Re: [PATCH v4 0/6] GDB: aarch64-linux: Some header fixes

[Thiago Jung Bauermann <thiago.bauermann@linaro.org>]

Hello Simon,

Simon Marchi <simark@simark.ca> writes:

> On 2026-03-03 00:06, Thiago Jung Bauermann wrote:
>> v4 adds two new patches, reorganizing MTE and SME definitions.
>> 
>> With this series, I think the AArch64 definitions are correctly organized
>> between the headers in gdb/arch/ and gdb/nat/.
>> 
>> if the series is approved I will propose a version of the last patch for
>> the GDB 17 branch which doesn't depend on the previous patches, so that the
>> build failure on aarch64-linux-musl can be fixed with minimal changes.
>> 
>> I didn't include Luis' Approved-by to patches, since they have some changes
>> in them.
>> 
>> I also build-tested this version on aarch64-freebsd14.3 in addition to
>> aarch64-linux-gnu and aarch64-linux-musl.
>> 
>> Previous versions of the patches:
>> 
>> v1: https://inbox.sourceware.org/gdb-patches/20260211045853.1436881-1-thiago.bauermann@linaro.org/
>> v2: https://inbox.sourceware.org/gdb-patches/20260214045504.361392-1-thiago.bauermann@linaro.org/
>> v3: https://inbox.sourceware.org/gdb-patches/20260217060106.1906312-1-thiago.bauermann@linaro.org/
>
> I skimmed the code and read the commit messages, it looks good to me,
> but I will leave Luis give the approval.  Thanks for the cleanup, I
> think things are organized in a systematic and predictable way now.

Thanks! and thank you for the prompt review.

> Here are some things that copilot had to say, I checked and they seemed
> true:

They are true indeed.

>  - gdb/nat/aarch64-pauth-linux.h: HWCAP_PACA is defined unconditionally,
>    unlike every other HWCAP fallback in the series (HWCAP2_MTE,
>    HWCAP2_FPMR, HWCAP2_SME, HWCAP_GCS) which all use #ifndef guards. On
>    any modern aarch64 Linux system HWCAP_PACA is already defined by
>    <asm/hwcap.h>, so including this header after any header that pulls
>    it in will cause a redefinition error. Wrap it in #ifndef HWCAP_PACA
>    / #endif.

Ah, I missed that. Fixed locally.

>    (Simon: not sure it will necessarily be a error, I think it's possible to
>    redefine macros if they have exactly the same value, but still it
>    would be good to add the #ifndef.)

I tested and you are correct. There's a build error only if the
definitions differ.

In the process of testing this, I noticed that the #ifndef implies that
<asm/hwcap.h> is included, so I amended the nat/*.h files that define
HWCAP macros to also include it explicitly. And also <asm/ptrace.h> in
the case of one header which defines a couple of ptrace constants. I'll
detail it in the respective patches.

>  - FPMR: the commit message appears to be a copy-paste of the pauth one
>    that wasn't updated — it references AARCH64_HWCAP2_FPRM, HWCAP2_FPRM,
>    and gdb/arch/aarch64-pauth-linux.h / gdb/nat/aarch64-pauth-linux.h,
>    when all four should refer to the FPMR variants.

Argh. Fixed.

>    (Simon: notice the FPRM -> FPMR typos in there)

Argh again. Fixed again.

>  - GCS: the commit message says "create
>    gdb/nat/aarch64-gcs-linux-ptrace.h" but the file actually created is
>    gdb/nat/aarch64-gcs-linux.h.

Yes, I originally called the file ...-ptrace.h but then I changed my
mind and forgot to adjust the commit message. Fixed.

-- 
Thiago