xref: /linux/drivers/pci/pcie/dpc.c (revision a8b70ccf10e38775785d9cb12ead916474549f99)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * PCI Express Downstream Port Containment services driver
4  * Author: Keith Busch <keith.busch@intel.com>
5  *
6  * Copyright (C) 2016 Intel Corp.
7  */
8 
9 #include <linux/delay.h>
10 #include <linux/interrupt.h>
11 #include <linux/init.h>
12 #include <linux/pci.h>
13 
14 #include "portdrv.h"
15 #include "../pci.h"
16 #include "aer/aerdrv.h"
17 
18 struct dpc_dev {
19 	struct pcie_device	*dev;
20 	struct work_struct	work;
21 	u16			cap_pos;
22 	bool			rp_extensions;
23 	u32			rp_pio_status;
24 	u8			rp_log_size;
25 };
26 
27 static const char * const rp_pio_error_string[] = {
28 	"Configuration Request received UR Completion",	 /* Bit Position 0  */
29 	"Configuration Request received CA Completion",	 /* Bit Position 1  */
30 	"Configuration Request Completion Timeout",	 /* Bit Position 2  */
31 	NULL,
32 	NULL,
33 	NULL,
34 	NULL,
35 	NULL,
36 	"I/O Request received UR Completion",		 /* Bit Position 8  */
37 	"I/O Request received CA Completion",		 /* Bit Position 9  */
38 	"I/O Request Completion Timeout",		 /* Bit Position 10 */
39 	NULL,
40 	NULL,
41 	NULL,
42 	NULL,
43 	NULL,
44 	"Memory Request received UR Completion",	 /* Bit Position 16 */
45 	"Memory Request received CA Completion",	 /* Bit Position 17 */
46 	"Memory Request Completion Timeout",		 /* Bit Position 18 */
47 };
48 
49 static int dpc_wait_rp_inactive(struct dpc_dev *dpc)
50 {
51 	unsigned long timeout = jiffies + HZ;
52 	struct pci_dev *pdev = dpc->dev->port;
53 	struct device *dev = &dpc->dev->device;
54 	u16 cap = dpc->cap_pos, status;
55 
56 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
57 	while (status & PCI_EXP_DPC_RP_BUSY &&
58 					!time_after(jiffies, timeout)) {
59 		msleep(10);
60 		pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
61 	}
62 	if (status & PCI_EXP_DPC_RP_BUSY) {
63 		dev_warn(dev, "DPC root port still busy\n");
64 		return -EBUSY;
65 	}
66 	return 0;
67 }
68 
69 static void dpc_wait_link_inactive(struct dpc_dev *dpc)
70 {
71 	unsigned long timeout = jiffies + HZ;
72 	struct pci_dev *pdev = dpc->dev->port;
73 	struct device *dev = &dpc->dev->device;
74 	u16 lnk_status;
75 
76 	pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
77 	while (lnk_status & PCI_EXP_LNKSTA_DLLLA &&
78 					!time_after(jiffies, timeout)) {
79 		msleep(10);
80 		pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
81 	}
82 	if (lnk_status & PCI_EXP_LNKSTA_DLLLA)
83 		dev_warn(dev, "Link state not disabled for DPC event\n");
84 }
85 
86 static void dpc_work(struct work_struct *work)
87 {
88 	struct dpc_dev *dpc = container_of(work, struct dpc_dev, work);
89 	struct pci_dev *dev, *temp, *pdev = dpc->dev->port;
90 	struct pci_bus *parent = pdev->subordinate;
91 	u16 cap = dpc->cap_pos, ctl;
92 
93 	pci_lock_rescan_remove();
94 	list_for_each_entry_safe_reverse(dev, temp, &parent->devices,
95 					 bus_list) {
96 		pci_dev_get(dev);
97 		pci_dev_set_disconnected(dev, NULL);
98 		if (pci_has_subordinate(dev))
99 			pci_walk_bus(dev->subordinate,
100 				     pci_dev_set_disconnected, NULL);
101 		pci_stop_and_remove_bus_device(dev);
102 		pci_dev_put(dev);
103 	}
104 	pci_unlock_rescan_remove();
105 
106 	dpc_wait_link_inactive(dpc);
107 	if (dpc->rp_extensions && dpc_wait_rp_inactive(dpc))
108 		return;
109 	if (dpc->rp_extensions && dpc->rp_pio_status) {
110 		pci_write_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS,
111 				       dpc->rp_pio_status);
112 		dpc->rp_pio_status = 0;
113 	}
114 
115 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
116 		PCI_EXP_DPC_STATUS_TRIGGER | PCI_EXP_DPC_STATUS_INTERRUPT);
117 
118 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_CTL, &ctl);
119 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_CTL,
120 			      ctl | PCI_EXP_DPC_CTL_INT_EN);
121 }
122 
123 static void dpc_process_rp_pio_error(struct dpc_dev *dpc)
124 {
125 	struct device *dev = &dpc->dev->device;
126 	struct pci_dev *pdev = dpc->dev->port;
127 	u16 cap = dpc->cap_pos, dpc_status, first_error;
128 	u32 status, mask, sev, syserr, exc, dw0, dw1, dw2, dw3, log, prefix;
129 	int i;
130 
131 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, &status);
132 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_MASK, &mask);
133 	dev_err(dev, "rp_pio_status: %#010x, rp_pio_mask: %#010x\n",
134 		status, mask);
135 
136 	dpc->rp_pio_status = status;
137 
138 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SEVERITY, &sev);
139 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SYSERROR, &syserr);
140 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_EXCEPTION, &exc);
141 	dev_err(dev, "RP PIO severity=%#010x, syserror=%#010x, exception=%#010x\n",
142 		sev, syserr, exc);
143 
144 	/* Get First Error Pointer */
145 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &dpc_status);
146 	first_error = (dpc_status & 0x1f00) >> 8;
147 
148 	status &= ~mask;
149 	for (i = 0; i < ARRAY_SIZE(rp_pio_error_string); i++) {
150 		if (status & (1 << i))
151 			dev_err(dev, "[%2d] %s%s\n", i, rp_pio_error_string[i],
152 				first_error == i ? " (First)" : "");
153 	}
154 
155 	if (dpc->rp_log_size < 4)
156 		return;
157 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG,
158 			      &dw0);
159 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 4,
160 			      &dw1);
161 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 8,
162 			      &dw2);
163 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 12,
164 			      &dw3);
165 	dev_err(dev, "TLP Header: %#010x %#010x %#010x %#010x\n",
166 		dw0, dw1, dw2, dw3);
167 
168 	if (dpc->rp_log_size < 5)
169 		return;
170 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_IMPSPEC_LOG, &log);
171 	dev_err(dev, "RP PIO ImpSpec Log %#010x\n", log);
172 
173 	for (i = 0; i < dpc->rp_log_size - 5; i++) {
174 		pci_read_config_dword(pdev,
175 			cap + PCI_EXP_DPC_RP_PIO_TLPPREFIX_LOG, &prefix);
176 		dev_err(dev, "TLP Prefix Header: dw%d, %#010x\n", i, prefix);
177 	}
178 }
179 
180 static irqreturn_t dpc_irq(int irq, void *context)
181 {
182 	struct dpc_dev *dpc = (struct dpc_dev *)context;
183 	struct pci_dev *pdev = dpc->dev->port;
184 	struct device *dev = &dpc->dev->device;
185 	u16 cap = dpc->cap_pos, ctl, status, source, reason, ext_reason;
186 
187 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_CTL, &ctl);
188 
189 	if (!(ctl & PCI_EXP_DPC_CTL_INT_EN) || ctl == (u16)(~0))
190 		return IRQ_NONE;
191 
192 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
193 
194 	if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT))
195 		return IRQ_NONE;
196 
197 	if (!(status & PCI_EXP_DPC_STATUS_TRIGGER)) {
198 		pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
199 				      PCI_EXP_DPC_STATUS_INTERRUPT);
200 		return IRQ_HANDLED;
201 	}
202 
203 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_CTL,
204 			      ctl & ~PCI_EXP_DPC_CTL_INT_EN);
205 
206 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID,
207 			     &source);
208 
209 	dev_info(dev, "DPC containment event, status:%#06x source:%#06x\n",
210 		status, source);
211 
212 	reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN) >> 1;
213 	ext_reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT) >> 5;
214 
215 	dev_warn(dev, "DPC %s detected, remove downstream devices\n",
216 		 (reason == 0) ? "unmasked uncorrectable error" :
217 		 (reason == 1) ? "ERR_NONFATAL" :
218 		 (reason == 2) ? "ERR_FATAL" :
219 		 (ext_reason == 0) ? "RP PIO error" :
220 		 (ext_reason == 1) ? "software trigger" :
221 				     "reserved error");
222 	/* show RP PIO error detail information */
223 	if (dpc->rp_extensions && reason == 3 && ext_reason == 0)
224 		dpc_process_rp_pio_error(dpc);
225 
226 	schedule_work(&dpc->work);
227 
228 	return IRQ_HANDLED;
229 }
230 
231 #define FLAG(x, y) (((x) & (y)) ? '+' : '-')
232 static int dpc_probe(struct pcie_device *dev)
233 {
234 	struct dpc_dev *dpc;
235 	struct pci_dev *pdev = dev->port;
236 	struct device *device = &dev->device;
237 	int status;
238 	u16 ctl, cap;
239 
240 	if (pcie_aer_get_firmware_first(pdev))
241 		return -ENOTSUPP;
242 
243 	dpc = devm_kzalloc(device, sizeof(*dpc), GFP_KERNEL);
244 	if (!dpc)
245 		return -ENOMEM;
246 
247 	dpc->cap_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC);
248 	dpc->dev = dev;
249 	INIT_WORK(&dpc->work, dpc_work);
250 	set_service_data(dev, dpc);
251 
252 	status = devm_request_irq(device, dev->irq, dpc_irq, IRQF_SHARED,
253 				  "pcie-dpc", dpc);
254 	if (status) {
255 		dev_warn(device, "request IRQ%d failed: %d\n", dev->irq,
256 			 status);
257 		return status;
258 	}
259 
260 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CAP, &cap);
261 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
262 
263 	dpc->rp_extensions = (cap & PCI_EXP_DPC_CAP_RP_EXT);
264 	if (dpc->rp_extensions) {
265 		dpc->rp_log_size = (cap & PCI_EXP_DPC_RP_PIO_LOG_SIZE) >> 8;
266 		if (dpc->rp_log_size < 4 || dpc->rp_log_size > 9) {
267 			dev_err(device, "RP PIO log size %u is invalid\n",
268 				dpc->rp_log_size);
269 			dpc->rp_log_size = 0;
270 		}
271 	}
272 
273 	ctl = (ctl & 0xfff4) | PCI_EXP_DPC_CTL_EN_NONFATAL | PCI_EXP_DPC_CTL_INT_EN;
274 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
275 
276 	dev_info(device, "DPC error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n",
277 		cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT),
278 		FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP),
279 		FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), dpc->rp_log_size,
280 		FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE));
281 	return status;
282 }
283 
284 static void dpc_remove(struct pcie_device *dev)
285 {
286 	struct dpc_dev *dpc = get_service_data(dev);
287 	struct pci_dev *pdev = dev->port;
288 	u16 ctl;
289 
290 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
291 	ctl &= ~(PCI_EXP_DPC_CTL_EN_NONFATAL | PCI_EXP_DPC_CTL_INT_EN);
292 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
293 }
294 
295 static struct pcie_port_service_driver dpcdriver = {
296 	.name		= "dpc",
297 	.port_type	= PCIE_ANY_PORT,
298 	.service	= PCIE_PORT_SERVICE_DPC,
299 	.probe		= dpc_probe,
300 	.remove		= dpc_remove,
301 };
302 
303 static int __init dpc_service_init(void)
304 {
305 	return pcie_port_service_register(&dpcdriver);
306 }
307 device_initcall(dpc_service_init);
308