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/bitfield.h>
13#include <linux/delay.h>
14#include <linux/interrupt.h>
15#include <linux/init.h>
16#include <linux/pci.h>
17
18#include "portdrv.h"
19#include "../pci.h"
20
21#define PCI_EXP_DPC_CTL_EN_MASK (PCI_EXP_DPC_CTL_EN_FATAL | \
22 PCI_EXP_DPC_CTL_EN_NONFATAL)
23
24static const char * const rp_pio_error_string[] = {
25 "Configuration Request received UR Completion", /* Bit Position 0 */
26 "Configuration Request received CA Completion", /* Bit Position 1 */
27 "Configuration Request Completion Timeout", /* Bit Position 2 */
28 NULL,
29 NULL,
30 NULL,
31 NULL,
32 NULL,
33 "I/O Request received UR Completion", /* Bit Position 8 */
34 "I/O Request received CA Completion", /* Bit Position 9 */
35 "I/O Request Completion Timeout", /* Bit Position 10 */
36 NULL,
37 NULL,
38 NULL,
39 NULL,
40 NULL,
41 "Memory Request received UR Completion", /* Bit Position 16 */
42 "Memory Request received CA Completion", /* Bit Position 17 */
43 "Memory Request Completion Timeout", /* Bit Position 18 */
44};
45
46void pci_save_dpc_state(struct pci_dev *dev)
47{
48 struct pci_cap_saved_state *save_state;
49 u16 *cap;
50
51 if (!pci_is_pcie(dev))
52 return;
53
54 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC);
55 if (!save_state)
56 return;
57
58 cap = (u16 *)&save_state->cap.data[0];
59 pci_read_config_word(dev, where: dev->dpc_cap + PCI_EXP_DPC_CTL, val: cap);
60}
61
62void pci_restore_dpc_state(struct pci_dev *dev)
63{
64 struct pci_cap_saved_state *save_state;
65 u16 *cap;
66
67 if (!pci_is_pcie(dev))
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_write_config_word(dev, where: dev->dpc_cap + PCI_EXP_DPC_CTL, val: *cap);
76}
77
78static DECLARE_WAIT_QUEUE_HEAD(dpc_completed_waitqueue);
79
80#ifdef CONFIG_HOTPLUG_PCI_PCIE
81static bool dpc_completed(struct pci_dev *pdev)
82{
83 u16 status;
84
85 pci_read_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_STATUS, val: &status);
86 if ((!PCI_POSSIBLE_ERROR(status)) && (status & PCI_EXP_DPC_STATUS_TRIGGER))
87 return false;
88
89 if (test_bit(PCI_DPC_RECOVERING, &pdev->priv_flags))
90 return false;
91
92 return true;
93}
94
95/**
96 * pci_dpc_recovered - whether DPC triggered and has recovered successfully
97 * @pdev: PCI device
98 *
99 * Return true if DPC was triggered for @pdev and has recovered successfully.
100 * Wait for recovery if it hasn't completed yet. Called from the PCIe hotplug
101 * driver to recognize and ignore Link Down/Up events caused by DPC.
102 */
103bool pci_dpc_recovered(struct pci_dev *pdev)
104{
105 struct pci_host_bridge *host;
106
107 if (!pdev->dpc_cap)
108 return false;
109
110 /*
111 * Synchronization between hotplug and DPC is not supported
112 * if DPC is owned by firmware and EDR is not enabled.
113 */
114 host = pci_find_host_bridge(bus: pdev->bus);
115 if (!host->native_dpc && !IS_ENABLED(CONFIG_PCIE_EDR))
116 return false;
117
118 /*
119 * Need a timeout in case DPC never completes due to failure of
120 * dpc_wait_rp_inactive(). The spec doesn't mandate a time limit,
121 * but reports indicate that DPC completes within 4 seconds.
122 */
123 wait_event_timeout(dpc_completed_waitqueue, dpc_completed(pdev),
124 msecs_to_jiffies(4000));
125
126 return test_and_clear_bit(PCI_DPC_RECOVERED, addr: &pdev->priv_flags);
127}
128#endif /* CONFIG_HOTPLUG_PCI_PCIE */
129
130static int dpc_wait_rp_inactive(struct pci_dev *pdev)
131{
132 unsigned long timeout = jiffies + HZ;
133 u16 cap = pdev->dpc_cap, status;
134
135 pci_read_config_word(dev: pdev, where: cap + PCI_EXP_DPC_STATUS, val: &status);
136 while (status & PCI_EXP_DPC_RP_BUSY &&
137 !time_after(jiffies, timeout)) {
138 msleep(msecs: 10);
139 pci_read_config_word(dev: pdev, where: cap + PCI_EXP_DPC_STATUS, val: &status);
140 }
141 if (status & PCI_EXP_DPC_RP_BUSY) {
142 pci_warn(pdev, "root port still busy\n");
143 return -EBUSY;
144 }
145 return 0;
146}
147
148pci_ers_result_t dpc_reset_link(struct pci_dev *pdev)
149{
150 pci_ers_result_t ret;
151 u16 cap;
152
153 set_bit(PCI_DPC_RECOVERING, addr: &pdev->priv_flags);
154
155 /*
156 * DPC disables the Link automatically in hardware, so it has
157 * already been reset by the time we get here.
158 */
159 cap = pdev->dpc_cap;
160
161 /*
162 * Wait until the Link is inactive, then clear DPC Trigger Status
163 * to allow the Port to leave DPC.
164 */
165 if (!pcie_wait_for_link(pdev, active: false))
166 pci_info(pdev, "Data Link Layer Link Active not cleared in 1000 msec\n");
167
168 if (pdev->dpc_rp_extensions && dpc_wait_rp_inactive(pdev)) {
169 clear_bit(PCI_DPC_RECOVERED, addr: &pdev->priv_flags);
170 ret = PCI_ERS_RESULT_DISCONNECT;
171 goto out;
172 }
173
174 pci_write_config_word(dev: pdev, where: cap + PCI_EXP_DPC_STATUS,
175 PCI_EXP_DPC_STATUS_TRIGGER);
176
177 if (pci_bridge_wait_for_secondary_bus(dev: pdev, reset_type: "DPC")) {
178 clear_bit(PCI_DPC_RECOVERED, addr: &pdev->priv_flags);
179 ret = PCI_ERS_RESULT_DISCONNECT;
180 } else {
181 set_bit(PCI_DPC_RECOVERED, addr: &pdev->priv_flags);
182 ret = PCI_ERS_RESULT_RECOVERED;
183 }
184out:
185 clear_bit(PCI_DPC_RECOVERING, addr: &pdev->priv_flags);
186 wake_up_all(&dpc_completed_waitqueue);
187 return ret;
188}
189
190static void dpc_process_rp_pio_error(struct pci_dev *pdev)
191{
192 u16 cap = pdev->dpc_cap, dpc_status, first_error;
193 u32 status, mask, sev, syserr, exc, log, prefix;
194 struct pcie_tlp_log tlp_log;
195 int i;
196
197 pci_read_config_dword(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_STATUS, val: &status);
198 pci_read_config_dword(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_MASK, val: &mask);
199 pci_err(pdev, "rp_pio_status: %#010x, rp_pio_mask: %#010x\n",
200 status, mask);
201
202 pci_read_config_dword(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_SEVERITY, val: &sev);
203 pci_read_config_dword(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_SYSERROR, val: &syserr);
204 pci_read_config_dword(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_EXCEPTION, val: &exc);
205 pci_err(pdev, "RP PIO severity=%#010x, syserror=%#010x, exception=%#010x\n",
206 sev, syserr, exc);
207
208 /* Get First Error Pointer */
209 pci_read_config_word(dev: pdev, where: cap + PCI_EXP_DPC_STATUS, val: &dpc_status);
210 first_error = FIELD_GET(PCI_EXP_DPC_RP_PIO_FEP, dpc_status);
211
212 for (i = 0; i < ARRAY_SIZE(rp_pio_error_string); i++) {
213 if ((status & ~mask) & (1 << i))
214 pci_err(pdev, "[%2d] %s%s\n", i, rp_pio_error_string[i],
215 first_error == i ? " (First)" : "");
216 }
217
218 if (pdev->dpc_rp_log_size < 4)
219 goto clear_status;
220 pcie_read_tlp_log(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG, log: &tlp_log);
221 pci_err(pdev, "TLP Header: %#010x %#010x %#010x %#010x\n",
222 tlp_log.dw[0], tlp_log.dw[1], tlp_log.dw[2], tlp_log.dw[3]);
223
224 if (pdev->dpc_rp_log_size < 5)
225 goto clear_status;
226 pci_read_config_dword(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_IMPSPEC_LOG, val: &log);
227 pci_err(pdev, "RP PIO ImpSpec Log %#010x\n", log);
228
229 for (i = 0; i < pdev->dpc_rp_log_size - 5; i++) {
230 pci_read_config_dword(dev: pdev,
231 where: cap + PCI_EXP_DPC_RP_PIO_TLPPREFIX_LOG + i * 4, val: &prefix);
232 pci_err(pdev, "TLP Prefix Header: dw%d, %#010x\n", i, prefix);
233 }
234 clear_status:
235 pci_write_config_dword(dev: pdev, where: cap + PCI_EXP_DPC_RP_PIO_STATUS, val: status);
236}
237
238static int dpc_get_aer_uncorrect_severity(struct pci_dev *dev,
239 struct aer_err_info *info)
240{
241 int pos = dev->aer_cap;
242 u32 status, mask, sev;
243
244 pci_read_config_dword(dev, where: pos + PCI_ERR_UNCOR_STATUS, val: &status);
245 pci_read_config_dword(dev, where: pos + PCI_ERR_UNCOR_MASK, val: &mask);
246 status &= ~mask;
247 if (!status)
248 return 0;
249
250 pci_read_config_dword(dev, where: pos + PCI_ERR_UNCOR_SEVER, val: &sev);
251 status &= sev;
252 if (status)
253 info->severity = AER_FATAL;
254 else
255 info->severity = AER_NONFATAL;
256
257 return 1;
258}
259
260void dpc_process_error(struct pci_dev *pdev)
261{
262 u16 cap = pdev->dpc_cap, status, source, reason, ext_reason;
263 struct aer_err_info info;
264
265 pci_read_config_word(dev: pdev, where: cap + PCI_EXP_DPC_STATUS, val: &status);
266 pci_read_config_word(dev: pdev, where: cap + PCI_EXP_DPC_SOURCE_ID, val: &source);
267
268 pci_info(pdev, "containment event, status:%#06x source:%#06x\n",
269 status, source);
270
271 reason = status & PCI_EXP_DPC_STATUS_TRIGGER_RSN;
272 ext_reason = status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT;
273 pci_warn(pdev, "%s detected\n",
274 (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_UNCOR) ?
275 "unmasked uncorrectable error" :
276 (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_NFE) ?
277 "ERR_NONFATAL" :
278 (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_FE) ?
279 "ERR_FATAL" :
280 (ext_reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_RP_PIO) ?
281 "RP PIO error" :
282 (ext_reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_SW_TRIGGER) ?
283 "software trigger" :
284 "reserved error");
285
286 /* show RP PIO error detail information */
287 if (pdev->dpc_rp_extensions &&
288 reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_IN_EXT &&
289 ext_reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_RP_PIO)
290 dpc_process_rp_pio_error(pdev);
291 else if (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_UNCOR &&
292 dpc_get_aer_uncorrect_severity(dev: pdev, info: &info) &&
293 aer_get_device_error_info(dev: pdev, info: &info)) {
294 aer_print_error(dev: pdev, info: &info);
295 pci_aer_clear_nonfatal_status(dev: pdev);
296 pci_aer_clear_fatal_status(dev: pdev);
297 }
298}
299
300static void pci_clear_surpdn_errors(struct pci_dev *pdev)
301{
302 if (pdev->dpc_rp_extensions)
303 pci_write_config_dword(dev: pdev, where: pdev->dpc_cap +
304 PCI_EXP_DPC_RP_PIO_STATUS, val: ~0);
305
306 /*
307 * In practice, Surprise Down errors have been observed to also set
308 * error bits in the Status Register as well as the Fatal Error
309 * Detected bit in the Device Status Register.
310 */
311 pci_write_config_word(dev: pdev, PCI_STATUS, val: 0xffff);
312
313 pcie_capability_write_word(dev: pdev, PCI_EXP_DEVSTA, PCI_EXP_DEVSTA_FED);
314}
315
316static void dpc_handle_surprise_removal(struct pci_dev *pdev)
317{
318 if (!pcie_wait_for_link(pdev, active: false)) {
319 pci_info(pdev, "Data Link Layer Link Active not cleared in 1000 msec\n");
320 goto out;
321 }
322
323 if (pdev->dpc_rp_extensions && dpc_wait_rp_inactive(pdev))
324 goto out;
325
326 pci_aer_raw_clear_status(dev: pdev);
327 pci_clear_surpdn_errors(pdev);
328
329 pci_write_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_STATUS,
330 PCI_EXP_DPC_STATUS_TRIGGER);
331
332out:
333 clear_bit(PCI_DPC_RECOVERED, addr: &pdev->priv_flags);
334 wake_up_all(&dpc_completed_waitqueue);
335}
336
337static bool dpc_is_surprise_removal(struct pci_dev *pdev)
338{
339 u16 status;
340
341 if (!pdev->is_hotplug_bridge)
342 return false;
343
344 if (pci_read_config_word(dev: pdev, where: pdev->aer_cap + PCI_ERR_UNCOR_STATUS,
345 val: &status))
346 return false;
347
348 return status & PCI_ERR_UNC_SURPDN;
349}
350
351static irqreturn_t dpc_handler(int irq, void *context)
352{
353 struct pci_dev *pdev = context;
354
355 /*
356 * According to PCIe r6.0 sec 6.7.6, errors are an expected side effect
357 * of async removal and should be ignored by software.
358 */
359 if (dpc_is_surprise_removal(pdev)) {
360 dpc_handle_surprise_removal(pdev);
361 return IRQ_HANDLED;
362 }
363
364 dpc_process_error(pdev);
365
366 /* We configure DPC so it only triggers on ERR_FATAL */
367 pcie_do_recovery(dev: pdev, state: pci_channel_io_frozen, reset_subordinates: dpc_reset_link);
368
369 return IRQ_HANDLED;
370}
371
372static irqreturn_t dpc_irq(int irq, void *context)
373{
374 struct pci_dev *pdev = context;
375 u16 cap = pdev->dpc_cap, status;
376
377 pci_read_config_word(dev: pdev, where: cap + PCI_EXP_DPC_STATUS, val: &status);
378
379 if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT) || PCI_POSSIBLE_ERROR(status))
380 return IRQ_NONE;
381
382 pci_write_config_word(dev: pdev, where: cap + PCI_EXP_DPC_STATUS,
383 PCI_EXP_DPC_STATUS_INTERRUPT);
384 if (status & PCI_EXP_DPC_STATUS_TRIGGER)
385 return IRQ_WAKE_THREAD;
386 return IRQ_HANDLED;
387}
388
389void pci_dpc_init(struct pci_dev *pdev)
390{
391 u16 cap;
392
393 pdev->dpc_cap = pci_find_ext_capability(dev: pdev, PCI_EXT_CAP_ID_DPC);
394 if (!pdev->dpc_cap)
395 return;
396
397 pci_read_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_CAP, val: &cap);
398 if (!(cap & PCI_EXP_DPC_CAP_RP_EXT))
399 return;
400
401 pdev->dpc_rp_extensions = true;
402
403 /* Quirks may set dpc_rp_log_size if device or firmware is buggy */
404 if (!pdev->dpc_rp_log_size) {
405 pdev->dpc_rp_log_size =
406 FIELD_GET(PCI_EXP_DPC_RP_PIO_LOG_SIZE, cap);
407 if (pdev->dpc_rp_log_size < 4 || pdev->dpc_rp_log_size > 9) {
408 pci_err(pdev, "RP PIO log size %u is invalid\n",
409 pdev->dpc_rp_log_size);
410 pdev->dpc_rp_log_size = 0;
411 }
412 }
413}
414
415#define FLAG(x, y) (((x) & (y)) ? '+' : '-')
416static int dpc_probe(struct pcie_device *dev)
417{
418 struct pci_dev *pdev = dev->port;
419 struct device *device = &dev->device;
420 int status;
421 u16 ctl, cap;
422
423 if (!pcie_aer_is_native(dev: pdev) && !pcie_ports_dpc_native)
424 return -ENOTSUPP;
425
426 status = devm_request_threaded_irq(dev: device, irq: dev->irq, handler: dpc_irq,
427 thread_fn: dpc_handler, IRQF_SHARED,
428 devname: "pcie-dpc", dev_id: pdev);
429 if (status) {
430 pci_warn(pdev, "request IRQ%d failed: %d\n", dev->irq,
431 status);
432 return status;
433 }
434
435 pci_read_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_CAP, val: &cap);
436
437 pci_read_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_CTL, val: &ctl);
438 ctl &= ~PCI_EXP_DPC_CTL_EN_MASK;
439 ctl |= PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN;
440 pci_write_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_CTL, val: ctl);
441
442 pci_info(pdev, "enabled with IRQ %d\n", dev->irq);
443 pci_info(pdev, "error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n",
444 cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT),
445 FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP),
446 FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), pdev->dpc_rp_log_size,
447 FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE));
448
449 pci_add_ext_cap_save_buffer(dev: pdev, PCI_EXT_CAP_ID_DPC, size: sizeof(u16));
450 return status;
451}
452
453static void dpc_remove(struct pcie_device *dev)
454{
455 struct pci_dev *pdev = dev->port;
456 u16 ctl;
457
458 pci_read_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_CTL, val: &ctl);
459 ctl &= ~(PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN);
460 pci_write_config_word(dev: pdev, where: pdev->dpc_cap + PCI_EXP_DPC_CTL, val: ctl);
461}
462
463static struct pcie_port_service_driver dpcdriver = {
464 .name = "dpc",
465 .port_type = PCIE_ANY_PORT,
466 .service = PCIE_PORT_SERVICE_DPC,
467 .probe = dpc_probe,
468 .remove = dpc_remove,
469};
470
471int __init pcie_dpc_init(void)
472{
473 return pcie_port_service_register(new: &dpcdriver);
474}
475

source code of linux/drivers/pci/pcie/dpc.c