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 
47 typedef 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 */                                                 \
106 union __name##_sring_entry {                                            \
107     __req_t req;                                                        \
108     __rsp_t rsp;                                                        \
109 };                                                                      \
110                                                                         \
111 /* Shared ring page */                                                  \
112 struct __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 */                                   \
120 struct __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 */                                    \
128 struct __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)                                            \
365 static inline RING_IDX name##_mask(RING_IDX idx, RING_IDX ring_size)          \
366 {                                                                             \
367     return idx & (ring_size - 1);                                             \
368 }                                                                             \
369                                                                               \
370 static 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                                                                               \
377 static 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                                                                               \
395 static 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                                                                               \
413 static 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                                                                               \
435 struct 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)                                   \
441 struct 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 };                                                                            \
453 DEFINE_XEN_FLEX_RING(name)
454 
455 #endif /* __XEN_PUBLIC_IO_RING_H__ */
456