On Sun, Apr 6, 2014 at 7:08 AM, Jonas Gorski <jogo@xxxxxxxxxxx> wrote:
> On Sun, Apr 6, 2014 at 12:52 AM, Manuel Lauss <manuel.lauss@xxxxxxxxx> wrote:
>> Add a config option to disable the FPU emulator.
>> Saves around 56kB on 32bit, and kills FPU users with SIGILL.
>
> Looks similar to what we do in OpenWrt[1]. Main differences are we
> also stub out process_fpemu_return and emit a notice into the kernel
> log that the FPU emulator is disabled.
Does Florian have any plans to upstream it?
I kept process_fpuemu_return() because it actually emits the SIGILL I want,
and that was the quickest way to get the result I want.
>> --- a/arch/mips/Kbuild
>> +++ b/arch/mips/Kbuild
>> @@ -16,7 +16,9 @@ obj- := $(platform-)
>>
>> obj-y += kernel/
>> obj-y += mm/
>> +ifdef CONFIG_MIPS_FPU_EMULATOR
>> obj-y += math-emu/
>> +endif
>
> You can write this as obj-$(CONFIG_MIPS_FPU_EMULATOR) += math-emu/
Done...
>> --- a/arch/mips/include/asm/fpu.h
>> +++ b/arch/mips/include/asm/fpu.h
>> @@ -161,7 +161,9 @@ static inline int init_fpu(void)
>> if (!ret)
>> _init_fpu();
>> } else {
>> +#ifdef CONFIG_MIPS_FPU_EMULATOR
>> fpu_emulator_init_fpu();
>> +#endif
>
> Maybe do something like
>
> } else if (IS_ENABLED(CONFIG_MIPS_FPU_EMULATOR)) {
> fpu_emulator_init_fpu();
>
> which kooks a bit nicer IMHO.
and done.
>> --- a/arch/mips/include/asm/fpu_emulator.h
>> +++ b/arch/mips/include/asm/fpu_emulator.h
>> @@ -53,13 +53,35 @@ do { \
[...]
>> +static inline int mm_isBranchInstr(struct pt_regs *regs,
>> + struct mm_decoded_insn dec_insn,
>> + unsigned long *contpc)
>> +{
>> + return 0;
>> +}
>
> Won't this break micromips support? mm_isBranchInstr is called from a
> few other places outside the FPU emulator.
True. I get the impression that it doesn't really belong in the fpu emu code,
it looks like generic umips support code for e.g. kernel/branch.c
Thanks!
Mano
