xref: /linux/drivers/pci/pcie/dpc.c (revision 801f5e1ac783df9fafff8899ef2d5511bd4dbdcb)
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 	struct pci_dev *pdev = dpc->dev->port;
72 
73 	pcie_wait_for_link(pdev, false);
74 }
75 
76 static pci_ers_result_t dpc_reset_link(struct pci_dev *pdev)
77 {
78 	struct dpc_dev *dpc;
79 	struct pcie_device *pciedev;
80 	struct device *devdpc;
81 	u16 cap, ctl;
82 
83 	/*
84 	 * DPC disables the Link automatically in hardware, so it has
85 	 * already been reset by the time we get here.
86 	 */
87 	devdpc = pcie_port_find_device(pdev, PCIE_PORT_SERVICE_DPC);
88 	pciedev = to_pcie_device(devdpc);
89 	dpc = get_service_data(pciedev);
90 	cap = dpc->cap_pos;
91 
92 	/*
93 	 * Wait until the Link is inactive, then clear DPC Trigger Status
94 	 * to allow the Port to leave DPC.
95 	 */
96 	dpc_wait_link_inactive(dpc);
97 
98 	if (dpc->rp_extensions && dpc_wait_rp_inactive(dpc))
99 		return PCI_ERS_RESULT_DISCONNECT;
100 	if (dpc->rp_extensions && dpc->rp_pio_status) {
101 		pci_write_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS,
102 				       dpc->rp_pio_status);
103 		dpc->rp_pio_status = 0;
104 	}
105 
106 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
107 			      PCI_EXP_DPC_STATUS_TRIGGER);
108 
109 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_CTL, &ctl);
110 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_CTL,
111 			      ctl | PCI_EXP_DPC_CTL_INT_EN);
112 
113 	return PCI_ERS_RESULT_RECOVERED;
114 }
115 
116 static void dpc_work(struct work_struct *work)
117 {
118 	struct dpc_dev *dpc = container_of(work, struct dpc_dev, work);
119 	struct pci_dev *pdev = dpc->dev->port;
120 
121 	/* We configure DPC so it only triggers on ERR_FATAL */
122 	pcie_do_fatal_recovery(pdev, PCIE_PORT_SERVICE_DPC);
123 }
124 
125 static void dpc_process_rp_pio_error(struct dpc_dev *dpc)
126 {
127 	struct device *dev = &dpc->dev->device;
128 	struct pci_dev *pdev = dpc->dev->port;
129 	u16 cap = dpc->cap_pos, dpc_status, first_error;
130 	u32 status, mask, sev, syserr, exc, dw0, dw1, dw2, dw3, log, prefix;
131 	int i;
132 
133 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, &status);
134 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_MASK, &mask);
135 	dev_err(dev, "rp_pio_status: %#010x, rp_pio_mask: %#010x\n",
136 		status, mask);
137 
138 	dpc->rp_pio_status = status;
139 
140 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SEVERITY, &sev);
141 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SYSERROR, &syserr);
142 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_EXCEPTION, &exc);
143 	dev_err(dev, "RP PIO severity=%#010x, syserror=%#010x, exception=%#010x\n",
144 		sev, syserr, exc);
145 
146 	/* Get First Error Pointer */
147 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &dpc_status);
148 	first_error = (dpc_status & 0x1f00) >> 8;
149 
150 	status &= ~mask;
151 	for (i = 0; i < ARRAY_SIZE(rp_pio_error_string); i++) {
152 		if (status & (1 << i))
153 			dev_err(dev, "[%2d] %s%s\n", i, rp_pio_error_string[i],
154 				first_error == i ? " (First)" : "");
155 	}
156 
157 	if (dpc->rp_log_size < 4)
158 		return;
159 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG,
160 			      &dw0);
161 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 4,
162 			      &dw1);
163 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 8,
164 			      &dw2);
165 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 12,
166 			      &dw3);
167 	dev_err(dev, "TLP Header: %#010x %#010x %#010x %#010x\n",
168 		dw0, dw1, dw2, dw3);
169 
170 	if (dpc->rp_log_size < 5)
171 		return;
172 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_IMPSPEC_LOG, &log);
173 	dev_err(dev, "RP PIO ImpSpec Log %#010x\n", log);
174 
175 	for (i = 0; i < dpc->rp_log_size - 5; i++) {
176 		pci_read_config_dword(pdev,
177 			cap + PCI_EXP_DPC_RP_PIO_TLPPREFIX_LOG, &prefix);
178 		dev_err(dev, "TLP Prefix Header: dw%d, %#010x\n", i, prefix);
179 	}
180 }
181 
182 static irqreturn_t dpc_irq(int irq, void *context)
183 {
184 	struct dpc_dev *dpc = (struct dpc_dev *)context;
185 	struct pci_dev *pdev = dpc->dev->port;
186 	struct device *dev = &dpc->dev->device;
187 	u16 cap = dpc->cap_pos, ctl, status, source, reason, ext_reason;
188 
189 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_CTL, &ctl);
190 
191 	if (!(ctl & PCI_EXP_DPC_CTL_INT_EN) || ctl == (u16)(~0))
192 		return IRQ_NONE;
193 
194 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
195 
196 	if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT))
197 		return IRQ_NONE;
198 
199 	if (!(status & PCI_EXP_DPC_STATUS_TRIGGER)) {
200 		pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
201 				      PCI_EXP_DPC_STATUS_INTERRUPT);
202 		return IRQ_HANDLED;
203 	}
204 
205 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_CTL,
206 			      ctl & ~PCI_EXP_DPC_CTL_INT_EN);
207 
208 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID,
209 			     &source);
210 
211 	dev_info(dev, "DPC containment event, status:%#06x source:%#06x\n",
212 		status, source);
213 
214 	reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN) >> 1;
215 	ext_reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT) >> 5;
216 
217 	dev_warn(dev, "DPC %s detected, remove downstream devices\n",
218 		 (reason == 0) ? "unmasked uncorrectable error" :
219 		 (reason == 1) ? "ERR_NONFATAL" :
220 		 (reason == 2) ? "ERR_FATAL" :
221 		 (ext_reason == 0) ? "RP PIO error" :
222 		 (ext_reason == 1) ? "software trigger" :
223 				     "reserved error");
224 	/* show RP PIO error detail information */
225 	if (dpc->rp_extensions && reason == 3 && ext_reason == 0)
226 		dpc_process_rp_pio_error(dpc);
227 
228 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
229 			      PCI_EXP_DPC_STATUS_INTERRUPT);
230 
231 	schedule_work(&dpc->work);
232 
233 	return IRQ_HANDLED;
234 }
235 
236 #define FLAG(x, y) (((x) & (y)) ? '+' : '-')
237 static int dpc_probe(struct pcie_device *dev)
238 {
239 	struct dpc_dev *dpc;
240 	struct pci_dev *pdev = dev->port;
241 	struct device *device = &dev->device;
242 	int status;
243 	u16 ctl, cap;
244 
245 	if (pcie_aer_get_firmware_first(pdev))
246 		return -ENOTSUPP;
247 
248 	dpc = devm_kzalloc(device, sizeof(*dpc), GFP_KERNEL);
249 	if (!dpc)
250 		return -ENOMEM;
251 
252 	dpc->cap_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC);
253 	dpc->dev = dev;
254 	INIT_WORK(&dpc->work, dpc_work);
255 	set_service_data(dev, dpc);
256 
257 	status = devm_request_irq(device, dev->irq, dpc_irq, IRQF_SHARED,
258 				  "pcie-dpc", dpc);
259 	if (status) {
260 		dev_warn(device, "request IRQ%d failed: %d\n", dev->irq,
261 			 status);
262 		return status;
263 	}
264 
265 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CAP, &cap);
266 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
267 
268 	dpc->rp_extensions = (cap & PCI_EXP_DPC_CAP_RP_EXT);
269 	if (dpc->rp_extensions) {
270 		dpc->rp_log_size = (cap & PCI_EXP_DPC_RP_PIO_LOG_SIZE) >> 8;
271 		if (dpc->rp_log_size < 4 || dpc->rp_log_size > 9) {
272 			dev_err(device, "RP PIO log size %u is invalid\n",
273 				dpc->rp_log_size);
274 			dpc->rp_log_size = 0;
275 		}
276 	}
277 
278 	ctl = (ctl & 0xfff4) | PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN;
279 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
280 
281 	dev_info(device, "DPC error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n",
282 		cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT),
283 		FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP),
284 		FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), dpc->rp_log_size,
285 		FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE));
286 	return status;
287 }
288 
289 static void dpc_remove(struct pcie_device *dev)
290 {
291 	struct dpc_dev *dpc = get_service_data(dev);
292 	struct pci_dev *pdev = dev->port;
293 	u16 ctl;
294 
295 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
296 	ctl &= ~(PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN);
297 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
298 }
299 
300 static struct pcie_port_service_driver dpcdriver = {
301 	.name		= "dpc",
302 	.port_type	= PCIE_ANY_PORT,
303 	.service	= PCIE_PORT_SERVICE_DPC,
304 	.probe		= dpc_probe,
305 	.remove		= dpc_remove,
306 	.reset_link	= dpc_reset_link,
307 };
308 
309 static int __init dpc_service_init(void)
310 {
311 	return pcie_port_service_register(&dpcdriver);
312 }
313 device_initcall(dpc_service_init);
314