1 Objtool
2 =======
3
4 The kernel CONFIG_OBJTOOL option enables a host tool named 'objtool'
5 which runs at compile time. It can do various validations and
6 transformations on .o files.
7
8 Objtool has become an integral part of the x86-64 kernel toolchain. The
9 kernel depends on it for a variety of security and performance features
10 (and other types of features as well).
11
12
13 Features
14 --------
15
16 Objtool has the following features:
17
18 - Stack unwinding metadata validation -- useful for helping to ensure
19 stack traces are reliable for live patching
20
21 - ORC unwinder metadata generation -- a faster and more precise
22 alternative to frame pointer based unwinding
23
24 - Retpoline validation -- ensures that all indirect calls go through
25 retpoline thunks, for Spectre v2 mitigations
26
27 - Retpoline call site annotation -- annotates all retpoline thunk call
28 sites, enabling the kernel to patch them inline, to prevent "thunk
29 funneling" for both security and performance reasons
30
31 - Non-instrumentation validation -- validates non-instrumentable
32 ("noinstr") code rules, preventing instrumentation in low-level C
33 entry code
34
35 - Static call annotation -- annotates static call sites, enabling the
36 kernel to implement inline static calls, a faster alternative to some
37 indirect branches
38
39 - Uaccess validation -- validates uaccess rules for a proper
40 implementation of Supervisor Mode Access Protection (SMAP)
41
42 - Straight Line Speculation validation -- validates certain SLS
43 mitigations
44
45 - Indirect Branch Tracking validation -- validates Intel CET IBT rules
46 to ensure that all functions referenced by function pointers have
47 corresponding ENDBR instructions
48
49 - Indirect Branch Tracking annotation -- annotates unused ENDBR
50 instruction sites, enabling the kernel to "seal" them (replace them
51 with NOPs) to further harden IBT
52
53 - Function entry annotation -- annotates function entries, enabling
54 kernel function tracing
55
56 - Other toolchain hacks which will go unmentioned at this time...
57
58 Each feature can be enabled individually or in combination using the
59 objtool cmdline.
60
61
62 Objects
63 -------
64
65 Typically, objtool runs on every translation unit (TU, aka ".o file") in
66 the kernel. If a TU is part of a kernel module, the '--module' option
67 is added.
68
69 However:
70
71 - If noinstr validation is enabled, it also runs on vmlinux.o, with all
72 options removed and '--noinstr' added.
73
74 - If IBT or LTO is enabled, it doesn't run on TUs at all. Instead it
75 runs on vmlinux.o and linked modules, with all options.
76
77 In summary:
78
79 A) Legacy mode:
80 TU: objtool [--module] <options>
81 vmlinux: N/A
82 module: N/A
83
84 B) CONFIG_NOINSTR_VALIDATION=y && !(CONFIG_X86_KERNEL_IBT=y || CONFIG_LTO=y):
85 TU: objtool [--module] <options> // no --noinstr
86 vmlinux: objtool --noinstr // other options removed
87 module: N/A
88
89 C) CONFIG_X86_KERNEL_IBT=y || CONFIG_LTO=y:
90 TU: N/A
91 vmlinux: objtool --noinstr <options>
92 module: objtool --module --noinstr <options>
93
94
95 Stack validation
96 ----------------
97
98 Objtool's stack validation feature analyzes every .o file and ensures
99 the validity of its stack metadata. It enforces a set of rules on asm
100 code and C inline assembly code so that stack traces can be reliable.
101
102 For each function, it recursively follows all possible code paths and
103 validates the correct frame pointer state at each instruction.
104
105 It also follows code paths involving special sections, like
106 .altinstructions, __jump_table, and __ex_table, which can add
107 alternative execution paths to a given instruction (or set of
108 instructions). Similarly, it knows how to follow switch statements, for
109 which gcc sometimes uses jump tables.
110
111 Here are some of the benefits of validating stack metadata:
112
113 a) More reliable stack traces for frame pointer enabled kernels
114
115 Frame pointers are used for debugging purposes. They allow runtime
116 code and debug tools to be able to walk the stack to determine the
117 chain of function call sites that led to the currently executing
118 code.
119
120 For some architectures, frame pointers are enabled by
121 CONFIG_FRAME_POINTER. For some other architectures they may be
122 required by the ABI (sometimes referred to as "backchain pointers").
123
124 For C code, gcc automatically generates instructions for setting up
125 frame pointers when the -fno-omit-frame-pointer option is used.
126
127 But for asm code, the frame setup instructions have to be written by
128 hand, which most people don't do. So the end result is that
129 CONFIG_FRAME_POINTER is honored for C code but not for most asm code.
130
131 For stack traces based on frame pointers to be reliable, all
132 functions which call other functions must first create a stack frame
133 and update the frame pointer. If a first function doesn't properly
134 create a stack frame before calling a second function, the *caller*
135 of the first function will be skipped on the stack trace.
136
137 For example, consider the following example backtrace with frame
138 pointers enabled:
139
140 [<ffffffff81812584>] dump_stack+0x4b/0x63
141 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30
142 [<ffffffff8127f568>] seq_read+0x108/0x3e0
143 [<ffffffff812cce62>] proc_reg_read+0x42/0x70
144 [<ffffffff81256197>] __vfs_read+0x37/0x100
145 [<ffffffff81256b16>] vfs_read+0x86/0x130
146 [<ffffffff81257898>] SyS_read+0x58/0xd0
147 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76
148
149 It correctly shows that the caller of cmdline_proc_show() is
150 seq_read().
151
152 If we remove the frame pointer logic from cmdline_proc_show() by
153 replacing the frame pointer related instructions with nops, here's
154 what it looks like instead:
155
156 [<ffffffff81812584>] dump_stack+0x4b/0x63
157 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30
158 [<ffffffff812cce62>] proc_reg_read+0x42/0x70
159 [<ffffffff81256197>] __vfs_read+0x37/0x100
160 [<ffffffff81256b16>] vfs_read+0x86/0x130
161 [<ffffffff81257898>] SyS_read+0x58/0xd0
162 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76
163
164 Notice that cmdline_proc_show()'s caller, seq_read(), has been
165 skipped. Instead the stack trace seems to show that
166 cmdline_proc_show() was called by proc_reg_read().
167
168 The benefit of objtool here is that because it ensures that *all*
169 functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be
170 skipped on a stack trace.
171
172 [*] unless an interrupt or exception has occurred at the very
173 beginning of a function before the stack frame has been created,
174 or at the very end of the function after the stack frame has been
175 destroyed. This is an inherent limitation of frame pointers.
176
177 b) ORC (Oops Rewind Capability) unwind table generation
178
179 An alternative to frame pointers and DWARF, ORC unwind data can be
180 used to walk the stack. Unlike frame pointers, ORC data is out of
181 band. So it doesn't affect runtime performance and it can be
182 reliable even when interrupts or exceptions are involved.
183
184 For more details, see Documentation/arch/x86/orc-unwinder.rst.
185
186 c) Higher live patching compatibility rate
187
188 Livepatch has an optional "consistency model", which is needed for
189 more complex patches. In order for the consistency model to work,
190 stack traces need to be reliable (or an unreliable condition needs to
191 be detectable). Objtool makes that possible.
192
193 For more details, see the livepatch documentation in the Linux kernel
194 source tree at Documentation/livepatch/livepatch.rst.
195
196 To achieve the validation, objtool enforces the following rules:
197
198 1. Each callable function must be annotated as such with the ELF
199 function type. In asm code, this is typically done using the
200 ENTRY/ENDPROC macros. If objtool finds a return instruction
201 outside of a function, it flags an error since that usually indicates
202 callable code which should be annotated accordingly.
203
204 This rule is needed so that objtool can properly identify each
205 callable function in order to analyze its stack metadata.
206
207 2. Conversely, each section of code which is *not* callable should *not*
208 be annotated as an ELF function. The ENDPROC macro shouldn't be used
209 in this case.
210
211 This rule is needed so that objtool can ignore non-callable code.
212 Such code doesn't have to follow any of the other rules.
213
214 3. Each callable function which calls another function must have the
215 correct frame pointer logic, if required by CONFIG_FRAME_POINTER or
216 the architecture's back chain rules. This can by done in asm code
217 with the FRAME_BEGIN/FRAME_END macros.
218
219 This rule ensures that frame pointer based stack traces will work as
220 designed. If function A doesn't create a stack frame before calling
221 function B, the _caller_ of function A will be skipped on the stack
222 trace.
223
224 4. Dynamic jumps and jumps to undefined symbols are only allowed if:
225
226 a) the jump is part of a switch statement; or
227
228 b) the jump matches sibling call semantics and the frame pointer has
229 the same value it had on function entry.
230
231 This rule is needed so that objtool can reliably analyze all of a
232 function's code paths. If a function jumps to code in another file,
233 and it's not a sibling call, objtool has no way to follow the jump
234 because it only analyzes a single file at a time.
235
236 5. A callable function may not execute kernel entry/exit instructions.
237 The only code which needs such instructions is kernel entry code,
238 which shouldn't be be in callable functions anyway.
239
240 This rule is just a sanity check to ensure that callable functions
241 return normally.
242
243
244 Objtool warnings
245 ----------------
246
247 NOTE: When requesting help with an objtool warning, please recreate with
248 OBJTOOL_VERBOSE=1 (e.g., "make OBJTOOL_VERBOSE=1") and send the full
249 output, including any disassembly or backtrace below the warning, to the
250 objtool maintainers.
251
252 For asm files, if you're getting an error which doesn't make sense,
253 first make sure that the affected code follows the above rules.
254
255 For C files, the common culprits are inline asm statements and calls to
256 "noreturn" functions. See below for more details.
257
258 Another possible cause for errors in C code is if the Makefile removes
259 -fno-omit-frame-pointer or adds -fomit-frame-pointer to the gcc options.
260
261 Here are some examples of common warnings reported by objtool, what
262 they mean, and suggestions for how to fix them. When in doubt, ping
263 the objtool maintainers.
264
265
266 1. file.o: warning: objtool: func()+0x128: call without frame pointer save/setup
267
268 The func() function made a function call without first saving and/or
269 updating the frame pointer, and CONFIG_FRAME_POINTER is enabled.
270
271 If the error is for an asm file, and func() is indeed a callable
272 function, add proper frame pointer logic using the FRAME_BEGIN and
273 FRAME_END macros. Otherwise, if it's not a callable function, remove
274 its ELF function annotation by changing ENDPROC to END, and instead
275 use the manual unwind hint macros in asm/unwind_hints.h.
276
277 If it's a GCC-compiled .c file, the error may be because the function
278 uses an inline asm() statement which has a "call" instruction. An
279 asm() statement with a call instruction must declare the use of the
280 stack pointer in its output operand. On x86_64, this means adding
281 the ASM_CALL_CONSTRAINT as an output constraint:
282
283 asm volatile("call func" : ASM_CALL_CONSTRAINT);
284
285 Otherwise the stack frame may not get created before the call.
286
287 objtool can help with pinpointing the exact function where it happens:
288
289 $ OBJTOOL_ARGS="--verbose" make arch/x86/kvm/
290
291 arch/x86/kvm/kvm.o: warning: objtool: .altinstr_replacement+0xc5: call without frame pointer save/setup
292 arch/x86/kvm/kvm.o: warning: objtool: em_loop.part.0+0x29: (alt)
293 arch/x86/kvm/kvm.o: warning: objtool: em_loop.part.0+0x0: <=== (sym)
294 LD [M] arch/x86/kvm/kvm-intel.o
295 0000 0000000000028220 <em_loop.part.0>:
296 0000 28220: 0f b6 47 61 movzbl 0x61(%rdi),%eax
297 0004 28224: 3c e2 cmp $0xe2,%al
298 0006 28226: 74 2c je 28254 <em_loop.part.0+0x34>
299 0008 28228: 48 8b 57 10 mov 0x10(%rdi),%rdx
300 000c 2822c: 83 f0 05 xor $0x5,%eax
301 000f 2822f: 48 c1 e0 04 shl $0x4,%rax
302 0013 28233: 25 f0 00 00 00 and $0xf0,%eax
303 0018 28238: 81 e2 d5 08 00 00 and $0x8d5,%edx
304 001e 2823e: 80 ce 02 or $0x2,%dh
305 ...
306
307 2. file.o: warning: objtool: .text+0x53: unreachable instruction
308
309 Objtool couldn't find a code path to reach the instruction.
310
311 If the error is for an asm file, and the instruction is inside (or
312 reachable from) a callable function, the function should be annotated
313 with the ENTRY/ENDPROC macros (ENDPROC is the important one).
314 Otherwise, the code should probably be annotated with the unwind hint
315 macros in asm/unwind_hints.h so objtool and the unwinder can know the
316 stack state associated with the code.
317
318 If you're 100% sure the code won't affect stack traces, or if you're
319 a just a bad person, you can tell objtool to ignore it. See the
320 "Adding exceptions" section below.
321
322 If it's not actually in a callable function (e.g. kernel entry code),
323 change ENDPROC to END.
324
325 3. file.o: warning: objtool: foo+0x48c: bar() is missing a __noreturn annotation
326
327 The call from foo() to bar() doesn't return, but bar() is missing the
328 __noreturn annotation. NOTE: In addition to annotating the function
329 with __noreturn, please also add it to tools/objtool/noreturns.h.
330
331 4. file.o: warning: objtool: func(): can't find starting instruction
332 or
333 file.o: warning: objtool: func()+0x11dd: can't decode instruction
334
335 Does the file have data in a text section? If so, that can confuse
336 objtool's instruction decoder. Move the data to a more appropriate
337 section like .data or .rodata.
338
339
340 5. file.o: warning: objtool: func()+0x6: unsupported instruction in callable function
341
342 This is a kernel entry/exit instruction like sysenter or iret. Such
343 instructions aren't allowed in a callable function, and are most
344 likely part of the kernel entry code. They should usually not have
345 the callable function annotation (ENDPROC) and should always be
346 annotated with the unwind hint macros in asm/unwind_hints.h.
347
348
349 6. file.o: warning: objtool: func()+0x26: sibling call from callable instruction with modified stack frame
350
351 This is a dynamic jump or a jump to an undefined symbol. Objtool
352 assumed it's a sibling call and detected that the frame pointer
353 wasn't first restored to its original state.
354
355 If it's not really a sibling call, you may need to move the
356 destination code to the local file.
357
358 If the instruction is not actually in a callable function (e.g.
359 kernel entry code), change ENDPROC to END and annotate manually with
360 the unwind hint macros in asm/unwind_hints.h.
361
362
363 7. file: warning: objtool: func()+0x5c: stack state mismatch
364
365 The instruction's frame pointer state is inconsistent, depending on
366 which execution path was taken to reach the instruction.
367
368 Make sure that, when CONFIG_FRAME_POINTER is enabled, the function
369 pushes and sets up the frame pointer (for x86_64, this means rbp) at
370 the beginning of the function and pops it at the end of the function.
371 Also make sure that no other code in the function touches the frame
372 pointer.
373
374 Another possibility is that the code has some asm or inline asm which
375 does some unusual things to the stack or the frame pointer. In such
376 cases it's probably appropriate to use the unwind hint macros in
377 asm/unwind_hints.h.
378
379
380 8. file.o: warning: objtool: funcA() falls through to next function funcB()
381
382 This means that funcA() doesn't end with a return instruction or an
383 unconditional jump, and that objtool has determined that the function
384 can fall through into the next function. There could be different
385 reasons for this:
386
387 1) funcA()'s last instruction is a call to a "noreturn" function like
388 panic(). In this case the noreturn function needs to be added to
389 objtool's hard-coded global_noreturns array. Feel free to bug the
390 objtool maintainer, or you can submit a patch.
391
392 2) funcA() uses the unreachable() annotation in a section of code
393 that is actually reachable.
394
395 3) If funcA() calls an inline function, the object code for funcA()
396 might be corrupt due to a gcc bug. For more details, see:
397 https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70646
398
399 9. file.o: warning: objtool: funcA() call to funcB() with UACCESS enabled
400
401 This means that an unexpected call to a non-whitelisted function exists
402 outside of arch-specific guards.
403 X86: SMAP (stac/clac): __uaccess_begin()/__uaccess_end()
404 ARM: PAN: uaccess_enable()/uaccess_disable()
405
406 These functions should be called to denote a minimal critical section around
407 access to __user variables. See also: https://lwn.net/Articles/517475/
408
409 The intention of the warning is to prevent calls to funcB() from eventually
410 calling schedule(), potentially leaking the AC flags state, and not
411 restoring them correctly.
412
413 It also helps verify that there are no unexpected calls to funcB() which may
414 access user space pages with protections against doing so disabled.
415
416 To fix, either:
417 1) remove explicit calls to funcB() from funcA().
418 2) add the correct guards before and after calls to low level functions like
419 __get_user_size()/__put_user_size().
420 3) add funcB to uaccess_safe_builtin whitelist in tools/objtool/check.c, if
421 funcB obviously does not call schedule(), and is marked notrace (since
422 function tracing inserts additional calls, which is not obvious from the
423 sources).
424
425 10. file.o: warning: func()+0x5c: stack layout conflict in alternatives
426
427 This means that in the use of the alternative() or ALTERNATIVE()
428 macro, the code paths have conflicting modifications to the stack.
429 The problem is that there is only one ORC unwind table, which means
430 that the ORC unwind entries must be consistent for all possible
431 instruction boundaries regardless of which code has been patched.
432 This limitation can be overcome by massaging the alternatives with
433 NOPs to shift the stack changes around so they no longer conflict.
434
435 11. file.o: warning: unannotated intra-function call
436
437 This warning means that a direct call is done to a destination which
438 is not at the beginning of a function. If this is a legit call, you
439 can remove this warning by putting the ANNOTATE_INTRA_FUNCTION_CALL
440 directive right before the call.
441
442 12. file.o: warning: func(): not an indirect call target
443
444 This means that objtool is running with --ibt and a function expected
445 to be an indirect call target is not. In particular, this happens for
446 init_module() or cleanup_module() if a module relies on these special
447 names and does not use module_init() / module_exit() macros to create
448 them.
449
450
451 If the error doesn't seem to make sense, it could be a bug in objtool.
452 Feel free to ask the objtool maintainer for help.
453
454
455 Adding exceptions
456 -----------------
457
458 If you _really_ need objtool to ignore something, and are 100% sure
459 that it won't affect kernel stack traces, you can tell objtool to
460 ignore it:
461
462 - To skip validation of a function, use the STACK_FRAME_NON_STANDARD
463 macro.
464
465 - To skip validation of a file, add
466
467 OBJECT_FILES_NON_STANDARD_filename.o := y
468
469 to the Makefile.
470
471 - To skip validation of a directory, add
472
473 OBJECT_FILES_NON_STANDARD := y
474
475 to the Makefile.
476
477 NOTE: OBJECT_FILES_NON_STANDARD doesn't work for link time validation of
478 vmlinux.o or a linked module. So it should only be used for files which
479 aren't linked into vmlinux or a module.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.