This patch adds a frontend and backend static latency to the DRAM
controller by delaying the responses. Two parameters expressing the
frontend and backend contributions in absolute time are added to the
controller, and the appropriate latency is added to the responses when
adding them to the (infinite) queued port for sending.
For writes and reads that hit in the write buffer, only the frontend
latency is added. For reads that are serviced by the DRAM, the static
latency is the sum of the pipeline latencies of the entire frontend,
backend and PHY. The default values are chosen based on having roughly
10 pipeline stages in total at 500 MHz.
In the future, it would be sensible to make the controller use its
clock and convert these latencies (and a few of the DRAM timings) to
cycles.
This patch does some minor tidying up of the MSHR and MSHRQueue. The
clean up started as part of some ad-hoc tracing and debugging, but
seems worthwhile enough to go in as a separate patch.
The highlights of the changes are reduced scoping (private) members
where possible, avoiding redundant new/delete, and constructor
initialisation to please static code analyzers.
This patch introduces a mirrored internal snoop port to facilitate
easy addition of flow control for the snoop responses that are turned
into normal responses on their return. To perform this, the slave
ports of the coherent bus are wrapped in internal master ports that
are passed as the source ports to the response layer in question.
As a result of this patch, there is more contention for the response
resources, and as such system performance will decrease slightly.
A consequence of the mirrored internal port is that the port the bus
tells to retry (the internal one) and the port actually retrying (the
mirrored) one are not the same. Thus, the existing check in tryTiming
is not longer correct. In fact, the test is redundant as the layer is
only in the retry state while calling sendRetry on the waiting port,
and if the latter does not immediately call the bus then the retry
state is left. Consequently the check is removed.
This patch makes the buses multi layered, and effectively creates a
crossbar structure with distributed contention ports at the
destination ports. Before this patch, a bus could have a single
request, response and snoop response in flight at any time, and with
these changes there can be as many requests as connected slaves (bus
master ports), and as many responses as connected masters (bus slave
ports).
Together with address interleaving, this patch enables us to create
high-throughput memory interconnects, e.g. 50+ GByte/s.
This patch makes the flow control and state updates of the coherent
bus more clear by separating the two cases, i.e. forward as a snoop
response, or turn it into a normal response.
With this change it is also more clear what resources are being
occupied, and that we effectively bypass the busy check for the second
case. As a result of the change in resource usage some stats change.
This patch does some minor housekeeping on the bus code, removing
redundant code, and moving the extraction of the destination id to the
top of the functions using it.
This patch adds a basic set of stats which are hard to impossible to
implement using only communication monitors, and are needed for
insight such as bus utilization, transactions through the bus etc.
Stats added include throughput and transaction distribution, and also
a two-dimensional vector capturing how many packets and how much data
is exchanged between the masters and slaves connected to the bus.
This patch changes the set used to track outstanding requests to an
unordered set (part of C++11 STL). There is no need to maintain the
order, and hopefully there might even be a small performance benefit.
This patch adds a typical (leaning towards fast) LPDDR3 configuration
based on publically available data. As expected, it looks very similar
to the LPDDR2-S4 configuration, only with a slightly lower burst time.
This patch adapts the existing LPDDR2 configuration to make use of the
multi-channel functionality. Thus, to get a x64 interface two
controllers should be instantiated using the makeMultiChannel method.
The page size and ranks are also adapted to better suit with a typical
LPDDR2 part.
This patch removes the explicit memset as it is redundant and causes
the simulator to touch the entire space, forcing the host system to
allocate the pages.
Anonymous pages are mapped on the first access, and the page-fault
handler is responsible for zeroing them. Thus, the pages are still
zeroed, but we avoid touching the entire allocated space which enables
us to use much larger memory sizes as long as not all the memory is
actually used.
This patch changes the way cache statistics are collected in ruby.
As of now, there is separate entity called CacheProfiler which holds
statistical variables for caches. The CacheMemory class defines different
functions for accessing the CacheProfiler. These functions are then invoked
in the .sm files. I find this approach opaque and prone to error. Secondly,
we probably should not be paying the cost of a function call for recording
statistics.
Instead, this patch allows for accessing statistical variables in the
.sm files. The collection would become transparent. Secondly, it would happen
in place, so no function calls. The patch also removes the CacheProfiler class.
--HG--
rename : src/mem/slicc/ast/InfixOperatorExprAST.py => src/mem/slicc/ast/OperatorExprAST.py
The existing implementation can read uninitialized data or stale information
from the cached PageTable entries.
1) Add a valid bit for the cache entries. Simply using zero for the virtual
address to signify invalid entries is not sufficient. Speculative, wrong-path
accesses frequently access page zero. The current implementation would return
a uninitialized TLB entry when address zero was accessed and the PageTable
cache entry was invalid.
2) When unmapping/mapping/remaping a page, invalidate the corresponding
PageTable cache entry if one already exists.
Due to recent changes to clocking system in Ruby and the way Ruby restores
state from a checkpoint, garnet was failing to run from a checkpointed state.
The problem is that Ruby resets the time to zero while warming up the caches.
If any component records a local copy of the time (read calls curCycle())
before the simulation has started, then that component will not operate until
that time is reached. In the context of this particular patch, the Garnet
Network class calls curCycle() at multiple places. Any non-operational
component can block in requests in the memory system, which the system
interprets as a deadlock. This patch makes changes so that Garnet can
successfully run from checkpointed state.
It adds a globally visible time at which the actual execution started. This
time is initialized in RubySystem::startup() function. This variable is only
meant for components with in Ruby. This replaces the private variable that
was maintained within Garnet since it is not possible to figure out the
correct time when the value of this variable can be set.
The patch also does away with all cases where curCycle() is called with in
some Ruby component before the system has actually started executing. This
is required due to the quirky manner in which ruby restores from a checkpoint.
This patch adds an address mapping scheme where the channel
interleaving takes place on a cache line granularity. It is similar to
the existing RaBaChCo that interleaves on a DRAM page, but should give
higher performance when there is less locality in the address
stream.
This patch changes the slightly ambigious names used for the address
mapping scheme to be more descriptive, and actually spell out what
they do. With this patch we also open up for adding more flavours of
open- and close-type mappings, i.e. interleaving across channels with
the open map.
This patch adds a WideIO 200 MHz configuration that can be used as a
baseline to compare with DDRx and LPDDRx. Note that it is a single
channel and that it should be replicated 4 times. It is based on
publically available information and attempts to capture an envisioned
8 Gbit single-die part (i.e. without TSVs).
This patch provides useful printouts throughut the memory system. This
includes pretty-printed cache tags and function call messages
(call-stack like).
This patch changes the SimpleTimingPort and RubyPort to panic on
inhibited requests as this should never happen in either of the
cases. The SimpleTimingPort is only used for the I/O devices PIO port
and the DMA devices config port and should thus never see an inhibited
request. Similarly, the SimpleTimingPort is also used for the
MessagePort in x86, and there should also not be any cases where the
port sees an inhibited request.
Previously, nextCycle() could return the *current* cycle if the current tick was
already aligned with the clock edge. This behavior is not only confusing (not
quite what the function name implies), but also caused problems in the
drainResume() function. When exiting/re-entering the sim loop (e.g., to take
checkpoints), the CPUs will drain and resume. Due to the previous behavior of
nextCycle(), the CPU tick events were being rescheduled in the same ticks that
were already processed before draining. This caused divergence from runs that
did not exit/re-entered the sim loop. (Initially a cycle difference, but a
significant impact later on.)
This patch separates out the two behaviors (nextCycle() and clockEdge()),
uses nextCycle() in drainResume, and uses clockEdge() everywhere else.
Nothing (other than name) should change except for the drainResume timing.
When using the o3 or inorder CPUs with many Ruby protocols, the caches may
need to forward invalidations to the CPUs. The RubyPort was instantiating a
packet to be sent to the CPUs to signal the eviction, but the packets were
not being freed by the CPUs. Consistent with the classic memory model, stack
allocate the packet and heap allocate the request so on
ruby_eviction_callback() completion, the packet deconstructor is called, and
deletes the request (*Note: stack allocating the request causes double
deletion, since it will be deleted in the packet destructor). This results in
the least memory allocations without memory errors.
When warming up caches in Ruby, the CacheRecorder sends fetch requests into
Ruby Sequencers with packet types that require responses. Since responses are
never generated for these CacheRecorder requests, the requests are not deleted
in the packet destructor called from the Ruby hit callback. Free the request.
This allows you to have (i.e.) an L2 cache that is not named "L2Cache"
but is still a GenericMachineType_L2Cache. This is particularly
helpful if the protocol has multiple L2 controllers.
When Ruby stats are printed for events and transitions, they include stats
for all of the controllers of the same type, but they are not necessarily
printed in order of the controller ID "version", because of the way the
profilers were added to the profiler vector. This patch fixes the push order
problem so that the stats are printed in ascending order 0->(# controllers),
so statistics parsers may correctly assume the controller to which the stats
belong.
When connecting message buffers between Ruby controllers, it is
easy to mistakenly connect multiple controllers to the same message
buffer. This patch prints a more descriptive fatal message than the
previous assert statement in order to facilitate easier debugging.
The cache trace variables are array allocated uint8_t* in the RubySystem and
the Ruby CacheRecorder, but the code used delete to free the memory, resulting
in Valgrind memory errors. Change these deletes to delete [] to get rid of the
errors.
Fixes a latency calculation bug for accesses during a cache line fill.
Under a cache miss, before the line is filled, accesses to the cache are
associated with a MSHR and marked as targets. Once the line fill completes,
MSHR target packets pay an additional latency of
"responseLatency + busSerializationLatency". However, the "whenReady"
field of the cache line is only set to an additional delay of
"busSerializationLatency". This lacks the responseLatency component of
the fill. It is possible for accesses that occur on the cycle of
(or briefly after) the line fill to respond without properly paying the
responseLatency. This also creates the situation where two accesses to the
same address may be serviced in an order opposite of how they were received
by the cache. For stores to the same address, this means that although the
cache performs the stores in the order they were received, acknowledgements
may be sent in a different order.
Adding the responseLatency component to the whenReady field preserves the
penalty that should be paid and prevents these ordering issues.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
This patch solves the corner case scenario where the sendRetryEvent could be
scheduled twice, when an io device stresses the IOcache in the system. This
should not be possible in the cache system.
This patch splits the retryList into a list of ports that are waiting
for the bus itself to become available, and a map that tracks the
ports where forwarding failed due to a peer not accepting the
packet. Thus, when a retry reaches the bus, it can be sent to the
appropriate port that initiated that transaction.
As a consequence of this patch, only ports that are really ready to go
will get a retry, thus reducing the amount of redundant failed
attempts. This patch also makes it easier to reason about the order of
servicing requests as the ports waiting for the bus are now clearly
FIFO and much easier to change if desired.
This patch introduces a variable to keep track of the retrying port
instead of relying on it being the front of the retryList.
Besides the improvement in readability, this patch is a step towards
separating out the two cases where a port is waiting for the bus to be
free, and where the forwarding did not succeed and the bus is waiting
for a retry to pass on to the original initiator of the transaction.
The changes made are currently such that the regressions are not
affected. This is ensured by always prioritizing the currently
retrying port and putting it back at the front of the retry list.
This patch adds an optional flags field to the packet trace to encode
the request flags that contain information about whether the request
is (un)cacheable, instruction fetch, preftech etc.
A recent set of patches added support for multiple clock domains to ruby.
I had made some errors while writing those patches. The sender was using
the receiver side clock while enqueuing a message in the buffer. Those
errors became visible while creating (or restoring from) checkpoints. The
errors also become visible when a multi eventq scenario occurs.
The message buffer node used to keep time in terms of Cycles. Since the
sender and the receiver can have different clock periods, storing node
time in cycles requires some conversion. Instead store the time directly
in Ticks.
A set of patches was recently committed to allow multiple clock domains
in ruby. In those patches, I had inadvertently made an incorrect use of
the clocks. Suppose object A needs to schedule an event on object B. It
was possible that A accesses B's clock to schedule the event. This is not
possible in actual system. Hence, changes are being to the Consumer class
so as to avoid such happenings. Note that in a multi eventq simulation,
this can possibly lead to an incorrect simulation.
There are two functions in the Consumer class that are used for scheduling
events. The first function takes in the relative delay over the current time
as the argument and adds the current time to it for scheduling the event.
The second function takes in the absolute time (in ticks) for scheduling the
event. The first function is now being moved to protected section of the
class so that only objects of the derived classes can use it. All other
objects will have to specify absolute time while scheduling an event
for some consumer.
The histogram for tracking outstanding counts per cycle is maintained
in the profiler. For a parallel implementation of the memory system, we
need that this histogram is maintained locally. Hence it will now be
kept in the sequencer itself. The resulting histograms will be merged
when the stats are printed.
These functions are currently implemented in one of the files related to Slicc.
Since these are purely C++ functions, they are better suited to be in the base
class.
This patch modifies ruby so that two controllers can be connected to each
other with only message buffers in between. Before this patch, all the
controllers had to be connected to the network for them to communicate
with each other. With this patch, one can have protocols where a controller
is not connected to the network, but communicates with another controller
through a message buffer.
The Topology class in Ruby does not need to inherit from SimObject class.
This patch turns it into a regular class. The topology object is now created
in the constructor of the Network class. All the parameters for the topology
class have been moved to the network class.
The functional write code was assuming that all writes are block sized,
which may not be true for Ruby Requests. This bug can lead to a buffer
overflow.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
This check covers a case where a retry is called from the SimpleDRAM
causing a new request to appear before the DRAM itself schedules a
nextReqEvent. By adding this check, the event is not scheduled twice.
This patch adds a class method that allows easy creation of
channel-interleaved multi-channel DRAM configurations. It is enabled
by a class method to allow customisation of the class independent of
the channel configuration. For example, the user can create a MyDDR
subclass of e.g. SimpleDDR3, and then create a four-channel
configuration of the subclass by calling MyDDR.makeMultiChannel(4,
mem_start, mem_size).
This patch fixes a number of small cosmetic issues in the SimpleDRAM
module. The most important change is to move the accounting of
received packets to after the check is made if the packet should be
retried or not. Thus, packets are only counted if they are actually
accepted.
This patch adds support for multi-channel instances of the DRAM
controller model by stripping away the channel bits in the address
decoding. The patch relies on the availiability of address
interleaving and, at this time, it is up to the user to configure the
interleaving appropriately. At the moment it is assumed that the
channel interleaving bits are immediately following the column bits
(smallest sensible interleaving). Convenience methods for building
multi-channel configurations will be added later.
This patch adds merging of interleaved ranges before creating the
backing stores. The backing stores are always a contigous chunk of the
address space, and with this patch it is possible to have interleaved
memories in the system.
This patch adds basic merging of address ranges to the bus, such that
interleaved ranges are merged together before being passed on by the
bus. As such, the bus aggregates the address ranges of the connected
slave ports and then passes on the merged ranges through its master
ports. The bus thus hides the complexity of the interleaved ranges and
only exposes contigous ranges to the surrounding system.
As part of this patch, the bus ranges are also cached for any future
queries.
The MESI CMP directory coherence protocol, while transitioning from SM to IM,
did not invalidate the lock that it might have taken on a cache line. This
patch adds an action for doing so.
The problem was found by Dibakar, but I was not happy with his proposed
solution. So I implemented a different solution.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
This patch fixes a newly introduced bug where the sender state was
popped before checking that it should be. Amazingly all regressions
pass, but Linux fails to boot on the detailed CPU with caches enabled.
This patch fixes the warnings that clang3.2svn emit due to the "-Wall"
flag. There is one case of an uninitialised value in the ARM neon ISA
description, and then a whole range of unused private fields that are
pruned.
This patch address the most important name shadowing warnings (as
produced when using gcc/clang with -Wshadow). There are many
locations where constructor parameters and function parameters shadow
local variables, but these are left unchanged.
This patch adds a check to ensure that the delay incurred by
the bus is not simply disregarded, but accounted for by someone. At
this point, all the modules do is to zero it out, and no additional
time is spent. This highlights where the bus timing is simply dropped
instead of being paid for.
As a follow up, the locations identified in this patch should add this
additional time to the packets in one way or another. For now it
simply acts as a sanity check and highlights where the delay is simply
ignored.
Since no time is added, all regressions remain the same.
This patch changes the names of the cache accessor functions to be in
line with those used by the ports. This is done to avoid confusion and
get closer to a one-to-one correspondence between the interface of the
memory object (the cache in this case) and the port itself.
The member function timingAccess has been split into a snoop/non-snoop
part to avoid branching on the isResponse() of the packet.
This patch changes the bus-related time accounting done in the packet
to be relative. Besides making it easier to align the cache timing to
cache clock cycles, it also makes it possible to create a Last-Level
Cache (LLC) directly to a memory controller without a bus inbetween.
The bus is unique in that it does not ever make the packets wait to
reflect the time spent forwarding them. Instead, the cache is
currently responsible for making the packets wait. Thus, the bus
annotates the packets with the time needed for the first word to
appear, and also the last word. The cache then delays the packets in
its queues before passing them on. It is worth noting that every
object attached to a bus (devices, memories, bridges, etc) should be
doing this if we opt for keeping this way of accounting for the bus
timing.
This patch removes the time field from the packet as it was only used
by the preftecher. Similar to the packet queue, the prefetcher now
wraps the packet in a deferred packet, which also has a tick
representing the absolute time when the packet should be sent.
This patch makes the clock member private to the ClockedObject and
forces all children to access it using clockPeriod(). This makes it
impossible to inadvertently change the clock, and also makes it easier
to transition to a situation where the clock is derived from e.g. a
clock domain, or through a multiplier.
This patch fixes a potential deadlock in the caches. This deadlock
could occur when more than one cache is used in a system, and
pkt->senderState is modified in between the two caches. This happened
as the caches relied on the senderState remaining unchanged, and used
it for instantaneous upstream communication with other caches.
This issue has been addressed by iterating over the linked list of
senderStates until we are either able to cast to a MSHR* or
senderState is NULL. If the cast is successful, we know that the
packet has previously passed through another cache, and therefore
update the downstreamPending flag accordingly. Otherwise, we do
nothing.
This patch adds a predecessor field to the SenderState base class to
make the process of linking them up more uniform, and enable a
traversal of the stack without knowing the specific type of the
subclasses.
There are a number of simplifications done as part of changing the
SenderState, particularly in the RubyTest.
This patch fixes a bug in the CommMonitor caused by the packet being
modified before it is captured in the trace. By recording the fields
before passing the packet on, and then putting these values in the
trace we ensure that even if the packet is modified the trace captures
what the CommMonitor saw.
This patch merely adopts a more strict use of const for the cache
member functions and variables, and also moves a large portion of the
member functions from public to protected.
Virtualized CPUs and the fastmem mode of the atomic CPU require direct
access to physical memory. We currently require caches to be disabled
when using them to prevent chaos. This is not ideal when switching
between hardware virutalized CPUs and other CPU models as it would
require a configuration change on each switch. This changeset
introduces a new version of the atomic memory mode,
'atomic_noncaching', where memory accesses are inserted into the
memory system as atomic accesses, but bypass caches.
To make memory mode tests cleaner, the following methods are added to
the System class:
* isAtomicMode() -- True if the memory mode is 'atomic' or 'direct'.
* isTimingMode() -- True if the memory mode is 'timing'.
* bypassCaches() -- True if caches should be bypassed.
The old getMemoryMode() and setMemoryMode() methods should never be
used from the C++ world anymore.
The transition for state MII and event Store was found missing during testing.
The transition is being added. The controller will not stall the Store request
in state MII
This patch allows ruby to have multiple clock domains. As I understand
with this patch, controllers can have different frequencies. The entire
network needs to run at a single frequency.
The idea is that with in an object, time is treated in terms of cycles.
But the messages that are passed from one entity to another should contain
the time in Ticks. As of now, this is only true for the message buffers,
but not for the links in the network. As I understand the code, all the
entities in different networks (simple, garnet-fixed, garnet-flexible) should
be clocked at the same frequency.
Another problem is that the directory controller has to operate at the same
frequency as the ruby system. This is because the memory controller does
not make use of the Message Buffer, and instead implements a buffer of its
own. So, it has no idea of the frequency at which the directory controller
is operating and uses ruby system's frequency for scheduling events.
This patch is as of now the final patch in the series of patches that replace
Time with Cycles.This patch further replaces Time with Cycles in Sequencer,
Profiler, different protocols and related entities.
Though Time has not been completely removed, the places where it is in use
seem benign as of now.
The patch started of with replacing Time with Cycles in the Consumer class.
But to get ruby to compile, the rest of the changes had to be carried out.
Subsequent patches will further this process, till we completely replace
Time with Cycles.
This patch modifies the Histogram class' add() function so that it can add
linear histograms as well. The function assumes that the left end point of
the ranges of the two histograms are the same. It also assumes that when
the ranges of the two histogram are changed to accomodate an element not in
the range, the factor used in changing the range is same for both the
histograms.
This function is then used in removing one of the calls to the global
profiler*. The histograms for recording the delays incurred in processing
different requests are now maintained by the controllers. The profiler
adds these histograms when it needs to print the stats.
This patch does several things. First, the counter for fully busy cycles for a
controller is now kept with in the controller, instead of being part of the profiler.
Second, the topology class no longer keeps an array of controllers which was only
used for printing stats. Instead, ruby system will now ask each controller to print
the stats. Thirdly, the statistical variable for recording how many different types
were created is being moved in to the controller from the profiler. Note that for
printing, the profiler will collate results from different controllers.
The number of bits required for an address was set to floorLog2(memory size).
This is correct under the assumption that the memory size is a power of 2,
which is not always true. Hence, floorLog2 is being replaced with ceilLog2.
This patch moves the default DRAM parameters from the SimpleDRAM class
to two different subclasses, one for DDR3 and one for LPDDR2. More can
be added as we go forward.
The regressions that previously used the SimpleDRAM are now using
SimpleDDR3 as this is the most similar configuration.
This patch adds two additional scheduling constraints to the DRAM
controller model, to constrain the activation rate. The two metrics
are determine the size of the activation window in terms of the number
of activates and the minimum time required for that number of
activates. This maps to current DDRx, LPDDRx and WIOx standards that
have either tFAW (4 activate window) or tTAW (2 activate window)
scheduling constraints.
This patch changes how the data bus busy time is calculated such that
it is delayed to the actual scheduling time of the request as opposed
to being done as soon as possible.
This patch changes a bunch of statistics, and the stats update is
bundled together with the introruction of tFAW/tTAW and the named DRAM
configurations like DDR3 and LPDDR2.
the cache drainManager is set but never cleared, this is because
the cache itself does not need to be drained and thus never
triggers a signalDrainDone(). because the drainManager variable
is not used properly and does not appear to be necessary it has
been removed with this patch.
The panic() function already prints the current tick value. This call to
curCycle() is as such redundant. Since we are trying to move towards multiple
clock domains, this call will print misleading time.
This patch converts the panic() print outs in the Sequencer::wakeup()
call from ruby cycles to Ticks(). This makes it easier to debug deadlocks
with the ProtocolTrace flag so the issue time indicated in the panic message
can be quickly searched for.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
This patch was initiated so as to remove reference to g_system_ptr,
the pointer to Ruby System that is used for getting the current time.
That simple change actual requires changing a lot many things in slicc and
garnet. All these changes are related to how time is handled.
In most of the places, g_system_ptr has been replaced by another clock
object. The changes have been done under the assumption that all the
components in the memory system are on the same clock frequency, but the
actual clocks might be distributed.
Many Ruby structures inherit from the Consumer, which is used for scheduling
events. The Consumer used to relay on an Event Manager for scheduling events
and on g_system_ptr for time. With this patch, the Consumer will now use a
ClockedObject to schedule events and to query for current time. This resulted
in several structures being converted from SimObjects to ClockedObjects. Also,
the MessageBuffer class now requires a pointer to a ClockedObject so as to
query for time.
The current implementation in gem5 just keeps a list of locks per cacheline.
Due to this, a store to a non-overlapping portion of the cacheline can cause an
LL/SC pair to fail. This patch simply adds an address range to the lock
structure, so that the lock is only invalidated if the store overlaps the lock
range.
Running with valgrind I noticed a use after free originating from
simple_mem.cc. It looks like this is a known issue and this additional call
site was missed in an earlier patch.
When the classic gem5 cache sees an uncacheable memory access, it used
to ignore it or silently drop the cache line in case of a
write. Normally, there shouldn't be any data in the cache belonging to
an uncacheable address range. However, since some architecture models
don't implement cache maintenance instructions, there might be some
dirty data in the cache that is discarded when this happens. The
reason it has mostly worked before is because such cache lines were
most likely evicted by normal memory activity before a TLB flush was
requested by the OS.
Previously, the cache model would invalidate cache lines when they
were accessed by an uncacheable write. This changeset alters this
behavior so all uncacheable memory accesses cause a cache flush with
an associated writeback if necessary. This is implemented by reusing
the cache flushing machinery used when draining the cache, which
implies that writebacks are performed using functional accesses.
The IIC replacement policy seems to be unused and has probably
gathered too much bit rot to be useful. This patch removes the IIC and
its associated cache parameters.
This patch adds basic merging of address ranges when determining which
address ranges should be reported in the configuration table. By
performing this merging it is possible to distribute an address range
across many memory channels (controllers). This is essential to enable
address interleaving.
This patch adds support for interleaving bits for the address
ranges. What was previously just a start and end address, now has an
additional three fields, for the high bit, and number of bits to use
for interleaving, and a match value to compare against. If the number
of interleaving bits is set to zero it is effectively disabled.
A number of convenience functions are added to the range to enquire
about the interleaving, its granularity and the number of stripes it
is part of.
This patch cleans up the AddrRangeMap in preparation for the addition
of interleaving by removing unused code. The non-const editions of
find are never used, and hence the duplication is not needed.
This patch tidies up a number of the bus DPRINTFs related to range
manipulation. In particular, it shifts the message about range changes
to the start of the member function, and also adds information about
when all ranges are received.
This patch makes the address mapper less stringent about checking the
before and after ranges, i.e. the original and remapped ranges. The
checks were not really necessary, and there are situations when the
previous checks were too strict.
This patch makes the start and end address private in a move to
prevent direct manipulation and matching of ranges based on these
fields. This is done so that a transition to ranges with interleaving
support is possible.
As a result of hiding the start and end, a number of member functions
are needed to perform the comparisons and manipulations that
previously took place directly on the members. An accessor function is
provided for the start address, and a function is added to test if an
address is within a range. As a result of the latter the != and ==
operator is also removed in favour of the member function. A member
function that returns a string representation is also created to allow
debug printing.
In general, this patch does not add any functionality, but it does
take us closer to a situation where interleaving (and more cleverness)
can be added under the bonnet without exposing it to the user. More on
that in a later patch.
This patch temporarily removes the joining of ranges when creating the
backing store, to reserve this functionality for the interleaved
ranges that are about to be introduced.
When creating the mmaps for the backing store, there is no point in
creating larger contigous chunks that what is necessary. The larger
chunks will only make life more difficult for the host.
Merging will be re-added later, but then only for interleaved ranges.
This patch adds packet tracing to the communication monitor using a
protobuf as the mechanism for creating the trace.
If no file is specified, then the tracing is disabled. If a file is
specified, then for every packet that is successfully sent, a protobuf
message is serialized to the file.
This patch adds a basic check to ensure that the packet queue does not
grow absurdly large. The queue should only be used to store packets
that were delayed due to blocking from the neighbouring port, and not
for actual storage. Thus, a limit of 100 has been chosen for now
(which is already quite substantial).
This patch addresses a warning related to the swig interface
generation for the Switch class. The cxx_header is now specified
correctly, and the header in question has got a few includes added to
make it all compile.
This patch fixes a bug that caused multiple systems to overwrite each
other physical memory. The system name is now included in the filename
such that this is avoided.
The patch adds support to slicc for recognizing arguments that should be
passed to the constructor of a class. I did not like the fact that an explicit
check was being carried on the type 'TBETable' to figure out the arguments to
be passed to the constructor.
The patch also moves some of the member variables that are declared for all
the controllers to the base class AbstractController.
This patch adds a prefetcher for the ruby memory system. The prefetcher
is based on a prefetcher implemented by others (well, I don't know
who wrote the original). The prefetcher does stride-based prefetching,
both unit and non-unit. It obseves the misses in the cache and trains on
these. After the training period is over, the prefetcher starts issuing
prefetch requests to the controller.
This patch adds a _curTick variable to an eventq. This variable is updated
whenever an event is serviced in function serviceOne(), or all events upto
a particular time are processed in function serviceEvents(). This change
helps when there are eventqs that do not make use of curTick for scheduling
events.
Recent changes to functionalRead() in the memory system was not correct.
The change allowed for returning data from the first message found in
the buffers of the memory system. This is not correct since it is possible
that a timing message has data from an older state of the block.
The changes are being reverted.
This patch fixes the draining of the SimpleDRAM controller model. The
controller performs buffering of writes and normally there is no need
to ever empty the write buffer (if you have a fast on-chip memory,
then use it). The patch adds checks to ensure the write buffer is
drained when the controller is asked to do so.
This patch adds support to ruby so that the statistics maintained by ruby
are reset/dumped when the statistics for the rest of the system are
reset/dumped. For resetting the statistics, ruby now provides the
resetStats() function that a sim object can provide. As a consequence, the
clearStats() function has been removed from RubySystem. For dumping stats,
Ruby now adds a callback event to the dumpStatsQueue. The exit callback that
ruby used to add earlier is being removed.
Created by: Hamid Reza Khaleghzadeh.
Improved by: Lluc Alvarez, Nilay Vaish
Committed by: Nilay Vaish
This patch adds support for the following optional drain methods in
the classical memory system's cache model:
memWriteback() - Write back all dirty cache lines to memory using
functional accesses.
memInvalidate() - Invalidate all cache lines. Dirty cache lines
are lost unless a writeback is requested.
Since memWriteback() is called when checkpointing systems, this patch
adds support for checkpointing systems with caches. The serialization
code now checks whether there are any dirty lines in the cache. If
there are dirty lines in the cache, the checkpoint is flagged as bad
and a warning is printed.
This patch moves the draining interface from SimObject to a separate
class that can be used by any object needing draining. However,
objects not visible to the Python code (i.e., objects not deriving
from SimObject) still depend on their parents informing them when to
drain. This patch also gets rid of the CountedDrainEvent (which isn't
really an event) and replaces it with a DrainManager.
When casting objects in the generated SWIG interfaces, SWIG uses
classical C-style casts ( (Foo *)bar; ). In some cases, this can
degenerate into the equivalent of a reinterpret_cast (mainly if only a
forward declaration of the type is available). This usually works for
most compilers, but it is known to break if multiple inheritance is
used anywhere in the object hierarchy.
This patch introduces the cxx_header attribute to Python SimObject
definitions, which should be used to specify a header to include in
the SWIG interface. The header should include the declaration of the
wrapped object. We currently don't enforce header the use of the
header attribute, but a warning will be generated for objects that do
not use it.
This patch enables dumping statistics and Linux process information on
context switch boundaries (__switch_to() calls) that are used for
Streamline integration (a graphical statistics viewer from ARM).
This patch changes the default system clock from 1THz to 1GHz. This
clock is used by all modules that do not override the default (parent
clock), and primarily affects the IO subsystem. Every DMA device uses
its clock to schedule the next transfer, and the change will thus
cause this inter-transfer delay to be longer.
The default clock of the bus is removed, as the clock inherited from
the system provides exactly the same value.
A follow-on patch will bump the stats.
I had forgotten to change the network test protocol while making changes to
ruby for supporting functional accesses. This patch updates the protocol so
that it can compile correctly.
This patch adds support to different entities in the ruby memory system
for more reliable functional read/write accesses. Only the simple network
has been augmented as of now. Later on Garnet will also support functional
accesses.
The patch adds functional access code to all the different types of messages
that protocols can send around. These messages are functionally accessed
by going through the buffers maintained by the network entities.
The patch also rectifies some of the bugs found in coherence protocols while
testing the patch.
With this patch applied, functional writes always succeed. But functional
reads can still fail.
Currently the Ruby System maintains pointer to only one of the memory
controllers. But there can be multiple controllers in the system. This
patch adds a vector of memory controllers.
This patch adds support for function definitions to appear in slicc structs.
This is required for supporting functional accesses for different types of
messages. Subsequent patches will use this to development.
It seems unecessary that the BankedArray class needs to schedule an event
to figure out when the access ends. Instead only the time for the end of access
needs to be tracked.
Ruby system was recently converted to a clocked object. Such objects maintain
state related to the time that has passed so far. During the cache warmup, Ruby
system changes its own time and the global time. Later on, the global time is
restored. So Ruby system also needs to reset its own time.
This patch adds an additional level of ports in the inheritance
hierarchy, separating out the protocol-specific and protocl-agnostic
parts. All the functionality related to the binding of ports is now
confined to use BaseMaster/BaseSlavePorts, and all the
protocol-specific parts stay in the Master/SlavePort. In the future it
will be possible to add other protocol-specific implementations.
The functions used in the binding of ports, i.e. getMaster/SlavePort
now use the base classes, and the index parameter is updated to use
the PortID typedef with the symbolic InvalidPortID as the default.
This patch moves all the memory backing store operations from the
independent memory controllers to the global physical memory. The main
reason for this patch is to allow address striping in a future set of
patches, but at this point it already provides some useful
functionality in that it is now possible to change the number of
memory controllers and their address mapping in combination with
checkpointing. Thus, the host and guest view of the memory backing
store are now completely separate.
With this patch, the individual memory controllers are far simpler as
all responsibility for serializing/unserializing is moved to the
physical memory. Currently, the functionality is more or less moved
from AbstractMemory to PhysicalMemory without any major
changes. However, in a future patch the physical memory will also
resolve any ranges that are interleaved and properly assign the
backing store to the memory controllers, and keep the host memory as a
single contigous chunk per address range.
Functionality for future extensions which involve CPU virtualization
also enable the host to get pointers to the backing store.
This patch changes the data structure used to keep track of ports that
should be told to retry. As the bus is doing this in an FCFS way,
there is no point having a list. A deque is a better match (and is at
least in theory a better choice from a performance point of view).
This patch addresses a number of smaller issues identified by the code
inspection utility cppcheck. There are a number of identified leaks in
the arm/linux/system.cc (although the function only get's called once
so it is not a major problem), a few deletes in dev/x86/i8042.cc that
were not array deletes, and sprintfs where the character array had one
element less than needed. In the IIC tags there was a function
allocating an array of longs which is in fact never used.
This patch changes the cache-related latencies from an absolute time
expressed in Ticks, to a number of cycles that can be scaled with the
clock period of the caches. Ultimately this patch serves to enable
future work that involves dynamic frequency scaling. As an immediate
benefit it also makes it more convenient to specify cache performance
without implicitly assuming a specific CPU core operating frequency.
The stat blocked_cycles that actually counter in ticks is now updated
to count in cycles.
As the timing is now rounded to the clock edges of the cache, there
are some regressions that change. Plenty of them have very minor
changes, whereas some regressions with a short run-time are perturbed
quite significantly. A follow-on patch updates all the statistics for
the regressions.
This patch transitions the bus to use the AddrRange operations instead
of directly accessing the start and end. The change facilitates the
move to a more elaborate AddrRange class that also supports address
striping in the bus by specifying interleaving bits in the ranges.
Two new functions are added to the AddrRange to determine if two
ranges intersect, and if one is a subset of another. The bus
propagation of address ranges is also tweaked such that an update is
only propagated if the bus received information from all the
downstream slave modules. This avoids the iteration and need for the
cycle-breaking scheme that was previously used.
This patch moves the block size computation from findBlockSize to
initialisation time, once all the neighbouring ports are connected.
There is no need to dynamically update the block size, and the caching
of the value effectively avoided that anyhow. This is very similar to
what was already in place, just with a slightly leaner implementation.
This patch removes the parts of slicc that were required for multi-chip
protocols. Going ahead, it seems multi-chip protocols would be implemented
by playing with the network itself.
This patch moves the code for functional accesses to ruby system. This is
because the subsequent patches add support for making functional accesses
to the messages in the interconnect. Making those accesses from the ruby port
would be cumbersome.
In the current caches the hit latency is paid twice on a miss. This patch lets
a configurable response latency be set of the cache for the backward path.
Fix the drain functionality of the RubyPort to only call drain on child ports
during a system-wide drain process, instead of calling each time that a
ruby_hit_callback is executed.
This fixes the issue of the RubyPort ports being reawakened during the drain
simulation, possibly with work they didn't previously have to complete. If
they have new work, they may call process on the drain event that they had
not registered work for, causing an assertion failure when completing the
drain event.
Also, in RubyPort, set the drainEvent to NULL when there are no events
to be drained. If not set to NULL, the drain loop can result in stale
drainEvents used.
This patch introduces a high-level model of a DRAM controller, with a
basic read/write buffer structure, a selectable and customisable
arbiter, a few address mapping options, and the basic DRAM timing
constraints. The parameters make it possible to turn this model into
any desired DDRx/LPDDRx/WideIOx memory controller.
The intention is not to be cycle accurate or capture every aspect of a
DDR DRAM interface, but rather to enable exploring of the high-level
knobs with a good simulation speed. Thus, contrary to e.g. DRAMSim
this module emphasizes simulation speed with a good-enough accuracy.
This module is merely a starting point, and there are plenty additions
and improvements to come. A notable addition is the support for
address-striping in the bus to enable a multi-channel DRAM
controller. Also note that there are still a few "todo's" in the code
base that will be addressed as we go along.
A follow-up patch will add basic performance regressions that use the
traffic generator to exercise a few well-defined corner cases.
This patch removes the unused file parameter from the
AbstractMemory. The patch serves to make it easier to transition to a
separation of the actual contigious host memory backing store, and the
gem5 memory controllers.
Without the file parameter it becomes easier to hide the creation of
the mmap in the PhysicalMemory, as there are no longer any reasons to
expose the actual contigious ranges to the user.
To the best of my knowledge there is no use of the parameter, so the
change should not affect anyone.
This patch takes the final plunge and transitions from the templated
Range class to the more specific AddrRange. In doing so it changes the
obvious Range<Addr> to AddrRange, and also bumps the range_map to be
AddrRangeMap.
In addition to the obvious changes, including the removal of redundant
includes, this patch also does some house keeping in preparing for the
introduction of address interleaving support in the ranges. The Range
class is also stripped of all the functionality that is never used.
--HG--
rename : src/base/range.hh => src/base/addr_range.hh
rename : src/base/range_map.hh => src/base/addr_range_map.hh
This patch removes the use of g_system_ptr for event scheduling. Each consumer
object now needs to specify upfront an EventManager object it would use for
scheduling events. This makes the ruby memory system more amenable for a
multi-threaded simulation.
This patch makes a minor addition to the SimpleMemory by enforcing a
maximum data rate. The bandwidth is configurable, and a reasonable
value (12.8GB/s) has been choosen as the default.
The changes do add some complexity to the SimpleMemory, but they
should definitely be justifiable as this enables a far more realistic
setup using even this simple memory controller.
The rate regulation is done for reads and writes combined to reflect
the bidirectional data busses used by most (if not all) relevant
memories. Moreover, the regulation is done per packet as opposed to
long term, as it is the short term data rate (data bus width times
frequency) that is the limiting factor.
A follow-up patch bumps the stats for the regressions.
This patch shifts the version of gcc for which we enable c++0x from
4.6 to 4.4 The more long term plan is to see what the c++0x features
can bring and what level of support would be enabled simply by bumping
the required version of gcc from 4.3 to 4.4.
A few minor things had to be fixed in the code base, most notably the
choice of a hashmap implementation. In the Ruby Sequencer there were
also a few minor issues that gcc 4.4 was not too happy about.
This patch addresses a few minor issues reported by the clang static
analyzer.
The analysis was run with:
scan-build -disable-checker deadcode \
-enable-checker experimental.core \
-disable-checker experimental.core.CastToStruct \
-enable-checker experimental.cpluscplus
This seperates the functionality to clear the state in a block into
blk.hh and the functionality to udpate the tag information into the
tags. This gets rid of the case where calling invalidateBlk on an
already-invalid block does something different than calling it on a
valid block, which was confusing.
If I understand correctly, this was put in place so that a debugger can be
attached when the protocol aborts. While this sounds useful, it is a problem
when the simulation is not being actively monitored. I think it is better to
remove this.
Despite gzwrite taking an unsigned for length, it returns an int for
bytes written; gzwrite fails if (int)len < 0. Because of this, call
gzwrite with len no larger than INT_MAX: write in blocks of INT_MAX if
data to be written is larger than INT_MAX.
This patch is a first step to using Cycles as a parameter type. The
main affected modules are the CPUs and the Ruby caches. There are
definitely plenty more places that are affected, but this patch serves
as a starting point to making the transition.
An important part of this patch is to actually enable parameters to be
specified as Param.Cycles which involves some changes to params.py.
The =operator for the DataBlock class was incorrectly interpreting the class
member m_alloc. This variable stands for whether the assigned memory for the
data block needs to be freed or not by the class itself. It seems that the
=operator interpreted the variable as whether the memory is assigned to the
data block. This wrong interpretation was causing values not to propagate
to RubySystem::m_mem_vec_ptr. This caused major issues with restoring from
checkpoints when using a protocol which verified that the cache data was
consistent with the backing store (i.e. MOESI-hammer).
This patch addresses the comments and feedback on the preceding patch
that reworks the clocks and now more clearly shows where cycles
(relative cycle counts) are used to express time.
Instead of bumping the existing patch I chose to make this a separate
patch, merely to try and focus the discussion around a smaller set of
changes. The two patches will be pushed together though.
This changes done as part of this patch are mostly following directly
from the introduction of the wrapper class, and change enough code to
make things compile and run again. There are definitely more places
where int/uint/Tick is still used to represent cycles, and it will
take some time to chase them all down. Similarly, a lot of parameters
should be changed from Param.Tick and Param.Unsigned to
Param.Cycles.
In addition, the use of curTick is questionable as there should not be
an absolute cycle. Potential solutions can be built on top of this
patch. There is a similar situation in the o3 CPU where
lastRunningCycle is currently counting in Cycles, and is still an
absolute time. More discussion to be had in other words.
An additional change that would be appropriate in the future is to
perform a similar wrapping of Tick and probably also introduce a
Ticks class along with suitable operators for all these classes.
This patch tightens up the semantics around port binding and checks
that the ports that are being bound are currently not connected, and
similarly connected before unbind is called.
The patch consequently also changes the order of the unbind and bind
for the switching of CPUs to ensure that the rules are adhered
to. Previously the ports would be "over-written" without any check.
There are no changes in behaviour due to this patch, and the only
place where the unbind functionality is used is in the CPU.
The memory size variable was a 32-bit int. This meant that the size of the
memory was limited to 4GB. This patch changes the type of the variable to
64-bit to support larger memory sizes. Thanks to Raghuraman Balasubramanian
for bringing this to notice.
This patch removes the NACK frrom the packet as there is no longer any
module in the system that issues them (the bridge was the only one and
the previous patch removes that).
The handling of NACKs was mostly avoided throughout the code base, by
using e.g. panic or assert false, but in a few locations the NACKs
were actually dealt with (although NACKs never occured in any of the
regressions). Most notably, the DMA port will now never receive a NACK
and the backoff time is thus never changed. As a consequence, the
entire backoff mechanism (similar to a PCI bus) is now removed and the
DMA port entirely relies on the bus performing the arbitration and
issuing a retry when appropriate. This is more in line with e.g. PCIe.
Surprisingly, this patch has no impact on any of the regressions. As
mentioned in the patch that removes the NACK from the bridge, a
follow-up patch should change the request and response buffer size for
at least one regression to also verify that the system behaves as
expected when the bridge fills up.
This patch removes the NACKing in the bridge, as the split
request/response busses now ensure that protocol deadlocks do not
occur, i.e. the message-dependency chain is broken by always allowing
responses to make progress without being stalled by requests. The
NACKs had limited support in the system with most components ignoring
their use (with a suitable call to panic), and as the NACKs are no
longer needed to avoid protocol deadlocks, the cleanest way is to
simply remove them.
The bridge is the starting point as this is the only place where the
NACKs are created. A follow-up patch will remove the code that deals
with NACKs in the endpoints, e.g. the X86 table walker and DMA
port. Ultimately the type of packet can be complete removed (until
someone sees a need for modelling more complex protocols, which can
now be done in parts of the system since the port and interface is
split).
As a consequence of the NACK removal, the bridge now has to send a
retry to a master if the request or response queue was full on the
first attempt. This change also makes the bridge ports very similar to
QueuedPorts, and a later patch will change the bridge to use these. A
first step in this direction is taken by aligning the name of the
member functions, as done by this patch.
A bit of tidying up has also been done as part of the simplifications.
Surprisingly, this patch has no impact on any of the
regressions. Hence, there was never any NACKs issued. In a follow-up
patch I would suggest changing the size of the bridge buffers set in
FSConfig.py to also test the situation where the bridge fills up.
This patch extends the queued port interfaces with methods for
scheduling the transmission of a timing request/response. The methods
are named similar to the corresponding sendTiming(Snoop)Req/Resp,
replacing the "send" with "sched". As the queues are currently
unbounded, the methods always succeed and hence do not return a value.
This functionality was previously provided in the subclasses by
calling PacketQueue::schedSendTiming with the appropriate
parameters. With this change, there is no need to introduce these
extra methods in the subclasses, and the use of the queued interface
is more uniform and explicit.
This patch allows packets to be enqueued in the same tick as they are
intended to be sent. This does not imply they actually are sent that
tick, although that is possible.
This change is useful for module that use the queued ports primarly to
avoid handling the flow control involved in sending and retrying
packets.
This patch moves the clock of the CPU, bus, and numerous devices to
the new class ClockedObject, that sits in between the SimObject and
MemObject in the class hierarchy. Although there are currently a fair
amount of MemObjects that do not make use of the clock, they
potentially should do so, e.g. the caches should at some point have
the same clock as the CPU, potentially with a 1:n ratio. This patch
does not introduce any new clock objects or object hierarchies
(clusters, clock domains etc), but is still a step in the direction of
having a more structured approach clock domains.
The most contentious part of this patch is the serialisation of clocks
that some of the modules (but not all) did previously. This
serialisation should not be needed as the clock is set through the
parameters even when restoring from the checkpoint. In other words,
the state is "stored" in the Python code that creates the modules.
The nextCycle methods are also simplified and the clock phase
parameter of the CPU is removed (this could be part of a clock object
once they are introduced).
This patch fixes some problems with the drain/switchout functionality
for the O3 cpu and for the ARM ISA and adds some useful debug print
statements.
This is an incremental fix as there are still a few bugs/mem leaks with the
switchout code. Particularly when switching from an O3CPU to a
TimingSimpleCPU. However, when switching from O3 to O3 cores with the ARM ISA
I haven't encountered any more assertion failures; now the kernel will
typically panic inside of simulation.
This patch moves instantiateTopology into Ruby.py and removes the
mem/ruby/network/topologies directory. It also adds some extra inheritance to
the topologies to clean up some issues in the existing topologies.
Off-by-one loop termination meant that we were stuffing
the terminating '\0' into the std::string value, which
makes for difficult-to-debug string comparison failures.
removes the optimization that forwards an exclusive copy to a requester on a
read, only for the i-cache. this optimization isn't necessary because we
typically won't be writing to the i-cache.
This patch changes the simple memory to have a single slave port
rather than a vector port. The simple memory makes no attempts at
modelling the contention between multiple ports, and any such
multiplexing and demultiplexing could be done in a bus (or crossbar)
outside the memory controller. This scenario also matches with the
ongoing work on a SimpleDRAM model, which will be a single-ported
single-channel controller that can be used in conjunction with a bus
(or crossbar) to create a multi-port multi-channel controller.
There are only very few regressions that make use of the vector port,
and these are all for functional accesses only. To facilitate these
cases, memtest and memtest-ruby have been updated to also have a
"functional" bus to perform the (de)multiplexing of the functional
memory accesses.
This patch removes printConfig() functions from all structures in Ruby.
Most of the information is already part of config.ini, and where ever it
is not, it would become in due course.
This patch models a cache as separate tag and data arrays. The patch exposes
the banked array as another resource that is checked by SLICC before a
transition is allowed to execute. This is similar to how TBE entries and slots
in output ports are modeled.
Updates to Ruby to support statistics counting of cache accesses. This feature
serves multiple purposes beyond simple stats collection. It provides the
foundation for ruby to model the cache tag and data arrays as physical
resources, as well as provide the necessary input data for McPAT power
modeling.
Instead of just passing a list of controllers to the makeTopology function
in src/mem/ruby/network/topologies/<Topo>.py we pass in a function pointer
which knows how to make the topology, possibly with some extra state set
in the configs/ruby/<protocol>.py file. Thus, we can move all of the files
from network/topologies to configs/topologies. A new class BaseTopology
is added which all topologies in configs/topologies must inheirit from and
follow its API.
--HG--
rename : src/mem/ruby/network/topologies/Crossbar.py => configs/topologies/Crossbar.py
rename : src/mem/ruby/network/topologies/Mesh.py => configs/topologies/Mesh.py
rename : src/mem/ruby/network/topologies/MeshDirCorners.py => configs/topologies/MeshDirCorners.py
rename : src/mem/ruby/network/topologies/Pt2Pt.py => configs/topologies/Pt2Pt.py
rename : src/mem/ruby/network/topologies/Torus.py => configs/topologies/Torus.py
This patch makes the queue implementation in the SimpleTimingPort
private to avoid confusion with the protected member queue in the
QueuedSlavePort. The SimpleTimingPort provides the queue_impl to the
QueuedSlavePort and it can be accessed via the reference in the base
class. The use of the member name queue is thus no longer overloaded.
This patch is a first step to align the port names used in the Python
world and the C++ world. Ultimately it serves to make the use of
config.json together with output from the simulation easier, including
post-processing of statistics.
Most notably, the CPU, cache, and bus is addressed in this patch, and
there might be other ports that should be updated accordingly. The
dash name separator has also been replaced with a "." which is what is
used to concatenate the names in python, and a separation is made
between the master and slave port in the bus.
This patch changes the default bus width to a more sensible 8 bytes
(64 bits), which is in line with most on-chip buses. Although there
are cases where a wider or narrower bus is useful, the 8 bytes is a
good compromise to serve as the default.
This patch changes essentially all statistics, and will be bundled
with the outstanding changes to the bus.
This patch splits the existing buses into multiple layers. The
non-coherent bus is split into a request and a response layer, and the
coherent bus adds an additional layer for the snoop responses. The
layer is modified to be templatised on the port type, such that the
different layers can have retryLists with either master or slave
ports. This patch also removes the dynamic cast from the retry, as
previously promised when moving the recvRetry from the port base class
to the master/slave port respectively.
Overall, the split bus more closely reflects any modern on-chip bus
and should be at step in the right direction. From this point, it
would be reasonable straight forward to add separate layers (and thus
contention points and arbitration) for each port and thus create a
true crossbar.
The regressions all produce the correct output, but have varying
degrees of changes to their statistics. A separate patch will be
pushed with the updates to the reference statistics.
This patch moves all flow control, arbitration and state information
into a bus layer. The layer is thus responsible for all the state
transitions, and for keeping hold of the retry list. Consequently the
layer is also responsible for the draining.
With this change, the non-coherent and coherent bus are given a single
layer to avoid changing any temporal behaviour, but the patch opens up
for adding more layers.
This patch adds a state enum and member variable in the bus, tracking
the bus state, thus eliminating the need for tickNextIdle and inRetry,
and fixing an issue that allowed the bus to be occupied by multiple
packets at once (hopefully it also makes it easier to understand the
code).
The bus, in its current form, uses tickNextIdle and inRetry to keep
track of the state of the bus. However, it only updates tickNextIdle
_after_ forwarding a packet using sendTiming, and the result is that
the bus is still seen as idle, and a module that receives the packet
and starts transmitting new packets in zero time will still see the
bus as idle (and this is done by a number of DMA devices). The issue
can also be seen in isOccupied where the bus calls reschedule on an
event instead of schedule.
This patch addresses the problem by marking the bus as _not_ idle
already by the time we conclude that the bus is not occupied and we
will deal with the packet.
As a result of not allowing multiple packets to occupy the bus, some
regressions have slight changes in their statistics. A separate patch
updates these accordingly.
Further ahead, a follow-on patch will introduce a separate state
variable for request/responses/snoop responses, and thus implement a
split request/response bus with separate flow control for the
different message types (even further ahead it will introduce a
multi-layer bus).
This patch makes getAddrRanges const throughout the code base. There
is no reason why it should not be, and making it const prevents adding
any unintentional side-effects.
This patch adds getAddrRanges to the master port, and thus avoids
going through getSlavePort to be able to ask the slave. Similar to the
previous patch that added isSnooping to the SlavePort, this patch aims
to introduce an additional level of hierarchy in the ports (base port
being protocol-agnostic) and getSlave/MasterPort will return port
pointers to these base classes.
The function is named getAddrRanges also on the master port, but does
nothing besides asking the connected slave port. The slave port, as
before, has to provide an implementation and actually produce a list
of address ranges. The initial design used the name getSlaveAddrRanges
for the new function, but the more verbose name was later changed.
This patch adds isSnooping to the slave port, and thus avoids going
through getMasterPort to be able to ask the master. Over the course of
the next few patches, all getMasterPort/getSlavePort in Port and
MemObject are to be protocol agnostic, and the snooping is part of the
protocol layer.
The function is already present on the master port, where it is
implemented by the module itself, e.g. a cache. On the slave side, it
is merely asking the connected master port. The same name is used by
both functions despite their difference in behaviour. The initial
design used isMasterSnooping on the slave port side, but the more
verbose function name was later changed.
This patch is the last part of moving all protocol-related
functionality out of the Port base class. All the send/recv functions
are already moved, and the retry (which still governs all the timing
transport functions) is the only part that remained in the base class.
The only point where this currently causes a bit of inconvenience is
in the bus where the retry list is global and holds Port pointers (not
Master/SlavePort). This is about to change with the split into a
request/response bus and will soon be removed anyway.
The patch has no impact on any regressions.
This patch is the result of static analysis identifying a number of
memory leaks. The leaks are all benign as they are a result of not
deallocating memory in the desctructor. The fix still has value as it
removes false positives in the static analysis.
The LRU policy always evicted the least recently touched way, even if it
contained valid data and another way was invalid, as can happen if a block has
been invalidated by coherance. This can result in caches never warming up even
though they are replacing blocks. This modifies the LRU policy to move blocks
to LRU position on invalidation.
Currently when multiple CPUs perform a load-linked/store-conditional sequence,
the loads all create a list of reservations which is then scanned when the
stores occur. A reservation matching the context and address of the store is
sought, BUT all reservations matching the address are also erased at this point.
The upshot is that a store-conditional will remove all reservations even if the
store itself does not succeed. A livelock was observed using 7-8 CPUs where a
thread would erase the reservations of other threads, not succeed, loop and put
its own reservation in again only to have it blown by another thread that
unsuccessfully now tries to store-conditional -- no forward progress was made,
hanging the system.
The correct way to do this is to only blow a reservation when a store
(conditional or not) actually /occurs/ to its address. One thread always wins
(the one that does the store-conditional first).
This patch is a temporary fix until Andreas' four-phase patches
get reviewed and committed. Removing FastAlloc seems to have exposed
an issue which previously was reasonable rare in which packets are freed
before the sending cache is done with them. This change puts incoming packets
no a pendingDelete queue which are deleted at the start of the next call and
thus breaks the dependency between when the caller returns true and when the
packet is actually used by the sending cache.
Running valgrind on a multi-core linux boot and the memtester results in no
valgrind warnings.
While FastAlloc provides a small performance increase (~1.5%) over regular malloc it isn't thread safe.
After removing FastAlloc and using tcmalloc I've seen a performance increase of 12% over libc malloc
when running twolf for ARM.
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
This patch merely remove the Packet* from the isOccupied member
function. Historically this was used to check if the packet was an
express snoop, but this is now done outside this function (where
relevant).
The main aim of this patch is to arrive at a suitable port interface
for vector ports, including both the packet and the port id. This
patch changes the bus transport functions
(recvFunctional/Atomic/Timing) to require a PortId parameter
indicating the source port. Previously this information was passed by
setting the source field of the packet, and this is only required in
the case of a timing request.
With this patch, the use of the source and destination field is also
more restrictive, as they are only needed for timing accesses. The
modifications to these fields for atomic snoops is now removed
entirely, also making minor modifications to the cache.
This patch removes the Packet::NodeID typedef and unifies it with the
Port::PortId. The src and dest fields in the packet are used to hold a
port id (e.g. in the bus), and thus the two should actually be the
same.
The typedef PortID is now global (in base/types.hh) and aligned with
the ThreadID in terms of capitalisation and naming of the
InvalidPortID constant.
Before this patch, two flags were used for valid destination and
source, rather than relying on a named value (InvalidPortID), and
this is now redundant, as the src and dest field themselves are
sufficient to tell whether the current value is a valid port
identifier or not. Consequently, the VALID_SRC and VALID_DST are
removed.
As part of the cleaning up, a number of int parameters and local
variables are updated to use PortID.
Note that Ruby still has its own NodeID typedef. Furthermore, the
MemObject getMaster/SlavePort still has an int idx parameter with a
default value of -1 which should eventually change to PortID idx =
InvalidPortID.
This patch updates the comments for the src and dest fields to reflect
their actual use. Due to a number of patches (e.g. removing the
Broadcast flag), the old comments are no longer indicative of the
current usage.
This patch splits the PacketBuffer class into a RequestState and a
DeferredRequest and DeferredResponse. Only the requests need a
SenderState, and the deferred requests and responses only need an
associated point in time for the request and the response queue.
Besides the cleaning up, the goal is to simplify the transition to a
new port handshake, and with these changes, the two packet queues are
starting to look very similar to the generic packet queue, but
currently they do a few unique things relating to the NACK and
counting of requests/responses that the packet queue cannot be
conveniently used. This will be addressed in a later patch.
This patch removes unused commands and attributes from the packet to
avoid any confusion. It is part of an effort to clear up how and where
different commands and attributes are used.
The scheduling of the deadlock check event was being done incorrectly as the
clock was not being multiplied, so as to convert the time into ticks. This
patch removes that bug.
