xref: /linux/drivers/pci/pcie/dpc.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
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 static const char * const rp_pio_error_string[] = {
21 	"Configuration Request received UR Completion",	 /* Bit Position 0  */
22 	"Configuration Request received CA Completion",	 /* Bit Position 1  */
23 	"Configuration Request Completion Timeout",	 /* Bit Position 2  */
24 	NULL,
25 	NULL,
26 	NULL,
27 	NULL,
28 	NULL,
29 	"I/O Request received UR Completion",		 /* Bit Position 8  */
30 	"I/O Request received CA Completion",		 /* Bit Position 9  */
31 	"I/O Request Completion Timeout",		 /* Bit Position 10 */
32 	NULL,
33 	NULL,
34 	NULL,
35 	NULL,
36 	NULL,
37 	"Memory Request received UR Completion",	 /* Bit Position 16 */
38 	"Memory Request received CA Completion",	 /* Bit Position 17 */
39 	"Memory Request Completion Timeout",		 /* Bit Position 18 */
40 };
41 
42 void pci_save_dpc_state(struct pci_dev *dev)
43 {
44 	struct pci_cap_saved_state *save_state;
45 	u16 *cap;
46 
47 	if (!pci_is_pcie(dev))
48 		return;
49 
50 	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
51 	if (!save_state)
52 		return;
53 
54 	cap = (u16 *)&save_state->cap.data[0];
55 	pci_read_config_word(dev, dev->dpc_cap + PCI_EXP_DPC_CTL, cap);
56 }
57 
58 void pci_restore_dpc_state(struct pci_dev *dev)
59 {
60 	struct pci_cap_saved_state *save_state;
61 	u16 *cap;
62 
63 	if (!pci_is_pcie(dev))
64 		return;
65 
66 	save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
67 	if (!save_state)
68 		return;
69 
70 	cap = (u16 *)&save_state->cap.data[0];
71 	pci_write_config_word(dev, dev->dpc_cap + PCI_EXP_DPC_CTL, *cap);
72 }
73 
74 static int dpc_wait_rp_inactive(struct pci_dev *pdev)
75 {
76 	unsigned long timeout = jiffies + HZ;
77 	u16 cap = pdev->dpc_cap, status;
78 
79 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
80 	while (status & PCI_EXP_DPC_RP_BUSY &&
81 					!time_after(jiffies, timeout)) {
82 		msleep(10);
83 		pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
84 	}
85 	if (status & PCI_EXP_DPC_RP_BUSY) {
86 		pci_warn(pdev, "root port still busy\n");
87 		return -EBUSY;
88 	}
89 	return 0;
90 }
91 
92 pci_ers_result_t dpc_reset_link(struct pci_dev *pdev)
93 {
94 	u16 cap;
95 
96 	/*
97 	 * DPC disables the Link automatically in hardware, so it has
98 	 * already been reset by the time we get here.
99 	 */
100 	cap = pdev->dpc_cap;
101 
102 	/*
103 	 * Wait until the Link is inactive, then clear DPC Trigger Status
104 	 * to allow the Port to leave DPC.
105 	 */
106 	if (!pcie_wait_for_link(pdev, false))
107 		pci_info(pdev, "Data Link Layer Link Active not cleared in 1000 msec\n");
108 
109 	if (pdev->dpc_rp_extensions && dpc_wait_rp_inactive(pdev))
110 		return PCI_ERS_RESULT_DISCONNECT;
111 
112 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
113 			      PCI_EXP_DPC_STATUS_TRIGGER);
114 
115 	if (!pcie_wait_for_link(pdev, true)) {
116 		pci_info(pdev, "Data Link Layer Link Active not set in 1000 msec\n");
117 		return PCI_ERS_RESULT_DISCONNECT;
118 	}
119 
120 	return PCI_ERS_RESULT_RECOVERED;
121 }
122 
123 static void dpc_process_rp_pio_error(struct pci_dev *pdev)
124 {
125 	u16 cap = pdev->dpc_cap, dpc_status, first_error;
126 	u32 status, mask, sev, syserr, exc, dw0, dw1, dw2, dw3, log, prefix;
127 	int i;
128 
129 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, &status);
130 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_MASK, &mask);
131 	pci_err(pdev, "rp_pio_status: %#010x, rp_pio_mask: %#010x\n",
132 		status, mask);
133 
134 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SEVERITY, &sev);
135 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SYSERROR, &syserr);
136 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_EXCEPTION, &exc);
137 	pci_err(pdev, "RP PIO severity=%#010x, syserror=%#010x, exception=%#010x\n",
138 		sev, syserr, exc);
139 
140 	/* Get First Error Pointer */
141 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &dpc_status);
142 	first_error = (dpc_status & 0x1f00) >> 8;
143 
144 	for (i = 0; i < ARRAY_SIZE(rp_pio_error_string); i++) {
145 		if ((status & ~mask) & (1 << i))
146 			pci_err(pdev, "[%2d] %s%s\n", i, rp_pio_error_string[i],
147 				first_error == i ? " (First)" : "");
148 	}
149 
150 	if (pdev->dpc_rp_log_size < 4)
151 		goto clear_status;
152 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG,
153 			      &dw0);
154 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 4,
155 			      &dw1);
156 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 8,
157 			      &dw2);
158 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG + 12,
159 			      &dw3);
160 	pci_err(pdev, "TLP Header: %#010x %#010x %#010x %#010x\n",
161 		dw0, dw1, dw2, dw3);
162 
163 	if (pdev->dpc_rp_log_size < 5)
164 		goto clear_status;
165 	pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_IMPSPEC_LOG, &log);
166 	pci_err(pdev, "RP PIO ImpSpec Log %#010x\n", log);
167 
168 	for (i = 0; i < pdev->dpc_rp_log_size - 5; i++) {
169 		pci_read_config_dword(pdev,
170 			cap + PCI_EXP_DPC_RP_PIO_TLPPREFIX_LOG, &prefix);
171 		pci_err(pdev, "TLP Prefix Header: dw%d, %#010x\n", i, prefix);
172 	}
173  clear_status:
174 	pci_write_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, status);
175 }
176 
177 static int dpc_get_aer_uncorrect_severity(struct pci_dev *dev,
178 					  struct aer_err_info *info)
179 {
180 	int pos = dev->aer_cap;
181 	u32 status, mask, sev;
182 
183 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
184 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
185 	status &= ~mask;
186 	if (!status)
187 		return 0;
188 
189 	pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &sev);
190 	status &= sev;
191 	if (status)
192 		info->severity = AER_FATAL;
193 	else
194 		info->severity = AER_NONFATAL;
195 
196 	return 1;
197 }
198 
199 void dpc_process_error(struct pci_dev *pdev)
200 {
201 	u16 cap = pdev->dpc_cap, status, source, reason, ext_reason;
202 	struct aer_err_info info;
203 
204 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
205 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID, &source);
206 
207 	pci_info(pdev, "containment event, status:%#06x source:%#06x\n",
208 		 status, source);
209 
210 	reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN) >> 1;
211 	ext_reason = (status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT) >> 5;
212 	pci_warn(pdev, "%s detected\n",
213 		 (reason == 0) ? "unmasked uncorrectable error" :
214 		 (reason == 1) ? "ERR_NONFATAL" :
215 		 (reason == 2) ? "ERR_FATAL" :
216 		 (ext_reason == 0) ? "RP PIO error" :
217 		 (ext_reason == 1) ? "software trigger" :
218 				     "reserved error");
219 
220 	/* show RP PIO error detail information */
221 	if (pdev->dpc_rp_extensions && reason == 3 && ext_reason == 0)
222 		dpc_process_rp_pio_error(pdev);
223 	else if (reason == 0 &&
224 		 dpc_get_aer_uncorrect_severity(pdev, &info) &&
225 		 aer_get_device_error_info(pdev, &info)) {
226 		aer_print_error(pdev, &info);
227 		pci_aer_clear_nonfatal_status(pdev);
228 		pci_aer_clear_fatal_status(pdev);
229 	}
230 }
231 
232 static irqreturn_t dpc_handler(int irq, void *context)
233 {
234 	struct pci_dev *pdev = context;
235 
236 	dpc_process_error(pdev);
237 
238 	/* We configure DPC so it only triggers on ERR_FATAL */
239 	pcie_do_recovery(pdev, pci_channel_io_frozen, dpc_reset_link);
240 
241 	return IRQ_HANDLED;
242 }
243 
244 static irqreturn_t dpc_irq(int irq, void *context)
245 {
246 	struct pci_dev *pdev = context;
247 	u16 cap = pdev->dpc_cap, status;
248 
249 	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);
250 
251 	if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT) || status == (u16)(~0))
252 		return IRQ_NONE;
253 
254 	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
255 			      PCI_EXP_DPC_STATUS_INTERRUPT);
256 	if (status & PCI_EXP_DPC_STATUS_TRIGGER)
257 		return IRQ_WAKE_THREAD;
258 	return IRQ_HANDLED;
259 }
260 
261 void pci_dpc_init(struct pci_dev *pdev)
262 {
263 	u16 cap;
264 
265 	pdev->dpc_cap = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC);
266 	if (!pdev->dpc_cap)
267 		return;
268 
269 	pci_read_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CAP, &cap);
270 	if (!(cap & PCI_EXP_DPC_CAP_RP_EXT))
271 		return;
272 
273 	pdev->dpc_rp_extensions = true;
274 	pdev->dpc_rp_log_size = (cap & PCI_EXP_DPC_RP_PIO_LOG_SIZE) >> 8;
275 	if (pdev->dpc_rp_log_size < 4 || pdev->dpc_rp_log_size > 9) {
276 		pci_err(pdev, "RP PIO log size %u is invalid\n",
277 			pdev->dpc_rp_log_size);
278 		pdev->dpc_rp_log_size = 0;
279 	}
280 }
281 
282 #define FLAG(x, y) (((x) & (y)) ? '+' : '-')
283 static int dpc_probe(struct pcie_device *dev)
284 {
285 	struct pci_dev *pdev = dev->port;
286 	struct device *device = &dev->device;
287 	int status;
288 	u16 ctl, cap;
289 
290 	if (!pcie_aer_is_native(pdev) && !pcie_ports_dpc_native)
291 		return -ENOTSUPP;
292 
293 	status = devm_request_threaded_irq(device, dev->irq, dpc_irq,
294 					   dpc_handler, IRQF_SHARED,
295 					   "pcie-dpc", pdev);
296 	if (status) {
297 		pci_warn(pdev, "request IRQ%d failed: %d\n", dev->irq,
298 			 status);
299 		return status;
300 	}
301 
302 	pci_read_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CAP, &cap);
303 	pci_read_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CTL, &ctl);
304 
305 	ctl = (ctl & 0xfff4) | PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN;
306 	pci_write_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CTL, ctl);
307 	pci_info(pdev, "enabled with IRQ %d\n", dev->irq);
308 
309 	pci_info(pdev, "error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n",
310 		 cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT),
311 		 FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP),
312 		 FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), pdev->dpc_rp_log_size,
313 		 FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE));
314 
315 	pci_add_ext_cap_save_buffer(pdev, PCI_EXT_CAP_ID_DPC, sizeof(u16));
316 	return status;
317 }
318 
319 static void dpc_remove(struct pcie_device *dev)
320 {
321 	struct pci_dev *pdev = dev->port;
322 	u16 ctl;
323 
324 	pci_read_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CTL, &ctl);
325 	ctl &= ~(PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN);
326 	pci_write_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CTL, ctl);
327 }
328 
329 static struct pcie_port_service_driver dpcdriver = {
330 	.name		= "dpc",
331 	.port_type	= PCIE_ANY_PORT,
332 	.service	= PCIE_PORT_SERVICE_DPC,
333 	.probe		= dpc_probe,
334 	.remove		= dpc_remove,
335 };
336 
337 int __init pcie_dpc_init(void)
338 {
339 	return pcie_port_service_register(&dpcdriver);
340 }
341