diff options
| author |   Steven <sluong@cisco.com> | 2018-11-02 16:28:52 -0700 |
|---|---|---|
| committer |   Damjan Marion <dmarion@me.com> | 2018-11-07 01:13:01 +0000 |
| commit | 005e4d50416d15368a6687076c7c934f89470f72 (patch) | |
| tree | 0bc050d6c369e6b2298dbe44003605979bfa7cf0 /src/plugins/vmxnet3/output.c | |
| parent | 6ccff02def39f314765f82a39f5dea9a667e47c8 (diff) | |
vmxnet3: reduce calling vmxnet3_reg_write_inline
In output.c, we buffer the descriptors and call vmxnet3_reg_write_inline
once outside the loop. This change improves the performance dramatically.
When refilling the ring, there is no need to inform the device unless
explicitly specified by the device (ctrl.update_prod == 1)
Change-Id: I7031d58bff0d249e913d14236d416c91eb6ab94a
Signed-off-by: Steven <sluong@cisco.com>
Diffstat (limited to 'src/plugins/vmxnet3/output.c')
| -rw-r--r-- | src/plugins/vmxnet3/output.c | 11 |
1 files changed, 7 insertions, 4 deletions
diff --git a/src/plugins/vmxnet3/output.c b/src/plugins/vmxnet3/output.c index c80ad0706cc..2886c31edf7 100644 --- a/src/plugins/vmxnet3/output.c +++ b/src/plugins/vmxnet3/output.c @@ -109,7 +109,7 @@ VNET_DEVICE_CLASS_TX_FN (vmxnet3_device_class) (vlib_main_t * vm, u16 n_left = frame->n_vectors; vmxnet3_txq_t *txq; u32 thread_index = vm->thread_index; - u16 qid = thread_index; + u16 qid = thread_index, produce; if (PREDICT_FALSE (!(vd->flags & VMXNET3_DEVICE_F_LINK_UP))) { @@ -124,7 +124,8 @@ VNET_DEVICE_CLASS_TX_FN (vmxnet3_device_class) (vlib_main_t * vm, vmxnet3_txq_release (vm, vd, txq); - while (n_left) + produce = txq->tx_ring.produce; + while (PREDICT_TRUE (n_left)) { u16 space_needed = 1, i; vlib_buffer_t *b; @@ -195,13 +196,15 @@ VNET_DEVICE_CLASS_TX_FN (vmxnet3_device_class) (vlib_main_t * vm, * Device can start reading the packet */ txq->tx_desc[first_idx].flags[0] ^= VMXNET3_TXF_GEN; - vmxnet3_reg_write_inline (vd, 0, VMXNET3_REG_TXPROD, - txq->tx_ring.produce); buffers++; n_left--; } + if (PREDICT_TRUE (produce != txq->tx_ring.produce)) + vmxnet3_reg_write_inline (vd, 0, VMXNET3_REG_TXPROD, + txq->tx_ring.produce); + clib_spinlock_unlock_if_init (&txq->lock); return (frame->n_vectors - n_left); |