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 /* 70 * sync_set_bit() and friends must be unsigned long aligned on non-x86 71 * platforms. 72 */ 73 #if !defined(CONFIG_X86) && BITS_PER_LONG > 32 74 75 #define BM(w) (unsigned long *)((unsigned long)w & ~0x7UL) 76 #define EVTCHN_FIFO_BIT(b, w) \ 77 (((unsigned long)w & 0x4UL) ? (EVTCHN_FIFO_ ##b + 32) : EVTCHN_FIFO_ ##b) 78 79 #else 80 81 #define BM(w) ((unsigned long *)(w)) 82 #define EVTCHN_FIFO_BIT(b, w) EVTCHN_FIFO_ ##b 83 84 #endif 85 86 static inline event_word_t *event_word_from_port(unsigned port) 87 { 88 unsigned i = port / EVENT_WORDS_PER_PAGE; 89 90 return event_array[i] + port % EVENT_WORDS_PER_PAGE; 91 } 92 93 static unsigned evtchn_fifo_max_channels(void) 94 { 95 return EVTCHN_FIFO_NR_CHANNELS; 96 } 97 98 static unsigned evtchn_fifo_nr_channels(void) 99 { 100 return event_array_pages * EVENT_WORDS_PER_PAGE; 101 } 102 103 static void free_unused_array_pages(void) 104 { 105 unsigned i; 106 107 for (i = event_array_pages; i < MAX_EVENT_ARRAY_PAGES; i++) { 108 if (!event_array[i]) 109 break; 110 free_page((unsigned long)event_array[i]); 111 event_array[i] = NULL; 112 } 113 } 114 115 static void init_array_page(event_word_t *array_page) 116 { 117 unsigned i; 118 119 for (i = 0; i < EVENT_WORDS_PER_PAGE; i++) 120 array_page[i] = 1 << EVTCHN_FIFO_MASKED; 121 } 122 123 static int evtchn_fifo_setup(struct irq_info *info) 124 { 125 unsigned port = info->evtchn; 126 unsigned new_array_pages; 127 int ret; 128 129 new_array_pages = port / EVENT_WORDS_PER_PAGE + 1; 130 131 if (new_array_pages > MAX_EVENT_ARRAY_PAGES) 132 return -EINVAL; 133 134 while (event_array_pages < new_array_pages) { 135 void *array_page; 136 struct evtchn_expand_array expand_array; 137 138 /* Might already have a page if we've resumed. */ 139 array_page = event_array[event_array_pages]; 140 if (!array_page) { 141 array_page = (void *)__get_free_page(GFP_KERNEL); 142 if (array_page == NULL) { 143 ret = -ENOMEM; 144 goto error; 145 } 146 event_array[event_array_pages] = array_page; 147 } 148 149 /* Mask all events in this page before adding it. */ 150 init_array_page(array_page); 151 152 expand_array.array_gfn = virt_to_mfn(array_page); 153 154 ret = HYPERVISOR_event_channel_op(EVTCHNOP_expand_array, &expand_array); 155 if (ret < 0) 156 goto error; 157 158 event_array_pages++; 159 } 160 return 0; 161 162 error: 163 if (event_array_pages == 0) 164 panic("xen: unable to expand event array with initial page (%d)\n", ret); 165 else 166 pr_err("unable to expand event array (%d)\n", ret); 167 free_unused_array_pages(); 168 return ret; 169 } 170 171 static void evtchn_fifo_bind_to_cpu(struct irq_info *info, unsigned cpu) 172 { 173 /* no-op */ 174 } 175 176 static void evtchn_fifo_clear_pending(unsigned port) 177 { 178 event_word_t *word = event_word_from_port(port); 179 sync_clear_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word)); 180 } 181 182 static void evtchn_fifo_set_pending(unsigned port) 183 { 184 event_word_t *word = event_word_from_port(port); 185 sync_set_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word)); 186 } 187 188 static bool evtchn_fifo_is_pending(unsigned port) 189 { 190 event_word_t *word = event_word_from_port(port); 191 return sync_test_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word)); 192 } 193 194 static bool evtchn_fifo_test_and_set_mask(unsigned port) 195 { 196 event_word_t *word = event_word_from_port(port); 197 return sync_test_and_set_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word)); 198 } 199 200 static void evtchn_fifo_mask(unsigned port) 201 { 202 event_word_t *word = event_word_from_port(port); 203 sync_set_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word)); 204 } 205 206 static bool evtchn_fifo_is_masked(unsigned port) 207 { 208 event_word_t *word = event_word_from_port(port); 209 return sync_test_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word)); 210 } 211 /* 212 * Clear MASKED, spinning if BUSY is set. 213 */ 214 static void clear_masked(volatile event_word_t *word) 215 { 216 event_word_t new, old, w; 217 218 w = *word; 219 220 do { 221 old = w & ~(1 << EVTCHN_FIFO_BUSY); 222 new = old & ~(1 << EVTCHN_FIFO_MASKED); 223 w = sync_cmpxchg(word, old, new); 224 } while (w != old); 225 } 226 227 static void evtchn_fifo_unmask(unsigned port) 228 { 229 event_word_t *word = event_word_from_port(port); 230 231 BUG_ON(!irqs_disabled()); 232 233 clear_masked(word); 234 if (evtchn_fifo_is_pending(port)) { 235 struct evtchn_unmask unmask = { .port = port }; 236 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask); 237 } 238 } 239 240 static uint32_t clear_linked(volatile event_word_t *word) 241 { 242 event_word_t new, old, w; 243 244 w = *word; 245 246 do { 247 old = w; 248 new = (w & ~((1 << EVTCHN_FIFO_LINKED) 249 | EVTCHN_FIFO_LINK_MASK)); 250 } while ((w = sync_cmpxchg(word, old, new)) != old); 251 252 return w & EVTCHN_FIFO_LINK_MASK; 253 } 254 255 static void handle_irq_for_port(unsigned port) 256 { 257 int irq; 258 259 irq = get_evtchn_to_irq(port); 260 if (irq != -1) 261 generic_handle_irq(irq); 262 } 263 264 static void consume_one_event(unsigned cpu, 265 struct evtchn_fifo_control_block *control_block, 266 unsigned priority, unsigned long *ready) 267 { 268 struct evtchn_fifo_queue *q = &per_cpu(cpu_queue, cpu); 269 uint32_t head; 270 unsigned port; 271 event_word_t *word; 272 273 head = q->head[priority]; 274 275 /* 276 * Reached the tail last time? Read the new HEAD from the 277 * control block. 278 */ 279 if (head == 0) { 280 rmb(); /* Ensure word is up-to-date before reading head. */ 281 head = control_block->head[priority]; 282 } 283 284 port = head; 285 word = event_word_from_port(port); 286 head = clear_linked(word); 287 288 /* 289 * If the link is non-zero, there are more events in the 290 * queue, otherwise the queue is empty. 291 * 292 * If the queue is empty, clear this priority from our local 293 * copy of the ready word. 294 */ 295 if (head == 0) 296 clear_bit(priority, ready); 297 298 if (evtchn_fifo_is_pending(port) && !evtchn_fifo_is_masked(port)) 299 handle_irq_for_port(port); 300 301 q->head[priority] = head; 302 } 303 304 static void evtchn_fifo_handle_events(unsigned cpu) 305 { 306 struct evtchn_fifo_control_block *control_block; 307 unsigned long ready; 308 unsigned q; 309 310 control_block = per_cpu(cpu_control_block, cpu); 311 312 ready = xchg(&control_block->ready, 0); 313 314 while (ready) { 315 q = find_first_bit(BM(&ready), EVTCHN_FIFO_MAX_QUEUES); 316 consume_one_event(cpu, control_block, q, &ready); 317 ready |= xchg(&control_block->ready, 0); 318 } 319 } 320 321 static void evtchn_fifo_resume(void) 322 { 323 unsigned cpu; 324 325 for_each_possible_cpu(cpu) { 326 void *control_block = per_cpu(cpu_control_block, cpu); 327 struct evtchn_init_control init_control; 328 int ret; 329 330 if (!control_block) 331 continue; 332 333 /* 334 * If this CPU is offline, take the opportunity to 335 * free the control block while it is not being 336 * used. 337 */ 338 if (!cpu_online(cpu)) { 339 free_page((unsigned long)control_block); 340 per_cpu(cpu_control_block, cpu) = NULL; 341 continue; 342 } 343 344 init_control.control_gfn = virt_to_mfn(control_block); 345 init_control.offset = 0; 346 init_control.vcpu = cpu; 347 348 ret = HYPERVISOR_event_channel_op(EVTCHNOP_init_control, 349 &init_control); 350 if (ret < 0) 351 BUG(); 352 } 353 354 /* 355 * The event array starts out as empty again and is extended 356 * as normal when events are bound. The existing pages will 357 * be reused. 358 */ 359 event_array_pages = 0; 360 } 361 362 static const struct evtchn_ops evtchn_ops_fifo = { 363 .max_channels = evtchn_fifo_max_channels, 364 .nr_channels = evtchn_fifo_nr_channels, 365 .setup = evtchn_fifo_setup, 366 .bind_to_cpu = evtchn_fifo_bind_to_cpu, 367 .clear_pending = evtchn_fifo_clear_pending, 368 .set_pending = evtchn_fifo_set_pending, 369 .is_pending = evtchn_fifo_is_pending, 370 .test_and_set_mask = evtchn_fifo_test_and_set_mask, 371 .mask = evtchn_fifo_mask, 372 .unmask = evtchn_fifo_unmask, 373 .handle_events = evtchn_fifo_handle_events, 374 .resume = evtchn_fifo_resume, 375 }; 376 377 static int evtchn_fifo_init_control_block(unsigned cpu) 378 { 379 struct page *control_block = NULL; 380 struct evtchn_init_control init_control; 381 int ret = -ENOMEM; 382 383 control_block = alloc_page(GFP_KERNEL|__GFP_ZERO); 384 if (control_block == NULL) 385 goto error; 386 387 init_control.control_gfn = virt_to_mfn(page_address(control_block)); 388 init_control.offset = 0; 389 init_control.vcpu = cpu; 390 391 ret = HYPERVISOR_event_channel_op(EVTCHNOP_init_control, &init_control); 392 if (ret < 0) 393 goto error; 394 395 per_cpu(cpu_control_block, cpu) = page_address(control_block); 396 397 return 0; 398 399 error: 400 __free_page(control_block); 401 return ret; 402 } 403 404 static int evtchn_fifo_cpu_notification(struct notifier_block *self, 405 unsigned long action, 406 void *hcpu) 407 { 408 int cpu = (long)hcpu; 409 int ret = 0; 410 411 switch (action) { 412 case CPU_UP_PREPARE: 413 if (!per_cpu(cpu_control_block, cpu)) 414 ret = evtchn_fifo_init_control_block(cpu); 415 break; 416 default: 417 break; 418 } 419 return ret < 0 ? NOTIFY_BAD : NOTIFY_OK; 420 } 421 422 static struct notifier_block evtchn_fifo_cpu_notifier = { 423 .notifier_call = evtchn_fifo_cpu_notification, 424 }; 425 426 int __init xen_evtchn_fifo_init(void) 427 { 428 int cpu = get_cpu(); 429 int ret; 430 431 ret = evtchn_fifo_init_control_block(cpu); 432 if (ret < 0) 433 goto out; 434 435 pr_info("Using FIFO-based ABI\n"); 436 437 evtchn_ops = &evtchn_ops_fifo; 438 439 register_cpu_notifier(&evtchn_fifo_cpu_notifier); 440 out: 441 put_cpu(); 442 return ret; 443 } 444