1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
4  * Copyright (c) 2005, 2006, 2007 Cisco Systems, Inc.  All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <linux/sched.h>
36 #include <linux/slab.h>
37 #include <linux/export.h>
38 #include <linux/pci.h>
39 #include <linux/errno.h>
40 
41 #include <linux/mlx4/cmd.h>
42 #include <linux/mlx4/device.h>
43 #include <linux/semaphore.h>
44 #include <rdma/ib_smi.h>
45 #include <linux/delay.h>
46 
47 #include <asm/io.h>
48 
49 #include "mlx4.h"
50 #include "fw.h"
51 #include "fw_qos.h"
52 #include "mlx4_stats.h"
53 
54 #define CMD_POLL_TOKEN 0xffff
55 #define INBOX_MASK	0xffffffffffffff00ULL
56 
57 #define CMD_CHAN_VER 1
58 #define CMD_CHAN_IF_REV 1
59 
60 enum {
61 	/* command completed successfully: */
62 	CMD_STAT_OK		= 0x00,
63 	/* Internal error (such as a bus error) occurred while processing command: */
64 	CMD_STAT_INTERNAL_ERR	= 0x01,
65 	/* Operation/command not supported or opcode modifier not supported: */
66 	CMD_STAT_BAD_OP		= 0x02,
67 	/* Parameter not supported or parameter out of range: */
68 	CMD_STAT_BAD_PARAM	= 0x03,
69 	/* System not enabled or bad system state: */
70 	CMD_STAT_BAD_SYS_STATE	= 0x04,
71 	/* Attempt to access reserved or unallocaterd resource: */
72 	CMD_STAT_BAD_RESOURCE	= 0x05,
73 	/* Requested resource is currently executing a command, or is otherwise busy: */
74 	CMD_STAT_RESOURCE_BUSY	= 0x06,
75 	/* Required capability exceeds device limits: */
76 	CMD_STAT_EXCEED_LIM	= 0x08,
77 	/* Resource is not in the appropriate state or ownership: */
78 	CMD_STAT_BAD_RES_STATE	= 0x09,
79 	/* Index out of range: */
80 	CMD_STAT_BAD_INDEX	= 0x0a,
81 	/* FW image corrupted: */
82 	CMD_STAT_BAD_NVMEM	= 0x0b,
83 	/* Error in ICM mapping (e.g. not enough auxiliary ICM pages to execute command): */
84 	CMD_STAT_ICM_ERROR	= 0x0c,
85 	/* Attempt to modify a QP/EE which is not in the presumed state: */
86 	CMD_STAT_BAD_QP_STATE   = 0x10,
87 	/* Bad segment parameters (Address/Size): */
88 	CMD_STAT_BAD_SEG_PARAM	= 0x20,
89 	/* Memory Region has Memory Windows bound to: */
90 	CMD_STAT_REG_BOUND	= 0x21,
91 	/* HCA local attached memory not present: */
92 	CMD_STAT_LAM_NOT_PRE	= 0x22,
93 	/* Bad management packet (silently discarded): */
94 	CMD_STAT_BAD_PKT	= 0x30,
95 	/* More outstanding CQEs in CQ than new CQ size: */
96 	CMD_STAT_BAD_SIZE	= 0x40,
97 	/* Multi Function device support required: */
98 	CMD_STAT_MULTI_FUNC_REQ	= 0x50,
99 };
100 
101 enum {
102 	HCR_IN_PARAM_OFFSET	= 0x00,
103 	HCR_IN_MODIFIER_OFFSET	= 0x08,
104 	HCR_OUT_PARAM_OFFSET	= 0x0c,
105 	HCR_TOKEN_OFFSET	= 0x14,
106 	HCR_STATUS_OFFSET	= 0x18,
107 
108 	HCR_OPMOD_SHIFT		= 12,
109 	HCR_T_BIT		= 21,
110 	HCR_E_BIT		= 22,
111 	HCR_GO_BIT		= 23
112 };
113 
114 enum {
115 	GO_BIT_TIMEOUT_MSECS	= 10000
116 };
117 
118 enum mlx4_vlan_transition {
119 	MLX4_VLAN_TRANSITION_VST_VST = 0,
120 	MLX4_VLAN_TRANSITION_VST_VGT = 1,
121 	MLX4_VLAN_TRANSITION_VGT_VST = 2,
122 	MLX4_VLAN_TRANSITION_VGT_VGT = 3,
123 };
124 
125 
126 struct mlx4_cmd_context {
127 	struct completion	done;
128 	int			result;
129 	int			next;
130 	u64			out_param;
131 	u16			token;
132 	u8			fw_status;
133 };
134 
135 static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
136 				    struct mlx4_vhcr_cmd *in_vhcr);
137 
138 static int mlx4_status_to_errno(u8 status)
139 {
140 	static const int trans_table[] = {
141 		[CMD_STAT_INTERNAL_ERR]	  = -EIO,
142 		[CMD_STAT_BAD_OP]	  = -EPERM,
143 		[CMD_STAT_BAD_PARAM]	  = -EINVAL,
144 		[CMD_STAT_BAD_SYS_STATE]  = -ENXIO,
145 		[CMD_STAT_BAD_RESOURCE]	  = -EBADF,
146 		[CMD_STAT_RESOURCE_BUSY]  = -EBUSY,
147 		[CMD_STAT_EXCEED_LIM]	  = -ENOMEM,
148 		[CMD_STAT_BAD_RES_STATE]  = -EBADF,
149 		[CMD_STAT_BAD_INDEX]	  = -EBADF,
150 		[CMD_STAT_BAD_NVMEM]	  = -EFAULT,
151 		[CMD_STAT_ICM_ERROR]	  = -ENFILE,
152 		[CMD_STAT_BAD_QP_STATE]   = -EINVAL,
153 		[CMD_STAT_BAD_SEG_PARAM]  = -EFAULT,
154 		[CMD_STAT_REG_BOUND]	  = -EBUSY,
155 		[CMD_STAT_LAM_NOT_PRE]	  = -EAGAIN,
156 		[CMD_STAT_BAD_PKT]	  = -EINVAL,
157 		[CMD_STAT_BAD_SIZE]	  = -ENOMEM,
158 		[CMD_STAT_MULTI_FUNC_REQ] = -EACCES,
159 	};
160 
161 	if (status >= ARRAY_SIZE(trans_table) ||
162 	    (status != CMD_STAT_OK && trans_table[status] == 0))
163 		return -EIO;
164 
165 	return trans_table[status];
166 }
167 
168 static u8 mlx4_errno_to_status(int errno)
169 {
170 	switch (errno) {
171 	case -EPERM:
172 		return CMD_STAT_BAD_OP;
173 	case -EINVAL:
174 		return CMD_STAT_BAD_PARAM;
175 	case -ENXIO:
176 		return CMD_STAT_BAD_SYS_STATE;
177 	case -EBUSY:
178 		return CMD_STAT_RESOURCE_BUSY;
179 	case -ENOMEM:
180 		return CMD_STAT_EXCEED_LIM;
181 	case -ENFILE:
182 		return CMD_STAT_ICM_ERROR;
183 	default:
184 		return CMD_STAT_INTERNAL_ERR;
185 	}
186 }
187 
188 static int mlx4_internal_err_ret_value(struct mlx4_dev *dev, u16 op,
189 				       u8 op_modifier)
190 {
191 	switch (op) {
192 	case MLX4_CMD_UNMAP_ICM:
193 	case MLX4_CMD_UNMAP_ICM_AUX:
194 	case MLX4_CMD_UNMAP_FA:
195 	case MLX4_CMD_2RST_QP:
196 	case MLX4_CMD_HW2SW_EQ:
197 	case MLX4_CMD_HW2SW_CQ:
198 	case MLX4_CMD_HW2SW_SRQ:
199 	case MLX4_CMD_HW2SW_MPT:
200 	case MLX4_CMD_CLOSE_HCA:
201 	case MLX4_QP_FLOW_STEERING_DETACH:
202 	case MLX4_CMD_FREE_RES:
203 	case MLX4_CMD_CLOSE_PORT:
204 		return CMD_STAT_OK;
205 
206 	case MLX4_CMD_QP_ATTACH:
207 		/* On Detach case return success */
208 		if (op_modifier == 0)
209 			return CMD_STAT_OK;
210 		return mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
211 
212 	default:
213 		return mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
214 	}
215 }
216 
217 static int mlx4_closing_cmd_fatal_error(u16 op, u8 fw_status)
218 {
219 	/* Any error during the closing commands below is considered fatal */
220 	if (op == MLX4_CMD_CLOSE_HCA ||
221 	    op == MLX4_CMD_HW2SW_EQ ||
222 	    op == MLX4_CMD_HW2SW_CQ ||
223 	    op == MLX4_CMD_2RST_QP ||
224 	    op == MLX4_CMD_HW2SW_SRQ ||
225 	    op == MLX4_CMD_SYNC_TPT ||
226 	    op == MLX4_CMD_UNMAP_ICM ||
227 	    op == MLX4_CMD_UNMAP_ICM_AUX ||
228 	    op == MLX4_CMD_UNMAP_FA)
229 		return 1;
230 	/* Error on MLX4_CMD_HW2SW_MPT is fatal except when fw status equals
231 	  * CMD_STAT_REG_BOUND.
232 	  * This status indicates that memory region has memory windows bound to it
233 	  * which may result from invalid user space usage and is not fatal.
234 	  */
235 	if (op == MLX4_CMD_HW2SW_MPT && fw_status != CMD_STAT_REG_BOUND)
236 		return 1;
237 	return 0;
238 }
239 
240 static int mlx4_cmd_reset_flow(struct mlx4_dev *dev, u16 op, u8 op_modifier,
241 			       int err)
242 {
243 	/* Only if reset flow is really active return code is based on
244 	  * command, otherwise current error code is returned.
245 	  */
246 	if (mlx4_internal_err_reset) {
247 		mlx4_enter_error_state(dev->persist);
248 		err = mlx4_internal_err_ret_value(dev, op, op_modifier);
249 	}
250 
251 	return err;
252 }
253 
254 static int comm_pending(struct mlx4_dev *dev)
255 {
256 	struct mlx4_priv *priv = mlx4_priv(dev);
257 	u32 status = readl(&priv->mfunc.comm->slave_read);
258 
259 	return (swab32(status) >> 31) != priv->cmd.comm_toggle;
260 }
261 
262 static int mlx4_comm_cmd_post(struct mlx4_dev *dev, u8 cmd, u16 param)
263 {
264 	struct mlx4_priv *priv = mlx4_priv(dev);
265 	u32 val;
266 
267 	/* To avoid writing to unknown addresses after the device state was
268 	 * changed to internal error and the function was rest,
269 	 * check the INTERNAL_ERROR flag which is updated under
270 	 * device_state_mutex lock.
271 	 */
272 	mutex_lock(&dev->persist->device_state_mutex);
273 
274 	if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
275 		mutex_unlock(&dev->persist->device_state_mutex);
276 		return -EIO;
277 	}
278 
279 	priv->cmd.comm_toggle ^= 1;
280 	val = param | (cmd << 16) | (priv->cmd.comm_toggle << 31);
281 	__raw_writel((__force u32) cpu_to_be32(val),
282 		     &priv->mfunc.comm->slave_write);
283 	mmiowb();
284 	mutex_unlock(&dev->persist->device_state_mutex);
285 	return 0;
286 }
287 
288 static int mlx4_comm_cmd_poll(struct mlx4_dev *dev, u8 cmd, u16 param,
289 		       unsigned long timeout)
290 {
291 	struct mlx4_priv *priv = mlx4_priv(dev);
292 	unsigned long end;
293 	int err = 0;
294 	int ret_from_pending = 0;
295 
296 	/* First, verify that the master reports correct status */
297 	if (comm_pending(dev)) {
298 		mlx4_warn(dev, "Communication channel is not idle - my toggle is %d (cmd:0x%x)\n",
299 			  priv->cmd.comm_toggle, cmd);
300 		return -EAGAIN;
301 	}
302 
303 	/* Write command */
304 	down(&priv->cmd.poll_sem);
305 	if (mlx4_comm_cmd_post(dev, cmd, param)) {
306 		/* Only in case the device state is INTERNAL_ERROR,
307 		 * mlx4_comm_cmd_post returns with an error
308 		 */
309 		err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
310 		goto out;
311 	}
312 
313 	end = msecs_to_jiffies(timeout) + jiffies;
314 	while (comm_pending(dev) && time_before(jiffies, end))
315 		cond_resched();
316 	ret_from_pending = comm_pending(dev);
317 	if (ret_from_pending) {
318 		/* check if the slave is trying to boot in the middle of
319 		 * FLR process. The only non-zero result in the RESET command
320 		 * is MLX4_DELAY_RESET_SLAVE*/
321 		if ((MLX4_COMM_CMD_RESET == cmd)) {
322 			err = MLX4_DELAY_RESET_SLAVE;
323 			goto out;
324 		} else {
325 			mlx4_warn(dev, "Communication channel command 0x%x timed out\n",
326 				  cmd);
327 			err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
328 		}
329 	}
330 
331 	if (err)
332 		mlx4_enter_error_state(dev->persist);
333 out:
334 	up(&priv->cmd.poll_sem);
335 	return err;
336 }
337 
338 static int mlx4_comm_cmd_wait(struct mlx4_dev *dev, u8 vhcr_cmd,
339 			      u16 param, u16 op, unsigned long timeout)
340 {
341 	struct mlx4_cmd *cmd = &mlx4_priv(dev)->cmd;
342 	struct mlx4_cmd_context *context;
343 	unsigned long end;
344 	int err = 0;
345 
346 	down(&cmd->event_sem);
347 
348 	spin_lock(&cmd->context_lock);
349 	BUG_ON(cmd->free_head < 0);
350 	context = &cmd->context[cmd->free_head];
351 	context->token += cmd->token_mask + 1;
352 	cmd->free_head = context->next;
353 	spin_unlock(&cmd->context_lock);
354 
355 	reinit_completion(&context->done);
356 
357 	if (mlx4_comm_cmd_post(dev, vhcr_cmd, param)) {
358 		/* Only in case the device state is INTERNAL_ERROR,
359 		 * mlx4_comm_cmd_post returns with an error
360 		 */
361 		err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
362 		goto out;
363 	}
364 
365 	if (!wait_for_completion_timeout(&context->done,
366 					 msecs_to_jiffies(timeout))) {
367 		mlx4_warn(dev, "communication channel command 0x%x (op=0x%x) timed out\n",
368 			  vhcr_cmd, op);
369 		goto out_reset;
370 	}
371 
372 	err = context->result;
373 	if (err && context->fw_status != CMD_STAT_MULTI_FUNC_REQ) {
374 		mlx4_err(dev, "command 0x%x failed: fw status = 0x%x\n",
375 			 vhcr_cmd, context->fw_status);
376 		if (mlx4_closing_cmd_fatal_error(op, context->fw_status))
377 			goto out_reset;
378 	}
379 
380 	/* wait for comm channel ready
381 	 * this is necessary for prevention the race
382 	 * when switching between event to polling mode
383 	 * Skipping this section in case the device is in FATAL_ERROR state,
384 	 * In this state, no commands are sent via the comm channel until
385 	 * the device has returned from reset.
386 	 */
387 	if (!(dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)) {
388 		end = msecs_to_jiffies(timeout) + jiffies;
389 		while (comm_pending(dev) && time_before(jiffies, end))
390 			cond_resched();
391 	}
392 	goto out;
393 
394 out_reset:
395 	err = mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
396 	mlx4_enter_error_state(dev->persist);
397 out:
398 	spin_lock(&cmd->context_lock);
399 	context->next = cmd->free_head;
400 	cmd->free_head = context - cmd->context;
401 	spin_unlock(&cmd->context_lock);
402 
403 	up(&cmd->event_sem);
404 	return err;
405 }
406 
407 int mlx4_comm_cmd(struct mlx4_dev *dev, u8 cmd, u16 param,
408 		  u16 op, unsigned long timeout)
409 {
410 	if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
411 		return mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
412 
413 	if (mlx4_priv(dev)->cmd.use_events)
414 		return mlx4_comm_cmd_wait(dev, cmd, param, op, timeout);
415 	return mlx4_comm_cmd_poll(dev, cmd, param, timeout);
416 }
417 
418 static int cmd_pending(struct mlx4_dev *dev)
419 {
420 	u32 status;
421 
422 	if (pci_channel_offline(dev->persist->pdev))
423 		return -EIO;
424 
425 	status = readl(mlx4_priv(dev)->cmd.hcr + HCR_STATUS_OFFSET);
426 
427 	return (status & swab32(1 << HCR_GO_BIT)) ||
428 		(mlx4_priv(dev)->cmd.toggle ==
429 		 !!(status & swab32(1 << HCR_T_BIT)));
430 }
431 
432 static int mlx4_cmd_post(struct mlx4_dev *dev, u64 in_param, u64 out_param,
433 			 u32 in_modifier, u8 op_modifier, u16 op, u16 token,
434 			 int event)
435 {
436 	struct mlx4_cmd *cmd = &mlx4_priv(dev)->cmd;
437 	u32 __iomem *hcr = cmd->hcr;
438 	int ret = -EIO;
439 	unsigned long end;
440 
441 	mutex_lock(&dev->persist->device_state_mutex);
442 	/* To avoid writing to unknown addresses after the device state was
443 	  * changed to internal error and the chip was reset,
444 	  * check the INTERNAL_ERROR flag which is updated under
445 	  * device_state_mutex lock.
446 	  */
447 	if (pci_channel_offline(dev->persist->pdev) ||
448 	    (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)) {
449 		/*
450 		 * Device is going through error recovery
451 		 * and cannot accept commands.
452 		 */
453 		goto out;
454 	}
455 
456 	end = jiffies;
457 	if (event)
458 		end += msecs_to_jiffies(GO_BIT_TIMEOUT_MSECS);
459 
460 	while (cmd_pending(dev)) {
461 		if (pci_channel_offline(dev->persist->pdev)) {
462 			/*
463 			 * Device is going through error recovery
464 			 * and cannot accept commands.
465 			 */
466 			goto out;
467 		}
468 
469 		if (time_after_eq(jiffies, end)) {
470 			mlx4_err(dev, "%s:cmd_pending failed\n", __func__);
471 			goto out;
472 		}
473 		cond_resched();
474 	}
475 
476 	/*
477 	 * We use writel (instead of something like memcpy_toio)
478 	 * because writes of less than 32 bits to the HCR don't work
479 	 * (and some architectures such as ia64 implement memcpy_toio
480 	 * in terms of writeb).
481 	 */
482 	__raw_writel((__force u32) cpu_to_be32(in_param >> 32),		  hcr + 0);
483 	__raw_writel((__force u32) cpu_to_be32(in_param & 0xfffffffful),  hcr + 1);
484 	__raw_writel((__force u32) cpu_to_be32(in_modifier),		  hcr + 2);
485 	__raw_writel((__force u32) cpu_to_be32(out_param >> 32),	  hcr + 3);
486 	__raw_writel((__force u32) cpu_to_be32(out_param & 0xfffffffful), hcr + 4);
487 	__raw_writel((__force u32) cpu_to_be32(token << 16),		  hcr + 5);
488 
489 	/* __raw_writel may not order writes. */
490 	wmb();
491 
492 	__raw_writel((__force u32) cpu_to_be32((1 << HCR_GO_BIT)		|
493 					       (cmd->toggle << HCR_T_BIT)	|
494 					       (event ? (1 << HCR_E_BIT) : 0)	|
495 					       (op_modifier << HCR_OPMOD_SHIFT) |
496 					       op), hcr + 6);
497 
498 	/*
499 	 * Make sure that our HCR writes don't get mixed in with
500 	 * writes from another CPU starting a FW command.
501 	 */
502 	mmiowb();
503 
504 	cmd->toggle = cmd->toggle ^ 1;
505 
506 	ret = 0;
507 
508 out:
509 	if (ret)
510 		mlx4_warn(dev, "Could not post command 0x%x: ret=%d, in_param=0x%llx, in_mod=0x%x, op_mod=0x%x\n",
511 			  op, ret, in_param, in_modifier, op_modifier);
512 	mutex_unlock(&dev->persist->device_state_mutex);
513 
514 	return ret;
515 }
516 
517 static int mlx4_slave_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
518 			  int out_is_imm, u32 in_modifier, u8 op_modifier,
519 			  u16 op, unsigned long timeout)
520 {
521 	struct mlx4_priv *priv = mlx4_priv(dev);
522 	struct mlx4_vhcr_cmd *vhcr = priv->mfunc.vhcr;
523 	int ret;
524 
525 	mutex_lock(&priv->cmd.slave_cmd_mutex);
526 
527 	vhcr->in_param = cpu_to_be64(in_param);
528 	vhcr->out_param = out_param ? cpu_to_be64(*out_param) : 0;
529 	vhcr->in_modifier = cpu_to_be32(in_modifier);
530 	vhcr->opcode = cpu_to_be16((((u16) op_modifier) << 12) | (op & 0xfff));
531 	vhcr->token = cpu_to_be16(CMD_POLL_TOKEN);
532 	vhcr->status = 0;
533 	vhcr->flags = !!(priv->cmd.use_events) << 6;
534 
535 	if (mlx4_is_master(dev)) {
536 		ret = mlx4_master_process_vhcr(dev, dev->caps.function, vhcr);
537 		if (!ret) {
538 			if (out_is_imm) {
539 				if (out_param)
540 					*out_param =
541 						be64_to_cpu(vhcr->out_param);
542 				else {
543 					mlx4_err(dev, "response expected while output mailbox is NULL for command 0x%x\n",
544 						 op);
545 					vhcr->status = CMD_STAT_BAD_PARAM;
546 				}
547 			}
548 			ret = mlx4_status_to_errno(vhcr->status);
549 		}
550 		if (ret &&
551 		    dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
552 			ret = mlx4_internal_err_ret_value(dev, op, op_modifier);
553 	} else {
554 		ret = mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_POST, 0, op,
555 				    MLX4_COMM_TIME + timeout);
556 		if (!ret) {
557 			if (out_is_imm) {
558 				if (out_param)
559 					*out_param =
560 						be64_to_cpu(vhcr->out_param);
561 				else {
562 					mlx4_err(dev, "response expected while output mailbox is NULL for command 0x%x\n",
563 						 op);
564 					vhcr->status = CMD_STAT_BAD_PARAM;
565 				}
566 			}
567 			ret = mlx4_status_to_errno(vhcr->status);
568 		} else {
569 			if (dev->persist->state &
570 			    MLX4_DEVICE_STATE_INTERNAL_ERROR)
571 				ret = mlx4_internal_err_ret_value(dev, op,
572 								  op_modifier);
573 			else
574 				mlx4_err(dev, "failed execution of VHCR_POST command opcode 0x%x\n", op);
575 		}
576 	}
577 
578 	mutex_unlock(&priv->cmd.slave_cmd_mutex);
579 	return ret;
580 }
581 
582 static int mlx4_cmd_poll(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
583 			 int out_is_imm, u32 in_modifier, u8 op_modifier,
584 			 u16 op, unsigned long timeout)
585 {
586 	struct mlx4_priv *priv = mlx4_priv(dev);
587 	void __iomem *hcr = priv->cmd.hcr;
588 	int err = 0;
589 	unsigned long end;
590 	u32 stat;
591 
592 	down(&priv->cmd.poll_sem);
593 
594 	if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
595 		/*
596 		 * Device is going through error recovery
597 		 * and cannot accept commands.
598 		 */
599 		err = mlx4_internal_err_ret_value(dev, op, op_modifier);
600 		goto out;
601 	}
602 
603 	if (out_is_imm && !out_param) {
604 		mlx4_err(dev, "response expected while output mailbox is NULL for command 0x%x\n",
605 			 op);
606 		err = -EINVAL;
607 		goto out;
608 	}
609 
610 	err = mlx4_cmd_post(dev, in_param, out_param ? *out_param : 0,
611 			    in_modifier, op_modifier, op, CMD_POLL_TOKEN, 0);
612 	if (err)
613 		goto out_reset;
614 
615 	end = msecs_to_jiffies(timeout) + jiffies;
616 	while (cmd_pending(dev) && time_before(jiffies, end)) {
617 		if (pci_channel_offline(dev->persist->pdev)) {
618 			/*
619 			 * Device is going through error recovery
620 			 * and cannot accept commands.
621 			 */
622 			err = -EIO;
623 			goto out_reset;
624 		}
625 
626 		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR) {
627 			err = mlx4_internal_err_ret_value(dev, op, op_modifier);
628 			goto out;
629 		}
630 
631 		cond_resched();
632 	}
633 
634 	if (cmd_pending(dev)) {
635 		mlx4_warn(dev, "command 0x%x timed out (go bit not cleared)\n",
636 			  op);
637 		err = -EIO;
638 		goto out_reset;
639 	}
640 
641 	if (out_is_imm)
642 		*out_param =
643 			(u64) be32_to_cpu((__force __be32)
644 					  __raw_readl(hcr + HCR_OUT_PARAM_OFFSET)) << 32 |
645 			(u64) be32_to_cpu((__force __be32)
646 					  __raw_readl(hcr + HCR_OUT_PARAM_OFFSET + 4));
647 	stat = be32_to_cpu((__force __be32)
648 			   __raw_readl(hcr + HCR_STATUS_OFFSET)) >> 24;
649 	err = mlx4_status_to_errno(stat);
650 	if (err) {
651 		mlx4_err(dev, "command 0x%x failed: fw status = 0x%x\n",
652 			 op, stat);
653 		if (mlx4_closing_cmd_fatal_error(op, stat))
654 			goto out_reset;
655 		goto out;
656 	}
657 
658 out_reset:
659 	if (err)
660 		err = mlx4_cmd_reset_flow(dev, op, op_modifier, err);
661 out:
662 	up(&priv->cmd.poll_sem);
663 	return err;
664 }
665 
666 void mlx4_cmd_event(struct mlx4_dev *dev, u16 token, u8 status, u64 out_param)
667 {
668 	struct mlx4_priv *priv = mlx4_priv(dev);
669 	struct mlx4_cmd_context *context =
670 		&priv->cmd.context[token & priv->cmd.token_mask];
671 
672 	/* previously timed out command completing at long last */
673 	if (token != context->token)
674 		return;
675 
676 	context->fw_status = status;
677 	context->result    = mlx4_status_to_errno(status);
678 	context->out_param = out_param;
679 
680 	complete(&context->done);
681 }
682 
683 static int mlx4_cmd_wait(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
684 			 int out_is_imm, u32 in_modifier, u8 op_modifier,
685 			 u16 op, unsigned long timeout)
686 {
687 	struct mlx4_cmd *cmd = &mlx4_priv(dev)->cmd;
688 	struct mlx4_cmd_context *context;
689 	int err = 0;
690 
691 	down(&cmd->event_sem);
692 
693 	spin_lock(&cmd->context_lock);
694 	BUG_ON(cmd->free_head < 0);
695 	context = &cmd->context[cmd->free_head];
696 	context->token += cmd->token_mask + 1;
697 	cmd->free_head = context->next;
698 	spin_unlock(&cmd->context_lock);
699 
700 	if (out_is_imm && !out_param) {
701 		mlx4_err(dev, "response expected while output mailbox is NULL for command 0x%x\n",
702 			 op);
703 		err = -EINVAL;
704 		goto out;
705 	}
706 
707 	reinit_completion(&context->done);
708 
709 	err = mlx4_cmd_post(dev, in_param, out_param ? *out_param : 0,
710 			    in_modifier, op_modifier, op, context->token, 1);
711 	if (err)
712 		goto out_reset;
713 
714 	if (!wait_for_completion_timeout(&context->done,
715 					 msecs_to_jiffies(timeout))) {
716 		mlx4_warn(dev, "command 0x%x timed out (go bit not cleared)\n",
717 			  op);
718 		if (op == MLX4_CMD_NOP) {
719 			err = -EBUSY;
720 			goto out;
721 		} else {
722 			err = -EIO;
723 			goto out_reset;
724 		}
725 	}
726 
727 	err = context->result;
728 	if (err) {
729 		/* Since we do not want to have this error message always
730 		 * displayed at driver start when there are ConnectX2 HCAs
731 		 * on the host, we deprecate the error message for this
732 		 * specific command/input_mod/opcode_mod/fw-status to be debug.
733 		 */
734 		if (op == MLX4_CMD_SET_PORT &&
735 		    (in_modifier == 1 || in_modifier == 2) &&
736 		    op_modifier == MLX4_SET_PORT_IB_OPCODE &&
737 		    context->fw_status == CMD_STAT_BAD_SIZE)
738 			mlx4_dbg(dev, "command 0x%x failed: fw status = 0x%x\n",
739 				 op, context->fw_status);
740 		else
741 			mlx4_err(dev, "command 0x%x failed: fw status = 0x%x\n",
742 				 op, context->fw_status);
743 		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
744 			err = mlx4_internal_err_ret_value(dev, op, op_modifier);
745 		else if (mlx4_closing_cmd_fatal_error(op, context->fw_status))
746 			goto out_reset;
747 
748 		goto out;
749 	}
750 
751 	if (out_is_imm)
752 		*out_param = context->out_param;
753 
754 out_reset:
755 	if (err)
756 		err = mlx4_cmd_reset_flow(dev, op, op_modifier, err);
757 out:
758 	spin_lock(&cmd->context_lock);
759 	context->next = cmd->free_head;
760 	cmd->free_head = context - cmd->context;
761 	spin_unlock(&cmd->context_lock);
762 
763 	up(&cmd->event_sem);
764 	return err;
765 }
766 
767 int __mlx4_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
768 	       int out_is_imm, u32 in_modifier, u8 op_modifier,
769 	       u16 op, unsigned long timeout, int native)
770 {
771 	if (pci_channel_offline(dev->persist->pdev))
772 		return mlx4_cmd_reset_flow(dev, op, op_modifier, -EIO);
773 
774 	if (!mlx4_is_mfunc(dev) || (native && mlx4_is_master(dev))) {
775 		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
776 			return mlx4_internal_err_ret_value(dev, op,
777 							  op_modifier);
778 		if (mlx4_priv(dev)->cmd.use_events)
779 			return mlx4_cmd_wait(dev, in_param, out_param,
780 					     out_is_imm, in_modifier,
781 					     op_modifier, op, timeout);
782 		else
783 			return mlx4_cmd_poll(dev, in_param, out_param,
784 					     out_is_imm, in_modifier,
785 					     op_modifier, op, timeout);
786 	}
787 	return mlx4_slave_cmd(dev, in_param, out_param, out_is_imm,
788 			      in_modifier, op_modifier, op, timeout);
789 }
790 EXPORT_SYMBOL_GPL(__mlx4_cmd);
791 
792 
793 int mlx4_ARM_COMM_CHANNEL(struct mlx4_dev *dev)
794 {
795 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_ARM_COMM_CHANNEL,
796 			MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
797 }
798 
799 static int mlx4_ACCESS_MEM(struct mlx4_dev *dev, u64 master_addr,
800 			   int slave, u64 slave_addr,
801 			   int size, int is_read)
802 {
803 	u64 in_param;
804 	u64 out_param;
805 
806 	if ((slave_addr & 0xfff) | (master_addr & 0xfff) |
807 	    (slave & ~0x7f) | (size & 0xff)) {
808 		mlx4_err(dev, "Bad access mem params - slave_addr:0x%llx master_addr:0x%llx slave_id:%d size:%d\n",
809 			 slave_addr, master_addr, slave, size);
810 		return -EINVAL;
811 	}
812 
813 	if (is_read) {
814 		in_param = (u64) slave | slave_addr;
815 		out_param = (u64) dev->caps.function | master_addr;
816 	} else {
817 		in_param = (u64) dev->caps.function | master_addr;
818 		out_param = (u64) slave | slave_addr;
819 	}
820 
821 	return mlx4_cmd_imm(dev, in_param, &out_param, size, 0,
822 			    MLX4_CMD_ACCESS_MEM,
823 			    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
824 }
825 
826 static int query_pkey_block(struct mlx4_dev *dev, u8 port, u16 index, u16 *pkey,
827 			       struct mlx4_cmd_mailbox *inbox,
828 			       struct mlx4_cmd_mailbox *outbox)
829 {
830 	struct ib_smp *in_mad = (struct ib_smp *)(inbox->buf);
831 	struct ib_smp *out_mad = (struct ib_smp *)(outbox->buf);
832 	int err;
833 	int i;
834 
835 	if (index & 0x1f)
836 		return -EINVAL;
837 
838 	in_mad->attr_mod = cpu_to_be32(index / 32);
839 
840 	err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, port, 3,
841 			   MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C,
842 			   MLX4_CMD_NATIVE);
843 	if (err)
844 		return err;
845 
846 	for (i = 0; i < 32; ++i)
847 		pkey[i] = be16_to_cpu(((__be16 *) out_mad->data)[i]);
848 
849 	return err;
850 }
851 
852 static int get_full_pkey_table(struct mlx4_dev *dev, u8 port, u16 *table,
853 			       struct mlx4_cmd_mailbox *inbox,
854 			       struct mlx4_cmd_mailbox *outbox)
855 {
856 	int i;
857 	int err;
858 
859 	for (i = 0; i < dev->caps.pkey_table_len[port]; i += 32) {
860 		err = query_pkey_block(dev, port, i, table + i, inbox, outbox);
861 		if (err)
862 			return err;
863 	}
864 
865 	return 0;
866 }
867 #define PORT_CAPABILITY_LOCATION_IN_SMP 20
868 #define PORT_STATE_OFFSET 32
869 
870 static enum ib_port_state vf_port_state(struct mlx4_dev *dev, int port, int vf)
871 {
872 	if (mlx4_get_slave_port_state(dev, vf, port) == SLAVE_PORT_UP)
873 		return IB_PORT_ACTIVE;
874 	else
875 		return IB_PORT_DOWN;
876 }
877 
878 static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave,
879 				struct mlx4_vhcr *vhcr,
880 				struct mlx4_cmd_mailbox *inbox,
881 				struct mlx4_cmd_mailbox *outbox,
882 				struct mlx4_cmd_info *cmd)
883 {
884 	struct ib_smp *smp = inbox->buf;
885 	u32 index;
886 	u8 port, slave_port;
887 	u8 opcode_modifier;
888 	u16 *table;
889 	int err;
890 	int vidx, pidx;
891 	int network_view;
892 	struct mlx4_priv *priv = mlx4_priv(dev);
893 	struct ib_smp *outsmp = outbox->buf;
894 	__be16 *outtab = (__be16 *)(outsmp->data);
895 	__be32 slave_cap_mask;
896 	__be64 slave_node_guid;
897 
898 	slave_port = vhcr->in_modifier;
899 	port = mlx4_slave_convert_port(dev, slave, slave_port);
900 
901 	/* network-view bit is for driver use only, and should not be passed to FW */
902 	opcode_modifier = vhcr->op_modifier & ~0x8; /* clear netw view bit */
903 	network_view = !!(vhcr->op_modifier & 0x8);
904 
905 	if (smp->base_version == 1 &&
906 	    smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED &&
907 	    smp->class_version == 1) {
908 		/* host view is paravirtualized */
909 		if (!network_view && smp->method == IB_MGMT_METHOD_GET) {
910 			if (smp->attr_id == IB_SMP_ATTR_PKEY_TABLE) {
911 				index = be32_to_cpu(smp->attr_mod);
912 				if (port < 1 || port > dev->caps.num_ports)
913 					return -EINVAL;
914 				table = kcalloc((dev->caps.pkey_table_len[port] / 32) + 1,
915 						sizeof(*table) * 32, GFP_KERNEL);
916 
917 				if (!table)
918 					return -ENOMEM;
919 				/* need to get the full pkey table because the paravirtualized
920 				 * pkeys may be scattered among several pkey blocks.
921 				 */
922 				err = get_full_pkey_table(dev, port, table, inbox, outbox);
923 				if (!err) {
924 					for (vidx = index * 32; vidx < (index + 1) * 32; ++vidx) {
925 						pidx = priv->virt2phys_pkey[slave][port - 1][vidx];
926 						outtab[vidx % 32] = cpu_to_be16(table[pidx]);
927 					}
928 				}
929 				kfree(table);
930 				return err;
931 			}
932 			if (smp->attr_id == IB_SMP_ATTR_PORT_INFO) {
933 				/*get the slave specific caps:*/
934 				/*do the command */
935 				smp->attr_mod = cpu_to_be32(port);
936 				err = mlx4_cmd_box(dev, inbox->dma, outbox->dma,
937 					    port, opcode_modifier,
938 					    vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
939 				/* modify the response for slaves */
940 				if (!err && slave != mlx4_master_func_num(dev)) {
941 					u8 *state = outsmp->data + PORT_STATE_OFFSET;
942 
943 					*state = (*state & 0xf0) | vf_port_state(dev, port, slave);
944 					slave_cap_mask = priv->mfunc.master.slave_state[slave].ib_cap_mask[port];
945 					memcpy(outsmp->data + PORT_CAPABILITY_LOCATION_IN_SMP, &slave_cap_mask, 4);
946 				}
947 				return err;
948 			}
949 			if (smp->attr_id == IB_SMP_ATTR_GUID_INFO) {
950 				__be64 guid = mlx4_get_admin_guid(dev, slave,
951 								  port);
952 
953 				/* set the PF admin guid to the FW/HW burned
954 				 * GUID, if it wasn't yet set
955 				 */
956 				if (slave == 0 && guid == 0) {
957 					smp->attr_mod = 0;
958 					err = mlx4_cmd_box(dev,
959 							   inbox->dma,
960 							   outbox->dma,
961 							   vhcr->in_modifier,
962 							   opcode_modifier,
963 							   vhcr->op,
964 							   MLX4_CMD_TIME_CLASS_C,
965 							   MLX4_CMD_NATIVE);
966 					if (err)
967 						return err;
968 					mlx4_set_admin_guid(dev,
969 							    *(__be64 *)outsmp->
970 							    data, slave, port);
971 				} else {
972 					memcpy(outsmp->data, &guid, 8);
973 				}
974 
975 				/* clean all other gids */
976 				memset(outsmp->data + 8, 0, 56);
977 				return 0;
978 			}
979 			if (smp->attr_id == IB_SMP_ATTR_NODE_INFO) {
980 				err = mlx4_cmd_box(dev, inbox->dma, outbox->dma,
981 					     port, opcode_modifier,
982 					     vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
983 				if (!err) {
984 					slave_node_guid =  mlx4_get_slave_node_guid(dev, slave);
985 					memcpy(outsmp->data + 12, &slave_node_guid, 8);
986 				}
987 				return err;
988 			}
989 		}
990 	}
991 
992 	/* Non-privileged VFs are only allowed "host" view LID-routed 'Get' MADs.
993 	 * These are the MADs used by ib verbs (such as ib_query_gids).
994 	 */
995 	if (slave != mlx4_master_func_num(dev) &&
996 	    !mlx4_vf_smi_enabled(dev, slave, port)) {
997 		if (!(smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED &&
998 		      smp->method == IB_MGMT_METHOD_GET) || network_view) {
999 			mlx4_err(dev, "Unprivileged slave %d is trying to execute a Subnet MGMT MAD, class 0x%x, method 0x%x, view=%s for attr 0x%x. Rejecting\n",
1000 				 slave, smp->method, smp->mgmt_class,
1001 				 network_view ? "Network" : "Host",
1002 				 be16_to_cpu(smp->attr_id));
1003 			return -EPERM;
1004 		}
1005 	}
1006 
1007 	return mlx4_cmd_box(dev, inbox->dma, outbox->dma,
1008 				    vhcr->in_modifier, opcode_modifier,
1009 				    vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
1010 }
1011 
1012 static int mlx4_CMD_EPERM_wrapper(struct mlx4_dev *dev, int slave,
1013 		     struct mlx4_vhcr *vhcr,
1014 		     struct mlx4_cmd_mailbox *inbox,
1015 		     struct mlx4_cmd_mailbox *outbox,
1016 		     struct mlx4_cmd_info *cmd)
1017 {
1018 	return -EPERM;
1019 }
1020 
1021 int mlx4_DMA_wrapper(struct mlx4_dev *dev, int slave,
1022 		     struct mlx4_vhcr *vhcr,
1023 		     struct mlx4_cmd_mailbox *inbox,
1024 		     struct mlx4_cmd_mailbox *outbox,
1025 		     struct mlx4_cmd_info *cmd)
1026 {
1027 	u64 in_param;
1028 	u64 out_param;
1029 	int err;
1030 
1031 	in_param = cmd->has_inbox ? (u64) inbox->dma : vhcr->in_param;
1032 	out_param = cmd->has_outbox ? (u64) outbox->dma : vhcr->out_param;
1033 	if (cmd->encode_slave_id) {
1034 		in_param &= 0xffffffffffffff00ll;
1035 		in_param |= slave;
1036 	}
1037 
1038 	err = __mlx4_cmd(dev, in_param, &out_param, cmd->out_is_imm,
1039 			 vhcr->in_modifier, vhcr->op_modifier, vhcr->op,
1040 			 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1041 
1042 	if (cmd->out_is_imm)
1043 		vhcr->out_param = out_param;
1044 
1045 	return err;
1046 }
1047 
1048 static struct mlx4_cmd_info cmd_info[] = {
1049 	{
1050 		.opcode = MLX4_CMD_QUERY_FW,
1051 		.has_inbox = false,
1052 		.has_outbox = true,
1053 		.out_is_imm = false,
1054 		.encode_slave_id = false,
1055 		.verify = NULL,
1056 		.wrapper = mlx4_QUERY_FW_wrapper
1057 	},
1058 	{
1059 		.opcode = MLX4_CMD_QUERY_HCA,
1060 		.has_inbox = false,
1061 		.has_outbox = true,
1062 		.out_is_imm = false,
1063 		.encode_slave_id = false,
1064 		.verify = NULL,
1065 		.wrapper = NULL
1066 	},
1067 	{
1068 		.opcode = MLX4_CMD_QUERY_DEV_CAP,
1069 		.has_inbox = false,
1070 		.has_outbox = true,
1071 		.out_is_imm = false,
1072 		.encode_slave_id = false,
1073 		.verify = NULL,
1074 		.wrapper = mlx4_QUERY_DEV_CAP_wrapper
1075 	},
1076 	{
1077 		.opcode = MLX4_CMD_QUERY_FUNC_CAP,
1078 		.has_inbox = false,
1079 		.has_outbox = true,
1080 		.out_is_imm = false,
1081 		.encode_slave_id = false,
1082 		.verify = NULL,
1083 		.wrapper = mlx4_QUERY_FUNC_CAP_wrapper
1084 	},
1085 	{
1086 		.opcode = MLX4_CMD_QUERY_ADAPTER,
1087 		.has_inbox = false,
1088 		.has_outbox = true,
1089 		.out_is_imm = false,
1090 		.encode_slave_id = false,
1091 		.verify = NULL,
1092 		.wrapper = NULL
1093 	},
1094 	{
1095 		.opcode = MLX4_CMD_INIT_PORT,
1096 		.has_inbox = false,
1097 		.has_outbox = false,
1098 		.out_is_imm = false,
1099 		.encode_slave_id = false,
1100 		.verify = NULL,
1101 		.wrapper = mlx4_INIT_PORT_wrapper
1102 	},
1103 	{
1104 		.opcode = MLX4_CMD_CLOSE_PORT,
1105 		.has_inbox = false,
1106 		.has_outbox = false,
1107 		.out_is_imm  = false,
1108 		.encode_slave_id = false,
1109 		.verify = NULL,
1110 		.wrapper = mlx4_CLOSE_PORT_wrapper
1111 	},
1112 	{
1113 		.opcode = MLX4_CMD_QUERY_PORT,
1114 		.has_inbox = false,
1115 		.has_outbox = true,
1116 		.out_is_imm = false,
1117 		.encode_slave_id = false,
1118 		.verify = NULL,
1119 		.wrapper = mlx4_QUERY_PORT_wrapper
1120 	},
1121 	{
1122 		.opcode = MLX4_CMD_SET_PORT,
1123 		.has_inbox = true,
1124 		.has_outbox = false,
1125 		.out_is_imm = false,
1126 		.encode_slave_id = false,
1127 		.verify = NULL,
1128 		.wrapper = mlx4_SET_PORT_wrapper
1129 	},
1130 	{
1131 		.opcode = MLX4_CMD_MAP_EQ,
1132 		.has_inbox = false,
1133 		.has_outbox = false,
1134 		.out_is_imm = false,
1135 		.encode_slave_id = false,
1136 		.verify = NULL,
1137 		.wrapper = mlx4_MAP_EQ_wrapper
1138 	},
1139 	{
1140 		.opcode = MLX4_CMD_SW2HW_EQ,
1141 		.has_inbox = true,
1142 		.has_outbox = false,
1143 		.out_is_imm = false,
1144 		.encode_slave_id = true,
1145 		.verify = NULL,
1146 		.wrapper = mlx4_SW2HW_EQ_wrapper
1147 	},
1148 	{
1149 		.opcode = MLX4_CMD_HW_HEALTH_CHECK,
1150 		.has_inbox = false,
1151 		.has_outbox = false,
1152 		.out_is_imm = false,
1153 		.encode_slave_id = false,
1154 		.verify = NULL,
1155 		.wrapper = NULL
1156 	},
1157 	{
1158 		.opcode = MLX4_CMD_NOP,
1159 		.has_inbox = false,
1160 		.has_outbox = false,
1161 		.out_is_imm = false,
1162 		.encode_slave_id = false,
1163 		.verify = NULL,
1164 		.wrapper = NULL
1165 	},
1166 	{
1167 		.opcode = MLX4_CMD_CONFIG_DEV,
1168 		.has_inbox = false,
1169 		.has_outbox = true,
1170 		.out_is_imm = false,
1171 		.encode_slave_id = false,
1172 		.verify = NULL,
1173 		.wrapper = mlx4_CONFIG_DEV_wrapper
1174 	},
1175 	{
1176 		.opcode = MLX4_CMD_ALLOC_RES,
1177 		.has_inbox = false,
1178 		.has_outbox = false,
1179 		.out_is_imm = true,
1180 		.encode_slave_id = false,
1181 		.verify = NULL,
1182 		.wrapper = mlx4_ALLOC_RES_wrapper
1183 	},
1184 	{
1185 		.opcode = MLX4_CMD_FREE_RES,
1186 		.has_inbox = false,
1187 		.has_outbox = false,
1188 		.out_is_imm = false,
1189 		.encode_slave_id = false,
1190 		.verify = NULL,
1191 		.wrapper = mlx4_FREE_RES_wrapper
1192 	},
1193 	{
1194 		.opcode = MLX4_CMD_SW2HW_MPT,
1195 		.has_inbox = true,
1196 		.has_outbox = false,
1197 		.out_is_imm = false,
1198 		.encode_slave_id = true,
1199 		.verify = NULL,
1200 		.wrapper = mlx4_SW2HW_MPT_wrapper
1201 	},
1202 	{
1203 		.opcode = MLX4_CMD_QUERY_MPT,
1204 		.has_inbox = false,
1205 		.has_outbox = true,
1206 		.out_is_imm = false,
1207 		.encode_slave_id = false,
1208 		.verify = NULL,
1209 		.wrapper = mlx4_QUERY_MPT_wrapper
1210 	},
1211 	{
1212 		.opcode = MLX4_CMD_HW2SW_MPT,
1213 		.has_inbox = false,
1214 		.has_outbox = false,
1215 		.out_is_imm = false,
1216 		.encode_slave_id = false,
1217 		.verify = NULL,
1218 		.wrapper = mlx4_HW2SW_MPT_wrapper
1219 	},
1220 	{
1221 		.opcode = MLX4_CMD_READ_MTT,
1222 		.has_inbox = false,
1223 		.has_outbox = true,
1224 		.out_is_imm = false,
1225 		.encode_slave_id = false,
1226 		.verify = NULL,
1227 		.wrapper = NULL
1228 	},
1229 	{
1230 		.opcode = MLX4_CMD_WRITE_MTT,
1231 		.has_inbox = true,
1232 		.has_outbox = false,
1233 		.out_is_imm = false,
1234 		.encode_slave_id = false,
1235 		.verify = NULL,
1236 		.wrapper = mlx4_WRITE_MTT_wrapper
1237 	},
1238 	{
1239 		.opcode = MLX4_CMD_SYNC_TPT,
1240 		.has_inbox = true,
1241 		.has_outbox = false,
1242 		.out_is_imm = false,
1243 		.encode_slave_id = false,
1244 		.verify = NULL,
1245 		.wrapper = NULL
1246 	},
1247 	{
1248 		.opcode = MLX4_CMD_HW2SW_EQ,
1249 		.has_inbox = false,
1250 		.has_outbox = false,
1251 		.out_is_imm = false,
1252 		.encode_slave_id = true,
1253 		.verify = NULL,
1254 		.wrapper = mlx4_HW2SW_EQ_wrapper
1255 	},
1256 	{
1257 		.opcode = MLX4_CMD_QUERY_EQ,
1258 		.has_inbox = false,
1259 		.has_outbox = true,
1260 		.out_is_imm = false,
1261 		.encode_slave_id = true,
1262 		.verify = NULL,
1263 		.wrapper = mlx4_QUERY_EQ_wrapper
1264 	},
1265 	{
1266 		.opcode = MLX4_CMD_SW2HW_CQ,
1267 		.has_inbox = true,
1268 		.has_outbox = false,
1269 		.out_is_imm = false,
1270 		.encode_slave_id = true,
1271 		.verify = NULL,
1272 		.wrapper = mlx4_SW2HW_CQ_wrapper
1273 	},
1274 	{
1275 		.opcode = MLX4_CMD_HW2SW_CQ,
1276 		.has_inbox = false,
1277 		.has_outbox = false,
1278 		.out_is_imm = false,
1279 		.encode_slave_id = false,
1280 		.verify = NULL,
1281 		.wrapper = mlx4_HW2SW_CQ_wrapper
1282 	},
1283 	{
1284 		.opcode = MLX4_CMD_QUERY_CQ,
1285 		.has_inbox = false,
1286 		.has_outbox = true,
1287 		.out_is_imm = false,
1288 		.encode_slave_id = false,
1289 		.verify = NULL,
1290 		.wrapper = mlx4_QUERY_CQ_wrapper
1291 	},
1292 	{
1293 		.opcode = MLX4_CMD_MODIFY_CQ,
1294 		.has_inbox = true,
1295 		.has_outbox = false,
1296 		.out_is_imm = true,
1297 		.encode_slave_id = false,
1298 		.verify = NULL,
1299 		.wrapper = mlx4_MODIFY_CQ_wrapper
1300 	},
1301 	{
1302 		.opcode = MLX4_CMD_SW2HW_SRQ,
1303 		.has_inbox = true,
1304 		.has_outbox = false,
1305 		.out_is_imm = false,
1306 		.encode_slave_id = true,
1307 		.verify = NULL,
1308 		.wrapper = mlx4_SW2HW_SRQ_wrapper
1309 	},
1310 	{
1311 		.opcode = MLX4_CMD_HW2SW_SRQ,
1312 		.has_inbox = false,
1313 		.has_outbox = false,
1314 		.out_is_imm = false,
1315 		.encode_slave_id = false,
1316 		.verify = NULL,
1317 		.wrapper = mlx4_HW2SW_SRQ_wrapper
1318 	},
1319 	{
1320 		.opcode = MLX4_CMD_QUERY_SRQ,
1321 		.has_inbox = false,
1322 		.has_outbox = true,
1323 		.out_is_imm = false,
1324 		.encode_slave_id = false,
1325 		.verify = NULL,
1326 		.wrapper = mlx4_QUERY_SRQ_wrapper
1327 	},
1328 	{
1329 		.opcode = MLX4_CMD_ARM_SRQ,
1330 		.has_inbox = false,
1331 		.has_outbox = false,
1332 		.out_is_imm = false,
1333 		.encode_slave_id = false,
1334 		.verify = NULL,
1335 		.wrapper = mlx4_ARM_SRQ_wrapper
1336 	},
1337 	{
1338 		.opcode = MLX4_CMD_RST2INIT_QP,
1339 		.has_inbox = true,
1340 		.has_outbox = false,
1341 		.out_is_imm = false,
1342 		.encode_slave_id = true,
1343 		.verify = NULL,
1344 		.wrapper = mlx4_RST2INIT_QP_wrapper
1345 	},
1346 	{
1347 		.opcode = MLX4_CMD_INIT2INIT_QP,
1348 		.has_inbox = true,
1349 		.has_outbox = false,
1350 		.out_is_imm = false,
1351 		.encode_slave_id = false,
1352 		.verify = NULL,
1353 		.wrapper = mlx4_INIT2INIT_QP_wrapper
1354 	},
1355 	{
1356 		.opcode = MLX4_CMD_INIT2RTR_QP,
1357 		.has_inbox = true,
1358 		.has_outbox = false,
1359 		.out_is_imm = false,
1360 		.encode_slave_id = false,
1361 		.verify = NULL,
1362 		.wrapper = mlx4_INIT2RTR_QP_wrapper
1363 	},
1364 	{
1365 		.opcode = MLX4_CMD_RTR2RTS_QP,
1366 		.has_inbox = true,
1367 		.has_outbox = false,
1368 		.out_is_imm = false,
1369 		.encode_slave_id = false,
1370 		.verify = NULL,
1371 		.wrapper = mlx4_RTR2RTS_QP_wrapper
1372 	},
1373 	{
1374 		.opcode = MLX4_CMD_RTS2RTS_QP,
1375 		.has_inbox = true,
1376 		.has_outbox = false,
1377 		.out_is_imm = false,
1378 		.encode_slave_id = false,
1379 		.verify = NULL,
1380 		.wrapper = mlx4_RTS2RTS_QP_wrapper
1381 	},
1382 	{
1383 		.opcode = MLX4_CMD_SQERR2RTS_QP,
1384 		.has_inbox = true,
1385 		.has_outbox = false,
1386 		.out_is_imm = false,
1387 		.encode_slave_id = false,
1388 		.verify = NULL,
1389 		.wrapper = mlx4_SQERR2RTS_QP_wrapper
1390 	},
1391 	{
1392 		.opcode = MLX4_CMD_2ERR_QP,
1393 		.has_inbox = false,
1394 		.has_outbox = false,
1395 		.out_is_imm = false,
1396 		.encode_slave_id = false,
1397 		.verify = NULL,
1398 		.wrapper = mlx4_GEN_QP_wrapper
1399 	},
1400 	{
1401 		.opcode = MLX4_CMD_RTS2SQD_QP,
1402 		.has_inbox = false,
1403 		.has_outbox = false,
1404 		.out_is_imm = false,
1405 		.encode_slave_id = false,
1406 		.verify = NULL,
1407 		.wrapper = mlx4_GEN_QP_wrapper
1408 	},
1409 	{
1410 		.opcode = MLX4_CMD_SQD2SQD_QP,
1411 		.has_inbox = true,
1412 		.has_outbox = false,
1413 		.out_is_imm = false,
1414 		.encode_slave_id = false,
1415 		.verify = NULL,
1416 		.wrapper = mlx4_SQD2SQD_QP_wrapper
1417 	},
1418 	{
1419 		.opcode = MLX4_CMD_SQD2RTS_QP,
1420 		.has_inbox = true,
1421 		.has_outbox = false,
1422 		.out_is_imm = false,
1423 		.encode_slave_id = false,
1424 		.verify = NULL,
1425 		.wrapper = mlx4_SQD2RTS_QP_wrapper
1426 	},
1427 	{
1428 		.opcode = MLX4_CMD_2RST_QP,
1429 		.has_inbox = false,
1430 		.has_outbox = false,
1431 		.out_is_imm = false,
1432 		.encode_slave_id = false,
1433 		.verify = NULL,
1434 		.wrapper = mlx4_2RST_QP_wrapper
1435 	},
1436 	{
1437 		.opcode = MLX4_CMD_QUERY_QP,
1438 		.has_inbox = false,
1439 		.has_outbox = true,
1440 		.out_is_imm = false,
1441 		.encode_slave_id = false,
1442 		.verify = NULL,
1443 		.wrapper = mlx4_GEN_QP_wrapper
1444 	},
1445 	{
1446 		.opcode = MLX4_CMD_SUSPEND_QP,
1447 		.has_inbox = false,
1448 		.has_outbox = false,
1449 		.out_is_imm = false,
1450 		.encode_slave_id = false,
1451 		.verify = NULL,
1452 		.wrapper = mlx4_GEN_QP_wrapper
1453 	},
1454 	{
1455 		.opcode = MLX4_CMD_UNSUSPEND_QP,
1456 		.has_inbox = false,
1457 		.has_outbox = false,
1458 		.out_is_imm = false,
1459 		.encode_slave_id = false,
1460 		.verify = NULL,
1461 		.wrapper = mlx4_GEN_QP_wrapper
1462 	},
1463 	{
1464 		.opcode = MLX4_CMD_UPDATE_QP,
1465 		.has_inbox = true,
1466 		.has_outbox = false,
1467 		.out_is_imm = false,
1468 		.encode_slave_id = false,
1469 		.verify = NULL,
1470 		.wrapper = mlx4_UPDATE_QP_wrapper
1471 	},
1472 	{
1473 		.opcode = MLX4_CMD_GET_OP_REQ,
1474 		.has_inbox = false,
1475 		.has_outbox = false,
1476 		.out_is_imm = false,
1477 		.encode_slave_id = false,
1478 		.verify = NULL,
1479 		.wrapper = mlx4_CMD_EPERM_wrapper,
1480 	},
1481 	{
1482 		.opcode = MLX4_CMD_ALLOCATE_VPP,
1483 		.has_inbox = false,
1484 		.has_outbox = true,
1485 		.out_is_imm = false,
1486 		.encode_slave_id = false,
1487 		.verify = NULL,
1488 		.wrapper = mlx4_CMD_EPERM_wrapper,
1489 	},
1490 	{
1491 		.opcode = MLX4_CMD_SET_VPORT_QOS,
1492 		.has_inbox = false,
1493 		.has_outbox = true,
1494 		.out_is_imm = false,
1495 		.encode_slave_id = false,
1496 		.verify = NULL,
1497 		.wrapper = mlx4_CMD_EPERM_wrapper,
1498 	},
1499 	{
1500 		.opcode = MLX4_CMD_CONF_SPECIAL_QP,
1501 		.has_inbox = false,
1502 		.has_outbox = false,
1503 		.out_is_imm = false,
1504 		.encode_slave_id = false,
1505 		.verify = NULL, /* XXX verify: only demux can do this */
1506 		.wrapper = NULL
1507 	},
1508 	{
1509 		.opcode = MLX4_CMD_MAD_IFC,
1510 		.has_inbox = true,
1511 		.has_outbox = true,
1512 		.out_is_imm = false,
1513 		.encode_slave_id = false,
1514 		.verify = NULL,
1515 		.wrapper = mlx4_MAD_IFC_wrapper
1516 	},
1517 	{
1518 		.opcode = MLX4_CMD_MAD_DEMUX,
1519 		.has_inbox = false,
1520 		.has_outbox = false,
1521 		.out_is_imm = false,
1522 		.encode_slave_id = false,
1523 		.verify = NULL,
1524 		.wrapper = mlx4_CMD_EPERM_wrapper
1525 	},
1526 	{
1527 		.opcode = MLX4_CMD_QUERY_IF_STAT,
1528 		.has_inbox = false,
1529 		.has_outbox = true,
1530 		.out_is_imm = false,
1531 		.encode_slave_id = false,
1532 		.verify = NULL,
1533 		.wrapper = mlx4_QUERY_IF_STAT_wrapper
1534 	},
1535 	{
1536 		.opcode = MLX4_CMD_ACCESS_REG,
1537 		.has_inbox = true,
1538 		.has_outbox = true,
1539 		.out_is_imm = false,
1540 		.encode_slave_id = false,
1541 		.verify = NULL,
1542 		.wrapper = mlx4_ACCESS_REG_wrapper,
1543 	},
1544 	{
1545 		.opcode = MLX4_CMD_CONGESTION_CTRL_OPCODE,
1546 		.has_inbox = false,
1547 		.has_outbox = false,
1548 		.out_is_imm = false,
1549 		.encode_slave_id = false,
1550 		.verify = NULL,
1551 		.wrapper = mlx4_CMD_EPERM_wrapper,
1552 	},
1553 	/* Native multicast commands are not available for guests */
1554 	{
1555 		.opcode = MLX4_CMD_QP_ATTACH,
1556 		.has_inbox = true,
1557 		.has_outbox = false,
1558 		.out_is_imm = false,
1559 		.encode_slave_id = false,
1560 		.verify = NULL,
1561 		.wrapper = mlx4_QP_ATTACH_wrapper
1562 	},
1563 	{
1564 		.opcode = MLX4_CMD_PROMISC,
1565 		.has_inbox = false,
1566 		.has_outbox = false,
1567 		.out_is_imm = false,
1568 		.encode_slave_id = false,
1569 		.verify = NULL,
1570 		.wrapper = mlx4_PROMISC_wrapper
1571 	},
1572 	/* Ethernet specific commands */
1573 	{
1574 		.opcode = MLX4_CMD_SET_VLAN_FLTR,
1575 		.has_inbox = true,
1576 		.has_outbox = false,
1577 		.out_is_imm = false,
1578 		.encode_slave_id = false,
1579 		.verify = NULL,
1580 		.wrapper = mlx4_SET_VLAN_FLTR_wrapper
1581 	},
1582 	{
1583 		.opcode = MLX4_CMD_SET_MCAST_FLTR,
1584 		.has_inbox = false,
1585 		.has_outbox = false,
1586 		.out_is_imm = false,
1587 		.encode_slave_id = false,
1588 		.verify = NULL,
1589 		.wrapper = mlx4_SET_MCAST_FLTR_wrapper
1590 	},
1591 	{
1592 		.opcode = MLX4_CMD_DUMP_ETH_STATS,
1593 		.has_inbox = false,
1594 		.has_outbox = true,
1595 		.out_is_imm = false,
1596 		.encode_slave_id = false,
1597 		.verify = NULL,
1598 		.wrapper = mlx4_DUMP_ETH_STATS_wrapper
1599 	},
1600 	{
1601 		.opcode = MLX4_CMD_INFORM_FLR_DONE,
1602 		.has_inbox = false,
1603 		.has_outbox = false,
1604 		.out_is_imm = false,
1605 		.encode_slave_id = false,
1606 		.verify = NULL,
1607 		.wrapper = NULL
1608 	},
1609 	/* flow steering commands */
1610 	{
1611 		.opcode = MLX4_QP_FLOW_STEERING_ATTACH,
1612 		.has_inbox = true,
1613 		.has_outbox = false,
1614 		.out_is_imm = true,
1615 		.encode_slave_id = false,
1616 		.verify = NULL,
1617 		.wrapper = mlx4_QP_FLOW_STEERING_ATTACH_wrapper
1618 	},
1619 	{
1620 		.opcode = MLX4_QP_FLOW_STEERING_DETACH,
1621 		.has_inbox = false,
1622 		.has_outbox = false,
1623 		.out_is_imm = false,
1624 		.encode_slave_id = false,
1625 		.verify = NULL,
1626 		.wrapper = mlx4_QP_FLOW_STEERING_DETACH_wrapper
1627 	},
1628 	{
1629 		.opcode = MLX4_FLOW_STEERING_IB_UC_QP_RANGE,
1630 		.has_inbox = false,
1631 		.has_outbox = false,
1632 		.out_is_imm = false,
1633 		.encode_slave_id = false,
1634 		.verify = NULL,
1635 		.wrapper = mlx4_CMD_EPERM_wrapper
1636 	},
1637 	{
1638 		.opcode = MLX4_CMD_VIRT_PORT_MAP,
1639 		.has_inbox = false,
1640 		.has_outbox = false,
1641 		.out_is_imm = false,
1642 		.encode_slave_id = false,
1643 		.verify = NULL,
1644 		.wrapper = mlx4_CMD_EPERM_wrapper
1645 	},
1646 };
1647 
1648 static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
1649 				    struct mlx4_vhcr_cmd *in_vhcr)
1650 {
1651 	struct mlx4_priv *priv = mlx4_priv(dev);
1652 	struct mlx4_cmd_info *cmd = NULL;
1653 	struct mlx4_vhcr_cmd *vhcr_cmd = in_vhcr ? in_vhcr : priv->mfunc.vhcr;
1654 	struct mlx4_vhcr *vhcr;
1655 	struct mlx4_cmd_mailbox *inbox = NULL;
1656 	struct mlx4_cmd_mailbox *outbox = NULL;
1657 	u64 in_param;
1658 	u64 out_param;
1659 	int ret = 0;
1660 	int i;
1661 	int err = 0;
1662 
1663 	/* Create sw representation of Virtual HCR */
1664 	vhcr = kzalloc(sizeof(struct mlx4_vhcr), GFP_KERNEL);
1665 	if (!vhcr)
1666 		return -ENOMEM;
1667 
1668 	/* DMA in the vHCR */
1669 	if (!in_vhcr) {
1670 		ret = mlx4_ACCESS_MEM(dev, priv->mfunc.vhcr_dma, slave,
1671 				      priv->mfunc.master.slave_state[slave].vhcr_dma,
1672 				      ALIGN(sizeof(struct mlx4_vhcr_cmd),
1673 					    MLX4_ACCESS_MEM_ALIGN), 1);
1674 		if (ret) {
1675 			if (!(dev->persist->state &
1676 			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
1677 				mlx4_err(dev, "%s: Failed reading vhcr ret: 0x%x\n",
1678 					 __func__, ret);
1679 			kfree(vhcr);
1680 			return ret;
1681 		}
1682 	}
1683 
1684 	/* Fill SW VHCR fields */
1685 	vhcr->in_param = be64_to_cpu(vhcr_cmd->in_param);
1686 	vhcr->out_param = be64_to_cpu(vhcr_cmd->out_param);
1687 	vhcr->in_modifier = be32_to_cpu(vhcr_cmd->in_modifier);
1688 	vhcr->token = be16_to_cpu(vhcr_cmd->token);
1689 	vhcr->op = be16_to_cpu(vhcr_cmd->opcode) & 0xfff;
1690 	vhcr->op_modifier = (u8) (be16_to_cpu(vhcr_cmd->opcode) >> 12);
1691 	vhcr->e_bit = vhcr_cmd->flags & (1 << 6);
1692 
1693 	/* Lookup command */
1694 	for (i = 0; i < ARRAY_SIZE(cmd_info); ++i) {
1695 		if (vhcr->op == cmd_info[i].opcode) {
1696 			cmd = &cmd_info[i];
1697 			break;
1698 		}
1699 	}
1700 	if (!cmd) {
1701 		mlx4_err(dev, "Unknown command:0x%x accepted from slave:%d\n",
1702 			 vhcr->op, slave);
1703 		vhcr_cmd->status = CMD_STAT_BAD_PARAM;
1704 		goto out_status;
1705 	}
1706 
1707 	/* Read inbox */
1708 	if (cmd->has_inbox) {
1709 		vhcr->in_param &= INBOX_MASK;
1710 		inbox = mlx4_alloc_cmd_mailbox(dev);
1711 		if (IS_ERR(inbox)) {
1712 			vhcr_cmd->status = CMD_STAT_BAD_SIZE;
1713 			inbox = NULL;
1714 			goto out_status;
1715 		}
1716 
1717 		ret = mlx4_ACCESS_MEM(dev, inbox->dma, slave,
1718 				      vhcr->in_param,
1719 				      MLX4_MAILBOX_SIZE, 1);
1720 		if (ret) {
1721 			if (!(dev->persist->state &
1722 			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
1723 				mlx4_err(dev, "%s: Failed reading inbox (cmd:0x%x)\n",
1724 					 __func__, cmd->opcode);
1725 			vhcr_cmd->status = CMD_STAT_INTERNAL_ERR;
1726 			goto out_status;
1727 		}
1728 	}
1729 
1730 	/* Apply permission and bound checks if applicable */
1731 	if (cmd->verify && cmd->verify(dev, slave, vhcr, inbox)) {
1732 		mlx4_warn(dev, "Command:0x%x from slave: %d failed protection checks for resource_id:%d\n",
1733 			  vhcr->op, slave, vhcr->in_modifier);
1734 		vhcr_cmd->status = CMD_STAT_BAD_OP;
1735 		goto out_status;
1736 	}
1737 
1738 	/* Allocate outbox */
1739 	if (cmd->has_outbox) {
1740 		outbox = mlx4_alloc_cmd_mailbox(dev);
1741 		if (IS_ERR(outbox)) {
1742 			vhcr_cmd->status = CMD_STAT_BAD_SIZE;
1743 			outbox = NULL;
1744 			goto out_status;
1745 		}
1746 	}
1747 
1748 	/* Execute the command! */
1749 	if (cmd->wrapper) {
1750 		err = cmd->wrapper(dev, slave, vhcr, inbox, outbox,
1751 				   cmd);
1752 		if (cmd->out_is_imm)
1753 			vhcr_cmd->out_param = cpu_to_be64(vhcr->out_param);
1754 	} else {
1755 		in_param = cmd->has_inbox ? (u64) inbox->dma :
1756 			vhcr->in_param;
1757 		out_param = cmd->has_outbox ? (u64) outbox->dma :
1758 			vhcr->out_param;
1759 		err = __mlx4_cmd(dev, in_param, &out_param,
1760 				 cmd->out_is_imm, vhcr->in_modifier,
1761 				 vhcr->op_modifier, vhcr->op,
1762 				 MLX4_CMD_TIME_CLASS_A,
1763 				 MLX4_CMD_NATIVE);
1764 
1765 		if (cmd->out_is_imm) {
1766 			vhcr->out_param = out_param;
1767 			vhcr_cmd->out_param = cpu_to_be64(vhcr->out_param);
1768 		}
1769 	}
1770 
1771 	if (err) {
1772 		if (!(dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR))
1773 			mlx4_warn(dev, "vhcr command:0x%x slave:%d failed with error:%d, status %d\n",
1774 				  vhcr->op, slave, vhcr->errno, err);
1775 		vhcr_cmd->status = mlx4_errno_to_status(err);
1776 		goto out_status;
1777 	}
1778 
1779 
1780 	/* Write outbox if command completed successfully */
1781 	if (cmd->has_outbox && !vhcr_cmd->status) {
1782 		ret = mlx4_ACCESS_MEM(dev, outbox->dma, slave,
1783 				      vhcr->out_param,
1784 				      MLX4_MAILBOX_SIZE, MLX4_CMD_WRAPPED);
1785 		if (ret) {
1786 			/* If we failed to write back the outbox after the
1787 			 *command was successfully executed, we must fail this
1788 			 * slave, as it is now in undefined state */
1789 			if (!(dev->persist->state &
1790 			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
1791 				mlx4_err(dev, "%s:Failed writing outbox\n", __func__);
1792 			goto out;
1793 		}
1794 	}
1795 
1796 out_status:
1797 	/* DMA back vhcr result */
1798 	if (!in_vhcr) {
1799 		ret = mlx4_ACCESS_MEM(dev, priv->mfunc.vhcr_dma, slave,
1800 				      priv->mfunc.master.slave_state[slave].vhcr_dma,
1801 				      ALIGN(sizeof(struct mlx4_vhcr),
1802 					    MLX4_ACCESS_MEM_ALIGN),
1803 				      MLX4_CMD_WRAPPED);
1804 		if (ret)
1805 			mlx4_err(dev, "%s:Failed writing vhcr result\n",
1806 				 __func__);
1807 		else if (vhcr->e_bit &&
1808 			 mlx4_GEN_EQE(dev, slave, &priv->mfunc.master.cmd_eqe))
1809 				mlx4_warn(dev, "Failed to generate command completion eqe for slave %d\n",
1810 					  slave);
1811 	}
1812 
1813 out:
1814 	kfree(vhcr);
1815 	mlx4_free_cmd_mailbox(dev, inbox);
1816 	mlx4_free_cmd_mailbox(dev, outbox);
1817 	return ret;
1818 }
1819 
1820 static int mlx4_master_immediate_activate_vlan_qos(struct mlx4_priv *priv,
1821 					    int slave, int port)
1822 {
1823 	struct mlx4_vport_oper_state *vp_oper;
1824 	struct mlx4_vport_state *vp_admin;
1825 	struct mlx4_vf_immed_vlan_work *work;
1826 	struct mlx4_dev *dev = &(priv->dev);
1827 	int err;
1828 	int admin_vlan_ix = NO_INDX;
1829 
1830 	vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
1831 	vp_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
1832 
1833 	if (vp_oper->state.default_vlan == vp_admin->default_vlan &&
1834 	    vp_oper->state.default_qos == vp_admin->default_qos &&
1835 	    vp_oper->state.link_state == vp_admin->link_state &&
1836 	    vp_oper->state.qos_vport == vp_admin->qos_vport)
1837 		return 0;
1838 
1839 	if (!(priv->mfunc.master.slave_state[slave].active &&
1840 	      dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_UPDATE_QP)) {
1841 		/* even if the UPDATE_QP command isn't supported, we still want
1842 		 * to set this VF link according to the admin directive
1843 		 */
1844 		vp_oper->state.link_state = vp_admin->link_state;
1845 		return -1;
1846 	}
1847 
1848 	mlx4_dbg(dev, "updating immediately admin params slave %d port %d\n",
1849 		 slave, port);
1850 	mlx4_dbg(dev, "vlan %d QoS %d link down %d\n",
1851 		 vp_admin->default_vlan, vp_admin->default_qos,
1852 		 vp_admin->link_state);
1853 
1854 	work = kzalloc(sizeof(*work), GFP_KERNEL);
1855 	if (!work)
1856 		return -ENOMEM;
1857 
1858 	if (vp_oper->state.default_vlan != vp_admin->default_vlan) {
1859 		if (MLX4_VGT != vp_admin->default_vlan) {
1860 			err = __mlx4_register_vlan(&priv->dev, port,
1861 						   vp_admin->default_vlan,
1862 						   &admin_vlan_ix);
1863 			if (err) {
1864 				kfree(work);
1865 				mlx4_warn(&priv->dev,
1866 					  "No vlan resources slave %d, port %d\n",
1867 					  slave, port);
1868 				return err;
1869 			}
1870 		} else {
1871 			admin_vlan_ix = NO_INDX;
1872 		}
1873 		work->flags |= MLX4_VF_IMMED_VLAN_FLAG_VLAN;
1874 		mlx4_dbg(&priv->dev,
1875 			 "alloc vlan %d idx  %d slave %d port %d\n",
1876 			 (int)(vp_admin->default_vlan),
1877 			 admin_vlan_ix, slave, port);
1878 	}
1879 
1880 	/* save original vlan ix and vlan id */
1881 	work->orig_vlan_id = vp_oper->state.default_vlan;
1882 	work->orig_vlan_ix = vp_oper->vlan_idx;
1883 
1884 	/* handle new qos */
1885 	if (vp_oper->state.default_qos != vp_admin->default_qos)
1886 		work->flags |= MLX4_VF_IMMED_VLAN_FLAG_QOS;
1887 
1888 	if (work->flags & MLX4_VF_IMMED_VLAN_FLAG_VLAN)
1889 		vp_oper->vlan_idx = admin_vlan_ix;
1890 
1891 	vp_oper->state.default_vlan = vp_admin->default_vlan;
1892 	vp_oper->state.default_qos = vp_admin->default_qos;
1893 	vp_oper->state.link_state = vp_admin->link_state;
1894 	vp_oper->state.qos_vport = vp_admin->qos_vport;
1895 
1896 	if (vp_admin->link_state == IFLA_VF_LINK_STATE_DISABLE)
1897 		work->flags |= MLX4_VF_IMMED_VLAN_FLAG_LINK_DISABLE;
1898 
1899 	/* iterate over QPs owned by this slave, using UPDATE_QP */
1900 	work->port = port;
1901 	work->slave = slave;
1902 	work->qos = vp_oper->state.default_qos;
1903 	work->qos_vport = vp_oper->state.qos_vport;
1904 	work->vlan_id = vp_oper->state.default_vlan;
1905 	work->vlan_ix = vp_oper->vlan_idx;
1906 	work->priv = priv;
1907 	INIT_WORK(&work->work, mlx4_vf_immed_vlan_work_handler);
1908 	queue_work(priv->mfunc.master.comm_wq, &work->work);
1909 
1910 	return 0;
1911 }
1912 
1913 static void mlx4_set_default_port_qos(struct mlx4_dev *dev, int port)
1914 {
1915 	struct mlx4_qos_manager *port_qos_ctl;
1916 	struct mlx4_priv *priv = mlx4_priv(dev);
1917 
1918 	port_qos_ctl = &priv->mfunc.master.qos_ctl[port];
1919 	bitmap_zero(port_qos_ctl->priority_bm, MLX4_NUM_UP);
1920 
1921 	/* Enable only default prio at PF init routine */
1922 	set_bit(MLX4_DEFAULT_QOS_PRIO, port_qos_ctl->priority_bm);
1923 }
1924 
1925 static void mlx4_allocate_port_vpps(struct mlx4_dev *dev, int port)
1926 {
1927 	int i;
1928 	int err;
1929 	int num_vfs;
1930 	u16 availible_vpp;
1931 	u8 vpp_param[MLX4_NUM_UP];
1932 	struct mlx4_qos_manager *port_qos;
1933 	struct mlx4_priv *priv = mlx4_priv(dev);
1934 
1935 	err = mlx4_ALLOCATE_VPP_get(dev, port, &availible_vpp, vpp_param);
1936 	if (err) {
1937 		mlx4_info(dev, "Failed query availible VPPs\n");
1938 		return;
1939 	}
1940 
1941 	port_qos = &priv->mfunc.master.qos_ctl[port];
1942 	num_vfs = (availible_vpp /
1943 		   bitmap_weight(port_qos->priority_bm, MLX4_NUM_UP));
1944 
1945 	for (i = 0; i < MLX4_NUM_UP; i++) {
1946 		if (test_bit(i, port_qos->priority_bm))
1947 			vpp_param[i] = num_vfs;
1948 	}
1949 
1950 	err = mlx4_ALLOCATE_VPP_set(dev, port, vpp_param);
1951 	if (err) {
1952 		mlx4_info(dev, "Failed allocating VPPs\n");
1953 		return;
1954 	}
1955 
1956 	/* Query actual allocated VPP, just to make sure */
1957 	err = mlx4_ALLOCATE_VPP_get(dev, port, &availible_vpp, vpp_param);
1958 	if (err) {
1959 		mlx4_info(dev, "Failed query availible VPPs\n");
1960 		return;
1961 	}
1962 
1963 	port_qos->num_of_qos_vfs = num_vfs;
1964 	mlx4_dbg(dev, "Port %d Availible VPPs %d\n", port, availible_vpp);
1965 
1966 	for (i = 0; i < MLX4_NUM_UP; i++)
1967 		mlx4_dbg(dev, "Port %d UP %d Allocated %d VPPs\n", port, i,
1968 			 vpp_param[i]);
1969 }
1970 
1971 static int mlx4_master_activate_admin_state(struct mlx4_priv *priv, int slave)
1972 {
1973 	int port, err;
1974 	struct mlx4_vport_state *vp_admin;
1975 	struct mlx4_vport_oper_state *vp_oper;
1976 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(
1977 			&priv->dev, slave);
1978 	int min_port = find_first_bit(actv_ports.ports,
1979 				      priv->dev.caps.num_ports) + 1;
1980 	int max_port = min_port - 1 +
1981 		bitmap_weight(actv_ports.ports, priv->dev.caps.num_ports);
1982 
1983 	for (port = min_port; port <= max_port; port++) {
1984 		if (!test_bit(port - 1, actv_ports.ports))
1985 			continue;
1986 		priv->mfunc.master.vf_oper[slave].smi_enabled[port] =
1987 			priv->mfunc.master.vf_admin[slave].enable_smi[port];
1988 		vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
1989 		vp_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
1990 		vp_oper->state = *vp_admin;
1991 		if (MLX4_VGT != vp_admin->default_vlan) {
1992 			err = __mlx4_register_vlan(&priv->dev, port,
1993 						   vp_admin->default_vlan, &(vp_oper->vlan_idx));
1994 			if (err) {
1995 				vp_oper->vlan_idx = NO_INDX;
1996 				mlx4_warn(&priv->dev,
1997 					  "No vlan resources slave %d, port %d\n",
1998 					  slave, port);
1999 				return err;
2000 			}
2001 			mlx4_dbg(&priv->dev, "alloc vlan %d idx  %d slave %d port %d\n",
2002 				 (int)(vp_oper->state.default_vlan),
2003 				 vp_oper->vlan_idx, slave, port);
2004 		}
2005 		if (vp_admin->spoofchk) {
2006 			vp_oper->mac_idx = __mlx4_register_mac(&priv->dev,
2007 							       port,
2008 							       vp_admin->mac);
2009 			if (0 > vp_oper->mac_idx) {
2010 				err = vp_oper->mac_idx;
2011 				vp_oper->mac_idx = NO_INDX;
2012 				mlx4_warn(&priv->dev,
2013 					  "No mac resources slave %d, port %d\n",
2014 					  slave, port);
2015 				return err;
2016 			}
2017 			mlx4_dbg(&priv->dev, "alloc mac %llx idx  %d slave %d port %d\n",
2018 				 vp_oper->state.mac, vp_oper->mac_idx, slave, port);
2019 		}
2020 	}
2021 	return 0;
2022 }
2023 
2024 static void mlx4_master_deactivate_admin_state(struct mlx4_priv *priv, int slave)
2025 {
2026 	int port;
2027 	struct mlx4_vport_oper_state *vp_oper;
2028 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(
2029 			&priv->dev, slave);
2030 	int min_port = find_first_bit(actv_ports.ports,
2031 				      priv->dev.caps.num_ports) + 1;
2032 	int max_port = min_port - 1 +
2033 		bitmap_weight(actv_ports.ports, priv->dev.caps.num_ports);
2034 
2035 
2036 	for (port = min_port; port <= max_port; port++) {
2037 		if (!test_bit(port - 1, actv_ports.ports))
2038 			continue;
2039 		priv->mfunc.master.vf_oper[slave].smi_enabled[port] =
2040 			MLX4_VF_SMI_DISABLED;
2041 		vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
2042 		if (NO_INDX != vp_oper->vlan_idx) {
2043 			__mlx4_unregister_vlan(&priv->dev,
2044 					       port, vp_oper->state.default_vlan);
2045 			vp_oper->vlan_idx = NO_INDX;
2046 		}
2047 		if (NO_INDX != vp_oper->mac_idx) {
2048 			__mlx4_unregister_mac(&priv->dev, port, vp_oper->state.mac);
2049 			vp_oper->mac_idx = NO_INDX;
2050 		}
2051 	}
2052 	return;
2053 }
2054 
2055 static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd,
2056 			       u16 param, u8 toggle)
2057 {
2058 	struct mlx4_priv *priv = mlx4_priv(dev);
2059 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
2060 	u32 reply;
2061 	u8 is_going_down = 0;
2062 	int i;
2063 	unsigned long flags;
2064 
2065 	slave_state[slave].comm_toggle ^= 1;
2066 	reply = (u32) slave_state[slave].comm_toggle << 31;
2067 	if (toggle != slave_state[slave].comm_toggle) {
2068 		mlx4_warn(dev, "Incorrect toggle %d from slave %d. *** MASTER STATE COMPROMISED ***\n",
2069 			  toggle, slave);
2070 		goto reset_slave;
2071 	}
2072 	if (cmd == MLX4_COMM_CMD_RESET) {
2073 		mlx4_warn(dev, "Received reset from slave:%d\n", slave);
2074 		slave_state[slave].active = false;
2075 		slave_state[slave].old_vlan_api = false;
2076 		mlx4_master_deactivate_admin_state(priv, slave);
2077 		for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i) {
2078 				slave_state[slave].event_eq[i].eqn = -1;
2079 				slave_state[slave].event_eq[i].token = 0;
2080 		}
2081 		/*check if we are in the middle of FLR process,
2082 		if so return "retry" status to the slave*/
2083 		if (MLX4_COMM_CMD_FLR == slave_state[slave].last_cmd)
2084 			goto inform_slave_state;
2085 
2086 		mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_SHUTDOWN, slave);
2087 
2088 		/* write the version in the event field */
2089 		reply |= mlx4_comm_get_version();
2090 
2091 		goto reset_slave;
2092 	}
2093 	/*command from slave in the middle of FLR*/
2094 	if (cmd != MLX4_COMM_CMD_RESET &&
2095 	    MLX4_COMM_CMD_FLR == slave_state[slave].last_cmd) {
2096 		mlx4_warn(dev, "slave:%d is Trying to run cmd(0x%x) in the middle of FLR\n",
2097 			  slave, cmd);
2098 		return;
2099 	}
2100 
2101 	switch (cmd) {
2102 	case MLX4_COMM_CMD_VHCR0:
2103 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_RESET)
2104 			goto reset_slave;
2105 		slave_state[slave].vhcr_dma = ((u64) param) << 48;
2106 		priv->mfunc.master.slave_state[slave].cookie = 0;
2107 		break;
2108 	case MLX4_COMM_CMD_VHCR1:
2109 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR0)
2110 			goto reset_slave;
2111 		slave_state[slave].vhcr_dma |= ((u64) param) << 32;
2112 		break;
2113 	case MLX4_COMM_CMD_VHCR2:
2114 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR1)
2115 			goto reset_slave;
2116 		slave_state[slave].vhcr_dma |= ((u64) param) << 16;
2117 		break;
2118 	case MLX4_COMM_CMD_VHCR_EN:
2119 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR2)
2120 			goto reset_slave;
2121 		slave_state[slave].vhcr_dma |= param;
2122 		if (mlx4_master_activate_admin_state(priv, slave))
2123 				goto reset_slave;
2124 		slave_state[slave].active = true;
2125 		mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_INIT, slave);
2126 		break;
2127 	case MLX4_COMM_CMD_VHCR_POST:
2128 		if ((slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_EN) &&
2129 		    (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_POST)) {
2130 			mlx4_warn(dev, "slave:%d is out of sync, cmd=0x%x, last command=0x%x, reset is needed\n",
2131 				  slave, cmd, slave_state[slave].last_cmd);
2132 			goto reset_slave;
2133 		}
2134 
2135 		mutex_lock(&priv->cmd.slave_cmd_mutex);
2136 		if (mlx4_master_process_vhcr(dev, slave, NULL)) {
2137 			mlx4_err(dev, "Failed processing vhcr for slave:%d, resetting slave\n",
2138 				 slave);
2139 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2140 			goto reset_slave;
2141 		}
2142 		mutex_unlock(&priv->cmd.slave_cmd_mutex);
2143 		break;
2144 	default:
2145 		mlx4_warn(dev, "Bad comm cmd:%d from slave:%d\n", cmd, slave);
2146 		goto reset_slave;
2147 	}
2148 	spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
2149 	if (!slave_state[slave].is_slave_going_down)
2150 		slave_state[slave].last_cmd = cmd;
2151 	else
2152 		is_going_down = 1;
2153 	spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
2154 	if (is_going_down) {
2155 		mlx4_warn(dev, "Slave is going down aborting command(%d) executing from slave:%d\n",
2156 			  cmd, slave);
2157 		return;
2158 	}
2159 	__raw_writel((__force u32) cpu_to_be32(reply),
2160 		     &priv->mfunc.comm[slave].slave_read);
2161 	mmiowb();
2162 
2163 	return;
2164 
2165 reset_slave:
2166 	/* cleanup any slave resources */
2167 	if (dev->persist->interface_state & MLX4_INTERFACE_STATE_UP)
2168 		mlx4_delete_all_resources_for_slave(dev, slave);
2169 
2170 	if (cmd != MLX4_COMM_CMD_RESET) {
2171 		mlx4_warn(dev, "Turn on internal error to force reset, slave=%d, cmd=0x%x\n",
2172 			  slave, cmd);
2173 		/* Turn on internal error letting slave reset itself immeditaly,
2174 		 * otherwise it might take till timeout on command is passed
2175 		 */
2176 		reply |= ((u32)COMM_CHAN_EVENT_INTERNAL_ERR);
2177 	}
2178 
2179 	spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
2180 	if (!slave_state[slave].is_slave_going_down)
2181 		slave_state[slave].last_cmd = MLX4_COMM_CMD_RESET;
2182 	spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
2183 	/*with slave in the middle of flr, no need to clean resources again.*/
2184 inform_slave_state:
2185 	memset(&slave_state[slave].event_eq, 0,
2186 	       sizeof(struct mlx4_slave_event_eq_info));
2187 	__raw_writel((__force u32) cpu_to_be32(reply),
2188 		     &priv->mfunc.comm[slave].slave_read);
2189 	wmb();
2190 }
2191 
2192 /* master command processing */
2193 void mlx4_master_comm_channel(struct work_struct *work)
2194 {
2195 	struct mlx4_mfunc_master_ctx *master =
2196 		container_of(work,
2197 			     struct mlx4_mfunc_master_ctx,
2198 			     comm_work);
2199 	struct mlx4_mfunc *mfunc =
2200 		container_of(master, struct mlx4_mfunc, master);
2201 	struct mlx4_priv *priv =
2202 		container_of(mfunc, struct mlx4_priv, mfunc);
2203 	struct mlx4_dev *dev = &priv->dev;
2204 	__be32 *bit_vec;
2205 	u32 comm_cmd;
2206 	u32 vec;
2207 	int i, j, slave;
2208 	int toggle;
2209 	int served = 0;
2210 	int reported = 0;
2211 	u32 slt;
2212 
2213 	bit_vec = master->comm_arm_bit_vector;
2214 	for (i = 0; i < COMM_CHANNEL_BIT_ARRAY_SIZE; i++) {
2215 		vec = be32_to_cpu(bit_vec[i]);
2216 		for (j = 0; j < 32; j++) {
2217 			if (!(vec & (1 << j)))
2218 				continue;
2219 			++reported;
2220 			slave = (i * 32) + j;
2221 			comm_cmd = swab32(readl(
2222 					  &mfunc->comm[slave].slave_write));
2223 			slt = swab32(readl(&mfunc->comm[slave].slave_read))
2224 				     >> 31;
2225 			toggle = comm_cmd >> 31;
2226 			if (toggle != slt) {
2227 				if (master->slave_state[slave].comm_toggle
2228 				    != slt) {
2229 					pr_info("slave %d out of sync. read toggle %d, state toggle %d. Resynching.\n",
2230 						slave, slt,
2231 						master->slave_state[slave].comm_toggle);
2232 					master->slave_state[slave].comm_toggle =
2233 						slt;
2234 				}
2235 				mlx4_master_do_cmd(dev, slave,
2236 						   comm_cmd >> 16 & 0xff,
2237 						   comm_cmd & 0xffff, toggle);
2238 				++served;
2239 			}
2240 		}
2241 	}
2242 
2243 	if (reported && reported != served)
2244 		mlx4_warn(dev, "Got command event with bitmask from %d slaves but %d were served\n",
2245 			  reported, served);
2246 
2247 	if (mlx4_ARM_COMM_CHANNEL(dev))
2248 		mlx4_warn(dev, "Failed to arm comm channel events\n");
2249 }
2250 
2251 static int sync_toggles(struct mlx4_dev *dev)
2252 {
2253 	struct mlx4_priv *priv = mlx4_priv(dev);
2254 	u32 wr_toggle;
2255 	u32 rd_toggle;
2256 	unsigned long end;
2257 
2258 	wr_toggle = swab32(readl(&priv->mfunc.comm->slave_write));
2259 	if (wr_toggle == 0xffffffff)
2260 		end = jiffies + msecs_to_jiffies(30000);
2261 	else
2262 		end = jiffies + msecs_to_jiffies(5000);
2263 
2264 	while (time_before(jiffies, end)) {
2265 		rd_toggle = swab32(readl(&priv->mfunc.comm->slave_read));
2266 		if (wr_toggle == 0xffffffff || rd_toggle == 0xffffffff) {
2267 			/* PCI might be offline */
2268 			msleep(100);
2269 			wr_toggle = swab32(readl(&priv->mfunc.comm->
2270 					   slave_write));
2271 			continue;
2272 		}
2273 
2274 		if (rd_toggle >> 31 == wr_toggle >> 31) {
2275 			priv->cmd.comm_toggle = rd_toggle >> 31;
2276 			return 0;
2277 		}
2278 
2279 		cond_resched();
2280 	}
2281 
2282 	/*
2283 	 * we could reach here if for example the previous VM using this
2284 	 * function misbehaved and left the channel with unsynced state. We
2285 	 * should fix this here and give this VM a chance to use a properly
2286 	 * synced channel
2287 	 */
2288 	mlx4_warn(dev, "recovering from previously mis-behaved VM\n");
2289 	__raw_writel((__force u32) 0, &priv->mfunc.comm->slave_read);
2290 	__raw_writel((__force u32) 0, &priv->mfunc.comm->slave_write);
2291 	priv->cmd.comm_toggle = 0;
2292 
2293 	return 0;
2294 }
2295 
2296 int mlx4_multi_func_init(struct mlx4_dev *dev)
2297 {
2298 	struct mlx4_priv *priv = mlx4_priv(dev);
2299 	struct mlx4_slave_state *s_state;
2300 	int i, j, err, port;
2301 
2302 	if (mlx4_is_master(dev))
2303 		priv->mfunc.comm =
2304 		ioremap(pci_resource_start(dev->persist->pdev,
2305 					   priv->fw.comm_bar) +
2306 			priv->fw.comm_base, MLX4_COMM_PAGESIZE);
2307 	else
2308 		priv->mfunc.comm =
2309 		ioremap(pci_resource_start(dev->persist->pdev, 2) +
2310 			MLX4_SLAVE_COMM_BASE, MLX4_COMM_PAGESIZE);
2311 	if (!priv->mfunc.comm) {
2312 		mlx4_err(dev, "Couldn't map communication vector\n");
2313 		goto err_vhcr;
2314 	}
2315 
2316 	if (mlx4_is_master(dev)) {
2317 		struct mlx4_vf_oper_state *vf_oper;
2318 		struct mlx4_vf_admin_state *vf_admin;
2319 
2320 		priv->mfunc.master.slave_state =
2321 			kzalloc(dev->num_slaves *
2322 				sizeof(struct mlx4_slave_state), GFP_KERNEL);
2323 		if (!priv->mfunc.master.slave_state)
2324 			goto err_comm;
2325 
2326 		priv->mfunc.master.vf_admin =
2327 			kzalloc(dev->num_slaves *
2328 				sizeof(struct mlx4_vf_admin_state), GFP_KERNEL);
2329 		if (!priv->mfunc.master.vf_admin)
2330 			goto err_comm_admin;
2331 
2332 		priv->mfunc.master.vf_oper =
2333 			kzalloc(dev->num_slaves *
2334 				sizeof(struct mlx4_vf_oper_state), GFP_KERNEL);
2335 		if (!priv->mfunc.master.vf_oper)
2336 			goto err_comm_oper;
2337 
2338 		for (i = 0; i < dev->num_slaves; ++i) {
2339 			vf_admin = &priv->mfunc.master.vf_admin[i];
2340 			vf_oper = &priv->mfunc.master.vf_oper[i];
2341 			s_state = &priv->mfunc.master.slave_state[i];
2342 			s_state->last_cmd = MLX4_COMM_CMD_RESET;
2343 			mutex_init(&priv->mfunc.master.gen_eqe_mutex[i]);
2344 			for (j = 0; j < MLX4_EVENT_TYPES_NUM; ++j)
2345 				s_state->event_eq[j].eqn = -1;
2346 			__raw_writel((__force u32) 0,
2347 				     &priv->mfunc.comm[i].slave_write);
2348 			__raw_writel((__force u32) 0,
2349 				     &priv->mfunc.comm[i].slave_read);
2350 			mmiowb();
2351 			for (port = 1; port <= MLX4_MAX_PORTS; port++) {
2352 				struct mlx4_vport_state *admin_vport;
2353 				struct mlx4_vport_state *oper_vport;
2354 
2355 				s_state->vlan_filter[port] =
2356 					kzalloc(sizeof(struct mlx4_vlan_fltr),
2357 						GFP_KERNEL);
2358 				if (!s_state->vlan_filter[port]) {
2359 					if (--port)
2360 						kfree(s_state->vlan_filter[port]);
2361 					goto err_slaves;
2362 				}
2363 
2364 				admin_vport = &vf_admin->vport[port];
2365 				oper_vport = &vf_oper->vport[port].state;
2366 				INIT_LIST_HEAD(&s_state->mcast_filters[port]);
2367 				admin_vport->default_vlan = MLX4_VGT;
2368 				oper_vport->default_vlan = MLX4_VGT;
2369 				admin_vport->qos_vport =
2370 						MLX4_VPP_DEFAULT_VPORT;
2371 				oper_vport->qos_vport = MLX4_VPP_DEFAULT_VPORT;
2372 				vf_oper->vport[port].vlan_idx = NO_INDX;
2373 				vf_oper->vport[port].mac_idx = NO_INDX;
2374 				mlx4_set_random_admin_guid(dev, i, port);
2375 			}
2376 			spin_lock_init(&s_state->lock);
2377 		}
2378 
2379 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QOS_VPP) {
2380 			for (port = 1; port <= dev->caps.num_ports; port++) {
2381 				if (mlx4_is_eth(dev, port)) {
2382 					mlx4_set_default_port_qos(dev, port);
2383 					mlx4_allocate_port_vpps(dev, port);
2384 				}
2385 			}
2386 		}
2387 
2388 		memset(&priv->mfunc.master.cmd_eqe, 0, dev->caps.eqe_size);
2389 		priv->mfunc.master.cmd_eqe.type = MLX4_EVENT_TYPE_CMD;
2390 		INIT_WORK(&priv->mfunc.master.comm_work,
2391 			  mlx4_master_comm_channel);
2392 		INIT_WORK(&priv->mfunc.master.slave_event_work,
2393 			  mlx4_gen_slave_eqe);
2394 		INIT_WORK(&priv->mfunc.master.slave_flr_event_work,
2395 			  mlx4_master_handle_slave_flr);
2396 		spin_lock_init(&priv->mfunc.master.slave_state_lock);
2397 		spin_lock_init(&priv->mfunc.master.slave_eq.event_lock);
2398 		priv->mfunc.master.comm_wq =
2399 			create_singlethread_workqueue("mlx4_comm");
2400 		if (!priv->mfunc.master.comm_wq)
2401 			goto err_slaves;
2402 
2403 		if (mlx4_init_resource_tracker(dev))
2404 			goto err_thread;
2405 
2406 	} else {
2407 		err = sync_toggles(dev);
2408 		if (err) {
2409 			mlx4_err(dev, "Couldn't sync toggles\n");
2410 			goto err_comm;
2411 		}
2412 	}
2413 	return 0;
2414 
2415 err_thread:
2416 	flush_workqueue(priv->mfunc.master.comm_wq);
2417 	destroy_workqueue(priv->mfunc.master.comm_wq);
2418 err_slaves:
2419 	while (--i) {
2420 		for (port = 1; port <= MLX4_MAX_PORTS; port++)
2421 			kfree(priv->mfunc.master.slave_state[i].vlan_filter[port]);
2422 	}
2423 	kfree(priv->mfunc.master.vf_oper);
2424 err_comm_oper:
2425 	kfree(priv->mfunc.master.vf_admin);
2426 err_comm_admin:
2427 	kfree(priv->mfunc.master.slave_state);
2428 err_comm:
2429 	iounmap(priv->mfunc.comm);
2430 err_vhcr:
2431 	dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
2432 			  priv->mfunc.vhcr,
2433 			  priv->mfunc.vhcr_dma);
2434 	priv->mfunc.vhcr = NULL;
2435 	return -ENOMEM;
2436 }
2437 
2438 int mlx4_cmd_init(struct mlx4_dev *dev)
2439 {
2440 	struct mlx4_priv *priv = mlx4_priv(dev);
2441 	int flags = 0;
2442 
2443 	if (!priv->cmd.initialized) {
2444 		mutex_init(&priv->cmd.slave_cmd_mutex);
2445 		sema_init(&priv->cmd.poll_sem, 1);
2446 		priv->cmd.use_events = 0;
2447 		priv->cmd.toggle     = 1;
2448 		priv->cmd.initialized = 1;
2449 		flags |= MLX4_CMD_CLEANUP_STRUCT;
2450 	}
2451 
2452 	if (!mlx4_is_slave(dev) && !priv->cmd.hcr) {
2453 		priv->cmd.hcr = ioremap(pci_resource_start(dev->persist->pdev,
2454 					0) + MLX4_HCR_BASE, MLX4_HCR_SIZE);
2455 		if (!priv->cmd.hcr) {
2456 			mlx4_err(dev, "Couldn't map command register\n");
2457 			goto err;
2458 		}
2459 		flags |= MLX4_CMD_CLEANUP_HCR;
2460 	}
2461 
2462 	if (mlx4_is_mfunc(dev) && !priv->mfunc.vhcr) {
2463 		priv->mfunc.vhcr = dma_alloc_coherent(&dev->persist->pdev->dev,
2464 						      PAGE_SIZE,
2465 						      &priv->mfunc.vhcr_dma,
2466 						      GFP_KERNEL);
2467 		if (!priv->mfunc.vhcr)
2468 			goto err;
2469 
2470 		flags |= MLX4_CMD_CLEANUP_VHCR;
2471 	}
2472 
2473 	if (!priv->cmd.pool) {
2474 		priv->cmd.pool = pci_pool_create("mlx4_cmd",
2475 						 dev->persist->pdev,
2476 						 MLX4_MAILBOX_SIZE,
2477 						 MLX4_MAILBOX_SIZE, 0);
2478 		if (!priv->cmd.pool)
2479 			goto err;
2480 
2481 		flags |= MLX4_CMD_CLEANUP_POOL;
2482 	}
2483 
2484 	return 0;
2485 
2486 err:
2487 	mlx4_cmd_cleanup(dev, flags);
2488 	return -ENOMEM;
2489 }
2490 
2491 void mlx4_report_internal_err_comm_event(struct mlx4_dev *dev)
2492 {
2493 	struct mlx4_priv *priv = mlx4_priv(dev);
2494 	int slave;
2495 	u32 slave_read;
2496 
2497 	/* Report an internal error event to all
2498 	 * communication channels.
2499 	 */
2500 	for (slave = 0; slave < dev->num_slaves; slave++) {
2501 		slave_read = swab32(readl(&priv->mfunc.comm[slave].slave_read));
2502 		slave_read |= (u32)COMM_CHAN_EVENT_INTERNAL_ERR;
2503 		__raw_writel((__force u32)cpu_to_be32(slave_read),
2504 			     &priv->mfunc.comm[slave].slave_read);
2505 		/* Make sure that our comm channel write doesn't
2506 		 * get mixed in with writes from another CPU.
2507 		 */
2508 		mmiowb();
2509 	}
2510 }
2511 
2512 void mlx4_multi_func_cleanup(struct mlx4_dev *dev)
2513 {
2514 	struct mlx4_priv *priv = mlx4_priv(dev);
2515 	int i, port;
2516 
2517 	if (mlx4_is_master(dev)) {
2518 		flush_workqueue(priv->mfunc.master.comm_wq);
2519 		destroy_workqueue(priv->mfunc.master.comm_wq);
2520 		for (i = 0; i < dev->num_slaves; i++) {
2521 			for (port = 1; port <= MLX4_MAX_PORTS; port++)
2522 				kfree(priv->mfunc.master.slave_state[i].vlan_filter[port]);
2523 		}
2524 		kfree(priv->mfunc.master.slave_state);
2525 		kfree(priv->mfunc.master.vf_admin);
2526 		kfree(priv->mfunc.master.vf_oper);
2527 		dev->num_slaves = 0;
2528 	}
2529 
2530 	iounmap(priv->mfunc.comm);
2531 }
2532 
2533 void mlx4_cmd_cleanup(struct mlx4_dev *dev, int cleanup_mask)
2534 {
2535 	struct mlx4_priv *priv = mlx4_priv(dev);
2536 
2537 	if (priv->cmd.pool && (cleanup_mask & MLX4_CMD_CLEANUP_POOL)) {
2538 		pci_pool_destroy(priv->cmd.pool);
2539 		priv->cmd.pool = NULL;
2540 	}
2541 
2542 	if (!mlx4_is_slave(dev) && priv->cmd.hcr &&
2543 	    (cleanup_mask & MLX4_CMD_CLEANUP_HCR)) {
2544 		iounmap(priv->cmd.hcr);
2545 		priv->cmd.hcr = NULL;
2546 	}
2547 	if (mlx4_is_mfunc(dev) && priv->mfunc.vhcr &&
2548 	    (cleanup_mask & MLX4_CMD_CLEANUP_VHCR)) {
2549 		dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
2550 				  priv->mfunc.vhcr, priv->mfunc.vhcr_dma);
2551 		priv->mfunc.vhcr = NULL;
2552 	}
2553 	if (priv->cmd.initialized && (cleanup_mask & MLX4_CMD_CLEANUP_STRUCT))
2554 		priv->cmd.initialized = 0;
2555 }
2556 
2557 /*
2558  * Switch to using events to issue FW commands (can only be called
2559  * after event queue for command events has been initialized).
2560  */
2561 int mlx4_cmd_use_events(struct mlx4_dev *dev)
2562 {
2563 	struct mlx4_priv *priv = mlx4_priv(dev);
2564 	int i;
2565 	int err = 0;
2566 
2567 	priv->cmd.context = kmalloc(priv->cmd.max_cmds *
2568 				   sizeof (struct mlx4_cmd_context),
2569 				   GFP_KERNEL);
2570 	if (!priv->cmd.context)
2571 		return -ENOMEM;
2572 
2573 	for (i = 0; i < priv->cmd.max_cmds; ++i) {
2574 		priv->cmd.context[i].token = i;
2575 		priv->cmd.context[i].next  = i + 1;
2576 		/* To support fatal error flow, initialize all
2577 		 * cmd contexts to allow simulating completions
2578 		 * with complete() at any time.
2579 		 */
2580 		init_completion(&priv->cmd.context[i].done);
2581 	}
2582 
2583 	priv->cmd.context[priv->cmd.max_cmds - 1].next = -1;
2584 	priv->cmd.free_head = 0;
2585 
2586 	sema_init(&priv->cmd.event_sem, priv->cmd.max_cmds);
2587 	spin_lock_init(&priv->cmd.context_lock);
2588 
2589 	for (priv->cmd.token_mask = 1;
2590 	     priv->cmd.token_mask < priv->cmd.max_cmds;
2591 	     priv->cmd.token_mask <<= 1)
2592 		; /* nothing */
2593 	--priv->cmd.token_mask;
2594 
2595 	down(&priv->cmd.poll_sem);
2596 	priv->cmd.use_events = 1;
2597 
2598 	return err;
2599 }
2600 
2601 /*
2602  * Switch back to polling (used when shutting down the device)
2603  */
2604 void mlx4_cmd_use_polling(struct mlx4_dev *dev)
2605 {
2606 	struct mlx4_priv *priv = mlx4_priv(dev);
2607 	int i;
2608 
2609 	priv->cmd.use_events = 0;
2610 
2611 	for (i = 0; i < priv->cmd.max_cmds; ++i)
2612 		down(&priv->cmd.event_sem);
2613 
2614 	kfree(priv->cmd.context);
2615 
2616 	up(&priv->cmd.poll_sem);
2617 }
2618 
2619 struct mlx4_cmd_mailbox *mlx4_alloc_cmd_mailbox(struct mlx4_dev *dev)
2620 {
2621 	struct mlx4_cmd_mailbox *mailbox;
2622 
2623 	mailbox = kmalloc(sizeof *mailbox, GFP_KERNEL);
2624 	if (!mailbox)
2625 		return ERR_PTR(-ENOMEM);
2626 
2627 	mailbox->buf = pci_pool_alloc(mlx4_priv(dev)->cmd.pool, GFP_KERNEL,
2628 				      &mailbox->dma);
2629 	if (!mailbox->buf) {
2630 		kfree(mailbox);
2631 		return ERR_PTR(-ENOMEM);
2632 	}
2633 
2634 	memset(mailbox->buf, 0, MLX4_MAILBOX_SIZE);
2635 
2636 	return mailbox;
2637 }
2638 EXPORT_SYMBOL_GPL(mlx4_alloc_cmd_mailbox);
2639 
2640 void mlx4_free_cmd_mailbox(struct mlx4_dev *dev,
2641 			   struct mlx4_cmd_mailbox *mailbox)
2642 {
2643 	if (!mailbox)
2644 		return;
2645 
2646 	pci_pool_free(mlx4_priv(dev)->cmd.pool, mailbox->buf, mailbox->dma);
2647 	kfree(mailbox);
2648 }
2649 EXPORT_SYMBOL_GPL(mlx4_free_cmd_mailbox);
2650 
2651 u32 mlx4_comm_get_version(void)
2652 {
2653 	 return ((u32) CMD_CHAN_IF_REV << 8) | (u32) CMD_CHAN_VER;
2654 }
2655 
2656 static int mlx4_get_slave_indx(struct mlx4_dev *dev, int vf)
2657 {
2658 	if ((vf < 0) || (vf >= dev->persist->num_vfs)) {
2659 		mlx4_err(dev, "Bad vf number:%d (number of activated vf: %d)\n",
2660 			 vf, dev->persist->num_vfs);
2661 		return -EINVAL;
2662 	}
2663 
2664 	return vf+1;
2665 }
2666 
2667 int mlx4_get_vf_indx(struct mlx4_dev *dev, int slave)
2668 {
2669 	if (slave < 1 || slave > dev->persist->num_vfs) {
2670 		mlx4_err(dev,
2671 			 "Bad slave number:%d (number of activated slaves: %lu)\n",
2672 			 slave, dev->num_slaves);
2673 		return -EINVAL;
2674 	}
2675 	return slave - 1;
2676 }
2677 
2678 void mlx4_cmd_wake_completions(struct mlx4_dev *dev)
2679 {
2680 	struct mlx4_priv *priv = mlx4_priv(dev);
2681 	struct mlx4_cmd_context *context;
2682 	int i;
2683 
2684 	spin_lock(&priv->cmd.context_lock);
2685 	if (priv->cmd.context) {
2686 		for (i = 0; i < priv->cmd.max_cmds; ++i) {
2687 			context = &priv->cmd.context[i];
2688 			context->fw_status = CMD_STAT_INTERNAL_ERR;
2689 			context->result    =
2690 				mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
2691 			complete(&context->done);
2692 		}
2693 	}
2694 	spin_unlock(&priv->cmd.context_lock);
2695 }
2696 
2697 struct mlx4_active_ports mlx4_get_active_ports(struct mlx4_dev *dev, int slave)
2698 {
2699 	struct mlx4_active_ports actv_ports;
2700 	int vf;
2701 
2702 	bitmap_zero(actv_ports.ports, MLX4_MAX_PORTS);
2703 
2704 	if (slave == 0) {
2705 		bitmap_fill(actv_ports.ports, dev->caps.num_ports);
2706 		return actv_ports;
2707 	}
2708 
2709 	vf = mlx4_get_vf_indx(dev, slave);
2710 	if (vf < 0)
2711 		return actv_ports;
2712 
2713 	bitmap_set(actv_ports.ports, dev->dev_vfs[vf].min_port - 1,
2714 		   min((int)dev->dev_vfs[mlx4_get_vf_indx(dev, slave)].n_ports,
2715 		   dev->caps.num_ports));
2716 
2717 	return actv_ports;
2718 }
2719 EXPORT_SYMBOL_GPL(mlx4_get_active_ports);
2720 
2721 int mlx4_slave_convert_port(struct mlx4_dev *dev, int slave, int port)
2722 {
2723 	unsigned n;
2724 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
2725 	unsigned m = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2726 
2727 	if (port <= 0 || port > m)
2728 		return -EINVAL;
2729 
2730 	n = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2731 	if (port <= n)
2732 		port = n + 1;
2733 
2734 	return port;
2735 }
2736 EXPORT_SYMBOL_GPL(mlx4_slave_convert_port);
2737 
2738 int mlx4_phys_to_slave_port(struct mlx4_dev *dev, int slave, int port)
2739 {
2740 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
2741 	if (test_bit(port - 1, actv_ports.ports))
2742 		return port -
2743 			find_first_bit(actv_ports.ports, dev->caps.num_ports);
2744 
2745 	return -1;
2746 }
2747 EXPORT_SYMBOL_GPL(mlx4_phys_to_slave_port);
2748 
2749 struct mlx4_slaves_pport mlx4_phys_to_slaves_pport(struct mlx4_dev *dev,
2750 						   int port)
2751 {
2752 	unsigned i;
2753 	struct mlx4_slaves_pport slaves_pport;
2754 
2755 	bitmap_zero(slaves_pport.slaves, MLX4_MFUNC_MAX);
2756 
2757 	if (port <= 0 || port > dev->caps.num_ports)
2758 		return slaves_pport;
2759 
2760 	for (i = 0; i < dev->persist->num_vfs + 1; i++) {
2761 		struct mlx4_active_ports actv_ports =
2762 			mlx4_get_active_ports(dev, i);
2763 		if (test_bit(port - 1, actv_ports.ports))
2764 			set_bit(i, slaves_pport.slaves);
2765 	}
2766 
2767 	return slaves_pport;
2768 }
2769 EXPORT_SYMBOL_GPL(mlx4_phys_to_slaves_pport);
2770 
2771 struct mlx4_slaves_pport mlx4_phys_to_slaves_pport_actv(
2772 		struct mlx4_dev *dev,
2773 		const struct mlx4_active_ports *crit_ports)
2774 {
2775 	unsigned i;
2776 	struct mlx4_slaves_pport slaves_pport;
2777 
2778 	bitmap_zero(slaves_pport.slaves, MLX4_MFUNC_MAX);
2779 
2780 	for (i = 0; i < dev->persist->num_vfs + 1; i++) {
2781 		struct mlx4_active_ports actv_ports =
2782 			mlx4_get_active_ports(dev, i);
2783 		if (bitmap_equal(crit_ports->ports, actv_ports.ports,
2784 				 dev->caps.num_ports))
2785 			set_bit(i, slaves_pport.slaves);
2786 	}
2787 
2788 	return slaves_pport;
2789 }
2790 EXPORT_SYMBOL_GPL(mlx4_phys_to_slaves_pport_actv);
2791 
2792 static int mlx4_slaves_closest_port(struct mlx4_dev *dev, int slave, int port)
2793 {
2794 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
2795 	int min_port = find_first_bit(actv_ports.ports, dev->caps.num_ports)
2796 			+ 1;
2797 	int max_port = min_port +
2798 		bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2799 
2800 	if (port < min_port)
2801 		port = min_port;
2802 	else if (port >= max_port)
2803 		port = max_port - 1;
2804 
2805 	return port;
2806 }
2807 
2808 static int mlx4_set_vport_qos(struct mlx4_priv *priv, int slave, int port,
2809 			      int max_tx_rate)
2810 {
2811 	int i;
2812 	int err;
2813 	struct mlx4_qos_manager *port_qos;
2814 	struct mlx4_dev *dev = &priv->dev;
2815 	struct mlx4_vport_qos_param vpp_qos[MLX4_NUM_UP];
2816 
2817 	port_qos = &priv->mfunc.master.qos_ctl[port];
2818 	memset(vpp_qos, 0, sizeof(struct mlx4_vport_qos_param) * MLX4_NUM_UP);
2819 
2820 	if (slave > port_qos->num_of_qos_vfs) {
2821 		mlx4_info(dev, "No availible VPP resources for this VF\n");
2822 		return -EINVAL;
2823 	}
2824 
2825 	/* Query for default QoS values from Vport 0 is needed */
2826 	err = mlx4_SET_VPORT_QOS_get(dev, port, 0, vpp_qos);
2827 	if (err) {
2828 		mlx4_info(dev, "Failed to query Vport 0 QoS values\n");
2829 		return err;
2830 	}
2831 
2832 	for (i = 0; i < MLX4_NUM_UP; i++) {
2833 		if (test_bit(i, port_qos->priority_bm) && max_tx_rate) {
2834 			vpp_qos[i].max_avg_bw = max_tx_rate;
2835 			vpp_qos[i].enable = 1;
2836 		} else {
2837 			/* if user supplied tx_rate == 0, meaning no rate limit
2838 			 * configuration is required. so we are leaving the
2839 			 * value of max_avg_bw as queried from Vport 0.
2840 			 */
2841 			vpp_qos[i].enable = 0;
2842 		}
2843 	}
2844 
2845 	err = mlx4_SET_VPORT_QOS_set(dev, port, slave, vpp_qos);
2846 	if (err) {
2847 		mlx4_info(dev, "Failed to set Vport %d QoS values\n", slave);
2848 		return err;
2849 	}
2850 
2851 	return 0;
2852 }
2853 
2854 static bool mlx4_is_vf_vst_and_prio_qos(struct mlx4_dev *dev, int port,
2855 					struct mlx4_vport_state *vf_admin)
2856 {
2857 	struct mlx4_qos_manager *info;
2858 	struct mlx4_priv *priv = mlx4_priv(dev);
2859 
2860 	if (!mlx4_is_master(dev) ||
2861 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QOS_VPP))
2862 		return false;
2863 
2864 	info = &priv->mfunc.master.qos_ctl[port];
2865 
2866 	if (vf_admin->default_vlan != MLX4_VGT &&
2867 	    test_bit(vf_admin->default_qos, info->priority_bm))
2868 		return true;
2869 
2870 	return false;
2871 }
2872 
2873 static bool mlx4_valid_vf_state_change(struct mlx4_dev *dev, int port,
2874 				       struct mlx4_vport_state *vf_admin,
2875 				       int vlan, int qos)
2876 {
2877 	struct mlx4_vport_state dummy_admin = {0};
2878 
2879 	if (!mlx4_is_vf_vst_and_prio_qos(dev, port, vf_admin) ||
2880 	    !vf_admin->tx_rate)
2881 		return true;
2882 
2883 	dummy_admin.default_qos = qos;
2884 	dummy_admin.default_vlan = vlan;
2885 
2886 	/* VF wants to move to other VST state which is valid with current
2887 	 * rate limit. Either differnt default vlan in VST or other
2888 	 * supported QoS priority. Otherwise we don't allow this change when
2889 	 * the TX rate is still configured.
2890 	 */
2891 	if (mlx4_is_vf_vst_and_prio_qos(dev, port, &dummy_admin))
2892 		return true;
2893 
2894 	mlx4_info(dev, "Cannot change VF state to %s while rate is set\n",
2895 		  (vlan == MLX4_VGT) ? "VGT" : "VST");
2896 
2897 	if (vlan != MLX4_VGT)
2898 		mlx4_info(dev, "VST priority %d not supported for QoS\n", qos);
2899 
2900 	mlx4_info(dev, "Please set rate to 0 prior to this VF state change\n");
2901 
2902 	return false;
2903 }
2904 
2905 int mlx4_set_vf_mac(struct mlx4_dev *dev, int port, int vf, u64 mac)
2906 {
2907 	struct mlx4_priv *priv = mlx4_priv(dev);
2908 	struct mlx4_vport_state *s_info;
2909 	int slave;
2910 
2911 	if (!mlx4_is_master(dev))
2912 		return -EPROTONOSUPPORT;
2913 
2914 	slave = mlx4_get_slave_indx(dev, vf);
2915 	if (slave < 0)
2916 		return -EINVAL;
2917 
2918 	port = mlx4_slaves_closest_port(dev, slave, port);
2919 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
2920 	s_info->mac = mac;
2921 	mlx4_info(dev, "default mac on vf %d port %d to %llX will take effect only after vf restart\n",
2922 		  vf, port, s_info->mac);
2923 	return 0;
2924 }
2925 EXPORT_SYMBOL_GPL(mlx4_set_vf_mac);
2926 
2927 
2928 int mlx4_set_vf_vlan(struct mlx4_dev *dev, int port, int vf, u16 vlan, u8 qos)
2929 {
2930 	struct mlx4_priv *priv = mlx4_priv(dev);
2931 	struct mlx4_vport_state *vf_admin;
2932 	int slave;
2933 
2934 	if ((!mlx4_is_master(dev)) ||
2935 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_VLAN_CONTROL))
2936 		return -EPROTONOSUPPORT;
2937 
2938 	if ((vlan > 4095) || (qos > 7))
2939 		return -EINVAL;
2940 
2941 	slave = mlx4_get_slave_indx(dev, vf);
2942 	if (slave < 0)
2943 		return -EINVAL;
2944 
2945 	port = mlx4_slaves_closest_port(dev, slave, port);
2946 	vf_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
2947 
2948 	if (!mlx4_valid_vf_state_change(dev, port, vf_admin, vlan, qos))
2949 		return -EPERM;
2950 
2951 	if ((0 == vlan) && (0 == qos))
2952 		vf_admin->default_vlan = MLX4_VGT;
2953 	else
2954 		vf_admin->default_vlan = vlan;
2955 	vf_admin->default_qos = qos;
2956 
2957 	/* If rate was configured prior to VST, we saved the configured rate
2958 	 * in vf_admin->rate and now, if priority supported we enforce the QoS
2959 	 */
2960 	if (mlx4_is_vf_vst_and_prio_qos(dev, port, vf_admin) &&
2961 	    vf_admin->tx_rate)
2962 		vf_admin->qos_vport = slave;
2963 
2964 	if (mlx4_master_immediate_activate_vlan_qos(priv, slave, port))
2965 		mlx4_info(dev,
2966 			  "updating vf %d port %d config will take effect on next VF restart\n",
2967 			  vf, port);
2968 	return 0;
2969 }
2970 EXPORT_SYMBOL_GPL(mlx4_set_vf_vlan);
2971 
2972 int mlx4_set_vf_rate(struct mlx4_dev *dev, int port, int vf, int min_tx_rate,
2973 		     int max_tx_rate)
2974 {
2975 	int err;
2976 	int slave;
2977 	struct mlx4_vport_state *vf_admin;
2978 	struct mlx4_priv *priv = mlx4_priv(dev);
2979 
2980 	if (!mlx4_is_master(dev) ||
2981 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QOS_VPP))
2982 		return -EPROTONOSUPPORT;
2983 
2984 	if (min_tx_rate) {
2985 		mlx4_info(dev, "Minimum BW share not supported\n");
2986 		return -EPROTONOSUPPORT;
2987 	}
2988 
2989 	slave = mlx4_get_slave_indx(dev, vf);
2990 	if (slave < 0)
2991 		return -EINVAL;
2992 
2993 	port = mlx4_slaves_closest_port(dev, slave, port);
2994 	vf_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
2995 
2996 	err = mlx4_set_vport_qos(priv, slave, port, max_tx_rate);
2997 	if (err) {
2998 		mlx4_info(dev, "vf %d failed to set rate %d\n", vf,
2999 			  max_tx_rate);
3000 		return err;
3001 	}
3002 
3003 	vf_admin->tx_rate = max_tx_rate;
3004 	/* if VF is not in supported mode (VST with supported prio),
3005 	 * we do not change vport configuration for its QPs, but save
3006 	 * the rate, so it will be enforced when it moves to supported
3007 	 * mode next time.
3008 	 */
3009 	if (!mlx4_is_vf_vst_and_prio_qos(dev, port, vf_admin)) {
3010 		mlx4_info(dev,
3011 			  "rate set for VF %d when not in valid state\n", vf);
3012 
3013 		if (vf_admin->default_vlan != MLX4_VGT)
3014 			mlx4_info(dev, "VST priority not supported by QoS\n");
3015 		else
3016 			mlx4_info(dev, "VF in VGT mode (needed VST)\n");
3017 
3018 		mlx4_info(dev,
3019 			  "rate %d take affect when VF moves to valid state\n",
3020 			  max_tx_rate);
3021 		return 0;
3022 	}
3023 
3024 	/* If user sets rate 0 assigning default vport for its QPs */
3025 	vf_admin->qos_vport = max_tx_rate ? slave : MLX4_VPP_DEFAULT_VPORT;
3026 
3027 	if (priv->mfunc.master.slave_state[slave].active &&
3028 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_UPDATE_QP)
3029 		mlx4_master_immediate_activate_vlan_qos(priv, slave, port);
3030 
3031 	return 0;
3032 }
3033 EXPORT_SYMBOL_GPL(mlx4_set_vf_rate);
3034 
3035  /* mlx4_get_slave_default_vlan -
3036  * return true if VST ( default vlan)
3037  * if VST, will return vlan & qos (if not NULL)
3038  */
3039 bool mlx4_get_slave_default_vlan(struct mlx4_dev *dev, int port, int slave,
3040 				 u16 *vlan, u8 *qos)
3041 {
3042 	struct mlx4_vport_oper_state *vp_oper;
3043 	struct mlx4_priv *priv;
3044 
3045 	priv = mlx4_priv(dev);
3046 	port = mlx4_slaves_closest_port(dev, slave, port);
3047 	vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
3048 
3049 	if (MLX4_VGT != vp_oper->state.default_vlan) {
3050 		if (vlan)
3051 			*vlan = vp_oper->state.default_vlan;
3052 		if (qos)
3053 			*qos = vp_oper->state.default_qos;
3054 		return true;
3055 	}
3056 	return false;
3057 }
3058 EXPORT_SYMBOL_GPL(mlx4_get_slave_default_vlan);
3059 
3060 int mlx4_set_vf_spoofchk(struct mlx4_dev *dev, int port, int vf, bool setting)
3061 {
3062 	struct mlx4_priv *priv = mlx4_priv(dev);
3063 	struct mlx4_vport_state *s_info;
3064 	int slave;
3065 
3066 	if ((!mlx4_is_master(dev)) ||
3067 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FSM))
3068 		return -EPROTONOSUPPORT;
3069 
3070 	slave = mlx4_get_slave_indx(dev, vf);
3071 	if (slave < 0)
3072 		return -EINVAL;
3073 
3074 	port = mlx4_slaves_closest_port(dev, slave, port);
3075 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
3076 	s_info->spoofchk = setting;
3077 
3078 	return 0;
3079 }
3080 EXPORT_SYMBOL_GPL(mlx4_set_vf_spoofchk);
3081 
3082 int mlx4_get_vf_config(struct mlx4_dev *dev, int port, int vf, struct ifla_vf_info *ivf)
3083 {
3084 	struct mlx4_priv *priv = mlx4_priv(dev);
3085 	struct mlx4_vport_state *s_info;
3086 	int slave;
3087 
3088 	if (!mlx4_is_master(dev))
3089 		return -EPROTONOSUPPORT;
3090 
3091 	slave = mlx4_get_slave_indx(dev, vf);
3092 	if (slave < 0)
3093 		return -EINVAL;
3094 
3095 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
3096 	ivf->vf = vf;
3097 
3098 	/* need to convert it to a func */
3099 	ivf->mac[0] = ((s_info->mac >> (5*8)) & 0xff);
3100 	ivf->mac[1] = ((s_info->mac >> (4*8)) & 0xff);
3101 	ivf->mac[2] = ((s_info->mac >> (3*8)) & 0xff);
3102 	ivf->mac[3] = ((s_info->mac >> (2*8)) & 0xff);
3103 	ivf->mac[4] = ((s_info->mac >> (1*8)) & 0xff);
3104 	ivf->mac[5] = ((s_info->mac)  & 0xff);
3105 
3106 	ivf->vlan		= s_info->default_vlan;
3107 	ivf->qos		= s_info->default_qos;
3108 
3109 	if (mlx4_is_vf_vst_and_prio_qos(dev, port, s_info))
3110 		ivf->max_tx_rate = s_info->tx_rate;
3111 	else
3112 		ivf->max_tx_rate = 0;
3113 
3114 	ivf->min_tx_rate	= 0;
3115 	ivf->spoofchk		= s_info->spoofchk;
3116 	ivf->linkstate		= s_info->link_state;
3117 
3118 	return 0;
3119 }
3120 EXPORT_SYMBOL_GPL(mlx4_get_vf_config);
3121 
3122 int mlx4_set_vf_link_state(struct mlx4_dev *dev, int port, int vf, int link_state)
3123 {
3124 	struct mlx4_priv *priv = mlx4_priv(dev);
3125 	struct mlx4_vport_state *s_info;
3126 	int slave;
3127 	u8 link_stat_event;
3128 
3129 	slave = mlx4_get_slave_indx(dev, vf);
3130 	if (slave < 0)
3131 		return -EINVAL;
3132 
3133 	port = mlx4_slaves_closest_port(dev, slave, port);
3134 	switch (link_state) {
3135 	case IFLA_VF_LINK_STATE_AUTO:
3136 		/* get current link state */
3137 		if (!priv->sense.do_sense_port[port])
3138 			link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_ACTIVE;
3139 		else
3140 			link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_DOWN;
3141 	    break;
3142 
3143 	case IFLA_VF_LINK_STATE_ENABLE:
3144 		link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_ACTIVE;
3145 	    break;
3146 
3147 	case IFLA_VF_LINK_STATE_DISABLE:
3148 		link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_DOWN;
3149 	    break;
3150 
3151 	default:
3152 		mlx4_warn(dev, "unknown value for link_state %02x on slave %d port %d\n",
3153 			  link_state, slave, port);
3154 		return -EINVAL;
3155 	};
3156 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
3157 	s_info->link_state = link_state;
3158 
3159 	/* send event */
3160 	mlx4_gen_port_state_change_eqe(dev, slave, port, link_stat_event);
3161 
3162 	if (mlx4_master_immediate_activate_vlan_qos(priv, slave, port))
3163 		mlx4_dbg(dev,
3164 			 "updating vf %d port %d no link state HW enforcment\n",
3165 			 vf, port);
3166 	return 0;
3167 }
3168 EXPORT_SYMBOL_GPL(mlx4_set_vf_link_state);
3169 
3170 int mlx4_get_counter_stats(struct mlx4_dev *dev, int counter_index,
3171 			   struct mlx4_counter *counter_stats, int reset)
3172 {
3173 	struct mlx4_cmd_mailbox *mailbox = NULL;
3174 	struct mlx4_counter *tmp_counter;
3175 	int err;
3176 	u32 if_stat_in_mod;
3177 
3178 	if (!counter_stats)
3179 		return -EINVAL;
3180 
3181 	if (counter_index == MLX4_SINK_COUNTER_INDEX(dev))
3182 		return 0;
3183 
3184 	mailbox = mlx4_alloc_cmd_mailbox(dev);
3185 	if (IS_ERR(mailbox))
3186 		return PTR_ERR(mailbox);
3187 
3188 	memset(mailbox->buf, 0, sizeof(struct mlx4_counter));
3189 	if_stat_in_mod = counter_index;
3190 	if (reset)
3191 		if_stat_in_mod |= MLX4_QUERY_IF_STAT_RESET;
3192 	err = mlx4_cmd_box(dev, 0, mailbox->dma,
3193 			   if_stat_in_mod, 0,
3194 			   MLX4_CMD_QUERY_IF_STAT,
3195 			   MLX4_CMD_TIME_CLASS_C,
3196 			   MLX4_CMD_NATIVE);
3197 	if (err) {
3198 		mlx4_dbg(dev, "%s: failed to read statistics for counter index %d\n",
3199 			 __func__, counter_index);
3200 		goto if_stat_out;
3201 	}
3202 	tmp_counter = (struct mlx4_counter *)mailbox->buf;
3203 	counter_stats->counter_mode = tmp_counter->counter_mode;
3204 	if (counter_stats->counter_mode == 0) {
3205 		counter_stats->rx_frames =
3206 			cpu_to_be64(be64_to_cpu(counter_stats->rx_frames) +
3207 				    be64_to_cpu(tmp_counter->rx_frames));
3208 		counter_stats->tx_frames =
3209 			cpu_to_be64(be64_to_cpu(counter_stats->tx_frames) +
3210 				    be64_to_cpu(tmp_counter->tx_frames));
3211 		counter_stats->rx_bytes =
3212 			cpu_to_be64(be64_to_cpu(counter_stats->rx_bytes) +
3213 				    be64_to_cpu(tmp_counter->rx_bytes));
3214 		counter_stats->tx_bytes =
3215 			cpu_to_be64(be64_to_cpu(counter_stats->tx_bytes) +
3216 				    be64_to_cpu(tmp_counter->tx_bytes));
3217 	}
3218 
3219 if_stat_out:
3220 	mlx4_free_cmd_mailbox(dev, mailbox);
3221 
3222 	return err;
3223 }
3224 EXPORT_SYMBOL_GPL(mlx4_get_counter_stats);
3225 
3226 int mlx4_get_vf_stats(struct mlx4_dev *dev, int port, int vf_idx,
3227 		      struct ifla_vf_stats *vf_stats)
3228 {
3229 	struct mlx4_counter tmp_vf_stats;
3230 	int slave;
3231 	int err = 0;
3232 
3233 	if (!vf_stats)
3234 		return -EINVAL;
3235 
3236 	if (!mlx4_is_master(dev))
3237 		return -EPROTONOSUPPORT;
3238 
3239 	slave = mlx4_get_slave_indx(dev, vf_idx);
3240 	if (slave < 0)
3241 		return -EINVAL;
3242 
3243 	port = mlx4_slaves_closest_port(dev, slave, port);
3244 	err = mlx4_calc_vf_counters(dev, slave, port, &tmp_vf_stats);
3245 	if (!err && tmp_vf_stats.counter_mode == 0) {
3246 		vf_stats->rx_packets = be64_to_cpu(tmp_vf_stats.rx_frames);
3247 		vf_stats->tx_packets = be64_to_cpu(tmp_vf_stats.tx_frames);
3248 		vf_stats->rx_bytes = be64_to_cpu(tmp_vf_stats.rx_bytes);
3249 		vf_stats->tx_bytes = be64_to_cpu(tmp_vf_stats.tx_bytes);
3250 	}
3251 
3252 	return err;
3253 }
3254 EXPORT_SYMBOL_GPL(mlx4_get_vf_stats);
3255 
3256 int mlx4_vf_smi_enabled(struct mlx4_dev *dev, int slave, int port)
3257 {
3258 	struct mlx4_priv *priv = mlx4_priv(dev);
3259 
3260 	if (slave < 1 || slave >= dev->num_slaves ||
3261 	    port < 1 || port > MLX4_MAX_PORTS)
3262 		return 0;
3263 
3264 	return priv->mfunc.master.vf_oper[slave].smi_enabled[port] ==
3265 		MLX4_VF_SMI_ENABLED;
3266 }
3267 EXPORT_SYMBOL_GPL(mlx4_vf_smi_enabled);
3268 
3269 int mlx4_vf_get_enable_smi_admin(struct mlx4_dev *dev, int slave, int port)
3270 {
3271 	struct mlx4_priv *priv = mlx4_priv(dev);
3272 
3273 	if (slave == mlx4_master_func_num(dev))
3274 		return 1;
3275 
3276 	if (slave < 1 || slave >= dev->num_slaves ||
3277 	    port < 1 || port > MLX4_MAX_PORTS)
3278 		return 0;
3279 
3280 	return priv->mfunc.master.vf_admin[slave].enable_smi[port] ==
3281 		MLX4_VF_SMI_ENABLED;
3282 }
3283 EXPORT_SYMBOL_GPL(mlx4_vf_get_enable_smi_admin);
3284 
3285 int mlx4_vf_set_enable_smi_admin(struct mlx4_dev *dev, int slave, int port,
3286 				 int enabled)
3287 {
3288 	struct mlx4_priv *priv = mlx4_priv(dev);
3289 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(
3290 			&priv->dev, slave);
3291 	int min_port = find_first_bit(actv_ports.ports,
3292 				      priv->dev.caps.num_ports) + 1;
3293 	int max_port = min_port - 1 +
3294 		bitmap_weight(actv_ports.ports, priv->dev.caps.num_ports);
3295 
3296 	if (slave == mlx4_master_func_num(dev))
3297 		return 0;
3298 
3299 	if (slave < 1 || slave >= dev->num_slaves ||
3300 	    port < 1 || port > MLX4_MAX_PORTS ||
3301 	    enabled < 0 || enabled > 1)
3302 		return -EINVAL;
3303 
3304 	if (min_port == max_port && dev->caps.num_ports > 1) {
3305 		mlx4_info(dev, "SMI access disallowed for single ported VFs\n");
3306 		return -EPROTONOSUPPORT;
3307 	}
3308 
3309 	priv->mfunc.master.vf_admin[slave].enable_smi[port] = enabled;
3310 	return 0;
3311 }
3312 EXPORT_SYMBOL_GPL(mlx4_vf_set_enable_smi_admin);
3313