Commit graph

62 commits

Author SHA1 Message Date
Steve Reinhardt
5fbf3aa471 Turn cache MissQueue/BlockingBuffer into virtual object
instead of template parameter.

--HG--
extra : convert_revision : fce0fbd041149b9c781eb23f480ba84fddbfd4a0
2006-12-04 09:10:53 -08:00
Ron Dreslinski
28fd4ab39f Do a functional access to levels above on a read as a temporary solution for L2's in FS
Fix a small writeback bug when missing in the L2 in atomic mode

src/mem/bus.cc:
    Fix a comment to make sense
src/mem/cache/cache_impl.hh:
    Do a functional access to levels above on a read as a temporary solution for L2's in FS
    Also fix a small writeback miss in L2 issue
src/mem/cache/coherence/simple_coherence.hh:
src/mem/cache/coherence/uni_coherence.cc:
src/mem/cache/coherence/uni_coherence.hh:
    Do a functional access to levels above on a read as a temporary solution for L2's in FS
tests/quick/00.hello/ref/alpha/linux/o3-timing/m5stats.txt:
tests/quick/00.hello/ref/alpha/linux/simple-timing/m5stats.txt:
tests/quick/01.hello-2T-smt/ref/alpha/linux/o3-timing/m5stats.txt:
    Update ref's for writeback changes

--HG--
extra : convert_revision : 937febd577b16b7fd97a5a68acaf53541828a251
2006-11-22 20:20:38 -05:00
Ron Dreslinski
a200bccc20 Handle packets being deleted by lower level properly.
Fixes for Mem Leak associated with Writebacks.

src/mem/cache/miss/mshr_queue.cc:
    Fixes for Mem Leak associated with Writebacks. (Double Delete removed)

--HG--
extra : convert_revision : 7a52ddd57da35995896f2c4438a58aa53f762416
2006-11-12 09:06:15 -05:00
Ron Dreslinski
11accacf7c Move code before a early return to make sure it is executed on all paths
--HG--
extra : convert_revision : cfdd5b6911422fbb733677c43d027aa4407fbc85
2006-11-12 06:44:05 -05:00
Ron Dreslinski
b22d390721 Yet another small bug in mem system related to flow control
src/mem/cache/cache_impl.hh:
    When upgrades change to readEx make sure to allocate the block
    Fix dprintf

--HG--
extra : convert_revision : 8700a7e47ad042c8708302620b907849c4bfdded
2006-11-12 06:36:33 -05:00
Ron Dreslinski
f876bc2bf0 More fixes for functional accesses. It now makes the writeback memory leak to crash all configs.
Working on that now.

src/mem/cache/base_cache.cc:
    Keep a list of the responders so we can search them on functional accesses.
src/mem/cache/base_cache.hh:
    Properly put things on a list for responses so we can search the list.
    Also, be sure to check the outgoing ports lists on a functional access (factor some common code out there)
src/mem/cache/cache_impl.hh:
    Properly return when the first read hit on a functional access.
    Make sure to call to check the other ports list of packets before forwarding it out.

--HG--
extra : convert_revision : 1d21cb55ff29c15716617efc48441329707c088a
2006-11-10 22:45:50 -05:00
Steve Reinhardt
e321a21e27 Clean up cache DPRINTFs
--HG--
extra : convert_revision : f836e77efd40e25259d7794dd148696586b79a09
2006-10-22 21:07:38 -07:00
Steve Reinhardt
1b21d9ba5e s/pktuest/request/ (all in comments)
--HG--
extra : convert_revision : 7ce779242a15245a20322c0b6c40d02c8ddd15ad
2006-10-22 20:38:34 -07:00
Steve Reinhardt
5e2263fc52 Small bug fixes for timing LL/SC. Better now but
not necessarily 100% there yet.

src/mem/cache/cache_impl.hh:
    Generate response packet on failed store conditional.
src/mem/packet.hh:
    Clear packet flags when reinitializing.
    (SATISFIED in particular is one we don't want to leave set.)

--HG--
extra : convert_revision : 29207c8a09afcbce43f41c480ad0c1b21d47454f
2006-10-21 23:35:00 -07:00
Steve Reinhardt
883ed108e4 Just give up if a store conditional misses completely
in the cache (don't treat as normal write miss).

--HG--
extra : convert_revision : c030eb6ba25318cae422e4da31e3b802049c8c74
2006-10-21 17:19:33 -07:00
Ron Dreslinski
54ed57cc4c Merge zizzer:/bk/newmem
into  zazzer.eecs.umich.edu:/z/rdreslin/m5bk/newmemcleanest

src/mem/tport.cc:
    Merge PacketPtr changes

--HG--
extra : convert_revision : 0329c5803a3df67af3dda89bd9d4753fd1a286d1
2006-10-20 13:04:59 -04:00
Ron Dreslinski
28e9641c2c Use fixPacket function everywhere.
Fix fixPacket assert function.
Stop timing port from forwarding the request if a response was found in its queue on a read.

src/cpu/memtest/memtest.cc:
src/cpu/memtest/memtest.hh:
src/python/m5/objects/MemTest.py:
    Add parameter to configure what percentage of mem accesses are functional
src/mem/cache/base_cache.cc:
src/mem/cache/cache_impl.hh:
    Use fix Packet function
src/mem/packet.cc:
    Fix an assert that was checking the wrong thing
src/mem/tport.cc:
    Properly detect if we need to do the access to the functional device

--HG--
extra : convert_revision : 447cc1a9a65ddd2a41e937fb09dc0e7c74e9c75e
2006-10-20 13:01:21 -04:00
Nathan Binkert
a4c6f0d69e Use PacketPtr everywhere
--HG--
extra : convert_revision : d9eb83ab77ffd2d725961f295b1733137e187711
2006-10-20 00:10:12 -07:00
Ron Dreslinski
780aa0a0eb Fix corner case on assertion.
I need to move over to using the fixPacket function so I don't have to make the same changes everywhere.
Still a functional access bug someplace I need to track down in timing mode.

src/mem/cache/base_cache.cc:
src/mem/cache/cache_impl.hh:
    Fix corner case on assertion
tests/configs/memtest.py:
    Updated memtester with uncacheable addresses and functional accesses

--HG--
extra : convert_revision : e6fa851621700ff9227b83cc5cac20af4fc8444f
2006-10-19 21:26:46 -04:00
Ron Dreslinski
cc1feb9f6d Fix memtester to use functional access, fix cache to work functionally now that we could test it.
src/cpu/memtest/memtest.cc:
    Fix memtest to do functional accesses
src/mem/cache/cache_impl.hh:
    Fix cache to handle functional accesses properly based on memtester changes
    Still need to fix functional accesses in timing mode now that the memtester can test it.

--HG--
extra : convert_revision : a6dbca4dc23763ca13560fbf5d41a23ddf021113
2006-10-19 21:07:53 -04:00
Ron Dreslinski
210e73f2a2 Small changes:
?? doesn't compile in warn statements
Should have been false, where I had a true.

src/cpu/o3/lsq_impl.hh:
    Apparently you can't have ?? in a warn statement (Something about trigraphs)
src/mem/cache/cache_impl.hh:
    Forgot to signal atomic mode in snoopProbe

--HG--
extra : convert_revision : c75cb76e193e852284564993440c8ea39e6de426
2006-10-19 20:18:17 -04:00
Ron Dreslinski
e34e564f79 Fixes to get single level uni-coherence to work.
Now to try L2 caches in FS.

src/mem/cache/base_cache.hh:
    Fix uni-coherence for atomic accesses in coherence protocol access to port
src/mem/cache/cache_impl.hh:
    Properly handle uni-coherence
src/mem/cache/coherence/simple_coherence.hh:
    Properly forward invalidates (not done for MSI+ protocols (assumed top level for now)
src/mem/cache/coherence/uni_coherence.cc:
src/mem/cache/coherence/uni_coherence.hh:
    Properly forward invalidates in atomic/timing uni-coherence

--HG--
extra : convert_revision : f0f11315e8e7f32c19d92287f6f9c27b079c96f7
2006-10-19 20:02:57 -04:00
Ron Dreslinski
9cf063eb8e Merge zizzer:/bk/newmem
into  zazzer.eecs.umich.edu:/z/rdreslin/m5bk/newmemcleanest

--HG--
extra : convert_revision : c6611b32537918f5bf183788227ddf69a9a9a069
2006-10-19 19:00:43 -04:00
Ron Dreslinski
39d24f7241 Always get the functional access from the highest level of cache first.
src/mem/cache/cache_impl.hh:
    Get the read data from the highest level of cache on a functional access

--HG--
extra : convert_revision : 7437ac46fb40f3ea3b42197a1aa8aec62af60181
2006-10-19 19:00:27 -04:00
Steve Reinhardt
bba3dfb0d3 First cut at LL/SC support in caches (atomic mode only).
configs/example/fs.py:
    Add MOESI protocol to caches (uni coherence not quite working w/FS yet).

--HG--
extra : convert_revision : 7bef7d9c5b24bf7241cc810df692408837b06b86
2006-10-19 00:33:33 -07:00
Steve Reinhardt
0e2561710b Break a lot of overly long lines.
Factor out some asserts that were on both
sides of an if/else.

--HG--
extra : convert_revision : 78f0c2d76a81a98216b2f281159c6b6ea0147731
2006-10-18 08:41:05 -07:00
Steve Reinhardt
6cd187e1f0 Get rid of obsolete in-cache copy support.
--HG--
extra : convert_revision : a701ed9d078c67718a33f4284c0403a8aaac7b25
2006-10-18 08:16:22 -07:00
Ron Dreslinski
4fff6d4603 Fixes to cache eliminating the assumption that the Packet is still valid after sending out a request.
Still need to rework upgrades into this system, but works for now.

src/mem/cache/base_cache.cc:
    Re order code to be more readable
src/mem/cache/base_cache.hh:
    Be sure to delete the copy on a bus block
src/mem/cache/cache_impl.hh:
    Be sure to remove the copy on a writeback success
src/mem/cache/miss/mshr_queue.cc:
    Demorgans to make it easier to understand
src/mem/tport.cc:
    Delete writebacks

--HG--
extra : convert_revision : 9519fb37b46ead781d340de29bb342a322a6a92e
2006-10-17 16:47:22 -04:00
Ron Dreslinski
6e8bfa4e63 Properly chack the pkt pointer on upgrades to insure no segfaults when writebacks delete the packet.
--HG--
extra : convert_revision : 72b1c6296a16319f4d16c62bc7038365654dbc40
2006-10-17 15:07:40 -04:00
Ron Dreslinski
288b98eb69 Fix it so that the cache does not assume to gave the packet it sent out via sendTiming.
Still need to fix upgrades to use this path

src/mem/cache/base_cache.cc:
    Copy the pkt to the MSHR before issuing the sendTiming where it may be changed/consumed
src/mem/cache/cache_impl.hh:
    Use copy of packet, because sendTiming may have changed the pkt
    Also, delete the copy when the time comes

--HG--
extra : convert_revision : 635cde6b4f08d010affde310c46b1caf50fbe424
2006-10-17 15:05:21 -04:00
Ron Dreslinski
a17afb1649 Fix for DMA's in FS caches.
Fix CSHR's for flow control.
Fix for Bus Bridges reusing packets (clean flags up)

Now both timing/atomic caches with MOESI in UP fail at same point.

src/dev/io_device.hh:
    DMA's should send WriteInvalidates
src/mem/bridge.cc:
    Reusing packet, clean flags in the packet set by bus.
src/mem/cache/base_cache.cc:
src/mem/cache/base_cache.hh:
src/mem/cache/cache.hh:
src/mem/cache/cache_impl.hh:
src/mem/cache/coherence/simple_coherence.hh:
src/mem/cache/coherence/uni_coherence.cc:
src/mem/cache/coherence/uni_coherence.hh:
    Fix CSHR's for flow control.
src/mem/packet.hh:
    Make a writeInvalidateResp, since the DMA expects responses to it's writes

--HG--
extra : convert_revision : 59fd6658bcc0d076f4b143169caca946472a86cd
2006-10-13 15:47:05 -04:00
Ron Dreslinski
fe230ddb8f Remove bus and top level parameters from cache
src/mem/cache/base_cache.hh:
    Remove top level param from cache
src/mem/cache/coherence/uni_coherence.cc:
    Remove top level parameters from the cache

--HG--
extra : convert_revision : 4437aeedc20866869de7f9ab123dfa7baeebedf0
2006-10-12 14:21:25 -04:00
Ron Dreslinski
ba4c224c39 Fix problems with unCacheable addresses in timing-coherence
src/base/traceflags.py:
src/mem/physical.cc:
    Add debug falgs fro physical memory accesses
src/mem/cache/cache_impl.hh:
    Snoops to uncacheable blocks should not happen
src/mem/cache/miss/miss_queue.cc:
    Set the size properly on unCacheable accesses

--HG--
extra : convert_revision : fc78192863afb11fc7c591fba169021b9e127d16
2006-10-12 13:33:21 -04:00
Ron Dreslinski
567afbf6ce More cache fixes. Atomic coherence now works as well.
src/cpu/memtest/memtest.cc:
src/cpu/memtest/memtest.hh:
    Make Memtester able to test atomic as well
src/mem/bus.cc:
src/mem/bus.hh:
    Handle atomic snoops properly for cache->cache transfers
src/mem/cache/cache_impl.hh:
    Debug output.
    Clean up memleak in atomic mode.
    Set hitLatency.
    Still need to send back reasonable number for atomic return value.
src/mem/packet.cc:
    Add command strings for new commands
src/python/m5/objects/MemTest.py:
    Add param to test atomic memory.

--HG--
extra : convert_revision : 43f880e29215776167c16ea90793ebf8122c785b
2006-10-11 18:28:33 -04:00
Ron Dreslinski
995146ead7 Fix some more mem leaks, still some left
Update retry mechanism

src/mem/cache/base_cache.cc:
    Rework the retry mechanism
src/mem/cache/base_cache.hh:
    Rework the retry mechanism
    Try to fix memory bug
src/mem/cache/cache_impl.hh:
    Rework upgrades to not be blocked by slave
src/mem/cache/miss/mshr_queue.cc:
    Fix mem leak on writebacks

--HG--
extra : convert_revision : 3cec234ee441edf398ec8d0f51a0c5d7ada1e2be
2006-10-10 17:10:56 -04:00
Ron Dreslinski
cc78d86661 Fix several bugs pertaining to upgrades/mem leaks.
src/mem/cache/base_cache.cc:
    Fix a bug about not having a request to send
src/mem/cache/base_cache.hh:
    Fix a bug with the blocking code
src/mem/cache/cache.hh:
    AFix a bug with snoop hits in WB buffer
src/mem/cache/cache_impl.hh:
    Fix a bug with snoop hits in WB buffer
    Also, add better DPRINTF's
src/mem/cache/miss/miss_queue.cc:
    Fix a bug with upgrades (Need to clean it up later)
src/mem/cache/miss/mshr.cc:
    Fix a memory leak bug, still some outstanding with writebacks not being deleted
src/mem/cache/miss/mshr_queue.cc:
    Fix a bug about upgrades (need to clean up later)
src/mem/packet.hh:
    Fix for newly added cmd attribute for upgrades
tests/configs/memtest.py:
    More interesting testcase

--HG--
extra : convert_revision : fcb4f17dd58b537bb4f67a8c835f50e455e8c688
2006-10-10 01:32:18 -04:00
Ron Dreslinski
ec8a437b2c Handle NACK's that occur from devices on the same bus.
Not fully implemented yet, but good enough for single level cache coherence

src/mem/packet.hh:
    Add a bit to distinguish invalidates and upgrades

--HG--
extra : convert_revision : 5bf50d535857cea37fbdaf7993915d1332cb757e
2006-10-09 20:18:00 -04:00
Ron Dreslinski
e03b9c9939 Fix how upgrades work.
Remove some dead code.

src/mem/cache/cache_impl.hh:
    Upgrades don't need a response.
    Moved satisfied check into bus so removed some dead code.
src/mem/cache/coherence/coherence_protocol.cc:
src/mem/packet.hh:
    Upgrades don't require a response

--HG--
extra : convert_revision : dee0440ff19ba4c9e51bf9a47a5b0991265cfc1d
2006-10-09 19:15:24 -04:00
Ron Dreslinski
13ac9a419d One step closet to having NACK's work.
src/cpu/memtest/memtest.cc:
    Fix functional return path
src/cpu/memtest/memtest.hh:
    Add snoop ranges in
src/mem/cache/base_cache.cc:
    Properly signal NACKED
src/mem/cache/cache_impl.hh:
    Catch nacked packet and panic for now

--HG--
extra : convert_revision : 59a64e82254dfa206681c5f987e6939167754d67
2006-10-09 18:52:20 -04:00
Ron Dreslinski
c4dba7a8ed Fix a typo in the printf
--HG--
extra : convert_revision : bfa8ffae0a9bef25ceca168ff376ba816abf23f3
2006-10-09 17:30:54 -04:00
Ron Dreslinski
fd27c229b6 Fix a bitwise operation that was accidentally a logical operation.
--HG--
extra : convert_revision : 30f64bcb6bea47fd8cd6d77b0df17eff04dbbad0
2006-10-09 17:18:34 -04:00
Ron Dreslinski
d7c1557e7e Merge zizzer:/z/m5/Bitkeeper/newmem
into  zazzer.eecs.umich.edu:/z/rdreslin/m5bk/newmemcleanest

--HG--
extra : convert_revision : b4cb1702ffa2fca298cfde47683cac019e1da900
2006-10-09 16:48:58 -04:00
Ron Dreslinski
45732376f6 Add more DPRINTF's fix a supply condition.
src/mem/cache/cache_impl.hh:
    Add more usefull DPRINTF's
    REmove the PC to get rid of asserts

--HG--
extra : convert_revision : 3f6d832b138d058dbe79bb5f42bd2db9c50b35b5
2006-10-09 16:47:55 -04:00
Ron Dreslinski
afce51d10a Set size properly on uncache accesses
Don't use the senderState after you get a succesful sendTiming.  Not guarnteed to be correct

src/mem/cache/base_cache.cc:
src/mem/cache/base_cache.hh:
src/mem/cache/cache.hh:
src/mem/cache/cache_impl.hh:
src/mem/cache/miss/blocking_buffer.cc:
src/mem/cache/miss/blocking_buffer.hh:
src/mem/cache/miss/miss_queue.hh:
    Don't use the senderState after you get a succesful sendTiming.  Not guarnteed to be correct

--HG--
extra : convert_revision : 2e8e812bf7fd3ba2b4cba7f7173cb41862f761af
2006-10-09 16:37:02 -04:00
Ron Dreslinski
0087061681 Don't create a response if one isn't needed.
--HG--
extra : convert_revision : 37bd230f527f64eb12779157869aae9dcfdde7fd
2006-10-09 00:27:41 -04:00
Ron Dreslinski
4cfddc0d77 Make sure to propogate sendFunctional calls with functional not atomic.
src/mem/cache/cache_impl.hh:
    Fix a error case by putting a panic in.
    Make sure to propogate sendFunctional calls with functional not atomic.

--HG--
extra : convert_revision : 05d03f729a40cfa3ecb68bcba172eb560b24e897
2006-10-08 20:47:50 -04:00
Ron Dreslinski
8a539a774f Merge zizzer:/z/m5/Bitkeeper/newmem
into  zazzer.eecs.umich.edu:/z/rdreslin/m5bk/newmemcleanest

--HG--
extra : convert_revision : f3067efb7f3ff30158d541dfc52de4ea8edae576
2006-10-08 18:49:30 -04:00
Ron Dreslinski
1345183a89 Move away from using the statusChange function on snoops. Clean up snooping code in general.
--HG--
extra : convert_revision : 5a57bfd7742a212047fc32e8cae0dc602fdc915c
2006-10-08 18:48:03 -04:00
Steve Reinhardt
5df93cc1cd Replace tests of LOCKED/UNCACHEABLE flags with isLocked()/isUncacheable().
--HG--
extra : convert_revision : f22ce3221d270ecf8631d3dcaed05753accd5461
2006-10-08 14:48:24 -07:00
Ron Dreslinski
178d114fa5 Fix infinite writebacks bug in cache.
src/mem/cache/cache_impl.hh:
    Make sure to pop the list.  Fixes infinite writeback bug.
src/mem/cache/miss/mshr_queue.cc:
    Add an assert as sanity check in case .full() stops working again.

--HG--
extra : convert_revision : d847e49a397eeb0b7c5ac060fcfc3eaeac921311
2006-10-07 11:36:55 -04:00
Ron Dreslinski
c42a7bc4f6 Merge zizzer:/z/m5/Bitkeeper/newmem
into  zazzer.eecs.umich.edu:/z/rdreslin/m5bk/newmemcleanest

--HG--
extra : convert_revision : 2f1bbe84c92879fd1bfa579adc62a367ece1cddd
2006-10-06 09:28:16 -04:00
Lisa Hsu
9c901225f8 there are two main thrusts of this changeset.
1) return the periodicity of checkpoints back into the code (i.e. make m5 checkpoint n m meaningful again).
2) to do this, i had to much around with being able to repeatedly schedule and SimLoopExitEvent, which led to changes in how exit simloop events are handled to make this easier.

src/arch/alpha/isa/decoder.isa:
src/mem/cache/cache_impl.hh:
    modify arg. order for new calling convention of exitSimLoop.
src/cpu/base.cc:
src/sim/main.cc:
src/sim/pseudo_inst.cc:
src/sim/root.cc:
    now, instead of creating a new SimLoopExitEvent, call a wrapper schedExitSimLoop which handles all the default args.
src/sim/sim_events.cc:
src/sim/sim_events.hh:
src/sim/sim_exit.hh:
    add the periodicity of checkpointing back into the code.

    to facilitate this, there are now two wrappers (instead of just overloading exitSimLoop).  exitSimLoop is only for exiting NOW (i.e. at curTick), while schedExitSimLoop schedules and exit event for the future.

--HG--
extra : convert_revision : c61f4bf05517172edd2c83368fd10bb0f0678029
2006-10-06 01:27:02 -04:00
Ron Dreslinski
212c5aefb5 Fixes for functional accesses to use the snoop path.
And small other tweaks to snooping coherence.

src/mem/cache/base_cache.hh:
    Make timing response at the time of send.
src/mem/cache/cache.hh:
src/mem/cache/cache_impl.hh:
    Update probe interface to be bi-directional for functional accesses
src/mem/packet.hh:
    Add the function to create an atomic response to a given request

--HG--
extra : convert_revision : 04075a117cf30a7df16e6d3ce485543cc77d4ca6
2006-10-05 23:28:03 -04:00
Ron Dreslinski
45f881a4ce First pass at snooping stuff that compiles and doesn't break.
Still need:
-Handle NACK's on the recieve side
-Distinguish top level caches
-Handle repsonses from caches failing the fast path
-Handle BusError and propogate it
-Fix the invalidate packet associated with snooping in the cache

src/mem/bus.cc:
    Make sure to snoop on functional accesses
src/mem/cache/base_cache.cc:
    Wait to make a request into a response until it is ready to be issued
src/mem/cache/base_cache.hh:
    Support range changes for snoops
    Set up snoop responses for cache->cache transfers
src/mem/cache/cache_impl.hh:
    Only access the cache if it wasn't satisfied by cache->cache transfer
    Handle snoop phases (detect block, then snoop)
    Fix functional access to work properly (still need to fix snoop path for functional accesses)

--HG--
extra : convert_revision : 4c25f11d7a996c1f56f4f7b55dde87a344e5fdf8
2006-10-05 21:10:03 -04:00
Ron Dreslinski
95e7b85c8b Still need LL/SC support in cache, add hack to always return success for now
--HG--
extra : convert_revision : b354bd91be8c1bbb3aca7b4ba9e7e3e117ced164
2006-08-22 16:09:34 -04:00