blob: ac346b9ec29a67b609f1cbd9656b844569aa500b [file] [log] [blame]
njn43b9a8a2005-05-10 04:37:01 +00001
2/*--------------------------------------------------------------------*/
3/*--- The core/tool interface. pub_tool_tooliface.h ---*/
4/*--------------------------------------------------------------------*/
5
6/*
7 This file is part of Valgrind, a dynamic binary instrumentation
8 framework.
9
sewardj9eecbbb2010-05-03 21:37:12 +000010 Copyright (C) 2000-2010 Julian Seward
njn43b9a8a2005-05-10 04:37:01 +000011 jseward@acm.org
12
13 This program is free software; you can redistribute it and/or
14 modify it under the terms of the GNU General Public License as
15 published by the Free Software Foundation; either version 2 of the
16 License, or (at your option) any later version.
17
18 This program is distributed in the hope that it will be useful, but
19 WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 General Public License for more details.
22
23 You should have received a copy of the GNU General Public License
24 along with this program; if not, write to the Free Software
25 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
26 02111-1307, USA.
27
28 The GNU General Public License is contained in the file COPYING.
29*/
30
31#ifndef __PUB_TOOL_TOOLIFACE_H
32#define __PUB_TOOL_TOOLIFACE_H
33
njnacd885a2005-05-16 20:40:51 +000034#include "pub_tool_errormgr.h" // for Error, Supp
njn0fc5cbd2006-10-18 21:50:26 +000035#include "libvex.h" // for all Vex stuff
njnacd885a2005-05-16 20:40:51 +000036
njn43b9a8a2005-05-10 04:37:01 +000037/* ------------------------------------------------------------------ */
38/* The interface version */
39
njn08ce7b32009-02-27 03:38:28 +000040/* Initialise tool. Must do the following:
41 - initialise the `details' struct, via the VG_(details_*)() functions
42 - register the basic tool functions, via VG_(basic_tool_funcs)().
43 May do the following:
44 - initialise the `needs' struct to indicate certain requirements, via
45 the VG_(needs_*)() functions
46 - any other tool-specific initialisation
47*/
48extern void (*VG_(tl_pre_clo_init)) ( void );
njn43b9a8a2005-05-10 04:37:01 +000049
njn08ce7b32009-02-27 03:38:28 +000050/* Every tool must include this macro somewhere, exactly once. The
51 interface version is no longer relevant, but we kept the same name
52 to avoid requiring changes to tools.
53*/
54#define VG_DETERMINE_INTERFACE_VERSION(pre_clo_init) \
55 void (*VG_(tl_pre_clo_init)) ( void ) = pre_clo_init;
njn43b9a8a2005-05-10 04:37:01 +000056
57/* ------------------------------------------------------------------ */
58/* Basic tool functions */
59
sewardj461df9c2006-01-17 02:06:39 +000060/* The tool_instrument function is passed as a callback to
sewardj7ce62392006-10-15 12:48:18 +000061 LibVEX_Translate. VgCallbackClosure carries additional info
sewardj461df9c2006-01-17 02:06:39 +000062 which the instrumenter might like to know, but which is opaque to
63 Vex.
64*/
65typedef
66 struct {
67 Addr64 nraddr; /* non-redirected guest address */
68 Addr64 readdr; /* redirected guest address */
69 ThreadId tid; /* tid requesting translation */
70 }
71 VgCallbackClosure;
72
njn43b9a8a2005-05-10 04:37:01 +000073extern void VG_(basic_tool_funcs)(
74 // Do any initialisation that can only be done after command line
75 // processing.
76 void (*post_clo_init)(void),
77
sewardj4ba057c2005-10-18 12:04:18 +000078 // Instrument a basic block. Must be a true function, ie. the same
79 // input always results in the same output, because basic blocks
sewardj7ce62392006-10-15 12:48:18 +000080 // can be retranslated, unless you're doing something really
81 // strange. Anyway, the arguments. Mostly they are straightforward
82 // except for the distinction between redirected and non-redirected
83 // guest code addresses, which is important to understand.
84 //
85 // VgCallBackClosure* closure contains extra arguments passed
86 // from Valgrind to the instrumenter, which Vex doesn't know about.
87 // You are free to look inside this structure.
88 //
89 // * closure->tid is the ThreadId of the thread requesting the
90 // translation. Not sure why this is here; perhaps callgrind
91 // uses it.
92 //
93 // * closure->nraddr is the non-redirected guest address of the
94 // start of the translation. In other words, the translation is
95 // being constructed because the guest program jumped to
96 // closure->nraddr but no translation of it was found.
97 //
98 // * closure->readdr is the redirected guest address, from which
99 // the translation was really made.
100 //
101 // To clarify this, consider what happens when, in Memcheck, the
102 // first call to malloc() happens. The guest program will be
103 // trying to jump to malloc() in libc; hence ->nraddr will contain
104 // that address. However, Memcheck intercepts and replaces
105 // malloc, hence ->readdr will be the address of Memcheck's
106 // malloc replacement in
107 // coregrind/m_replacemalloc/vg_replacemalloc.c. It follows
108 // that the first IMark in the translation will be labelled as
109 // from ->readdr rather than ->nraddr.
110 //
111 // Since most functions are not redirected, the majority of the
112 // time ->nraddr will be the same as ->readdr. However, you
113 // cannot assume this: if your tool has metadata associated
114 // with code addresses it will get into deep trouble if it does
115 // make this assumption.
116 //
sewardj0b9d74a2006-12-24 02:24:11 +0000117 // IRSB* sb_in is the incoming superblock to be instrumented,
118 // in flat IR form.
sewardj7ce62392006-10-15 12:48:18 +0000119 //
120 // VexGuestLayout* layout contains limited info on the layout of
121 // the guest state: where the stack pointer and program counter
122 // are, and which fields should be regarded as 'always defined'.
123 // Memcheck uses this.
124 //
125 // VexGuestExtents* vge points to a structure which states the
126 // precise byte ranges of original code from which this translation
127 // was made (there may be up to three different ranges involved).
128 // Note again that these are the real addresses from which the code
129 // came. And so it should be the case that closure->readdr is the
130 // same as vge->base[0]; indeed Cachegrind contains this assertion.
131 //
132 // Tools which associate shadow data with code addresses
133 // (cachegrind, callgrind) need to be particularly clear about
134 // whether they are making the association with redirected or
135 // non-redirected code addresses. Both approaches are viable
136 // but you do need to understand what's going on. See comments
137 // below on discard_basic_block_info().
138 //
139 // IRType gWordTy and IRType hWordTy contain the types of native
140 // words on the guest (simulated) and host (real) CPUs. They will
141 // by either Ity_I32 or Ity_I64. So far we have never built a
142 // cross-architecture Valgrind so they should always be the same.
143 //
sewardjc87b5ec2006-10-15 13:46:18 +0000144 /* --- Further comments about the IR that your --- */
145 /* --- instrumentation function will receive. --- */
146 /*
njna3cc29f2007-02-05 23:23:55 +0000147 In the incoming IRSB, the IR for each instruction begins with an
sewardjc87b5ec2006-10-15 13:46:18 +0000148 IRStmt_IMark, which states the address and length of the
149 instruction from which this IR came. This makes it easy for
150 profiling-style tools to know precisely which guest code
151 addresses are being executed.
152
153 However, before the first IRStmt_IMark, there may be other IR
154 statements -- a preamble. In most cases this preamble is empty,
155 but when it isn't, what it contains is some supporting IR that
156 the JIT uses to ensure control flow works correctly. This
157 preamble does not modify any architecturally defined guest state
158 (registers or memory) and so does not contain anything that will
159 be of interest to your tool.
160
161 You should therefore
162
163 (1) copy any IR preceding the first IMark verbatim to the start
njna3cc29f2007-02-05 23:23:55 +0000164 of the output IRSB.
sewardjc87b5ec2006-10-15 13:46:18 +0000165
166 (2) not try to instrument it or modify it in any way.
167
168 For the record, stuff that may be in the preamble at
169 present is:
170
171 - A self-modifying-code check has been requested for this block.
172 The preamble will contain instructions to checksum the block,
173 compare against the expected value, and exit the dispatcher
174 requesting a discard (hence forcing a retranslation) if they
175 don't match.
176
177 - This block is known to be the entry point of a wrapper of some
sewardj358ebea2006-10-15 13:47:43 +0000178 function F. In this case the preamble contains code to write
sewardjc87b5ec2006-10-15 13:46:18 +0000179 the address of the original F (the fn being wrapped) into a
180 'hidden' guest state register _NRADDR. The wrapper can later
181 read this register using a client request and make a
182 non-redirected call to it using another client-request-like
183 magic macro.
184
185 - For platforms that use the AIX ABI (including ppc64-linux), it
sewardj358ebea2006-10-15 13:47:43 +0000186 is necessary to have a preamble even for replacement functions
187 (not just for wrappers), because it is necessary to switch the
188 R2 register (constant-pool pointer) to a different value when
189 swizzling the program counter.
sewardjc87b5ec2006-10-15 13:46:18 +0000190
191 Hence the preamble pushes both R2 and LR (the return address)
192 on a small 16-entry stack in the guest state and sets R2 to an
193 appropriate value for the wrapper/replacement fn. LR is then
194 set so that the wrapper/replacement fn returns to a magic IR
195 stub which restores R2 and LR and returns.
196
197 It's all hugely ugly and fragile. And it places a stringent
198 requirement on m_debuginfo to find out the correct R2 (toc
199 pointer) value for the wrapper/replacement function. So much
200 so that m_redir will refuse to honour a redirect-to-me request
201 if it cannot find (by asking m_debuginfo) a plausible R2 value
202 for 'me'.
203
204 Because this mechanism maintains a shadow stack of (R2,LR)
205 pairs in the guest state, it will fail if the
206 wrapper/redirection function, or anything it calls, longjumps
207 out past the wrapper, because then the magic return stub will
208 not be run and so the shadow stack will not be popped. So it
209 will quickly fill up. Fortunately none of this applies to
210 {x86,amd64,ppc32}-linux; on those platforms, wrappers can
211 longjump and recurse arbitrarily and everything should work
212 fine.
sewardjf1962d32006-10-19 13:22:16 +0000213
214 Note that copying the preamble verbatim may cause complications
215 for your instrumenter if you shadow IR temporaries. See big
216 comment in MC_(instrument) in memcheck/mc_translate.c for
217 details.
sewardjc87b5ec2006-10-15 13:46:18 +0000218 */
sewardj0b9d74a2006-12-24 02:24:11 +0000219 IRSB*(*instrument)(VgCallbackClosure* closure,
220 IRSB* sb_in,
sewardj7ce62392006-10-15 12:48:18 +0000221 VexGuestLayout* layout,
222 VexGuestExtents* vge,
223 IRType gWordTy,
224 IRType hWordTy),
njn43b9a8a2005-05-10 04:37:01 +0000225
226 // Finish up, print out any results, etc. `exitcode' is program's exit
227 // code. The shadow can be found with VG_(get_exit_status_shadow)().
228 void (*fini)(Int)
229);
230
231/* ------------------------------------------------------------------ */
232/* Details */
233
234/* Default value for avg_translations_sizeB (in bytes), indicating typical
235 code expansion of about 6:1. */
sewardj9644cfd2006-10-17 02:25:50 +0000236#define VG_DEFAULT_TRANS_SIZEB 172
njn43b9a8a2005-05-10 04:37:01 +0000237
238/* Information used in the startup message. `name' also determines the
239 string used for identifying suppressions in a suppression file as
240 belonging to this tool. `version' can be NULL, in which case (not
241 surprisingly) no version info is printed; this mechanism is designed for
242 tools distributed with Valgrind that share a version number with
243 Valgrind. Other tools not distributed as part of Valgrind should
244 probably have their own version number. */
245extern void VG_(details_name) ( Char* name );
246extern void VG_(details_version) ( Char* version );
247extern void VG_(details_description) ( Char* description );
248extern void VG_(details_copyright_author) ( Char* copyright_author );
249
250/* Average size of a translation, in bytes, so that the translation
251 storage machinery can allocate memory appropriately. Not critical,
252 setting is optional. */
253extern void VG_(details_avg_translation_sizeB) ( UInt size );
254
255/* String printed if an `tl_assert' assertion fails or VG_(tool_panic)
256 is called. Should probably be an email address. */
257extern void VG_(details_bug_reports_to) ( Char* bug_reports_to );
258
259/* ------------------------------------------------------------------ */
260/* Needs */
261
njn43b9a8a2005-05-10 04:37:01 +0000262/* Should __libc_freeres() be run? Bugs in it can crash the tool. */
263extern void VG_(needs_libc_freeres) ( void );
264
265/* Want to have errors detected by Valgrind's core reported? Includes:
njn0087c502005-07-01 04:15:36 +0000266 - pthread API errors (many; eg. unlocking a non-locked mutex)
267 [currently disabled]
268 - invalid file descriptors to syscalls like read() and write()
njn43b9a8a2005-05-10 04:37:01 +0000269 - bad signal numbers passed to sigaction()
270 - attempt to install signal handler for SIGKILL or SIGSTOP */
271extern void VG_(needs_core_errors) ( void );
272
273/* Booleans that indicate extra operations are defined; if these are True,
274 the corresponding template functions (given below) must be defined. A
275 lot like being a member of a type class. */
276
277/* Want to report errors from tool? This implies use of suppressions, too. */
278extern void VG_(needs_tool_errors) (
njna5453e92009-07-27 22:21:22 +0000279 // Identify if two errors are equal, or close enough. This function is
280 // only called if e1 and e2 will have the same error kind. `res' indicates
281 // how close is "close enough". `res' should be passed on as necessary,
282 // eg. if the Error's `extra' part contains an ExeContext, `res' should be
njn43b9a8a2005-05-10 04:37:01 +0000283 // passed to VG_(eq_ExeContext)() if the ExeContexts are considered. Other
284 // than that, probably don't worry about it unless you have lots of very
285 // similar errors occurring.
286 Bool (*eq_Error)(VgRes res, Error* e1, Error* e2),
287
sewardj5d9dc752009-07-15 14:52:18 +0000288 // We give tools a chance to have a look at errors
289 // just before they are printed. That is, before_pp_Error is
290 // called just before pp_Error itself. This gives the tool a
291 // chance to look at the just-about-to-be-printed error, so as to
292 // emit any arbitrary output if wants to, before the error itself
293 // is printed. This functionality was added to allow Helgrind to
294 // print thread-announcement messages immediately before the
295 // errors that refer to them.
296 void (*before_pp_Error)(Error* err),
297
njn43b9a8a2005-05-10 04:37:01 +0000298 // Print error context.
299 void (*pp_Error)(Error* err),
300
sewardjadb102f2007-11-09 23:21:44 +0000301 // Should the core indicate which ThreadId each error comes from?
302 Bool show_ThreadIDs_for_errors,
303
njn43b9a8a2005-05-10 04:37:01 +0000304 // Should fill in any details that could be postponed until after the
305 // decision whether to ignore the error (ie. details not affecting the
306 // result of VG_(tdict).tool_eq_Error()). This saves time when errors
307 // are ignored.
308 // Yuk.
309 // Return value: must be the size of the `extra' part in bytes -- used by
310 // the core to make a copy.
311 UInt (*update_extra)(Error* err),
312
313 // Return value indicates recognition. If recognised, must set skind using
314 // VG_(set_supp_kind)().
315 Bool (*recognised_suppression)(Char* name, Supp* su),
316
317 // Read any extra info for this suppression kind. Most likely for filling
318 // in the `extra' and `string' parts (with VG_(set_supp_{extra, string})())
319 // of a suppression if necessary. Should return False if a syntax error
njn35db56c2009-07-24 07:38:29 +0000320 // occurred, True otherwise. bufpp and nBufp are the same as for
321 // VG_(get_line).
322 Bool (*read_extra_suppression_info)(Int fd, Char** bufpp, SizeT* nBufp,
323 Supp* su),
njn43b9a8a2005-05-10 04:37:01 +0000324
325 // This should just check the kinds match and maybe some stuff in the
326 // `string' and `extra' field if appropriate (using VG_(get_supp_*)() to
327 // get the relevant suppression parts).
328 Bool (*error_matches_suppression)(Error* err, Supp* su),
329
330 // This should return the suppression name, for --gen-suppressions, or NULL
331 // if that error type cannot be suppressed. This is the inverse of
332 // VG_(tdict).tool_recognised_suppression().
333 Char* (*get_error_name)(Error* err),
334
sewardj588adef2009-08-15 22:41:51 +0000335 // This should print into buf[0..nBuf-1] any extra info for the
336 // error, for --gen-suppressions, but not including any leading
337 // spaces nor a trailing newline. When called, buf[0 .. nBuf-1]
338 // will be zero filled, and it is expected and checked that the
339 // last element is still zero after the call. In other words the
340 // tool may not overrun the buffer, and this is checked for. If
341 // there is any info printed in the buffer, return True, otherwise
342 // do nothing, and return False. This function is the inverse of
njn43b9a8a2005-05-10 04:37:01 +0000343 // VG_(tdict).tool_read_extra_suppression_info().
sewardj588adef2009-08-15 22:41:51 +0000344 Bool (*print_extra_suppression_info)(Error* err,
345 /*OUT*/Char* buf, Int nBuf)
njn43b9a8a2005-05-10 04:37:01 +0000346);
347
sewardj5155dec2005-10-12 10:09:23 +0000348/* Is information kept by the tool about specific instructions or
349 translations? (Eg. for cachegrind there are cost-centres for every
350 instruction, stored in a per-translation fashion.) If so, the info
351 may have to be discarded when translations are unloaded (eg. due to
352 .so unloading, or otherwise at the discretion of m_transtab, eg
353 when the table becomes too full) to avoid stale information being
354 reused for new translations. */
sewardj0b9d74a2006-12-24 02:24:11 +0000355extern void VG_(needs_superblock_discards) (
sewardj5155dec2005-10-12 10:09:23 +0000356 // Discard any information that pertains to specific translations
sewardj4ba057c2005-10-18 12:04:18 +0000357 // or instructions within the address range given. There are two
358 // possible approaches.
359 // - If info is being stored at a per-translation level, use orig_addr
360 // to identify which translation is being discarded. Each translation
361 // will be discarded exactly once.
sewardj7ce62392006-10-15 12:48:18 +0000362 // This orig_addr will match the closure->nraddr which was passed to
363 // to instrument() (see extensive comments above) when this
364 // translation was made. Note that orig_addr won't necessarily be
365 // the same as the first address in "extents".
sewardj5155dec2005-10-12 10:09:23 +0000366 // - If info is being stored at a per-instruction level, you can get
367 // the address range(s) being discarded by stepping through "extents".
368 // Note that any single instruction may belong to more than one
369 // translation, and so could be covered by the "extents" of more than
370 // one call to this function.
371 // Doing it the first way (as eg. Cachegrind does) is probably easier.
sewardj0b9d74a2006-12-24 02:24:11 +0000372 void (*discard_superblock_info)(Addr64 orig_addr, VexGuestExtents extents)
njn43b9a8a2005-05-10 04:37:01 +0000373);
374
375/* Tool defines its own command line options? */
376extern void VG_(needs_command_line_options) (
njnb1cc5d62010-07-06 04:05:23 +0000377 // Return True if option was recognised, False if it wasn't (but also see
378 // below). Presumably sets some state to record the option as well.
379 //
380 // Nb: tools can assume that the argv will never disappear. So they can,
381 // for example, store a pointer to a string within an option, rather than
382 // having to make a copy.
383 //
384 // Options (and combinations of options) should be checked in this function
385 // if possible rather than in post_clo_init(), and if they are bad then
386 // VG_(fmsg_bad_option)() should be called. This ensures that the
387 // messaging is consistent with command line option errors from the core.
njn43b9a8a2005-05-10 04:37:01 +0000388 Bool (*process_cmd_line_option)(Char* argv),
389
390 // Print out command line usage for options for normal tool operation.
391 void (*print_usage)(void),
392
393 // Print out command line usage for options for debugging the tool.
394 void (*print_debug_usage)(void)
395);
396
397/* Tool defines its own client requests? */
398extern void VG_(needs_client_requests) (
399 // If using client requests, the number of the first request should be equal
400 // to VG_USERREQ_TOOL_BASE('X', 'Y'), where 'X' and 'Y' form a suitable two
401 // character identification for the string. The second and subsequent
402 // requests should follow.
403 //
404 // This function should use the VG_IS_TOOL_USERREQ macro (in
405 // include/valgrind.h) to first check if it's a request for this tool. Then
406 // should handle it if it's recognised (and return True), or return False if
407 // not recognised. arg_block[0] holds the request number, any further args
408 // from the request are in arg_block[1..]. 'ret' is for the return value...
409 // it should probably be filled, if only with 0.
410 Bool (*handle_client_request)(ThreadId tid, UWord* arg_block, UWord* ret)
411);
412
413/* Tool does stuff before and/or after system calls? */
414// Nb: If either of the pre_ functions malloc() something to return, the
415// corresponding post_ function had better free() it!
sewardj1c0ce7a2009-07-01 08:10:49 +0000416// Also, the args are the 'original args' -- that is, it may be
417// that the syscall pre-wrapper will modify the args before the
418// syscall happens. So these args are the original, un-modified
419// args. Finally, nArgs merely indicates the length of args[..],
420// it does not indicate how many of those values are actually
421// relevant to the syscall. args[0 .. nArgs-1] is guaranteed
422// to be defined and to contain all the args for this syscall,
423// possibly including some trailing zeroes.
njn43b9a8a2005-05-10 04:37:01 +0000424extern void VG_(needs_syscall_wrapper) (
sewardj1c0ce7a2009-07-01 08:10:49 +0000425 void (* pre_syscall)(ThreadId tid, UInt syscallno,
426 UWord* args, UInt nArgs),
427 void (*post_syscall)(ThreadId tid, UInt syscallno,
428 UWord* args, UInt nArgs, SysRes res)
njn43b9a8a2005-05-10 04:37:01 +0000429);
430
431/* Are tool-state sanity checks performed? */
432// Can be useful for ensuring a tool's correctness. cheap_sanity_check()
433// is called very frequently; expensive_sanity_check() is called less
434// frequently and can be more involved.
435extern void VG_(needs_sanity_checks) (
436 Bool(*cheap_sanity_check)(void),
437 Bool(*expensive_sanity_check)(void)
438);
439
sewardjb8b79ad2008-03-03 01:35:41 +0000440/* Do we need to see variable type and location information? */
441extern void VG_(needs_var_info) ( void );
njn43b9a8a2005-05-10 04:37:01 +0000442
njn09ca09b2005-10-16 17:48:09 +0000443/* Does the tool replace malloc() and friends with its own versions?
444 This has to be combined with the use of a vgpreload_<tool>.so module
445 or it won't work. See massif/Makefile.am for how to build it. */
njn43b9a8a2005-05-10 04:37:01 +0000446// The 'p' prefix avoids GCC complaints about overshadowing global names.
njnfc51f8d2005-06-21 03:20:17 +0000447extern void VG_(needs_malloc_replacement)(
njn43b9a8a2005-05-10 04:37:01 +0000448 void* (*pmalloc) ( ThreadId tid, SizeT n ),
449 void* (*p__builtin_new) ( ThreadId tid, SizeT n ),
450 void* (*p__builtin_vec_new) ( ThreadId tid, SizeT n ),
451 void* (*pmemalign) ( ThreadId tid, SizeT align, SizeT n ),
452 void* (*pcalloc) ( ThreadId tid, SizeT nmemb, SizeT size1 ),
453 void (*pfree) ( ThreadId tid, void* p ),
454 void (*p__builtin_delete) ( ThreadId tid, void* p ),
455 void (*p__builtin_vec_delete) ( ThreadId tid, void* p ),
456 void* (*prealloc) ( ThreadId tid, void* p, SizeT new_size ),
njn8b140de2009-02-17 04:31:18 +0000457 SizeT (*pmalloc_usable_size) ( ThreadId tid, void* p),
njn43b9a8a2005-05-10 04:37:01 +0000458 SizeT client_malloc_redzone_szB
459);
460
njnca54af32006-04-16 10:25:43 +0000461/* Can the tool do XML output? This is a slight misnomer, because the tool
462 * is not requesting the core to do anything, rather saying "I can handle
463 * it". */
sewardj5d9dc752009-07-15 14:52:18 +0000464extern void VG_(needs_xml_output) ( void );
njnca54af32006-04-16 10:25:43 +0000465
sewardj81651dc2007-08-28 06:05:20 +0000466/* Does the tool want to have one final pass over the IR after tree
467 building but before instruction selection? If so specify the
468 function here. */
469extern void VG_(needs_final_IR_tidy_pass) ( IRSB*(*final_tidy)(IRSB*) );
470
471
njn43b9a8a2005-05-10 04:37:01 +0000472/* ------------------------------------------------------------------ */
473/* Core events to track */
474
475/* Part of the core from which this call was made. Useful for determining
476 what kind of error message should be emitted. */
477typedef
sewardj9c606bd2008-09-18 18:12:50 +0000478 enum { Vg_CoreStartup=1, Vg_CoreSignal, Vg_CoreSysCall,
njnf76d27a2009-05-28 01:53:07 +0000479 // This is for platforms where syscall args are passed on the
480 // stack; although pre_mem_read is the callback that will be
481 // called, such an arg should be treated (with respect to
482 // presenting information to the user) as if it was passed in a
483 // register, ie. like pre_reg_read.
484 Vg_CoreSysCallArgInMem,
485 Vg_CoreTranslate, Vg_CoreClientReq
486 } CorePart;
njn43b9a8a2005-05-10 04:37:01 +0000487
488/* Events happening in core to track. To be notified, pass a callback
489 function to the appropriate function. To ignore an event, don't do
490 anything (the default is for events to be ignored).
491
492 Note that most events aren't passed a ThreadId. If the event is one called
493 from generated code (eg. new_mem_stack_*), you can use
494 VG_(get_running_tid)() to find it. Otherwise, it has to be passed in,
495 as in pre_mem_read, and so the event signature will require changing.
496
497 Memory events (Nb: to track heap allocation/freeing, a tool must replace
498 malloc() et al. See above how to do this.)
499
sewardj7cf4e6b2008-05-01 20:24:26 +0000500 These ones occur at startup, upon some signals, and upon some syscalls.
501
sewardj9c606bd2008-09-18 18:12:50 +0000502 For new_mem_brk and new_mem_stack_signal, the supplied ThreadId
sewardj7cf4e6b2008-05-01 20:24:26 +0000503 indicates the thread for whom the new memory is being allocated.
sewardj9c606bd2008-09-18 18:12:50 +0000504
505 For new_mem_startup and new_mem_mmap, the di_handle argument is a
506 handle which can be used to retrieve debug info associated with the
507 mapping or allocation (because it is of a file that Valgrind has
508 decided to read debug info from). If the value is zero, there is
509 no associated debug info. If the value exceeds zero, it can be
510 supplied as an argument to selected queries in m_debuginfo.
sewardj7cf4e6b2008-05-01 20:24:26 +0000511*/
njn43b9a8a2005-05-10 04:37:01 +0000512void VG_(track_new_mem_startup) (void(*f)(Addr a, SizeT len,
sewardj9c606bd2008-09-18 18:12:50 +0000513 Bool rr, Bool ww, Bool xx,
514 ULong di_handle));
sewardj7cf4e6b2008-05-01 20:24:26 +0000515void VG_(track_new_mem_stack_signal)(void(*f)(Addr a, SizeT len, ThreadId tid));
516void VG_(track_new_mem_brk) (void(*f)(Addr a, SizeT len, ThreadId tid));
njn43b9a8a2005-05-10 04:37:01 +0000517void VG_(track_new_mem_mmap) (void(*f)(Addr a, SizeT len,
sewardj9c606bd2008-09-18 18:12:50 +0000518 Bool rr, Bool ww, Bool xx,
519 ULong di_handle));
njn43b9a8a2005-05-10 04:37:01 +0000520
521void VG_(track_copy_mem_remap) (void(*f)(Addr from, Addr to, SizeT len));
522void VG_(track_change_mem_mprotect) (void(*f)(Addr a, SizeT len,
523 Bool rr, Bool ww, Bool xx));
524void VG_(track_die_mem_stack_signal)(void(*f)(Addr a, SizeT len));
525void VG_(track_die_mem_brk) (void(*f)(Addr a, SizeT len));
526void VG_(track_die_mem_munmap) (void(*f)(Addr a, SizeT len));
527
528/* These ones are called when SP changes. A tool could track these itself
529 (except for ban_mem_stack) but it's much easier to use the core's help.
530
531 The specialised ones are called in preference to the general one, if they
532 are defined. These functions are called a lot if they are used, so
533 specialising can optimise things significantly. If any of the
534 specialised cases are defined, the general case must be defined too.
535
njnaf839f52005-06-23 03:27:57 +0000536 Nb: all the specialised ones must use the VG_REGPARM(n) attribute.
sewardj7cf4e6b2008-05-01 20:24:26 +0000537
538 For the _new functions, a tool may specify with with-ECU
539 (ExeContext Unique) or without-ECU version for each size, but not
540 both. If the with-ECU version is supplied, then the core will
541 arrange to pass, as the ecu argument, a 32-bit int which uniquely
542 identifies the instruction moving the stack pointer down. This
543 32-bit value is as obtained from VG_(get_ECU_from_ExeContext).
544 VG_(get_ExeContext_from_ECU) can then be used to retrieve the
545 associated depth-1 ExeContext for the location. All this
546 complexity is provided to support origin tracking in Memcheck.
547*/
548void VG_(track_new_mem_stack_4_w_ECU) (VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
549void VG_(track_new_mem_stack_8_w_ECU) (VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
550void VG_(track_new_mem_stack_12_w_ECU) (VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
551void VG_(track_new_mem_stack_16_w_ECU) (VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
552void VG_(track_new_mem_stack_32_w_ECU) (VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
553void VG_(track_new_mem_stack_112_w_ECU)(VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
554void VG_(track_new_mem_stack_128_w_ECU)(VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
555void VG_(track_new_mem_stack_144_w_ECU)(VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
556void VG_(track_new_mem_stack_160_w_ECU)(VG_REGPARM(2) void(*f)(Addr new_ESP, UInt ecu));
557void VG_(track_new_mem_stack_w_ECU) (void(*f)(Addr a, SizeT len,
558 UInt ecu));
559
sewardjf5c8e372006-02-12 15:42:20 +0000560void VG_(track_new_mem_stack_4) (VG_REGPARM(1) void(*f)(Addr new_ESP));
561void VG_(track_new_mem_stack_8) (VG_REGPARM(1) void(*f)(Addr new_ESP));
562void VG_(track_new_mem_stack_12) (VG_REGPARM(1) void(*f)(Addr new_ESP));
563void VG_(track_new_mem_stack_16) (VG_REGPARM(1) void(*f)(Addr new_ESP));
564void VG_(track_new_mem_stack_32) (VG_REGPARM(1) void(*f)(Addr new_ESP));
565void VG_(track_new_mem_stack_112)(VG_REGPARM(1) void(*f)(Addr new_ESP));
566void VG_(track_new_mem_stack_128)(VG_REGPARM(1) void(*f)(Addr new_ESP));
567void VG_(track_new_mem_stack_144)(VG_REGPARM(1) void(*f)(Addr new_ESP));
568void VG_(track_new_mem_stack_160)(VG_REGPARM(1) void(*f)(Addr new_ESP));
569void VG_(track_new_mem_stack) (void(*f)(Addr a, SizeT len));
njn43b9a8a2005-05-10 04:37:01 +0000570
sewardjf5c8e372006-02-12 15:42:20 +0000571void VG_(track_die_mem_stack_4) (VG_REGPARM(1) void(*f)(Addr die_ESP));
572void VG_(track_die_mem_stack_8) (VG_REGPARM(1) void(*f)(Addr die_ESP));
573void VG_(track_die_mem_stack_12) (VG_REGPARM(1) void(*f)(Addr die_ESP));
574void VG_(track_die_mem_stack_16) (VG_REGPARM(1) void(*f)(Addr die_ESP));
575void VG_(track_die_mem_stack_32) (VG_REGPARM(1) void(*f)(Addr die_ESP));
576void VG_(track_die_mem_stack_112)(VG_REGPARM(1) void(*f)(Addr die_ESP));
577void VG_(track_die_mem_stack_128)(VG_REGPARM(1) void(*f)(Addr die_ESP));
578void VG_(track_die_mem_stack_144)(VG_REGPARM(1) void(*f)(Addr die_ESP));
579void VG_(track_die_mem_stack_160)(VG_REGPARM(1) void(*f)(Addr die_ESP));
580void VG_(track_die_mem_stack) (void(*f)(Addr a, SizeT len));
njn43b9a8a2005-05-10 04:37:01 +0000581
582/* Used for redzone at end of thread stacks */
583void VG_(track_ban_mem_stack) (void(*f)(Addr a, SizeT len));
584
585/* These ones occur around syscalls, signal handling, etc */
586void VG_(track_pre_mem_read) (void(*f)(CorePart part, ThreadId tid,
587 Char* s, Addr a, SizeT size));
588void VG_(track_pre_mem_read_asciiz)(void(*f)(CorePart part, ThreadId tid,
589 Char* s, Addr a));
590void VG_(track_pre_mem_write) (void(*f)(CorePart part, ThreadId tid,
591 Char* s, Addr a, SizeT size));
592void VG_(track_post_mem_write) (void(*f)(CorePart part, ThreadId tid,
593 Addr a, SizeT size));
594
595/* Register events. Use VG_(set_shadow_state_area)() to set the shadow regs
596 for these events. */
597void VG_(track_pre_reg_read) (void(*f)(CorePart part, ThreadId tid,
njnc4431bf2009-01-15 21:29:24 +0000598 Char* s, PtrdiffT guest_state_offset,
njn43b9a8a2005-05-10 04:37:01 +0000599 SizeT size));
600void VG_(track_post_reg_write)(void(*f)(CorePart part, ThreadId tid,
njnc4431bf2009-01-15 21:29:24 +0000601 PtrdiffT guest_state_offset,
njn43b9a8a2005-05-10 04:37:01 +0000602 SizeT size));
603
604/* This one is called for malloc() et al if they are replaced by a tool. */
605void VG_(track_post_reg_write_clientcall_return)(
njnc4431bf2009-01-15 21:29:24 +0000606 void(*f)(ThreadId tid, PtrdiffT guest_state_offset, SizeT size, Addr f));
njn43b9a8a2005-05-10 04:37:01 +0000607
608
609/* Scheduler events (not exhaustive) */
sewardj97561812006-12-23 01:21:12 +0000610
611/* Called when 'tid' starts or stops running client code blocks.
sewardjadb102f2007-11-09 23:21:44 +0000612 Gives the total dispatched block count at that event. Note, this
613 is not the same as 'tid' holding the BigLock (the lock that ensures
614 that only one thread runs at a time): a thread can hold the lock
615 for other purposes (making translations, etc) yet not be running
616 client blocks. Obviously though, a thread must hold the lock in
617 order to run client code blocks, so the times bracketed by
618 'start_client_code'..'stop_client_code' are a subset of the times
619 when thread 'tid' holds the cpu lock.
sewardj97561812006-12-23 01:21:12 +0000620*/
njn3e32c872006-12-24 07:51:17 +0000621void VG_(track_start_client_code)(
622 void(*f)(ThreadId tid, ULong blocks_dispatched)
623 );
624void VG_(track_stop_client_code)(
625 void(*f)(ThreadId tid, ULong blocks_dispatched)
sewardj97561812006-12-23 01:21:12 +0000626 );
njn43b9a8a2005-05-10 04:37:01 +0000627
628
629/* Thread events (not exhaustive)
630
sewardjadb102f2007-11-09 23:21:44 +0000631 ll_create: low level thread creation. Called before the new thread
632 has run any instructions (or touched any memory). In fact, called
633 immediately before the new thread has come into existence; the new
634 thread can be assumed to exist when notified by this call.
635
636 ll_exit: low level thread exit. Called after the exiting thread
637 has run its last instruction.
sewardj29d68f72007-11-10 22:13:03 +0000638
639 The _ll_ part makes it clear these events are not to do with
640 pthread_create or pthread_exit/pthread_join (etc), which are a
sewardj391ddf82007-11-10 22:19:42 +0000641 higher level abstraction synthesised by libpthread. What you can
642 be sure of from _ll_create/_ll_exit is the absolute limits of each
sewardj29d68f72007-11-10 22:13:03 +0000643 thread's lifetime, and hence be assured that all memory references
644 made by the thread fall inside the _ll_create/_ll_exit pair. This
645 is important for tools that need a 100% accurate account of which
646 thread is responsible for every memory reference in the process.
647
sewardj391ddf82007-11-10 22:19:42 +0000648 pthread_create/join/exit do not give this property. Calls/returns
649 to/from them happen arbitrarily far away from the relevant
650 low-level thread create/quit event. In general a few hundred
651 instructions; hence a few hundred(ish) memory references could get
652 misclassified each time.
sewardj7a387ea2007-11-25 14:06:06 +0000653
654 pre_thread_first_insn: is called when the thread is all set up and
655 ready to go (stack in place, etc) but has not executed its first
656 instruction yet. Gives threading tools a chance to ask questions
657 about the thread (eg, what is its initial client stack pointer)
658 that are not easily answered at pre_thread_ll_create time.
659
660 For a given thread, the call sequence is:
661 ll_create (in the parent's context)
662 first_insn (in the child's context)
663 ll_exit (in the child's context)
sewardjadb102f2007-11-09 23:21:44 +0000664*/
sewardj7a387ea2007-11-25 14:06:06 +0000665void VG_(track_pre_thread_ll_create) (void(*f)(ThreadId tid, ThreadId child));
666void VG_(track_pre_thread_first_insn)(void(*f)(ThreadId tid));
667void VG_(track_pre_thread_ll_exit) (void(*f)(ThreadId tid));
njn43b9a8a2005-05-10 04:37:01 +0000668
njn3e32c872006-12-24 07:51:17 +0000669
njn43b9a8a2005-05-10 04:37:01 +0000670/* Signal events (not exhaustive)
671
672 ... pre_send_signal, post_send_signal ...
673
674 Called before a signal is delivered; `alt_stack' indicates if it is
675 delivered on an alternative stack. */
676void VG_(track_pre_deliver_signal) (void(*f)(ThreadId tid, Int sigNo,
677 Bool alt_stack));
678/* Called after a signal is delivered. Nb: unfortunately, if the signal
679 handler longjmps, this won't be called. */
680void VG_(track_post_deliver_signal)(void(*f)(ThreadId tid, Int sigNo));
681
njn43b9a8a2005-05-10 04:37:01 +0000682#endif // __PUB_TOOL_TOOLIFACE_H
683
684/*--------------------------------------------------------------------*/
685/*--- end ---*/
686/*--------------------------------------------------------------------*/