Skip to content

Commit

Permalink
avf: Handle chain buffer in TX properly
Browse files Browse the repository at this point in the history 
For chain buffer, need to traverse b->next_buffer to transmit all buffers
in the chain. Only set EOP at the last descriptor in the chain to signal
this is a chain descriptor.

Introduce slow path to handle ring wrap. This is needed because chain
buffer may consist of multiple pieces and it may span from near the end of
the ring to the beginning of the ring.

Type: fix

Signed-off-by: Steven Luong <sluong@cisco.com>
Change-Id: Id7c872f3e39e09f3566aa63f3cdba8f40736d508
  • Loading branch information
Steven Luong authored and Dave Barach committed Feb 25, 2020
1 parent 2da3971 commit f7ab7b2
Show file tree
Hide file tree
Showing 2 changed files with 140 additions and 7 deletions.
3 changes: 2 additions & 1 deletion src/plugins/avf/avf.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -333,7 +333,8 @@ typedef struct
} avf_input_trace_t;

#define foreach_avf_tx_func_error \
_(NO_FREE_SLOTS, "no free tx slots")
_(SEGMENT_SIZE_EXCEEDED, "segment size exceeded") \
_(NO_FREE_SLOTS, "no free tx slots")

typedef enum
{
Expand Down
144 changes: 138 additions & 6 deletions src/plugins/avf/output.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -31,8 +31,8 @@ avf_tx_desc_get_dtyp (avf_tx_desc_t * d)
}

static_always_inline u16
avf_tx_enqueue (vlib_main_t * vm, avf_txq_t * txq, u32 * buffers,
u32 n_packets, int use_va_dma)
avf_tx_enqueue (vlib_main_t * vm, vlib_node_runtime_t * node, avf_txq_t * txq,
u32 * buffers, u32 n_packets, int use_va_dma)
{
u16 next = txq->next;
u64 bits = AVF_TXD_CMD_EOP | AVF_TXD_CMD_RSV;
Expand All @@ -41,13 +41,16 @@ avf_tx_enqueue (vlib_main_t * vm, avf_txq_t * txq, u32 * buffers,
u16 mask = txq->size - 1;
vlib_buffer_t *b[4];
avf_tx_desc_t *d = txq->descs + next;
u16 n_desc_needed;
vlib_buffer_t *b0;

/* avoid ring wrap */
n_desc_left = txq->size - clib_max (txq->next, txq->n_enqueued + 8);

if (n_desc_left == 0)
return 0;

/* Fast path, no ring wrap */
while (n_packets_left && n_desc_left)
{
u32 or_flags;
Expand Down Expand Up @@ -103,6 +106,57 @@ avf_tx_enqueue (vlib_main_t * vm, avf_txq_t * txq, u32 * buffers,
txq->bufs[next] = buffers[0];
b[0] = vlib_get_buffer (vm, buffers[0]);

/* Deal with chain buffer if present */
if (b[0]->flags & VLIB_BUFFER_NEXT_PRESENT)
{
n_desc_needed = 1;
b0 = b[0];

/* Wish there were a buffer count for chain buffer */
while (b0->flags & VLIB_BUFFER_NEXT_PRESENT)
{
b0 = vlib_get_buffer (vm, b0->next_buffer);
n_desc_needed++;
}

/* spec says data descriptor is limited to 8 segments */
if (PREDICT_FALSE (n_desc_needed > 8))
{
vlib_buffer_free_one (vm, buffers[0]);
vlib_error_count (vm, node->node_index,
AVF_TX_ERROR_SEGMENT_SIZE_EXCEEDED, 1);
n_packets_left -= 1;
buffers += 1;
continue;
}

if (PREDICT_FALSE (n_desc_left < n_desc_needed))
/*
* Slow path may be able to to deal with this since it can handle
* ring wrap
*/
break;

while (b[0]->flags & VLIB_BUFFER_NEXT_PRESENT)
{
if (use_va_dma)
d[0].qword[0] = vlib_buffer_get_current_va (b[0]);
else
d[0].qword[0] = vlib_buffer_get_current_pa (vm, b[0]);

d[0].qword[1] = (((u64) b[0]->current_length) << 34) |
AVF_TXD_CMD_RSV;

next += 1;
n_desc += 1;
n_desc_left -= 1;
d += 1;

txq->bufs[next] = b[0]->next_buffer;
b[0] = vlib_get_buffer (vm, b[0]->next_buffer);
}
}

if (use_va_dma)
d[0].qword[0] = vlib_buffer_get_current_va (b[0]);
else
Expand All @@ -118,6 +172,84 @@ avf_tx_enqueue (vlib_main_t * vm, avf_txq_t * txq, u32 * buffers,
d += 1;
}

/* Slow path to support ring wrap */
if (PREDICT_FALSE (n_packets_left))
{
txq->n_enqueued += n_desc;

n_desc = 0;
d = txq->descs + (next & mask);

/* +8 to be consistent with fast path */
n_desc_left = txq->size - (txq->n_enqueued + 8);

while (n_packets_left && n_desc_left)
{
txq->bufs[next & mask] = buffers[0];
b[0] = vlib_get_buffer (vm, buffers[0]);

/* Deal with chain buffer if present */
if (b[0]->flags & VLIB_BUFFER_NEXT_PRESENT)
{
n_desc_needed = 1;
b0 = b[0];

while (b0->flags & VLIB_BUFFER_NEXT_PRESENT)
{
b0 = vlib_get_buffer (vm, b0->next_buffer);
n_desc_needed++;
}

/* Spec says data descriptor is limited to 8 segments */
if (PREDICT_FALSE (n_desc_needed > 8))
{
vlib_buffer_free_one (vm, buffers[0]);
vlib_error_count (vm, node->node_index,
AVF_TX_ERROR_SEGMENT_SIZE_EXCEEDED, 1);
n_packets_left -= 1;
buffers += 1;
continue;
}

if (PREDICT_FALSE (n_desc_left < n_desc_needed))
break;

while (b[0]->flags & VLIB_BUFFER_NEXT_PRESENT)
{
if (use_va_dma)
d[0].qword[0] = vlib_buffer_get_current_va (b[0]);
else
d[0].qword[0] = vlib_buffer_get_current_pa (vm, b[0]);

d[0].qword[1] = (((u64) b[0]->current_length) << 34) |
AVF_TXD_CMD_RSV;

next += 1;
n_desc += 1;
n_desc_left -= 1;
d = txq->descs + (next & mask);

txq->bufs[next & mask] = b[0]->next_buffer;
b[0] = vlib_get_buffer (vm, b[0]->next_buffer);
}
}

if (use_va_dma)
d[0].qword[0] = vlib_buffer_get_current_va (b[0]);
else
d[0].qword[0] = vlib_buffer_get_current_pa (vm, b[0]);

d[0].qword[1] = (((u64) b[0]->current_length) << 34) | bits;

next += 1;
n_desc += 1;
buffers += 1;
n_packets_left -= 1;
n_desc_left -= 1;
d = txq->descs + (next & mask);
}
}

if ((slot = clib_ring_enq (txq->rs_slots)))
{
u16 rs_slot = slot[0] = (next - 1) & mask;
Expand Down Expand Up @@ -177,15 +309,15 @@ VNET_DEVICE_CLASS_TX_FN (avf_device_class) (vlib_main_t * vm,
n_free = (complete_slot + 1 - first) & mask;

txq->n_enqueued -= n_free;
vlib_buffer_free_from_ring (vm, txq->bufs, first, txq->size,
n_free);
vlib_buffer_free_from_ring_no_next (vm, txq->bufs, first, txq->size,
n_free);
}
}

if (ad->flags & AVF_DEVICE_F_VA_DMA)
n_enq = avf_tx_enqueue (vm, txq, buffers, n_left, 1);
n_enq = avf_tx_enqueue (vm, node, txq, buffers, n_left, 1);
else
n_enq = avf_tx_enqueue (vm, txq, buffers, n_left, 0);
n_enq = avf_tx_enqueue (vm, node, txq, buffers, n_left, 0);

n_left -= n_enq;

Expand Down

0 comments on commit f7ab7b2

Please sign in to comment.