On Wed, 8 Sept 2021 at 00:00, Keith Packard <keithpac@xxxxxxxxxx> wrote:
>
> We're going to store TPIDRPRW here instead
>
?
> Signed-off-by: Keith Packard <keithpac@xxxxxxxxxx>
I'd much prefer to keep using TPIDIRPRW for the per-CPU offsets, and
use the user space TLS register for current.
There are several reasons for this:
- arm64 does the same - as someone who still cares about ARM while
many have moved on to arm64 or RISC-V, I am still trying to maintain
parity between ARM and arm64 where possible.
- efficiency: loading the per-CPU offset using a CPU id stored in
memory, which is then used to index the per-CPU offsets array in
memory adds two additional loads to every load/store of a per-CPU
variable
- 'current' usually does not change value under the code's feet,
whereas per-CPU offsets might change at any time. Given the fact that
the CPU offset load is visible to the compiler as a memory access, I
suppose this should be safe, but I would still prefer per-CPU access
to avoid going via current if possible.
> ---
> arch/arm/include/asm/percpu.h | 31 -------------------------------
> arch/arm/kernel/setup.c | 7 -------
> arch/arm/kernel/smp.c | 3 ---
> 3 files changed, 41 deletions(-)
>
> diff --git a/arch/arm/include/asm/percpu.h b/arch/arm/include/asm/percpu.h
> index e2fcb3cfd3de..eeafcd6a3e01 100644
> --- a/arch/arm/include/asm/percpu.h
> +++ b/arch/arm/include/asm/percpu.h
> @@ -7,37 +7,6 @@
>
> register unsigned long current_stack_pointer asm ("sp");
>
> -/*
> - * Same as asm-generic/percpu.h, except that we store the per cpu offset
> - * in the TPIDRPRW. TPIDRPRW only exists on V6K and V7
> - */
> -#if defined(CONFIG_SMP) && !defined(CONFIG_CPU_V6)
> -static inline void set_my_cpu_offset(unsigned long off)
> -{
> - /* Set TPIDRPRW */
> - asm volatile("mcr p15, 0, %0, c13, c0, 4" : : "r" (off) : "memory");
> -}
> -
> -static inline unsigned long __my_cpu_offset(void)
> -{
> - unsigned long off;
> -
> - /*
> - * Read TPIDRPRW.
> - * We want to allow caching the value, so avoid using volatile and
> - * instead use a fake stack read to hazard against barrier().
> - */
> - asm("mrc p15, 0, %0, c13, c0, 4" : "=r" (off)
> - : "Q" (*(const unsigned long *)current_stack_pointer));
> -
> - return off;
> -}
> -#define __my_cpu_offset __my_cpu_offset()
> -#else
> -#define set_my_cpu_offset(x) do {} while(0)
> -
> -#endif /* CONFIG_SMP */
> -
> #include <asm-generic/percpu.h>
>
> #endif /* _ASM_ARM_PERCPU_H_ */
> diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c
> index ca0201635fac..d0dc60afe54f 100644
> --- a/arch/arm/kernel/setup.c
> +++ b/arch/arm/kernel/setup.c
> @@ -590,13 +590,6 @@ void __init smp_setup_processor_id(void)
> for (i = 1; i < nr_cpu_ids; ++i)
> cpu_logical_map(i) = i == cpu ? 0 : i;
>
> - /*
> - * clear __my_cpu_offset on boot CPU to avoid hang caused by
> - * using percpu variable early, for example, lockdep will
> - * access percpu variable inside lock_release
> - */
> - set_my_cpu_offset(0);
> -
> pr_info("Booting Linux on physical CPU 0x%x\n", mpidr);
> }
>
> diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
> index 5e999f1f1aea..8ccf10b34f08 100644
> --- a/arch/arm/kernel/smp.c
> +++ b/arch/arm/kernel/smp.c
> @@ -410,8 +410,6 @@ asmlinkage void secondary_start_kernel(unsigned int cpu, struct task_struct *tas
> {
> struct mm_struct *mm = &init_mm;
>
> - set_my_cpu_offset(per_cpu_offset(cpu));
> -
> secondary_biglittle_init();
>
> /*
> @@ -495,7 +493,6 @@ void __init smp_cpus_done(unsigned int max_cpus)
>
> void __init smp_prepare_boot_cpu(void)
> {
> - set_my_cpu_offset(per_cpu_offset(smp_processor_id()));
> }
>
> void __init smp_prepare_cpus(unsigned int max_cpus)
> --
> 2.33.0
>
