On 2015-12-24 13:57, James Hogan wrote:
> On Thu, Dec 24, 2015 at 12:48:12PM +0000, James Hogan wrote:
>> Hi Guenter,
>>
>> On Wed, Dec 23, 2015 at 09:04:31PM -0800, Guenter Roeck wrote:
>>> Commit 2a037f310bab ("MIPS: VDSO: Fix build error") tries to fix a build
>>> error seen with binutils 2.24 and earlier. However, the fix does not work,
>>> and again results in the already known build errors if the kernel is built
>>> with an earlier version of binutils.
>>>
>>> CC arch/mips/vdso/gettimeofday.o
>>> /tmp/ccnOVbHT.s: Assembler messages:
>>> /tmp/ccnOVbHT.s:50: Error: can't resolve `_start' {*UND* section} - `L0 {.text section}
>>> /tmp/ccnOVbHT.s:374: Error: can't resolve `_start' {*UND* section} - `L0 {.text section}
>>> scripts/Makefile.build:258: recipe for target 'arch/mips/vdso/gettimeofday.o' failed
>>> make[2]: *** [arch/mips/vdso/gettimeofday.o] Error 1
>>>
>>> Fixes: 2a037f310bab ("MIPS: VDSO: Fix build error")
>>> Cc: Qais Yousef <qais.yousef@xxxxxxxxxx>
>>> Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>
>>> ---
>>> Tested with binutils 2.25 and 2.22.
>>>
>>> arch/mips/vdso/Makefile | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/arch/mips/vdso/Makefile b/arch/mips/vdso/Makefile
>>> index 018f8c7b94f2..14568900fc1d 100644
>>> --- a/arch/mips/vdso/Makefile
>>> +++ b/arch/mips/vdso/Makefile
>>> @@ -26,7 +26,7 @@ aflags-vdso := $(ccflags-vdso) \
>>> # the comments on that file.
>>> #
>>> ifndef CONFIG_CPU_MIPSR6
>>> - ifeq ($(call ld-ifversion, -lt, 22500000, y),)
>>> + ifeq ($(call ld-ifversion, -lt, 22500000, y),y)
>>
>> I agree this is semantically correct, but there is something more evil
>> going on here.
>>
>> Originally the check was version <= 2.24
>> Qais' patch changed it to version >= 2.25 (intending version < 2.25)
>> Your patch changes it to version < 2.25
>>
>> I think the reason this fixed the problem for Qais is actually that he
>> probably had a similar toolchain version to what I'm using:
>>
>> GNU ld (Codescape GNU Tools 2015.06-05 for MIPS MTI Linux) 2.24.90
>>
>> ./scripts/ld-version.sh does this:
>>
>> print a[1]*10000000 + a[2]*100000 + a[3]*10000 + a[4]*100 + a[5];
>>
>> which changes that version number into:
>> 20000000
>> + 2400000
>> + 900000 = 23300000
>>
>> I.e. it doesn't expect a[3] to be >= 10.
>>
>> Should we do something like this (increase multipliers on a[1] and
>> a[2])?:
>>
>> diff --git a/scripts/ld-version.sh b/scripts/ld-version.sh
>> index 198580d245e0..0b67edc5bc6f 100755
>> --- a/scripts/ld-version.sh
>> +++ b/scripts/ld-version.sh
>> @@ -3,6 +3,6 @@
>> {
>> gsub(".*)", "");
>> split($1,a, ".");
>> - print a[1]*10000000 + a[2]*100000 + a[3]*10000 + a[4]*100 + a[5];
>> + print a[1]*100000000 + a[2]*1000000 + a[3]*10000 + a[4]*100 + a[5];
>> exit
>> }
>>
>> which gives 2.24.90 => 224900000.
>>
>> All call sites would need updating too to add the extra 0, but a quick
>> git grep isn't showing any other ones than this one.
>
> Actually, linux-next includes this commit which uses ld-ifversion too:
>
> 19a3cc83353e3bb4bc28769f8606139a3d350d2d
> "Kbuild, lto: Add Link Time Optimization support v3"
That commit needs updating for other reasons, so feel free to fix
ld-ifversion and its usage in arch/mips.
Michal
