[PATCH] libgloss: arm: Add hardfault handler for M-profile

[Richard Earnshaw]

On 12/06/2026 16:53, Christophe Lyon wrote:
> After reaching a hardfault, the simulator enters an infinite loop
> unless we map the handler to 'abort'.
> 
> The patch adds this minimal setup for M-profile in
> libgloss/arm/cpu-init/rdimon-aem.S and introduces the new
> libgloss/arm/elf-mprofile-mps2.specs to add the mapping to the right
> memory address.
> 
> 2026-05-29  Andre Vieira <andre.simoesdiasvieira@arm.com>
> 	Christophe Lyon  <christophe.lyon@arm.com>
> 
> 	libgloss/
> 	* Makefile.in: Add entry for arm/mprofile-mps2.specs.
> 	* arm/Makefile.inc: Add entry for mprofile-mps2.specs.
> 	* arm/cpu-init/rdimon-aem.S: Update comment. Add hardfault
> 	handler.
> 	* arm/elf-mprofile-mps2.specs: New file.
> ---
>   libgloss/Makefile.in                 |  1 +
>   libgloss/arm/Makefile.inc            |  3 ++-
>   libgloss/arm/cpu-init/rdimon-aem.S   | 11 +++++++++--
>   libgloss/arm/elf-mprofile-mps2.specs | 26 ++++++++++++++++++++++++++
>   4 files changed, 38 insertions(+), 3 deletions(-)
>   create mode 100644 libgloss/arm/elf-mprofile-mps2.specs
> 
> diff --git a/libgloss/Makefile.in b/libgloss/Makefile.in
> index 5146a7880..c03ea4e6b 100644
> --- a/libgloss/Makefile.in
> +++ b/libgloss/Makefile.in
> @@ -2255,6 +2255,7 @@ TEXINFO_TEX = ../texinfo/texinfo.tex
>   @CONFIG_ARM_TRUE@	arm/aprofile-validation-v2m.specs \
>   @CONFIG_ARM_TRUE@	arm/aprofile-ve-v2m.specs \
>   @CONFIG_ARM_TRUE@	arm/aprofile-ve.specs arm/pid.specs \
> +@CONFIG_ARM_TRUE@       arm/mprofile-mps2.specs \
>   @CONFIG_ARM_TRUE@	arm/iq80310.specs
>   @CONFIG_ARM_TRUE@arm_libgloss_linux_a_SOURCES = \
>   @CONFIG_ARM_TRUE@	arm/linux-syscalls0.c \
> diff --git a/libgloss/arm/Makefile.inc b/libgloss/arm/Makefile.inc
> index 41fce3c99..5c740edcb 100644
> --- a/libgloss/arm/Makefile.inc
> +++ b/libgloss/arm/Makefile.inc
> @@ -47,7 +47,8 @@ multilibtool_LIBRARIES += %D%/librdpmon.a
>   	%D%/aprofile-validation.specs \
>   	%D%/aprofile-validation-v2m.specs \
>   	%D%/aprofile-ve-v2m.specs \
> -	%D%/aprofile-ve.specs
> +	%D%/aprofile-ve.specs \
> +	%D%/mprofile-mps2.specs
>   multilibtool_DATA += \
>   	%D%/rdimon-crt0.o \
>   	%D%/rdimon-crt0-v2m.o
> diff --git a/libgloss/arm/cpu-init/rdimon-aem.S b/libgloss/arm/cpu-init/rdimon-aem.S
> index 2aacbeba9..c960924d4 100644
> --- a/libgloss/arm/cpu-init/rdimon-aem.S
> +++ b/libgloss/arm/cpu-init/rdimon-aem.S
It looks to me as though there's no shared code in this file compared to 
the a-profile variant, which makes me think that perhaps they should 
just be separate files, rather than munged together when there's no 
commonality beyond 'this is startup code'.

Personally, I think it would be better to rename the existing file as 
something like rdimon-aem-a.S and then add a new one (say, 
rdimon-aem-m.S) for the m-profile code.

> @@ -25,7 +25,9 @@
>   
>   /* This file gives a basic initialisation of a Cortex-A series core.  It is
>      the bare minimum required to get Cortex-A core running with a semihosting
> -   interface.
> +   interface.  For M-profile we only define a hardfault handler, required to
> +   prevent the simulator from entering into an infinite loop.  The comments
> +   below apply to A-profile only.

If I've missed something and this code does belong in this file, then 
this whole comment needs properly restructuring.  There should be a 
single top-level comment that just says something like "Provide the 
basic CPU initialization for Armv7 and later CPUs.  We currently support 
A-profile and M-profile.  Then the profile-specific details should be 
moved inside the relevant ifdef sections.  It would be good to get rid 
of the reference to 'Cortex-A' here as well (replace it with A-profile) 
as there's nothing here that is specific to Arm's Cortex-* products: 
this is defined by the architecture.

Finally, if you're keeping this in one place, the copyright dates at the 
top of the file need updating.

>   
>      It sets up a basic 1:1 phsyical address to virtual address mapping;
>      turns the MMU on; enables branch prediction; activates any integrated
> @@ -539,4 +541,9 @@ is_a15_a7:
>   page_tables:
>        PT7(0x1c0e)
>   
> -#endif //#if defined(__ARM_ARCH_7A__) || __ARM_ARCH_PROFILE == 'A'
> +#elif  (__ARM_ARCH_PROFILE == 'M')
> +    /* On M-profile, call abort in case of hardfault: this causes the simulator to
> +       abort execution instead of going into an infinite loop.  */
> +    .section .hardfault_handler_addr, "a"
> +    .word abort
> +#endif //#if defined(__ARM_ARCH_7A__) || (__ARM_ARCH_PROFILE == 'A')
> diff --git a/libgloss/arm/elf-mprofile-mps2.specs b/libgloss/arm/elf-mprofile-mps2.specs
> new file mode 100644
> index 000000000..edaf71917
> --- /dev/null
> +++ b/libgloss/arm/elf-mprofile-mps2.specs
> @@ -0,0 +1,26 @@
> +# rdimon.specs
> +#
> +# Spec file for AArch32 M profile baremetal newlib with version 2 of
> +# the AngelAPI semi-hosting using the SVC trap instruction.
> +#
> +# This version supports extensibility through an extension mechanism.
> +
> +%rename link_gcc_c_sequence                rdimon_link_gcc_c_sequence
> +
> +%rename link old_link
> +
> +*link:
> +--section-start=.hardfault_handler_addr=0xc %(old_link)
> +
> +*rdimon_libc:
> +%{!specs=nano.specs:-lc} %{specs=nano.specs:-lc_nano}
> +
> +*rdimon_libgloss:
> +%{!specs=nano.specs:-lrdimon} %{specs=nano.specs:-lrdimon_nano}
> +
> +*link_gcc_c_sequence:
> +%(rdimon_link_gcc_c_sequence) --start-group %G %(rdimon_libc) %(rdimon_libgloss) --end-group
> +
> +*startfile:
> +crti%O%s crtbegin%O%s cpu-init/rdimon-aem.o%s %{!pg:rdimon-crt0%O%s} %{pg:rdimon-crt0%O%s}
> +