diff options
author | Michael Neuling <mikey@neuling.org> | 2008-06-25 14:07:18 +1000 |
---|---|---|
committer | Paul Mackerras <paulus@samba.org> | 2008-07-01 11:28:50 +1000 |
commit | ce48b2100785e5ca629fb3aa8e3b50aca808f692 (patch) | |
tree | 63532ff7cc68b18ca4902bd10e03fcbaaf01cade /arch/powerpc/kernel/process.c | |
parent | 72ffff5b1792b0fa4d40a8e2f3276fff999820ec (diff) |
powerpc: Add VSX context save/restore, ptrace and signal support
This patch extends the floating point save and restore code to use the
VSX load/stores when VSX is available. This will make FP context
save/restore marginally slower on FP only code, when VSX is available,
as it has to load/store 128bits rather than just 64bits.
Mixing FP, VMX and VSX code will get constant architected state.
The signals interface is extended to enable access to VSR 0-31
doubleword 1 after discussions with tool chain maintainers. Backward
compatibility is maintained.
The ptrace interface is also extended to allow access to VSR 0-31 full
registers.
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Diffstat (limited to 'arch/powerpc/kernel/process.c')
-rw-r--r-- | arch/powerpc/kernel/process.c | 107 |
1 files changed, 106 insertions, 1 deletions
diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c index 582df70439c..d52ded366f1 100644 --- a/arch/powerpc/kernel/process.c +++ b/arch/powerpc/kernel/process.c @@ -53,6 +53,7 @@ extern unsigned long _get_SP(void); #ifndef CONFIG_SMP struct task_struct *last_task_used_math = NULL; struct task_struct *last_task_used_altivec = NULL; +struct task_struct *last_task_used_vsx = NULL; struct task_struct *last_task_used_spe = NULL; #endif @@ -106,11 +107,23 @@ EXPORT_SYMBOL(enable_kernel_fp); int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs) { +#ifdef CONFIG_VSX + int i; + elf_fpreg_t *reg; +#endif + if (!tsk->thread.regs) return 0; flush_fp_to_thread(current); +#ifdef CONFIG_VSX + reg = (elf_fpreg_t *)fpregs; + for (i = 0; i < ELF_NFPREG - 1; i++, reg++) + *reg = tsk->thread.TS_FPR(i); + memcpy(reg, &tsk->thread.fpscr, sizeof(elf_fpreg_t)); +#else memcpy(fpregs, &tsk->thread.TS_FPR(0), sizeof(*fpregs)); +#endif return 1; } @@ -149,7 +162,7 @@ void flush_altivec_to_thread(struct task_struct *tsk) } } -int dump_task_altivec(struct task_struct *tsk, elf_vrregset_t *vrregs) +int dump_task_altivec(struct task_struct *tsk, elf_vrreg_t *vrregs) { /* ELF_NVRREG includes the VSCR and VRSAVE which we need to save * separately, see below */ @@ -179,6 +192,80 @@ int dump_task_altivec(struct task_struct *tsk, elf_vrregset_t *vrregs) } #endif /* CONFIG_ALTIVEC */ +#ifdef CONFIG_VSX +#if 0 +/* not currently used, but some crazy RAID module might want to later */ +void enable_kernel_vsx(void) +{ + WARN_ON(preemptible()); + +#ifdef CONFIG_SMP + if (current->thread.regs && (current->thread.regs->msr & MSR_VSX)) + giveup_vsx(current); + else + giveup_vsx(NULL); /* just enable vsx for kernel - force */ +#else + giveup_vsx(last_task_used_vsx); +#endif /* CONFIG_SMP */ +} +EXPORT_SYMBOL(enable_kernel_vsx); +#endif + +void flush_vsx_to_thread(struct task_struct *tsk) +{ + if (tsk->thread.regs) { + preempt_disable(); + if (tsk->thread.regs->msr & MSR_VSX) { +#ifdef CONFIG_SMP + BUG_ON(tsk != current); +#endif + giveup_vsx(tsk); + } + preempt_enable(); + } +} + +/* + * This dumps the lower half 64bits of the first 32 VSX registers. + * This needs to be called with dump_task_fp and dump_task_altivec to + * get all the VSX state. + */ +int dump_task_vsx(struct task_struct *tsk, elf_vrreg_t *vrregs) +{ + elf_vrreg_t *reg; + double buf[32]; + int i; + + if (tsk == current) + flush_vsx_to_thread(tsk); + + reg = (elf_vrreg_t *)vrregs; + + for (i = 0; i < 32 ; i++) + buf[i] = current->thread.fpr[i][TS_VSRLOWOFFSET]; + memcpy(reg, buf, sizeof(buf)); + + return 1; +} +#endif /* CONFIG_VSX */ + +int dump_task_vector(struct task_struct *tsk, elf_vrregset_t *vrregs) +{ + int rc = 0; + elf_vrreg_t *regs = (elf_vrreg_t *)vrregs; +#ifdef CONFIG_ALTIVEC + rc = dump_task_altivec(tsk, regs); + if (rc) + return rc; + regs += ELF_NVRREG; +#endif + +#ifdef CONFIG_VSX + rc = dump_task_vsx(tsk, regs); +#endif + return rc; +} + #ifdef CONFIG_SPE void enable_kernel_spe(void) @@ -233,6 +320,10 @@ void discard_lazy_cpu_state(void) if (last_task_used_altivec == current) last_task_used_altivec = NULL; #endif /* CONFIG_ALTIVEC */ +#ifdef CONFIG_VSX + if (last_task_used_vsx == current) + last_task_used_vsx = NULL; +#endif /* CONFIG_VSX */ #ifdef CONFIG_SPE if (last_task_used_spe == current) last_task_used_spe = NULL; @@ -297,6 +388,10 @@ struct task_struct *__switch_to(struct task_struct *prev, if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC)) giveup_altivec(prev); #endif /* CONFIG_ALTIVEC */ +#ifdef CONFIG_VSX + if (prev->thread.regs && (prev->thread.regs->msr & MSR_VSX)) + giveup_vsx(prev); +#endif /* CONFIG_VSX */ #ifdef CONFIG_SPE /* * If the previous thread used spe in the last quantum @@ -317,6 +412,10 @@ struct task_struct *__switch_to(struct task_struct *prev, if (new->thread.regs && last_task_used_altivec == new) new->thread.regs->msr |= MSR_VEC; #endif /* CONFIG_ALTIVEC */ +#ifdef CONFIG_VSX + if (new->thread.regs && last_task_used_vsx == new) + new->thread.regs->msr |= MSR_VSX; +#endif /* CONFIG_VSX */ #ifdef CONFIG_SPE /* Avoid the trap. On smp this this never happens since * we don't set last_task_used_spe @@ -417,6 +516,8 @@ static struct regbit { {MSR_EE, "EE"}, {MSR_PR, "PR"}, {MSR_FP, "FP"}, + {MSR_VEC, "VEC"}, + {MSR_VSX, "VSX"}, {MSR_ME, "ME"}, {MSR_IR, "IR"}, {MSR_DR, "DR"}, @@ -534,6 +635,7 @@ void prepare_to_copy(struct task_struct *tsk) { flush_fp_to_thread(current); flush_altivec_to_thread(current); + flush_vsx_to_thread(current); flush_spe_to_thread(current); } @@ -689,6 +791,9 @@ void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp) #endif discard_lazy_cpu_state(); +#ifdef CONFIG_VSX + current->thread.used_vsr = 0; +#endif memset(current->thread.fpr, 0, sizeof(current->thread.fpr)); current->thread.fpscr.val = 0; #ifdef CONFIG_ALTIVEC |