aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorPierre Pfister <ppfister@cisco.com>2016-07-27 16:38:20 +0100
committerDamjan Marion <dmarion.lists@gmail.com>2016-09-07 17:19:11 +0000
commitba1d04641454beaae04a6c9ca7146d5f3205068f (patch)
tree0cc15d5505a88b426045fd683b03da7d6c01e54d
parent24704851584a2b1db4545a2a3cd3fc73dff2ce5f (diff)
Vhost-User: Implement Indirect Descriptors
One of the main issue with virtio is the small queue size. VPP can quickly fill the queue before Linux could empty it, resulting in packet losses. Virtio indirect descriptors double the number of packet which may fit in the virtio queue (From 128 to 256 with standard qemu which hardcodes queue size to 256). Linux will not use such descriptors if the virtio MRG feature is enabled. You may add mrg_rxbuff=off option to qemu commandline when starting the VM such as to disable this feature. This way, Linux will use indirect buffers to receive packets. But it seems that Linux never uses indirect buffers to send packets. This patch also includes some optimization modifications. Change-Id: I26940d41ce6b7e3a08d5516018a8c46c5316ce1c Signed-off-by: Pierre Pfister <ppfister@cisco.com>
-rw-r--r--vnet/vnet/devices/virtio/vhost-user.c274
-rw-r--r--vnet/vnet/devices/virtio/vhost-user.h2
2 files changed, 182 insertions, 94 deletions
diff --git a/vnet/vnet/devices/virtio/vhost-user.c b/vnet/vnet/devices/virtio/vhost-user.c
index 6fa1c652c8f..59daf871200 100644
--- a/vnet/vnet/devices/virtio/vhost-user.c
+++ b/vnet/vnet/devices/virtio/vhost-user.c
@@ -64,7 +64,9 @@ vlib_node_registration_t vhost_user_input_node;
_(NONE, "no error") \
_(NOT_READY, "vhost user state error") \
_(PKT_DROP_NOBUF, "tx packet drops (no available descriptors)") \
- _(MMAP_FAIL, "mmap failure")
+ _(PKT_DROP_NOMRG, "tx packet drops (cannot merge descriptors)") \
+ _(MMAP_FAIL, "mmap failure") \
+ _(INDIRECT_OVERFLOW, "indirect descriptor table overflow")
typedef enum
{
@@ -84,7 +86,9 @@ static char *vhost_user_tx_func_error_strings[] = {
_(NO_ERROR, "no error") \
_(NO_BUFFER, "no available buffer") \
_(MMAP_FAIL, "mmap failure") \
- _(UNDERSIZED_FRAME, "undersized ethernet frame received (< 14 bytes)")
+ _(INDIRECT_OVERFLOW, "indirect descriptor overflows table") \
+ _(UNDERSIZED_FRAME, "undersized ethernet frame received (< 14 bytes)") \
+ _(FULL_RX_QUEUE, "full rx queue (possible driver tx drop)")
typedef enum
{
@@ -383,6 +387,7 @@ vhost_user_socket_read (unix_file_t * uf)
msg.flags |= 4;
msg.u64 = (1 << FEAT_VIRTIO_NET_F_MRG_RXBUF) |
(1 << FEAT_VIRTIO_F_ANY_LAYOUT) |
+ (1 << FEAT_VIRTIO_F_INDIRECT_DESC) |
(1 << FEAT_VHOST_F_LOG_ALL) |
(1 << FEAT_VIRTIO_NET_F_GUEST_ANNOUNCE) |
(1 << FEAT_VHOST_USER_F_PROTOCOL_FEATURES);
@@ -957,14 +962,18 @@ vhost_user_if_input (vlib_main_t * vm,
if (PREDICT_FALSE (txvq->avail->flags & 0xFFFE))
return 0;
+ n_left = (u16) (txvq->avail->idx - txvq->last_avail_idx);
+
/* nothing to do */
- if (txvq->avail->idx == txvq->last_avail_idx)
+ if (PREDICT_FALSE (n_left == 0))
return 0;
- if (PREDICT_TRUE (txvq->avail->idx > txvq->last_avail_idx))
- n_left = txvq->avail->idx - txvq->last_avail_idx;
- else /* wrapped */
- n_left = (u16) - 1 - txvq->last_avail_idx + txvq->avail->idx;
+ if (PREDICT_FALSE (n_left == txvq->qsz))
+ {
+ //Informational error logging when VPP is not receiving packets fast enough
+ vlib_error_count (vm, node->node_index,
+ VHOST_USER_INPUT_FUNC_ERROR_FULL_RX_QUEUE, 1);
+ }
if (PREDICT_FALSE (!vui->admin_up))
{
@@ -977,9 +986,6 @@ vhost_user_if_input (vlib_main_t * vm,
return 0;
}
- if (PREDICT_FALSE (n_left > txvq->qsz))
- return 0;
-
qsz_mask = txvq->qsz - 1;
cpu_index = os_get_cpu_number ();
drops = 0;
@@ -997,7 +1003,7 @@ vhost_user_if_input (vlib_main_t * vm,
*/
if (PREDICT_FALSE (!vum->rx_buffers[cpu_index]))
{
- vec_alloc (vum->rx_buffers[cpu_index], VLIB_FRAME_SIZE);
+ vec_alloc (vum->rx_buffers[cpu_index], 2 * VLIB_FRAME_SIZE);
if (PREDICT_FALSE (!vum->rx_buffers[cpu_index]))
flush = n_left; //Drop all input
@@ -1005,14 +1011,12 @@ vhost_user_if_input (vlib_main_t * vm,
if (PREDICT_FALSE (_vec_len (vum->rx_buffers[cpu_index]) < n_left))
{
+ u32 curr_len = _vec_len (vum->rx_buffers[cpu_index]);
_vec_len (vum->rx_buffers[cpu_index]) +=
vlib_buffer_alloc_from_free_list (vm,
vum->rx_buffers[cpu_index] +
- _vec_len (vum->rx_buffers
- [cpu_index]),
- VLIB_FRAME_SIZE -
- _vec_len (vum->rx_buffers
- [cpu_index]),
+ curr_len,
+ 2 * VLIB_FRAME_SIZE - curr_len,
VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX);
if (PREDICT_FALSE (n_left > _vec_len (vum->rx_buffers[cpu_index])))
@@ -1053,6 +1057,20 @@ vhost_user_if_input (vlib_main_t * vm,
u16 desc_chain_head, desc_current;
u8 error = VHOST_USER_INPUT_FUNC_ERROR_NO_ERROR;
+ if (PREDICT_TRUE (n_left > 1))
+ {
+ u32 next_desc =
+ txvq->avail->ring[(txvq->last_avail_idx + 1) & qsz_mask];
+ void *buffer_addr =
+ map_guest_mem (vui, txvq->desc[next_desc].addr);
+ if (PREDICT_TRUE (buffer_addr != 0))
+ CLIB_PREFETCH (buffer_addr, 64, STORE);
+
+ u32 bi = vum->rx_buffers[cpu_index][rx_len - 2];
+ vlib_prefetch_buffer_with_index (vm, bi, STORE);
+ CLIB_PREFETCH (vlib_get_buffer (vm, bi)->data, 128, STORE);
+ }
+
desc_chain_head = desc_current =
txvq->avail->ring[txvq->last_avail_idx & qsz_mask];
bi_head = bi_current = vum->rx_buffers[cpu_index][--rx_len];
@@ -1061,7 +1079,8 @@ vhost_user_if_input (vlib_main_t * vm,
uword offset;
if (PREDICT_TRUE (vui->is_any_layout) ||
- !(txvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT))
+ (!(txvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT) &&
+ !(txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT)))
{
/* ANYLAYOUT or single buffer */
offset = vui->virtio_net_hdr_sz;
@@ -1072,14 +1091,35 @@ vhost_user_if_input (vlib_main_t * vm,
offset = txvq->desc[desc_current].len;
}
+ vring_desc_t *desc_table = txvq->desc;
+ u32 desc_index = desc_current;
+
+ if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT)
+ {
+ desc_table = map_guest_mem (vui, txvq->desc[desc_current].addr);
+ desc_index = 0;
+ if (PREDICT_FALSE (desc_table == 0))
+ {
+ error = VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL;
+ goto out;
+ }
+ }
+
while (1)
{
void *buffer_addr =
- map_guest_mem (vui, txvq->desc[desc_current].addr);
+ map_guest_mem (vui, desc_table[desc_index].addr);
if (PREDICT_FALSE (buffer_addr == 0))
{
error = VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL;
- break;
+ goto out;
+ }
+
+ if (PREDICT_TRUE
+ (desc_table[desc_index].flags & VIRTQ_DESC_F_NEXT))
+ {
+ CLIB_PREFETCH (&desc_table[desc_table[desc_index].next],
+ sizeof (vring_desc_t), STORE);
}
#if VHOST_USER_COPY_TX_HDR == 1
@@ -1087,9 +1127,9 @@ vhost_user_if_input (vlib_main_t * vm,
clib_memcpy (b->pre_data, buffer_addr, sizeof (virtio_net_hdr_t)); /* 12 byte hdr is not used on tx */
#endif
- if (txvq->desc[desc_current].len > offset)
+ if (desc_table[desc_index].len > offset)
{
- u16 len = txvq->desc[desc_current].len - offset;
+ u16 len = desc_table[desc_index].len - offset;
u16 copied = vlib_buffer_chain_append_data_with_alloc (vm,
VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX,
b_head,
@@ -1098,7 +1138,6 @@ vhost_user_if_input (vlib_main_t * vm,
+
offset,
len);
-
if (copied != len)
{
error = VHOST_USER_INPUT_FUNC_ERROR_NO_BUFFER;
@@ -1108,11 +1147,12 @@ vhost_user_if_input (vlib_main_t * vm,
offset = 0;
/* if next flag is set, take next desc in the chain */
- if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT)
- desc_current = txvq->desc[desc_current].next;
+ if ((desc_table[desc_index].flags & VIRTQ_DESC_F_NEXT))
+ desc_index = desc_table[desc_index].next;
else
- break;
+ goto out;
}
+ out:
/* consume the descriptor and return it as used */
txvq->last_avail_idx++;
@@ -1123,11 +1163,14 @@ vhost_user_if_input (vlib_main_t * vm,
ring[txvq->last_used_idx & qsz_mask]);
txvq->last_used_idx++;
+ //It is important to free RX as fast as possible such that the TX
+ //process does not drop packets
+ if ((txvq->last_used_idx & 0x3f) == 0) // Every 64 packets
+ txvq->used->idx = txvq->last_used_idx;
+
if (PREDICT_FALSE (b_head->current_length < 14 &&
error == VHOST_USER_INPUT_FUNC_ERROR_NO_ERROR))
- {
- error = VHOST_USER_INPUT_FUNC_ERROR_UNDERSIZED_FRAME;
- }
+ error = VHOST_USER_INPUT_FUNC_ERROR_UNDERSIZED_FRAME;
VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b_head);
@@ -1162,6 +1205,7 @@ vhost_user_if_input (vlib_main_t * vm,
}
vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+
}
if (PREDICT_TRUE (vum->rx_buffers[cpu_index] != 0))
@@ -1264,7 +1308,6 @@ vhost_user_intfc_tx (vlib_main_t * vm,
{
u32 *buffers = vlib_frame_args (frame);
u32 n_left = 0;
- u16 used_index;
vhost_user_main_t *vum = &vhost_user_main;
uword n_packets = 0;
vnet_interface_output_runtime_t *rd = (void *) node->runtime_data;
@@ -1274,6 +1317,8 @@ vhost_user_intfc_tx (vlib_main_t * vm,
u16 qsz_mask;
u8 error = VHOST_USER_TX_FUNC_ERROR_NONE;
+ n_left = n_packets = frame->n_vectors;
+
if (PREDICT_FALSE (!vui->is_up))
goto done2;
@@ -1304,23 +1349,18 @@ vhost_user_intfc_tx (vlib_main_t * vm,
goto done2;
}
- n_left = n_packets = frame->n_vectors;
- used_index = rxvq->used->idx;
qsz_mask = rxvq->qsz - 1; /* qsz is always power of 2 */
while (n_left > 0)
{
vlib_buffer_t *b0, *current_b0;
- u16 desc_chain_head, desc_current, desc_len;
+ u16 desc_head, desc_index, desc_len;
+ vring_desc_t *desc_table;
void *buffer_addr;
- uword offset;
-
- if (n_left >= 2)
- vlib_prefetch_buffer_with_index (vm, buffers[1], LOAD);
+ u32 buffer_len;
b0 = vlib_get_buffer (vm, buffers[0]);
buffers++;
- n_left--;
if (PREDICT_FALSE (rxvq->last_avail_idx == rxvq->avail->idx))
{
@@ -1328,41 +1368,54 @@ vhost_user_intfc_tx (vlib_main_t * vm,
goto done;
}
- desc_current = desc_chain_head =
+ desc_table = rxvq->desc;
+ desc_head = desc_index =
rxvq->avail->ring[rxvq->last_avail_idx & qsz_mask];
- offset = vui->virtio_net_hdr_sz;
- desc_len = offset;
+ if (rxvq->desc[desc_head].flags & VIRTQ_DESC_F_INDIRECT)
+ {
+ if (PREDICT_FALSE
+ (rxvq->desc[desc_head].len < sizeof (vring_desc_t)))
+ {
+ error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
+ goto done;
+ }
+ if (PREDICT_FALSE
+ (!(desc_table =
+ map_guest_mem (vui, rxvq->desc[desc_index].addr))))
+ {
+ error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
+ goto done;
+ }
+ desc_index = 0;
+ }
+
+ desc_len = vui->virtio_net_hdr_sz;
+
if (PREDICT_FALSE
- (!(buffer_addr =
- map_guest_mem (vui, rxvq->desc[desc_current].addr))))
+ (!(buffer_addr = map_guest_mem (vui, desc_table[desc_index].addr))))
{
error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
goto done;
}
- CLIB_PREFETCH (buffer_addr, clib_min (rxvq->desc[desc_current].len,
- 4 * CLIB_CACHE_LINE_BYTES),
- STORE);
+ buffer_len = desc_table[desc_index].len;
+
+ CLIB_PREFETCH (buffer_addr,
+ clib_min (buffer_len, 2 * CLIB_CACHE_LINE_BYTES), STORE);
virtio_net_hdr_mrg_rxbuf_t *hdr =
(virtio_net_hdr_mrg_rxbuf_t *) buffer_addr;
hdr->hdr.flags = 0;
hdr->hdr.gso_type = 0;
-
- vhost_user_log_dirty_pages (vui, rxvq->desc[desc_current].addr,
- vui->virtio_net_hdr_sz);
-
if (vui->virtio_net_hdr_sz == 12)
hdr->num_buffers = 1;
+ vhost_user_log_dirty_pages (vui, desc_table[desc_index].addr,
+ vui->virtio_net_hdr_sz);
+
u16 bytes_left = b0->current_length;
- buffer_addr += offset;
+ buffer_addr += vui->virtio_net_hdr_sz;
+ buffer_len -= vui->virtio_net_hdr_sz;
current_b0 = b0;
-
- //FIXME: This was in the code but I don't think it is valid
- /*if (PREDICT_FALSE(!vui->is_any_layout && (rxvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT))) {
- rxvq->desc[desc_current].len = vui->virtio_net_hdr_sz;
- } */
-
while (1)
{
if (!bytes_left)
@@ -1379,99 +1432,132 @@ vhost_user_intfc_tx (vlib_main_t * vm,
}
}
- if (rxvq->desc[desc_current].len <= offset)
+ if (buffer_len == 0)
{ //Get new output
- if (rxvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT)
+ if (desc_table[desc_index].flags & VIRTQ_DESC_F_NEXT)
{
- offset = 0;
- desc_current = rxvq->desc[desc_current].next;
+ //Next one is chained
+ desc_index = desc_table[desc_index].next;
if (PREDICT_FALSE
(!(buffer_addr =
- map_guest_mem (vui, rxvq->desc[desc_current].addr))))
+ map_guest_mem (vui, desc_table[desc_index].addr))))
{
- used_index -= hdr->num_buffers - 1;
+ rxvq->last_used_idx -= hdr->num_buffers - 1;
rxvq->last_avail_idx -= hdr->num_buffers - 1;
error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
goto done;
}
+ buffer_len = desc_table[desc_index].len;
}
- else if (vui->virtio_net_hdr_sz == 12)
- { //MRG is available
-
+ else if (vui->virtio_net_hdr_sz == 12) //MRG is available
+ {
//Move from available to used buffer
- rxvq->used->ring[used_index & qsz_mask].id =
- desc_chain_head;
- rxvq->used->ring[used_index & qsz_mask].len = desc_len;
+ rxvq->used->ring[rxvq->last_used_idx & qsz_mask].id =
+ desc_head;
+ rxvq->used->ring[rxvq->last_used_idx & qsz_mask].len =
+ desc_len;
vhost_user_log_dirty_ring (vui, rxvq,
- ring[used_index & qsz_mask]);
+ ring[rxvq->last_used_idx &
+ qsz_mask]);
rxvq->last_avail_idx++;
- used_index++;
+ rxvq->last_used_idx++;
hdr->num_buffers++;
if (PREDICT_FALSE
(rxvq->last_avail_idx == rxvq->avail->idx))
{
//Dequeue queued descriptors for this packet
- used_index -= hdr->num_buffers - 1;
+ rxvq->last_used_idx -= hdr->num_buffers - 1;
rxvq->last_avail_idx -= hdr->num_buffers - 1;
error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
goto done;
}
- //Look at next one
- desc_chain_head =
+ desc_table = rxvq->desc;
+ desc_head = desc_index =
rxvq->avail->ring[rxvq->last_avail_idx & qsz_mask];
- desc_current = desc_chain_head;
- desc_len = 0;
- offset = 0;
+ if (PREDICT_FALSE
+ (rxvq->desc[desc_head].flags & VIRTQ_DESC_F_INDIRECT))
+ {
+ //It is seriously unlikely that a driver will put indirect descriptor
+ //after non-indirect descriptor.
+ if (PREDICT_FALSE
+ (rxvq->desc[desc_head].len < sizeof (vring_desc_t)))
+ {
+ error = VHOST_USER_TX_FUNC_ERROR_INDIRECT_OVERFLOW;
+ goto done;
+ }
+ if (PREDICT_FALSE
+ (!(desc_table =
+ map_guest_mem (vui,
+ rxvq->desc[desc_index].addr))))
+ {
+ error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
+ goto done;
+ }
+ desc_index = 0;
+ }
+
if (PREDICT_FALSE
(!(buffer_addr =
- map_guest_mem (vui, rxvq->desc[desc_current].addr))))
+ map_guest_mem (vui, desc_table[desc_index].addr))))
{
- //Dequeue queued descriptors for this packet
- used_index -= hdr->num_buffers - 1;
- rxvq->last_avail_idx -= hdr->num_buffers - 1;
error = VHOST_USER_TX_FUNC_ERROR_MMAP_FAIL;
goto done;
}
+ buffer_len = desc_table[desc_index].len;
+ CLIB_PREFETCH (buffer_addr,
+ clib_min (buffer_len,
+ 2 * CLIB_CACHE_LINE_BYTES), STORE);
}
else
{
- error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOBUF;
+ error = VHOST_USER_TX_FUNC_ERROR_PKT_DROP_NOMRG;
goto done;
}
}
- u16 bytes_to_copy =
- bytes_left >
- (rxvq->desc[desc_current].len -
- offset) ? (rxvq->desc[desc_current].len - offset) : bytes_left;
+ u16 bytes_to_copy = bytes_left;
+ bytes_to_copy =
+ (bytes_to_copy > buffer_len) ? buffer_len : bytes_to_copy;
clib_memcpy (buffer_addr,
vlib_buffer_get_current (current_b0) +
current_b0->current_length - bytes_left,
bytes_to_copy);
vhost_user_log_dirty_pages (vui,
- rxvq->desc[desc_current].addr + offset,
+ desc_table[desc_index].addr +
+ desc_table[desc_index].len -
+ bytes_left - bytes_to_copy,
bytes_to_copy);
+
bytes_left -= bytes_to_copy;
- offset += bytes_to_copy;
+ buffer_len -= bytes_to_copy;
buffer_addr += bytes_to_copy;
desc_len += bytes_to_copy;
}
+ if (PREDICT_TRUE (n_left >= 2))
+ {
+ vlib_prefetch_buffer_with_index (vm, buffers[1], STORE);
+ CLIB_PREFETCH (&n_left, sizeof (n_left), STORE);
+ }
+
//Move from available to used ring
- rxvq->used->ring[used_index & qsz_mask].id = desc_chain_head;
- rxvq->used->ring[used_index & qsz_mask].len = desc_len;
- vhost_user_log_dirty_ring (vui, rxvq, ring[used_index & qsz_mask]);
+ rxvq->used->ring[rxvq->last_used_idx & qsz_mask].id = desc_head;
+ rxvq->used->ring[rxvq->last_used_idx & qsz_mask].len = desc_len;
+ vhost_user_log_dirty_ring (vui, rxvq,
+ ring[rxvq->last_used_idx & qsz_mask]);
rxvq->last_avail_idx++;
- used_index++;
+ rxvq->last_used_idx++;
+
+ n_left--; //At the end for error counting when 'goto done' is invoked
}
done:
CLIB_MEMORY_BARRIER ();
- rxvq->used->idx = used_index;
+ rxvq->used->idx = rxvq->last_used_idx;
vhost_user_log_dirty_ring (vui, rxvq, idx);
/* interrupt (call) handling */
@@ -2221,8 +2307,8 @@ show_vhost_user_command_fn (vlib_main_t * vm,
vui->vrings[q].desc[j].next,
pointer_to_uword (map_guest_mem
(vui,
- vui->vrings[q].
- desc[j].addr)));
+ vui->vrings[q].desc[j].
+ addr)));
}
}
}
diff --git a/vnet/vnet/devices/virtio/vhost-user.h b/vnet/vnet/devices/virtio/vhost-user.h
index 4934a8c0c1e..fa41403874b 100644
--- a/vnet/vnet/devices/virtio/vhost-user.h
+++ b/vnet/vnet/devices/virtio/vhost-user.h
@@ -24,6 +24,7 @@
#define VHOST_NET_VRING_NUM 2
#define VIRTQ_DESC_F_NEXT 1
+#define VIRTQ_DESC_F_INDIRECT 4
#define VHOST_USER_REPLY_MASK (0x1 << 2)
#define VHOST_USER_PROTOCOL_F_MQ 0
@@ -51,6 +52,7 @@
#define foreach_virtio_net_feature \
_ (VIRTIO_NET_F_MRG_RXBUF, 15) \
_ (VIRTIO_F_ANY_LAYOUT, 27) \
+ _ (VIRTIO_F_INDIRECT_DESC, 28) \
_ (VHOST_F_LOG_ALL, 26) \
_ (VIRTIO_NET_F_GUEST_ANNOUNCE, 21) \
_ (VHOST_USER_F_PROTOCOL_FEATURES, 30)