2 % (c) The OBFUSCATION-THROUGH-GRATUITOUS-PREPROCESSOR-ABUSE Project,
3 % Glasgow University, 1990-1994
8 % o I think it would be worth making the connection with CPS explicit.
9 % Now that we have explicit activation records (on the stack), we can
10 % explain the whole system in terms of CPS and tail calls --- with the
11 % one requirement that we carefuly distinguish stack-allocated objects
12 % from heap-allocated objects.
15 % \documentstyle[preprint]{acmconf}
16 \documentclass[11pt]{article}
17 \oddsidemargin 0.1 in % Note that \oddsidemargin = \evensidemargin
18 \evensidemargin 0.1 in
19 \marginparwidth 0.85in % Narrow margins require narrower marginal notes
23 \newcommand{\note}[1]{{\em Note: #1}}
33 \setlength{\parskip}{0.15cm}
34 \setlength{\parsep}{0.15cm}
35 \setlength{\topsep}{0cm} % Reduces space before and after verbatim,
36 % which is implemented using trivlist
37 \setlength{\parindent}{0cm}
39 \renewcommand{\textfraction}{0.2}
40 \renewcommand{\floatpagefraction}{0.7}
44 \newcommand{\ToDo}[1]{{{\bf ToDo:}\sl #1}}
45 \newcommand{\Arg}[1]{\mbox{${\tt arg}_{#1}$}}
46 \newcommand{\bottom}{bottom} % foo, can't remember the symbol name
48 \title{The STG runtime system (revised)}
49 \author{Simon Peyton Jones \\ Glasgow University and Oregon Graduate Institute \and
50 Alastair Reid \\ Yale University}
57 \section{Introduction}
59 This document describes the GHC/Hugs run-time system. It serves as
60 a Glasgow/Yale/Nottingham ``contract'' about what the RTS does.
62 \subsection{New features compared to GHC 2.04}
65 \item The RTS supports mixed compiled/interpreted execution, so
66 that a program can consist of a mixture of GHC-compiled and Hugs-interpreted
69 \item CAFs are only retained if they are
70 reachable. Since they are referred to by implicit references buried
71 in code, this means that the garbage collector must traverse the whole
72 accessible code tree. This feature eliminates a whole class of painful
75 \item A running thread has only one stack, which contains a mixture
76 of pointers and non-pointers. Section~\ref{sect:stacks} describes how
77 we find out which is which. (GHC has used two stacks for some while.
78 Using one stack instead of two reduces register pressure, reduces the
79 size of update frames, and eliminates
80 ``stack-stubbing'' instructions.)
84 \subsection{Wish list}
86 Here's a list of things we'd like to support in the future.
88 \item Interrupts, speculative computation.
91 The SM could tune the size of the allocation arena, the number of
92 generations, etc taking into account residency, GC rate and page fault
96 There should be no need to specify the amnount of stack/heap space to
97 allocate when you started a program - let it just take as much or as
98 little as it wants. (It might be useful to be able to specify maximum
99 sizes and to be able to suggest an initial size.)
102 We could trigger a GC when all threads are blocked waiting for IO if
103 the allocation arena (or some of the generations) are nearly full.
107 \subsection{Configuration}
109 Some of the above features are expensive or less portable, so we
110 envision building a number of different configurations supporting
111 different subsets of the above features.
113 You can make the following choices:
116 Support for concurrency or parallelism. There are four
117 mutually-exclusive choices.
120 \item[@SEQUENTIAL@] No concurrency or parallelism support.
121 This configuration might not support interrupt recovery.
122 \item[@CONCURRENT@] Support for concurrency but not for parallelism.
123 \item[@CONCURRENT@+@GRANSIM@] Concurrency support and simulated parallelism.
124 \item[@CONCURRENT@+@PARALLEL@] Concurrency support and real parallelism.
127 \item @PROFILING@ adds cost-centre profiling.
129 \item @TICKY@ gathers internal statistics (often known as ``ticky-ticky'' code).
131 \item @DEBUG@ does internal consistency checks.
133 \item Persistence. (well, not yet).
136 Which garbage collector to use. At the moment we
137 only anticipate one, however.
140 \subsection{Terminology}
144 \item A {\em word} is (at least) 32 bits and can hold either a signed
147 \item A {\em pointer} is (at least) 32 bits and big enough to hold a
148 function pointer or a data pointer.
150 \item A {\em closure} is a (representation of) a value of a {\em pointed}
151 type. It may be in HNF or it may be unevaluated --- in either case, you can
152 try to evaluate it again.
154 \item A {\em thunk} is a (representation of) a value of a {\em pointed}
155 type which is {\em not} in HNF.
159 Occasionally, a field of a data structure must hold either a word or a
160 pointer. In such circumstances, it is {\em not safe} to assume that
161 words and pointers are the same size.
164 % More terminology to mention.
167 There are a few other system invariants which need to be mentioned ---
168 though not necessarily here:
172 \item The garbage collector never expands an object when it promotes
173 it to the old generation. This is important because the GC avoids
174 performing heap overflow checks by assuming that the amount added to
175 the old generation is no bigger than the current new generation.
180 \section{The Scheduler}
182 The Scheduler is the heart of the run-time system. A running program
183 consists of a single running thread, and a list of runnable and
184 blocked threads. The running thread returns to the scheduler when any
185 of the following conditions arises:
188 \item A heap check fails, and a garbage collection is required
189 \item Compiled code needs to switch to interpreted code, and vice
191 \item The thread becomes blocked.
192 \item The thread is preempted.
195 A running system has a global state, consisting of
198 \item @Hp@, the current heap pointer, which points to the next
199 available address in the Heap.
200 \item @HpLim@, the heap limit pointer, which points to the end of the
202 \item The Thread Preemption Flag, which is set whenever the currently
203 running thread should be preempted at the next opportunity.
206 Each thread has a thread-local state, which consists of
209 \item @TSO@, the Thread State Object for this thread. This is a heap
210 object that is used to store the current thread state when the thread
211 is blocked or sleeping.
212 \item @Sp@, the current stack pointer.
213 \item @Su@, the current stack update frame pointer. This register
214 points to the most recent update frame on the stack, and is used to
215 calculate the number of arguments available when entering a function.
216 \item @SpLim@, the stack limit pointer. This points to the end of the
218 \item Several general purpose registers, used for passing arguments to
222 \noindent and various other bits of information used in specialised
223 circumstances, such as profiling and parallel execution. These are
224 described in the appropriate sections.
226 The following is pseudo-code for the inner loop of the scheduler
230 while (threads_exist) {
231 // handle global problems: GC, parallelism, etc
233 if (external_message) service_message();
234 // deal with other urgent stuff
236 pick a runnable thread;
238 switch (thread->whatNext) {
239 case (RunGHC pc): status=runGHC(pc); break;
240 case (RunHugs bc): status=runHugs(bc); break;
242 switch (status) { // handle local problems
243 case (StackOverflow): enlargeStack; break;
244 case (Error e) : error(thread,e); break;
245 case (ExitWith e) : exit(e); break;
246 case (Yield) : break;
248 } while (thread_runnable);
252 Optimisations to avoid excess trampolining from Hugs into itself.
253 How do we invoke GC, ccalls, etc.
254 General ccall (@ccall-GC@) and optimised ccall.
258 This section describes the framework in which compiled code evaluates
259 expressions. Only at certain points will compiled code need to be
260 able to talk to the interpreted world; these are discussed in Section
261 \ref{sec:hugs-ghc-interaction}.
263 \subsection{Calling conventions}
265 \subsubsection{The call/return registers}
267 One of the problems in designing a virtual machine is that we want it
268 abstract away from tedious machine details but still reveal enough of
269 the underlying hardware that we can make sensible decisions about code
270 generation. A major problem area is the use of registers in
271 call/return conventions. On a machine with lots of registers, it's
272 cheaper to pass arguments and results in registers than to pass them
273 on the stack. On a machine with very few registers, it's cheaper to
274 pass arguments and results on the stack than to use ``virtual
275 registers'' in memory. We therefore use a hybrid system: the first
276 $n$ arguments or results are passed in registers; and the remaining
277 arguments or results are passed on the stack. For register-poor
278 architectures, it is important that we allow $n=0$.
280 We'll label the arguments and results \Arg{1} \ldots \Arg{m} --- with
281 the understanding that \Arg{1} \ldots \Arg{n} are in registers and
282 \Arg{n+1} \ldots \Arg{m} are on top of the stack.
284 Note that the mapping of arguments \Arg{1} \ldots \Arg{n} to machine
285 registers depends on the {\em kinds} of the arguments. For example,
286 if the first argument is a Float, we might pass it in a different
287 register from if it is an Int. In fact, we might find that a given
288 architecture lets us pass varying numbers of arguments according to
289 their types. For example, if a CPU has 2 Int registers and 2 Float
290 registers then we could pass between 2 and 4 arguments in machine
291 registers --- depending on whether they all have the same kind or they
292 have different kinds.
294 \subsubsection{Entering closures}
296 To evaluate a closure we jump to the entry code for the closure
297 passing a pointer to the closure in \Arg{1} so that the entry code can
298 access its environment.
300 \subsubsection{Function call}
302 The function-call mechanism is obviously crucial. There are five different
306 \item {\em Known combinator (function with no free variables) and enough arguments.}
308 A fast call can be made: push excess arguments onto stack and jump to
309 function's {\em fast entry point} passing arguments in \Arg{1} \ldots
312 The {\em fast entry point} is only called with exactly the right
313 number of arguments (in \Arg{1} \ldots \Arg{m}) so it can instantly
314 start doing useful work without first testing whether it has enough
315 registers or having to pop them off the stack first.
317 \item {\em Known combinator and insufficient arguments.}
319 A slow call can be made: push all arguments onto stack and jump to
320 function's {\em slow entry point}.
322 Any unpointed arguments which are pushed on the stack must be tagged.
323 This means pushing an extra word on the stack below the unpointed
324 words, containing the number of unpointed words above it.
326 %Todo: forward ref about tagging?
329 The {\em slow entry point} might be called with insufficient arguments
330 and so it must test whether there are enough arguments on the stack.
331 This {\em argument satisfaction check} consists of checking that
332 @Su-Sp@ is big enough to hold all the arguments (including any tags).
336 \item If the argument satisfaction check fails, it is because there is
337 one or more update frames on the stack before the rest of the
338 arguments that the function needs. In this case, we construct a PAP
339 (partial application, section~\ref{sect:PAP}) containing the arguments
340 which are on the stack. The PAP construction code will return to the
341 update frame with the address of the PAP in \Arg{1}.
343 \item If the argument satisfaction check succeeds, we jump to the fast
344 entry point with the arguments in \Arg{1} \ldots \Arg{arity}.
346 If the fast entry point expects to receive some of \Arg{i} on the
347 stack, we can reduce the amount of movement required by making the
348 stack layout for the fast entry point look like the stack layout for
349 the slow entry point. Since the slow entry point is entered with the
350 first argument on the top of the stack and with tags in front of any
351 unpointed arguments, this means that if \Arg{i} is unpointed, there
352 should be space below it for a tag and that the highest numbered
353 argument should be passed on the top of the stack.
355 We usually arrange that the fast entry point is placed immediately
356 after the slow entry point --- so we can just ``fall through'' to the
357 fast entry point without performing a jump.
362 \item {\em Known function closure (function with free variables) and enough arguments.}
364 A fast call can be made: push excess arguments onto stack and jump to
365 function's {\em fast entry point} passing a pointer to closure in
366 \Arg{1} and arguments in \Arg{2} \ldots \Arg{m+1}.
368 Like the fast entry point for a combinator, the fast entry point for a
369 closure is only called with appropriate values in \Arg{1} \ldots
370 \Arg{m+1} so we can start work straight away. The pointer to the
371 closure is used to access the free variables of the closure.
374 \item {\em Known function closure and insufficient arguments.}
376 A slow call can be made: push all arguments onto stack and jump to the
377 closure's slow entry point passing a pointer to the closure in \Arg{1}.
379 Again, the slow entry point performs an argument satisfaction check
380 and either builds a PAP or pops the arguments off the stack into
381 \Arg{2} \ldots \Arg{m+1} and jumps to the fast entry point.
384 \item {\em Unknown function closure or thunk.}
386 Sometimes, the function being called is not statically identifiable.
387 Consider, for example, the @compose@ function:
389 compose f g x = f (g x)
391 Since @f@ and @g@ are passed as arguments to @compose@, the latter has
392 to make a heap call. In a heap call the arguments are pushed onto the
393 stack, and the closure bound to the function is entered. In the
394 example, a thunk for @(g x)@ will be allocated, (a pointer to it)
395 pushed on the stack, and the closure bound to @f@ will be
396 entered. That is, we will jump to @f@s entry point passing @f@ in
397 \Arg{1}. If \Arg{1} is passed on the stack, it is pushed on top of
398 the thunk for @(g x)@.
400 The {\em entry code} for an updateable thunk (which must have arity 0)
401 pushes an update frame on the stack and starts executing the body of
402 the closure --- using \Arg{1} to access any free variables. This is
403 described in more detail in section~\ref{sect:data-updates}.
405 The {\em entry code} for a non-updateable closure is just the
406 closure's slow entry point.
410 In addition to the above considerations, if there are \emph{too many}
411 arguments then the extra arguments are simply pushed on the stack with
414 To summarise, a closure's standard (slow) entry point performs the following:
416 \item[Argument satisfaction check.] (function closure only)
417 \item[Stack overflow check.]
418 \item[Heap overflow check.]
419 \item[Copy free variables out of closure.] %Todo: why?
420 \item[Eager black holing.] (updateable thunk only) %Todo: forward ref.
421 \item[Push update frame.]
422 \item[Evaluate body of closure.]
426 \subsection{Case expressions and return conventions}
427 \label{sect:return-conventions}
429 The {\em evaluation} of a thunk is always initiated by
430 a @case@ expression. For example:
435 The code for a @case@ expression looks like this:
438 \item Push the free variables of the branches on the stack (fv(@E@) in
440 \item Push a \emph{return address} on the stack.
441 \item Evaluate the scrutinee (@x@ in this case).
444 Once evaluation of the scrutinee is complete, execution resumes at the
445 return address, which points to the code for the expression @E@.
447 When execution resumes at the return point, there must be some {\em
448 return convention} that defines where the components of the pair, @a@
449 and @b@, can be found. The return convention varies according to the
450 type of the scrutinee @x@:
456 (A space for) the return address is left on the top of the stack.
457 Leaving the return address on the stack ensures that the top of the
458 stack contains a valid activation record
459 (section~\ref{sect:activation-records}) --- should a garbage collection
462 \item If @x@ has a pointed type (e.g.~a data constructor or a function),
463 a pointer to @x@ is returned in \Arg{1}.
465 \ToDo{Warn that components of E should be extracted as soon as
466 possible to avoid a space leak.}
468 \item If @x@ is an unpointed type (e.g.~@Int#@ or @Float#@), @x@ is
471 \item If @x@ is an unpointed tuple constructor, the components of @x@
472 are returned in \Arg{1} \ldots \Arg{n} but no object is constructed in
473 the heap. Unboxed tuple constructors are useful for functions which
474 want to return multiple values such as those used in an (explicitly
475 encoded) state monad:
477 \ToDo{Move stuff about unboxed tuples to a separate section}
480 data unpointed AAndState a = AnS a State
481 type S a = State -> AAndState a
483 bindS m k s0 = case m s0 of { AnS s1 a -> k s1 a }
484 returnS a s = AnS a s
487 Note that unboxed datatypes can only have one constructor and that
488 thunks never have unboxed types --- so we'll never try to update an
489 unboxed constructor. Unboxed tuples are \emph{never} built on the
492 When passing an unboxed tuple to a function, the components are
493 flattened out and passed in \Arg{1} \ldots \Arg{n} as usual.
497 \subsection{Vectored Returns}
499 Many algebraic data types have more than one constructor. For
500 example, the @Maybe@ type is defined like this:
502 data Maybe a = Nothing | Just a
504 How does the return convention encode which of the two constructors is
505 being returned? A @case@ expression scrutinising a value of @Maybe@
506 type would look like this:
512 Rather than pushing a return address before evaluating the scrutinee,
513 @E@, the @case@ expression pushes (a pointer to) a {\em return
514 vector}, a static table consisting of two code pointers: one for the
515 @Just@ alternative, and one for the @Nothing@ alternative.
521 The constructor @Nothing@ returns by jumping to the first item in the
522 return vector with a pointer to a (statically built) Nothing closure
525 It might seem that we could avoid loading \Arg{1} in this case since the
526 first item in the return vector will know that @Nothing@ was returned
527 (and can easily access the Nothing closure in the (unlikely) event
528 that it needs it. The only reason we load \Arg{1} is in case we have to
529 perform an update (section~\ref{sect:data-updates}).
533 The constructor @Just@ returns by jumping to the second element of the
534 return vector with a pointer to the closure in \Arg{1}.
538 In this way no test need be made to see which constructor returns;
539 instead, execution resumes immediately in the appropriate branch of
542 \subsection{Direct Returns}
544 When a datatype has a large number of constructors, it may be
545 inappropriate to use vectored returns. The vector tables may be
546 large and sparse, and it may be better to identify the constructor
547 using a test-and-branch sequence on the tag. For this reason, we
548 provide an alternative return convention, called a \emph{direct
551 In a direct return, the return address pushed on the stack really is a
552 code pointer. The returning code loads a pointer to the closure being
553 returned in \Arg{1} as usual, and also loads the tag into \Arg{2}.
554 The code at the return address will test the tag and jump to the
555 appropriate code for the case branch.
557 The choice of whether to use a vectored return or a direct return is
558 made on a type-by-type basis --- up to a certain maximum number of
559 constructors imposed by the update mechanism
560 (section~\ref{sect:data-updates}).
562 Single-constructor data types also use direct returns, although in
563 that case there is no need to return a tag in \Arg{2}.
565 \ToDo{Say whether we pop the return address before returning}
567 \ToDo{Stack stubbing?}
570 \label{sect:data-updates}
572 The entry code for an updatable thunk (which must also be of arity 0):
575 \item copies the free variables out of the thunk into registers or
577 \item pushes an {\em update frame} onto the stack.
579 An update frame is a small activation record consisting of
581 \begin{tabular}{|l|l|l|}
583 {\em Fixed header} & {\em Update Frame link} & {\em Updatee} \\
588 \note{In the semantics part of the STG paper (section 5.6), an update
589 frame consists of everything down to the last update frame on the
590 stack. This would make sense too --- and would fit in nicely with
591 what we're going to do when we add support for speculative
593 \ToDo{I think update frames contain cost centres sometimes}
596 If we are doing ``eager blackholing,'' we then overwrite the thunk
597 with a black hole. Otherwise, we leave it to the garbage collector to
598 black hole the thunk.
601 Start evaluating the body of the expression.
605 When the expression finishes evaluation, it will enter the update
606 frame on the top of the stack. Since the returner doesn't know
607 whether it is entering a normal return address/vector or an update
608 frame, we follow exactly the same conventions as return addresses and
609 return vectors. That is, on entering the update frame:
612 \item The value of the thunk is in \Arg{1}. (Recall that only thunks
613 are updateable and that thunks return just one value.)
615 \item If the data type is a direct-return type rather than a
616 vectored-return type, then the tag is in \Arg{2}.
618 \item The update frame is still on the stack.
621 We can safely share a single statically-compiled update function
622 between all types. However, the code must be able to handle both
623 vectored and direct-return datatypes. This is done by arranging that
624 the update code looks like this:
632 |---------------| <- update code pointer
637 Each entry in the return vector (which is large enough to cover the
638 largest vectored-return type) points to the update code.
642 \item overwrites the {\em updatee} with an indirection to \Arg{1};
643 \item loads @Su@ from the Update Frame link;
644 \item removes the update frame from the stack; and
645 \item enters \Arg{1}.
648 We enter \Arg{1} again, having probably just come from there, because
649 it knows whether to perform a direct or vectored return. This could
650 be optimised by compiling special update code for each slot in the
651 return vector, which performs the correct return.
653 \subsection{Semi-tagging}
654 \label{sect:semi-tagging}
656 When a @case@ expression evaluates a variable that might be bound
657 to a thunk it is often the case that the scrutinee is already evaluated.
658 In this case we have paid the penalty of (a) pushing the return address (or
659 return vector address) on the stack, (b) jumping through the info pointer
660 of the scrutinee, and (c) returning by an indirect jump through the
661 return address on the stack.
663 If we knew that the scrutinee was already evaluated we could generate
664 (better) code which simply jumps to the appropriate branch of the @case@
665 with a pointer to the scrutinee in \Arg{1}.
667 An obvious idea, therefore, is to test dynamically whether the heap
668 closure is a value (using the tag in the info table). If not, we
669 enter the closure as usual; if so, we jump straight to the appropriate
670 alternative. Here, for example, is pseudo-code for the expression
671 @(case x of { (a,_,c) -> E }@:
673 \Arg{1} = <pointer to x>;
674 tag = \Arg{1}->entry->tag;
676 Sp--; \\ insert space for return address
682 <info table for return address goes here>
683 ret: a = \Arg{1}->data1; \\ suck out a and c to avoid space leak
687 and here is the code for the expression @(case x of { [] -> E1; x:xs -> E2 }@:
689 \Arg{1} = <pointer to x>;
690 tag = \Arg{1}->entry->tag;
692 Sp--; \\ insert space for return address
700 retvec: \\ reversed return vector
701 <return info table for case goes here>
703 panic("Direct return into vectored case");
707 ret2: x = \Arg{1}->head;
711 There is an obvious cost in compiled code size (but none in the size
712 of the bytecodes). There is also a cost in execution time if we enter
713 more thunks than data constructors.
715 Both the direct and vectored returns are easily modified to chase chains
716 of indirections too. In the vectored case, this is most easily done by
717 making sure that @IND = TAG_1 - 1@, and adding an extra field to every
718 return vector. In the above example, the indirection code would be
720 ind: \Arg{1} = \Arg{1}->next;
723 where @ind_loop@ is the second line of code.
725 Note that we have to leave space for a return address since the return
726 address expects to find one. If the body of the expression requires a
727 heap check, we will actually have to write the return address before
728 entering the garbage collector.
731 \subsection{Heap and Stack Checks}
733 \note{I reckon these deserve a subsection of their own}
735 Don't move heap pointer before success occurs.
736 Talk about how stack check looks ahead into the branches of case expressions.
738 \subsection{Handling interrupts/signals}
741 May have to keep C stack pointer in register to placate OS?
742 May have to revert black holes - ouch!
745 \section{Switching Worlds}
747 Because this is a combined compiled/interpreted system, the
748 interpreter will sometimes encounter compiled code, and vice-versa.
750 There are six cases we need to consider:
753 \item A GHC thread enters a Hugs-built thunk.
754 \item A GHC thread calls a Hugs-compiled function.
755 \item A GHC thread returns to a Hugs-compiled return address.
756 \item A Hugs thread enters a GHC-built thunk.
757 \item A Hugs thread calls a GHC-compiled function.
758 \item A Hugs thread returns to a Hugs-compiled return address.
761 \subsection{A GHC thread enters a Hugs-built thunk}
763 A Hugs-built thunk looks like this:
766 \begin{tabular}{|l|l|}
768 \emph{Hugs} & \emph{Hugs-specific information} \\
773 \noindent where \emph{Hugs} is a pointer to a small
774 statically-compiled piece of code that does the following:
777 \item Push the address of the thunk on the stack.
778 \item Push @entertop@ on the stack.
779 \item Save the current state of the thread in the TSO.
780 \item Return to the scheduler, with the @whatNext@ field set to
784 \noindent where @entertop@ is a small statically-compiled piece of
785 code that does the following:
788 \item pop the return address from the stack.
789 \item pop the next word off the stack into \Arg{1}.
793 The infotable for @entertop@ has some byte-codes attached that do
794 essentially the same thing if the code is entered from Hugs.
796 \subsection{A GHC thread calls a Hugs-compiled function}
800 \subsection{A GHC thread returns to a Hugs-compiled return address}
802 When Hugs pushes return addresses on the stack, they look like this:
807 | | -----> bytecode object
810 |_______________| |___ GHC-friendly return code
820 If GHC is returning, it will return to the address at the top of the
821 stack. The code at this address
824 \item saves the thread state in the TSO
825 \item returns to the scheduler with a @whatNext@ field of @RunHugs@.
828 If Hugs is returning to one of these addresses, it can spot the
829 special return address at the top and instead jump to the bytecodes
830 pointed to by the second word on the stack.
832 \subsection{A Hugs thread enters a GHC-compiled thunk}
834 When Hugs is called on to enter a non-Hugs closure (these are
835 recognisable by the lack of a \emph{Hugs} pointer at the front), the
836 following sequence of instructions is executed:
839 \item Push the address of the thunk on the stack.
840 \item Push @entertop@ on the stack.
841 \item Save the current state of the thread in the TSO.
842 \item Return to the scheduler, with the @whatNext@ field set to
846 \subsection{A Hugs thread calls a GHC-compiled function}
848 Hugs never calls GHC-functions directly, it only enters closures
849 (which point to the slow entry point for the function). Hence in this
850 case, we just push the arguments on the stack and proceed as for a
853 \subsection{A Hugs thread returns to a GHC-compiled return address}
855 \section{Heap objects}
856 \label{sect:fixed-header}
858 \ToDo{Fix this picture}
868 Every {\em heap object}, or {\em closure} is a contiguous block
869 of memory, consisting of a fixed-format {\em header} followed
870 by zero or more {\em data words}.
871 The header consists of
872 the following fields:
874 \item A one-word {\em info pointer}, which points to
875 the object's static {\em info table}.
876 \item Zero or more {\em admin words} that support
878 \item Profiling (notably a {\em cost centre} word).
879 \note{We could possibly omit the cost centre word from some
880 administrative objects.}
881 \item Parallelism (e.g. GranSim keeps the object's global address here,
882 though GUM keeps a separate hash table).
883 \item Statistics (e.g. a word to track how many times a thunk is entered.).
885 We add a Ticky word to the fixed-header part of closures. This is
886 used to record indicate if a closure has been updated but not yet
887 entered. It is set when the closure is updated and cleared when
888 subsequently entered.
890 NB: It is {\em not} an ``entry count'', it is an
891 ``entries-after-update count.'' The commoning up of @CONST@,
892 @CHARLIKE@ and @INTLIKE@ closures is turned off(?) if this is
893 required. This has only been done for 2s collection.
899 Most of the RTS is completely insensitive to the number of admin words.
900 The total size of the fixed header is @FIXED_HS@.
902 Many heap objects contain fields allowing them to be inserted onto lists
903 during evaluation or during garbage collection. The lists required by
904 the evaluator and storage manager are as follows.
907 \item 2 lists of threads: runnable threads and sleeping threads.
909 \item The {\em static object list} is a list of all statically
910 allocated objects which might contain pointers into the heap.
911 (Section~\ref{sect:static-objects}.)
913 \item The {\em updated thunk list} is a list of all thunks in the old
914 generation which have been updated with an indirection.
915 (Section~\ref{sect:IND_OLDGEN}.)
917 \item The {\em mutables list} is a list of all other objects in the
918 old generation which might contain pointers into the new generation.
919 Most of the object on this list are ``mutable.''
920 (Section~\ref{sect:mutables}.)
922 \item The {\em Foreign Object list} is a list of all foreign objects
923 which have not yet been deallocated. (Section~\ref{sect:FOREIGN}.)
925 \item The {\em Spark pool} is a doubly(?) linked list of Spark objects
926 maintained by the parallel system. (Section~\ref{sect:SPARK}.)
928 \item The {\em Blocked Fetch list} (or
929 lists?). (Section~\ref{sect:BLOCKED_FETCH}.)
931 \item For each thread, there is a list of all update frames on the
932 stack. (Section~\ref{sect:data-updates}.)
937 \ToDo{The links for these fields are usually inserted immediately
938 after the fixed header except ...}
940 \subsection{Info Tables}
942 An {\em info table} is a contiguous block of memory, {\em laid out
943 backwards}. That is, the first field in the list that follows
944 occupies the highest memory address, and the successive fields occupy
945 successive decreasing memory addresses.
949 \hline Parallelism Info
950 \\ \hline Profile Info
952 \\ \hline Tag/bytecode pointer
953 \\ \hline Static reference table
954 \\ \hline Storage manager layout info
955 \\ \hline Closure type
956 \\ \hline entry code \ldots
960 An info table has the following contents (working backwards in memory
963 \item The {\em entry code} for the closure.
964 This code appears literally as the (large) last entry in the
965 info table, immediately preceded by the rest of the info table.
966 An {\em info pointer} always points to the first byte of the entry code.
968 \item A one-word {\em closure type field}, @INFO_TYPE@, identifies what kind
969 of closure the object is. The various types of closure are described
970 in Section~\ref{sect:closures}.
971 In some configurations, some useful properties of
972 closures (is it a HNF? can it be sparked?)
973 are represented as high-order bits so they can be tested quickly.
975 \item A single pointer or word --- the {\em storage manager info field},
976 @INFO_SM@, contains auxiliary information describing the closure's
977 precise layout, for the benefit of the garbage collector and the code
978 that stuffs graph into packets for transmission over the network.
980 \item A one-pointer {\em Static Reference Table (SRT) pointer}, @INFO_SRT@, points to
981 a table which enables the garbage collector to identify all accessible
982 code and CAFs. They are fully described in Section~\ref{sect:srt}.
984 \item A one-pointer {\em tag/bytecode-pointer} field, @INFO_TAG@ or @INFO_BC@.
985 For data constructors this field contains the constructor tag, in the
986 range $0..n-1$ where $n$ is the number of constructors.
988 For other objects that can be entered this field points to the byte
989 codes for the object. For the constructor case you can think of the
990 tag as the name of a a suitable bytecode sequence but it can also be used to
991 implement semi-tagging (section~\ref{sect:semi-tagging}).
993 One awkward question (which may not belong here) is ``how does the
994 bytecode interpreter know whether to do a vectored return?'' The
995 answer is it examines the @INFO_TYPE@ field of the return address:
996 @RET_VEC_@$sz$ requires a vectored return and @RET_@$sz$ requires a
999 \item {\em Profiling info\/}
1001 Closure category records are attached to the info table of the
1002 closure. They are declared with the info table. We put pointers to
1003 these ClCat things in info tables. We need these ClCat things because
1004 they are mutable, whereas info tables are immutable. Hashing will map
1005 similar categories to the same hash value allowing statistics to be
1006 grouped by closure category.
1008 Cost Centres and Closure Categories are hashed to provide indexes
1009 against which arbitrary information can be stored. These indexes are
1010 memoised in the appropriate cost centre or category record and
1011 subsequent hashes avoided by the index routine (it simply returns the
1014 There are different features which can be hashed allowing information
1015 to be stored for different groupings. Cost centres have the cost
1016 centre recorded (using the pointer), module and group. Closure
1017 categories have the closure description and the type
1018 description. Records with the same feature will be hashed to the same
1021 The initialisation routines, @init_index_<feature>@, allocate a hash
1022 table in which the cost centre / category records are stored. The
1023 lower bound for the table size is taken from @max_<feature>_no@. They
1024 return the actual table size used (the next power of 2). Unused
1025 locations in the hash table are indicated by a 0 entry. Successive
1026 @init_index_<feature>@ calls just return the actual table size.
1028 Calls to @index_<feature>@ will insert the cost centre / category
1029 record in the @<feature>@ hash table, if not already inserted. The hash
1030 index is memoised in the record and returned.
1032 CURRENTLY ONLY ONE MEMOISATION SLOT IS AVILABLE IN EACH RECORD SO
1033 HASHING CAN ONLY BE DONE ON ONE FEATURE FOR EACH RECORD. This can be
1034 easily relaxed at the expense of extra memoisation space or continued
1037 The initialisation routines must be called before initialisation of
1038 the stacks and heap as they require to allocate storage. It is also
1039 expected that the caller may want to allocate additional storage in
1040 which to store profiling information based on the return table size
1044 \begin{tabular}{|l|}
1048 \\ \hline Description String
1049 \\ \hline Type String
1055 \item[Hash Index] Memoised copy
1057 Is this category selected (-1 == not memoised, selected? 0 or 1)
1059 One of the following values (defined in CostCentre.lh):
1067 A partial application.
1069 A thunk, or suspension.
1074 \item[@ForeignObj_K@]
1075 A Foreign object (non-Haskell heap resident).
1077 The Stable Pointer table. (There should only be one of these but it
1078 represents a form of weak space leak since it can't shrink to meet
1079 non-demand so it may be worth watching separately? ADR)
1080 \item[@INTERNAL_KIND@]
1081 Something internal to the runtime system.
1085 \item[Description] Source derived string detailing closure description.
1086 \item[Type] Source derived string detailing closure type.
1089 \item {\em Parallelism info\/}
1092 \item {\em Debugging info\/}
1099 \section{Kinds of Heap Object}
1100 \label{sect:closures}
1102 Heap objects can be classified in several ways, but one useful one is
1106 {\em Static closures} occupy fixed, statically-allocated memory
1107 locations, with globally known addresses.
1110 {\em Dynamic closures} are individually allocated in the heap.
1113 {\em Stack closures} are closures allocated within a thread's stack
1114 (which is itself a heap object). Unlike other closures, there are
1115 never any pointers to stack closures. Stack closures are discussed in
1116 Section~\ref{sect:stacks}.
1119 A second useful classification is this:
1122 {\em Executive closures}, such as thunks and data constructors,
1123 participate directly in a program's execution. They can be subdivided into
1124 two kinds of objects according to their type:
1127 {\em Pointed objects}, represent values of a {\em pointed} type
1128 (<.pointed types launchbury.>) --i.e.~a type that includes $\bottom$ such as @Int@ or @Int# -> Int#@.
1130 \item {\em Unpointed objects}, represent values of a {\em unpointed} type --i.e.~a type that does not include $\bottom$ such as @Int#@ or @Array#@.
1132 \item {\em Activation frames}, represent ``continuations''. They are
1133 always stored on the stack and are never pointed to by heap objects or
1134 passed as arguments. \note{It's not clear if this will still be true
1135 once we support speculative evaluation.}
1139 \item {\em Administrative closures}, such as stack objects and thread
1140 state objects, do not represent values in the original program.
1143 Only pointed objects can be entered. All pointed objects share a
1144 common header format: the ``pointed header''; while all unpointed
1145 objects share a common header format: the ``unpointed header''.
1146 \ToDo{Describe the difference and update the diagrams to mention
1147 an appropriate header type.}
1149 This section enumerates all the kinds of heap objects in the system.
1150 Each is identified by a distinct @INFO_TYPE@ tag in its info table.
1152 \ToDo{Check this table very carefully}
1154 \begin{tabular}{|l|l|l|l|l|l|l|l|l|l|l|}
1157 closure kind & HNF & UPD & NS & STA & THU & MUT & UPT & BH & IND & Section \\
1163 @CONSTR@ & 1 & & 1 & & & & & & & \ref{sect:CONSTR} \\
1164 @CONSTR_STATIC@ & 1 & & 1 & 1 & & & & & & \ref{sect:CONSTR} \\
1165 @CONSTR_STATIC_NOCAF@ & 1 & & 1 & 1 & & & & & & \ref{sect:CONSTR} \\
1167 @FUN@ & 1 & & ? & & & & & & & \ref{sect:FUN} \\
1168 @FUN_STATIC@ & 1 & & ? & 1 & & & & & & \ref{sect:FUN} \\
1170 @THUNK@ & 1 & 1 & & & 1 & & & & & \ref{sect:THUNK} \\
1171 @THUNK_STATIC@ & 1 & 1 & & 1 & 1 & & & & & \ref{sect:THUNK} \\
1172 @THUNK_SELECTOR@ & 1 & 1 & 1 & & 1 & & & & & \ref{sect:THUNK_SEL} \\
1174 @PAP@ & 1 & & ? & & & & & & & \ref{sect:PAP} \\
1176 @IND@ & & & 1 & & ? & & & & 1 & \ref{sect:IND} \\
1177 @IND_OLDGEN@ & 1 & & 1 & & ? & & & & 1 & \ref{sect:IND} \\
1178 @IND_PERM@ & & & 1 & & ? & & & & 1 & \ref{sect:IND} \\
1179 @IND_OLDGEN_PERM@ & 1 & & 1 & & ? & & & & 1 & \ref{sect:IND} \\
1180 @IND_STATIC@ & ? & & 1 & 1 & ? & & & & 1 & \ref{sect:IND} \\
1187 @ARR_WORDS@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:ARR_WORDS1},\ref{sect:ARR_WORDS2} \\
1188 @ARR_PTRS@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:ARR_PTRS} \\
1189 @MUTVAR@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:MUTVAR} \\
1190 @MUTARR_PTRS@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:MUTARR_PTRS} \\
1191 @MUTARR_PTRS_FROZEN@ & 1 & & 1 & & & 1 & 1 & & & \ref{sect:MUTARR_PTRS_FROZEN} \\
1193 @FOREIGN@ & 1 & & 1 & & & & 1 & & & \ref{sect:FOREIGN} \\
1195 @BH@ & ? & 0/1 & 1 & & ? & ? & & 1 & ? & \ref{sect:BH} \\
1196 @MVAR@ & & & & & & & & & & \ref{sect:MVAR} \\
1197 @IVAR@ & & & & & & & & & & \ref{sect:IVAR} \\
1198 @FETCHME@ & & & & & & & & & & \ref{sect:FETCHME} \\
1202 Activation frames do not live (directly) on the heap --- but they have
1203 a similar organisation. The classification bits are all zero in
1206 \begin{tabular}{|l|l|}\hline
1207 closure kind & Section \\ \hline
1208 @RET_SMALL@ & \ref{sect:activation-records} \\
1209 @RET_VEC_SMALL@ & \ref{sect:activation-records} \\
1210 @RET_BIG@ & \ref{sect:activation-records} \\
1211 @RET_VEC_BIG@ & \ref{sect:activation-records} \\
1212 @UPDATE_FRAME@ & \ref{sect:activation-records} \\
1216 There are also a number of administrative objects. The classification bits are
1217 all zero in administrative objects.
1219 \begin{tabular}{|l|l|}\hline
1220 closure kind & Section \\ \hline
1221 @TSO@ & \ref{sect:TSO} \\
1222 @STACK_OBJECT@ & \ref{sect:STACK_OBJECT} \\
1223 @STABLEPTR_TABLE@ & \ref{sect:STABLEPTR_TABLE} \\
1224 @SPARK_OBJECT@ & \ref{sect:SPARK} \\
1225 @BLOCKED_FETCH@ & \ref{sect:BLOCKED_FETCH} \\
1229 \ToDo{I guess the parallel system has something like a stable ptr
1230 table. Is there any opportunity for sharing code/data structures
1234 \subsection{Classification bits}
1236 The top bits of the @INFO_TYPE@ tag tells what sort of animal the
1239 \begin{tabular}{|l|l|l|} \hline
1240 Abbrev & Bit & Interpretation \\ \hline
1241 HNF & 0 & 1 $\Rightarrow$ Head normal form \\
1242 UPD & 4 & 1 $\Rightarrow$ May be updated (inconsistent with being a HNF) \\
1243 NS & 1 & 1 $\Rightarrow$ Don't spark me (Any HNF will have this set to 1)\\
1244 STA & 2 & 1 $\Rightarrow$ This is a static closure \\
1245 THU & 8 & 1 $\Rightarrow$ Is a thunk \\
1246 MUT & 3 & 1 $\Rightarrow$ Has mutable pointer fields \\
1247 UPT & 5 & 1 $\Rightarrow$ Has an unpointed type (eg a primitive array) \\
1248 BH & 6 & 1 $\Rightarrow$ Is a black hole \\
1249 IND & 7 & 1 $\Rightarrow$ Is an indirection \\
1253 Updatable structures (@_UP@) are thunks that may be shared. Primitive
1254 arrays (@_BM@ -- Big Mothers) are structures that are always held
1255 in-memory (basically extensions of a closure). Because there may be
1256 offsets into these arrays, a primitive array cannot be handled as a
1257 FetchMe in the parallel system, but must be shipped in its entirety if
1258 its parent closure is shipped.
1260 The other bits in the info-type field simply give a unique bit-pattern
1261 to identify the closure type.
1265 #define _NF 0x0001 /* Normal form */
1266 #define _NS 0x0002 /* Don't spark */
1267 #define _ST 0x0004 /* Is static */
1268 #define _MU 0x0008 /* Is mutable */
1269 #define _UP 0x0010 /* Is updatable (but not mutable) */
1270 #define _BM 0x0020 /* Is a "primitive" array */
1271 #define _BH 0x0040 /* Is a black hole */
1272 #define _IN 0x0080 /* Is an indirection */
1273 #define _TH 0x0100 /* Is a thunk */
1278 SPEC_N SPEC | _NF | _NS
1280 SPEC_U SPEC | _UP | _TH
1283 GEN_N GEN | _NF | _NS
1285 GEN_U GEN | _UP | _TH
1288 TUPLE _NF | _NS | _BM
1289 DATA _NF | _NS | _BM
1290 MUTUPLE _NF | _NS | _MU | _BM
1291 IMMUTUPLE _NF | _NS | _BM
1303 CAF _NF | _NS | _ST | _IN
1312 STKO_DYNAMIC STKO | _MU
1313 STKO_STATIC STKO | _ST
1315 SPEC_RBH _NS | _MU | _BH
1316 GEN_RBH _NS | _MU | _BH
1325 An indirection either points to HNF (post update); or is result of
1326 overwriting a FetchMe, in which case the thing fetched is either
1327 under evaluation (BH), or by now an HNF. Thus, indirections get NoSpark flag.
1331 \subsection{Pointed Objects}
1333 All pointed objects can be entered.
1335 \subsubsection{Function closures}\label{sect:FUN}
1337 Function closures represent lambda abstractions. For example,
1338 consider the top-level declaration:
1340 f = \x -> let g = \y -> x+y
1343 Both @f@ and @g@ are represented by function closures. The closure
1344 for @f@ is {\em static} while that for @g@ is {\em dynamic}.
1346 The layout of a function closure is as follows:
1348 \begin{tabular}{|l|l|l|l|}\hline
1349 {\em Fixed header} & {\em Pointers} & {\em Non-pointers} \\ \hline
1352 The data words (pointers and non-pointers) are the free variables of
1353 the function closure.
1354 The number of pointers
1355 and number of non-pointers are stored in the @INFO_SM@ word, in the least significant
1356 and most significant half-word respectively.
1358 There are several different sorts of function closure, distinguished
1359 by their @INFO_TYPE@ field:
1361 \item @FUN@: a vanilla, dynamically allocated on the heap.
1363 \item $@FUN_@p@_@np$: to speed up garbage collection a number of
1364 specialised forms of @FUN@ are provided, for particular $(p,np)$ pairs,
1365 where $p$ is the number of pointers and $np$ the number of non-pointers.
1367 \item @FUN_STATIC@. Top-level, static, function closures (such as
1368 @f@ above) have a different
1369 layout than dynamic ones:
1371 \begin{tabular}{|l|l|l|}\hline
1372 {\em Fixed header} & {\em Static object link} \\ \hline
1375 Static function closurs have no free variables. (However they may refer to other
1376 static closures; these references are recorded in the function closure's SRT.)
1377 They have one field that is not present in dynamic closures, the {\em static object
1378 link} field. This is used by the garbage collector in the same way that to-space
1379 is, to gather closures that have been determined to be live but that have not yet
1381 \note{Static function closures that have no static references, and hence
1382 a null SRT pointer, don't need the static object link field. Is it worth
1383 taking advantage of this? See @CONSTR_STATIC_NOCAF@.}
1386 Each lambda abstraction, $f$, in the STG program has its own private info table.
1387 The following labels are relevant:
1389 \item $f$@_info@ is $f$'s info table.
1390 \item $f$@_entry@ is $f$'s slow entry point (i.e. the entry code of its
1391 info table; so it will label the same byte as $f$@_info@).
1392 \item $f@_fast_@k$ is $f$'s fast entry point. $k$ is the number of arguments
1393 $f$ takes; encoding this number in the fast-entry label occasionally catches some nasty
1394 code-generation errors.
1397 \subsubsection{Data Constructors}\label{sect:CONSTR}
1399 Data-constructor closures represent values constructed with
1400 algebraic data type constructors.
1401 The general layout of data constructors is the same as that for function
1404 \begin{tabular}{|l|l|l|l|}\hline
1405 {\em Fixed header} & {\em Pointers} & {\em Non-pointers} \\ \hline
1409 The SRT pointer in a data constructor's info table is never used --- the
1410 code for a constructor does not make any static references.
1411 \note{Use it for something else?? E.g. tag?}
1413 There are several different sorts of constructor:
1415 \item @CONSTR@: a vanilla, dynamically allocated constructor.
1416 \item @CONSTR_@$p$@_@$np$: just like $@FUN_@p@_@np$.
1417 \item @CONSTR_INTLIKE@.
1418 A dynamically-allocated heap object that looks just like an @Int@. The
1419 garbage collector checks to see if it can common it up with one of a fixed
1420 set of static int-like closures, thus getting it out of the dynamic heap
1423 \item @CONSTR_CHARLIKE@: same deal, but for @Char@.
1425 \item @CONSTR_STATIC@ is similar to @FUN_STATIC@, with the complication that
1426 the layout of the constructor must mimic that of a dynamic constructor,
1427 because a static constructor might be returned to some code that unpacks it.
1428 So its layout is like this:
1430 \begin{tabular}{|l|l|l|l|l|}\hline
1431 {\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Static object link}\\ \hline
1434 The static object link, at the end of the closure, serves the same purpose
1435 as that for @FUN_STATIC@. The pointers in the static constructor can point
1436 only to other static closures.
1438 The static object link occurs last in the closure so that static
1439 constructors can store their data fields in exactly the same place as
1440 dynamic constructors.
1442 \item @CONSTR_STATIC_NOCAF@. A statically allocated data constructor
1443 that guarantees not to point (directly or indirectly) to any CAF
1444 (section~\ref{sect:CAF}). This means it does not need a static object
1445 link field. Since we expect that there might be quite a lot of static
1446 constructors this optimisation makes sense. Furthermore, the @NOCAF@
1447 tag allows the compiler to indicate that no CAFs can be reached
1448 anywhere {\em even indirectly}.
1453 For each data constructor $Con$, two
1454 info tables are generated:
1456 \item $Con$@_info@ labels $Con$'s dynamic info table,
1457 shared by all dynamic instances of the constructor.
1458 \item $Con$@_static@ labels $Con$'s static info table,
1459 shared by all static instances of the constructor.
1462 \subsubsection{Thunks}
1464 \label{sect:THUNK_SEL}
1466 A thunk represents an expression that is not obviously in head normal
1467 form. For example, consider the following top-level definitions:
1469 range = between 1 10
1470 f = \x -> let ys = take x range
1473 Here the right-hand sides of @range@ and @ys@ are both thunks; the former
1474 is static while the latter is dynamic.
1476 The layout of a thunk is the same as that for a function closure,
1477 except that it may have some words of ``slop'' at the end to make sure
1479 at least @MIN_UPD_PAYLOAD@ words in addition to its
1482 \begin{tabular}{|l|l|l|l|l|}\hline
1483 {\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Slop} \\ \hline
1486 The @INFO_SM@ word contains the same information as for function
1487 closures; that is, number of pointers and number of non-pointers (excluding slop).
1489 A thunk differs from a function closure in that it can be updated.
1491 There are several forms of thunk:
1493 \item @THUNK@: a vanilla, dynamically allocated thunk.
1494 The garbage collection code for thunks whose
1495 pointer + non-pointer words is less than @MIN_UPD_PAYLOAD@ differs from
1496 that for function closures and data constructors, because the GC code
1497 has to account for the slop.
1498 \item $@THUNK_@p@_@np$. Similar comments apply.
1499 \item @THUNK_STATIC@. A static thunk is also known as
1500 a {\em constant applicative form}, or {\em CAF}.
1503 \begin{tabular}{|l|l|l|l|l|}\hline
1504 {\em Fixed header} & {\em Pointers} & {\em Non-pointers} & {\em Slop} & {\em Static object link}\\ \hline
1508 \item @THUNK_SELECTOR@ is a (dynamically allocated) thunk
1509 whose entry code performs a simple selection operation from
1510 a data constructor drawn from a single-constructor type. For example,
1513 x = case y of (a,b) -> a
1515 is a selector thunk. A selector thunk is laid out like this:
1517 \begin{tabular}{|l|l|l|l|}\hline
1518 {\em Fixed header} & {\em Selectee pointer} \\ \hline
1521 The @INFO_SM@ word contains the byte offset of the desired word in
1522 the selectee. Note that this is different from all other thunks.
1524 The garbage collector ``peeks'' at the selectee's
1525 tag (in its info table). If it is evaluated, then it goes ahead and do
1526 the selection, and then behaves just as if the selector thunk was an
1527 indirection to the selected field.
1529 evaluated, it treats the selector thunk like any other thunk of that
1530 shape. [Implementation notes.
1531 Copying: only the evacuate routine needs to be special.
1532 Compacting: only the PRStart (marking) routine needs to be special.]
1536 The only label associated with a thunk is its info table:
1538 \item[$f$@_info@] is $f$'s info table.
1542 \subsubsection{Partial applications (PAPs)}\label{sect:PAP}
1544 A partial application (PAP) represents a function applied to too few arguments.
1545 It is only built as a result of updating after an argument-satisfaction
1546 check failure. A PAP has the following shape:
1548 \begin{tabular}{|l|l|l|l|}\hline
1549 {\em Fixed header} & {\em No of arg words} & {\em Function closure} & {\em Arg stack} \\ \hline
1552 The ``arg stack'' is a copy of of the chunk of stack above the update
1553 frame; ``no of arg words'' tells how many words it consists of. The
1554 function closure is (a pointer to) the closure for the function whose
1555 argument-satisfaction check failed.
1557 There is just one standard form of PAP with @INFO_TYPE@ = @PAP@.
1558 There is just one info table too, called @PAP_info@.
1559 Its entry code simply copies the arg stack chunk back on top of the
1560 stack and enters the function closure. (It has to do a stack overflow test first.)
1562 There are no static PAPs.
1564 \subsubsection{Indirections}
1566 \label{sect:IND_OLDGEN}
1568 Indirection closures just point to other closures. They are introduced
1569 when a thunk is updated to point to its value.
1570 The entry code for all indirections simply enters the closure it points to.
1572 There are several forms of indirection:
1574 \item[@IND@] is the vanilla, dynamically-allocated indirection.
1575 It is removed by the garbage collector. It has the following
1578 \begin{tabular}{|l|l|l|}\hline
1579 {\em Fixed header} & {\em Target closure} \\ \hline
1583 \item[@IND_OLDGEN@] is the indirection used to update an old-generation
1584 thunk. Its shape is like this:
1586 \begin{tabular}{|l|l|l|}\hline
1587 {\em Fixed header} & {\em Mutable link field} & {\em Target closure} \\ \hline
1590 It contains a {\em mutable link field} that is used to string together
1591 all old-generation indirections that might have a pointer into
1595 \item[@IND_PERMANENT@ and @IND_OLDGEN_PERMANENT@.]
1596 for lexical profiling, it is necessary to maintain cost centre
1597 information in an indirection, so ``permanent indirections'' are
1598 retained forever. Otherwise they are just like vanilla indirections.
1599 \note{If a permanent indirection points to another permanent
1600 indirection or a @CONST@ closure, it is possible to elide the indirection
1601 since it will have no effect on the profiler.}
1602 \note{Do we still need @IND@ and @IND_OLDGEN@
1603 in the profiling build, or can we just make
1604 do with one pair whose behaviour changes when profiling is built?}
1606 \item[@IND_STATIC@] is used for overwriting CAFs when they have been
1607 evaluated. Static indirections are not removed by the garbage
1608 collector; and are statically allocated outside the heap (and should
1609 stay there). Their static object link field is used just as for
1610 @FUN_STATIC@ closures.
1613 \begin{tabular}{|l|l|l|}
1615 {\em Fixed header} & {\em Target closure} & {\em Static object link} \\
1622 \subsubsection{Activation Records}
1624 Activation records are parts of the stack described by return address
1625 info tables (closures with @INFO_TYPE@ values of @RET_SMALL@,
1626 @RET_VEC_SMALL@, @RET_BIG@ and @RET_VEC_BIG@). They are described in
1627 section~\ref{sect:activation-records}.
1630 \subsubsection{Black holes, MVars and IVars}
1635 Black hole closures are used to overwrite closures currently being
1636 evaluated. They inform the garbage collector that there are no live
1637 roots in the closure, thus removing a potential space leak.
1639 Black holes also become synchronization points in the threaded world.
1640 They contain a pointer to a list of blocked threads to be awakened
1641 when the black hole is updated (or @NULL@ if the list is empty).
1643 \begin{tabular}{|l|l|l|}
1645 {\em Fixed header} & {\em Mutable link} & {\em Blocked thread link} \\
1649 The {\em Blocked thread link} points to the TSO of the first thread
1650 waiting for the value of this thunk. All subsequent TSOs in the list
1651 are linked together using their @TSO_LINK@ field.
1653 When the blocking queue is non-@NULL@, the black hole must be added to
1654 the mutables list since the TSOs on the list may contain pointers into
1655 the new generation. There is no need to clutter up the mutables list
1656 with black holes with empty blocking queues.
1661 \subsubsection{FetchMes}\label{sect:FETCHME}
1663 In the parallel systems, FetchMes are used to represent pointers into
1664 the global heap. When evaluated, the value they point to is read from
1667 \ToDo{Describe layout}
1670 \subsection{Unpointed Objects}
1672 A variable of unpointed type is always bound to a {\em value}, never to a {\em thunk}.
1673 For this reason, unpointed objects cannot be entered.
1675 A {\em value} may be:
1677 \item {\em Boxed}, i.e.~represented indirectly by a pointer to a heap object (e.g.~foreign objects, arrays); or
1678 \item {\em Unboxed}, i.e.~represented directly by a bit-pattern in one or more registers (e.g.~@Int#@ and @Float#@).
1680 All {\em pointed} values are {\em boxed}.
1682 \subsubsection{Immutable Objects}
1683 \label{sect:ARR_WORDS1}
1684 \label{sect:ARR_PTRS}
1687 \item[@ARR_WORDS@] is a variable-sized object consisting solely of
1688 non-pointers. It is used for arrays of all
1689 sorts of things (bytes, words, floats, doubles... it doesn't matter).
1691 \begin{tabular}{|c|c|c|c|}
1693 {\em Fixed Hdr} & {\em No of non-pointers} & {\em Non-pointers\ldots} \\ \hline
1697 \item[@ARR_PTRS@] is an immutable, variable sized array of pointers.
1699 \begin{tabular}{|c|c|c|c|}
1701 {\em Fixed Hdr} & {\em Mutable link} & {\em No of pointers} & {\em Pointers\ldots} \\ \hline
1704 The mutable link is present so that we can easily freeze and thaw an
1705 array (by changing the header and adding/removing the array to the
1710 \subsubsection{Mutable Objects}
1711 \label{sect:mutables}
1712 \label{sect:ARR_WORDS2}
1714 \label{sect:MUTARR_PTRS}
1715 \label{sect:MUTARR_PTRS_FROZEN}
1717 Some of these objects are {\em mutable}; they represent objects which
1718 are explicitly mutated by Haskell code through the @ST@ monad.
1719 They're not used for thunks which are updated precisely once.
1720 Depending on the garbage collector, mutable closures may contain extra
1721 header information which allows a generational collector to implement
1722 the ``write barrier.''
1726 \item[@ARR_WORDS@] is also used to represent {\em mutable} arrays of
1727 bytes, words, floats, doubles, etc. It's possible to use the same
1728 object type because even generational collectors don't need to
1731 \item[@MUTVAR@] is a mutable variable.
1733 \begin{tabular}{|c|c|c|}
1735 {\em Fixed Hdr} & {\em Mutable link} & {\em Pointer} \\ \hline
1739 \item[@MUTARR_PTRS@] is a mutable array of pointers.
1740 Such an array may be {\em frozen}, becoming an @SM_MUTARR_PTRS_FROZEN@, with a
1741 different info-table.
1743 \begin{tabular}{|c|c|c|c|}
1745 {\em Fixed Hdr} & {\em Mutable link} & {\em No of ptrs} & {\em Pointers\ldots} \\ \hline
1749 \item[@MUTARR_PTRS_FROZEN@] is a frozen @MUTARR_PTRS@ closure.
1750 The garbage collector converts @MUTARR_PTRS_FROZEN@ to @ARR_PTRS@ as it removes them from
1756 \subsubsection{Foreign Objects}\label{sect:FOREIGN}
1758 Here's what a ForeignObj looks like:
1761 \begin{tabular}{|l|l|l|l|}
1763 {\em Fixed header} & {\em Data} & {\em Free Routine} & {\em Foreign object link} \\
1768 The @FreeRoutine@ is a reference to the finalisation routine to call
1769 when the @ForeignObj@ becomes garbage. If @freeForeignObject@ is
1770 called on a Foreign Object, the @FreeRoutine@ is set to zero and the
1771 garbage collector will not attempt to call @FreeRoutine@ when the
1772 object becomes garbage.
1774 The Foreign object link is a link to the next foreign object in the
1775 list. This list is traversed at the end of garbage collection: if an
1776 object is about to be deallocated (e.g.~it was not marked or
1777 evacuated), the free routine is called and the object is deleted from
1781 The remaining objects types are all administrative --- none of them may be entered.
1783 \subsection{Thread State Objects (TSOs)}\label{sect:TSO}
1785 In the multi-threaded system, the state of a suspended thread is
1786 packed up into a Thread State Object (TSO) which contains all the
1787 information needed to restart the thread and for the garbage collector
1788 to find all reachable objects. When a thread is running, it may be
1789 ``unpacked'' into machine registers and various other memory locations
1790 to provide faster access.
1792 Single-threaded systems don't really {\em need\/} TSOs --- but they do
1793 need some way to tell the storage manager about live roots so it is
1794 convenient to use a single TSO to store the mutator state even in
1795 single-threaded systems.
1797 Rather than manage TSOs' alloc/dealloc, etc., in some {\em ad hoc}
1798 way, we instead alloc/dealloc/etc them in the heap; then we can use
1799 all the standard garbage-collection/fetching/flushing/etc machinery on
1800 them. So that's why TSOs are ``heap objects,'' albeit very special
1803 \begin{tabular}{|l|l|}
1804 \hline {\em Fixed header}
1805 \\ \hline @TSO_LINK@
1806 \\ \hline @TSO_WHATNEXT@
1807 \\ \hline @TSO_WHATNEXT_INFO@
1808 \\ \hline @TSO_STACK@
1809 \\ \hline {\em Exception Handlers}
1810 \\ \hline {\em Ticky Info}
1811 \\ \hline {\em Profiling Info}
1812 \\ \hline {\em Parallel Info}
1813 \\ \hline {\em GranSim Info}
1817 The contents of a TSO are:
1820 \item A pointer (@TSO_LINK@) used to maintain a list of threads with a similar
1821 state (e.g.~all runable, all sleeping, all blocked on the same black
1822 hole, all blocked on the same MVar, etc.)
1824 \item A word (@TSO_WHATNEXT@) which is in suspended threads to record
1825 how to awaken it. This typically requires a program counter which is stored
1826 in the pointer @TSO_WHATNEXT_INFO@
1828 \item A pointer (@TSO_STACK@) to the top stack block.
1830 \item Optional information for ``Ticky Ticky'' statistics: @TSO_STK_HWM@ is
1831 the maximum number of words allocated to this thread.
1833 \item Optional information for profiling:
1834 @TSO_CCC@ is the current cost centre.
1836 \item Optional information for parallel execution:
1839 \item The types of threads (@TSO_TYPE@):
1841 \item[@T_MAIN@] Must be executed locally.
1842 \item[@T_REQUIRED@] A required thread -- may be exported.
1843 \item[@T_ADVISORY@] An advisory thread -- may be exported.
1844 \item[@T_FAIL@] A failure thread -- may be exported.
1847 \item I've no idea what else
1851 \item Optional information for GranSim execution:
1868 \item clock (gransim light only)
1872 Here are the various queues for GrAnSim-type events.
1883 \subsection{Other weird objects}
1885 \label{sect:BLOCKED_FETCH}
1888 \item[@BlockedFetch@ heap objects (`closures')] (parallel only)
1890 @BlockedFetch@s are inbound fetch messages blocked on local closures.
1891 They arise as entries in a local blocking queue when a fetch has been
1892 received for a local black hole. When awakened, we look at their
1893 contents to figure out where to send a resume.
1895 A @BlockedFetch@ closure has the form:
1897 \begin{tabular}{|l|l|l|l|l|l|}\hline
1898 {\em Fixed header} & link & node & gtid & slot & weight \\ \hline
1902 \item[Spark Closures] (parallel only)
1904 Spark closures are used to link together all closures in the spark pool. When
1905 the current processor is idle, it may choose to speculatively evaluate some of
1906 the closures in the pool. It may also choose to delete sparks from the pool.
1908 \begin{tabular}{|l|l|l|l|l|l|}\hline
1909 {\em Fixed header} & {\em Spark pool link} & {\em Sparked closure} \\ \hline
1917 \subsection{Stack Objects}
1918 \label{sect:STACK_OBJECT}
1921 These are ``stack objects,'' which are used in the threaded world as
1922 the stack for each thread is allocated from the heap in smallish
1923 chunks. (The stack in the sequential world is allocated outside of
1926 Each reduction thread has to have its own stack space. As there may
1927 be many such threads, and as any given one may need quite a big stack,
1928 a naive give-'em-a-big-stack-and-let-'em-run approach will cost a {\em
1931 Our approach is to give a thread a small stack space, and then link
1932 on/off extra ``chunks'' as the need arises. Again, this is a
1933 storage-management problem, and, yet again, we choose to graft the
1934 whole business onto the existing heap-management machinery. So stack
1935 objects will live in the heap, be garbage collected, etc., etc..
1937 A stack object is laid out like this:
1940 \begin{tabular}{|l|}
1944 {\em Link to next stack object (0 for last)}
1946 {\em N, the payload size in words}
1948 {\em @Sp@ (byte offset from head of object)}
1950 {\em @Su@ (byte offset from head of object)}
1952 {\em Payload (N words)}
1957 \ToDo{Are stack objects on the mutable list?}
1959 The stack grows downwards, towards decreasing
1960 addresses. This makes it easier to print out the stack
1961 when debugging, and it means that a return address is
1962 at the lowest address of the chunk of stack it ``knows about''
1963 just like an info pointer on a closure.
1965 The garbage collector needs to be able to find all the
1966 pointers in a stack. How does it do this?
1970 \item Within the stack there are return addresses, pushed
1971 by @case@ expressions. Below a return address (i.e. at higher
1972 memory addresses, since the stack grows downwards) is a chunk
1973 of stack that the return address ``knows about'', namely the
1974 activation record of the currently running function.
1976 \item Below each such activation record is a {\em pending-argument
1977 section}, a chunk of
1978 zero or more words that are the arguments to which the result
1979 of the function should be applied. The return address does not
1981 ``know'' how many pending arguments there are, or their types.
1982 (For example, the function might return a result of type $\alpha$.)
1984 \item Below each pending-argument section is another return address,
1985 and so on. Actually, there might be an update frame instead, but we
1986 can consider update frames as a special case of a return address with
1987 a well-defined activation record.
1989 \ToDo{Doesn't it {\em have} to be an update frame? After all, the arg
1990 satisfaction check is @Su - Sp >= ...@.}
1994 The game plan is this. The garbage collector
1995 walks the stack from the top, traversing pending-argument sections and
1996 activation records alternately. Next we discuss how it finds
1997 the pointers in each of these two stack regions.
2000 \subsubsection{Activation records}\label{sect:activation-records}
2002 An {\em activation record} is a contiguous chunk of stack,
2003 with a return address as its first word, followed by as many
2004 data words as the return address ``knows about''. The return
2005 address is actually a fully-fledged info pointer. It points
2006 to an info table, replete with:
2009 \item entry code (i.e. the code to return to).
2010 \item @INFO_TYPE@ is either @RET_SMALL/RET_VEC_SMALL@ or @RET_BIG/RET_VEC_BIG@, depending
2011 on whether the activation record has more than 32 data words (\note{64 for 8-byte-word architectures}) and on whether
2012 to use a direct or a vectored return.
2013 \item @INFO_SM@ for @RET_SMALL@ is a bitmap telling the layout
2014 of the activation record, one bit per word. The least-significant bit
2015 describes the first data word of the record (adjacent to the fixed
2016 header) and so on. A ``@1@'' indicates a non-pointer, a ``@0@''
2018 a pointer. We don't need to indicate exactly how many words there
2020 because when we get to all zeros we can treat the rest of the
2021 activation record as part of the next pending-argument region.
2023 For @RET_BIG@ the @INFO_SM@ field points to a block of bitmap
2024 words, starting with a word that tells how many words are in
2027 \item @INFO_SRT@ is the Static Reference Table for the return
2028 address (Section~\ref{sect:srt}).
2031 The activation record is a fully fledged closure too.
2032 As well as an info pointer, it has all the other attributes of
2033 a fixed header (Section~\ref{sect:fixed-header}) including a saved cost
2034 centre which is reloaded when the return address is entered.
2036 In other words, all the attributes of closures are needed for
2037 activation records, so it's very convenient to make them look alike.
2040 \subsubsection{Pending arguments}
2042 So that the garbage collector can correctly identify pointers
2043 in pending-argument sections we explicitly tag all non-pointers.
2044 Every non-pointer in a pending-argument section is preceded
2045 (at the next lower memory word) by a one-word byte count that
2046 says how many bytes to skip over (excluding the tag word).
2048 The garbage collector traverses a pending argument section from
2049 the top (i.e. lowest memory address). It looks at each word in turn:
2052 \item If it is less than or equal to a small constant @MAX_STACK_TAG@
2054 it treats it as a tag heralding zero or more words of non-pointers,
2055 so it just skips over them.
2057 \item If it points to the code segment, it must be a return
2058 address, so we have come to the end of the pending-argument section.
2060 \item Otherwise it must be a bona fide heap pointer.
2064 \subsection{The Stable Pointer Table}\label{sect:STABLEPTR_TABLE}
2066 A stable pointer is a name for a Haskell object which can be passed to
2067 the external world. It is ``stable'' in the sense that the name does
2068 not change when the Haskell garbage collector runs---in contrast to
2069 the address of the object which may well change.
2071 A stable pointer is represented by an index into the
2072 @StablePointerTable@. The Haskell garbage collector treats the
2073 @StablePointerTable@ as a source of roots for GC.
2075 In order to provide efficient access to stable pointers and to be able
2076 to cope with any number of stable pointers (eg $0 \ldots 100000$), the
2077 table of stable pointers is an array stored on the heap and can grow
2078 when it overflows. (Since we cannot compact the table by moving
2079 stable pointers about, it seems unlikely that a half-empty table can
2080 be reduced in size---this could be fixed if necessary by using a
2081 hash table of some sort.)
2083 In general a stable pointer table closure looks like this:
2086 \begin{tabular}{|l|l|l|l|l|l|l|l|l|l|l|}
2088 {\em Fixed header} & {\em No of pointers} & {\em Free} & $SP_0$ & \ldots & $SP_{n-1}$
2096 \item[@NPtrs@:] number of (stable) pointers.
2098 \item[@Free@:] the byte offset (from the first byte of the object) of the first free stable pointer.
2100 \item[$SP_i$:] A stable pointer slot. If this entry is in use, it is
2101 an ``unstable'' pointer to a closure. If this entry is not in use, it
2102 is a byte offset of the next free stable pointer slot.
2106 When a stable pointer table is evacuated
2108 \item the free list entries are all set to @NULL@ so that the evacuation
2109 code knows they're not pointers;
2111 \item The stable pointer slots are scanned linearly: non-@NULL@ slots
2112 are evacuated and @NULL@-values are chained together to form a new free list.
2115 There's no need to link the stable pointer table onto the mutable
2116 list because we always treat it as a root.
2120 \section{The Storage Manager}
2122 The generational collector remembers the depth of the last generation
2123 collected and the value of the heap pointer at the end of the last GC.
2124 If the mutator has not moved the heap pointer, that means that the
2125 amount of space recovered is insufficient to satisfy even one request
2126 and it is time to collect an older generation or report a heap overflow.
2128 A deeper collection is also triggered when a minor collection fails to
2129 recover at least @...@ bytes of space.
2131 When can a GC happen?
2134 - During updates (ie during returns)
2135 - When a heap check fails
2136 - When a stack check fails (concurrent system only)
2137 - When a context switch happens (concurrent system only)
2139 When do heap checks occur?
2140 - Immediately after entering a thunk
2141 - Immediately after entering a case alternative
2143 When do stack checks occur?
2144 - We calculate the worst-case stack usage of an entire
2145 thunk so there's no need to put a check inside each alternative.
2146 - Immediately after entering a thunk
2147 We can't make a similar worst-case calculation for heap usage
2148 because the heap isn't used in a stacklike manner so any
2149 evaluation inside a case might steal some of the heap we've
2153 - Threads can be blocked
2154 - Threads can be put to sleep
2155 - Heap may move while we sleep
2156 - Black holing may happen while we sleep (ie during GC)
2159 \subsection{The SM state}
2161 Contains @Hp@, @HpLim@, @StablePtrTable@ plus version-specific info.
2165 \item Static Object list
2166 \item Foreign Object list
2167 \item Stable Pointer Table
2171 In addition, the generational collector requires:
2175 \item Old Generation Indirection list
2176 \item Mutables list --- list of mutable objects in the old generation.
2177 \item @OldLim@ --- the boundary between the generations
2178 \item Old Foreign Object list --- foreign objects in the old generation
2182 It is passed a table of {\em roots\/} containing
2186 \item All runnable TSOs
2191 In the parallel system, there must be some extra magic associated with
2194 \subsection{The SM interface}
2196 @initSM@ finalizes any runtime parameters of the storage manager.
2198 @exitSM@ does any cleaning up required by the storage manager before
2199 the program is executed. Its main purpose is to print any summary
2202 @initHeap@ allocates the heap. It initialises the @hp@ and @hplim@
2203 fields of @sm@ to represent an empty heap for the compiled-in garbage
2204 collector. It also initialises @CAFlist@ to be the empty list. If we
2205 are using Appel's collector it also initialises the @OldLim@ field.
2206 It also initialises the stable pointer table and the @ForeignObjList@
2207 (and @OldForeignObjList@) fields.
2209 @collectHeap@ invokes the garbage collector. @collectHeap@ requires
2210 all the fields of @sm@ to be initialised appropriately (from the
2211 STG-machine registers). The following are identified as heap roots:
2213 \item The updated CAFs recorded in @CAFlist@.
2214 \item Any pointers found on the stack.
2215 \item All runnable and sleeping TSOs.
2216 \item The stable pointer table.
2219 There are two possible results from a garbage collection:
2222 The garbage collector is unable to free up any more space.
2225 The garbage collector managed to free up more space.
2228 \item @hp@ and @hplim@ will indicate the new space available for
2231 \item The elements of @CAFlist@ and the stable pointers will be
2232 updated to point to the new locations of the closures they reference.
2234 \item Any members of @ForeignObjList@ which became garbage should have
2235 been reported (by calling their finalising routines; and the
2236 @(Old)ForeignObjList@ updated to contain only those Foreign objects
2237 which are still live.
2244 %************************************************************************
2246 \subsection{``What really happens in a garbage collection?''}
2248 %************************************************************************
2250 This is a brief tutorial on ``what really happens'' going to/from the
2251 storage manager in a garbage collection.
2254 %------------------------------------------------------------------------
2255 \item[The heap check:]
2259 If you gaze into the C output of GHC, you see many macros calls like:
2261 HEAP_CHK_2PtrsLive((_FHS+2));
2264 This expands into the C (roughly speaking...):
2266 Hp = Hp + (_FHS+2); /* optimistically move heap pointer forward */
2268 GC_WHILE_OR_IF (HEAP_OVERFLOW_OP(Hp, HpLim) OR_INTERVAL_EXPIRED) {
2269 STGCALL2_GC(PerformGC, <liveness-bits>, (_FHS+2));
2273 In the parallel world, where we will need to re-try the heap check,
2274 @GC_WHILE_OR_IF@ will be a ``while''; in the sequential world, it will
2277 The ``heap lookahead'' checks, which are similar and used for
2278 multi-precision @Integer@ ops, have some further complications. See
2279 the commentary there (@StgMacros.lh@).
2281 %------------------------------------------------------------------------
2282 \item[Into @callWrapper_GC@...:]
2284 When we failed the heap check (above), we were inside the
2285 GCC-registerised ``threaded world.'' @callWrapper_GC@ is all about
2286 getting in and out of the threaded world. On SPARCs, with register
2287 windows, the name of the game is not shifting windows until we have
2288 what we want out of the old one. In tricky cases like this, it's best
2289 written in assembly language.
2291 Performing a GC (potentially) means giving up the thread of control.
2292 So we must fill in the thread-state-object (TSO) [and its associated
2293 stk object] with enough information for later resumption:
2296 Save the return address in the TSO's PC field.
2298 Save the machine registers used in the STG threaded world in their
2299 corresponding TSO fields. We also save the pointer-liveness
2300 information in the TSO.
2302 The registers that are not thread-specific, notably @Hp@ and
2303 @HpLim@, are saved in the @StorageMgrInfo@ structure.
2305 Call the routine it was asked to call; in this example, call
2306 @PerformGC@ with arguments @<liveness>@ and @_FHS+2@ (some constant)...
2310 %------------------------------------------------------------------------
2311 \item[Into the heap overflow wrapper, @PerformGC@ [parallel]:]
2313 Most information has already been saved in the TSO.
2317 The first argument (@<liveness>@, in our example) say what registers
2318 are live, i.e., are ``roots'' the storage manager needs to know.
2320 StorageMgrInfo.rootno = 2;
2321 StorageMgrInfo.roots[0] = (P_) Ret1_SAVE;
2322 StorageMgrInfo.roots[1] = (P_) Ret2_SAVE;
2326 We move the heap-pointer back [we had optimistically
2327 advanced it, in the initial heap check]
2330 We load up the @smInfo@ data from the STG registers' @*_SAVE@ locations.
2333 We mark on the scheduler's big ``blackboard'' that a GC is
2337 We reschedule, i.e., this thread gives up control. (The scheduler
2338 will presumably initiate a garbage-collection, but it may have to do
2339 any number of other things---flushing, for example---before ``normal
2340 execution'' resumes; and it most certainly may not be this thread that
2341 resumes at that point!)
2344 IT IS AT THIS POINT THAT THE WORLD IS COMPLETELY TIDY.
2346 %------------------------------------------------------------------------
2347 \item[Out of @callWrapper_GC@ [parallel]:]
2349 When this thread is finally resumed after GC (and who knows what
2350 else), it will restart by the normal enter-TSO/enter-stack-object
2351 sequence, which has the effect of re-loading the registers, etc.,
2352 (i.e., restoring the state).
2354 Because the address we saved in the TSO's PC field was that at the end
2355 of the heap check, and because the check is a while-loop in the
2356 parallel system, we will now loop back around, and make sure there is
2357 enough space before continuing.
2362 \subsection{Static Reference Tables (SRTs)}
2365 \label{sect:static-objects}
2367 In the above, we assumed that objects always contained pointers to all
2368 their free variables. In fact, this isn't quite true: GHC omits
2369 pointers to top-level objects and allocates their closures in static
2370 memory. This optimisation reduces the number of free variables in
2371 heap objects - reducing memory usage and the effort needed to put them
2372 into heap objects. However, this optimisation comes at a cost: we
2373 need to complicate the garbage collector with machinery for tracing
2374 these static references.
2376 Early versions of GHC used a very simple algorithm: it treated all
2377 static objects as roots. This is safe in the sense that no object is
2378 ever deallocated if there's a chance that it might be required later
2379 but can lead to some terrible space leaks. For example, this program
2380 ought to be able to run in constant space but, because @xs@ is never
2381 deallocated, it runs in linear space.
2388 The correct behaviour is for the garbage collector to keep a static
2389 object alive iff it might be required later in execution. That is, if
2390 it is reachable from any live heap objects {\em or\/} from any return
2391 addresses found on the stack or from the current program counter.
2392 Since it is obviously infeasible for the garbage collector to scan
2393 machine code looking for static references, the code generator must
2394 generate a table of all static references in any piece of code (and we
2395 must place a pointer to this table next to any piece of code we
2398 Here's what the SRT has to contain:
2404 Here's how we represent it:
2408 must be able to handle 0 references well
2415 o Generational GC trickery
2418 \subsection{Space leaks and black holes}
2419 \label{sect:black-hole}
2423 \ToDo{Insert text stolen from update paper}
2427 A program exhibits a {\em space leak} if it retains storage that is
2428 sure not to be used again. Space leaks are becoming increasingly
2429 common in imperative programs that @malloc@ storage and fail
2430 subsequently to @free@ it. They are, however, also common in
2431 garbage-collected systems, especially where lazy evaluation is
2432 used.[.wadler leak, runciman heap profiling jfp.]
2434 Quite a bit of experience has now accumulated suggesting that
2435 implementors must be very conscientious about avoiding gratuitous
2436 space leaks --- that is, ones which are an accidental artefact of some
2437 implementation technique.[.appel book.] The update mechanism is
2438 a case in point, as <.jones jfp leak.> points out. Consider a thunk for
2441 let xs = [1..1000] in last xs
2443 where @last@ is a function that returns the last element of its
2444 argument list. When the thunk is entered it will call @last@, which
2445 will consume @xs@ until it finds the last element. Since the list
2446 @[1..1000]@ is produced lazily one might reasonably expect the
2447 expression to evaluate in constant space. But {\em until the moment
2448 of update, the thunk itself still retains a pointer to the beginning
2449 of the list @xs@}. So, until the update takes place the whole list
2452 Of course, this is completely gratuitous. The pointer to @xs@ in the
2453 thunk will never be used again. In <.peyton stock hardware.> the solution to
2454 this problem that we advocated was to overwrite a thunk's info with a
2455 fixed ``black hole'' info pointer, {\em at the moment of entry}. The
2456 storage management information attached to a black-hole info pointer
2457 tells the garbage collector that the closure contains no pointers,
2458 thereby plugging the space leak.
2460 \subsubsection{Lazy black-holing}
2462 Black-holing is a well-known idea. The trouble is that it is
2463 gallingly expensive. To avoid the occasional space leak, for every
2464 single thunk entry we have to load a full-word literal constant into a
2465 register (often two instructions) and then store that register into a
2468 Fortunately, this cost can easily be avoided. The
2469 idea is simple: {\em instead of black-holing every thunk on entry,
2470 wait until the garbage collector is called, and then black-hole all
2471 (and only) the thunks whose evaluation is in progress at that moment}.
2472 There is no benefit in black-holing a thunk that is updated before
2473 garbage collection strikes! In effect, the idea is to perform the
2474 black-holing operation lazily, only when it is needed. This
2475 dramatically cuts down the number of black-holing operations, as our
2476 results show {\em forward ref}.
2478 How can we find all the thunks whose evaluation is in progress? They
2479 are precisely the ones for which update frames are on the stack. So
2480 all we need do is find all the update frames (via the @Su@ chain) and
2481 black-hole their thunks right at the start of garbage collection.
2482 Notice that it is not enough to refrain from treating update frames as
2483 roots: firstly because the thunks to which they point may need to be
2484 moved in a copying collector, but more importantly because the thunk
2485 might be accessible via some other route.
2487 \subsubsection{Detecting loops}
2489 Black-holing has a second minor advantage: evaluation of a thunk whose
2490 value depends on itself will cause a black hole closure to be entered,
2491 which can cause a suitable error message to be displayed. For example,
2492 consider the definition
2496 The code to evaluate @x@'s right hand side will evaluate @x@. In the
2497 absence of black-holing, the result will be a stack overflow, as the
2498 evaluator repeatedly pushes a return address and enters @x@. If
2499 thunks are black-holed on entry, then this infinite loop can be caught
2502 With our new method of lazy black-holing, a self-referential program
2503 might cause either stack overflow or a black-hole error message,
2504 depending on exactly when garbage collection strikes. It is quite
2505 easy to conceal these differences, however. If stack overflow occurs,
2506 all we need do is examine the update frames on the stack to see if
2507 more than one refers to the same thunk. If so, there is a loop that
2508 would have been detected by eager black-holing.
2510 \subsubsection{Lazy locking}
2513 In a parallel implementation, it is necessary somehow to ``lock'' a
2514 thunk that is under evaluation, so that other parallel evaluators
2515 cannot simultaneously evaluate it and thereby duplicate work.
2516 Instead, an evaluator that enters a locked thunk should be blocked,
2517 and made runnable again when the thunk is updated.
2519 This locking is readily arranged in the same way as black-holing, by
2520 overwriting the thunk's info pointer with a special ``locked'' info
2521 pointer, at the moment of entry. If another evaluator enters the
2522 thunk before it has been updated, it will land in the entry code for
2523 the ``locked'' info pointer, which blocks the evaluator and queues it
2524 on the locked thunk.
2526 The details are given by <.portable parallel trinder.>. However, the close similarity
2527 between locking and black holing suggests the following question: can
2528 locking be done lazily too? The answer is that it can, except that
2529 locking can be postponed only until the next {\em context switch},
2530 rather than the next {\em garbage collection}. We are assuming here
2531 that the parallel implementation does not use shared memory to allow
2532 two processors to access the same closure. If such access is
2533 permitted then every thunk entry requires a hardware lock, and becomes
2536 Is lazy locking worth while, given that it requires extra work every
2537 context switch? We believe it is, because contexts switches are
2538 relatively infrequent, and thousands of thunk-entries typically take
2541 {\em Measurements elsewhere. Omit this section? If so, fix cross refs to here.}
2546 \subsection{Squeezing identical updates}
2550 \ToDo{Insert text stolen from update paper}
2554 Consider the following Haskell definition of the standard
2555 function @partition@ that divides a list into two, those elements
2556 that satisfy a predicate @p@ and those that do not:
2558 partition :: (a->Bool) -> [a] -> ([a],[a])
2559 partition p [] = ([],[])
2560 partition p (x:xs) = if p x then (x:ys, zs)
2563 (ys,zs) = partition p xs
2565 By the time this definition has been desugared, it looks like this:
2575 if p x then (x:ys,zs)
2578 Lazy evaluation demands that the recursive call is bound to an
2579 intermediate variable, @t@, from which @ys@ and @zs@ are lazily
2580 selected. (The functions @fst@ and @snd@ select the first and second
2581 elements of a pair, respectively.)
2583 Now, suppose that @partition@ is applied to a list @[x1,x2]@,
2585 elements satisfy @p@. We can get a good idea of what will happen
2586 at runtime by unrolling the recursion a few times in our heads.
2587 Unrolling once, and remembering that @(p x1)@ is @True@, we get this:
2591 let t1 = partition [x2]
2596 Unrolling the rest of the way gives this:
2608 Now consider what happens if @zs1@ is evaluated. It is bound to a
2609 thunk, which will push an update frame before evaluating the
2610 expression @snd t1@. This expression in turn forces evaluation of
2611 @zs2@, which pushes an update frame before evaluating @snd t2@.
2612 Indeed the stack of update frames will grow as deep as the list is
2613 long when @zs1@ is evaluated. This is rather galling, since all the
2614 thunks @zs1@, @zs2@, and so on, have the same value.
2616 \ToDo{Describe the state-transformer case in which we get a space leak from
2617 pending update frames.}
2619 The solution is simple. The garbage collector, which is going to traverse the
2620 update stack in any case, can easily identify two update frames that are directly
2621 on top of each other. The second of these will update its target with the same
2622 value as the first. Therefore, the garbage collector can perform the update
2623 right away, by overwriting one update target with an indirection to the second,
2624 and eliminate the corresponding update frame. In this way ever-growing stacks of
2625 update frames are reduced to a single representative at garbage collection time.
2626 If this is done at the start of garbage collection then, if it turns out that
2627 some of these update targets are garbage they will be collected right away.
2631 \subsection{Space leaks and selectors}
2635 \ToDo{Insert text stolen from update paper}
2639 In 1987, Wadler identified an important source of space leaks in
2640 lazy functional programs. Consider the Haskell function definition:
2642 f p = (g1 a, g2 b) where (a,b) = p
2644 The pattern-matching in the @where@ clause is known as
2645 {\em lazy pattern-matching}, because it is performed only if @a@
2646 or @b@ is actually evaluated. The desugarer translates lazy pattern matching
2647 to the use of selectors, @fst@ and @snd@ in this case:
2654 Now suppose that the second component of the pair @(f p)@, namely @a@,
2655 is evaluated and discarded, but the first is not although it remains
2656 reachable. The garbage collector will find that the thunk for @b@ refers
2657 to @p@ and hence to @a@. Thus, although @a@ cannot ever be used again, its
2658 space is retained. It turns out that this space leak can have a very bad effect
2659 indeed on a program's space behaviour (Section~\ref{sect:selector-results}).
2661 Wadler's paper also proposed a solution: if the garbage collector
2662 encounters a thunk of the form @snd p@, where @p@ is evaluated, then
2663 the garbage collector should perform the selection and overwrite the
2664 thunk with a pointer to the second component of the pair. In effect, the
2665 garbage collector thereby performs a bounded amount of as-yet-undemanded evaluation
2666 in the hope of improving space behaviour.
2667 We implement this idea directly, by making the garbage collector
2668 eagerly execute all selector thunks\footnote{A word of caution: it is rather easy
2669 to make a mistake in the implementation, especially if the garbage collector
2670 uses pointer reversal to traverse the reachable graph.},
2672 reported in Section~\ref{sect:THUNK_SEL}.
2674 One could easily imagine generalisations of this idea, with the garbage
2675 collector performing bounded amounts of space-saving work. One example is
2679 f x (y:ys) = f (x+1) ys
2681 Most lazy evaluators will build up a chain of thunks for the accumulating
2682 parameter, @x@, each of which increments @x@. It is not safe to evaluate
2683 any of these thunks eagerly, since @f@ is not strict in @x@, and we know nothing
2684 about the value of @x@ passed in the initial call to @f@.
2685 On the other hand, if the garbage collector found a thunk @(x+1)@ where
2686 @x@ happened to be evaluated, then it could ``execute'' it eagerly.
2687 If done carefully, the entire chain could be eliminated in a single
2688 garbage collection. We have not (yet) implemented this idea.
2689 A very similar idea, dubbed ``stingy evaluation'', is described
2692 <.sparud lazy pattern matching.> describes another solution to the
2693 lazy-pattern-matching
2694 problem. His solution involves adding code to the two thunks for
2695 @a@ and @b@ so that if either is evaluated it arranges to update the
2696 other as well as itself. The garbage-collector solution is a little
2697 more general, since it applies whether or not the selectors were
2698 generated by lazy pattern matching, and in our setting it was easier
2699 to implement than Sparud's.
2704 \subsection{Internal workings of the Compacting Collector}
2706 \subsection{Internal workings of the Copying Collector}
2708 \subsection{Internal workings of the Generational Collector}
2714 Registering costs centres looks awkward - can we tidy it up?
2716 \section{Parallelism}
2718 Something about global GC, inter-process messages and fetchmes.
2722 \section{Ticky Ticky profiling}
2724 Measure what proportion of ...:
2727 ... Enters are to data values, function values, thunks.
2729 ... allocations are for data values, functions values, thunks.
2731 ... updates are for data values, function values.
2735 ... return-in-heap (dynamic)
2737 ... vectored return (dynamic)
2739 ... updates are wasted (never re-entered).
2741 ... constructor returns get away without hitting an update.
2744 %************************************************************************
2746 \subsubsection[ticky-stk-heap-use]{Stack and heap usage}
2748 %************************************************************************
2750 Things we are interested in here:
2753 How many times we do a heap check and move @Hp@; comparing this with
2754 the allocations gives an indication of how many things we get per trip
2757 If we do a ``heap lookahead,'' we haven't really allocated any
2758 heap, so we need to undo the effects of an @ALLOC_HEAP@:
2761 The stack high-water mark.
2764 Re-use of stack slots, and stubbing of stack slots:
2768 %************************************************************************
2770 \subsubsection[ticky-allocs]{Allocations}
2772 %************************************************************************
2774 We count things every time we allocate something in the dynamic heap.
2775 For each, we count the number of words of (1)~``admin'' (header),
2776 (2)~good stuff (useful pointers and data), and (3)~``slop'' (extra
2777 space, in hopes it will allow an in-place update).
2779 The first five macros are inserted when the compiler generates code
2780 to allocate something; the categories correspond to the @ClosureClass@
2781 datatype (manifest functions, thunks, constructors, big tuples, and
2782 partial applications).
2784 We may also allocate space when we do an update, and there isn't
2785 enough space. These macros suffice (for: updating with a partial
2786 application and a constructor):
2788 In the threaded world, we allocate space for the spark pool, stack objects,
2789 and thread state objects.
2791 The histogrammy bit is fairly straightforward; the @-2@ is: one for
2792 0-origin C arrays; the other one because we do {\em no} one-word
2793 allocations, so we would never inc that histogram slot; so we shift
2794 everything over by one.
2796 Some hard-to-account-for words are allocated by/for primitives,
2797 includes Integer support. @ALLOC_PRIM2@ tells us about these. We
2798 count everything as ``goods'', which is not strictly correct.
2799 (@ALLOC_PRIM@ is the same sort of stuff, but we know the
2800 admin/goods/slop breakdown.)
2802 %************************************************************************
2804 \subsubsection[ticky-enters]{Enters}
2806 %************************************************************************
2808 We do more magical things with @ENT_FUN_DIRECT@. Besides simply knowing
2809 how many ``fast-entry-point'' enters there were, we'd like {\em simple}
2810 information about where those enters were, and the properties thereof.
2812 struct ent_counter {
2813 unsigned registeredp:16, /* 0 == no, 1 == yes */
2814 arity:16, /* arity (static info) */
2815 Astk_args:16, /* # of args off A stack */
2816 Bstk_args:16; /* # of args off B stack */
2817 /* (rest of args are in registers) */
2818 StgChar *f_str; /* name of the thing */
2819 StgChar *f_arg_kinds; /* info about the args types */
2820 StgChar *wrap_str; /* name of its wrapper (if any) */
2821 StgChar *wrap_arg_kinds;/* info about the orig wrapper's arg types */
2822 I_ ctr; /* the actual counter */
2823 struct ent_counter *link; /* link to chain them all together */
2827 %************************************************************************
2829 \subsubsection[ticky-returns]{Returns}
2831 %************************************************************************
2833 Whenever a ``return'' occurs, it is returning the constituent parts of
2834 a data constructor. The parts can be returned either in registers, or
2835 by allocating some heap to put it in (the @ALLOC_*@ macros account for
2836 the allocation). The constructor can either be an existing one
2837 (@*OLD*@) or we could have {\em just} figured out this stuff
2840 Here's some special magic that Simon wants [edited to match names
2844 From: Simon L Peyton Jones <simonpj>
2845 To: partain, simonpj
2846 Subject: counting updates
2847 Date: Wed, 25 Mar 92 08:39:48 +0000
2849 I'd like to count how many times we update in place when actually Node
2850 points to the thing. Here's how:
2852 @RET_OLD_IN_REGS@ sets the variable @ReturnInRegsNodeValid@ to @True@;
2853 @RET_NEW_IN_REGS@ sets it to @False@.
2855 @RET_SEMI_???@ sets it to??? ToDo [WDP]
2857 @UPD_CON_IN_PLACE@ tests the variable, and increments @UPD_IN_PLACE_COPY_ctr@
2860 Then we need to report it along with the update-in-place info.
2864 Of all the returns (sum of four categories above), how many were
2865 vectored? (The rest were obviously unvectored).
2867 %************************************************************************
2869 \subsubsection[ticky-update-frames]{Update frames}
2871 %************************************************************************
2873 These macros count up the following update information.
2875 \begin{tabular}{|l|l|} \hline
2876 Macro & Counts \\ \hline
2878 @UPDF_STD_PUSHED@ & Update frame pushed \\
2879 @UPDF_CON_PUSHED@ & Constructor update frame pushed \\
2880 @UPDF_HOLE_PUSHED@ & An update frame to update a black hole \\
2881 @UPDF_OMITTED@ & A thunk decided not to push an update frame \\
2882 & (all subsets of @ENT_THK@) \\
2883 @UPDF_RCC_PUSHED@ & Cost Centre restore frame pushed \\
2884 @UPDF_RCC_OMITTED@ & Cost Centres not required -- not pushed \\\hline
2887 %************************************************************************
2889 \subsubsection[ticky-updates]{Updates}
2891 %************************************************************************
2893 These macros record information when we do an update. We always
2894 update either with a data constructor (CON) or a partial application
2897 \begin{tabular}{|l|l|}\hline
2898 Macro & Where \\ \hline
2900 @UPD_EXISTING@ & Updating with an indirection to something \\
2901 & already in the heap \\
2902 @UPD_SQUEEZED@ & Same as @UPD_EXISTING@ but because \\
2903 & of stack-squeezing \\
2904 @UPD_CON_W_NODE@ & Updating with a CON: by indirecting to Node \\
2905 @UPD_CON_IN_PLACE@ & Ditto, but in place \\
2906 @UPD_CON_IN_NEW@ & Ditto, but allocating the object \\
2907 @UPD_PAP_IN_PLACE@ & Same, but updating w/ a PAP \\
2908 @UPD_PAP_IN_NEW@ & \\\hline
2911 %************************************************************************
2913 \subsubsection[ticky-selectors]{Doing selectors at GC time}
2915 %************************************************************************
2917 @GC_SEL_ABANDONED@: we could've done the selection, but we gave up
2918 (e.g., to avoid overflowing the C stack); @GC_SEL_MINOR@: did a
2919 selection in a minor GC; @GC_SEL_MAJOR@: ditto, but major GC.
2925 We're nuking the following:
2932 Return in registers.
2933 This lets us remove update code pointers from info tables,
2934 removes the need for phantom info tables, simplifies
2939 Careful analysis suggests that it doesn't buy us very much
2940 and it is hard to work with.
2942 Eliminating threaded GCs eliminates the desire to share SMReps
2943 so they are (once more) part of the Info table.
2947 Doesn't buy us anything on a register-poor architecture and
2948 isn't so important if we have semi-tagging.
2951 - Probably bad on register poor architecture
2952 - Can avoid need to write return address to stack on reg rich arch.
2953 - when a function does a small amount of work, doesn't
2954 enter any other thunks and then returns.
2955 eg entering a known constructor (but semitagging will catch this)
2956 - Adds complications
2962 This lets us drop CONST closures and CHARLIKE closures (assuming we
2963 don't support Unicode). The only point of these closures was to
2964 avoid updating with an indirection.
2966 We also drop @MIN_UPD_SIZE@ --- all we need is space to insert an
2967 indirection or a black hole.
2970 STATIC SMReps are now called CONST
2975 \item The profiling ``kind'' field is now encoded in the @INFO_TYPE@ field.
2976 This identifies the general sort of the closure for profiling purposes.
2978 \item Various papers describe deleting update frames for unreachable objects.
2979 This has never been implemented and we don't plan to anytime soon.
2983 \section{Old tricks}
2987 These are statically defined closures which have been updated with a
2988 heap-allocated result. Initially these are exactly the same as a
2989 @STATIC@ closure but with special entry code. On entering the closure
2990 the entry code must:
2993 \item Allocate a black hole in the heap which will be updated with
2995 \item Overwrite the static closure with a special @CAF@ indirection.
2997 \item Link the static indirection onto the list of updated @CAF@s.
3000 The indirection and the link field require the initial @STATIC@
3001 closure to be of at least size @MIN_UPD_SIZE@ (excluding the fixed
3004 @CAF@s are treated as special garbage collection roots. These roots
3005 are explicitly collected by the garbage collector, since they may
3006 appear in code even if they are not linked with the main heap. They
3007 consequently represent potentially enormous space-leaks. A @CAF@
3008 closure retains a fixed location in statically allocated data space.
3009 When updated, the contents of the @CAF@ indirection are changed to
3010 reflect the new closure. @CAF@ indirections require special garbage
3013 \section{Old stuff about SRTs}
3015 Garbage collection of @CAF@s is tricky. We have to cope with explicit
3016 collection from the @CAFlist@ as well as potential references from the
3017 stack and heap which will cause the @CAF@ evacuation code to be
3018 called. They are treated like indirections which are shorted out.
3019 However they must also be updated to point to the new location of the
3020 new closure as the @CAF@ may still be used by references which
3023 {\bf Copying Collection}
3025 A first scheme might use evacuation code which evacuates the reference
3026 and updates the indirection. This is no good as subsequent evacuations
3027 will result in an already evacuated closure being evacuated. This will
3028 leave a forward reference in to-space!
3030 An alternative scheme evacuates the @CAFlist@ first. The closures
3031 referenced are evacuated and the @CAF@ indirection updated to point to
3032 the evacuated closure. The @CAF@ evacuation code simply returns the
3033 updated indirection pointer --- the pointer to the evacuated closure.
3034 Unfortunately the closure the @CAF@ references may be a static
3035 closure, in fact, it may be another @CAF@. This will cause the second
3036 @CAF@'s evacuation code to be called before the @CAF@ has been
3037 evacuated, returning an unevacuated pointer.
3039 Another scheme leaves updating the @CAF@ indirections to the end of
3040 the garbage collection. All the references are evacuated and
3041 scavenged as usual (including the @CAFlist@). Once collection is
3042 complete the @CAFlist@ is traversed updating the @CAF@ references with
3043 the result of evacuating the referenced closure again. This will
3044 immediately return as it must be a forward reference, a static
3045 closure, or a @CAF@ which will indirect by evacuating its reference.
3047 The crux of the problem is that the @CAF@ evacuation code needs to
3048 know if its reference has already been evacuated and updated. If not,
3049 then the reference can be evacuated, updated and returned safely
3050 (possibly evacuating another @CAF@). If it has, then the updated
3051 reference can be returned. This can be done using two @CAF@
3052 info-tables. At the start of a collection the @CAFlist@ is traversed
3053 and set to an internal {\em evacuate and update} info-table. During
3054 collection, evacution of such a @CAF@ also results in the info-table
3055 being reset back to the standard @CAF@ info-table. Thus subsequent
3056 evacuations will simply return the updated reference. On completion of
3057 the collection all @CAF@s will have {\em return reference} info-tables
3060 This is the scheme we adopt. A @CAF@ indirection has evacuation code
3061 which returns the evacuated and updated reference. During garbage
3062 collection, all the @CAF@s are overwritten with an internal @CAF@ info
3063 table which has evacuation code which performs this evacuate and
3064 update and restores the original @CAF@ code. At some point during the
3065 collection we must ensure that all the @CAF@s are indeed evacuated.
3067 The only potential problem with this scheme is a cyclic list of @CAF@s
3068 all directly referencing (possibly via indirections) another @CAF@!
3069 Evacuation of the first @CAF@ will fail in an infinite loop of @CAF@
3070 evacuations. This is solved by ensuring that the @CAF@ info-table is
3071 updated to a {\em return reference} info-table before performing the
3072 evacuate and update. If this {\em return reference} evacuation code is
3073 called before the actual evacuation is complete it must be because
3074 such a cycle of references exists. Returning the still unevacuated
3075 reference is OK --- all the @CAF@s will now reference the same
3076 @CAF@ which will reference itself! Construction of such a structure
3077 indicates the program must be in an infinite loop.
3079 {\bf Compacting Collector}
3081 When shorting out a @CAF@, its reference must be marked. A first
3082 attempt might explicitly mark the @CAF@s, updating the reference with
3083 the marked reference (possibly short circuting indirections). The
3084 actual @CAF@ marking code can indicate that they have already been
3085 marked (though this might not have actually been done yet) and return
3086 the indirection pointer so it is shorted out. Unfortunately the @CAF@
3087 reference might point to an indirection which will be subsequently
3088 shorted out. Rather than returning the @CAF@ reference we treat the
3089 @CAF@ as an indirection, calling the mark code of the reference, which
3090 will return the appropriately shorted reference.
3092 Problem: Cyclic list of @CAF@s all directly referencing (possibly via
3093 indirections) another @CAF@!
3095 Before compacting, the locations of the @CAF@ references are
3096 explicitly linked to the closures they reference (if they reference
3097 heap allocated closures) so that the compacting process will update
3098 them to the closure's new location. Unfortunately these locations'
3099 @CAF@ indirections are static. This causes premature termination
3100 since the test to find the info pointer at the end of the location
3101 list will match more than one value. This can be solved by using an
3102 auxiliary dynamic array (on the top of the A stack). One location for
3103 each @CAF@ indirection is linked to the closure that the @CAF@
3104 references. Once collection is complete this array is traversed and
3105 the corresponding @CAF@ is then updated with the updated pointer from
3106 the auxiliary array.
3109 It is possible to use an alternative marking scheme, using a similar
3110 idea to the copying solution. This scheme avoids the need to update
3111 the @CAF@ references explicitly. We introduce an auxillary {\em mark
3112 and update} @CAF@ info-table which is used to update all @CAF@s at the
3113 start of a collection. The new code marks the @CAF@ reference,
3114 updating it with the returned reference. The returned reference is
3115 itself returned so the @CAF@ is shorted out. The code also modifies the
3116 @CAF@ info-table to be a {\em return reference}. Subsequent attempts to
3117 mark the @CAF@ simply return the updated reference.
3119 A cyclic @CAF@ reference will result in an attempt to mark the @CAF@
3120 before the marking has been completed and the reference updated. We
3121 cannot start marking the @CAF@ as it is already being marked. Nor can
3122 we return the reference as it has not yet been updated. Neither can we
3123 treat the CAF as an indirection since the @CAF@ reference has been
3124 obscured by the pointer reversal stack. All we can do is return the
3125 @CAF@ itself. This will result in some @CAF@ references not being
3128 This scheme has not been adopted but has been implemented. The code is
3129 commented out with @#if 0@.
3131 \subsection{The virtual register set}
3133 We refer to any (atomic) part of the virtual machine state as a ``register.''
3134 These ``registers'' may be shared between all threads in the system or may be
3135 specific to each thread.
3141 Thread preemption flag
3146 TSO - pointer to the TSO for when we have to pack thread away
3149 Su - used to calculate number of arguments on stack
3150 - this is a more convenient representation
3151 Call/return registers (aka General purpose registers)
3152 Cost centre (and other debug/profile info)
3153 Statistic gathering (not in production system)
3155 Heap overflow - possible global?
3156 Stack overflow - possibly global?
3157 Pattern match failure
3158 maybe a failWith handler?
3159 maybe an exitWith handler?
3163 Some of these virtual ``registers'' are used very frequently and should
3164 be mapped onto machine registers if at all possible. Others are used
3165 very infrequently and can be kept in memory to free up registers for
3168 On register-poor architectures, we can play a few tricks to reduce the
3169 number of virtual registers which need to be accessed on a regular
3174 - Grow stack and heap towards each other (single-threaded system only)
3175 - We might need to keep the C stack pointer in a register if that
3176 is what the OS expects when a signal occurs.
3177 - Preemption flag trick
3178 - If any of the frequently accessed registers cannot be mapped onto
3179 machine registers we should keep the TSO in a machine register to
3180 allow faster access to all the other non-machine registers.