap(mdev->uar_map);
372 err_uar:
373 	mlx4_uar_free(dev, &mdev->priv_uar);
374 err_pd:
375 	mlx4_pd_free(dev, mdev->priv_pdn);
376 err_free_dev:
377 	kfree(mdev);
378 err_free_res:
379 	return err;
380 }
381 
382 static const struct auxiliary_device_id mlx4_en_id_table[] = {
383 	{ .name = MLX4_ADEV_NAME ".eth" },
384 	{},
385 };
386 
387 MODULE_DEVICE_TABLE(auxiliary, mlx4_en_id_table);
388 
389 static struct mlx4_adrv mlx4_en_adrv = {
390 	.adrv = {
391 		.name	= "eth",
392 		.probe	= mlx4_en_probe,
393 		.remove	= mlx4_en_remove,
394 		.id_table = mlx4_en_id_table,
395 	},
396 	.protocol	= MLX4_PROT_ETH,
397 };
398 
399 static void mlx4_en_verify_params(void)
400 {
401 	if (pfctx > MAX_PFC_TX) {
402 		pr_warn("mlx4_en: WARNING: illegal module parameter pfctx 0x%x - should be in range 0-0x%x, will be changed to default (0)\n",
403 			pfctx, MAX_PFC_TX);
404 		pfctx = 0;
405 	}
406 
407 	if (pfcrx > MAX_PFC_RX) {
408 		pr_warn("mlx4_en: WARNING: illegal module parameter pfcrx 0x%x - should be in range 0-0x%x, will be changed to default (0)\n",
409 			pfcrx, MAX_PFC_RX);
410 		pfcrx = 0;
411 	}
412 
413 	if (inline_thold < MIN_PKT_LEN || inline_thold > MAX_INLINE) {
414 		pr_warn("mlx4_en: WARNING: illegal module parameter inline_thold %d - should be in range %d-%d, will be changed to default (%d)\n",
415 			inline_thold, MIN_PKT_LEN, MAX_INLINE, MAX_INLINE);
416 		inline_thold = MAX_INLINE;
417 	}
418 }
419 
420 static int __init mlx4_en_init(void)
421 {
422 	mlx4_en_verify_params();
423 	mlx4_en_init_ptys2ethtool_map();
424 
425 	return mlx4_register_auxiliary_driver(&mlx4_en_adrv);
426 }
427 
428 static void __exit mlx4_en_cleanup(void)
429 {
430 	mlx4_unregister_auxiliary_driver(&mlx4_en_adrv);
431 }
432 
433 module_init(mlx4_en_init);
434 module_exit(mlx4_en_cleanup);
435 
436