sim: tag-based checkpoint versioning

This commit addresses gem5 checkpoints' linear versioning bottleneck.
Since development is distributed across many private trees, there exists
a sort of 'race' for checkpoint version numbers: internally a checkpoint
version may be used but then resynchronizing with the external tree causes
a conflict on that version.  This change replaces the linear version number
with a set of unique strings called tags.  Now the only conflicts that can
arise are of tag names, where collisions are much easier to avoid.

The checkpoint upgrader (util/cpt_upgrader.py) upgrades the version
representation, as one would expect. Each tag version implements its
upgrader code in a python file in the util/cpt_upgraders directory
rather than adding a function to the upgrader script itself.

The version tags are stored in the 'Globals' section rather than 'root'
(as the version was previously) because 'Globals' gets unserialized
first and can provide a warning before any other unserialization errors
can occur.

1 files changed, 135 insertions, 636 deletions

diff --git a/util/cpt_upgrader.py b/util/cpt_upgrader.py
index a0d1b94cb..ffd5d701b 100755
--- a/util/cpt_upgrader.py
+++ b/util/cpt_upgrader.py
@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-# Copyright (c) 2012-2013 ARM Limited
+# Copyright (c) 2012-2013,2015 ARM Limited
 # All rights reserved
 #
 # The license below extends only to copyright in the software and shall
@@ -36,6 +36,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #
 # Authors: Ali Saidi
+#          Curtis Dunham
 #
 
 # This python code is used to migrate checkpoints that were created in one
@@ -45,624 +46,22 @@
 # both time consuming and error-prone.
 
 # This script provides a way to migrate checkpoints to the newer repository in
-# a programatic way. It can be imported into another script or used on the
+# a programmatic way. It can be imported into another script or used on the
 # command line. From the command line the script will either migrate every
 # checkpoint it finds recursively (-r option) or a single checkpoint. When a
-# change is made to the gem5 repository that breaks previous checkpoints a
-# from_N() method should be implemented here and the gem5CheckpointVersion
-# variable in src/sim/serialize.hh should be incremented. For each version
-# between the checkpoints current version and the new version the from_N()
-# method will be run, passing in a ConfigParser object which contains the open
-# file. As these operations can be isa specific the method can verify the isa
-# and use regexes to find the correct sections that need to be updated.
+# change is made to the gem5 repository that breaks previous checkpoints an
+# upgrade() method should be implemented in its own .py file and placed in
+# src/util/cpt_upgraders/.  For each upgrader whose tag is not present in
+# the checkpoint tag list, the upgrade() method will be run, passing in a
+# ConfigParser object which contains the open file. As these operations can
+# be isa specific the method can verify the isa and use regexes to find the
+# correct sections that need to be updated.
 
 
 import ConfigParser
-import sys, os
+import glob, types, sys, os
 import os.path as osp
 
-# An example of a translator
-def from_0(cpt):
-    if cpt.get('root','isa') == 'arm':
-        for sec in cpt.sections():
-            import re
-            # Search for all the execution contexts
-            if re.search('.*sys.*\.cpu.*\.x.\..*', sec):
-                # Update each one
-                mr = cpt.get(sec, 'miscRegs').split()
-                #mr.insert(21,0)
-                #mr.insert(26,0)
-                cpt.set(sec, 'miscRegs', ' '.join(str(x) for x in mr))
-
-# The backing store supporting the memories in the system has changed
-# in that it is now stored globally per address range. As a result the
-# actual storage is separate from the memory controllers themselves.
-def from_1(cpt):
-    for sec in cpt.sections():
-        import re
-        # Search for a physical memory
-        if re.search('.*sys.*\.physmem$', sec):
-            # Add the number of stores attribute to the global physmem
-            cpt.set(sec, 'nbr_of_stores', '1')
-
-            # Get the filename and size as this is moving to the
-            # specific backing store
-            mem_filename = cpt.get(sec, 'filename')
-            mem_size = cpt.get(sec, '_size')
-            cpt.remove_option(sec, 'filename')
-            cpt.remove_option(sec, '_size')
-
-            # Get the name so that we can create the new section
-            system_name = str(sec).split('.')[0]
-            section_name = system_name + '.physmem.store0'
-            cpt.add_section(section_name)
-            cpt.set(section_name, 'store_id', '0')
-            cpt.set(section_name, 'range_size', mem_size)
-            cpt.set(section_name, 'filename', mem_filename)
-        elif re.search('.*sys.*\.\w*mem$', sec):
-            # Due to the lack of information about a start address,
-            # this migration only works if there is a single memory in
-            # the system, thus starting at 0
-            raise ValueError("more than one memory detected (" + sec + ")")
-
-def from_2(cpt):
-    for sec in cpt.sections():
-        import re
-        # Search for a CPUs
-        if re.search('.*sys.*cpu', sec):
-            try:
-                junk = cpt.get(sec, 'instCnt')
-                cpt.set(sec, '_pid', '0')
-            except ConfigParser.NoOptionError:
-                pass
-
-# The ISA is now a separate SimObject, which means that we serialize
-# it in a separate section instead of as a part of the ThreadContext.
-def from_3(cpt):
-    isa = cpt.get('root','isa')
-    isa_fields = {
-        "alpha" : ( "fpcr", "uniq", "lock_flag", "lock_addr", "ipr" ),
-        "arm" : ( "miscRegs" ),
-        "sparc" : ( "asi", "tick", "fprs", "gsr", "softint", "tick_cmpr",
-                    "stick", "stick_cmpr", "tpc", "tnpc", "tstate", "tt",
-                    "tba", "pstate", "tl", "pil", "cwp", "gl", "hpstate",
-                    "htstate", "hintp", "htba", "hstick_cmpr",
-                    "strandStatusReg", "fsr", "priContext", "secContext",
-                    "partId", "lsuCtrlReg", "scratchPad",
-                    "cpu_mondo_head", "cpu_mondo_tail",
-                    "dev_mondo_head", "dev_mondo_tail",
-                    "res_error_head", "res_error_tail",
-                    "nres_error_head", "nres_error_tail",
-                    "tick_intr_sched",
-                    "cpu", "tc_num", "tick_cmp", "stick_cmp", "hstick_cmp"),
-        "x86" : ( "regVal" ),
-        }
-
-    isa_fields = isa_fields.get(isa, [])
-    isa_sections = []
-    for sec in cpt.sections():
-        import re
-
-        re_cpu_match = re.match('^(.*sys.*\.cpu[^.]*)\.xc\.(.+)$', sec)
-        # Search for all the execution contexts
-        if not re_cpu_match:
-            continue
-
-        if re_cpu_match.group(2) != "0":
-            # This shouldn't happen as we didn't support checkpointing
-            # of in-order and O3 CPUs.
-            raise ValueError("Don't know how to migrate multi-threaded CPUs "
-                             "from version 1")
-
-        isa_section = []
-        for fspec in isa_fields:
-            for (key, value) in cpt.items(sec, raw=True):
-                if key in isa_fields:
-                    isa_section.append((key, value))
-
-        name = "%s.isa" % re_cpu_match.group(1)
-        isa_sections.append((name, isa_section))
-
-        for (key, value) in isa_section:
-            cpt.remove_option(sec, key)
-
-    for (sec, options) in isa_sections:
-        # Some intermediate versions of gem5 have empty ISA sections
-        # (after we made the ISA a SimObject, but before we started to
-        # serialize into a separate ISA section).
-        if not cpt.has_section(sec):
-            cpt.add_section(sec)
-        else:
-            if cpt.items(sec):
-                raise ValueError("Unexpected populated ISA section in old "
-                                 "checkpoint")
-
-        for (key, value) in options:
-            cpt.set(sec, key, value)
-
-# Version 5 of the checkpoint format removes the MISCREG_CPSR_MODE
-# register from the ARM register file.
-def from_4(cpt):
-    if cpt.get('root','isa') == 'arm':
-        for sec in cpt.sections():
-            import re
-            # Search for all ISA sections
-            if re.search('.*sys.*\.cpu.*\.isa', sec):
-                mr = cpt.get(sec, 'miscRegs').split()
-                # Remove MISCREG_CPSR_MODE
-                del mr[137]
-                cpt.set(sec, 'miscRegs', ' '.join(str(x) for x in mr))
-
-# Version 6 of the checkpoint format adds tlb to x86 checkpoints
-def from_5(cpt):
-    if cpt.get('root','isa') == 'x86':
-        for sec in cpt.sections():
-            import re
-            # Search for all ISA sections
-            if re.search('.*sys.*\.cpu.*\.dtb$', sec):
-                cpt.set(sec, '_size', '0')
-                cpt.set(sec, 'lruSeq', '0')
-
-            if re.search('.*sys.*\.cpu.*\.itb$', sec):
-                cpt.set(sec, '_size', '0')
-                cpt.set(sec, 'lruSeq', '0')
-    else:
-        print "ISA is not x86"
-
-# Version 7 of the checkpoint adds support for the IDE dmaAbort flag
-def from_6(cpt):
-    # Update IDE disk devices with dmaAborted
-    for sec in cpt.sections():
-        # curSector only exists in IDE devices, so key on that attribute
-        if cpt.has_option(sec, "curSector"):
-            cpt.set(sec, "dmaAborted", "false")
-
-# Version 8 of the checkpoint adds an ARM MISCREG
-def from_7(cpt):
-    if cpt.get('root','isa') == 'arm':
-        for sec in cpt.sections():
-            import re
-            # Search for all ISA sections
-            if re.search('.*sys.*\.cpu.*\.isa', sec):
-                mr = cpt.get(sec, 'miscRegs').split()
-                if len(mr) == 161:
-                    print "MISCREG_TEEHBR already seems to be inserted."
-                else:
-                    # Add dummy value for MISCREG_TEEHBR
-                    mr.insert(51,0);
-                    cpt.set(sec, 'miscRegs', ' '.join(str(x) for x in mr))
-
-# Version 9 of the checkpoint adds an all ARMv8 state
-def from_8(cpt):
-    if cpt.get('root','isa') != 'arm':
-        return
-    import re
-    print "Warning: The size of the FP register file has changed. "\
-          "To get similar results you need to adjust the number of "\
-          "physical registers in the CPU you're restoring into by "\
-          "NNNN."
-    # Find the CPU context's and upgrade their registers
-    for sec in cpt.sections():
-        re_xc_match = re.match('^.*?sys.*?\.cpu(\d+)*\.xc\.*', sec)
-        if not re_xc_match:
-            continue
-
-        # Update floating point regs
-        fpr = cpt.get(sec, 'floatRegs.i').split()
-        # v8 has 128 normal fp and 32 special fp regs compared
-        # to v7's 64 normal fp and 8 special fp regs.
-        # Insert the extra normal fp registers at end of v7 normal fp regs
-        for x in xrange(64):
-            fpr.insert(64, "0")
-        # Append the extra special registers
-        for x in xrange(24):
-            fpr.append("0")
-        cpt.set(sec, 'floatRegs.i', ' '.join(str(x) for x in fpr))
-
-        ir = cpt.get(sec, 'intRegs').split()
-        # Add in v8 int reg state
-        # Splice in R13_HYP
-        ir.insert(20, "0")
-        # Splice in INTREG_DUMMY and SP0 - SP3
-        ir.extend(["0", "0", "0", "0", "0"])
-        cpt.set(sec, 'intRegs', ' '.join(str(x) for x in ir))
-
-    # Update the cpu interrupt field
-    for sec in cpt.sections():
-        re_int_match = re.match("^.*?sys.*?\.cpu(\d+)*$", sec)
-        if not re_int_match:
-            continue
-
-        irqs = cpt.get(sec, "interrupts").split()
-        irqs.append("false")
-        irqs.append("false")
-        cpt.set(sec, "interrupts", ' '.join(str(x) for x in irqs))
-
-    # Update the per cpu interrupt structure
-    for sec in cpt.sections():
-        re_int_match = re.match("^.*?sys.*?\.cpu(\d+)*\.interrupts$", sec)
-        if not re_int_match:
-            continue
-
-        irqs = cpt.get(sec, "interrupts").split()
-        irqs.append("false")
-        irqs.append("false")
-        cpt.set(sec, "interrupts", ' '.join(str(x) for x in irqs))
-
-    # Update the misc regs and add in new isa specific fields
-    for sec in cpt.sections():
-        re_isa_match = re.match("^.*?sys.*?\.cpu(\d+)*\.isa$", sec)
-        if not re_isa_match:
-            continue
-
-        cpt.set(sec, 'haveSecurity', 'false')
-        cpt.set(sec, 'haveLPAE', 'false')
-        cpt.set(sec, 'haveVirtualization', 'false')
-        cpt.set(sec, 'haveLargeAsid64', 'false')
-        cpt.set(sec, 'physAddrRange64', '40')
-
-        # splice in the new misc registers, ~200 -> 605 registers,
-        # ordering does not remain consistent
-        mr_old = cpt.get(sec, 'miscRegs').split()
-        mr_new = [ '0' for x in xrange(605) ]
-
-        # map old v7 miscRegs to new v8 miscRegs
-        mr_new[0] = mr_old[0] # CPSR
-        mr_new[16] = mr_old[1] # CPSR_Q
-        mr_new[1] = mr_old[2] # SPSR
-        mr_new[2] = mr_old[3] # SPSR_FIQ
-        mr_new[3] = mr_old[4] # SPSR_IRQ
-        mr_new[4] = mr_old[5] # SPSR_SVC
-        mr_new[5] = mr_old[6] # SPSR_MON
-        mr_new[8] = mr_old[7] # SPSR_UND
-        mr_new[6] = mr_old[8] # SPSR_ABT
-        mr_new[432] = mr_old[9] # FPSR
-        mr_new[10] = mr_old[10] # FPSID
-        mr_new[11] = mr_old[11] # FPSCR
-        mr_new[18] = mr_old[12] # FPSCR_QC
-        mr_new[17] = mr_old[13] # FPSCR_EXC
-        mr_new[14] = mr_old[14] # FPEXC
-        mr_new[13] = mr_old[15] # MVFR0
-        mr_new[12] = mr_old[16] # MVFR1
-        mr_new[28] = mr_old[17] # SCTLR_RST,
-        mr_new[29] = mr_old[18] # SEV_MAILBOX,
-        mr_new[30] = mr_old[19] # DBGDIDR
-        mr_new[31] = mr_old[20] # DBGDSCR_INT,
-        mr_new[33] = mr_old[21] # DBGDTRRX_INT,
-        mr_new[34] = mr_old[22] # DBGTRTX_INT,
-        mr_new[35] = mr_old[23] # DBGWFAR,
-        mr_new[36] = mr_old[24] # DBGVCR,
-        #mr_new[] = mr_old[25] # DBGECR -> UNUSED,
-        #mr_new[] = mr_old[26] # DBGDSCCR -> UNUSED,
-        #mr_new[] = mr_old[27] # DBGSMCR -> UNUSED,
-        mr_new[37] = mr_old[28] # DBGDTRRX_EXT,
-        mr_new[38] = mr_old[29] # DBGDSCR_EXT,
-        mr_new[39] = mr_old[30] # DBGDTRTX_EXT,
-        #mr_new[] = mr_old[31] # DBGDRCR -> UNUSED,
-        mr_new[41] = mr_old[32] # DBGBVR,
-        mr_new[47] = mr_old[33] # DBGBCR,
-        #mr_new[] = mr_old[34] # DBGBVR_M -> UNUSED,
-        #mr_new[] = mr_old[35] # DBGBCR_M -> UNUSED,
-        mr_new[61] = mr_old[36] # DBGDRAR,
-        #mr_new[] = mr_old[37] # DBGBXVR_M -> UNUSED,
-        mr_new[64] = mr_old[38] # DBGOSLAR,
-        #mr_new[] = mr_old[39] # DBGOSSRR -> UNUSED,
-        mr_new[66] = mr_old[40] # DBGOSDLR,
-        mr_new[67] = mr_old[41] # DBGPRCR,
-        #mr_new[] = mr_old[42] # DBGPRSR -> UNUSED,
-        mr_new[68] = mr_old[43] # DBGDSAR,
-        #mr_new[] = mr_old[44] # DBGITCTRL -> UNUSED,
-        mr_new[69] = mr_old[45] # DBGCLAIMSET,
-        mr_new[70] = mr_old[46] # DBGCLAIMCLR,
-        mr_new[71] = mr_old[47] # DBGAUTHSTATUS,
-        mr_new[72] = mr_old[48] # DBGDEVID2,
-        mr_new[73] = mr_old[49] # DBGDEVID1,
-        mr_new[74] = mr_old[50] # DBGDEVID,
-        mr_new[77] = mr_old[51] # TEEHBR,
-        mr_new[109] = mr_old[52] # v7 SCTLR -> aarc32 SCTLR_NS
-        mr_new[189] = mr_old[53] # DCCISW,
-        mr_new[188] = mr_old[54] # DCCIMVAC,
-        mr_new[183] = mr_old[55] # DCCMVAC,
-        mr_new[271] = mr_old[56] # v7 CONTEXTIDR -> aarch32 CONTEXTIDR_NS,
-        mr_new[274] = mr_old[57] # v7 TPIDRURW -> aarch32 TPIDRURW_NS,
-        mr_new[277] = mr_old[58] # v7 TPIDRURO -> aarch32 TPIDRURO_NS,
-        mr_new[280] = mr_old[59] # v7 TPIDRPRW -> aarch32 TPIDRPRW_NS,
-        mr_new[170] = mr_old[60] # CP15ISB,
-        mr_new[185] = mr_old[61] # CP15DSB,
-        mr_new[186] = mr_old[62] # CP15DMB,
-        mr_new[114] = mr_old[63] # CPACR,
-        mr_new[101] = mr_old[64] # CLIDR,
-        mr_new[100] = mr_old[65] # CCSIDR,
-        mr_new[104] = mr_old[66] # v7 CSSELR -> aarch32 CSSELR_NS,
-        mr_new[163] = mr_old[67] # ICIALLUIS,
-        mr_new[168] = mr_old[68] # ICIALLU,
-        mr_new[169] = mr_old[69] # ICIMVAU,
-        mr_new[172] = mr_old[70] # BPIMVA,
-        mr_new[164] = mr_old[71] # BPIALLIS,
-        mr_new[171] = mr_old[72] # BPIALL,
-        mr_new[80] = mr_old[73] # MIDR,
-        mr_new[126] = mr_old[74] # v7 TTBR0 -> aarch32 TTBR0_NS,
-        mr_new[129] = mr_old[75] # v7 TTBR1 -> aarch32 TTBR1_NS,
-        mr_new[83] = mr_old[76] # TLBTR,
-        mr_new[137] = mr_old[77] # v7 DACR -> aarch32 DACR_NS,
-        mr_new[192] = mr_old[78] # TLBIALLIS,
-        mr_new[193] = mr_old[79] # TLBIMVAIS,
-        mr_new[194] = mr_old[80] # TLBIASIDIS,
-        mr_new[195] = mr_old[81] # TLBIMVAAIS,
-        mr_new[198] = mr_old[82] # ITLBIALL,
-        mr_new[199] = mr_old[83] # ITLBIMVA,
-        mr_new[200] = mr_old[84] # ITLBIASID,
-        mr_new[201] = mr_old[85] # DTLBIALL,
-        mr_new[202] = mr_old[86] # DTLBIMVA,
-        mr_new[203] = mr_old[87] # DTLBIASID,
-        mr_new[204] = mr_old[88] # TLBIALL,
-        mr_new[205] = mr_old[89] # TLBIMVA,
-        mr_new[206] = mr_old[90] # TLBIASID,
-        mr_new[207] = mr_old[91] # TLBIMVAA,
-        mr_new[140] = mr_old[92] # v7 DFSR -> aarch32 DFSR_NS,
-        mr_new[143] = mr_old[93] # v7 IFSR -> aarch32 IFSR_NS,
-        mr_new[155] = mr_old[94] # v7 DFAR -> aarch32 DFAR_NS,
-        mr_new[158] = mr_old[95] # v7 IFAR -> aarch32 IFAR_NS,
-        mr_new[84] = mr_old[96] # MPIDR,
-        mr_new[241] = mr_old[97] # v7 PRRR -> aarch32 PRRR_NS,
-        mr_new[247] = mr_old[98] # v7 NMRR -> aarch32 NMRR_NS,
-        mr_new[131] = mr_old[99] # TTBCR,
-        mr_new[86] = mr_old[100] # ID_PFR0,
-        mr_new[81] = mr_old[101] # CTR,
-        mr_new[115] = mr_old[102] # SCR,
-        # Set the non-secure bit
-        scr = int(mr_new[115])
-        scr = scr | 0x1
-        mr_new[115] = str(scr)
-        ###
-        mr_new[116] = mr_old[103] # SDER,
-        mr_new[165] = mr_old[104] # PAR,
-        mr_new[175] = mr_old[105] # V2PCWPR -> ATS1CPR,
-        mr_new[176] = mr_old[106] # V2PCWPW -> ATS1CPW,
-        mr_new[177] = mr_old[107] # V2PCWUR -> ATS1CUR,
-        mr_new[178] = mr_old[108] # V2PCWUW -> ATS1CUW,
-        mr_new[179] = mr_old[109] # V2POWPR -> ATS12NSOPR,
-        mr_new[180] = mr_old[110] # V2POWPW -> ATS12NSOPW,
-        mr_new[181] = mr_old[111] # V2POWUR -> ATS12NSOUR,
-        mr_new[182] = mr_old[112] # V2POWUW -> ATS12NWOUW,
-        mr_new[90] = mr_old[113] # ID_MMFR0,
-        mr_new[92] = mr_old[114] # ID_MMFR2,
-        mr_new[93] = mr_old[115] # ID_MMFR3,
-        mr_new[112] = mr_old[116] # v7 ACTLR -> aarch32 ACTLR_NS
-        mr_new[222] = mr_old[117] # PMCR,
-        mr_new[230] = mr_old[118] # PMCCNTR,
-        mr_new[223] = mr_old[119] # PMCNTENSET,
-        mr_new[224] = mr_old[120] # PMCNTENCLR,
-        mr_new[225] = mr_old[121] # PMOVSR,
-        mr_new[226] = mr_old[122] # PMSWINC,
-        mr_new[227] = mr_old[123] # PMSELR,
-        mr_new[228] = mr_old[124] # PMCEID0,
-        mr_new[229] = mr_old[125] # PMCEID1,
-        mr_new[231] = mr_old[126] # PMXEVTYPER,
-        mr_new[233] = mr_old[127] # PMXEVCNTR,
-        mr_new[234] = mr_old[128] # PMUSERENR,
-        mr_new[235] = mr_old[129] # PMINTENSET,
-        mr_new[236] = mr_old[130] # PMINTENCLR,
-        mr_new[94] = mr_old[131] # ID_ISAR0,
-        mr_new[95] = mr_old[132] # ID_ISAR1,
-        mr_new[96] = mr_old[133] # ID_ISAR2,
-        mr_new[97] = mr_old[134] # ID_ISAR3,
-        mr_new[98] = mr_old[135] # ID_ISAR4,
-        mr_new[99] = mr_old[136] # ID_ISAR5,
-        mr_new[20] = mr_old[137] # LOCKFLAG,
-        mr_new[19] = mr_old[138] # LOCKADDR,
-        mr_new[87] = mr_old[139] # ID_PFR1,
-        # Set up the processor features register
-        pfr = int(mr_new[87])
-        pfr = pfr | 0x1011
-        mr_new[87] = str(pfr)
-        ###
-        mr_new[238] = mr_old[140] # L2CTLR,
-        mr_new[82] = mr_old[141] # TCMTR
-        mr_new[88] = mr_old[142] # ID_DFR0,
-        mr_new[89] = mr_old[143] # ID_AFR0,
-        mr_new[91] = mr_old[144] # ID_MMFR1,
-        mr_new[102] = mr_old[145] # AIDR,
-        mr_new[146] = mr_old[146] # v7 ADFSR -> aarch32 ADFSR_NS,
-        mr_new[148] = mr_old[147] # AIFSR,
-        mr_new[173] = mr_old[148] # DCIMVAC,
-        mr_new[174] = mr_old[149] # DCISW,
-        mr_new[184] = mr_old[150] # MCCSW -> DCCSW,
-        mr_new[187] = mr_old[151] # DCCMVAU,
-        mr_new[117] = mr_old[152] # NSACR,
-        mr_new[262] = mr_old[153] # VBAR,
-        mr_new[265] = mr_old[154] # MVBAR,
-        mr_new[267] = mr_old[155] # ISR,
-        mr_new[269] = mr_old[156] # FCEIDR -> FCSEIDR,
-        #mr_new[] = mr_old[157] # L2LATENCY -> UNUSED,
-        #mr_new[] = mr_old[158] # CRN15 -> UNUSED,
-        mr_new[599] = mr_old[159] # NOP
-        mr_new[600] = mr_old[160] # RAZ,
-
-        # Set the new miscRegs structure
-        cpt.set(sec, 'miscRegs', ' '.join(str(x) for x in mr_new))
-
-    cpu_prefix = {}
-    # Add in state for ITB/DTB
-    for sec in cpt.sections():
-        re_tlb_match = re.match('(^.*?sys.*?\.cpu(\d+)*)\.(dtb|itb)$', sec)
-        if not re_tlb_match:
-            continue
-
-        cpu_prefix[re_tlb_match.group(1)] = True # Save off prefix to add
-        # Set the non-secure bit (bit 9) to 1 for attributes
-        attr = int(cpt.get(sec, '_attr'))
-        attr = attr | 0x200
-        cpt.set(sec, '_attr', str(attr))
-        cpt.set(sec, 'haveLPAE', 'false')
-        cpt.set(sec, 'directToStage2', 'false')
-        cpt.set(sec, 'stage2Req', 'false')
-        cpt.set(sec, 'bootUncacheability', 'true')
-
-    # Add in extra state for the new TLB Entries
-    for sec in cpt.sections():
-        re_tlbentry_match = re.match('(^.*?sys.*?\.cpu(\d+)*)\.(dtb|itb).TlbEntry\d+$', sec)
-        if not re_tlbentry_match:
-            continue
-
-        # Add in the new entries
-        cpt.set(sec, 'longDescFormat', 'false')
-        cpt.set(sec, 'vmid', '0')
-        cpt.set(sec, 'isHyp', 'false')
-        valid = cpt.get(sec, 'valid')
-        if valid == 'true':
-            cpt.set(sec, 'ns', 'true')
-            cpt.set(sec, 'nstid', 'true')
-            cpt.set(sec, 'pxn', 'true')
-            cpt.set(sec, 'hap', '3')
-            # All v7 code used 2 level page tables
-            cpt.set(sec, 'lookupLevel', '2')
-            attr = int(cpt.get(sec, 'attributes'))
-            # set the non-secure bit (bit 9) to 1
-            # as no previous v7 code used secure code
-            attr = attr | 0x200
-            cpt.set(sec, 'attributes', str(attr))
-        else:
-            cpt.set(sec, 'ns', 'false')
-            cpt.set(sec, 'nstid', 'false')
-            cpt.set(sec, 'pxn', 'false')
-            cpt.set(sec, 'hap', '0')
-            cpt.set(sec, 'lookupLevel', '0')
-        cpt.set(sec, 'outerShareable', 'false')
-
-    # Add d/istage2_mmu and d/istage2_mmu.stage2_tlb
-    for key in cpu_prefix:
-        for suffix in ['.istage2_mmu', '.dstage2_mmu']:
-            new_sec = key + suffix
-            cpt.add_section(new_sec)
-            new_sec = key + suffix + ".stage2_tlb"
-            cpt.add_section(new_sec)
-            # Fill in tlb info with some defaults
-            cpt.set(new_sec, '_attr', '0')
-            cpt.set(new_sec, 'haveLPAE', 'false')
-            cpt.set(new_sec, 'directToStage2', 'false')
-            cpt.set(new_sec, 'stage2Req', 'false')
-            cpt.set(new_sec, 'bootUncacheability', 'false')
-            cpt.set(new_sec, 'num_entries', '0')
-
-# Version 10 adds block_size_bytes to system.ruby
-def from_9(cpt):
-    for sec in cpt.sections():
-        if sec == 'system.ruby':
-            # Use Gem5's default of 64; this should be changed if the to be
-            # upgraded checkpoints were not taken with block-size 64!
-            cpt.set(sec, 'block_size_bytes', '64')
-
-# Checkpoint version 11 (0xB) adds the perfLevel variable in the clock domain
-# and voltage domain simObjects used for DVFS and is serialized and
-# unserialized.
-def from_A(cpt):
-    for sec in cpt.sections():
-        import re
-
-        if re.match('^.*sys.*[._]clk_domain$', sec):
-            # Make _perfLevel equal to 0 which means best performance
-            cpt.set(sec, '_perfLevel', ' '.join('0'))
-        elif re.match('^.*sys.*[._]voltage_domain$', sec):
-            # Make _perfLevel equal to 0 which means best performance
-            cpt.set(sec, '_perfLevel', ' '.join('0'))
-        else:
-            continue
-
-# The change between versions C and D is the addition of support for multiple
-# event queues, so for old checkpoints we must specify that there's only one.
-def from_B(cpt):
-    cpt.set('Globals', 'numMainEventQueues', '1')
-
-# Checkpoint version D uses condition code registers for the ARM
-# architecture; previously the integer register file was used for these
-# registers. To upgrade, we move those 5 integer registers to the ccRegs
-# register file.
-def from_C(cpt):
-    if cpt.get('root','isa') == 'arm':
-        for sec in cpt.sections():
-            import re
-
-            re_cpu_match = re.match('^(.*sys.*\.cpu[^.]*)\.xc\.(.+)$', sec)
-            # Search for all the execution contexts
-            if not re_cpu_match:
-                continue
-
-            items = []
-            for (item,value) in cpt.items(sec):
-                items.append(item)
-            if 'ccRegs' not in items:
-                intRegs = cpt.get(sec, 'intRegs').split()
-
-                ccRegs = intRegs[38:43]
-                del      intRegs[38:43]
-
-                ccRegs.append('0') # CCREG_ZERO
-
-                cpt.set(sec, 'intRegs', ' '.join(intRegs))
-                cpt.set(sec, 'ccRegs',  ' '.join(ccRegs))
-
-# Checkpoint version E adds the ARM CONTEXTIDR_EL2 miscreg.
-def from_D(cpt):
-    if cpt.get('root','isa') == 'arm':
-        for sec in cpt.sections():
-            import re
-            # Search for all ISA sections
-            if re.search('.*sys.*\.cpu.*\.isa$', sec):
-                miscRegs = cpt.get(sec, 'miscRegs').split()
-                # CONTEXTIDR_EL2 defaults to 0b11111100000000000001
-                miscRegs[599:599] = [0xFC001]
-                cpt.set(sec, 'miscRegs', ' '.join(str(x) for x in miscRegs))
-
-# Checkpoint version F renames an internal member of Process class.
-def from_E(cpt):
-    import re
-    for sec in cpt.sections():
-        fdm = 'FdMap'
-        fde = 'FDEntry'
-        if re.match('.*\.%s.*' % fdm, sec):
-            rename = re.sub(fdm, fde, sec)
-            split = re.split(fde, rename)
-
-            # rename the section and add the 'mode' field
-            rename_section(cpt, sec, rename)
-            cpt.set(rename, 'mode', "0") # no proper value to set :(
-
-            # add in entries 257 to 1023
-            if split[1] == "0":
-                for x in range(257, 1024):
-                    seq = (split[0], fde, "%s" % x)
-                    section = "".join(seq)
-                    cpt.add_section(section)
-                    cpt.set(section, 'fd', '-1')
-
-
-migrations = []
-migrations.append(from_0)
-migrations.append(from_1)
-migrations.append(from_2)
-migrations.append(from_3)
-migrations.append(from_4)
-migrations.append(from_5)
-migrations.append(from_6)
-migrations.append(from_7)
-migrations.append(from_8)
-migrations.append(from_9)
-migrations.append(from_A)
-migrations.append(from_B)
-migrations.append(from_C)
-migrations.append(from_D)
-migrations.append(from_E)
-
-# http://stackoverflow.com/questions/15069127/python-configparser-module-\
-# rename-a-section
-def rename_section(cp, section_from, section_to):
-    items = cp.items(section_from)
-    cp.add_section(section_to)
-    for item in items:
-        cp.set(section_to, item[0], item[1])
-    cp.remove_section(section_from)
-
 verbose_print = False
 
 def verboseprint(*args):
@@ -672,6 +71,62 @@ def verboseprint(*args):
         print arg,
     print
 
+class Upgrader:
+    tag_set = set()
+    by_tag = {}
+    legacy = {}
+    def __init__(self, filename):
+        self.filename = filename
+        execfile(filename, {}, self.__dict__)
+
+        if not hasattr(self, 'tag'):
+            self.tag = osp.basename(filename)[:-3]
+        if not hasattr(self, 'depends'):
+            self.depends = []
+        elif isinstance(self.depends, str):
+            self.depends = [self.depends]
+
+        if not hasattr(self, 'upgrader'):
+            print "Error: no upgrader method for", self.tag
+            sys.exit(1)
+        elif not isinstance(self.upgrader, types.FunctionType):
+            print "Error: 'upgrader' for %s is %s, not function", \
+                self.tag, type(self)
+            sys.exit(1)
+
+        if hasattr(self, 'legacy_version'):
+            Upgrader.legacy[self.legacy_version] = self
+
+        Upgrader.by_tag[self.tag] = self
+        Upgrader.tag_set.add(self.tag)
+
+    def ready(self, tags):
+        for dep in self.depends:
+            if dep not in tags:
+                return False
+        return True
+
+    def upgrade(self, cpt):
+        (self.upgrader)(cpt)
+        verboseprint("applied upgrade for", self.tag)
+
+    @staticmethod
+    def get(tag):
+        return Upgrader.by_tag[tag]
+
+    @staticmethod
+    def load_all():
+        util_dir = osp.dirname(osp.abspath(__file__))
+
+        for py in glob.glob(util_dir + '/cpt_upgraders/*.py'):
+            Upgrader(py)
+
+        # make linear dependences for legacy versions
+        i = 3
+        while i in Upgrader.legacy:
+            Upgrader.legacy[i].depends = [Upgrader.legacy[i-1].tag]
+            i = i + 1
+
 def process_file(path, **kwargs):
     if not osp.isfile(path):
         import errno
@@ -693,35 +148,64 @@ def process_file(path, **kwargs):
     cpt.readfp(cpt_file)
     cpt_file.close()
 
-    # Make sure we know what we're starting from
-    if not cpt.has_option('root','cpt_ver'):
-        raise LookupError("cannot determine version of checkpoint")
-
-    cpt_ver = cpt.getint('root','cpt_ver')
-
-    # If the current checkpoint is longer than the migrations list, we have a problem
-    # and someone didn't update this file
-    if cpt_ver > len(migrations):
-        raise ValueError("upgrade script is too old and needs updating")
-
-    verboseprint("\t...file is at version %#x" % cpt_ver)
+    change = False
 
-    if cpt_ver == len(migrations):
-        verboseprint("\t...nothing to do")
+    # Make sure we know what we're starting from
+    if cpt.has_option('root','cpt_ver'):
+        cpt_ver = cpt.getint('root','cpt_ver')
+
+        # Legacy linear checkpoint version
+        # convert to list of tags before proceeding
+        tags = set([])
+        for i in xrange(2, cpt_ver+1):
+            tags.add(Upgrader.legacy[i].tag)
+        verboseprint("performed legacy version -> tags conversion")
+        change = True
+
+        cpt.remove_option('root', 'cpt_ver')
+    elif cpt.has_option('Globals','version_tags'):
+        tags = set((''.join(cpt.get('Globals','version_tags'))).split())
+    else:
+        print "fatal: no version information in checkpoint"
+        exit(1)
+
+    verboseprint("has tags", ' '.join(tags))
+    # If the current checkpoint has a tag we don't know about, we have
+    # a divergence that (in general) must be addressed by (e.g.) merging
+    # simulator support for its changes.
+    unknown_tags = tags - Upgrader.tag_set
+    if unknown_tags:
+        print "warning: upgrade script does not recognize the following "\
+              "tags in this checkpoint:", ' '.join(unknown_tags)
+
+    # Apply migrations for tags not in checkpoint, respecting dependences
+    to_apply = Upgrader.tag_set - tags
+    while to_apply:
+        ready = set([ t for t in to_apply if Upgrader.get(t).ready(tags) ])
+        if not ready:
+            print "could not apply these upgrades:", ' '.join(to_apply)
+            print "upgrade dependences impossible to resolve; aborting"
+            exit(1)
+
+        for tag in ready:
+            Upgrader.get(tag).upgrade(cpt)
+            tags.add(tag)
+            change = True
+
+        to_apply -= ready
+
+    if not change:
+        verboseprint("...nothing to do")
         return
 
-    # Walk through every function from now until the end fixing the checkpoint
-    for v in xrange(cpt_ver,len(migrations)):
-        verboseprint("\t...migrating to version %#x" %  (v + 1))
-        migrations[v](cpt)
-        cpt.set('root','cpt_ver', str(v + 1))
+    cpt.set('Globals', 'version_tags', ' '.join(tags))
 
     # Write the old data back
-    verboseprint("\t...completed")
+    verboseprint("...completed")
     cpt.write(file(path, 'w'))
 
 if __name__ == '__main__':
-    from optparse import OptionParser
+    from optparse import OptionParser, SUPPRESS_HELP
     parser = OptionParser("usage: %prog [options] <filename or directory>")
     parser.add_option("-r", "--recurse", action="store_true",
                       help="Recurse through all subdirectories modifying "\
@@ -731,14 +215,29 @@ if __name__ == '__main__':
                       help="Do no backup each checkpoint before modifying it")
     parser.add_option("-v", "--verbose", action="store_true",
                       help="Print out debugging information as")
+    parser.add_option("--get-cc-file", action="store_true",
+                      # used during build; generate src/sim/tags.cc and exit
+                      help=SUPPRESS_HELP)
 
     (options, args) = parser.parse_args()
-    if len(args) != 1:
+    verbose_print = options.verbose
+
+    Upgrader.load_all()
+
+    if options.get_cc_file:
+        print "// this file is auto-generated by util/cpt_upgrader.py"
+        print "#include <string>"
+        print "#include <set>"
+        print
+        print "std::set<std::string> version_tags = {"
+        for tag in Upgrader.tag_set:
+            print "  \"%s\"," % tag
+        print "};"
+        exit(0)
+    elif len(args) != 1:
         parser.error("You must specify a checkpoint file to modify or a "\
                      "directory of checkpoints to recursively update")
 
-    verbose_print = options.verbose
-
     # Deal with shell variables and ~
     path = osp.expandvars(osp.expanduser(args[0]))