1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * This file implements the error recovery as a core part of PCIe error
4  * reporting. When a PCIe error is delivered, an error message will be
5  * collected and printed to console, then, an error recovery procedure
6  * will be executed by following the PCI error recovery rules.
7  *
8  * Copyright (C) 2006 Intel Corp.
9  *	Tom Long Nguyen (tom.l.nguyen@intel.com)
10  *	Zhang Yanmin (yanmin.zhang@intel.com)
11  */
12 
13 #define dev_fmt(fmt) "AER: " fmt
14 
15 #include <linux/pci.h>
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/aer.h>
20 #include "portdrv.h"
21 #include "../pci.h"
22 
merge_result(enum pci_ers_result orig,enum pci_ers_result new)23 static pci_ers_result_t merge_result(enum pci_ers_result orig,
24 				  enum pci_ers_result new)
25 {
26 	if (new == PCI_ERS_RESULT_NO_AER_DRIVER)
27 		return PCI_ERS_RESULT_NO_AER_DRIVER;
28 
29 	if (new == PCI_ERS_RESULT_NONE)
30 		return orig;
31 
32 	switch (orig) {
33 	case PCI_ERS_RESULT_CAN_RECOVER:
34 	case PCI_ERS_RESULT_RECOVERED:
35 		orig = new;
36 		break;
37 	case PCI_ERS_RESULT_DISCONNECT:
38 		if (new == PCI_ERS_RESULT_NEED_RESET)
39 			orig = PCI_ERS_RESULT_NEED_RESET;
40 		break;
41 	default:
42 		break;
43 	}
44 
45 	return orig;
46 }
47 
report_error_detected(struct pci_dev * dev,pci_channel_state_t state,enum pci_ers_result * result)48 static int report_error_detected(struct pci_dev *dev,
49 				 pci_channel_state_t state,
50 				 enum pci_ers_result *result)
51 {
52 	struct pci_driver *pdrv;
53 	pci_ers_result_t vote;
54 	const struct pci_error_handlers *err_handler;
55 
56 	device_lock(&dev->dev);
57 	pdrv = dev->driver;
58 	if (!pci_dev_set_io_state(dev, state) ||
59 		!pdrv ||
60 		!pdrv->err_handler ||
61 		!pdrv->err_handler->error_detected) {
62 		/*
63 		 * If any device in the subtree does not have an error_detected
64 		 * callback, PCI_ERS_RESULT_NO_AER_DRIVER prevents subsequent
65 		 * error callbacks of "any" device in the subtree, and will
66 		 * exit in the disconnected error state.
67 		 */
68 		if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) {
69 			vote = PCI_ERS_RESULT_NO_AER_DRIVER;
70 			pci_info(dev, "can't recover (no error_detected callback)\n");
71 		} else {
72 			vote = PCI_ERS_RESULT_NONE;
73 		}
74 	} else {
75 		err_handler = pdrv->err_handler;
76 		vote = err_handler->error_detected(dev, state);
77 	}
78 	pci_uevent_ers(dev, vote);
79 	*result = merge_result(*result, vote);
80 	device_unlock(&dev->dev);
81 	return 0;
82 }
83 
report_frozen_detected(struct pci_dev * dev,void * data)84 static int report_frozen_detected(struct pci_dev *dev, void *data)
85 {
86 	return report_error_detected(dev, pci_channel_io_frozen, data);
87 }
88 
report_normal_detected(struct pci_dev * dev,void * data)89 static int report_normal_detected(struct pci_dev *dev, void *data)
90 {
91 	return report_error_detected(dev, pci_channel_io_normal, data);
92 }
93 
report_mmio_enabled(struct pci_dev * dev,void * data)94 static int report_mmio_enabled(struct pci_dev *dev, void *data)
95 {
96 	struct pci_driver *pdrv;
97 	pci_ers_result_t vote, *result = data;
98 	const struct pci_error_handlers *err_handler;
99 
100 	device_lock(&dev->dev);
101 	pdrv = dev->driver;
102 	if (!pdrv ||
103 		!pdrv->err_handler ||
104 		!pdrv->err_handler->mmio_enabled)
105 		goto out;
106 
107 	err_handler = pdrv->err_handler;
108 	vote = err_handler->mmio_enabled(dev);
109 	*result = merge_result(*result, vote);
110 out:
111 	device_unlock(&dev->dev);
112 	return 0;
113 }
114 
report_slot_reset(struct pci_dev * dev,void * data)115 static int report_slot_reset(struct pci_dev *dev, void *data)
116 {
117 	struct pci_driver *pdrv;
118 	pci_ers_result_t vote, *result = data;
119 	const struct pci_error_handlers *err_handler;
120 
121 	device_lock(&dev->dev);
122 	pdrv = dev->driver;
123 	if (!pdrv ||
124 		!pdrv->err_handler ||
125 		!pdrv->err_handler->slot_reset)
126 		goto out;
127 
128 	err_handler = pdrv->err_handler;
129 	vote = err_handler->slot_reset(dev);
130 	*result = merge_result(*result, vote);
131 out:
132 	device_unlock(&dev->dev);
133 	return 0;
134 }
135 
report_resume(struct pci_dev * dev,void * data)136 static int report_resume(struct pci_dev *dev, void *data)
137 {
138 	struct pci_driver *pdrv;
139 	const struct pci_error_handlers *err_handler;
140 
141 	device_lock(&dev->dev);
142 	pdrv = dev->driver;
143 	if (!pci_dev_set_io_state(dev, pci_channel_io_normal) ||
144 		!pdrv ||
145 		!pdrv->err_handler ||
146 		!pdrv->err_handler->resume)
147 		goto out;
148 
149 	err_handler = pdrv->err_handler;
150 	err_handler->resume(dev);
151 out:
152 	pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED);
153 	device_unlock(&dev->dev);
154 	return 0;
155 }
156 
157 /**
158  * pci_walk_bridge - walk bridges potentially AER affected
159  * @bridge:	bridge which may be a Port, an RCEC, or an RCiEP
160  * @cb:		callback to be called for each device found
161  * @userdata:	arbitrary pointer to be passed to callback
162  *
163  * If the device provided is a bridge, walk the subordinate bus, including
164  * any bridged devices on buses under this bus.  Call the provided callback
165  * on each device found.
166  *
167  * If the device provided has no subordinate bus, e.g., an RCEC or RCiEP,
168  * call the callback on the device itself.
169  */
pci_walk_bridge(struct pci_dev * bridge,int (* cb)(struct pci_dev *,void *),void * userdata)170 static void pci_walk_bridge(struct pci_dev *bridge,
171 			    int (*cb)(struct pci_dev *, void *),
172 			    void *userdata)
173 {
174 	if (bridge->subordinate)
175 		pci_walk_bus(bridge->subordinate, cb, userdata);
176 	else
177 		cb(bridge, userdata);
178 }
179 
pcie_do_recovery(struct pci_dev * dev,pci_channel_state_t state,pci_ers_result_t (* reset_subordinates)(struct pci_dev * pdev))180 pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
181 		pci_channel_state_t state,
182 		pci_ers_result_t (*reset_subordinates)(struct pci_dev *pdev))
183 {
184 	int type = pci_pcie_type(dev);
185 	struct pci_dev *bridge;
186 	pci_ers_result_t status = PCI_ERS_RESULT_CAN_RECOVER;
187 	struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
188 
189 	/*
190 	 * If the error was detected by a Root Port, Downstream Port, RCEC,
191 	 * or RCiEP, recovery runs on the device itself.  For Ports, that
192 	 * also includes any subordinate devices.
193 	 *
194 	 * If it was detected by another device (Endpoint, etc), recovery
195 	 * runs on the device and anything else under the same Port, i.e.,
196 	 * everything under "bridge".
197 	 */
198 	if (type == PCI_EXP_TYPE_ROOT_PORT ||
199 	    type == PCI_EXP_TYPE_DOWNSTREAM ||
200 	    type == PCI_EXP_TYPE_RC_EC ||
201 	    type == PCI_EXP_TYPE_RC_END)
202 		bridge = dev;
203 	else
204 		bridge = pci_upstream_bridge(dev);
205 
206 	pci_dbg(bridge, "broadcast error_detected message\n");
207 	if (state == pci_channel_io_frozen) {
208 		pci_walk_bridge(bridge, report_frozen_detected, &status);
209 		if (reset_subordinates(bridge) != PCI_ERS_RESULT_RECOVERED) {
210 			pci_warn(bridge, "subordinate device reset failed\n");
211 			goto failed;
212 		}
213 	} else {
214 		pci_walk_bridge(bridge, report_normal_detected, &status);
215 	}
216 
217 	if (status == PCI_ERS_RESULT_CAN_RECOVER) {
218 		status = PCI_ERS_RESULT_RECOVERED;
219 		pci_dbg(bridge, "broadcast mmio_enabled message\n");
220 		pci_walk_bridge(bridge, report_mmio_enabled, &status);
221 	}
222 
223 	if (status == PCI_ERS_RESULT_NEED_RESET) {
224 		/*
225 		 * TODO: Should call platform-specific
226 		 * functions to reset slot before calling
227 		 * drivers' slot_reset callbacks?
228 		 */
229 		status = PCI_ERS_RESULT_RECOVERED;
230 		pci_dbg(bridge, "broadcast slot_reset message\n");
231 		pci_walk_bridge(bridge, report_slot_reset, &status);
232 	}
233 
234 	if (status != PCI_ERS_RESULT_RECOVERED)
235 		goto failed;
236 
237 	pci_dbg(bridge, "broadcast resume message\n");
238 	pci_walk_bridge(bridge, report_resume, &status);
239 
240 	/*
241 	 * If we have native control of AER, clear error status in the device
242 	 * that detected the error.  If the platform retained control of AER,
243 	 * it is responsible for clearing this status.  In that case, the
244 	 * signaling device may not even be visible to the OS.
245 	 */
246 	if (host->native_aer || pcie_ports_native) {
247 		pcie_clear_device_status(dev);
248 		pci_aer_clear_nonfatal_status(dev);
249 	}
250 	pci_info(bridge, "device recovery successful\n");
251 	return status;
252 
253 failed:
254 	pci_uevent_ers(bridge, PCI_ERS_RESULT_DISCONNECT);
255 
256 	/* TODO: Should kernel panic here? */
257 	pci_info(bridge, "device recovery failed\n");
258 
259 	return status;
260 }
261