[PATCH v2] bfd: Grow in-memory BFD buffers geometrically
Jan Beulich
jbeulich@suse.com
Fri Nov 13 09:47:45 GMT 2026
On 06.11.2026 21:33, Oleg Tolmatcev wrote:
> memory_bwrite and memory_bseek grew the buffer in 128-byte steps, so
> building a large in-memory BFD out of many small writes reallocated and
> copied the whole buffer repeatedly. This made writing PDB streams
> quadratic: linking a 235 MB executable with --pdb spent more than half
> of its time in realloc.
>
> Add a separate capacity field and grow the buffer by one eighth of its
> capacity, with a minimum increase of 128 bytes. If the over-allocation
> fails, retry with the exact required size.
>
> bfd/
> * libbfd-in.h (bfd_in_memory): Add capacity.
> * libbfd.h: Regenerate.
> * bfdio.c (memory_grow): New function.
> (memory_bwrite, memory_bseek): Use it.
> * coff-alpha.c (alpha_ecoff_get_elt_at_filepos): Initialize capacity.
> * coff-bfd.c (import_make_member): Likewise.
> * elfcode.h (NAME(_bfd_elf, bfd_from_remote_memory)): Likewise.
> * opncls.c (bfd_make_writable): Likewise.
> * peicode.h (pe_ILF_build_a_bfd): Likewise.
> * xcofflink.c (bfd_xcoff_link_generate_rtinit): Likewise.
Trying to apply this to run through a full round of testing, I find that ...
> --- a/bfd/coff-bfd.c
> +++ b/bfd/coff-bfd.c
> @@ -241,6 +241,7 @@ import_make_member (bfd *parent, const char *name,
>
> bim->buffer = data;
> bim->size = size;
> + bim->capacity = size;
>
> abfd->format = bfd_unknown;
> abfd->tdata.any = NULL;
... the function being patched here doesn't exist at all, and ...
> --- a/bfd/peicode.h
> +++ b/bfd/peicode.h
> @@ -881,6 +881,7 @@ pe_ILF_build_a_bfd (bfd * abfd,
> ptr = (bfd_byte *) bfd_zmalloc (raw_space + ILF_DATA_SIZE);
> vars.bim->buffer = ptr;
> vars.bim->size = raw_space + ILF_DATA_SIZE;
> + vars.bim->capacity = raw_space + ILF_DATA_SIZE;
> if (ptr == NULL)
> goto error_return;
... this chunk of code also looks different (no raw_space anywhere in
sight).
Please can you make sure that you send patches which actually apply?
This time I'll make the necessary adjustments, but this isn't a promise
that I would continue doing so.
Jan
More information about the Binutils
mailing list