Date: Tue, 6 Dec 2016 21:43:55 -0500
From: Steven Rostedt <rostedt@goodmis.org>
To: Abel Vesa <abelvesa@linux.com>
Cc: linux@armlinux.org.uk, jpoimboe@redhat.com, jeyu@redhat.com,
        jikos@kernel.org, mbenes@suse.cz, pmladek@suse.com, mingo@redhat.com,
        gregkh@linuxfoundation.org, geert+renesas@glider.be,
        davem@davemloft.net, akpm@linux-foundation.org,
        emil.l.velikov@gmail.com, mchehab@kernel.org, linux@roeck-us.net,
        ard.biesheuvel@linaro.org, jens.wiklander@linaro.org,
        jean-philippe.brucker@arm.com, viro@zeniv.linux.org.uk,
        stefano.stabellini@eu.citrix.com, chris.brandt@renesas.com,
        linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        live-patching@vger.kernel.org
Subject: Re: [PATCH 4/7] arm: Add ftrace with regs support
Message-ID: <20161206214355.268d2791@grimm.local.home>
In-Reply-To: <1481043967-15602-5-git-send-email-abelvesa@linux.com>
References: <1481043967-15602-1-git-send-email-abelvesa@linux.com>
        <1481043967-15602-5-git-send-email-abelvesa@linux.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2258
Lines: 95

On Tue,  6 Dec 2016 17:06:04 +0000
Abel Vesa <abelvesa@linux.com> wrote:

> This adds __ftrace_regs_caller which, unlike __ftrace_caller,
> adds register saving/restoring and livepatch handling if
> the pc register gets modified by klp_ftrace_handler.
> 
> Signed-off-by: Abel Vesa <abelvesa@linux.com>
> ---
>  arch/arm/kernel/entry-ftrace.S | 49
> ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 49
> insertions(+)
> 
> diff --git a/arch/arm/kernel/entry-ftrace.S
> b/arch/arm/kernel/entry-ftrace.S index c73c403..b6ada5c 100644
> --- a/arch/arm/kernel/entry-ftrace.S
> +++ b/arch/arm/kernel/entry-ftrace.S
> @@ -92,6 +92,46 @@
>  2:	mcount_exit
>  .endm
>  
> +#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
> +
> +.macro __ftrace_regs_caller suffix
> +
> +	stmdb	sp!, {r0-r15}
> +	mov	r3, sp
> +
> +	ldr	r10, [sp, #60]
> +
> +	mcount_get_lr   r1                      @ lr of instrumented
> func
> +	mcount_adjust_addr	r0, lr		@
> instrumented function +
> +	.globl ftrace_regs_call\suffix
> +ftrace_regs_call\suffix:
> +	bl	ftrace_stub
> +
> +#ifdef CONFIG_FUNCTION_GRAPH_TRACER
> +	.globl ftrace_regs_graph_call\suffix
> +ftrace_regs_graph_call\suffix:
> +	mov	r0, r0
> +#endif

So basically what the below does is to undo the mcount prologue and
recall the new function as the old function is called.

> +#ifdef CONFIG_LIVEPATCH
> +	ldr	r0, [sp, #60]
> +	cmp	r0, r10
> +	beq	ftrace_regs_caller_end
> +	ldmia	sp!, {r0-r12}
> +	add	sp, sp, #8
> +	ldmia	sp!, {r11}
> +	sub	sp, r12, #16
> +	str	r11, [sp, #12]
> +	ldmia	sp!, {r11, r12, lr, pc}

But the above really could do with a lot of comments to explain exactly
what it is doing.

I don't condemn or condone this code. I'm just happy I don't have to
maintain it.

-- Steve


> +#endif
> +ftrace_regs_caller_end\suffix:
> +	ldmia	sp!, {r0-r14}
> +	add	sp, sp, #4
> +	mov	pc, lr
> +.endm
> +
> +#endif
> +
>  .macro __ftrace_caller suffix
>  	mcount_enter
>  
> @@ -212,6 +252,15 @@ UNWIND(.fnstart)
>  	__ftrace_caller
>  UNWIND(.fnend)
>  ENDPROC(ftrace_caller)
> +
> +#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
> +ENTRY(ftrace_regs_caller)
> +UNWIND(.fnstart)
> +	__ftrace_regs_caller
> +UNWIND(.fnend)
> +ENDPROC(ftrace_regs_caller)
> +#endif
> +
>  #endif
>  
>  #ifdef CONFIG_FUNCTION_GRAPH_TRACER