1 /****************************************************************************** 2 * ring.h 3 * 4 * Shared producer-consumer ring macros. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a copy 7 * of this software and associated documentation files (the "Software"), to 8 * deal in the Software without restriction, including without limitation the 9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or 10 * sell copies of the Software, and to permit persons to whom the Software is 11 * furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER 22 * DEALINGS IN THE SOFTWARE. 23 * 24 * Tim Deegan and Andrew Warfield November 2004. 25 */ 26 27 #ifndef __XEN_PUBLIC_IO_RING_H__ 28 #define __XEN_PUBLIC_IO_RING_H__ 29 30 /* 31 * When #include'ing this header, you need to provide the following 32 * declaration upfront: 33 * - standard integers types (uint8_t, uint16_t, etc) 34 * They are provided by stdint.h of the standard headers. 35 * 36 * Before using the different macros, you need to provide the following 37 * macros: 38 * - xen_mb() a memory barrier 39 * - xen_rmb() a read memory barrier 40 * - xen_wmb() a write memory barrier 41 * Example of those can be found in xenctrl.h. 42 * 43 * In addition, if you intend to use the FLEX macros, you also need to 44 * provide the following, before invoking the FLEX macros: 45 * - size_t 46 * - memcpy 47 * - grant_ref_t 48 * These declarations are provided by string.h of the standard headers, 49 * and grant_table.h from the Xen public headers. 50 */ 51 52 typedef unsigned int RING_IDX; 53 54 /* Round a 32-bit unsigned constant down to the nearest power of two. */ 55 #define __RD2(_x) (((_x) & 0x00000002) ? 0x2 : ((_x) & 0x1)) 56 #define __RD4(_x) (((_x) & 0x0000000c) ? __RD2((_x)>>2)<<2 : __RD2(_x)) 57 #define __RD8(_x) (((_x) & 0x000000f0) ? __RD4((_x)>>4)<<4 : __RD4(_x)) 58 #define __RD16(_x) (((_x) & 0x0000ff00) ? __RD8((_x)>>8)<<8 : __RD8(_x)) 59 #define __RD32(_x) (((_x) & 0xffff0000) ? __RD16((_x)>>16)<<16 : __RD16(_x)) 60 61 /* 62 * Calculate size of a shared ring, given the total available space for the 63 * ring and indexes (_sz), and the name tag of the request/response structure. 64 * A ring contains as many entries as will fit, rounded down to the nearest 65 * power of two (so we can mask with (size-1) to loop around). 66 */ 67 #define __CONST_RING_SIZE(_s, _sz) \ 68 (__RD32(((_sz) - offsetof(struct _s##_sring, ring)) / \ 69 sizeof_field(struct _s##_sring, ring[0]))) 70 /* 71 * The same for passing in an actual pointer instead of a name tag. 72 */ 73 #define __RING_SIZE(_s, _sz) \ 74 (__RD32(((_sz) - (long)(_s)->ring + (long)(_s)) / sizeof((_s)->ring[0]))) 75 76 /* 77 * Macros to make the correct C datatypes for a new kind of ring. 78 * 79 * To make a new ring datatype, you need to have two message structures, 80 * let's say request_t, and response_t already defined. 81 * 82 * In a header where you want the ring datatype declared, you then do: 83 * 84 * DEFINE_RING_TYPES(mytag, request_t, response_t); 85 * 86 * These expand out to give you a set of types, as you can see below. 87 * The most important of these are: 88 * 89 * mytag_sring_t - The shared ring. 90 * mytag_front_ring_t - The 'front' half of the ring. 91 * mytag_back_ring_t - The 'back' half of the ring. 92 * 93 * To initialize a ring in your code you need to know the location and size 94 * of the shared memory area (PAGE_SIZE, for instance). To initialise 95 * the front half: 96 * 97 * mytag_front_ring_t front_ring; 98 * SHARED_RING_INIT((mytag_sring_t *)shared_page); 99 * FRONT_RING_INIT(&front_ring, (mytag_sring_t *)shared_page, PAGE_SIZE); 100 * 101 * Initializing the back follows similarly (note that only the front 102 * initializes the shared ring): 103 * 104 * mytag_back_ring_t back_ring; 105 * BACK_RING_INIT(&back_ring, (mytag_sring_t *)shared_page, PAGE_SIZE); 106 */ 107 108 #define DEFINE_RING_TYPES(__name, __req_t, __rsp_t) \ 109 \ 110 /* Shared ring entry */ \ 111 union __name##_sring_entry { \ 112 __req_t req; \ 113 __rsp_t rsp; \ 114 }; \ 115 \ 116 /* Shared ring page */ \ 117 struct __name##_sring { \ 118 RING_IDX req_prod, req_event; \ 119 RING_IDX rsp_prod, rsp_event; \ 120 union { \ 121 struct { \ 122 uint8_t smartpoll_active; \ 123 } netif; \ 124 struct { \ 125 uint8_t msg; \ 126 } tapif_user; \ 127 uint8_t pvt_pad[4]; \ 128 } pvt; \ 129 uint8_t __pad[44]; \ 130 union __name##_sring_entry ring[1]; /* variable-length */ \ 131 }; \ 132 \ 133 /* "Front" end's private variables */ \ 134 struct __name##_front_ring { \ 135 RING_IDX req_prod_pvt; \ 136 RING_IDX rsp_cons; \ 137 unsigned int nr_ents; \ 138 struct __name##_sring *sring; \ 139 }; \ 140 \ 141 /* "Back" end's private variables */ \ 142 struct __name##_back_ring { \ 143 RING_IDX rsp_prod_pvt; \ 144 RING_IDX req_cons; \ 145 unsigned int nr_ents; \ 146 struct __name##_sring *sring; \ 147 }; \ 148 \ 149 /* Syntactic sugar */ \ 150 typedef struct __name##_sring __name##_sring_t; \ 151 typedef struct __name##_front_ring __name##_front_ring_t; \ 152 typedef struct __name##_back_ring __name##_back_ring_t 153 154 /* 155 * Macros for manipulating rings. 156 * 157 * FRONT_RING_whatever works on the "front end" of a ring: here 158 * requests are pushed on to the ring and responses taken off it. 159 * 160 * BACK_RING_whatever works on the "back end" of a ring: here 161 * requests are taken off the ring and responses put on. 162 * 163 * N.B. these macros do NO INTERLOCKS OR FLOW CONTROL. 164 * This is OK in 1-for-1 request-response situations where the 165 * requestor (front end) never has more than RING_SIZE()-1 166 * outstanding requests. 167 */ 168 169 /* Initialising empty rings */ 170 #define SHARED_RING_INIT(_s) do { \ 171 (_s)->req_prod = (_s)->rsp_prod = 0; \ 172 (_s)->req_event = (_s)->rsp_event = 1; \ 173 (void)memset((_s)->pvt.pvt_pad, 0, sizeof((_s)->pvt.pvt_pad)); \ 174 (void)memset((_s)->__pad, 0, sizeof((_s)->__pad)); \ 175 } while(0) 176 177 #define FRONT_RING_INIT(_r, _s, __size) do { \ 178 (_r)->req_prod_pvt = 0; \ 179 (_r)->rsp_cons = 0; \ 180 (_r)->nr_ents = __RING_SIZE(_s, __size); \ 181 (_r)->sring = (_s); \ 182 } while (0) 183 184 #define BACK_RING_INIT(_r, _s, __size) do { \ 185 (_r)->rsp_prod_pvt = 0; \ 186 (_r)->req_cons = 0; \ 187 (_r)->nr_ents = __RING_SIZE(_s, __size); \ 188 (_r)->sring = (_s); \ 189 } while (0) 190 191 /* How big is this ring? */ 192 #define RING_SIZE(_r) \ 193 ((_r)->nr_ents) 194 195 /* Number of free requests (for use on front side only). */ 196 #define RING_FREE_REQUESTS(_r) \ 197 (RING_SIZE(_r) - ((_r)->req_prod_pvt - (_r)->rsp_cons)) 198 199 /* Test if there is an empty slot available on the front ring. 200 * (This is only meaningful from the front. ) 201 */ 202 #define RING_FULL(_r) \ 203 (RING_FREE_REQUESTS(_r) == 0) 204 205 /* Test if there are outstanding messages to be processed on a ring. */ 206 #define RING_HAS_UNCONSUMED_RESPONSES(_r) \ 207 ((_r)->sring->rsp_prod - (_r)->rsp_cons) 208 209 #ifdef __GNUC__ 210 #define RING_HAS_UNCONSUMED_REQUESTS(_r) ({ \ 211 unsigned int req = (_r)->sring->req_prod - (_r)->req_cons; \ 212 unsigned int rsp = RING_SIZE(_r) - \ 213 ((_r)->req_cons - (_r)->rsp_prod_pvt); \ 214 req < rsp ? req : rsp; \ 215 }) 216 #else 217 /* Same as above, but without the nice GCC ({ ... }) syntax. */ 218 #define RING_HAS_UNCONSUMED_REQUESTS(_r) \ 219 ((((_r)->sring->req_prod - (_r)->req_cons) < \ 220 (RING_SIZE(_r) - ((_r)->req_cons - (_r)->rsp_prod_pvt))) ? \ 221 ((_r)->sring->req_prod - (_r)->req_cons) : \ 222 (RING_SIZE(_r) - ((_r)->req_cons - (_r)->rsp_prod_pvt))) 223 #endif 224 225 /* Direct access to individual ring elements, by index. */ 226 #define RING_GET_REQUEST(_r, _idx) \ 227 (&((_r)->sring->ring[((_idx) & (RING_SIZE(_r) - 1))].req)) 228 229 /* 230 * Get a local copy of a request. 231 * 232 * Use this in preference to RING_GET_REQUEST() so all processing is 233 * done on a local copy that cannot be modified by the other end. 234 * 235 * Note that https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58145 may cause this 236 * to be ineffective where _req is a struct which consists of only bitfields. 237 */ 238 #define RING_COPY_REQUEST(_r, _idx, _req) do { \ 239 /* Use volatile to force the copy into _req. */ \ 240 *(_req) = *(volatile typeof(_req))RING_GET_REQUEST(_r, _idx); \ 241 } while (0) 242 243 #define RING_GET_RESPONSE(_r, _idx) \ 244 (&((_r)->sring->ring[((_idx) & (RING_SIZE(_r) - 1))].rsp)) 245 246 /* Loop termination condition: Would the specified index overflow the ring? */ 247 #define RING_REQUEST_CONS_OVERFLOW(_r, _cons) \ 248 (((_cons) - (_r)->rsp_prod_pvt) >= RING_SIZE(_r)) 249 250 /* Ill-behaved frontend determination: Can there be this many requests? */ 251 #define RING_REQUEST_PROD_OVERFLOW(_r, _prod) \ 252 (((_prod) - (_r)->rsp_prod_pvt) > RING_SIZE(_r)) 253 254 #define RING_PUSH_REQUESTS(_r) do { \ 255 xen_wmb(); /* back sees requests /before/ updated producer index */ \ 256 (_r)->sring->req_prod = (_r)->req_prod_pvt; \ 257 } while (0) 258 259 #define RING_PUSH_RESPONSES(_r) do { \ 260 xen_wmb(); /* front sees resps /before/ updated producer index */ \ 261 (_r)->sring->rsp_prod = (_r)->rsp_prod_pvt; \ 262 } while (0) 263 264 /* 265 * Notification hold-off (req_event and rsp_event): 266 * 267 * When queueing requests or responses on a shared ring, it may not always be 268 * necessary to notify the remote end. For example, if requests are in flight 269 * in a backend, the front may be able to queue further requests without 270 * notifying the back (if the back checks for new requests when it queues 271 * responses). 272 * 273 * When enqueuing requests or responses: 274 * 275 * Use RING_PUSH_{REQUESTS,RESPONSES}_AND_CHECK_NOTIFY(). The second argument 276 * is a boolean return value. True indicates that the receiver requires an 277 * asynchronous notification. 278 * 279 * After dequeuing requests or responses (before sleeping the connection): 280 * 281 * Use RING_FINAL_CHECK_FOR_REQUESTS() or RING_FINAL_CHECK_FOR_RESPONSES(). 282 * The second argument is a boolean return value. True indicates that there 283 * are pending messages on the ring (i.e., the connection should not be put 284 * to sleep). 285 * 286 * These macros will set the req_event/rsp_event field to trigger a 287 * notification on the very next message that is enqueued. If you want to 288 * create batches of work (i.e., only receive a notification after several 289 * messages have been enqueued) then you will need to create a customised 290 * version of the FINAL_CHECK macro in your own code, which sets the event 291 * field appropriately. 292 */ 293 294 #define RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(_r, _notify) do { \ 295 RING_IDX __old = (_r)->sring->req_prod; \ 296 RING_IDX __new = (_r)->req_prod_pvt; \ 297 xen_wmb(); /* back sees requests /before/ updated producer index */ \ 298 (_r)->sring->req_prod = __new; \ 299 xen_mb(); /* back sees new requests /before/ we check req_event */ \ 300 (_notify) = ((RING_IDX)(__new - (_r)->sring->req_event) < \ 301 (RING_IDX)(__new - __old)); \ 302 } while (0) 303 304 #define RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(_r, _notify) do { \ 305 RING_IDX __old = (_r)->sring->rsp_prod; \ 306 RING_IDX __new = (_r)->rsp_prod_pvt; \ 307 xen_wmb(); /* front sees resps /before/ updated producer index */ \ 308 (_r)->sring->rsp_prod = __new; \ 309 xen_mb(); /* front sees new resps /before/ we check rsp_event */ \ 310 (_notify) = ((RING_IDX)(__new - (_r)->sring->rsp_event) < \ 311 (RING_IDX)(__new - __old)); \ 312 } while (0) 313 314 #define RING_FINAL_CHECK_FOR_REQUESTS(_r, _work_to_do) do { \ 315 (_work_to_do) = RING_HAS_UNCONSUMED_REQUESTS(_r); \ 316 if (_work_to_do) break; \ 317 (_r)->sring->req_event = (_r)->req_cons + 1; \ 318 xen_mb(); \ 319 (_work_to_do) = RING_HAS_UNCONSUMED_REQUESTS(_r); \ 320 } while (0) 321 322 #define RING_FINAL_CHECK_FOR_RESPONSES(_r, _work_to_do) do { \ 323 (_work_to_do) = RING_HAS_UNCONSUMED_RESPONSES(_r); \ 324 if (_work_to_do) break; \ 325 (_r)->sring->rsp_event = (_r)->rsp_cons + 1; \ 326 xen_mb(); \ 327 (_work_to_do) = RING_HAS_UNCONSUMED_RESPONSES(_r); \ 328 } while (0) 329 330 331 /* 332 * DEFINE_XEN_FLEX_RING_AND_INTF defines two monodirectional rings and 333 * functions to check if there is data on the ring, and to read and 334 * write to them. 335 * 336 * DEFINE_XEN_FLEX_RING is similar to DEFINE_XEN_FLEX_RING_AND_INTF, but 337 * does not define the indexes page. As different protocols can have 338 * extensions to the basic format, this macro allow them to define their 339 * own struct. 340 * 341 * XEN_FLEX_RING_SIZE 342 * Convenience macro to calculate the size of one of the two rings 343 * from the overall order. 344 * 345 * $NAME_mask 346 * Function to apply the size mask to an index, to reduce the index 347 * within the range [0-size]. 348 * 349 * $NAME_read_packet 350 * Function to read data from the ring. The amount of data to read is 351 * specified by the "size" argument. 352 * 353 * $NAME_write_packet 354 * Function to write data to the ring. The amount of data to write is 355 * specified by the "size" argument. 356 * 357 * $NAME_get_ring_ptr 358 * Convenience function that returns a pointer to read/write to the 359 * ring at the right location. 360 * 361 * $NAME_data_intf 362 * Indexes page, shared between frontend and backend. It also 363 * contains the array of grant refs. 364 * 365 * $NAME_queued 366 * Function to calculate how many bytes are currently on the ring, 367 * ready to be read. It can also be used to calculate how much free 368 * space is currently on the ring (XEN_FLEX_RING_SIZE() - 369 * $NAME_queued()). 370 */ 371 372 #ifndef XEN_PAGE_SHIFT 373 /* The PAGE_SIZE for ring protocols and hypercall interfaces is always 374 * 4K, regardless of the architecture, and page granularity chosen by 375 * operating systems. 376 */ 377 #define XEN_PAGE_SHIFT 12 378 #endif 379 #define XEN_FLEX_RING_SIZE(order) \ 380 (1UL << ((order) + XEN_PAGE_SHIFT - 1)) 381 382 #define DEFINE_XEN_FLEX_RING(name) \ 383 static inline RING_IDX name##_mask(RING_IDX idx, RING_IDX ring_size) \ 384 { \ 385 return idx & (ring_size - 1); \ 386 } \ 387 \ 388 static inline unsigned char *name##_get_ring_ptr(unsigned char *buf, \ 389 RING_IDX idx, \ 390 RING_IDX ring_size) \ 391 { \ 392 return buf + name##_mask(idx, ring_size); \ 393 } \ 394 \ 395 static inline void name##_read_packet(void *opaque, \ 396 const unsigned char *buf, \ 397 size_t size, \ 398 RING_IDX masked_prod, \ 399 RING_IDX *masked_cons, \ 400 RING_IDX ring_size) \ 401 { \ 402 if (*masked_cons < masked_prod || \ 403 size <= ring_size - *masked_cons) { \ 404 memcpy(opaque, buf + *masked_cons, size); \ 405 } else { \ 406 memcpy(opaque, buf + *masked_cons, ring_size - *masked_cons); \ 407 memcpy((unsigned char *)opaque + ring_size - *masked_cons, buf, \ 408 size - (ring_size - *masked_cons)); \ 409 } \ 410 *masked_cons = name##_mask(*masked_cons + size, ring_size); \ 411 } \ 412 \ 413 static inline void name##_write_packet(unsigned char *buf, \ 414 const void *opaque, \ 415 size_t size, \ 416 RING_IDX *masked_prod, \ 417 RING_IDX masked_cons, \ 418 RING_IDX ring_size) \ 419 { \ 420 if (*masked_prod < masked_cons || \ 421 size <= ring_size - *masked_prod) { \ 422 memcpy(buf + *masked_prod, opaque, size); \ 423 } else { \ 424 memcpy(buf + *masked_prod, opaque, ring_size - *masked_prod); \ 425 memcpy(buf, (unsigned char *)opaque + (ring_size - *masked_prod), \ 426 size - (ring_size - *masked_prod)); \ 427 } \ 428 *masked_prod = name##_mask(*masked_prod + size, ring_size); \ 429 } \ 430 \ 431 static inline RING_IDX name##_queued(RING_IDX prod, \ 432 RING_IDX cons, \ 433 RING_IDX ring_size) \ 434 { \ 435 RING_IDX size; \ 436 \ 437 if (prod == cons) \ 438 return 0; \ 439 \ 440 prod = name##_mask(prod, ring_size); \ 441 cons = name##_mask(cons, ring_size); \ 442 \ 443 if (prod == cons) \ 444 return ring_size; \ 445 \ 446 if (prod > cons) \ 447 size = prod - cons; \ 448 else \ 449 size = ring_size - (cons - prod); \ 450 return size; \ 451 } \ 452 \ 453 struct name##_data { \ 454 unsigned char *in; /* half of the allocation */ \ 455 unsigned char *out; /* half of the allocation */ \ 456 } 457 458 #define DEFINE_XEN_FLEX_RING_AND_INTF(name) \ 459 struct name##_data_intf { \ 460 RING_IDX in_cons, in_prod; \ 461 \ 462 uint8_t pad1[56]; \ 463 \ 464 RING_IDX out_cons, out_prod; \ 465 \ 466 uint8_t pad2[56]; \ 467 \ 468 RING_IDX ring_order; \ 469 grant_ref_t ref[]; \ 470 }; \ 471 DEFINE_XEN_FLEX_RING(name) 472 473 #endif /* __XEN_PUBLIC_IO_RING_H__ */ 474 475 /* 476 * Local variables: 477 * mode: C 478 * c-file-style: "BSD" 479 * c-basic-offset: 4 480 * tab-width: 4 481 * indent-tabs-mode: nil 482 * End: 483 */ 484