Re: vdso(7): new man page

[Mike Frysinger <vapier-aBrp7R+bbdUdnm+yROfE0A@public.gmane.org>]

[-- Attachment #1: Type: Text/Plain, Size: 10972 bytes --]

here's v2 w/Andy's feedback
-mike

.\" Written by Mike Frysinger <vapier-aBrp7R+bbdUdnm+yROfE0A@public.gmane.org>
.\"
.\" %%%LICENSE_START(PUBLIC_DOMAIN)
.\" This page is in the public domain.  Suck it.
.\" %%%LICENSE_END
.\"
.TH VDSO 7 2013-04-09 "Linux" "Linux Programmer's Manual"
.SH NAME
vDSO \- overview of the virtual ELF dynamic shared object
.SH SYNOPSIS
.B #include <sys/auxv.h>

.B void *vdso = (uintptr_t)getauxval(AT_SYSINFO_EHDR);
.SH DESCRIPTION
The "vDSO" is a small shared library that the kernel automatically maps into the
address space of all userspace applications.
Applications themselves usually need not concern themselves with this as it is
most commonly called by the C library.
This way you can write using standard functions and the C library will take care
of using any available functionality.

Why does this object exist at all?
There are some facilities the kernel provides that userspace ends up using
frequently to the point that such calls can dominate overall performance.
This is due both to the frequency of the call as well as the context overhead
from exiting userspace and entering the kernel.

The rest of this documentation is geared towards the curious and/or C library
writers rather than general developers.
If you're trying to call the vDSO in your own application rather than using
the C library, you're most likely doing it wrong.
.SS Example Background
Making syscalls can be slow.
In x86 32bit systems, you can trigger a software interrupt (int $0x80) to tell
the kernel you wish to make a syscall.
However, this instruction is expensive: it goes through the full interrupt
handling paths in the processor's microcode as well as in the kernel.
Newer processors have faster (but backwards incompatible) instructions to
initiate system calls.
Rather than require the C library to figure out if this functionality is
available at runtime itself, it can use functions provided by the kernel in
the vDSO.
Note that the terminology can be confusing.
On x86 systems, the vDSO function is named "__kernel_vsyscall", but on x86_64,
the term "vsyscall" also refers to an obsolete way to ask the kernel what time
it is or what cpu the caller is on.

Another frequent system call is gettimeofday().
This is called both directly by userspace applications as well as indirectly by
the C library.
Think timestamps or timing loops or polling -- all of these frequently need to
know what time it is right now.
This information is also not secret -- any application in any privilege mode
(root or any user) will get the same answer.
Thus the kernel arranges for the information required to answer this question
to be placed in memory the process can access.
Now a call to gettimeofday() changes from a syscall to a normal function call
and a few memory accesses.
.SS Finding The vDSO
The base address of the vDSO (if one exists) is passed by the kernel to each
program in the initial auxiliary vector.
Specifically, via the
.B AT_SYSINFO_EHDR
tag.

You must not assume the vDSO is mapped at any particular location in the
user's memory map.
The base address will usually be randomized at runtime every time a new is
processed (at
.BR execve (2)
time).
This is done for security reasons to prevent standard "return-to-libc" attacks.

For some architectures, there is also a
.B AT_SYSINFO
tag.
This is used only for locating the vsyscall entry point and is frequently
omitted or set to 0 (meaning it's not available).
It is a throw back to the initial vDSO work (see
.IR HISTORY
below) and should be avoided.

Refer to
.BR getauxval (3)
for more details on accessing these fields.
.SS File Format
Since the vDSO is a fully formed ELF, you can do symbol lookups on it.
This allows new symbols to be added with newer kernel releases, and for the
C library to detect available functionality at runtime when running under
different kernel versions.
Often times the C library will do detection with the first call and then
cache the result for subsequent calls.

All symbols are also versioned (using the GNU version format).
This allows the kernel (in the very unlikely situation) to update the function
signature without breaking backwards compatibility.
This means changing the arguments that it accepts as well as the return value.
When looking up a symbol in the vDSO, you must always include the version you
are writing against.

Typically the vDSO follows the naming convention of prefixing all symbols with
"__vdso_" or "__kernel_" so as to distinguish from other standard symbols.
e.g. The "gettimeofday" function is named "__vdso_gettimeofday".

You use the standard C calling conventions when calling any of these functions.
No need to worry about weird register or stack behavior.
.SH NOTES
.SS Source
When you compile the kernel, it will automatically compile and link the vDSO
code for you.
You will frequently find it under the arch specific dir:
.br
find arch/$ARCH/ -name '*vdso*.so*' -o -name '*gate*.so*'

Note that the vDSO that is used is based on the ABI of your userspace code
and not the ABI of the kernel.
i.e. If you run an i386 32bit ELF under an i386 32bit kernel or under an
x86_64 64bit kernel, you'll get the same vDSO.
So when referring to sections below, use the userspace ABI.
.SS vDSO Names
The name of this shared object varies across architectures.
It will often show up in things like glibc's `ldd` output.
The exact name should not matter to any code, so please do not hardcode it.
.if t \{\
.ft CW
\}
.TS
l l.
user ABI	vDSO name
_
aarch64	linux-vdso.so.1
ia64	linux-gate.so.1
ppc/32	linux-vdso32.so.1
ppc/64	linux-vdso64.so.1
s390	linux-vdso32.so.1
s390x	linux-vdso64.so.1
sh	linux-gate.so.1
i386	linux-gate.so.1
x86_64	linux-vdso.so.1
x86/x32	linux-vdso.so.1
.TE
.if t \{\
.in
.ft P
\}
.SS aarch64 functions
.\" See linux/arch/arm64/kernel/vdso/vdso.lds.S
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_rt_sigreturn	LINUX_2.6.39
__kernel_gettimeofday	LINUX_2.6.39
__kernel_clock_gettime	LINUX_2.6.39
__kernel_clock_getres	LINUX_2.6.39
.TE
.if t \{\
.in
.ft P
\}
.SS bfin (Blackfin) functions
.\" See linux/arch/blackfin/kernel/fixed_code.S
.\" See http://docs.blackfin.uclinux.org/doku.php?id=linux-kernel:fixed-code
As this cpu lacks a MMU, it doesn't setup a vDSO in the normal sense.
Instead, it maps at boot time a few raw functions into a fixed location in
memory.
Userspace apps then call directly into that.
There is no provision for backwards compatibility beyond sniffing raw opcodes,
but as this is an embedded CPU, it can get away with things -- some of the
object formats it runs aren't even ELF based (they're bFLT/FLAT).

For documentation on this format, it's better you refer to the public docs:
.br
http://docs.blackfin.uclinux.org/doku.php?id=linux-kernel:fixed-code
.SS ia64 (Itanium) functions
.\" See linux/arch/ia64/kernel/gate.lds.S
.\" Also linux/arch/ia64/kernel/fsys.S and linux/Documentation/ia64/fsys.txt
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_sigtramp	LINUX_2.5
__kernel_syscall_via_break	LINUX_2.5
__kernel_syscall_via_epc	LINUX_2.5
.TE
.if t \{\
.in
.ft P
\}

The Itanium port actually likes to get tricky.
In addition to the vDSO above, it also has "light-weight system calls" aka
"fast syscalls" aka "fsys".
You can invoke these via the __kernel_syscall_via_epc vDSO helper.
The system calls listed here have the same semantics as if you called them
directly via
.BR syscall (3),
so refer to the relevant
documentation for each.
The table below lists the functions available via this mechanism.
.if t \{\
.ft CW
\}
.TS
l.
function
_
clock_gettime
getcpu
getpid
getppid
gettimeofday
set_tid_address
.TE
.if t \{\
.in
.ft P
\}
.SS ppc/32 functions
.\" See linux/arch/powerpc/kernel/vdso32/vdso32.lds.S
The functions marked with a
.I *
below are only available when the kernel is
a powerpc64 (64bit) kernel.
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_clock_getres	LINUX_2.6.15
__kernel_clock_gettime	LINUX_2.6.15
__kernel_datapage_offset	LINUX_2.6.15
__kernel_get_syscall_map	LINUX_2.6.15
__kernel_get_tbfreq	LINUX_2.6.15
__kernel_getcpu \fI*\fR	LINUX_2.6.15
__kernel_gettimeofday	LINUX_2.6.15
__kernel_sigtramp_rt32	LINUX_2.6.15
__kernel_sigtramp32	LINUX_2.6.15
__kernel_sync_dicache	LINUX_2.6.15
__kernel_sync_dicache_p5	LINUX_2.6.15
.TE
.if t \{\
.in
.ft P
\}
.SS ppc/64 functions
.\" See linux/arch/powerpc/kernel/vdso64/vdso64.lds.S
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_clock_getres	LINUX_2.6.15
__kernel_clock_gettime	LINUX_2.6.15
__kernel_datapage_offset	LINUX_2.6.15
__kernel_get_syscall_map	LINUX_2.6.15
__kernel_get_tbfreq	LINUX_2.6.15
__kernel_getcpu	LINUX_2.6.15
__kernel_gettimeofday	LINUX_2.6.15
__kernel_sigtramp_rt64	LINUX_2.6.15
__kernel_sync_dicache	LINUX_2.6.15
__kernel_sync_dicache_p5	LINUX_2.6.15
.TE
.if t \{\
.in
.ft P
\}
.SS s390 functions
.\" See linux/arch/s390/kernel/vdso32/vdso32.lds.S
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_clock_getres	LINUX_2.6.29
__kernel_clock_gettime	LINUX_2.6.29
__kernel_gettimeofday	LINUX_2.6.29
.TE
.if t \{\
.in
.ft P
\}
.SS s390x functions
.\" See linux/arch/s390/kernel/vdso64/vdso64.lds.S
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_clock_getres	LINUX_2.6.29
__kernel_clock_gettime	LINUX_2.6.29
__kernel_gettimeofday	LINUX_2.6.29
.TE
.if t \{\
.in
.ft P
\}
.SS sh (SuperH) functions
.\" See linux/arch/sh/kernel/vsyscall/vsyscall.lds.S
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_rt_sigreturn	LINUX_2.6
__kernel_sigreturn	LINUX_2.6
__kernel_vsyscall	LINUX_2.6
.TE
.if t \{\
.in
.ft P
\}
.SS i386 functions
.\" See linux/arch/x86/vdso/vdso32/vdso32.lds.S
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__kernel_sigreturn	LINUX_2.5
__kernel_rt_sigreturn	LINUX_2.5
__kernel_vsyscall	LINUX_2.5
.TE
.if t \{\
.in
.ft P
\}
.SS x86_64 functions
.\" See linux/arch/x86/vdso/vdso.lds.S
Each of these symbols are also available without the "__vdso_" prefix, but
you should ignore those and stick to the names below.
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__vdso_clock_gettime	LINUX_2.6
__vdso_getcpu	LINUX_2.6
__vdso_gettimeofday	LINUX_2.6
__vdso_time	LINUX_2.6
.TE
.if t \{\
.in
.ft P
\}
.SS x86/x32 functions
.\" See linux/arch/x86/vdso/vdso32.lds.S
.if t \{\
.ft CW
\}
.TS
l l.
symbol	version
_
__vdso_clock_gettime	LINUX_2.6
__vdso_getcpu	LINUX_2.6
__vdso_gettimeofday	LINUX_2.6
__vdso_time	LINUX_2.6
.TE
.if t \{\
.in
.ft P
\}
.SH HISTORY
The vDSO was originally just a single function -- the vsyscall.
In older kernels, you might see that in a process's memory map rather than vdso.
Overtime, people realized that this was a great way to pass more functionality
to userspace, so it was reconceived as a vDSO in the current format.
.SH SEE ALSO
.BR syscalls (2),
.BR getauxval (3),
.BR proc (5)

The docs/examples/sources in the Linux sources:
.nf
Documentation/ABI/stable/vdso
linux/Documentation/ia64/fsys.txt
Documentation/vDSO/* (includes examples of using the vDSO)
find arch/ -iname '*vdso*' -o -iname '*gate*'
.fi

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]