]>
git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/drivers/bus/vmbus/vmbus_channel.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright (c) 2018, Microsoft Corporation.
12 #include <rte_tailq.h>
14 #include <rte_malloc.h>
16 #include <rte_atomic.h>
17 #include <rte_memory.h>
18 #include <rte_bus_vmbus.h>
23 vmbus_sync_set_bit(volatile uint32_t *addr
, uint32_t mask
)
25 /* Use GCC builtin which atomic does atomic OR operation */
26 __sync_or_and_fetch(addr
, mask
);
30 vmbus_send_interrupt(const struct rte_vmbus_device
*dev
, uint32_t relid
)
35 int_addr
= dev
->int_page
+ relid
/ 32;
36 int_mask
= 1u << (relid
% 32);
38 vmbus_sync_set_bit(int_addr
, int_mask
);
42 vmbus_set_monitor(const struct rte_vmbus_device
*dev
, uint32_t monitor_id
)
44 uint32_t *monitor_addr
, monitor_mask
;
45 unsigned int trigger_index
;
47 trigger_index
= monitor_id
/ HV_MON_TRIG_LEN
;
48 monitor_mask
= 1u << (monitor_id
% HV_MON_TRIG_LEN
);
50 monitor_addr
= &dev
->monitor_page
->trigs
[trigger_index
].pending
;
51 vmbus_sync_set_bit(monitor_addr
, monitor_mask
);
55 vmbus_set_event(const struct rte_vmbus_device
*dev
,
56 const struct vmbus_channel
*chan
)
58 vmbus_send_interrupt(dev
, chan
->relid
);
59 vmbus_set_monitor(dev
, chan
->monitor_id
);
63 * Notify host that there are data pending on our TX bufring.
65 * Since this in userspace, rely on the monitor page.
66 * Can't do a hypercall from userspace.
69 rte_vmbus_chan_signal_tx(const struct vmbus_channel
*chan
)
71 const struct rte_vmbus_device
*dev
= chan
->device
;
72 const struct vmbus_br
*tbr
= &chan
->txbr
;
74 /* Make sure all updates are done before signaling host */
77 /* If host is ignoring interrupts? */
81 vmbus_set_event(dev
, chan
);
85 /* Do a simple send directly using transmit ring. */
86 int rte_vmbus_chan_send(struct vmbus_channel
*chan
, uint16_t type
,
87 void *data
, uint32_t dlen
,
88 uint64_t xactid
, uint32_t flags
, bool *need_sig
)
90 struct vmbus_chanpkt pkt
;
91 unsigned int pktlen
, pad_pktlen
;
92 const uint32_t hlen
= sizeof(pkt
);
93 bool send_evt
= false;
99 pad_pktlen
= RTE_ALIGN(pktlen
, sizeof(uint64_t));
102 pkt
.hdr
.flags
= flags
;
103 pkt
.hdr
.hlen
= hlen
>> VMBUS_CHANPKT_SIZE_SHIFT
;
104 pkt
.hdr
.tlen
= pad_pktlen
>> VMBUS_CHANPKT_SIZE_SHIFT
;
105 pkt
.hdr
.xactid
= xactid
;
107 iov
[0].iov_base
= &pkt
;
108 iov
[0].iov_len
= hlen
;
109 iov
[1].iov_base
= data
;
110 iov
[1].iov_len
= dlen
;
111 iov
[2].iov_base
= &pad
;
112 iov
[2].iov_len
= pad_pktlen
- pktlen
;
114 error
= vmbus_txbr_write(&chan
->txbr
, iov
, 3, &send_evt
);
117 * caller sets need_sig to non-NULL if it will handle
118 * signaling if required later.
119 * if need_sig is NULL, signal now if needed.
122 *need_sig
|= send_evt
;
123 else if (error
== 0 && send_evt
)
124 rte_vmbus_chan_signal_tx(chan
);
128 /* Do a scatter/gather send where the descriptor points to data. */
129 int rte_vmbus_chan_send_sglist(struct vmbus_channel
*chan
,
130 struct vmbus_gpa sg
[], uint32_t sglen
,
131 void *data
, uint32_t dlen
,
132 uint64_t xactid
, bool *need_sig
)
134 struct vmbus_chanpkt_sglist pkt
;
135 unsigned int pktlen
, pad_pktlen
, hlen
;
136 bool send_evt
= false;
141 hlen
= offsetof(struct vmbus_chanpkt_sglist
, gpa
[sglen
]);
142 pktlen
= hlen
+ dlen
;
143 pad_pktlen
= RTE_ALIGN(pktlen
, sizeof(uint64_t));
145 pkt
.hdr
.type
= VMBUS_CHANPKT_TYPE_GPA
;
146 pkt
.hdr
.flags
= VMBUS_CHANPKT_FLAG_RC
;
147 pkt
.hdr
.hlen
= hlen
>> VMBUS_CHANPKT_SIZE_SHIFT
;
148 pkt
.hdr
.tlen
= pad_pktlen
>> VMBUS_CHANPKT_SIZE_SHIFT
;
149 pkt
.hdr
.xactid
= xactid
;
153 iov
[0].iov_base
= &pkt
;
154 iov
[0].iov_len
= sizeof(pkt
);
155 iov
[1].iov_base
= sg
;
156 iov
[1].iov_len
= sizeof(struct vmbus_gpa
) * sglen
;
157 iov
[2].iov_base
= data
;
158 iov
[2].iov_len
= dlen
;
159 iov
[3].iov_base
= &pad
;
160 iov
[3].iov_len
= pad_pktlen
- pktlen
;
162 error
= vmbus_txbr_write(&chan
->txbr
, iov
, 4, &send_evt
);
164 /* if caller is batching, just propagate the status */
166 *need_sig
|= send_evt
;
167 else if (error
== 0 && send_evt
)
168 rte_vmbus_chan_signal_tx(chan
);
172 bool rte_vmbus_chan_rx_empty(const struct vmbus_channel
*channel
)
174 const struct vmbus_br
*br
= &channel
->rxbr
;
176 return br
->vbr
->rindex
== br
->vbr
->windex
;
179 /* Signal host after reading N bytes */
180 void rte_vmbus_chan_signal_read(struct vmbus_channel
*chan
, uint32_t bytes_read
)
182 struct vmbus_br
*rbr
= &chan
->rxbr
;
183 uint32_t write_sz
, pending_sz
;
185 /* No need for signaling on older versions */
186 if (!rbr
->vbr
->feature_bits
.feat_pending_send_sz
)
189 /* Make sure reading of pending happens after new read index */
192 pending_sz
= rbr
->vbr
->pending_send
;
197 write_sz
= vmbus_br_availwrite(rbr
, rbr
->vbr
->windex
);
199 /* If there was space before then host was not blocked */
200 if (write_sz
- bytes_read
> pending_sz
)
203 /* If pending write will not fit */
204 if (write_sz
<= pending_sz
)
207 vmbus_set_event(chan
->device
, chan
);
210 int rte_vmbus_chan_recv(struct vmbus_channel
*chan
, void *data
, uint32_t *len
,
211 uint64_t *request_id
)
213 struct vmbus_chanpkt_hdr pkt
;
214 uint32_t dlen
, hlen
, bufferlen
= *len
;
219 error
= vmbus_rxbr_peek(&chan
->rxbr
, &pkt
, sizeof(pkt
));
223 if (unlikely(pkt
.hlen
< VMBUS_CHANPKT_HLEN_MIN
)) {
224 VMBUS_LOG(ERR
, "VMBUS recv, invalid hlen %u", pkt
.hlen
);
225 /* XXX this channel is dead actually. */
229 if (unlikely(pkt
.hlen
> pkt
.tlen
)) {
230 VMBUS_LOG(ERR
, "VMBUS recv,invalid hlen %u and tlen %u",
235 /* Length are in quad words */
236 hlen
= pkt
.hlen
<< VMBUS_CHANPKT_SIZE_SHIFT
;
237 dlen
= (pkt
.tlen
<< VMBUS_CHANPKT_SIZE_SHIFT
) - hlen
;
240 /* If caller buffer is not large enough */
241 if (unlikely(dlen
> bufferlen
))
245 *request_id
= pkt
.xactid
;
247 /* Read data and skip packet header */
248 error
= vmbus_rxbr_read(&chan
->rxbr
, data
, dlen
, hlen
);
252 rte_vmbus_chan_signal_read(chan
, dlen
+ hlen
+ sizeof(uint64_t));
256 /* TODO: replace this with inplace ring buffer (no copy) */
257 int rte_vmbus_chan_recv_raw(struct vmbus_channel
*chan
,
258 void *data
, uint32_t *len
)
260 struct vmbus_chanpkt_hdr pkt
;
261 uint32_t dlen
, bufferlen
= *len
;
264 error
= vmbus_rxbr_peek(&chan
->rxbr
, &pkt
, sizeof(pkt
));
268 if (unlikely(pkt
.hlen
< VMBUS_CHANPKT_HLEN_MIN
)) {
269 VMBUS_LOG(ERR
, "VMBUS recv, invalid hlen %u", pkt
.hlen
);
270 /* XXX this channel is dead actually. */
274 if (unlikely(pkt
.hlen
> pkt
.tlen
)) {
275 VMBUS_LOG(ERR
, "VMBUS recv,invalid hlen %u and tlen %u",
280 /* Length are in quad words */
281 dlen
= pkt
.tlen
<< VMBUS_CHANPKT_SIZE_SHIFT
;
284 /* If caller buffer is not large enough */
285 if (unlikely(dlen
> bufferlen
))
288 /* Read data and skip packet header */
289 error
= vmbus_rxbr_read(&chan
->rxbr
, data
, dlen
, 0);
293 /* Return the number of bytes read */
294 return dlen
+ sizeof(uint64_t);
297 int vmbus_chan_create(const struct rte_vmbus_device
*device
,
298 uint16_t relid
, uint16_t subid
, uint8_t monitor_id
,
299 struct vmbus_channel
**new_chan
)
301 struct vmbus_channel
*chan
;
304 chan
= rte_zmalloc_socket("VMBUS", sizeof(*chan
), RTE_CACHE_LINE_SIZE
,
305 device
->device
.numa_node
);
309 STAILQ_INIT(&chan
->subchannel_list
);
310 chan
->device
= device
;
311 chan
->subchannel_id
= subid
;
313 chan
->monitor_id
= monitor_id
;
316 err
= vmbus_uio_map_rings(chan
);
325 /* Setup the primary channel */
326 int rte_vmbus_chan_open(struct rte_vmbus_device
*device
,
327 struct vmbus_channel
**new_chan
)
331 err
= vmbus_chan_create(device
, device
->relid
, 0,
332 device
->monitor_id
, new_chan
);
334 device
->primary
= *new_chan
;
339 int rte_vmbus_max_channels(const struct rte_vmbus_device
*device
)
341 if (vmbus_uio_subchannels_supported(device
, device
->primary
))
342 return VMBUS_MAX_CHANNELS
;
347 /* Setup secondary channel */
348 int rte_vmbus_subchan_open(struct vmbus_channel
*primary
,
349 struct vmbus_channel
**new_chan
)
351 struct vmbus_channel
*chan
;
354 err
= vmbus_uio_get_subchan(primary
, &chan
);
358 STAILQ_INSERT_TAIL(&primary
->subchannel_list
, chan
, next
);
363 uint16_t rte_vmbus_sub_channel_index(const struct vmbus_channel
*chan
)
365 return chan
->subchannel_id
;
368 void rte_vmbus_chan_close(struct vmbus_channel
*chan
)
370 const struct rte_vmbus_device
*device
= chan
->device
;
371 struct vmbus_channel
*primary
= device
->primary
;
374 STAILQ_REMOVE(&primary
->subchannel_list
, chan
,
375 vmbus_channel
, next
);
380 static void vmbus_dump_ring(FILE *f
, const char *id
, const struct vmbus_br
*br
)
382 const struct vmbus_bufring
*vbr
= br
->vbr
;
383 struct vmbus_chanpkt_hdr pkt
;
385 fprintf(f
, "%s windex=%u rindex=%u mask=%u pending=%u feature=%#x\n",
386 id
, vbr
->windex
, vbr
->rindex
, vbr
->imask
,
387 vbr
->pending_send
, vbr
->feature_bits
.value
);
388 fprintf(f
, " size=%u avail write=%u read=%u\n",
389 br
->dsize
, vmbus_br_availwrite(br
, vbr
->windex
),
390 vmbus_br_availread(br
));
392 if (vmbus_rxbr_peek(br
, &pkt
, sizeof(pkt
)) == 0)
393 fprintf(f
, " pkt type %#x len %u flags %#x xactid %#"PRIx64
"\n",
395 pkt
.tlen
<< VMBUS_CHANPKT_SIZE_SHIFT
,
396 pkt
.flags
, pkt
.xactid
);
399 void rte_vmbus_chan_dump(FILE *f
, const struct vmbus_channel
*chan
)
401 fprintf(f
, "channel[%u] relid=%u monitor=%u\n",
402 chan
->subchannel_id
, chan
->relid
, chan
->monitor_id
);
403 vmbus_dump_ring(f
, "rxbr", &chan
->rxbr
);
404 vmbus_dump_ring(f
, "txbr", &chan
->txbr
);