[igt-dev] [PATCH i-g-t] tools: Fix running intel_dp_compliance if it is found in the search path($PATH env variable)

[Petri Latvala]

On Wed, Apr 07, 2021 at 05:21:58PM -0700, Khaled Almahallawy wrote:
> When intel_dp_compliance binary location is part of PATH environment variable, running it from the shell without relative or full path, cause the following assertion failure because chdir:
> 
> $:~# intel_dp_compliance
> (intel_dp_compliance:3070) CRITICAL: Test assertion failure function enter_exec_path, file ../tools/igt_compliance_utils.c:56:
> (intel_dp_compliance:3070) CRITICAL: Failed assertion: ret == 0
> (intel_dp_compliance:3070) CRITICAL: Last errno: 2, No such file or directory
> (intel_dp_compliance:3070) CRITICAL: error: -1 != 0
> Stack trace:
>   #0 ../lib/igt_core.c:1726 __igt_fail_assert()
>   #1 [enter_exec_path+0x8b]
>   #2 ../tools/intel_dp_compliance.c:843 main()
>   #3 ../csu/libc-start.c:342 __libc_start_main()
>   #4 [_start+0x2e]
> Test (null) failed.
> **** DEBUG ****
> (intel_dp_compliance:3070) CRITICAL: Test assertion failure function enter_exec_path, file ../tools/igt_compliance_utils.c:56:
> (intel_dp_compliance:3070) CRITICAL: Failed assertion: ret == 0
> (intel_dp_compliance:3070) CRITICAL: Last errno: 2, No such file or directory
> (intel_dp_compliance:3070) CRITICAL: error: -1 != 0
> (intel_dp_compliance:3070) igt_core-INFO: Stack trace:
> (intel_dp_compliance:3070) igt_core-INFO:   #0 ../lib/igt_core.c:1726 __igt_fail_assert()
> (intel_dp_compliance:3070) igt_core-INFO:   #1 [enter_exec_path+0x8b]
> (intel_dp_compliance:3070) igt_core-INFO:   #2 ../tools/intel_dp_compliance.c:843 main()
> (intel_dp_compliance:3070) igt_core-INFO:   #3 ../csu/libc-start.c:342 __libc_start_main()
> (intel_dp_compliance:3070) igt_core-INFO:   #4 [_start+0x2e]
> ****  END  ****
> FAIL (-1.000s)
> 
> This patch fixes this issue.
> 
> Signed-off-by: Khaled Almahallawy <khaled.almahallawy at intel.com>
> ---
>  tools/igt_compliance_utils.c | 14 ++++++--------
>  1 file changed, 6 insertions(+), 8 deletions(-)
> 
> diff --git a/tools/igt_compliance_utils.c b/tools/igt_compliance_utils.c
> index d6a2e4e529d9..bcc707ae83c1 100644
> --- a/tools/igt_compliance_utils.c
> +++ b/tools/igt_compliance_utils.c
> @@ -39,22 +39,20 @@ struct termios saved_tio;
>  
>  void enter_exec_path(char **argv)
>  {
> -	char *exec_path = NULL;
> +	char exec_path[PATH_MAX];
>  	char *pos = NULL;
> -	short len_path = 0;
>  	int ret;
>  
> -	len_path = strlen(argv[0]);
> -	exec_path = (char *) malloc(len_path);
> +	memset(exec_path, 0, sizeof(exec_path));
> +	ret = readlink("/proc/self/exe", exec_path, PATH_MAX);

Does this need to be PATH_MAX - 1? A bit of a weird corner case of
having the path be exactly PATH_MAX long but would be consistent with
how readlink() is called elsewhere in IGT.

> +	igt_assert(ret != -1);
>  
> -	memcpy(exec_path, argv[0], len_path);
>  	pos = strrchr(exec_path, '/');
> -	if (pos != NULL)
> -		*(pos+1) = '\0';
> +	if(pos != NULL)
> +		*pos = '\0';

Could use dirname() for this but that's just a nitpick.



With or without those above,
Reviewed-by: Petri Latvala <petri.latvala at intel.com>