Re: [PATCH v2 0/8] Improve ld.lld-21+ compatibility when building i386-pc target

[Nicholas Vinson <nvinson234@gmail.com>]

On 2/10/26 10:47, Daniel Kiper wrote:
> On Mon, Feb 09, 2026 at 07:43:28PM -0500, Nicholas Vinson wrote:
>> Starting with ld.llvm-21, any attempt create a non-relocatable binary and set
>> one more secton addresses below 0x400000 results in a linker error. Furthermore,
>> the differences between ld.bfd and ld.lld made finding a proper set of
>> command-line flags tht worked with both linkers and bypass the image base
>> address restriction difficult. Therefore, the approach of using a custom linker
>> script was adopted to solve the issue.
>>
>> This approach was tested using:
>>
>> ../configure CC=clang CXX=clang++ LDFLAGS="-fuse-ld=lld" TARGET_LDFLAGS="-fuse-ld=lld" --with-platform=pc
>> ../configure CC=clang CXX=clang++ --with-platform=pc (both with ld.lld as the default and ld.bfd as the default)
>> ../configure CC=gcc CXX=g++ --with-platform=pc
>>
>> and a VM was used for testing. To build the disk images the VM was booted with,
>> the following scripts were used:
> 
> [...]
> 
>> In all cases, the VM successfully booted to the standard GRUB prompt.
>>
>> Nicholas Vinson (8):
>>    i386/pc/int.h: conditionally apply regparm attr.
>>    grub-core: Update kernel image generation
>>    i386-cygwin-img-ld.sc -> i386-cygwin-img.lds
>>    Revert "configure: Print a more helpful error if autoconf-archive is
>>      not installed"
>>    Revert "configure: Check linker for --image-base support"
>>    Revert "INSTALL: Add note that the GNU Autoconf Archive may be needed"
>>    configure: drop -Ttext checks for i386-pc
> 
> For all these patches Reviewed-by: Daniel Kiper <daniel.kiper@oracle.com>...
> 
>>    C23 fixes: fix strchr() and strrchr() handling
> 
> It seems to me that this patch does more than commit message says.
> I think it has to be split into more parts or commit message has to be
> improved.

I admit the message is a bit terse, but I kept the changes to a minimum. 
I did, however, forget to mention strstr().

Put simply, with C23, if the first argument to strstr(), strchr(), or 
strrchar() is a const char *, the result is a const char *. C23 also 
made this change to about 9 or so other functions.

As a result, most of the changes were just changing a variable from 
'char *' to 'const char *'.

The changes for ofpath.c are a bit more involved because the original 
code would modify the string 'ed' pointed to. With C23, ed needs to be a 
'const char *' because sysfs_path is a 'const char *'. This also means 
that the line "*q = '\0'" is no longer valid because you cannot safely 
modify a const char string.

I saw two options to fix this, the first was to modify the line to be q 
= strndup(...);, add the requisite error checking, and then make sure to 
free q just before every subsequent return. The second option, is what I 
did, find the difference between q and ed, save it off, and then replace 
the strlen(ed) calls with that difference.

I believe the second approach is the better of the two. It did have the 
ancillary effect of updating a few snprintf() calls and the path_size 
calculation; however, these changes are required with this approach.

If you'd prefer, I can re-submit patch 8 with these details in the 
commit message. Alternatively, if you would prefer the strndup() 
approach, I will respin this patch series without patch 8, and then 
submit a re-worked C23 patch as a separate request.

Thanks,
Nicholas Vinson

> 
> Daniel


_______________________________________________
Grub-devel mailing list
Grub-devel@gnu.org
https://lists.gnu.org/mailman/listinfo/grub-devel