1 /*
2  * Copyright (c) 2014, Mellanox Technologies inc.  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/pci.h>
34 #include <linux/mlx5/driver.h>
35 #include <linux/mlx5/vport.h>
36 #include "mlx5_core.h"
37 #include "mlx5_irq.h"
38 #include "eswitch.h"
39 
sriov_restore_guids(struct mlx5_core_dev * dev,int vf)40 static int sriov_restore_guids(struct mlx5_core_dev *dev, int vf)
41 {
42 	struct mlx5_core_sriov *sriov = &dev->priv.sriov;
43 	struct mlx5_hca_vport_context *in;
44 	int err = 0;
45 
46 	/* Restore sriov guid and policy settings */
47 	if (sriov->vfs_ctx[vf].node_guid ||
48 	    sriov->vfs_ctx[vf].port_guid ||
49 	    sriov->vfs_ctx[vf].policy != MLX5_POLICY_INVALID) {
50 		in = kzalloc(sizeof(*in), GFP_KERNEL);
51 		if (!in)
52 			return -ENOMEM;
53 
54 		in->node_guid = sriov->vfs_ctx[vf].node_guid;
55 		in->port_guid = sriov->vfs_ctx[vf].port_guid;
56 		in->policy = sriov->vfs_ctx[vf].policy;
57 		in->field_select =
58 			!!(in->port_guid) * MLX5_HCA_VPORT_SEL_PORT_GUID |
59 			!!(in->node_guid) * MLX5_HCA_VPORT_SEL_NODE_GUID |
60 			!!(in->policy) * MLX5_HCA_VPORT_SEL_STATE_POLICY;
61 
62 		err = mlx5_core_modify_hca_vport_context(dev, 1, 1, vf + 1, in);
63 		if (err)
64 			mlx5_core_warn(dev, "modify vport context failed, unable to restore VF %d settings\n", vf);
65 
66 		kfree(in);
67 	}
68 
69 	return err;
70 }
71 
mlx5_device_enable_sriov(struct mlx5_core_dev * dev,int num_vfs)72 static int mlx5_device_enable_sriov(struct mlx5_core_dev *dev, int num_vfs)
73 {
74 	struct mlx5_core_sriov *sriov = &dev->priv.sriov;
75 	int err, vf, num_msix_count;
76 
77 	if (!MLX5_ESWITCH_MANAGER(dev))
78 		goto enable_vfs_hca;
79 
80 	err = mlx5_eswitch_enable(dev->priv.eswitch, num_vfs);
81 	if (err) {
82 		mlx5_core_warn(dev,
83 			       "failed to enable eswitch SRIOV (%d)\n", err);
84 		return err;
85 	}
86 
87 enable_vfs_hca:
88 	num_msix_count = mlx5_get_default_msix_vec_count(dev, num_vfs);
89 	for (vf = 0; vf < num_vfs; vf++) {
90 		/* Notify the VF before its enablement to let it set
91 		 * some stuff.
92 		 */
93 		blocking_notifier_call_chain(&sriov->vfs_ctx[vf].notifier,
94 					     MLX5_PF_NOTIFY_ENABLE_VF, dev);
95 		err = mlx5_core_enable_hca(dev, vf + 1);
96 		if (err) {
97 			mlx5_core_warn(dev, "failed to enable VF %d (%d)\n", vf, err);
98 			continue;
99 		}
100 
101 		err = mlx5_set_msix_vec_count(dev, vf + 1, num_msix_count);
102 		if (err) {
103 			mlx5_core_warn(dev,
104 				       "failed to set MSI-X vector counts VF %d, err %d\n",
105 				       vf, err);
106 			continue;
107 		}
108 
109 		sriov->vfs_ctx[vf].enabled = 1;
110 		if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) {
111 			err = sriov_restore_guids(dev, vf);
112 			if (err) {
113 				mlx5_core_warn(dev,
114 					       "failed to restore VF %d settings, err %d\n",
115 					       vf, err);
116 				continue;
117 			}
118 		}
119 		mlx5_core_dbg(dev, "successfully enabled VF* %d\n", vf);
120 	}
121 
122 	return 0;
123 }
124 
125 static void
mlx5_device_disable_sriov(struct mlx5_core_dev * dev,int num_vfs,bool clear_vf)126 mlx5_device_disable_sriov(struct mlx5_core_dev *dev, int num_vfs, bool clear_vf)
127 {
128 	struct mlx5_core_sriov *sriov = &dev->priv.sriov;
129 	int err;
130 	int vf;
131 
132 	for (vf = num_vfs - 1; vf >= 0; vf--) {
133 		if (!sriov->vfs_ctx[vf].enabled)
134 			continue;
135 		/* Notify the VF before its disablement to let it clean
136 		 * some resources.
137 		 */
138 		blocking_notifier_call_chain(&sriov->vfs_ctx[vf].notifier,
139 					     MLX5_PF_NOTIFY_DISABLE_VF, dev);
140 		err = mlx5_core_disable_hca(dev, vf + 1);
141 		if (err) {
142 			mlx5_core_warn(dev, "failed to disable VF %d\n", vf);
143 			continue;
144 		}
145 		sriov->vfs_ctx[vf].enabled = 0;
146 	}
147 
148 	mlx5_eswitch_disable_sriov(dev->priv.eswitch, clear_vf);
149 
150 	/* For ECPFs, skip waiting for host VF pages until ECPF is destroyed */
151 	if (mlx5_core_is_ecpf(dev))
152 		return;
153 
154 	if (mlx5_wait_for_pages(dev, &dev->priv.page_counters[MLX5_VF]))
155 		mlx5_core_warn(dev, "timeout reclaiming VFs pages\n");
156 }
157 
mlx5_sriov_enable(struct pci_dev * pdev,int num_vfs)158 static int mlx5_sriov_enable(struct pci_dev *pdev, int num_vfs)
159 {
160 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
161 	struct devlink *devlink = priv_to_devlink(dev);
162 	int err;
163 
164 	devl_lock(devlink);
165 	err = mlx5_device_enable_sriov(dev, num_vfs);
166 	devl_unlock(devlink);
167 	if (err) {
168 		mlx5_core_warn(dev, "mlx5_device_enable_sriov failed : %d\n", err);
169 		return err;
170 	}
171 
172 	err = pci_enable_sriov(pdev, num_vfs);
173 	if (err) {
174 		mlx5_core_warn(dev, "pci_enable_sriov failed : %d\n", err);
175 		mlx5_device_disable_sriov(dev, num_vfs, true);
176 	}
177 	return err;
178 }
179 
mlx5_sriov_disable(struct pci_dev * pdev)180 void mlx5_sriov_disable(struct pci_dev *pdev)
181 {
182 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
183 	struct devlink *devlink = priv_to_devlink(dev);
184 	int num_vfs = pci_num_vf(dev->pdev);
185 
186 	pci_disable_sriov(pdev);
187 	devl_lock(devlink);
188 	mlx5_device_disable_sriov(dev, num_vfs, true);
189 	devl_unlock(devlink);
190 }
191 
mlx5_core_sriov_configure(struct pci_dev * pdev,int num_vfs)192 int mlx5_core_sriov_configure(struct pci_dev *pdev, int num_vfs)
193 {
194 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
195 	struct mlx5_core_sriov *sriov = &dev->priv.sriov;
196 	int err = 0;
197 
198 	mlx5_core_dbg(dev, "requested num_vfs %d\n", num_vfs);
199 
200 	if (num_vfs)
201 		err = mlx5_sriov_enable(pdev, num_vfs);
202 	else
203 		mlx5_sriov_disable(pdev);
204 
205 	if (!err)
206 		sriov->num_vfs = num_vfs;
207 	return err ? err : num_vfs;
208 }
209 
mlx5_core_sriov_set_msix_vec_count(struct pci_dev * vf,int msix_vec_count)210 int mlx5_core_sriov_set_msix_vec_count(struct pci_dev *vf, int msix_vec_count)
211 {
212 	struct pci_dev *pf = pci_physfn(vf);
213 	struct mlx5_core_sriov *sriov;
214 	struct mlx5_core_dev *dev;
215 	int num_vf_msix, id;
216 
217 	dev = pci_get_drvdata(pf);
218 	num_vf_msix = MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix);
219 	if (!num_vf_msix)
220 		return -EOPNOTSUPP;
221 
222 	if (!msix_vec_count)
223 		msix_vec_count =
224 			mlx5_get_default_msix_vec_count(dev, pci_num_vf(pf));
225 
226 	sriov = &dev->priv.sriov;
227 	id = pci_iov_vf_id(vf);
228 	if (id < 0 || !sriov->vfs_ctx[id].enabled)
229 		return -EINVAL;
230 
231 	return mlx5_set_msix_vec_count(dev, id + 1, msix_vec_count);
232 }
233 
mlx5_sriov_attach(struct mlx5_core_dev * dev)234 int mlx5_sriov_attach(struct mlx5_core_dev *dev)
235 {
236 	if (!mlx5_core_is_pf(dev) || !pci_num_vf(dev->pdev))
237 		return 0;
238 
239 	/* If sriov VFs exist in PCI level, enable them in device level */
240 	return mlx5_device_enable_sriov(dev, pci_num_vf(dev->pdev));
241 }
242 
mlx5_sriov_detach(struct mlx5_core_dev * dev)243 void mlx5_sriov_detach(struct mlx5_core_dev *dev)
244 {
245 	if (!mlx5_core_is_pf(dev))
246 		return;
247 
248 	mlx5_device_disable_sriov(dev, pci_num_vf(dev->pdev), false);
249 }
250 
mlx5_get_max_vfs(struct mlx5_core_dev * dev)251 static u16 mlx5_get_max_vfs(struct mlx5_core_dev *dev)
252 {
253 	u16 host_total_vfs;
254 	const u32 *out;
255 
256 	if (mlx5_core_is_ecpf_esw_manager(dev)) {
257 		out = mlx5_esw_query_functions(dev);
258 
259 		/* Old FW doesn't support getting total_vfs from esw func
260 		 * but supports getting it from pci_sriov.
261 		 */
262 		if (IS_ERR(out))
263 			goto done;
264 		host_total_vfs = MLX5_GET(query_esw_functions_out, out,
265 					  host_params_context.host_total_vfs);
266 		kvfree(out);
267 		if (host_total_vfs)
268 			return host_total_vfs;
269 	}
270 
271 done:
272 	return pci_sriov_get_totalvfs(dev->pdev);
273 }
274 
mlx5_sriov_init(struct mlx5_core_dev * dev)275 int mlx5_sriov_init(struct mlx5_core_dev *dev)
276 {
277 	struct mlx5_core_sriov *sriov = &dev->priv.sriov;
278 	struct pci_dev *pdev = dev->pdev;
279 	int total_vfs, i;
280 
281 	if (!mlx5_core_is_pf(dev))
282 		return 0;
283 
284 	total_vfs = pci_sriov_get_totalvfs(pdev);
285 	sriov->max_vfs = mlx5_get_max_vfs(dev);
286 	sriov->num_vfs = pci_num_vf(pdev);
287 	sriov->vfs_ctx = kcalloc(total_vfs, sizeof(*sriov->vfs_ctx), GFP_KERNEL);
288 	if (!sriov->vfs_ctx)
289 		return -ENOMEM;
290 
291 	for (i = 0; i < total_vfs; i++)
292 		BLOCKING_INIT_NOTIFIER_HEAD(&sriov->vfs_ctx[i].notifier);
293 
294 	return 0;
295 }
296 
mlx5_sriov_cleanup(struct mlx5_core_dev * dev)297 void mlx5_sriov_cleanup(struct mlx5_core_dev *dev)
298 {
299 	struct mlx5_core_sriov *sriov = &dev->priv.sriov;
300 
301 	if (!mlx5_core_is_pf(dev))
302 		return;
303 
304 	kfree(sriov->vfs_ctx);
305 }
306 
307 /**
308  * mlx5_sriov_blocking_notifier_unregister - Unregister a VF from
309  * a notification block chain.
310  *
311  * @mdev: The mlx5 core device.
312  * @vf_id: The VF id.
313  * @nb: The notifier block to be unregistered.
314  */
mlx5_sriov_blocking_notifier_unregister(struct mlx5_core_dev * mdev,int vf_id,struct notifier_block * nb)315 void mlx5_sriov_blocking_notifier_unregister(struct mlx5_core_dev *mdev,
316 					     int vf_id,
317 					     struct notifier_block *nb)
318 {
319 	struct mlx5_vf_context *vfs_ctx;
320 	struct mlx5_core_sriov *sriov;
321 
322 	sriov = &mdev->priv.sriov;
323 	if (WARN_ON(vf_id < 0 || vf_id >= sriov->num_vfs))
324 		return;
325 
326 	vfs_ctx = &sriov->vfs_ctx[vf_id];
327 	blocking_notifier_chain_unregister(&vfs_ctx->notifier, nb);
328 }
329 EXPORT_SYMBOL(mlx5_sriov_blocking_notifier_unregister);
330 
331 /**
332  * mlx5_sriov_blocking_notifier_register - Register a VF notification
333  * block chain.
334  *
335  * @mdev: The mlx5 core device.
336  * @vf_id: The VF id.
337  * @nb: The notifier block to be called upon the VF events.
338  *
339  * Returns 0 on success or an error code.
340  */
mlx5_sriov_blocking_notifier_register(struct mlx5_core_dev * mdev,int vf_id,struct notifier_block * nb)341 int mlx5_sriov_blocking_notifier_register(struct mlx5_core_dev *mdev,
342 					  int vf_id,
343 					  struct notifier_block *nb)
344 {
345 	struct mlx5_vf_context *vfs_ctx;
346 	struct mlx5_core_sriov *sriov;
347 
348 	sriov = &mdev->priv.sriov;
349 	if (vf_id < 0 || vf_id >= sriov->num_vfs)
350 		return -EINVAL;
351 
352 	vfs_ctx = &sriov->vfs_ctx[vf_id];
353 	return blocking_notifier_chain_register(&vfs_ctx->notifier, nb);
354 }
355 EXPORT_SYMBOL(mlx5_sriov_blocking_notifier_register);
356