1 // SPDX-License-Identifier: (GPL-2.0 OR MIT)
2 /* Google virtual Ethernet (gve) driver
3  *
4  * Copyright (C) 2015-2021 Google, Inc.
5  */
6 
7 #include <linux/etherdevice.h>
8 #include <linux/pci.h>
9 #include "gve.h"
10 #include "gve_adminq.h"
11 #include "gve_register.h"
12 
13 #define GVE_MAX_ADMINQ_RELEASE_CHECK	500
14 #define GVE_ADMINQ_SLEEP_LEN		20
15 #define GVE_MAX_ADMINQ_EVENT_COUNTER_CHECK	100
16 
17 #define GVE_DEVICE_OPTION_ERROR_FMT "%s option error:\n" \
18 "Expected: length=%d, feature_mask=%x.\n" \
19 "Actual: length=%d, feature_mask=%x.\n"
20 
21 #define GVE_DEVICE_OPTION_TOO_BIG_FMT "Length of %s option larger than expected. Possible older version of guest driver.\n"
22 
23 static
24 struct gve_device_option *gve_get_next_option(struct gve_device_descriptor *descriptor,
25 					      struct gve_device_option *option)
26 {
27 	void *option_end, *descriptor_end;
28 
29 	option_end = (void *)(option + 1) + be16_to_cpu(option->option_length);
30 	descriptor_end = (void *)descriptor + be16_to_cpu(descriptor->total_length);
31 
32 	return option_end > descriptor_end ? NULL : (struct gve_device_option *)option_end;
33 }
34 
35 static
36 void gve_parse_device_option(struct gve_priv *priv,
37 			     struct gve_device_descriptor *device_descriptor,
38 			     struct gve_device_option *option,
39 			     struct gve_device_option_gqi_rda **dev_op_gqi_rda,
40 			     struct gve_device_option_gqi_qpl **dev_op_gqi_qpl,
41 			     struct gve_device_option_dqo_rda **dev_op_dqo_rda)
42 {
43 	u32 req_feat_mask = be32_to_cpu(option->required_features_mask);
44 	u16 option_length = be16_to_cpu(option->option_length);
45 	u16 option_id = be16_to_cpu(option->option_id);
46 
47 	/* If the length or feature mask doesn't match, continue without
48 	 * enabling the feature.
49 	 */
50 	switch (option_id) {
51 	case GVE_DEV_OPT_ID_GQI_RAW_ADDRESSING:
52 		if (option_length != GVE_DEV_OPT_LEN_GQI_RAW_ADDRESSING ||
53 		    req_feat_mask != GVE_DEV_OPT_REQ_FEAT_MASK_GQI_RAW_ADDRESSING) {
54 			dev_warn(&priv->pdev->dev, GVE_DEVICE_OPTION_ERROR_FMT,
55 				 "Raw Addressing",
56 				 GVE_DEV_OPT_LEN_GQI_RAW_ADDRESSING,
57 				 GVE_DEV_OPT_REQ_FEAT_MASK_GQI_RAW_ADDRESSING,
58 				 option_length, req_feat_mask);
59 			break;
60 		}
61 
62 		dev_info(&priv->pdev->dev,
63 			 "Gqi raw addressing device option enabled.\n");
64 		priv->queue_format = GVE_GQI_RDA_FORMAT;
65 		break;
66 	case GVE_DEV_OPT_ID_GQI_RDA:
67 		if (option_length < sizeof(**dev_op_gqi_rda) ||
68 		    req_feat_mask != GVE_DEV_OPT_REQ_FEAT_MASK_GQI_RDA) {
69 			dev_warn(&priv->pdev->dev, GVE_DEVICE_OPTION_ERROR_FMT,
70 				 "GQI RDA", (int)sizeof(**dev_op_gqi_rda),
71 				 GVE_DEV_OPT_REQ_FEAT_MASK_GQI_RDA,
72 				 option_length, req_feat_mask);
73 			break;
74 		}
75 
76 		if (option_length > sizeof(**dev_op_gqi_rda)) {
77 			dev_warn(&priv->pdev->dev,
78 				 GVE_DEVICE_OPTION_TOO_BIG_FMT, "GQI RDA");
79 		}
80 		*dev_op_gqi_rda = (void *)(option + 1);
81 		break;
82 	case GVE_DEV_OPT_ID_GQI_QPL:
83 		if (option_length < sizeof(**dev_op_gqi_qpl) ||
84 		    req_feat_mask != GVE_DEV_OPT_REQ_FEAT_MASK_GQI_QPL) {
85 			dev_warn(&priv->pdev->dev, GVE_DEVICE_OPTION_ERROR_FMT,
86 				 "GQI QPL", (int)sizeof(**dev_op_gqi_qpl),
87 				 GVE_DEV_OPT_REQ_FEAT_MASK_GQI_QPL,
88 				 option_length, req_feat_mask);
89 			break;
90 		}
91 
92 		if (option_length > sizeof(**dev_op_gqi_qpl)) {
93 			dev_warn(&priv->pdev->dev,
94 				 GVE_DEVICE_OPTION_TOO_BIG_FMT, "GQI QPL");
95 		}
96 		*dev_op_gqi_qpl = (void *)(option + 1);
97 		break;
98 	case GVE_DEV_OPT_ID_DQO_RDA:
99 		if (option_length < sizeof(**dev_op_dqo_rda) ||
100 		    req_feat_mask != GVE_DEV_OPT_REQ_FEAT_MASK_DQO_RDA) {
101 			dev_warn(&priv->pdev->dev, GVE_DEVICE_OPTION_ERROR_FMT,
102 				 "DQO RDA", (int)sizeof(**dev_op_dqo_rda),
103 				 GVE_DEV_OPT_REQ_FEAT_MASK_DQO_RDA,
104 				 option_length, req_feat_mask);
105 			break;
106 		}
107 
108 		if (option_length > sizeof(**dev_op_dqo_rda)) {
109 			dev_warn(&priv->pdev->dev,
110 				 GVE_DEVICE_OPTION_TOO_BIG_FMT, "DQO RDA");
111 		}
112 		*dev_op_dqo_rda = (void *)(option + 1);
113 		break;
114 	default:
115 		/* If we don't recognize the option just continue
116 		 * without doing anything.
117 		 */
118 		dev_dbg(&priv->pdev->dev, "Unrecognized device option 0x%hx not enabled.\n",
119 			option_id);
120 	}
121 }
122 
123 /* Process all device options for a given describe device call. */
124 static int
125 gve_process_device_options(struct gve_priv *priv,
126 			   struct gve_device_descriptor *descriptor,
127 			   struct gve_device_option_gqi_rda **dev_op_gqi_rda,
128 			   struct gve_device_option_gqi_qpl **dev_op_gqi_qpl,
129 			   struct gve_device_option_dqo_rda **dev_op_dqo_rda)
130 {
131 	const int num_options = be16_to_cpu(descriptor->num_device_options);
132 	struct gve_device_option *dev_opt;
133 	int i;
134 
135 	/* The options struct directly follows the device descriptor. */
136 	dev_opt = (void *)(descriptor + 1);
137 	for (i = 0; i < num_options; i++) {
138 		struct gve_device_option *next_opt;
139 
140 		next_opt = gve_get_next_option(descriptor, dev_opt);
141 		if (!next_opt) {
142 			dev_err(&priv->dev->dev,
143 				"options exceed device_descriptor's total length.\n");
144 			return -EINVAL;
145 		}
146 
147 		gve_parse_device_option(priv, descriptor, dev_opt,
148 					dev_op_gqi_rda, dev_op_gqi_qpl,
149 					dev_op_dqo_rda);
150 		dev_opt = next_opt;
151 	}
152 
153 	return 0;
154 }
155 
156 int gve_adminq_alloc(struct device *dev, struct gve_priv *priv)
157 {
158 	priv->adminq = dma_alloc_coherent(dev, PAGE_SIZE,
159 					  &priv->adminq_bus_addr, GFP_KERNEL);
160 	if (unlikely(!priv->adminq))
161 		return -ENOMEM;
162 
163 	priv->adminq_mask = (PAGE_SIZE / sizeof(union gve_adminq_command)) - 1;
164 	priv->adminq_prod_cnt = 0;
165 	priv->adminq_cmd_fail = 0;
166 	priv->adminq_timeouts = 0;
167 	priv->adminq_describe_device_cnt = 0;
168 	priv->adminq_cfg_device_resources_cnt = 0;
169 	priv->adminq_register_page_list_cnt = 0;
170 	priv->adminq_unregister_page_list_cnt = 0;
171 	priv->adminq_create_tx_queue_cnt = 0;
172 	priv->adminq_create_rx_queue_cnt = 0;
173 	priv->adminq_destroy_tx_queue_cnt = 0;
174 	priv->adminq_destroy_rx_queue_cnt = 0;
175 	priv->adminq_dcfg_device_resources_cnt = 0;
176 	priv->adminq_set_driver_parameter_cnt = 0;
177 	priv->adminq_report_stats_cnt = 0;
178 	priv->adminq_report_link_speed_cnt = 0;
179 	priv->adminq_get_ptype_map_cnt = 0;
180 
181 	/* Setup Admin queue with the device */
182 	iowrite32be(priv->adminq_bus_addr / PAGE_SIZE,
183 		    &priv->reg_bar0->adminq_pfn);
184 
185 	gve_set_admin_queue_ok(priv);
186 	return 0;
187 }
188 
189 void gve_adminq_release(struct gve_priv *priv)
190 {
191 	int i = 0;
192 
193 	/* Tell the device the adminq is leaving */
194 	iowrite32be(0x0, &priv->reg_bar0->adminq_pfn);
195 	while (ioread32be(&priv->reg_bar0->adminq_pfn)) {
196 		/* If this is reached the device is unrecoverable and still
197 		 * holding memory. Continue looping to avoid memory corruption,
198 		 * but WARN so it is visible what is going on.
199 		 */
200 		if (i == GVE_MAX_ADMINQ_RELEASE_CHECK)
201 			WARN(1, "Unrecoverable platform error!");
202 		i++;
203 		msleep(GVE_ADMINQ_SLEEP_LEN);
204 	}
205 	gve_clear_device_rings_ok(priv);
206 	gve_clear_device_resources_ok(priv);
207 	gve_clear_admin_queue_ok(priv);
208 }
209 
210 void gve_adminq_free(struct device *dev, struct gve_priv *priv)
211 {
212 	if (!gve_get_admin_queue_ok(priv))
213 		return;
214 	gve_adminq_release(priv);
215 	dma_free_coherent(dev, PAGE_SIZE, priv->adminq, priv->adminq_bus_addr);
216 	gve_clear_admin_queue_ok(priv);
217 }
218 
219 static void gve_adminq_kick_cmd(struct gve_priv *priv, u32 prod_cnt)
220 {
221 	iowrite32be(prod_cnt, &priv->reg_bar0->adminq_doorbell);
222 }
223 
224 static bool gve_adminq_wait_for_cmd(struct gve_priv *priv, u32 prod_cnt)
225 {
226 	int i;
227 
228 	for (i = 0; i < GVE_MAX_ADMINQ_EVENT_COUNTER_CHECK; i++) {
229 		if (ioread32be(&priv->reg_bar0->adminq_event_counter)
230 		    == prod_cnt)
231 			return true;
232 		msleep(GVE_ADMINQ_SLEEP_LEN);
233 	}
234 
235 	return false;
236 }
237 
238 static int gve_adminq_parse_err(struct gve_priv *priv, u32 status)
239 {
240 	if (status != GVE_ADMINQ_COMMAND_PASSED &&
241 	    status != GVE_ADMINQ_COMMAND_UNSET) {
242 		dev_err(&priv->pdev->dev, "AQ command failed with status %d\n", status);
243 		priv->adminq_cmd_fail++;
244 	}
245 	switch (status) {
246 	case GVE_ADMINQ_COMMAND_PASSED:
247 		return 0;
248 	case GVE_ADMINQ_COMMAND_UNSET:
249 		dev_err(&priv->pdev->dev, "parse_aq_err: err and status both unset, this should not be possible.\n");
250 		return -EINVAL;
251 	case GVE_ADMINQ_COMMAND_ERROR_ABORTED:
252 	case GVE_ADMINQ_COMMAND_ERROR_CANCELLED:
253 	case GVE_ADMINQ_COMMAND_ERROR_DATALOSS:
254 	case GVE_ADMINQ_COMMAND_ERROR_FAILED_PRECONDITION:
255 	case GVE_ADMINQ_COMMAND_ERROR_UNAVAILABLE:
256 		return -EAGAIN;
257 	case GVE_ADMINQ_COMMAND_ERROR_ALREADY_EXISTS:
258 	case GVE_ADMINQ_COMMAND_ERROR_INTERNAL_ERROR:
259 	case GVE_ADMINQ_COMMAND_ERROR_INVALID_ARGUMENT:
260 	case GVE_ADMINQ_COMMAND_ERROR_NOT_FOUND:
261 	case GVE_ADMINQ_COMMAND_ERROR_OUT_OF_RANGE:
262 	case GVE_ADMINQ_COMMAND_ERROR_UNKNOWN_ERROR:
263 		return -EINVAL;
264 	case GVE_ADMINQ_COMMAND_ERROR_DEADLINE_EXCEEDED:
265 		return -ETIME;
266 	case GVE_ADMINQ_COMMAND_ERROR_PERMISSION_DENIED:
267 	case GVE_ADMINQ_COMMAND_ERROR_UNAUTHENTICATED:
268 		return -EACCES;
269 	case GVE_ADMINQ_COMMAND_ERROR_RESOURCE_EXHAUSTED:
270 		return -ENOMEM;
271 	case GVE_ADMINQ_COMMAND_ERROR_UNIMPLEMENTED:
272 		return -ENOTSUPP;
273 	default:
274 		dev_err(&priv->pdev->dev, "parse_aq_err: unknown status code %d\n", status);
275 		return -EINVAL;
276 	}
277 }
278 
279 /* Flushes all AQ commands currently queued and waits for them to complete.
280  * If there are failures, it will return the first error.
281  */
282 static int gve_adminq_kick_and_wait(struct gve_priv *priv)
283 {
284 	u32 tail, head;
285 	int i;
286 
287 	tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
288 	head = priv->adminq_prod_cnt;
289 
290 	gve_adminq_kick_cmd(priv, head);
291 	if (!gve_adminq_wait_for_cmd(priv, head)) {
292 		dev_err(&priv->pdev->dev, "AQ commands timed out, need to reset AQ\n");
293 		priv->adminq_timeouts++;
294 		return -ENOTRECOVERABLE;
295 	}
296 
297 	for (i = tail; i < head; i++) {
298 		union gve_adminq_command *cmd;
299 		u32 status, err;
300 
301 		cmd = &priv->adminq[i & priv->adminq_mask];
302 		status = be32_to_cpu(READ_ONCE(cmd->status));
303 		err = gve_adminq_parse_err(priv, status);
304 		if (err)
305 			// Return the first error if we failed.
306 			return err;
307 	}
308 
309 	return 0;
310 }
311 
312 /* This function is not threadsafe - the caller is responsible for any
313  * necessary locks.
314  */
315 static int gve_adminq_issue_cmd(struct gve_priv *priv,
316 				union gve_adminq_command *cmd_orig)
317 {
318 	union gve_adminq_command *cmd;
319 	u32 opcode;
320 	u32 tail;
321 
322 	tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
323 
324 	// Check if next command will overflow the buffer.
325 	if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) ==
326 	    (tail & priv->adminq_mask)) {
327 		int err;
328 
329 		// Flush existing commands to make room.
330 		err = gve_adminq_kick_and_wait(priv);
331 		if (err)
332 			return err;
333 
334 		// Retry.
335 		tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
336 		if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) ==
337 		    (tail & priv->adminq_mask)) {
338 			// This should never happen. We just flushed the
339 			// command queue so there should be enough space.
340 			return -ENOMEM;
341 		}
342 	}
343 
344 	cmd = &priv->adminq[priv->adminq_prod_cnt & priv->adminq_mask];
345 	priv->adminq_prod_cnt++;
346 
347 	memcpy(cmd, cmd_orig, sizeof(*cmd_orig));
348 	opcode = be32_to_cpu(READ_ONCE(cmd->opcode));
349 
350 	switch (opcode) {
351 	case GVE_ADMINQ_DESCRIBE_DEVICE:
352 		priv->adminq_describe_device_cnt++;
353 		break;
354 	case GVE_ADMINQ_CONFIGURE_DEVICE_RESOURCES:
355 		priv->adminq_cfg_device_resources_cnt++;
356 		break;
357 	case GVE_ADMINQ_REGISTER_PAGE_LIST:
358 		priv->adminq_register_page_list_cnt++;
359 		break;
360 	case GVE_ADMINQ_UNREGISTER_PAGE_LIST:
361 		priv->adminq_unregister_page_list_cnt++;
362 		break;
363 	case GVE_ADMINQ_CREATE_TX_QUEUE:
364 		priv->adminq_create_tx_queue_cnt++;
365 		break;
366 	case GVE_ADMINQ_CREATE_RX_QUEUE:
367 		priv->adminq_create_rx_queue_cnt++;
368 		break;
369 	case GVE_ADMINQ_DESTROY_TX_QUEUE:
370 		priv->adminq_destroy_tx_queue_cnt++;
371 		break;
372 	case GVE_ADMINQ_DESTROY_RX_QUEUE:
373 		priv->adminq_destroy_rx_queue_cnt++;
374 		break;
375 	case GVE_ADMINQ_DECONFIGURE_DEVICE_RESOURCES:
376 		priv->adminq_dcfg_device_resources_cnt++;
377 		break;
378 	case GVE_ADMINQ_SET_DRIVER_PARAMETER:
379 		priv->adminq_set_driver_parameter_cnt++;
380 		break;
381 	case GVE_ADMINQ_REPORT_STATS:
382 		priv->adminq_report_stats_cnt++;
383 		break;
384 	case GVE_ADMINQ_REPORT_LINK_SPEED:
385 		priv->adminq_report_link_speed_cnt++;
386 		break;
387 	case GVE_ADMINQ_GET_PTYPE_MAP:
388 		priv->adminq_get_ptype_map_cnt++;
389 		break;
390 	default:
391 		dev_err(&priv->pdev->dev, "unknown AQ command opcode %d\n", opcode);
392 	}
393 
394 	return 0;
395 }
396 
397 /* This function is not threadsafe - the caller is responsible for any
398  * necessary locks.
399  * The caller is also responsible for making sure there are no commands
400  * waiting to be executed.
401  */
402 static int gve_adminq_execute_cmd(struct gve_priv *priv,
403 				  union gve_adminq_command *cmd_orig)
404 {
405 	u32 tail, head;
406 	int err;
407 
408 	tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
409 	head = priv->adminq_prod_cnt;
410 	if (tail != head)
411 		// This is not a valid path
412 		return -EINVAL;
413 
414 	err = gve_adminq_issue_cmd(priv, cmd_orig);
415 	if (err)
416 		return err;
417 
418 	return gve_adminq_kick_and_wait(priv);
419 }
420 
421 /* The device specifies that the management vector can either be the first irq
422  * or the last irq. ntfy_blk_msix_base_idx indicates the first irq assigned to
423  * the ntfy blks. It if is 0 then the management vector is last, if it is 1 then
424  * the management vector is first.
425  *
426  * gve arranges the msix vectors so that the management vector is last.
427  */
428 #define GVE_NTFY_BLK_BASE_MSIX_IDX	0
429 int gve_adminq_configure_device_resources(struct gve_priv *priv,
430 					  dma_addr_t counter_array_bus_addr,
431 					  u32 num_counters,
432 					  dma_addr_t db_array_bus_addr,
433 					  u32 num_ntfy_blks)
434 {
435 	union gve_adminq_command cmd;
436 
437 	memset(&cmd, 0, sizeof(cmd));
438 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_CONFIGURE_DEVICE_RESOURCES);
439 	cmd.configure_device_resources =
440 		(struct gve_adminq_configure_device_resources) {
441 		.counter_array = cpu_to_be64(counter_array_bus_addr),
442 		.num_counters = cpu_to_be32(num_counters),
443 		.irq_db_addr = cpu_to_be64(db_array_bus_addr),
444 		.num_irq_dbs = cpu_to_be32(num_ntfy_blks),
445 		.irq_db_stride = cpu_to_be32(sizeof(priv->ntfy_blocks[0])),
446 		.ntfy_blk_msix_base_idx =
447 					cpu_to_be32(GVE_NTFY_BLK_BASE_MSIX_IDX),
448 		.queue_format = priv->queue_format,
449 	};
450 
451 	return gve_adminq_execute_cmd(priv, &cmd);
452 }
453 
454 int gve_adminq_deconfigure_device_resources(struct gve_priv *priv)
455 {
456 	union gve_adminq_command cmd;
457 
458 	memset(&cmd, 0, sizeof(cmd));
459 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_DECONFIGURE_DEVICE_RESOURCES);
460 
461 	return gve_adminq_execute_cmd(priv, &cmd);
462 }
463 
464 static int gve_adminq_create_tx_queue(struct gve_priv *priv, u32 queue_index)
465 {
466 	struct gve_tx_ring *tx = &priv->tx[queue_index];
467 	union gve_adminq_command cmd;
468 
469 	memset(&cmd, 0, sizeof(cmd));
470 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_CREATE_TX_QUEUE);
471 	cmd.create_tx_queue = (struct gve_adminq_create_tx_queue) {
472 		.queue_id = cpu_to_be32(queue_index),
473 		.queue_resources_addr =
474 			cpu_to_be64(tx->q_resources_bus),
475 		.tx_ring_addr = cpu_to_be64(tx->bus),
476 		.ntfy_id = cpu_to_be32(tx->ntfy_id),
477 	};
478 
479 	if (gve_is_gqi(priv)) {
480 		u32 qpl_id = priv->queue_format == GVE_GQI_RDA_FORMAT ?
481 			GVE_RAW_ADDRESSING_QPL_ID : tx->tx_fifo.qpl->id;
482 
483 		cmd.create_tx_queue.queue_page_list_id = cpu_to_be32(qpl_id);
484 	} else {
485 		cmd.create_tx_queue.tx_ring_size =
486 			cpu_to_be16(priv->tx_desc_cnt);
487 		cmd.create_tx_queue.tx_comp_ring_addr =
488 			cpu_to_be64(tx->complq_bus_dqo);
489 		cmd.create_tx_queue.tx_comp_ring_size =
490 			cpu_to_be16(priv->options_dqo_rda.tx_comp_ring_entries);
491 	}
492 
493 	return gve_adminq_issue_cmd(priv, &cmd);
494 }
495 
496 int gve_adminq_create_tx_queues(struct gve_priv *priv, u32 num_queues)
497 {
498 	int err;
499 	int i;
500 
501 	for (i = 0; i < num_queues; i++) {
502 		err = gve_adminq_create_tx_queue(priv, i);
503 		if (err)
504 			return err;
505 	}
506 
507 	return gve_adminq_kick_and_wait(priv);
508 }
509 
510 static int gve_adminq_create_rx_queue(struct gve_priv *priv, u32 queue_index)
511 {
512 	struct gve_rx_ring *rx = &priv->rx[queue_index];
513 	union gve_adminq_command cmd;
514 
515 	memset(&cmd, 0, sizeof(cmd));
516 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_CREATE_RX_QUEUE);
517 	cmd.create_rx_queue = (struct gve_adminq_create_rx_queue) {
518 		.queue_id = cpu_to_be32(queue_index),
519 		.ntfy_id = cpu_to_be32(rx->ntfy_id),
520 		.queue_resources_addr = cpu_to_be64(rx->q_resources_bus),
521 	};
522 
523 	if (gve_is_gqi(priv)) {
524 		u32 qpl_id = priv->queue_format == GVE_GQI_RDA_FORMAT ?
525 			GVE_RAW_ADDRESSING_QPL_ID : rx->data.qpl->id;
526 
527 		cmd.create_rx_queue.rx_desc_ring_addr =
528 			cpu_to_be64(rx->desc.bus),
529 		cmd.create_rx_queue.rx_data_ring_addr =
530 			cpu_to_be64(rx->data.data_bus),
531 		cmd.create_rx_queue.index = cpu_to_be32(queue_index);
532 		cmd.create_rx_queue.queue_page_list_id = cpu_to_be32(qpl_id);
533 	} else {
534 		cmd.create_rx_queue.rx_ring_size =
535 			cpu_to_be16(priv->rx_desc_cnt);
536 		cmd.create_rx_queue.rx_desc_ring_addr =
537 			cpu_to_be64(rx->dqo.complq.bus);
538 		cmd.create_rx_queue.rx_data_ring_addr =
539 			cpu_to_be64(rx->dqo.bufq.bus);
540 		cmd.create_rx_queue.packet_buffer_size =
541 			cpu_to_be16(priv->data_buffer_size_dqo);
542 		cmd.create_rx_queue.rx_buff_ring_size =
543 			cpu_to_be16(priv->options_dqo_rda.rx_buff_ring_entries);
544 		cmd.create_rx_queue.enable_rsc =
545 			!!(priv->dev->features & NETIF_F_LRO);
546 	}
547 
548 	return gve_adminq_issue_cmd(priv, &cmd);
549 }
550 
551 int gve_adminq_create_rx_queues(struct gve_priv *priv, u32 num_queues)
552 {
553 	int err;
554 	int i;
555 
556 	for (i = 0; i < num_queues; i++) {
557 		err = gve_adminq_create_rx_queue(priv, i);
558 		if (err)
559 			return err;
560 	}
561 
562 	return gve_adminq_kick_and_wait(priv);
563 }
564 
565 static int gve_adminq_destroy_tx_queue(struct gve_priv *priv, u32 queue_index)
566 {
567 	union gve_adminq_command cmd;
568 	int err;
569 
570 	memset(&cmd, 0, sizeof(cmd));
571 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_DESTROY_TX_QUEUE);
572 	cmd.destroy_tx_queue = (struct gve_adminq_destroy_tx_queue) {
573 		.queue_id = cpu_to_be32(queue_index),
574 	};
575 
576 	err = gve_adminq_issue_cmd(priv, &cmd);
577 	if (err)
578 		return err;
579 
580 	return 0;
581 }
582 
583 int gve_adminq_destroy_tx_queues(struct gve_priv *priv, u32 num_queues)
584 {
585 	int err;
586 	int i;
587 
588 	for (i = 0; i < num_queues; i++) {
589 		err = gve_adminq_destroy_tx_queue(priv, i);
590 		if (err)
591 			return err;
592 	}
593 
594 	return gve_adminq_kick_and_wait(priv);
595 }
596 
597 static int gve_adminq_destroy_rx_queue(struct gve_priv *priv, u32 queue_index)
598 {
599 	union gve_adminq_command cmd;
600 	int err;
601 
602 	memset(&cmd, 0, sizeof(cmd));
603 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_DESTROY_RX_QUEUE);
604 	cmd.destroy_rx_queue = (struct gve_adminq_destroy_rx_queue) {
605 		.queue_id = cpu_to_be32(queue_index),
606 	};
607 
608 	err = gve_adminq_issue_cmd(priv, &cmd);
609 	if (err)
610 		return err;
611 
612 	return 0;
613 }
614 
615 int gve_adminq_destroy_rx_queues(struct gve_priv *priv, u32 num_queues)
616 {
617 	int err;
618 	int i;
619 
620 	for (i = 0; i < num_queues; i++) {
621 		err = gve_adminq_destroy_rx_queue(priv, i);
622 		if (err)
623 			return err;
624 	}
625 
626 	return gve_adminq_kick_and_wait(priv);
627 }
628 
629 static int gve_set_desc_cnt(struct gve_priv *priv,
630 			    struct gve_device_descriptor *descriptor)
631 {
632 	priv->tx_desc_cnt = be16_to_cpu(descriptor->tx_queue_entries);
633 	if (priv->tx_desc_cnt * sizeof(priv->tx->desc[0]) < PAGE_SIZE) {
634 		dev_err(&priv->pdev->dev, "Tx desc count %d too low\n",
635 			priv->tx_desc_cnt);
636 		return -EINVAL;
637 	}
638 	priv->rx_desc_cnt = be16_to_cpu(descriptor->rx_queue_entries);
639 	if (priv->rx_desc_cnt * sizeof(priv->rx->desc.desc_ring[0])
640 	    < PAGE_SIZE) {
641 		dev_err(&priv->pdev->dev, "Rx desc count %d too low\n",
642 			priv->rx_desc_cnt);
643 		return -EINVAL;
644 	}
645 	return 0;
646 }
647 
648 static int
649 gve_set_desc_cnt_dqo(struct gve_priv *priv,
650 		     const struct gve_device_descriptor *descriptor,
651 		     const struct gve_device_option_dqo_rda *dev_op_dqo_rda)
652 {
653 	priv->tx_desc_cnt = be16_to_cpu(descriptor->tx_queue_entries);
654 	priv->options_dqo_rda.tx_comp_ring_entries =
655 		be16_to_cpu(dev_op_dqo_rda->tx_comp_ring_entries);
656 	priv->rx_desc_cnt = be16_to_cpu(descriptor->rx_queue_entries);
657 	priv->options_dqo_rda.rx_buff_ring_entries =
658 		be16_to_cpu(dev_op_dqo_rda->rx_buff_ring_entries);
659 
660 	return 0;
661 }
662 
663 int gve_adminq_describe_device(struct gve_priv *priv)
664 {
665 	struct gve_device_option_gqi_rda *dev_op_gqi_rda = NULL;
666 	struct gve_device_option_gqi_qpl *dev_op_gqi_qpl = NULL;
667 	struct gve_device_option_dqo_rda *dev_op_dqo_rda = NULL;
668 	struct gve_device_descriptor *descriptor;
669 	union gve_adminq_command cmd;
670 	dma_addr_t descriptor_bus;
671 	int err = 0;
672 	u8 *mac;
673 	u16 mtu;
674 
675 	memset(&cmd, 0, sizeof(cmd));
676 	descriptor = dma_alloc_coherent(&priv->pdev->dev, PAGE_SIZE,
677 					&descriptor_bus, GFP_KERNEL);
678 	if (!descriptor)
679 		return -ENOMEM;
680 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_DESCRIBE_DEVICE);
681 	cmd.describe_device.device_descriptor_addr =
682 						cpu_to_be64(descriptor_bus);
683 	cmd.describe_device.device_descriptor_version =
684 			cpu_to_be32(GVE_ADMINQ_DEVICE_DESCRIPTOR_VERSION);
685 	cmd.describe_device.available_length = cpu_to_be32(PAGE_SIZE);
686 
687 	err = gve_adminq_execute_cmd(priv, &cmd);
688 	if (err)
689 		goto free_device_descriptor;
690 
691 	err = gve_process_device_options(priv, descriptor, &dev_op_gqi_rda,
692 					 &dev_op_gqi_qpl, &dev_op_dqo_rda);
693 	if (err)
694 		goto free_device_descriptor;
695 
696 	/* If the GQI_RAW_ADDRESSING option is not enabled and the queue format
697 	 * is not set to GqiRda, choose the queue format in a priority order:
698 	 * DqoRda, GqiRda, GqiQpl. Use GqiQpl as default.
699 	 */
700 	if (priv->queue_format == GVE_GQI_RDA_FORMAT) {
701 		dev_info(&priv->pdev->dev,
702 			 "Driver is running with GQI RDA queue format.\n");
703 	} else if (dev_op_dqo_rda) {
704 		priv->queue_format = GVE_DQO_RDA_FORMAT;
705 		dev_info(&priv->pdev->dev,
706 			 "Driver is running with DQO RDA queue format.\n");
707 	} else if (dev_op_gqi_rda) {
708 		priv->queue_format = GVE_GQI_RDA_FORMAT;
709 		dev_info(&priv->pdev->dev,
710 			 "Driver is running with GQI RDA queue format.\n");
711 	} else {
712 		priv->queue_format = GVE_GQI_QPL_FORMAT;
713 		dev_info(&priv->pdev->dev,
714 			 "Driver is running with GQI QPL queue format.\n");
715 	}
716 	if (gve_is_gqi(priv)) {
717 		err = gve_set_desc_cnt(priv, descriptor);
718 	} else {
719 		/* DQO supports LRO. */
720 		priv->dev->hw_features |= NETIF_F_LRO;
721 		err = gve_set_desc_cnt_dqo(priv, descriptor, dev_op_dqo_rda);
722 	}
723 	if (err)
724 		goto free_device_descriptor;
725 
726 	priv->max_registered_pages =
727 				be64_to_cpu(descriptor->max_registered_pages);
728 	mtu = be16_to_cpu(descriptor->mtu);
729 	if (mtu < ETH_MIN_MTU) {
730 		dev_err(&priv->pdev->dev, "MTU %d below minimum MTU\n", mtu);
731 		err = -EINVAL;
732 		goto free_device_descriptor;
733 	}
734 	priv->dev->max_mtu = mtu;
735 	priv->num_event_counters = be16_to_cpu(descriptor->counters);
736 	ether_addr_copy(priv->dev->dev_addr, descriptor->mac);
737 	mac = descriptor->mac;
738 	dev_info(&priv->pdev->dev, "MAC addr: %pM\n", mac);
739 	priv->tx_pages_per_qpl = be16_to_cpu(descriptor->tx_pages_per_qpl);
740 	priv->rx_data_slot_cnt = be16_to_cpu(descriptor->rx_pages_per_qpl);
741 
742 	if (gve_is_gqi(priv) && priv->rx_data_slot_cnt < priv->rx_desc_cnt) {
743 		dev_err(&priv->pdev->dev, "rx_data_slot_cnt cannot be smaller than rx_desc_cnt, setting rx_desc_cnt down to %d.\n",
744 			priv->rx_data_slot_cnt);
745 		priv->rx_desc_cnt = priv->rx_data_slot_cnt;
746 	}
747 	priv->default_num_queues = be16_to_cpu(descriptor->default_num_queues);
748 
749 free_device_descriptor:
750 	dma_free_coherent(&priv->pdev->dev, PAGE_SIZE, descriptor,
751 			  descriptor_bus);
752 	return err;
753 }
754 
755 int gve_adminq_register_page_list(struct gve_priv *priv,
756 				  struct gve_queue_page_list *qpl)
757 {
758 	struct device *hdev = &priv->pdev->dev;
759 	u32 num_entries = qpl->num_entries;
760 	u32 size = num_entries * sizeof(qpl->page_buses[0]);
761 	union gve_adminq_command cmd;
762 	dma_addr_t page_list_bus;
763 	__be64 *page_list;
764 	int err;
765 	int i;
766 
767 	memset(&cmd, 0, sizeof(cmd));
768 	page_list = dma_alloc_coherent(hdev, size, &page_list_bus, GFP_KERNEL);
769 	if (!page_list)
770 		return -ENOMEM;
771 
772 	for (i = 0; i < num_entries; i++)
773 		page_list[i] = cpu_to_be64(qpl->page_buses[i]);
774 
775 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_REGISTER_PAGE_LIST);
776 	cmd.reg_page_list = (struct gve_adminq_register_page_list) {
777 		.page_list_id = cpu_to_be32(qpl->id),
778 		.num_pages = cpu_to_be32(num_entries),
779 		.page_address_list_addr = cpu_to_be64(page_list_bus),
780 	};
781 
782 	err = gve_adminq_execute_cmd(priv, &cmd);
783 	dma_free_coherent(hdev, size, page_list, page_list_bus);
784 	return err;
785 }
786 
787 int gve_adminq_unregister_page_list(struct gve_priv *priv, u32 page_list_id)
788 {
789 	union gve_adminq_command cmd;
790 
791 	memset(&cmd, 0, sizeof(cmd));
792 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_UNREGISTER_PAGE_LIST);
793 	cmd.unreg_page_list = (struct gve_adminq_unregister_page_list) {
794 		.page_list_id = cpu_to_be32(page_list_id),
795 	};
796 
797 	return gve_adminq_execute_cmd(priv, &cmd);
798 }
799 
800 int gve_adminq_set_mtu(struct gve_priv *priv, u64 mtu)
801 {
802 	union gve_adminq_command cmd;
803 
804 	memset(&cmd, 0, sizeof(cmd));
805 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_SET_DRIVER_PARAMETER);
806 	cmd.set_driver_param = (struct gve_adminq_set_driver_parameter) {
807 		.parameter_type = cpu_to_be32(GVE_SET_PARAM_MTU),
808 		.parameter_value = cpu_to_be64(mtu),
809 	};
810 
811 	return gve_adminq_execute_cmd(priv, &cmd);
812 }
813 
814 int gve_adminq_report_stats(struct gve_priv *priv, u64 stats_report_len,
815 			    dma_addr_t stats_report_addr, u64 interval)
816 {
817 	union gve_adminq_command cmd;
818 
819 	memset(&cmd, 0, sizeof(cmd));
820 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_REPORT_STATS);
821 	cmd.report_stats = (struct gve_adminq_report_stats) {
822 		.stats_report_len = cpu_to_be64(stats_report_len),
823 		.stats_report_addr = cpu_to_be64(stats_report_addr),
824 		.interval = cpu_to_be64(interval),
825 	};
826 
827 	return gve_adminq_execute_cmd(priv, &cmd);
828 }
829 
830 int gve_adminq_report_link_speed(struct gve_priv *priv)
831 {
832 	union gve_adminq_command gvnic_cmd;
833 	dma_addr_t link_speed_region_bus;
834 	__be64 *link_speed_region;
835 	int err;
836 
837 	link_speed_region =
838 		dma_alloc_coherent(&priv->pdev->dev, sizeof(*link_speed_region),
839 				   &link_speed_region_bus, GFP_KERNEL);
840 
841 	if (!link_speed_region)
842 		return -ENOMEM;
843 
844 	memset(&gvnic_cmd, 0, sizeof(gvnic_cmd));
845 	gvnic_cmd.opcode = cpu_to_be32(GVE_ADMINQ_REPORT_LINK_SPEED);
846 	gvnic_cmd.report_link_speed.link_speed_address =
847 		cpu_to_be64(link_speed_region_bus);
848 
849 	err = gve_adminq_execute_cmd(priv, &gvnic_cmd);
850 
851 	priv->link_speed = be64_to_cpu(*link_speed_region);
852 	dma_free_coherent(&priv->pdev->dev, sizeof(*link_speed_region), link_speed_region,
853 			  link_speed_region_bus);
854 	return err;
855 }
856 
857 int gve_adminq_get_ptype_map_dqo(struct gve_priv *priv,
858 				 struct gve_ptype_lut *ptype_lut)
859 {
860 	struct gve_ptype_map *ptype_map;
861 	union gve_adminq_command cmd;
862 	dma_addr_t ptype_map_bus;
863 	int err = 0;
864 	int i;
865 
866 	memset(&cmd, 0, sizeof(cmd));
867 	ptype_map = dma_alloc_coherent(&priv->pdev->dev, sizeof(*ptype_map),
868 				       &ptype_map_bus, GFP_KERNEL);
869 	if (!ptype_map)
870 		return -ENOMEM;
871 
872 	cmd.opcode = cpu_to_be32(GVE_ADMINQ_GET_PTYPE_MAP);
873 	cmd.get_ptype_map = (struct gve_adminq_get_ptype_map) {
874 		.ptype_map_len = cpu_to_be64(sizeof(*ptype_map)),
875 		.ptype_map_addr = cpu_to_be64(ptype_map_bus),
876 	};
877 
878 	err = gve_adminq_execute_cmd(priv, &cmd);
879 	if (err)
880 		goto err;
881 
882 	/* Populate ptype_lut. */
883 	for (i = 0; i < GVE_NUM_PTYPES; i++) {
884 		ptype_lut->ptypes[i].l3_type =
885 			ptype_map->ptypes[i].l3_type;
886 		ptype_lut->ptypes[i].l4_type =
887 			ptype_map->ptypes[i].l4_type;
888 	}
889 err:
890 	dma_free_coherent(&priv->pdev->dev, sizeof(*ptype_map), ptype_map,
891 			  ptype_map_bus);
892 	return err;
893 }
894