xref: /linux/drivers/pci/pcie/dpc.c (revision 037fc3368be46dc1a2a90f6e50c8cbce49d75fd6)
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/aer.h>
10 #include <linux/delay.h>
11 #include <linux/interrupt.h>
12 #include <linux/init.h>
13 #include <linux/pci.h>
14 
15 #include "portdrv.h"
16 #include "../pci.h"
17 
18 struct dpc_dev {
19 	struct pcie_device	*dev;
20 	u16			cap_pos;
21 	bool			rp_extensions;
22 	u8			rp_log_size;
23 };
24 
25 static const char * const rp_pio_error_string[] = {
26 	"Configuration Request received UR Completion",	 /* Bit Position 0  */
27 	"Configuration Request received CA Completion",	 /* Bit Position 1  */
28 	"Configuration Request Completion Timeout",	 /* Bit Position 2  */
29 	NULL,
30 	NULL,
31 	NULL,
32 	NULL,
33 	NULL,
34 	"I/O Request received UR Completion",		 /* Bit Position 8  */
35 	"I/O Request received CA Completion",		 /* Bit Position 9  */
36 	"I/O Request Completion Timeout",		 /* Bit Position 10 */
37 	NULL,
38 	NULL,
39 	NULL,
40 	NULL,
41 	NULL,
42 	"Memory Request received UR Completion",	 /* Bit Position 16 */
43 	"Memory Request received CA Completion",	 /* Bit Position 17 */
44 	"Memory Request Completion Timeout",		 /* Bit Position 18 */
45 };
46 
47 static struct dpc_dev *to_dpc_dev(struct pci_dev *dev)
48 {
49 	struct device *device;
50 
51 	device = pcie_port_find_device(dev, PCIE_PORT_SERVICE_DPC);
52 	if (!device)
53 		return NULL;
54 	return get_service_data(to_pcie_device(device));
55 }
56 
57 void pci_save_dpc_state(struct pci_dev *dev)
58 {
59 	struct dpc_dev *dpc;
60 	struct pci_cap_saved_state *save_state;
61 	u16 *cap;
62 
63 	if (!pci_is_pcie(dev))
64 		return;
65 
66 	dpc = to_dpc_dev(dev);
67 	if (!dpc)
68 		return;
69 
70 	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
71 	if (!save_state)
72 		return;
73 
74 	cap = (u16 *)&save_state->cap.data[0];
75 	pci_read_config_word(dev, dpc->cap_pos + PCI_EXP_DPC_CTL, cap);
76 }
77 
78 void pci_restore_dpc_state(struct pci_dev *dev)
79 {
80 	struct dpc_dev *dpc;
81 	struct pci_cap_saved_state *save_state;
82 	u16 *cap;
83 
84 	if (!pci_is_pcie(dev))
85 		return;
86 
87 	dpc = to_dpc_dev(dev);
88 	if (!dpc)
89 		return;
90 
91 	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
92 	if (!save_state)
93 		return;
94 
95 	cap = (u16 *)&save_state->cap.data[0];
96 	pci_write_config_word(dev, dpc->cap_pos + PCI_EXP_DPC_CTL, *cap);
97 }
98 
99 static int dpc_wait_rp_inactive(struct dpc_dev *dpc)
100 {
101 	unsigned long timeout = jiffies + HZ;
102 	struct pci_dev *pdev = dpc->dev->port;
103 	struct device *dev = &dpc->dev->device;
104 	u16 cap = dpc->cap_pos, status;
105 
106 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
107 	while (status & PCI_EXP_DPC_RP_BUSY &&
108 					!time_after(jiffies, timeout)) {
109 		msleep(10);
110 		pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
111 	}
112 	if (status & PCI_EXP_DPC_RP_BUSY) {
113 		dev_warn(dev, "DPC root port still busy\n");
114 		return -EBUSY;
115 	}
116 	return 0;
117 }
118 
119 static pci_ers_result_t dpc_reset_link(struct pci_dev *pdev)
120 {
121 	struct dpc_dev *dpc;
122 	u16 cap;
123 
124 	/*
125 	 * DPC disables the Link automatically in hardware, so it has
126 	 * already been reset by the time we get here.
127 	 */
128 	dpc = to_dpc_dev(pdev);
129 	cap = dpc->cap_pos;
130 
131 	/*
132 	 * Wait until the Link is inactive, then clear DPC Trigger Status
133 	 * to allow the Port to leave DPC.
134 	 */
135 	pcie_wait_for_link(pdev, false);
136 
137 	if (dpc->rp_extensions && dpc_wait_rp_inactive(dpc))
138 		return PCI_ERS_RESULT_DISCONNECT;
139 
140 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
141 			      PCI_EXP_DPC_STATUS_TRIGGER);
142 
143 	if (!pcie_wait_for_link(pdev, true))
144 		return PCI_ERS_RESULT_DISCONNECT;
145 
146 	return PCI_ERS_RESULT_RECOVERED;
147 }
148 
149 static void dpc_process_rp_pio_error(struct dpc_dev *dpc)
150 {
151 	struct device *dev = &dpc->dev->device;
152 	struct pci_dev *pdev = dpc->dev->port;
153 	u16 cap = dpc->cap_pos, dpc_status, first_error;
154 	u32 status, mask, sev, syserr, exc, dw0, dw1, dw2, dw3, log, prefix;
155 	int i;
156 
157 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, &status);
158 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_MASK, &mask);
159 	dev_err(dev, "rp_pio_status: %#010x, rp_pio_mask: %#010x\n",
160 		status, mask);
161 
162 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SEVERITY, &sev);
163 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SYSERROR, &syserr);
164 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_EXCEPTION, &exc);
165 	dev_err(dev, "RP PIO severity=%#010x, syserror=%#010x, exception=%#010x\n",
166 		sev, syserr, exc);
167 
168 	/* Get First Error Pointer */
169 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &dpc_status);
170 	first_error = (dpc_status & 0x1f00) >> 8;
171 
172 	for (i = 0; i < ARRAY_SIZE(rp_pio_error_string); i++) {
173 		if ((status & ~mask) & (1 << i))
174 			dev_err(dev, "[%2d] %s%s\n", i, rp_pio_error_string[i],
175 				first_error == i ? " (First)" : "");
176 	}
177 
178 	if (dpc->rp_log_size < 4)
179 		goto clear_status;
180 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG,
181 			      &dw0);
182 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 4,
183 			      &dw1);
184 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 8,
185 			      &dw2);
186 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 12,
187 			      &dw3);
188 	dev_err(dev, "TLP Header: %#010x %#010x %#010x %#010x\n",
189 		dw0, dw1, dw2, dw3);
190 
191 	if (dpc->rp_log_size < 5)
192 		goto clear_status;
193 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_IMPSPEC_LOG, &log);
194 	dev_err(dev, "RP PIO ImpSpec Log %#010x\n", log);
195 
196 	for (i = 0; i < dpc->rp_log_size - 5; i++) {
197 		pci_read_config_dword(pdev,
198 			cap + PCI_EXP_DPC_RP_PIO_TLPPREFIX_LOG, &prefix);
199 		dev_err(dev, "TLP Prefix Header: dw%d, %#010x\n", i, prefix);
200 	}
201  clear_status:
202 	pci_write_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, status);
203 }
204 
205 static int dpc_get_aer_uncorrect_severity(struct pci_dev *dev,
206 					  struct aer_err_info *info)
207 {
208 	int pos = dev->aer_cap;
209 	u32 status, mask, sev;
210 
211 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
212 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
213 	status &= ~mask;
214 	if (!status)
215 		return 0;
216 
217 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &sev);
218 	status &= sev;
219 	if (status)
220 		info->severity = AER_FATAL;
221 	else
222 		info->severity = AER_NONFATAL;
223 
224 	return 1;
225 }
226 
227 static irqreturn_t dpc_handler(int irq, void *context)
228 {
229 	struct aer_err_info info;
230 	struct dpc_dev *dpc = context;
231 	struct pci_dev *pdev = dpc->dev->port;
232 	struct device *dev = &dpc->dev->device;
233 	u16 cap = dpc->cap_pos, status, source, reason, ext_reason;
234 
235 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
236 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID, &source);
237 
238 	dev_info(dev, "DPC containment event, status:%#06x source:%#06x\n",
239 		 status, source);
240 
241 	reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN) >> 1;
242 	ext_reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT) >> 5;
243 	dev_warn(dev, "DPC %s detected\n",
244 		 (reason == 0) ? "unmasked uncorrectable error" :
245 		 (reason == 1) ? "ERR_NONFATAL" :
246 		 (reason == 2) ? "ERR_FATAL" :
247 		 (ext_reason == 0) ? "RP PIO error" :
248 		 (ext_reason == 1) ? "software trigger" :
249 				     "reserved error");
250 
251 	/* show RP PIO error detail information */
252 	if (dpc->rp_extensions && reason == 3 && ext_reason == 0)
253 		dpc_process_rp_pio_error(dpc);
254 	else if (reason == 0 &&
255 		 dpc_get_aer_uncorrect_severity(pdev, &info) &&
256 		 aer_get_device_error_info(pdev, &info)) {
257 		aer_print_error(pdev, &info);
258 		pci_cleanup_aer_uncorrect_error_status(pdev);
259 		pci_aer_clear_fatal_status(pdev);
260 	}
261 
262 	/* We configure DPC so it only triggers on ERR_FATAL */
263 	pcie_do_recovery(pdev, pci_channel_io_frozen, PCIE_PORT_SERVICE_DPC);
264 
265 	return IRQ_HANDLED;
266 }
267 
268 static irqreturn_t dpc_irq(int irq, void *context)
269 {
270 	struct dpc_dev *dpc = (struct dpc_dev *)context;
271 	struct pci_dev *pdev = dpc->dev->port;
272 	u16 cap = dpc->cap_pos, status;
273 
274 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
275 
276 	if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT) || status == (u16)(~0))
277 		return IRQ_NONE;
278 
279 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
280 			      PCI_EXP_DPC_STATUS_INTERRUPT);
281 	if (status & PCI_EXP_DPC_STATUS_TRIGGER)
282 		return IRQ_WAKE_THREAD;
283 	return IRQ_HANDLED;
284 }
285 
286 #define FLAG(x, y) (((x) & (y)) ? '+' : '-')
287 static int dpc_probe(struct pcie_device *dev)
288 {
289 	struct dpc_dev *dpc;
290 	struct pci_dev *pdev = dev->port;
291 	struct device *device = &dev->device;
292 	int status;
293 	u16 ctl, cap;
294 
295 	if (pcie_aer_get_firmware_first(pdev))
296 		return -ENOTSUPP;
297 
298 	dpc = devm_kzalloc(device, sizeof(*dpc), GFP_KERNEL);
299 	if (!dpc)
300 		return -ENOMEM;
301 
302 	dpc->cap_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC);
303 	dpc->dev = dev;
304 	set_service_data(dev, dpc);
305 
306 	status = devm_request_threaded_irq(device, dev->irq, dpc_irq,
307 					   dpc_handler, IRQF_SHARED,
308 					   "pcie-dpc", dpc);
309 	if (status) {
310 		dev_warn(device, "request IRQ%d failed: %d\n", dev->irq,
311 			 status);
312 		return status;
313 	}
314 
315 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CAP, &cap);
316 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
317 
318 	dpc->rp_extensions = (cap & PCI_EXP_DPC_CAP_RP_EXT);
319 	if (dpc->rp_extensions) {
320 		dpc->rp_log_size = (cap & PCI_EXP_DPC_RP_PIO_LOG_SIZE) >> 8;
321 		if (dpc->rp_log_size < 4 || dpc->rp_log_size > 9) {
322 			dev_err(device, "RP PIO log size %u is invalid\n",
323 				dpc->rp_log_size);
324 			dpc->rp_log_size = 0;
325 		}
326 	}
327 
328 	ctl = (ctl & 0xfff4) | PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN;
329 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
330 
331 	dev_info(device, "DPC error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n",
332 		cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT),
333 		FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP),
334 		FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), dpc->rp_log_size,
335 		FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE));
336 
337 	pci_add_ext_cap_save_buffer(pdev, PCI_EXT_CAP_ID_DPC, sizeof(u16));
338 	return status;
339 }
340 
341 static void dpc_remove(struct pcie_device *dev)
342 {
343 	struct dpc_dev *dpc = get_service_data(dev);
344 	struct pci_dev *pdev = dev->port;
345 	u16 ctl;
346 
347 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
348 	ctl &= ~(PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN);
349 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
350 }
351 
352 static struct pcie_port_service_driver dpcdriver = {
353 	.name		= "dpc",
354 	.port_type	= PCIE_ANY_PORT,
355 	.service	= PCIE_PORT_SERVICE_DPC,
356 	.probe		= dpc_probe,
357 	.remove		= dpc_remove,
358 	.reset_link	= dpc_reset_link,
359 };
360 
361 int __init pcie_dpc_init(void)
362 {
363 	return pcie_port_service_register(&dpcdriver);
364 }
365