Hi Manuel, Sorry for taking a while to respond to you, from a toolchain perspective I think this patch is looking good with one extra change as below. I'll leave the kernel guys to comment further. Manuel Lauss <manuel.lauss@xxxxxxxxx> writes: > diff --git a/arch/mips/kernel/r4k_fpu.S b/arch/mips/kernel/r4k_fpu.S > index 8352523..f21c046 100644 > --- a/arch/mips/kernel/r4k_fpu.S > +++ b/arch/mips/kernel/r4k_fpu.S > @@ -33,10 +34,14 @@ > .set arch=r4000 > > LEAF(_save_fp_context) > + .set push > + SET_HARDFLOAT > cfc1 t1, fcr31 > + .set pop > > #if defined(CONFIG_64BIT) || defined(CONFIG_CPU_MIPS32_R2) > .set push > + SET_HARDFLOAT > #ifdef CONFIG_CPU_MIPS32_R2 > .set mips64r2 I know you haven't modified the lines with .set mips64r2 on them but they are not strictly correct. Instead of mips64r2 (which is dangerous to put in 32-bit code for a variety of reasons unless absolutely critical)... this should be: .set mips32r2 .set fp=64 There should be no need to protect this change based on assembler version as I believe .set fp=... support predates .set mips64r2. This change protects the kernel source from failing to build depending on how future toolchains are configured and it would be ideal to include in this patch to keep all these tweaks in one place. For the record the issue is that .set mips64r2 is getting used with the O32 ABI which is OK albeit weird. Under some circumstances the use of .set mips64r2 will result in implicitly setting fp=64 which is required for the SDC1 instructions which follow, however it will not always set fp=64. Markos has a set of tools which are configured such that .set mips64r2 used with an O32 ABI will not switch to 64-bit FP registers which is the reason he has seen a problem which you can't reproduce. > @@ -150,6 +161,7 @@ LEAF(_restore_fp_context) > > #if defined(CONFIG_64BIT) || defined(CONFIG_CPU_MIPS32_R2) > .set push > + SET_HARDFLOAT > #ifdef CONFIG_CPU_MIPS32_R2 > .set mips64r2 Likewise. Thanks, Matthew
