xref: /openbmc/linux/drivers/pci/hotplug/pciehp_hpc.c (revision a1117495)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * PCI Express PCI Hot Plug Driver
4  *
5  * Copyright (C) 1995,2001 Compaq Computer Corporation
6  * Copyright (C) 2001 Greg Kroah-Hartman (greg@kroah.com)
7  * Copyright (C) 2001 IBM Corp.
8  * Copyright (C) 2003-2004 Intel Corporation
9  *
10  * All rights reserved.
11  *
12  * Send feedback to <greg@kroah.com>,<kristen.c.accardi@intel.com>
13  */
14 
15 #define dev_fmt(fmt) "pciehp: " fmt
16 
17 #include <linux/dmi.h>
18 #include <linux/kernel.h>
19 #include <linux/types.h>
20 #include <linux/jiffies.h>
21 #include <linux/kthread.h>
22 #include <linux/pci.h>
23 #include <linux/pm_runtime.h>
24 #include <linux/interrupt.h>
25 #include <linux/slab.h>
26 
27 #include "../pci.h"
28 #include "pciehp.h"
29 
30 static const struct dmi_system_id inband_presence_disabled_dmi_table[] = {
31 	/*
32 	 * Match all Dell systems, as some Dell systems have inband
33 	 * presence disabled on NVMe slots (but don't support the bit to
34 	 * report it). Setting inband presence disabled should have no
35 	 * negative effect, except on broken hotplug slots that never
36 	 * assert presence detect--and those will still work, they will
37 	 * just have a bit of extra delay before being probed.
38 	 */
39 	{
40 		.ident = "Dell System",
41 		.matches = {
42 			DMI_MATCH(DMI_OEM_STRING, "Dell System"),
43 		},
44 	},
45 	{}
46 };
47 
48 static inline struct pci_dev *ctrl_dev(struct controller *ctrl)
49 {
50 	return ctrl->pcie->port;
51 }
52 
53 static irqreturn_t pciehp_isr(int irq, void *dev_id);
54 static irqreturn_t pciehp_ist(int irq, void *dev_id);
55 static int pciehp_poll(void *data);
56 
57 static inline int pciehp_request_irq(struct controller *ctrl)
58 {
59 	int retval, irq = ctrl->pcie->irq;
60 
61 	if (pciehp_poll_mode) {
62 		ctrl->poll_thread = kthread_run(&pciehp_poll, ctrl,
63 						"pciehp_poll-%s",
64 						slot_name(ctrl));
65 		return PTR_ERR_OR_ZERO(ctrl->poll_thread);
66 	}
67 
68 	/* Installs the interrupt handler */
69 	retval = request_threaded_irq(irq, pciehp_isr, pciehp_ist,
70 				      IRQF_SHARED, "pciehp", ctrl);
71 	if (retval)
72 		ctrl_err(ctrl, "Cannot get irq %d for the hotplug controller\n",
73 			 irq);
74 	return retval;
75 }
76 
77 static inline void pciehp_free_irq(struct controller *ctrl)
78 {
79 	if (pciehp_poll_mode)
80 		kthread_stop(ctrl->poll_thread);
81 	else
82 		free_irq(ctrl->pcie->irq, ctrl);
83 }
84 
85 static int pcie_poll_cmd(struct controller *ctrl, int timeout)
86 {
87 	struct pci_dev *pdev = ctrl_dev(ctrl);
88 	u16 slot_status;
89 
90 	do {
91 		pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &slot_status);
92 		if (slot_status == (u16) ~0) {
93 			ctrl_info(ctrl, "%s: no response from device\n",
94 				  __func__);
95 			return 0;
96 		}
97 
98 		if (slot_status & PCI_EXP_SLTSTA_CC) {
99 			pcie_capability_write_word(pdev, PCI_EXP_SLTSTA,
100 						   PCI_EXP_SLTSTA_CC);
101 			return 1;
102 		}
103 		msleep(10);
104 		timeout -= 10;
105 	} while (timeout >= 0);
106 	return 0;	/* timeout */
107 }
108 
109 static void pcie_wait_cmd(struct controller *ctrl)
110 {
111 	unsigned int msecs = pciehp_poll_mode ? 2500 : 1000;
112 	unsigned long duration = msecs_to_jiffies(msecs);
113 	unsigned long cmd_timeout = ctrl->cmd_started + duration;
114 	unsigned long now, timeout;
115 	int rc;
116 
117 	/*
118 	 * If the controller does not generate notifications for command
119 	 * completions, we never need to wait between writes.
120 	 */
121 	if (NO_CMD_CMPL(ctrl))
122 		return;
123 
124 	if (!ctrl->cmd_busy)
125 		return;
126 
127 	/*
128 	 * Even if the command has already timed out, we want to call
129 	 * pcie_poll_cmd() so it can clear PCI_EXP_SLTSTA_CC.
130 	 */
131 	now = jiffies;
132 	if (time_before_eq(cmd_timeout, now))
133 		timeout = 1;
134 	else
135 		timeout = cmd_timeout - now;
136 
137 	if (ctrl->slot_ctrl & PCI_EXP_SLTCTL_HPIE &&
138 	    ctrl->slot_ctrl & PCI_EXP_SLTCTL_CCIE)
139 		rc = wait_event_timeout(ctrl->queue, !ctrl->cmd_busy, timeout);
140 	else
141 		rc = pcie_poll_cmd(ctrl, jiffies_to_msecs(timeout));
142 
143 	if (!rc)
144 		ctrl_info(ctrl, "Timeout on hotplug command %#06x (issued %u msec ago)\n",
145 			  ctrl->slot_ctrl,
146 			  jiffies_to_msecs(jiffies - ctrl->cmd_started));
147 }
148 
149 #define CC_ERRATUM_MASK		(PCI_EXP_SLTCTL_PCC |	\
150 				 PCI_EXP_SLTCTL_PIC |	\
151 				 PCI_EXP_SLTCTL_AIC |	\
152 				 PCI_EXP_SLTCTL_EIC)
153 
154 static void pcie_do_write_cmd(struct controller *ctrl, u16 cmd,
155 			      u16 mask, bool wait)
156 {
157 	struct pci_dev *pdev = ctrl_dev(ctrl);
158 	u16 slot_ctrl_orig, slot_ctrl;
159 
160 	mutex_lock(&ctrl->ctrl_lock);
161 
162 	/*
163 	 * Always wait for any previous command that might still be in progress
164 	 */
165 	pcie_wait_cmd(ctrl);
166 
167 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &slot_ctrl);
168 	if (slot_ctrl == (u16) ~0) {
169 		ctrl_info(ctrl, "%s: no response from device\n", __func__);
170 		goto out;
171 	}
172 
173 	slot_ctrl_orig = slot_ctrl;
174 	slot_ctrl &= ~mask;
175 	slot_ctrl |= (cmd & mask);
176 	ctrl->cmd_busy = 1;
177 	smp_mb();
178 	ctrl->slot_ctrl = slot_ctrl;
179 	pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, slot_ctrl);
180 	ctrl->cmd_started = jiffies;
181 
182 	/*
183 	 * Controllers with the Intel CF118 and similar errata advertise
184 	 * Command Completed support, but they only set Command Completed
185 	 * if we change the "Control" bits for power, power indicator,
186 	 * attention indicator, or interlock.  If we only change the
187 	 * "Enable" bits, they never set the Command Completed bit.
188 	 */
189 	if (pdev->broken_cmd_compl &&
190 	    (slot_ctrl_orig & CC_ERRATUM_MASK) == (slot_ctrl & CC_ERRATUM_MASK))
191 		ctrl->cmd_busy = 0;
192 
193 	/*
194 	 * Optionally wait for the hardware to be ready for a new command,
195 	 * indicating completion of the above issued command.
196 	 */
197 	if (wait)
198 		pcie_wait_cmd(ctrl);
199 
200 out:
201 	mutex_unlock(&ctrl->ctrl_lock);
202 }
203 
204 /**
205  * pcie_write_cmd - Issue controller command
206  * @ctrl: controller to which the command is issued
207  * @cmd:  command value written to slot control register
208  * @mask: bitmask of slot control register to be modified
209  */
210 static void pcie_write_cmd(struct controller *ctrl, u16 cmd, u16 mask)
211 {
212 	pcie_do_write_cmd(ctrl, cmd, mask, true);
213 }
214 
215 /* Same as above without waiting for the hardware to latch */
216 static void pcie_write_cmd_nowait(struct controller *ctrl, u16 cmd, u16 mask)
217 {
218 	pcie_do_write_cmd(ctrl, cmd, mask, false);
219 }
220 
221 /**
222  * pciehp_check_link_active() - Is the link active
223  * @ctrl: PCIe hotplug controller
224  *
225  * Check whether the downstream link is currently active. Note it is
226  * possible that the card is removed immediately after this so the
227  * caller may need to take it into account.
228  *
229  * If the hotplug controller itself is not available anymore returns
230  * %-ENODEV.
231  */
232 int pciehp_check_link_active(struct controller *ctrl)
233 {
234 	struct pci_dev *pdev = ctrl_dev(ctrl);
235 	u16 lnk_status;
236 	int ret;
237 
238 	ret = pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
239 	if (ret == PCIBIOS_DEVICE_NOT_FOUND || lnk_status == (u16)~0)
240 		return -ENODEV;
241 
242 	ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
243 	ctrl_dbg(ctrl, "%s: lnk_status = %x\n", __func__, lnk_status);
244 
245 	return ret;
246 }
247 
248 static bool pci_bus_check_dev(struct pci_bus *bus, int devfn)
249 {
250 	u32 l;
251 	int count = 0;
252 	int delay = 1000, step = 20;
253 	bool found = false;
254 
255 	do {
256 		found = pci_bus_read_dev_vendor_id(bus, devfn, &l, 0);
257 		count++;
258 
259 		if (found)
260 			break;
261 
262 		msleep(step);
263 		delay -= step;
264 	} while (delay > 0);
265 
266 	if (count > 1)
267 		pr_debug("pci %04x:%02x:%02x.%d id reading try %d times with interval %d ms to get %08x\n",
268 			pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
269 			PCI_FUNC(devfn), count, step, l);
270 
271 	return found;
272 }
273 
274 static void pcie_wait_for_presence(struct pci_dev *pdev)
275 {
276 	int timeout = 1250;
277 	u16 slot_status;
278 
279 	do {
280 		pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &slot_status);
281 		if (slot_status & PCI_EXP_SLTSTA_PDS)
282 			return;
283 		msleep(10);
284 		timeout -= 10;
285 	} while (timeout > 0);
286 }
287 
288 int pciehp_check_link_status(struct controller *ctrl)
289 {
290 	struct pci_dev *pdev = ctrl_dev(ctrl);
291 	bool found;
292 	u16 lnk_status;
293 
294 	if (!pcie_wait_for_link(pdev, true)) {
295 		ctrl_info(ctrl, "Slot(%s): No link\n", slot_name(ctrl));
296 		return -1;
297 	}
298 
299 	if (ctrl->inband_presence_disabled)
300 		pcie_wait_for_presence(pdev);
301 
302 	found = pci_bus_check_dev(ctrl->pcie->port->subordinate,
303 					PCI_DEVFN(0, 0));
304 
305 	/* ignore link or presence changes up to this point */
306 	if (found)
307 		atomic_and(~(PCI_EXP_SLTSTA_DLLSC | PCI_EXP_SLTSTA_PDC),
308 			   &ctrl->pending_events);
309 
310 	pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lnk_status);
311 	ctrl_dbg(ctrl, "%s: lnk_status = %x\n", __func__, lnk_status);
312 	if ((lnk_status & PCI_EXP_LNKSTA_LT) ||
313 	    !(lnk_status & PCI_EXP_LNKSTA_NLW)) {
314 		ctrl_info(ctrl, "Slot(%s): Cannot train link: status %#06x\n",
315 			  slot_name(ctrl), lnk_status);
316 		return -1;
317 	}
318 
319 	pcie_update_link_speed(ctrl->pcie->port->subordinate, lnk_status);
320 
321 	if (!found) {
322 		ctrl_info(ctrl, "Slot(%s): No device found\n",
323 			  slot_name(ctrl));
324 		return -1;
325 	}
326 
327 	return 0;
328 }
329 
330 static int __pciehp_link_set(struct controller *ctrl, bool enable)
331 {
332 	struct pci_dev *pdev = ctrl_dev(ctrl);
333 	u16 lnk_ctrl;
334 
335 	pcie_capability_read_word(pdev, PCI_EXP_LNKCTL, &lnk_ctrl);
336 
337 	if (enable)
338 		lnk_ctrl &= ~PCI_EXP_LNKCTL_LD;
339 	else
340 		lnk_ctrl |= PCI_EXP_LNKCTL_LD;
341 
342 	pcie_capability_write_word(pdev, PCI_EXP_LNKCTL, lnk_ctrl);
343 	ctrl_dbg(ctrl, "%s: lnk_ctrl = %x\n", __func__, lnk_ctrl);
344 	return 0;
345 }
346 
347 static int pciehp_link_enable(struct controller *ctrl)
348 {
349 	return __pciehp_link_set(ctrl, true);
350 }
351 
352 int pciehp_get_raw_indicator_status(struct hotplug_slot *hotplug_slot,
353 				    u8 *status)
354 {
355 	struct controller *ctrl = to_ctrl(hotplug_slot);
356 	struct pci_dev *pdev = ctrl_dev(ctrl);
357 	u16 slot_ctrl;
358 
359 	pci_config_pm_runtime_get(pdev);
360 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &slot_ctrl);
361 	pci_config_pm_runtime_put(pdev);
362 	*status = (slot_ctrl & (PCI_EXP_SLTCTL_AIC | PCI_EXP_SLTCTL_PIC)) >> 6;
363 	return 0;
364 }
365 
366 int pciehp_get_attention_status(struct hotplug_slot *hotplug_slot, u8 *status)
367 {
368 	struct controller *ctrl = to_ctrl(hotplug_slot);
369 	struct pci_dev *pdev = ctrl_dev(ctrl);
370 	u16 slot_ctrl;
371 
372 	pci_config_pm_runtime_get(pdev);
373 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &slot_ctrl);
374 	pci_config_pm_runtime_put(pdev);
375 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x, value read %x\n", __func__,
376 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL, slot_ctrl);
377 
378 	switch (slot_ctrl & PCI_EXP_SLTCTL_AIC) {
379 	case PCI_EXP_SLTCTL_ATTN_IND_ON:
380 		*status = 1;	/* On */
381 		break;
382 	case PCI_EXP_SLTCTL_ATTN_IND_BLINK:
383 		*status = 2;	/* Blink */
384 		break;
385 	case PCI_EXP_SLTCTL_ATTN_IND_OFF:
386 		*status = 0;	/* Off */
387 		break;
388 	default:
389 		*status = 0xFF;
390 		break;
391 	}
392 
393 	return 0;
394 }
395 
396 void pciehp_get_power_status(struct controller *ctrl, u8 *status)
397 {
398 	struct pci_dev *pdev = ctrl_dev(ctrl);
399 	u16 slot_ctrl;
400 
401 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &slot_ctrl);
402 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x value read %x\n", __func__,
403 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL, slot_ctrl);
404 
405 	switch (slot_ctrl & PCI_EXP_SLTCTL_PCC) {
406 	case PCI_EXP_SLTCTL_PWR_ON:
407 		*status = 1;	/* On */
408 		break;
409 	case PCI_EXP_SLTCTL_PWR_OFF:
410 		*status = 0;	/* Off */
411 		break;
412 	default:
413 		*status = 0xFF;
414 		break;
415 	}
416 }
417 
418 void pciehp_get_latch_status(struct controller *ctrl, u8 *status)
419 {
420 	struct pci_dev *pdev = ctrl_dev(ctrl);
421 	u16 slot_status;
422 
423 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &slot_status);
424 	*status = !!(slot_status & PCI_EXP_SLTSTA_MRLSS);
425 }
426 
427 /**
428  * pciehp_card_present() - Is the card present
429  * @ctrl: PCIe hotplug controller
430  *
431  * Function checks whether the card is currently present in the slot and
432  * in that case returns true. Note it is possible that the card is
433  * removed immediately after the check so the caller may need to take
434  * this into account.
435  *
436  * It the hotplug controller itself is not available anymore returns
437  * %-ENODEV.
438  */
439 int pciehp_card_present(struct controller *ctrl)
440 {
441 	struct pci_dev *pdev = ctrl_dev(ctrl);
442 	u16 slot_status;
443 	int ret;
444 
445 	ret = pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &slot_status);
446 	if (ret == PCIBIOS_DEVICE_NOT_FOUND || slot_status == (u16)~0)
447 		return -ENODEV;
448 
449 	return !!(slot_status & PCI_EXP_SLTSTA_PDS);
450 }
451 
452 /**
453  * pciehp_card_present_or_link_active() - whether given slot is occupied
454  * @ctrl: PCIe hotplug controller
455  *
456  * Unlike pciehp_card_present(), which determines presence solely from the
457  * Presence Detect State bit, this helper also returns true if the Link Active
458  * bit is set.  This is a concession to broken hotplug ports which hardwire
459  * Presence Detect State to zero, such as Wilocity's [1ae9:0200].
460  *
461  * Returns: %1 if the slot is occupied and %0 if it is not. If the hotplug
462  *	    port is not present anymore returns %-ENODEV.
463  */
464 int pciehp_card_present_or_link_active(struct controller *ctrl)
465 {
466 	int ret;
467 
468 	ret = pciehp_card_present(ctrl);
469 	if (ret)
470 		return ret;
471 
472 	return pciehp_check_link_active(ctrl);
473 }
474 
475 int pciehp_query_power_fault(struct controller *ctrl)
476 {
477 	struct pci_dev *pdev = ctrl_dev(ctrl);
478 	u16 slot_status;
479 
480 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &slot_status);
481 	return !!(slot_status & PCI_EXP_SLTSTA_PFD);
482 }
483 
484 int pciehp_set_raw_indicator_status(struct hotplug_slot *hotplug_slot,
485 				    u8 status)
486 {
487 	struct controller *ctrl = to_ctrl(hotplug_slot);
488 	struct pci_dev *pdev = ctrl_dev(ctrl);
489 
490 	pci_config_pm_runtime_get(pdev);
491 	pcie_write_cmd_nowait(ctrl, status << 6,
492 			      PCI_EXP_SLTCTL_AIC | PCI_EXP_SLTCTL_PIC);
493 	pci_config_pm_runtime_put(pdev);
494 	return 0;
495 }
496 
497 /**
498  * pciehp_set_indicators() - set attention indicator, power indicator, or both
499  * @ctrl: PCIe hotplug controller
500  * @pwr: one of:
501  *	PCI_EXP_SLTCTL_PWR_IND_ON
502  *	PCI_EXP_SLTCTL_PWR_IND_BLINK
503  *	PCI_EXP_SLTCTL_PWR_IND_OFF
504  * @attn: one of:
505  *	PCI_EXP_SLTCTL_ATTN_IND_ON
506  *	PCI_EXP_SLTCTL_ATTN_IND_BLINK
507  *	PCI_EXP_SLTCTL_ATTN_IND_OFF
508  *
509  * Either @pwr or @attn can also be INDICATOR_NOOP to leave that indicator
510  * unchanged.
511  */
512 void pciehp_set_indicators(struct controller *ctrl, int pwr, int attn)
513 {
514 	u16 cmd = 0, mask = 0;
515 
516 	if (PWR_LED(ctrl) && pwr != INDICATOR_NOOP) {
517 		cmd |= (pwr & PCI_EXP_SLTCTL_PIC);
518 		mask |= PCI_EXP_SLTCTL_PIC;
519 	}
520 
521 	if (ATTN_LED(ctrl) && attn != INDICATOR_NOOP) {
522 		cmd |= (attn & PCI_EXP_SLTCTL_AIC);
523 		mask |= PCI_EXP_SLTCTL_AIC;
524 	}
525 
526 	if (cmd) {
527 		pcie_write_cmd_nowait(ctrl, cmd, mask);
528 		ctrl_dbg(ctrl, "%s: SLOTCTRL %x write cmd %x\n", __func__,
529 			 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL, cmd);
530 	}
531 }
532 
533 int pciehp_power_on_slot(struct controller *ctrl)
534 {
535 	struct pci_dev *pdev = ctrl_dev(ctrl);
536 	u16 slot_status;
537 	int retval;
538 
539 	/* Clear power-fault bit from previous power failures */
540 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &slot_status);
541 	if (slot_status & PCI_EXP_SLTSTA_PFD)
542 		pcie_capability_write_word(pdev, PCI_EXP_SLTSTA,
543 					   PCI_EXP_SLTSTA_PFD);
544 	ctrl->power_fault_detected = 0;
545 
546 	pcie_write_cmd(ctrl, PCI_EXP_SLTCTL_PWR_ON, PCI_EXP_SLTCTL_PCC);
547 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x write cmd %x\n", __func__,
548 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL,
549 		 PCI_EXP_SLTCTL_PWR_ON);
550 
551 	retval = pciehp_link_enable(ctrl);
552 	if (retval)
553 		ctrl_err(ctrl, "%s: Can not enable the link!\n", __func__);
554 
555 	return retval;
556 }
557 
558 void pciehp_power_off_slot(struct controller *ctrl)
559 {
560 	pcie_write_cmd(ctrl, PCI_EXP_SLTCTL_PWR_OFF, PCI_EXP_SLTCTL_PCC);
561 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x write cmd %x\n", __func__,
562 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL,
563 		 PCI_EXP_SLTCTL_PWR_OFF);
564 }
565 
566 static void pciehp_ignore_dpc_link_change(struct controller *ctrl,
567 					  struct pci_dev *pdev, int irq)
568 {
569 	/*
570 	 * Ignore link changes which occurred while waiting for DPC recovery.
571 	 * Could be several if DPC triggered multiple times consecutively.
572 	 */
573 	synchronize_hardirq(irq);
574 	atomic_and(~PCI_EXP_SLTSTA_DLLSC, &ctrl->pending_events);
575 	if (pciehp_poll_mode)
576 		pcie_capability_write_word(pdev, PCI_EXP_SLTSTA,
577 					   PCI_EXP_SLTSTA_DLLSC);
578 	ctrl_info(ctrl, "Slot(%s): Link Down/Up ignored (recovered by DPC)\n",
579 		  slot_name(ctrl));
580 
581 	/*
582 	 * If the link is unexpectedly down after successful recovery,
583 	 * the corresponding link change may have been ignored above.
584 	 * Synthesize it to ensure that it is acted on.
585 	 */
586 	down_read(&ctrl->reset_lock);
587 	if (!pciehp_check_link_active(ctrl))
588 		pciehp_request(ctrl, PCI_EXP_SLTSTA_DLLSC);
589 	up_read(&ctrl->reset_lock);
590 }
591 
592 static irqreturn_t pciehp_isr(int irq, void *dev_id)
593 {
594 	struct controller *ctrl = (struct controller *)dev_id;
595 	struct pci_dev *pdev = ctrl_dev(ctrl);
596 	struct device *parent = pdev->dev.parent;
597 	u16 status, events = 0;
598 
599 	/*
600 	 * Interrupts only occur in D3hot or shallower and only if enabled
601 	 * in the Slot Control register (PCIe r4.0, sec 6.7.3.4).
602 	 */
603 	if (pdev->current_state == PCI_D3cold ||
604 	    (!(ctrl->slot_ctrl & PCI_EXP_SLTCTL_HPIE) && !pciehp_poll_mode))
605 		return IRQ_NONE;
606 
607 	/*
608 	 * Keep the port accessible by holding a runtime PM ref on its parent.
609 	 * Defer resume of the parent to the IRQ thread if it's suspended.
610 	 * Mask the interrupt until then.
611 	 */
612 	if (parent) {
613 		pm_runtime_get_noresume(parent);
614 		if (!pm_runtime_active(parent)) {
615 			pm_runtime_put(parent);
616 			disable_irq_nosync(irq);
617 			atomic_or(RERUN_ISR, &ctrl->pending_events);
618 			return IRQ_WAKE_THREAD;
619 		}
620 	}
621 
622 read_status:
623 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &status);
624 	if (status == (u16) ~0) {
625 		ctrl_info(ctrl, "%s: no response from device\n", __func__);
626 		if (parent)
627 			pm_runtime_put(parent);
628 		return IRQ_NONE;
629 	}
630 
631 	/*
632 	 * Slot Status contains plain status bits as well as event
633 	 * notification bits; right now we only want the event bits.
634 	 */
635 	status &= PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD |
636 		  PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_CC |
637 		  PCI_EXP_SLTSTA_DLLSC;
638 
639 	/*
640 	 * If we've already reported a power fault, don't report it again
641 	 * until we've done something to handle it.
642 	 */
643 	if (ctrl->power_fault_detected)
644 		status &= ~PCI_EXP_SLTSTA_PFD;
645 
646 	events |= status;
647 	if (!events) {
648 		if (parent)
649 			pm_runtime_put(parent);
650 		return IRQ_NONE;
651 	}
652 
653 	if (status) {
654 		pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, events);
655 
656 		/*
657 		 * In MSI mode, all event bits must be zero before the port
658 		 * will send a new interrupt (PCIe Base Spec r5.0 sec 6.7.3.4).
659 		 * So re-read the Slot Status register in case a bit was set
660 		 * between read and write.
661 		 */
662 		if (pci_dev_msi_enabled(pdev) && !pciehp_poll_mode)
663 			goto read_status;
664 	}
665 
666 	ctrl_dbg(ctrl, "pending interrupts %#06x from Slot Status\n", events);
667 	if (parent)
668 		pm_runtime_put(parent);
669 
670 	/*
671 	 * Command Completed notifications are not deferred to the
672 	 * IRQ thread because it may be waiting for their arrival.
673 	 */
674 	if (events & PCI_EXP_SLTSTA_CC) {
675 		ctrl->cmd_busy = 0;
676 		smp_mb();
677 		wake_up(&ctrl->queue);
678 
679 		if (events == PCI_EXP_SLTSTA_CC)
680 			return IRQ_HANDLED;
681 
682 		events &= ~PCI_EXP_SLTSTA_CC;
683 	}
684 
685 	if (pdev->ignore_hotplug) {
686 		ctrl_dbg(ctrl, "ignoring hotplug event %#06x\n", events);
687 		return IRQ_HANDLED;
688 	}
689 
690 	/* Save pending events for consumption by IRQ thread. */
691 	atomic_or(events, &ctrl->pending_events);
692 	return IRQ_WAKE_THREAD;
693 }
694 
695 static irqreturn_t pciehp_ist(int irq, void *dev_id)
696 {
697 	struct controller *ctrl = (struct controller *)dev_id;
698 	struct pci_dev *pdev = ctrl_dev(ctrl);
699 	irqreturn_t ret;
700 	u32 events;
701 
702 	ctrl->ist_running = true;
703 	pci_config_pm_runtime_get(pdev);
704 
705 	/* rerun pciehp_isr() if the port was inaccessible on interrupt */
706 	if (atomic_fetch_and(~RERUN_ISR, &ctrl->pending_events) & RERUN_ISR) {
707 		ret = pciehp_isr(irq, dev_id);
708 		enable_irq(irq);
709 		if (ret != IRQ_WAKE_THREAD)
710 			goto out;
711 	}
712 
713 	synchronize_hardirq(irq);
714 	events = atomic_xchg(&ctrl->pending_events, 0);
715 	if (!events) {
716 		ret = IRQ_NONE;
717 		goto out;
718 	}
719 
720 	/* Check Attention Button Pressed */
721 	if (events & PCI_EXP_SLTSTA_ABP) {
722 		ctrl_info(ctrl, "Slot(%s): Attention button pressed\n",
723 			  slot_name(ctrl));
724 		pciehp_handle_button_press(ctrl);
725 	}
726 
727 	/* Check Power Fault Detected */
728 	if ((events & PCI_EXP_SLTSTA_PFD) && !ctrl->power_fault_detected) {
729 		ctrl->power_fault_detected = 1;
730 		ctrl_err(ctrl, "Slot(%s): Power fault\n", slot_name(ctrl));
731 		pciehp_set_indicators(ctrl, PCI_EXP_SLTCTL_PWR_IND_OFF,
732 				      PCI_EXP_SLTCTL_ATTN_IND_ON);
733 	}
734 
735 	/*
736 	 * Ignore Link Down/Up events caused by Downstream Port Containment
737 	 * if recovery from the error succeeded.
738 	 */
739 	if ((events & PCI_EXP_SLTSTA_DLLSC) && pci_dpc_recovered(pdev) &&
740 	    ctrl->state == ON_STATE) {
741 		events &= ~PCI_EXP_SLTSTA_DLLSC;
742 		pciehp_ignore_dpc_link_change(ctrl, pdev, irq);
743 	}
744 
745 	/*
746 	 * Disable requests have higher priority than Presence Detect Changed
747 	 * or Data Link Layer State Changed events.
748 	 */
749 	down_read(&ctrl->reset_lock);
750 	if (events & DISABLE_SLOT)
751 		pciehp_handle_disable_request(ctrl);
752 	else if (events & (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC))
753 		pciehp_handle_presence_or_link_change(ctrl, events);
754 	up_read(&ctrl->reset_lock);
755 
756 	ret = IRQ_HANDLED;
757 out:
758 	pci_config_pm_runtime_put(pdev);
759 	ctrl->ist_running = false;
760 	wake_up(&ctrl->requester);
761 	return ret;
762 }
763 
764 static int pciehp_poll(void *data)
765 {
766 	struct controller *ctrl = data;
767 
768 	schedule_timeout_idle(10 * HZ); /* start with 10 sec delay */
769 
770 	while (!kthread_should_stop()) {
771 		/* poll for interrupt events or user requests */
772 		while (pciehp_isr(IRQ_NOTCONNECTED, ctrl) == IRQ_WAKE_THREAD ||
773 		       atomic_read(&ctrl->pending_events))
774 			pciehp_ist(IRQ_NOTCONNECTED, ctrl);
775 
776 		if (pciehp_poll_time <= 0 || pciehp_poll_time > 60)
777 			pciehp_poll_time = 2; /* clamp to sane value */
778 
779 		schedule_timeout_idle(pciehp_poll_time * HZ);
780 	}
781 
782 	return 0;
783 }
784 
785 static void pcie_enable_notification(struct controller *ctrl)
786 {
787 	u16 cmd, mask;
788 
789 	/*
790 	 * TBD: Power fault detected software notification support.
791 	 *
792 	 * Power fault detected software notification is not enabled
793 	 * now, because it caused power fault detected interrupt storm
794 	 * on some machines. On those machines, power fault detected
795 	 * bit in the slot status register was set again immediately
796 	 * when it is cleared in the interrupt service routine, and
797 	 * next power fault detected interrupt was notified again.
798 	 */
799 
800 	/*
801 	 * Always enable link events: thus link-up and link-down shall
802 	 * always be treated as hotplug and unplug respectively. Enable
803 	 * presence detect only if Attention Button is not present.
804 	 */
805 	cmd = PCI_EXP_SLTCTL_DLLSCE;
806 	if (ATTN_BUTTN(ctrl))
807 		cmd |= PCI_EXP_SLTCTL_ABPE;
808 	else
809 		cmd |= PCI_EXP_SLTCTL_PDCE;
810 	if (!pciehp_poll_mode)
811 		cmd |= PCI_EXP_SLTCTL_HPIE | PCI_EXP_SLTCTL_CCIE;
812 
813 	mask = (PCI_EXP_SLTCTL_PDCE | PCI_EXP_SLTCTL_ABPE |
814 		PCI_EXP_SLTCTL_PFDE |
815 		PCI_EXP_SLTCTL_HPIE | PCI_EXP_SLTCTL_CCIE |
816 		PCI_EXP_SLTCTL_DLLSCE);
817 
818 	pcie_write_cmd_nowait(ctrl, cmd, mask);
819 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x write cmd %x\n", __func__,
820 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL, cmd);
821 }
822 
823 static void pcie_disable_notification(struct controller *ctrl)
824 {
825 	u16 mask;
826 
827 	mask = (PCI_EXP_SLTCTL_PDCE | PCI_EXP_SLTCTL_ABPE |
828 		PCI_EXP_SLTCTL_MRLSCE | PCI_EXP_SLTCTL_PFDE |
829 		PCI_EXP_SLTCTL_HPIE | PCI_EXP_SLTCTL_CCIE |
830 		PCI_EXP_SLTCTL_DLLSCE);
831 	pcie_write_cmd(ctrl, 0, mask);
832 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x write cmd %x\n", __func__,
833 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL, 0);
834 }
835 
836 void pcie_clear_hotplug_events(struct controller *ctrl)
837 {
838 	pcie_capability_write_word(ctrl_dev(ctrl), PCI_EXP_SLTSTA,
839 				   PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
840 }
841 
842 void pcie_enable_interrupt(struct controller *ctrl)
843 {
844 	u16 mask;
845 
846 	mask = PCI_EXP_SLTCTL_HPIE | PCI_EXP_SLTCTL_DLLSCE;
847 	pcie_write_cmd(ctrl, mask, mask);
848 }
849 
850 void pcie_disable_interrupt(struct controller *ctrl)
851 {
852 	u16 mask;
853 
854 	/*
855 	 * Mask hot-plug interrupt to prevent it triggering immediately
856 	 * when the link goes inactive (we still get PME when any of the
857 	 * enabled events is detected). Same goes with Link Layer State
858 	 * changed event which generates PME immediately when the link goes
859 	 * inactive so mask it as well.
860 	 */
861 	mask = PCI_EXP_SLTCTL_HPIE | PCI_EXP_SLTCTL_DLLSCE;
862 	pcie_write_cmd(ctrl, 0, mask);
863 }
864 
865 /**
866  * pciehp_slot_reset() - ignore link event caused by error-induced hot reset
867  * @dev: PCI Express port service device
868  *
869  * Called from pcie_portdrv_slot_reset() after AER or DPC initiated a reset
870  * further up in the hierarchy to recover from an error.  The reset was
871  * propagated down to this hotplug port.  Ignore the resulting link flap.
872  * If the link failed to retrain successfully, synthesize the ignored event.
873  * Surprise removal during reset is detected through Presence Detect Changed.
874  */
875 int pciehp_slot_reset(struct pcie_device *dev)
876 {
877 	struct controller *ctrl = get_service_data(dev);
878 
879 	if (ctrl->state != ON_STATE)
880 		return 0;
881 
882 	pcie_capability_write_word(dev->port, PCI_EXP_SLTSTA,
883 				   PCI_EXP_SLTSTA_DLLSC);
884 
885 	if (!pciehp_check_link_active(ctrl))
886 		pciehp_request(ctrl, PCI_EXP_SLTSTA_DLLSC);
887 
888 	return 0;
889 }
890 
891 /*
892  * pciehp has a 1:1 bus:slot relationship so we ultimately want a secondary
893  * bus reset of the bridge, but at the same time we want to ensure that it is
894  * not seen as a hot-unplug, followed by the hot-plug of the device. Thus,
895  * disable link state notification and presence detection change notification
896  * momentarily, if we see that they could interfere. Also, clear any spurious
897  * events after.
898  */
899 int pciehp_reset_slot(struct hotplug_slot *hotplug_slot, bool probe)
900 {
901 	struct controller *ctrl = to_ctrl(hotplug_slot);
902 	struct pci_dev *pdev = ctrl_dev(ctrl);
903 	u16 stat_mask = 0, ctrl_mask = 0;
904 	int rc;
905 
906 	if (probe)
907 		return 0;
908 
909 	down_write(&ctrl->reset_lock);
910 
911 	if (!ATTN_BUTTN(ctrl)) {
912 		ctrl_mask |= PCI_EXP_SLTCTL_PDCE;
913 		stat_mask |= PCI_EXP_SLTSTA_PDC;
914 	}
915 	ctrl_mask |= PCI_EXP_SLTCTL_DLLSCE;
916 	stat_mask |= PCI_EXP_SLTSTA_DLLSC;
917 
918 	pcie_write_cmd(ctrl, 0, ctrl_mask);
919 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x write cmd %x\n", __func__,
920 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL, 0);
921 
922 	rc = pci_bridge_secondary_bus_reset(ctrl->pcie->port);
923 
924 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, stat_mask);
925 	pcie_write_cmd_nowait(ctrl, ctrl_mask, ctrl_mask);
926 	ctrl_dbg(ctrl, "%s: SLOTCTRL %x write cmd %x\n", __func__,
927 		 pci_pcie_cap(ctrl->pcie->port) + PCI_EXP_SLTCTL, ctrl_mask);
928 
929 	up_write(&ctrl->reset_lock);
930 	return rc;
931 }
932 
933 int pcie_init_notification(struct controller *ctrl)
934 {
935 	if (pciehp_request_irq(ctrl))
936 		return -1;
937 	pcie_enable_notification(ctrl);
938 	ctrl->notification_enabled = 1;
939 	return 0;
940 }
941 
942 void pcie_shutdown_notification(struct controller *ctrl)
943 {
944 	if (ctrl->notification_enabled) {
945 		pcie_disable_notification(ctrl);
946 		pciehp_free_irq(ctrl);
947 		ctrl->notification_enabled = 0;
948 	}
949 }
950 
951 static inline void dbg_ctrl(struct controller *ctrl)
952 {
953 	struct pci_dev *pdev = ctrl->pcie->port;
954 	u16 reg16;
955 
956 	ctrl_dbg(ctrl, "Slot Capabilities      : 0x%08x\n", ctrl->slot_cap);
957 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &reg16);
958 	ctrl_dbg(ctrl, "Slot Status            : 0x%04x\n", reg16);
959 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &reg16);
960 	ctrl_dbg(ctrl, "Slot Control           : 0x%04x\n", reg16);
961 }
962 
963 #define FLAG(x, y)	(((x) & (y)) ? '+' : '-')
964 
965 struct controller *pcie_init(struct pcie_device *dev)
966 {
967 	struct controller *ctrl;
968 	u32 slot_cap, slot_cap2, link_cap;
969 	u8 poweron;
970 	struct pci_dev *pdev = dev->port;
971 	struct pci_bus *subordinate = pdev->subordinate;
972 
973 	ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL);
974 	if (!ctrl)
975 		return NULL;
976 
977 	ctrl->pcie = dev;
978 	pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, &slot_cap);
979 
980 	if (pdev->hotplug_user_indicators)
981 		slot_cap &= ~(PCI_EXP_SLTCAP_AIP | PCI_EXP_SLTCAP_PIP);
982 
983 	/*
984 	 * We assume no Thunderbolt controllers support Command Complete events,
985 	 * but some controllers falsely claim they do.
986 	 */
987 	if (pdev->is_thunderbolt)
988 		slot_cap |= PCI_EXP_SLTCAP_NCCS;
989 
990 	ctrl->slot_cap = slot_cap;
991 	mutex_init(&ctrl->ctrl_lock);
992 	mutex_init(&ctrl->state_lock);
993 	init_rwsem(&ctrl->reset_lock);
994 	init_waitqueue_head(&ctrl->requester);
995 	init_waitqueue_head(&ctrl->queue);
996 	INIT_DELAYED_WORK(&ctrl->button_work, pciehp_queue_pushbutton_work);
997 	dbg_ctrl(ctrl);
998 
999 	down_read(&pci_bus_sem);
1000 	ctrl->state = list_empty(&subordinate->devices) ? OFF_STATE : ON_STATE;
1001 	up_read(&pci_bus_sem);
1002 
1003 	pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP2, &slot_cap2);
1004 	if (slot_cap2 & PCI_EXP_SLTCAP2_IBPD) {
1005 		pcie_write_cmd_nowait(ctrl, PCI_EXP_SLTCTL_IBPD_DISABLE,
1006 				      PCI_EXP_SLTCTL_IBPD_DISABLE);
1007 		ctrl->inband_presence_disabled = 1;
1008 	}
1009 
1010 	if (dmi_first_match(inband_presence_disabled_dmi_table))
1011 		ctrl->inband_presence_disabled = 1;
1012 
1013 	/* Check if Data Link Layer Link Active Reporting is implemented */
1014 	pcie_capability_read_dword(pdev, PCI_EXP_LNKCAP, &link_cap);
1015 
1016 	/* Clear all remaining event bits in Slot Status register. */
1017 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA,
1018 		PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD |
1019 		PCI_EXP_SLTSTA_MRLSC | PCI_EXP_SLTSTA_CC |
1020 		PCI_EXP_SLTSTA_DLLSC | PCI_EXP_SLTSTA_PDC);
1021 
1022 	ctrl_info(ctrl, "Slot #%d AttnBtn%c PwrCtrl%c MRL%c AttnInd%c PwrInd%c HotPlug%c Surprise%c Interlock%c NoCompl%c IbPresDis%c LLActRep%c%s\n",
1023 		(slot_cap & PCI_EXP_SLTCAP_PSN) >> 19,
1024 		FLAG(slot_cap, PCI_EXP_SLTCAP_ABP),
1025 		FLAG(slot_cap, PCI_EXP_SLTCAP_PCP),
1026 		FLAG(slot_cap, PCI_EXP_SLTCAP_MRLSP),
1027 		FLAG(slot_cap, PCI_EXP_SLTCAP_AIP),
1028 		FLAG(slot_cap, PCI_EXP_SLTCAP_PIP),
1029 		FLAG(slot_cap, PCI_EXP_SLTCAP_HPC),
1030 		FLAG(slot_cap, PCI_EXP_SLTCAP_HPS),
1031 		FLAG(slot_cap, PCI_EXP_SLTCAP_EIP),
1032 		FLAG(slot_cap, PCI_EXP_SLTCAP_NCCS),
1033 		FLAG(slot_cap2, PCI_EXP_SLTCAP2_IBPD),
1034 		FLAG(link_cap, PCI_EXP_LNKCAP_DLLLARC),
1035 		pdev->broken_cmd_compl ? " (with Cmd Compl erratum)" : "");
1036 
1037 	/*
1038 	 * If empty slot's power status is on, turn power off.  The IRQ isn't
1039 	 * requested yet, so avoid triggering a notification with this command.
1040 	 */
1041 	if (POWER_CTRL(ctrl)) {
1042 		pciehp_get_power_status(ctrl, &poweron);
1043 		if (!pciehp_card_present_or_link_active(ctrl) && poweron) {
1044 			pcie_disable_notification(ctrl);
1045 			pciehp_power_off_slot(ctrl);
1046 		}
1047 	}
1048 
1049 	return ctrl;
1050 }
1051 
1052 void pciehp_release_ctrl(struct controller *ctrl)
1053 {
1054 	cancel_delayed_work_sync(&ctrl->button_work);
1055 	kfree(ctrl);
1056 }
1057 
1058 static void quirk_cmd_compl(struct pci_dev *pdev)
1059 {
1060 	u32 slot_cap;
1061 
1062 	if (pci_is_pcie(pdev)) {
1063 		pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, &slot_cap);
1064 		if (slot_cap & PCI_EXP_SLTCAP_HPC &&
1065 		    !(slot_cap & PCI_EXP_SLTCAP_NCCS))
1066 			pdev->broken_cmd_compl = 1;
1067 	}
1068 }
1069 DECLARE_PCI_FIXUP_CLASS_EARLY(PCI_VENDOR_ID_INTEL, PCI_ANY_ID,
1070 			      PCI_CLASS_BRIDGE_PCI, 8, quirk_cmd_compl);
1071 DECLARE_PCI_FIXUP_CLASS_EARLY(PCI_VENDOR_ID_QCOM, 0x0400,
1072 			      PCI_CLASS_BRIDGE_PCI, 8, quirk_cmd_compl);
1073 DECLARE_PCI_FIXUP_CLASS_EARLY(PCI_VENDOR_ID_QCOM, 0x0401,
1074 			      PCI_CLASS_BRIDGE_PCI, 8, quirk_cmd_compl);
1075 DECLARE_PCI_FIXUP_CLASS_EARLY(PCI_VENDOR_ID_HXT, 0x0401,
1076 			      PCI_CLASS_BRIDGE_PCI, 8, quirk_cmd_compl);
1077