2022-01-14 01:05:43

by Nick Desaulniers

[permalink] [raw]
Subject: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

commit dcce50e6cc4d ("compiler.h: Fix annotation macro misplacement with Clang")
mentions:

> 'volatile' is ignored for some reason and Clang feels free to move the
> reachable() annotation away from its intended location.

Indeed, volatile is not a compiler barrier. Particularly once `-march=`
flags are used under certain configs, LLVM's machine-scheduler can be
observed moving instructions across the asm statement meant to point to
known reachable or unreachable code, as reported by 0day bot.

Prefer a memory clobber which is a compiler barrier that prevents these
re-orderings and remove the volatile qualifier.

Looking closer, the use of __COUNTER__ seems to have been used to
prevent de-duplication of these asm statements. The GCC manual mentions:

> Under certain circumstances, GCC may duplicate (or remove duplicates
> of) your assembly code when optimizing. This can lead to unexpected
> duplicate symbol errors during compilation if your asm code defines
> symbols or labels. Using ‘%=’ (see AssemblerTemplate) may help resolve
> this problem.
>
> ‘%=’ Outputs a number that is unique to each instance of the asm
> statement in the entire compilation. This option is useful when
> creating local labels and referring to them multiple times in a single
> template that generates multiple assembler instructions.

commit 3d1e236022cc ("objtool: Prevent GCC from merging annotate_unreachable()")

Mentions that

> The inline asm ‘%=’ token could be used for that, but unfortunately
> older versions of GCC don't support it.

From testing all versions of GCC available on godbolt.org, GCC 4.1+
seems to support 4.1. Since the minimum supported version of GCC at the
moment is GCC 5.1, it sounds like this is no longer a concern.

Prefer the %= assembler template to having to stringify __COUNTER__.

This commit is effectively a revert of the following commits:
commit dcce50e6cc4d ("compiler.h: Fix annotation macro misplacement with Clang")
commit f1069a8756b9 ("compiler.h: Avoid using inline asm operand modifiers")
commit c199f64ff93c ("instrumentation.h: Avoid using inline asm operand modifiers")
commit d0c2e691d1cb ("objtool: Add a comment for the unreachable annotation macros")
commit ec1e1b610917 ("objtool: Prevent GCC from merging annotate_unreachable(), take 2")
commit 3d1e236022cc ("objtool: Prevent GCC from merging annotate_unreachable()")

Link: https://github.com/ClangBuiltLinux/linux/issues/1566
Link: https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Volatile
Link: https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#AssemblerTemplate
Link: https://lore.kernel.org/llvm/[email protected]/
Link: https://lore.kernel.org/llvm/[email protected]/
Reported-by: kernel test robot <[email protected]>
Debugged-by: Nathan Chancellor <[email protected]>
Signed-off-by: Nick Desaulniers <[email protected]>
---
include/linux/compiler.h | 31 +++++++++++--------------------
include/linux/instrumentation.h | 24 ++++++++++--------------
2 files changed, 21 insertions(+), 34 deletions(-)

diff --git a/include/linux/compiler.h b/include/linux/compiler.h
index 429dcebe2b99..3ac21b888d20 100644
--- a/include/linux/compiler.h
+++ b/include/linux/compiler.h
@@ -108,30 +108,21 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
# define barrier_before_unreachable() do { } while (0)
#endif

-/* Unreachable code */
+/* These macros help objtool understand GCC code flow for unreachable code. */
#ifdef CONFIG_STACK_VALIDATION
-/*
- * These macros help objtool understand GCC code flow for unreachable code.
- * The __COUNTER__ based labels are a hack to make each instance of the macros
- * unique, to convince GCC not to merge duplicate inline asm statements.
- */
-#define __stringify_label(n) #n
-
-#define __annotate_reachable(c) ({ \
- asm volatile(__stringify_label(c) ":\n\t" \
- ".pushsection .discard.reachable\n\t" \
- ".long " __stringify_label(c) "b - .\n\t" \
- ".popsection\n\t" : : "i" (c)); \
+#define annotate_reachable() ({ \
+ asm (".Lreachable%=:\n\t" \
+ ".pushsection .discard.reachable\n\t" \
+ ".long .Lreachable%= - .\n\t" \
+ ".popsection\n\t" ::: "memory"); \
})
-#define annotate_reachable() __annotate_reachable(__COUNTER__)

-#define __annotate_unreachable(c) ({ \
- asm volatile(__stringify_label(c) ":\n\t" \
- ".pushsection .discard.unreachable\n\t" \
- ".long " __stringify_label(c) "b - .\n\t" \
- ".popsection\n\t" : : "i" (c)); \
+#define annotate_unreachable() ({ \
+ asm (".Lunreachable%=:\n\t" \
+ ".pushsection .discard.unreachable\n\t" \
+ ".long .Lunreachable%= - .\n\t" \
+ ".popsection\n\t" ::: "memory"); \
})
-#define annotate_unreachable() __annotate_unreachable(__COUNTER__)

#define ASM_UNREACHABLE \
"999:\n\t" \
diff --git a/include/linux/instrumentation.h b/include/linux/instrumentation.h
index 24359b4a9605..0dae9c08764f 100644
--- a/include/linux/instrumentation.h
+++ b/include/linux/instrumentation.h
@@ -4,16 +4,13 @@

#if defined(CONFIG_DEBUG_ENTRY) && defined(CONFIG_STACK_VALIDATION)

-#include <linux/stringify.h>
-
/* Begin/end of an instrumentation safe region */
-#define __instrumentation_begin(c) ({ \
- asm volatile(__stringify(c) ": nop\n\t" \
- ".pushsection .discard.instr_begin\n\t" \
- ".long " __stringify(c) "b - .\n\t" \
- ".popsection\n\t" : : "i" (c)); \
+#define instrumentation_begin() ({ \
+ asm (".Linstbegin%=: nop\n\t" \
+ ".pushsection .discard.instr_begin\n\t" \
+ ".long .Linstbegin%= - .\n\t" \
+ ".popsection\n\t" ::: "memory"); \
})
-#define instrumentation_begin() __instrumentation_begin(__COUNTER__)

/*
* Because instrumentation_{begin,end}() can nest, objtool validation considers
@@ -46,13 +43,12 @@
* To avoid this, have _end() be a NOP instruction, this ensures it will be
* part of the condition block and does not escape.
*/
-#define __instrumentation_end(c) ({ \
- asm volatile(__stringify(c) ": nop\n\t" \
- ".pushsection .discard.instr_end\n\t" \
- ".long " __stringify(c) "b - .\n\t" \
- ".popsection\n\t" : : "i" (c)); \
+#define instrumentation_end() ({ \
+ asm (".Linstend%=: nop\n\t" \
+ ".pushsection .discard.instr_end\n\t" \
+ ".long .Linstend%= - .\n\t" \
+ ".popsection\n\t" ::: "memory"); \
})
-#define instrumentation_end() __instrumentation_end(__COUNTER__)
#else
# define instrumentation_begin() do { } while(0)
# define instrumentation_end() do { } while(0)

base-commit: dcce50e6cc4d86a63dc0a9a6ee7d4f948ccd53a1
--
2.34.1.703.g22d0c6ccf7-goog



2022-01-14 23:09:39

by Nathan Chancellor

[permalink] [raw]
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

On Thu, Jan 13, 2022 at 05:05:26PM -0800, Nick Desaulniers wrote:
> commit dcce50e6cc4d ("compiler.h: Fix annotation macro misplacement with Clang")
> mentions:
>
> > 'volatile' is ignored for some reason and Clang feels free to move the
> > reachable() annotation away from its intended location.

Perhaps it is worth mentioning in the next paragraph that this does not
contradict GCC's own documentation that you link to below, which
mentions that asm volatile statements can be reordered.

"Note that the compiler can move even volatile asm instructions relative
to other code, including across jump instructions."

No strong opinion though.

> Indeed, volatile is not a compiler barrier. Particularly once `-march=`
> flags are used under certain configs, LLVM's machine-scheduler can be
> observed moving instructions across the asm statement meant to point to
> known reachable or unreachable code, as reported by 0day bot.
>
> Prefer a memory clobber which is a compiler barrier that prevents these
> re-orderings and remove the volatile qualifier.
>
> Looking closer, the use of __COUNTER__ seems to have been used to
> prevent de-duplication of these asm statements. The GCC manual mentions:
>
> > Under certain circumstances, GCC may duplicate (or remove duplicates
> > of) your assembly code when optimizing. This can lead to unexpected
> > duplicate symbol errors during compilation if your asm code defines
> > symbols or labels. Using ‘%=’ (see AssemblerTemplate) may help resolve
> > this problem.
> >
> > ‘%=’ Outputs a number that is unique to each instance of the asm
> > statement in the entire compilation. This option is useful when
> > creating local labels and referring to them multiple times in a single
> > template that generates multiple assembler instructions.
>
> commit 3d1e236022cc ("objtool: Prevent GCC from merging annotate_unreachable()")
>
> Mentions that
>
> > The inline asm ‘%=’ token could be used for that, but unfortunately
> > older versions of GCC don't support it.
>
> From testing all versions of GCC available on godbolt.org, GCC 4.1+
> seems to support 4.1. Since the minimum supported version of GCC at the
> moment is GCC 5.1, it sounds like this is no longer a concern.
>
> Prefer the %= assembler template to having to stringify __COUNTER__.
>
> This commit is effectively a revert of the following commits:
> commit dcce50e6cc4d ("compiler.h: Fix annotation macro misplacement with Clang")
> commit f1069a8756b9 ("compiler.h: Avoid using inline asm operand modifiers")
> commit c199f64ff93c ("instrumentation.h: Avoid using inline asm operand modifiers")
> commit d0c2e691d1cb ("objtool: Add a comment for the unreachable annotation macros")
> commit ec1e1b610917 ("objtool: Prevent GCC from merging annotate_unreachable(), take 2")
> commit 3d1e236022cc ("objtool: Prevent GCC from merging annotate_unreachable()")
>
> Link: https://github.com/ClangBuiltLinux/linux/issues/1566
> Link: https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Volatile
> Link: https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#AssemblerTemplate
> Link: https://lore.kernel.org/llvm/[email protected]/
> Link: https://lore.kernel.org/llvm/[email protected]/
> Reported-by: kernel test robot <[email protected]>
> Debugged-by: Nathan Chancellor <[email protected]>
> Signed-off-by: Nick Desaulniers <[email protected]>

This resolves the original spew of warnings and does not regress a
couple of other configurations that I tested.

Tested-by: Nathan Chancellor <[email protected]>

Not sure I am qualified enough to give a reviewed-by.

> ---
> include/linux/compiler.h | 31 +++++++++++--------------------
> include/linux/instrumentation.h | 24 ++++++++++--------------
> 2 files changed, 21 insertions(+), 34 deletions(-)
>
> diff --git a/include/linux/compiler.h b/include/linux/compiler.h
> index 429dcebe2b99..3ac21b888d20 100644
> --- a/include/linux/compiler.h
> +++ b/include/linux/compiler.h
> @@ -108,30 +108,21 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
> # define barrier_before_unreachable() do { } while (0)
> #endif
>
> -/* Unreachable code */
> +/* These macros help objtool understand GCC code flow for unreachable code. */
> #ifdef CONFIG_STACK_VALIDATION
> -/*
> - * These macros help objtool understand GCC code flow for unreachable code.
> - * The __COUNTER__ based labels are a hack to make each instance of the macros
> - * unique, to convince GCC not to merge duplicate inline asm statements.
> - */
> -#define __stringify_label(n) #n
> -
> -#define __annotate_reachable(c) ({ \
> - asm volatile(__stringify_label(c) ":\n\t" \
> - ".pushsection .discard.reachable\n\t" \
> - ".long " __stringify_label(c) "b - .\n\t" \
> - ".popsection\n\t" : : "i" (c)); \
> +#define annotate_reachable() ({ \
> + asm (".Lreachable%=:\n\t" \
> + ".pushsection .discard.reachable\n\t" \
> + ".long .Lreachable%= - .\n\t" \
> + ".popsection\n\t" ::: "memory"); \
> })
> -#define annotate_reachable() __annotate_reachable(__COUNTER__)
>
> -#define __annotate_unreachable(c) ({ \
> - asm volatile(__stringify_label(c) ":\n\t" \
> - ".pushsection .discard.unreachable\n\t" \
> - ".long " __stringify_label(c) "b - .\n\t" \
> - ".popsection\n\t" : : "i" (c)); \
> +#define annotate_unreachable() ({ \
> + asm (".Lunreachable%=:\n\t" \
> + ".pushsection .discard.unreachable\n\t" \
> + ".long .Lunreachable%= - .\n\t" \
> + ".popsection\n\t" ::: "memory"); \
> })
> -#define annotate_unreachable() __annotate_unreachable(__COUNTER__)
>
> #define ASM_UNREACHABLE \
> "999:\n\t" \
> diff --git a/include/linux/instrumentation.h b/include/linux/instrumentation.h
> index 24359b4a9605..0dae9c08764f 100644
> --- a/include/linux/instrumentation.h
> +++ b/include/linux/instrumentation.h
> @@ -4,16 +4,13 @@
>
> #if defined(CONFIG_DEBUG_ENTRY) && defined(CONFIG_STACK_VALIDATION)
>
> -#include <linux/stringify.h>
> -
> /* Begin/end of an instrumentation safe region */
> -#define __instrumentation_begin(c) ({ \
> - asm volatile(__stringify(c) ": nop\n\t" \
> - ".pushsection .discard.instr_begin\n\t" \
> - ".long " __stringify(c) "b - .\n\t" \
> - ".popsection\n\t" : : "i" (c)); \
> +#define instrumentation_begin() ({ \
> + asm (".Linstbegin%=: nop\n\t" \
> + ".pushsection .discard.instr_begin\n\t" \
> + ".long .Linstbegin%= - .\n\t" \
> + ".popsection\n\t" ::: "memory"); \
> })
> -#define instrumentation_begin() __instrumentation_begin(__COUNTER__)
>
> /*
> * Because instrumentation_{begin,end}() can nest, objtool validation considers
> @@ -46,13 +43,12 @@
> * To avoid this, have _end() be a NOP instruction, this ensures it will be
> * part of the condition block and does not escape.
> */
> -#define __instrumentation_end(c) ({ \
> - asm volatile(__stringify(c) ": nop\n\t" \
> - ".pushsection .discard.instr_end\n\t" \
> - ".long " __stringify(c) "b - .\n\t" \
> - ".popsection\n\t" : : "i" (c)); \
> +#define instrumentation_end() ({ \
> + asm (".Linstend%=: nop\n\t" \
> + ".pushsection .discard.instr_end\n\t" \
> + ".long .Linstend%= - .\n\t" \
> + ".popsection\n\t" ::: "memory"); \
> })
> -#define instrumentation_end() __instrumentation_end(__COUNTER__)
> #else
> # define instrumentation_begin() do { } while(0)
> # define instrumentation_end() do { } while(0)
>
> base-commit: dcce50e6cc4d86a63dc0a9a6ee7d4f948ccd53a1
> --
> 2.34.1.703.g22d0c6ccf7-goog
>

2022-01-17 07:10:12

by Borislav Petkov

[permalink] [raw]
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

Hi Nick,

thanks for taking the time.

Here are a bunch of nitpicks:

On Thu, Jan 13, 2022 at 05:05:26PM -0800, Nick Desaulniers wrote:
> Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

You probably wanna formulate that subject a little less cryptic and more
to the point what the patch does:

objtool: Use a memory clobber to prevent reachability annotations from getting merged

or so.

> commit dcce50e6cc4d ("compiler.h: Fix annotation macro misplacement with Clang")
> mentions:
>
> > 'volatile' is ignored for some reason and Clang feels free to move the
> > reachable() annotation away from its intended location.
>
> Indeed, volatile is not a compiler barrier. Particularly once `-march=`
> flags are used under certain configs, LLVM's machine-scheduler can be
> observed moving instructions across the asm statement meant to point to
> known reachable or unreachable code, as reported by 0day bot.
>
> Prefer a memory clobber which is a compiler barrier that prevents these
> re-orderings and remove the volatile qualifier.

Yah, Nathan's note about talking about gcc reordering volatile inline
asms might make sense here as, apparently, that could have failed on gcc
too but it didn't for whatever reason and the most important thing is
that the memory clobber will prevent such reordering.

> Looking closer, the use of __COUNTER__ seems to have been used to

"... it seems, __COUNTER__ has been used... " or so. Simpler.

> prevent de-duplication of these asm statements. The GCC manual mentions:
>
> > Under certain circumstances, GCC may duplicate (or remove duplicates
> > of) your assembly code when optimizing. This can lead to unexpected
> > duplicate symbol errors during compilation if your asm code defines
> > symbols or labels. Using ‘%=’ (see AssemblerTemplate) may help resolve
> > this problem.
> >
> > ‘%=’ Outputs a number that is unique to each instance of the asm
> > statement in the entire compilation. This option is useful when
> > creating local labels and referring to them multiple times in a single
> > template that generates multiple assembler instructions.
>
> commit 3d1e236022cc ("objtool: Prevent GCC from merging annotate_unreachable()")
>
> Mentions that
>
> > The inline asm ‘%=’ token could be used for that, but unfortunately
> > older versions of GCC don't support it.
>
> From testing all versions of GCC available on godbolt.org, GCC 4.1+
> seems to support 4.1. Since the minimum supported version of GCC at the

You mean "seems to support %=." here perhaps?

> moment is GCC 5.1, it sounds like this is no longer a concern.
>
> Prefer the %= assembler template to having to stringify __COUNTER__.
>
> This commit is effectively a revert of the following commits:
> commit dcce50e6cc4d ("compiler.h: Fix annotation macro misplacement with Clang")
> commit f1069a8756b9 ("compiler.h: Avoid using inline asm operand modifiers")
> commit c199f64ff93c ("instrumentation.h: Avoid using inline asm operand modifiers")
> commit d0c2e691d1cb ("objtool: Add a comment for the unreachable annotation macros")
> commit ec1e1b610917 ("objtool: Prevent GCC from merging annotate_unreachable(), take 2")
> commit 3d1e236022cc ("objtool: Prevent GCC from merging annotate_unreachable()")
>
> Link: https://github.com/ClangBuiltLinux/linux/issues/1566
> Link: https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Volatile
> Link: https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#AssemblerTemplate
> Link: https://lore.kernel.org/llvm/[email protected]/
> Link: https://lore.kernel.org/llvm/[email protected]/
> Reported-by: kernel test robot <[email protected]>
> Debugged-by: Nathan Chancellor <[email protected]>
> Signed-off-by: Nick Desaulniers <[email protected]>

I very much definitely appreciate the extensive and comprehensive
writeup - thanks for taking the time!

> ---
> include/linux/compiler.h | 31 +++++++++++--------------------
> include/linux/instrumentation.h | 24 ++++++++++--------------
> 2 files changed, 21 insertions(+), 34 deletions(-)
>
> diff --git a/include/linux/compiler.h b/include/linux/compiler.h
> index 429dcebe2b99..3ac21b888d20 100644
> --- a/include/linux/compiler.h
> +++ b/include/linux/compiler.h
> @@ -108,30 +108,21 @@ void ftrace_likely_update(struct ftrace_likely_data *f, int val,
> # define barrier_before_unreachable() do { } while (0)
> #endif
>
> -/* Unreachable code */
> +/* These macros help objtool understand GCC code flow for unreachable code. */
> #ifdef CONFIG_STACK_VALIDATION
> -/*
> - * These macros help objtool understand GCC code flow for unreachable code.
> - * The __COUNTER__ based labels are a hack to make each instance of the macros
> - * unique, to convince GCC not to merge duplicate inline asm statements.
> - */
> -#define __stringify_label(n) #n
> -
> -#define __annotate_reachable(c) ({ \
> - asm volatile(__stringify_label(c) ":\n\t" \
> - ".pushsection .discard.reachable\n\t" \
> - ".long " __stringify_label(c) "b - .\n\t" \
> - ".popsection\n\t" : : "i" (c)); \
> +#define annotate_reachable() ({ \
> + asm (".Lreachable%=:\n\t" \
^

no need for that space.


> + ".pushsection .discard.reachable\n\t" \
> + ".long .Lreachable%= - .\n\t" \
> + ".popsection\n\t" ::: "memory"); \

Btw, you don't need the "\t" on the last lines of the macros here. The
asm output has a stray tab this way.

As to the change itself, before it would look like this:

# arch/x86/kernel/reboot.c:130: unreachable();
# 130 "arch/x86/kernel/reboot.c" 1
333:
.pushsection .discard.unreachable
.long 333b - .
.popsection



and after

# arch/x86/kernel/reboot.c:130: unreachable();
# 130 "arch/x86/kernel/reboot.c" 1
.Lunreachable44:
.pushsection .discard.unreachable
.long .Lunreachable44 - .
.popsection


so I like the local label and how it is more readable this way.

So, provided the memory clobber works (I wonder here if Josh has some
concrete failing cases which could be tested with your version) and
after the nitpicks have been addressed

Acked-by: Borislav Petkov <[email protected]>

Thx.

--
Regards/Gruss,
Boris.

https://people.kernel.org/tglx/notes-about-netiquette

2022-01-21 05:25:26

by Josh Poimboeuf

[permalink] [raw]
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

On Sun, Jan 16, 2022 at 02:32:59PM +0100, Borislav Petkov wrote:
> so I like the local label and how it is more readable this way.
>
> So, provided the memory clobber works (I wonder here if Josh has some
> concrete failing cases which could be tested with your version) and
> after the nitpicks have been addressed
>
> Acked-by: Borislav Petkov <[email protected]>

I think Nick was already able to recreate the original issue. I'll run
it through some more testing.

I wanted to make this change years ago, but couldn't because of legacy
toolchains. Here's hoping this is the final solution for those @#$%^
macros.

Boris, thanks for looping Nick in, I should have done so to begin with.

--
Josh

2022-01-21 11:13:15

by Josh Poimboeuf

[permalink] [raw]
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

On Tue, Jan 18, 2022 at 11:22:59AM -0800, Josh Poimboeuf wrote:
> On Sun, Jan 16, 2022 at 02:32:59PM +0100, Borislav Petkov wrote:
> > so I like the local label and how it is more readable this way.
> >
> > So, provided the memory clobber works (I wonder here if Josh has some
> > concrete failing cases which could be tested with your version) and
> > after the nitpicks have been addressed
> >
> > Acked-by: Borislav Petkov <[email protected]>
>
> I think Nick was already able to recreate the original issue. I'll run
> it through some more testing.
>
> I wanted to make this change years ago, but couldn't because of legacy
> toolchains. Here's hoping this is the final solution for those @#$%^
> macros.
>
> Boris, thanks for looping Nick in, I should have done so to begin with.

Apparently this patch isn't going to work after all :-(

https://lkml.kernel.org/r/[email protected]

With the two WARN_ONs in media_request_object_complete(), GCC apparently
considers the two reachable() asm statements as duplicates, and it
removes the second one.

--
Josh

2022-01-21 11:52:31

by Borislav Petkov

[permalink] [raw]
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

On Tue, Jan 18, 2022 at 03:01:20PM -0800, Josh Poimboeuf wrote:
> With the two WARN_ONs in media_request_object_complete(), GCC apparently
> considers the two reachable() asm statements as duplicates, and it
> removes the second one.

Could that be the same thing:

net/xfrm/xfrm_output.o: warning: objtool: xfrm_output_resume()+0x2e0: unreachable instruction

I see two WARN_ONs in asm output there too...

--
Regards/Gruss,
Boris.

https://people.kernel.org/tglx/notes-about-netiquette

2022-01-21 11:53:53

by Josh Poimboeuf

[permalink] [raw]
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

On Wed, Jan 19, 2022 at 12:33:02AM +0100, Borislav Petkov wrote:
> On Tue, Jan 18, 2022 at 03:01:20PM -0800, Josh Poimboeuf wrote:
> > With the two WARN_ONs in media_request_object_complete(), GCC apparently
> > considers the two reachable() asm statements as duplicates, and it
> > removes the second one.
>
> Could that be the same thing:
>
> net/xfrm/xfrm_output.o: warning: objtool: xfrm_output_resume()+0x2e0: unreachable instruction
>
> I see two WARN_ONs in asm output there too...

If one of the '__bug_table' asm snippets isn't immediately followed by
the .L[un]reachable asm, then yeah, it's the same issue.


For example it's supposed to look something like this:


# 472 "net/xfrm/xfrm_output.c" 1
1: .byte 0x0f, 0x0b
.pushsection __bug_table,"aw"
2: .long 1b - 2b # bug_entry::bug_addr
.long .LC4 - 2b # bug_entry::file #
.word 472 # bug_entry::line #
.word 2307 # bug_entry::flags #
.org 2b+12 #
.popsection
# 0 "" 2
# 472 "net/xfrm/xfrm_output.c" 1
.Lreachable1666:
.pushsection .discard.reachable
.long .Lreachable1666 - .
.popsection


NOT just this:


# 472 "net/xfrm/xfrm_output.c" 1
1: .byte 0x0f, 0x0b
.pushsection __bug_table,"aw"
2: .long 1b - 2b # bug_entry::bug_addr
.long .LC4 - 2b # bug_entry::file #
.word 472 # bug_entry::line #
.word 2307 # bug_entry::flags #
.org 2b+12 #
.popsection
# some other code here...


There's a bunch of those throughout the code base. The current
annotate_[un]reachable() implementations are carefully written to avoid
that happening.

--
Josh

2022-01-21 19:06:25

by Borislav Petkov

[permalink] [raw]
Subject: Re: [PATCH] objtool: prefer memory clobber & %= to volatile & __COUNTER__

On Tue, Jan 18, 2022 at 04:03:27PM -0800, Josh Poimboeuf wrote:
> If one of the '__bug_table' asm snippets isn't immediately followed by
> the .L[un]reachable asm, then yeah, it's the same issue.

Found one.

AFAICT, that's the WARN_ON_ONCE(1) catch-all in the default: label of
the switch-case in nf_hook(). That thing is followed by other gunk and
no *reachable label near it.

Damn - that was too good to be true. Gotta love those compilers. :-P

# ./include/linux/netfilter.h:252: WARN_ON_ONCE(1);
#APP
# 252 "./include/linux/netfilter.h" 1
1: .byte 0x0f, 0x0b
.pushsection __bug_table,"aw"
2: .long 1b - 2b # bug_entry::bug_addr
.long .LC5 - 2b # bug_entry::file #
.word 252 # bug_entry::line #
.word 2307 # bug_entry::flags #
.org 2b+12 #
.popsection
# 0 "" 2
#NO_APP
.p2align 4,,3
jmp .L344 #
.L265:
# ./include/linux/netfilter.h:229: hook_head = rcu_dereference(net->nf.hooks_ipv4[hook]);

--
Regards/Gruss,
Boris.

https://people.kernel.org/tglx/notes-about-netiquette