Re: [PATCH] m68k: don't use lea on profiling

[Maxim Kuvyrkov <maxim@codesourcery.com>]

Jeff Law wrote:
> On 11/19/09 06:50, Maxim Kuvyrkov wrote:
>> Sebastian Andrzej Siewior wrote:
>>> this will not work on large binaries:
>>>
>>> |bigeasy@ayoka:~/cc$ echo 'int main(void) { return 0; }' > file.c
>>> |bigeasy@ayoka:~/cc$ m68k-linux-gnu-gcc -o file file.c -static -pg
>>> |/tmp/ccw33VYP.o: In function `main':
>>> |file.c:(.text+0x6): relocation truncated to fit: R_68K_PC16 against 
>>> `.data'
>>> |collect2: ld returned 1 exit status
>>>
>>> This is reported as #36047. It appears that the generated label is 
>>> unused.
>>
>> As I noted in the PR 
>> <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36047> neither GLIBC nor 
>> uClibc need the argument so we may simply remove it (thus getting rid 
>> of the PC-relative relocation that causes the problem).
>>
>> Andreas, does this look OK to you?
> My areas of concern would be:
> 
>   1. Does anyone hook the mcount call and expect the a0/a1 to have 
> reasonable values.  For example, the kernel tracing code?

I don't think so.  M68K kernel does not define any target-specific code 
for profiling.

> 
>   2. When did glibc change to its new approach?  If it was recent, then 
> we probably can't assume the newer implementation.

I looked through revision history of glibc-port and didn't find any 
mentions of _mcount implementation in the m68k directories since GLIBC 2.3.

-- 
Maxim Kuvyrkov
CodeSourcery
maxim@codesourcery.com
(650) 331-3385 x724