Currently, we open the system base frequency file, but never close it, which results in a memory leak. Coverity issue: 369693 Fixes: 8a5febaac4f7 ("power: fix P-state base frequency handling") Cc: david.hunt@intel.com Cc: reshma.pattan@intel.com Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- Notes: Ideally, the close should be added at the end, but there's a bunch of ERR_RET macros before that, so addressing that would put us dangerously close to refactoring, which is not what we want to do so close to the release. This issue was already "fixed", but because the variable naming and the flow of code is confusing, the fix was addressing a different variable. There is a patch for 21.08 that will address the code flow and make it less confusing. lib/librte_power/power_pstate_cpufreq.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/lib/librte_power/power_pstate_cpufreq.c b/lib/librte_power/power_pstate_cpufreq.c index ec745153d3..1451a024be 100644 --- a/lib/librte_power/power_pstate_cpufreq.c +++ b/lib/librte_power/power_pstate_cpufreq.c @@ -185,6 +185,9 @@ power_init_for_setting_freq(struct pstate_power_info *pi) base_max_ratio = strtoul(buf_base, NULL, POWER_CONVERT_TO_DECIMAL) / BUS_FREQ; + /* not needed any more */ + fclose(f_base_max); + f_base_max = NULL; } snprintf(fullpath_min, sizeof(fullpath_min), POWER_SYSFILE_MIN_FREQ, -- 2.25.1
> -----Original Message-----
> From: Burakov, Anatoly <anatoly.burakov@intel.com>
> base_max_ratio =
> strtoul(buf_base, NULL,
> POWER_CONVERT_TO_DECIMAL)
> / BUS_FREQ;
> + /* not needed any more */
> + fclose(f_base_max);
> + f_base_max = NULL;
Should this be moved before FOPS_OR_NULL_GOTO() call ?
On 14-Apr-21 10:24 AM, Pattan, Reshma wrote: > > >> -----Original Message----- >> From: Burakov, Anatoly <anatoly.burakov@intel.com> >> base_max_ratio = >> strtoul(buf_base, NULL, >> POWER_CONVERT_TO_DECIMAL) >> / BUS_FREQ; >> +/* not needed any more */ >> +fclose(f_base_max); >> +f_base_max = NULL; > > Should this be moved before FOPS_OR_NULL_GOTO() call ? > No, otherwise we wouldn't be able to read the data. It *could* be moved to the end, but then we'd have to modify the rest of the logic as well, because right after this there are unconditional returns there. All of this is addressed in a refactor patch [1], this is just fixing a bug and nothing else. [1] http://patches.dpdk.org/project/dpdk/patch/20210402092701.258316-1-anatoly.burakov@intel.com/ -- Thanks, Anatoly
On 14-Apr-21 11:00 AM, Burakov, Anatoly wrote:
> On 14-Apr-21 10:24 AM, Pattan, Reshma wrote:
>>
>>
>>> -----Original Message-----
>>> From: Burakov, Anatoly <anatoly.burakov@intel.com>
>>> base_max_ratio =
>>> strtoul(buf_base, NULL,
>>> POWER_CONVERT_TO_DECIMAL)
>>> / BUS_FREQ;
>>> +/* not needed any more */
>>> +fclose(f_base_max);
>>> +f_base_max = NULL;
>>
>> Should this be moved before FOPS_OR_NULL_GOTO() call ?
>>
>
> No, otherwise we wouldn't be able to read the data. It *could* be moved
> to the end, but then we'd have to modify the rest of the logic as well,
> because right after this there are unconditional returns there. All of
> this is addressed in a refactor patch [1], this is just fixing a bug and
> nothing else.
>
> [1]
> http://patches.dpdk.org/project/dpdk/patch/20210402092701.258316-1-anatoly.burakov@intel.com/
>
>
Actually, no, you're right! This does have to be moved till before
FOPS_OR_NULL_GOTO, because this is where we actually read the data (if
we fail to read data, we leak). I hate this code :D
--
Thanks,
Anatoly
Currently, we open the system base frequency file, but never close it, which results in a memory leak. Coverity issue: 369693 Fixes: 8a5febaac4f7 ("power: fix P-state base frequency handling") Cc: david.hunt@intel.com Cc: reshma.pattan@intel.com Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> --- Notes: Ideally, the close should be added at the end, but there's a bunch of ERR_RET macros before that, so addressing that would put us dangerously close to refactoring, which is not what we want to do so close to the release. This issue was already "fixed", but because the variable naming and the flow of code is confusing, the fix was addressing a different variable. There is a patch for 21.08 that will address the code flow and make it less confusing. v2: - Move the fd close to before we check read data lib/librte_power/power_pstate_cpufreq.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/lib/librte_power/power_pstate_cpufreq.c b/lib/librte_power/power_pstate_cpufreq.c index ec745153d3..2cfc54acf3 100644 --- a/lib/librte_power/power_pstate_cpufreq.c +++ b/lib/librte_power/power_pstate_cpufreq.c @@ -175,6 +175,11 @@ power_init_for_setting_freq(struct pstate_power_info *pi) FOPEN_OR_ERR_RET(f_base_max, -1); if (f_base_max != NULL) { s_base_max = fgets(buf_base, sizeof(buf_base), f_base_max); + + /* close the file unconditionally */ + fclose(f_base_max); + f_base_max = NULL; + FOPS_OR_NULL_GOTO(s_base_max, out); buf_base[BUFSIZ-1] = '\0'; -- 2.25.1
> -----Original Message-----
> From: Burakov, Anatoly <anatoly.burakov@intel.com>
> Coverity issue: 369693
>
> Fixes: 8a5febaac4f7 ("power: fix P-state base frequency handling")
> v2:
> - Move the fd close to before we check read data
>
Acked-By: Reshma Pattan <reshma.pattan@intel.com>
> > From: Burakov, Anatoly <anatoly.burakov@intel.com>
> > Coverity issue: 369693
> >
> > Fixes: 8a5febaac4f7 ("power: fix P-state base frequency handling")
>
> Acked-By: Reshma Pattan <reshma.pattan@intel.com>
Applied, thanks