xref: /linux/drivers/net/ethernet/meta/fbnic/fbnic_pci.c (revision 55d0969c451159cff86949b38c39171cab962069)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) Meta Platforms, Inc. and affiliates. */
3 
4 #include <linux/init.h>
5 #include <linux/module.h>
6 #include <linux/pci.h>
7 #include <linux/rtnetlink.h>
8 #include <linux/types.h>
9 
10 #include "fbnic.h"
11 #include "fbnic_drvinfo.h"
12 #include "fbnic_netdev.h"
13 
14 char fbnic_driver_name[] = DRV_NAME;
15 
16 MODULE_DESCRIPTION(DRV_SUMMARY);
17 MODULE_LICENSE("GPL");
18 
19 static const struct fbnic_info fbnic_asic_info = {
20 	.max_num_queues = FBNIC_MAX_QUEUES,
21 	.bar_mask = BIT(0) | BIT(4)
22 };
23 
24 static const struct fbnic_info *fbnic_info_tbl[] = {
25 	[fbnic_board_asic] = &fbnic_asic_info,
26 };
27 
28 static const struct pci_device_id fbnic_pci_tbl[] = {
29 	{ PCI_DEVICE_DATA(META, FBNIC_ASIC, fbnic_board_asic) },
30 	/* Required last entry */
31 	{0, }
32 };
33 MODULE_DEVICE_TABLE(pci, fbnic_pci_tbl);
34 
35 u32 fbnic_rd32(struct fbnic_dev *fbd, u32 reg)
36 {
37 	u32 __iomem *csr = READ_ONCE(fbd->uc_addr0);
38 	u32 value;
39 
40 	if (!csr)
41 		return ~0U;
42 
43 	value = readl(csr + reg);
44 
45 	/* If any bits are 0 value should be valid */
46 	if (~value)
47 		return value;
48 
49 	/* All 1's may be valid if ZEROs register still works */
50 	if (reg != FBNIC_MASTER_SPARE_0 && ~readl(csr + FBNIC_MASTER_SPARE_0))
51 		return value;
52 
53 	/* Hardware is giving us all 1's reads, assume it is gone */
54 	WRITE_ONCE(fbd->uc_addr0, NULL);
55 	WRITE_ONCE(fbd->uc_addr4, NULL);
56 
57 	dev_err(fbd->dev,
58 		"Failed read (idx 0x%x AKA addr 0x%x), disabled CSR access, awaiting reset\n",
59 		reg, reg << 2);
60 
61 	/* Notify stack that device has lost (PCIe) link */
62 	if (!fbnic_init_failure(fbd))
63 		netif_device_detach(fbd->netdev);
64 
65 	return ~0U;
66 }
67 
68 bool fbnic_fw_present(struct fbnic_dev *fbd)
69 {
70 	return !!READ_ONCE(fbd->uc_addr4);
71 }
72 
73 void fbnic_fw_wr32(struct fbnic_dev *fbd, u32 reg, u32 val)
74 {
75 	u32 __iomem *csr = READ_ONCE(fbd->uc_addr4);
76 
77 	if (csr)
78 		writel(val, csr + reg);
79 }
80 
81 u32 fbnic_fw_rd32(struct fbnic_dev *fbd, u32 reg)
82 {
83 	u32 __iomem *csr = READ_ONCE(fbd->uc_addr4);
84 	u32 value;
85 
86 	if (!csr)
87 		return ~0U;
88 
89 	value = readl(csr + reg);
90 
91 	/* If any bits are 0 value should be valid */
92 	if (~value)
93 		return value;
94 
95 	/* All 1's may be valid if ZEROs register still works */
96 	if (reg != FBNIC_FW_ZERO_REG && ~readl(csr + FBNIC_FW_ZERO_REG))
97 		return value;
98 
99 	/* Hardware is giving us all 1's reads, assume it is gone */
100 	WRITE_ONCE(fbd->uc_addr0, NULL);
101 	WRITE_ONCE(fbd->uc_addr4, NULL);
102 
103 	dev_err(fbd->dev,
104 		"Failed read (idx 0x%x AKA addr 0x%x), disabled CSR access, awaiting reset\n",
105 		reg, reg << 2);
106 
107 	/* Notify stack that device has lost (PCIe) link */
108 	if (!fbnic_init_failure(fbd))
109 		netif_device_detach(fbd->netdev);
110 
111 	return ~0U;
112 }
113 
114 static void fbnic_service_task_start(struct fbnic_net *fbn)
115 {
116 	struct fbnic_dev *fbd = fbn->fbd;
117 
118 	schedule_delayed_work(&fbd->service_task, HZ);
119 	phylink_resume(fbn->phylink);
120 }
121 
122 static void fbnic_service_task_stop(struct fbnic_net *fbn)
123 {
124 	struct fbnic_dev *fbd = fbn->fbd;
125 
126 	phylink_suspend(fbn->phylink, fbnic_bmc_present(fbd));
127 	cancel_delayed_work(&fbd->service_task);
128 }
129 
130 void fbnic_up(struct fbnic_net *fbn)
131 {
132 	fbnic_enable(fbn);
133 
134 	fbnic_fill(fbn);
135 
136 	fbnic_rss_reinit_hw(fbn->fbd, fbn);
137 
138 	__fbnic_set_rx_mode(fbn->netdev);
139 
140 	/* Enable Tx/Rx processing */
141 	fbnic_napi_enable(fbn);
142 	netif_tx_start_all_queues(fbn->netdev);
143 
144 	fbnic_service_task_start(fbn);
145 }
146 
147 static void fbnic_down_noidle(struct fbnic_net *fbn)
148 {
149 	fbnic_service_task_stop(fbn);
150 
151 	/* Disable Tx/Rx Processing */
152 	fbnic_napi_disable(fbn);
153 	netif_tx_disable(fbn->netdev);
154 
155 	fbnic_clear_rx_mode(fbn->netdev);
156 	fbnic_clear_rules(fbn->fbd);
157 	fbnic_rss_disable_hw(fbn->fbd);
158 	fbnic_disable(fbn);
159 }
160 
161 void fbnic_down(struct fbnic_net *fbn)
162 {
163 	fbnic_down_noidle(fbn);
164 
165 	fbnic_wait_all_queues_idle(fbn->fbd, false);
166 
167 	fbnic_flush(fbn);
168 }
169 
170 static void fbnic_health_check(struct fbnic_dev *fbd)
171 {
172 	struct fbnic_fw_mbx *tx_mbx = &fbd->mbx[FBNIC_IPC_MBX_TX_IDX];
173 
174 	/* As long as the heart is beating the FW is healty */
175 	if (fbd->fw_heartbeat_enabled)
176 		return;
177 
178 	/* If the Tx mailbox still has messages sitting in it then there likely
179 	 * isn't anything we can do. We will wait until the mailbox is empty to
180 	 * report the fault so we can collect the crashlog.
181 	 */
182 	if (tx_mbx->head != tx_mbx->tail)
183 		return;
184 
185 	/* TBD: Need to add a more thorough recovery here.
186 	 *	Specifically I need to verify what all the firmware will have
187 	 *	changed since we had setup and it rebooted. May just need to
188 	 *	perform a down/up. For now we will just reclaim ownership so
189 	 *	the heartbeat can catch the next fault.
190 	 */
191 	fbnic_fw_xmit_ownership_msg(fbd, true);
192 }
193 
194 static void fbnic_service_task(struct work_struct *work)
195 {
196 	struct fbnic_dev *fbd = container_of(to_delayed_work(work),
197 					     struct fbnic_dev, service_task);
198 
199 	rtnl_lock();
200 
201 	fbnic_fw_check_heartbeat(fbd);
202 
203 	fbnic_health_check(fbd);
204 
205 	if (netif_carrier_ok(fbd->netdev))
206 		fbnic_napi_depletion_check(fbd->netdev);
207 
208 	if (netif_running(fbd->netdev))
209 		schedule_delayed_work(&fbd->service_task, HZ);
210 
211 	rtnl_unlock();
212 }
213 
214 /**
215  * fbnic_probe - Device Initialization Routine
216  * @pdev: PCI device information struct
217  * @ent: entry in fbnic_pci_tbl
218  *
219  * Initializes a PCI device identified by a pci_dev structure.
220  * The OS initialization, configuring of the adapter private structure,
221  * and a hardware reset occur.
222  *
223  * Return: 0 on success, negative on failure
224  **/
225 static int fbnic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
226 {
227 	const struct fbnic_info *info = fbnic_info_tbl[ent->driver_data];
228 	struct net_device *netdev;
229 	struct fbnic_dev *fbd;
230 	int err;
231 
232 	if (pdev->error_state != pci_channel_io_normal) {
233 		dev_err(&pdev->dev,
234 			"PCI device still in an error state. Unable to load...\n");
235 		return -EIO;
236 	}
237 
238 	err = pcim_enable_device(pdev);
239 	if (err) {
240 		dev_err(&pdev->dev, "PCI enable device failed: %d\n", err);
241 		return err;
242 	}
243 
244 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(46));
245 	if (err)
246 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
247 	if (err) {
248 		dev_err(&pdev->dev, "DMA configuration failed: %d\n", err);
249 		return err;
250 	}
251 
252 	err = pcim_iomap_regions(pdev, info->bar_mask, fbnic_driver_name);
253 	if (err) {
254 		dev_err(&pdev->dev,
255 			"pci_request_selected_regions failed: %d\n", err);
256 		return err;
257 	}
258 
259 	fbd = fbnic_devlink_alloc(pdev);
260 	if (!fbd) {
261 		dev_err(&pdev->dev, "Devlink allocation failed\n");
262 		return -ENOMEM;
263 	}
264 
265 	/* Populate driver with hardware-specific info and handlers */
266 	fbd->max_num_queues = info->max_num_queues;
267 
268 	pci_set_master(pdev);
269 	pci_save_state(pdev);
270 
271 	INIT_DELAYED_WORK(&fbd->service_task, fbnic_service_task);
272 
273 	err = fbnic_alloc_irqs(fbd);
274 	if (err)
275 		goto free_fbd;
276 
277 	err = fbnic_mac_init(fbd);
278 	if (err) {
279 		dev_err(&pdev->dev, "Failed to initialize MAC: %d\n", err);
280 		goto free_irqs;
281 	}
282 
283 	err = fbnic_fw_enable_mbx(fbd);
284 	if (err) {
285 		dev_err(&pdev->dev,
286 			"Firmware mailbox initialization failure\n");
287 		goto free_irqs;
288 	}
289 
290 	fbnic_devlink_register(fbd);
291 
292 	if (!fbd->dsn) {
293 		dev_warn(&pdev->dev, "Reading serial number failed\n");
294 		goto init_failure_mode;
295 	}
296 
297 	netdev = fbnic_netdev_alloc(fbd);
298 	if (!netdev) {
299 		dev_err(&pdev->dev, "Netdev allocation failed\n");
300 		goto init_failure_mode;
301 	}
302 
303 	err = fbnic_netdev_register(netdev);
304 	if (err) {
305 		dev_err(&pdev->dev, "Netdev registration failed: %d\n", err);
306 		goto ifm_free_netdev;
307 	}
308 
309 	return 0;
310 
311 ifm_free_netdev:
312 	fbnic_netdev_free(fbd);
313 init_failure_mode:
314 	dev_warn(&pdev->dev, "Probe error encountered, entering init failure mode. Normal networking functionality will not be available.\n");
315 	 /* Always return 0 even on error so devlink is registered to allow
316 	  * firmware updates for fixes.
317 	  */
318 	return 0;
319 free_irqs:
320 	fbnic_free_irqs(fbd);
321 free_fbd:
322 	pci_disable_device(pdev);
323 	fbnic_devlink_free(fbd);
324 
325 	return err;
326 }
327 
328 /**
329  * fbnic_remove - Device Removal Routine
330  * @pdev: PCI device information struct
331  *
332  * Called by the PCI subsystem to alert the driver that it should release
333  * a PCI device.  The could be caused by a Hot-Plug event, or because the
334  * driver is going to be removed from memory.
335  **/
336 static void fbnic_remove(struct pci_dev *pdev)
337 {
338 	struct fbnic_dev *fbd = pci_get_drvdata(pdev);
339 
340 	if (!fbnic_init_failure(fbd)) {
341 		struct net_device *netdev = fbd->netdev;
342 
343 		fbnic_netdev_unregister(netdev);
344 		cancel_delayed_work_sync(&fbd->service_task);
345 		fbnic_netdev_free(fbd);
346 	}
347 
348 	fbnic_devlink_unregister(fbd);
349 	fbnic_fw_disable_mbx(fbd);
350 	fbnic_free_irqs(fbd);
351 
352 	pci_disable_device(pdev);
353 	fbnic_devlink_free(fbd);
354 }
355 
356 static int fbnic_pm_suspend(struct device *dev)
357 {
358 	struct fbnic_dev *fbd = dev_get_drvdata(dev);
359 	struct net_device *netdev = fbd->netdev;
360 
361 	if (fbnic_init_failure(fbd))
362 		goto null_uc_addr;
363 
364 	rtnl_lock();
365 
366 	netif_device_detach(netdev);
367 
368 	if (netif_running(netdev))
369 		netdev->netdev_ops->ndo_stop(netdev);
370 
371 	rtnl_unlock();
372 
373 null_uc_addr:
374 	fbnic_fw_disable_mbx(fbd);
375 
376 	/* Free the IRQs so they aren't trying to occupy sleeping CPUs */
377 	fbnic_free_irqs(fbd);
378 
379 	/* Hardware is about to go away, so switch off MMIO access internally */
380 	WRITE_ONCE(fbd->uc_addr0, NULL);
381 	WRITE_ONCE(fbd->uc_addr4, NULL);
382 
383 	return 0;
384 }
385 
386 static int __fbnic_pm_resume(struct device *dev)
387 {
388 	struct fbnic_dev *fbd = dev_get_drvdata(dev);
389 	struct net_device *netdev = fbd->netdev;
390 	void __iomem * const *iomap_table;
391 	struct fbnic_net *fbn;
392 	int err;
393 
394 	/* Restore MMIO access */
395 	iomap_table = pcim_iomap_table(to_pci_dev(dev));
396 	fbd->uc_addr0 = iomap_table[0];
397 	fbd->uc_addr4 = iomap_table[4];
398 
399 	/* Rerequest the IRQs */
400 	err = fbnic_alloc_irqs(fbd);
401 	if (err)
402 		goto err_invalidate_uc_addr;
403 
404 	fbd->mac->init_regs(fbd);
405 
406 	/* Re-enable mailbox */
407 	err = fbnic_fw_enable_mbx(fbd);
408 	if (err)
409 		goto err_free_irqs;
410 
411 	/* No netdev means there isn't a network interface to bring up */
412 	if (fbnic_init_failure(fbd))
413 		return 0;
414 
415 	fbn = netdev_priv(netdev);
416 
417 	/* Reset the queues if needed */
418 	fbnic_reset_queues(fbn, fbn->num_tx_queues, fbn->num_rx_queues);
419 
420 	rtnl_lock();
421 
422 	if (netif_running(netdev)) {
423 		err = __fbnic_open(fbn);
424 		if (err)
425 			goto err_disable_mbx;
426 	}
427 
428 	rtnl_unlock();
429 
430 	return 0;
431 err_disable_mbx:
432 	rtnl_unlock();
433 	fbnic_fw_disable_mbx(fbd);
434 err_free_irqs:
435 	fbnic_free_irqs(fbd);
436 err_invalidate_uc_addr:
437 	WRITE_ONCE(fbd->uc_addr0, NULL);
438 	WRITE_ONCE(fbd->uc_addr4, NULL);
439 	return err;
440 }
441 
442 static void __fbnic_pm_attach(struct device *dev)
443 {
444 	struct fbnic_dev *fbd = dev_get_drvdata(dev);
445 	struct net_device *netdev = fbd->netdev;
446 	struct fbnic_net *fbn;
447 
448 	if (fbnic_init_failure(fbd))
449 		return;
450 
451 	fbn = netdev_priv(netdev);
452 
453 	if (netif_running(netdev))
454 		fbnic_up(fbn);
455 
456 	netif_device_attach(netdev);
457 }
458 
459 static int __maybe_unused fbnic_pm_resume(struct device *dev)
460 {
461 	int err;
462 
463 	err = __fbnic_pm_resume(dev);
464 	if (!err)
465 		__fbnic_pm_attach(dev);
466 
467 	return err;
468 }
469 
470 static const struct dev_pm_ops fbnic_pm_ops = {
471 	SET_SYSTEM_SLEEP_PM_OPS(fbnic_pm_suspend, fbnic_pm_resume)
472 };
473 
474 static void fbnic_shutdown(struct pci_dev *pdev)
475 {
476 	fbnic_pm_suspend(&pdev->dev);
477 }
478 
479 static pci_ers_result_t fbnic_err_error_detected(struct pci_dev *pdev,
480 						 pci_channel_state_t state)
481 {
482 	/* Disconnect device if failure is not recoverable via reset */
483 	if (state == pci_channel_io_perm_failure)
484 		return PCI_ERS_RESULT_DISCONNECT;
485 
486 	fbnic_pm_suspend(&pdev->dev);
487 
488 	/* Request a slot reset */
489 	return PCI_ERS_RESULT_NEED_RESET;
490 }
491 
492 static pci_ers_result_t fbnic_err_slot_reset(struct pci_dev *pdev)
493 {
494 	int err;
495 
496 	pci_set_power_state(pdev, PCI_D0);
497 	pci_restore_state(pdev);
498 	pci_save_state(pdev);
499 
500 	if (pci_enable_device_mem(pdev)) {
501 		dev_err(&pdev->dev,
502 			"Cannot re-enable PCI device after reset.\n");
503 		return PCI_ERS_RESULT_DISCONNECT;
504 	}
505 
506 	/* Restore device to previous state */
507 	err = __fbnic_pm_resume(&pdev->dev);
508 
509 	return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
510 }
511 
512 static void fbnic_err_resume(struct pci_dev *pdev)
513 {
514 	__fbnic_pm_attach(&pdev->dev);
515 }
516 
517 static const struct pci_error_handlers fbnic_err_handler = {
518 	.error_detected	= fbnic_err_error_detected,
519 	.slot_reset	= fbnic_err_slot_reset,
520 	.resume		= fbnic_err_resume,
521 };
522 
523 static struct pci_driver fbnic_driver = {
524 	.name		= fbnic_driver_name,
525 	.id_table	= fbnic_pci_tbl,
526 	.probe		= fbnic_probe,
527 	.remove		= fbnic_remove,
528 	.driver.pm	= &fbnic_pm_ops,
529 	.shutdown	= fbnic_shutdown,
530 	.err_handler	= &fbnic_err_handler,
531 };
532 
533 /**
534  * fbnic_init_module - Driver Registration Routine
535  *
536  * The first routine called when the driver is loaded.  All it does is
537  * register with the PCI subsystem.
538  *
539  * Return: 0 on success, negative on failure
540  **/
541 static int __init fbnic_init_module(void)
542 {
543 	int err;
544 
545 	err = pci_register_driver(&fbnic_driver);
546 	if (err)
547 		goto out;
548 
549 	pr_info(DRV_SUMMARY " (%s)", fbnic_driver.name);
550 out:
551 	return err;
552 }
553 module_init(fbnic_init_module);
554 
555 /**
556  * fbnic_exit_module - Driver Exit Cleanup Routine
557  *
558  * Called just before the driver is removed from memory.
559  **/
560 static void __exit fbnic_exit_module(void)
561 {
562 	pci_unregister_driver(&fbnic_driver);
563 }
564 module_exit(fbnic_exit_module);
565