| Message ID | 1452860937-4413-1-git-send-email-ard.biesheuvel@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Fri, Jan 15, 2016 at 01:28:57PM +0100, Ard Biesheuvel wrote: > Commit e8f3010f7326 ("arm64/efi: isolate EFI stub from the kernel > proper") isolated the EFI stub code from the kernel proper by prefixing > all of its symbols with __efistub_, and selectively allowing access to > core kernel symbols from the stub by emitting __efistub_ aliases for > functions and variables that the stub can access legally. > > As an unintended side effect, these aliases are emitted into the > kallsyms symbol table, which means they may turn up in backtraces, > e.g., > > ... > PC is at __efistub_memset+0x108/0x200 > LR is at fixup_init+0x3c/0x48 > ... > [<ffffff8008328608>] __efistub_memset+0x108/0x200 > [<ffffff8008094dcc>] free_initmem+0x2c/0x40 > [<ffffff8008645198>] kernel_init+0x20/0xe0 > [<ffffff8008085cd0>] ret_from_fork+0x10/0x40 > > The backtrace in question has nothing to do with the EFI stub, but > simply returns one of the several aliases of memset() that have been > recorded in the kallsyms table. This is undesirable, since it may > suggest to people who are not aware of this that the issue they are > seeing is somehow EFI related. > > So hide the __efistub_ aliases from kallsyms, by emitting them as > absolute linker symbols explicitly. The distinction between those > and section relative symbols is completely irrelevant to these > definitions, and to the final link we are performing when these > definitions are being taken into account (the distinction is only > relevant to symbols defined inside a section definition when performing > a partial link), and so the resulting values are identical to the > original ones. Since absolute symbols are ignored by kallsyms, this > will result in these values to be omitted from its symbol table. > > After this patch, the backtrace generated from the same address looks > like this: > ... > PC is at __memset+0x108/0x200 > LR is at fixup_init+0x3c/0x48 > ... > [<ffffff8008328608>] __memset+0x108/0x200 > [<ffffff8008094dcc>] free_initmem+0x2c/0x40 > [<ffffff8008645198>] kernel_init+0x20/0xe0 > [<ffffff8008085cd0>] ret_from_fork+0x10/0x40 > > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> > --- > arch/arm64/kernel/image.h | 40 +++++++++++++++++++++++++--------------- > 1 file changed, 25 insertions(+), 15 deletions(-) > > diff --git a/arch/arm64/kernel/image.h b/arch/arm64/kernel/image.h > index ddd61e2d9e3e..ce6f02c56b3d 100644 > --- a/arch/arm64/kernel/image.h > +++ b/arch/arm64/kernel/image.h > @@ -81,6 +81,16 @@ > #ifdef CONFIG_EFI > > /* > + * Prevent the symbol aliases below from being emitted into the kallsyms > + * table, by forcing them to be absolute symbols (which are conveniently > + * ignored by scripts/kallsyms) rather than section relative symbols. > + * The distinction is only relevant for partial linking, and only for symbols > + * that are defined within a section declaration (which is not the case for > + * the definitions below) so the resulting values will be identical. > + */ > +#define KALLSYMS_HIDE(sym) ABSOLUTE(sym) From looking at the GNU LD manual recently, I discovered PROVIDE_HIDDEN(sym), which sounds closer to what we want semantically, even if ABSOLUTE(sym) acheives the same thing. If that works, that might be preferable. Either way this looks sensible to me, so: Acked-by: Mark Rutland <mark.rutland@arm.com> Mark. > + > +/* > * The EFI stub has its own symbol namespace prefixed by __efistub_, to > * isolate it from the kernel proper. The following symbols are legally > * accessed by the stub, so provide some aliases to make them accessible. > @@ -89,25 +99,25 @@ > * linked at. The routines below are all implemented in assembler in a > * position independent manner > */ > -__efistub_memcmp = __pi_memcmp; > -__efistub_memchr = __pi_memchr; > -__efistub_memcpy = __pi_memcpy; > -__efistub_memmove = __pi_memmove; > -__efistub_memset = __pi_memset; > -__efistub_strlen = __pi_strlen; > -__efistub_strcmp = __pi_strcmp; > -__efistub_strncmp = __pi_strncmp; > -__efistub___flush_dcache_area = __pi___flush_dcache_area; > +__efistub_memcmp = KALLSYMS_HIDE(__pi_memcmp); > +__efistub_memchr = KALLSYMS_HIDE(__pi_memchr); > +__efistub_memcpy = KALLSYMS_HIDE(__pi_memcpy); > +__efistub_memmove = KALLSYMS_HIDE(__pi_memmove); > +__efistub_memset = KALLSYMS_HIDE(__pi_memset); > +__efistub_strlen = KALLSYMS_HIDE(__pi_strlen); > +__efistub_strcmp = KALLSYMS_HIDE(__pi_strcmp); > +__efistub_strncmp = KALLSYMS_HIDE(__pi_strncmp); > +__efistub___flush_dcache_area = KALLSYMS_HIDE(__pi___flush_dcache_area); > > #ifdef CONFIG_KASAN > -__efistub___memcpy = __pi_memcpy; > -__efistub___memmove = __pi_memmove; > -__efistub___memset = __pi_memset; > +__efistub___memcpy = KALLSYMS_HIDE(__pi_memcpy); > +__efistub___memmove = KALLSYMS_HIDE(__pi_memmove); > +__efistub___memset = KALLSYMS_HIDE(__pi_memset); > #endif > > -__efistub__text = _text; > -__efistub__end = _end; > -__efistub__edata = _edata; > +__efistub__text = KALLSYMS_HIDE(_text); > +__efistub__end = KALLSYMS_HIDE(_end); > +__efistub__edata = KALLSYMS_HIDE(_edata); > > #endif > > -- > 2.5.0 >
On 15 January 2016 at 13:41, Mark Rutland <mark.rutland@arm.com> wrote: > On Fri, Jan 15, 2016 at 01:28:57PM +0100, Ard Biesheuvel wrote: >> Commit e8f3010f7326 ("arm64/efi: isolate EFI stub from the kernel >> proper") isolated the EFI stub code from the kernel proper by prefixing >> all of its symbols with __efistub_, and selectively allowing access to >> core kernel symbols from the stub by emitting __efistub_ aliases for >> functions and variables that the stub can access legally. >> >> As an unintended side effect, these aliases are emitted into the >> kallsyms symbol table, which means they may turn up in backtraces, >> e.g., >> >> ... >> PC is at __efistub_memset+0x108/0x200 >> LR is at fixup_init+0x3c/0x48 >> ... >> [<ffffff8008328608>] __efistub_memset+0x108/0x200 >> [<ffffff8008094dcc>] free_initmem+0x2c/0x40 >> [<ffffff8008645198>] kernel_init+0x20/0xe0 >> [<ffffff8008085cd0>] ret_from_fork+0x10/0x40 >> >> The backtrace in question has nothing to do with the EFI stub, but >> simply returns one of the several aliases of memset() that have been >> recorded in the kallsyms table. This is undesirable, since it may >> suggest to people who are not aware of this that the issue they are >> seeing is somehow EFI related. >> >> So hide the __efistub_ aliases from kallsyms, by emitting them as >> absolute linker symbols explicitly. The distinction between those >> and section relative symbols is completely irrelevant to these >> definitions, and to the final link we are performing when these >> definitions are being taken into account (the distinction is only >> relevant to symbols defined inside a section definition when performing >> a partial link), and so the resulting values are identical to the >> original ones. Since absolute symbols are ignored by kallsyms, this >> will result in these values to be omitted from its symbol table. >> >> After this patch, the backtrace generated from the same address looks >> like this: >> ... >> PC is at __memset+0x108/0x200 >> LR is at fixup_init+0x3c/0x48 >> ... >> [<ffffff8008328608>] __memset+0x108/0x200 >> [<ffffff8008094dcc>] free_initmem+0x2c/0x40 >> [<ffffff8008645198>] kernel_init+0x20/0xe0 >> [<ffffff8008085cd0>] ret_from_fork+0x10/0x40 >> >> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> >> --- >> arch/arm64/kernel/image.h | 40 +++++++++++++++++++++++++--------------- >> 1 file changed, 25 insertions(+), 15 deletions(-) >> >> diff --git a/arch/arm64/kernel/image.h b/arch/arm64/kernel/image.h >> index ddd61e2d9e3e..ce6f02c56b3d 100644 >> --- a/arch/arm64/kernel/image.h >> +++ b/arch/arm64/kernel/image.h >> @@ -81,6 +81,16 @@ >> #ifdef CONFIG_EFI >> >> /* >> + * Prevent the symbol aliases below from being emitted into the kallsyms >> + * table, by forcing them to be absolute symbols (which are conveniently >> + * ignored by scripts/kallsyms) rather than section relative symbols. >> + * The distinction is only relevant for partial linking, and only for symbols >> + * that are defined within a section declaration (which is not the case for >> + * the definitions below) so the resulting values will be identical. >> + */ >> +#define KALLSYMS_HIDE(sym) ABSOLUTE(sym) > > From looking at the GNU LD manual recently, I discovered > PROVIDE_HIDDEN(sym), which sounds closer to what we want semantically, > even if ABSOLUTE(sym) acheives the same thing. > > If that works, that might be preferable. > No, it doesn't. PROVIDE_HIDDEN() turns it into a local symbol, but preserves the type, i.e., Before: ffffff8008080000 T __efistub__text With this patch ffffff8008080000 A __efistub__text With PROVIDE_HIDDEN ffffff8008080000 t __efistub__text I am not crazy about this change myself, but at least it keeps the ugliness confined to image.h in a well annotated manner, rather than in every backtrace that involves any of these functions. > Either way this looks sensible to me, so: > > Acked-by: Mark Rutland <mark.rutland@arm.com> > Thanks, Ard. >> + >> +/* >> * The EFI stub has its own symbol namespace prefixed by __efistub_, to >> * isolate it from the kernel proper. The following symbols are legally >> * accessed by the stub, so provide some aliases to make them accessible. >> @@ -89,25 +99,25 @@ >> * linked at. The routines below are all implemented in assembler in a >> * position independent manner >> */ >> -__efistub_memcmp = __pi_memcmp; >> -__efistub_memchr = __pi_memchr; >> -__efistub_memcpy = __pi_memcpy; >> -__efistub_memmove = __pi_memmove; >> -__efistub_memset = __pi_memset; >> -__efistub_strlen = __pi_strlen; >> -__efistub_strcmp = __pi_strcmp; >> -__efistub_strncmp = __pi_strncmp; >> -__efistub___flush_dcache_area = __pi___flush_dcache_area; >> +__efistub_memcmp = KALLSYMS_HIDE(__pi_memcmp); >> +__efistub_memchr = KALLSYMS_HIDE(__pi_memchr); >> +__efistub_memcpy = KALLSYMS_HIDE(__pi_memcpy); >> +__efistub_memmove = KALLSYMS_HIDE(__pi_memmove); >> +__efistub_memset = KALLSYMS_HIDE(__pi_memset); >> +__efistub_strlen = KALLSYMS_HIDE(__pi_strlen); >> +__efistub_strcmp = KALLSYMS_HIDE(__pi_strcmp); >> +__efistub_strncmp = KALLSYMS_HIDE(__pi_strncmp); >> +__efistub___flush_dcache_area = KALLSYMS_HIDE(__pi___flush_dcache_area); >> >> #ifdef CONFIG_KASAN >> -__efistub___memcpy = __pi_memcpy; >> -__efistub___memmove = __pi_memmove; >> -__efistub___memset = __pi_memset; >> +__efistub___memcpy = KALLSYMS_HIDE(__pi_memcpy); >> +__efistub___memmove = KALLSYMS_HIDE(__pi_memmove); >> +__efistub___memset = KALLSYMS_HIDE(__pi_memset); >> #endif >> >> -__efistub__text = _text; >> -__efistub__end = _end; >> -__efistub__edata = _edata; >> +__efistub__text = KALLSYMS_HIDE(_text); >> +__efistub__end = KALLSYMS_HIDE(_end); >> +__efistub__edata = KALLSYMS_HIDE(_edata); >> >> #endif >> >> -- >> 2.5.0 >>
diff --git a/arch/arm64/kernel/image.h b/arch/arm64/kernel/image.h index ddd61e2d9e3e..ce6f02c56b3d 100644 --- a/arch/arm64/kernel/image.h +++ b/arch/arm64/kernel/image.h @@ -81,6 +81,16 @@ #ifdef CONFIG_EFI /* + * Prevent the symbol aliases below from being emitted into the kallsyms + * table, by forcing them to be absolute symbols (which are conveniently + * ignored by scripts/kallsyms) rather than section relative symbols. + * The distinction is only relevant for partial linking, and only for symbols + * that are defined within a section declaration (which is not the case for + * the definitions below) so the resulting values will be identical. + */ +#define KALLSYMS_HIDE(sym) ABSOLUTE(sym) + +/* * The EFI stub has its own symbol namespace prefixed by __efistub_, to * isolate it from the kernel proper. The following symbols are legally * accessed by the stub, so provide some aliases to make them accessible. @@ -89,25 +99,25 @@ * linked at. The routines below are all implemented in assembler in a * position independent manner */ -__efistub_memcmp = __pi_memcmp; -__efistub_memchr = __pi_memchr; -__efistub_memcpy = __pi_memcpy; -__efistub_memmove = __pi_memmove; -__efistub_memset = __pi_memset; -__efistub_strlen = __pi_strlen; -__efistub_strcmp = __pi_strcmp; -__efistub_strncmp = __pi_strncmp; -__efistub___flush_dcache_area = __pi___flush_dcache_area; +__efistub_memcmp = KALLSYMS_HIDE(__pi_memcmp); +__efistub_memchr = KALLSYMS_HIDE(__pi_memchr); +__efistub_memcpy = KALLSYMS_HIDE(__pi_memcpy); +__efistub_memmove = KALLSYMS_HIDE(__pi_memmove); +__efistub_memset = KALLSYMS_HIDE(__pi_memset); +__efistub_strlen = KALLSYMS_HIDE(__pi_strlen); +__efistub_strcmp = KALLSYMS_HIDE(__pi_strcmp); +__efistub_strncmp = KALLSYMS_HIDE(__pi_strncmp); +__efistub___flush_dcache_area = KALLSYMS_HIDE(__pi___flush_dcache_area); #ifdef CONFIG_KASAN -__efistub___memcpy = __pi_memcpy; -__efistub___memmove = __pi_memmove; -__efistub___memset = __pi_memset; +__efistub___memcpy = KALLSYMS_HIDE(__pi_memcpy); +__efistub___memmove = KALLSYMS_HIDE(__pi_memmove); +__efistub___memset = KALLSYMS_HIDE(__pi_memset); #endif -__efistub__text = _text; -__efistub__end = _end; -__efistub__edata = _edata; +__efistub__text = KALLSYMS_HIDE(_text); +__efistub__end = KALLSYMS_HIDE(_end); +__efistub__edata = KALLSYMS_HIDE(_edata); #endif
Commit e8f3010f7326 ("arm64/efi: isolate EFI stub from the kernel proper") isolated the EFI stub code from the kernel proper by prefixing all of its symbols with __efistub_, and selectively allowing access to core kernel symbols from the stub by emitting __efistub_ aliases for functions and variables that the stub can access legally. As an unintended side effect, these aliases are emitted into the kallsyms symbol table, which means they may turn up in backtraces, e.g., ... PC is at __efistub_memset+0x108/0x200 LR is at fixup_init+0x3c/0x48 ... [<ffffff8008328608>] __efistub_memset+0x108/0x200 [<ffffff8008094dcc>] free_initmem+0x2c/0x40 [<ffffff8008645198>] kernel_init+0x20/0xe0 [<ffffff8008085cd0>] ret_from_fork+0x10/0x40 The backtrace in question has nothing to do with the EFI stub, but simply returns one of the several aliases of memset() that have been recorded in the kallsyms table. This is undesirable, since it may suggest to people who are not aware of this that the issue they are seeing is somehow EFI related. So hide the __efistub_ aliases from kallsyms, by emitting them as absolute linker symbols explicitly. The distinction between those and section relative symbols is completely irrelevant to these definitions, and to the final link we are performing when these definitions are being taken into account (the distinction is only relevant to symbols defined inside a section definition when performing a partial link), and so the resulting values are identical to the original ones. Since absolute symbols are ignored by kallsyms, this will result in these values to be omitted from its symbol table. After this patch, the backtrace generated from the same address looks like this: ... PC is at __memset+0x108/0x200 LR is at fixup_init+0x3c/0x48 ... [<ffffff8008328608>] __memset+0x108/0x200 [<ffffff8008094dcc>] free_initmem+0x2c/0x40 [<ffffff8008645198>] kernel_init+0x20/0xe0 [<ffffff8008085cd0>] ret_from_fork+0x10/0x40 Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> --- arch/arm64/kernel/image.h | 40 +++++++++++++++++++++++++--------------- 1 file changed, 25 insertions(+), 15 deletions(-)