mklockrank.go

     1  // Copyright 2022 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  //go:build ignore
     6  
     7  // mklockrank records the static rank graph of the locks in the
     8  // runtime and generates the rank checking structures in lockrank.go.
     9  package main
    10  
    11  import (
    12  	"bytes"
    13  	"flag"
    14  	"fmt"
    15  	"go/format"
    16  	"internal/dag"
    17  	"io"
    18  	"log"
    19  	"os"
    20  	"strings"
    21  )
    22  
    23  // ranks describes the lock rank graph. See "go doc internal/dag" for
    24  // the syntax.
    25  //
    26  // "a < b" means a must be acquired before b if both are held
    27  // (or, if b is held, a cannot be acquired).
    28  //
    29  // "NONE < a" means no locks may be held when a is acquired.
    30  //
    31  // If a lock is not given a rank, then it is assumed to be a leaf
    32  // lock, which means no other lock can be acquired while it is held.
    33  // Therefore, leaf locks do not need to be given an explicit rank.
    34  //
    35  // Ranks in all caps are pseudo-nodes that help define order, but do
    36  // not actually define a rank.
    37  //
    38  // TODO: It's often hard to correlate rank names to locks. Change
    39  // these to be more consistent with the locks they label.
    40  const ranks = `
    41  # Sysmon
    42  NONE
    43  < sysmon
    44  < scavenge, forcegc;
    45  
    46  # Defer
    47  NONE < defer;
    48  
    49  # GC
    50  NONE <
    51    sweepWaiters,
    52    assistQueue,
    53    strongFromWeakQueue,
    54    sweep;
    55  
    56  # Test only
    57  NONE < testR, testW;
    58  
    59  NONE < timerSend;
    60  
    61  # Scheduler, timers, netpoll
    62  NONE < allocmW, execW, cpuprof, pollCache, pollDesc, wakeableSleep;
    63  scavenge, sweep, testR, wakeableSleep, timerSend < hchan;
    64  assistQueue,
    65    cpuprof,
    66    forcegc,
    67    hchan,
    68    pollDesc, # pollDesc can interact with timers, which can lock sched.
    69    scavenge,
    70    strongFromWeakQueue,
    71    sweep,
    72    sweepWaiters,
    73    testR,
    74    wakeableSleep
    75  # Above SCHED are things that can call into the scheduler.
    76  < SCHED
    77  # Below SCHED is the scheduler implementation.
    78  < allocmR,
    79    execR;
    80  allocmR, execR, hchan < sched;
    81  sched < allg, allp;
    82  
    83  # Channels
    84  NONE < notifyList;
    85  hchan, notifyList < sudog;
    86  
    87  hchan, pollDesc, wakeableSleep < timers;
    88  timers, timerSend < timer < netpollInit;
    89  
    90  # Semaphores
    91  NONE < root;
    92  
    93  # Itabs
    94  NONE
    95  < itab
    96  < reflectOffs;
    97  
    98  # User arena state
    99  NONE < userArenaState;
   100  
   101  # Tracing without a P uses a global trace buffer.
   102  scavenge
   103  # Above TRACEGLOBAL can emit a trace event without a P.
   104  < TRACEGLOBAL
   105  # Below TRACEGLOBAL manages the global tracing buffer.
   106  # Note that traceBuf eventually chains to MALLOC, but we never get that far
   107  # in the situation where there's no P.
   108  < traceBuf;
   109  # Starting/stopping tracing traces strings.
   110  traceBuf < traceStrings;
   111  
   112  # Malloc
   113  allg,
   114    allocmR,
   115    allp, # procresize
   116    execR, # May grow stack
   117    execW, # May allocate after BeforeFork
   118    hchan,
   119    notifyList,
   120    reflectOffs,
   121    timer,
   122    traceStrings,
   123    userArenaState
   124  # Above MALLOC are things that can allocate memory.
   125  < MALLOC
   126  # Below MALLOC is the malloc implementation.
   127  < fin,
   128    spanSetSpine,
   129    mspanSpecial,
   130    traceTypeTab,
   131    MPROF;
   132  
   133  # We can acquire gcBitsArenas for pinner bits, and
   134  # it's guarded by mspanSpecial.
   135  MALLOC, mspanSpecial < gcBitsArenas;
   136  
   137  # Memory profiling
   138  MPROF < profInsert, profBlock, profMemActive;
   139  profMemActive < profMemFuture;
   140  
   141  # Stack allocation and copying
   142  gcBitsArenas,
   143    netpollInit,
   144    profBlock,
   145    profInsert,
   146    profMemFuture,
   147    spanSetSpine,
   148    fin,
   149    root
   150  # Anything that can grow the stack can acquire STACKGROW.
   151  # (Most higher layers imply STACKGROW, like MALLOC.)
   152  < STACKGROW
   153  # Below STACKGROW is the stack allocator/copying implementation.
   154  < gscan;
   155  gscan < stackpool;
   156  gscan < stackLarge;
   157  # Generally, hchan must be acquired before gscan. But in one case,
   158  # where we suspend a G and then shrink its stack, syncadjustsudogs
   159  # can acquire hchan locks while holding gscan. To allow this case,
   160  # we use hchanLeaf instead of hchan.
   161  gscan < hchanLeaf;
   162  
   163  # Write barrier
   164  defer,
   165    gscan,
   166    mspanSpecial,
   167    pollCache,
   168    sudog,
   169    timer
   170  # Anything that can have write barriers can acquire WB.
   171  # Above WB, we can have write barriers.
   172  < WB
   173  # Below WB is the write barrier implementation.
   174  < wbufSpans;
   175  
   176  # Span allocator
   177  stackLarge,
   178    stackpool,
   179    wbufSpans
   180  # Above mheap is anything that can call the span allocator.
   181  < mheap;
   182  # Below mheap is the span allocator implementation.
   183  #
   184  # Specials: we're allowed to allocate a special while holding
   185  # an mspanSpecial lock, and they're part of the malloc implementation.
   186  # Pinner bits might be freed by the span allocator.
   187  mheap, mspanSpecial < mheapSpecial;
   188  mheap, mheapSpecial < globalAlloc;
   189  
   190  # Execution tracer events (with a P)
   191  hchan,
   192    mheap,
   193    root,
   194    sched,
   195    traceStrings,
   196    notifyList,
   197    fin
   198  # Above TRACE is anything that can create a trace event
   199  < TRACE
   200  < trace
   201  < traceStackTab;
   202  
   203  # panic is handled specially. It is implicitly below all other locks.
   204  NONE < panic;
   205  # deadlock is not acquired while holding panic, but it also needs to be
   206  # below all other locks.
   207  panic < deadlock;
   208  # raceFini is only held while exiting.
   209  panic < raceFini;
   210  
   211  # RWMutex internal read lock
   212  
   213  allocmR,
   214    allocmW
   215  < allocmRInternal;
   216  
   217  execR,
   218    execW
   219  < execRInternal;
   220  
   221  testR,
   222    testW
   223  < testRInternal;
   224  `
   225  
   226  // cyclicRanks lists lock ranks that allow multiple locks of the same
   227  // rank to be acquired simultaneously. The runtime enforces ordering
   228  // within these ranks using a separate mechanism.
   229  var cyclicRanks = map[string]bool{
   230  	// Multiple timers are locked simultaneously in destroy().
   231  	"timers": true,
   232  	// Multiple hchans are acquired in hchan.sortkey() order in
   233  	// select.
   234  	"hchan": true,
   235  	// Multiple hchanLeafs are acquired in hchan.sortkey() order in
   236  	// syncadjustsudogs().
   237  	"hchanLeaf": true,
   238  	// The point of the deadlock lock is to deadlock.
   239  	"deadlock": true,
   240  }
   241  
   242  func main() {
   243  	flagO := flag.String("o", "", "write to `file` instead of stdout")
   244  	flagDot := flag.Bool("dot", false, "emit graphviz output instead of Go")
   245  	flag.Parse()
   246  	if flag.NArg() != 0 {
   247  		fmt.Fprintf(os.Stderr, "too many arguments")
   248  		os.Exit(2)
   249  	}
   250  
   251  	g, err := dag.Parse(ranks)
   252  	if err != nil {
   253  		log.Fatal(err)
   254  	}
   255  
   256  	var out []byte
   257  	if *flagDot {
   258  		var b bytes.Buffer
   259  		g.TransitiveReduction()
   260  		// Add cyclic edges for visualization.
   261  		for k := range cyclicRanks {
   262  			g.AddEdge(k, k)
   263  		}
   264  		// Reverse the graph. It's much easier to read this as
   265  		// a "<" partial order than a ">" partial order. This
   266  		// ways, locks are acquired from the top going down
   267  		// and time moves forward over the edges instead of
   268  		// backward.
   269  		g.Transpose()
   270  		generateDot(&b, g)
   271  		out = b.Bytes()
   272  	} else {
   273  		var b bytes.Buffer
   274  		generateGo(&b, g)
   275  		out, err = format.Source(b.Bytes())
   276  		if err != nil {
   277  			log.Fatal(err)
   278  		}
   279  	}
   280  
   281  	if *flagO != "" {
   282  		err = os.WriteFile(*flagO, out, 0666)
   283  	} else {
   284  		_, err = os.Stdout.Write(out)
   285  	}
   286  	if err != nil {
   287  		log.Fatal(err)
   288  	}
   289  }
   290  
   291  func generateGo(w io.Writer, g *dag.Graph) {
   292  	fmt.Fprintf(w, `// Code generated by mklockrank.go; DO NOT EDIT.
   293  
   294  package runtime
   295  
   296  type lockRank int
   297  
   298  `)
   299  
   300  	// Create numeric ranks.
   301  	topo := g.Topo()
   302  	for i, j := 0, len(topo)-1; i < j; i, j = i+1, j-1 {
   303  		topo[i], topo[j] = topo[j], topo[i]
   304  	}
   305  	fmt.Fprintf(w, `
   306  // Constants representing the ranks of all non-leaf runtime locks, in rank order.
   307  // Locks with lower rank must be taken before locks with higher rank,
   308  // in addition to satisfying the partial order in lockPartialOrder.
   309  // A few ranks allow self-cycles, which are specified in lockPartialOrder.
   310  const (
   311  	lockRankUnknown lockRank = iota
   312  
   313  `)
   314  	for _, rank := range topo {
   315  		if isPseudo(rank) {
   316  			fmt.Fprintf(w, "\t// %s\n", rank)
   317  		} else {
   318  			fmt.Fprintf(w, "\t%s\n", cname(rank))
   319  		}
   320  	}
   321  	fmt.Fprintf(w, `)
   322  
   323  // lockRankLeafRank is the rank of lock that does not have a declared rank,
   324  // and hence is a leaf lock.
   325  const lockRankLeafRank lockRank = 1000
   326  `)
   327  
   328  	// Create string table.
   329  	fmt.Fprintf(w, `
   330  // lockNames gives the names associated with each of the above ranks.
   331  var lockNames = []string{
   332  `)
   333  	for _, rank := range topo {
   334  		if !isPseudo(rank) {
   335  			fmt.Fprintf(w, "\t%s: %q,\n", cname(rank), rank)
   336  		}
   337  	}
   338  	fmt.Fprintf(w, `}
   339  
   340  func (rank lockRank) String() string {
   341  	if rank == 0 {
   342  		return "UNKNOWN"
   343  	}
   344  	if rank == lockRankLeafRank {
   345  		return "LEAF"
   346  	}
   347  	if rank < 0 || int(rank) >= len(lockNames) {
   348  		return "BAD RANK"
   349  	}
   350  	return lockNames[rank]
   351  }
   352  `)
   353  
   354  	// Create partial order structure.
   355  	fmt.Fprintf(w, `
   356  // lockPartialOrder is the transitive closure of the lock rank graph.
   357  // An entry for rank X lists all of the ranks that can already be held
   358  // when rank X is acquired.
   359  //
   360  // Lock ranks that allow self-cycles list themselves.
   361  var lockPartialOrder [][]lockRank = [][]lockRank{
   362  `)
   363  	for _, rank := range topo {
   364  		if isPseudo(rank) {
   365  			continue
   366  		}
   367  		list := []string{}
   368  		for _, before := range g.Edges(rank) {
   369  			if !isPseudo(before) {
   370  				list = append(list, cname(before))
   371  			}
   372  		}
   373  		if cyclicRanks[rank] {
   374  			list = append(list, cname(rank))
   375  		}
   376  
   377  		fmt.Fprintf(w, "\t%s: {%s},\n", cname(rank), strings.Join(list, ", "))
   378  	}
   379  	fmt.Fprintf(w, "}\n")
   380  }
   381  
   382  // cname returns the Go const name for the given lock rank label.
   383  func cname(label string) string {
   384  	return "lockRank" + strings.ToUpper(label[:1]) + label[1:]
   385  }
   386  
   387  func isPseudo(label string) bool {
   388  	return strings.ToUpper(label) == label
   389  }
   390  
   391  // generateDot emits a Graphviz dot representation of g to w.
   392  func generateDot(w io.Writer, g *dag.Graph) {
   393  	fmt.Fprintf(w, "digraph g {\n")
   394  
   395  	// Define all nodes.
   396  	for _, node := range g.Nodes {
   397  		fmt.Fprintf(w, "%q;\n", node)
   398  	}
   399  
   400  	// Create edges.
   401  	for _, node := range g.Nodes {
   402  		for _, to := range g.Edges(node) {
   403  			fmt.Fprintf(w, "%q -> %q;\n", node, to)
   404  		}
   405  	}
   406  
   407  	fmt.Fprintf(w, "}\n")
   408  }
   409
View as plain text