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 	long ret_wait;
690 	int err = 0;
691 
692 	down(&cmd->event_sem);
693 
694 	spin_lock(&cmd->context_lock);
695 	BUG_ON(cmd->free_head < 0);
696 	context = &cmd->context[cmd->free_head];
697 	context->token += cmd->token_mask + 1;
698 	cmd->free_head = context->next;
699 	spin_unlock(&cmd->context_lock);
700 
701 	if (out_is_imm && !out_param) {
702 		mlx4_err(dev, "response expected while output mailbox is NULL for command 0x%x\n",
703 			 op);
704 		err = -EINVAL;
705 		goto out;
706 	}
707 
708 	reinit_completion(&context->done);
709 
710 	err = mlx4_cmd_post(dev, in_param, out_param ? *out_param : 0,
711 			    in_modifier, op_modifier, op, context->token, 1);
712 	if (err)
713 		goto out_reset;
714 
715 	if (op == MLX4_CMD_SENSE_PORT) {
716 		ret_wait =
717 			wait_for_completion_interruptible_timeout(&context->done,
718 								  msecs_to_jiffies(timeout));
719 		if (ret_wait < 0) {
720 			context->fw_status = 0;
721 			context->out_param = 0;
722 			context->result = 0;
723 		}
724 	} else {
725 		ret_wait = (long)wait_for_completion_timeout(&context->done,
726 							     msecs_to_jiffies(timeout));
727 	}
728 	if (!ret_wait) {
729 		mlx4_warn(dev, "command 0x%x timed out (go bit not cleared)\n",
730 			  op);
731 		if (op == MLX4_CMD_NOP) {
732 			err = -EBUSY;
733 			goto out;
734 		} else {
735 			err = -EIO;
736 			goto out_reset;
737 		}
738 	}
739 
740 	err = context->result;
741 	if (err) {
742 		/* Since we do not want to have this error message always
743 		 * displayed at driver start when there are ConnectX2 HCAs
744 		 * on the host, we deprecate the error message for this
745 		 * specific command/input_mod/opcode_mod/fw-status to be debug.
746 		 */
747 		if (op == MLX4_CMD_SET_PORT &&
748 		    (in_modifier == 1 || in_modifier == 2) &&
749 		    op_modifier == MLX4_SET_PORT_IB_OPCODE &&
750 		    context->fw_status == CMD_STAT_BAD_SIZE)
751 			mlx4_dbg(dev, "command 0x%x failed: fw status = 0x%x\n",
752 				 op, context->fw_status);
753 		else
754 			mlx4_err(dev, "command 0x%x failed: fw status = 0x%x\n",
755 				 op, context->fw_status);
756 		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
757 			err = mlx4_internal_err_ret_value(dev, op, op_modifier);
758 		else if (mlx4_closing_cmd_fatal_error(op, context->fw_status))
759 			goto out_reset;
760 
761 		goto out;
762 	}
763 
764 	if (out_is_imm)
765 		*out_param = context->out_param;
766 
767 out_reset:
768 	if (err)
769 		err = mlx4_cmd_reset_flow(dev, op, op_modifier, err);
770 out:
771 	spin_lock(&cmd->context_lock);
772 	context->next = cmd->free_head;
773 	cmd->free_head = context - cmd->context;
774 	spin_unlock(&cmd->context_lock);
775 
776 	up(&cmd->event_sem);
777 	return err;
778 }
779 
780 int __mlx4_cmd(struct mlx4_dev *dev, u64 in_param, u64 *out_param,
781 	       int out_is_imm, u32 in_modifier, u8 op_modifier,
782 	       u16 op, unsigned long timeout, int native)
783 {
784 	if (pci_channel_offline(dev->persist->pdev))
785 		return mlx4_cmd_reset_flow(dev, op, op_modifier, -EIO);
786 
787 	if (!mlx4_is_mfunc(dev) || (native && mlx4_is_master(dev))) {
788 		if (dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR)
789 			return mlx4_internal_err_ret_value(dev, op,
790 							  op_modifier);
791 		if (mlx4_priv(dev)->cmd.use_events)
792 			return mlx4_cmd_wait(dev, in_param, out_param,
793 					     out_is_imm, in_modifier,
794 					     op_modifier, op, timeout);
795 		else
796 			return mlx4_cmd_poll(dev, in_param, out_param,
797 					     out_is_imm, in_modifier,
798 					     op_modifier, op, timeout);
799 	}
800 	return mlx4_slave_cmd(dev, in_param, out_param, out_is_imm,
801 			      in_modifier, op_modifier, op, timeout);
802 }
803 EXPORT_SYMBOL_GPL(__mlx4_cmd);
804 
805 
806 int mlx4_ARM_COMM_CHANNEL(struct mlx4_dev *dev)
807 {
808 	return mlx4_cmd(dev, 0, 0, 0, MLX4_CMD_ARM_COMM_CHANNEL,
809 			MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
810 }
811 
812 static int mlx4_ACCESS_MEM(struct mlx4_dev *dev, u64 master_addr,
813 			   int slave, u64 slave_addr,
814 			   int size, int is_read)
815 {
816 	u64 in_param;
817 	u64 out_param;
818 
819 	if ((slave_addr & 0xfff) | (master_addr & 0xfff) |
820 	    (slave & ~0x7f) | (size & 0xff)) {
821 		mlx4_err(dev, "Bad access mem params - slave_addr:0x%llx master_addr:0x%llx slave_id:%d size:%d\n",
822 			 slave_addr, master_addr, slave, size);
823 		return -EINVAL;
824 	}
825 
826 	if (is_read) {
827 		in_param = (u64) slave | slave_addr;
828 		out_param = (u64) dev->caps.function | master_addr;
829 	} else {
830 		in_param = (u64) dev->caps.function | master_addr;
831 		out_param = (u64) slave | slave_addr;
832 	}
833 
834 	return mlx4_cmd_imm(dev, in_param, &out_param, size, 0,
835 			    MLX4_CMD_ACCESS_MEM,
836 			    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
837 }
838 
839 static int query_pkey_block(struct mlx4_dev *dev, u8 port, u16 index, u16 *pkey,
840 			       struct mlx4_cmd_mailbox *inbox,
841 			       struct mlx4_cmd_mailbox *outbox)
842 {
843 	struct ib_smp *in_mad = (struct ib_smp *)(inbox->buf);
844 	struct ib_smp *out_mad = (struct ib_smp *)(outbox->buf);
845 	int err;
846 	int i;
847 
848 	if (index & 0x1f)
849 		return -EINVAL;
850 
851 	in_mad->attr_mod = cpu_to_be32(index / 32);
852 
853 	err = mlx4_cmd_box(dev, inbox->dma, outbox->dma, port, 3,
854 			   MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C,
855 			   MLX4_CMD_NATIVE);
856 	if (err)
857 		return err;
858 
859 	for (i = 0; i < 32; ++i)
860 		pkey[i] = be16_to_cpu(((__be16 *) out_mad->data)[i]);
861 
862 	return err;
863 }
864 
865 static int get_full_pkey_table(struct mlx4_dev *dev, u8 port, u16 *table,
866 			       struct mlx4_cmd_mailbox *inbox,
867 			       struct mlx4_cmd_mailbox *outbox)
868 {
869 	int i;
870 	int err;
871 
872 	for (i = 0; i < dev->caps.pkey_table_len[port]; i += 32) {
873 		err = query_pkey_block(dev, port, i, table + i, inbox, outbox);
874 		if (err)
875 			return err;
876 	}
877 
878 	return 0;
879 }
880 #define PORT_CAPABILITY_LOCATION_IN_SMP 20
881 #define PORT_STATE_OFFSET 32
882 
883 static enum ib_port_state vf_port_state(struct mlx4_dev *dev, int port, int vf)
884 {
885 	if (mlx4_get_slave_port_state(dev, vf, port) == SLAVE_PORT_UP)
886 		return IB_PORT_ACTIVE;
887 	else
888 		return IB_PORT_DOWN;
889 }
890 
891 static int mlx4_MAD_IFC_wrapper(struct mlx4_dev *dev, int slave,
892 				struct mlx4_vhcr *vhcr,
893 				struct mlx4_cmd_mailbox *inbox,
894 				struct mlx4_cmd_mailbox *outbox,
895 				struct mlx4_cmd_info *cmd)
896 {
897 	struct ib_smp *smp = inbox->buf;
898 	u32 index;
899 	u8 port, slave_port;
900 	u8 opcode_modifier;
901 	u16 *table;
902 	int err;
903 	int vidx, pidx;
904 	int network_view;
905 	struct mlx4_priv *priv = mlx4_priv(dev);
906 	struct ib_smp *outsmp = outbox->buf;
907 	__be16 *outtab = (__be16 *)(outsmp->data);
908 	__be32 slave_cap_mask;
909 	__be64 slave_node_guid;
910 
911 	slave_port = vhcr->in_modifier;
912 	port = mlx4_slave_convert_port(dev, slave, slave_port);
913 
914 	/* network-view bit is for driver use only, and should not be passed to FW */
915 	opcode_modifier = vhcr->op_modifier & ~0x8; /* clear netw view bit */
916 	network_view = !!(vhcr->op_modifier & 0x8);
917 
918 	if (smp->base_version == 1 &&
919 	    smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED &&
920 	    smp->class_version == 1) {
921 		/* host view is paravirtualized */
922 		if (!network_view && smp->method == IB_MGMT_METHOD_GET) {
923 			if (smp->attr_id == IB_SMP_ATTR_PKEY_TABLE) {
924 				index = be32_to_cpu(smp->attr_mod);
925 				if (port < 1 || port > dev->caps.num_ports)
926 					return -EINVAL;
927 				table = kcalloc((dev->caps.pkey_table_len[port] / 32) + 1,
928 						sizeof(*table) * 32, GFP_KERNEL);
929 
930 				if (!table)
931 					return -ENOMEM;
932 				/* need to get the full pkey table because the paravirtualized
933 				 * pkeys may be scattered among several pkey blocks.
934 				 */
935 				err = get_full_pkey_table(dev, port, table, inbox, outbox);
936 				if (!err) {
937 					for (vidx = index * 32; vidx < (index + 1) * 32; ++vidx) {
938 						pidx = priv->virt2phys_pkey[slave][port - 1][vidx];
939 						outtab[vidx % 32] = cpu_to_be16(table[pidx]);
940 					}
941 				}
942 				kfree(table);
943 				return err;
944 			}
945 			if (smp->attr_id == IB_SMP_ATTR_PORT_INFO) {
946 				/*get the slave specific caps:*/
947 				/*do the command */
948 				smp->attr_mod = cpu_to_be32(port);
949 				err = mlx4_cmd_box(dev, inbox->dma, outbox->dma,
950 					    port, opcode_modifier,
951 					    vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
952 				/* modify the response for slaves */
953 				if (!err && slave != mlx4_master_func_num(dev)) {
954 					u8 *state = outsmp->data + PORT_STATE_OFFSET;
955 
956 					*state = (*state & 0xf0) | vf_port_state(dev, port, slave);
957 					slave_cap_mask = priv->mfunc.master.slave_state[slave].ib_cap_mask[port];
958 					memcpy(outsmp->data + PORT_CAPABILITY_LOCATION_IN_SMP, &slave_cap_mask, 4);
959 				}
960 				return err;
961 			}
962 			if (smp->attr_id == IB_SMP_ATTR_GUID_INFO) {
963 				__be64 guid = mlx4_get_admin_guid(dev, slave,
964 								  port);
965 
966 				/* set the PF admin guid to the FW/HW burned
967 				 * GUID, if it wasn't yet set
968 				 */
969 				if (slave == 0 && guid == 0) {
970 					smp->attr_mod = 0;
971 					err = mlx4_cmd_box(dev,
972 							   inbox->dma,
973 							   outbox->dma,
974 							   vhcr->in_modifier,
975 							   opcode_modifier,
976 							   vhcr->op,
977 							   MLX4_CMD_TIME_CLASS_C,
978 							   MLX4_CMD_NATIVE);
979 					if (err)
980 						return err;
981 					mlx4_set_admin_guid(dev,
982 							    *(__be64 *)outsmp->
983 							    data, slave, port);
984 				} else {
985 					memcpy(outsmp->data, &guid, 8);
986 				}
987 
988 				/* clean all other gids */
989 				memset(outsmp->data + 8, 0, 56);
990 				return 0;
991 			}
992 			if (smp->attr_id == IB_SMP_ATTR_NODE_INFO) {
993 				err = mlx4_cmd_box(dev, inbox->dma, outbox->dma,
994 					     port, opcode_modifier,
995 					     vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
996 				if (!err) {
997 					slave_node_guid =  mlx4_get_slave_node_guid(dev, slave);
998 					memcpy(outsmp->data + 12, &slave_node_guid, 8);
999 				}
1000 				return err;
1001 			}
1002 		}
1003 	}
1004 
1005 	/* Non-privileged VFs are only allowed "host" view LID-routed 'Get' MADs.
1006 	 * These are the MADs used by ib verbs (such as ib_query_gids).
1007 	 */
1008 	if (slave != mlx4_master_func_num(dev) &&
1009 	    !mlx4_vf_smi_enabled(dev, slave, port)) {
1010 		if (!(smp->mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED &&
1011 		      smp->method == IB_MGMT_METHOD_GET) || network_view) {
1012 			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",
1013 				 slave, smp->mgmt_class, smp->method,
1014 				 network_view ? "Network" : "Host",
1015 				 be16_to_cpu(smp->attr_id));
1016 			return -EPERM;
1017 		}
1018 	}
1019 
1020 	return mlx4_cmd_box(dev, inbox->dma, outbox->dma,
1021 				    vhcr->in_modifier, opcode_modifier,
1022 				    vhcr->op, MLX4_CMD_TIME_CLASS_C, MLX4_CMD_NATIVE);
1023 }
1024 
1025 static int mlx4_CMD_EPERM_wrapper(struct mlx4_dev *dev, int slave,
1026 		     struct mlx4_vhcr *vhcr,
1027 		     struct mlx4_cmd_mailbox *inbox,
1028 		     struct mlx4_cmd_mailbox *outbox,
1029 		     struct mlx4_cmd_info *cmd)
1030 {
1031 	return -EPERM;
1032 }
1033 
1034 int mlx4_DMA_wrapper(struct mlx4_dev *dev, int slave,
1035 		     struct mlx4_vhcr *vhcr,
1036 		     struct mlx4_cmd_mailbox *inbox,
1037 		     struct mlx4_cmd_mailbox *outbox,
1038 		     struct mlx4_cmd_info *cmd)
1039 {
1040 	u64 in_param;
1041 	u64 out_param;
1042 	int err;
1043 
1044 	in_param = cmd->has_inbox ? (u64) inbox->dma : vhcr->in_param;
1045 	out_param = cmd->has_outbox ? (u64) outbox->dma : vhcr->out_param;
1046 	if (cmd->encode_slave_id) {
1047 		in_param &= 0xffffffffffffff00ll;
1048 		in_param |= slave;
1049 	}
1050 
1051 	err = __mlx4_cmd(dev, in_param, &out_param, cmd->out_is_imm,
1052 			 vhcr->in_modifier, vhcr->op_modifier, vhcr->op,
1053 			 MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1054 
1055 	if (cmd->out_is_imm)
1056 		vhcr->out_param = out_param;
1057 
1058 	return err;
1059 }
1060 
1061 static struct mlx4_cmd_info cmd_info[] = {
1062 	{
1063 		.opcode = MLX4_CMD_QUERY_FW,
1064 		.has_inbox = false,
1065 		.has_outbox = true,
1066 		.out_is_imm = false,
1067 		.encode_slave_id = false,
1068 		.verify = NULL,
1069 		.wrapper = mlx4_QUERY_FW_wrapper
1070 	},
1071 	{
1072 		.opcode = MLX4_CMD_QUERY_HCA,
1073 		.has_inbox = false,
1074 		.has_outbox = true,
1075 		.out_is_imm = false,
1076 		.encode_slave_id = false,
1077 		.verify = NULL,
1078 		.wrapper = NULL
1079 	},
1080 	{
1081 		.opcode = MLX4_CMD_QUERY_DEV_CAP,
1082 		.has_inbox = false,
1083 		.has_outbox = true,
1084 		.out_is_imm = false,
1085 		.encode_slave_id = false,
1086 		.verify = NULL,
1087 		.wrapper = mlx4_QUERY_DEV_CAP_wrapper
1088 	},
1089 	{
1090 		.opcode = MLX4_CMD_QUERY_FUNC_CAP,
1091 		.has_inbox = false,
1092 		.has_outbox = true,
1093 		.out_is_imm = false,
1094 		.encode_slave_id = false,
1095 		.verify = NULL,
1096 		.wrapper = mlx4_QUERY_FUNC_CAP_wrapper
1097 	},
1098 	{
1099 		.opcode = MLX4_CMD_QUERY_ADAPTER,
1100 		.has_inbox = false,
1101 		.has_outbox = true,
1102 		.out_is_imm = false,
1103 		.encode_slave_id = false,
1104 		.verify = NULL,
1105 		.wrapper = NULL
1106 	},
1107 	{
1108 		.opcode = MLX4_CMD_INIT_PORT,
1109 		.has_inbox = false,
1110 		.has_outbox = false,
1111 		.out_is_imm = false,
1112 		.encode_slave_id = false,
1113 		.verify = NULL,
1114 		.wrapper = mlx4_INIT_PORT_wrapper
1115 	},
1116 	{
1117 		.opcode = MLX4_CMD_CLOSE_PORT,
1118 		.has_inbox = false,
1119 		.has_outbox = false,
1120 		.out_is_imm  = false,
1121 		.encode_slave_id = false,
1122 		.verify = NULL,
1123 		.wrapper = mlx4_CLOSE_PORT_wrapper
1124 	},
1125 	{
1126 		.opcode = MLX4_CMD_QUERY_PORT,
1127 		.has_inbox = false,
1128 		.has_outbox = true,
1129 		.out_is_imm = false,
1130 		.encode_slave_id = false,
1131 		.verify = NULL,
1132 		.wrapper = mlx4_QUERY_PORT_wrapper
1133 	},
1134 	{
1135 		.opcode = MLX4_CMD_SET_PORT,
1136 		.has_inbox = true,
1137 		.has_outbox = false,
1138 		.out_is_imm = false,
1139 		.encode_slave_id = false,
1140 		.verify = NULL,
1141 		.wrapper = mlx4_SET_PORT_wrapper
1142 	},
1143 	{
1144 		.opcode = MLX4_CMD_MAP_EQ,
1145 		.has_inbox = false,
1146 		.has_outbox = false,
1147 		.out_is_imm = false,
1148 		.encode_slave_id = false,
1149 		.verify = NULL,
1150 		.wrapper = mlx4_MAP_EQ_wrapper
1151 	},
1152 	{
1153 		.opcode = MLX4_CMD_SW2HW_EQ,
1154 		.has_inbox = true,
1155 		.has_outbox = false,
1156 		.out_is_imm = false,
1157 		.encode_slave_id = true,
1158 		.verify = NULL,
1159 		.wrapper = mlx4_SW2HW_EQ_wrapper
1160 	},
1161 	{
1162 		.opcode = MLX4_CMD_HW_HEALTH_CHECK,
1163 		.has_inbox = false,
1164 		.has_outbox = false,
1165 		.out_is_imm = false,
1166 		.encode_slave_id = false,
1167 		.verify = NULL,
1168 		.wrapper = NULL
1169 	},
1170 	{
1171 		.opcode = MLX4_CMD_NOP,
1172 		.has_inbox = false,
1173 		.has_outbox = false,
1174 		.out_is_imm = false,
1175 		.encode_slave_id = false,
1176 		.verify = NULL,
1177 		.wrapper = NULL
1178 	},
1179 	{
1180 		.opcode = MLX4_CMD_CONFIG_DEV,
1181 		.has_inbox = false,
1182 		.has_outbox = true,
1183 		.out_is_imm = false,
1184 		.encode_slave_id = false,
1185 		.verify = NULL,
1186 		.wrapper = mlx4_CONFIG_DEV_wrapper
1187 	},
1188 	{
1189 		.opcode = MLX4_CMD_ALLOC_RES,
1190 		.has_inbox = false,
1191 		.has_outbox = false,
1192 		.out_is_imm = true,
1193 		.encode_slave_id = false,
1194 		.verify = NULL,
1195 		.wrapper = mlx4_ALLOC_RES_wrapper
1196 	},
1197 	{
1198 		.opcode = MLX4_CMD_FREE_RES,
1199 		.has_inbox = false,
1200 		.has_outbox = false,
1201 		.out_is_imm = false,
1202 		.encode_slave_id = false,
1203 		.verify = NULL,
1204 		.wrapper = mlx4_FREE_RES_wrapper
1205 	},
1206 	{
1207 		.opcode = MLX4_CMD_SW2HW_MPT,
1208 		.has_inbox = true,
1209 		.has_outbox = false,
1210 		.out_is_imm = false,
1211 		.encode_slave_id = true,
1212 		.verify = NULL,
1213 		.wrapper = mlx4_SW2HW_MPT_wrapper
1214 	},
1215 	{
1216 		.opcode = MLX4_CMD_QUERY_MPT,
1217 		.has_inbox = false,
1218 		.has_outbox = true,
1219 		.out_is_imm = false,
1220 		.encode_slave_id = false,
1221 		.verify = NULL,
1222 		.wrapper = mlx4_QUERY_MPT_wrapper
1223 	},
1224 	{
1225 		.opcode = MLX4_CMD_HW2SW_MPT,
1226 		.has_inbox = false,
1227 		.has_outbox = false,
1228 		.out_is_imm = false,
1229 		.encode_slave_id = false,
1230 		.verify = NULL,
1231 		.wrapper = mlx4_HW2SW_MPT_wrapper
1232 	},
1233 	{
1234 		.opcode = MLX4_CMD_READ_MTT,
1235 		.has_inbox = false,
1236 		.has_outbox = true,
1237 		.out_is_imm = false,
1238 		.encode_slave_id = false,
1239 		.verify = NULL,
1240 		.wrapper = NULL
1241 	},
1242 	{
1243 		.opcode = MLX4_CMD_WRITE_MTT,
1244 		.has_inbox = true,
1245 		.has_outbox = false,
1246 		.out_is_imm = false,
1247 		.encode_slave_id = false,
1248 		.verify = NULL,
1249 		.wrapper = mlx4_WRITE_MTT_wrapper
1250 	},
1251 	{
1252 		.opcode = MLX4_CMD_SYNC_TPT,
1253 		.has_inbox = true,
1254 		.has_outbox = false,
1255 		.out_is_imm = false,
1256 		.encode_slave_id = false,
1257 		.verify = NULL,
1258 		.wrapper = NULL
1259 	},
1260 	{
1261 		.opcode = MLX4_CMD_HW2SW_EQ,
1262 		.has_inbox = false,
1263 		.has_outbox = false,
1264 		.out_is_imm = false,
1265 		.encode_slave_id = true,
1266 		.verify = NULL,
1267 		.wrapper = mlx4_HW2SW_EQ_wrapper
1268 	},
1269 	{
1270 		.opcode = MLX4_CMD_QUERY_EQ,
1271 		.has_inbox = false,
1272 		.has_outbox = true,
1273 		.out_is_imm = false,
1274 		.encode_slave_id = true,
1275 		.verify = NULL,
1276 		.wrapper = mlx4_QUERY_EQ_wrapper
1277 	},
1278 	{
1279 		.opcode = MLX4_CMD_SW2HW_CQ,
1280 		.has_inbox = true,
1281 		.has_outbox = false,
1282 		.out_is_imm = false,
1283 		.encode_slave_id = true,
1284 		.verify = NULL,
1285 		.wrapper = mlx4_SW2HW_CQ_wrapper
1286 	},
1287 	{
1288 		.opcode = MLX4_CMD_HW2SW_CQ,
1289 		.has_inbox = false,
1290 		.has_outbox = false,
1291 		.out_is_imm = false,
1292 		.encode_slave_id = false,
1293 		.verify = NULL,
1294 		.wrapper = mlx4_HW2SW_CQ_wrapper
1295 	},
1296 	{
1297 		.opcode = MLX4_CMD_QUERY_CQ,
1298 		.has_inbox = false,
1299 		.has_outbox = true,
1300 		.out_is_imm = false,
1301 		.encode_slave_id = false,
1302 		.verify = NULL,
1303 		.wrapper = mlx4_QUERY_CQ_wrapper
1304 	},
1305 	{
1306 		.opcode = MLX4_CMD_MODIFY_CQ,
1307 		.has_inbox = true,
1308 		.has_outbox = false,
1309 		.out_is_imm = true,
1310 		.encode_slave_id = false,
1311 		.verify = NULL,
1312 		.wrapper = mlx4_MODIFY_CQ_wrapper
1313 	},
1314 	{
1315 		.opcode = MLX4_CMD_SW2HW_SRQ,
1316 		.has_inbox = true,
1317 		.has_outbox = false,
1318 		.out_is_imm = false,
1319 		.encode_slave_id = true,
1320 		.verify = NULL,
1321 		.wrapper = mlx4_SW2HW_SRQ_wrapper
1322 	},
1323 	{
1324 		.opcode = MLX4_CMD_HW2SW_SRQ,
1325 		.has_inbox = false,
1326 		.has_outbox = false,
1327 		.out_is_imm = false,
1328 		.encode_slave_id = false,
1329 		.verify = NULL,
1330 		.wrapper = mlx4_HW2SW_SRQ_wrapper
1331 	},
1332 	{
1333 		.opcode = MLX4_CMD_QUERY_SRQ,
1334 		.has_inbox = false,
1335 		.has_outbox = true,
1336 		.out_is_imm = false,
1337 		.encode_slave_id = false,
1338 		.verify = NULL,
1339 		.wrapper = mlx4_QUERY_SRQ_wrapper
1340 	},
1341 	{
1342 		.opcode = MLX4_CMD_ARM_SRQ,
1343 		.has_inbox = false,
1344 		.has_outbox = false,
1345 		.out_is_imm = false,
1346 		.encode_slave_id = false,
1347 		.verify = NULL,
1348 		.wrapper = mlx4_ARM_SRQ_wrapper
1349 	},
1350 	{
1351 		.opcode = MLX4_CMD_RST2INIT_QP,
1352 		.has_inbox = true,
1353 		.has_outbox = false,
1354 		.out_is_imm = false,
1355 		.encode_slave_id = true,
1356 		.verify = NULL,
1357 		.wrapper = mlx4_RST2INIT_QP_wrapper
1358 	},
1359 	{
1360 		.opcode = MLX4_CMD_INIT2INIT_QP,
1361 		.has_inbox = true,
1362 		.has_outbox = false,
1363 		.out_is_imm = false,
1364 		.encode_slave_id = false,
1365 		.verify = NULL,
1366 		.wrapper = mlx4_INIT2INIT_QP_wrapper
1367 	},
1368 	{
1369 		.opcode = MLX4_CMD_INIT2RTR_QP,
1370 		.has_inbox = true,
1371 		.has_outbox = false,
1372 		.out_is_imm = false,
1373 		.encode_slave_id = false,
1374 		.verify = NULL,
1375 		.wrapper = mlx4_INIT2RTR_QP_wrapper
1376 	},
1377 	{
1378 		.opcode = MLX4_CMD_RTR2RTS_QP,
1379 		.has_inbox = true,
1380 		.has_outbox = false,
1381 		.out_is_imm = false,
1382 		.encode_slave_id = false,
1383 		.verify = NULL,
1384 		.wrapper = mlx4_RTR2RTS_QP_wrapper
1385 	},
1386 	{
1387 		.opcode = MLX4_CMD_RTS2RTS_QP,
1388 		.has_inbox = true,
1389 		.has_outbox = false,
1390 		.out_is_imm = false,
1391 		.encode_slave_id = false,
1392 		.verify = NULL,
1393 		.wrapper = mlx4_RTS2RTS_QP_wrapper
1394 	},
1395 	{
1396 		.opcode = MLX4_CMD_SQERR2RTS_QP,
1397 		.has_inbox = true,
1398 		.has_outbox = false,
1399 		.out_is_imm = false,
1400 		.encode_slave_id = false,
1401 		.verify = NULL,
1402 		.wrapper = mlx4_SQERR2RTS_QP_wrapper
1403 	},
1404 	{
1405 		.opcode = MLX4_CMD_2ERR_QP,
1406 		.has_inbox = false,
1407 		.has_outbox = false,
1408 		.out_is_imm = false,
1409 		.encode_slave_id = false,
1410 		.verify = NULL,
1411 		.wrapper = mlx4_GEN_QP_wrapper
1412 	},
1413 	{
1414 		.opcode = MLX4_CMD_RTS2SQD_QP,
1415 		.has_inbox = false,
1416 		.has_outbox = false,
1417 		.out_is_imm = false,
1418 		.encode_slave_id = false,
1419 		.verify = NULL,
1420 		.wrapper = mlx4_GEN_QP_wrapper
1421 	},
1422 	{
1423 		.opcode = MLX4_CMD_SQD2SQD_QP,
1424 		.has_inbox = true,
1425 		.has_outbox = false,
1426 		.out_is_imm = false,
1427 		.encode_slave_id = false,
1428 		.verify = NULL,
1429 		.wrapper = mlx4_SQD2SQD_QP_wrapper
1430 	},
1431 	{
1432 		.opcode = MLX4_CMD_SQD2RTS_QP,
1433 		.has_inbox = true,
1434 		.has_outbox = false,
1435 		.out_is_imm = false,
1436 		.encode_slave_id = false,
1437 		.verify = NULL,
1438 		.wrapper = mlx4_SQD2RTS_QP_wrapper
1439 	},
1440 	{
1441 		.opcode = MLX4_CMD_2RST_QP,
1442 		.has_inbox = false,
1443 		.has_outbox = false,
1444 		.out_is_imm = false,
1445 		.encode_slave_id = false,
1446 		.verify = NULL,
1447 		.wrapper = mlx4_2RST_QP_wrapper
1448 	},
1449 	{
1450 		.opcode = MLX4_CMD_QUERY_QP,
1451 		.has_inbox = false,
1452 		.has_outbox = true,
1453 		.out_is_imm = false,
1454 		.encode_slave_id = false,
1455 		.verify = NULL,
1456 		.wrapper = mlx4_GEN_QP_wrapper
1457 	},
1458 	{
1459 		.opcode = MLX4_CMD_SUSPEND_QP,
1460 		.has_inbox = false,
1461 		.has_outbox = false,
1462 		.out_is_imm = false,
1463 		.encode_slave_id = false,
1464 		.verify = NULL,
1465 		.wrapper = mlx4_GEN_QP_wrapper
1466 	},
1467 	{
1468 		.opcode = MLX4_CMD_UNSUSPEND_QP,
1469 		.has_inbox = false,
1470 		.has_outbox = false,
1471 		.out_is_imm = false,
1472 		.encode_slave_id = false,
1473 		.verify = NULL,
1474 		.wrapper = mlx4_GEN_QP_wrapper
1475 	},
1476 	{
1477 		.opcode = MLX4_CMD_UPDATE_QP,
1478 		.has_inbox = true,
1479 		.has_outbox = false,
1480 		.out_is_imm = false,
1481 		.encode_slave_id = false,
1482 		.verify = NULL,
1483 		.wrapper = mlx4_UPDATE_QP_wrapper
1484 	},
1485 	{
1486 		.opcode = MLX4_CMD_GET_OP_REQ,
1487 		.has_inbox = false,
1488 		.has_outbox = false,
1489 		.out_is_imm = false,
1490 		.encode_slave_id = false,
1491 		.verify = NULL,
1492 		.wrapper = mlx4_CMD_EPERM_wrapper,
1493 	},
1494 	{
1495 		.opcode = MLX4_CMD_ALLOCATE_VPP,
1496 		.has_inbox = false,
1497 		.has_outbox = true,
1498 		.out_is_imm = false,
1499 		.encode_slave_id = false,
1500 		.verify = NULL,
1501 		.wrapper = mlx4_CMD_EPERM_wrapper,
1502 	},
1503 	{
1504 		.opcode = MLX4_CMD_SET_VPORT_QOS,
1505 		.has_inbox = false,
1506 		.has_outbox = true,
1507 		.out_is_imm = false,
1508 		.encode_slave_id = false,
1509 		.verify = NULL,
1510 		.wrapper = mlx4_CMD_EPERM_wrapper,
1511 	},
1512 	{
1513 		.opcode = MLX4_CMD_CONF_SPECIAL_QP,
1514 		.has_inbox = false,
1515 		.has_outbox = false,
1516 		.out_is_imm = false,
1517 		.encode_slave_id = false,
1518 		.verify = NULL, /* XXX verify: only demux can do this */
1519 		.wrapper = NULL
1520 	},
1521 	{
1522 		.opcode = MLX4_CMD_MAD_IFC,
1523 		.has_inbox = true,
1524 		.has_outbox = true,
1525 		.out_is_imm = false,
1526 		.encode_slave_id = false,
1527 		.verify = NULL,
1528 		.wrapper = mlx4_MAD_IFC_wrapper
1529 	},
1530 	{
1531 		.opcode = MLX4_CMD_MAD_DEMUX,
1532 		.has_inbox = false,
1533 		.has_outbox = false,
1534 		.out_is_imm = false,
1535 		.encode_slave_id = false,
1536 		.verify = NULL,
1537 		.wrapper = mlx4_CMD_EPERM_wrapper
1538 	},
1539 	{
1540 		.opcode = MLX4_CMD_QUERY_IF_STAT,
1541 		.has_inbox = false,
1542 		.has_outbox = true,
1543 		.out_is_imm = false,
1544 		.encode_slave_id = false,
1545 		.verify = NULL,
1546 		.wrapper = mlx4_QUERY_IF_STAT_wrapper
1547 	},
1548 	{
1549 		.opcode = MLX4_CMD_ACCESS_REG,
1550 		.has_inbox = true,
1551 		.has_outbox = true,
1552 		.out_is_imm = false,
1553 		.encode_slave_id = false,
1554 		.verify = NULL,
1555 		.wrapper = mlx4_ACCESS_REG_wrapper,
1556 	},
1557 	{
1558 		.opcode = MLX4_CMD_CONGESTION_CTRL_OPCODE,
1559 		.has_inbox = false,
1560 		.has_outbox = false,
1561 		.out_is_imm = false,
1562 		.encode_slave_id = false,
1563 		.verify = NULL,
1564 		.wrapper = mlx4_CMD_EPERM_wrapper,
1565 	},
1566 	/* Native multicast commands are not available for guests */
1567 	{
1568 		.opcode = MLX4_CMD_QP_ATTACH,
1569 		.has_inbox = true,
1570 		.has_outbox = false,
1571 		.out_is_imm = false,
1572 		.encode_slave_id = false,
1573 		.verify = NULL,
1574 		.wrapper = mlx4_QP_ATTACH_wrapper
1575 	},
1576 	{
1577 		.opcode = MLX4_CMD_PROMISC,
1578 		.has_inbox = false,
1579 		.has_outbox = false,
1580 		.out_is_imm = false,
1581 		.encode_slave_id = false,
1582 		.verify = NULL,
1583 		.wrapper = mlx4_PROMISC_wrapper
1584 	},
1585 	/* Ethernet specific commands */
1586 	{
1587 		.opcode = MLX4_CMD_SET_VLAN_FLTR,
1588 		.has_inbox = true,
1589 		.has_outbox = false,
1590 		.out_is_imm = false,
1591 		.encode_slave_id = false,
1592 		.verify = NULL,
1593 		.wrapper = mlx4_SET_VLAN_FLTR_wrapper
1594 	},
1595 	{
1596 		.opcode = MLX4_CMD_SET_MCAST_FLTR,
1597 		.has_inbox = false,
1598 		.has_outbox = false,
1599 		.out_is_imm = false,
1600 		.encode_slave_id = false,
1601 		.verify = NULL,
1602 		.wrapper = mlx4_SET_MCAST_FLTR_wrapper
1603 	},
1604 	{
1605 		.opcode = MLX4_CMD_DUMP_ETH_STATS,
1606 		.has_inbox = false,
1607 		.has_outbox = true,
1608 		.out_is_imm = false,
1609 		.encode_slave_id = false,
1610 		.verify = NULL,
1611 		.wrapper = mlx4_DUMP_ETH_STATS_wrapper
1612 	},
1613 	{
1614 		.opcode = MLX4_CMD_INFORM_FLR_DONE,
1615 		.has_inbox = false,
1616 		.has_outbox = false,
1617 		.out_is_imm = false,
1618 		.encode_slave_id = false,
1619 		.verify = NULL,
1620 		.wrapper = NULL
1621 	},
1622 	/* flow steering commands */
1623 	{
1624 		.opcode = MLX4_QP_FLOW_STEERING_ATTACH,
1625 		.has_inbox = true,
1626 		.has_outbox = false,
1627 		.out_is_imm = true,
1628 		.encode_slave_id = false,
1629 		.verify = NULL,
1630 		.wrapper = mlx4_QP_FLOW_STEERING_ATTACH_wrapper
1631 	},
1632 	{
1633 		.opcode = MLX4_QP_FLOW_STEERING_DETACH,
1634 		.has_inbox = false,
1635 		.has_outbox = false,
1636 		.out_is_imm = false,
1637 		.encode_slave_id = false,
1638 		.verify = NULL,
1639 		.wrapper = mlx4_QP_FLOW_STEERING_DETACH_wrapper
1640 	},
1641 	{
1642 		.opcode = MLX4_FLOW_STEERING_IB_UC_QP_RANGE,
1643 		.has_inbox = false,
1644 		.has_outbox = false,
1645 		.out_is_imm = false,
1646 		.encode_slave_id = false,
1647 		.verify = NULL,
1648 		.wrapper = mlx4_CMD_EPERM_wrapper
1649 	},
1650 	{
1651 		.opcode = MLX4_CMD_VIRT_PORT_MAP,
1652 		.has_inbox = false,
1653 		.has_outbox = false,
1654 		.out_is_imm = false,
1655 		.encode_slave_id = false,
1656 		.verify = NULL,
1657 		.wrapper = mlx4_CMD_EPERM_wrapper
1658 	},
1659 };
1660 
1661 static int mlx4_master_process_vhcr(struct mlx4_dev *dev, int slave,
1662 				    struct mlx4_vhcr_cmd *in_vhcr)
1663 {
1664 	struct mlx4_priv *priv = mlx4_priv(dev);
1665 	struct mlx4_cmd_info *cmd = NULL;
1666 	struct mlx4_vhcr_cmd *vhcr_cmd = in_vhcr ? in_vhcr : priv->mfunc.vhcr;
1667 	struct mlx4_vhcr *vhcr;
1668 	struct mlx4_cmd_mailbox *inbox = NULL;
1669 	struct mlx4_cmd_mailbox *outbox = NULL;
1670 	u64 in_param;
1671 	u64 out_param;
1672 	int ret = 0;
1673 	int i;
1674 	int err = 0;
1675 
1676 	/* Create sw representation of Virtual HCR */
1677 	vhcr = kzalloc(sizeof(struct mlx4_vhcr), GFP_KERNEL);
1678 	if (!vhcr)
1679 		return -ENOMEM;
1680 
1681 	/* DMA in the vHCR */
1682 	if (!in_vhcr) {
1683 		ret = mlx4_ACCESS_MEM(dev, priv->mfunc.vhcr_dma, slave,
1684 				      priv->mfunc.master.slave_state[slave].vhcr_dma,
1685 				      ALIGN(sizeof(struct mlx4_vhcr_cmd),
1686 					    MLX4_ACCESS_MEM_ALIGN), 1);
1687 		if (ret) {
1688 			if (!(dev->persist->state &
1689 			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
1690 				mlx4_err(dev, "%s: Failed reading vhcr ret: 0x%x\n",
1691 					 __func__, ret);
1692 			kfree(vhcr);
1693 			return ret;
1694 		}
1695 	}
1696 
1697 	/* Fill SW VHCR fields */
1698 	vhcr->in_param = be64_to_cpu(vhcr_cmd->in_param);
1699 	vhcr->out_param = be64_to_cpu(vhcr_cmd->out_param);
1700 	vhcr->in_modifier = be32_to_cpu(vhcr_cmd->in_modifier);
1701 	vhcr->token = be16_to_cpu(vhcr_cmd->token);
1702 	vhcr->op = be16_to_cpu(vhcr_cmd->opcode) & 0xfff;
1703 	vhcr->op_modifier = (u8) (be16_to_cpu(vhcr_cmd->opcode) >> 12);
1704 	vhcr->e_bit = vhcr_cmd->flags & (1 << 6);
1705 
1706 	/* Lookup command */
1707 	for (i = 0; i < ARRAY_SIZE(cmd_info); ++i) {
1708 		if (vhcr->op == cmd_info[i].opcode) {
1709 			cmd = &cmd_info[i];
1710 			break;
1711 		}
1712 	}
1713 	if (!cmd) {
1714 		mlx4_err(dev, "Unknown command:0x%x accepted from slave:%d\n",
1715 			 vhcr->op, slave);
1716 		vhcr_cmd->status = CMD_STAT_BAD_PARAM;
1717 		goto out_status;
1718 	}
1719 
1720 	/* Read inbox */
1721 	if (cmd->has_inbox) {
1722 		vhcr->in_param &= INBOX_MASK;
1723 		inbox = mlx4_alloc_cmd_mailbox(dev);
1724 		if (IS_ERR(inbox)) {
1725 			vhcr_cmd->status = CMD_STAT_BAD_SIZE;
1726 			inbox = NULL;
1727 			goto out_status;
1728 		}
1729 
1730 		ret = mlx4_ACCESS_MEM(dev, inbox->dma, slave,
1731 				      vhcr->in_param,
1732 				      MLX4_MAILBOX_SIZE, 1);
1733 		if (ret) {
1734 			if (!(dev->persist->state &
1735 			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
1736 				mlx4_err(dev, "%s: Failed reading inbox (cmd:0x%x)\n",
1737 					 __func__, cmd->opcode);
1738 			vhcr_cmd->status = CMD_STAT_INTERNAL_ERR;
1739 			goto out_status;
1740 		}
1741 	}
1742 
1743 	/* Apply permission and bound checks if applicable */
1744 	if (cmd->verify && cmd->verify(dev, slave, vhcr, inbox)) {
1745 		mlx4_warn(dev, "Command:0x%x from slave: %d failed protection checks for resource_id:%d\n",
1746 			  vhcr->op, slave, vhcr->in_modifier);
1747 		vhcr_cmd->status = CMD_STAT_BAD_OP;
1748 		goto out_status;
1749 	}
1750 
1751 	/* Allocate outbox */
1752 	if (cmd->has_outbox) {
1753 		outbox = mlx4_alloc_cmd_mailbox(dev);
1754 		if (IS_ERR(outbox)) {
1755 			vhcr_cmd->status = CMD_STAT_BAD_SIZE;
1756 			outbox = NULL;
1757 			goto out_status;
1758 		}
1759 	}
1760 
1761 	/* Execute the command! */
1762 	if (cmd->wrapper) {
1763 		err = cmd->wrapper(dev, slave, vhcr, inbox, outbox,
1764 				   cmd);
1765 		if (cmd->out_is_imm)
1766 			vhcr_cmd->out_param = cpu_to_be64(vhcr->out_param);
1767 	} else {
1768 		in_param = cmd->has_inbox ? (u64) inbox->dma :
1769 			vhcr->in_param;
1770 		out_param = cmd->has_outbox ? (u64) outbox->dma :
1771 			vhcr->out_param;
1772 		err = __mlx4_cmd(dev, in_param, &out_param,
1773 				 cmd->out_is_imm, vhcr->in_modifier,
1774 				 vhcr->op_modifier, vhcr->op,
1775 				 MLX4_CMD_TIME_CLASS_A,
1776 				 MLX4_CMD_NATIVE);
1777 
1778 		if (cmd->out_is_imm) {
1779 			vhcr->out_param = out_param;
1780 			vhcr_cmd->out_param = cpu_to_be64(vhcr->out_param);
1781 		}
1782 	}
1783 
1784 	if (err) {
1785 		if (!(dev->persist->state & MLX4_DEVICE_STATE_INTERNAL_ERROR))
1786 			mlx4_warn(dev, "vhcr command:0x%x slave:%d failed with error:%d, status %d\n",
1787 				  vhcr->op, slave, vhcr->errno, err);
1788 		vhcr_cmd->status = mlx4_errno_to_status(err);
1789 		goto out_status;
1790 	}
1791 
1792 
1793 	/* Write outbox if command completed successfully */
1794 	if (cmd->has_outbox && !vhcr_cmd->status) {
1795 		ret = mlx4_ACCESS_MEM(dev, outbox->dma, slave,
1796 				      vhcr->out_param,
1797 				      MLX4_MAILBOX_SIZE, MLX4_CMD_WRAPPED);
1798 		if (ret) {
1799 			/* If we failed to write back the outbox after the
1800 			 *command was successfully executed, we must fail this
1801 			 * slave, as it is now in undefined state */
1802 			if (!(dev->persist->state &
1803 			    MLX4_DEVICE_STATE_INTERNAL_ERROR))
1804 				mlx4_err(dev, "%s:Failed writing outbox\n", __func__);
1805 			goto out;
1806 		}
1807 	}
1808 
1809 out_status:
1810 	/* DMA back vhcr result */
1811 	if (!in_vhcr) {
1812 		ret = mlx4_ACCESS_MEM(dev, priv->mfunc.vhcr_dma, slave,
1813 				      priv->mfunc.master.slave_state[slave].vhcr_dma,
1814 				      ALIGN(sizeof(struct mlx4_vhcr),
1815 					    MLX4_ACCESS_MEM_ALIGN),
1816 				      MLX4_CMD_WRAPPED);
1817 		if (ret)
1818 			mlx4_err(dev, "%s:Failed writing vhcr result\n",
1819 				 __func__);
1820 		else if (vhcr->e_bit &&
1821 			 mlx4_GEN_EQE(dev, slave, &priv->mfunc.master.cmd_eqe))
1822 				mlx4_warn(dev, "Failed to generate command completion eqe for slave %d\n",
1823 					  slave);
1824 	}
1825 
1826 out:
1827 	kfree(vhcr);
1828 	mlx4_free_cmd_mailbox(dev, inbox);
1829 	mlx4_free_cmd_mailbox(dev, outbox);
1830 	return ret;
1831 }
1832 
1833 static int mlx4_master_immediate_activate_vlan_qos(struct mlx4_priv *priv,
1834 					    int slave, int port)
1835 {
1836 	struct mlx4_vport_oper_state *vp_oper;
1837 	struct mlx4_vport_state *vp_admin;
1838 	struct mlx4_vf_immed_vlan_work *work;
1839 	struct mlx4_dev *dev = &(priv->dev);
1840 	int err;
1841 	int admin_vlan_ix = NO_INDX;
1842 
1843 	vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
1844 	vp_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
1845 
1846 	if (vp_oper->state.default_vlan == vp_admin->default_vlan &&
1847 	    vp_oper->state.default_qos == vp_admin->default_qos &&
1848 	    vp_oper->state.link_state == vp_admin->link_state &&
1849 	    vp_oper->state.qos_vport == vp_admin->qos_vport)
1850 		return 0;
1851 
1852 	if (!(priv->mfunc.master.slave_state[slave].active &&
1853 	      dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_UPDATE_QP)) {
1854 		/* even if the UPDATE_QP command isn't supported, we still want
1855 		 * to set this VF link according to the admin directive
1856 		 */
1857 		vp_oper->state.link_state = vp_admin->link_state;
1858 		return -1;
1859 	}
1860 
1861 	mlx4_dbg(dev, "updating immediately admin params slave %d port %d\n",
1862 		 slave, port);
1863 	mlx4_dbg(dev, "vlan %d QoS %d link down %d\n",
1864 		 vp_admin->default_vlan, vp_admin->default_qos,
1865 		 vp_admin->link_state);
1866 
1867 	work = kzalloc(sizeof(*work), GFP_KERNEL);
1868 	if (!work)
1869 		return -ENOMEM;
1870 
1871 	if (vp_oper->state.default_vlan != vp_admin->default_vlan) {
1872 		if (MLX4_VGT != vp_admin->default_vlan) {
1873 			err = __mlx4_register_vlan(&priv->dev, port,
1874 						   vp_admin->default_vlan,
1875 						   &admin_vlan_ix);
1876 			if (err) {
1877 				kfree(work);
1878 				mlx4_warn(&priv->dev,
1879 					  "No vlan resources slave %d, port %d\n",
1880 					  slave, port);
1881 				return err;
1882 			}
1883 		} else {
1884 			admin_vlan_ix = NO_INDX;
1885 		}
1886 		work->flags |= MLX4_VF_IMMED_VLAN_FLAG_VLAN;
1887 		mlx4_dbg(&priv->dev,
1888 			 "alloc vlan %d idx  %d slave %d port %d\n",
1889 			 (int)(vp_admin->default_vlan),
1890 			 admin_vlan_ix, slave, port);
1891 	}
1892 
1893 	/* save original vlan ix and vlan id */
1894 	work->orig_vlan_id = vp_oper->state.default_vlan;
1895 	work->orig_vlan_ix = vp_oper->vlan_idx;
1896 
1897 	/* handle new qos */
1898 	if (vp_oper->state.default_qos != vp_admin->default_qos)
1899 		work->flags |= MLX4_VF_IMMED_VLAN_FLAG_QOS;
1900 
1901 	if (work->flags & MLX4_VF_IMMED_VLAN_FLAG_VLAN)
1902 		vp_oper->vlan_idx = admin_vlan_ix;
1903 
1904 	vp_oper->state.default_vlan = vp_admin->default_vlan;
1905 	vp_oper->state.default_qos = vp_admin->default_qos;
1906 	vp_oper->state.link_state = vp_admin->link_state;
1907 	vp_oper->state.qos_vport = vp_admin->qos_vport;
1908 
1909 	if (vp_admin->link_state == IFLA_VF_LINK_STATE_DISABLE)
1910 		work->flags |= MLX4_VF_IMMED_VLAN_FLAG_LINK_DISABLE;
1911 
1912 	/* iterate over QPs owned by this slave, using UPDATE_QP */
1913 	work->port = port;
1914 	work->slave = slave;
1915 	work->qos = vp_oper->state.default_qos;
1916 	work->qos_vport = vp_oper->state.qos_vport;
1917 	work->vlan_id = vp_oper->state.default_vlan;
1918 	work->vlan_ix = vp_oper->vlan_idx;
1919 	work->priv = priv;
1920 	INIT_WORK(&work->work, mlx4_vf_immed_vlan_work_handler);
1921 	queue_work(priv->mfunc.master.comm_wq, &work->work);
1922 
1923 	return 0;
1924 }
1925 
1926 static void mlx4_set_default_port_qos(struct mlx4_dev *dev, int port)
1927 {
1928 	struct mlx4_qos_manager *port_qos_ctl;
1929 	struct mlx4_priv *priv = mlx4_priv(dev);
1930 
1931 	port_qos_ctl = &priv->mfunc.master.qos_ctl[port];
1932 	bitmap_zero(port_qos_ctl->priority_bm, MLX4_NUM_UP);
1933 
1934 	/* Enable only default prio at PF init routine */
1935 	set_bit(MLX4_DEFAULT_QOS_PRIO, port_qos_ctl->priority_bm);
1936 }
1937 
1938 static void mlx4_allocate_port_vpps(struct mlx4_dev *dev, int port)
1939 {
1940 	int i;
1941 	int err;
1942 	int num_vfs;
1943 	u16 availible_vpp;
1944 	u8 vpp_param[MLX4_NUM_UP];
1945 	struct mlx4_qos_manager *port_qos;
1946 	struct mlx4_priv *priv = mlx4_priv(dev);
1947 
1948 	err = mlx4_ALLOCATE_VPP_get(dev, port, &availible_vpp, vpp_param);
1949 	if (err) {
1950 		mlx4_info(dev, "Failed query availible VPPs\n");
1951 		return;
1952 	}
1953 
1954 	port_qos = &priv->mfunc.master.qos_ctl[port];
1955 	num_vfs = (availible_vpp /
1956 		   bitmap_weight(port_qos->priority_bm, MLX4_NUM_UP));
1957 
1958 	for (i = 0; i < MLX4_NUM_UP; i++) {
1959 		if (test_bit(i, port_qos->priority_bm))
1960 			vpp_param[i] = num_vfs;
1961 	}
1962 
1963 	err = mlx4_ALLOCATE_VPP_set(dev, port, vpp_param);
1964 	if (err) {
1965 		mlx4_info(dev, "Failed allocating VPPs\n");
1966 		return;
1967 	}
1968 
1969 	/* Query actual allocated VPP, just to make sure */
1970 	err = mlx4_ALLOCATE_VPP_get(dev, port, &availible_vpp, vpp_param);
1971 	if (err) {
1972 		mlx4_info(dev, "Failed query availible VPPs\n");
1973 		return;
1974 	}
1975 
1976 	port_qos->num_of_qos_vfs = num_vfs;
1977 	mlx4_dbg(dev, "Port %d Availible VPPs %d\n", port, availible_vpp);
1978 
1979 	for (i = 0; i < MLX4_NUM_UP; i++)
1980 		mlx4_dbg(dev, "Port %d UP %d Allocated %d VPPs\n", port, i,
1981 			 vpp_param[i]);
1982 }
1983 
1984 static int mlx4_master_activate_admin_state(struct mlx4_priv *priv, int slave)
1985 {
1986 	int port, err;
1987 	struct mlx4_vport_state *vp_admin;
1988 	struct mlx4_vport_oper_state *vp_oper;
1989 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(
1990 			&priv->dev, slave);
1991 	int min_port = find_first_bit(actv_ports.ports,
1992 				      priv->dev.caps.num_ports) + 1;
1993 	int max_port = min_port - 1 +
1994 		bitmap_weight(actv_ports.ports, priv->dev.caps.num_ports);
1995 
1996 	for (port = min_port; port <= max_port; port++) {
1997 		if (!test_bit(port - 1, actv_ports.ports))
1998 			continue;
1999 		priv->mfunc.master.vf_oper[slave].smi_enabled[port] =
2000 			priv->mfunc.master.vf_admin[slave].enable_smi[port];
2001 		vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
2002 		vp_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
2003 		vp_oper->state = *vp_admin;
2004 		if (MLX4_VGT != vp_admin->default_vlan) {
2005 			err = __mlx4_register_vlan(&priv->dev, port,
2006 						   vp_admin->default_vlan, &(vp_oper->vlan_idx));
2007 			if (err) {
2008 				vp_oper->vlan_idx = NO_INDX;
2009 				mlx4_warn(&priv->dev,
2010 					  "No vlan resources slave %d, port %d\n",
2011 					  slave, port);
2012 				return err;
2013 			}
2014 			mlx4_dbg(&priv->dev, "alloc vlan %d idx  %d slave %d port %d\n",
2015 				 (int)(vp_oper->state.default_vlan),
2016 				 vp_oper->vlan_idx, slave, port);
2017 		}
2018 		if (vp_admin->spoofchk) {
2019 			vp_oper->mac_idx = __mlx4_register_mac(&priv->dev,
2020 							       port,
2021 							       vp_admin->mac);
2022 			if (0 > vp_oper->mac_idx) {
2023 				err = vp_oper->mac_idx;
2024 				vp_oper->mac_idx = NO_INDX;
2025 				mlx4_warn(&priv->dev,
2026 					  "No mac resources slave %d, port %d\n",
2027 					  slave, port);
2028 				return err;
2029 			}
2030 			mlx4_dbg(&priv->dev, "alloc mac %llx idx  %d slave %d port %d\n",
2031 				 vp_oper->state.mac, vp_oper->mac_idx, slave, port);
2032 		}
2033 	}
2034 	return 0;
2035 }
2036 
2037 static void mlx4_master_deactivate_admin_state(struct mlx4_priv *priv, int slave)
2038 {
2039 	int port;
2040 	struct mlx4_vport_oper_state *vp_oper;
2041 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(
2042 			&priv->dev, slave);
2043 	int min_port = find_first_bit(actv_ports.ports,
2044 				      priv->dev.caps.num_ports) + 1;
2045 	int max_port = min_port - 1 +
2046 		bitmap_weight(actv_ports.ports, priv->dev.caps.num_ports);
2047 
2048 
2049 	for (port = min_port; port <= max_port; port++) {
2050 		if (!test_bit(port - 1, actv_ports.ports))
2051 			continue;
2052 		priv->mfunc.master.vf_oper[slave].smi_enabled[port] =
2053 			MLX4_VF_SMI_DISABLED;
2054 		vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
2055 		if (NO_INDX != vp_oper->vlan_idx) {
2056 			__mlx4_unregister_vlan(&priv->dev,
2057 					       port, vp_oper->state.default_vlan);
2058 			vp_oper->vlan_idx = NO_INDX;
2059 		}
2060 		if (NO_INDX != vp_oper->mac_idx) {
2061 			__mlx4_unregister_mac(&priv->dev, port, vp_oper->state.mac);
2062 			vp_oper->mac_idx = NO_INDX;
2063 		}
2064 	}
2065 	return;
2066 }
2067 
2068 static void mlx4_master_do_cmd(struct mlx4_dev *dev, int slave, u8 cmd,
2069 			       u16 param, u8 toggle)
2070 {
2071 	struct mlx4_priv *priv = mlx4_priv(dev);
2072 	struct mlx4_slave_state *slave_state = priv->mfunc.master.slave_state;
2073 	u32 reply;
2074 	u8 is_going_down = 0;
2075 	int i;
2076 	unsigned long flags;
2077 
2078 	slave_state[slave].comm_toggle ^= 1;
2079 	reply = (u32) slave_state[slave].comm_toggle << 31;
2080 	if (toggle != slave_state[slave].comm_toggle) {
2081 		mlx4_warn(dev, "Incorrect toggle %d from slave %d. *** MASTER STATE COMPROMISED ***\n",
2082 			  toggle, slave);
2083 		goto reset_slave;
2084 	}
2085 	if (cmd == MLX4_COMM_CMD_RESET) {
2086 		mlx4_warn(dev, "Received reset from slave:%d\n", slave);
2087 		slave_state[slave].active = false;
2088 		slave_state[slave].old_vlan_api = false;
2089 		mlx4_master_deactivate_admin_state(priv, slave);
2090 		for (i = 0; i < MLX4_EVENT_TYPES_NUM; ++i) {
2091 				slave_state[slave].event_eq[i].eqn = -1;
2092 				slave_state[slave].event_eq[i].token = 0;
2093 		}
2094 		/*check if we are in the middle of FLR process,
2095 		if so return "retry" status to the slave*/
2096 		if (MLX4_COMM_CMD_FLR == slave_state[slave].last_cmd)
2097 			goto inform_slave_state;
2098 
2099 		mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_SHUTDOWN, slave);
2100 
2101 		/* write the version in the event field */
2102 		reply |= mlx4_comm_get_version();
2103 
2104 		goto reset_slave;
2105 	}
2106 	/*command from slave in the middle of FLR*/
2107 	if (cmd != MLX4_COMM_CMD_RESET &&
2108 	    MLX4_COMM_CMD_FLR == slave_state[slave].last_cmd) {
2109 		mlx4_warn(dev, "slave:%d is Trying to run cmd(0x%x) in the middle of FLR\n",
2110 			  slave, cmd);
2111 		return;
2112 	}
2113 
2114 	switch (cmd) {
2115 	case MLX4_COMM_CMD_VHCR0:
2116 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_RESET)
2117 			goto reset_slave;
2118 		slave_state[slave].vhcr_dma = ((u64) param) << 48;
2119 		priv->mfunc.master.slave_state[slave].cookie = 0;
2120 		break;
2121 	case MLX4_COMM_CMD_VHCR1:
2122 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR0)
2123 			goto reset_slave;
2124 		slave_state[slave].vhcr_dma |= ((u64) param) << 32;
2125 		break;
2126 	case MLX4_COMM_CMD_VHCR2:
2127 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR1)
2128 			goto reset_slave;
2129 		slave_state[slave].vhcr_dma |= ((u64) param) << 16;
2130 		break;
2131 	case MLX4_COMM_CMD_VHCR_EN:
2132 		if (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR2)
2133 			goto reset_slave;
2134 		slave_state[slave].vhcr_dma |= param;
2135 		if (mlx4_master_activate_admin_state(priv, slave))
2136 				goto reset_slave;
2137 		slave_state[slave].active = true;
2138 		mlx4_dispatch_event(dev, MLX4_DEV_EVENT_SLAVE_INIT, slave);
2139 		break;
2140 	case MLX4_COMM_CMD_VHCR_POST:
2141 		if ((slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_EN) &&
2142 		    (slave_state[slave].last_cmd != MLX4_COMM_CMD_VHCR_POST)) {
2143 			mlx4_warn(dev, "slave:%d is out of sync, cmd=0x%x, last command=0x%x, reset is needed\n",
2144 				  slave, cmd, slave_state[slave].last_cmd);
2145 			goto reset_slave;
2146 		}
2147 
2148 		mutex_lock(&priv->cmd.slave_cmd_mutex);
2149 		if (mlx4_master_process_vhcr(dev, slave, NULL)) {
2150 			mlx4_err(dev, "Failed processing vhcr for slave:%d, resetting slave\n",
2151 				 slave);
2152 			mutex_unlock(&priv->cmd.slave_cmd_mutex);
2153 			goto reset_slave;
2154 		}
2155 		mutex_unlock(&priv->cmd.slave_cmd_mutex);
2156 		break;
2157 	default:
2158 		mlx4_warn(dev, "Bad comm cmd:%d from slave:%d\n", cmd, slave);
2159 		goto reset_slave;
2160 	}
2161 	spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
2162 	if (!slave_state[slave].is_slave_going_down)
2163 		slave_state[slave].last_cmd = cmd;
2164 	else
2165 		is_going_down = 1;
2166 	spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
2167 	if (is_going_down) {
2168 		mlx4_warn(dev, "Slave is going down aborting command(%d) executing from slave:%d\n",
2169 			  cmd, slave);
2170 		return;
2171 	}
2172 	__raw_writel((__force u32) cpu_to_be32(reply),
2173 		     &priv->mfunc.comm[slave].slave_read);
2174 	mmiowb();
2175 
2176 	return;
2177 
2178 reset_slave:
2179 	/* cleanup any slave resources */
2180 	if (dev->persist->interface_state & MLX4_INTERFACE_STATE_UP)
2181 		mlx4_delete_all_resources_for_slave(dev, slave);
2182 
2183 	if (cmd != MLX4_COMM_CMD_RESET) {
2184 		mlx4_warn(dev, "Turn on internal error to force reset, slave=%d, cmd=0x%x\n",
2185 			  slave, cmd);
2186 		/* Turn on internal error letting slave reset itself immeditaly,
2187 		 * otherwise it might take till timeout on command is passed
2188 		 */
2189 		reply |= ((u32)COMM_CHAN_EVENT_INTERNAL_ERR);
2190 	}
2191 
2192 	spin_lock_irqsave(&priv->mfunc.master.slave_state_lock, flags);
2193 	if (!slave_state[slave].is_slave_going_down)
2194 		slave_state[slave].last_cmd = MLX4_COMM_CMD_RESET;
2195 	spin_unlock_irqrestore(&priv->mfunc.master.slave_state_lock, flags);
2196 	/*with slave in the middle of flr, no need to clean resources again.*/
2197 inform_slave_state:
2198 	memset(&slave_state[slave].event_eq, 0,
2199 	       sizeof(struct mlx4_slave_event_eq_info));
2200 	__raw_writel((__force u32) cpu_to_be32(reply),
2201 		     &priv->mfunc.comm[slave].slave_read);
2202 	wmb();
2203 }
2204 
2205 /* master command processing */
2206 void mlx4_master_comm_channel(struct work_struct *work)
2207 {
2208 	struct mlx4_mfunc_master_ctx *master =
2209 		container_of(work,
2210 			     struct mlx4_mfunc_master_ctx,
2211 			     comm_work);
2212 	struct mlx4_mfunc *mfunc =
2213 		container_of(master, struct mlx4_mfunc, master);
2214 	struct mlx4_priv *priv =
2215 		container_of(mfunc, struct mlx4_priv, mfunc);
2216 	struct mlx4_dev *dev = &priv->dev;
2217 	__be32 *bit_vec;
2218 	u32 comm_cmd;
2219 	u32 vec;
2220 	int i, j, slave;
2221 	int toggle;
2222 	int served = 0;
2223 	int reported = 0;
2224 	u32 slt;
2225 
2226 	bit_vec = master->comm_arm_bit_vector;
2227 	for (i = 0; i < COMM_CHANNEL_BIT_ARRAY_SIZE; i++) {
2228 		vec = be32_to_cpu(bit_vec[i]);
2229 		for (j = 0; j < 32; j++) {
2230 			if (!(vec & (1 << j)))
2231 				continue;
2232 			++reported;
2233 			slave = (i * 32) + j;
2234 			comm_cmd = swab32(readl(
2235 					  &mfunc->comm[slave].slave_write));
2236 			slt = swab32(readl(&mfunc->comm[slave].slave_read))
2237 				     >> 31;
2238 			toggle = comm_cmd >> 31;
2239 			if (toggle != slt) {
2240 				if (master->slave_state[slave].comm_toggle
2241 				    != slt) {
2242 					pr_info("slave %d out of sync. read toggle %d, state toggle %d. Resynching.\n",
2243 						slave, slt,
2244 						master->slave_state[slave].comm_toggle);
2245 					master->slave_state[slave].comm_toggle =
2246 						slt;
2247 				}
2248 				mlx4_master_do_cmd(dev, slave,
2249 						   comm_cmd >> 16 & 0xff,
2250 						   comm_cmd & 0xffff, toggle);
2251 				++served;
2252 			}
2253 		}
2254 	}
2255 
2256 	if (reported && reported != served)
2257 		mlx4_warn(dev, "Got command event with bitmask from %d slaves but %d were served\n",
2258 			  reported, served);
2259 
2260 	if (mlx4_ARM_COMM_CHANNEL(dev))
2261 		mlx4_warn(dev, "Failed to arm comm channel events\n");
2262 }
2263 
2264 static int sync_toggles(struct mlx4_dev *dev)
2265 {
2266 	struct mlx4_priv *priv = mlx4_priv(dev);
2267 	u32 wr_toggle;
2268 	u32 rd_toggle;
2269 	unsigned long end;
2270 
2271 	wr_toggle = swab32(readl(&priv->mfunc.comm->slave_write));
2272 	if (wr_toggle == 0xffffffff)
2273 		end = jiffies + msecs_to_jiffies(30000);
2274 	else
2275 		end = jiffies + msecs_to_jiffies(5000);
2276 
2277 	while (time_before(jiffies, end)) {
2278 		rd_toggle = swab32(readl(&priv->mfunc.comm->slave_read));
2279 		if (wr_toggle == 0xffffffff || rd_toggle == 0xffffffff) {
2280 			/* PCI might be offline */
2281 			msleep(100);
2282 			wr_toggle = swab32(readl(&priv->mfunc.comm->
2283 					   slave_write));
2284 			continue;
2285 		}
2286 
2287 		if (rd_toggle >> 31 == wr_toggle >> 31) {
2288 			priv->cmd.comm_toggle = rd_toggle >> 31;
2289 			return 0;
2290 		}
2291 
2292 		cond_resched();
2293 	}
2294 
2295 	/*
2296 	 * we could reach here if for example the previous VM using this
2297 	 * function misbehaved and left the channel with unsynced state. We
2298 	 * should fix this here and give this VM a chance to use a properly
2299 	 * synced channel
2300 	 */
2301 	mlx4_warn(dev, "recovering from previously mis-behaved VM\n");
2302 	__raw_writel((__force u32) 0, &priv->mfunc.comm->slave_read);
2303 	__raw_writel((__force u32) 0, &priv->mfunc.comm->slave_write);
2304 	priv->cmd.comm_toggle = 0;
2305 
2306 	return 0;
2307 }
2308 
2309 int mlx4_multi_func_init(struct mlx4_dev *dev)
2310 {
2311 	struct mlx4_priv *priv = mlx4_priv(dev);
2312 	struct mlx4_slave_state *s_state;
2313 	int i, j, err, port;
2314 
2315 	if (mlx4_is_master(dev))
2316 		priv->mfunc.comm =
2317 		ioremap(pci_resource_start(dev->persist->pdev,
2318 					   priv->fw.comm_bar) +
2319 			priv->fw.comm_base, MLX4_COMM_PAGESIZE);
2320 	else
2321 		priv->mfunc.comm =
2322 		ioremap(pci_resource_start(dev->persist->pdev, 2) +
2323 			MLX4_SLAVE_COMM_BASE, MLX4_COMM_PAGESIZE);
2324 	if (!priv->mfunc.comm) {
2325 		mlx4_err(dev, "Couldn't map communication vector\n");
2326 		goto err_vhcr;
2327 	}
2328 
2329 	if (mlx4_is_master(dev)) {
2330 		struct mlx4_vf_oper_state *vf_oper;
2331 		struct mlx4_vf_admin_state *vf_admin;
2332 
2333 		priv->mfunc.master.slave_state =
2334 			kzalloc(dev->num_slaves *
2335 				sizeof(struct mlx4_slave_state), GFP_KERNEL);
2336 		if (!priv->mfunc.master.slave_state)
2337 			goto err_comm;
2338 
2339 		priv->mfunc.master.vf_admin =
2340 			kzalloc(dev->num_slaves *
2341 				sizeof(struct mlx4_vf_admin_state), GFP_KERNEL);
2342 		if (!priv->mfunc.master.vf_admin)
2343 			goto err_comm_admin;
2344 
2345 		priv->mfunc.master.vf_oper =
2346 			kzalloc(dev->num_slaves *
2347 				sizeof(struct mlx4_vf_oper_state), GFP_KERNEL);
2348 		if (!priv->mfunc.master.vf_oper)
2349 			goto err_comm_oper;
2350 
2351 		for (i = 0; i < dev->num_slaves; ++i) {
2352 			vf_admin = &priv->mfunc.master.vf_admin[i];
2353 			vf_oper = &priv->mfunc.master.vf_oper[i];
2354 			s_state = &priv->mfunc.master.slave_state[i];
2355 			s_state->last_cmd = MLX4_COMM_CMD_RESET;
2356 			mutex_init(&priv->mfunc.master.gen_eqe_mutex[i]);
2357 			for (j = 0; j < MLX4_EVENT_TYPES_NUM; ++j)
2358 				s_state->event_eq[j].eqn = -1;
2359 			__raw_writel((__force u32) 0,
2360 				     &priv->mfunc.comm[i].slave_write);
2361 			__raw_writel((__force u32) 0,
2362 				     &priv->mfunc.comm[i].slave_read);
2363 			mmiowb();
2364 			for (port = 1; port <= MLX4_MAX_PORTS; port++) {
2365 				struct mlx4_vport_state *admin_vport;
2366 				struct mlx4_vport_state *oper_vport;
2367 
2368 				s_state->vlan_filter[port] =
2369 					kzalloc(sizeof(struct mlx4_vlan_fltr),
2370 						GFP_KERNEL);
2371 				if (!s_state->vlan_filter[port]) {
2372 					if (--port)
2373 						kfree(s_state->vlan_filter[port]);
2374 					goto err_slaves;
2375 				}
2376 
2377 				admin_vport = &vf_admin->vport[port];
2378 				oper_vport = &vf_oper->vport[port].state;
2379 				INIT_LIST_HEAD(&s_state->mcast_filters[port]);
2380 				admin_vport->default_vlan = MLX4_VGT;
2381 				oper_vport->default_vlan = MLX4_VGT;
2382 				admin_vport->qos_vport =
2383 						MLX4_VPP_DEFAULT_VPORT;
2384 				oper_vport->qos_vport = MLX4_VPP_DEFAULT_VPORT;
2385 				vf_oper->vport[port].vlan_idx = NO_INDX;
2386 				vf_oper->vport[port].mac_idx = NO_INDX;
2387 				mlx4_set_random_admin_guid(dev, i, port);
2388 			}
2389 			spin_lock_init(&s_state->lock);
2390 		}
2391 
2392 		if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QOS_VPP) {
2393 			for (port = 1; port <= dev->caps.num_ports; port++) {
2394 				if (mlx4_is_eth(dev, port)) {
2395 					mlx4_set_default_port_qos(dev, port);
2396 					mlx4_allocate_port_vpps(dev, port);
2397 				}
2398 			}
2399 		}
2400 
2401 		memset(&priv->mfunc.master.cmd_eqe, 0, sizeof(struct mlx4_eqe));
2402 		priv->mfunc.master.cmd_eqe.type = MLX4_EVENT_TYPE_CMD;
2403 		INIT_WORK(&priv->mfunc.master.comm_work,
2404 			  mlx4_master_comm_channel);
2405 		INIT_WORK(&priv->mfunc.master.slave_event_work,
2406 			  mlx4_gen_slave_eqe);
2407 		INIT_WORK(&priv->mfunc.master.slave_flr_event_work,
2408 			  mlx4_master_handle_slave_flr);
2409 		spin_lock_init(&priv->mfunc.master.slave_state_lock);
2410 		spin_lock_init(&priv->mfunc.master.slave_eq.event_lock);
2411 		priv->mfunc.master.comm_wq =
2412 			create_singlethread_workqueue("mlx4_comm");
2413 		if (!priv->mfunc.master.comm_wq)
2414 			goto err_slaves;
2415 
2416 		if (mlx4_init_resource_tracker(dev))
2417 			goto err_thread;
2418 
2419 	} else {
2420 		err = sync_toggles(dev);
2421 		if (err) {
2422 			mlx4_err(dev, "Couldn't sync toggles\n");
2423 			goto err_comm;
2424 		}
2425 	}
2426 	return 0;
2427 
2428 err_thread:
2429 	flush_workqueue(priv->mfunc.master.comm_wq);
2430 	destroy_workqueue(priv->mfunc.master.comm_wq);
2431 err_slaves:
2432 	while (i--) {
2433 		for (port = 1; port <= MLX4_MAX_PORTS; port++)
2434 			kfree(priv->mfunc.master.slave_state[i].vlan_filter[port]);
2435 	}
2436 	kfree(priv->mfunc.master.vf_oper);
2437 err_comm_oper:
2438 	kfree(priv->mfunc.master.vf_admin);
2439 err_comm_admin:
2440 	kfree(priv->mfunc.master.slave_state);
2441 err_comm:
2442 	iounmap(priv->mfunc.comm);
2443 err_vhcr:
2444 	dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
2445 			  priv->mfunc.vhcr,
2446 			  priv->mfunc.vhcr_dma);
2447 	priv->mfunc.vhcr = NULL;
2448 	return -ENOMEM;
2449 }
2450 
2451 int mlx4_cmd_init(struct mlx4_dev *dev)
2452 {
2453 	struct mlx4_priv *priv = mlx4_priv(dev);
2454 	int flags = 0;
2455 
2456 	if (!priv->cmd.initialized) {
2457 		mutex_init(&priv->cmd.slave_cmd_mutex);
2458 		sema_init(&priv->cmd.poll_sem, 1);
2459 		priv->cmd.use_events = 0;
2460 		priv->cmd.toggle     = 1;
2461 		priv->cmd.initialized = 1;
2462 		flags |= MLX4_CMD_CLEANUP_STRUCT;
2463 	}
2464 
2465 	if (!mlx4_is_slave(dev) && !priv->cmd.hcr) {
2466 		priv->cmd.hcr = ioremap(pci_resource_start(dev->persist->pdev,
2467 					0) + MLX4_HCR_BASE, MLX4_HCR_SIZE);
2468 		if (!priv->cmd.hcr) {
2469 			mlx4_err(dev, "Couldn't map command register\n");
2470 			goto err;
2471 		}
2472 		flags |= MLX4_CMD_CLEANUP_HCR;
2473 	}
2474 
2475 	if (mlx4_is_mfunc(dev) && !priv->mfunc.vhcr) {
2476 		priv->mfunc.vhcr = dma_alloc_coherent(&dev->persist->pdev->dev,
2477 						      PAGE_SIZE,
2478 						      &priv->mfunc.vhcr_dma,
2479 						      GFP_KERNEL);
2480 		if (!priv->mfunc.vhcr)
2481 			goto err;
2482 
2483 		flags |= MLX4_CMD_CLEANUP_VHCR;
2484 	}
2485 
2486 	if (!priv->cmd.pool) {
2487 		priv->cmd.pool = pci_pool_create("mlx4_cmd",
2488 						 dev->persist->pdev,
2489 						 MLX4_MAILBOX_SIZE,
2490 						 MLX4_MAILBOX_SIZE, 0);
2491 		if (!priv->cmd.pool)
2492 			goto err;
2493 
2494 		flags |= MLX4_CMD_CLEANUP_POOL;
2495 	}
2496 
2497 	return 0;
2498 
2499 err:
2500 	mlx4_cmd_cleanup(dev, flags);
2501 	return -ENOMEM;
2502 }
2503 
2504 void mlx4_report_internal_err_comm_event(struct mlx4_dev *dev)
2505 {
2506 	struct mlx4_priv *priv = mlx4_priv(dev);
2507 	int slave;
2508 	u32 slave_read;
2509 
2510 	/* Report an internal error event to all
2511 	 * communication channels.
2512 	 */
2513 	for (slave = 0; slave < dev->num_slaves; slave++) {
2514 		slave_read = swab32(readl(&priv->mfunc.comm[slave].slave_read));
2515 		slave_read |= (u32)COMM_CHAN_EVENT_INTERNAL_ERR;
2516 		__raw_writel((__force u32)cpu_to_be32(slave_read),
2517 			     &priv->mfunc.comm[slave].slave_read);
2518 		/* Make sure that our comm channel write doesn't
2519 		 * get mixed in with writes from another CPU.
2520 		 */
2521 		mmiowb();
2522 	}
2523 }
2524 
2525 void mlx4_multi_func_cleanup(struct mlx4_dev *dev)
2526 {
2527 	struct mlx4_priv *priv = mlx4_priv(dev);
2528 	int i, port;
2529 
2530 	if (mlx4_is_master(dev)) {
2531 		flush_workqueue(priv->mfunc.master.comm_wq);
2532 		destroy_workqueue(priv->mfunc.master.comm_wq);
2533 		for (i = 0; i < dev->num_slaves; i++) {
2534 			for (port = 1; port <= MLX4_MAX_PORTS; port++)
2535 				kfree(priv->mfunc.master.slave_state[i].vlan_filter[port]);
2536 		}
2537 		kfree(priv->mfunc.master.slave_state);
2538 		kfree(priv->mfunc.master.vf_admin);
2539 		kfree(priv->mfunc.master.vf_oper);
2540 		dev->num_slaves = 0;
2541 	}
2542 
2543 	iounmap(priv->mfunc.comm);
2544 }
2545 
2546 void mlx4_cmd_cleanup(struct mlx4_dev *dev, int cleanup_mask)
2547 {
2548 	struct mlx4_priv *priv = mlx4_priv(dev);
2549 
2550 	if (priv->cmd.pool && (cleanup_mask & MLX4_CMD_CLEANUP_POOL)) {
2551 		pci_pool_destroy(priv->cmd.pool);
2552 		priv->cmd.pool = NULL;
2553 	}
2554 
2555 	if (!mlx4_is_slave(dev) && priv->cmd.hcr &&
2556 	    (cleanup_mask & MLX4_CMD_CLEANUP_HCR)) {
2557 		iounmap(priv->cmd.hcr);
2558 		priv->cmd.hcr = NULL;
2559 	}
2560 	if (mlx4_is_mfunc(dev) && priv->mfunc.vhcr &&
2561 	    (cleanup_mask & MLX4_CMD_CLEANUP_VHCR)) {
2562 		dma_free_coherent(&dev->persist->pdev->dev, PAGE_SIZE,
2563 				  priv->mfunc.vhcr, priv->mfunc.vhcr_dma);
2564 		priv->mfunc.vhcr = NULL;
2565 	}
2566 	if (priv->cmd.initialized && (cleanup_mask & MLX4_CMD_CLEANUP_STRUCT))
2567 		priv->cmd.initialized = 0;
2568 }
2569 
2570 /*
2571  * Switch to using events to issue FW commands (can only be called
2572  * after event queue for command events has been initialized).
2573  */
2574 int mlx4_cmd_use_events(struct mlx4_dev *dev)
2575 {
2576 	struct mlx4_priv *priv = mlx4_priv(dev);
2577 	int i;
2578 	int err = 0;
2579 
2580 	priv->cmd.context = kmalloc(priv->cmd.max_cmds *
2581 				   sizeof (struct mlx4_cmd_context),
2582 				   GFP_KERNEL);
2583 	if (!priv->cmd.context)
2584 		return -ENOMEM;
2585 
2586 	for (i = 0; i < priv->cmd.max_cmds; ++i) {
2587 		priv->cmd.context[i].token = i;
2588 		priv->cmd.context[i].next  = i + 1;
2589 		/* To support fatal error flow, initialize all
2590 		 * cmd contexts to allow simulating completions
2591 		 * with complete() at any time.
2592 		 */
2593 		init_completion(&priv->cmd.context[i].done);
2594 	}
2595 
2596 	priv->cmd.context[priv->cmd.max_cmds - 1].next = -1;
2597 	priv->cmd.free_head = 0;
2598 
2599 	sema_init(&priv->cmd.event_sem, priv->cmd.max_cmds);
2600 
2601 	for (priv->cmd.token_mask = 1;
2602 	     priv->cmd.token_mask < priv->cmd.max_cmds;
2603 	     priv->cmd.token_mask <<= 1)
2604 		; /* nothing */
2605 	--priv->cmd.token_mask;
2606 
2607 	down(&priv->cmd.poll_sem);
2608 	priv->cmd.use_events = 1;
2609 
2610 	return err;
2611 }
2612 
2613 /*
2614  * Switch back to polling (used when shutting down the device)
2615  */
2616 void mlx4_cmd_use_polling(struct mlx4_dev *dev)
2617 {
2618 	struct mlx4_priv *priv = mlx4_priv(dev);
2619 	int i;
2620 
2621 	priv->cmd.use_events = 0;
2622 
2623 	for (i = 0; i < priv->cmd.max_cmds; ++i)
2624 		down(&priv->cmd.event_sem);
2625 
2626 	kfree(priv->cmd.context);
2627 
2628 	up(&priv->cmd.poll_sem);
2629 }
2630 
2631 struct mlx4_cmd_mailbox *mlx4_alloc_cmd_mailbox(struct mlx4_dev *dev)
2632 {
2633 	struct mlx4_cmd_mailbox *mailbox;
2634 
2635 	mailbox = kmalloc(sizeof *mailbox, GFP_KERNEL);
2636 	if (!mailbox)
2637 		return ERR_PTR(-ENOMEM);
2638 
2639 	mailbox->buf = pci_pool_alloc(mlx4_priv(dev)->cmd.pool, GFP_KERNEL,
2640 				      &mailbox->dma);
2641 	if (!mailbox->buf) {
2642 		kfree(mailbox);
2643 		return ERR_PTR(-ENOMEM);
2644 	}
2645 
2646 	memset(mailbox->buf, 0, MLX4_MAILBOX_SIZE);
2647 
2648 	return mailbox;
2649 }
2650 EXPORT_SYMBOL_GPL(mlx4_alloc_cmd_mailbox);
2651 
2652 void mlx4_free_cmd_mailbox(struct mlx4_dev *dev,
2653 			   struct mlx4_cmd_mailbox *mailbox)
2654 {
2655 	if (!mailbox)
2656 		return;
2657 
2658 	pci_pool_free(mlx4_priv(dev)->cmd.pool, mailbox->buf, mailbox->dma);
2659 	kfree(mailbox);
2660 }
2661 EXPORT_SYMBOL_GPL(mlx4_free_cmd_mailbox);
2662 
2663 u32 mlx4_comm_get_version(void)
2664 {
2665 	 return ((u32) CMD_CHAN_IF_REV << 8) | (u32) CMD_CHAN_VER;
2666 }
2667 
2668 static int mlx4_get_slave_indx(struct mlx4_dev *dev, int vf)
2669 {
2670 	if ((vf < 0) || (vf >= dev->persist->num_vfs)) {
2671 		mlx4_err(dev, "Bad vf number:%d (number of activated vf: %d)\n",
2672 			 vf, dev->persist->num_vfs);
2673 		return -EINVAL;
2674 	}
2675 
2676 	return vf+1;
2677 }
2678 
2679 int mlx4_get_vf_indx(struct mlx4_dev *dev, int slave)
2680 {
2681 	if (slave < 1 || slave > dev->persist->num_vfs) {
2682 		mlx4_err(dev,
2683 			 "Bad slave number:%d (number of activated slaves: %lu)\n",
2684 			 slave, dev->num_slaves);
2685 		return -EINVAL;
2686 	}
2687 	return slave - 1;
2688 }
2689 
2690 void mlx4_cmd_wake_completions(struct mlx4_dev *dev)
2691 {
2692 	struct mlx4_priv *priv = mlx4_priv(dev);
2693 	struct mlx4_cmd_context *context;
2694 	int i;
2695 
2696 	spin_lock(&priv->cmd.context_lock);
2697 	if (priv->cmd.context) {
2698 		for (i = 0; i < priv->cmd.max_cmds; ++i) {
2699 			context = &priv->cmd.context[i];
2700 			context->fw_status = CMD_STAT_INTERNAL_ERR;
2701 			context->result    =
2702 				mlx4_status_to_errno(CMD_STAT_INTERNAL_ERR);
2703 			complete(&context->done);
2704 		}
2705 	}
2706 	spin_unlock(&priv->cmd.context_lock);
2707 }
2708 
2709 struct mlx4_active_ports mlx4_get_active_ports(struct mlx4_dev *dev, int slave)
2710 {
2711 	struct mlx4_active_ports actv_ports;
2712 	int vf;
2713 
2714 	bitmap_zero(actv_ports.ports, MLX4_MAX_PORTS);
2715 
2716 	if (slave == 0) {
2717 		bitmap_fill(actv_ports.ports, dev->caps.num_ports);
2718 		return actv_ports;
2719 	}
2720 
2721 	vf = mlx4_get_vf_indx(dev, slave);
2722 	if (vf < 0)
2723 		return actv_ports;
2724 
2725 	bitmap_set(actv_ports.ports, dev->dev_vfs[vf].min_port - 1,
2726 		   min((int)dev->dev_vfs[mlx4_get_vf_indx(dev, slave)].n_ports,
2727 		   dev->caps.num_ports));
2728 
2729 	return actv_ports;
2730 }
2731 EXPORT_SYMBOL_GPL(mlx4_get_active_ports);
2732 
2733 int mlx4_slave_convert_port(struct mlx4_dev *dev, int slave, int port)
2734 {
2735 	unsigned n;
2736 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
2737 	unsigned m = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2738 
2739 	if (port <= 0 || port > m)
2740 		return -EINVAL;
2741 
2742 	n = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2743 	if (port <= n)
2744 		port = n + 1;
2745 
2746 	return port;
2747 }
2748 EXPORT_SYMBOL_GPL(mlx4_slave_convert_port);
2749 
2750 int mlx4_phys_to_slave_port(struct mlx4_dev *dev, int slave, int port)
2751 {
2752 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
2753 	if (test_bit(port - 1, actv_ports.ports))
2754 		return port -
2755 			find_first_bit(actv_ports.ports, dev->caps.num_ports);
2756 
2757 	return -1;
2758 }
2759 EXPORT_SYMBOL_GPL(mlx4_phys_to_slave_port);
2760 
2761 struct mlx4_slaves_pport mlx4_phys_to_slaves_pport(struct mlx4_dev *dev,
2762 						   int port)
2763 {
2764 	unsigned i;
2765 	struct mlx4_slaves_pport slaves_pport;
2766 
2767 	bitmap_zero(slaves_pport.slaves, MLX4_MFUNC_MAX);
2768 
2769 	if (port <= 0 || port > dev->caps.num_ports)
2770 		return slaves_pport;
2771 
2772 	for (i = 0; i < dev->persist->num_vfs + 1; i++) {
2773 		struct mlx4_active_ports actv_ports =
2774 			mlx4_get_active_ports(dev, i);
2775 		if (test_bit(port - 1, actv_ports.ports))
2776 			set_bit(i, slaves_pport.slaves);
2777 	}
2778 
2779 	return slaves_pport;
2780 }
2781 EXPORT_SYMBOL_GPL(mlx4_phys_to_slaves_pport);
2782 
2783 struct mlx4_slaves_pport mlx4_phys_to_slaves_pport_actv(
2784 		struct mlx4_dev *dev,
2785 		const struct mlx4_active_ports *crit_ports)
2786 {
2787 	unsigned i;
2788 	struct mlx4_slaves_pport slaves_pport;
2789 
2790 	bitmap_zero(slaves_pport.slaves, MLX4_MFUNC_MAX);
2791 
2792 	for (i = 0; i < dev->persist->num_vfs + 1; i++) {
2793 		struct mlx4_active_ports actv_ports =
2794 			mlx4_get_active_ports(dev, i);
2795 		if (bitmap_equal(crit_ports->ports, actv_ports.ports,
2796 				 dev->caps.num_ports))
2797 			set_bit(i, slaves_pport.slaves);
2798 	}
2799 
2800 	return slaves_pport;
2801 }
2802 EXPORT_SYMBOL_GPL(mlx4_phys_to_slaves_pport_actv);
2803 
2804 static int mlx4_slaves_closest_port(struct mlx4_dev *dev, int slave, int port)
2805 {
2806 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(dev, slave);
2807 	int min_port = find_first_bit(actv_ports.ports, dev->caps.num_ports)
2808 			+ 1;
2809 	int max_port = min_port +
2810 		bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2811 
2812 	if (port < min_port)
2813 		port = min_port;
2814 	else if (port >= max_port)
2815 		port = max_port - 1;
2816 
2817 	return port;
2818 }
2819 
2820 static int mlx4_set_vport_qos(struct mlx4_priv *priv, int slave, int port,
2821 			      int max_tx_rate)
2822 {
2823 	int i;
2824 	int err;
2825 	struct mlx4_qos_manager *port_qos;
2826 	struct mlx4_dev *dev = &priv->dev;
2827 	struct mlx4_vport_qos_param vpp_qos[MLX4_NUM_UP];
2828 
2829 	port_qos = &priv->mfunc.master.qos_ctl[port];
2830 	memset(vpp_qos, 0, sizeof(struct mlx4_vport_qos_param) * MLX4_NUM_UP);
2831 
2832 	if (slave > port_qos->num_of_qos_vfs) {
2833 		mlx4_info(dev, "No availible VPP resources for this VF\n");
2834 		return -EINVAL;
2835 	}
2836 
2837 	/* Query for default QoS values from Vport 0 is needed */
2838 	err = mlx4_SET_VPORT_QOS_get(dev, port, 0, vpp_qos);
2839 	if (err) {
2840 		mlx4_info(dev, "Failed to query Vport 0 QoS values\n");
2841 		return err;
2842 	}
2843 
2844 	for (i = 0; i < MLX4_NUM_UP; i++) {
2845 		if (test_bit(i, port_qos->priority_bm) && max_tx_rate) {
2846 			vpp_qos[i].max_avg_bw = max_tx_rate;
2847 			vpp_qos[i].enable = 1;
2848 		} else {
2849 			/* if user supplied tx_rate == 0, meaning no rate limit
2850 			 * configuration is required. so we are leaving the
2851 			 * value of max_avg_bw as queried from Vport 0.
2852 			 */
2853 			vpp_qos[i].enable = 0;
2854 		}
2855 	}
2856 
2857 	err = mlx4_SET_VPORT_QOS_set(dev, port, slave, vpp_qos);
2858 	if (err) {
2859 		mlx4_info(dev, "Failed to set Vport %d QoS values\n", slave);
2860 		return err;
2861 	}
2862 
2863 	return 0;
2864 }
2865 
2866 static bool mlx4_is_vf_vst_and_prio_qos(struct mlx4_dev *dev, int port,
2867 					struct mlx4_vport_state *vf_admin)
2868 {
2869 	struct mlx4_qos_manager *info;
2870 	struct mlx4_priv *priv = mlx4_priv(dev);
2871 
2872 	if (!mlx4_is_master(dev) ||
2873 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QOS_VPP))
2874 		return false;
2875 
2876 	info = &priv->mfunc.master.qos_ctl[port];
2877 
2878 	if (vf_admin->default_vlan != MLX4_VGT &&
2879 	    test_bit(vf_admin->default_qos, info->priority_bm))
2880 		return true;
2881 
2882 	return false;
2883 }
2884 
2885 static bool mlx4_valid_vf_state_change(struct mlx4_dev *dev, int port,
2886 				       struct mlx4_vport_state *vf_admin,
2887 				       int vlan, int qos)
2888 {
2889 	struct mlx4_vport_state dummy_admin = {0};
2890 
2891 	if (!mlx4_is_vf_vst_and_prio_qos(dev, port, vf_admin) ||
2892 	    !vf_admin->tx_rate)
2893 		return true;
2894 
2895 	dummy_admin.default_qos = qos;
2896 	dummy_admin.default_vlan = vlan;
2897 
2898 	/* VF wants to move to other VST state which is valid with current
2899 	 * rate limit. Either differnt default vlan in VST or other
2900 	 * supported QoS priority. Otherwise we don't allow this change when
2901 	 * the TX rate is still configured.
2902 	 */
2903 	if (mlx4_is_vf_vst_and_prio_qos(dev, port, &dummy_admin))
2904 		return true;
2905 
2906 	mlx4_info(dev, "Cannot change VF state to %s while rate is set\n",
2907 		  (vlan == MLX4_VGT) ? "VGT" : "VST");
2908 
2909 	if (vlan != MLX4_VGT)
2910 		mlx4_info(dev, "VST priority %d not supported for QoS\n", qos);
2911 
2912 	mlx4_info(dev, "Please set rate to 0 prior to this VF state change\n");
2913 
2914 	return false;
2915 }
2916 
2917 int mlx4_set_vf_mac(struct mlx4_dev *dev, int port, int vf, u64 mac)
2918 {
2919 	struct mlx4_priv *priv = mlx4_priv(dev);
2920 	struct mlx4_vport_state *s_info;
2921 	int slave;
2922 
2923 	if (!mlx4_is_master(dev))
2924 		return -EPROTONOSUPPORT;
2925 
2926 	slave = mlx4_get_slave_indx(dev, vf);
2927 	if (slave < 0)
2928 		return -EINVAL;
2929 
2930 	port = mlx4_slaves_closest_port(dev, slave, port);
2931 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
2932 	s_info->mac = mac;
2933 	mlx4_info(dev, "default mac on vf %d port %d to %llX will take effect only after vf restart\n",
2934 		  vf, port, s_info->mac);
2935 	return 0;
2936 }
2937 EXPORT_SYMBOL_GPL(mlx4_set_vf_mac);
2938 
2939 
2940 int mlx4_set_vf_vlan(struct mlx4_dev *dev, int port, int vf, u16 vlan, u8 qos)
2941 {
2942 	struct mlx4_priv *priv = mlx4_priv(dev);
2943 	struct mlx4_vport_state *vf_admin;
2944 	int slave;
2945 
2946 	if ((!mlx4_is_master(dev)) ||
2947 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_VLAN_CONTROL))
2948 		return -EPROTONOSUPPORT;
2949 
2950 	if ((vlan > 4095) || (qos > 7))
2951 		return -EINVAL;
2952 
2953 	slave = mlx4_get_slave_indx(dev, vf);
2954 	if (slave < 0)
2955 		return -EINVAL;
2956 
2957 	port = mlx4_slaves_closest_port(dev, slave, port);
2958 	vf_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
2959 
2960 	if (!mlx4_valid_vf_state_change(dev, port, vf_admin, vlan, qos))
2961 		return -EPERM;
2962 
2963 	if ((0 == vlan) && (0 == qos))
2964 		vf_admin->default_vlan = MLX4_VGT;
2965 	else
2966 		vf_admin->default_vlan = vlan;
2967 	vf_admin->default_qos = qos;
2968 
2969 	/* If rate was configured prior to VST, we saved the configured rate
2970 	 * in vf_admin->rate and now, if priority supported we enforce the QoS
2971 	 */
2972 	if (mlx4_is_vf_vst_and_prio_qos(dev, port, vf_admin) &&
2973 	    vf_admin->tx_rate)
2974 		vf_admin->qos_vport = slave;
2975 
2976 	if (mlx4_master_immediate_activate_vlan_qos(priv, slave, port))
2977 		mlx4_info(dev,
2978 			  "updating vf %d port %d config will take effect on next VF restart\n",
2979 			  vf, port);
2980 	return 0;
2981 }
2982 EXPORT_SYMBOL_GPL(mlx4_set_vf_vlan);
2983 
2984 int mlx4_set_vf_rate(struct mlx4_dev *dev, int port, int vf, int min_tx_rate,
2985 		     int max_tx_rate)
2986 {
2987 	int err;
2988 	int slave;
2989 	struct mlx4_vport_state *vf_admin;
2990 	struct mlx4_priv *priv = mlx4_priv(dev);
2991 
2992 	if (!mlx4_is_master(dev) ||
2993 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_QOS_VPP))
2994 		return -EPROTONOSUPPORT;
2995 
2996 	if (min_tx_rate) {
2997 		mlx4_info(dev, "Minimum BW share not supported\n");
2998 		return -EPROTONOSUPPORT;
2999 	}
3000 
3001 	slave = mlx4_get_slave_indx(dev, vf);
3002 	if (slave < 0)
3003 		return -EINVAL;
3004 
3005 	port = mlx4_slaves_closest_port(dev, slave, port);
3006 	vf_admin = &priv->mfunc.master.vf_admin[slave].vport[port];
3007 
3008 	err = mlx4_set_vport_qos(priv, slave, port, max_tx_rate);
3009 	if (err) {
3010 		mlx4_info(dev, "vf %d failed to set rate %d\n", vf,
3011 			  max_tx_rate);
3012 		return err;
3013 	}
3014 
3015 	vf_admin->tx_rate = max_tx_rate;
3016 	/* if VF is not in supported mode (VST with supported prio),
3017 	 * we do not change vport configuration for its QPs, but save
3018 	 * the rate, so it will be enforced when it moves to supported
3019 	 * mode next time.
3020 	 */
3021 	if (!mlx4_is_vf_vst_and_prio_qos(dev, port, vf_admin)) {
3022 		mlx4_info(dev,
3023 			  "rate set for VF %d when not in valid state\n", vf);
3024 
3025 		if (vf_admin->default_vlan != MLX4_VGT)
3026 			mlx4_info(dev, "VST priority not supported by QoS\n");
3027 		else
3028 			mlx4_info(dev, "VF in VGT mode (needed VST)\n");
3029 
3030 		mlx4_info(dev,
3031 			  "rate %d take affect when VF moves to valid state\n",
3032 			  max_tx_rate);
3033 		return 0;
3034 	}
3035 
3036 	/* If user sets rate 0 assigning default vport for its QPs */
3037 	vf_admin->qos_vport = max_tx_rate ? slave : MLX4_VPP_DEFAULT_VPORT;
3038 
3039 	if (priv->mfunc.master.slave_state[slave].active &&
3040 	    dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_UPDATE_QP)
3041 		mlx4_master_immediate_activate_vlan_qos(priv, slave, port);
3042 
3043 	return 0;
3044 }
3045 EXPORT_SYMBOL_GPL(mlx4_set_vf_rate);
3046 
3047  /* mlx4_get_slave_default_vlan -
3048  * return true if VST ( default vlan)
3049  * if VST, will return vlan & qos (if not NULL)
3050  */
3051 bool mlx4_get_slave_default_vlan(struct mlx4_dev *dev, int port, int slave,
3052 				 u16 *vlan, u8 *qos)
3053 {
3054 	struct mlx4_vport_oper_state *vp_oper;
3055 	struct mlx4_priv *priv;
3056 
3057 	priv = mlx4_priv(dev);
3058 	port = mlx4_slaves_closest_port(dev, slave, port);
3059 	vp_oper = &priv->mfunc.master.vf_oper[slave].vport[port];
3060 
3061 	if (MLX4_VGT != vp_oper->state.default_vlan) {
3062 		if (vlan)
3063 			*vlan = vp_oper->state.default_vlan;
3064 		if (qos)
3065 			*qos = vp_oper->state.default_qos;
3066 		return true;
3067 	}
3068 	return false;
3069 }
3070 EXPORT_SYMBOL_GPL(mlx4_get_slave_default_vlan);
3071 
3072 int mlx4_set_vf_spoofchk(struct mlx4_dev *dev, int port, int vf, bool setting)
3073 {
3074 	struct mlx4_priv *priv = mlx4_priv(dev);
3075 	struct mlx4_vport_state *s_info;
3076 	int slave;
3077 
3078 	if ((!mlx4_is_master(dev)) ||
3079 	    !(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FSM))
3080 		return -EPROTONOSUPPORT;
3081 
3082 	slave = mlx4_get_slave_indx(dev, vf);
3083 	if (slave < 0)
3084 		return -EINVAL;
3085 
3086 	port = mlx4_slaves_closest_port(dev, slave, port);
3087 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
3088 	s_info->spoofchk = setting;
3089 
3090 	return 0;
3091 }
3092 EXPORT_SYMBOL_GPL(mlx4_set_vf_spoofchk);
3093 
3094 int mlx4_get_vf_config(struct mlx4_dev *dev, int port, int vf, struct ifla_vf_info *ivf)
3095 {
3096 	struct mlx4_priv *priv = mlx4_priv(dev);
3097 	struct mlx4_vport_state *s_info;
3098 	int slave;
3099 
3100 	if (!mlx4_is_master(dev))
3101 		return -EPROTONOSUPPORT;
3102 
3103 	slave = mlx4_get_slave_indx(dev, vf);
3104 	if (slave < 0)
3105 		return -EINVAL;
3106 
3107 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
3108 	ivf->vf = vf;
3109 
3110 	/* need to convert it to a func */
3111 	ivf->mac[0] = ((s_info->mac >> (5*8)) & 0xff);
3112 	ivf->mac[1] = ((s_info->mac >> (4*8)) & 0xff);
3113 	ivf->mac[2] = ((s_info->mac >> (3*8)) & 0xff);
3114 	ivf->mac[3] = ((s_info->mac >> (2*8)) & 0xff);
3115 	ivf->mac[4] = ((s_info->mac >> (1*8)) & 0xff);
3116 	ivf->mac[5] = ((s_info->mac)  & 0xff);
3117 
3118 	ivf->vlan		= s_info->default_vlan;
3119 	ivf->qos		= s_info->default_qos;
3120 
3121 	if (mlx4_is_vf_vst_and_prio_qos(dev, port, s_info))
3122 		ivf->max_tx_rate = s_info->tx_rate;
3123 	else
3124 		ivf->max_tx_rate = 0;
3125 
3126 	ivf->min_tx_rate	= 0;
3127 	ivf->spoofchk		= s_info->spoofchk;
3128 	ivf->linkstate		= s_info->link_state;
3129 
3130 	return 0;
3131 }
3132 EXPORT_SYMBOL_GPL(mlx4_get_vf_config);
3133 
3134 int mlx4_set_vf_link_state(struct mlx4_dev *dev, int port, int vf, int link_state)
3135 {
3136 	struct mlx4_priv *priv = mlx4_priv(dev);
3137 	struct mlx4_vport_state *s_info;
3138 	int slave;
3139 	u8 link_stat_event;
3140 
3141 	slave = mlx4_get_slave_indx(dev, vf);
3142 	if (slave < 0)
3143 		return -EINVAL;
3144 
3145 	port = mlx4_slaves_closest_port(dev, slave, port);
3146 	switch (link_state) {
3147 	case IFLA_VF_LINK_STATE_AUTO:
3148 		/* get current link state */
3149 		if (!priv->sense.do_sense_port[port])
3150 			link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_ACTIVE;
3151 		else
3152 			link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_DOWN;
3153 	    break;
3154 
3155 	case IFLA_VF_LINK_STATE_ENABLE:
3156 		link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_ACTIVE;
3157 	    break;
3158 
3159 	case IFLA_VF_LINK_STATE_DISABLE:
3160 		link_stat_event = MLX4_PORT_CHANGE_SUBTYPE_DOWN;
3161 	    break;
3162 
3163 	default:
3164 		mlx4_warn(dev, "unknown value for link_state %02x on slave %d port %d\n",
3165 			  link_state, slave, port);
3166 		return -EINVAL;
3167 	};
3168 	s_info = &priv->mfunc.master.vf_admin[slave].vport[port];
3169 	s_info->link_state = link_state;
3170 
3171 	/* send event */
3172 	mlx4_gen_port_state_change_eqe(dev, slave, port, link_stat_event);
3173 
3174 	if (mlx4_master_immediate_activate_vlan_qos(priv, slave, port))
3175 		mlx4_dbg(dev,
3176 			 "updating vf %d port %d no link state HW enforcment\n",
3177 			 vf, port);
3178 	return 0;
3179 }
3180 EXPORT_SYMBOL_GPL(mlx4_set_vf_link_state);
3181 
3182 int mlx4_get_counter_stats(struct mlx4_dev *dev, int counter_index,
3183 			   struct mlx4_counter *counter_stats, int reset)
3184 {
3185 	struct mlx4_cmd_mailbox *mailbox = NULL;
3186 	struct mlx4_counter *tmp_counter;
3187 	int err;
3188 	u32 if_stat_in_mod;
3189 
3190 	if (!counter_stats)
3191 		return -EINVAL;
3192 
3193 	if (counter_index == MLX4_SINK_COUNTER_INDEX(dev))
3194 		return 0;
3195 
3196 	mailbox = mlx4_alloc_cmd_mailbox(dev);
3197 	if (IS_ERR(mailbox))
3198 		return PTR_ERR(mailbox);
3199 
3200 	memset(mailbox->buf, 0, sizeof(struct mlx4_counter));
3201 	if_stat_in_mod = counter_index;
3202 	if (reset)
3203 		if_stat_in_mod |= MLX4_QUERY_IF_STAT_RESET;
3204 	err = mlx4_cmd_box(dev, 0, mailbox->dma,
3205 			   if_stat_in_mod, 0,
3206 			   MLX4_CMD_QUERY_IF_STAT,
3207 			   MLX4_CMD_TIME_CLASS_C,
3208 			   MLX4_CMD_NATIVE);
3209 	if (err) {
3210 		mlx4_dbg(dev, "%s: failed to read statistics for counter index %d\n",
3211 			 __func__, counter_index);
3212 		goto if_stat_out;
3213 	}
3214 	tmp_counter = (struct mlx4_counter *)mailbox->buf;
3215 	counter_stats->counter_mode = tmp_counter->counter_mode;
3216 	if (counter_stats->counter_mode == 0) {
3217 		counter_stats->rx_frames =
3218 			cpu_to_be64(be64_to_cpu(counter_stats->rx_frames) +
3219 				    be64_to_cpu(tmp_counter->rx_frames));
3220 		counter_stats->tx_frames =
3221 			cpu_to_be64(be64_to_cpu(counter_stats->tx_frames) +
3222 				    be64_to_cpu(tmp_counter->tx_frames));
3223 		counter_stats->rx_bytes =
3224 			cpu_to_be64(be64_to_cpu(counter_stats->rx_bytes) +
3225 				    be64_to_cpu(tmp_counter->rx_bytes));
3226 		counter_stats->tx_bytes =
3227 			cpu_to_be64(be64_to_cpu(counter_stats->tx_bytes) +
3228 				    be64_to_cpu(tmp_counter->tx_bytes));
3229 	}
3230 
3231 if_stat_out:
3232 	mlx4_free_cmd_mailbox(dev, mailbox);
3233 
3234 	return err;
3235 }
3236 EXPORT_SYMBOL_GPL(mlx4_get_counter_stats);
3237 
3238 int mlx4_get_vf_stats(struct mlx4_dev *dev, int port, int vf_idx,
3239 		      struct ifla_vf_stats *vf_stats)
3240 {
3241 	struct mlx4_counter tmp_vf_stats;
3242 	int slave;
3243 	int err = 0;
3244 
3245 	if (!vf_stats)
3246 		return -EINVAL;
3247 
3248 	if (!mlx4_is_master(dev))
3249 		return -EPROTONOSUPPORT;
3250 
3251 	slave = mlx4_get_slave_indx(dev, vf_idx);
3252 	if (slave < 0)
3253 		return -EINVAL;
3254 
3255 	port = mlx4_slaves_closest_port(dev, slave, port);
3256 	err = mlx4_calc_vf_counters(dev, slave, port, &tmp_vf_stats);
3257 	if (!err && tmp_vf_stats.counter_mode == 0) {
3258 		vf_stats->rx_packets = be64_to_cpu(tmp_vf_stats.rx_frames);
3259 		vf_stats->tx_packets = be64_to_cpu(tmp_vf_stats.tx_frames);
3260 		vf_stats->rx_bytes = be64_to_cpu(tmp_vf_stats.rx_bytes);
3261 		vf_stats->tx_bytes = be64_to_cpu(tmp_vf_stats.tx_bytes);
3262 	}
3263 
3264 	return err;
3265 }
3266 EXPORT_SYMBOL_GPL(mlx4_get_vf_stats);
3267 
3268 int mlx4_vf_smi_enabled(struct mlx4_dev *dev, int slave, int port)
3269 {
3270 	struct mlx4_priv *priv = mlx4_priv(dev);
3271 
3272 	if (slave < 1 || slave >= dev->num_slaves ||
3273 	    port < 1 || port > MLX4_MAX_PORTS)
3274 		return 0;
3275 
3276 	return priv->mfunc.master.vf_oper[slave].smi_enabled[port] ==
3277 		MLX4_VF_SMI_ENABLED;
3278 }
3279 EXPORT_SYMBOL_GPL(mlx4_vf_smi_enabled);
3280 
3281 int mlx4_vf_get_enable_smi_admin(struct mlx4_dev *dev, int slave, int port)
3282 {
3283 	struct mlx4_priv *priv = mlx4_priv(dev);
3284 
3285 	if (slave == mlx4_master_func_num(dev))
3286 		return 1;
3287 
3288 	if (slave < 1 || slave >= dev->num_slaves ||
3289 	    port < 1 || port > MLX4_MAX_PORTS)
3290 		return 0;
3291 
3292 	return priv->mfunc.master.vf_admin[slave].enable_smi[port] ==
3293 		MLX4_VF_SMI_ENABLED;
3294 }
3295 EXPORT_SYMBOL_GPL(mlx4_vf_get_enable_smi_admin);
3296 
3297 int mlx4_vf_set_enable_smi_admin(struct mlx4_dev *dev, int slave, int port,
3298 				 int enabled)
3299 {
3300 	struct mlx4_priv *priv = mlx4_priv(dev);
3301 	struct mlx4_active_ports actv_ports = mlx4_get_active_ports(
3302 			&priv->dev, slave);
3303 	int min_port = find_first_bit(actv_ports.ports,
3304 				      priv->dev.caps.num_ports) + 1;
3305 	int max_port = min_port - 1 +
3306 		bitmap_weight(actv_ports.ports, priv->dev.caps.num_ports);
3307 
3308 	if (slave == mlx4_master_func_num(dev))
3309 		return 0;
3310 
3311 	if (slave < 1 || slave >= dev->num_slaves ||
3312 	    port < 1 || port > MLX4_MAX_PORTS ||
3313 	    enabled < 0 || enabled > 1)
3314 		return -EINVAL;
3315 
3316 	if (min_port == max_port && dev->caps.num_ports > 1) {
3317 		mlx4_info(dev, "SMI access disallowed for single ported VFs\n");
3318 		return -EPROTONOSUPPORT;
3319 	}
3320 
3321 	priv->mfunc.master.vf_admin[slave].enable_smi[port] = enabled;
3322 	return 0;
3323 }
3324 EXPORT_SYMBOL_GPL(mlx4_vf_set_enable_smi_admin);
3325