d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
3e0522bc8f
freebsd-nq/sys/net/netisr.c

1386 lines
40 KiB
C
Raw Normal View History

Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
/*-
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* Copyright (c) 2007-2009 Robert N. M. Watson
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
* Copyright (c) 2010-2011 Juniper Networks, Inc.
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
* All rights reserved.
*
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
* This software was developed by Robert N. M. Watson under contract
* to Juniper Networks, Inc.
*
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +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
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
* notice, this list of conditions and the following disclaimer.
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
* 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.
*
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
* 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.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
* netisr is a packet dispatch service, allowing synchronous (directly
* dispatched) and asynchronous (deferred dispatch) processing of packets by
* registered protocol handlers. Callers pass a protocol identifier and
* packet to netisr, along with a direct dispatch hint, and work will either
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* be immediately processed by the registered handler, or passed to a
* software interrupt (SWI) thread for deferred dispatch. Callers will
* generally select one or the other based on:
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* - Whether directly dispatching a netisr handler lead to code reentrance or
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* lock recursion, such as entering the socket code from the socket code.
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* - Whether directly dispatching a netisr handler lead to recursive
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* processing, such as when decapsulating several wrapped layers of tunnel
* information (IPSEC within IPSEC within ...).
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
*
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* Maintaining ordering for protocol streams is a critical design concern.
* Enforcing ordering limits the opportunity for concurrency, but maintains
* the strong ordering requirements found in some protocols, such as TCP. Of
* related concern is CPU affinity--it is desirable to process all data
* associated with a particular stream on the same CPU over time in order to
* avoid acquiring locks associated with the connection on different CPUs,
* keep connection data in one cache, and to generally encourage associated
* user threads to live on the same CPU as the stream. It's also desirable
* to avoid lock migration and contention where locks are associated with
* more than one flow.
*
* netisr supports several policy variations, represented by the
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* NETISR_POLICY_* constants, allowing protocols to play various roles in
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* identifying flows, assigning work to CPUs, etc. These are described in
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* netisr.h.
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
*/
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#include "opt_ddb.h"
- Don't pollute opt_global.h with DEVICE_POLLING and introduce opt_device_polling.h - Include opt_device_polling.h into appropriate files. - Embrace with HAVE_KERNEL_OPTION_HEADERS the include in the files that can be compiled as loadable modules. Reviewed by: bde
2005-10-05 10:09:17 +00:00
#include "opt_device_polling.h"
Change the default disposition of debug.mpsafenet from 0 to 1, which will cause the network stack to operate without the Giant lock by default. This change has the potential to improve performance by increasing parallelism and decreasing latency in network processing. Due to the potential exposure of existing or new bugs, the following compatibility functionality is maintained: - It is still possible to disable Giant-free operation by setting debug.mpsafenet to 0 in loader.conf. - Add "options NET_WITH_GIANT", which will restore the default value of debug.mpsafenet to 0, and is intended for use on systems compiled with known unsafe components, or where a more conservative configuration is desired. - Add a new declaration, NET_NEEDS_GIANT("componentname"), which permits kernel components to declare dependence on Giant over the network stack. If the declaration is made by a preloaded module or a compiled in component, the disposition of debug.mpsafenet will be set to 0 and a warning concerning performance degraded operation printed to the console. If it is declared by a loadable kernel module after boot, a warning is displayed but the disposition cannot be changed. This is implemented by defining a new SYSINIT() value, SI_SUB_SETTINGS, which is intended for the processing of configuration choices after tunables are read in and the console is available to generate errors, but before much else gets going. This compatibility behavior will go away when we've finished the last of the locking work and are confident that operation is correct.
2004-08-28 15:11:13 +00:00
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
#include <sys/param.h>
#include <sys/bus.h>
#include <sys/kernel.h>
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
#include <sys/kthread.h>
These files were getting sys/malloc.h and vm/uma.h with header pollution via sys/mbuf.h
2016-02-01 17:41:21 +00:00
#include <sys/malloc.h>
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#include <sys/interrupt.h>
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
#include <sys/lock.h>
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#include <sys/mbuf.h>
#include <sys/mutex.h>
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
#include <sys/pcpu.h>
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
#include <sys/proc.h>
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#include <sys/rmlock.h>
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/socket.h>
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
#include <sys/sysctl.h>
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#include <sys/systm.h>
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#ifdef DDB
#include <ddb/ddb.h>
#endif
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
#define _WANT_NETISR_INTERNAL /* Enable definitions from netisr_internal.h */
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
#include <net/if.h>
#include <net/if_var.h>
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
#include <net/netisr.h>
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
#include <net/netisr_internal.h>
Merge the remainder of kern_vimage.c and vimage.h into vnet.c and vnet.h, we now use jails (rather than vimages) as the abstraction for virtualization management, and what remained was specific to virtual network stacks. Minor cleanups are done in the process, and comments updated to reflect these changes. Reviewed by: bz Approved by: re (vimage blanket)
2009-08-01 19:26:27 +00:00
#include <net/vnet.h>
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*-
* Synchronize use and modification of the registered netisr data structures;
* acquire a read lock while modifying the set of registered protocols to
* prevent partially registered or unregistered protocols from being run.
*
* The following data structures and fields are protected by this lock:
*
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
* - The netisr_proto array, including all fields of struct netisr_proto.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* - The nws array, including all fields of struct netisr_worker.
* - The nws_array array.
*
* Note: the NETISR_LOCKING define controls whether read locks are acquired
* in packet processing paths requiring netisr registration stability. This
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* is disabled by default as it can lead to measurable performance
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* degradation even with rmlocks (3%-6% for loopback ping-pong traffic), and
* because netisr registration and unregistration is extremely rare at
* runtime. If it becomes more common, this decision should be revisited.
*
* XXXRW: rmlocks don't support assertions.
*/
static struct rmlock netisr_rmlock;
#define NETISR_LOCK_INIT() rm_init_flags(&netisr_rmlock, "netisr", \
RM_NOWITNESS)
#define NETISR_LOCK_ASSERT()
#define NETISR_RLOCK(tracker) rm_rlock(&netisr_rmlock, (tracker))
#define NETISR_RUNLOCK(tracker) rm_runlock(&netisr_rmlock, (tracker))
#define NETISR_WLOCK() rm_wlock(&netisr_rmlock)
#define NETISR_WUNLOCK() rm_wunlock(&netisr_rmlock)
/* #define NETISR_LOCKING */
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
static SYSCTL_NODE(_net, OID_AUTO, isr, CTLFLAG_RW, 0, "netisr");
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*-
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
* Three global direct dispatch policies are supported:
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*
Fix macro name in comment.
2013-11-26 15:23:56 +00:00
* NETISR_DISPATCH_DEFERRED: All work is deferred for a netisr, regardless of
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
* context (may be overriden by protocols).
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
* NETISR_DISPATCH_HYBRID: If the executing context allows direct dispatch,
* and we're running on the CPU the work would be performed on, then direct
* dispatch it if it wouldn't violate ordering constraints on the workstream.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
* NETISR_DISPATCH_DIRECT: If the executing context allows direct dispatch,
* always direct dispatch. (The default.)
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*
* Notice that changing the global policy could lead to short periods of
* misordered processing, but this is considered acceptable as compared to
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
* the complexity of enforcing ordering during policy changes. Protocols can
* override the global policy (when they're not doing that, they select
* NETISR_DISPATCH_DEFAULT).
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
#define NETISR_DISPATCH_POLICY_DEFAULT NETISR_DISPATCH_DIRECT
#define NETISR_DISPATCH_POLICY_MAXSTR 20 /* Used for temporary buffers. */
static u_int netisr_dispatch_policy = NETISR_DISPATCH_POLICY_DEFAULT;
static int sysctl_netisr_dispatch_policy(SYSCTL_HANDLER_ARGS);
Pull in r267961 and r267973 again. Fix for issues reported will follow.
2014-06-28 03:56:17 +00:00
SYSCTL_PROC(_net_isr, OID_AUTO, dispatch, CTLTYPE_STRING | CTLFLAG_RWTUN,
0, 0, sysctl_netisr_dispatch_policy, "A",
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
"netisr dispatch policy");
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* Allow the administrator to limit the number of threads (CPUs) to use for
* netisr. We don't check netisr_maxthreads before creating the thread for
Currently there is no easy way to specify net.isr.maxthreads = all cpus. We need to specify exact number of cpus in loader.conf which get annoying when you have mix of machines which don't have equal number of total cpus. I propose "-1" as that value. When loader.conf has net.isr.maxthreads = -1, netisr will use all available cpus. In collaboration with: davide Reviewed by: gnn Differential Revision: https://reviews.freebsd.org/D2318 MFC after: 2 weeks Sponsored by: Limelight Networks
2015-04-25 16:12:06 +00:00
* CPU 0. This must be set at boot. We will create at most one thread per CPU.
* By default we initialize this to 1 which would assign just 1 cpu (cpu0) and
* therefore only 1 workstream. If set to -1, netisr would use all cpus
* (mp_ncpus) and therefore would have those many workstreams. One workstream
* per thread (CPU).
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
Currently there is no easy way to specify net.isr.maxthreads = all cpus. We need to specify exact number of cpus in loader.conf which get annoying when you have mix of machines which don't have equal number of total cpus. I propose "-1" as that value. When loader.conf has net.isr.maxthreads = -1, netisr will use all available cpus. In collaboration with: davide Reviewed by: gnn Differential Revision: https://reviews.freebsd.org/D2318 MFC after: 2 weeks Sponsored by: Limelight Networks
2015-04-25 16:12:06 +00:00
static int netisr_maxthreads = 1; /* Max number of threads. */
Mark various sysctls also as tunables. Reviewed by: rwatson MFC after: 1 week
2010-02-15 09:19:07 +00:00
SYSCTL_INT(_net_isr, OID_AUTO, maxthreads, CTLFLAG_RDTUN,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
&netisr_maxthreads, 0,
"Use at most this many CPUs for netisr processing");
static int netisr_bindthreads = 0; /* Bind threads to CPUs. */
Mark various sysctls also as tunables. Reviewed by: rwatson MFC after: 1 week
2010-02-15 09:19:07 +00:00
SYSCTL_INT(_net_isr, OID_AUTO, bindthreads, CTLFLAG_RDTUN,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
&netisr_bindthreads, 0, "Bind netisr threads to CPUs.");
/*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* Limit per-workstream mbuf queue limits s to at most net.isr.maxqlimit,
* both for initial configuration and later modification using
* netisr_setqlimit().
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
#define NETISR_DEFAULT_MAXQLIMIT 10240
static u_int netisr_maxqlimit = NETISR_DEFAULT_MAXQLIMIT;
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the net* piece.
2011-01-12 19:53:50 +00:00
SYSCTL_UINT(_net_isr, OID_AUTO, maxqlimit, CTLFLAG_RDTUN,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
&netisr_maxqlimit, 0,
"Maximum netisr per-protocol, per-CPU queue depth.");
/*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* The default per-workstream mbuf queue limit for protocols that don't
* initialize the nh_qlimit field of their struct netisr_handler. If this is
* set above netisr_maxqlimit, we truncate it to the maximum during boot.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
#define NETISR_DEFAULT_DEFAULTQLIMIT 256
static u_int netisr_defaultqlimit = NETISR_DEFAULT_DEFAULTQLIMIT;
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the net* piece.
2011-01-12 19:53:50 +00:00
SYSCTL_UINT(_net_isr, OID_AUTO, defaultqlimit, CTLFLAG_RDTUN,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
&netisr_defaultqlimit, 0,
"Default netisr per-protocol, per-CPU queue limit if not set by protocol");
/*
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
* Store and export the compile-time constant NETISR_MAXPROT limit on the
* number of protocols that can register with netisr at a time. This is
* required for crashdump analysis, as it sizes netisr_proto[].
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
static u_int netisr_maxprot = NETISR_MAXPROT;
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the net* piece.
2011-01-12 19:53:50 +00:00
SYSCTL_UINT(_net_isr, OID_AUTO, maxprot, CTLFLAG_RD,
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
&netisr_maxprot, 0,
"Compile-time limit on the number of protocols supported by netisr.");
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
* The netisr_proto array describes all registered protocols, indexed by
* protocol number. See netisr_internal.h for more details.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
static struct netisr_proto netisr_proto[NETISR_MAXPROT];
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
* Per-CPU workstream data. See netisr_internal.h for more details.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
DPCPU_DEFINE(struct netisr_workstream, nws);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* Map contiguous values between 0 and nws_count into CPU IDs appropriate for
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
* accessing workstreams. This allows constructions of the form
* DPCPU_ID_GET(nws_array[arbitraryvalue % nws_count], nws).
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
static u_int nws_array[MAXCPU];
/*
* Number of registered workstreams. Will be at most the number of running
* CPUs once fully started.
*/
static u_int nws_count;
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the net* piece.
2011-01-12 19:53:50 +00:00
SYSCTL_UINT(_net_isr, OID_AUTO, numthreads, CTLFLAG_RD,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
&nws_count, 0, "Number of extant netisr threads.");
/*
* Synchronization for each workstream: a mutex protects all mutable fields
* in each stream, including per-protocol state (mbuf queues). The SWI is
* woken up if asynchronous dispatch is required.
*/
#define NWS_LOCK(s) mtx_lock(&(s)->nws_mtx)
#define NWS_LOCK_ASSERT(s) mtx_assert(&(s)->nws_mtx, MA_OWNED)
#define NWS_UNLOCK(s) mtx_unlock(&(s)->nws_mtx)
#define NWS_SIGNAL(s) swi_sched((s)->nws_swi_cookie, 0)
/*
* Utility routines for protocols that implement their own mapping of flows
* to CPUs.
*/
u_int
netisr_get_cpucount(void)
{
return (nws_count);
}
u_int
netisr_get_cpuid(u_int cpunumber)
{
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
KASSERT(cpunumber < nws_count, ("%s: %u > %u", __func__, cpunumber,
nws_count));
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
return (nws_array[cpunumber]);
}
/*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* The default implementation of flow -> CPU ID mapping.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*
* Non-static so that protocols can use it to map their own work to specific
* CPUs in a manner consistent to netisr for affinity purposes.
*/
u_int
netisr_default_flow2cpu(u_int flowid)
{
return (nws_array[flowid % nws_count]);
}
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
/*
* Dispatch tunable and sysctl configuration.
*/
struct netisr_dispatch_table_entry {
u_int ndte_policy;
const char *ndte_policy_str;
};
static const struct netisr_dispatch_table_entry netisr_dispatch_table[] = {
{ NETISR_DISPATCH_DEFAULT, "default" },
{ NETISR_DISPATCH_DEFERRED, "deferred" },
{ NETISR_DISPATCH_HYBRID, "hybrid" },
{ NETISR_DISPATCH_DIRECT, "direct" },
};
static void
netisr_dispatch_policy_to_str(u_int dispatch_policy, char *buffer,
u_int buflen)
{
const struct netisr_dispatch_table_entry *ndtep;
const char *str;
u_int i;
str = "unknown";
Remove slightly used const values that can be replaced with nitems(). Suggested by: jhb
2016-04-21 15:38:28 +00:00
for (i = 0; i < nitems(netisr_dispatch_table); i++) {
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
ndtep = &netisr_dispatch_table[i];
if (ndtep->ndte_policy == dispatch_policy) {
str = ndtep->ndte_policy_str;
break;
}
}
snprintf(buffer, buflen, "%s", str);
}
static int
netisr_dispatch_policy_from_str(const char *str, u_int *dispatch_policyp)
{
const struct netisr_dispatch_table_entry *ndtep;
u_int i;
Remove slightly used const values that can be replaced with nitems(). Suggested by: jhb
2016-04-21 15:38:28 +00:00
for (i = 0; i < nitems(netisr_dispatch_table); i++) {
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
ndtep = &netisr_dispatch_table[i];
if (strcmp(ndtep->ndte_policy_str, str) == 0) {
*dispatch_policyp = ndtep->ndte_policy;
return (0);
}
}
return (EINVAL);
}
static int
sysctl_netisr_dispatch_policy(SYSCTL_HANDLER_ARGS)
{
char tmp[NETISR_DISPATCH_POLICY_MAXSTR];
u_int dispatch_policy;
int error;
netisr_dispatch_policy_to_str(netisr_dispatch_policy, tmp,
sizeof(tmp));
error = sysctl_handle_string(oidp, tmp, sizeof(tmp), req);
if (error == 0 && req->newptr != NULL) {
error = netisr_dispatch_policy_from_str(tmp,
&dispatch_policy);
if (error == 0 && dispatch_policy == NETISR_DISPATCH_DEFAULT)
error = EINVAL;
Retire netisr.netisr_direct and netisr.netisr_direct_force sysctls. These were used to control/export dispatch policy but they're not anymore. This commit cannot be MFC'ed to 9 because old netstat(9) binary relies on such sysctl to work. On the other hand, there's no real reason to keep'em around in 10.
2013-09-06 21:02:43 +00:00
if (error == 0)
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
netisr_dispatch_policy = dispatch_policy;
}
return (error);
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* Register a new netisr handler, which requires initializing per-protocol
* fields for each workstream. All netisr work is briefly suspended while
* the protocol is installed.
*/
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
void
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_register(const struct netisr_handler *nhp)
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
{
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
struct netisr_work *npwp;
const char *name;
u_int i, proto;
proto = nhp->nh_proto;
name = nhp->nh_name;
/*
* Test that the requested registration is valid.
*/
KASSERT(nhp->nh_name != NULL,
("%s: nh_name NULL for %u", __func__, proto));
KASSERT(nhp->nh_handler != NULL,
("%s: nh_handler NULL for %s", __func__, name));
KASSERT(nhp->nh_policy == NETISR_POLICY_SOURCE ||
nhp->nh_policy == NETISR_POLICY_FLOW ||
nhp->nh_policy == NETISR_POLICY_CPU,
("%s: unsupported nh_policy %u for %s", __func__,
nhp->nh_policy, name));
KASSERT(nhp->nh_policy == NETISR_POLICY_FLOW ||
nhp->nh_m2flow == NULL,
("%s: nh_policy != FLOW but m2flow defined for %s", __func__,
name));
KASSERT(nhp->nh_policy == NETISR_POLICY_CPU || nhp->nh_m2cpuid == NULL,
("%s: nh_policy != CPU but m2cpuid defined for %s", __func__,
name));
KASSERT(nhp->nh_policy != NETISR_POLICY_CPU || nhp->nh_m2cpuid != NULL,
("%s: nh_policy == CPU but m2cpuid not defined for %s", __func__,
name));
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
KASSERT(nhp->nh_dispatch == NETISR_DISPATCH_DEFAULT ||
nhp->nh_dispatch == NETISR_DISPATCH_DEFERRED ||
nhp->nh_dispatch == NETISR_DISPATCH_HYBRID ||
nhp->nh_dispatch == NETISR_DISPATCH_DIRECT,
("%s: invalid nh_dispatch (%u)", __func__, nhp->nh_dispatch));
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
KASSERT(proto < NETISR_MAXPROT,
("%s(%u, %s): protocol too big", __func__, proto, name));
/*
* Test that no existing registration exists for this protocol.
*/
NETISR_WLOCK();
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_name == NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s(%u, %s): name present", __func__, proto, name));
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_handler == NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s(%u, %s): handler present", __func__, proto, name));
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_name = name;
netisr_proto[proto].np_handler = nhp->nh_handler;
netisr_proto[proto].np_m2flow = nhp->nh_m2flow;
netisr_proto[proto].np_m2cpuid = nhp->nh_m2cpuid;
netisr_proto[proto].np_drainedcpu = nhp->nh_drainedcpu;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
if (nhp->nh_qlimit == 0)
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_qlimit = netisr_defaultqlimit;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
else if (nhp->nh_qlimit > netisr_maxqlimit) {
printf("%s: %s requested queue limit %u capped to "
"net.isr.maxqlimit %u\n", __func__, name, nhp->nh_qlimit,
netisr_maxqlimit);
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_qlimit = netisr_maxqlimit;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
} else
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_qlimit = nhp->nh_qlimit;
netisr_proto[proto].np_policy = nhp->nh_policy;
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
netisr_proto[proto].np_dispatch = nhp->nh_dispatch;
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(i) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto];
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
bzero(npwp, sizeof(*npwp));
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
npwp->nw_qlimit = netisr_proto[proto].np_qlimit;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
}
NETISR_WUNLOCK();
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* Clear drop counters across all workstreams for a protocol.
*/
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
void
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_clearqdrops(const struct netisr_handler *nhp)
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
{
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
struct netisr_work *npwp;
#ifdef INVARIANTS
const char *name;
#endif
u_int i, proto;
proto = nhp->nh_proto;
#ifdef INVARIANTS
name = nhp->nh_name;
#endif
KASSERT(proto < NETISR_MAXPROT,
("%s(%u): protocol too big for %s", __func__, proto, name));
NETISR_WLOCK();
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_handler != NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s(%u): protocol not registered for %s", __func__, proto,
name));
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(i) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto];
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
npwp->nw_qdrops = 0;
}
NETISR_WUNLOCK();
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* Query current drop counters across all workstreams for a protocol.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
void
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_getqdrops(const struct netisr_handler *nhp, u_int64_t *qdropp)
{
struct netisr_work *npwp;
struct rm_priotracker tracker;
#ifdef INVARIANTS
const char *name;
#endif
u_int i, proto;
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*qdropp = 0;
proto = nhp->nh_proto;
#ifdef INVARIANTS
name = nhp->nh_name;
#endif
KASSERT(proto < NETISR_MAXPROT,
("%s(%u): protocol too big for %s", __func__, proto, name));
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
NETISR_RLOCK(&tracker);
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_handler != NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s(%u): protocol not registered for %s", __func__, proto,
name));
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(i) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto];
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*qdropp += npwp->nw_qdrops;
}
NETISR_RUNLOCK(&tracker);
}
/*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* Query current per-workstream queue limit for a protocol.
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*/
void
netisr_getqlimit(const struct netisr_handler *nhp, u_int *qlimitp)
{
struct rm_priotracker tracker;
#ifdef INVARIANTS
const char *name;
#endif
u_int proto;
proto = nhp->nh_proto;
#ifdef INVARIANTS
name = nhp->nh_name;
#endif
KASSERT(proto < NETISR_MAXPROT,
("%s(%u): protocol too big for %s", __func__, proto, name));
NETISR_RLOCK(&tracker);
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_handler != NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s(%u): protocol not registered for %s", __func__, proto,
name));
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
*qlimitp = netisr_proto[proto].np_qlimit;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
NETISR_RUNLOCK(&tracker);
}
/*
* Update the queue limit across per-workstream queues for a protocol. We
* simply change the limits, and don't drain overflowed packets as they will
* (hopefully) take care of themselves shortly.
*/
int
netisr_setqlimit(const struct netisr_handler *nhp, u_int qlimit)
{
struct netisr_work *npwp;
#ifdef INVARIANTS
const char *name;
#endif
u_int i, proto;
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
if (qlimit > netisr_maxqlimit)
return (EINVAL);
proto = nhp->nh_proto;
#ifdef INVARIANTS
name = nhp->nh_name;
#endif
KASSERT(proto < NETISR_MAXPROT,
("%s(%u): protocol too big for %s", __func__, proto, name));
NETISR_WLOCK();
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_handler != NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s(%u): protocol not registered for %s", __func__, proto,
name));
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_qlimit = qlimit;
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(i) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto];
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
npwp->nw_qlimit = qlimit;
}
NETISR_WUNLOCK();
return (0);
}
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
When direct dispatching an netisr (net.isr.enable=1), if there are already any queued packets for the isr, process those packets before the newly submitted packet, maintaining ordering of all packets being delivered to the netisr. Remove the bypass counter since we don't bypass anymore. Leave the comment about possible problems and options since later performance optimization may change the strategy for addressing ordering problems here. Specifically, this maintains the strong isr ordering guarantee; additional parallelism and lower latency may be possible by moving to weaker guarantees (per-interface, for example). We will probably at some point also want to remove the one instance netisr dispatch limit currently enforced by a mutex, but it's not clear that's 100% safe yet, even in the netperf branch. Reviewed by: sam, others
2003-10-03 18:27:24 +00:00
/*
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* Drain all packets currently held in a particular protocol work queue.
When direct dispatching an netisr (net.isr.enable=1), if there are already any queued packets for the isr, process those packets before the newly submitted packet, maintaining ordering of all packets being delivered to the netisr. Remove the bypass counter since we don't bypass anymore. Leave the comment about possible problems and options since later performance optimization may change the strategy for addressing ordering problems here. Specifically, this maintains the strong isr ordering guarantee; additional parallelism and lower latency may be possible by moving to weaker guarantees (per-interface, for example). We will probably at some point also want to remove the one instance netisr dispatch limit currently enforced by a mutex, but it's not clear that's 100% safe yet, even in the netperf branch. Reviewed by: sam, others
2003-10-03 18:27:24 +00:00
*/
static void
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_drain_proto(struct netisr_work *npwp)
When direct dispatching an netisr (net.isr.enable=1), if there are already any queued packets for the isr, process those packets before the newly submitted packet, maintaining ordering of all packets being delivered to the netisr. Remove the bypass counter since we don't bypass anymore. Leave the comment about possible problems and options since later performance optimization may change the strategy for addressing ordering problems here. Specifically, this maintains the strong isr ordering guarantee; additional parallelism and lower latency may be possible by moving to weaker guarantees (per-interface, for example). We will probably at some point also want to remove the one instance netisr dispatch limit currently enforced by a mutex, but it's not clear that's 100% safe yet, even in the netperf branch. Reviewed by: sam, others
2003-10-03 18:27:24 +00:00
{
struct mbuf *m;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* We would assert the lock on the workstream but it's not passed in.
*/
while ((m = npwp->nw_head) != NULL) {
npwp->nw_head = m->m_nextpkt;
m->m_nextpkt = NULL;
if (npwp->nw_head == NULL)
npwp->nw_tail = NULL;
npwp->nw_len--;
m_freem(m);
When direct dispatching an netisr (net.isr.enable=1), if there are already any queued packets for the isr, process those packets before the newly submitted packet, maintaining ordering of all packets being delivered to the netisr. Remove the bypass counter since we don't bypass anymore. Leave the comment about possible problems and options since later performance optimization may change the strategy for addressing ordering problems here. Specifically, this maintains the strong isr ordering guarantee; additional parallelism and lower latency may be possible by moving to weaker guarantees (per-interface, for example). We will probably at some point also want to remove the one instance netisr dispatch limit currently enforced by a mutex, but it's not clear that's 100% safe yet, even in the netperf branch. Reviewed by: sam, others
2003-10-03 18:27:24 +00:00
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
KASSERT(npwp->nw_tail == NULL, ("%s: tail", __func__));
KASSERT(npwp->nw_len == 0, ("%s: len", __func__));
When direct dispatching an netisr (net.isr.enable=1), if there are already any queued packets for the isr, process those packets before the newly submitted packet, maintaining ordering of all packets being delivered to the netisr. Remove the bypass counter since we don't bypass anymore. Leave the comment about possible problems and options since later performance optimization may change the strategy for addressing ordering problems here. Specifically, this maintains the strong isr ordering guarantee; additional parallelism and lower latency may be possible by moving to weaker guarantees (per-interface, for example). We will probably at some point also want to remove the one instance netisr dispatch limit currently enforced by a mutex, but it's not clear that's 100% safe yet, even in the netperf branch. Reviewed by: sam, others
2003-10-03 18:27:24 +00:00
}
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
/*
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* Remove the registration of a network protocol, which requires clearing
* per-protocol fields across all workstreams, including freeing all mbufs in
* the queues at time of unregister. All work in netisr is briefly suspended
* while this takes place.
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
*/
void
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_unregister(const struct netisr_handler *nhp)
{
struct netisr_work *npwp;
#ifdef INVARIANTS
const char *name;
#endif
u_int i, proto;
proto = nhp->nh_proto;
#ifdef INVARIANTS
name = nhp->nh_name;
#endif
KASSERT(proto < NETISR_MAXPROT,
("%s(%u): protocol too big for %s", __func__, proto, name));
NETISR_WLOCK();
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_handler != NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s(%u): protocol not registered for %s", __func__, proto,
name));
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_name = NULL;
netisr_proto[proto].np_handler = NULL;
netisr_proto[proto].np_m2flow = NULL;
netisr_proto[proto].np_m2cpuid = NULL;
netisr_proto[proto].np_qlimit = 0;
netisr_proto[proto].np_policy = 0;
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(i) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto];
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_drain_proto(npwp);
bzero(npwp, sizeof(*npwp));
}
NETISR_WUNLOCK();
}
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
/*
* Compose the global and per-protocol policies on dispatch, and return the
* dispatch policy to use.
*/
static u_int
netisr_get_dispatch(struct netisr_proto *npp)
{
/*
* Protocol-specific configuration overrides the global default.
*/
if (npp->np_dispatch != NETISR_DISPATCH_DEFAULT)
return (npp->np_dispatch);
return (netisr_dispatch_policy);
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* Look up the workstream given a packet and source identifier. Do this by
* checking the protocol's policy, and optionally call out to the protocol
* for assistance if required.
*/
static struct mbuf *
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
netisr_select_cpuid(struct netisr_proto *npp, u_int dispatch_policy,
uintptr_t source, struct mbuf *m, u_int *cpuidp)
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
{
struct ifnet *ifp;
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
u_int policy;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
NETISR_LOCK_ASSERT();
/*
* In the event we have only one worker, shortcut and deliver to it
* without further ado.
*/
if (nws_count == 1) {
*cpuidp = nws_array[0];
return (m);
Discard the packet if the netisr queue is null instead of panicing, for the benefit of modules which are compiled differently than the kernel.
2003-03-08 22:12:32 +00:00
}
Remove NETISR_MPSAFE, which allows specific netisr handlers to be directly dispatched without Giant, and add NETISR_FORCEQUEUE, which allows specific netisr handlers to always be dispatched via a queue (deferred). Mark the usb and if_ppp netisr handlers as NETISR_FORCEQUEUE, and explicitly acquire Giant in those handlers. Previously, any netisr handler not marked NETISR_MPSAFE would necessarily run deferred and with Giant acquired. This change removes Giant scaffolding from the netisr infrastructure, but NETISR_FORCEQUEUE allows non-MPSAFE handlers to continue to force deferred dispatch so as to avoid lock order reversals between their acqusition of Giant and any calling context. It is likely we will be able to remove NETISR_FORCEQUEUE once IFF_NEEDSGIANT is removed, as non-MPSAFE usb and if_ppp drivers will no longer be supported. Reviewed by: bz MFC after: 1 month X-MFC note: We can't remove NETISR_MPSAFE from stable/7 for KPI reasons, but the rest can go back.
2008-07-04 00:21:38 +00:00
o add a flags parameter to netisr_register that is used to specify whether or not the isr needs to hold Giant when running; Giant-less operation is also controlled by the setting of debug_mpsafenet o mark all netisr's except NETISR_IP as needing Giant o add a GIANT_REQUIRED assertion to the top of netisr's that need Giant o pickup Giant (when debug_mpsafenet is 1) inside ip_input before calling up with a packet o change netisr handling so swi_net runs w/o Giant; instead we grab Giant before invoking handlers based on whether the handler needs Giant o change netisr handling so that netisr's that are marked MPSAFE may have multiple instances active at a time o add netisr statistics for packets dropped because the isr is inactive Supported by: FreeBSD Foundation
2003-11-08 22:28:40 +00:00
/*
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* What happens next depends on the policy selected by the protocol.
* If we want to support per-interface policies, we should do that
* here first.
o add a flags parameter to netisr_register that is used to specify whether or not the isr needs to hold Giant when running; Giant-less operation is also controlled by the setting of debug_mpsafenet o mark all netisr's except NETISR_IP as needing Giant o add a GIANT_REQUIRED assertion to the top of netisr's that need Giant o pickup Giant (when debug_mpsafenet is 1) inside ip_input before calling up with a packet o change netisr handling so swi_net runs w/o Giant; instead we grab Giant before invoking handlers based on whether the handler needs Giant o change netisr handling so that netisr's that are marked MPSAFE may have multiple instances active at a time o add netisr statistics for packets dropped because the isr is inactive Supported by: FreeBSD Foundation
2003-11-08 22:28:40 +00:00
*/
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
policy = npp->np_policy;
if (policy == NETISR_POLICY_CPU) {
m = npp->np_m2cpuid(m, source, cpuidp);
if (m == NULL)
return (NULL);
/*
* It's possible for a protocol not to have a good idea about
* where to process a packet, in which case we fall back on
* the netisr code to decide. In the hybrid case, return the
* current CPU ID, which will force an immediate direct
* dispatch. In the queued case, fall back on the SOURCE
* policy.
*/
if (*cpuidp != NETISR_CPUID_NONE)
return (m);
if (dispatch_policy == NETISR_DISPATCH_HYBRID) {
*cpuidp = curcpu;
return (m);
}
policy = NETISR_POLICY_SOURCE;
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
if (policy == NETISR_POLICY_FLOW) {
Start process of removing the use of the deprecated "M_FLOWID" flag 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
2014-12-01 11:45:24 +00:00
if (M_HASHTYPE_GET(m) == M_HASHTYPE_NONE &&
npp->np_m2flow != NULL) {
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
m = npp->np_m2flow(m, source);
if (m == NULL)
return (NULL);
}
Start process of removing the use of the deprecated "M_FLOWID" flag 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
2014-12-01 11:45:24 +00:00
if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) {
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
*cpuidp =
netisr_default_flow2cpu(m->m_pkthdr.flowid);
return (m);
}
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
policy = NETISR_POLICY_SOURCE;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
}
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
KASSERT(policy == NETISR_POLICY_SOURCE,
("%s: invalid policy %u for %s", __func__, npp->np_policy,
npp->np_name));
ifp = m->m_pkthdr.rcvif;
if (ifp != NULL)
*cpuidp = nws_array[(ifp->if_index + source) % nws_count];
else
*cpuidp = nws_array[source % nws_count];
return (m);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
}
/*
* Process packets associated with a workstream and protocol. For reasons of
* fairness, we process up to one complete netisr queue at a time, moving the
* queue to a stack-local queue for processing, but do not loop refreshing
* from the global queue. The caller is responsible for deciding whether to
* loop, and for setting the NWS_RUNNING flag. The passed workstream will be
* locked on entry and relocked before return, but will be released while
* processing. The number of packets processed is returned.
*/
static u_int
netisr_process_workstream_proto(struct netisr_workstream *nwsp, u_int proto)
{
struct netisr_work local_npw, *npwp;
u_int handled;
struct mbuf *m;
NETISR_LOCK_ASSERT();
NWS_LOCK_ASSERT(nwsp);
KASSERT(nwsp->nws_flags & NWS_RUNNING,
("%s(%u): not running", __func__, proto));
KASSERT(proto >= 0 && proto < NETISR_MAXPROT,
("%s(%u): invalid proto\n", __func__, proto));
npwp = &nwsp->nws_work[proto];
if (npwp->nw_len == 0)
return (0);
/*
* Move the global work queue to a thread-local work queue.
*
* Notice that this means the effective maximum length of the queue
* is actually twice that of the maximum queue length specified in
* the protocol registration call.
*/
handled = npwp->nw_len;
local_npw = *npwp;
npwp->nw_head = NULL;
npwp->nw_tail = NULL;
npwp->nw_len = 0;
nwsp->nws_pendingbits &= ~(1 << proto);
NWS_UNLOCK(nwsp);
while ((m = local_npw.nw_head) != NULL) {
local_npw.nw_head = m->m_nextpkt;
m->m_nextpkt = NULL;
if (local_npw.nw_head == NULL)
local_npw.nw_tail = NULL;
local_npw.nw_len--;
Mfp4 CH=177255: Make VNET_ASSERT() available with either VNET_DEBUG or INVARIANTS. Change the syntax to match KASSERT() to allow more flexible panic messages rather than having a printf with hardcoded arguments before panic. Adjust the few assertions we have to the new format (and enhance the output). Sponsored by: The FreeBSD Foundation Sponsored by: CK Software GmbH Reviewed by: jhb MFC after: 2 weeks
2011-02-11 13:27:00 +00:00
VNET_ASSERT(m->m_pkthdr.rcvif != NULL,
("%s:%d rcvif == NULL: m=%p", __func__, __LINE__, m));
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
CURVNET_SET(m->m_pkthdr.rcvif->if_vnet);
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_handler(m);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
CURVNET_RESTORE();
}
KASSERT(local_npw.nw_len == 0,
("%s(%u): len %u", __func__, proto, local_npw.nw_len));
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
if (netisr_proto[proto].np_drainedcpu)
netisr_proto[proto].np_drainedcpu(nwsp->nws_cpu);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
NWS_LOCK(nwsp);
npwp->nw_handled += handled;
return (handled);
}
/*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* SWI handler for netisr -- processes packets in a set of workstreams that
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* it owns, woken up by calls to NWS_SIGNAL(). If this workstream is already
* being direct dispatched, go back to sleep and wait for the dispatching
* thread to wake us up again.
*/
static void
swi_net(void *arg)
{
#ifdef NETISR_LOCKING
struct rm_priotracker tracker;
#endif
struct netisr_workstream *nwsp;
u_int bits, prot;
nwsp = arg;
#ifdef DEVICE_POLLING
Revert 281276 as unnecessary. Proper change to be committed to the base polling code in a subsequent commit. Pointed out by: glebius Sponsored by: Rubicon Communications (NetGate)
2015-04-09 14:44:30 +00:00
KASSERT(nws_count == 1,
("%s: device_polling but nws_count != 1", __func__));
netisr_poll();
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#endif
#ifdef NETISR_LOCKING
NETISR_RLOCK(&tracker);
#endif
NWS_LOCK(nwsp);
KASSERT(!(nwsp->nws_flags & NWS_RUNNING), ("swi_net: running"));
if (nwsp->nws_flags & NWS_DISPATCHING)
goto out;
nwsp->nws_flags |= NWS_RUNNING;
nwsp->nws_flags &= ~NWS_SCHEDULED;
while ((bits = nwsp->nws_pendingbits) != 0) {
while ((prot = ffs(bits)) != 0) {
prot--;
bits &= ~(1 << prot);
(void)netisr_process_workstream_proto(nwsp, prot);
}
}
nwsp->nws_flags &= ~NWS_RUNNING;
out:
NWS_UNLOCK(nwsp);
#ifdef NETISR_LOCKING
NETISR_RUNLOCK(&tracker);
#endif
#ifdef DEVICE_POLLING
Revert 281276 as unnecessary. Proper change to be committed to the base polling code in a subsequent commit. Pointed out by: glebius Sponsored by: Rubicon Communications (NetGate)
2015-04-09 14:44:30 +00:00
netisr_pollmore();
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#endif
}
static int
netisr_queue_workstream(struct netisr_workstream *nwsp, u_int proto,
struct netisr_work *npwp, struct mbuf *m, int *dosignalp)
{
NWS_LOCK_ASSERT(nwsp);
*dosignalp = 0;
if (npwp->nw_len < npwp->nw_qlimit) {
m->m_nextpkt = NULL;
if (npwp->nw_head == NULL) {
npwp->nw_head = m;
npwp->nw_tail = m;
} else {
npwp->nw_tail->m_nextpkt = m;
npwp->nw_tail = m;
}
npwp->nw_len++;
if (npwp->nw_len > npwp->nw_watermark)
npwp->nw_watermark = npwp->nw_len;
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
/*
* We must set the bit regardless of NWS_RUNNING, so that
* swi_net() keeps calling netisr_process_workstream_proto().
*/
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
nwsp->nws_pendingbits |= (1 << proto);
if (!(nwsp->nws_flags &
(NWS_RUNNING | NWS_DISPATCHING | NWS_SCHEDULED))) {
nwsp->nws_flags |= NWS_SCHEDULED;
*dosignalp = 1; /* Defer until unlocked. */
}
npwp->nw_queued++;
return (0);
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
} else {
In case we cannot queue a packet reaching the queue limit, retain the semantics netisr_queue() always had and free the mbuf along with returning the error. Reviewed by: rwatson Approved by: re (kensmith)
2009-06-30 05:21:00 +00:00
m_freem(m);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
npwp->nw_qdrops++;
return (ENOBUFS);
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
}
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
static int
netisr_queue_internal(u_int proto, struct mbuf *m, u_int cpuid)
{
struct netisr_workstream *nwsp;
struct netisr_work *npwp;
int dosignal, error;
#ifdef NETISR_LOCKING
NETISR_LOCK_ASSERT();
#endif
In light of DPCPU use by netisr, revise various for loops from using MAXCPU to mp_maxid, and handling and reporting of requests to use more threads than we have CPUs to run them on. Reviewed by: bz Approved by: re (kib) MFC after: 6 weeks
2009-06-26 20:39:36 +00:00
KASSERT(cpuid <= mp_maxid, ("%s: cpuid too big (%u, %u)", __func__,
cpuid, mp_maxid));
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__, cpuid));
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
dosignal = 0;
error = 0;
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
nwsp = DPCPU_ID_PTR(cpuid, nws);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
npwp = &nwsp->nws_work[proto];
NWS_LOCK(nwsp);
error = netisr_queue_workstream(nwsp, proto, npwp, m, &dosignal);
NWS_UNLOCK(nwsp);
if (dosignal)
NWS_SIGNAL(nwsp);
return (error);
}
int
netisr_queue_src(u_int proto, uintptr_t source, struct mbuf *m)
{
#ifdef NETISR_LOCKING
struct rm_priotracker tracker;
#endif
u_int cpuid;
int error;
KASSERT(proto < NETISR_MAXPROT,
("%s: invalid proto %u", __func__, proto));
#ifdef NETISR_LOCKING
NETISR_RLOCK(&tracker);
#endif
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
KASSERT(netisr_proto[proto].np_handler != NULL,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
("%s: invalid proto %u", __func__, proto));
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
m = netisr_select_cpuid(&netisr_proto[proto], NETISR_DISPATCH_DEFERRED,
source, m, &cpuid);
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
if (m != NULL) {
KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__,
cpuid));
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
error = netisr_queue_internal(proto, m, cpuid);
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
} else
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
error = ENOBUFS;
#ifdef NETISR_LOCKING
NETISR_RUNLOCK(&tracker);
#endif
return (error);
}
int
netisr_queue(u_int proto, struct mbuf *m)
{
return (netisr_queue_src(proto, 0, m));
}
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
/*
Refine netisr.c comments a bit.
2009-12-23 12:31:27 +00:00
* Dispatch a packet for netisr processing; direct dispatch is permitted by
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
* calling context.
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
*/
int
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_dispatch_src(u_int proto, uintptr_t source, struct mbuf *m)
{
#ifdef NETISR_LOCKING
struct rm_priotracker tracker;
#endif
struct netisr_workstream *nwsp;
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
struct netisr_proto *npp;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
struct netisr_work *npwp;
int dosignal, error;
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
u_int cpuid, dispatch_policy;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
KASSERT(proto < NETISR_MAXPROT,
("%s: invalid proto %u", __func__, proto));
#ifdef NETISR_LOCKING
NETISR_RLOCK(&tracker);
#endif
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
npp = &netisr_proto[proto];
KASSERT(npp->np_handler != NULL, ("%s: invalid proto %u", __func__,
proto));
dispatch_policy = netisr_get_dispatch(npp);
if (dispatch_policy == NETISR_DISPATCH_DEFERRED)
return (netisr_queue_src(proto, source, m));
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* If direct dispatch is forced, then unconditionally dispatch
* without a formal CPU selection. Borrow the current CPU's stats,
* even if there's no worker on it. In this case we don't update
* nws_flags because all netisr processing will be source ordered due
* to always being forced to directly dispatch.
*/
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
if (dispatch_policy == NETISR_DISPATCH_DIRECT) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
nwsp = DPCPU_PTR(nws);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
npwp = &nwsp->nws_work[proto];
npwp->nw_dispatched++;
npwp->nw_handled++;
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_handler(m);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
error = 0;
goto out_unlock;
Discard the packet if the netisr queue is null instead of panicing, for the benefit of modules which are compiled differently than the kernel.
2003-03-08 22:12:32 +00:00
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
KASSERT(dispatch_policy == NETISR_DISPATCH_HYBRID,
("%s: unknown dispatch policy (%u)", __func__, dispatch_policy));
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* Otherwise, we execute in a hybrid mode where we will try to direct
* dispatch if we're on the right CPU and the netisr worker isn't
* already running.
*/
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
sched_pin();
m = netisr_select_cpuid(&netisr_proto[proto], NETISR_DISPATCH_HYBRID,
source, m, &cpuid);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
if (m == NULL) {
error = ENOBUFS;
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
goto out_unpin;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
}
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__, cpuid));
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
if (cpuid != curcpu)
goto queue_fallback;
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
nwsp = DPCPU_PTR(nws);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
npwp = &nwsp->nws_work[proto];
/*-
* We are willing to direct dispatch only if three conditions hold:
*
* (1) The netisr worker isn't already running,
* (2) Another thread isn't already directly dispatching, and
* (3) The netisr hasn't already been woken up.
*/
NWS_LOCK(nwsp);
if (nwsp->nws_flags & (NWS_RUNNING | NWS_DISPATCHING | NWS_SCHEDULED)) {
error = netisr_queue_workstream(nwsp, proto, npwp, m,
&dosignal);
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
NWS_UNLOCK(nwsp);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
if (dosignal)
NWS_SIGNAL(nwsp);
goto out_unpin;
}
/*
* The current thread is now effectively the netisr worker, so set
* the dispatching flag to prevent concurrent processing of the
* stream from another thread (even the netisr worker), which could
* otherwise lead to effective misordering of the stream.
*/
nwsp->nws_flags |= NWS_DISPATCHING;
NWS_UNLOCK(nwsp);
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_handler(m);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
NWS_LOCK(nwsp);
nwsp->nws_flags &= ~NWS_DISPATCHING;
npwp->nw_handled++;
npwp->nw_hybrid_dispatched++;
/*
* If other work was enqueued by another thread while we were direct
* dispatching, we need to signal the netisr worker to do that work.
* In the future, we might want to do some of that work in the
* current thread, rather than trigger further context switches. If
* so, we'll want to establish a reasonable bound on the work done in
* the "borrowed" context.
*/
if (nwsp->nws_pendingbits != 0) {
nwsp->nws_flags |= NWS_SCHEDULED;
dosignal = 1;
} else
dosignal = 0;
NWS_UNLOCK(nwsp);
if (dosignal)
NWS_SIGNAL(nwsp);
error = 0;
goto out_unpin;
queue_fallback:
error = netisr_queue_internal(proto, m, cpuid);
out_unpin:
sched_unpin();
out_unlock:
#ifdef NETISR_LOCKING
NETISR_RUNLOCK(&tracker);
#endif
return (error);
}
int
netisr_dispatch(u_int proto, struct mbuf *m)
{
return (netisr_dispatch_src(proto, 0, m));
}
#ifdef DEVICE_POLLING
/*
* Kernel polling borrows a netisr thread to run interface polling in; this
* function allows kernel polling to request that the netisr thread be
* scheduled even if no packets are pending for protocols.
*/
void
netisr_sched_poll(void)
{
struct netisr_workstream *nwsp;
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
nwsp = DPCPU_ID_PTR(nws_array[0], nws);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
NWS_SIGNAL(nwsp);
Update netisr handling; Each SWI now registers its queue, and all queue drain routines are done by swi_net, which allows for better queue control at some future point. Packets may also be directly dispatched to a netisr instead of queued, this may be of interest at some installations, but currently defaults to off. Reviewed by: hsu, silby, jayanth, sam Sponsored by: DARPA, NAI Labs
2003-03-04 23:19:55 +00:00
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#endif
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
static void
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_start_swi(u_int cpuid, struct pcpu *pc)
{
char swiname[12];
struct netisr_workstream *nwsp;
int error;
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__, cpuid));
nwsp = DPCPU_ID_PTR(cpuid, nws);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
mtx_init(&nwsp->nws_mtx, "netisr_mtx", NULL, MTX_DEF);
nwsp->nws_cpu = cpuid;
snprintf(swiname, sizeof(swiname), "netisr %u", cpuid);
error = swi_add(&nwsp->nws_intr_event, swiname, swi_net, nwsp,
SWI_NET, INTR_MPSAFE, &nwsp->nws_swi_cookie);
if (error)
panic("%s: swi_add %d", __func__, error);
pc->pc_netisr = nwsp->nws_intr_event;
if (netisr_bindthreads) {
error = intr_event_bind(nwsp->nws_intr_event, cpuid);
if (error != 0)
printf("%s: cpu %u: intr_event_bind: %d", __func__,
cpuid, error);
}
NETISR_WLOCK();
nws_array[nws_count] = nwsp->nws_cpu;
nws_count++;
NETISR_WUNLOCK();
}
/*
* Initialize the netisr subsystem. We rely on BSS and static initialization
* of most fields in global data structures.
*
* Start a worker thread for the boot CPU so that we can support network
* traffic immediately in case the network stack is used before additional
* CPUs are started (for example, diskless boot).
*/
static void
netisr_init(void *arg)
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
{
Add an EARLY_AP_STARTUP option to start APs earlier during boot. Currently, Application Processors (non-boot CPUs) are started by MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until SI_SUB_SMP at which point they are released to run kernel threads. SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter the scheduler and start running threads until fairly late in the boot. This change moves SI_SUB_SMP up to just before software interrupt threads are created allowing the APs to start executing kernel threads much sooner (before any devices are probed). This allows several initialization routines that need to perform initialization on all CPUs to now perform that initialization in one step rather than having to defer the AP initialization to a second SYSINIT run at SI_SUB_SMP. It also permits all CPUs to be available for handling interrupts before any devices are probed. This last feature fixes a problem on with interrupt vector exhaustion. Specifically, in the old model all device interrupts were routed onto the boot CPU during boot. Later after the APs were released at SI_SUB_SMP, interrupts were redistributed across all CPUs. However, several drivers for multiqueue hardware allocate N interrupts per CPU in the system. In a system with many CPUs, just a few drivers doing this could exhaust the available pool of interrupt vectors on the boot CPU as each driver was allocating N * mp_ncpu vectors on the boot CPU. Now, drivers will allocate interrupts on their desired CPUs during boot meaning that only N interrupts are allocated from the boot CPU instead of N * mp_ncpu. Some other bits of code can also be simplified as smp_started is now true much earlier and will now always be true for these bits of code. This removes the need to treat the single-CPU boot environment as a special case. As a transition aid, the new behavior is available under a new kernel option (EARLY_AP_STARTUP). This will allow the option to be turned off if need be during initial testing. I plan to enable this on x86 by default in a followup commit in the next few days and to have all platforms moved over before 11.0. Once the transition is complete, the option will be removed along with the !EARLY_AP_STARTUP code. These changes have only been tested on x86. Other platform maintainers are encouraged to port their architectures over as well. The main things to check for are any uses of smp_started in MD code that can be simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in the EARLY_AP_STARTUP case (e.g. the interrupt shuffling). PR: kern/199321 Reviewed by: markj, gnn, kib Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
#ifdef EARLY_AP_STARTUP
struct pcpu *pc;
#endif
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
KASSERT(curcpu == 0, ("%s: not on CPU 0", __func__));
NETISR_LOCK_INIT();
Currently there is no easy way to specify net.isr.maxthreads = all cpus. We need to specify exact number of cpus in loader.conf which get annoying when you have mix of machines which don't have equal number of total cpus. I propose "-1" as that value. When loader.conf has net.isr.maxthreads = -1, netisr will use all available cpus. In collaboration with: davide Reviewed by: gnn Differential Revision: https://reviews.freebsd.org/D2318 MFC after: 2 weeks Sponsored by: Limelight Networks
2015-04-25 16:12:06 +00:00
if (netisr_maxthreads == 0 || netisr_maxthreads < -1 )
netisr_maxthreads = 1; /* default behavior */
else if (netisr_maxthreads == -1)
netisr_maxthreads = mp_ncpus; /* use max cpus */
In light of DPCPU use by netisr, revise various for loops from using MAXCPU to mp_maxid, and handling and reporting of requests to use more threads than we have CPUs to run them on. Reviewed by: bz Approved by: re (kib) MFC after: 6 weeks
2009-06-26 20:39:36 +00:00
if (netisr_maxthreads > mp_ncpus) {
When warning about possible netisr configuration problems during boot, report using "netisr_init" rather than "netisr2", which was the development name for the project. MFC after: 3 days
2009-12-23 12:33:59 +00:00
printf("netisr_init: forcing maxthreads from %d to %d\n",
In light of DPCPU use by netisr, revise various for loops from using MAXCPU to mp_maxid, and handling and reporting of requests to use more threads than we have CPUs to run them on. Reviewed by: bz Approved by: re (kib) MFC after: 6 weeks
2009-06-26 20:39:36 +00:00
netisr_maxthreads, mp_ncpus);
netisr_maxthreads = mp_ncpus;
Garbage collect NETISR_POLL and NETISR_POLLMORE, which are no longer required for options DEVICE_POLLING. De-fragment the NETISR_ constant space and lower NETISR_MAXPROT from 32 to 16 -- when sizing queue arrays using this compile-time constant, significant amounts of memory are saved. Warn on the console when tunable values for netisr are automatically adjusted during boot due to exceeding limits, invalid values, or as a result of DEVICE_POLLING.
2009-06-01 15:03:58 +00:00
}
if (netisr_defaultqlimit > netisr_maxqlimit) {
When warning about possible netisr configuration problems during boot, report using "netisr_init" rather than "netisr2", which was the development name for the project. MFC after: 3 days
2009-12-23 12:33:59 +00:00
printf("netisr_init: forcing defaultqlimit from %d to %d\n",
In light of DPCPU use by netisr, revise various for loops from using MAXCPU to mp_maxid, and handling and reporting of requests to use more threads than we have CPUs to run them on. Reviewed by: bz Approved by: re (kib) MFC after: 6 weeks
2009-06-26 20:39:36 +00:00
netisr_defaultqlimit, netisr_maxqlimit);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_defaultqlimit = netisr_maxqlimit;
Garbage collect NETISR_POLL and NETISR_POLLMORE, which are no longer required for options DEVICE_POLLING. De-fragment the NETISR_ constant space and lower NETISR_MAXPROT from 32 to 16 -- when sizing queue arrays using this compile-time constant, significant amounts of memory are saved. Warn on the console when tunable values for netisr are automatically adjusted during boot due to exceeding limits, invalid values, or as a result of DEVICE_POLLING.
2009-06-01 15:03:58 +00:00
}
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
#ifdef DEVICE_POLLING
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* The device polling code is not yet aware of how to deal with
* multiple netisr threads, so for the time being compiling in device
* polling disables parallel netisr workers.
*/
Revert 281276 as unnecessary. Proper change to be committed to the base polling code in a subsequent commit. Pointed out by: glebius Sponsored by: Rubicon Communications (NetGate)
2015-04-09 14:44:30 +00:00
if (netisr_maxthreads != 1 || netisr_bindthreads != 0) {
When warning about possible netisr configuration problems during boot, report using "netisr_init" rather than "netisr2", which was the development name for the project. MFC after: 3 days
2009-12-23 12:33:59 +00:00
printf("netisr_init: forcing maxthreads to 1 and "
"bindthreads to 0 for device polling\n");
Garbage collect NETISR_POLL and NETISR_POLLMORE, which are no longer required for options DEVICE_POLLING. De-fragment the NETISR_ constant space and lower NETISR_MAXPROT from 32 to 16 -- when sizing queue arrays using this compile-time constant, significant amounts of memory are saved. Warn on the console when tunable values for netisr are automatically adjusted during boot due to exceeding limits, invalid values, or as a result of DEVICE_POLLING.
2009-06-01 15:03:58 +00:00
netisr_maxthreads = 1;
netisr_bindthreads = 0;
}
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
#endif
Add an EARLY_AP_STARTUP option to start APs earlier during boot. Currently, Application Processors (non-boot CPUs) are started by MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until SI_SUB_SMP at which point they are released to run kernel threads. SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter the scheduler and start running threads until fairly late in the boot. This change moves SI_SUB_SMP up to just before software interrupt threads are created allowing the APs to start executing kernel threads much sooner (before any devices are probed). This allows several initialization routines that need to perform initialization on all CPUs to now perform that initialization in one step rather than having to defer the AP initialization to a second SYSINIT run at SI_SUB_SMP. It also permits all CPUs to be available for handling interrupts before any devices are probed. This last feature fixes a problem on with interrupt vector exhaustion. Specifically, in the old model all device interrupts were routed onto the boot CPU during boot. Later after the APs were released at SI_SUB_SMP, interrupts were redistributed across all CPUs. However, several drivers for multiqueue hardware allocate N interrupts per CPU in the system. In a system with many CPUs, just a few drivers doing this could exhaust the available pool of interrupt vectors on the boot CPU as each driver was allocating N * mp_ncpu vectors on the boot CPU. Now, drivers will allocate interrupts on their desired CPUs during boot meaning that only N interrupts are allocated from the boot CPU instead of N * mp_ncpu. Some other bits of code can also be simplified as smp_started is now true much earlier and will now always be true for these bits of code. This removes the need to treat the single-CPU boot environment as a special case. As a transition aid, the new behavior is available under a new kernel option (EARLY_AP_STARTUP). This will allow the option to be turned off if need be during initial testing. I plan to enable this on x86 by default in a followup commit in the next few days and to have all platforms moved over before 11.0. Once the transition is complete, the option will be removed along with the !EARLY_AP_STARTUP code. These changes have only been tested on x86. Other platform maintainers are encouraged to port their architectures over as well. The main things to check for are any uses of smp_started in MD code that can be simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in the EARLY_AP_STARTUP case (e.g. the interrupt shuffling). PR: kern/199321 Reviewed by: markj, gnn, kib Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
#ifdef EARLY_AP_STARTUP
STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
if (nws_count >= netisr_maxthreads)
break;
netisr_start_swi(pc->pc_cpuid, pc);
}
#else
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_start_swi(curcpu, pcpu_find(curcpu));
Add an EARLY_AP_STARTUP option to start APs earlier during boot. Currently, Application Processors (non-boot CPUs) are started by MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until SI_SUB_SMP at which point they are released to run kernel threads. SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter the scheduler and start running threads until fairly late in the boot. This change moves SI_SUB_SMP up to just before software interrupt threads are created allowing the APs to start executing kernel threads much sooner (before any devices are probed). This allows several initialization routines that need to perform initialization on all CPUs to now perform that initialization in one step rather than having to defer the AP initialization to a second SYSINIT run at SI_SUB_SMP. It also permits all CPUs to be available for handling interrupts before any devices are probed. This last feature fixes a problem on with interrupt vector exhaustion. Specifically, in the old model all device interrupts were routed onto the boot CPU during boot. Later after the APs were released at SI_SUB_SMP, interrupts were redistributed across all CPUs. However, several drivers for multiqueue hardware allocate N interrupts per CPU in the system. In a system with many CPUs, just a few drivers doing this could exhaust the available pool of interrupt vectors on the boot CPU as each driver was allocating N * mp_ncpu vectors on the boot CPU. Now, drivers will allocate interrupts on their desired CPUs during boot meaning that only N interrupts are allocated from the boot CPU instead of N * mp_ncpu. Some other bits of code can also be simplified as smp_started is now true much earlier and will now always be true for these bits of code. This removes the need to treat the single-CPU boot environment as a special case. As a transition aid, the new behavior is available under a new kernel option (EARLY_AP_STARTUP). This will allow the option to be turned off if need be during initial testing. I plan to enable this on x86 by default in a followup commit in the next few days and to have all platforms moved over before 11.0. Once the transition is complete, the option will be removed along with the !EARLY_AP_STARTUP code. These changes have only been tested on x86. Other platform maintainers are encouraged to port their architectures over as well. The main things to check for are any uses of smp_started in MD code that can be simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in the EARLY_AP_STARTUP case (e.g. the interrupt shuffling). PR: kern/199321 Reviewed by: markj, gnn, kib Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
#endif
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
SYSINIT(netisr_init, SI_SUB_SOFTINTR, SI_ORDER_FIRST, netisr_init, NULL);
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
Add an EARLY_AP_STARTUP option to start APs earlier during boot. Currently, Application Processors (non-boot CPUs) are started by MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until SI_SUB_SMP at which point they are released to run kernel threads. SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter the scheduler and start running threads until fairly late in the boot. This change moves SI_SUB_SMP up to just before software interrupt threads are created allowing the APs to start executing kernel threads much sooner (before any devices are probed). This allows several initialization routines that need to perform initialization on all CPUs to now perform that initialization in one step rather than having to defer the AP initialization to a second SYSINIT run at SI_SUB_SMP. It also permits all CPUs to be available for handling interrupts before any devices are probed. This last feature fixes a problem on with interrupt vector exhaustion. Specifically, in the old model all device interrupts were routed onto the boot CPU during boot. Later after the APs were released at SI_SUB_SMP, interrupts were redistributed across all CPUs. However, several drivers for multiqueue hardware allocate N interrupts per CPU in the system. In a system with many CPUs, just a few drivers doing this could exhaust the available pool of interrupt vectors on the boot CPU as each driver was allocating N * mp_ncpu vectors on the boot CPU. Now, drivers will allocate interrupts on their desired CPUs during boot meaning that only N interrupts are allocated from the boot CPU instead of N * mp_ncpu. Some other bits of code can also be simplified as smp_started is now true much earlier and will now always be true for these bits of code. This removes the need to treat the single-CPU boot environment as a special case. As a transition aid, the new behavior is available under a new kernel option (EARLY_AP_STARTUP). This will allow the option to be turned off if need be during initial testing. I plan to enable this on x86 by default in a followup commit in the next few days and to have all platforms moved over before 11.0. Once the transition is complete, the option will be removed along with the !EARLY_AP_STARTUP code. These changes have only been tested on x86. Other platform maintainers are encouraged to port their architectures over as well. The main things to check for are any uses of smp_started in MD code that can be simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in the EARLY_AP_STARTUP case (e.g. the interrupt shuffling). PR: kern/199321 Reviewed by: markj, gnn, kib Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
#ifndef EARLY_AP_STARTUP
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
/*
* Start worker threads for additional CPUs. No attempt to gracefully handle
* work reassignment, we don't yet support dynamic reconfiguration.
*/
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
static void
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
netisr_start(void *arg)
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
{
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
struct pcpu *pc;
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
On multi-core, multi-threaded PPC systems, it is important that the threads be brought up in the order they are enumerated in the device tree (in particular, that thread 0 on each core be brought up first). The SLIST through which we loop to start the CPUs has all of its entries added with SLIST_INSERT_HEAD(), which means it is in reverse order of enumeration and so AP startup would always fail in such situations (causing a machine check or RTAS failure). Fix this by changing the SLIST into an STAILQ, and inserting new CPUs at the end. Reviewed by: jhb
2011-05-31 15:11:43 +00:00
STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
if (nws_count >= netisr_maxthreads)
break;
/* Worker will already be present for boot CPU. */
if (pc->pc_netisr != NULL)
continue;
netisr_start_swi(pc->pc_cpuid, pc);
}
Moved netisr code from kern/kern_intr.c to net/netisr.c as threatened in a comment.
2002-09-22 05:56:41 +00:00
}
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
SYSINIT(netisr_start, SI_SUB_SMP, SI_ORDER_MIDDLE, netisr_start, NULL);
Add an EARLY_AP_STARTUP option to start APs earlier during boot. Currently, Application Processors (non-boot CPUs) are started by MD code at SI_SUB_CPU, but they are kept waiting in a "pen" until SI_SUB_SMP at which point they are released to run kernel threads. SI_SUB_SMP is one of the last SYSINIT levels, so APs don't enter the scheduler and start running threads until fairly late in the boot. This change moves SI_SUB_SMP up to just before software interrupt threads are created allowing the APs to start executing kernel threads much sooner (before any devices are probed). This allows several initialization routines that need to perform initialization on all CPUs to now perform that initialization in one step rather than having to defer the AP initialization to a second SYSINIT run at SI_SUB_SMP. It also permits all CPUs to be available for handling interrupts before any devices are probed. This last feature fixes a problem on with interrupt vector exhaustion. Specifically, in the old model all device interrupts were routed onto the boot CPU during boot. Later after the APs were released at SI_SUB_SMP, interrupts were redistributed across all CPUs. However, several drivers for multiqueue hardware allocate N interrupts per CPU in the system. In a system with many CPUs, just a few drivers doing this could exhaust the available pool of interrupt vectors on the boot CPU as each driver was allocating N * mp_ncpu vectors on the boot CPU. Now, drivers will allocate interrupts on their desired CPUs during boot meaning that only N interrupts are allocated from the boot CPU instead of N * mp_ncpu. Some other bits of code can also be simplified as smp_started is now true much earlier and will now always be true for these bits of code. This removes the need to treat the single-CPU boot environment as a special case. As a transition aid, the new behavior is available under a new kernel option (EARLY_AP_STARTUP). This will allow the option to be turned off if need be during initial testing. I plan to enable this on x86 by default in a followup commit in the next few days and to have all platforms moved over before 11.0. Once the transition is complete, the option will be removed along with the !EARLY_AP_STARTUP code. These changes have only been tested on x86. Other platform maintainers are encouraged to port their architectures over as well. The main things to check for are any uses of smp_started in MD code that can be simplified and SI_SUB_SMP SYSINITs in MD code that can be removed in the EARLY_AP_STARTUP case (e.g. the interrupt shuffling). PR: kern/199321 Reviewed by: markj, gnn, kib Sponsored by: Netflix
2016-05-14 18:22:52 +00:00
#endif
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
/*
* Sysctl monitoring for netisr: query a list of registered protocols.
*/
static int
sysctl_netisr_proto(SYSCTL_HANDLER_ARGS)
{
struct rm_priotracker tracker;
struct sysctl_netisr_proto *snpp, *snp_array;
struct netisr_proto *npp;
u_int counter, proto;
int error;
if (req->newptr != NULL)
return (EINVAL);
snp_array = malloc(sizeof(*snp_array) * NETISR_MAXPROT, M_TEMP,
M_ZERO | M_WAITOK);
counter = 0;
NETISR_RLOCK(&tracker);
for (proto = 0; proto < NETISR_MAXPROT; proto++) {
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
npp = &netisr_proto[proto];
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
if (npp->np_name == NULL)
continue;
snpp = &snp_array[counter];
snpp->snp_version = sizeof(*snpp);
strlcpy(snpp->snp_name, npp->np_name, NETISR_NAMEMAXLEN);
snpp->snp_proto = proto;
snpp->snp_qlimit = npp->np_qlimit;
snpp->snp_policy = npp->np_policy;
Rework netisr policy mechanism so that per-protocol dispatch policies can be represented: - A single policy namespace is defined, consisting of four possible policies: "default" to use the global default, "deferred" to force deferred dispatch, "direct" to employ direct dispatch where possible, and "hybrid" which makes a dynamic decision based on CPU affinity, ordering, etc. Routines are implemented to convert between strings and an integer namespace. - A new global variable, netisr_dispatch_policy, subsumes existing global variables for direct dispatch, forced direct dispatch, etc, and is used for explicit policy interpretation and composition. Old variables remain so that they can be exported by legacy sysctls for use by old netstat(1) binaries. A new sysctl and tunable, netisr.dispatch.policy, accepts the above strings for specifying a global policy default. - The protocol registration structure, netisr_handler, grows an nh_dispatch field, which accepts a per-policy policy override. The default value is '0', which corresponds to "default", meaning that protocols will accept the global default policy unless otherwise specified. - Policies are now interpreted and composed explicitly at various points in packet dispatch; protocol policies override global policies. - Protocols grow the ability to express a non-opinion about affinity even when implenting m2cpuid by returning NETISR_CPUID_NONE. In that case, the framework falls back on source ordering, rather than simply using the current CPU. These changes are in support of allowing link layer re-dispatch based on RSS or similar hashes provided by NICs, especially in the case where the number of hardware receive queues matches hardware core count, rather than hardware thread count, requiring further software redistributeon. (i.e., on RMI XLR). MFC after: 3 weeks Reviewed by: bz Sponsored by: Juniper Networks, Inc.
2011-05-24 12:34:19 +00:00
snpp->snp_dispatch = npp->np_dispatch;
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
if (npp->np_m2flow != NULL)
snpp->snp_flags |= NETISR_SNP_FLAGS_M2FLOW;
if (npp->np_m2cpuid != NULL)
snpp->snp_flags |= NETISR_SNP_FLAGS_M2CPUID;
if (npp->np_drainedcpu != NULL)
snpp->snp_flags |= NETISR_SNP_FLAGS_DRAINEDCPU;
counter++;
}
NETISR_RUNLOCK(&tracker);
Fix edge cases in several KASSERTs: use <= rather than < when testing that counters have not gone about MAXCPU or NETISR_MAXPROT. These problems caused panics on UP kernels with INVARIANTS when using sysctl -a, but would also have caused problems for 32-core boxes or if the netisr protocol vector was fully populated. Reported by: nwhitehorn, Neel Natu <neelnatu@gmail.com> MFC after: 4 days
2010-02-25 09:51:14 +00:00
KASSERT(counter <= NETISR_MAXPROT,
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
("sysctl_netisr_proto: counter too big (%d)", counter));
error = SYSCTL_OUT(req, snp_array, sizeof(*snp_array) * counter);
free(snp_array, M_TEMP);
return (error);
}
SYSCTL_PROC(_net_isr, OID_AUTO, proto,
CTLFLAG_RD|CTLTYPE_STRUCT|CTLFLAG_MPSAFE, 0, 0, sysctl_netisr_proto,
"S,sysctl_netisr_proto",
"Return list of protocols registered with netisr");
/*
* Sysctl monitoring for netisr: query a list of workstreams.
*/
static int
sysctl_netisr_workstream(SYSCTL_HANDLER_ARGS)
{
struct rm_priotracker tracker;
struct sysctl_netisr_workstream *snwsp, *snws_array;
struct netisr_workstream *nwsp;
u_int counter, cpuid;
int error;
if (req->newptr != NULL)
return (EINVAL);
snws_array = malloc(sizeof(*snws_array) * MAXCPU, M_TEMP,
M_ZERO | M_WAITOK);
counter = 0;
NETISR_RLOCK(&tracker);
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(cpuid) {
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
nwsp = DPCPU_ID_PTR(cpuid, nws);
if (nwsp->nws_intr_event == NULL)
continue;
NWS_LOCK(nwsp);
snwsp = &snws_array[counter];
snwsp->snws_version = sizeof(*snwsp);
/*
* For now, we equate workstream IDs and CPU IDs in the
* kernel, but expose them independently to userspace in case
* that assumption changes in the future.
*/
snwsp->snws_wsid = cpuid;
snwsp->snws_cpu = cpuid;
if (nwsp->nws_intr_event != NULL)
snwsp->snws_flags |= NETISR_SNWS_FLAGS_INTR;
NWS_UNLOCK(nwsp);
counter++;
}
NETISR_RUNLOCK(&tracker);
Fix edge cases in several KASSERTs: use <= rather than < when testing that counters have not gone about MAXCPU or NETISR_MAXPROT. These problems caused panics on UP kernels with INVARIANTS when using sysctl -a, but would also have caused problems for 32-core boxes or if the netisr protocol vector was fully populated. Reported by: nwhitehorn, Neel Natu <neelnatu@gmail.com> MFC after: 4 days
2010-02-25 09:51:14 +00:00
KASSERT(counter <= MAXCPU,
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
("sysctl_netisr_workstream: counter too big (%d)", counter));
error = SYSCTL_OUT(req, snws_array, sizeof(*snws_array) * counter);
free(snws_array, M_TEMP);
return (error);
}
SYSCTL_PROC(_net_isr, OID_AUTO, workstream,
CTLFLAG_RD|CTLTYPE_STRUCT|CTLFLAG_MPSAFE, 0, 0, sysctl_netisr_workstream,
"S,sysctl_netisr_workstream",
"Return list of workstreams implemented by netisr");
/*
* Sysctl monitoring for netisr: query per-protocol data across all
* workstreams.
*/
static int
sysctl_netisr_work(SYSCTL_HANDLER_ARGS)
{
struct rm_priotracker tracker;
struct sysctl_netisr_work *snwp, *snw_array;
struct netisr_workstream *nwsp;
struct netisr_proto *npp;
struct netisr_work *nwp;
u_int counter, cpuid, proto;
int error;
if (req->newptr != NULL)
return (EINVAL);
snw_array = malloc(sizeof(*snw_array) * MAXCPU * NETISR_MAXPROT,
M_TEMP, M_ZERO | M_WAITOK);
counter = 0;
NETISR_RLOCK(&tracker);
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(cpuid) {
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
nwsp = DPCPU_ID_PTR(cpuid, nws);
if (nwsp->nws_intr_event == NULL)
continue;
NWS_LOCK(nwsp);
for (proto = 0; proto < NETISR_MAXPROT; proto++) {
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
npp = &netisr_proto[proto];
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
if (npp->np_name == NULL)
continue;
nwp = &nwsp->nws_work[proto];
snwp = &snw_array[counter];
snwp->snw_version = sizeof(*snwp);
snwp->snw_wsid = cpuid; /* See comment above. */
snwp->snw_proto = proto;
snwp->snw_len = nwp->nw_len;
snwp->snw_watermark = nwp->nw_watermark;
snwp->snw_dispatched = nwp->nw_dispatched;
snwp->snw_hybrid_dispatched =
nwp->nw_hybrid_dispatched;
snwp->snw_qdrops = nwp->nw_qdrops;
snwp->snw_queued = nwp->nw_queued;
snwp->snw_handled = nwp->nw_handled;
counter++;
}
NWS_UNLOCK(nwsp);
}
Fix edge cases in several KASSERTs: use <= rather than < when testing that counters have not gone about MAXCPU or NETISR_MAXPROT. These problems caused panics on UP kernels with INVARIANTS when using sysctl -a, but would also have caused problems for 32-core boxes or if the netisr protocol vector was fully populated. Reported by: nwhitehorn, Neel Natu <neelnatu@gmail.com> MFC after: 4 days
2010-02-25 09:51:14 +00:00
KASSERT(counter <= MAXCPU * NETISR_MAXPROT,
Export netisr configuration and statistics to userspace via sysctl(9). MFC after: 1 week Sponsored by: Juniper Networks
2010-02-22 15:03:16 +00:00
("sysctl_netisr_work: counter too big (%d)", counter));
NETISR_RUNLOCK(&tracker);
error = SYSCTL_OUT(req, snw_array, sizeof(*snw_array) * counter);
free(snw_array, M_TEMP);
return (error);
}
SYSCTL_PROC(_net_isr, OID_AUTO, work,
CTLFLAG_RD|CTLTYPE_STRUCT|CTLFLAG_MPSAFE, 0, 0, sysctl_netisr_work,
"S,sysctl_netisr_work",
"Return list of per-workstream, per-protocol work in netisr");
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
#ifdef DDB
DB_SHOW_COMMAND(netisr, db_show_netisr)
{
struct netisr_workstream *nwsp;
struct netisr_work *nwp;
int first, proto;
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
u_int cpuid;
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
db_printf("%3s %6s %5s %5s %5s %8s %8s %8s %8s\n", "CPU", "Proto",
"Len", "WMark", "Max", "Disp", "HDisp", "Drop", "Queue");
Update several places that iterate over CPUs to use CPU_FOREACH().
2010-06-11 18:46:34 +00:00
CPU_FOREACH(cpuid) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
nwsp = DPCPU_ID_PTR(cpuid, nws);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
if (nwsp->nws_intr_event == NULL)
continue;
first = 1;
for (proto = 0; proto < NETISR_MAXPROT; proto++) {
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
if (netisr_proto[proto].np_handler == NULL)
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
continue;
nwp = &nwsp->nws_work[proto];
if (first) {
Convert netisr to use dynamic per-CPU storage (DPCPU) instead of sizing arrays to [MAXCPU], offering moderate memory savings. In some places, this requires using CPU_ABSENT() to handle less common platforms with sparse CPU IDs. In several places, assert that the selected CPUID for work placement or statistics is not CPU_ABSENT() to be on the safe side. Discussed with: bz, jeff
2009-06-26 00:19:25 +00:00
db_printf("%3d ", cpuid);
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
first = 0;
} else
db_printf("%3s ", "");
db_printf(
"%6s %5d %5d %5d %8ju %8ju %8ju %8ju\n",
Changes to support crashdump analysis of netisr: - Rename the netisr protocol registration array, 'np' to 'netisr_proto', in order to reduce the chances of symbol name collisions. It remains statically defined, but it will be looked up by netstat(1). - Move certain internal structure definitions from netisr.c to netisr_internal.h so that netstat(1) can find them. They remain private, and should not be used for any other purpose (for example, they should not be used by kernel modules, which must instead use the public interfaces in netisr.h). - Store a kernel-compiled version of NETISR_MAXPROT in the global variable netisr_maxprot, and export via a sysctl, so that it is available for use by netstat(1). This is especially important for crashdump interpretation, where the size of the workstream structure is determined by the maximum number of protocols compiled into the kernel. MFC after: 1 week Sponsored by: Juniper Networks
2010-03-01 00:42:36 +00:00
netisr_proto[proto].np_name, nwp->nw_len,
Reimplement the netisr framework in order to support parallel netisr threads: - Support up to one netisr thread per CPU, each processings its own workstream, or set of per-protocol queues. Threads may be bound to specific CPUs, or allowed to migrate, based on a global policy. In the future it would be desirable to support topology-centric policies, such as "one netisr per package". - Allow each protocol to advertise an ordering policy, which can currently be one of: NETISR_POLICY_SOURCE: packets must maintain ordering with respect to an implicit or explicit source (such as an interface or socket). NETISR_POLICY_FLOW: make use of mbuf flow identifiers to place work, as well as allowing protocols to provide a flow generation function for mbufs without flow identifers (m2flow). Falls back on NETISR_POLICY_SOURCE if now flow ID is available. NETISR_POLICY_CPU: allow protocols to inspect and assign a CPU for each packet handled by netisr (m2cpuid). - Provide utility functions for querying the number of workstreams being used, as well as a mapping function from workstream to CPU ID, which protocols may use in work placement decisions. - Add explicit interfaces to get and set per-protocol queue limits, and get and clear drop counters, which query data or apply changes across all workstreams. - Add a more extensible netisr registration interface, in which protocols declare 'struct netisr_handler' structures for each registered NETISR_ type. These include name, handler function, optional mbuf to flow ID function, optional mbuf to CPU ID function, queue limit, and ordering policy. Padding is present to allow these to be expanded in the future. If no queue limit is declared, then a default is used. - Queue limits are now per-workstream, and raised from the previous IFQ_MAXLEN default of 50 to 256. - All protocols are updated to use the new registration interface, and with the exception of netnatm, default queue limits. Most protocols register as NETISR_POLICY_SOURCE, except IPv4 and IPv6, which use NETISR_POLICY_FLOW, and will therefore take advantage of driver- generated flow IDs if present. - Formalize a non-packet based interface between interface polling and the netisr, rather than having polling pretend to be two protocols. Provide two explicit hooks in the netisr worker for start and end events for runs: netisr_poll() and netisr_pollmore(), as well as a function, netisr_sched_poll(), to allow the polling code to schedule netisr execution. DEVICE_POLLING still embeds single-netisr assumptions in its implementation, so for now if it is compiled into the kernel, a single and un-bound netisr thread is enforced regardless of tunable configuration. In the default configuration, the new netisr implementation maintains the same basic assumptions as the previous implementation: a single, un-bound worker thread processes all deferred work, and direct dispatch is enabled by default wherever possible. Performance measurement shows a marginal performance improvement over the old implementation due to the use of batched dequeue. An rmlock is used to synchronize use and registration/unregistration using the framework; currently, synchronized use is disabled (replicating current netisr policy) due to a measurable 3%-6% hit in ping-pong micro-benchmarking. It will be enabled once further rmlock optimization has taken place. However, in practice, netisrs are rarely registered or unregistered at runtime. A new man page for netisr will follow, but since one doesn't currently exist, it hasn't been updated. This change is not appropriate for MFC, although the polling shutdown handler should be merged to 7-STABLE. Bump __FreeBSD_version. Reviewed by: bz
2009-06-01 10:41:38 +00:00
nwp->nw_watermark, nwp->nw_qlimit,
nwp->nw_dispatched, nwp->nw_hybrid_dispatched,
nwp->nw_qdrops, nwp->nw_queued);
}
}
}
#endif
Reference in New Issue Copy Permalink