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