| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <aK3NbFkHn5jx7JJ2@antec>
Date: Tue, 26 Aug 2025 16:06:20 +0100
From: Stafford Horne <shorne@...il.com>
To: ChenMiao <chenmiao.ku@...il.com>
Cc: Linux Kernel <linux-kernel@...r.kernel.org>,
Linux OpenRISC <linux-openrisc@...r.kernel.org>,
Jonas Bonn <jonas@...thpole.se>,
Stefan Kristiansson <stefan.kristiansson@...nalahti.fi>,
Arnd Bergmann <arnd@...db.de>,
"Mike Rapoport (Microsoft)" <rppt@...nel.org>,
Andrew Morton <akpm@...ux-foundation.org>,
Luis Chamberlain <mcgrof@...nel.org>,
Geert Uytterhoeven <geert@...ux-m68k.org>,
Sahil Siddiq <sahilcdq0@...il.com>,
Johannes Berg <johannes@...solutions.net>,
Nicolas Schier <nicolas.schier@...ux.dev>,
Masahiro Yamada <masahiroy@...nel.org>,
Dave Hansen <dave.hansen@...ux.intel.com>
Subject: Re: [PATCH v3 1/2] openrisc: Add text patching API support
On Thu, Aug 14, 2025 at 03:27:01AM +0000, ChenMiao wrote:
> From: chenmiao <chenmiao.ku@...il.com>
>
> We need a text patching mechanism to ensure that in the subsequent
> implementation of jump_label, the code can be modified to the correct
> location. Therefore, FIX_TEXT_POKE0 has been added as a mapping area.
>
> Among these changes, we implement patch_map and support the
> patch_insn_write API for single instruction writing.
I think the description here could be improved a bit.
Something like:
Add text patching api's to use in subsequent jump_label implementation. We use
a new fixmap FIX_TEXT_POKE0 entry to temporarily override MMU mappings to allow
read only text pages to be written to.
A new function patch_insn_write is exposed to allow single instruction patching.
> Link: https://lore.kernel.org/openrisc/aJIC8o1WmVHol9RY@antec/T/#t
>
> Signed-off-by: chenmiao <chenmiao.ku@...il.com>
>
> ---
> Changes in V3:
> - Removed the unimplemented and unsupported is_exit_text, added
> comments for the set_fixmap modification explaining why __init was
> removed, and added new comments for patch_insn_write.
>
> Changes in V2:
> - We modify the patch_insn_write(void *addr, const void *insn) API to
> patch_insn_write(void *addr, u32 insn), derectly support a single u32
> instruction write to map memory.
> - Create a new file named insn-def.h to define the or1k insn macro
> size and more define in the future.
>
> Signed-off-by: chenmiao <chenmiao.ku@...il.com>
> ---
> arch/openrisc/include/asm/Kbuild | 1 -
> arch/openrisc/include/asm/fixmap.h | 1 +
> arch/openrisc/include/asm/insn-def.h | 12 ++++
> arch/openrisc/include/asm/text-patching.h | 13 ++++
> arch/openrisc/kernel/Makefile | 1 +
> arch/openrisc/kernel/patching.c | 78 +++++++++++++++++++++++
> arch/openrisc/mm/init.c | 10 ++-
> 7 files changed, 114 insertions(+), 2 deletions(-)
> create mode 100644 arch/openrisc/include/asm/insn-def.h
> create mode 100644 arch/openrisc/include/asm/text-patching.h
> create mode 100644 arch/openrisc/kernel/patching.c
>
> diff --git a/arch/openrisc/include/asm/Kbuild b/arch/openrisc/include/asm/Kbuild
> index 2b1a6b00cdac..cef49d60d74c 100644
> --- a/arch/openrisc/include/asm/Kbuild
> +++ b/arch/openrisc/include/asm/Kbuild
> @@ -9,4 +9,3 @@ generic-y += spinlock.h
> generic-y += qrwlock_types.h
> generic-y += qrwlock.h
> generic-y += user.h
> -generic-y += text-patching.h
> diff --git a/arch/openrisc/include/asm/fixmap.h b/arch/openrisc/include/asm/fixmap.h
> index aaa6a26a3e92..74000215064d 100644
> --- a/arch/openrisc/include/asm/fixmap.h
> +++ b/arch/openrisc/include/asm/fixmap.h
> @@ -28,6 +28,7 @@
>
> enum fixed_addresses {
> FIX_EARLYCON_MEM_BASE,
> + FIX_TEXT_POKE0,
> __end_of_fixed_addresses
> };
>
> diff --git a/arch/openrisc/include/asm/insn-def.h b/arch/openrisc/include/asm/insn-def.h
> new file mode 100644
> index 000000000000..dc8d16db1579
> --- /dev/null
> +++ b/arch/openrisc/include/asm/insn-def.h
> @@ -0,0 +1,12 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * Copyright (C) 2025 Chen Miao
> + */
> +
> +#ifndef __ASM_INSN_DEF_H
> +#define __ASM_INSN_DEF_H
> +
> +/* or1k instructions are always 32 bits. */
> +#define OPENRISC_INSN_SIZE 4
> +
> +#endif /* __ASM_INSN_DEF_H */
> diff --git a/arch/openrisc/include/asm/text-patching.h b/arch/openrisc/include/asm/text-patching.h
> new file mode 100644
> index 000000000000..bffe828288c3
> --- /dev/null
> +++ b/arch/openrisc/include/asm/text-patching.h
> @@ -0,0 +1,13 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * Copyright (C) 2025 Chen Miao
> + */
> +
> +#ifndef _ASM_PATCHING_H_
> +#define _ASM_PATCHING_H_
>
The ifdef here and above a incosistent. Can you try to use a consistent
convention to what is already used in openrisc?
Mostly we use: __ASM_OPENRISC_*_H
> +#include <linux/types.h>
> +
> +int patch_insn_write(void *addr, u32 insn);
> +
> +#endif /* _ASM_PATCHING_H_ */
> diff --git a/arch/openrisc/kernel/Makefile b/arch/openrisc/kernel/Makefile
> index 58e6a1b525b7..f0957ce16d6b 100644
> --- a/arch/openrisc/kernel/Makefile
> +++ b/arch/openrisc/kernel/Makefile
> @@ -13,5 +13,6 @@ obj-$(CONFIG_SMP) += smp.o sync-timer.o
> obj-$(CONFIG_STACKTRACE) += stacktrace.o
> obj-$(CONFIG_MODULES) += module.o
> obj-$(CONFIG_OF) += prom.o
> +obj-y += patching.o
>
> clean:
> diff --git a/arch/openrisc/kernel/patching.c b/arch/openrisc/kernel/patching.c
> new file mode 100644
> index 000000000000..73ae449c6c4e
> --- /dev/null
> +++ b/arch/openrisc/kernel/patching.c
> @@ -0,0 +1,78 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/* Copyright (C) 2020 SiFive
> + * Copyright (C) 2025 Chen Miao
> + */
> +
> +#include <linux/mm.h>
> +#include <linux/kernel.h>
> +#include <linux/spinlock.h>
> +#include <linux/uaccess.h>
> +
> +#include <asm/insn-def.h>
> +#include <asm/cacheflush.h>
> +#include <asm/page.h>
> +#include <asm/fixmap.h>
> +#include <asm/text-patching.h>
> +#include <asm/sections.h>
> +
> +static DEFINE_RAW_SPINLOCK(patch_lock);
> +
> +static __always_inline void *patch_map(void *addr, int fixmap)
> +{
> + uintptr_t uaddr = (uintptr_t) addr;
> + phys_addr_t phys;
> +
> + if (core_kernel_text(uaddr)) {
> + phys = __pa_symbol(addr);
> + } else {
> + struct page *page = vmalloc_to_page(addr);
> + BUG_ON(!page);
> + phys = page_to_phys(page) + offset_in_page(addr);
> + }
> +
> + return (void *)set_fixmap_offset(fixmap, phys);
> +}
> +
> +static void patch_unmap(int fixmap)
> +{
> + clear_fixmap(fixmap);
> +}
> +
> +static int __patch_insn_write(void *addr, u32 insn)
> +{
> + void *waddr = addr;
> + unsigned long flags = 0;
> + int ret;
> +
> + raw_spin_lock_irqsave(&patch_lock, flags);
> +
> + waddr = patch_map(addr, FIX_TEXT_POKE0);
> +
> + ret = copy_to_kernel_nofault(waddr, &insn, OPENRISC_INSN_SIZE);
> + local_icache_range_inv((unsigned long)waddr,
> + (unsigned long)waddr + OPENRISC_INSN_SIZE);
> +
> + patch_unmap(FIX_TEXT_POKE0);
> +
> + raw_spin_unlock_irqrestore(&patch_lock, flags);
> +
> + return ret;
> +}
> +
> +/* patch_insn_write - Write a single instruction to a specified memory location
> + * This API provides a single-instruction patching, primarily used for runtime
> + * code modification.
> + * By the way, the insn size must be 4 bytes.
> + */
The commit style is a bit off. In the kernel multi line comments usually start
with a lone '/*'. e.g.
/*
* patch_insn_write - Write ...
* ...
*/
> +int patch_insn_write(void *addr, u32 insn)
> +{
> + u32 *tp = addr;
> + int ret;
> +
> + if ((uintptr_t) tp & 0x3)
> + return -EINVAL;
> +
> + ret = __patch_insn_write(tp, insn);
> +
> + return ret;
> +}
> diff --git a/arch/openrisc/mm/init.c b/arch/openrisc/mm/init.c
> index e4904ca6f0a0..ac256c3d9c7a 100644
> --- a/arch/openrisc/mm/init.c
> +++ b/arch/openrisc/mm/init.c
> @@ -226,7 +226,15 @@ static int __init map_page(unsigned long va, phys_addr_t pa, pgprot_t prot)
> return 0;
> }
>
> -void __init __set_fixmap(enum fixed_addresses idx,
> +/* Removing __init is necessary. Before supporting FIX_TEXT_POKE0,
> + * __init here indicates that it is valid during the initialization phase
> + * and is used for FIX_EARLYCON_MEM_BASE. However, attempting to support
> + * FIX_TEXT_POKE0 would introduce a bug. FIX_TEXT_POKE0 is used after the
> + * initialization phase, so __init would cause the function to become invalid.
> + * At that point, using set_fixmap would lead to accessing dirty data,
> + * which is invalid.
> + */
The comment about __init does not belong in the source here. Perhaps your
comment could mention hwo the __set_fixmap function is used for both EARLYCON
and TEXT_POKE mappings. Also, the comment style need to be updated.
> +void __set_fixmap(enum fixed_addresses idx,
> phys_addr_t phys, pgprot_t prot)
> {
> unsigned long address = __fix_to_virt(idx);
> --
> 2.45.2
>
Powered by blists - more mailing lists