xref: /linux/drivers/pci/pcie/dpc.c (revision 9a6b55ac)
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 #define dev_fmt(fmt) "DPC: " fmt
10 
11 #include <linux/aer.h>
12 #include <linux/delay.h>
13 #include <linux/interrupt.h>
14 #include <linux/init.h>
15 #include <linux/pci.h>
16 
17 #include "portdrv.h"
18 #include "../pci.h"
19 
20 struct dpc_dev {
21 	struct pcie_device	*dev;
22 	u16			cap_pos;
23 	bool			rp_extensions;
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 struct dpc_dev *to_dpc_dev(struct pci_dev *dev)
50 {
51 	struct device *device;
52 
53 	device = pcie_port_find_device(dev, PCIE_PORT_SERVICE_DPC);
54 	if (!device)
55 		return NULL;
56 	return get_service_data(to_pcie_device(device));
57 }
58 
59 void pci_save_dpc_state(struct pci_dev *dev)
60 {
61 	struct dpc_dev *dpc;
62 	struct pci_cap_saved_state *save_state;
63 	u16 *cap;
64 
65 	if (!pci_is_pcie(dev))
66 		return;
67 
68 	dpc = to_dpc_dev(dev);
69 	if (!dpc)
70 		return;
71 
72 	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
73 	if (!save_state)
74 		return;
75 
76 	cap = (u16 *)&save_state->cap.data[0];
77 	pci_read_config_word(dev, dpc->cap_pos + PCI_EXP_DPC_CTL, cap);
78 }
79 
80 void pci_restore_dpc_state(struct pci_dev *dev)
81 {
82 	struct dpc_dev *dpc;
83 	struct pci_cap_saved_state *save_state;
84 	u16 *cap;
85 
86 	if (!pci_is_pcie(dev))
87 		return;
88 
89 	dpc = to_dpc_dev(dev);
90 	if (!dpc)
91 		return;
92 
93 	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
94 	if (!save_state)
95 		return;
96 
97 	cap = (u16 *)&save_state->cap.data[0];
98 	pci_write_config_word(dev, dpc->cap_pos + PCI_EXP_DPC_CTL, *cap);
99 }
100 
101 static int dpc_wait_rp_inactive(struct dpc_dev *dpc)
102 {
103 	unsigned long timeout = jiffies + HZ;
104 	struct pci_dev *pdev = dpc->dev->port;
105 	u16 cap = dpc->cap_pos, status;
106 
107 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
108 	while (status & PCI_EXP_DPC_RP_BUSY &&
109 					!time_after(jiffies, timeout)) {
110 		msleep(10);
111 		pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
112 	}
113 	if (status & PCI_EXP_DPC_RP_BUSY) {
114 		pci_warn(pdev, "root port still busy\n");
115 		return -EBUSY;
116 	}
117 	return 0;
118 }
119 
120 static pci_ers_result_t dpc_reset_link(struct pci_dev *pdev)
121 {
122 	struct dpc_dev *dpc;
123 	u16 cap;
124 
125 	/*
126 	 * DPC disables the Link automatically in hardware, so it has
127 	 * already been reset by the time we get here.
128 	 */
129 	dpc = to_dpc_dev(pdev);
130 	cap = dpc->cap_pos;
131 
132 	/*
133 	 * Wait until the Link is inactive, then clear DPC Trigger Status
134 	 * to allow the Port to leave DPC.
135 	 */
136 	pcie_wait_for_link(pdev, false);
137 
138 	if (dpc->rp_extensions && dpc_wait_rp_inactive(dpc))
139 		return PCI_ERS_RESULT_DISCONNECT;
140 
141 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
142 			      PCI_EXP_DPC_STATUS_TRIGGER);
143 
144 	if (!pcie_wait_for_link(pdev, true))
145 		return PCI_ERS_RESULT_DISCONNECT;
146 
147 	return PCI_ERS_RESULT_RECOVERED;
148 }
149 
150 static void dpc_process_rp_pio_error(struct dpc_dev *dpc)
151 {
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 	pci_err(pdev, "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 	pci_err(pdev, "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 			pci_err(pdev, "[%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 	pci_err(pdev, "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 	pci_err(pdev, "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 		pci_err(pdev, "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 	u16 cap = dpc->cap_pos, status, source, reason, ext_reason;
233 
234 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
235 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID, &source);
236 
237 	pci_info(pdev, "containment event, status:%#06x source:%#06x\n",
238 		 status, source);
239 
240 	reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN) >> 1;
241 	ext_reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT) >> 5;
242 	pci_warn(pdev, "%s detected\n",
243 		 (reason == 0) ? "unmasked uncorrectable error" :
244 		 (reason == 1) ? "ERR_NONFATAL" :
245 		 (reason == 2) ? "ERR_FATAL" :
246 		 (ext_reason == 0) ? "RP PIO error" :
247 		 (ext_reason == 1) ? "software trigger" :
248 				     "reserved error");
249 
250 	/* show RP PIO error detail information */
251 	if (dpc->rp_extensions && reason == 3 && ext_reason == 0)
252 		dpc_process_rp_pio_error(dpc);
253 	else if (reason == 0 &&
254 		 dpc_get_aer_uncorrect_severity(pdev, &info) &&
255 		 aer_get_device_error_info(pdev, &info)) {
256 		aer_print_error(pdev, &info);
257 		pci_cleanup_aer_uncorrect_error_status(pdev);
258 		pci_aer_clear_fatal_status(pdev);
259 	}
260 
261 	/* We configure DPC so it only triggers on ERR_FATAL */
262 	pcie_do_recovery(pdev, pci_channel_io_frozen, PCIE_PORT_SERVICE_DPC);
263 
264 	return IRQ_HANDLED;
265 }
266 
267 static irqreturn_t dpc_irq(int irq, void *context)
268 {
269 	struct dpc_dev *dpc = (struct dpc_dev *)context;
270 	struct pci_dev *pdev = dpc->dev->port;
271 	u16 cap = dpc->cap_pos, status;
272 
273 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
274 
275 	if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT) || status == (u16)(~0))
276 		return IRQ_NONE;
277 
278 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
279 			      PCI_EXP_DPC_STATUS_INTERRUPT);
280 	if (status & PCI_EXP_DPC_STATUS_TRIGGER)
281 		return IRQ_WAKE_THREAD;
282 	return IRQ_HANDLED;
283 }
284 
285 #define FLAG(x, y) (((x) & (y)) ? '+' : '-')
286 static int dpc_probe(struct pcie_device *dev)
287 {
288 	struct dpc_dev *dpc;
289 	struct pci_dev *pdev = dev->port;
290 	struct device *device = &dev->device;
291 	int status;
292 	u16 ctl, cap;
293 
294 	if (pcie_aer_get_firmware_first(pdev) && !pcie_ports_dpc_native)
295 		return -ENOTSUPP;
296 
297 	dpc = devm_kzalloc(device, sizeof(*dpc), GFP_KERNEL);
298 	if (!dpc)
299 		return -ENOMEM;
300 
301 	dpc->cap_pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC);
302 	dpc->dev = dev;
303 	set_service_data(dev, dpc);
304 
305 	status = devm_request_threaded_irq(device, dev->irq, dpc_irq,
306 					   dpc_handler, IRQF_SHARED,
307 					   "pcie-dpc", dpc);
308 	if (status) {
309 		pci_warn(pdev, "request IRQ%d failed: %d\n", dev->irq,
310 			 status);
311 		return status;
312 	}
313 
314 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CAP, &cap);
315 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
316 
317 	dpc->rp_extensions = (cap & PCI_EXP_DPC_CAP_RP_EXT);
318 	if (dpc->rp_extensions) {
319 		dpc->rp_log_size = (cap & PCI_EXP_DPC_RP_PIO_LOG_SIZE) >> 8;
320 		if (dpc->rp_log_size < 4 || dpc->rp_log_size > 9) {
321 			pci_err(pdev, "RP PIO log size %u is invalid\n",
322 				dpc->rp_log_size);
323 			dpc->rp_log_size = 0;
324 		}
325 	}
326 
327 	ctl = (ctl & 0xfff4) | PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN;
328 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
329 
330 	pci_info(pdev, "error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n",
331 		 cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT),
332 		 FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP),
333 		 FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), dpc->rp_log_size,
334 		 FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE));
335 
336 	pci_add_ext_cap_save_buffer(pdev, PCI_EXT_CAP_ID_DPC, sizeof(u16));
337 	return status;
338 }
339 
340 static void dpc_remove(struct pcie_device *dev)
341 {
342 	struct dpc_dev *dpc = get_service_data(dev);
343 	struct pci_dev *pdev = dev->port;
344 	u16 ctl;
345 
346 	pci_read_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, &ctl);
347 	ctl &= ~(PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN);
348 	pci_write_config_word(pdev, dpc->cap_pos + PCI_EXP_DPC_CTL, ctl);
349 }
350 
351 static struct pcie_port_service_driver dpcdriver = {
352 	.name		= "dpc",
353 	.port_type	= PCIE_ANY_PORT,
354 	.service	= PCIE_PORT_SERVICE_DPC,
355 	.probe		= dpc_probe,
356 	.remove		= dpc_remove,
357 	.reset_link	= dpc_reset_link,
358 };
359 
360 int __init pcie_dpc_init(void)
361 {
362 	return pcie_port_service_register(&dpcdriver);
363 }
364