Commit Graph

2132 Commits

Author SHA1 Message Date
Ali Saidi
64528df38d In the case that we generate a fault (e.g. a tlb miss) on a microcoded instruction set curMacroStaticInst to null
This way we'll jump immediately to the handler

--HG--
extra : convert_revision : 36218d3a5c2342337e66e1229ea2219533efd41e
2007-01-16 19:12:33 -05:00
Ali Saidi
d6c92cdb3c Fix legion lock code a bit so that if we jump out of a micro coded instruction (because of a fault on the first op) we don't lose sync with legion
Only print TLB if there is a tlb difference

--HG--
extra : convert_revision : f3baf667ca466d6b8efcaccd186ecec14498229d
2007-01-16 19:08:21 -05:00
Ali Saidi
ecfd628ecd Modify ISA and staticInst to support a IsFirstMicroOp flag
Increment instruction count on first micro-op instead of last

src/arch/sparc/isa/decoder.isa:
    Implement a twin load for ASI_LDTX_P(0xe2)
src/arch/sparc/isa/formats/mem/blockmem.isa:
    set the new flag IsFirstMicroOp when needed
src/cpu/simple/atomic.cc:
    Increment instruction count on first micro-op instead of last (because if we take a fault on a micro coded instruction it should be counted twice acording to legion)
src/cpu/static_inst.hh:
    Add IsFirstMicroop flag to static insts

--HG--
extra : convert_revision : 02bea93d38c03bbafe4570665eb4c01c11caa2fc
2007-01-16 19:06:05 -05:00
Lisa Hsu
032ea9b2db the way i understand it, interrupts in m5 is a little bloated. the usage of CPU->checkInterrupts bool is inconsistent, and i think should eventually be phased out. For now, I've just assumed that CPU->checkInterrupts() is the way to fast path a CPU if you have no interrupts by having a simple bitfield in each ISA to determine whether interrupts are pending. getInterrupts has been mostly filled in.
src/arch/sparc/interrupts.hh:
    fill in how we do interrupts on sparc a little bit.

    1) create a bitfield for interrupts, and check that in checkInterrupts() to fast path CPU.
    2) fill in getInterrupts() a little bit.

    also, update the bitfield access to be HPSTATE::hpriv, etc.
src/arch/sparc/ua2005.cc:
    1) update formatting
    2) change the way interrupts are done to use the new way to tickle the CPU.
src/cpu/base.cc:
src/cpu/base.hh:
    overload the post_interrupt function for SPARC interrupts - which are only denoted by a single int value.

--HG--
extra : convert_revision : 9074a003eff37a40dcce78f56d20f6cbcc453eb5
2007-01-08 18:18:28 -05:00
Ali Saidi
a8b2d66661 change when legion-lock causes the simulation to die. It now happens after two consuctive differences since we compare stuff
at slightly different times interrupts are seen the cycle before they happen in m5 so the pc gets changed early.

--HG--
extra : convert_revision : f237363eababb2aad67e5b41670cf40be048a042
2007-01-08 17:11:10 -05:00
Nathan Binkert
e9a395c2ce Formatting
--HG--
extra : convert_revision : bf1eae73995f772a4343c8ebcb254818eeb5d949
2007-01-03 10:13:45 -08:00
Gabe Black
8840ebcb00 Merge zizzer:/bk/newmem
into  zower.eecs.umich.edu:/eecshome/m5/newmem

--HG--
extra : convert_revision : f4a05accb8fa24d425dd818b1b7f268378180e99
2007-01-03 00:52:30 -05:00
Kevin Lim
7d7f3d0e99 Fix up previous commit to proper logic.
src/cpu/o3/commit_impl.hh:
    Oops, changed the logic a little bit.  Fix it up to how it used to be.

--HG--
extra : convert_revision : df7f69b0997207b611374c3c92880f3a405e88be
2006-12-30 13:21:25 -05:00
Gabe Black
a0e8aa6737 Fixes to get non-delay slot ISAs (Alpha) working again, and pulling some debug output out of ifdefs.
--HG--
extra : convert_revision : 29d0969e2d3e809aac32262ba20907e6e4ef1a42
2006-12-28 14:35:31 -05:00
Gabe Black
3f2b25d997 Phased out DelaySlotInfo.
--HG--
extra : convert_revision : ab48db10caf38137300da63078aa9360f46b9631
2006-12-28 14:33:45 -05:00
Gabe Black
d24f60788f Some fixes for decode stage branches without delay slots. This will need some work to be compatible with delay slots too. Also changed some direct variable uses to use an accessor function.
--HG--
extra : convert_revision : b291292600e9d3e7e4a8255daf54342b736c7e35
2006-12-28 14:32:41 -05:00
Gabe Black
15df0a27bb Make sure the value of PC is actually updated now that the instruction target isn't set explicitly.
--HG--
extra : convert_revision : 4c00a219ac1d82abea78e4e8d70f529a435fdfe2
2006-12-28 14:29:17 -05:00
Gabe Black
b642ad00eb Implement a stub nnpc for alpha that is read only as npc+4.
--HG--
extra : convert_revision : d08b740d32757fa5471c9bcde9084d59a1d8102d
2006-12-28 14:27:45 -05:00
Ali Saidi
b48a8fb347 Merge zizzer:/bk/newmem
into  zeep.pool:/z/saidi/work/m5.newmem

--HG--
extra : convert_revision : afd4266bd494bb8f127c06985f343219ded4f637
2006-12-27 14:38:22 -05:00
Ali Saidi
ff88f3b13a Compare legion and m5 tlbs for differences
Only print faults instructions that aren't traps or faulting loads

src/cpu/exetrace.cc:
    Compare the legion and m5 tlbs and printout any differences
    Only show differences if the instruction isn't a trap and isn't a memory
    operation that changes the trap level (a fault)
src/cpu/m5legion_interface.h:
    update the m5<->legion interface to add tlb data

--HG--
extra : convert_revision : 6963b64ca1012604e6b1d3c5e0e5f5282fd0164e
2006-12-27 14:35:23 -05:00
Kevin Lim
0bd7518480 Remove some #if FULL_SYSTEMs so MP stuff works even in SE mode.
--HG--
extra : convert_revision : 5c334ec806305451b3883c7fd0ed9cd695c038bc
2006-12-26 01:43:18 -05:00
Nathan Binkert
ba191d85c2 style
--HG--
extra : convert_revision : 6bbaaa88a608081eebf706ff30293f38729415aa
2006-12-21 22:34:19 -08:00
Nathan Binkert
9aecfb3e3b don't use (*activeThreads).begin(), use activeThreads->blah().
Also don't call (*activeThreads).end() over and over.  Just
call activeThreads->end() once and save the result.
Make sure we always check that there are elements in the list
before we grab the first one.

--HG--
extra : convert_revision : d769d8ed52da99532d57a9bbc93e92ddf22b7e58
2006-12-20 22:20:11 -08:00
Nathan Binkert
4b3538b609 Merge zizzer.eecs.umich.edu:/bk/newmem
into  iceaxe.:/Volumes/work/m5/incoming

--HG--
extra : convert_revision : c1724538f27091e16ca495c8fdf2df06f55f7668
2006-12-20 21:46:39 -08:00
Nathan Binkert
6487d358a4 <scold> Make sure that variables are always initalized! </scold>
--HG--
extra : convert_revision : 1e946d9b1e1def36f9b8a73986dabf1b77096327
2006-12-20 21:46:16 -08:00
Gabe Black
68a0e6f2e9 Fixes to get MIPS_SE to compile.
--HG--
extra : convert_revision : d173f212841341e436e9a38dcd3006d27886c1b8
2006-12-20 22:14:40 -05:00
Gabe Black
327f451eb7 Fixes to get ALPHA_FS and ALPHA_SE to compile again.
--HG--
extra : convert_revision : 6e0913903d4cbda6f31bec3b5d725b9c08dc1419
2006-12-20 20:44:06 -05:00
Gabe Black
f13155393d Initial work to make remote gdb available in SE mode. This is completely untested.
--HG--
extra : convert_revision : 3ad9a3368961d5e9e71f702da84ffe293fe8adc8
2006-12-20 18:39:40 -05:00
Ali Saidi
5e9d8795f2 fix twinx loads a little bit
bugfixes and demap implementation in tlb
ignore some more differencs for one cycle

src/arch/sparc/isa/formats/mem/blockmem.isa:
    twinx has 2 micro-ops
src/arch/sparc/isa/formats/mem/util.isa:
    fix the fault check for twinx
src/arch/sparc/tlb.cc:
    tlb bugfixes and write demapping code
src/cpu/exetrace.cc:
    don't halt on a couple more instruction (ldx, stx) when things differ
    beacuse of the way tlb faults are handled in legion.

--HG--
extra : convert_revision : 1e156dead6ebd58b257213625ed63c3793ef4b71
2006-12-19 02:11:33 -05:00
Gabe Black
5b41ab694c Fix a place where the wrong width parameter was used, and set the nextNPC correctly on memory squashes.
--HG--
extra : convert_revision : 7914a48ea953607c48f93984e3b043098f0d7c62
2006-12-18 18:20:13 -05:00
Gabe Black
dfafe6741f Make sure you only handle branch delay slots specially when there actually was a branch.
--HG--
extra : convert_revision : ea6d33b1b9c2ba5c24225af4b10a9bd25558f1dd
2006-12-18 18:18:37 -05:00
Steve Reinhardt
968048f56a Convert Alpha (and finish converting MIPS) to new
InstObjParam interface.

src/arch/alpha/isa/branch.isa:
src/arch/alpha/isa/fp.isa:
src/arch/alpha/isa/int.isa:
src/arch/alpha/isa/main.isa:
src/arch/alpha/isa/mem.isa:
src/arch/alpha/isa/pal.isa:
src/arch/mips/isa/formats/mem.isa:
src/arch/mips/isa/formats/util.isa:
    Get rid of CodeBlock calls to adapt to new InstObjParam interface.
src/arch/isa_parser.py:
    Check template code for operands (in addition to snippets).
src/cpu/o3/alpha/dyn_inst.hh:
    Add (read|write)MiscRegOperand calls to Alpha DynInst.

--HG--
extra : convert_revision : 332caf1bee19b014cb62c1ed9e793e793334c8ee
2006-12-17 19:27:50 -08:00
Gabe Black
9d0ca61b7e Merge zizzer:/bk/newmem
into  zower.eecs.umich.edu:/eecshome/m5/newmem

src/arch/isa_parser.py:
src/arch/sparc/isa/formats/mem/basicmem.isa:
src/arch/sparc/isa/formats/mem/blockmem.isa:
src/arch/sparc/isa/formats/mem/util.isa:
src/arch/sparc/miscregfile.cc:
src/arch/sparc/miscregfile.hh:
src/cpu/o3/iew_impl.hh:
    Hand Merge

--HG--
extra : convert_revision : ae1b25cde85ab8ec275a09d554acd372887d4d47
2006-12-16 11:35:40 -05:00
Gabe Black
f4f00c5ae9 Switch the endianness of data that's forwarded. This is the same sort of problem that was happening when stores went all the way to memory and back.
--HG--
extra : convert_revision : 09fece7ae934f542e51046d33505df3f7ec0b919
2006-12-16 09:35:09 -05:00
Gabe Black
96e5086c81 Make fetch detect when a branch is happening, rather than trying to compute when.
--HG--
extra : convert_revision : 1a8edc004570abb48e6c4cdf1b43c5699866838e
2006-12-16 09:34:20 -05:00
Gabe Black
a6eb16adb4 Accidently "cleaned" away the NPC parameter to the constructor.
--HG--
extra : convert_revision : 46670ee86000dfb171d327eb8f58555a4afb2360
2006-12-16 07:47:33 -05:00
Gabe Black
f410d5f4e0 Don't have "predict" set the predicted target of the instruction. Do that explicitly when you use predict.
--HG--
extra : convert_revision : 8b613bb365b31ffaef1cea9fd789abe46219bdcf
2006-12-16 07:39:44 -05:00
Gabe Black
c6944e320c Add in capability to return to unblocking after a squash. This is needed because if you don't squash -all- the instructions, you need to keep clearing out whatever is left in the skid buffer.
--HG--
extra : convert_revision : 7308eda27f4366348cf5fce71ddfa4b217bc172d
2006-12-16 07:35:56 -05:00
Gabe Black
244506ae12 Make sure endian conversion is done on the memory data when it's just set to an existing buffer.
--HG--
extra : convert_revision : 5a890091b6a31b5414acbf68f19e28d7122a98d7
2006-12-16 07:34:34 -05:00
Gabe Black
6413b74e4f Make the decoder use the new setup in the dyninsts for branch prediction.
--HG--
extra : convert_revision : 9a6d6c93e5b40a55774891df54d290ff557b322c
2006-12-16 07:33:08 -05:00
Gabe Black
37b9966eb4 Made branch delay slots get squashed, and passed back an NPC and NNPC to start fetching from.
--HG--
extra : convert_revision : a2e4845fedf113b5a2fd92d3d28ce5b006278103
2006-12-16 07:32:06 -05:00
Gabe Black
4d66ddbe35 Added a predicted NPC field, explicitly stored whether the instruction was predicted taken or not.
--HG--
extra : convert_revision : ba668af302ca4d8a3a032e907d5058e1477f462a
2006-12-16 07:22:19 -05:00
Gabe Black
6aa06a26b7 Changes to the isa_parser and affected files to fix an indexing problem with split execute instructions and miscregs aliasing with integer registers.
src/arch/isa_parser.py:
    Rearranged things so that classes with more than one execute function treat operands properly.
    1. Eliminated the CodeBlock class
    2. Created a SubOperandList
    3. Redefined how InstObjParams is constructed

    To define an InstObjParam, you can either pass in a single code literal which will be named "code", or you can pass in a dictionary of code snippets which will be substituted into the Templates. In order to get this to work, there is a new restriction that each template has only one function in it. These changes should only affect memory instructions which have regular and split execute functions.

    Also changed the MiscRegs so that they use the instrunctions srcReg and destReg arrays.
src/arch/sparc/isa/formats/basic.isa:
src/arch/sparc/isa/formats/branch.isa:
src/arch/sparc/isa/formats/integerop.isa:
src/arch/sparc/isa/formats/mem/basicmem.isa:
src/arch/sparc/isa/formats/mem/blockmem.isa:
src/arch/sparc/isa/formats/mem/util.isa:
src/arch/sparc/isa/formats/nop.isa:
src/arch/sparc/isa/formats/priv.isa:
src/arch/sparc/isa/formats/trap.isa:
    Rearranged to work with new InstObjParam scheme.
src/cpu/o3/sparc/dyn_inst.hh:
    Added functions to access the miscregs using the indexes from instructions srcReg and destReg arrays. Also changed the names of the other accessors so that they have the suffix "Operand" if they use those arrays.
src/cpu/simple/base.hh:
    Added functions to access the miscregs using the indexes from instructions srcReg and destReg arrays.

--HG--
extra : convert_revision : c91e1073138b72bcf4113a721e0ed40ec600cf2e
2006-12-16 07:10:04 -05:00
Lisa Hsu
551ba56ae2 little fixes i noticed while searching for reason for address range issues (but these weren't the cause of the problem).
RangeSize as a function takes a start address, and a SIZE, and will make the range (start, start+size-1) for you.

src/cpu/memtest/memtest.hh:
src/cpu/o3/fetch.hh:
src/cpu/o3/lsq.hh:
src/cpu/ozone/front_end.hh:
src/cpu/ozone/lw_lsq.hh:
src/cpu/simple/atomic.hh:
src/cpu/simple/timing.hh:
    Fix RangeSize arguments
src/dev/alpha/tsunami_cchip.cc:
src/dev/alpha/tsunami_io.cc:
src/dev/alpha/tsunami_pchip.cc:
src/dev/baddev.cc:
    pioSize indicates SIZE, not a mask

--HG--
extra : convert_revision : d385521fcfe58f8dffc8622260937e668a47a948
2006-12-15 17:55:47 -05:00
Lisa Hsu
991146218d Merge zizzer:/bk/sparcfs
into  zed.eecs.umich.edu:/z/hsul/work/sparc/m5

--HG--
extra : convert_revision : 2f11b5f9fa6356cbf9f98c8cd7d4f6fbfaf9d24d
2006-12-15 13:27:53 -05:00
Lisa Hsu
573d59441e some small general fixes to make everythign work nicely with other ISAs, now we can merge back with newmem.
exetrace.cc:
wrap this variable between FULL_SYSTEM #ifs
mmaped_ipr.hh:
fix for build
miscregfile.cc:
fixes for HPSTATE access during SE mode

src/arch/sparc/miscregfile.cc:
    fixes for HPSTATE access during SE mode
src/arch/mips/mmaped_ipr.hh:
    fix for build
src/cpu/exetrace.cc:
    wrap this variable between FULL_SYSTEM #ifs

--HG--
extra : convert_revision : c5b9d56ab99018a91d04de47ba1d5ca7768590bb
2006-12-15 13:05:46 -05:00
Lisa Hsu
0fa30e579e Merge zizzer:/bk/newmem
into  zed.eecs.umich.edu:/z/hsul/work/sparc/m5

--HG--
extra : convert_revision : 82733f9c7bf833cf6bbfbd2aad292f69f52d21bc
2006-12-13 14:33:59 -05:00
Lisa Hsu
a983c4968c Merge zizzer:/bk/sparcfs
into  zed.eecs.umich.edu:/z/hsul/work/sparc/m5

--HG--
extra : convert_revision : c6d174716641f0b8286b8478bcb9053b3eec54e3
2006-12-13 14:33:32 -05:00
Lisa Hsu
5d42fd836b Merge zizzer:/bk/newmem
into  zed.eecs.umich.edu:/z/hsul/work/sparc/m5

--HG--
extra : convert_revision : 6e58629b1e51f1fc493a89f16c3f2e676dc5d191
2006-12-12 21:19:51 -05:00
Gabe Black
90907f6b3c Merge zizzer:/bk/newmem/
into  zower.eecs.umich.edu:/eecshome/m5/newmem

--HG--
extra : convert_revision : 17d6c49ee15af5d192dedf82871159219d4277cd
2006-12-12 18:10:00 -05:00
Kevin Lim
bc05f5982e Merge ktlim@zizzer:/bk/newmem
into  zamp.eecs.umich.edu:/z/ktlim2/clean/tmp/test-regress

--HG--
extra : convert_revision : d420ee86454b72b0e5d3a98bac3b496f172c1788
2006-12-12 17:55:50 -05:00
Ali Saidi
139519ef87 Fix bugs in tlbmap (and thus rangemap since the code is nearly identical)
Deal with block initializing stores (by doing nothing, at some point we might want to do the write hint 64 like thing)
Fix tcc instruction igoner in legion-lock stuff to be correct in all cases
Have console interrupts warn rather than panicing until we figure out what to do with interrupts

src/arch/sparc/miscregfile.cc:
src/arch/sparc/miscregfile.hh:
    add a magic miscreg which reads all the bits the tlb needs in one go
src/arch/sparc/tlb.cc:
    initialized the context type and id to reasonable values and handle block init stores
src/arch/sparc/tlb_map.hh:
    fix bug in tlb map code
src/base/range_map.hh:
    fix bug in rangemap code and add range_multimap
    (these are probably useful for bus range stuff)
src/cpu/exetrace.cc:
    fixup tcc ignore code to be correct
src/dev/sparc/t1000.cc:
    make console interrupt stuff warn instead of panicing until we get interrupt stuff figured out
src/unittest/rangemaptest.cc:
    fix up the rangemap unit test to catch the missing case

--HG--
extra : convert_revision : 70604a8b5d0553aa0b0bd7649f775a0cfa8267a5
2006-12-12 17:55:27 -05:00
Kevin Lim
c7ad7b44eb Allow for multiple redirects to happen on a single cycle (only the one for the oldest instruction is passed on to commit).
This fixes a minor bug when multiple FU completions come back out of order (due to the order in which the FUs are freed up), and the oldest redirect isn't recorded properly.  The eon benchmark should run now.

src/cpu/o3/iew_impl.hh:
    Allow for multiple redirects to happen on a single cycle (only the one for the oldest instruction is passed on to commit).

--HG--
extra : convert_revision : b7d202dee1754539ed814f0fac59adb8c6328ee1
2006-12-12 17:35:46 -05:00
Steve Reinhardt
6c8c86f2f9 Rename the StaticInst-based (read|set)(Int|Float)Reg methods to (read|set)(Int|Float)RegOperand to distinguish from non-StaticInst version.
--HG--
extra : convert_revision : b33ce0ebe2fee86cc791c00a35d8c6e395e1380c
2006-12-12 09:58:40 -08:00
Steve Reinhardt
cdc3e5bc22 Get rid of unused lock code.
--HG--
extra : convert_revision : a8030132268662ca54f487b8d32d09ba224317a8
2006-12-12 02:21:03 -05:00