[PATCH v2] arm64: hw_breakpoint: warn on invalid breakpoint length

[Will Deacon]

On Sat, Dec 13, 2025 at 12:23:35PM +0100, Osama Abdelkader wrote:
> Some tools (e.g. perf) incorrectly assume that breakpoints should be
> sizeof(long), but this is wrong for AArch64 where breakpoints must be
> 4 bytes. Add a warning when we silently fix up the parameter so tool
> developers can get notified.
> 
> This addresses the FIXME comment by adding diagnostic output rather
> than breaking existing tools by returning -EINVAL.
> 
> Signed-off-by: Osama Abdelkader <osama.abdelkader at gmail.com>
> ---
> v2: fix warning messageline splitting
> ---
>  arch/arm64/kernel/hw_breakpoint.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/arm64/kernel/hw_breakpoint.c b/arch/arm64/kernel/hw_breakpoint.c
> index ab76b36dce82..cce306145d78 100644
> --- a/arch/arm64/kernel/hw_breakpoint.c
> +++ b/arch/arm64/kernel/hw_breakpoint.c
> @@ -475,11 +475,13 @@ static int arch_build_bp_info(struct perf_event *bp,
>  				return -EINVAL;
>  		} else if (hw->ctrl.len != ARM_BREAKPOINT_LEN_4) {
>  			/*
> -			 * FIXME: Some tools (I'm looking at you perf) assume
> -			 *	  that breakpoints should be sizeof(long). This
> -			 *	  is nonsense. For now, we fix up the parameter
> -			 *	  but we should probably return -EINVAL instead.
> +			 * Some tools (e.g. perf) incorrectly assume that
> +			 * breakpoints should be sizeof(long). This is wrong
> +			 * for AArch64 where breakpoints must be 4 bytes.
> +			 * Warn the user and fix up the parameter.
>  			 */
> +			pr_warn_once("hw_breakpoint: invalid AArch64 breakpoint length %d, fixing to 4 bytes\n",
> +				     hw->ctrl.len);
>  			hw->ctrl.len = ARM_BREAKPOINT_LEN_4;

This is user triggerable so I don't see why printing a message is a good
idea. Just remove the FIXME tag if it bothers you -- we can't really
change the behaviour at this point.

Will