• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 This file contains various notes/ideas/history/... related
2 to gdbserver in valgrind.
3 
4 How to use Valgrind gdbserver ?
5 -------------------------------
6 This is described in the Valgrind user manual.
7 Before reading the below, you better read the user manual first.
8 
9 What is gdbserver ?
10 -------------------
11 gdb debugger typically is used to debug a process running
12 on the same machine : gdb uses system calls (such as ptrace)
13 to fetch data from the process being debugged
14 or to change data in the process
15 or interrupt the process
16 or ...
17 
18 gdb can also debug processes running in a different computer
19 (e.g. it can debug a process running on a small real time
20 board).
21 
22 gdb does this by sending some commands (e.g. using tcp/ip) to a piece
23 of code running on the remote computer. This piece of code (called a
24 gdb stub in small boards, or gdbserver when the remote computer runs
25 an OS such as GNU/linux) will provide a set of commands allowing gdb
26 to remotely debug the process.  Examples of commands are: "get the
27 registers", "get the list of running threads", "read xxx bytes at
28 address yyyyyyyy", etc.  The definition of all these commands and the
29 associated replies is the gdb remote serial protocol, which is
30 documented in Appendix D of gdb user manual.
31 
32 The standard gdb distribution has a standalone gdbserver (a small
33 executable) which implements this protocol and the needed system calls
34 to allow gdb to remotely debug process running on a linux or MacOS or
35 ...
36 
37 Activation of gdbserver code inside valgrind
38 --------------------------------------------
39 The gdbserver code (from gdb 6.6, GPL2+) has been modified so as to
40 link it with valgrind and allow the valgrind guest process to be
41 debugged by a gdb speaking to this gdbserver embedded in valgrind.
42 The ptrace system calls inside gdbserver have been replaced by reading
43 the state of the guest.
44 
45 The gdbserver functionality is activated with valgrind command line
46 options. If gdbserver is not enabled, then the impact on valgrind
47 runtime is minimal: basically it just checks at startup the command
48 line option to see that there is nothing to do for what concerns gdb
49 server: there is a "if gdbserver is active" check in the translate
50 function of translate.c and an "if" in the valgrind scheduler.
51 If the valgrind gdbserver is activated (--vgdb=yes), the impact
52 is minimal (from time to time, the valgrind scheduler checks a counter
53 in memory). Option --vgdb-poll=yyyyy controls how often the scheduler
54 will do a (somewhat) more heavy check to see if gdbserver needs to
55 stop execution of the guest to allow debugging.
56 If valgrind gdbserver is activated with --vgdb=full, then
57 each instruction is instrumented with an additional call to a dirty
58 helper.
59 
60 How does gdbserver code interacts with valgrind ?
61 -------------------------------------------------
62 When an error is reported, the gdbserver code is called.  It reads
63 commands from gdb using read system call on a FIFO (e.g. a command
64 such as "get the registers").  It executes the command (e.g. fetches
65 the registers from the guest state) and writes the reply (e.g. a
66 packet containing the register data).  When gdb instructs gdbserver to
67 "continue", the control is returned to valgrind, which then continues
68 to execute guest code.  The FIFOs used to communication between
69 valgrind and gdb are created at startup if gdbserver is activated
70 according to the --vgdb=no/yes/full command line option.
71 
72 How are signals "handled" ?
73 ---------------------------
74 When a signal is to be given to the guest, valgrind core first calls
75 gdbserver (if a gdb is currently connected to valgrind, otherwise the
76 signal is delivered immediately). If gdb instructs to give the signal
77 to the process, the signal is delivered to the guest.  Otherwise, the
78 signal is ignored (not given to the guest). The user can
79 with gdb further decide to pass (or not pass) the signal.
80 Note that some (fatal) signals cannot be ignored.
81 
82 How are "break/step/stepi/next/..." implemented ?
83 -------------------------------------------------
84 When a break is put by gdb on an instruction, a command is sent to the
85 gdbserver in valgrind. This causes the basic block of this instruction
86 to be discarded and then re-instrumented so as to insert calls to a
87 dirty helper which calls the gdb server code.  When a block is
88 instrumented for gdbserver, all the "jump targets" of this block are
89 invalidated, so as to allow step/stepi/next to properly work: these
90 blocks will themselves automatically be re-instrumented for gdbserver
91 if they are jumped to.
92 The valgrind gdbserver remembers which blocks have been instrumented
93 due to this "lazy 'jump targets' debugging instrumentation" so as to
94 discard these "debugging translation" when gdb instructs to continue
95 the execution normally.
96 The blocks in which an explicit break has been put by the user
97 are kept instrumented for gdbserver.
98 (but note that by default, gdb removes all breaks when the
99 process is stopped, and re-inserts all breaks when the process
100 is continued). This behaviour can be changed using the gdb
101 command 'set breakpoint always-inserted'.
102 
103 How are watchpoints implemented ?
104 ---------------------------------
105 Watchpoints implies support from the tool to detect that
106 a location is read and/or written. Currently, only memcheck
107 supports this : when a watchpoint is placed, memcheck changes
108 the addressability bits of the watched memory zone to be unacessible.
109 Before an access, memcheck then detects an error, but sees this error
110 is due to a watchpoint and gives the control back to gdb.
111 Stopping on the exact instruction for a write watchpoint implies
112 to use --vgdb=full. This is because the error is detected by memcheck
113 before modifying the value. gdb checks that the value has not changed
114 and so "does not believe" the information that the write watchpoint
115 was triggered, and continues the execution. At the next watchpoint
116 occurence, gdb sees the value has changed. But the watchpoints are all
117 reported "off by one". To avoid this, Valgrind gdbserver must
118 terminate the current instruction before reporting the write watchpoint.
119 Terminating precisely the current instruction implies to have
120 instrumented all the instructions of the block for gdbserver even
121 if there is no break in this block. This is ensured by --vgdb=full.
122 See m_gdbserver.c Bool VG_(is_watched) where watchpoint handling
123 is implemented.
124 
125 How is the Valgrind gdbserver receiving commands/packets from gdb ?
126 -------------------------------------------------------------------
127 The embedded gdbserver reads gdb commands on a named pipe having
128 (by default) the name   /tmp/vgdb-pipe-from-vgdb-to-PID-by-USER-on-HOST
129 where PID, USER, and HOST will be replaced by the actual pid, the user id,
130 and the host name, respectively.
131 The embedded gdbserver will reply to gdb commands on a named pipe
132 /tmp/vgdb-pipe-to-vgdb-from-PID-by-USER-on-HOST
133 
134 gdb does not speak directly with gdbserver in valgrind: a relay application
135 called vgdb is needed between gdb and the valgrind-ified process.
136 gdb writes commands on the stdin of vgdb. vgdb reads these
137 commands and writes them on FIFO /tmp/vgdb-pipe-from-vgdb-to-PID-by-USER-on-HOST.
138 vgdb reads replies on FIFO /tmp/vgdb-pipe-to-vgdb-from-PID-by-USER-on-HOST
139 and writes them on its stdout.
140 
141 Note: The solution of named pipes was preferred to tcp ip connections as
142 it allows a discovery of which valgrind-ified processes are ready to accept
143 command by looking at files starting with the /tmp/vgdb-pipe- prefix
144 (changeable by a command line option).
145 Also, the usual unix protections are protecting
146 the valgrind process against other users sending commands.
147 The relay process also takes into account the wake up of the valgrind
148 process in case all threads are blocked in a system call.
149 The relay process can also be used in a shell to send commands
150 without a gdb (this allows to have a standard mechanism to control
151 valgrind tools from the command line, rather than specialized mechanism
152 e.g. in callgrind).
153 
154 How is gdbserver activated if all Valgrind threads are blocked in a syscall ?
155 -----------------------------------------------------------------------------
156 vgdb relays characters from gdb to valgrind. The scheduler will from
157 time to time check if gdbserver has to handle incoming characters.
158 (the check is efficient i.e. most of the time consists in checking
159 a counter in (shared) memory).
160 
161 However, it might be that all the threads in the valgrind process are
162 blocked in a system call. In such a case, no polling will be done by
163 the valgrind scheduler (as no activity takes place).  By default, vgdb
164 will check after 100ms if the characters it has written have been read
165 by valgrind. If not, vgdb will force the invocation of the gdbserver
166 code inside the valgrind process.
167 
168 On Linux, this forced invocation is implemented using the ptrace system call:
169 using ptrace, vgdb will cause the valgrind process to call the
170 gdbserver code.
171 
172 This wake up is *not* done using signals as this would imply to
173 implement a syscall restart logic in valgrind for all system
174 calls. When using ptrace as above, the linux kernel is responsible to
175 restart the system call.
176 
177 This wakeup is also *not* implemented by having a "system thread"
178 started by valgrind as this would transform all non-threaded programs
179 in threaded programs when running under valgrind. Also, such a 'system
180 thread' for gdbserver was tried by Greg Parker in the early MacOS
181 port, and was unreliable.
182 
183 So, the ptrace based solution was chosen instead.
184 
185 There used to be some bugs in the kernel when using ptrace on
186 a process blocked in a system call : the symptom is that the system
187 call fails with an unknown errno 512. This typically happens
188 with a vgdb in 64bits ptrace-ing a 32 bits process.
189 A bypass for old kernels has been integrated in vgdb.c (sign extend
190 register rax).
191 
192 At least on a fedora core 12 (kernel 2.6.32), syscall restart of read
193 and select are working ok and red-hat 5.3 (an old kernel), everything
194 works properly.
195 
196 Need to investigate if darwin can similarly do syscall
197 restart with ptrace.
198 
199 The vgdb argument --max-invoke-ms=xxx allows to control the nr of
200 milli-seconds after which vgdb will force the invocation of gdbserver
201 code.  If xxx is 0, this disables the forced invocation.
202 Also, disabling this ptrace mechanism is necessary in case you are
203 debugging the valgrind code at the same time as debugging the guest
204 process using gdbserver.
205 
206 Do not kill -9 vgdb while it has interrupted the valgrind process,
207 otherwise the valgrind process will very probably stay stopped or die.
208 
209 
210 Implementation is based on the gdbserver code from gdb 6.6
211 ----------------------------------------------------------
212 The gdbserver implementation is derived from the gdbserver included
213 in the gdb distribution.
214 The files originating from gdb are : inferiors.c, regcache.[ch],
215 regdef.h, remote-utils.c, server.[ch], signals.c, target.[ch], utils.c,
216 version.c.
217 valgrind-low-* are inspired from gdb files.
218 
219 This code had to be changed to integrate properly within valgrind
220 (e.g. no libc usage).  Some of these changes have been ensured by
221 using the preprocessor to replace calls by valgrind equivalent,
222 e.g. #define strcmp(...) VG_(strcmp) (...).
223 
224 Some "control flow" changes are due to the fact that gdbserver inside
225 valgrind must return the control to valgrind when the 'debugged'
226 process has to run, while in a classical gdbserver usage, the
227 gdbserver process waits for a debugged process to stop on a break or
228 similar.  This has implied to have some variables to remember the
229 state of gdbserver before returning to valgrind (search for
230 resume_packet_needed in server.c) and "goto" the place where gdbserver
231 expects a stopped process to return control to gdbserver.
232 
233 How does a tool need to be changed to be "debuggable" ?
234 -------------------------------------------------------
235 There is no need to modify a tool to have it "debuggable" via
236 gdbserver : e.g. reports of errors, break etc will work "out of the
237 box".  If an interactive usage of tool client requests or similar is
238 desired for a tool, then simple code can be written for that via a
239 specific client request VG_USERREQ__GDB_MONITOR_COMMAND code. The tool
240 function "handle_client_request" must then parse the string received
241 in argument and call the expected valgrind or tool code.  See
242 e.g. massif ms_handle_client_request as an example.
243 
244 
245 Automatic regression tests:
246 ---------------------------
247 Automatic Valgrind gdbserver tests are in the directory
248 $(top_srcdir)/gdbserver_tests.
249 Read $(top_srcdir)/gdbserver_tests/README_DEVELOPERS for more
250 info about testing.
251 
252 How to integrate support for a new architecture xxx?
253 ----------------------------------------------------
254 Let's imagine a new architecture hal9000 has to be supported.
255 
256 Mandatory:
257 The main thing to do is to make a file valgrind-low-hal9000.c.
258 Start from an existing file (e.g. valgrind-low-x86.c).
259 The data structures 'struct reg regs'
260 and 'const char *expedite_regs' are build from files
261 in the gdb sources, e.g. for an new arch hal9000
262    cd gdb/regformats
263    sh ./regdat.sh reg-hal9000.dat hal9000
264 
265 From the generated file hal9000, you copy/paste in
266 valgrind-low-hal9000.c the two needed data structures and change their
267 name to 'regs' and 'expedite_regs'
268 
269 Then adapt the set of functions needed to initialize the structure
270 'static struct valgrind_target_ops low_target'.
271 
272 Optional but heavily recommended:
273 To have a proper wake up of a Valgrind process with all threads
274 blocked in a system call, some architecture specific code
275 has to be done in vgdb-invoker-*.c.
276 Typically, for a linux system supporting ptrace, you have to modify
277 vgdb-invoker-ptrace.c.
278 
279 For Linux based platforms, all the ptrace calls in vgdb-invoker-ptrace.c
280 should be ok.
281 The only thing needed is the code needed to "push a dummy call" on the stack,
282 i.e. assign the relevant registers in the struct user_regs_struct, and push
283 values on the stack according to the ABI.
284 
285 For other platforms (i.e. Macos), more work is needed as the ptrace calls
286 on Macos are either different and/or incomplete (and so, 'Mach' specific
287 things are needed e.g. to attach to threads etc).
288 A courageous Mac aficionado is welcome on this aspect.
289 
290 Optional:
291 To let gdb see the Valgrind shadow registers, xml description
292 files have to be provided + valgrind-low-hal9000.c has
293 to give the top xml file.
294 Start from the xml files found in the gdb distribution directory
295 gdb/features. You need to duplicate and modify these files to provide
296 shadow1 and shadow2 register sets description.
297 
298 Modify coregrind/Makefile.am:
299     add valgrind-low-hal9000.c
300     If you have target xml description, also add them to GDBSERVER_XML_FILES
301 
302 
303 TODO and/or additional nice things to have
304 ------------------------------------------
305 * many options can be changed on-line without problems.
306   => would be nice to have a v.option command that would evaluate
307   its arguments like the  startup options of m_main.c and tool clo processing.
308 
309 * have a memcheck monitor command
310   show_dangling_pointers [last_n_recently_released_blocks]
311   showing which of the n last recently released blocks are still
312   referenced. These references are (potential) dangling pointers.
313 
314 * some GDBTD in the code
315 
316 (GDBTD = GDB To Do = something still to look at and/or a question)
317 
318 * All architectures and platforms are done.
319   But there are still some "GDBTD" to convert between gdb registers
320   and VEX registers :
321   e.g. some registers in x86 or amd64 that I could not
322   translate to VEX registers. Someone with a good knowledge
323   of these architectures might complete this
324   (see the GDBTD in valgrind-low-*.c)
325 
326 * Currently, at least on recent linux kernel, vgdb can properly wake
327   up a valgrind process which is blocked in system calls. Maybe we
328   need to see till which kernel version the ptrace + syscall restart
329   is broken, and put the default value of --max-invoke-ms to 0 in this
330   case.
331 
332 * more client requests can be programmed in various tools.  Currently,
333   there are only a few standard valgrind or memcheck client requests
334   implemented.
335   v.suppression [generate|add|delete] might be an interesting command:
336      generate would output a suppression, add/delete would add a suppression
337      in memory for the last (or selected?) error.
338   v.break on fn calls/entry/exit + commands associated to it
339     (such as search leaks)?
340 
341 
342 * currently jump(s) and inferior call(s) are somewhat dangerous
343   when called from a block not yet instrumented : instead
344   of continuing till the next Imark, where there will be a
345   debugger call that can properly jump at an instruction boundary,
346   the jump/call will quit the "middle" of an instruction.
347   We could detect if the current block is instrumented by a trick
348   like this:
349      /* Each time helperc_CallDebugger is called, we will store
350         the address from which is it called and the nr of bbs_done
351         when called. This allows to detect that gdbserver is called
352         from a block which is instrumented. */
353      static HWord CallDebugger_addr;
354      static ULong CallDebugger_bbs_done;
355 
356      Bool VG_(gdbserver_current_IP_instrumented) (ThreadId tid)
357      {
358         if (VG_(get_IP) (tid) != CallDebugger_addr
359             || CallDebugger_bbs_done != VG_(bbs_done)())
360            return False;
361         return True;
362      }
363 
364   Alternatively, we ensure we can re-instrument the current
365   block for gdbserver while executing it.
366   Something like:
367   keep current block till the end of the current instruction, then
368   go back to scheduler.
369   Unsure if and how this is do-able.
370 
371 
372 * ensure that all non static symbols of gdbserver files are #define
373   xxxxx VG_(xxxxx) ???? Is this really needed ? I have tried to put in
374   a test program variables and functions with the same name as valgrind
375   stuff, and everything seems to be ok.
376   I see that all exported symbols in valgrind have a unique prefix
377   created with VG_ or MC_ or ...
378   This is not done for the "gdb gdbserver code", where I have kept
379   the original names. Is this a problem ? I could not create
380   a "symbol" collision between the user symbol and the valgrind
381   core gdbserver symbol.
382 
383 * currently, gdbserver can only stop/continue the whole process. It
384   might be interesting to have a fine-grained thread control (vCont
385   packet) maybe for tools such as helgrind, drd.  This would allow the
386   user to stop/resume specific threads.  Also, maybe this would solve
387   the following problem: wait for a breakpoint to be encountered,
388   switch thread, next. This sometimes causes an internal error in gdb,
389   probably because gdb believes the current thread will be continued ?
390 
391 * would be nice to have some more tests.
392 
393 * better valgrind target support in gdb (see comments of Tom Tromey).
394 
395 
396 -------- description of how gdb invokes a function in the inferior
397 to call a function in the inferior (below is for x86):
398 gdb writes ESP and EBP to have some more stack space
399 push a return address equal to  0x8048390 <_start>
400 puts a break                at  0x8048390
401 put address of the function to call (e.g. hello_world in EIP (0x8048444))
402 continue
403 break encountered at 0x8048391 (90 after decrement)
404   => report stop to gdb
405   => gdb restores esp/ebp/eip to what it was (eg. 0x804848C)
406   => gdb "s" => causes the EIP to go to the new EIP (i.e. 0x804848C)
407      gdbserver tells "resuming from 0x804848c"
408                      "stop pc is 0x8048491" => informed gdb of this
409 
410