[PATCH v4 1/1] malloc: aarch64: Add ifuncs for malloc functions

[Adhemerval Zanella Netto]

On 04/06/26 08:21, Yury Khrustalev wrote:
> On Wed, Jun 03, 2026 at 10:46:16AM -0300, Adhemerval Zanella Netto wrote:
>>
>> On 01/06/26 07:30, Yury Khrustalev wrote:
>>> Introduce ifuncs and resolvers for functions pertinent to the
>>> malloc interface on the AArch64 target: malloc, calloc, free,
>>> realloc, memalign, valloc, pvalloc, posix_memalign, aligned_alloc,
>>> free_sized, free_aligned_sized, malloc_usable_size.
>>>
>>> A target can define the USE_MULTIARCH_MALLOC macro. In this case
>>> it must provide alternative aliases for the malloc functions that
>>> point to the ifuncs.
>>>
>>> This implementation respects the --disable-multi-arch configure
>>> flag. If multi-arch support is disabled, the generic aliases
>>> are used on aarch64.
>>>
>>> This patch contains aarch64-specific resolvers. At this moment they
>>> return core implementations but in the future they can be changed
>>> to support for features, e.g. to handle memory tagging.
>>
>> The patch looks ok, some comments below. I think the only missing piece 
>> is to check if external malloc-intercepting tools like valgrind, ASan/TSan, 
>> heap profilers, and gdb's inferior call malloc(...) handle the ifunc change
>> (as we discussed on the weekly call).
> 
> I've checked it with gdb and lldb as well as with valgrind memcheck and
> massif tools, address, hwadderess, thread and leak sanitizers and also
> using the heaptrack heap profiler. The output before and after looks the
> same except for 2 things I describe below. The heap usage, memory leaks,
> and out of buffer accesses are reported in the same way before and after
> the change.
> 
> One difference in the valgrind output is that it detects that the malloc
> functions are ifuncs now and you see
> 
>   REDIR: 0x4958810 (libc.so.6:aligned_alloc) redirected to 0x48507cc (_vgnU_ifunc_wrapper)
>   REDIR: 0x4956c84 (libc.so.6:__aligned_alloc) redirected to 0x488d708 (aligned_alloc)
> 
> instead of
> 
>   REDIR: 0x4956c84 (libc.so.6:aligned_alloc) redirected to 0x488d708 (aligned_alloc)
> 
> This is expected I guess, and it doesn't affect valgrind's operation.
> 
> The second thing in the debuggers and in thread stacks in sanitizers the
> malloc functions appear as "__lib_malloc" instead of "malloc" (for example)
> which is also expected as a tool would show the symbol that the corresponding
> ifunc resolves to. When we have "__libc_malloc" and "__libc_malloc_mte", it
> would show the appropriate symbol which is how it should be, I think.
> 
> So, I think that this patch should not introduce any issues for sanitizers,
> profiling and debugging tools.

Sounds good, thanks for checking this.

> 
>>
>> ...
>>
>>> +weak_alias (__free_sized, free_sized)
>>> +weak_alias (__free_aligned_sized, free_aligned_sized)
>>> +#endif /* !USE_MULTIARCH_MALLOC */
>>>  
>>>  weak_alias (__malloc_info, malloc_info)
>>>  weak_alias (__libc_mallinfo, mallinfo)
>>
>> We discussed this privately; the cfree bypasses the redirect: compat_symbol (..., cfree, GLIBC_2_0)
>> is kept outside the USE_MULTIARCH_MALLOC guard.
>>
>> It does not matter for this patch, because aarch64 does not support multiple, 
>> potentially incompatible malloc implementations.
>>
>> But it might become an issue (though highly unlikely, since cfree will be present 
>> only in older binaries). Maybe add a route it through __libc_free_redirect too.
> 
> Yes, I will fix it in the next version of the patch.
> 
>>
>> ...
>>
>>> diff --git a/sysdeps/aarch64/multiarch/malloc-ifuncs.h b/sysdeps/aarch64/multiarch/malloc-ifuncs.h
>>> new file mode 100644
>>> index 0000000000..98e913d49c
>>> --- /dev/null
>>> +++ b/sysdeps/aarch64/multiarch/malloc-ifuncs.h
>>> +
>>
>> ...
>>
>>> +#define USE_MULTIARCH_MALLOC 1
>>> +
>>> +#include <stddef.h>
>>> +#include <sys/cdefs.h>
>>> +
>>> +/* Core implementations of malloc functions.  An ifunc resolver must
>>> +   use this implementations as a fallback option.  Other implementations
>>
>> s/this/these
>>
>>> +   may internally call these core function.  */
>>
>> s/function/functions
>>
> 
> OK
> 
>> ...
>>
>>> diff --git a/sysdeps/generic/malloc-ifuncs.h b/sysdeps/generic/malloc-ifuncs.h
>>> new file mode 100644
>>> index 0000000000..aedc56ae33
>>> --- /dev/null
>>> +++ b/sysdeps/generic/malloc-ifuncs.h
>>
>> ...
>>
>> This header is already included iff USE_MULTIARCH is defined:
>>
>>  27 #if defined(USE_MULTIARCH)
>>  28 # include <malloc-ifuncs.h>
>>  29 #endif
>>  30
>>
> 
> OK
> 
>>> +
>>> +#endif /* GENERIC_MALLOC_IFUNCS_H */
>>
> 
> Thanks,
> Yury
>