On Thu, Dec 21, 2017 at 12:52:01AM +1000, Nicholas Piggin wrote:
> Shifted left by 16 bits, so the low 16 bits of r14 remain available.
> This allows per-cpu pointers to be dereferenced with a single extra
> shift whereas previously it was a load and add.
> ---
> arch/powerpc/include/asm/paca.h | 5 +++++
> arch/powerpc/include/asm/percpu.h | 2 +-
> arch/powerpc/kernel/entry_64.S | 5 -----
> arch/powerpc/kernel/head_64.S | 5 +----
> arch/powerpc/kernel/setup_64.c | 11 +++++++++--
> 5 files changed, 16 insertions(+), 12 deletions(-)
>
> diff --git a/arch/powerpc/include/asm/paca.h b/arch/powerpc/include/asm/paca.h
> index cd6a9a010895..4dd4ac69e84f 100644
> --- a/arch/powerpc/include/asm/paca.h
> +++ b/arch/powerpc/include/asm/paca.h
> @@ -35,6 +35,11 @@
>
> register struct paca_struct *local_paca asm("r13");
> #ifdef CONFIG_PPC_BOOK3S
> +/*
> + * The top 32-bits of r14 is used as the per-cpu offset, shifted by
> PAGE_SHIFT.
Top 32, really? It's 48 in later comments.
Gabriel
> + * The per-cpu could be moved completely to vmalloc space if we had large
> + * vmalloc page mapping? (no, must access it in real mode).
> + */
> register u64 local_r14 asm("r14");
> #endif
>
> diff --git a/arch/powerpc/include/asm/percpu.h
> b/arch/powerpc/include/asm/percpu.h
> index dce863a7635c..1e0d79d30eac 100644
> --- a/arch/powerpc/include/asm/percpu.h
> +++ b/arch/powerpc/include/asm/percpu.h
> @@ -12,7 +12,7 @@
>
> #include <asm/paca.h>
>
> -#define __my_cpu_offset local_paca->data_offset
> +#define __my_cpu_offset (local_r14 >> 16)
>
> #endif /* CONFIG_SMP */
> #endif /* __powerpc64__ */
> diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
> index 592e4b36065f..6b0e3ac311e8 100644
> --- a/arch/powerpc/kernel/entry_64.S
> +++ b/arch/powerpc/kernel/entry_64.S
> @@ -262,11 +262,6 @@ system_call_exit:
> BEGIN_FTR_SECTION
> stdcx. r0,0,r1 /* to clear the reservation */
> END_FTR_SECTION_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS)
> - LOAD_REG_IMMEDIATE(r10, 0xdeadbeefULL << 32)
> - mfspr r11,SPRN_PIR
> - or r10,r10,r11
> - tdne r10,r14
> -
> andi. r6,r8,MSR_PR
> ld r4,_LINK(r1)
>
> diff --git a/arch/powerpc/kernel/head_64.S b/arch/powerpc/kernel/head_64.S
> index 5a9ec06eab14..cdb710f43681 100644
> --- a/arch/powerpc/kernel/head_64.S
> +++ b/arch/powerpc/kernel/head_64.S
> @@ -413,10 +413,7 @@ generic_secondary_common_init:
> b kexec_wait /* next kernel might do better */
>
> 2: SET_PACA(r13)
> - LOAD_REG_IMMEDIATE(r14, 0xdeadbeef << 32)
> - mfspr r3,SPRN_PIR
> - or r14,r14,r3
> - std r14,PACA_R14(r13)
> + ld r14,PACA_R14(r13)
>
> #ifdef CONFIG_PPC_BOOK3E
> addi r12,r13,PACA_EXTLB /* and TLB exc frame in another */
> diff --git a/arch/powerpc/kernel/setup_64.c b/arch/powerpc/kernel/setup_64.c
> index 9a4c5bf35d92..f4a96ebb523a 100644
> --- a/arch/powerpc/kernel/setup_64.c
> +++ b/arch/powerpc/kernel/setup_64.c
> @@ -192,8 +192,8 @@ static void __init fixup_boot_paca(void)
> get_paca()->data_offset = 0;
> /* Mark interrupts disabled in PACA */
> irq_soft_mask_set(IRQ_SOFT_MASK_STD);
> - /* Set r14 and paca_r14 to debug value */
> - get_paca()->r14 = (0xdeadbeefULL << 32) | mfspr(SPRN_PIR);
> + /* Set r14 and paca_r14 to zero */
> + get_paca()->r14 = 0;
> local_r14 = get_paca()->r14;
> }
>
> @@ -761,7 +761,14 @@ void __init setup_per_cpu_areas(void)
> for_each_possible_cpu(cpu) {
> __per_cpu_offset[cpu] = delta + pcpu_unit_offsets[cpu];
> paca[cpu].data_offset = __per_cpu_offset[cpu];
> +
> + BUG_ON(paca[cpu].data_offset & (PAGE_SIZE-1));
> + BUG_ON(paca[cpu].data_offset >= (1UL << (64 - 16)));
> +
> + /* The top 48 bits are used for per-cpu data */
> + paca[cpu].r14 |= paca[cpu].data_offset << 16;
> }
> + local_r14 = paca[smp_processor_id()].r14;
> }
> #endif
>
> --
> 2.15.0
