Re: [PATCH v2 14/20] arm64: percpu: Implement preemptible read/write ops

[David Laight <david.laight.linux@gmail.com>]

On Wed, 5 Aug 2026 10:24:04 +0100
Ryan Roberts <ryan.roberts@arm.com> wrote:

> On 04/08/2026 18:04, Mark Rutland wrote:
> > Use the PCPU GPR infrastructure to implement preemptible this_cpu_read()
> > and this_cpu_write().
> > 
> > This change means that this_cpu_read() will always use a plain LDR, even
> > in LTO configurations where READ_ONCE() will use LDA[P]R. Using plain
> > LDR is preferable, given that the rationale for using LDA[P]R in
> > READ_ONCE() was to retain address dependencies against values written by
> > other CPUs, which isn't expected usage for this_cpu_read(). Using plain
> > LDR will enforce fewer ordering constraints.
> > 
> > Test case:
> > 
> > | void outline_this_cpu_write_u64(u64 __percpu *p, u64 v)
> > | {
> > | 	this_cpu_write(*p, v);
> > | }
> > 
> > Generated code before this patch (v7.2-rc4):
> > 
> > | <outline_this_cpu_write_u64>:
> > |        paciasp
> > |        stp     x29, x30, [sp, #-16]!
> > |        mrs     x2, sp_el0
> > |        mov     x29, sp
> > |        ldr     w3, [x2, #8]
> > |        add     w3, w3, #0x1
> > |        str     w3, [x2, #8]
> > |        mrs     x3, tpidr_el1
> > |        str     x1, [x0, x3]
> > |        ldr     x0, [x2, #8]
> > |        sub     x0, x0, #0x1
> > |        str     w0, [x2, #8]
> > |        cbz     x0, 1f
> > |        ldr     x0, [x2, #8]
> > |        cbnz    x0, 2f
> > | 1:     bl      preempt_schedule_notrace
> > | 2:     ldp     x29, x30, [sp], #16
> > |        autiasp
> > |        ret
> > 
> > Generated code after this patch:
> > 
> > | <outline_this_cpu_write_u64>:
> > |        mrs     x2, sp_el0
> > |        mov     w3, #0x7c60
> > |        strh    w3, [x2, #20]
> > |        mrs     x3, tpidr_el1
> > |        str     x1, [x0, x3]
> > |        strh    wzr, [x2, #20]
> > |        ret
> > 
> > Signed-off-by: Mark Rutland <mark.rutland@arm.com>
> > Cc: Ada Couprie Diaz <ada.coupriediaz@arm.com>
> > Cc: Ard Biesheuvel <ardb@kernel.org>
> > Cc: Catalin Marinas <catalin.marinas@arm.com>
> > Cc: James Morse <james.morse@arm.com>
> > Cc: Jinjie Ruan <ruanjinjie@huawei.com>
> > Cc: Marc Zyngier <maz@kernel.org>
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Vladimir Murzin <vladimir.murzin@arm.com>
> > Cc: Will Deacon <will@kernel.org>
> > Cc: Yang Shi <yang@os.amperecomputing.com>
> > ---  
> FYI I'm seeing build warnings caused by this patch (with ftrace enabled - based 
> on the warnings, I'm guessing that's the key bit), using:
> 
> aarch64-linux-gnu-gcc (Debian 14.2.0-19) 14.2.0, GNU ld (GNU Binutils for Debian) 2.44
> 
> I haven't investigated the cause.

It'll be the earlier patch that changed the casts.
The code is doing:
		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
			       FTRACE_PID_IGNORE);
where the constant is -1.
I suspect the warning messages are coming from the switch case that are
optimised away because the size if wrong (ftrace_ignore_pid seems to
be a pid number to ignore and is 32 bits.

That does seem a long-winded way to access per-cpu data.
To make any sense the code must be running with preemption disabled and
be caching info in per-cpu memory.
In which case it can just access it directly.
OTOH it could save the 'global' address of the per-cpu data and then access
it using the pointer so that access would be a normal one.

	David

> 
> Thanks,
> Ryan
> 
> ---8<---
> In file included from linux/arch/arm64/include/asm/spectre.h:17,                                                                                                                                                                       
>                  from linux/arch/arm64/include/asm/processor.h:47,
>                  from linux/include/linux/sched.h:13,
>                  from linux/include/linux/ratelimit.h:6,
>                  from linux/include/linux/dev_printk.h:16,
>                  from linux/include/linux/device.h:15,
>                  from linux/include/linux/node.h:18,
>                  from linux/include/linux/cpu.h:17,
>                  from linux/include/linux/stop_machine.h:5,
>                  from linux/kernel/trace/ftrace.c:17:
> linux/kernel/trace/ftrace.c: In function 'ftrace_filter_pid_sched_switch_probe':
> linux/arch/arm64/include/asm/percpu.h:295:42: warning: conversion from 'long unsigned int' to 'u8' {aka 'unsigned char'} changes value from '18446744073709551615' to '255' [-Woverflow]                                               
>   295 |         _pcp_wrap(__percpu_write_8, pcp, (unsigned long)(val))
>       |                                          ^~~~~~~~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:277:20: note: in definition of macro '_pcp_wrap'
>   277 |         op(&(pcp), __VA_ARGS__);                                        \
>       |                    ^~~~~~~~~~~
> linux/include/linux/percpu-defs.h:369:25: note: in expansion of macro 'this_cpu_write_1'
>   369 |                 case 1: stem##1(variable, __VA_ARGS__);break;           \
>       |                         ^~~~
> linux/include/linux/percpu-defs.h:500:41: note: in expansion of macro '__pcpu_size_call'
>   500 | #define this_cpu_write(pcp, val)        __pcpu_size_call(this_cpu_write_, pcp, val)
>       |                                         ^~~~~~~~~~~~~~~~
> linux/kernel/trace/ftrace.c:8628:17: note: in expansion of macro 'this_cpu_write'
>  8628 |                 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
>       |                 ^~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:297:43: warning: conversion from 'long unsigned int' to 'u16' {aka 'short unsigned int'} changes value from '18446744073709551615' to '65535' [-Woverflow]
>   297 |         _pcp_wrap(__percpu_write_16, pcp, (unsigned long)(val))
>       |                                           ^~~~~~~~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:277:20: note: in definition of macro '_pcp_wrap'
>   277 |         op(&(pcp), __VA_ARGS__);                                        \
>       |                    ^~~~~~~~~~~
> linux/include/linux/percpu-defs.h:370:25: note: in expansion of macro 'this_cpu_write_2'
>   370 |                 case 2: stem##2(variable, __VA_ARGS__);break;           \
>       |                         ^~~~
> linux/include/linux/percpu-defs.h:500:41: note: in expansion of macro '__pcpu_size_call'
>   500 | #define this_cpu_write(pcp, val)        __pcpu_size_call(this_cpu_write_, pcp, val)
>       |                                         ^~~~~~~~~~~~~~~~
> linux/kernel/trace/ftrace.c:8628:17: note: in expansion of macro 'this_cpu_write'
>  8628 |                 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
>       |                 ^~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:299:43: warning: conversion from 'long unsigned int' to 'u32' {aka 'unsigned int'} changes value from '18446744073709551615' to '4294967295' [-Woverflow]
>   299 |         _pcp_wrap(__percpu_write_32, pcp, (unsigned long)(val))
>       |                                           ^~~~~~~~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:277:20: note: in definition of macro '_pcp_wrap'
>   277 |         op(&(pcp), __VA_ARGS__);                                        \
>       |                    ^~~~~~~~~~~
> linux/include/linux/percpu-defs.h:371:25: note: in expansion of macro 'this_cpu_write_4'
>   371 |                 case 4: stem##4(variable, __VA_ARGS__);break;           \
>       |                         ^~~~
> linux/include/linux/percpu-defs.h:500:41: note: in expansion of macro '__pcpu_size_call'
>   500 | #define this_cpu_write(pcp, val)        __pcpu_size_call(this_cpu_write_, pcp, val)
>       |                                         ^~~~~~~~~~~~~~~~
> linux/kernel/trace/ftrace.c:8628:17: note: in expansion of macro 'this_cpu_write'
>  8628 |                 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
>       |                 ^~~~~~~~~~~~~~
> linux/kernel/trace/ftrace.c: In function 'ignore_task_cpu':
> linux/arch/arm64/include/asm/percpu.h:295:42: warning: conversion from 'long unsigned int' to 'u8' {aka 'unsigned char'} changes value from '18446744073709551615' to '255' [-Woverflow]
>   295 |         _pcp_wrap(__percpu_write_8, pcp, (unsigned long)(val))
>       |                                          ^~~~~~~~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:277:20: note: in definition of macro '_pcp_wrap'
>   277 |         op(&(pcp), __VA_ARGS__);                                        \
>       |                    ^~~~~~~~~~~
> linux/include/linux/percpu-defs.h:369:25: note: in expansion of macro 'this_cpu_write_1'
>   369 |                 case 1: stem##1(variable, __VA_ARGS__);break;           \
>       |                         ^~~~
> linux/include/linux/percpu-defs.h:500:41: note: in expansion of macro '__pcpu_size_call'
>   500 | #define this_cpu_write(pcp, val)        __pcpu_size_call(this_cpu_write_, pcp, val)
>       |                                         ^~~~~~~~~~~~~~~~
> linux/kernel/trace/ftrace.c:8898:17: note: in expansion of macro 'this_cpu_write'
>  8898 |                 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
>       |                 ^~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:297:43: warning: conversion from 'long unsigned int' to 'u16' {aka 'short unsigned int'} changes value from '18446744073709551615' to '65535' [-Woverflow]
>   297 |         _pcp_wrap(__percpu_write_16, pcp, (unsigned long)(val))
>       |                                           ^~~~~~~~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:277:20: note: in definition of macro '_pcp_wrap'
>   277 |         op(&(pcp), __VA_ARGS__);                                        \
>       |                    ^~~~~~~~~~~
> linux/include/linux/percpu-defs.h:370:25: note: in expansion of macro 'this_cpu_write_2'
>   370 |                 case 2: stem##2(variable, __VA_ARGS__);break;           \
>       |                         ^~~~
> linux/include/linux/percpu-defs.h:500:41: note: in expansion of macro '__pcpu_size_call'
>   500 | #define this_cpu_write(pcp, val)        __pcpu_size_call(this_cpu_write_, pcp, val)
>       |                                         ^~~~~~~~~~~~~~~~
> linux/kernel/trace/ftrace.c:8898:17: note: in expansion of macro 'this_cpu_write'
>  8898 |                 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
>       |                 ^~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:299:43: warning: conversion from 'long unsigned int' to 'u32' {aka 'unsigned int'} changes value from '18446744073709551615' to '4294967295' [-Woverflow]
>   299 |         _pcp_wrap(__percpu_write_32, pcp, (unsigned long)(val))
>       |                                           ^~~~~~~~~~~~~~~~~~~~
> linux/arch/arm64/include/asm/percpu.h:277:20: note: in definition of macro '_pcp_wrap'
>   277 |         op(&(pcp), __VA_ARGS__);                                        \
>       |                    ^~~~~~~~~~~
> linux/include/linux/percpu-defs.h:371:25: note: in expansion of macro 'this_cpu_write_4'
>   371 |                 case 4: stem##4(variable, __VA_ARGS__);break;           \
>       |                         ^~~~
> linux/include/linux/percpu-defs.h:500:41: note: in expansion of macro '__pcpu_size_call'
>   500 | #define this_cpu_write(pcp, val)        __pcpu_size_call(this_cpu_write_, pcp, val)
>       |                                         ^~~~~~~~~~~~~~~~
> linux/kernel/trace/ftrace.c:8898:17: note: in expansion of macro 'this_cpu_write'
>  8898 |                 this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
>       |                 ^~~~~~~~~~~~~~
> ---8<---
> 
>