Re: [PATCH] Move nested functions in link_map.c to file scope.

[Chih-Hung Hsieh <chh at google dot com>]

Mark,

Thanks for pushing several of my patches during the holidays.
I have merged all of them and latest elfutils changes into Android open
source project.
We are getting closer to compile all needed elfutils files with clang/llvm
for Android.

I will continue to fix the remain elfutils issues after fixing a few other
urgent unrelated Android problems. Thanks.


On Sun, Jan 3, 2016 at 1:25 PM, Mark Wielaard <mjw@redhat.com> wrote:

> On Tue, 2015-11-17 at 14:45 -0800, Chih-Hung Hsieh wrote:
> > * In libdwfl/link_map.c, nested functions check64, check32,
>
> For check64 and check32 the define plus static function extra args call
> trick seems a good fit. I did push this part of the patch.
>
> >   release_buffer, read_addrs, and consider_phdr are moved
> >   to file scope to compile with clang.
>
> But for the rest I feel it makes the code unnecessary messy. So I didn't
> apply those. See below for some suggestions how to maybe get something
> that doesn't create new functions with lots and lots of arguments (which
> I find doesn't improve readability).
>
> > +static inline int
> > +do_release_buffer (int result, Dwfl *dwfl,
> > +                   void **pbuffer, size_t *pbuffer_available,
> > +                   void *memory_callback_arg,
> > +                   Dwfl_Memory_Callback *memory_callback)
> > +{
> > +  if (*pbuffer != NULL)
> > +    (void) (*memory_callback) (dwfl, -1, pbuffer, pbuffer_available, 0,
> 0,
> > +                            memory_callback_arg);
> > +  return result;
> > +}
> > +
> > +#define release_buffer(result) \
> > +  do_release_buffer (result, dwfl, &buffer, &buffer_available, \
> > +                     memory_callback_arg, memory_callback)
>
> This is just 3 lines, and it is either used as "return release_buffer
> (result)" to cleanup and return a result or as release_buffer (0) just
> to cleanup the buffer (if not NULL). It seems better to make the define
> just the function body. Assuming your compiler does know about statement
> expressions https://gcc.gnu.org/onlinedocs/gcc/Statement-Exprs.html
> Otherwise maybe just split it in two. Don't use an argument, just use it
> as "cleanup_buffer()" plus a separate return line.
>
> > +static inline bool
> > +do_read_addrs (GElf_Addr vaddr, size_t n, GElf_Addr *addrs,
> > +               uint_fast8_t elfdata, GElf_Addr *pread_vaddr,
> > +               int elfclass, Dwfl *dwfl,
> > +               void **pbuffer, size_t *pbuffer_available,
> > +               void *memory_callback_arg,
> > +               Dwfl_Memory_Callback *memory_callback)
> > +{
> > +  size_t nb = n * addrsize (elfclass); /* Address words -> bytes to
> read.  */
> > +
> > +  /* Read a new buffer if the old one doesn't cover these words.  */
> > +  if (*pbuffer == NULL
> > +      || vaddr < *pread_vaddr
> > +      || vaddr - *pread_vaddr + nb > *pbuffer_available)
> > +    {
> > +      do_release_buffer (0, dwfl, pbuffer, pbuffer_available,
> > +                         memory_callback_arg, memory_callback);
> > +
> > +      *pread_vaddr = vaddr;
> > +      int segndx = INTUSE(dwfl_addrsegment) (dwfl, vaddr, NULL);
> > +      if (unlikely (segndx < 0)
> > +       || unlikely (! (*memory_callback) (dwfl, segndx,
> > +                                          pbuffer, pbuffer_available,
> > +                                          vaddr, nb,
> memory_callback_arg)))
> > +     return true;
> > +    }
> > +
> > +  Elf32_Addr (*a32)[n] = vaddr - *pread_vaddr + *pbuffer;
> > +  Elf64_Addr (*a64)[n] = (void *) a32;
> > +
> > +  if (elfclass == ELFCLASS32)
> > +    {
> > +      if (elfdata == ELFDATA2MSB)
> > +     for (size_t i = 0; i < n; ++i)
> > +       addrs[i] = BE32 (read_4ubyte_unaligned_noncvt (&(*a32)[i]));
> > +      else
> > +     for (size_t i = 0; i < n; ++i)
> > +       addrs[i] = LE32 (read_4ubyte_unaligned_noncvt (&(*a32)[i]));
> > +    }
> > +  else
> > +    {
> > +      if (elfdata == ELFDATA2MSB)
> > +     for (size_t i = 0; i < n; ++i)
> > +       addrs[i] = BE64 (read_8ubyte_unaligned_noncvt (&(*a64)[i]));
> > +      else
> > +     for (size_t i = 0; i < n; ++i)
> > +       addrs[i] = LE64 (read_8ubyte_unaligned_noncvt (&(*a64)[i]));
> > +    }
> > +
> > +  return false;
> > +}
> > +
> > +#define read_addrs(vaddr, n) \
> > +  do_read_addrs (vaddr, n, addrs, elfdata, &read_vaddr, elfclass, dwfl,
> \
> > +                 &buffer, &buffer_available, \
> > +                 memory_callback_arg, memory_callback)
>
> This very generic, but only used twice. Once to read 1 address and once
> to read 4 addresses. Can't we make this a little simpler? Maybe just let
> it handle reading one address. It might be a bit more hairy than needs
> to because of the extra arguments needed to pass to the buffer cleanup.
>
> > +static inline bool
> > +do_consider_phdr (GElf_Word type, GElf_Addr vaddr, GElf_Xword filesz,
> > +                  Dwfl *dwfl, GElf_Addr phdr, GElf_Addr *pdyn_bias,
> > +                  GElf_Addr *pdyn_vaddr, GElf_Xword *pdyn_filesz)
> > +{
> > +  switch (type)
> > +    {
> > +    case PT_PHDR:
> > +      if (*pdyn_bias == (GElf_Addr) - 1
> > +       /* Do a sanity check on the putative address.  */
> > +       && ((vaddr & (dwfl->segment_align - 1))
> > +           == (phdr & (dwfl->segment_align - 1))))
> > +     {
> > +       *pdyn_bias = phdr - vaddr;
> > +       return *pdyn_vaddr != 0;
> > +     }
> > +      break;
> > +
> > +    case PT_DYNAMIC:
> > +      *pdyn_vaddr = vaddr;
> > +      *pdyn_filesz = filesz;
> > +      return *pdyn_bias != (GElf_Addr) - 1;
> > +    }
> > +
> > +  return false;
> > +}
> > +
> > +#define consider_phdr(type, vaddr, filesz) \
> > +  do_consider_phdr (type, vaddr, filesz, \
> > +                    dwfl, phdr, &dyn_bias, &dyn_vaddr, &dyn_filesz)
>
> This is just used twice in the function. Might be simpler to just inline
> it twice and simplifying the switch by an if type == PT_... check
> instead of inventing a new function with 8 arguments.
>
> Thanks,
>
> Mark
>
>

Mark,

Thanks for pushing several of my patches during the holidays.

I have merged all of them and latest elfutils changes into Android open source project.

We are getting closer to compile all needed elfutils files with clang/llvm for Android.

I will continue to fix the remain elfutils issues after fixing a few other urgent unrelated Android problems. Thanks.

On Sun, Jan 3, 2016 at 1:25 PM, Mark Wielaard <mjw@redhat.com> wrote:

On Tue, 2015-11-17 at 14:45 -0800, Chih-Hung Hsieh wrote:
> * In libdwfl/link_map.c, nested functions check64, check32,

For check64 and check32 the define plus static function extra args call
trick seems a good fit. I did push this part of the patch.

>   release_buffer, read_addrs, and consider_phdr are moved
>   to file scope to compile with clang.

But for the rest I feel it makes the code unnecessary messy. So I didn't
apply those. See below for some suggestions how to maybe get something
that doesn't create new functions with lots and lots of arguments (which
I find doesn't improve readability).

> +static inline int
> +do_release_buffer (int result, Dwfl *dwfl,
> +                   void **pbuffer, size_t *pbuffer_available,
> +                   void *memory_callback_arg,
> +                   Dwfl_Memory_Callback *memory_callback)
> +{
> +  if (*pbuffer != NULL)
> +    (void) (*memory_callback) (dwfl, -1, pbuffer, pbuffer_available, 0, 0,
> +                            memory_callback_arg);
> +  return result;
> +}
> +
> +#define release_buffer(result) \
> +  do_release_buffer (result, dwfl, &buffer, &buffer_available, \
> +                     memory_callback_arg, memory_callback)

This is just 3 lines, and it is either used as "return release_buffer
(result)" to cleanup and return a result or as release_buffer (0) just
to cleanup the buffer (if not NULL). It seems better to make the define
just the function body. Assuming your compiler does know about statement
expressions https://gcc.gnu.org/onlinedocs/gcc/Statement-Exprs.html
Otherwise maybe just split it in two. Don't use an argument, just use it
as "cleanup_buffer()" plus a separate return line.

> +static inline bool
> +do_read_addrs (GElf_Addr vaddr, size_t n, GElf_Addr *addrs,
> +               uint_fast8_t elfdata, GElf_Addr *pread_vaddr,
> +               int elfclass, Dwfl *dwfl,
> +               void **pbuffer, size_t *pbuffer_available,
> +               void *memory_callback_arg,
> +               Dwfl_Memory_Callback *memory_callback)
> +{
> +  size_t nb = n * addrsize (elfclass); /* Address words -> bytes to read.  */
> +
> +  /* Read a new buffer if the old one doesn't cover these words.  */
> +  if (*pbuffer == NULL
> +      || vaddr < *pread_vaddr
> +      || vaddr - *pread_vaddr + nb > *pbuffer_available)
> +    {
> +      do_release_buffer (0, dwfl, pbuffer, pbuffer_available,
> +                         memory_callback_arg, memory_callback);
> +
> +      *pread_vaddr = vaddr;
> +      int segndx = INTUSE(dwfl_addrsegment) (dwfl, vaddr, NULL);
> +      if (unlikely (segndx < 0)
> +       || unlikely (! (*memory_callback) (dwfl, segndx,
> +                                          pbuffer, pbuffer_available,
> +                                          vaddr, nb, memory_callback_arg)))
> +     return true;
> +    }
> +
> +  Elf32_Addr (*a32)[n] = vaddr - *pread_vaddr + *pbuffer;
> +  Elf64_Addr (*a64)[n] = (void *) a32;
> +
> +  if (elfclass == ELFCLASS32)
> +    {
> +      if (elfdata == ELFDATA2MSB)
> +     for (size_t i = 0; i < n; ++i)
> +       addrs[i] = BE32 (read_4ubyte_unaligned_noncvt (&(*a32)[i]));
> +      else
> +     for (size_t i = 0; i < n; ++i)
> +       addrs[i] = LE32 (read_4ubyte_unaligned_noncvt (&(*a32)[i]));
> +    }
> +  else
> +    {
> +      if (elfdata == ELFDATA2MSB)
> +     for (size_t i = 0; i < n; ++i)
> +       addrs[i] = BE64 (read_8ubyte_unaligned_noncvt (&(*a64)[i]));
> +      else
> +     for (size_t i = 0; i < n; ++i)
> +       addrs[i] = LE64 (read_8ubyte_unaligned_noncvt (&(*a64)[i]));
> +    }
> +
> +  return false;
> +}
> +
> +#define read_addrs(vaddr, n) \
> +  do_read_addrs (vaddr, n, addrs, elfdata, &read_vaddr, elfclass, dwfl, \
> +                 &buffer, &buffer_available, \
> +                 memory_callback_arg, memory_callback)

This very generic, but only used twice. Once to read 1 address and once
to read 4 addresses. Can't we make this a little simpler? Maybe just let
it handle reading one address. It might be a bit more hairy than needs
to because of the extra arguments needed to pass to the buffer cleanup.

> +static inline bool
> +do_consider_phdr (GElf_Word type, GElf_Addr vaddr, GElf_Xword filesz,
> +                  Dwfl *dwfl, GElf_Addr phdr, GElf_Addr *pdyn_bias,
> +                  GElf_Addr *pdyn_vaddr, GElf_Xword *pdyn_filesz)
> +{
> +  switch (type)
> +    {
> +    case PT_PHDR:
> +      if (*pdyn_bias == (GElf_Addr) - 1
> +       /* Do a sanity check on the putative address.  */
> +       && ((vaddr & (dwfl->segment_align - 1))
> +           == (phdr & (dwfl->segment_align - 1))))
> +     {
> +       *pdyn_bias = phdr - vaddr;
> +       return *pdyn_vaddr != 0;
> +     }
> +      break;
> +
> +    case PT_DYNAMIC:
> +      *pdyn_vaddr = vaddr;
> +      *pdyn_filesz = filesz;
> +      return *pdyn_bias != (GElf_Addr) - 1;
> +    }
> +
> +  return false;
> +}
> +
> +#define consider_phdr(type, vaddr, filesz) \
> +  do_consider_phdr (type, vaddr, filesz, \
> +                    dwfl, phdr, &dyn_bias, &dyn_vaddr, &dyn_filesz)

This is just used twice in the function. Might be simpler to just inline
it twice and simplifying the switch by an if type == PT_... check
instead of inventing a new function with 8 arguments.

Thanks,

Mark