from the FreeBSD network code. The flag is still kept around in the
"sys/mbuf.h" header file, but does no longer have any users. Instead
the "m_pkthdr.rsstype" field in the mbuf structure is now used to
decide the meaning of the "m_pkthdr.flowid" field. To modify the
"m_pkthdr.rsstype" field please use the existing "M_HASHTYPE_XXX"
macros as defined in the "sys/mbuf.h" header file.
This patch introduces new behaviour in the transmit direction.
Previously network drivers checked if "M_FLOWID" was set in "m_flags"
before using the "m_pkthdr.flowid" field. This check has now now been
replaced by checking if "M_HASHTYPE_GET(m)" is different from
"M_HASHTYPE_NONE". In the future more hashtypes will be added, for
example hashtypes for hardware dedicated flows.
"M_HASHTYPE_OPAQUE" indicates that the "m_pkthdr.flowid" value is
valid and has no particular type. This change removes the need for an
"if" statement in TCP transmit code checking for the presence of a
valid flowid value. The "if" statement mentioned above is now a direct
variable assignment which is then later checked by the respective
network drivers like before.
Additional notes:
- The SCTP code changes will be committed as a separate patch.
- Removal of the "M_FLOWID" flag will also be done separately.
- The FreeBSD version has been bumped.
MFC after: 1 month
Sponsored by: Mellanox Technologies
and egress.
* rss_mbuf_software_hash_v4 - look at the IPv4 mbuf to fetch the IPv4 details
+ direction to calculate a hash.
* rss_proto_software_hash_v4 - hash the given source/destination IPv4 address,
port and direction.
* rss_soft_m2cpuid - map the given mbuf to an RSS CPU ("bucket" for now)
These functions are intended to be used by the stack to support
the following:
* Not all NICs do RSS hashing, so we should support some way of doing
a hash in software;
* The NIC / driver may not hash frames the way we want (eg UDP 4-tuple
hashing when the stack is only doing 2-tuple hashing for UDP); so we
may need to re-hash frames;
* .. same with IPv4 fragments - they will need to be re-hashed after
reassembly;
* .. and same with things like IP tunneling and such;
* The transmit path for things like UDP, RAW and ICMP don't currently
have any RSS information attached to them - so they'll need an
RSS calculation performed before transmit.
TODO:
* Counters! Everywhere!
* Add a debug mode that software hashes received frames and compares them
to the hardware hash provided by the hardware to ensure they match.
The IPv6 part of this is missing - I'm going to do some re-juggling of
where various parts of the RSS framework live before I add the IPv6
code (read: the IPv6 code is going to go into netinet6/in6_rss.[ch],
rather than living here.)
Note: This API is still fluid. Please keep that in mind.
Differential Revision: https://reviews.freebsd.org/D527
Reviewed by: grehan
The rss_key[] array in netinet/in_rss.c has the bytes in incorrect
order. This results in the RSS test vectors in the Microsft RSS spec
and Intel NIC specs giving incorrect results, and making it difficult
to verify correct hash operation when RSS functionality is added to
new NICs.
CR: https://phabric.freebsd.org/D516
Reviewed by: adrian
by the stack.
Right now the stack isn't really setup for RSS with 4-tuple UDP hashing
for either IPv4 and IPv6.
The specifics:
* The UDP init path udp_init() and udplite_init() specify the hash as
2-tuple, so the PCBGROUPS code only tries a 2-tuple check;
* The PCBGROUPS and RSS code doesn't know about the UDP hash types
just yet, so they're never treated as valid hashes.
* For correctness, 4-tuple can't be enabled in the general case because
UDP datagrams can be more fragmented than IP datagrams may be.
Strictly speaking, TCP datagrams may also be fragmented and this could
cause issues with PCBGROUPS/RSS until the IP defragment path grows some
code to re-calculate the RSS hash.
I'll follow this commit up with awareness of the UDP 4-tuple for those
who wish to configure it, but for now it'll stay disabled.
No drivers (yet) know to use this function when RSS is enabled.
markedly better distribution of IPv6 address/ports than the previous key.
The previous key would hash large swaths of the port space for a given
source/destination IP address to the same low handful of bits, effectively
mapping them to the same queue. This made testing very .. special.
These changes prevent sysctl(8) from returning proper output,
such as:
1) no output from sysctl(8)
2) erroneously returning ENOMEM with tools like truss(1)
or uname(1)
truss: can not get etype: Cannot allocate memory
there is an environment variable which shall initialize the SYSCTL
during early boot. This works for all SYSCTL types both statically and
dynamically created ones, except for the SYSCTL NODE type and SYSCTLs
which belong to VNETs. A new flag, CTLFLAG_NOFETCH, has been added to
be used in the case a tunable sysctl has a custom initialisation
function allowing the sysctl to still be marked as a tunable. The
kernel SYSCTL API is mostly the same, with a few exceptions for some
special operations like iterating childrens of a static/extern SYSCTL
node. This operation should probably be made into a factored out
common macro, hence some device drivers use this. The reason for
changing the SYSCTL API was the need for a SYSCTL parent OID pointer
and not only the SYSCTL parent OID list pointer in order to quickly
generate the sysctl path. The motivation behind this patch is to avoid
parameter loading cludges inside the OFED driver subsystem. Instead of
adding special code to the OFED driver subsystem to post-load tunables
into dynamically created sysctls, we generalize this in the kernel.
Other changes:
- Corrected a possibly incorrect sysctl name from "hw.cbb.intr_mask"
to "hw.pcic.intr_mask".
- Removed redundant TUNABLE statements throughout the kernel.
- Some minor code rewrites in connection to removing not needed
TUNABLE statements.
- Added a missing SYSCTL_DECL().
- Wrapped two very long lines.
- Avoid malloc()/free() inside sysctl string handling, in case it is
called to initialize a sysctl from a tunable, hence malloc()/free() is
not ready when sysctls from the sysctl dataset are registered.
- Bumped FreeBSD version to indicate SYSCTL API change.
MFC after: 2 weeks
Sponsored by: Mellanox Technologies
There's 128 indirection table entries which correspond to the
low 7 bits of the 32 bit RSS hash. Each value will correspond
to an RSS bucket. (Then each RSS bucket currently will map
to a CPU.)
This is a more explicit way of figuring out which RSS bucket
is in each RSS indirection slot. It can be inferred by the other
methods but I'd rather drivers use something more simplified and
explicit.
mappings. Instead, they should be first mapping to an RSS bucket and
then querying the RSS bucket -> CPU ID mapping to figure out the target
CPU.
When (if?) RSS rebalancing is implemented or some other (non round-robin)
distribution of work from buckets to CPU IDs, various bits of code - both
userland and kernel - will need to know how this mapping works.
So, to support this:
* Add a new function rss_m2bucket() - this maps an mbuf to a given bucket.
Anything which is currently doing hash -> CPU work may instead wish to
do hash -> bucket, and then query the bucket->cpuid map for which
CPU it belongs on. Or, map it to a bucket, then re-pin that bucket ->
CPU during a rebalance operation.
* For userland applications which wish to exploit affinity to RSS buckets,
the bucket -> CPU ID mapping is now available via a sysctl.
net.inet.rss.bucket_mapping lists the bucket to CPU ID mapping via
a list of bucket:cpu pairs.
This is intended to be used by various places that wish to hash some
information about a TCP/UDP/IP flow but don't necessarily have a
live mbuf to do it with.
Refactor rss_m2cpuid() to use the refactored function.
linking NIC Receive Side Scaling (RSS) to the network stack's
connection-group implementation. This prototype (and derived patches)
are in use at Juniper and several other FreeBSD-using companies, so
despite some reservations about its maturity, merge the patch to the
base tree so that it can be iteratively refined in collaboration rather
than maintained as a set of gradually diverging patch sets.
(1) Merge a software implementation of the Toeplitz hash specified in
RSS implemented by David Malone. This is used to allow suitable
pcbgroup placement of connections before the first packet is
received from the NIC. Software hashing is generally avoided,
however, due to high cost of the hash on general-purpose CPUs.
(2) In in_rss.c, maintain authoritative versions of RSS state intended
to be pushed to each NIC, including keying material, hash
algorithm/ configuration, and buckets. Provide software-facing
interfaces to hash 2- and 4-tuples for IPv4 and IPv6 using both
the RSS standardised Toeplitz and a 'naive' variation with a hash
efficient in software but with poor distribution properties.
Implement rss_m2cpuid()to be used by netisr and other load
balancing code to look up the CPU on which an mbuf should be
processed.
(3) In the Ethernet link layer, allow netisr distribution using RSS as
a source of policy as an alternative to source ordering; continue
to default to direct dispatch (i.e., don't try and requeue packets
for processing on the 'right' CPU if they arrive in a directly
dispatchable context).
(4) Allow RSS to control tuning of connection groups in order to align
groups with RSS buckets. If a packet arrives on a protocol using
connection groups, and contains a suitable hardware-generated
hash, use that hash value to select the connection group for pcb
lookup for both IPv4 and IPv6. If no hardware-generated Toeplitz
hash is available, we fall back on regular PCB lookup risking
contention rather than pay the cost of Toeplitz in software --
this is a less scalable but, at my last measurement, faster
approach. As core counts go up, we may want to revise this
strategy despite CPU overhead.
Where device drivers suitably configure NICs, and connection groups /
RSS are enabled, this should avoid both lock and line contention during
connection lookup for TCP. This commit does not modify any device
drivers to tune device RSS configuration to the global RSS
configuration; patches are in circulation to do this for at least
Chelsio T3 and Intel 1G/10G drivers. Currently, the KPI for device
drivers is not particularly robust, nor aware of more advanced features
such as runtime reconfiguration/rebalancing. This will hopefully prove
a useful starting point for refinement.
No MFC is scheduled as we will first want to nail down a more mature
and maintainable KPI/KBI for device drivers.
Sponsored by: Juniper Networks (original work)
Sponsored by: EMC/Isilon (patch update and merge)
