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/smp.h> 40 #include <linux/percpu.h> 41 #include <linux/cpu.h> 42 43 #include <asm/barrier.h> 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 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. 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 85 static inline event_word_t *event_word_from_port(evtchn_port_t port) 86 { 87 unsigned i = port / EVENT_WORDS_PER_PAGE; 88 89 return event_array[i] + port % EVENT_WORDS_PER_PAGE; 90 } 91 92 static unsigned evtchn_fifo_max_channels(void) 93 { 94 return EVTCHN_FIFO_NR_CHANNELS; 95 } 96 97 static unsigned evtchn_fifo_nr_channels(void) 98 { 99 return event_array_pages * EVENT_WORDS_PER_PAGE; 100 } 101 102 static 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 = xen_vcpu_nr(cpu); 117 118 return HYPERVISOR_event_channel_op(EVTCHNOP_init_control, &init_control); 119 } 120 121 static 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 } 132 133 static 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 } 140 141 static int evtchn_fifo_setup(evtchn_port_t port) 142 { 143 unsigned new_array_pages; 144 int ret; 145 146 new_array_pages = port / EVENT_WORDS_PER_PAGE + 1; 147 148 if (new_array_pages > MAX_EVENT_ARRAY_PAGES) 149 return -EINVAL; 150 151 while (event_array_pages < new_array_pages) { 152 void *array_page; 153 struct evtchn_expand_array expand_array; 154 155 /* Might already have a page if we've resumed. */ 156 array_page = event_array[event_array_pages]; 157 if (!array_page) { 158 array_page = (void *)__get_free_page(GFP_KERNEL); 159 if (array_page == NULL) { 160 ret = -ENOMEM; 161 goto error; 162 } 163 event_array[event_array_pages] = array_page; 164 } 165 166 /* Mask all events in this page before adding it. */ 167 init_array_page(array_page); 168 169 expand_array.array_gfn = virt_to_gfn(array_page); 170 171 ret = HYPERVISOR_event_channel_op(EVTCHNOP_expand_array, &expand_array); 172 if (ret < 0) 173 goto error; 174 175 event_array_pages++; 176 } 177 return 0; 178 179 error: 180 if (event_array_pages == 0) 181 panic("xen: unable to expand event array with initial page (%d)\n", ret); 182 else 183 pr_err("unable to expand event array (%d)\n", ret); 184 free_unused_array_pages(); 185 return ret; 186 } 187 188 static void evtchn_fifo_bind_to_cpu(evtchn_port_t evtchn, unsigned int cpu, 189 unsigned int old_cpu) 190 { 191 /* no-op */ 192 } 193 194 static void evtchn_fifo_clear_pending(evtchn_port_t port) 195 { 196 event_word_t *word = event_word_from_port(port); 197 sync_clear_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word)); 198 } 199 200 static void evtchn_fifo_set_pending(evtchn_port_t port) 201 { 202 event_word_t *word = event_word_from_port(port); 203 sync_set_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word)); 204 } 205 206 static bool evtchn_fifo_is_pending(evtchn_port_t 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 } 211 212 static bool evtchn_fifo_test_and_set_mask(evtchn_port_t 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 } 217 218 static void evtchn_fifo_mask(evtchn_port_t port) 219 { 220 event_word_t *word = event_word_from_port(port); 221 sync_set_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word)); 222 } 223 224 static bool evtchn_fifo_is_masked(evtchn_port_t 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 if not PENDING, spinning if BUSY is set. 231 * Return true if mask was cleared. 232 */ 233 static bool clear_masked_cond(volatile event_word_t *word) 234 { 235 event_word_t new, old, w; 236 237 w = *word; 238 239 do { 240 if (!(w & (1 << EVTCHN_FIFO_MASKED))) 241 return true; 242 243 if (w & (1 << EVTCHN_FIFO_PENDING)) 244 return false; 245 246 old = w & ~(1 << EVTCHN_FIFO_BUSY); 247 new = old & ~(1 << EVTCHN_FIFO_MASKED); 248 w = sync_cmpxchg(word, old, new); 249 } while (w != old); 250 251 return true; 252 } 253 254 static void evtchn_fifo_unmask(evtchn_port_t port) 255 { 256 event_word_t *word = event_word_from_port(port); 257 258 BUG_ON(!irqs_disabled()); 259 260 if (!clear_masked_cond(word)) { 261 struct evtchn_unmask unmask = { .port = port }; 262 (void)HYPERVISOR_event_channel_op(EVTCHNOP_unmask, &unmask); 263 } 264 } 265 266 static uint32_t clear_linked(volatile event_word_t *word) 267 { 268 event_word_t new, old, w; 269 270 w = *word; 271 272 do { 273 old = w; 274 new = (w & ~((1 << EVTCHN_FIFO_LINKED) 275 | EVTCHN_FIFO_LINK_MASK)); 276 } while ((w = sync_cmpxchg(word, old, new)) != old); 277 278 return w & EVTCHN_FIFO_LINK_MASK; 279 } 280 281 static void consume_one_event(unsigned cpu, struct evtchn_loop_ctrl *ctrl, 282 struct evtchn_fifo_control_block *control_block, 283 unsigned priority, unsigned long *ready) 284 { 285 struct evtchn_fifo_queue *q = &per_cpu(cpu_queue, cpu); 286 uint32_t head; 287 evtchn_port_t port; 288 event_word_t *word; 289 290 head = q->head[priority]; 291 292 /* 293 * Reached the tail last time? Read the new HEAD from the 294 * control block. 295 */ 296 if (head == 0) { 297 virt_rmb(); /* Ensure word is up-to-date before reading head. */ 298 head = control_block->head[priority]; 299 } 300 301 port = head; 302 word = event_word_from_port(port); 303 head = clear_linked(word); 304 305 /* 306 * If the link is non-zero, there are more events in the 307 * queue, otherwise the queue is empty. 308 * 309 * If the queue is empty, clear this priority from our local 310 * copy of the ready word. 311 */ 312 if (head == 0) 313 clear_bit(priority, ready); 314 315 if (evtchn_fifo_is_pending(port) && !evtchn_fifo_is_masked(port)) { 316 if (unlikely(!ctrl)) 317 pr_warn("Dropping pending event for port %u\n", port); 318 else 319 handle_irq_for_port(port, ctrl); 320 } 321 322 q->head[priority] = head; 323 } 324 325 static void __evtchn_fifo_handle_events(unsigned cpu, 326 struct evtchn_loop_ctrl *ctrl) 327 { 328 struct evtchn_fifo_control_block *control_block; 329 unsigned long ready; 330 unsigned q; 331 332 control_block = per_cpu(cpu_control_block, cpu); 333 334 ready = xchg(&control_block->ready, 0); 335 336 while (ready) { 337 q = find_first_bit(&ready, EVTCHN_FIFO_MAX_QUEUES); 338 consume_one_event(cpu, ctrl, control_block, q, &ready); 339 ready |= xchg(&control_block->ready, 0); 340 } 341 } 342 343 static void evtchn_fifo_handle_events(unsigned cpu, 344 struct evtchn_loop_ctrl *ctrl) 345 { 346 __evtchn_fifo_handle_events(cpu, ctrl); 347 } 348 349 static 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 BUG_ON(ret < 0); 373 } 374 375 /* 376 * The event array starts out as empty again and is extended 377 * as normal when events are bound. The existing pages will 378 * be reused. 379 */ 380 event_array_pages = 0; 381 } 382 383 static int evtchn_fifo_alloc_control_block(unsigned cpu) 384 { 385 void *control_block = NULL; 386 int ret = -ENOMEM; 387 388 control_block = (void *)__get_free_page(GFP_KERNEL); 389 if (control_block == NULL) 390 goto error; 391 392 ret = init_control_block(cpu, control_block); 393 if (ret < 0) 394 goto error; 395 396 per_cpu(cpu_control_block, cpu) = control_block; 397 398 return 0; 399 400 error: 401 free_page((unsigned long)control_block); 402 return ret; 403 } 404 405 static int evtchn_fifo_percpu_init(unsigned int cpu) 406 { 407 if (!per_cpu(cpu_control_block, cpu)) 408 return evtchn_fifo_alloc_control_block(cpu); 409 return 0; 410 } 411 412 static int evtchn_fifo_percpu_deinit(unsigned int cpu) 413 { 414 __evtchn_fifo_handle_events(cpu, NULL); 415 return 0; 416 } 417 418 static const struct evtchn_ops evtchn_ops_fifo = { 419 .max_channels = evtchn_fifo_max_channels, 420 .nr_channels = evtchn_fifo_nr_channels, 421 .setup = evtchn_fifo_setup, 422 .bind_to_cpu = evtchn_fifo_bind_to_cpu, 423 .clear_pending = evtchn_fifo_clear_pending, 424 .set_pending = evtchn_fifo_set_pending, 425 .is_pending = evtchn_fifo_is_pending, 426 .test_and_set_mask = evtchn_fifo_test_and_set_mask, 427 .mask = evtchn_fifo_mask, 428 .unmask = evtchn_fifo_unmask, 429 .handle_events = evtchn_fifo_handle_events, 430 .resume = evtchn_fifo_resume, 431 .percpu_init = evtchn_fifo_percpu_init, 432 .percpu_deinit = evtchn_fifo_percpu_deinit, 433 }; 434 435 int __init xen_evtchn_fifo_init(void) 436 { 437 int cpu = smp_processor_id(); 438 int ret; 439 440 ret = evtchn_fifo_alloc_control_block(cpu); 441 if (ret < 0) 442 return ret; 443 444 pr_info("Using FIFO-based ABI\n"); 445 446 evtchn_ops = &evtchn_ops_fifo; 447 448 return ret; 449 } 450