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 #include <asm/xen/page.h> 48 49 #include <xen/xen.h> 50 #include <xen/xen-ops.h> 51 #include <xen/events.h> 52 #include <xen/interface/xen.h> 53 #include <xen/interface/event_channel.h> 54 55 #include "events_internal.h" 56 57 #define EVENT_WORDS_PER_PAGE (PAGE_SIZE / sizeof(event_word_t)) 58 #define MAX_EVENT_ARRAY_PAGES (EVTCHN_FIFO_NR_CHANNELS / EVENT_WORDS_PER_PAGE) 59 60 struct evtchn_fifo_queue { 61 uint32_t head[EVTCHN_FIFO_MAX_QUEUES]; 62 }; 63 64 static DEFINE_PER_CPU(struct evtchn_fifo_control_block *, cpu_control_block); 65 static DEFINE_PER_CPU(struct evtchn_fifo_queue, cpu_queue); 66 static event_word_t *event_array[MAX_EVENT_ARRAY_PAGES] __read_mostly; 67 static unsigned event_array_pages __read_mostly; 68 69 #define BM(w) ((unsigned long *)(w)) 70 71 static inline event_word_t *event_word_from_port(unsigned port) 72 { 73 unsigned i = port / EVENT_WORDS_PER_PAGE; 74 75 return event_array[i] + port % EVENT_WORDS_PER_PAGE; 76 } 77 78 static unsigned evtchn_fifo_max_channels(void) 79 { 80 return EVTCHN_FIFO_NR_CHANNELS; 81 } 82 83 static unsigned evtchn_fifo_nr_channels(void) 84 { 85 return event_array_pages * EVENT_WORDS_PER_PAGE; 86 } 87 88 static void free_unused_array_pages(void) 89 { 90 unsigned i; 91 92 for (i = event_array_pages; i < MAX_EVENT_ARRAY_PAGES; i++) { 93 if (!event_array[i]) 94 break; 95 free_page((unsigned long)event_array[i]); 96 event_array[i] = NULL; 97 } 98 } 99 100 static void init_array_page(event_word_t *array_page) 101 { 102 unsigned i; 103 104 for (i = 0; i < EVENT_WORDS_PER_PAGE; i++) 105 array_page[i] = 1 << EVTCHN_FIFO_MASKED; 106 } 107 108 static int evtchn_fifo_setup(struct irq_info *info) 109 { 110 unsigned port = info->evtchn; 111 unsigned new_array_pages; 112 int ret; 113 114 new_array_pages = port / EVENT_WORDS_PER_PAGE + 1; 115 116 if (new_array_pages > MAX_EVENT_ARRAY_PAGES) 117 return -EINVAL; 118 119 while (event_array_pages < new_array_pages) { 120 void *array_page; 121 struct evtchn_expand_array expand_array; 122 123 /* Might already have a page if we've resumed. */ 124 array_page = event_array[event_array_pages]; 125 if (!array_page) { 126 array_page = (void *)__get_free_page(GFP_KERNEL); 127 if (array_page == NULL) { 128 ret = -ENOMEM; 129 goto error; 130 } 131 event_array[event_array_pages] = array_page; 132 } 133 134 /* Mask all events in this page before adding it. */ 135 init_array_page(array_page); 136 137 expand_array.array_gfn = virt_to_mfn(array_page); 138 139 ret = HYPERVISOR_event_channel_op(EVTCHNOP_expand_array, &expand_array); 140 if (ret < 0) 141 goto error; 142 143 event_array_pages++; 144 } 145 return 0; 146 147 error: 148 if (event_array_pages == 0) 149 panic("xen: unable to expand event array with initial page (%d)\n", ret); 150 else 151 pr_err("unable to expand event array (%d)\n", ret); 152 free_unused_array_pages(); 153 return ret; 154 } 155 156 static void evtchn_fifo_bind_to_cpu(struct irq_info *info, unsigned cpu) 157 { 158 /* no-op */ 159 } 160 161 static void evtchn_fifo_clear_pending(unsigned port) 162 { 163 event_word_t *word = event_word_from_port(port); 164 sync_clear_bit(EVTCHN_FIFO_PENDING, BM(word)); 165 } 166 167 static void evtchn_fifo_set_pending(unsigned port) 168 { 169 event_word_t *word = event_word_from_port(port); 170 sync_set_bit(EVTCHN_FIFO_PENDING, BM(word)); 171 } 172 173 static bool evtchn_fifo_is_pending(unsigned port) 174 { 175 event_word_t *word = event_word_from_port(port); 176 return sync_test_bit(EVTCHN_FIFO_PENDING, BM(word)); 177 } 178 179 static bool evtchn_fifo_test_and_set_mask(unsigned port) 180 { 181 event_word_t *word = event_word_from_port(port); 182 return sync_test_and_set_bit(EVTCHN_FIFO_MASKED, BM(word)); 183 } 184 185 static void evtchn_fifo_mask(unsigned port) 186 { 187 event_word_t *word = event_word_from_port(port); 188 sync_set_bit(EVTCHN_FIFO_MASKED, BM(word)); 189 } 190 191 /* 192 * Clear MASKED, spinning if BUSY is set. 193 */ 194 static void clear_masked(volatile event_word_t *word) 195 { 196 event_word_t new, old, w; 197 198 w = *word; 199 200 do { 201 old = w & ~(1 << EVTCHN_FIFO_BUSY); 202 new = old & ~(1 << EVTCHN_FIFO_MASKED); 203 w = sync_cmpxchg(word, old, new); 204 } while (w != old); 205 } 206 207 static void evtchn_fifo_unmask(unsigned port) 208 { 209 event_word_t *word = event_word_from_port(port); 210 211 BUG_ON(!irqs_disabled()); 212 213 clear_masked(word); 214 if (sync_test_bit(EVTCHN_FIFO_PENDING, BM(word))) { 215 struct evtchn_unmask unmask = { .port = port }; 216 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask); 217 } 218 } 219 220 static uint32_t clear_linked(volatile event_word_t *word) 221 { 222 event_word_t new, old, w; 223 224 w = *word; 225 226 do { 227 old = w; 228 new = (w & ~((1 << EVTCHN_FIFO_LINKED) 229 | EVTCHN_FIFO_LINK_MASK)); 230 } while ((w = sync_cmpxchg(word, old, new)) != old); 231 232 return w & EVTCHN_FIFO_LINK_MASK; 233 } 234 235 static void handle_irq_for_port(unsigned port) 236 { 237 int irq; 238 239 irq = get_evtchn_to_irq(port); 240 if (irq != -1) 241 generic_handle_irq(irq); 242 } 243 244 static void consume_one_event(unsigned cpu, 245 struct evtchn_fifo_control_block *control_block, 246 unsigned priority, uint32_t *ready) 247 { 248 struct evtchn_fifo_queue *q = &per_cpu(cpu_queue, cpu); 249 uint32_t head; 250 unsigned port; 251 event_word_t *word; 252 253 head = q->head[priority]; 254 255 /* 256 * Reached the tail last time? Read the new HEAD from the 257 * control block. 258 */ 259 if (head == 0) { 260 rmb(); /* Ensure word is up-to-date before reading head. */ 261 head = control_block->head[priority]; 262 } 263 264 port = head; 265 word = event_word_from_port(port); 266 head = clear_linked(word); 267 268 /* 269 * If the link is non-zero, there are more events in the 270 * queue, otherwise the queue is empty. 271 * 272 * If the queue is empty, clear this priority from our local 273 * copy of the ready word. 274 */ 275 if (head == 0) 276 clear_bit(priority, BM(ready)); 277 278 if (sync_test_bit(EVTCHN_FIFO_PENDING, BM(word)) 279 && !sync_test_bit(EVTCHN_FIFO_MASKED, BM(word))) 280 handle_irq_for_port(port); 281 282 q->head[priority] = head; 283 } 284 285 static void evtchn_fifo_handle_events(unsigned cpu) 286 { 287 struct evtchn_fifo_control_block *control_block; 288 uint32_t ready; 289 unsigned q; 290 291 control_block = per_cpu(cpu_control_block, cpu); 292 293 ready = xchg(&control_block->ready, 0); 294 295 while (ready) { 296 q = find_first_bit(BM(&ready), EVTCHN_FIFO_MAX_QUEUES); 297 consume_one_event(cpu, control_block, q, &ready); 298 ready |= xchg(&control_block->ready, 0); 299 } 300 } 301 302 static void evtchn_fifo_resume(void) 303 { 304 unsigned cpu; 305 306 for_each_possible_cpu(cpu) { 307 void *control_block = per_cpu(cpu_control_block, cpu); 308 struct evtchn_init_control init_control; 309 int ret; 310 311 if (!control_block) 312 continue; 313 314 /* 315 * If this CPU is offline, take the opportunity to 316 * free the control block while it is not being 317 * used. 318 */ 319 if (!cpu_online(cpu)) { 320 free_page((unsigned long)control_block); 321 per_cpu(cpu_control_block, cpu) = NULL; 322 continue; 323 } 324 325 init_control.control_gfn = virt_to_mfn(control_block); 326 init_control.offset = 0; 327 init_control.vcpu = cpu; 328 329 ret = HYPERVISOR_event_channel_op(EVTCHNOP_init_control, 330 &init_control); 331 if (ret < 0) 332 BUG(); 333 } 334 335 /* 336 * The event array starts out as empty again and is extended 337 * as normal when events are bound. The existing pages will 338 * be reused. 339 */ 340 event_array_pages = 0; 341 } 342 343 static const struct evtchn_ops evtchn_ops_fifo = { 344 .max_channels = evtchn_fifo_max_channels, 345 .nr_channels = evtchn_fifo_nr_channels, 346 .setup = evtchn_fifo_setup, 347 .bind_to_cpu = evtchn_fifo_bind_to_cpu, 348 .clear_pending = evtchn_fifo_clear_pending, 349 .set_pending = evtchn_fifo_set_pending, 350 .is_pending = evtchn_fifo_is_pending, 351 .test_and_set_mask = evtchn_fifo_test_and_set_mask, 352 .mask = evtchn_fifo_mask, 353 .unmask = evtchn_fifo_unmask, 354 .handle_events = evtchn_fifo_handle_events, 355 .resume = evtchn_fifo_resume, 356 }; 357 358 static int evtchn_fifo_init_control_block(unsigned cpu) 359 { 360 struct page *control_block = NULL; 361 struct evtchn_init_control init_control; 362 int ret = -ENOMEM; 363 364 control_block = alloc_page(GFP_KERNEL|__GFP_ZERO); 365 if (control_block == NULL) 366 goto error; 367 368 init_control.control_gfn = virt_to_mfn(page_address(control_block)); 369 init_control.offset = 0; 370 init_control.vcpu = cpu; 371 372 ret = HYPERVISOR_event_channel_op(EVTCHNOP_init_control, &init_control); 373 if (ret < 0) 374 goto error; 375 376 per_cpu(cpu_control_block, cpu) = page_address(control_block); 377 378 return 0; 379 380 error: 381 __free_page(control_block); 382 return ret; 383 } 384 385 static int evtchn_fifo_cpu_notification(struct notifier_block *self, 386 unsigned long action, 387 void *hcpu) 388 { 389 int cpu = (long)hcpu; 390 int ret = 0; 391 392 switch (action) { 393 case CPU_UP_PREPARE: 394 if (!per_cpu(cpu_control_block, cpu)) 395 ret = evtchn_fifo_init_control_block(cpu); 396 break; 397 default: 398 break; 399 } 400 return ret < 0 ? NOTIFY_BAD : NOTIFY_OK; 401 } 402 403 static struct notifier_block evtchn_fifo_cpu_notifier = { 404 .notifier_call = evtchn_fifo_cpu_notification, 405 }; 406 407 int __init xen_evtchn_fifo_init(void) 408 { 409 int cpu = get_cpu(); 410 int ret; 411 412 ret = evtchn_fifo_init_control_block(cpu); 413 if (ret < 0) 414 goto out; 415 416 pr_info("Using FIFO-based ABI\n"); 417 418 evtchn_ops = &evtchn_ops_fifo; 419 420 register_cpu_notifier(&evtchn_fifo_cpu_notifier); 421 out: 422 put_cpu(); 423 return ret; 424 } 425