1 %************************************************************************
3 \section[SMinterface.lh]{Main storage manager interface}
5 %************************************************************************
7 %% I have changed most of the text here, in an attempt to understand
8 %% what's going on. Please let me know about any mistakes, so that
9 %% I can correct them! KH@15/10/92 (UK)
11 %% I have also split the original monster into SMinterface.lh,
12 %% SMClosures.lh and SMInfoTables.lh. The latter two are
15 This describes the interface used between the STG-machine
16 reducer and the storage manager. The overriding goal is to isolate
17 the implementation details of each from the other.
19 Multi-slurp protection:
26 {}\input{epsf} % Uses encapsulated PostScript diagrams
29 %************************************************************************
31 \subsection[SM-calling-interface]{Calling interface}
33 %************************************************************************
35 The @smInfo@ structure is used to pass all information back and forth
36 between the storage manager and the STG world.
38 WARNING: If you modify this structure, you {\em must} modify the
39 native-code generator as well, because the offsets for various fields
40 are hard-coded into the NCG. (In nativeGen/StixMacro.lhs).
44 P_ hp; /* last successfully allocated word */
45 P_ hplim; /* last allocatable word */
47 I_ rootno; /* No of heap roots stored in roots */
48 P_ *roots; /* Array of heap roots -- must be allocated (not static) */
49 P_ CAFlist; /* List of updated CAF's */
51 #if defined(GCap) || defined(GCgn)
52 P_ OldMutables; /* List of old generation mutable closures */
53 P_ OldLim; /* Ptr to end of the old generation */
57 P_ MallocPtrList; /* List of all Malloc Pointers (in new generation) */
59 #if defined(GCap) || defined(GCgn)
60 P_ OldMallocPtrList; /* List of all Malloc Pointers in old generation */
63 P_ StablePointerTable;
64 /* Heap allocated table used to store stable pointers in */
67 I_ hardHpOverflowSize; /* Some slop at the top of the heap which
68 (hopefully) provides enough space to let
69 us recover from heap overflow exceptions */
72 extern smInfo StorageMgrInfo;
76 Maximum number of roots storable in the heap roots array.
77 Question: Where are the stable pointer roots? (JSM)
78 Answer: They're on the heap in a "Stable Pointer Table". (ADR)
81 # define SM_MAXROOTS 8 /* 8 Vanilla Regs */
85 # define SM_MAXROOTS (10 + (MAX_PROC*4) + 2 + (MAX_PROC*2) + MAX_SPARKS)
86 /* unthreaded + spark/thread queues + Current/Main TSOs
89 # define SM_MAXROOTS 5 /* See c-as-asm/HpOverflow.lc */
92 # define SM_MAXROOTS 6 /* See c-as-asm/HpOverflow.lc */
97 The storage manager is accessed exclusively through these routines:
99 IF_RTS(I_ initSM PROTO((I_ rts_argc, char **rts_argv, FILE *statsfile));)
100 IF_RTS(I_ exitSM PROTO((smInfo *sm));)
101 IF_RTS(I_ initStacks PROTO((smInfo *sm));)
102 IF_RTS(I_ initHeap PROTO((smInfo *sm));)
104 IF_RTS(rtsBool initThreadPool PROTO((I_ size));)
107 IF_RTS(void init_gr_profiling PROTO((int, char **, int, char **));)
110 I_ collectHeap PROTO((W_ reqsize, smInfo *sm, rtsBool do_full_collection));
112 IF_RTS(void unmapMiddleStackPage PROTO((char *, int));) /* char * == caddr_t ? */
114 #if defined(USE_COST_CENTRES) || defined(GUM)
115 IF_RTS(void handle_tick_serial(STG_NO_ARGS);)
116 IF_RTS(void handle_tick_noserial(STG_NO_ARGS);)
119 /* EXTFUN(_startMarkWorld); */
121 StgDouble usertime(STG_NO_ARGS);
122 StgDouble elapsedtime(STG_NO_ARGS);
123 void start_time(STG_NO_ARGS);
124 void end_init(STG_NO_ARGS);
127 void EvacuateLocalGAs PROTO((rtsBool full));
128 void RebuildGAtables PROTO((rtsBool full));
133 @initSM@ processes any runtime parameters directed towards the storage
134 manager. The @statsfile@ parameter is an open file, which will contain
135 any garbage collection statistics requested by the user. This file
136 must be opened for writing.
138 @exitSM@ does any cleaning up required by the storage manager before
139 the program is executed. Its main purpose is to print any summary
142 @initStacks@ allocates the A and B stacks (sequential only). It
143 initialises the @spa@, @spb@, @sua@, and @sub@ fields of @sm@
144 appropriately for empty stacks. Successive calls to @initStacks@
145 re-initialise the stacks.
147 @initHeap@ allocates the heap. It initialises the @hp@ and @hplim@
148 fields of @sm@ to represent an empty heap for the compiled-in garbage
149 collector. It also allocates the @roots@ array for later use within
150 @collectHeap@, and initialises @CAFlist@ to be the empty list. The
151 @roots@ array must be large enough to hold at least @SM_MAXROOTS@
152 roots. If we are using Appel's collector it also initialises the
155 In the sequential system, it also initialises the stable pointer table
156 and the @MallocPtr@ (and @OldMallocPtrList@) fields.
158 @collectHeap@ invokes the garbage collector that was requested at
159 compile time. @reqsize@ is the size of the request (in words) that
160 resulted in the overflow. If the garbage collection succeeds, then at
161 least @reqsize@ words will be available. @collectHeap@ requires all
162 the fields of @sm@ to be initialised appropriately (from the
163 STG-machine registers). The following are identified as
166 \item The @roots@ array.
167 \item The updated CAFs recorded in @CAFlist@.
169 \item Update frames on the B Stack. These may be ``squeezed'' out
170 if they are the only reference to a closure --- thus avoiding the
172 \item The stable pointer table. (In sequential system.)
175 There are three possible results from a garbage collection:
177 \item[\tr{GC_HARD_LIMIT_EXCEEDED} (\tr{reqsize > hplim - hp})]
178 The heap size exceeds the hard heap limit: we report an error and
181 \item[\tr{GC_SOFT_LIMIT_EXCEEDED} (\tr{reqsize + hardHpOverflowSize > hplim - hp})]
182 The heap size exceeds the soft heap limit: set \tr{hardHpOverflowSize}
183 to \tr{0} so that we can use the overflow space, unwind the stack and
184 call an appropriate piece of Haskell to handle the error.
186 \item[\tr{GC_SUCCESS} (\tr{reqsize + hardHpOverflowSize <= hplim - hp})]
187 The heap size is less than the soft heap limit.
190 \item @hp@ and @hplim@ will indicate the new space available for
191 allocation. But we'll subtract \tr{hardHpOverflowSize} from
192 \tr{hplim} so that we'll GC when we hit the soft limit.
194 \item The elements of the @roots@ array will point to the new
195 locations of the closures.
197 \item @spb@ and @sub@ will be updated to reflect the new state of the
198 B stack arising from any update frame ``squeezing'' [sequential only].
200 \item The elements of @CAFlist@ and the stable pointers will be
201 updated to point to the new locations of the closures they reference.
203 \item Any members of @MallocPtrList@ which became garbage should have
204 been reported (by calling @FreeMallocPtr@; and the @(Old)MallocPtrList@
205 updated to contain only those Malloc Pointers which are still live.
211 #define GC_HARD_LIMIT_EXCEEDED 0
212 #define GC_SOFT_LIMIT_EXCEEDED 1
216 %************************************************************************
218 \subsection[SM-what-really-happens]{``What really happens in a garbage collection?''}
220 %************************************************************************
222 This is a brief tutorial on ``what really happens'' going to/from the
223 storage manager in a garbage collection.
226 %------------------------------------------------------------------------
227 \item[The heap check:]
231 If you gaze into the C output of GHC, you see many macros calls like:
233 HEAP_CHK_2PtrsLive((_FHS+2));
236 This expands into the C (roughly speaking...):
238 Hp = Hp + (_FHS+2); /* optimistically move heap pointer forward */
240 GC_WHILE_OR_IF (HEAP_OVERFLOW_OP(Hp, HpLim) OR_INTERVAL_EXPIRED) {
241 STGCALL2_GC(PerformGC, <liveness-bits>, (_FHS+2));
242 /* Heap full. Call "PerformGC" with 2 arguments, "<liveness>",
243 (info about what ptrs are live) and "_FHS+2" (words
244 requested), via the magical routine "callWrapper_GC",
245 which indicates ``I am calling a routine in which GC
246 may happen'' (a safe bet for `PerformGC').
251 In the parallel world, where we will need to re-try the heap check,
252 @GC_WHILE_OR_IF@ will be a ``while''; in the sequential world, it will
255 The ``heap lookahead'' checks, which are similar and used for
256 multi-precision @Integer@ ops, have some further complications. See
257 the commentary there (\tr{StgMacros.lh}).
259 %------------------------------------------------------------------------
260 \item[Into @callWrapper_GC@...:]
262 When we failed the heap check (above), we were inside the
263 GCC-registerised ``threaded world.'' @callWrapper_GC@ is all about
264 getting in and out of the threaded world. On SPARCs, with register
265 windows, the name of the game is not shifting windows until we have
266 what we want out of the old one. In tricky cases like this, it's best
267 written in assembly language.
269 Though the principle of ``save everything away'' is the same in both
270 the sequential and parallel worlds, the details are different.
272 For the sequential world:
275 @callWrapper_GC@ saves the return address.
277 It saves the arguments passed to it (so it doesn't get lost).
279 Save the machine registers used in the STG threaded world in their
280 \tr{*_SAVE} global-variable backup locations. E.g., register \tr{Hp}
281 is saved into \tr{Hp_SAVE}.
283 Call the routine it was asked to call; in this example, call
284 @PerformGC@ with arguments \tr{<liveness>}, and @_FHS+2@ (some constant)...
287 For the parallel world, a GC means giving up the thread of control.
288 So we must fill in the thread-state-object (TSO) [and its associated
289 stk object] with enough information for later resumption:
292 Save the return address in the TSO's PC field.
294 Save the machine registers used in the STG threaded world in their
295 corresponding TSO fields. We also save the pointer-liveness
296 information in the TSO.
298 The registers that are not thread-specific, notably \tr{Hp} and
299 \tr{HpLim}, are saved in the @StorageMgrInfo@ structure.
301 Call the routine it was asked to call; in this example, call
302 @PerformGC@ with arguments \tr{<liveness>} and @_FHS+2@ (some constant)...
304 (In the parallel world, we don't expect it to return...)
307 %------------------------------------------------------------------------
308 \item[Into the heap overflow wrapper, @PerformGC@ [sequential]:]
310 The first argument (\tr{<liveness>}, in our example) say what registers
311 are live, i.e., are ``roots'' the storage manager needs to know.
313 StorageMgrInfo.rootno = 2;
314 StorageMgrInfo.roots[0] = (P_) Ret1_SAVE;
315 StorageMgrInfo.roots[1] = (P_) Ret2_SAVE;
318 We further: (a)~move the heap-pointer back [we had optimistically
319 advanced it, in the initial heap check], (b)~load up the @smInfo@ data
320 from the STG registers' \tr{*_SAVE} locations, and (c)~FINALLY: call
323 IT IS AT THIS POINT THAT THE WORLD IS COMPLETELY TIDY.
325 %------------------------------------------------------------------------
326 \item[Into the heap overflow wrapper, @PerformGC@ [parallel]:]
328 Parallel execution is only slightly different. Most information has
329 already been saved in the TSO.
333 We still need to set up the storage manager's @roots@ array.
335 We mark on the scheduler's big ``blackboard'' that a GC is
338 We reschedule, i.e., this thread gives up control. (The scheduler
339 will presumably initiate a garbage-collection, but it may have to do
340 any number of other things---flushing, for example---before ``normal
341 execution'' resumes; and it most certainly may not be this thread that
342 resumes at that point!)
345 %------------------------------------------------------------------------
346 \item[Into/out of @collectHeap@ [sequential only]:]
348 @collectHeap@ does the business and reports back whether it freed up
351 %------------------------------------------------------------------------
352 \item[Out of the heap overflow wrapper, @PerformGC@ [sequential only]:]
354 We begin our return back to doing useful work by: (a)~reloading the
355 appropriate STG-register \tr{*_SAVE} locations from (presumably
356 changed) @smInfo@; (b) re-advance the heap-pointer---which we've been
357 trying to do for a week or two---now that there is enough space.
359 We must further restore appropriate @Ret?@ registers from the storage
360 manager's roots array; in this example:
363 Ret1_SAVE = (W_) StorageMgrInfo.roots[0];
364 Ret2_SAVE = (W_) StorageMgrInfo.roots[1];
367 %------------------------------------------------------------------------
368 \item[Out of @callWrapper_GC@ [sequential]:]
370 We pop out of heap-overflow code and are ready to resume STG
371 ``threaded world'' stuff.
373 The main thing is to re-load up the GCC-ised machine registers from
374 the relevant \tr{*_SAVE} locations; e.g., \tr{SpA} from \tr{SpA_SAVE}.
376 To conclude, @callWrapper_GC@ merely {\em jumps} back to the return
377 address which it was given originally.
379 WE'RE BACK IN (SEQUENTIAL) BUSINESS.
381 %------------------------------------------------------------------------
382 \item[Out of @callWrapper_GC@ [parallel]:]
384 When this thread is finally resumed after GC (and who knows what
385 else), it will restart by the normal enter-TSO/enter-stack-object
386 sequence, which has the effect of re-loading the registers, etc.,
387 (i.e., restoring the state).
389 Because the address we saved in the TSO's PC field was that at the end
390 of the heap check, and because the check is a while-loop in the
391 parallel system, we will now loop back around, and make sure there is
392 enough space before continuing.
395 %************************************************************************
397 \subsection[SM-stack-info]{Stacks}
399 %************************************************************************
401 There are two stacks, as in the STG paper \cite{new-stg-paper}.
404 The A stack contains only closure pointers.
406 The B stack contains, basic values, return addresses, and update
409 The A stack and B stack grow towards each other, so they overflow when
410 they collide. Currently the A stack grows downward (towards lower
411 addresses); the B stack grows upward. (We localise the stuff which
412 uses this information within macros defined in @StgDirections.h@)
414 During reduction, SpA and SpB point to the topmost allocated word of
415 the corresponding stack (though they may not be up to date in the
416 middle of a basic block).
418 Each stack also has a {\em stack update pointer}, SuA and SuB, which
419 point to the topmost word of the most recent update frame in the
420 corresponding stack. (Colloquially, SuA and Sub point to the first
421 items on their respective stacks ``that you cannot have.'')
423 A standard update frame (on the B stack) looks like this
424 (stack grows downward in this picture):
426 \mbox{\epsffile{update-frame.ps}}
428 The SuB therefore points to the Update return vector component of
429 the topmost update frame.
432 A {\em constructor} update frame, which is pushed only by closures
433 which know they will evaluate to a data object, looks just the
434 same, but without the saved SuA pointer.
436 We store the following information concerning the stacks in a global
437 structure. (sequential only).
441 PP_ botA; /* Points to bottom-most word of A stack */
442 P_ botB; /* Points to bottom-most word of B stack */
445 extern stackData stackInfo;
448 %************************************************************************
450 \subsection[SM-choose-flavour]{Deciding which GC flavour is in force...}
452 %************************************************************************
454 Each garbage collector requires different garbage collection entries
459 #define _INFO_COPYING
463 #define _INFO_COMPACTING
464 #define _INFO_MARKING
467 #if defined(GCdu) || defined (GCap) || defined (GCgn)
468 #define _INFO_COPYING
469 #define _INFO_COMPACTING
470 #define _INFO_MARKING
473 /* NO_INFO_SPECIFIED (ToDo: an #error ???) */
479 %************************************************************************
481 %\subsection[Info.lh]{Info Pointer Definitions}
483 %************************************************************************
489 %************************************************************************
491 %\subsection[Parallel.lh]{Parallel Machine Definitions}
493 %************************************************************************
500 %************************************************************************
502 %\subsection[CostCentre.lh]{Profiling Definitions}
504 %************************************************************************
507 \input{CostCentre.lh}
511 %************************************************************************
513 %\subsection[SM-closures]{Closure-Related Definitions}
515 %************************************************************************
518 \input{SMClosures.lh}
523 %************************************************************************
525 %\subsection[SM-info-tables]{Info-table Related Definitions}
527 %************************************************************************
530 \input{SMInfoTables.lh}
534 End multi-slurp protection:
536 #endif /* SMinterface_H */