xref: /openbmc/linux/drivers/pci/pcie/dpc.c (revision ba61bb17)
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 
17 struct dpc_dev {
18 	struct pcie_device	*dev;
19 	struct work_struct	work;
20 	u16			cap_pos;
21 	bool			rp_extensions;
22 	u32			rp_pio_status;
23 	u8			rp_log_size;
24 };
25 
26 static const char * const rp_pio_error_string[] = {
27 	"Configuration Request received UR Completion",	 /* Bit Position 0  */
28 	"Configuration Request received CA Completion",	 /* Bit Position 1  */
29 	"Configuration Request Completion Timeout",	 /* Bit Position 2  */
30 	NULL,
31 	NULL,
32 	NULL,
33 	NULL,
34 	NULL,
35 	"I/O Request received UR Completion",		 /* Bit Position 8  */
36 	"I/O Request received CA Completion",		 /* Bit Position 9  */
37 	"I/O Request Completion Timeout",		 /* Bit Position 10 */
38 	NULL,
39 	NULL,
40 	NULL,
41 	NULL,
42 	NULL,
43 	"Memory Request received UR Completion",	 /* Bit Position 16 */
44 	"Memory Request received CA Completion",	 /* Bit Position 17 */
45 	"Memory Request Completion Timeout",		 /* Bit Position 18 */
46 };
47 
48 static int dpc_wait_rp_inactive(struct dpc_dev *dpc)
49 {
50 	unsigned long timeout = jiffies + HZ;
51 	struct pci_dev *pdev = dpc->dev->port;
52 	struct device *dev = &dpc->dev->device;
53 	u16 cap = dpc->cap_pos, status;
54 
55 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
56 	while (status & PCI_EXP_DPC_RP_BUSY &&
57 					!time_after(jiffies, timeout)) {
58 		msleep(10);
59 		pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
60 	}
61 	if (status & PCI_EXP_DPC_RP_BUSY) {
62 		dev_warn(dev, "DPC root port still busy\n");
63 		return -EBUSY;
64 	}
65 	return 0;
66 }
67 
68 static void dpc_wait_link_inactive(struct dpc_dev *dpc)
69 {
70 	struct pci_dev *pdev = dpc->dev->port;
71 
72 	pcie_wait_for_link(pdev, false);
73 }
74 
75 static pci_ers_result_t dpc_reset_link(struct pci_dev *pdev)
76 {
77 	struct dpc_dev *dpc;
78 	struct pcie_device *pciedev;
79 	struct device *devdpc;
80 	u16 cap, ctl;
81 
82 	/*
83 	 * DPC disables the Link automatically in hardware, so it has
84 	 * already been reset by the time we get here.
85 	 */
86 	devdpc = pcie_port_find_device(pdev, PCIE_PORT_SERVICE_DPC);
87 	pciedev = to_pcie_device(devdpc);
88 	dpc = get_service_data(pciedev);
89 	cap = dpc->cap_pos;
90 
91 	/*
92 	 * Wait until the Link is inactive, then clear DPC Trigger Status
93 	 * to allow the Port to leave DPC.
94 	 */
95 	dpc_wait_link_inactive(dpc);
96 
97 	if (dpc->rp_extensions && dpc_wait_rp_inactive(dpc))
98 		return PCI_ERS_RESULT_DISCONNECT;
99 	if (dpc->rp_extensions && dpc->rp_pio_status) {
100 		pci_write_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS,
101 				       dpc->rp_pio_status);
102 		dpc->rp_pio_status = 0;
103 	}
104 
105 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
106 			      PCI_EXP_DPC_STATUS_TRIGGER);
107 
108 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_CTL, &ctl);
109 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_CTL,
110 			      ctl | PCI_EXP_DPC_CTL_INT_EN);
111 
112 	return PCI_ERS_RESULT_RECOVERED;
113 }
114 
115 static void dpc_work(struct work_struct *work)
116 {
117 	struct dpc_dev *dpc = container_of(work, struct dpc_dev, work);
118 	struct pci_dev *pdev = dpc->dev->port;
119 
120 	/* We configure DPC so it only triggers on ERR_FATAL */
121 	pcie_do_fatal_recovery(pdev, PCIE_PORT_SERVICE_DPC);
122 }
123 
124 static void dpc_process_rp_pio_error(struct dpc_dev *dpc)
125 {
126 	struct device *dev = &dpc->dev->device;
127 	struct pci_dev *pdev = dpc->dev->port;
128 	u16 cap = dpc->cap_pos, dpc_status, first_error;
129 	u32 status, mask, sev, syserr, exc, dw0, dw1, dw2, dw3, log, prefix;
130 	int i;
131 
132 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, &status);
133 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_MASK, &mask);
134 	dev_err(dev, "rp_pio_status: %#010x, rp_pio_mask: %#010x\n",
135 		status, mask);
136 
137 	dpc->rp_pio_status = status;
138 
139 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SEVERITY, &sev);
140 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SYSERROR, &syserr);
141 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_EXCEPTION, &exc);
142 	dev_err(dev, "RP PIO severity=%#010x, syserror=%#010x, exception=%#010x\n",
143 		sev, syserr, exc);
144 
145 	/* Get First Error Pointer */
146 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &dpc_status);
147 	first_error = (dpc_status & 0x1f00) >> 8;
148 
149 	status &= ~mask;
150 	for (i = 0; i < ARRAY_SIZE(rp_pio_error_string); i++) {
151 		if (status & (1 << i))
152 			dev_err(dev, "[%2d] %s%s\n", i, rp_pio_error_string[i],
153 				first_error == i ? " (First)" : "");
154 	}
155 
156 	if (dpc->rp_log_size < 4)
157 		return;
158 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG,
159 			      &dw0);
160 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 4,
161 			      &dw1);
162 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 8,
163 			      &dw2);
164 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 12,
165 			      &dw3);
166 	dev_err(dev, "TLP Header: %#010x %#010x %#010x %#010x\n",
167 		dw0, dw1, dw2, dw3);
168 
169 	if (dpc->rp_log_size < 5)
170 		return;
171 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_IMPSPEC_LOG, &log);
172 	dev_err(dev, "RP PIO ImpSpec Log %#010x\n", log);
173 
174 	for (i = 0; i < dpc->rp_log_size - 5; i++) {
175 		pci_read_config_dword(pdev,
176 			cap + PCI_EXP_DPC_RP_PIO_TLPPREFIX_LOG, &prefix);
177 		dev_err(dev, "TLP Prefix Header: dw%d, %#010x\n", i, prefix);
178 	}
179 }
180 
181 static irqreturn_t dpc_irq(int irq, void *context)
182 {
183 	struct dpc_dev *dpc = (struct dpc_dev *)context;
184 	struct pci_dev *pdev = dpc->dev->port;
185 	struct device *dev = &dpc->dev->device;
186 	u16 cap = dpc->cap_pos, ctl, status, source, reason, ext_reason;
187 
188 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_CTL, &ctl);
189 
190 	if (!(ctl & PCI_EXP_DPC_CTL_INT_EN) || ctl == (u16)(~0))
191 		return IRQ_NONE;
192 
193 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
194 
195 	if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT))
196 		return IRQ_NONE;
197 
198 	if (!(status & PCI_EXP_DPC_STATUS_TRIGGER)) {
199 		pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
200 				      PCI_EXP_DPC_STATUS_INTERRUPT);
201 		return IRQ_HANDLED;
202 	}
203 
204 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_CTL,
205 			      ctl & ~PCI_EXP_DPC_CTL_INT_EN);
206 
207 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID,
208 			     &source);
209 
210 	dev_info(dev, "DPC containment event, status:%#06x source:%#06x\n",
211 		status, source);
212 
213 	reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN) >> 1;
214 	ext_reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT) >> 5;
215 
216 	dev_warn(dev, "DPC %s detected, remove downstream devices\n",
217 		 (reason == 0) ? "unmasked uncorrectable error" :
218 		 (reason == 1) ? "ERR_NONFATAL" :
219 		 (reason == 2) ? "ERR_FATAL" :
220 		 (ext_reason == 0) ? "RP PIO error" :
221 		 (ext_reason == 1) ? "software trigger" :
222 				     "reserved error");
223 	/* show RP PIO error detail information */
224 	if (dpc->rp_extensions && reason == 3 && ext_reason == 0)
225 		dpc_process_rp_pio_error(dpc);
226 
227 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
228 			      PCI_EXP_DPC_STATUS_INTERRUPT);
229 
230 	schedule_work(&dpc->work);
231 
232 	return IRQ_HANDLED;
233 }
234 
235 #define FLAG(x, y) (((x) & (y)) ? '+' : '-')
236 static int dpc_probe(struct pcie_device *dev)
237 {
238 	struct dpc_dev *dpc;
239 	struct pci_dev *pdev = dev->port;
240 	struct device *device = &dev->device;
241 	int status;
242 	u16 ctl, cap;
243 
244 	if (pcie_aer_get_firmware_first(pdev))
245 		return -ENOTSUPP;
246 
247 	dpc = devm_kzalloc(device, sizeof(*dpc), GFP_KERNEL);
248 	if (!dpc)
249 		return -ENOMEM;
250 
251 	dpc->cap_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC);
252 	dpc->dev = dev;
253 	INIT_WORK(&dpc->work, dpc_work);
254 	set_service_data(dev, dpc);
255 
256 	status = devm_request_irq(device, dev->irq, dpc_irq, IRQF_SHARED,
257 				  "pcie-dpc", dpc);
258 	if (status) {
259 		dev_warn(device, "request IRQ%d failed: %d\n", dev->irq,
260 			 status);
261 		return status;
262 	}
263 
264 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CAP, &cap);
265 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
266 
267 	dpc->rp_extensions = (cap & PCI_EXP_DPC_CAP_RP_EXT);
268 	if (dpc->rp_extensions) {
269 		dpc->rp_log_size = (cap & PCI_EXP_DPC_RP_PIO_LOG_SIZE) >> 8;
270 		if (dpc->rp_log_size < 4 || dpc->rp_log_size > 9) {
271 			dev_err(device, "RP PIO log size %u is invalid\n",
272 				dpc->rp_log_size);
273 			dpc->rp_log_size = 0;
274 		}
275 	}
276 
277 	ctl = (ctl & 0xfff4) | PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN;
278 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
279 
280 	dev_info(device, "DPC error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n",
281 		cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT),
282 		FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP),
283 		FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), dpc->rp_log_size,
284 		FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE));
285 	return status;
286 }
287 
288 static void dpc_remove(struct pcie_device *dev)
289 {
290 	struct dpc_dev *dpc = get_service_data(dev);
291 	struct pci_dev *pdev = dev->port;
292 	u16 ctl;
293 
294 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
295 	ctl &= ~(PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN);
296 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
297 }
298 
299 static struct pcie_port_service_driver dpcdriver = {
300 	.name		= "dpc",
301 	.port_type	= PCIE_ANY_PORT,
302 	.service	= PCIE_PORT_SERVICE_DPC,
303 	.probe		= dpc_probe,
304 	.remove		= dpc_remove,
305 	.reset_link	= dpc_reset_link,
306 };
307 
308 static int __init dpc_service_init(void)
309 {
310 	return pcie_port_service_register(&dpcdriver);
311 }
312 device_initcall(dpc_service_init);
313