3 * Copyright (c) 2009, Microsoft Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
16 * Place - Suite 330, Boston, MA 02111-1307 USA.
19 * Haiyang Zhang <haiyangz@microsoft.com>
20 * Hank Janssen <hjanssen@microsoft.com>
21 * K. Y. Srinivasan <kys@microsoft.com>
24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
26 #include <linux/kernel.h>
28 #include <linux/hyperv.h>
30 #include "hyperv_vmbus.h"
32 void hv_begin_read(struct hv_ring_buffer_info
*rbi
)
34 rbi
->ring_buffer
->interrupt_mask
= 1;
38 u32
hv_end_read(struct hv_ring_buffer_info
*rbi
)
43 rbi
->ring_buffer
->interrupt_mask
= 0;
47 * Now check to see if the ring buffer is still empty.
48 * If it is not, we raced and we need to process new
51 hv_get_ringbuffer_availbytes(rbi
, &read
, &write
);
57 * When we write to the ring buffer, check if the host needs to
58 * be signaled. Here is the details of this protocol:
60 * 1. The host guarantees that while it is draining the
61 * ring buffer, it will set the interrupt_mask to
62 * indicate it does not need to be interrupted when
65 * 2. The host guarantees that it will completely drain
66 * the ring buffer before exiting the read loop. Further,
67 * once the ring buffer is empty, it will clear the
68 * interrupt_mask and re-check to see if new data has
72 static bool hv_need_to_signal(u32 old_write
, struct hv_ring_buffer_info
*rbi
)
75 if (rbi
->ring_buffer
->interrupt_mask
)
78 /* check interrupt_mask before read_index */
81 * This is the only case we need to signal when the
82 * ring transitions from being empty to non-empty.
84 if (old_write
== rbi
->ring_buffer
->read_index
)
91 * To optimize the flow management on the send-side,
92 * when the sender is blocked because of lack of
93 * sufficient space in the ring buffer, potential the
94 * consumer of the ring buffer can signal the producer.
95 * This is controlled by the following parameters:
97 * 1. pending_send_sz: This is the size in bytes that the
98 * producer is trying to send.
99 * 2. The feature bit feat_pending_send_sz set to indicate if
100 * the consumer of the ring will signal when the ring
101 * state transitions from being full to a state where
102 * there is room for the producer to send the pending packet.
105 static bool hv_need_to_signal_on_read(u32 old_rd
,
106 struct hv_ring_buffer_info
*rbi
)
111 u32 write_loc
= rbi
->ring_buffer
->write_index
;
112 u32 read_loc
= rbi
->ring_buffer
->read_index
;
113 u32 pending_sz
= rbi
->ring_buffer
->pending_send_sz
;
116 * If the other end is not blocked on write don't bother.
121 r_size
= rbi
->ring_datasize
;
122 cur_write_sz
= write_loc
>= read_loc
? r_size
- (write_loc
- read_loc
) :
123 read_loc
- write_loc
;
125 prev_write_sz
= write_loc
>= old_rd
? r_size
- (write_loc
- old_rd
) :
129 if ((prev_write_sz
< pending_sz
) && (cur_write_sz
>= pending_sz
))
136 * hv_get_next_write_location()
138 * Get the next write location for the specified ring buffer
142 hv_get_next_write_location(struct hv_ring_buffer_info
*ring_info
)
144 u32 next
= ring_info
->ring_buffer
->write_index
;
150 * hv_set_next_write_location()
152 * Set the next write location for the specified ring buffer
156 hv_set_next_write_location(struct hv_ring_buffer_info
*ring_info
,
157 u32 next_write_location
)
159 ring_info
->ring_buffer
->write_index
= next_write_location
;
163 * hv_get_next_read_location()
165 * Get the next read location for the specified ring buffer
168 hv_get_next_read_location(struct hv_ring_buffer_info
*ring_info
)
170 u32 next
= ring_info
->ring_buffer
->read_index
;
176 * hv_get_next_readlocation_withoffset()
178 * Get the next read location + offset for the specified ring buffer.
179 * This allows the caller to skip
182 hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info
*ring_info
,
185 u32 next
= ring_info
->ring_buffer
->read_index
;
188 next
%= ring_info
->ring_datasize
;
195 * hv_set_next_read_location()
197 * Set the next read location for the specified ring buffer
201 hv_set_next_read_location(struct hv_ring_buffer_info
*ring_info
,
202 u32 next_read_location
)
204 ring_info
->ring_buffer
->read_index
= next_read_location
;
210 * hv_get_ring_buffer()
212 * Get the start of the ring buffer
215 hv_get_ring_buffer(struct hv_ring_buffer_info
*ring_info
)
217 return (void *)ring_info
->ring_buffer
->buffer
;
223 * hv_get_ring_buffersize()
225 * Get the size of the ring buffer
228 hv_get_ring_buffersize(struct hv_ring_buffer_info
*ring_info
)
230 return ring_info
->ring_datasize
;
235 * hv_get_ring_bufferindices()
237 * Get the read and write indices as u64 of the specified ring buffer
241 hv_get_ring_bufferindices(struct hv_ring_buffer_info
*ring_info
)
243 return (u64
)ring_info
->ring_buffer
->write_index
<< 32;
248 * hv_copyfrom_ringbuffer()
250 * Helper routine to copy to source from ring buffer.
251 * Assume there is enough room. Handles wrap-around in src case only!!
254 static u32
hv_copyfrom_ringbuffer(
255 struct hv_ring_buffer_info
*ring_info
,
258 u32 start_read_offset
)
260 void *ring_buffer
= hv_get_ring_buffer(ring_info
);
261 u32 ring_buffer_size
= hv_get_ring_buffersize(ring_info
);
265 /* wrap-around detected at the src */
266 if (destlen
> ring_buffer_size
- start_read_offset
) {
267 frag_len
= ring_buffer_size
- start_read_offset
;
269 memcpy(dest
, ring_buffer
+ start_read_offset
, frag_len
);
270 memcpy(dest
+ frag_len
, ring_buffer
, destlen
- frag_len
);
273 memcpy(dest
, ring_buffer
+ start_read_offset
, destlen
);
276 start_read_offset
+= destlen
;
277 start_read_offset
%= ring_buffer_size
;
279 return start_read_offset
;
285 * hv_copyto_ringbuffer()
287 * Helper routine to copy from source to ring buffer.
288 * Assume there is enough room. Handles wrap-around in dest case only!!
291 static u32
hv_copyto_ringbuffer(
292 struct hv_ring_buffer_info
*ring_info
,
293 u32 start_write_offset
,
297 void *ring_buffer
= hv_get_ring_buffer(ring_info
);
298 u32 ring_buffer_size
= hv_get_ring_buffersize(ring_info
);
301 /* wrap-around detected! */
302 if (srclen
> ring_buffer_size
- start_write_offset
) {
303 frag_len
= ring_buffer_size
- start_write_offset
;
304 memcpy(ring_buffer
+ start_write_offset
, src
, frag_len
);
305 memcpy(ring_buffer
, src
+ frag_len
, srclen
- frag_len
);
307 memcpy(ring_buffer
+ start_write_offset
, src
, srclen
);
309 start_write_offset
+= srclen
;
310 start_write_offset
%= ring_buffer_size
;
312 return start_write_offset
;
317 * hv_ringbuffer_get_debuginfo()
319 * Get various debug metrics for the specified ring buffer
322 void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info
*ring_info
,
323 struct hv_ring_buffer_debug_info
*debug_info
)
325 u32 bytes_avail_towrite
;
326 u32 bytes_avail_toread
;
328 if (ring_info
->ring_buffer
) {
329 hv_get_ringbuffer_availbytes(ring_info
,
331 &bytes_avail_towrite
);
333 debug_info
->bytes_avail_toread
= bytes_avail_toread
;
334 debug_info
->bytes_avail_towrite
= bytes_avail_towrite
;
335 debug_info
->current_read_index
=
336 ring_info
->ring_buffer
->read_index
;
337 debug_info
->current_write_index
=
338 ring_info
->ring_buffer
->write_index
;
339 debug_info
->current_interrupt_mask
=
340 ring_info
->ring_buffer
->interrupt_mask
;
346 * hv_ringbuffer_init()
348 *Initialize the ring buffer
351 int hv_ringbuffer_init(struct hv_ring_buffer_info
*ring_info
,
352 void *buffer
, u32 buflen
)
354 if (sizeof(struct hv_ring_buffer
) != PAGE_SIZE
)
357 memset(ring_info
, 0, sizeof(struct hv_ring_buffer_info
));
359 ring_info
->ring_buffer
= (struct hv_ring_buffer
*)buffer
;
360 ring_info
->ring_buffer
->read_index
=
361 ring_info
->ring_buffer
->write_index
= 0;
363 ring_info
->ring_size
= buflen
;
364 ring_info
->ring_datasize
= buflen
- sizeof(struct hv_ring_buffer
);
366 spin_lock_init(&ring_info
->ring_lock
);
373 * hv_ringbuffer_cleanup()
375 * Cleanup the ring buffer
378 void hv_ringbuffer_cleanup(struct hv_ring_buffer_info
*ring_info
)
384 * hv_ringbuffer_write()
386 * Write to the ring buffer
389 int hv_ringbuffer_write(struct hv_ring_buffer_info
*outring_info
,
390 struct scatterlist
*sglist
, u32 sgcount
, bool *signal
)
393 u32 bytes_avail_towrite
;
394 u32 bytes_avail_toread
;
395 u32 totalbytes_towrite
= 0;
397 struct scatterlist
*sg
;
398 u32 next_write_location
;
400 u64 prev_indices
= 0;
403 for_each_sg(sglist
, sg
, sgcount
, i
)
405 totalbytes_towrite
+= sg
->length
;
408 totalbytes_towrite
+= sizeof(u64
);
410 spin_lock_irqsave(&outring_info
->ring_lock
, flags
);
412 hv_get_ringbuffer_availbytes(outring_info
,
414 &bytes_avail_towrite
);
417 /* If there is only room for the packet, assume it is full. */
418 /* Otherwise, the next time around, we think the ring buffer */
419 /* is empty since the read index == write index */
420 if (bytes_avail_towrite
<= totalbytes_towrite
) {
421 spin_unlock_irqrestore(&outring_info
->ring_lock
, flags
);
425 /* Write to the ring buffer */
426 next_write_location
= hv_get_next_write_location(outring_info
);
428 old_write
= next_write_location
;
430 for_each_sg(sglist
, sg
, sgcount
, i
)
432 next_write_location
= hv_copyto_ringbuffer(outring_info
,
438 /* Set previous packet start */
439 prev_indices
= hv_get_ring_bufferindices(outring_info
);
441 next_write_location
= hv_copyto_ringbuffer(outring_info
,
446 /* Issue a full memory barrier before updating the write index */
449 /* Now, update the write location */
450 hv_set_next_write_location(outring_info
, next_write_location
);
453 spin_unlock_irqrestore(&outring_info
->ring_lock
, flags
);
455 *signal
= hv_need_to_signal(old_write
, outring_info
);
462 * hv_ringbuffer_peek()
464 * Read without advancing the read index
467 int hv_ringbuffer_peek(struct hv_ring_buffer_info
*Inring_info
,
468 void *Buffer
, u32 buflen
)
470 u32 bytes_avail_towrite
;
471 u32 bytes_avail_toread
;
472 u32 next_read_location
= 0;
475 spin_lock_irqsave(&Inring_info
->ring_lock
, flags
);
477 hv_get_ringbuffer_availbytes(Inring_info
,
479 &bytes_avail_towrite
);
481 /* Make sure there is something to read */
482 if (bytes_avail_toread
< buflen
) {
484 spin_unlock_irqrestore(&Inring_info
->ring_lock
, flags
);
489 /* Convert to byte offset */
490 next_read_location
= hv_get_next_read_location(Inring_info
);
492 next_read_location
= hv_copyfrom_ringbuffer(Inring_info
,
497 spin_unlock_irqrestore(&Inring_info
->ring_lock
, flags
);
505 * hv_ringbuffer_read()
507 * Read and advance the read index
510 int hv_ringbuffer_read(struct hv_ring_buffer_info
*inring_info
, void *buffer
,
511 u32 buflen
, u32 offset
, bool *signal
)
513 u32 bytes_avail_towrite
;
514 u32 bytes_avail_toread
;
515 u32 next_read_location
= 0;
516 u64 prev_indices
= 0;
523 spin_lock_irqsave(&inring_info
->ring_lock
, flags
);
525 hv_get_ringbuffer_availbytes(inring_info
,
527 &bytes_avail_towrite
);
529 old_read
= bytes_avail_toread
;
531 /* Make sure there is something to read */
532 if (bytes_avail_toread
< buflen
) {
533 spin_unlock_irqrestore(&inring_info
->ring_lock
, flags
);
539 hv_get_next_readlocation_withoffset(inring_info
, offset
);
541 next_read_location
= hv_copyfrom_ringbuffer(inring_info
,
546 next_read_location
= hv_copyfrom_ringbuffer(inring_info
,
551 /* Make sure all reads are done before we update the read index since */
552 /* the writer may start writing to the read area once the read index */
556 /* Update the read index */
557 hv_set_next_read_location(inring_info
, next_read_location
);
559 spin_unlock_irqrestore(&inring_info
->ring_lock
, flags
);
561 *signal
= hv_need_to_signal_on_read(old_read
, inring_info
);