2018-03-24 16:59:05

by Matthias Schiffer

[permalink] [raw]
Subject: [PATCH] mips: ftrace: fix static function graph tracing

ftrace_graph_caller was never run after calling ftrace_trace_function,
breaking the function graph tracer. Fix this, bringing it in line with the
x86 implementation.

While we're at it, also streamline the control flow of _mcount a bit to
reduce the number of branches.

This issue was reported before:
https://www.linux-mips.org/archives/linux-mips/2014-11/msg00295.html

Signed-off-by: Matthias Schiffer <[email protected]>
---

Caveats: I've only tested this on 32bit; it would be great if someone with
MIPS64 hardware or a working emulator setup could give it a spin. My test
device runs on kernel 4.9.y, but I don't expect any problems applying the
fix to newer kernels, given that this code is basically unchanged since
2014.

I'm not sure what the correct Fixes: line would be. I did not bother to
bisect the issue and could not find the commit introducing it by inspecting
the log (assuming the tracer worked at some point).


arch/mips/kernel/mcount.S | 27 ++++++++++++---------------
1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/arch/mips/kernel/mcount.S b/arch/mips/kernel/mcount.S
index f2ee7e1e3342..cff52b283e03 100644
--- a/arch/mips/kernel/mcount.S
+++ b/arch/mips/kernel/mcount.S
@@ -119,10 +119,20 @@ NESTED(_mcount, PT_SIZE, ra)
EXPORT_SYMBOL(_mcount)
PTR_LA t1, ftrace_stub
PTR_L t2, ftrace_trace_function /* Prepare t2 for (1) */
- bne t1, t2, static_trace
+ beq t1, t2, fgraph_trace
nop

+ MCOUNT_SAVE_REGS
+
+ move a0, ra /* arg1: self return address */
+ jalr t2 /* (1) call *ftrace_trace_function */
+ move a1, AT /* arg2: parent's return address */
+
+ MCOUNT_RESTORE_REGS
+
+fgraph_trace:
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
+ PTR_LA t1, ftrace_stub
PTR_L t3, ftrace_graph_return
bne t1, t3, ftrace_graph_caller
nop
@@ -131,24 +141,11 @@ EXPORT_SYMBOL(_mcount)
bne t1, t3, ftrace_graph_caller
nop
#endif
- b ftrace_stub
-#ifdef CONFIG_32BIT
- addiu sp, sp, 8
-#else
- nop
-#endif

-static_trace:
- MCOUNT_SAVE_REGS
-
- move a0, ra /* arg1: self return address */
- jalr t2 /* (1) call *ftrace_trace_function */
- move a1, AT /* arg2: parent's return address */
-
- MCOUNT_RESTORE_REGS
#ifdef CONFIG_32BIT
addiu sp, sp, 8
#endif
+
.globl ftrace_stub
ftrace_stub:
RETURN_BACK
--
2.16.2



2018-03-27 13:30:45

by Matt Redfearn

[permalink] [raw]
Subject: Re: [PATCH] mips: ftrace: fix static function graph tracing

Hi Matthias,

On 24/03/18 16:57, Matthias Schiffer wrote:
> ftrace_graph_caller was never run after calling ftrace_trace_function,
> breaking the function graph tracer. Fix this, bringing it in line with the
> x86 implementation.
>
> While we're at it, also streamline the control flow of _mcount a bit to
> reduce the number of branches.
>
> This issue was reported before:
> https://www.linux-mips.org/archives/linux-mips/2014-11/msg00295.html
>
> Signed-off-by: Matthias Schiffer <[email protected]>
> ---
>
> Caveats: I've only tested this on 32bit; it would be great if someone with
> MIPS64 hardware or a working emulator setup could give it a spin. My test

I've tested this patch fixes static function graph tracing on the
following platforms:
QEMU (MIPS32r2 P5600)
Creator Ci40 (MIPS32r2 InterAptiv)
Cavium Octeon II (MIPS64r2)
MIPS Boston FPGA with I6400 (MIPS64r6)

So
Tested-by: Matt Redfearn <[email protected]>

> device runs on kernel 4.9.y, but I don't expect any problems applying the
> fix to newer kernels, given that this code is basically unchanged since
> 2014.

Indeed, applies fine to v4.16-rc7 (which I mainly tested with)

>
> I'm not sure what the correct Fixes: line would be. I did not bother to
> bisect the issue and could not find the commit introducing it by inspecting
> the log (assuming the tracer worked at some point).

I verified that the patch applies to and fixes the issue as far back as
3.19 on the malta platform using qemu. Going further back I ran into
toolchain issues that prevented builds. The earliest version the patch
applies cleanly to is v3.17.
Like you, I don't see where this got broken from just looking at
mcount.S, and not being able to bisect with build / test makes it rather
difficult to find out.

Thanks,
Matt

>
>
> arch/mips/kernel/mcount.S | 27 ++++++++++++---------------
> 1 file changed, 12 insertions(+), 15 deletions(-)
>
> diff --git a/arch/mips/kernel/mcount.S b/arch/mips/kernel/mcount.S
> index f2ee7e1e3342..cff52b283e03 100644
> --- a/arch/mips/kernel/mcount.S
> +++ b/arch/mips/kernel/mcount.S
> @@ -119,10 +119,20 @@ NESTED(_mcount, PT_SIZE, ra)
> EXPORT_SYMBOL(_mcount)
> PTR_LA t1, ftrace_stub
> PTR_L t2, ftrace_trace_function /* Prepare t2 for (1) */
> - bne t1, t2, static_trace
> + beq t1, t2, fgraph_trace
> nop
>
> + MCOUNT_SAVE_REGS
> +
> + move a0, ra /* arg1: self return address */
> + jalr t2 /* (1) call *ftrace_trace_function */
> + move a1, AT /* arg2: parent's return address */
> +
> + MCOUNT_RESTORE_REGS
> +
> +fgraph_trace:
> #ifdef CONFIG_FUNCTION_GRAPH_TRACER
> + PTR_LA t1, ftrace_stub
> PTR_L t3, ftrace_graph_return
> bne t1, t3, ftrace_graph_caller
> nop
> @@ -131,24 +141,11 @@ EXPORT_SYMBOL(_mcount)
> bne t1, t3, ftrace_graph_caller
> nop
> #endif
> - b ftrace_stub
> -#ifdef CONFIG_32BIT
> - addiu sp, sp, 8
> -#else
> - nop
> -#endif
>
> -static_trace:
> - MCOUNT_SAVE_REGS
> -
> - move a0, ra /* arg1: self return address */
> - jalr t2 /* (1) call *ftrace_trace_function */
> - move a1, AT /* arg2: parent's return address */
> -
> - MCOUNT_RESTORE_REGS
> #ifdef CONFIG_32BIT
> addiu sp, sp, 8
> #endif
> +
> .globl ftrace_stub
> ftrace_stub:
> RETURN_BACK
>