Re: [PATCH RFC v2 09/11] stackdepot: add architecture hooks for compact frame storage

[Marco Elver]

On Tue, Sep 08, 2026 at 02:13PM +0100, Caleb Kan wrote:
> From: Caleb Kan <ckan@xxxxxxxxxxxxxx>
> 
> Path-compressed trie nodes can reduce their frame storage further when an
> architecture can represent kernel text and module addresses in 32 bits.
> Add architecture hooks that compress a frame only when decompression
> reproduces the original address exactly.
> 
> Store arm64 frames as signed 32-bit offsets from _text. This covers the
> 2 GB module relocation window without depending on a 4 GB high-bit
> boundary. On x86-64, store the low 32 bits when the upper 32 bits are all
> set. Keep other frames full-width, and provide a generic implementation
> that always rejects compression.
> 
> Make the generic header available to architectures without a specialized
> implementation and wire it explicitly for UML. Add KUnit coverage for raw
> fallback, arm64 boundary round trips, and native x86-64 prefix compression.
> 
> These hooks do not change stack depot behavior until a later patch adds
> trie storage.
> 
> Signed-off-by: Caleb Kan <ckan@xxxxxxxxxxxxxx>
> ---
>  arch/arm64/include/asm/stackdepot.h | 42 ++++++++++++++++++
>  arch/um/include/asm/Kbuild          |  1 +
>  arch/x86/include/asm/stackdepot.h   | 37 ++++++++++++++++
>  include/asm-generic/Kbuild          |  1 +
>  include/asm-generic/stackdepot.h    | 19 ++++++++
>  lib/tests/stackdepot_kunit.c        | 86 +++++++++++++++++++++++++++++++++++++
>  6 files changed, 186 insertions(+)
> 
> diff --git a/arch/arm64/include/asm/stackdepot.h b/arch/arm64/include/asm/stackdepot.h
> new file mode 100644
> index 000000000000..df8959d59336
> --- /dev/null
> +++ b/arch/arm64/include/asm/stackdepot.h
> @@ -0,0 +1,42 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef __ASM_STACKDEPOT_H
> +#define __ASM_STACKDEPOT_H
> +
> +#include <linux/types.h>
> +#include <asm/sections.h>
> +
> +/*
> + * Modules are allocated inside a 2 GB relocation window containing the
> + * kernel image. Store a signed 32-bit offset from _text so compression is
> + * independent of 4 GB high-bit boundaries crossed by that window.
> + */
> +static inline unsigned long arch_stack_depot_frame_from_payload(u32 payload)
> +{
> +	long offset;
> +
> +	offset = (s32)payload;
> +	if (offset < 0)
> +		return (unsigned long)_text - (unsigned long)(-offset);
> +	return (unsigned long)_text + (unsigned long)offset;
> +}

'(unsigned long)_text + (s64)(s32)payload' avoids the branch and the
potential UB if 'offset == S32_MIN' on a 32-bit 'long' (though arm64 is
LP64, sign-extending directly is simpler).

More importantly: why not use these 32-bit frame compression hooks
directly in the existing hash-backed 'struct stack_record'? That gives
most of the memory savings across *all* stackdepot users with almost no
added code.