[PATCH 2/2] malloc: Cleanup _int_memalign

[Wilco Dijkstra]

On 10/8/25 15:10, Adhemerval Zanella Netto wrote:

> It does not seems to happen on x86_64 or aarch64.

Indeed, I've reproduced these. It's due to the test running out of 
memory and then reverting to mmap even for small blocks. The newp - p 
offset calculation doesn't take into account that mmap_base_offset may 
be non-zero on x86. I've sent v2 that subtracts the mmap_base instead of p.

 >> +      p = mmap_set_chunk (mmap_base (p), mmap_size (p),
 >> +			  (char *)newp - (char *)p, mmap_is_hp (p));

Cheers,
Wilco