On 5/2/24 18:04, Puranjay Mohan wrote:
> Historically, arm64 implemented raw_smp_processor_id() as a read of
> current_thread_info()->cpu. This changed when arm64 moved thread_info
> into task struct, as at the time CONFIG_THREAD_INFO_IN_TASK made core
> code use thread_struct::cpu for the cpu number, and due to header
> dependencies prevented using this in raw_smp_processor_id(). As a
> workaround, we moved to using a percpu variable in commit:
>
> commit 57c82954e77f ("arm64: make cpu number a percpu variable")
>
> Since then, thread_info::cpu was reintroduced, and core code was made to
> use this in commits:
>
> commit 001430c1910d ("arm64: add CPU field to struct thread_info")
> commit bcf9033e5449 ("sched: move CPU field back into thread_info if
> THREAD_INFO_IN_TASK=y")
>
> Consequently it is possible to use current_thread_info()->cpu again.
This captures Mark's earlier suggestion on the commit message.
>
> This decreases the number of emitted instructions like in the following
> example:
>
> Dump of assembler code for function bpf_get_smp_processor_id:
> 0xffff8000802cd608 <+0>: nop
> 0xffff8000802cd60c <+4>: nop
> 0xffff8000802cd610 <+8>: adrp x0, 0xffff800082138000
> 0xffff8000802cd614 <+12>: mrs x1, tpidr_el1
> 0xffff8000802cd618 <+16>: add x0, x0, #0x8
> 0xffff8000802cd61c <+20>: ldrsw x0, [x0, x1]
> 0xffff8000802cd620 <+24>: ret
>
> After this patch:
>
> Dump of assembler code for function bpf_get_smp_processor_id:
> 0xffff8000802c9130 <+0>: nop
> 0xffff8000802c9134 <+4>: nop
> 0xffff8000802c9138 <+8>: mrs x0, sp_el0
> 0xffff8000802c913c <+12>: ldr w0, [x0, #24]
> 0xffff8000802c9140 <+16>: ret
>
> A microbenchmark[1] was built to measure the performance improvement
> provided by this change. It calls the following function given number of
> times and finds the runtime overhead:
>
> static noinline int get_cpu_id(void)
> {
> return smp_processor_id();
> }
>
> Run the benchmark like:
> modprobe smp_processor_id nr_function_calls=1000000000
>
> +--------------------------+------------------------+
> | | Number of Calls | Time taken |
> +--------+-----------------+------------------------+
> | Before | 1000000000 | 1602888401ns |
> +--------+-----------------+------------------------+
> | After | 1000000000 | 1206212658ns |
> +--------+-----------------+------------------------+
> | Difference (decrease) | 396675743ns (24.74%) |
> +---------------------------------------------------+
>
> Remove the percpu variable cpu_number as it is used only in
> set_smp_ipi_range() as a dummy variable to be passed to ipi_handler().
> Use irq_stat in place of cpu_number here.
>
> [1] https://github.com/puranjaymohan/linux/commit/77d3fdd
>
> Signed-off-by: Puranjay Mohan <puranjay@xxxxxxxxxx>
> ---
> Changes in v1 -> v2:
> v1: https://lore.kernel.org/all/20240501154236.10236-1-puranjay@xxxxxxxxxx/
> - Remove the percpu variable cpu_number
> - Add more information to the commit message.
> ---
> arch/arm64/include/asm/smp.h | 13 +------------
> arch/arm64/kernel/smp.c | 9 ++-------
> 2 files changed, 3 insertions(+), 19 deletions(-)
>
> diff --git a/arch/arm64/include/asm/smp.h b/arch/arm64/include/asm/smp.h
> index efb13112b408..2510eec026f7 100644
> --- a/arch/arm64/include/asm/smp.h
> +++ b/arch/arm64/include/asm/smp.h
> @@ -25,22 +25,11 @@
>
> #ifndef __ASSEMBLY__
>
> -#include <asm/percpu.h>
> -
> #include <linux/threads.h>
> #include <linux/cpumask.h>
> #include <linux/thread_info.h>
>
> -DECLARE_PER_CPU_READ_MOSTLY(int, cpu_number);
> -
> -/*
> - * We don't use this_cpu_read(cpu_number) as that has implicit writes to
> - * preempt_count, and associated (compiler) barriers, that we'd like to avoid
> - * the expense of. If we're preemptible, the value can be stale at use anyway.
> - * And we can't use this_cpu_ptr() either, as that winds up recursing back
> - * here under CONFIG_DEBUG_PREEMPT=y.
> - */
> -#define raw_smp_processor_id() (*raw_cpu_ptr(&cpu_number))
> +#define raw_smp_processor_id() (current_thread_info()->cpu)
>
> /*
> * Logical CPU mapping.
> diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> index 4ced34f62dab..98d4e352c3d0 100644
> --- a/arch/arm64/kernel/smp.c
> +++ b/arch/arm64/kernel/smp.c
> @@ -55,9 +55,6 @@
>
> #include <trace/events/ipi.h>
>
> -DEFINE_PER_CPU_READ_MOSTLY(int, cpu_number);
> -EXPORT_PER_CPU_SYMBOL(cpu_number);
> -
> /*
> * as from 2.5, kernels no longer have an init_tasks structure
> * so we need some other way of telling a new secondary core
> @@ -742,8 +739,6 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
> */
> for_each_possible_cpu(cpu) {
>
> - per_cpu(cpu_number, cpu) = cpu;
> -
> if (cpu == smp_processor_id())
> continue;
>
> @@ -1021,12 +1016,12 @@ void __init set_smp_ipi_range(int ipi_base, int n)
>
> if (ipi_should_be_nmi(i)) {
> err = request_percpu_nmi(ipi_base + i, ipi_handler,
> - "IPI", &cpu_number);
> + "IPI", &irq_stat);
> WARN(err, "Could not request IPI %d as NMI, err=%d\n",
> i, err);
> } else {
> err = request_percpu_irq(ipi_base + i, ipi_handler,
> - "IPI", &cpu_number);
> + "IPI", &irq_stat);
> WARN(err, "Could not request IPI %d as IRQ, err=%d\n",
> i, err);
> }
set_smp_ipi_range() now looks similar to what we have on arm32 platform.
Besides the arch/arm64/include/asm/smp.h changes (not sure if is just a
clean up, then might be a stand alone patch on its own) this patch LGTM.
