xennetback(4): Nix trailing whitespace.

No functional change intended.

xennetback(4): Omit needless membars in xennetback_connect.

xneti is a private data structure to which we have exclusive access
here; ordering the stores doesn't make sense.

xennetback(4): Fix membars in xennetback_rx_copy_process.

- No need for barrier around touching req_cons and rsp_prod_pvt,
which are private.

- RING_PUSH_RESPONSES_AND_CHECK_NOTIFY already issues

xennetback(4): Fix membars in xennetback_rx_copy_process.

- No need for barrier around touching req_cons and rsp_prod_pvt,
which are private.

- RING_PUSH_RESPONSES_AND_CHECK_NOTIFY already issues xen_wmb, no
need to add one explicitly.

- After pushing responses, must issue xen_wmb (not xen_rmb) before
hypervisor_notify_via_evtchn.

show more ...

xennetback(4): Fix xennetback_evthandler loop.

- After observing the other side has produced pending tx requests by
reading sring->req_prod, must issue xen_rmb before touching them.

Despite all

xennetback(4): Fix xennetback_evthandler loop.

- After observing the other side has produced pending tx requests by
reading sring->req_prod, must issue xen_rmb before touching them.

Despite all the effort to use the heavy-weight
RING_FINAL_CHECK_FOR_REQUESTS on each request in the loop, this
barrier was missing.

- No need to update req_cons at each iteration in the loop. It's
private. Just update it once at the end.

- After consuming requests, must issue xen_wmb before releasing the
slots with RING_FINAL_CHECK_FOR_REQUEST for the other side to
reuse.

XXX pullup-8 (requires patch; at least add xen_rmb between
RING_FINAL_CHECK_FOR_REQUESTS and RING_COPY_REQUEST)
XXX pullup-9 (requires patch; at least add xen_rmb between
RING_FINAL_CHECK_FOR_REQUESTS and RING_COPY_REQUEST)
XXX pullup-10

show more ...

Remove unnecessary inclusion of <net/netisr.h>.

in backend drivers, use xen_shm_(un)map for the rings instead of inline,
mostly duplicate code.

Add PVH support for backend drivers grant operation.
Now a domU in a PVH dom0 boots multiuser.

Make DOM0OPS build for PVH/PVHVM too

remove IPv4 csum offloading for xennet(4) - it's not complete, and even
if it was, it doesn't work with Linux Dom0 as it expects the IPv4 csum present

add support for scatter-gather also for frontend Rx path (backend -> frontend)

enable ETHERCAP_JUMBO_MTU and feature-sg

fix gref offset when setting up copy of multi-fragment packet so data
for second and further fragments gets copied into correct place

destroy dma maps in xennetback_xenbus_destroy() to avoid memory leak
when xvif(4) is destroyed

make the csum blank/undefer counters per interface

add support for scatter-gather when accepting packets on frontend Tx path
(frontend -> backend)

don't enable ETHERCAP_JUMBO_MTU nor feature-sg yet, need to implement
support also for the frontend Rx

add support for scatter-gather when accepting packets on frontend Tx path
(frontend -> backend)

don't enable ETHERCAP_JUMBO_MTU nor feature-sg yet, need to implement
support also for the frontend Rx side

show more ...

bump send queue to 2*NET_TX_RING_SIZE to make it less likely packets
are dropped on load

m_defrag() short packets before calling bus_dmamap_load_mbuf() -
if it's fragmented load_mbuf would fail anyw

bump send queue to 2*NET_TX_RING_SIZE to make it less likely packets
are dropped on load

m_defrag() short packets before calling bus_dmamap_load_mbuf() -
if it's fragmented load_mbuf would fail anyway, and even with
eventual feature-sg support it's way faster to pass the short packet
in single fragment

show more ...

use m_defrag() instead of local code now that it returns single mbuf

convert to bus_dma(9), no explicit xpmap_*() calls any more

as part of this move some global arrays into struct xnetback_instance,
and fix race for xnetif_lookup()

remove check for matching ethernet address on Tx - the higher levels do this
check anyway, and the check did not handle VLANs

change xenbus_read() interface so that caller supplies the buffer and it's
size, caller doesn't free(9) the returned value any more

pass and use feature-ipv6-csum-offload for ipv6 csum support, matches
Linux Dom0/DomU

remove support for legacy rx-flip mode for xennet(4)/xvif(4), making
rx-copy (first shipped in NetBSD 6.0 in 2012) the only supported
mode

this is mostly to simplify maintenance and future developme

remove support for legacy rx-flip mode for xennet(4)/xvif(4), making
rx-copy (first shipped in NetBSD 6.0 in 2012) the only supported
mode

this is mostly to simplify maintenance and future development

rx-flip is not supported by Linux Dom0/DomU, and NetBSD Dom0/DomU
defaults to rx-copy for over 8 years now too, so there is little
need to keep the support for compatibility

besides compatibility there is no other reason to keep rx-flip -
page transfer is generally slower than copy due to necessary MMU/TLB
manipulation, especially on MP systems

show more ...

enable Rx checkum offload (i.e. skip) for xen netback; it works well
as long as netfront in DomU is configured to fill the checksum

documentation will be updated to recommend how the flags should be

enable Rx checkum offload (i.e. skip) for xen netback; it works well
as long as netfront in DomU is configured to fill the checksum

documentation will be updated to recommend how the flags should be setup
on each side of the xennet(4)/xvif(4) pair

show more ...

don't check for IFF_OACTIVE, nothing sets it

batch the GNTTABOP_copy hypervisor calls in xvif(4) when processing
DomU requests in xennetback_evthandler()

some 25% speed increase observed over the single-copy version

completely g/c the #if 0'ed xennetback_tx_free(), it's not used any more