Re: [PATCH v6 1/9] ppc64 (le): prepare for -mprofile-kernel
From: Balbir Singh
Date: Thu Feb 04 2016 - 23:40:34 EST
On Thu, Feb 4, 2016 at 10:02 PM, Petr Mladek <pmladek@xxxxxxxx> wrote:
> On Thu 2016-02-04 18:31:40, AKASHI Takahiro wrote:
>> Jiri, Torsten
>>
>> Thank you for your explanation.
>>
>> On 02/03/2016 08:24 PM, Torsten Duwe wrote:
>> >On Wed, Feb 03, 2016 at 09:55:11AM +0100, Jiri Kosina wrote:
>> >>On Wed, 3 Feb 2016, AKASHI Takahiro wrote:
>> >>>those efforts, we are proposing[1] a new *generic* gcc option, -fprolog-add=N.
>> >>>This option will insert N nop instructions at the beginning of each function.
>> >
>> >>The interesting part of the story with ppc64 is that you indeed want to
>> >>create the callsite before the *most* of the prologue, but not really :)
>> >
>> >I was silently assuming that GCC would do this right on ppc64le; add the NOPs
>> >right after the TOC load. Or after TOC load and LR save? ...
>>
>> On arm/arm64, link register must be saved before any function call. So anyhow
>> we will have to add something, 3 instructions at the minimum, like:
>> save lr
>> branch _mcount
>> restore lr
>> <prologue>
>> ...
>> <body>
>> ...
>
> So, it is similar to PPC that has to handle LR as well.
>
>
>> >>The part of the prologue where TOC pointer is saved needs to happen before
>> >>the fentry/profiling call.
>> >
>> >Yes, any call, to any profiler/tracer/live patcher is potentially global
>> >and needs the _new_ TOC value.
>
> The code below is generated for PPC64LE with -mprofile-kernel using:
>
> $> gcc --version
> gcc (SUSE Linux) 6.0.0 20160121 (experimental) [trunk revision 232670]
> Copyright (C) 2016 Free Software Foundation, Inc.
> This is free software; see the source for copying conditions. There is NO
> warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
>
>
> 0000000000000050 <cmdline_proc_show>:
> 50: 00 00 4c 3c addis r2,r12,0
> 50: R_PPC64_REL16_HA .TOC.
> 54: 00 00 42 38 addi r2,r2,0
> 54: R_PPC64_REL16_LO .TOC.+0x4
> 58: a6 02 08 7c mflr r0
> 5c: 01 00 00 48 bl 5c <cmdline_proc_show+0xc>
> 5c: R_PPC64_REL24 _mcount
> 60: a6 02 08 7c mflr r0
> 64: 10 00 01 f8 std r0,16(r1)
> 68: a1 ff 21 f8 stdu r1,-96(r1)
> 6c: 00 00 22 3d addis r9,r2,0
> 6c: R_PPC64_TOC16_HA .toc
> 70: 00 00 82 3c addis r4,r2,0
> 70: R_PPC64_TOC16_HA .rodata.str1.8
> 74: 00 00 29 e9 ld r9,0(r9)
> 74: R_PPC64_TOC16_LO_DS .toc
> 78: 00 00 84 38 addi r4,r4,0
> 78: R_PPC64_TOC16_LO .rodata.str1.8
> 7c: 00 00 a9 e8 ld r5,0(r9)
> 80: 01 00 00 48 bl 80 <cmdline_proc_show+0x30>
> 80: R_PPC64_REL24 seq_printf
> 84: 00 00 00 60 nop
> 88: 00 00 60 38 li r3,0
> 8c: 60 00 21 38 addi r1,r1,96
> 90: 10 00 01 e8 ld r0,16(r1)
> 94: a6 03 08 7c mtlr r0
> 98: 20 00 80 4e blr
>
>
> And the same function compiled using:
>
> $> gcc --version
> gcc (SUSE Linux) 4.8.5
> Copyright (C) 2015 Free Software Foundation, Inc.
> This is free software; see the source for copying conditions. There is NO
> warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
>
>
> 0000000000000050 <cmdline_proc_show>:
> 50: 00 00 4c 3c addis r2,r12,0
> 50: R_PPC64_REL16_HA .TOC.
> 54: 00 00 42 38 addi r2,r2,0
> 54: R_PPC64_REL16_LO .TOC.+0x4
> 58: a6 02 08 7c mflr r0
> 5c: 10 00 01 f8 std r0,16(r1)
> 60: 01 00 00 48 bl 60 <cmdline_proc_show+0x10>
> 60: R_PPC64_REL24 _mcount
> 64: a6 02 08 7c mflr r0
> 68: 10 00 01 f8 std r0,16(r1)
> 6c: a1 ff 21 f8 stdu r1,-96(r1)
> 70: 00 00 42 3d addis r10,r2,0
> 70: R_PPC64_TOC16_HA .toc
> 74: 00 00 82 3c addis r4,r2,0
> 74: R_PPC64_TOC16_HA .rodata.str1.8
> 78: 00 00 2a e9 ld r9,0(r10)
> 78: R_PPC64_TOC16_LO_DS .toc
> 7c: 00 00 84 38 addi r4,r4,0
> 7c: R_PPC64_TOC16_LO .rodata.str1.8
> 80: 00 00 a9 e8 ld r5,0(r9)
> 84: 01 00 00 48 bl 84 <cmdline_proc_show+0x34>
> 84: R_PPC64_REL24 seq_printf
> 88: 00 00 00 60 nop
> 8c: 00 00 60 38 li r3,0
> 90: 60 00 21 38 addi r1,r1,96
> 94: 10 00 01 e8 ld r0,16(r1)
> 98: a6 03 08 7c mtlr r0
> 9c: 20 00 80 4e blr
>
>
> Please, note that are used either 3 or 4 instructions before the
> mcount location depending on the compiler version.
Thanks Petr
For big endian builds I saw
Dump of assembler code for function alloc_pages_current:
0xc000000000256f00 <+0>: mflr r0
0xc000000000256f04 <+4>: std r0,16(r1)
0xc000000000256f08 <+8>: bl 0xc000000000009e5c <.mcount>
0xc000000000256f0c <+12>: mflr r0
The offset is 8 bytes. Your earlier patch handled this by adding 16, I
suspect it needs revisiting
Balbir