Let the arch argument be overruled by bits. Otherwise, building of
external modules against a i386 target on a x86-64 host (and likely vice
versa as well) fails unless ARCH=i386 is explicitly passed to make.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
LKML-Reference: <4B4AFE10.8050109@siemens.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
With dynamic function tracer, by default, _mcount is defined as an
"empty" function, it returns directly without any more action . When
enabling it in user-space, it will jump to a real tracing
function(ftrace_caller), and do the real job for us.
Differ from the static function tracer, dynamic function tracer provides
two functions ftrace_make_call()/ftrace_make_nop() to enable/disable the
tracing of some indicated kernel functions(set_ftrace_filter).
In the -v4 version, the implementation of this support is basically the same as
X86 version does: _mcount is implemented as an empty function and ftrace_caller
is implemented as a real tracing function respectively.
But in this version, to support module tracing with the help of
-mlong-calls in arch/mips/Makefile:
MODFLAGS += -mlong-calls.
The stuff becomes a little more complex. We need to cope with two
different type of calling to _mcount.
For the kernel part, the calling to _mcount(result of "objdump -hdr
vmlinux"). is like this:
108: 03e0082d move at,ra
10c: 0c000000 jal 0 <fpcsr_pending>
10c: R_MIPS_26 _mcount
10c: R_MIPS_NONE *ABS*
10c: R_MIPS_NONE *ABS*
110: 00020021 nop
For the module with -mlong-calls, it looks like this:
c: 3c030000 lui v1,0x0
c: R_MIPS_HI16 _mcount
c: R_MIPS_NONE *ABS*
c: R_MIPS_NONE *ABS*
10: 64630000 daddiu v1,v1,0
10: R_MIPS_LO16 _mcount
10: R_MIPS_NONE *ABS*
10: R_MIPS_NONE *ABS*
14: 03e0082d move at,ra
18: 0060f809 jalr v1
In the kernel version, there is only one "_mcount" string for every
kernel function, so, we just need to match this one in mcount_regex of
scripts/recordmcount.pl, but in the module version, we need to choose
one of the two to match. Herein, I choose the first one with
"R_MIPS_HI16 _mcount".
and In the kernel verion, without module tracing support, we just need
to replace "jal _mcount" by "jal ftrace_caller" to do real tracing, and
filter the tracing of some kernel functions via replacing it by a nop
instruction.
but as we have described before, the instruction "jal ftrace_caller" only left
32bit length for the address of ftrace_caller, it will fail when calling from
the module space. so, herein, we must replace something else.
the basic idea is loading the address of ftrace_caller to v1 via changing these
two instructions:
lui v1,0x0
addiu v1,v1,0
If we want to enable the tracing, we need to replace the above instructions to:
lui v1, HI_16BIT_ftrace_caller
addiu v1, v1, LOW_16BIT_ftrace_caller
If we want to stop the tracing of the indicated kernel functions, we
just need to replace the "jalr v1" to a nop instruction. but we need to
replace two instructions and encode the above two instructions
oursevles.
Is there a simpler solution? Yes! Here it is, in this version, we put _mcount
and ftrace_caller together, which means the address of _mcount and
ftrace_caller is the same:
_mcount:
ftrace_caller:
j ftrace_stub
nop
...(do real tracing here)...
ftrace_stub:
jr ra
move ra, at
By default, the kernel functions call _mcount, and then jump to ftrace_stub and
return. and when we want to do real tracing, we just need to remove that "j
ftrace_stub", and it will run through the two "nop" instructions and then do
the real tracing job.
what about filtering job? we just need to do this:
lui v1, hi_16bit_of_mcount <--> b 1f (0x10000004)
addiu v1, v1, low_16bit_of_mcount
move at, ra
jalr v1
nop
1f: (rec->ip + 12)
In linux-mips64, there will be some local symbols, whose name are
prefixed by $L, which need to be filtered. thanks goes to Steven for
writing the mips64-specific function_regex.
In a conclusion, with RISC, things becomes easier with such a "stupid"
trick, RISC is something like K.I.S.S, and also, there are lots of
"simple" tricks in the whole ftrace support, thanks goes to Steven and
the other folks for providing such a wonderful tracing framework!
Signed-off-by: Wu Zhangjin <wuzhangjin@gmail.com>
Cc: Nicholas Mc Guire <der.herr@hofr.at>
Cc: zhangfx@lemote.com
Cc: Wu Zhangjin <wuzhangjin@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: linux-kernel@vger.kernel.org
Cc: linux-mips@linux-mips.org
Patchwork: http://patchwork.linux-mips.org/patch/675/
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
With dynamic function tracer, by default, _mcount is defined as an
"empty" function, it returns directly without any more action. When
enabling it in user-space, it will jump to a real tracing
function(ftrace_caller), and do the real job for us.
Differ from the static function tracer, dynamic function tracer provides
two functions ftrace_make_call()/ftrace_make_nop() to enable/disable the
tracing of some indicated kernel functions(set_ftrace_filter).
In the kernel version, there is only one "_mcount" string for every
kernel function, so, we just need to match this one in mcount_regex of
scripts/recordmcount.pl.
For more information please look at code and Documentation/trace folder.
Steven ACK that scripts/recordmcount.pl part.
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Michal Simek <monstr@monstr.eu>
If the user has an older version of objcopy, that can not handle
converting local symbols to global and vice versa, then some
functions will not be part of the dynamic function tracer. The current
code in recordmcount.pl will print a warning in this case. Unfortunately,
there exists lots of files that may have this issue with older objcopys
and this will cause a warning for every file compiled with this
issue.
This patch solves this overwhelming output by creating a
.tmp_quiet_recordmcount file on the first instance the warning is
encountered. The warning will not print if this file exists.
The temp file is deleted at the beginning of the compile to ensure that
the warning will happen once again on new compiles (because the issue
is still present).
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
If a weak function is used as a relocation reference for mcount callers
and that function is overridden, it will cause ftrace to fail at run time.
The current code should prevent a weak function from being used, but if
one is, the code should exit with an error to fail at compile time.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091028050743.GH30758@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Move all the condition validations into the function update_funcs().
Also update_funcs should not die if $ref_func is undefined for there may be
more than one valid section in an object file.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091028050703.GG30758@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Add a variable to contain the regex needed to find weak functions
in the 'nm' output. This will allow other archs to easily override it.
Also rename the regex variable $nm_regex to $local_regex to be more
descriptive.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091028050619.GF30758@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Move the mcount section check to the beginning of the objdump read loop.
This makes the code easier to follow since the search for the mcount
section is performed first before the mcount callers are processed.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091028050523.GE30758@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The current logic to check objcopy's version is incorrect. This patch
fixes the algorithm and disables the use of local functions as a reference
if the objcopy version does not support static to global conversions.
Also remove some usused variables.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091028050421.GD30758@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The ftrace.c file may reference the mcount function and this may interfere
with the recordmcount.pl processing. To avoid this, the code does not
process the kernel/trace/ftrace.o. But currently the check is against
a relative path. This patch modifies the check to succeed if the path
is an absolute path.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091028050332.GC30758@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The number of arguments passed into recordmcount.pl is 10, but the code
checks if only 7 are passed in.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091027065733.GB22032@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The documentation currently says we will use the first function in a section
as a reference. The actual algorithm is: choose the first global function we
meet as a reference. If there is none, choose the first local one.
Change the documentation to be consistent with the code.
Also add several other clarifications.
Signed-off-by: Li Hong <lihong.hi@gmail.com>
LKML-Reference: <20091028050138.GA30758@uhli>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Based on the commit:
a586df06 "x86: Support __attribute__((__cold__)) in gcc 4.3"
some of the functions goes to the ".text.unlikely" section.
Looks like there's not many of them (I found printk, panic,
__ssb_dma_not_implemented, fat_fs_error), but still worth to
include I think.
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <20091013203426.175845614@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Roland Dreier found that a section that contained only a weak
function in one of the staging drivers and this caused
recordmcount.pl to spit out a warning and fail.
Although it is strange that a driver would have a weak function, and
this function only be used in one place, it should not be something
to make recordmcount.pl fail.
This patch fixes the issue in a simple manner: if only weak
functions exist in a section, then that section will not be
recorded.
Reported-by: Roland Dreier <rdreier@cisco.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
On Wed, 5 Aug 2009, Ingo Molnar wrote:
> * Dave Airlie <airlied@gmail.com> wrote:
>
> > Hey,
> >
> > So I spent 3-4 hrs today (I'm stupid yes) tracking down a .o
> > breakage by blaming rawhide gcc/binutils as I was using make
> > V=1and seeing only the compiler chain running,
>
> Hm, is this that powerpc related build bug you just reported?
Well we tracked it down and it is powerpc64 specific.
Seems that in drivers/hwmon/lm93.c there's a function called:
LM93_IN_FROM_REG()
But PPC64 has function descriptors and the real function names (the ones
you see in objdump) start with a '.'. Thus this in objdump you have:
Disassembly of section .text:
0000000000000000 <.LM93_IN_FROM_REG>:
0: 7c 08 02 a6 mflr r0
4: fb 81 ff e0 std r28,-32(r1)
The function name used is .LM93_IN_FROM_REG. But gcc considers symbols
that start with ".L" as a special symbol that is used inside the assembly
stage.
The nm passed into recordmcount uses the --synthetic option which shows
the ".L" symbols (my runs outside of the build did not include the
--synthetic option, so my older patch worked). We see the function as a
local.
Now to capture all the locations that use "mcount" we need to have a
reference to link into the object file a list of mcount callers. We need a
reference that will not disappear. We try to use a global function and if
that does not work, we use a local function as a reference. But to relink
the section back into the object, we need to make it global. In this case,
we run objcopy using --globalize-symbol and --localize-symbol to convert
the symbol into a global symbol, link the mcount list, then convert it
back to a local symbol.
This works great except for this case. .L* symbols can not be converted
into a global symbol, and the mcount section referencing it will remain
unresolved.
Reported-by: Dave Airlie <airlied@gmail.com>
LKML-Reference: <alpine.DEB.2.00.0908052011590.5010@gandalf.stny.rr.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The value of $offset should be the offset of $ref_func from the
beginning of the object file. Therefore, we should set both variables
together.
This fixes a bug I was hitting on sh where $offset (which is used to
calcualte the addends for the __mcount_loc entries) was being set
multiple times and didn't correspond to $ref_func's offset in the object
file. The addends in __mcount_loc were calculated incorrectly, resulting
in ftrace dynamically modifying addresses that weren't mcount call
sites.
Signed-off-by: Matt Fleming <matt@console-pimps.org>
LKML-Reference: <1248365775-25196-2-git-send-email-matt@console-pimps.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Fix the conditional that checks if we already have a $ref_func and that
the new function is weak. The code as previously checking whether either
condition was false, and we really need to only update $ref_func is both
cconditions are false.
Signed-off-by: Matt Fleming <matt@console-pimps.org>
LKML-Reference: <1248365775-25196-1-git-send-email-matt@console-pimps.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
I was reading throught the recordmcount.pl starting comment,
and spotted a tiny discrepancy.
The second example is about my_func not being global, but the
example code has the ".globl my_func" statement just moved.
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: rostedt@goodmis.org
LKML-Reference: <1247773468-11594-4-git-send-email-jolsa@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The only references in the kernel to the .text.sched section are in
recordmcount.pl. Since the code it has is intended to be example code
it should refer to real kernel sections. So change it to .sched.text
instead.
[ Impact: consistency in comments ]
Signed-off-by: Tim Abbott <tabbott@mit.edu>
LKML-Reference: <1241136371-10768-1-git-send-email-tabbott@mit.edu>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Impact: fix failure of dynamic function tracer selftest
In a course of development, a developer does several makes on their
kernel. Sometimes, the make might do something abnormal. In the
case of running the recordmcount.pl script on an object twice,
the script will duplicate all the calls to mcount in the __mcount_loc
section.
On boot up, the dynamic function tracer is careful when it modifies
code, and performs several consistency checks. One is to not modify
the call site if it is not what it expects it to be. If a function
call site is listed twice, the first entry will convert the site
to a nop, and the second will fail because it expected to see a
call to mcount, but instead it sees a nop. Thus, the function tracer
is disabled.
Eric Sesterhenn reported seeing:
[ 1.055440] ftrace: converting mcount calls to 0f 1f 44 00 00
[ 1.055568] ftrace: allocating 29418 entries in 116 pages
[ 1.061000] ------------[ cut here ]------------
[ 1.061000] WARNING: at kernel/trace/ftrace.c:441
[...]
[ 1.060000] ---[ end trace 4eaa2a86a8e2da23 ]---
[ 1.060000] ftrace failed to modify [<c0118072>] check_corruption+0x3/0x2d
[ 1.060000] actual: 0f:1f:44:00:00
This warning shows that check_corruption+0x3 already had a nop in
its place (0x0f1f440000). After compiling another kernel the problem
went away.
Later Eric Paris notice the same type of issue. Luckily, he saved
the vmlinux file that caused it. In the file we found a bunch of
duplicate mcount call site records, which lead us to the script.
Perhaps this problem only happens to people named Eric.
This patch changes the script to test if the __mcount_loc already
exists in the object file, and if it does, it will print out
an error message and kill the compile.
Reported-by: Eric Sesterhenn <snakebyte@gmx.de>
Reported-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In IA64, a function pointer isn't a 'unsigned long' but a
'struct {unsigned long ip, unsigned long gp}'. MCOUNT_ADDR is determined
at link time not compile time, so explictly ignore kernel/trace/ftrace.o
in recordmcount.pl.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In IA64, module build and kernel build use different option.
Make recordmcount.pl differentiate the two cases.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: let the function-graph-tracer be aware of the irq entrypoints
Add a new .irqentry.text section to store the irq entrypoints functions
inside the same section. This way, the tracer will be able to signal
an interrupts triggering on output by recognizing these entrypoints.
Also, make this section recordable for dynamic tracing.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: widen the scope of recordmcount.pl
Besides .text section, there are three .text sections that won't
be freed after kernel booting. They are: .sched.text, .spinlock.text
and .kprobes.text, which contain functions we can trace. But the last
section ".kprobes.text" is particular, which has been marked as "notrace",
we ignore it. Thus we add other two sections.
Signed-off-by: Liming Wang <liming.wang@windriver.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: extend scripts/recordmcount.pl to ARM
Arm uses %progbits instead of @progbits and requires only 4 byte alignment.
[ Thanks to Sam Ravnborg for mentioning that ARM uses %progbits ]
Signed-off-by: Jim Radford <radford@galvanix.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: extend scripts/recordmcount.pl with default alignment for SH
Set $alignment=2 for the sh architecture so that a ".align 2" directive
will be emitted for all __mcount_loc sections. Fix a whitspace error
while I'm here (converted spaces to tabs).
Signed-off-by: Matt Fleming <mjf@gentoo.org>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup of recordmcount.pl
Now that more architectures are being ported to the MCOUNT_RECORD
method, there is no reason to have each declare their own arch
specific variable if most of them share the same value. This patch
creates a set of default values for the arch specific variables
based off of i386.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: Add PowerPC port to recordmcount.pl script
This patch updates the recordmcount.pl script to process
PowerPC.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
First cut at dynamic ftrace support.
[
Steven Rostedt - only updated the recordmcount.pl file.
There are updates for PowerPC that will conflict with this,
and we need to base off of these changes.
]
Signed-off-by: Matt Fleming <mjf@gentoo.org>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: add alignment option for recordmcount.pl script
Align the __mcount_loc sections so that architectures with strict
alignment requirements need not worry about performing unaligned
accesses.
This fixes an issue where I was seeing unaligned accesses, which are not
supported on our architecture (the results of an unaligned access are
undefined).
Signed-off-by: Matt Fleming <matthew.fleming@imgtec.com>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The text section stays in memory without ever leaving. With the exception
of modules, but modules know how to handle that case. With the dynamic
ftrace tracer, we need to make sure that it does not try to modify code
that no longer exists. The only safe section is .text.
This patch changes the recordmcount script to only record the mcount calls
in the .text sections.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The recordmcount script requires that the actual arch is passed in.
This works well when ARCH=i386 or ARCH=x86_64 but does not handle the
case of ARCH=x86.
This patch adds a parameter to the function to pass in the number of
bits of the architecture. So that it can determine if x86 should be
run for x86_64 or i386 archs.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The warning messages about old objcopy and local functions spam the
user quite drastically. Remove the warning until we can find a nicer
way of tell the user to upgrade their objcopy.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The --globalize-symbols option came out in objcopy version 2.17.
If the kernel is being compiled on a system with a lower version of
objcopy, then we can not use the globalize / localize trick to
link to symbols pointing to local functions.
This patch tests the version of objcopy and will only use the trick
if the version is greater than or equal to 2.17. Otherwise, if an
object has only local functions within a section, it will give a
nice warning and recommend the user to upgrade their objcopy.
Leaving the symbols unrecorded is not that big of a deal, since the
mcount record method changes the actual mcount code to be a simple
"ret" without recording registers or anything.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
During tests and checks, I've discovered that there were failures to
convert mcount callers into nops. Looking deeper into these failures,
code that was attempted to be changed was not an mcount caller.
The current code only updates if the code being changed is what it expects,
but I still investigate any time there is a failure.
What was happening is that a weak symbol was being used as a reference
for other mcount callers. That weak symbol was also referenced elsewhere
so the offsets were using the strong symbol and not the function symbol
that it was referenced from.
This patch changes the setting up of the mcount_loc section to search
for a global function that is not weak. It will pick a local over a weak
but if only a weak is found in a section, a warning is printed and the
mcount location is not recorded (just to be safe).
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
I'm trying to keep all the arch changes in recordmcount.pl in one place.
I moved your code into that area, by adding the flags to the commands
that were passed in.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
hack around:
ld: Relocatable linking with relocations from format elf32-i386 (init/.tmp_gl_calibrate.o) to format elf64-x86-64 (init/.tmp_mx_calibrate.o) i CC arch/x86/mm/extable.o
objcopy: 'init/.tmp_mx_calibrate.o': No such file
rm: cannot remove `init/.tmp_mx_calibrate.o': No such file or directory
ld: Relocatable linking with relocations from format elf32-i386 (arch/x86/mm/extable.o) to format elf64-x86-64 (arch/x86/mm/.tmp_mx_extable.o) is not supported
mv: cannot stat `arch/x86/mm/.tmp_mx_extable.o': No such file or directory
ld: Relocatable linking with relocations from format elf32-i386 (arch/x86/mm/fault.o) to format elf64-x86-64 (arch/x86/mm/.tmp_mx_fault.o) is not supported
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This patch creates a section in the kernel called "__mcount_loc".
This will hold a list of pointers to the mcount relocation for
each call site of mcount.
For example:
objdump -dr init/main.o
[...]
Disassembly of section .text:
0000000000000000 <do_one_initcall>:
0: 55 push %rbp
[...]
000000000000017b <init_post>:
17b: 55 push %rbp
17c: 48 89 e5 mov %rsp,%rbp
17f: 53 push %rbx
180: 48 83 ec 08 sub $0x8,%rsp
184: e8 00 00 00 00 callq 189 <init_post+0xe>
185: R_X86_64_PC32 mcount+0xfffffffffffffffc
[...]
We will add a section to point to each function call.
.section __mcount_loc,"a",@progbits
[...]
.quad .text + 0x185
[...]
The offset to of the mcount call site in init_post is an offset from
the start of the section, and not the start of the function init_post.
The mcount relocation is at the call site 0x185 from the start of the
.text section.
.text + 0x185 == init_post + 0xa
We need a way to add this __mcount_loc section in a way that we do not
lose the relocations after final link. The .text section here will
be attached to all other .text sections after final link and the
offsets will be meaningless. We need to keep track of where these
.text sections are.
To do this, we use the start of the first function in the section.
do_one_initcall. We can make a tmp.s file with this function as a reference
to the start of the .text section.
.section __mcount_loc,"a",@progbits
[...]
.quad do_one_initcall + 0x185
[...]
Then we can compile the tmp.s into a tmp.o
gcc -c tmp.s -o tmp.o
And link it into back into main.o.
ld -r main.o tmp.o -o tmp_main.o
mv tmp_main.o main.o
But we have a problem. What happens if the first function in a section
is not exported, and is a static function. The linker will not let
the tmp.o use it. This case exists in main.o as well.
Disassembly of section .init.text:
0000000000000000 <set_reset_devices>:
0: 55 push %rbp
1: 48 89 e5 mov %rsp,%rbp
4: e8 00 00 00 00 callq 9 <set_reset_devices+0x9>
5: R_X86_64_PC32 mcount+0xfffffffffffffffc
The first function in .init.text is a static function.
00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 t set_reset_devices
The lowercase 't' means that set_reset_devices is local and is not exported.
If we simply try to link the tmp.o with the set_reset_devices we end
up with two symbols: one local and one global.
.section __mcount_loc,"a",@progbits
.quad set_reset_devices + 0x10
00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 t set_reset_devices
U set_reset_devices
We still have an undefined reference to set_reset_devices, and if we try
to compile the kernel, we will end up with an undefined reference to
set_reset_devices, or even worst, it could be exported someplace else,
and then we will have a reference to the wrong location.
To handle this case, we make an intermediate step using objcopy.
We convert set_reset_devices into a global exported symbol before linking
it with tmp.o and set it back afterwards.
00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 T set_reset_devices
00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 T set_reset_devices
00000000000000a8 t __setup_set_reset_devices
000000000000105f t __setup_str_set_reset_devices
0000000000000000 t set_reset_devices
Now we have a section in main.o called __mcount_loc that we can place
somewhere in the kernel using vmlinux.ld.S and access it to convert
all these locations that call mcount into nops before starting SMP
and thus, eliminating the need to do this with kstop_machine.
Note, A well documented perl script (scripts/recordmcount.pl) is used
to do all this in one location.
Signed-off-by: Steven Rostedt <srostedt@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
