d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
e04e947299
freebsd-skq/sys/dev/netmap/netmap_generic.c

866 lines
24 KiB
C
Raw Normal View History

split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
/*
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
* Copyright (C) 2013-2014 Universita` di Pisa. All rights reserved.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*
* This module implements netmap support on top of standard,
* unmodified device drivers.
*
* A NIOCREGIF request is handled here if the device does not
* have native support. TX and RX rings are emulated as follows:
*
* NIOCREGIF
* We preallocate a block of TX mbufs (roughly as many as
* tx descriptors; the number is not critical) to speed up
* operation during transmissions. The refcount on most of
* these buffers is artificially bumped up so we can recycle
* them more easily. Also, the destructor is intercepted
* so we use it as an interrupt notification to wake up
* processes blocked on a poll().
*
* For each receive ring we allocate one "struct mbq"
* (an mbuf tailq plus a spinlock). We intercept packets
* (through if_input)
* on the receive path and put them in the mbq from which
* netmap receive routines can grab them.
*
* TX:
* in the generic_txsync() routine, netmap buffers are copied
* (or linked, in a future) to the preallocated mbufs
* and pushed to the transmit queue. Some of these mbufs
* (those with NS_REPORT, or otherwise every half ring)
* have the refcount=1, others have refcount=2.
* When the destructor is invoked, we take that as
* a notification that all mbufs up to that one in
* the specific ring have been completed, and generate
* the equivalent of a transmit interrupt.
*
* RX:
*
*/
#ifdef __FreeBSD__
#include <sys/cdefs.h> /* prerequisite */
__FBSDID("$FreeBSD$");
#include <sys/types.h>
#include <sys/errno.h>
#include <sys/malloc.h>
#include <sys/lock.h> /* PROT_EXEC */
#include <sys/rwlock.h>
#include <sys/socket.h> /* sockaddrs */
#include <sys/selinfo.h>
#include <net/if.h>
#include <net/if_var.h>
#include <machine/bus.h> /* bus_dmamap_* in netmap_kern.h */
// XXX temporary - D() defined here
#include <net/netmap.h>
#include <dev/netmap/netmap_kern.h>
#include <dev/netmap/netmap_mem2.h>
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
#define rtnl_lock() ND("rtnl_lock called")
#define rtnl_unlock() ND("rtnl_unlock called")
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#define MBUF_TXQ(m) ((m)->m_pkthdr.flowid)
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
#define MBUF_RXQ(m) ((m)->m_pkthdr.flowid)
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#define smp_mb()
/*
remove two debugging messages, align comments with the code in our development trunk
2014-06-06 14:57:16 +00:00
* FreeBSD mbuf allocator/deallocator in emulation mode:
*
* We allocate EXT_PACKET mbuf+clusters, but need to set M_NOFREE
* so that the destructor, if invoked, will not free the packet.
* In principle we should set the destructor only on demand,
* but since there might be a race we better do it on allocation.
* As a consequence, we also need to set the destructor or we
* would leak buffers.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*/
/*
remove two debugging messages, align comments with the code in our development trunk
2014-06-06 14:57:16 +00:00
* mbuf wrappers
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*/
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
/* mbuf destructor, also need to change the type to EXT_EXTREF,
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
* add an M_NOFREE flag, and then clear the flag and
* chain into uma_zfree(zone_pack, mf)
* (or reinstall the buffer ?)
*/
#define SET_MBUF_DESTRUCTOR(m, fn) do { \
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
(m)->m_ext.ext_free = (void *)fn; \
(m)->m_ext.ext_type = EXT_EXTREF; \
} while (0)
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
static void
change the netmap mbuf destructor so the same code works also on FreeBSD 9. For head and 10 this change has no effect, but on stable/9 it would cause panics when using emulated netmap on top of a standard device driver.
2014-06-10 16:06:59 +00:00
netmap_default_mbuf_destructor(struct mbuf *m)
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
{
/* restore original mbuf */
m->m_ext.ext_buf = m->m_data = m->m_ext.ext_arg1;
m->m_ext.ext_arg1 = NULL;
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
m->m_ext.ext_type = EXT_PACKET;
m->m_ext.ext_free = NULL;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
if (GET_MBUF_REFCNT(m) == 0)
SET_MBUF_REFCNT(m, 1);
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
uma_zfree(zone_pack, m);
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
}
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
static inline struct mbuf *
netmap_get_mbuf(int len)
{
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
struct mbuf *m;
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR | M_NOFREE);
if (m) {
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
m->m_ext.ext_arg1 = m->m_ext.ext_buf; // XXX save
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
m->m_ext.ext_free = (void *)netmap_default_mbuf_destructor;
m->m_ext.ext_type = EXT_EXTREF;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
ND(5, "create m %p refcnt %d", m, GET_MBUF_REFCNT(m));
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
}
return m;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#else /* linux */
#include "bsd_glue.h"
#include <linux/rtnetlink.h> /* rtnl_[un]lock() */
#include <linux/ethtool.h> /* struct ethtool_ops, get_ringparam */
#include <linux/hrtimer.h>
//#define REG_RESET
#endif /* linux */
/* Common headers. */
#include <net/netmap.h>
#include <dev/netmap/netmap_kern.h>
#include <dev/netmap/netmap_mem2.h>
/* ======================== usage stats =========================== */
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
#ifdef RATE_GENERIC
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#define IFRATE(x) x
struct rate_stats {
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
unsigned long txpkt;
unsigned long txsync;
unsigned long txirq;
unsigned long rxpkt;
unsigned long rxirq;
unsigned long rxsync;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
};
struct rate_context {
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
unsigned refcount;
struct timer_list timer;
struct rate_stats new;
struct rate_stats old;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
};
#define RATE_PRINTK(_NAME_) \
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
printk( #_NAME_ " = %lu Hz\n", (cur._NAME_ - ctx->old._NAME_)/RATE_PERIOD);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#define RATE_PERIOD 2
static void rate_callback(unsigned long arg)
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct rate_context * ctx = (struct rate_context *)arg;
struct rate_stats cur = ctx->new;
int r;
RATE_PRINTK(txpkt);
RATE_PRINTK(txsync);
RATE_PRINTK(txirq);
RATE_PRINTK(rxpkt);
RATE_PRINTK(rxsync);
RATE_PRINTK(rxirq);
printk("\n");
ctx->old = cur;
r = mod_timer(&ctx->timer, jiffies +
msecs_to_jiffies(RATE_PERIOD * 1000));
if (unlikely(r))
D("[v1000] Error: mod_timer()");
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
static struct rate_context rate_ctx;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
void generic_rate(int txp, int txs, int txi, int rxp, int rxs, int rxi)
{
if (txp) rate_ctx.new.txpkt++;
if (txs) rate_ctx.new.txsync++;
if (txi) rate_ctx.new.txirq++;
if (rxp) rate_ctx.new.rxpkt++;
if (rxs) rate_ctx.new.rxsync++;
if (rxi) rate_ctx.new.rxirq++;
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#else /* !RATE */
#define IFRATE(x)
#endif /* !RATE */
/* =============== GENERIC NETMAP ADAPTER SUPPORT ================= */
/*
* Wrapper used by the generic adapter layer to notify
* the poller threads. Differently from netmap_rx_irq(), we check
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
* only NAF_NETMAP_ON instead of NAF_NATIVE_ON to enable the irq.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*/
static void
netmap_generic_irq(struct ifnet *ifp, u_int q, u_int *work_done)
{
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
struct netmap_adapter *na = NA(ifp);
if (unlikely(!nm_netmap_on(na)))
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
return;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
netmap_common_irq(ifp, q, work_done);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
/* Enable/disable netmap mode for a generic network interface. */
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
static int
generic_netmap_register(struct netmap_adapter *na, int enable)
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct netmap_generic_adapter *gna = (struct netmap_generic_adapter *)na;
struct mbuf *m;
int error;
int i, r;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (!na)
return EINVAL;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#ifdef REG_RESET
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
error = ifp->netdev_ops->ndo_stop(ifp);
if (error) {
return error;
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#endif /* REG_RESET */
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (enable) { /* Enable netmap mode. */
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
/* Init the mitigation support on all the rx queues. */
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
gna->mit = malloc(na->num_rx_rings * sizeof(struct nm_generic_mit),
M_DEVBUF, M_NOWAIT | M_ZERO);
if (!gna->mit) {
D("mitigation allocation failed");
error = ENOMEM;
goto out;
}
for (r=0; r<na->num_rx_rings; r++)
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
netmap_mitigation_init(&gna->mit[r], r, na);
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Initialize the rx queue, as generic_rx_handler() can
* be called as soon as netmap_catch_rx() returns.
*/
for (r=0; r<na->num_rx_rings; r++) {
mbq_safe_init(&na->rx_rings[r].rx_queue);
}
/*
* Preallocate packet buffers for the tx rings.
*/
for (r=0; r<na->num_tx_rings; r++)
na->tx_rings[r].tx_pool = NULL;
for (r=0; r<na->num_tx_rings; r++) {
na->tx_rings[r].tx_pool = malloc(na->num_tx_desc * sizeof(struct mbuf *),
M_DEVBUF, M_NOWAIT | M_ZERO);
if (!na->tx_rings[r].tx_pool) {
D("tx_pool allocation failed");
error = ENOMEM;
goto free_tx_pools;
}
for (i=0; i<na->num_tx_desc; i++)
na->tx_rings[r].tx_pool[i] = NULL;
for (i=0; i<na->num_tx_desc; i++) {
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
m = netmap_get_mbuf(NETMAP_BUF_SIZE(na));
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (!m) {
D("tx_pool[%d] allocation failed", i);
error = ENOMEM;
goto free_tx_pools;
}
na->tx_rings[r].tx_pool[i] = m;
}
}
rtnl_lock();
/* Prepare to intercept incoming traffic. */
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
error = netmap_catch_rx(gna, 1);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (error) {
netmap_user.h: add separate rx/tx ring indexes add ring specifier in nm_open device name netmap.c, netmap_vale.c more consistent errno numbers netmap_generic.c correctly handle failure in registering interfaces. tools/tools/netmap/ massive cleanup of the example programs (a lot of common code is now in netmap_user.h.) nm_util.[ch] are going away soon. pcap.c will also go when i commit the native netmap support for libpcap.
2014-01-16 00:20:42 +00:00
D("netdev_rx_handler_register() failed (%d)", error);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
goto register_handler;
}
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
na->na_flags |= NAF_NETMAP_ON;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Make netmap control the packet steering. */
netmap_catch_tx(gna, 1);
rtnl_unlock();
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
#ifdef RATE_GENERIC
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (rate_ctx.refcount == 0) {
D("setup_timer()");
memset(&rate_ctx, 0, sizeof(rate_ctx));
setup_timer(&rate_ctx.timer, &rate_callback, (unsigned long)&rate_ctx);
if (mod_timer(&rate_ctx.timer, jiffies + msecs_to_jiffies(1500))) {
D("Error: mod_timer()");
}
}
rate_ctx.refcount++;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#endif /* RATE */
netmap_user.h: add separate rx/tx ring indexes add ring specifier in nm_open device name netmap.c, netmap_vale.c more consistent errno numbers netmap_generic.c correctly handle failure in registering interfaces. tools/tools/netmap/ massive cleanup of the example programs (a lot of common code is now in netmap_user.h.) nm_util.[ch] are going away soon. pcap.c will also go when i commit the native netmap support for libpcap.
2014-01-16 00:20:42 +00:00
} else if (na->tx_rings[0].tx_pool) {
/* Disable netmap mode. We enter here only if the previous
generic_netmap_register(na, 1) was successfull.
If it was not, na->tx_rings[0].tx_pool was set to NULL by the
error handling code below. */
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
rtnl_lock();
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
na->na_flags &= ~NAF_NETMAP_ON;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Release packet steering control. */
netmap_catch_tx(gna, 0);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Do not intercept packets on the rx path. */
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
netmap_catch_rx(gna, 0);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
rtnl_unlock();
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Free the mbufs going to the netmap rings */
for (r=0; r<na->num_rx_rings; r++) {
mbq_safe_purge(&na->rx_rings[r].rx_queue);
mbq_safe_destroy(&na->rx_rings[r].rx_queue);
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
for (r=0; r<na->num_rx_rings; r++)
netmap_mitigation_cleanup(&gna->mit[r]);
free(gna->mit, M_DEVBUF);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
for (r=0; r<na->num_tx_rings; r++) {
for (i=0; i<na->num_tx_desc; i++) {
m_freem(na->tx_rings[r].tx_pool[i]);
}
free(na->tx_rings[r].tx_pool, M_DEVBUF);
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
#ifdef RATE_GENERIC
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (--rate_ctx.refcount == 0) {
D("del_timer()");
del_timer(&rate_ctx.timer);
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#endif
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#ifdef REG_RESET
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
error = ifp->netdev_ops->ndo_open(ifp);
if (error) {
netmap_user.h: add separate rx/tx ring indexes add ring specifier in nm_open device name netmap.c, netmap_vale.c more consistent errno numbers netmap_generic.c correctly handle failure in registering interfaces. tools/tools/netmap/ massive cleanup of the example programs (a lot of common code is now in netmap_user.h.) nm_util.[ch] are going away soon. pcap.c will also go when i commit the native netmap support for libpcap.
2014-01-16 00:20:42 +00:00
goto free_tx_pools;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#endif
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
return 0;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
register_handler:
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
rtnl_unlock();
free_tx_pools:
for (r=0; r<na->num_tx_rings; r++) {
if (na->tx_rings[r].tx_pool == NULL)
continue;
for (i=0; i<na->num_tx_desc; i++)
if (na->tx_rings[r].tx_pool[i])
m_freem(na->tx_rings[r].tx_pool[i]);
free(na->tx_rings[r].tx_pool, M_DEVBUF);
netmap_user.h: add separate rx/tx ring indexes add ring specifier in nm_open device name netmap.c, netmap_vale.c more consistent errno numbers netmap_generic.c correctly handle failure in registering interfaces. tools/tools/netmap/ massive cleanup of the example programs (a lot of common code is now in netmap_user.h.) nm_util.[ch] are going away soon. pcap.c will also go when i commit the native netmap support for libpcap.
2014-01-16 00:20:42 +00:00
na->tx_rings[r].tx_pool = NULL;
}
for (r=0; r<na->num_rx_rings; r++) {
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
netmap_mitigation_cleanup(&gna->mit[r]);
netmap_user.h: add separate rx/tx ring indexes add ring specifier in nm_open device name netmap.c, netmap_vale.c more consistent errno numbers netmap_generic.c correctly handle failure in registering interfaces. tools/tools/netmap/ massive cleanup of the example programs (a lot of common code is now in netmap_user.h.) nm_util.[ch] are going away soon. pcap.c will also go when i commit the native netmap support for libpcap.
2014-01-16 00:20:42 +00:00
mbq_safe_destroy(&na->rx_rings[r].rx_queue);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
}
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
free(gna->mit, M_DEVBUF);
out:
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
return error;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
/*
* Callback invoked when the device driver frees an mbuf used
* by netmap to transmit a packet. This usually happens when
* the NIC notifies the driver that transmission is completed.
*/
static void
generic_mbuf_destructor(struct mbuf *m)
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
netmap_generic_irq(MBUF_IFP(m), MBUF_TXQ(m), NULL);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#ifdef __FreeBSD__
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
if (netmap_verbose)
RD(5, "Tx irq (%p) queue %d index %d" , m, MBUF_TXQ(m), (int)(uintptr_t)m->m_ext.ext_arg1);
netmap_default_mbuf_destructor(m);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
#endif /* __FreeBSD__ */
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
IFRATE(rate_ctx.new.txirq++);
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
extern int netmap_adaptive_io;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Record completed transmissions and update hwtail.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
* The oldest tx buffer not yet completed is at nr_hwtail + 1,
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
* nr_hwcur is the first unsent buffer.
*/
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
static u_int
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
generic_netmap_tx_clean(struct netmap_kring *kring)
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
u_int const lim = kring->nkr_num_slots - 1;
u_int nm_i = nm_next(kring->nr_hwtail, lim);
u_int hwcur = kring->nr_hwcur;
u_int n = 0;
struct mbuf **tx_pool = kring->tx_pool;
while (nm_i != hwcur) { /* buffers not completed */
struct mbuf *m = tx_pool[nm_i];
if (unlikely(m == NULL)) {
/* this is done, try to replenish the entry */
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
tx_pool[nm_i] = m = netmap_get_mbuf(NETMAP_BUF_SIZE(kring->na));
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (unlikely(m == NULL)) {
D("mbuf allocation failed, XXX error");
// XXX how do we proceed ? break ?
return -ENOMEM;
}
} else if (GET_MBUF_REFCNT(m) != 1) {
break; /* This mbuf is still busy: its refcnt is 2. */
}
n++;
nm_i = nm_next(nm_i, lim);
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
#if 0 /* rate adaptation */
if (netmap_adaptive_io > 1) {
if (n >= netmap_adaptive_io)
break;
} else if (netmap_adaptive_io) {
/* if hwcur - nm_i < lim/8 do an early break
* so we prevent the sender from stalling. See CVT.
*/
if (hwcur >= nm_i) {
if (hwcur - nm_i < lim/2)
break;
} else {
if (hwcur + lim + 1 - nm_i < lim/2)
break;
}
}
#endif
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
kring->nr_hwtail = nm_prev(nm_i, lim);
ND("tx completed [%d] -> hwtail %d", n, kring->nr_hwtail);
return n;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
/*
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
* We have pending packets in the driver between nr_hwtail +1 and hwcur.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
* Compute a position in the middle, to be used to generate
* a notification.
*/
static inline u_int
generic_tx_event_middle(struct netmap_kring *kring, u_int hwcur)
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
u_int n = kring->nkr_num_slots;
u_int ntc = nm_next(kring->nr_hwtail, n-1);
u_int e;
if (hwcur >= ntc) {
e = (hwcur + ntc) / 2;
} else { /* wrap around */
e = (hwcur + n + ntc) / 2;
if (e >= n) {
e -= n;
}
}
if (unlikely(e >= n)) {
D("This cannot happen");
e = 0;
}
return e;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
/*
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
* We have pending packets in the driver between nr_hwtail+1 and hwcur.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
* Schedule a notification approximately in the middle of the two.
* There is a race but this is only called within txsync which does
* a double check.
*/
static void
generic_set_tx_event(struct netmap_kring *kring, u_int hwcur)
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct mbuf *m;
u_int e;
if (nm_next(kring->nr_hwtail, kring->nkr_num_slots -1) == hwcur) {
return; /* all buffers are free */
}
e = generic_tx_event_middle(kring, hwcur);
m = kring->tx_pool[e];
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
ND(5, "Request Event at %d mbuf %p refcnt %d", e, m, m ? GET_MBUF_REFCNT(m) : -2 );
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (m == NULL) {
/* This can happen if there is already an event on the netmap
slot 'e': There is nothing to do. */
return;
}
kring->tx_pool[e] = NULL;
SET_MBUF_DESTRUCTOR(m, generic_mbuf_destructor);
// XXX wmb() ?
/* Decrement the refcount an free it if we have the last one. */
m_freem(m);
smp_mb();
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
/*
* generic_netmap_txsync() transforms netmap buffers into mbufs
* and passes them to the standard device driver
* (ndo_start_xmit() or ifp->if_transmit() ).
* On linux this is not done directly, but using dev_queue_xmit(),
* since it implements the TX flow control (and takes some locks).
*/
static int
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
generic_netmap_txsync(struct netmap_kring *kring, int flags)
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
{
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
struct netmap_adapter *na = kring->na;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct ifnet *ifp = na->ifp;
struct netmap_ring *ring = kring->ring;
u_int nm_i; /* index into the netmap ring */ // j
u_int const lim = kring->nkr_num_slots - 1;
u_int const head = kring->rhead;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
u_int ring_nr = kring->ring_id;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
IFRATE(rate_ctx.new.txsync++);
// TODO: handle the case of mbuf allocation failure
rmb();
/*
* First part: process new packets to send.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*/
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
nm_i = kring->nr_hwcur;
if (nm_i != head) { /* we have new packets to send */
while (nm_i != head) {
struct netmap_slot *slot = &ring->slot[nm_i];
u_int len = slot->len;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
void *addr = NMB(na, slot);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* device-specific */
struct mbuf *m;
int tx_ret;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
NM_CHECK_ADDR_LEN(na, addr, len);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Tale a mbuf from the tx pool and copy in the user packet. */
m = kring->tx_pool[nm_i];
if (unlikely(!m)) {
RD(5, "This should never happen");
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
kring->tx_pool[nm_i] = m = netmap_get_mbuf(NETMAP_BUF_SIZE(na));
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (unlikely(m == NULL)) {
D("mbuf allocation failed");
break;
}
}
/* XXX we should ask notifications when NS_REPORT is set,
* or roughly every half frame. We can optimize this
* by lazily requesting notifications only when a
* transmission fails. Probably the best way is to
* break on failures and set notifications when
* ring->cur == ring->tail || nm_i != cur
*/
tx_ret = generic_xmit_frame(ifp, m, addr, len, ring_nr);
if (unlikely(tx_ret)) {
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
ND(5, "start_xmit failed: err %d [nm_i %u, head %u, hwtail %u]",
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
tx_ret, nm_i, head, kring->nr_hwtail);
/*
* No room for this mbuf in the device driver.
* Request a notification FOR A PREVIOUS MBUF,
* then call generic_netmap_tx_clean(kring) to do the
* double check and see if we can free more buffers.
* If there is space continue, else break;
* NOTE: the double check is necessary if the problem
* occurs in the txsync call after selrecord().
* Also, we need some way to tell the caller that not
* all buffers were queued onto the device (this was
* not a problem with native netmap driver where space
* is preallocated). The bridge has a similar problem
* and we solve it there by dropping the excess packets.
*/
generic_set_tx_event(kring, nm_i);
if (generic_netmap_tx_clean(kring)) { /* space now available */
continue;
} else {
break;
}
}
slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED);
nm_i = nm_next(nm_i, lim);
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
IFRATE(rate_ctx.new.txpkt ++);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
}
/* Update hwcur to the next slot to transmit. */
kring->nr_hwcur = nm_i; /* not head, we could break early */
}
/*
* Second, reclaim completed buffers
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*/
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
if (flags & NAF_FORCE_RECLAIM || nm_kr_txempty(kring)) {
/* No more available slots? Set a notification event
* on a netmap slot that will be cleaned in the future.
* No doublecheck is performed, since txsync() will be
* called twice by netmap_poll().
*/
generic_set_tx_event(kring, nm_i);
}
ND("tx #%d, hwtail = %d", n, kring->nr_hwtail);
generic_netmap_tx_clean(kring);
return 0;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
/*
* This handler is registered (through netmap_catch_rx())
* within the attached network interface
* in the RX subsystem, so that every mbuf passed up by
* the driver can be stolen to the network stack.
* Stolen packets are put in a queue where the
* generic_netmap_rxsync() callback can extract them.
*/
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
void
generic_rx_handler(struct ifnet *ifp, struct mbuf *m)
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct netmap_adapter *na = NA(ifp);
struct netmap_generic_adapter *gna = (struct netmap_generic_adapter *)na;
u_int work_done;
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
u_int rr = MBUF_RXQ(m); // receive ring number
if (rr >= na->num_rx_rings) {
rr = rr % na->num_rx_rings; // XXX expensive...
}
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* limit the size of the queue */
if (unlikely(mbq_len(&na->rx_rings[rr].rx_queue) > 1024)) {
m_freem(m);
} else {
mbq_safe_enqueue(&na->rx_rings[rr].rx_queue, m);
}
if (netmap_generic_mit < 32768) {
/* no rx mitigation, pass notification up */
netmap_generic_irq(na->ifp, rr, &work_done);
IFRATE(rate_ctx.new.rxirq++);
} else {
/* same as send combining, filter notification if there is a
* pending timer, otherwise pass it up and start a timer.
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
if (likely(netmap_mitigation_active(&gna->mit[rr]))) {
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
/* Record that there is some pending work. */
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
gna->mit[rr].mit_pending = 1;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
} else {
netmap_generic_irq(na->ifp, rr, &work_done);
IFRATE(rate_ctx.new.rxirq++);
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
netmap_mitigation_start(&gna->mit[rr]);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
}
}
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
/*
* generic_netmap_rxsync() extracts mbufs from the queue filled by
* generic_netmap_rx_handler() and puts their content in the netmap
* receive ring.
* Access must be protected because the rx handler is asynchronous,
*/
static int
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
generic_netmap_rxsync(struct netmap_kring *kring, int flags)
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct netmap_ring *ring = kring->ring;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
struct netmap_adapter *na = kring->na;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
u_int nm_i; /* index into the netmap ring */ //j,
u_int n;
u_int const lim = kring->nkr_num_slots - 1;
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
u_int const head = kring->rhead;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR;
if (head > lim)
return netmap_ring_reinit(kring);
/*
* First part: import newly received packets.
*/
if (netmap_no_pendintr || force_update) {
/* extract buffers from the rx queue, stop at most one
* slot before nr_hwcur (stop_i)
*/
uint16_t slot_flags = kring->nkr_slot_flags;
u_int stop_i = nm_prev(kring->nr_hwcur, lim);
nm_i = kring->nr_hwtail; /* first empty slot in the receive ring */
for (n = 0; nm_i != stop_i; n++) {
int len;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
void *addr = NMB(na, &ring->slot[nm_i]);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct mbuf *m;
/* we only check the address here on generic rx rings */
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
if (addr == NETMAP_BUF_BASE(na)) { /* Bad buffer */
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
return netmap_ring_reinit(kring);
}
/*
* Call the locked version of the function.
* XXX Ideally we could grab a batch of mbufs at once
* and save some locking overhead.
*/
m = mbq_safe_dequeue(&kring->rx_queue);
if (!m) /* no more data */
break;
len = MBUF_LEN(m);
m_copydata(m, 0, len, addr);
ring->slot[nm_i].len = len;
ring->slot[nm_i].flags = slot_flags;
m_freem(m);
nm_i = nm_next(nm_i, lim);
}
if (n) {
kring->nr_hwtail = nm_i;
IFRATE(rate_ctx.new.rxpkt += n);
}
kring->nr_kflags &= ~NKR_PENDINTR;
}
// XXX should we invert the order ?
/*
* Second part: skip past packets that userspace has released.
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
*/
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
nm_i = kring->nr_hwcur;
if (nm_i != head) {
/* Userspace has released some packets. */
for (n = 0; nm_i != head; n++) {
struct netmap_slot *slot = &ring->slot[nm_i];
slot->flags &= ~NS_BUF_CHANGED;
nm_i = nm_next(nm_i, lim);
}
kring->nr_hwcur = head;
}
IFRATE(rate_ctx.new.rxsync++);
return 0;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
static void
generic_netmap_dtor(struct netmap_adapter *na)
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct netmap_generic_adapter *gna = (struct netmap_generic_adapter*)na;
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
struct ifnet *ifp = netmap_generic_getifp(gna);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct netmap_adapter *prev_na = gna->prev;
if (prev_na != NULL) {
D("Released generic NA %p", gna);
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
if_rele(ifp);
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
netmap_adapter_put(prev_na);
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
if (na->ifp == NULL) {
/*
* The driver has been removed without releasing
* the reference so we need to do it here.
*/
netmap_adapter_put(prev_na);
}
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
}
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
WNA(ifp) = prev_na;
D("Restored native NA %p", prev_na);
na->ifp = NULL;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
/*
* generic_netmap_attach() makes it possible to use netmap on
* a device without native netmap support.
* This is less performant than native support but potentially
* faster than raw sockets or similar schemes.
*
* In this "emulated" mode, netmap rings do not necessarily
* have the same size as those in the NIC. We use a default
* value and possibly override it if the OS has ways to fetch the
* actual configuration.
*/
int
generic_netmap_attach(struct ifnet *ifp)
{
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
struct netmap_adapter *na;
struct netmap_generic_adapter *gna;
int retval;
u_int num_tx_desc, num_rx_desc;
num_tx_desc = num_rx_desc = netmap_generic_ringsize; /* starting point */
sync a comment with our internal repo
2014-11-10 20:19:58 +00:00
generic_find_num_desc(ifp, &num_tx_desc, &num_rx_desc); /* ignore errors */
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
ND("Netmap ring size: TX = %d, RX = %d", num_tx_desc, num_rx_desc);
better handling of netmap emulation over standard device drivers: plug a potential mbuf leak, and detect bogus drivers that return ENOBUFS even when the packet has been queued. MFC after: 3 days
2014-06-06 18:36:02 +00:00
if (num_tx_desc == 0 || num_rx_desc == 0) {
D("Device has no hw slots (tx %u, rx %u)", num_tx_desc, num_rx_desc);
return EINVAL;
}
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
gna = malloc(sizeof(*gna), M_DEVBUF, M_NOWAIT | M_ZERO);
if (gna == NULL) {
D("no memory on attach, give up");
return ENOMEM;
}
na = (struct netmap_adapter *)gna;
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
strncpy(na->name, ifp->if_xname, sizeof(na->name));
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
na->ifp = ifp;
na->num_tx_desc = num_tx_desc;
na->num_rx_desc = num_rx_desc;
na->nm_register = &generic_netmap_register;
na->nm_txsync = &generic_netmap_txsync;
na->nm_rxsync = &generic_netmap_rxsync;
na->nm_dtor = &generic_netmap_dtor;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
/* when using generic, NAF_NETMAP_ON is set so we force
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
* NAF_SKIP_INTR to use the regular interrupt handler
*/
This new version of netmap brings you the following: - netmap pipes, providing bidirectional blocking I/O while moving 100+ Mpps between processes using shared memory channels (no mistake: over one hundred million. But mind you, i said *moving* not *processing*); - kqueue support (BHyVe needs it); - improved user library. Just the interface name lets you select a NIC, host port, VALE switch port, netmap pipe, and individual queues. The upcoming netmap-enabled libpcap will use this feature. - optional extra buffers associated to netmap ports, for applications that need to buffer data yet don't want to make copies. - segmentation offloading for the VALE switch, useful between VMs. and a number of bug fixes and performance improvements. My colleagues Giuseppe Lettieri and Vincenzo Maffione did a substantial amount of work on these features so we owe them a big thanks. There are some external repositories that can be of interest: https://code.google.com/p/netmap our public repository for netmap/VALE code, including linux versions and other stuff that does not belong here, such as python bindings. https://code.google.com/p/netmap-libpcap a clone of the libpcap repository with netmap support. With this any libpcap client has access to most netmap feature with no recompilation. E.g. tcpdump can filter packets at 10-15 Mpps. https://code.google.com/p/netmap-ipfw a userspace version of ipfw+dummynet which uses netmap to send/receive packets. Speed is up in the 7-10 Mpps range per core for simple rulesets. Both netmap-libpcap and netmap-ipfw will be merged upstream at some point, but while this happens it is useful to have access to them. And yes, this code will be merged soon. It is infinitely better than the version currently in 10 and 9. MFC after: 3 days
2014-02-15 04:53:04 +00:00
na->na_flags = NAF_SKIP_INTR | NAF_HOST_RINGS;
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
ND("[GNA] num_tx_queues(%d), real_num_tx_queues(%d), len(%lu)",
ifp->num_tx_queues, ifp->real_num_tx_queues,
ifp->tx_queue_len);
ND("[GNA] num_rx_queues(%d), real_num_rx_queues(%d)",
ifp->num_rx_queues, ifp->real_num_rx_queues);
generic_find_num_queues(ifp, &na->num_tx_rings, &na->num_rx_rings);
retval = netmap_attach_common(na);
if (retval) {
free(gna, M_DEVBUF);
}
return retval;
split netmap code according to functions: - netmap.c base code - netmap_freebsd.c FreeBSD-specific code - netmap_generic.c emulate netmap over standard drivers - netmap_mbq.c simple mbuf tailq - netmap_mem2.c memory management - netmap_vale.c VALE switch simplify devce-specific code
2013-12-15 08:37:24 +00:00
}
Reference in New Issue Copy Permalink