[gnutls-devel] libtasn1 | Reading CHOICE name confuses valgrind (#9)

[Development of GNU's TLS library]

Tim Rühsen commented:


Valgrind doens't complain when built with -O0, the assembler (L122/strcmp starts at the arrow):
```
   0x00005555555554f8 <+803>:   mov    $0x76,%esi
   0x00005555555554fd <+808>:   lea    0xb14(%rip),%rdi        # 0x555555556018
   0x0000555555555504 <+815>:   mov    $0x0,%eax
   0x0000555555555509 <+820>:   callq  0x555555555060 <printf at plt>
   0x000055555555550e <+825>:   mov    $0x1,%edi
   0x0000555555555513 <+830>:   callq  0x5555555550b0 <exit at plt>
=> 0x0000555555555518 <+835>:   lea    -0x4b0(%rbp),%rax
   0x000055555555551f <+842>:   lea    0xb4c(%rip),%rsi        # 0x555555556072
   0x0000555555555526 <+849>:   mov    %rax,%rdi
   0x0000555555555529 <+852>:   callq  0x555555555080 <strcmp at plt>
   0x000055555555552e <+857>:   test   %eax,%eax
   0x0000555555555530 <+859>:   jne    0x555555555552 <main+893>
   0x0000555555555532 <+861>:   mov    $0x7c,%esi
   0x0000555555555537 <+866>:   lea    0xada(%rip),%rdi        # 0x555555556018
```

In comparison, the code generated by gcc 9 with -O2 (see above) just accesses 16 bytes of stack memory for an optimized version of `strcmp`. So no wonder that valgrind complains, though it is a false positive.

To calm down valgrind, we can `*data = 0` before `asn1_read_value()`. This just 'marks' `data` as being initialized. Alternatively, we have to add/create a valgrind suppression.

-- 
Reply to this email directly or view it on GitLab: https://gitlab.com/gnutls/libtasn1/issues/9#note_237741131
You're receiving this email because of your account on gitlab.com.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.gnupg.org/pipermail/gnutls-devel/attachments/20191030/b80691c9/attachment.html>