thanks bruce.
I didn't know that PCI slots have direct socket affinity.
is it static or configurable through PCI configuration space?
well, my NUT, two node NUMA, seems always returns -1 on calling
rte_eth_dev_socket_id(portid) whenever portid is 0, 1, or other values.
I appreciate if you explain
In DPDK based switching enviroment, mostly vhost runs on a dedicated core
while virtio processing in guest VMs runs on different cores.
Take RX for example, with generic implementation, for each guest buffer,
a) virtio driver allocates a descriptor from free descriptor list
b) modify the entry of a
In DPDK based switching enviroment, mostly vhost runs on a dedicated core
while virtio processing in guest VMs runs on different cores.
Take RX for example, with generic implementation, for each guest buffer,
a) virtio driver allocates a descriptor from free descriptor list
b) modify the entry of a
In DPDK based switching enviroment, mostly vhost runs on a dedicated core
while virtio processing in guest VMs runs on different cores.
Take RX for example, with generic implementation, for each guest buffer,
a) virtio driver allocates a descriptor from free descriptor list
b) modify the entry of a
fill avail ring with blank mbufs in virtio_dev_vring_start
Signed-off-by: Huawei Xie
---
drivers/net/virtio/Makefile | 2 +-
drivers/net/virtio/virtio_rxtx.c| 6 ++-
drivers/net/virtio/virtio_rxtx.h| 3 ++
drivers/net/virtio/virtio_rxtx_simple.c | 84 ++
Would move all rx/tx related code into this header file in future.
Add RTE_VIRTIO_PMD_MAX_BURST.
Signed-off-by: Huawei Xie
---
drivers/net/virtio/virtio_ethdev.c | 1 +
drivers/net/virtio/virtio_rxtx.c | 1 +
drivers/net/virtio/virtio_rxtx.h | 34 ++
3 files
Add software RX ring in virtqueue.
Add fake_mbuf in virtqueue for wraparound processing.
Use global simple_rxtx to indicate whether simple rxtx is enabled
Signed-off-by: Huawei Xie
---
drivers/net/virtio/virtio_ethdev.c | 12
drivers/net/virtio/virtio_rxtx.c | 7 +++
drivers/
With fixed avail ring, we don't need to get desc idx from avail ring.
virtio driver only has to deal with desc ring.
This patch uses vector instruction to accelerate processing desc ring.
Signed-off-by: Huawei Xie
---
drivers/net/virtio/virtio_ethdev.h | 2 +
drivers/net/virtio/virtio_rxt
bulk free of mbufs when clean used ring.
shift operation of idx could be further saved if vq_free_cnt means
free slots rather than free descriptors.
Signed-off-by: Huawei Xie
---
drivers/net/virtio/virtio_ethdev.h | 3 ++
drivers/net/virtio/virtio_rxtx_simple.c | 95 +++
simple rx/tx func is enabled when user specifies single segment, no offload
support.
merge-able should be disabled to use simple rxtx.
Signed-off-by: Huawei Xie
---
drivers/net/virtio/virtio_rxtx.c | 12
1 file changed, 12 insertions(+)
diff --git a/drivers/net/virtio/virtio_rxtx.
On Fri, Oct 16, 2015 at 02:52:30PM +0100, Bruce Richardson wrote:
> On Thu, Oct 15, 2015 at 04:18:59PM +0300, Michael S. Tsirkin wrote:
> > On Thu, Oct 15, 2015 at 02:08:39PM +0300, Marcel Apfelbaum wrote:
> > > Make vhost-user virtio 1.0 compatible by adding it to the
> > > supported features and
On Sun, 18 Oct 2015 14:29:03 +0800
Huawei Xie wrote:
> bulk free of mbufs when clean used ring.
> shift operation of idx could be further saved if vq_free_cnt means
> free slots rather than free descriptors.
>
> Signed-off-by: Huawei Xie
Did you measure this. I finished my transmit optimizatio
On Sun, 18 Oct 2015 14:29:03 +0800
Huawei Xie wrote:
> +
> + for (i = 1; i < VIRTIO_TX_FREE_NR; i++) {
> + m = (struct rte_mbuf *)vq->vq_descx[desc_idx++].cookie;
> + if (likely(m->pool == free[0]->pool))
> + free[nb_free++] = m;
> + els
+static inline void __attribute__((always_inline))
+virtio_xmit_cleanup(struct virtqueue *vq)
+{
Please don't use always inline, frustrating the compiler isn't going
to help.
+ uint16_t i, desc_idx;
+ int nb_free = 0;
+ struct rte_mbuf *m, *free[VIRTIO_TX_MAX_FREE_BUF_SZ];
+
+
On Sun, 18 Oct 2015 14:28:59 +0800
Huawei Xie wrote:
> + if (vq->sw_ring)
> + rte_free(vq->sw_ring);
> +
Do not need to test for NULL before calling rte_free.
Better to just rely on the fact that rte_free(NULL) is documented
to be ok (no operation).
This is a tested version of the virtio Tx performance improvements
that I posted earlier on the list, and described at the DPDK Userspace
meeting in Dublin. Together they get a 25% performance improvement for
both small packet and large multi-segment packet case when testing
from DPDK guest applica
The space check for transmit ring only needs a single conditional.
I.e only need to recheck for space if there was no space in first check.
This can help performance and simplifies loop.
Signed-off-by: Stephen Hemminger
---
drivers/net/virtio/virtio_rxtx.c | 66 -
Don't use unlikely() for VLAN or ring getting full.
GCC will not optimize code in unlikely paths and since these can
happen with normal code that can hurt performance.
Signed-off-by: Stephen Hemminger
---
drivers/net/virtio/virtio_rxtx.c | 7 +++
1 file changed, 3 insertions(+), 4 deletions(
The virtio ring in QEMU/KVM is usually limited to 256 entries
and the normal way that virtio driver was queuing mbufs required
nsegs + 1 ring elements. By using the indirect ring element feature
if available, each packet will take only one ring slot even for
multi-segment packets.
Signed-off-by: S
Virtio supports a feature that allows sender to put transmit
header prepended to data. It requires that the mbuf be writeable, correct
alignment, and the feature has been negotiatied. If all this works out,
then it will be the optimum way to transmit a single segment packet.
Signed-off-by: Steph
All the error checks in virtqueue_enqueue_xmit are already done
by the caller. Therefore they can be removed to improve performance.
Signed-off-by: Stephen Hemminger
---
drivers/net/virtio/virtio_rxtx.c | 23 ++-
1 file changed, 2 insertions(+), 21 deletions(-)
diff --git a/
21 matches
Mail list logo
