1/*
2 *
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
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.
8 *
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
12 * more details.
13 *
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.
17 *
18 * Authors:
19 *   Haiyang Zhang <haiyangz@microsoft.com>
20 *   Hank Janssen  <hjanssen@microsoft.com>
21 *   K. Y. Srinivasan <kys@microsoft.com>
22 *
23 */
24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25
26#include <linux/kernel.h>
27#include <linux/mm.h>
28#include <linux/hyperv.h>
29#include <linux/uio.h>
30
31#include "hyperv_vmbus.h"
32
33void hv_begin_read(struct hv_ring_buffer_info *rbi)
34{
35	rbi->ring_buffer->interrupt_mask = 1;
36	mb();
37}
38
39u32 hv_end_read(struct hv_ring_buffer_info *rbi)
40{
41	u32 read;
42	u32 write;
43
44	rbi->ring_buffer->interrupt_mask = 0;
45	mb();
46
47	/*
48	 * Now check to see if the ring buffer is still empty.
49	 * If it is not, we raced and we need to process new
50	 * incoming messages.
51	 */
52	hv_get_ringbuffer_availbytes(rbi, &read, &write);
53
54	return read;
55}
56
57/*
58 * When we write to the ring buffer, check if the host needs to
59 * be signaled. Here is the details of this protocol:
60 *
61 *	1. The host guarantees that while it is draining the
62 *	   ring buffer, it will set the interrupt_mask to
63 *	   indicate it does not need to be interrupted when
64 *	   new data is placed.
65 *
66 *	2. The host guarantees that it will completely drain
67 *	   the ring buffer before exiting the read loop. Further,
68 *	   once the ring buffer is empty, it will clear the
69 *	   interrupt_mask and re-check to see if new data has
70 *	   arrived.
71 */
72
73static bool hv_need_to_signal(u32 old_write, struct hv_ring_buffer_info *rbi)
74{
75	mb();
76	if (rbi->ring_buffer->interrupt_mask)
77		return false;
78
79	/* check interrupt_mask before read_index */
80	rmb();
81	/*
82	 * This is the only case we need to signal when the
83	 * ring transitions from being empty to non-empty.
84	 */
85	if (old_write == rbi->ring_buffer->read_index)
86		return true;
87
88	return false;
89}
90
91/*
92 * To optimize the flow management on the send-side,
93 * when the sender is blocked because of lack of
94 * sufficient space in the ring buffer, potential the
95 * consumer of the ring buffer can signal the producer.
96 * This is controlled by the following parameters:
97 *
98 * 1. pending_send_sz: This is the size in bytes that the
99 *    producer is trying to send.
100 * 2. The feature bit feat_pending_send_sz set to indicate if
101 *    the consumer of the ring will signal when the ring
102 *    state transitions from being full to a state where
103 *    there is room for the producer to send the pending packet.
104 */
105
106static bool hv_need_to_signal_on_read(struct hv_ring_buffer_info *rbi)
107{
108	u32 cur_write_sz;
109	u32 r_size;
110	u32 write_loc;
111	u32 read_loc = rbi->ring_buffer->read_index;
112	u32 pending_sz;
113
114	/*
115	 * Issue a full memory barrier before making the signaling decision.
116	 * Here is the reason for having this barrier:
117	 * If the reading of the pend_sz (in this function)
118	 * were to be reordered and read before we commit the new read
119	 * index (in the calling function)  we could
120	 * have a problem. If the host were to set the pending_sz after we
121	 * have sampled pending_sz and go to sleep before we commit the
122	 * read index, we could miss sending the interrupt. Issue a full
123	 * memory barrier to address this.
124	 */
125	mb();
126
127	pending_sz = rbi->ring_buffer->pending_send_sz;
128	write_loc = rbi->ring_buffer->write_index;
129	/* If the other end is not blocked on write don't bother. */
130	if (pending_sz == 0)
131		return false;
132
133	r_size = rbi->ring_datasize;
134	cur_write_sz = write_loc >= read_loc ? r_size - (write_loc - read_loc) :
135			read_loc - write_loc;
136
137	if (cur_write_sz >= pending_sz)
138		return true;
139
140	return false;
141}
142
143/* Get the next write location for the specified ring buffer. */
144static inline u32
145hv_get_next_write_location(struct hv_ring_buffer_info *ring_info)
146{
147	u32 next = ring_info->ring_buffer->write_index;
148
149	return next;
150}
151
152/* Set the next write location for the specified ring buffer. */
153static inline void
154hv_set_next_write_location(struct hv_ring_buffer_info *ring_info,
155		     u32 next_write_location)
156{
157	ring_info->ring_buffer->write_index = next_write_location;
158}
159
160/* Get the next read location for the specified ring buffer. */
161static inline u32
162hv_get_next_read_location(struct hv_ring_buffer_info *ring_info)
163{
164	u32 next = ring_info->ring_buffer->read_index;
165
166	return next;
167}
168
169/*
170 * Get the next read location + offset for the specified ring buffer.
171 * This allows the caller to skip.
172 */
173static inline u32
174hv_get_next_readlocation_withoffset(struct hv_ring_buffer_info *ring_info,
175				 u32 offset)
176{
177	u32 next = ring_info->ring_buffer->read_index;
178
179	next += offset;
180	next %= ring_info->ring_datasize;
181
182	return next;
183}
184
185/* Set the next read location for the specified ring buffer. */
186static inline void
187hv_set_next_read_location(struct hv_ring_buffer_info *ring_info,
188		    u32 next_read_location)
189{
190	ring_info->ring_buffer->read_index = next_read_location;
191}
192
193
194/* Get the start of the ring buffer. */
195static inline void *
196hv_get_ring_buffer(struct hv_ring_buffer_info *ring_info)
197{
198	return (void *)ring_info->ring_buffer->buffer;
199}
200
201
202/* Get the size of the ring buffer. */
203static inline u32
204hv_get_ring_buffersize(struct hv_ring_buffer_info *ring_info)
205{
206	return ring_info->ring_datasize;
207}
208
209/* Get the read and write indices as u64 of the specified ring buffer. */
210static inline u64
211hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info)
212{
213	return (u64)ring_info->ring_buffer->write_index << 32;
214}
215
216/*
217 * Helper routine to copy to source from ring buffer.
218 * Assume there is enough room. Handles wrap-around in src case only!!
219 */
220static u32 hv_copyfrom_ringbuffer(
221	struct hv_ring_buffer_info	*ring_info,
222	void				*dest,
223	u32				destlen,
224	u32				start_read_offset)
225{
226	void *ring_buffer = hv_get_ring_buffer(ring_info);
227	u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
228
229	u32 frag_len;
230
231	/* wrap-around detected at the src */
232	if (destlen > ring_buffer_size - start_read_offset) {
233		frag_len = ring_buffer_size - start_read_offset;
234
235		memcpy(dest, ring_buffer + start_read_offset, frag_len);
236		memcpy(dest + frag_len, ring_buffer, destlen - frag_len);
237	} else
238
239		memcpy(dest, ring_buffer + start_read_offset, destlen);
240
241
242	start_read_offset += destlen;
243	start_read_offset %= ring_buffer_size;
244
245	return start_read_offset;
246}
247
248
249/*
250 * Helper routine to copy from source to ring buffer.
251 * Assume there is enough room. Handles wrap-around in dest case only!!
252 */
253static u32 hv_copyto_ringbuffer(
254	struct hv_ring_buffer_info	*ring_info,
255	u32				start_write_offset,
256	void				*src,
257	u32				srclen)
258{
259	void *ring_buffer = hv_get_ring_buffer(ring_info);
260	u32 ring_buffer_size = hv_get_ring_buffersize(ring_info);
261	u32 frag_len;
262
263	/* wrap-around detected! */
264	if (srclen > ring_buffer_size - start_write_offset) {
265		frag_len = ring_buffer_size - start_write_offset;
266		memcpy(ring_buffer + start_write_offset, src, frag_len);
267		memcpy(ring_buffer, src + frag_len, srclen - frag_len);
268	} else
269		memcpy(ring_buffer + start_write_offset, src, srclen);
270
271	start_write_offset += srclen;
272	start_write_offset %= ring_buffer_size;
273
274	return start_write_offset;
275}
276
277/* Get various debug metrics for the specified ring buffer. */
278void hv_ringbuffer_get_debuginfo(struct hv_ring_buffer_info *ring_info,
279			    struct hv_ring_buffer_debug_info *debug_info)
280{
281	u32 bytes_avail_towrite;
282	u32 bytes_avail_toread;
283
284	if (ring_info->ring_buffer) {
285		hv_get_ringbuffer_availbytes(ring_info,
286					&bytes_avail_toread,
287					&bytes_avail_towrite);
288
289		debug_info->bytes_avail_toread = bytes_avail_toread;
290		debug_info->bytes_avail_towrite = bytes_avail_towrite;
291		debug_info->current_read_index =
292			ring_info->ring_buffer->read_index;
293		debug_info->current_write_index =
294			ring_info->ring_buffer->write_index;
295		debug_info->current_interrupt_mask =
296			ring_info->ring_buffer->interrupt_mask;
297	}
298}
299
300/* Initialize the ring buffer. */
301int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info,
302		   void *buffer, u32 buflen)
303{
304	if (sizeof(struct hv_ring_buffer) != PAGE_SIZE)
305		return -EINVAL;
306
307	memset(ring_info, 0, sizeof(struct hv_ring_buffer_info));
308
309	ring_info->ring_buffer = (struct hv_ring_buffer *)buffer;
310	ring_info->ring_buffer->read_index =
311		ring_info->ring_buffer->write_index = 0;
312
313	/* Set the feature bit for enabling flow control. */
314	ring_info->ring_buffer->feature_bits.value = 1;
315
316	ring_info->ring_size = buflen;
317	ring_info->ring_datasize = buflen - sizeof(struct hv_ring_buffer);
318
319	spin_lock_init(&ring_info->ring_lock);
320
321	return 0;
322}
323
324/* Cleanup the ring buffer. */
325void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info)
326{
327}
328
329/* Write to the ring buffer. */
330int hv_ringbuffer_write(struct hv_ring_buffer_info *outring_info,
331		    struct kvec *kv_list, u32 kv_count, bool *signal)
332{
333	int i = 0;
334	u32 bytes_avail_towrite;
335	u32 bytes_avail_toread;
336	u32 totalbytes_towrite = 0;
337
338	u32 next_write_location;
339	u32 old_write;
340	u64 prev_indices = 0;
341	unsigned long flags;
342
343	for (i = 0; i < kv_count; i++)
344		totalbytes_towrite += kv_list[i].iov_len;
345
346	totalbytes_towrite += sizeof(u64);
347
348	spin_lock_irqsave(&outring_info->ring_lock, flags);
349
350	hv_get_ringbuffer_availbytes(outring_info,
351				&bytes_avail_toread,
352				&bytes_avail_towrite);
353
354	/*
355	 * If there is only room for the packet, assume it is full.
356	 * Otherwise, the next time around, we think the ring buffer
357	 * is empty since the read index == write index.
358	 */
359	if (bytes_avail_towrite <= totalbytes_towrite) {
360		spin_unlock_irqrestore(&outring_info->ring_lock, flags);
361		return -EAGAIN;
362	}
363
364	/* Write to the ring buffer */
365	next_write_location = hv_get_next_write_location(outring_info);
366
367	old_write = next_write_location;
368
369	for (i = 0; i < kv_count; i++) {
370		next_write_location = hv_copyto_ringbuffer(outring_info,
371						     next_write_location,
372						     kv_list[i].iov_base,
373						     kv_list[i].iov_len);
374	}
375
376	/* Set previous packet start */
377	prev_indices = hv_get_ring_bufferindices(outring_info);
378
379	next_write_location = hv_copyto_ringbuffer(outring_info,
380					     next_write_location,
381					     &prev_indices,
382					     sizeof(u64));
383
384	/* Issue a full memory barrier before updating the write index */
385	mb();
386
387	/* Now, update the write location */
388	hv_set_next_write_location(outring_info, next_write_location);
389
390
391	spin_unlock_irqrestore(&outring_info->ring_lock, flags);
392
393	*signal = hv_need_to_signal(old_write, outring_info);
394	return 0;
395}
396
397
398/* Read without advancing the read index. */
399int hv_ringbuffer_peek(struct hv_ring_buffer_info *Inring_info,
400		   void *Buffer, u32 buflen)
401{
402	u32 bytes_avail_towrite;
403	u32 bytes_avail_toread;
404	u32 next_read_location = 0;
405	unsigned long flags;
406
407	spin_lock_irqsave(&Inring_info->ring_lock, flags);
408
409	hv_get_ringbuffer_availbytes(Inring_info,
410				&bytes_avail_toread,
411				&bytes_avail_towrite);
412
413	/* Make sure there is something to read */
414	if (bytes_avail_toread < buflen) {
415
416		spin_unlock_irqrestore(&Inring_info->ring_lock, flags);
417
418		return -EAGAIN;
419	}
420
421	/* Convert to byte offset */
422	next_read_location = hv_get_next_read_location(Inring_info);
423
424	next_read_location = hv_copyfrom_ringbuffer(Inring_info,
425						Buffer,
426						buflen,
427						next_read_location);
428
429	spin_unlock_irqrestore(&Inring_info->ring_lock, flags);
430
431	return 0;
432}
433
434
435/* Read and advance the read index. */
436int hv_ringbuffer_read(struct hv_ring_buffer_info *inring_info, void *buffer,
437		   u32 buflen, u32 offset, bool *signal)
438{
439	u32 bytes_avail_towrite;
440	u32 bytes_avail_toread;
441	u32 next_read_location = 0;
442	u64 prev_indices = 0;
443	unsigned long flags;
444
445	if (buflen <= 0)
446		return -EINVAL;
447
448	spin_lock_irqsave(&inring_info->ring_lock, flags);
449
450	hv_get_ringbuffer_availbytes(inring_info,
451				&bytes_avail_toread,
452				&bytes_avail_towrite);
453
454	/* Make sure there is something to read */
455	if (bytes_avail_toread < buflen) {
456		spin_unlock_irqrestore(&inring_info->ring_lock, flags);
457
458		return -EAGAIN;
459	}
460
461	next_read_location =
462		hv_get_next_readlocation_withoffset(inring_info, offset);
463
464	next_read_location = hv_copyfrom_ringbuffer(inring_info,
465						buffer,
466						buflen,
467						next_read_location);
468
469	next_read_location = hv_copyfrom_ringbuffer(inring_info,
470						&prev_indices,
471						sizeof(u64),
472						next_read_location);
473
474	/*
475	 * Make sure all reads are done before we update the read index since
476	 * the writer may start writing to the read area once the read index
477	 * is updated.
478	 */
479	mb();
480
481	/* Update the read index */
482	hv_set_next_read_location(inring_info, next_read_location);
483
484	spin_unlock_irqrestore(&inring_info->ring_lock, flags);
485
486	*signal = hv_need_to_signal_on_read(inring_info);
487
488	return 0;
489}
490