xref: /NextBSD/sys/x86/xen/xen_intr.c (revision 4557fabb34e865d7f40be64b39c9e34fa41dbb60)
1 /******************************************************************************
2  * xen_intr.c
3  *
4  * Xen event and interrupt services for x86 HVM guests.
5  *
6  * Copyright (c) 2002-2005, K A Fraser
7  * Copyright (c) 2005, Intel Corporation <xiaofeng.ling@intel.com>
8  * Copyright (c) 2012, Spectra Logic Corporation
9  *
10  * This file may be distributed separately from the Linux kernel, or
11  * incorporated into other software packages, subject to the following license:
12  *
13  * Permission is hereby granted, free of charge, to any person obtaining a copy
14  * of this source file (the "Software"), to deal in the Software without
15  * restriction, including without limitation the rights to use, copy, modify,
16  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17  * and to permit persons to whom the Software is furnished to do so, subject to
18  * the following conditions:
19  *
20  * The above copyright notice and this permission notice shall be included in
21  * all copies or substantial portions of the Software.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29  * IN THE SOFTWARE.
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_ddb.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/bus.h>
40 #include <sys/malloc.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/interrupt.h>
46 #include <sys/pcpu.h>
47 #include <sys/smp.h>
48 
49 #include <vm/vm.h>
50 #include <vm/pmap.h>
51 
52 #include <machine/intr_machdep.h>
53 #include <x86/apicvar.h>
54 #include <x86/apicreg.h>
55 #include <machine/smp.h>
56 #include <machine/stdarg.h>
57 
58 #include <machine/xen/synch_bitops.h>
59 #include <machine/xen/xen-os.h>
60 
61 #include <xen/hypervisor.h>
62 #include <xen/xen_intr.h>
63 #include <xen/evtchn/evtchnvar.h>
64 
65 #include <dev/xen/xenpci/xenpcivar.h>
66 #include <dev/pci/pcivar.h>
67 
68 #ifdef DDB
69 #include <ddb/ddb.h>
70 #endif
71 
72 static MALLOC_DEFINE(M_XENINTR, "xen_intr", "Xen Interrupt Services");
73 
74 /**
75  * Per-cpu event channel processing state.
76  */
77 struct xen_intr_pcpu_data {
78 	/**
79 	 * The last event channel bitmap section (level one bit) processed.
80 	 * This is used to ensure we scan all ports before
81 	 * servicing an already servied port again.
82 	 */
83 	u_int	last_processed_l1i;
84 
85 	/**
86 	 * The last event channel processed within the event channel
87 	 * bitmap being scanned.
88 	 */
89 	u_int	last_processed_l2i;
90 
91 	/** Pointer to this CPU's interrupt statistic counter. */
92 	u_long *evtchn_intrcnt;
93 
94 	/**
95 	 * A bitmap of ports that can be serviced from this CPU.
96 	 * A set bit means interrupt handling is enabled.
97 	 */
98 	u_long	evtchn_enabled[sizeof(u_long) * 8];
99 };
100 
101 /*
102  * Start the scan at port 0 by initializing the last scanned
103  * location as the highest numbered event channel port.
104  */
105 static DPCPU_DEFINE(struct xen_intr_pcpu_data, xen_intr_pcpu) = {
106 	.last_processed_l1i = LONG_BIT - 1,
107 	.last_processed_l2i = LONG_BIT - 1
108 };
109 
110 DPCPU_DECLARE(struct vcpu_info *, vcpu_info);
111 
112 #define	XEN_EEXIST		17 /* Xen "already exists" error */
113 #define	XEN_ALLOCATE_VECTOR	0 /* Allocate a vector for this event channel */
114 #define	XEN_INVALID_EVTCHN	0 /* Invalid event channel */
115 
116 #define	is_valid_evtchn(x)	((x) != XEN_INVALID_EVTCHN)
117 
118 struct xenisrc {
119 	struct intsrc	xi_intsrc;
120 	enum evtchn_type xi_type;
121 	int		xi_cpu;		/* VCPU for delivery. */
122 	int		xi_vector;	/* Global isrc vector number. */
123 	evtchn_port_t	xi_port;
124 	int		xi_pirq;
125 	int		xi_virq;
126 	void		*xi_cookie;
127 	u_int		xi_close:1;	/* close on unbind? */
128 	u_int		xi_activehi:1;
129 	u_int		xi_edgetrigger:1;
130 	u_int		xi_masked:1;
131 };
132 
133 #define ARRAY_SIZE(a)	(sizeof(a) / sizeof(a[0]))
134 
135 static void	xen_intr_suspend(struct pic *);
136 static void	xen_intr_resume(struct pic *, bool suspend_cancelled);
137 static void	xen_intr_enable_source(struct intsrc *isrc);
138 static void	xen_intr_disable_source(struct intsrc *isrc, int eoi);
139 static void	xen_intr_eoi_source(struct intsrc *isrc);
140 static void	xen_intr_enable_intr(struct intsrc *isrc);
141 static void	xen_intr_disable_intr(struct intsrc *isrc);
142 static int	xen_intr_vector(struct intsrc *isrc);
143 static int	xen_intr_source_pending(struct intsrc *isrc);
144 static int	xen_intr_config_intr(struct intsrc *isrc,
145 		     enum intr_trigger trig, enum intr_polarity pol);
146 static int	xen_intr_assign_cpu(struct intsrc *isrc, u_int apic_id);
147 
148 static void	xen_intr_pirq_enable_source(struct intsrc *isrc);
149 static void	xen_intr_pirq_disable_source(struct intsrc *isrc, int eoi);
150 static void	xen_intr_pirq_eoi_source(struct intsrc *isrc);
151 static void	xen_intr_pirq_enable_intr(struct intsrc *isrc);
152 static void	xen_intr_pirq_disable_intr(struct intsrc *isrc);
153 static int	xen_intr_pirq_config_intr(struct intsrc *isrc,
154 		     enum intr_trigger trig, enum intr_polarity pol);
155 
156 /**
157  * PIC interface for all event channel port types except physical IRQs.
158  */
159 struct pic xen_intr_pic = {
160 	.pic_enable_source  = xen_intr_enable_source,
161 	.pic_disable_source = xen_intr_disable_source,
162 	.pic_eoi_source     = xen_intr_eoi_source,
163 	.pic_enable_intr    = xen_intr_enable_intr,
164 	.pic_disable_intr   = xen_intr_disable_intr,
165 	.pic_vector         = xen_intr_vector,
166 	.pic_source_pending = xen_intr_source_pending,
167 	.pic_suspend        = xen_intr_suspend,
168 	.pic_resume         = xen_intr_resume,
169 	.pic_config_intr    = xen_intr_config_intr,
170 	.pic_assign_cpu     = xen_intr_assign_cpu
171 };
172 
173 /**
174  * PIC interface for all event channel representing
175  * physical interrupt sources.
176  */
177 struct pic xen_intr_pirq_pic = {
178 	.pic_enable_source  = xen_intr_pirq_enable_source,
179 	.pic_disable_source = xen_intr_pirq_disable_source,
180 	.pic_eoi_source     = xen_intr_pirq_eoi_source,
181 	.pic_enable_intr    = xen_intr_pirq_enable_intr,
182 	.pic_disable_intr   = xen_intr_pirq_disable_intr,
183 	.pic_vector         = xen_intr_vector,
184 	.pic_source_pending = xen_intr_source_pending,
185 	.pic_config_intr    = xen_intr_pirq_config_intr,
186 	.pic_assign_cpu     = xen_intr_assign_cpu
187 };
188 
189 static struct mtx	 xen_intr_isrc_lock;
190 static int		 xen_intr_auto_vector_count;
191 static struct xenisrc	*xen_intr_port_to_isrc[NR_EVENT_CHANNELS];
192 static u_long		*xen_intr_pirq_eoi_map;
193 static boolean_t	 xen_intr_pirq_eoi_map_enabled;
194 
195 /*------------------------- Private Functions --------------------------------*/
196 /**
197  * Disable signal delivery for an event channel port on the
198  * specified CPU.
199  *
200  * \param port  The event channel port to mask.
201  *
202  * This API is used to manage the port<=>CPU binding of event
203  * channel handlers.
204  *
205  * \note  This operation does not preclude reception of an event
206  *        for this event channel on another CPU.  To mask the
207  *        event channel globally, use evtchn_mask().
208  */
209 static inline void
evtchn_cpu_mask_port(u_int cpu,evtchn_port_t port)210 evtchn_cpu_mask_port(u_int cpu, evtchn_port_t port)
211 {
212 	struct xen_intr_pcpu_data *pcpu;
213 
214 	pcpu = DPCPU_ID_PTR(cpu, xen_intr_pcpu);
215 	xen_clear_bit(port, pcpu->evtchn_enabled);
216 }
217 
218 /**
219  * Enable signal delivery for an event channel port on the
220  * specified CPU.
221  *
222  * \param port  The event channel port to unmask.
223  *
224  * This API is used to manage the port<=>CPU binding of event
225  * channel handlers.
226  *
227  * \note  This operation does not guarantee that event delivery
228  *        is enabled for this event channel port.  The port must
229  *        also be globally enabled.  See evtchn_unmask().
230  */
231 static inline void
evtchn_cpu_unmask_port(u_int cpu,evtchn_port_t port)232 evtchn_cpu_unmask_port(u_int cpu, evtchn_port_t port)
233 {
234 	struct xen_intr_pcpu_data *pcpu;
235 
236 	pcpu = DPCPU_ID_PTR(cpu, xen_intr_pcpu);
237 	xen_set_bit(port, pcpu->evtchn_enabled);
238 }
239 
240 /**
241  * Allocate and register a per-cpu Xen upcall interrupt counter.
242  *
243  * \param cpu  The cpu for which to register this interrupt count.
244  */
245 static void
xen_intr_intrcnt_add(u_int cpu)246 xen_intr_intrcnt_add(u_int cpu)
247 {
248 	char buf[MAXCOMLEN + 1];
249 	struct xen_intr_pcpu_data *pcpu;
250 
251 	pcpu = DPCPU_ID_PTR(cpu, xen_intr_pcpu);
252 	if (pcpu->evtchn_intrcnt != NULL)
253 		return;
254 
255 	snprintf(buf, sizeof(buf), "cpu%d:xen", cpu);
256 	intrcnt_add(buf, &pcpu->evtchn_intrcnt);
257 }
258 
259 /**
260  * Search for an already allocated but currently unused Xen interrupt
261  * source object.
262  *
263  * \param type  Restrict the search to interrupt sources of the given
264  *              type.
265  *
266  * \return  A pointer to a free Xen interrupt source object or NULL.
267  */
268 static struct xenisrc *
xen_intr_find_unused_isrc(enum evtchn_type type)269 xen_intr_find_unused_isrc(enum evtchn_type type)
270 {
271 	int isrc_idx;
272 
273 	KASSERT(mtx_owned(&xen_intr_isrc_lock), ("Evtchn isrc lock not held"));
274 
275 	for (isrc_idx = 0; isrc_idx < xen_intr_auto_vector_count; isrc_idx ++) {
276 		struct xenisrc *isrc;
277 		u_int vector;
278 
279 		vector = FIRST_EVTCHN_INT + isrc_idx;
280 		isrc = (struct xenisrc *)intr_lookup_source(vector);
281 		if (isrc != NULL
282 		 && isrc->xi_type == EVTCHN_TYPE_UNBOUND) {
283 			KASSERT(isrc->xi_intsrc.is_handlers == 0,
284 			    ("Free evtchn still has handlers"));
285 			isrc->xi_type = type;
286 			return (isrc);
287 		}
288 	}
289 	return (NULL);
290 }
291 
292 /**
293  * Allocate a Xen interrupt source object.
294  *
295  * \param type  The type of interrupt source to create.
296  *
297  * \return  A pointer to a newly allocated Xen interrupt source
298  *          object or NULL.
299  */
300 static struct xenisrc *
xen_intr_alloc_isrc(enum evtchn_type type,int vector)301 xen_intr_alloc_isrc(enum evtchn_type type, int vector)
302 {
303 	static int warned;
304 	struct xenisrc *isrc;
305 
306 	KASSERT(mtx_owned(&xen_intr_isrc_lock), ("Evtchn alloc lock not held"));
307 
308 	if (xen_intr_auto_vector_count > NR_EVENT_CHANNELS) {
309 		if (!warned) {
310 			warned = 1;
311 			printf("xen_intr_alloc: Event channels exhausted.\n");
312 		}
313 		return (NULL);
314 	}
315 
316 	if (type != EVTCHN_TYPE_PIRQ) {
317 		vector = FIRST_EVTCHN_INT + xen_intr_auto_vector_count;
318 		xen_intr_auto_vector_count++;
319 	}
320 
321 	KASSERT((intr_lookup_source(vector) == NULL),
322 	    ("Trying to use an already allocated vector"));
323 
324 	mtx_unlock(&xen_intr_isrc_lock);
325 	isrc = malloc(sizeof(*isrc), M_XENINTR, M_WAITOK | M_ZERO);
326 	isrc->xi_intsrc.is_pic =
327 	    (type == EVTCHN_TYPE_PIRQ) ? &xen_intr_pirq_pic : &xen_intr_pic;
328 	isrc->xi_vector = vector;
329 	isrc->xi_type = type;
330 	intr_register_source(&isrc->xi_intsrc);
331 	mtx_lock(&xen_intr_isrc_lock);
332 
333 	return (isrc);
334 }
335 
336 /**
337  * Attempt to free an active Xen interrupt source object.
338  *
339  * \param isrc  The interrupt source object to release.
340  *
341  * \returns  EBUSY if the source is still in use, otherwise 0.
342  */
343 static int
xen_intr_release_isrc(struct xenisrc * isrc)344 xen_intr_release_isrc(struct xenisrc *isrc)
345 {
346 
347 	mtx_lock(&xen_intr_isrc_lock);
348 	if (isrc->xi_intsrc.is_handlers != 0) {
349 		mtx_unlock(&xen_intr_isrc_lock);
350 		return (EBUSY);
351 	}
352 	evtchn_mask_port(isrc->xi_port);
353 	evtchn_clear_port(isrc->xi_port);
354 
355 	/* Rebind port to CPU 0. */
356 	evtchn_cpu_mask_port(isrc->xi_cpu, isrc->xi_port);
357 	evtchn_cpu_unmask_port(0, isrc->xi_port);
358 
359 	if (isrc->xi_close != 0 && is_valid_evtchn(isrc->xi_port)) {
360 		struct evtchn_close close = { .port = isrc->xi_port };
361 		if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close))
362 			panic("EVTCHNOP_close failed");
363 	}
364 
365 	xen_intr_port_to_isrc[isrc->xi_port] = NULL;
366 	isrc->xi_cpu = 0;
367 	isrc->xi_type = EVTCHN_TYPE_UNBOUND;
368 	isrc->xi_port = 0;
369 	isrc->xi_cookie = NULL;
370 	mtx_unlock(&xen_intr_isrc_lock);
371 	return (0);
372 }
373 
374 /**
375  * Associate an interrupt handler with an already allocated local Xen
376  * event channel port.
377  *
378  * \param isrcp       The returned Xen interrupt object associated with
379  *                    the specified local port.
380  * \param local_port  The event channel to bind.
381  * \param type        The event channel type of local_port.
382  * \param intr_owner  The device making this bind request.
383  * \param filter      An interrupt filter handler.  Specify NULL
384  *                    to always dispatch to the ithread handler.
385  * \param handler     An interrupt ithread handler.  Optional (can
386  *                    specify NULL) if all necessary event actions
387  *                    are performed by filter.
388  * \param arg         Argument to present to both filter and handler.
389  * \param irqflags    Interrupt handler flags.  See sys/bus.h.
390  * \param handlep     Pointer to an opaque handle used to manage this
391  *                    registration.
392  *
393  * \returns  0 on success, otherwise an errno.
394  */
395 static int
xen_intr_bind_isrc(struct xenisrc ** isrcp,evtchn_port_t local_port,enum evtchn_type type,device_t intr_owner,driver_filter_t filter,driver_intr_t handler,void * arg,enum intr_type flags,xen_intr_handle_t * port_handlep)396 xen_intr_bind_isrc(struct xenisrc **isrcp, evtchn_port_t local_port,
397     enum evtchn_type type, device_t intr_owner, driver_filter_t filter,
398     driver_intr_t handler, void *arg, enum intr_type flags,
399     xen_intr_handle_t *port_handlep)
400 {
401 	struct xenisrc *isrc;
402 	int error;
403 
404 	*isrcp = NULL;
405 	if (port_handlep == NULL) {
406 		device_printf(intr_owner,
407 			      "xen_intr_bind_isrc: Bad event handle\n");
408 		return (EINVAL);
409 	}
410 
411 	mtx_lock(&xen_intr_isrc_lock);
412 	isrc = xen_intr_find_unused_isrc(type);
413 	if (isrc == NULL) {
414 		isrc = xen_intr_alloc_isrc(type, XEN_ALLOCATE_VECTOR);
415 		if (isrc == NULL) {
416 			mtx_unlock(&xen_intr_isrc_lock);
417 			return (ENOSPC);
418 		}
419 	}
420 	isrc->xi_port = local_port;
421 	xen_intr_port_to_isrc[local_port] = isrc;
422 	mtx_unlock(&xen_intr_isrc_lock);
423 
424 	/* Assign the opaque handler (the event channel port) */
425 	*port_handlep = &isrc->xi_port;
426 
427 #ifdef SMP
428 	if (type == EVTCHN_TYPE_PORT) {
429 		/*
430 		 * By default all interrupts are assigned to vCPU#0
431 		 * unless specified otherwise, so shuffle them to balance
432 		 * the interrupt load.
433 		 */
434 		xen_intr_assign_cpu(&isrc->xi_intsrc, intr_next_cpu());
435 	}
436 #endif
437 
438 	if (filter == NULL && handler == NULL) {
439 		/*
440 		 * No filter/handler provided, leave the event channel
441 		 * masked and without a valid handler, the caller is
442 		 * in charge of setting that up.
443 		 */
444 		*isrcp = isrc;
445 		return (0);
446 	}
447 
448 	error = xen_intr_add_handler(intr_owner, filter, handler, arg, flags,
449 	    *port_handlep);
450 	if (error != 0) {
451 		xen_intr_release_isrc(isrc);
452 		return (error);
453 	}
454 	*isrcp = isrc;
455 	return (0);
456 }
457 
458 /**
459  * Lookup a Xen interrupt source object given an interrupt binding handle.
460  *
461  * \param handle  A handle initialized by a previous call to
462  *                xen_intr_bind_isrc().
463  *
464  * \returns  A pointer to the Xen interrupt source object associated
465  *           with the given interrupt handle.  NULL if no association
466  *           currently exists.
467  */
468 static struct xenisrc *
xen_intr_isrc(xen_intr_handle_t handle)469 xen_intr_isrc(xen_intr_handle_t handle)
470 {
471 	evtchn_port_t port;
472 
473 	if (handle == NULL)
474 		return (NULL);
475 
476 	port = *(evtchn_port_t *)handle;
477 	if (!is_valid_evtchn(port) || port >= NR_EVENT_CHANNELS)
478 		return (NULL);
479 
480 	return (xen_intr_port_to_isrc[port]);
481 }
482 
483 /**
484  * Determine the event channel ports at the given section of the
485  * event port bitmap which have pending events for the given cpu.
486  *
487  * \param pcpu  The Xen interrupt pcpu data for the cpu being querried.
488  * \param sh    The Xen shared info area.
489  * \param idx   The index of the section of the event channel bitmap to
490  *              inspect.
491  *
492  * \returns  A u_long with bits set for every event channel with pending
493  *           events.
494  */
495 static inline u_long
xen_intr_active_ports(struct xen_intr_pcpu_data * pcpu,shared_info_t * sh,u_int idx)496 xen_intr_active_ports(struct xen_intr_pcpu_data *pcpu, shared_info_t *sh,
497     u_int idx)
498 {
499 
500 	CTASSERT(sizeof(sh->evtchn_mask[0]) == sizeof(sh->evtchn_pending[0]));
501 	CTASSERT(sizeof(sh->evtchn_mask[0]) == sizeof(pcpu->evtchn_enabled[0]));
502 	CTASSERT(sizeof(sh->evtchn_mask) == sizeof(sh->evtchn_pending));
503 	CTASSERT(sizeof(sh->evtchn_mask) == sizeof(pcpu->evtchn_enabled));
504 	return (sh->evtchn_pending[idx]
505 	      & ~sh->evtchn_mask[idx]
506 	      & pcpu->evtchn_enabled[idx]);
507 }
508 
509 /**
510  * Interrupt handler for processing all Xen event channel events.
511  *
512  * \param trap_frame  The trap frame context for the current interrupt.
513  */
514 void
xen_intr_handle_upcall(struct trapframe * trap_frame)515 xen_intr_handle_upcall(struct trapframe *trap_frame)
516 {
517 	u_int l1i, l2i, port, cpu;
518 	u_long masked_l1, masked_l2;
519 	struct xenisrc *isrc;
520 	shared_info_t *s;
521 	vcpu_info_t *v;
522 	struct xen_intr_pcpu_data *pc;
523 	u_long l1, l2;
524 
525 	/*
526 	 * Disable preemption in order to always check and fire events
527 	 * on the right vCPU
528 	 */
529 	critical_enter();
530 
531 	cpu = PCPU_GET(cpuid);
532 	pc  = DPCPU_PTR(xen_intr_pcpu);
533 	s   = HYPERVISOR_shared_info;
534 	v   = DPCPU_GET(vcpu_info);
535 
536 	if (xen_hvm_domain() && !xen_vector_callback_enabled) {
537 		KASSERT((cpu == 0), ("Fired PCI event callback on wrong CPU"));
538 	}
539 
540 	v->evtchn_upcall_pending = 0;
541 
542 #if 0
543 #ifndef CONFIG_X86 /* No need for a barrier -- XCHG is a barrier on x86. */
544 	/* Clear master flag /before/ clearing selector flag. */
545 	wmb();
546 #endif
547 #endif
548 
549 	l1 = atomic_readandclear_long(&v->evtchn_pending_sel);
550 
551 	l1i = pc->last_processed_l1i;
552 	l2i = pc->last_processed_l2i;
553 	(*pc->evtchn_intrcnt)++;
554 
555 	while (l1 != 0) {
556 
557 		l1i = (l1i + 1) % LONG_BIT;
558 		masked_l1 = l1 & ((~0UL) << l1i);
559 
560 		if (masked_l1 == 0) {
561 			/*
562 			 * if we masked out all events, wrap around
563 			 * to the beginning.
564 			 */
565 			l1i = LONG_BIT - 1;
566 			l2i = LONG_BIT - 1;
567 			continue;
568 		}
569 		l1i = ffsl(masked_l1) - 1;
570 
571 		do {
572 			l2 = xen_intr_active_ports(pc, s, l1i);
573 
574 			l2i = (l2i + 1) % LONG_BIT;
575 			masked_l2 = l2 & ((~0UL) << l2i);
576 
577 			if (masked_l2 == 0) {
578 				/* if we masked out all events, move on */
579 				l2i = LONG_BIT - 1;
580 				break;
581 			}
582 			l2i = ffsl(masked_l2) - 1;
583 
584 			/* process port */
585 			port = (l1i * LONG_BIT) + l2i;
586 			synch_clear_bit(port, &s->evtchn_pending[0]);
587 
588 			isrc = xen_intr_port_to_isrc[port];
589 			if (__predict_false(isrc == NULL))
590 				continue;
591 
592 			/* Make sure we are firing on the right vCPU */
593 			KASSERT((isrc->xi_cpu == PCPU_GET(cpuid)),
594 				("Received unexpected event on vCPU#%d, event bound to vCPU#%d",
595 				PCPU_GET(cpuid), isrc->xi_cpu));
596 
597 			intr_execute_handlers(&isrc->xi_intsrc, trap_frame);
598 
599 			/*
600 			 * If this is the final port processed,
601 			 * we'll pick up here+1 next time.
602 			 */
603 			pc->last_processed_l1i = l1i;
604 			pc->last_processed_l2i = l2i;
605 
606 		} while (l2i != LONG_BIT - 1);
607 
608 		l2 = xen_intr_active_ports(pc, s, l1i);
609 		if (l2 == 0) {
610 			/*
611 			 * We handled all ports, so we can clear the
612 			 * selector bit.
613 			 */
614 			l1 &= ~(1UL << l1i);
615 		}
616 	}
617 	critical_exit();
618 }
619 
620 static int
xen_intr_init(void * dummy __unused)621 xen_intr_init(void *dummy __unused)
622 {
623 	shared_info_t *s = HYPERVISOR_shared_info;
624 	struct xen_intr_pcpu_data *pcpu;
625 	struct physdev_pirq_eoi_gmfn eoi_gmfn;
626 	int i, rc;
627 
628 	if (!xen_domain())
629 		return (0);
630 
631 	mtx_init(&xen_intr_isrc_lock, "xen-irq-lock", NULL, MTX_DEF);
632 
633 	/*
634 	 * Register interrupt count manually as we aren't
635 	 * guaranteed to see a call to xen_intr_assign_cpu()
636 	 * before our first interrupt. Also set the per-cpu
637 	 * mask of CPU#0 to enable all, since by default
638 	 * all event channels are bound to CPU#0.
639 	 */
640 	CPU_FOREACH(i) {
641 		pcpu = DPCPU_ID_PTR(i, xen_intr_pcpu);
642 		memset(pcpu->evtchn_enabled, i == 0 ? ~0 : 0,
643 		    sizeof(pcpu->evtchn_enabled));
644 		xen_intr_intrcnt_add(i);
645 	}
646 
647 	for (i = 0; i < nitems(s->evtchn_mask); i++)
648 		atomic_store_rel_long(&s->evtchn_mask[i], ~0);
649 
650 	/* Try to register PIRQ EOI map */
651 	xen_intr_pirq_eoi_map = malloc(PAGE_SIZE, M_XENINTR, M_WAITOK | M_ZERO);
652 	eoi_gmfn.gmfn = atop(vtophys(xen_intr_pirq_eoi_map));
653 	rc = HYPERVISOR_physdev_op(PHYSDEVOP_pirq_eoi_gmfn_v2, &eoi_gmfn);
654 	if (rc != 0 && bootverbose)
655 		printf("Xen interrupts: unable to register PIRQ EOI map\n");
656 	else
657 		xen_intr_pirq_eoi_map_enabled = true;
658 
659 	intr_register_pic(&xen_intr_pic);
660 	intr_register_pic(&xen_intr_pirq_pic);
661 
662 	if (bootverbose)
663 		printf("Xen interrupt system initialized\n");
664 
665 	return (0);
666 }
667 SYSINIT(xen_intr_init, SI_SUB_INTR, SI_ORDER_SECOND, xen_intr_init, NULL);
668 
669 /*--------------------------- Common PIC Functions ---------------------------*/
670 /**
671  * Prepare this PIC for system suspension.
672  */
673 static void
xen_intr_suspend(struct pic * unused)674 xen_intr_suspend(struct pic *unused)
675 {
676 }
677 
678 static void
xen_rebind_ipi(struct xenisrc * isrc)679 xen_rebind_ipi(struct xenisrc *isrc)
680 {
681 #ifdef SMP
682 	int cpu = isrc->xi_cpu;
683 	int vcpu_id = pcpu_find(cpu)->pc_vcpu_id;
684 	int error;
685 	struct evtchn_bind_ipi bind_ipi = { .vcpu = vcpu_id };
686 
687 	error = HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
688 	                                    &bind_ipi);
689 	if (error != 0)
690 		panic("unable to rebind xen IPI: %d", error);
691 
692 	isrc->xi_port = bind_ipi.port;
693 	isrc->xi_cpu = 0;
694 	xen_intr_port_to_isrc[bind_ipi.port] = isrc;
695 
696 	error = xen_intr_assign_cpu(&isrc->xi_intsrc,
697 	                            cpu_apic_ids[cpu]);
698 	if (error)
699 		panic("unable to bind xen IPI to CPU#%d: %d",
700 		      cpu, error);
701 
702 	evtchn_unmask_port(bind_ipi.port);
703 #else
704 	panic("Resume IPI event channel on UP");
705 #endif
706 }
707 
708 static void
xen_rebind_virq(struct xenisrc * isrc)709 xen_rebind_virq(struct xenisrc *isrc)
710 {
711 	int cpu = isrc->xi_cpu;
712 	int vcpu_id = pcpu_find(cpu)->pc_vcpu_id;
713 	int error;
714 	struct evtchn_bind_virq bind_virq = { .virq = isrc->xi_virq,
715 	                                      .vcpu = vcpu_id };
716 
717 	error = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
718 	                                    &bind_virq);
719 	if (error != 0)
720 		panic("unable to rebind xen VIRQ#%d: %d", isrc->xi_virq, error);
721 
722 	isrc->xi_port = bind_virq.port;
723 	isrc->xi_cpu = 0;
724 	xen_intr_port_to_isrc[bind_virq.port] = isrc;
725 
726 #ifdef SMP
727 	error = xen_intr_assign_cpu(&isrc->xi_intsrc,
728 	                            cpu_apic_ids[cpu]);
729 	if (error)
730 		panic("unable to bind xen VIRQ#%d to CPU#%d: %d",
731 		      isrc->xi_virq, cpu, error);
732 #endif
733 
734 	evtchn_unmask_port(bind_virq.port);
735 }
736 
737 /**
738  * Return this PIC to service after being suspended.
739  */
740 static void
xen_intr_resume(struct pic * unused,bool suspend_cancelled)741 xen_intr_resume(struct pic *unused, bool suspend_cancelled)
742 {
743 	shared_info_t *s = HYPERVISOR_shared_info;
744 	struct xenisrc *isrc;
745 	u_int isrc_idx;
746 	int i;
747 
748 	if (suspend_cancelled)
749 		return;
750 
751 	/* Reset the per-CPU masks */
752 	CPU_FOREACH(i) {
753 		struct xen_intr_pcpu_data *pcpu;
754 
755 		pcpu = DPCPU_ID_PTR(i, xen_intr_pcpu);
756 		memset(pcpu->evtchn_enabled, i == 0 ? ~0 : 0,
757 		    sizeof(pcpu->evtchn_enabled));
758 	}
759 
760 	/* Mask all event channels. */
761 	for (i = 0; i < nitems(s->evtchn_mask); i++)
762 		atomic_store_rel_long(&s->evtchn_mask[i], ~0);
763 
764 	/* Remove port -> isrc mappings */
765 	memset(xen_intr_port_to_isrc, 0, sizeof(xen_intr_port_to_isrc));
766 
767 	/* Free unused isrcs and rebind VIRQs and IPIs */
768 	for (isrc_idx = 0; isrc_idx < xen_intr_auto_vector_count; isrc_idx++) {
769 		u_int vector;
770 
771 		vector = FIRST_EVTCHN_INT + isrc_idx;
772 		isrc = (struct xenisrc *)intr_lookup_source(vector);
773 		if (isrc != NULL) {
774 			isrc->xi_port = 0;
775 			switch (isrc->xi_type) {
776 			case EVTCHN_TYPE_IPI:
777 				xen_rebind_ipi(isrc);
778 				break;
779 			case EVTCHN_TYPE_VIRQ:
780 				xen_rebind_virq(isrc);
781 				break;
782 			default:
783 				intr_remove_handler(isrc->xi_cookie);
784 				isrc->xi_cpu = 0;
785 				isrc->xi_type = EVTCHN_TYPE_UNBOUND;
786 				isrc->xi_cookie = NULL;
787 				break;
788 			}
789 		}
790 	}
791 }
792 
793 /**
794  * Disable a Xen interrupt source.
795  *
796  * \param isrc  The interrupt source to disable.
797  */
798 static void
xen_intr_disable_intr(struct intsrc * base_isrc)799 xen_intr_disable_intr(struct intsrc *base_isrc)
800 {
801 	struct xenisrc *isrc = (struct xenisrc *)base_isrc;
802 
803 	evtchn_mask_port(isrc->xi_port);
804 }
805 
806 /**
807  * Determine the global interrupt vector number for
808  * a Xen interrupt source.
809  *
810  * \param isrc  The interrupt source to query.
811  *
812  * \return  The vector number corresponding to the given interrupt source.
813  */
814 static int
xen_intr_vector(struct intsrc * base_isrc)815 xen_intr_vector(struct intsrc *base_isrc)
816 {
817 	struct xenisrc *isrc = (struct xenisrc *)base_isrc;
818 
819 	return (isrc->xi_vector);
820 }
821 
822 /**
823  * Determine whether or not interrupt events are pending on the
824  * the given interrupt source.
825  *
826  * \param isrc  The interrupt source to query.
827  *
828  * \returns  0 if no events are pending, otherwise non-zero.
829  */
830 static int
xen_intr_source_pending(struct intsrc * isrc)831 xen_intr_source_pending(struct intsrc *isrc)
832 {
833 	/*
834 	 * EventChannels are edge triggered and never masked.
835 	 * There can be no pending events.
836 	 */
837 	return (0);
838 }
839 
840 /**
841  * Perform configuration of an interrupt source.
842  *
843  * \param isrc  The interrupt source to configure.
844  * \param trig  Edge or level.
845  * \param pol   Active high or low.
846  *
847  * \returns  0 if no events are pending, otherwise non-zero.
848  */
849 static int
xen_intr_config_intr(struct intsrc * isrc,enum intr_trigger trig,enum intr_polarity pol)850 xen_intr_config_intr(struct intsrc *isrc, enum intr_trigger trig,
851     enum intr_polarity pol)
852 {
853 	/* Configuration is only possible via the evtchn apis. */
854 	return (ENODEV);
855 }
856 
857 /**
858  * Configure CPU affinity for interrupt source event delivery.
859  *
860  * \param isrc     The interrupt source to configure.
861  * \param apic_id  The apic id of the CPU for handling future events.
862  *
863  * \returns  0 if successful, otherwise an errno.
864  */
865 static int
xen_intr_assign_cpu(struct intsrc * base_isrc,u_int apic_id)866 xen_intr_assign_cpu(struct intsrc *base_isrc, u_int apic_id)
867 {
868 #ifdef SMP
869 	struct evtchn_bind_vcpu bind_vcpu;
870 	struct xenisrc *isrc;
871 	u_int to_cpu, vcpu_id;
872 	int error, masked;
873 
874 	if (xen_vector_callback_enabled == 0)
875 		return (EOPNOTSUPP);
876 
877 	to_cpu = apic_cpuid(apic_id);
878 	vcpu_id = pcpu_find(to_cpu)->pc_vcpu_id;
879 	xen_intr_intrcnt_add(to_cpu);
880 
881 	mtx_lock(&xen_intr_isrc_lock);
882 	isrc = (struct xenisrc *)base_isrc;
883 	if (!is_valid_evtchn(isrc->xi_port)) {
884 		mtx_unlock(&xen_intr_isrc_lock);
885 		return (EINVAL);
886 	}
887 
888 	/*
889 	 * Mask the event channel while binding it to prevent interrupt
890 	 * delivery with an inconsistent state in isrc->xi_cpu.
891 	 */
892 	masked = evtchn_test_and_set_mask(isrc->xi_port);
893 	if ((isrc->xi_type == EVTCHN_TYPE_VIRQ) ||
894 		(isrc->xi_type == EVTCHN_TYPE_IPI)) {
895 		/*
896 		 * Virtual IRQs are associated with a cpu by
897 		 * the Hypervisor at evtchn_bind_virq time, so
898 		 * all we need to do is update the per-CPU masks.
899 		 */
900 		evtchn_cpu_mask_port(isrc->xi_cpu, isrc->xi_port);
901 		isrc->xi_cpu = to_cpu;
902 		evtchn_cpu_unmask_port(isrc->xi_cpu, isrc->xi_port);
903 		goto out;
904 	}
905 
906 	bind_vcpu.port = isrc->xi_port;
907 	bind_vcpu.vcpu = vcpu_id;
908 
909 	error = HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu);
910 	if (isrc->xi_cpu != to_cpu) {
911 		if (error == 0) {
912 			/* Commit to new binding by removing the old one. */
913 			evtchn_cpu_mask_port(isrc->xi_cpu, isrc->xi_port);
914 			isrc->xi_cpu = to_cpu;
915 			evtchn_cpu_unmask_port(isrc->xi_cpu, isrc->xi_port);
916 		}
917 	}
918 
919 out:
920 	if (masked == 0)
921 		evtchn_unmask_port(isrc->xi_port);
922 	mtx_unlock(&xen_intr_isrc_lock);
923 	return (0);
924 #else
925 	return (EOPNOTSUPP);
926 #endif
927 }
928 
929 /*------------------- Virtual Interrupt Source PIC Functions -----------------*/
930 /*
931  * Mask a level triggered interrupt source.
932  *
933  * \param isrc  The interrupt source to mask (if necessary).
934  * \param eoi   If non-zero, perform any necessary end-of-interrupt
935  *              acknowledgements.
936  */
937 static void
xen_intr_disable_source(struct intsrc * base_isrc,int eoi)938 xen_intr_disable_source(struct intsrc *base_isrc, int eoi)
939 {
940 	struct xenisrc *isrc;
941 
942 	isrc = (struct xenisrc *)base_isrc;
943 
944 	/*
945 	 * NB: checking if the event channel is already masked is
946 	 * needed because the event channel user-space device
947 	 * masks event channels on it's filter as part of it's
948 	 * normal operation, and those shouldn't be automatically
949 	 * unmasked by the generic interrupt code. The event channel
950 	 * device will unmask them when needed.
951 	 */
952 	isrc->xi_masked = !!evtchn_test_and_set_mask(isrc->xi_port);
953 }
954 
955 /*
956  * Unmask a level triggered interrupt source.
957  *
958  * \param isrc  The interrupt source to unmask (if necessary).
959  */
960 static void
xen_intr_enable_source(struct intsrc * base_isrc)961 xen_intr_enable_source(struct intsrc *base_isrc)
962 {
963 	struct xenisrc *isrc;
964 
965 	isrc = (struct xenisrc *)base_isrc;
966 
967 	if (isrc->xi_masked == 0)
968 		evtchn_unmask_port(isrc->xi_port);
969 }
970 
971 /*
972  * Perform any necessary end-of-interrupt acknowledgements.
973  *
974  * \param isrc  The interrupt source to EOI.
975  */
976 static void
xen_intr_eoi_source(struct intsrc * base_isrc)977 xen_intr_eoi_source(struct intsrc *base_isrc)
978 {
979 }
980 
981 /*
982  * Enable and unmask the interrupt source.
983  *
984  * \param isrc  The interrupt source to enable.
985  */
986 static void
xen_intr_enable_intr(struct intsrc * base_isrc)987 xen_intr_enable_intr(struct intsrc *base_isrc)
988 {
989 	struct xenisrc *isrc = (struct xenisrc *)base_isrc;
990 
991 	evtchn_unmask_port(isrc->xi_port);
992 }
993 
994 /*------------------ Physical Interrupt Source PIC Functions -----------------*/
995 /*
996  * Mask a level triggered interrupt source.
997  *
998  * \param isrc  The interrupt source to mask (if necessary).
999  * \param eoi   If non-zero, perform any necessary end-of-interrupt
1000  *              acknowledgements.
1001  */
1002 static void
xen_intr_pirq_disable_source(struct intsrc * base_isrc,int eoi)1003 xen_intr_pirq_disable_source(struct intsrc *base_isrc, int eoi)
1004 {
1005 	struct xenisrc *isrc;
1006 
1007 	isrc = (struct xenisrc *)base_isrc;
1008 
1009 	if (isrc->xi_edgetrigger == 0)
1010 		evtchn_mask_port(isrc->xi_port);
1011 	if (eoi == PIC_EOI)
1012 		xen_intr_pirq_eoi_source(base_isrc);
1013 }
1014 
1015 /*
1016  * Unmask a level triggered interrupt source.
1017  *
1018  * \param isrc  The interrupt source to unmask (if necessary).
1019  */
1020 static void
xen_intr_pirq_enable_source(struct intsrc * base_isrc)1021 xen_intr_pirq_enable_source(struct intsrc *base_isrc)
1022 {
1023 	struct xenisrc *isrc;
1024 
1025 	isrc = (struct xenisrc *)base_isrc;
1026 
1027 	if (isrc->xi_edgetrigger == 0)
1028 		evtchn_unmask_port(isrc->xi_port);
1029 }
1030 
1031 /*
1032  * Perform any necessary end-of-interrupt acknowledgements.
1033  *
1034  * \param isrc  The interrupt source to EOI.
1035  */
1036 static void
xen_intr_pirq_eoi_source(struct intsrc * base_isrc)1037 xen_intr_pirq_eoi_source(struct intsrc *base_isrc)
1038 {
1039 	struct xenisrc *isrc;
1040 	int error;
1041 
1042 	isrc = (struct xenisrc *)base_isrc;
1043 
1044 	if (xen_test_bit(isrc->xi_pirq, xen_intr_pirq_eoi_map)) {
1045 		struct physdev_eoi eoi = { .irq = isrc->xi_pirq };
1046 
1047 		error = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi);
1048 		if (error != 0)
1049 			panic("Unable to EOI PIRQ#%d: %d\n",
1050 			    isrc->xi_pirq, error);
1051 	}
1052 }
1053 
1054 /*
1055  * Enable and unmask the interrupt source.
1056  *
1057  * \param isrc  The interrupt source to enable.
1058  */
1059 static void
xen_intr_pirq_enable_intr(struct intsrc * base_isrc)1060 xen_intr_pirq_enable_intr(struct intsrc *base_isrc)
1061 {
1062 	struct xenisrc *isrc;
1063 	struct evtchn_bind_pirq bind_pirq;
1064 	struct physdev_irq_status_query irq_status;
1065 	int error;
1066 
1067 	isrc = (struct xenisrc *)base_isrc;
1068 
1069 	if (!xen_intr_pirq_eoi_map_enabled) {
1070 		irq_status.irq = isrc->xi_pirq;
1071 		error = HYPERVISOR_physdev_op(PHYSDEVOP_irq_status_query,
1072 		    &irq_status);
1073 		if (error)
1074 			panic("unable to get status of IRQ#%d", isrc->xi_pirq);
1075 
1076 		if (irq_status.flags & XENIRQSTAT_needs_eoi) {
1077 			/*
1078 			 * Since the dynamic PIRQ EOI map is not available
1079 			 * mark the PIRQ as needing EOI unconditionally.
1080 			 */
1081 			xen_set_bit(isrc->xi_pirq, xen_intr_pirq_eoi_map);
1082 		}
1083 	}
1084 
1085 	bind_pirq.pirq = isrc->xi_pirq;
1086 	bind_pirq.flags = isrc->xi_edgetrigger ? 0 : BIND_PIRQ__WILL_SHARE;
1087 	error = HYPERVISOR_event_channel_op(EVTCHNOP_bind_pirq, &bind_pirq);
1088 	if (error)
1089 		panic("unable to bind IRQ#%d", isrc->xi_pirq);
1090 
1091 	isrc->xi_port = bind_pirq.port;
1092 
1093 	mtx_lock(&xen_intr_isrc_lock);
1094 	KASSERT((xen_intr_port_to_isrc[bind_pirq.port] == NULL),
1095 	    ("trying to override an already setup event channel port"));
1096 	xen_intr_port_to_isrc[bind_pirq.port] = isrc;
1097 	mtx_unlock(&xen_intr_isrc_lock);
1098 
1099 	evtchn_unmask_port(isrc->xi_port);
1100 }
1101 
1102 /*
1103  * Disable an interrupt source.
1104  *
1105  * \param isrc  The interrupt source to disable.
1106  */
1107 static void
xen_intr_pirq_disable_intr(struct intsrc * base_isrc)1108 xen_intr_pirq_disable_intr(struct intsrc *base_isrc)
1109 {
1110 	struct xenisrc *isrc;
1111 	struct evtchn_close close;
1112 	int error;
1113 
1114 	isrc = (struct xenisrc *)base_isrc;
1115 
1116 	evtchn_mask_port(isrc->xi_port);
1117 
1118 	close.port = isrc->xi_port;
1119 	error = HYPERVISOR_event_channel_op(EVTCHNOP_close, &close);
1120 	if (error)
1121 		panic("unable to close event channel %d IRQ#%d",
1122 		    isrc->xi_port, isrc->xi_pirq);
1123 
1124 	mtx_lock(&xen_intr_isrc_lock);
1125 	xen_intr_port_to_isrc[isrc->xi_port] = NULL;
1126 	mtx_unlock(&xen_intr_isrc_lock);
1127 
1128 	isrc->xi_port = 0;
1129 }
1130 
1131 /**
1132  * Perform configuration of an interrupt source.
1133  *
1134  * \param isrc  The interrupt source to configure.
1135  * \param trig  Edge or level.
1136  * \param pol   Active high or low.
1137  *
1138  * \returns  0 if no events are pending, otherwise non-zero.
1139  */
1140 static int
xen_intr_pirq_config_intr(struct intsrc * base_isrc,enum intr_trigger trig,enum intr_polarity pol)1141 xen_intr_pirq_config_intr(struct intsrc *base_isrc, enum intr_trigger trig,
1142     enum intr_polarity pol)
1143 {
1144 	struct xenisrc *isrc = (struct xenisrc *)base_isrc;
1145 	struct physdev_setup_gsi setup_gsi;
1146 	int error;
1147 
1148 	KASSERT(!(trig == INTR_TRIGGER_CONFORM || pol == INTR_POLARITY_CONFORM),
1149 	    ("%s: Conforming trigger or polarity\n", __func__));
1150 
1151 	setup_gsi.gsi = isrc->xi_pirq;
1152 	setup_gsi.triggering = trig == INTR_TRIGGER_EDGE ? 0 : 1;
1153 	setup_gsi.polarity = pol == INTR_POLARITY_HIGH ? 0 : 1;
1154 
1155 	error = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
1156 	if (error == -XEN_EEXIST) {
1157 		if ((isrc->xi_edgetrigger && (trig != INTR_TRIGGER_EDGE)) ||
1158 		    (isrc->xi_activehi && (pol != INTR_POLARITY_HIGH)))
1159 			panic("unable to reconfigure interrupt IRQ#%d",
1160 			    isrc->xi_pirq);
1161 		error = 0;
1162 	}
1163 	if (error)
1164 		panic("unable to configure IRQ#%d\n", isrc->xi_pirq);
1165 
1166 	isrc->xi_activehi = pol == INTR_POLARITY_HIGH ? 1 : 0;
1167 	isrc->xi_edgetrigger = trig == INTR_TRIGGER_EDGE ? 1 : 0;
1168 
1169 	return (0);
1170 }
1171 
1172 /*--------------------------- Public Functions -------------------------------*/
1173 /*------- API comments for these methods can be found in xen/xenintr.h -------*/
1174 int
xen_intr_bind_local_port(device_t dev,evtchn_port_t local_port,driver_filter_t filter,driver_intr_t handler,void * arg,enum intr_type flags,xen_intr_handle_t * port_handlep)1175 xen_intr_bind_local_port(device_t dev, evtchn_port_t local_port,
1176     driver_filter_t filter, driver_intr_t handler, void *arg,
1177     enum intr_type flags, xen_intr_handle_t *port_handlep)
1178 {
1179 	struct xenisrc *isrc;
1180 	int error;
1181 
1182 	error = xen_intr_bind_isrc(&isrc, local_port, EVTCHN_TYPE_PORT, dev,
1183 		    filter, handler, arg, flags, port_handlep);
1184 	if (error != 0)
1185 		return (error);
1186 
1187 	/*
1188 	 * The Event Channel API didn't open this port, so it is not
1189 	 * responsible for closing it automatically on unbind.
1190 	 */
1191 	isrc->xi_close = 0;
1192 	return (0);
1193 }
1194 
1195 int
xen_intr_alloc_and_bind_local_port(device_t dev,u_int remote_domain,driver_filter_t filter,driver_intr_t handler,void * arg,enum intr_type flags,xen_intr_handle_t * port_handlep)1196 xen_intr_alloc_and_bind_local_port(device_t dev, u_int remote_domain,
1197     driver_filter_t filter, driver_intr_t handler, void *arg,
1198     enum intr_type flags, xen_intr_handle_t *port_handlep)
1199 {
1200 	struct xenisrc *isrc;
1201 	struct evtchn_alloc_unbound alloc_unbound;
1202 	int error;
1203 
1204 	alloc_unbound.dom        = DOMID_SELF;
1205 	alloc_unbound.remote_dom = remote_domain;
1206 	error = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
1207 		    &alloc_unbound);
1208 	if (error != 0) {
1209 		/*
1210 		 * XXX Trap Hypercall error code Linuxisms in
1211 		 *     the HYPERCALL layer.
1212 		 */
1213 		return (-error);
1214 	}
1215 
1216 	error = xen_intr_bind_isrc(&isrc, alloc_unbound.port, EVTCHN_TYPE_PORT,
1217 				 dev, filter, handler, arg, flags,
1218 				 port_handlep);
1219 	if (error != 0) {
1220 		evtchn_close_t close = { .port = alloc_unbound.port };
1221 		if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close))
1222 			panic("EVTCHNOP_close failed");
1223 		return (error);
1224 	}
1225 
1226 	isrc->xi_close = 1;
1227 	return (0);
1228 }
1229 
1230 int
xen_intr_bind_remote_port(device_t dev,u_int remote_domain,u_int remote_port,driver_filter_t filter,driver_intr_t handler,void * arg,enum intr_type flags,xen_intr_handle_t * port_handlep)1231 xen_intr_bind_remote_port(device_t dev, u_int remote_domain,
1232     u_int remote_port, driver_filter_t filter, driver_intr_t handler,
1233     void *arg, enum intr_type flags, xen_intr_handle_t *port_handlep)
1234 {
1235 	struct xenisrc *isrc;
1236 	struct evtchn_bind_interdomain bind_interdomain;
1237 	int error;
1238 
1239 	bind_interdomain.remote_dom  = remote_domain;
1240 	bind_interdomain.remote_port = remote_port;
1241 	error = HYPERVISOR_event_channel_op(EVTCHNOP_bind_interdomain,
1242 					    &bind_interdomain);
1243 	if (error != 0) {
1244 		/*
1245 		 * XXX Trap Hypercall error code Linuxisms in
1246 		 *     the HYPERCALL layer.
1247 		 */
1248 		return (-error);
1249 	}
1250 
1251 	error = xen_intr_bind_isrc(&isrc, bind_interdomain.local_port,
1252 				 EVTCHN_TYPE_PORT, dev, filter, handler,
1253 				 arg, flags, port_handlep);
1254 	if (error) {
1255 		evtchn_close_t close = { .port = bind_interdomain.local_port };
1256 		if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close))
1257 			panic("EVTCHNOP_close failed");
1258 		return (error);
1259 	}
1260 
1261 	/*
1262 	 * The Event Channel API opened this port, so it is
1263 	 * responsible for closing it automatically on unbind.
1264 	 */
1265 	isrc->xi_close = 1;
1266 	return (0);
1267 }
1268 
1269 int
xen_intr_bind_virq(device_t dev,u_int virq,u_int cpu,driver_filter_t filter,driver_intr_t handler,void * arg,enum intr_type flags,xen_intr_handle_t * port_handlep)1270 xen_intr_bind_virq(device_t dev, u_int virq, u_int cpu,
1271     driver_filter_t filter, driver_intr_t handler, void *arg,
1272     enum intr_type flags, xen_intr_handle_t *port_handlep)
1273 {
1274 	int vcpu_id = pcpu_find(cpu)->pc_vcpu_id;
1275 	struct xenisrc *isrc;
1276 	struct evtchn_bind_virq bind_virq = { .virq = virq, .vcpu = vcpu_id };
1277 	int error;
1278 
1279 	/* Ensure the target CPU is ready to handle evtchn interrupts. */
1280 	xen_intr_intrcnt_add(cpu);
1281 
1282 	isrc = NULL;
1283 	error = HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq, &bind_virq);
1284 	if (error != 0) {
1285 		/*
1286 		 * XXX Trap Hypercall error code Linuxisms in
1287 		 *     the HYPERCALL layer.
1288 		 */
1289 		return (-error);
1290 	}
1291 
1292 	error = xen_intr_bind_isrc(&isrc, bind_virq.port, EVTCHN_TYPE_VIRQ, dev,
1293 				 filter, handler, arg, flags, port_handlep);
1294 
1295 #ifdef SMP
1296 	if (error == 0)
1297 		error = intr_event_bind(isrc->xi_intsrc.is_event, cpu);
1298 #endif
1299 
1300 	if (error != 0) {
1301 		evtchn_close_t close = { .port = bind_virq.port };
1302 
1303 		xen_intr_unbind(*port_handlep);
1304 		if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close))
1305 			panic("EVTCHNOP_close failed");
1306 		return (error);
1307 	}
1308 
1309 #ifdef SMP
1310 	if (isrc->xi_cpu != cpu) {
1311 		/*
1312 		 * Too early in the boot process for the generic interrupt
1313 		 * code to perform the binding.  Update our event channel
1314 		 * masks manually so events can't fire on the wrong cpu
1315 		 * during AP startup.
1316 		 */
1317 		xen_intr_assign_cpu(&isrc->xi_intsrc, cpu_apic_ids[cpu]);
1318 	}
1319 #endif
1320 
1321 	/*
1322 	 * The Event Channel API opened this port, so it is
1323 	 * responsible for closing it automatically on unbind.
1324 	 */
1325 	isrc->xi_close = 1;
1326 	isrc->xi_virq = virq;
1327 
1328 	return (0);
1329 }
1330 
1331 int
xen_intr_alloc_and_bind_ipi(device_t dev,u_int cpu,driver_filter_t filter,enum intr_type flags,xen_intr_handle_t * port_handlep)1332 xen_intr_alloc_and_bind_ipi(device_t dev, u_int cpu,
1333     driver_filter_t filter, enum intr_type flags,
1334     xen_intr_handle_t *port_handlep)
1335 {
1336 #ifdef SMP
1337 	int vcpu_id = pcpu_find(cpu)->pc_vcpu_id;
1338 	struct xenisrc *isrc;
1339 	struct evtchn_bind_ipi bind_ipi = { .vcpu = vcpu_id };
1340 	int error;
1341 
1342 	/* Ensure the target CPU is ready to handle evtchn interrupts. */
1343 	xen_intr_intrcnt_add(cpu);
1344 
1345 	isrc = NULL;
1346 	error = HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi, &bind_ipi);
1347 	if (error != 0) {
1348 		/*
1349 		 * XXX Trap Hypercall error code Linuxisms in
1350 		 *     the HYPERCALL layer.
1351 		 */
1352 		return (-error);
1353 	}
1354 
1355 	error = xen_intr_bind_isrc(&isrc, bind_ipi.port, EVTCHN_TYPE_IPI,
1356 	                           dev, filter, NULL, NULL, flags,
1357 	                           port_handlep);
1358 	if (error == 0)
1359 		error = intr_event_bind(isrc->xi_intsrc.is_event, cpu);
1360 
1361 	if (error != 0) {
1362 		evtchn_close_t close = { .port = bind_ipi.port };
1363 
1364 		xen_intr_unbind(*port_handlep);
1365 		if (HYPERVISOR_event_channel_op(EVTCHNOP_close, &close))
1366 			panic("EVTCHNOP_close failed");
1367 		return (error);
1368 	}
1369 
1370 	if (isrc->xi_cpu != cpu) {
1371 		/*
1372 		 * Too early in the boot process for the generic interrupt
1373 		 * code to perform the binding.  Update our event channel
1374 		 * masks manually so events can't fire on the wrong cpu
1375 		 * during AP startup.
1376 		 */
1377 		xen_intr_assign_cpu(&isrc->xi_intsrc, cpu_apic_ids[cpu]);
1378 	}
1379 
1380 	/*
1381 	 * The Event Channel API opened this port, so it is
1382 	 * responsible for closing it automatically on unbind.
1383 	 */
1384 	isrc->xi_close = 1;
1385 	return (0);
1386 #else
1387 	return (EOPNOTSUPP);
1388 #endif
1389 }
1390 
1391 int
xen_register_pirq(int vector,enum intr_trigger trig,enum intr_polarity pol)1392 xen_register_pirq(int vector, enum intr_trigger trig, enum intr_polarity pol)
1393 {
1394 	struct physdev_map_pirq map_pirq;
1395 	struct xenisrc *isrc;
1396 	int error;
1397 
1398 	if (vector == 0)
1399 		return (EINVAL);
1400 
1401 	if (bootverbose)
1402 		printf("xen: register IRQ#%d\n", vector);
1403 
1404 	map_pirq.domid = DOMID_SELF;
1405 	map_pirq.type = MAP_PIRQ_TYPE_GSI;
1406 	map_pirq.index = vector;
1407 	map_pirq.pirq = vector;
1408 
1409 	error = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_pirq);
1410 	if (error) {
1411 		printf("xen: unable to map IRQ#%d\n", vector);
1412 		return (error);
1413 	}
1414 
1415 	mtx_lock(&xen_intr_isrc_lock);
1416 	isrc = xen_intr_alloc_isrc(EVTCHN_TYPE_PIRQ, vector);
1417 	mtx_unlock(&xen_intr_isrc_lock);
1418 	KASSERT((isrc != NULL), ("xen: unable to allocate isrc for interrupt"));
1419 	isrc->xi_pirq = vector;
1420 	isrc->xi_activehi = pol == INTR_POLARITY_HIGH ? 1 : 0;
1421 	isrc->xi_edgetrigger = trig == INTR_TRIGGER_EDGE ? 1 : 0;
1422 
1423 	return (0);
1424 }
1425 
1426 int
xen_register_msi(device_t dev,int vector,int count)1427 xen_register_msi(device_t dev, int vector, int count)
1428 {
1429 	struct physdev_map_pirq msi_irq;
1430 	struct xenisrc *isrc;
1431 	int ret;
1432 
1433 	memset(&msi_irq, 0, sizeof(msi_irq));
1434 	msi_irq.domid = DOMID_SELF;
1435 	msi_irq.type = count == 1 ?
1436 	    MAP_PIRQ_TYPE_MSI_SEG : MAP_PIRQ_TYPE_MULTI_MSI;
1437 	msi_irq.index = -1;
1438 	msi_irq.pirq = -1;
1439 	msi_irq.bus = pci_get_bus(dev) | (pci_get_domain(dev) << 16);
1440 	msi_irq.devfn = (pci_get_slot(dev) << 3) | pci_get_function(dev);
1441 	msi_irq.entry_nr = count;
1442 
1443 	ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &msi_irq);
1444 	if (ret != 0)
1445 		return (ret);
1446 	if (count != msi_irq.entry_nr) {
1447 		panic("unable to setup all requested MSI vectors "
1448 		    "(expected %d got %d)", count, msi_irq.entry_nr);
1449 	}
1450 
1451 	mtx_lock(&xen_intr_isrc_lock);
1452 	for (int i = 0; i < count; i++) {
1453 		isrc = xen_intr_alloc_isrc(EVTCHN_TYPE_PIRQ, vector + i);
1454 		KASSERT(isrc != NULL,
1455 		    ("xen: unable to allocate isrc for interrupt"));
1456 		isrc->xi_pirq = msi_irq.pirq + i;
1457 		/* MSI interrupts are always edge triggered */
1458 		isrc->xi_edgetrigger = 1;
1459 	}
1460 	mtx_unlock(&xen_intr_isrc_lock);
1461 
1462 	return (0);
1463 }
1464 
1465 int
xen_release_msi(int vector)1466 xen_release_msi(int vector)
1467 {
1468 	struct physdev_unmap_pirq unmap;
1469 	struct xenisrc *isrc;
1470 	int ret;
1471 
1472 	isrc = (struct xenisrc *)intr_lookup_source(vector);
1473 	if (isrc == NULL)
1474 		return (ENXIO);
1475 
1476 	unmap.pirq = isrc->xi_pirq;
1477 	ret = HYPERVISOR_physdev_op(PHYSDEVOP_unmap_pirq, &unmap);
1478 	if (ret != 0)
1479 		return (ret);
1480 
1481 	xen_intr_release_isrc(isrc);
1482 
1483 	return (0);
1484 }
1485 
1486 int
xen_intr_describe(xen_intr_handle_t port_handle,const char * fmt,...)1487 xen_intr_describe(xen_intr_handle_t port_handle, const char *fmt, ...)
1488 {
1489 	char descr[MAXCOMLEN + 1];
1490 	struct xenisrc *isrc;
1491 	va_list ap;
1492 
1493 	isrc = xen_intr_isrc(port_handle);
1494 	if (isrc == NULL)
1495 		return (EINVAL);
1496 
1497 	va_start(ap, fmt);
1498 	vsnprintf(descr, sizeof(descr), fmt, ap);
1499 	va_end(ap);
1500 	return (intr_describe(isrc->xi_vector, isrc->xi_cookie, descr));
1501 }
1502 
1503 void
xen_intr_unbind(xen_intr_handle_t * port_handlep)1504 xen_intr_unbind(xen_intr_handle_t *port_handlep)
1505 {
1506 	struct xenisrc *isrc;
1507 
1508 	KASSERT(port_handlep != NULL,
1509 	    ("NULL xen_intr_handle_t passed to xen_intr_unbind"));
1510 
1511 	isrc = xen_intr_isrc(*port_handlep);
1512 	*port_handlep = NULL;
1513 	if (isrc == NULL)
1514 		return;
1515 
1516 	if (isrc->xi_cookie != NULL)
1517 		intr_remove_handler(isrc->xi_cookie);
1518 	xen_intr_release_isrc(isrc);
1519 }
1520 
1521 void
xen_intr_signal(xen_intr_handle_t handle)1522 xen_intr_signal(xen_intr_handle_t handle)
1523 {
1524 	struct xenisrc *isrc;
1525 
1526 	isrc = xen_intr_isrc(handle);
1527 	if (isrc != NULL) {
1528 		KASSERT(isrc->xi_type == EVTCHN_TYPE_PORT ||
1529 			isrc->xi_type == EVTCHN_TYPE_IPI,
1530 			("evtchn_signal on something other than a local port"));
1531 		struct evtchn_send send = { .port = isrc->xi_port };
1532 		(void)HYPERVISOR_event_channel_op(EVTCHNOP_send, &send);
1533 	}
1534 }
1535 
1536 evtchn_port_t
xen_intr_port(xen_intr_handle_t handle)1537 xen_intr_port(xen_intr_handle_t handle)
1538 {
1539 	struct xenisrc *isrc;
1540 
1541 	isrc = xen_intr_isrc(handle);
1542 	if (isrc == NULL)
1543 		return (0);
1544 
1545 	return (isrc->xi_port);
1546 }
1547 
1548 int
xen_intr_add_handler(device_t dev,driver_filter_t filter,driver_intr_t handler,void * arg,enum intr_type flags,xen_intr_handle_t handle)1549 xen_intr_add_handler(device_t dev, driver_filter_t filter,
1550     driver_intr_t handler, void *arg, enum intr_type flags,
1551     xen_intr_handle_t handle)
1552 {
1553 	struct xenisrc *isrc;
1554 	int error;
1555 
1556 	isrc = xen_intr_isrc(handle);
1557 	if (isrc == NULL || isrc->xi_cookie != NULL)
1558 		return (EINVAL);
1559 
1560 	error = intr_add_handler(device_get_nameunit(dev), isrc->xi_vector,
1561 	    filter, handler, arg, flags|INTR_EXCL, &isrc->xi_cookie);
1562 	if (error != 0) {
1563 		device_printf(dev,
1564 		    "xen_intr_add_handler: intr_add_handler failed: %d\n",
1565 		    error);
1566 	}
1567 
1568 	return (error);
1569 }
1570 
1571 #ifdef DDB
1572 static const char *
xen_intr_print_type(enum evtchn_type type)1573 xen_intr_print_type(enum evtchn_type type)
1574 {
1575 	static const char *evtchn_type_to_string[EVTCHN_TYPE_COUNT] = {
1576 		[EVTCHN_TYPE_UNBOUND]	= "UNBOUND",
1577 		[EVTCHN_TYPE_PIRQ]	= "PIRQ",
1578 		[EVTCHN_TYPE_VIRQ]	= "VIRQ",
1579 		[EVTCHN_TYPE_IPI]	= "IPI",
1580 		[EVTCHN_TYPE_PORT]	= "PORT",
1581 	};
1582 
1583 	if (type >= EVTCHN_TYPE_COUNT)
1584 		return ("UNKNOWN");
1585 
1586 	return (evtchn_type_to_string[type]);
1587 }
1588 
1589 static void
xen_intr_dump_port(struct xenisrc * isrc)1590 xen_intr_dump_port(struct xenisrc *isrc)
1591 {
1592 	struct xen_intr_pcpu_data *pcpu;
1593 	shared_info_t *s = HYPERVISOR_shared_info;
1594 	int i;
1595 
1596 	db_printf("Port %d Type: %s\n",
1597 	    isrc->xi_port, xen_intr_print_type(isrc->xi_type));
1598 	if (isrc->xi_type == EVTCHN_TYPE_PIRQ) {
1599 		db_printf("\tPirq: %d ActiveHi: %d EdgeTrigger: %d "
1600 		    "NeedsEOI: %d\n",
1601 		    isrc->xi_pirq, isrc->xi_activehi, isrc->xi_edgetrigger,
1602 		    !!xen_test_bit(isrc->xi_pirq, xen_intr_pirq_eoi_map));
1603 	}
1604 	if (isrc->xi_type == EVTCHN_TYPE_VIRQ)
1605 		db_printf("\tVirq: %d\n", isrc->xi_virq);
1606 
1607 	db_printf("\tMasked: %d Pending: %d\n",
1608 	    !!xen_test_bit(isrc->xi_port, &s->evtchn_mask[0]),
1609 	    !!xen_test_bit(isrc->xi_port, &s->evtchn_pending[0]));
1610 
1611 	db_printf("\tPer-CPU Masks: ");
1612 	CPU_FOREACH(i) {
1613 		pcpu = DPCPU_ID_PTR(i, xen_intr_pcpu);
1614 		db_printf("cpu#%d: %d ", i,
1615 		    !!xen_test_bit(isrc->xi_port, pcpu->evtchn_enabled));
1616 	}
1617 	db_printf("\n");
1618 }
1619 
DB_SHOW_COMMAND(xen_evtchn,db_show_xen_evtchn)1620 DB_SHOW_COMMAND(xen_evtchn, db_show_xen_evtchn)
1621 {
1622 	int i;
1623 
1624 	if (!xen_domain()) {
1625 		db_printf("Only available on Xen guests\n");
1626 		return;
1627 	}
1628 
1629 	for (i = 0; i < NR_EVENT_CHANNELS; i++) {
1630 		struct xenisrc *isrc;
1631 
1632 		isrc = xen_intr_port_to_isrc[i];
1633 		if (isrc == NULL)
1634 			continue;
1635 
1636 		xen_intr_dump_port(isrc);
1637 	}
1638 }
1639 #endif /* DDB */
1640