public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug gcov-profile/105535] New: GCOV analysis on 16-bit target systems impossible
@ 2022-05-09 13:11 peter-helfert@t-online.de
  2022-05-09 14:19 ` [Bug gcov-profile/105535] " marxin at gcc dot gnu.org
                   ` (6 more replies)
  0 siblings, 7 replies; 8+ messages in thread
From: peter-helfert@t-online.de @ 2022-05-09 13:11 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105535

            Bug ID: 105535
           Summary: GCOV analysis on 16-bit target systems impossible
           Product: gcc
           Version: 12.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: gcov-profile
          Assignee: unassigned at gcc dot gnu.org
          Reporter: peter-helfert@t-online.de
                CC: marxin at gcc dot gnu.org
  Target Milestone: ---

Created attachment 52944
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52944&action=edit
Proposed Patch file for libgcc/libgcov.h

Refers to 

GCC versions:

9.3.1 (oldest version I tested)

up to current release

12.1.0

GCOV analysis on 16-bit target systems impossible.

Length divergence of struct “gcov_info” member “n_functions”.

Dear GCC team,

Currently I am using GCC and its libgcov for on-target code coverage analysis
on a 16-bit target system (TI MSP-430). Unfortunately, during my GCOV test runs
I noticed an issue which breaks  code coverage analysis on 16-bit (and possibly
other) systems when using libgcov to assemble and write coverage information
directly on the target.

The issue is caused by a length divergence between the struct member
“n_functions” in the struct “gcov_info” (as declared in “libgcov.h”) and
statically allocated memory in generated object files.  The member
“n_functions” is declared as type “unsigned” in libgcov.h. The declared size
thus depends on the target system. On 32-bit systems, the “n_functions” member
declaration is 32 bits wide. However, on a 16-bit system, the size of the
“n_functions” element usually is reduced to 16 bits (sizeof(unsigned) == 2). 

This variable length, which depends on the compilation target, is not taken
into account when constructing the GCOV information stored as constant data in
the object file, generated at compile time. This information is generated and
included for every object file instrumented for GCOV analysis.  During
compilation, GCC generates an instance of the “gcov_info” struct in the
appropriate linker input section, where its member “n_functions” has a size of
(at least) 32 bits, irrespective of the compilation target. This size is not
compatible with the size declared in struct “gcov_info” on my 16-bit target, as
declared in the libgcov.h C-headers.

In gcc, during compilation of a source code file, the “n_functions” member is
generated by invoking “get_gcov_unsigned_t()” in function
“coverage_obj_init()”. The minimum size of an element generated by
get_gcov_usigned_t() is there hard-coded to 32 bits.

The size divergence of the “gcov_info” struct member “n_functions” causes a
shift of the subsequent struct member “functions" (pointer) by 2 bytes. This in
turn leads to a malfunction as soon as the pointer is accessed at runtime on
the 16-bit target. The issue generally prevents the target software from
collecting and outputting GCOV information on 16-bit systems and therefore
seems to be a quite severe problem, especially as it usually causes
unpredictable malfunction on software instrumented for GCOV analysis.

I wonder why that problem has not been reported before. Possibly, code coverage
analysis on 16-bit targets, using libgcov on target, is still not very common
these days.

Suggested solution:

In GCC source file “libgcov.h”:
In declaration of struct “gcov_info”:

Change: 

“unsigned n_functions” 

to 

“gcov_unsigned_t n_functions”

The gcov_unsigned_t is declared:

     typedef unsgined gcov_unsigned_t __attribute__ ((mode (SI)))

which ensures a 32-bit size for this type.

Other declarations in the context of libgcov and coverage, where the same
problem would otherwise appear, have been solved in the same fashion as my
proposed fix.

A proposed patch is attached to this bug report.

Thank you for considering my report.

Best regards,

Hans-Peter Helfert

^ permalink raw reply	[flat|nested] 8+ messages in thread
end of thread, other threads:[~2022-05-13 12:05 UTC | newest]

Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-05-09 13:11 [Bug gcov-profile/105535] New: GCOV analysis on 16-bit target systems impossible peter-helfert@t-online.de
2022-05-09 14:19 ` [Bug gcov-profile/105535] " marxin at gcc dot gnu.org
2022-05-09 14:19 ` marxin at gcc dot gnu.org
2022-05-09 15:27 ` peter-helfert@t-online.de
2022-05-10 10:47 ` cvs-commit at gcc dot gnu.org
2022-05-10 10:48 ` marxin at gcc dot gnu.org
2022-05-13 12:04 ` cvs-commit at gcc dot gnu.org
2022-05-13 12:05 ` marxin at gcc dot gnu.org

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).