1/*
2 * Xen event channels (FIFO-based ABI)
3 *
4 * Copyright (C) 2013 Citrix Systems R&D ltd.
5 *
6 * This source code is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation; either version 2 of the
9 * License, or (at your option) any later version.
10 *
11 * Or, when distributed separately from the Linux kernel or
12 * incorporated into other software packages, subject to the following
13 * license:
14 *
15 * Permission is hereby granted, free of charge, to any person obtaining a copy
16 * of this source file (the "Software"), to deal in the Software without
17 * restriction, including without limitation the rights to use, copy, modify,
18 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
19 * and to permit persons to whom the Software is furnished to do so, subject to
20 * the following conditions:
21 *
22 * The above copyright notice and this permission notice shall be included in
23 * all copies or substantial portions of the Software.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
26 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
27 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
28 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
29 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
30 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
31 * IN THE SOFTWARE.
32 */
33
34#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
35
36#include <linux/linkage.h>
37#include <linux/interrupt.h>
38#include <linux/irq.h>
39#include <linux/module.h>
40#include <linux/smp.h>
41#include <linux/percpu.h>
42#include <linux/cpu.h>
43
44#include <asm/sync_bitops.h>
45#include <asm/xen/hypercall.h>
46#include <asm/xen/hypervisor.h>
47
48#include <xen/xen.h>
49#include <xen/xen-ops.h>
50#include <xen/events.h>
51#include <xen/interface/xen.h>
52#include <xen/interface/event_channel.h>
53#include <xen/page.h>
54
55#include "events_internal.h"
56
57#define EVENT_WORDS_PER_PAGE (XEN_PAGE_SIZE / sizeof(event_word_t))
58#define MAX_EVENT_ARRAY_PAGES (EVTCHN_FIFO_NR_CHANNELS / EVENT_WORDS_PER_PAGE)
59
60struct evtchn_fifo_queue {
61	uint32_t head[EVTCHN_FIFO_MAX_QUEUES];
62};
63
64static DEFINE_PER_CPU(struct evtchn_fifo_control_block *, cpu_control_block);
65static DEFINE_PER_CPU(struct evtchn_fifo_queue, cpu_queue);
66static event_word_t *event_array[MAX_EVENT_ARRAY_PAGES] __read_mostly;
67static unsigned event_array_pages __read_mostly;
68
69/*
70 * sync_set_bit() and friends must be unsigned long aligned.
71 */
72#if BITS_PER_LONG > 32
73
74#define BM(w) (unsigned long *)((unsigned long)w & ~0x7UL)
75#define EVTCHN_FIFO_BIT(b, w) \
76    (((unsigned long)w & 0x4UL) ? (EVTCHN_FIFO_ ##b + 32) : EVTCHN_FIFO_ ##b)
77
78#else
79
80#define BM(w) ((unsigned long *)(w))
81#define EVTCHN_FIFO_BIT(b, w) EVTCHN_FIFO_ ##b
82
83#endif
84
85static inline event_word_t *event_word_from_port(unsigned port)
86{
87	unsigned i = port / EVENT_WORDS_PER_PAGE;
88
89	return event_array[i] + port % EVENT_WORDS_PER_PAGE;
90}
91
92static unsigned evtchn_fifo_max_channels(void)
93{
94	return EVTCHN_FIFO_NR_CHANNELS;
95}
96
97static unsigned evtchn_fifo_nr_channels(void)
98{
99	return event_array_pages * EVENT_WORDS_PER_PAGE;
100}
101
102static int init_control_block(int cpu,
103                              struct evtchn_fifo_control_block *control_block)
104{
105	struct evtchn_fifo_queue *q = &per_cpu(cpu_queue, cpu);
106	struct evtchn_init_control init_control;
107	unsigned int i;
108
109	/* Reset the control block and the local HEADs. */
110	clear_page(control_block);
111	for (i = 0; i < EVTCHN_FIFO_MAX_QUEUES; i++)
112		q->head[i] = 0;
113
114	init_control.control_gfn = virt_to_gfn(control_block);
115	init_control.offset      = 0;
116	init_control.vcpu        = cpu;
117
118	return HYPERVISOR_event_channel_op(EVTCHNOP_init_control, &init_control);
119}
120
121static void free_unused_array_pages(void)
122{
123	unsigned i;
124
125	for (i = event_array_pages; i < MAX_EVENT_ARRAY_PAGES; i++) {
126		if (!event_array[i])
127			break;
128		free_page((unsigned long)event_array[i]);
129		event_array[i] = NULL;
130	}
131}
132
133static void init_array_page(event_word_t *array_page)
134{
135	unsigned i;
136
137	for (i = 0; i < EVENT_WORDS_PER_PAGE; i++)
138		array_page[i] = 1 << EVTCHN_FIFO_MASKED;
139}
140
141static int evtchn_fifo_setup(struct irq_info *info)
142{
143	unsigned port = info->evtchn;
144	unsigned new_array_pages;
145	int ret;
146
147	new_array_pages = port / EVENT_WORDS_PER_PAGE + 1;
148
149	if (new_array_pages > MAX_EVENT_ARRAY_PAGES)
150		return -EINVAL;
151
152	while (event_array_pages < new_array_pages) {
153		void *array_page;
154		struct evtchn_expand_array expand_array;
155
156		/* Might already have a page if we've resumed. */
157		array_page = event_array[event_array_pages];
158		if (!array_page) {
159			array_page = (void *)__get_free_page(GFP_KERNEL);
160			if (array_page == NULL) {
161				ret = -ENOMEM;
162				goto error;
163			}
164			event_array[event_array_pages] = array_page;
165		}
166
167		/* Mask all events in this page before adding it. */
168		init_array_page(array_page);
169
170		expand_array.array_gfn = virt_to_gfn(array_page);
171
172		ret = HYPERVISOR_event_channel_op(EVTCHNOP_expand_array, &expand_array);
173		if (ret < 0)
174			goto error;
175
176		event_array_pages++;
177	}
178	return 0;
179
180  error:
181	if (event_array_pages == 0)
182		panic("xen: unable to expand event array with initial page (%d)\n", ret);
183	else
184		pr_err("unable to expand event array (%d)\n", ret);
185	free_unused_array_pages();
186	return ret;
187}
188
189static void evtchn_fifo_bind_to_cpu(struct irq_info *info, unsigned cpu)
190{
191	/* no-op */
192}
193
194static void evtchn_fifo_clear_pending(unsigned port)
195{
196	event_word_t *word = event_word_from_port(port);
197	sync_clear_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word));
198}
199
200static void evtchn_fifo_set_pending(unsigned port)
201{
202	event_word_t *word = event_word_from_port(port);
203	sync_set_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word));
204}
205
206static bool evtchn_fifo_is_pending(unsigned port)
207{
208	event_word_t *word = event_word_from_port(port);
209	return sync_test_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word));
210}
211
212static bool evtchn_fifo_test_and_set_mask(unsigned port)
213{
214	event_word_t *word = event_word_from_port(port);
215	return sync_test_and_set_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word));
216}
217
218static void evtchn_fifo_mask(unsigned port)
219{
220	event_word_t *word = event_word_from_port(port);
221	sync_set_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word));
222}
223
224static bool evtchn_fifo_is_masked(unsigned port)
225{
226	event_word_t *word = event_word_from_port(port);
227	return sync_test_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word));
228}
229/*
230 * Clear MASKED, spinning if BUSY is set.
231 */
232static void clear_masked(volatile event_word_t *word)
233{
234	event_word_t new, old, w;
235
236	w = *word;
237
238	do {
239		old = w & ~(1 << EVTCHN_FIFO_BUSY);
240		new = old & ~(1 << EVTCHN_FIFO_MASKED);
241		w = sync_cmpxchg(word, old, new);
242	} while (w != old);
243}
244
245static void evtchn_fifo_unmask(unsigned port)
246{
247	event_word_t *word = event_word_from_port(port);
248
249	BUG_ON(!irqs_disabled());
250
251	clear_masked(word);
252	if (evtchn_fifo_is_pending(port)) {
253		struct evtchn_unmask unmask = { .port = port };
254		(void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask);
255	}
256}
257
258static uint32_t clear_linked(volatile event_word_t *word)
259{
260	event_word_t new, old, w;
261
262	w = *word;
263
264	do {
265		old = w;
266		new = (w & ~((1 << EVTCHN_FIFO_LINKED)
267			     | EVTCHN_FIFO_LINK_MASK));
268	} while ((w = sync_cmpxchg(word, old, new)) != old);
269
270	return w & EVTCHN_FIFO_LINK_MASK;
271}
272
273static void handle_irq_for_port(unsigned port)
274{
275	int irq;
276
277	irq = get_evtchn_to_irq(port);
278	if (irq != -1)
279		generic_handle_irq(irq);
280}
281
282static void consume_one_event(unsigned cpu,
283			      struct evtchn_fifo_control_block *control_block,
284			      unsigned priority, unsigned long *ready,
285			      bool drop)
286{
287	struct evtchn_fifo_queue *q = &per_cpu(cpu_queue, cpu);
288	uint32_t head;
289	unsigned port;
290	event_word_t *word;
291
292	head = q->head[priority];
293
294	/*
295	 * Reached the tail last time?  Read the new HEAD from the
296	 * control block.
297	 */
298	if (head == 0) {
299		rmb(); /* Ensure word is up-to-date before reading head. */
300		head = control_block->head[priority];
301	}
302
303	port = head;
304	word = event_word_from_port(port);
305	head = clear_linked(word);
306
307	/*
308	 * If the link is non-zero, there are more events in the
309	 * queue, otherwise the queue is empty.
310	 *
311	 * If the queue is empty, clear this priority from our local
312	 * copy of the ready word.
313	 */
314	if (head == 0)
315		clear_bit(priority, ready);
316
317	if (evtchn_fifo_is_pending(port) && !evtchn_fifo_is_masked(port)) {
318		if (unlikely(drop))
319			pr_warn("Dropping pending event for port %u\n", port);
320		else
321			handle_irq_for_port(port);
322	}
323
324	q->head[priority] = head;
325}
326
327static void __evtchn_fifo_handle_events(unsigned cpu, bool drop)
328{
329	struct evtchn_fifo_control_block *control_block;
330	unsigned long ready;
331	unsigned q;
332
333	control_block = per_cpu(cpu_control_block, cpu);
334
335	ready = xchg(&control_block->ready, 0);
336
337	while (ready) {
338		q = find_first_bit(&ready, EVTCHN_FIFO_MAX_QUEUES);
339		consume_one_event(cpu, control_block, q, &ready, drop);
340		ready |= xchg(&control_block->ready, 0);
341	}
342}
343
344static void evtchn_fifo_handle_events(unsigned cpu)
345{
346	__evtchn_fifo_handle_events(cpu, false);
347}
348
349static void evtchn_fifo_resume(void)
350{
351	unsigned cpu;
352
353	for_each_possible_cpu(cpu) {
354		void *control_block = per_cpu(cpu_control_block, cpu);
355		int ret;
356
357		if (!control_block)
358			continue;
359
360		/*
361		 * If this CPU is offline, take the opportunity to
362		 * free the control block while it is not being
363		 * used.
364		 */
365		if (!cpu_online(cpu)) {
366			free_page((unsigned long)control_block);
367			per_cpu(cpu_control_block, cpu) = NULL;
368			continue;
369		}
370
371		ret = init_control_block(cpu, control_block);
372		if (ret < 0)
373			BUG();
374	}
375
376	/*
377	 * The event array starts out as empty again and is extended
378	 * as normal when events are bound.  The existing pages will
379	 * be reused.
380	 */
381	event_array_pages = 0;
382}
383
384static const struct evtchn_ops evtchn_ops_fifo = {
385	.max_channels      = evtchn_fifo_max_channels,
386	.nr_channels       = evtchn_fifo_nr_channels,
387	.setup             = evtchn_fifo_setup,
388	.bind_to_cpu       = evtchn_fifo_bind_to_cpu,
389	.clear_pending     = evtchn_fifo_clear_pending,
390	.set_pending       = evtchn_fifo_set_pending,
391	.is_pending        = evtchn_fifo_is_pending,
392	.test_and_set_mask = evtchn_fifo_test_and_set_mask,
393	.mask              = evtchn_fifo_mask,
394	.unmask            = evtchn_fifo_unmask,
395	.handle_events     = evtchn_fifo_handle_events,
396	.resume            = evtchn_fifo_resume,
397};
398
399static int evtchn_fifo_alloc_control_block(unsigned cpu)
400{
401	void *control_block = NULL;
402	int ret = -ENOMEM;
403
404	control_block = (void *)__get_free_page(GFP_KERNEL);
405	if (control_block == NULL)
406		goto error;
407
408	ret = init_control_block(cpu, control_block);
409	if (ret < 0)
410		goto error;
411
412	per_cpu(cpu_control_block, cpu) = control_block;
413
414	return 0;
415
416  error:
417	free_page((unsigned long)control_block);
418	return ret;
419}
420
421static int evtchn_fifo_cpu_notification(struct notifier_block *self,
422						  unsigned long action,
423						  void *hcpu)
424{
425	int cpu = (long)hcpu;
426	int ret = 0;
427
428	switch (action) {
429	case CPU_UP_PREPARE:
430		if (!per_cpu(cpu_control_block, cpu))
431			ret = evtchn_fifo_alloc_control_block(cpu);
432		break;
433	case CPU_DEAD:
434		__evtchn_fifo_handle_events(cpu, true);
435		break;
436	default:
437		break;
438	}
439	return ret < 0 ? NOTIFY_BAD : NOTIFY_OK;
440}
441
442static struct notifier_block evtchn_fifo_cpu_notifier = {
443	.notifier_call	= evtchn_fifo_cpu_notification,
444};
445
446int __init xen_evtchn_fifo_init(void)
447{
448	int cpu = get_cpu();
449	int ret;
450
451	ret = evtchn_fifo_alloc_control_block(cpu);
452	if (ret < 0)
453		goto out;
454
455	pr_info("Using FIFO-based ABI\n");
456
457	evtchn_ops = &evtchn_ops_fifo;
458
459	register_cpu_notifier(&evtchn_fifo_cpu_notifier);
460out:
461	put_cpu();
462	return ret;
463}
464