gem5/src/mem/xbar.cc

629 lines
22 KiB
C++
Raw Normal View History

/*
* Copyright (c) 2011-2015 ARM Limited
* All rights reserved
*
* The license below extends only to copyright in the software and shall
* not be construed as granting a license to any other intellectual
* property including but not limited to intellectual property relating
* to a hardware implementation of the functionality of the software
* licensed hereunder. You may use the software subject to the license
* terms below provided that you ensure that this notice is replicated
* unmodified and in its entirety in all distributions of the software,
* modified or unmodified, in source code or in binary form.
*
* Copyright (c) 2006 The Regents of The University of Michigan
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met: redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer;
* 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;
* neither the name of the copyright holders nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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 COPYRIGHT
* OWNER 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.
*
* Authors: Ali Saidi
* Andreas Hansson
* William Wang
*/
/**
* @file
* Definition of a crossbar object.
*/
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
#include "base/misc.hh"
#include "base/trace.hh"
#include "debug/AddrRanges.hh"
#include "debug/Drain.hh"
#include "debug/XBar.hh"
#include "mem/xbar.hh"
BaseXBar::BaseXBar(const BaseXBarParams *p)
: MemObject(p),
frontendLatency(p->frontend_latency),
forwardLatency(p->forward_latency),
responseLatency(p->response_latency),
width(p->width),
gotAddrRanges(p->port_default_connection_count +
p->port_master_connection_count, false),
gotAllAddrRanges(false), defaultPortID(InvalidPortID),
useDefaultRange(p->use_default_range)
{}
BaseXBar::~BaseXBar()
Bus: Split the bus into a non-coherent and coherent bus This patch introduces a class hierarchy of buses, a non-coherent one, and a coherent one, splitting the existing bus functionality. By doing so it also enables further specialisation of the two types of buses. A non-coherent bus connects a number of non-snooping masters and slaves, and routes the request and response packets based on the address. The request packets issued by the master connected to a non-coherent bus could still snoop in caches attached to a coherent bus, as is the case with the I/O bus and memory bus in most system configurations. No snoops will, however, reach any master on the non-coherent bus itself. The non-coherent bus can be used as a template for modelling PCI, PCIe, and non-coherent AMBA and OCP buses, and is typically used for the I/O buses. A coherent bus connects a number of (potentially) snooping masters and slaves, and routes the request and response packets based on the address, and also forwards all requests to the snoopers and deals with the snoop responses. The coherent bus can be used as a template for modelling QPI, HyperTransport, ACE and coherent OCP buses, and is typically used for the L1-to-L2 buses and as the main system interconnect. The configuration scripts are updated to use a NoncoherentBus for all peripheral and I/O buses. A bit of minor tidying up has also been done. --HG-- rename : src/mem/bus.cc => src/mem/coherent_bus.cc rename : src/mem/bus.hh => src/mem/coherent_bus.hh rename : src/mem/bus.cc => src/mem/noncoherent_bus.cc rename : src/mem/bus.hh => src/mem/noncoherent_bus.hh
2012-05-31 19:30:04 +02:00
{
for (auto m: masterPorts)
delete m;
for (auto s: slavePorts)
delete s;
}
void
BaseXBar::init()
{
}
BaseMasterPort &
BaseXBar::getMasterPort(const std::string &if_name, PortID idx)
{
MEM: Separate snoops and normal memory requests/responses This patch introduces port access methods that separates snoop request/responses from normal memory request/responses. The differentiation is made for functional, atomic and timing accesses and builds on the introduction of master and slave ports. Before the introduction of this patch, the packets belonging to the different phases of the protocol (request -> [forwarded snoop request -> snoop response]* -> response) all use the same port access functions, even though the snoop packets flow in the opposite direction to the normal packet. That is, a coherent master sends normal request and receives responses, but receives snoop requests and sends snoop responses (vice versa for the slave). These two distinct phases now use different access functions, as described below. Starting with the functional access, a master sends a request to a slave through sendFunctional, and the request packet is turned into a response before the call returns. In a system without cache coherence, this is all that is needed from the functional interface. For the cache-coherent scenario, a slave also sends snoop requests to coherent masters through sendFunctionalSnoop, with responses returned within the same packet pointer. This is currently used by the bus and caches, and the LSQ of the O3 CPU. The send/recvFunctional and send/recvFunctionalSnoop are moved from the Port super class to the appropriate subclass. Atomic accesses follow the same flow as functional accesses, with request being sent from master to slave through sendAtomic. In the case of cache-coherent ports, a slave can send snoop requests to a master through sendAtomicSnoop. Just as for the functional access methods, the atomic send and receive member functions are moved to the appropriate subclasses. The timing access methods are different from the functional and atomic in that requests and responses are separated in time and send/recvTiming are used for both directions. Hence, a master uses sendTiming to send a request to a slave, and a slave uses sendTiming to send a response back to a master, at a later point in time. Snoop requests and responses travel in the opposite direction, similar to what happens in functional and atomic accesses. With the introduction of this patch, it is possible to determine the direction of packets in the bus, and no longer necessary to look for both a master and a slave port with the requested port id. In contrast to the normal recvFunctional, recvAtomic and recvTiming that are pure virtual functions, the recvFunctionalSnoop, recvAtomicSnoop and recvTimingSnoop have a default implementation that calls panic. This is to allow non-coherent master and slave ports to not implement these functions.
2012-04-14 11:45:07 +02:00
if (if_name == "master" && idx < masterPorts.size()) {
// the master port index translates directly to the vector position
return *masterPorts[idx];
} else if (if_name == "default") {
return *masterPorts[defaultPortID];
} else {
return MemObject::getMasterPort(if_name, idx);
}
}
BaseSlavePort &
BaseXBar::getSlavePort(const std::string &if_name, PortID idx)
{
MEM: Separate snoops and normal memory requests/responses This patch introduces port access methods that separates snoop request/responses from normal memory request/responses. The differentiation is made for functional, atomic and timing accesses and builds on the introduction of master and slave ports. Before the introduction of this patch, the packets belonging to the different phases of the protocol (request -> [forwarded snoop request -> snoop response]* -> response) all use the same port access functions, even though the snoop packets flow in the opposite direction to the normal packet. That is, a coherent master sends normal request and receives responses, but receives snoop requests and sends snoop responses (vice versa for the slave). These two distinct phases now use different access functions, as described below. Starting with the functional access, a master sends a request to a slave through sendFunctional, and the request packet is turned into a response before the call returns. In a system without cache coherence, this is all that is needed from the functional interface. For the cache-coherent scenario, a slave also sends snoop requests to coherent masters through sendFunctionalSnoop, with responses returned within the same packet pointer. This is currently used by the bus and caches, and the LSQ of the O3 CPU. The send/recvFunctional and send/recvFunctionalSnoop are moved from the Port super class to the appropriate subclass. Atomic accesses follow the same flow as functional accesses, with request being sent from master to slave through sendAtomic. In the case of cache-coherent ports, a slave can send snoop requests to a master through sendAtomicSnoop. Just as for the functional access methods, the atomic send and receive member functions are moved to the appropriate subclasses. The timing access methods are different from the functional and atomic in that requests and responses are separated in time and send/recvTiming are used for both directions. Hence, a master uses sendTiming to send a request to a slave, and a slave uses sendTiming to send a response back to a master, at a later point in time. Snoop requests and responses travel in the opposite direction, similar to what happens in functional and atomic accesses. With the introduction of this patch, it is possible to determine the direction of packets in the bus, and no longer necessary to look for both a master and a slave port with the requested port id. In contrast to the normal recvFunctional, recvAtomic and recvTiming that are pure virtual functions, the recvFunctionalSnoop, recvAtomicSnoop and recvTimingSnoop have a default implementation that calls panic. This is to allow non-coherent master and slave ports to not implement these functions.
2012-04-14 11:45:07 +02:00
if (if_name == "slave" && idx < slavePorts.size()) {
// the slave port index translates directly to the vector position
return *slavePorts[idx];
} else {
return MemObject::getSlavePort(if_name, idx);
}
}
void
BaseXBar::calcPacketTiming(PacketPtr pkt, Tick header_delay)
{
// the crossbar will be called at a time that is not necessarily
// coinciding with its own clock, so start by determining how long
// until the next clock edge (could be zero)
Tick offset = clockEdge() - curTick();
// the header delay depends on the path through the crossbar, and
// we therefore rely on the caller to provide the actual
// value
pkt->headerDelay += offset + header_delay;
// note that we add the header delay to the existing value, and
// align it to the crossbar clock
// do a quick sanity check to ensure the timings are not being
// ignored, note that this specific value may cause problems for
// slower interconnects
panic_if(pkt->headerDelay > SimClock::Int::us,
"Encountered header delay exceeding 1 us\n");
if (pkt->hasData()) {
// the payloadDelay takes into account the relative time to
// deliver the payload of the packet, after the header delay,
// we take the maximum since the payload delay could already
// be longer than what this parcitular crossbar enforces.
pkt->payloadDelay = std::max<Tick>(pkt->payloadDelay,
divCeil(pkt->getSize(), width) *
clockPeriod());
}
// the payload delay is not paying for the clock offset as that is
// already done using the header delay, and the payload delay is
// also used to determine how long the crossbar layer is busy and
// thus regulates throughput
}
template <typename SrcType, typename DstType>
BaseXBar::Layer<SrcType,DstType>::Layer(DstType& _port, BaseXBar& _xbar,
const std::string& _name) :
port(_port), xbar(_xbar), _name(_name), state(IDLE),
waitingForPeer(NULL), releaseEvent(this)
{
}
template <typename SrcType, typename DstType>
void BaseXBar::Layer<SrcType,DstType>::occupyLayer(Tick until)
{
// ensure the state is busy at this point, as the layer should
// transition from idle as soon as it has decided to forward the
// packet to prevent any follow-on calls to sendTiming seeing an
// unoccupied layer
assert(state == BUSY);
// until should never be 0 as express snoops never occupy the layer
assert(until != 0);
xbar.schedule(releaseEvent, until);
// account for the occupied ticks
occupancy += until - curTick();
DPRINTF(BaseXBar, "The crossbar layer is now busy from tick %d to %d\n",
curTick(), until);
}
template <typename SrcType, typename DstType>
bool
BaseXBar::Layer<SrcType,DstType>::tryTiming(SrcType* src_port)
{
// if we are in the retry state, we will not see anything but the
// retrying port (or in the case of the snoop ports the snoop
// response port that mirrors the actual slave port) as we leave
// this state again in zero time if the peer does not immediately
// call the layer when receiving the retry
// first we see if the layer is busy, next we check if the
// destination port is already engaged in a transaction waiting
// for a retry from the peer
if (state == BUSY || waitingForPeer != NULL) {
// the port should not be waiting already
assert(std::find(waitingForLayer.begin(), waitingForLayer.end(),
src_port) == waitingForLayer.end());
// put the port at the end of the retry list waiting for the
// layer to be freed up (and in the case of a busy peer, for
// that transaction to go through, and then the layer to free
// up)
waitingForLayer.push_back(src_port);
return false;
MEM: Separate snoops and normal memory requests/responses This patch introduces port access methods that separates snoop request/responses from normal memory request/responses. The differentiation is made for functional, atomic and timing accesses and builds on the introduction of master and slave ports. Before the introduction of this patch, the packets belonging to the different phases of the protocol (request -> [forwarded snoop request -> snoop response]* -> response) all use the same port access functions, even though the snoop packets flow in the opposite direction to the normal packet. That is, a coherent master sends normal request and receives responses, but receives snoop requests and sends snoop responses (vice versa for the slave). These two distinct phases now use different access functions, as described below. Starting with the functional access, a master sends a request to a slave through sendFunctional, and the request packet is turned into a response before the call returns. In a system without cache coherence, this is all that is needed from the functional interface. For the cache-coherent scenario, a slave also sends snoop requests to coherent masters through sendFunctionalSnoop, with responses returned within the same packet pointer. This is currently used by the bus and caches, and the LSQ of the O3 CPU. The send/recvFunctional and send/recvFunctionalSnoop are moved from the Port super class to the appropriate subclass. Atomic accesses follow the same flow as functional accesses, with request being sent from master to slave through sendAtomic. In the case of cache-coherent ports, a slave can send snoop requests to a master through sendAtomicSnoop. Just as for the functional access methods, the atomic send and receive member functions are moved to the appropriate subclasses. The timing access methods are different from the functional and atomic in that requests and responses are separated in time and send/recvTiming are used for both directions. Hence, a master uses sendTiming to send a request to a slave, and a slave uses sendTiming to send a response back to a master, at a later point in time. Snoop requests and responses travel in the opposite direction, similar to what happens in functional and atomic accesses. With the introduction of this patch, it is possible to determine the direction of packets in the bus, and no longer necessary to look for both a master and a slave port with the requested port id. In contrast to the normal recvFunctional, recvAtomic and recvTiming that are pure virtual functions, the recvFunctionalSnoop, recvAtomicSnoop and recvTimingSnoop have a default implementation that calls panic. This is to allow non-coherent master and slave ports to not implement these functions.
2012-04-14 11:45:07 +02:00
}
state = BUSY;
return true;
MEM: Separate snoops and normal memory requests/responses This patch introduces port access methods that separates snoop request/responses from normal memory request/responses. The differentiation is made for functional, atomic and timing accesses and builds on the introduction of master and slave ports. Before the introduction of this patch, the packets belonging to the different phases of the protocol (request -> [forwarded snoop request -> snoop response]* -> response) all use the same port access functions, even though the snoop packets flow in the opposite direction to the normal packet. That is, a coherent master sends normal request and receives responses, but receives snoop requests and sends snoop responses (vice versa for the slave). These two distinct phases now use different access functions, as described below. Starting with the functional access, a master sends a request to a slave through sendFunctional, and the request packet is turned into a response before the call returns. In a system without cache coherence, this is all that is needed from the functional interface. For the cache-coherent scenario, a slave also sends snoop requests to coherent masters through sendFunctionalSnoop, with responses returned within the same packet pointer. This is currently used by the bus and caches, and the LSQ of the O3 CPU. The send/recvFunctional and send/recvFunctionalSnoop are moved from the Port super class to the appropriate subclass. Atomic accesses follow the same flow as functional accesses, with request being sent from master to slave through sendAtomic. In the case of cache-coherent ports, a slave can send snoop requests to a master through sendAtomicSnoop. Just as for the functional access methods, the atomic send and receive member functions are moved to the appropriate subclasses. The timing access methods are different from the functional and atomic in that requests and responses are separated in time and send/recvTiming are used for both directions. Hence, a master uses sendTiming to send a request to a slave, and a slave uses sendTiming to send a response back to a master, at a later point in time. Snoop requests and responses travel in the opposite direction, similar to what happens in functional and atomic accesses. With the introduction of this patch, it is possible to determine the direction of packets in the bus, and no longer necessary to look for both a master and a slave port with the requested port id. In contrast to the normal recvFunctional, recvAtomic and recvTiming that are pure virtual functions, the recvFunctionalSnoop, recvAtomicSnoop and recvTimingSnoop have a default implementation that calls panic. This is to allow non-coherent master and slave ports to not implement these functions.
2012-04-14 11:45:07 +02:00
}
template <typename SrcType, typename DstType>
MEM: Separate requests and responses for timing accesses This patch moves send/recvTiming and send/recvTimingSnoop from the Port base class to the MasterPort and SlavePort, and also splits them into separate member functions for requests and responses: send/recvTimingReq, send/recvTimingResp, and send/recvTimingSnoopReq, send/recvTimingSnoopResp. A master port sends requests and receives responses, and also receives snoop requests and sends snoop responses. A slave port has the reciprocal behaviour as it receives requests and sends responses, and sends snoop requests and receives snoop responses. For all MemObjects that have only master ports or slave ports (but not both), e.g. a CPU, or a PIO device, this patch merely adds more clarity to what kind of access is taking place. For example, a CPU port used to call sendTiming, and will now call sendTimingReq. Similarly, a response previously came back through recvTiming, which is now recvTimingResp. For the modules that have both master and slave ports, e.g. the bus, the behaviour was previously relying on branches based on pkt->isRequest(), and this is now replaced with a direct call to the apprioriate member function depending on the type of access. Please note that send/recvRetry is still shared by all the timing accessors and remains in the Port base class for now (to maintain the current bus functionality and avoid changing the statistics of all regressions). The packet queue is split into a MasterPort and SlavePort version to facilitate the use of the new timing accessors. All uses of the PacketQueue are updated accordingly. With this patch, the type of packet (request or response) is now well defined for each type of access, and asserts on pkt->isRequest() and pkt->isResponse() are now moved to the appropriate send member functions. It is also worth noting that sendTimingSnoopReq no longer returns a boolean, as the semantics do not alow snoop requests to be rejected or stalled. All these assumptions are now excplicitly part of the port interface itself.
2012-05-01 19:40:42 +02:00
void
BaseXBar::Layer<SrcType,DstType>::succeededTiming(Tick busy_time)
MEM: Separate snoops and normal memory requests/responses This patch introduces port access methods that separates snoop request/responses from normal memory request/responses. The differentiation is made for functional, atomic and timing accesses and builds on the introduction of master and slave ports. Before the introduction of this patch, the packets belonging to the different phases of the protocol (request -> [forwarded snoop request -> snoop response]* -> response) all use the same port access functions, even though the snoop packets flow in the opposite direction to the normal packet. That is, a coherent master sends normal request and receives responses, but receives snoop requests and sends snoop responses (vice versa for the slave). These two distinct phases now use different access functions, as described below. Starting with the functional access, a master sends a request to a slave through sendFunctional, and the request packet is turned into a response before the call returns. In a system without cache coherence, this is all that is needed from the functional interface. For the cache-coherent scenario, a slave also sends snoop requests to coherent masters through sendFunctionalSnoop, with responses returned within the same packet pointer. This is currently used by the bus and caches, and the LSQ of the O3 CPU. The send/recvFunctional and send/recvFunctionalSnoop are moved from the Port super class to the appropriate subclass. Atomic accesses follow the same flow as functional accesses, with request being sent from master to slave through sendAtomic. In the case of cache-coherent ports, a slave can send snoop requests to a master through sendAtomicSnoop. Just as for the functional access methods, the atomic send and receive member functions are moved to the appropriate subclasses. The timing access methods are different from the functional and atomic in that requests and responses are separated in time and send/recvTiming are used for both directions. Hence, a master uses sendTiming to send a request to a slave, and a slave uses sendTiming to send a response back to a master, at a later point in time. Snoop requests and responses travel in the opposite direction, similar to what happens in functional and atomic accesses. With the introduction of this patch, it is possible to determine the direction of packets in the bus, and no longer necessary to look for both a master and a slave port with the requested port id. In contrast to the normal recvFunctional, recvAtomic and recvTiming that are pure virtual functions, the recvFunctionalSnoop, recvAtomicSnoop and recvTimingSnoop have a default implementation that calls panic. This is to allow non-coherent master and slave ports to not implement these functions.
2012-04-14 11:45:07 +02:00
{
// we should have gone from idle or retry to busy in the tryTiming
// test
assert(state == BUSY);
// occupy the layer accordingly
occupyLayer(busy_time);
}
template <typename SrcType, typename DstType>
void
BaseXBar::Layer<SrcType,DstType>::failedTiming(SrcType* src_port,
Tick busy_time)
{
// ensure no one got in between and tried to send something to
// this port
assert(waitingForPeer == NULL);
// if the source port is the current retrying one or not, we have
// failed in forwarding and should track that we are now waiting
// for the peer to send a retry
waitingForPeer = src_port;
// we should have gone from idle or retry to busy in the tryTiming
// test
assert(state == BUSY);
// occupy the bus accordingly
occupyLayer(busy_time);
MEM: Separate snoops and normal memory requests/responses This patch introduces port access methods that separates snoop request/responses from normal memory request/responses. The differentiation is made for functional, atomic and timing accesses and builds on the introduction of master and slave ports. Before the introduction of this patch, the packets belonging to the different phases of the protocol (request -> [forwarded snoop request -> snoop response]* -> response) all use the same port access functions, even though the snoop packets flow in the opposite direction to the normal packet. That is, a coherent master sends normal request and receives responses, but receives snoop requests and sends snoop responses (vice versa for the slave). These two distinct phases now use different access functions, as described below. Starting with the functional access, a master sends a request to a slave through sendFunctional, and the request packet is turned into a response before the call returns. In a system without cache coherence, this is all that is needed from the functional interface. For the cache-coherent scenario, a slave also sends snoop requests to coherent masters through sendFunctionalSnoop, with responses returned within the same packet pointer. This is currently used by the bus and caches, and the LSQ of the O3 CPU. The send/recvFunctional and send/recvFunctionalSnoop are moved from the Port super class to the appropriate subclass. Atomic accesses follow the same flow as functional accesses, with request being sent from master to slave through sendAtomic. In the case of cache-coherent ports, a slave can send snoop requests to a master through sendAtomicSnoop. Just as for the functional access methods, the atomic send and receive member functions are moved to the appropriate subclasses. The timing access methods are different from the functional and atomic in that requests and responses are separated in time and send/recvTiming are used for both directions. Hence, a master uses sendTiming to send a request to a slave, and a slave uses sendTiming to send a response back to a master, at a later point in time. Snoop requests and responses travel in the opposite direction, similar to what happens in functional and atomic accesses. With the introduction of this patch, it is possible to determine the direction of packets in the bus, and no longer necessary to look for both a master and a slave port with the requested port id. In contrast to the normal recvFunctional, recvAtomic and recvTiming that are pure virtual functions, the recvFunctionalSnoop, recvAtomicSnoop and recvTimingSnoop have a default implementation that calls panic. This is to allow non-coherent master and slave ports to not implement these functions.
2012-04-14 11:45:07 +02:00
}
template <typename SrcType, typename DstType>
MEM: Separate snoops and normal memory requests/responses This patch introduces port access methods that separates snoop request/responses from normal memory request/responses. The differentiation is made for functional, atomic and timing accesses and builds on the introduction of master and slave ports. Before the introduction of this patch, the packets belonging to the different phases of the protocol (request -> [forwarded snoop request -> snoop response]* -> response) all use the same port access functions, even though the snoop packets flow in the opposite direction to the normal packet. That is, a coherent master sends normal request and receives responses, but receives snoop requests and sends snoop responses (vice versa for the slave). These two distinct phases now use different access functions, as described below. Starting with the functional access, a master sends a request to a slave through sendFunctional, and the request packet is turned into a response before the call returns. In a system without cache coherence, this is all that is needed from the functional interface. For the cache-coherent scenario, a slave also sends snoop requests to coherent masters through sendFunctionalSnoop, with responses returned within the same packet pointer. This is currently used by the bus and caches, and the LSQ of the O3 CPU. The send/recvFunctional and send/recvFunctionalSnoop are moved from the Port super class to the appropriate subclass. Atomic accesses follow the same flow as functional accesses, with request being sent from master to slave through sendAtomic. In the case of cache-coherent ports, a slave can send snoop requests to a master through sendAtomicSnoop. Just as for the functional access methods, the atomic send and receive member functions are moved to the appropriate subclasses. The timing access methods are different from the functional and atomic in that requests and responses are separated in time and send/recvTiming are used for both directions. Hence, a master uses sendTiming to send a request to a slave, and a slave uses sendTiming to send a response back to a master, at a later point in time. Snoop requests and responses travel in the opposite direction, similar to what happens in functional and atomic accesses. With the introduction of this patch, it is possible to determine the direction of packets in the bus, and no longer necessary to look for both a master and a slave port with the requested port id. In contrast to the normal recvFunctional, recvAtomic and recvTiming that are pure virtual functions, the recvFunctionalSnoop, recvAtomicSnoop and recvTimingSnoop have a default implementation that calls panic. This is to allow non-coherent master and slave ports to not implement these functions.
2012-04-14 11:45:07 +02:00
void
BaseXBar::Layer<SrcType,DstType>::releaseLayer()
Add a very poor implementation of dealing with retries on timing requests. It is especially slow with tracing on since it ends up being O(N^2). But it's probably going to have to change for the real bus anyway, so it should be rewritten then Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Port Blocked/Unblocked and replaced with sendRetry(). Remove possibility of packet mangling if packet is going to be refused anyway in bridge src/cpu/simple/atomic.cc: src/cpu/simple/atomic.hh: src/cpu/simple/timing.cc: src/cpu/simple/timing.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) src/dev/io_device.cc: src/dev/io_device.hh: Make DMA Timing requests/responses work. Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) src/mem/bridge.cc: src/mem/bridge.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Port Blocked/Unblocked and replaced with sendRetry(). Remove posibility of packet mangling if packet is going to be refused anyway. src/mem/bus.cc: src/mem/bus.hh: Add a very poor implementation of dealing with retries on timing requests. It is especially slow with tracing on since it ends up being O(N^2). But it's probably going to have to change for the real bus anyway, so it should be rewritten then src/mem/port.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Blocked/Unblocked port status, their functionality is really duplicated in the recvRetry() method --HG-- extra : convert_revision : fab613404be54bfa7a4c67572bae7b559169e573
2006-05-31 00:57:42 +02:00
{
// releasing the bus means we should now be idle
assert(state == BUSY);
assert(!releaseEvent.scheduled());
// update the state
state = IDLE;
// bus layer is now idle, so if someone is waiting we can retry
if (!waitingForLayer.empty()) {
// there is no point in sending a retry if someone is still
// waiting for the peer
if (waitingForPeer == NULL)
retryWaiting();
} else if (waitingForPeer == NULL && drainState() == DrainState::Draining) {
DPRINTF(Drain, "Crossbar done draining, signaling drain manager\n");
//If we weren't able to drain before, do it now.
signalDrainDone();
}
Add a very poor implementation of dealing with retries on timing requests. It is especially slow with tracing on since it ends up being O(N^2). But it's probably going to have to change for the real bus anyway, so it should be rewritten then Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Port Blocked/Unblocked and replaced with sendRetry(). Remove possibility of packet mangling if packet is going to be refused anyway in bridge src/cpu/simple/atomic.cc: src/cpu/simple/atomic.hh: src/cpu/simple/timing.cc: src/cpu/simple/timing.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) src/dev/io_device.cc: src/dev/io_device.hh: Make DMA Timing requests/responses work. Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) src/mem/bridge.cc: src/mem/bridge.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Port Blocked/Unblocked and replaced with sendRetry(). Remove posibility of packet mangling if packet is going to be refused anyway. src/mem/bus.cc: src/mem/bus.hh: Add a very poor implementation of dealing with retries on timing requests. It is especially slow with tracing on since it ends up being O(N^2). But it's probably going to have to change for the real bus anyway, so it should be rewritten then src/mem/port.hh: Change recvRetry() to not accept a packet. Sendtiming should be called again (and can respond with false or true) Removed Blocked/Unblocked port status, their functionality is really duplicated in the recvRetry() method --HG-- extra : convert_revision : fab613404be54bfa7a4c67572bae7b559169e573
2006-05-31 00:57:42 +02:00
}
template <typename SrcType, typename DstType>
void
BaseXBar::Layer<SrcType,DstType>::retryWaiting()
{
// this should never be called with no one waiting
assert(!waitingForLayer.empty());
// we always go to retrying from idle
assert(state == IDLE);
// update the state
state = RETRY;
// set the retrying port to the front of the retry list and pop it
// off the list
SrcType* retryingPort = waitingForLayer.front();
waitingForLayer.pop_front();
// tell the port to retry, which in some cases ends up calling the
// layer again
sendRetry(retryingPort);
// If the layer is still in the retry state, sendTiming wasn't
// called in zero time (e.g. the cache does this when a writeback
// is squashed)
if (state == RETRY) {
// update the state to busy and reset the retrying port, we
// have done our bit and sent the retry
state = BUSY;
// occupy the crossbar layer until the next clock edge
occupyLayer(xbar.clockEdge());
}
}
template <typename SrcType, typename DstType>
void
BaseXBar::Layer<SrcType,DstType>::recvRetry()
{
// we should never get a retry without having failed to forward
// something to this port
assert(waitingForPeer != NULL);
// add the port where the failed packet originated to the front of
// the waiting ports for the layer, this allows us to call retry
// on the port immediately if the crossbar layer is idle
waitingForLayer.push_front(waitingForPeer);
// we are no longer waiting for the peer
waitingForPeer = NULL;
// if the layer is idle, retry this port straight away, if we
// are busy, then simply let the port wait for its turn
if (state == IDLE) {
retryWaiting();
} else {
assert(state == BUSY);
}
}
PortID
BaseXBar::findPort(Addr addr)
{
// we should never see any address lookups before we've got the
// ranges of all connected slave modules
assert(gotAllAddrRanges);
// Check the cache
PortID dest_id = checkPortCache(addr);
if (dest_id != InvalidPortID)
return dest_id;
// Check the address map interval tree
auto i = portMap.find(addr);
if (i != portMap.end()) {
dest_id = i->second;
updatePortCache(dest_id, i->first);
return dest_id;
}
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
// Check if this matches the default range
if (useDefaultRange) {
if (defaultRange.contains(addr)) {
DPRINTF(AddrRanges, " found addr %#llx on default\n",
addr);
return defaultPortID;
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
}
} else if (defaultPortID != InvalidPortID) {
DPRINTF(AddrRanges, "Unable to find destination for addr %#llx, "
"will use default port\n", addr);
return defaultPortID;
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
}
// we should use the range for the default port and it did not
// match, or the default port is not set
fatal("Unable to find destination for addr %#llx on %s\n", addr,
name());
}
/** Function called by the port when the crossbar is receiving a range change.*/
void
BaseXBar::recvRangeChange(PortID master_port_id)
{
DPRINTF(AddrRanges, "Received range change from slave port %s\n",
masterPorts[master_port_id]->getSlavePort().name());
// remember that we got a range from this master port and thus the
// connected slave module
gotAddrRanges[master_port_id] = true;
// update the global flag
if (!gotAllAddrRanges) {
// take a logical AND of all the ports and see if we got
// ranges from everyone
gotAllAddrRanges = true;
std::vector<bool>::const_iterator r = gotAddrRanges.begin();
while (gotAllAddrRanges && r != gotAddrRanges.end()) {
gotAllAddrRanges &= *r++;
}
if (gotAllAddrRanges)
DPRINTF(AddrRanges, "Got address ranges from all slaves\n");
}
// note that we could get the range from the default port at any
// point in time, and we cannot assume that the default range is
// set before the other ones are, so we do additional checks once
// all ranges are provided
if (master_port_id == defaultPortID) {
// only update if we are indeed checking ranges for the
// default port since the port might not have a valid range
// otherwise
if (useDefaultRange) {
AddrRangeList ranges = masterPorts[master_port_id]->getAddrRanges();
if (ranges.size() != 1)
fatal("Crossbar %s may only have a single default range",
name());
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
defaultRange = ranges.front();
}
} else {
// the ports are allowed to update their address ranges
// dynamically, so remove any existing entries
if (gotAddrRanges[master_port_id]) {
for (auto p = portMap.begin(); p != portMap.end(); ) {
if (p->second == master_port_id)
// erasing invalidates the iterator, so advance it
// before the deletion takes place
portMap.erase(p++);
else
p++;
}
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
}
fixes for new memory system SConscript: comment out most devices add vport.cc arch/alpha/arguments.cc: arch/alpha/arguments.hh: push in alpha name space fix for new memory system arch/alpha/faults.cc: arch/alpha/faults.hh: Added an unimplemented fault that can be returned if a certain function isn't implemented arch/alpha/freebsd/system.cc: arch/alpha/linux/system.cc: arch/alpha/stacktrace.cc: arch/alpha/system.cc: arch/alpha/tlb.hh: arch/alpha/tru64/system.cc: fixed for new memory system arch/alpha/tlb.cc: fixed for new memory system removed code that seems to have no purpose arch/alpha/vtophys.cc: arch/alpha/vtophys.hh: fixed for new memory system put in namespace AlphaISA base/remote_gdb.cc: fix for new memory system cpu/cpu_exec_context.cc: cpu/cpu_exec_context.hh: cpu/exec_context.hh: create two ports one of physical accesses and one for superpage accesses Add functions getVirtPort() getPhysPort() delVirtPort(). To get statically allocated physical or virtual ports or if an execcontext is passed in get a dynamically allocated virtual port dev/alpha_console.cc: dev/alpha_console.hh: Redo for new memory system dev/io_device.cc: dev/io_device.hh: new I/O devices for new memory system kern/linux/events.cc: kern/linux/printk.cc: kern/linux/printk.hh: kern/tru64/dump_mbuf.hh: kern/tru64/printf.cc: kern/tru64/printf.hh: Arguments now in namespaces kern/tru64/tru64_events.cc: mem/bus.cc: fix for new memory syste mem/physical.hh: new addressranges function getPort should be public mem/port.hh: Add write/read methods to functional port update getDeviceAddrRanges to have a list of both snoops and response lists sim/pseudo_inst.cc: sim/system.cc: sim/system.hh: Update for new mem system sim/vptr.hh: comment out code and replace with panics This will need to be fixed at some point, but it's not easy. --HG-- extra : convert_revision : 41f41f422cfbab3751284d55cccb6ea64a7956e2
2006-04-06 06:51:46 +02:00
AddrRangeList ranges = masterPorts[master_port_id]->getAddrRanges();
for (const auto& r: ranges) {
DPRINTF(AddrRanges, "Adding range %s for id %d\n",
r.to_string(), master_port_id);
if (portMap.insert(r, master_port_id) == portMap.end()) {
PortID conflict_id = portMap.find(r)->second;
fatal("%s has two ports responding within range %s:\n\t%s\n\t%s\n",
name(),
r.to_string(),
masterPorts[master_port_id]->getSlavePort().name(),
masterPorts[conflict_id]->getSlavePort().name());
}
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
}
}
// if we have received ranges from all our neighbouring slave
// modules, go ahead and tell our connected master modules in
// turn, this effectively assumes a tree structure of the system
if (gotAllAddrRanges) {
DPRINTF(AddrRanges, "Aggregating address ranges\n");
xbarRanges.clear();
// start out with the default range
if (useDefaultRange) {
if (!gotAddrRanges[defaultPortID])
fatal("Crossbar %s uses default range, but none provided",
name());
xbarRanges.push_back(defaultRange);
DPRINTF(AddrRanges, "-- Adding default %s\n",
defaultRange.to_string());
}
// merge all interleaved ranges and add any range that is not
// a subset of the default range
std::vector<AddrRange> intlv_ranges;
for (const auto& r: portMap) {
// if the range is interleaved then save it for now
if (r.first.interleaved()) {
// if we already got interleaved ranges that are not
// part of the same range, then first do a merge
// before we add the new one
if (!intlv_ranges.empty() &&
!intlv_ranges.back().mergesWith(r.first)) {
DPRINTF(AddrRanges, "-- Merging range from %d ranges\n",
intlv_ranges.size());
AddrRange merged_range(intlv_ranges);
// next decide if we keep the merged range or not
if (!(useDefaultRange &&
merged_range.isSubset(defaultRange))) {
xbarRanges.push_back(merged_range);
DPRINTF(AddrRanges, "-- Adding merged range %s\n",
merged_range.to_string());
}
intlv_ranges.clear();
}
intlv_ranges.push_back(r.first);
} else {
// keep the current range if not a subset of the default
if (!(useDefaultRange &&
r.first.isSubset(defaultRange))) {
xbarRanges.push_back(r.first);
DPRINTF(AddrRanges, "-- Adding range %s\n",
r.first.to_string());
}
}
}
// if there is still interleaved ranges waiting to be merged,
// go ahead and do it
if (!intlv_ranges.empty()) {
DPRINTF(AddrRanges, "-- Merging range from %d ranges\n",
intlv_ranges.size());
AddrRange merged_range(intlv_ranges);
if (!(useDefaultRange && merged_range.isSubset(defaultRange))) {
xbarRanges.push_back(merged_range);
DPRINTF(AddrRanges, "-- Adding merged range %s\n",
merged_range.to_string());
}
}
// also check that no range partially overlaps with the
// default range, this has to be done after all ranges are set
// as there are no guarantees for when the default range is
// update with respect to the other ones
if (useDefaultRange) {
for (const auto& r: xbarRanges) {
// see if the new range is partially
// overlapping the default range
if (r.intersects(defaultRange) &&
!r.isSubset(defaultRange))
fatal("Range %s intersects the " \
"default range of %s but is not a " \
"subset\n", r.to_string(), name());
}
}
// tell all our neighbouring master ports that our address
// ranges have changed
for (const auto& s: slavePorts)
s->sendRangeChange();
}
Add default responder to bus Update configuration for new default responder on bus Update to devices to handle their own pci config space without pciconfigall Remove most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for Remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt Remove pciconfigspace from pci devices, and py files Add calcConfigAddr that returns address for config space based on bus/dev/function + offset configs/test/fs.py: Update configuration for new default responder on bus src/dev/ide_ctrl.cc: src/dev/ide_ctrl.hh: src/dev/ns_gige.cc: src/dev/ns_gige.hh: src/dev/pcidev.cc: src/dev/pcidev.hh: Update to handle it's own pci config space without pciconfigall src/dev/io_device.cc: src/dev/io_device.hh: change naming for pio port break out recvTiming into two functions to reuse code src/dev/pciconfigall.cc: src/dev/pciconfigall.hh: removing most of pciconfigall, it now is a dumbdevice which gets it's address based on the bus it's supposed to respond for src/dev/pcireg.h: add a max size for PCI config space (per PCI spec) src/dev/platform.cc: src/dev/platform.hh: remove need for pci config space from platform, add registerPciDevice function to prevent more than one device from having same bus:dev:func and interrupt src/dev/sinic.cc: remove pciconfigspace as it's no longer a needed parameter src/dev/tsunami.cc: src/dev/tsunami.hh: src/dev/tsunami_pchip.cc: src/dev/tsunami_pchip.hh: add calcConfigAddr that returns address for config space based on bus/dev/function + offset (per PCI spec) src/mem/bus.cc: src/mem/bus.hh: src/python/m5/objects/Bus.py: add idea of default responder to bus src/python/m5/objects/Pci.py: add config port for pci devices add latency, bus and size parameters for pci config all (min is 8MB, max is 256MB see pci spec) --HG-- extra : convert_revision : 99db43b0a3a077f86611d6eaff6664a3885da7c9
2006-07-06 20:41:01 +02:00
clearPortCache();
}
AddrRangeList
BaseXBar::getAddrRanges() const
{
// we should never be asked without first having sent a range
// change, and the latter is only done once we have all the ranges
// of the connected devices
assert(gotAllAddrRanges);
// at the moment, this never happens, as there are no cycles in
// the range queries and no devices on the master side of a crossbar
// (CPU, cache, bridge etc) actually care about the ranges of the
// ports they are connected to
DPRINTF(AddrRanges, "Received address range request\n");
return xbarRanges;
}
void
BaseXBar::regStats()
{
using namespace Stats;
transDist
.init(MemCmd::NUM_MEM_CMDS)
.name(name() + ".trans_dist")
.desc("Transaction distribution")
.flags(nozero);
// get the string representation of the commands
for (int i = 0; i < MemCmd::NUM_MEM_CMDS; i++) {
MemCmd cmd(i);
const std::string &cstr = cmd.toString();
transDist.subname(i, cstr);
}
pktCount
.init(slavePorts.size(), masterPorts.size())
.name(name() + ".pkt_count")
.desc("Packet count per connected master and slave (bytes)")
.flags(total | nozero | nonan);
pktSize
.init(slavePorts.size(), masterPorts.size())
.name(name() + ".pkt_size")
.desc("Cumulative packet size per connected master and slave (bytes)")
.flags(total | nozero | nonan);
// both the packet count and total size are two-dimensional
// vectors, indexed by slave port id and master port id, thus the
// neighbouring master and slave, they do not differentiate what
// came from the master and was forwarded to the slave (requests
// and snoop responses) and what came from the slave and was
// forwarded to the master (responses and snoop requests)
for (int i = 0; i < slavePorts.size(); i++) {
pktCount.subname(i, slavePorts[i]->getMasterPort().name());
pktSize.subname(i, slavePorts[i]->getMasterPort().name());
for (int j = 0; j < masterPorts.size(); j++) {
pktCount.ysubname(j, masterPorts[j]->getSlavePort().name());
pktSize.ysubname(j, masterPorts[j]->getSlavePort().name());
}
}
}
template <typename SrcType, typename DstType>
DrainState
BaseXBar::Layer<SrcType,DstType>::drain()
{
//We should check that we're not "doing" anything, and that noone is
//waiting. We might be idle but have someone waiting if the device we
//contacted for a retry didn't actually retry.
if (state != IDLE) {
DPRINTF(Drain, "Crossbar not drained\n");
return DrainState::Draining;
} else {
return DrainState::Drained;
}
}
template <typename SrcType, typename DstType>
void
BaseXBar::Layer<SrcType,DstType>::regStats()
{
using namespace Stats;
occupancy
.name(name() + ".occupancy")
.desc("Layer occupancy (ticks)")
.flags(nozero);
utilization
.name(name() + ".utilization")
.desc("Layer utilization (%)")
.precision(1)
.flags(nozero);
utilization = 100 * occupancy / simTicks;
}
/**
* Crossbar layer template instantiations. Could be removed with _impl.hh
* file, but since there are only two given options (MasterPort and
* SlavePort) it seems a bit excessive at this point.
*/
template class BaseXBar::Layer<SlavePort,MasterPort>;
template class BaseXBar::Layer<MasterPort,SlavePort>;