]>
Commit | Line | Data |
---|---|---|
3e7ee490 HJ |
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> | |
b2a5a585 | 21 | * K. Y. Srinivasan <kys@microsoft.com> |
3e7ee490 HJ |
22 | * |
23 | */ | |
0a46618d | 24 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
3e7ee490 | 25 | |
a0086dc5 GKH |
26 | #include <linux/kernel.h> |
27 | #include <linux/mm.h> | |
46a97191 | 28 | #include <linux/hyperv.h> |
011a7c3c | 29 | #include <linux/uio.h> |
9988ce68 VK |
30 | #include <linux/vmalloc.h> |
31 | #include <linux/slab.h> | |
8dd45f2a | 32 | #include <linux/prefetch.h> |
3f335ea2 | 33 | |
0f2a6619 | 34 | #include "hyperv_vmbus.h" |
3e7ee490 | 35 | |
f3dd3f47 | 36 | #define VMBUS_PKT_TRAILER 8 |
37 | ||
98fa8cf4 S |
38 | /* |
39 | * When we write to the ring buffer, check if the host needs to | |
40 | * be signaled. Here is the details of this protocol: | |
41 | * | |
42 | * 1. The host guarantees that while it is draining the | |
43 | * ring buffer, it will set the interrupt_mask to | |
44 | * indicate it does not need to be interrupted when | |
45 | * new data is placed. | |
46 | * | |
47 | * 2. The host guarantees that it will completely drain | |
48 | * the ring buffer before exiting the read loop. Further, | |
49 | * once the ring buffer is empty, it will clear the | |
50 | * interrupt_mask and re-check to see if new data has | |
51 | * arrived. | |
1f6ee4e7 S |
52 | * |
53 | * KYS: Oct. 30, 2016: | |
54 | * It looks like Windows hosts have logic to deal with DOS attacks that | |
55 | * can be triggered if it receives interrupts when it is not expecting | |
56 | * the interrupt. The host expects interrupts only when the ring | |
57 | * transitions from empty to non-empty (or full to non full on the guest | |
58 | * to host ring). | |
59 | * So, base the signaling decision solely on the ring state until the | |
60 | * host logic is fixed. | |
98fa8cf4 S |
61 | */ |
62 | ||
b103a56f | 63 | static void hv_signal_on_write(u32 old_write, struct vmbus_channel *channel) |
98fa8cf4 | 64 | { |
1f6ee4e7 S |
65 | struct hv_ring_buffer_info *rbi = &channel->outbound; |
66 | ||
dcd0eeca | 67 | virt_mb(); |
d45faaee | 68 | if (READ_ONCE(rbi->ring_buffer->interrupt_mask)) |
1f6ee4e7 | 69 | return; |
98fa8cf4 | 70 | |
e91e84fa | 71 | /* check interrupt_mask before read_index */ |
dcd0eeca | 72 | virt_rmb(); |
98fa8cf4 S |
73 | /* |
74 | * This is the only case we need to signal when the | |
75 | * ring transitions from being empty to non-empty. | |
76 | */ | |
d45faaee | 77 | if (old_write == READ_ONCE(rbi->ring_buffer->read_index)) |
1f6ee4e7 | 78 | vmbus_setevent(channel); |
98fa8cf4 S |
79 | } |
80 | ||
822f18d4 | 81 | /* Get the next write location for the specified ring buffer. */ |
4d643114 | 82 | static inline u32 |
2b8a912e | 83 | hv_get_next_write_location(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 84 | { |
fc8c72eb | 85 | u32 next = ring_info->ring_buffer->write_index; |
3e7ee490 | 86 | |
3e7ee490 HJ |
87 | return next; |
88 | } | |
89 | ||
822f18d4 | 90 | /* Set the next write location for the specified ring buffer. */ |
3e7ee490 | 91 | static inline void |
2b8a912e | 92 | hv_set_next_write_location(struct hv_ring_buffer_info *ring_info, |
fc8c72eb | 93 | u32 next_write_location) |
3e7ee490 | 94 | { |
fc8c72eb | 95 | ring_info->ring_buffer->write_index = next_write_location; |
3e7ee490 HJ |
96 | } |
97 | ||
822f18d4 | 98 | /* Set the next read location for the specified ring buffer. */ |
3e7ee490 | 99 | static inline void |
2b8a912e | 100 | hv_set_next_read_location(struct hv_ring_buffer_info *ring_info, |
fc8c72eb | 101 | u32 next_read_location) |
3e7ee490 | 102 | { |
fc8c72eb | 103 | ring_info->ring_buffer->read_index = next_read_location; |
ab028db4 | 104 | ring_info->priv_read_index = next_read_location; |
3e7ee490 HJ |
105 | } |
106 | ||
822f18d4 | 107 | /* Get the size of the ring buffer. */ |
4d643114 | 108 | static inline u32 |
e4165a0f | 109 | hv_get_ring_buffersize(const struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 110 | { |
fc8c72eb | 111 | return ring_info->ring_datasize; |
3e7ee490 HJ |
112 | } |
113 | ||
822f18d4 | 114 | /* Get the read and write indices as u64 of the specified ring buffer. */ |
59471438 | 115 | static inline u64 |
2b8a912e | 116 | hv_get_ring_bufferindices(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 117 | { |
fc8c72eb | 118 | return (u64)ring_info->ring_buffer->write_index << 32; |
3e7ee490 HJ |
119 | } |
120 | ||
7581578d | 121 | /* |
7581578d S |
122 | * Helper routine to copy from source to ring buffer. |
123 | * Assume there is enough room. Handles wrap-around in dest case only!! | |
7581578d S |
124 | */ |
125 | static u32 hv_copyto_ringbuffer( | |
fc8c72eb HZ |
126 | struct hv_ring_buffer_info *ring_info, |
127 | u32 start_write_offset, | |
e4165a0f | 128 | const void *src, |
7581578d S |
129 | u32 srclen) |
130 | { | |
131 | void *ring_buffer = hv_get_ring_buffer(ring_info); | |
132 | u32 ring_buffer_size = hv_get_ring_buffersize(ring_info); | |
f24f0b49 VK |
133 | |
134 | memcpy(ring_buffer + start_write_offset, src, srclen); | |
3e7ee490 | 135 | |
7581578d | 136 | start_write_offset += srclen; |
8d12f882 SH |
137 | if (start_write_offset >= ring_buffer_size) |
138 | start_write_offset -= ring_buffer_size; | |
7581578d S |
139 | |
140 | return start_write_offset; | |
141 | } | |
3e7ee490 | 142 | |
822f18d4 | 143 | /* Get various debug metrics for the specified ring buffer. */ |
e4165a0f SH |
144 | void hv_ringbuffer_get_debuginfo(const struct hv_ring_buffer_info *ring_info, |
145 | struct hv_ring_buffer_debug_info *debug_info) | |
3e7ee490 | 146 | { |
fc8c72eb HZ |
147 | u32 bytes_avail_towrite; |
148 | u32 bytes_avail_toread; | |
3e7ee490 | 149 | |
fc8c72eb | 150 | if (ring_info->ring_buffer) { |
2b8a912e | 151 | hv_get_ringbuffer_availbytes(ring_info, |
fc8c72eb HZ |
152 | &bytes_avail_toread, |
153 | &bytes_avail_towrite); | |
3e7ee490 | 154 | |
fc8c72eb HZ |
155 | debug_info->bytes_avail_toread = bytes_avail_toread; |
156 | debug_info->bytes_avail_towrite = bytes_avail_towrite; | |
82f8bd40 | 157 | debug_info->current_read_index = |
fc8c72eb | 158 | ring_info->ring_buffer->read_index; |
82f8bd40 | 159 | debug_info->current_write_index = |
fc8c72eb | 160 | ring_info->ring_buffer->write_index; |
82f8bd40 | 161 | debug_info->current_interrupt_mask = |
fc8c72eb | 162 | ring_info->ring_buffer->interrupt_mask; |
3e7ee490 HJ |
163 | } |
164 | } | |
4827ee1d | 165 | EXPORT_SYMBOL_GPL(hv_ringbuffer_get_debuginfo); |
3e7ee490 | 166 | |
822f18d4 | 167 | /* Initialize the ring buffer. */ |
72a95cbc | 168 | int hv_ringbuffer_init(struct hv_ring_buffer_info *ring_info, |
9988ce68 | 169 | struct page *pages, u32 page_cnt) |
3e7ee490 | 170 | { |
9988ce68 VK |
171 | int i; |
172 | struct page **pages_wraparound; | |
173 | ||
174 | BUILD_BUG_ON((sizeof(struct hv_ring_buffer) != PAGE_SIZE)); | |
3e7ee490 | 175 | |
fc8c72eb | 176 | memset(ring_info, 0, sizeof(struct hv_ring_buffer_info)); |
3e7ee490 | 177 | |
9988ce68 VK |
178 | /* |
179 | * First page holds struct hv_ring_buffer, do wraparound mapping for | |
180 | * the rest. | |
181 | */ | |
182 | pages_wraparound = kzalloc(sizeof(struct page *) * (page_cnt * 2 - 1), | |
183 | GFP_KERNEL); | |
184 | if (!pages_wraparound) | |
185 | return -ENOMEM; | |
186 | ||
187 | pages_wraparound[0] = pages; | |
188 | for (i = 0; i < 2 * (page_cnt - 1); i++) | |
189 | pages_wraparound[i + 1] = &pages[i % (page_cnt - 1) + 1]; | |
190 | ||
191 | ring_info->ring_buffer = (struct hv_ring_buffer *) | |
192 | vmap(pages_wraparound, page_cnt * 2 - 1, VM_MAP, PAGE_KERNEL); | |
193 | ||
194 | kfree(pages_wraparound); | |
195 | ||
196 | ||
197 | if (!ring_info->ring_buffer) | |
198 | return -ENOMEM; | |
199 | ||
fc8c72eb HZ |
200 | ring_info->ring_buffer->read_index = |
201 | ring_info->ring_buffer->write_index = 0; | |
3e7ee490 | 202 | |
822f18d4 | 203 | /* Set the feature bit for enabling flow control. */ |
046c7911 S |
204 | ring_info->ring_buffer->feature_bits.value = 1; |
205 | ||
9988ce68 VK |
206 | ring_info->ring_size = page_cnt << PAGE_SHIFT; |
207 | ring_info->ring_datasize = ring_info->ring_size - | |
208 | sizeof(struct hv_ring_buffer); | |
3e7ee490 | 209 | |
fc8c72eb | 210 | spin_lock_init(&ring_info->ring_lock); |
3e7ee490 HJ |
211 | |
212 | return 0; | |
213 | } | |
214 | ||
822f18d4 | 215 | /* Cleanup the ring buffer. */ |
2dba688b | 216 | void hv_ringbuffer_cleanup(struct hv_ring_buffer_info *ring_info) |
3e7ee490 | 217 | { |
9988ce68 | 218 | vunmap(ring_info->ring_buffer); |
3e7ee490 HJ |
219 | } |
220 | ||
822f18d4 | 221 | /* Write to the ring buffer. */ |
1f6ee4e7 | 222 | int hv_ringbuffer_write(struct vmbus_channel *channel, |
e4165a0f | 223 | const struct kvec *kv_list, u32 kv_count) |
3e7ee490 | 224 | { |
2c616a8b | 225 | int i; |
fc8c72eb | 226 | u32 bytes_avail_towrite; |
2c616a8b | 227 | u32 totalbytes_towrite = sizeof(u64); |
66a60543 | 228 | u32 next_write_location; |
98fa8cf4 | 229 | u32 old_write; |
2c616a8b SH |
230 | u64 prev_indices; |
231 | unsigned long flags; | |
1f6ee4e7 | 232 | struct hv_ring_buffer_info *outring_info = &channel->outbound; |
3e7ee490 | 233 | |
e7e97dd8 S |
234 | if (channel->rescind) |
235 | return -ENODEV; | |
236 | ||
011a7c3c S |
237 | for (i = 0; i < kv_count; i++) |
238 | totalbytes_towrite += kv_list[i].iov_len; | |
3e7ee490 | 239 | |
5529eaf6 | 240 | spin_lock_irqsave(&outring_info->ring_lock, flags); |
3e7ee490 | 241 | |
a6341f00 | 242 | bytes_avail_towrite = hv_get_bytes_to_write(outring_info); |
3e7ee490 | 243 | |
822f18d4 VK |
244 | /* |
245 | * If there is only room for the packet, assume it is full. | |
246 | * Otherwise, the next time around, we think the ring buffer | |
247 | * is empty since the read index == write index. | |
248 | */ | |
fc8c72eb | 249 | if (bytes_avail_towrite <= totalbytes_towrite) { |
5529eaf6 | 250 | spin_unlock_irqrestore(&outring_info->ring_lock, flags); |
d2598f01 | 251 | return -EAGAIN; |
3e7ee490 HJ |
252 | } |
253 | ||
454f18a9 | 254 | /* Write to the ring buffer */ |
2b8a912e | 255 | next_write_location = hv_get_next_write_location(outring_info); |
3e7ee490 | 256 | |
98fa8cf4 S |
257 | old_write = next_write_location; |
258 | ||
011a7c3c | 259 | for (i = 0; i < kv_count; i++) { |
2b8a912e | 260 | next_write_location = hv_copyto_ringbuffer(outring_info, |
fc8c72eb | 261 | next_write_location, |
011a7c3c S |
262 | kv_list[i].iov_base, |
263 | kv_list[i].iov_len); | |
3e7ee490 HJ |
264 | } |
265 | ||
454f18a9 | 266 | /* Set previous packet start */ |
2b8a912e | 267 | prev_indices = hv_get_ring_bufferindices(outring_info); |
3e7ee490 | 268 | |
2b8a912e | 269 | next_write_location = hv_copyto_ringbuffer(outring_info, |
fc8c72eb HZ |
270 | next_write_location, |
271 | &prev_indices, | |
b219b3f7 | 272 | sizeof(u64)); |
3e7ee490 | 273 | |
98fa8cf4 | 274 | /* Issue a full memory barrier before updating the write index */ |
dcd0eeca | 275 | virt_mb(); |
3e7ee490 | 276 | |
454f18a9 | 277 | /* Now, update the write location */ |
2b8a912e | 278 | hv_set_next_write_location(outring_info, next_write_location); |
3e7ee490 | 279 | |
3e7ee490 | 280 | |
5529eaf6 | 281 | spin_unlock_irqrestore(&outring_info->ring_lock, flags); |
98fa8cf4 | 282 | |
b103a56f | 283 | hv_signal_on_write(old_write, channel); |
e7e97dd8 S |
284 | |
285 | if (channel->rescind) | |
286 | return -ENODEV; | |
287 | ||
3e7ee490 HJ |
288 | return 0; |
289 | } | |
290 | ||
3372592a | 291 | int hv_ringbuffer_read(struct vmbus_channel *channel, |
940b68e2 | 292 | void *buffer, u32 buflen, u32 *buffer_actual_len, |
3372592a | 293 | u64 *requestid, bool raw) |
3e7ee490 | 294 | { |
4226ff69 SH |
295 | struct vmpacket_descriptor *desc; |
296 | u32 packetlen, offset; | |
297 | ||
298 | if (unlikely(buflen == 0)) | |
a16e1485 | 299 | return -EINVAL; |
3e7ee490 | 300 | |
940b68e2 VK |
301 | *buffer_actual_len = 0; |
302 | *requestid = 0; | |
303 | ||
454f18a9 | 304 | /* Make sure there is something to read */ |
4226ff69 SH |
305 | desc = hv_pkt_iter_first(channel); |
306 | if (desc == NULL) { | |
940b68e2 VK |
307 | /* |
308 | * No error is set when there is even no header, drivers are | |
309 | * supposed to analyze buffer_actual_len. | |
310 | */ | |
42dd2715 | 311 | return 0; |
940b68e2 | 312 | } |
3e7ee490 | 313 | |
4226ff69 SH |
314 | offset = raw ? 0 : (desc->offset8 << 3); |
315 | packetlen = (desc->len8 << 3) - offset; | |
940b68e2 | 316 | *buffer_actual_len = packetlen; |
4226ff69 | 317 | *requestid = desc->trans_id; |
940b68e2 | 318 | |
4226ff69 | 319 | if (unlikely(packetlen > buflen)) |
3eba9a77 | 320 | return -ENOBUFS; |
3e7ee490 | 321 | |
4226ff69 SH |
322 | /* since ring is double mapped, only one copy is necessary */ |
323 | memcpy(buffer, (const char *)desc + offset, packetlen); | |
3e7ee490 | 324 | |
4226ff69 SH |
325 | /* Advance ring index to next packet descriptor */ |
326 | __hv_pkt_iter_next(channel, desc); | |
3e7ee490 | 327 | |
4226ff69 SH |
328 | /* Notify host of update */ |
329 | hv_pkt_iter_close(channel); | |
c2b8e520 | 330 | |
42dd2715 | 331 | return 0; |
b5f53dde | 332 | } |
f3dd3f47 | 333 | |
334 | /* | |
335 | * Determine number of bytes available in ring buffer after | |
336 | * the current iterator (priv_read_index) location. | |
337 | * | |
338 | * This is similar to hv_get_bytes_to_read but with private | |
339 | * read index instead. | |
340 | */ | |
341 | static u32 hv_pkt_iter_avail(const struct hv_ring_buffer_info *rbi) | |
342 | { | |
343 | u32 priv_read_loc = rbi->priv_read_index; | |
344 | u32 write_loc = READ_ONCE(rbi->ring_buffer->write_index); | |
345 | ||
346 | if (write_loc >= priv_read_loc) | |
347 | return write_loc - priv_read_loc; | |
348 | else | |
349 | return (rbi->ring_datasize - priv_read_loc) + write_loc; | |
350 | } | |
351 | ||
352 | /* | |
353 | * Get first vmbus packet from ring buffer after read_index | |
354 | * | |
355 | * If ring buffer is empty, returns NULL and no other action needed. | |
356 | */ | |
357 | struct vmpacket_descriptor *hv_pkt_iter_first(struct vmbus_channel *channel) | |
358 | { | |
359 | struct hv_ring_buffer_info *rbi = &channel->inbound; | |
15e1674d | 360 | struct vmpacket_descriptor *desc; |
f3dd3f47 | 361 | |
f3dd3f47 | 362 | if (hv_pkt_iter_avail(rbi) < sizeof(struct vmpacket_descriptor)) |
363 | return NULL; | |
364 | ||
15e1674d SH |
365 | desc = hv_get_ring_buffer(rbi) + rbi->priv_read_index; |
366 | if (desc) | |
367 | prefetch((char *)desc + (desc->len8 << 3)); | |
368 | ||
369 | return desc; | |
f3dd3f47 | 370 | } |
371 | EXPORT_SYMBOL_GPL(hv_pkt_iter_first); | |
372 | ||
373 | /* | |
374 | * Get next vmbus packet from ring buffer. | |
375 | * | |
376 | * Advances the current location (priv_read_index) and checks for more | |
377 | * data. If the end of the ring buffer is reached, then return NULL. | |
378 | */ | |
379 | struct vmpacket_descriptor * | |
380 | __hv_pkt_iter_next(struct vmbus_channel *channel, | |
381 | const struct vmpacket_descriptor *desc) | |
382 | { | |
383 | struct hv_ring_buffer_info *rbi = &channel->inbound; | |
384 | u32 packetlen = desc->len8 << 3; | |
385 | u32 dsize = rbi->ring_datasize; | |
386 | ||
387 | /* bump offset to next potential packet */ | |
388 | rbi->priv_read_index += packetlen + VMBUS_PKT_TRAILER; | |
389 | if (rbi->priv_read_index >= dsize) | |
390 | rbi->priv_read_index -= dsize; | |
391 | ||
392 | /* more data? */ | |
05d00bc9 | 393 | return hv_pkt_iter_first(channel); |
f3dd3f47 | 394 | } |
395 | EXPORT_SYMBOL_GPL(__hv_pkt_iter_next); | |
396 | ||
397 | /* | |
398 | * Update host ring buffer after iterating over packets. | |
399 | */ | |
400 | void hv_pkt_iter_close(struct vmbus_channel *channel) | |
401 | { | |
402 | struct hv_ring_buffer_info *rbi = &channel->inbound; | |
05d00bc9 | 403 | u32 orig_write_sz = hv_get_bytes_to_write(rbi); |
f3dd3f47 | 404 | |
405 | /* | |
406 | * Make sure all reads are done before we update the read index since | |
407 | * the writer may start writing to the read area once the read index | |
408 | * is updated. | |
409 | */ | |
410 | virt_rmb(); | |
411 | rbi->ring_buffer->read_index = rbi->priv_read_index; | |
412 | ||
8dd45f2a SH |
413 | /* |
414 | * Issue a full memory barrier before making the signaling decision. | |
415 | * Here is the reason for having this barrier: | |
416 | * If the reading of the pend_sz (in this function) | |
417 | * were to be reordered and read before we commit the new read | |
418 | * index (in the calling function) we could | |
419 | * have a problem. If the host were to set the pending_sz after we | |
420 | * have sampled pending_sz and go to sleep before we commit the | |
421 | * read index, we could miss sending the interrupt. Issue a full | |
422 | * memory barrier to address this. | |
423 | */ | |
424 | virt_mb(); | |
425 | ||
03bad714 SH |
426 | /* If host has disabled notifications then skip */ |
427 | if (rbi->ring_buffer->interrupt_mask) | |
8dd45f2a SH |
428 | return; |
429 | ||
03bad714 SH |
430 | if (rbi->ring_buffer->feature_bits.feat_pending_send_sz) { |
431 | u32 pending_sz = READ_ONCE(rbi->ring_buffer->pending_send_sz); | |
8dd45f2a | 432 | |
03bad714 SH |
433 | /* |
434 | * If there was space before we began iteration, | |
435 | * then host was not blocked. Also handles case where | |
436 | * pending_sz is zero then host has nothing pending | |
437 | * and does not need to be signaled. | |
438 | */ | |
439 | if (orig_write_sz > pending_sz) | |
440 | return; | |
441 | ||
442 | /* If pending write will not fit, don't give false hope. */ | |
443 | if (hv_get_bytes_to_write(rbi) < pending_sz) | |
444 | return; | |
445 | } | |
446 | ||
447 | vmbus_setevent(channel); | |
f3dd3f47 | 448 | } |
449 | EXPORT_SYMBOL_GPL(hv_pkt_iter_close); |