| Message ID | 1431693789-9679-1-git-send-email-colin.king@canonical.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
* Colin King <colin.king@canonical.com> wrote: > From: Colin Ian King <colin.king@canonical.com> > > Recent commit 3b8786ff7a1b31645ae2c26a2ec32dbd42ac1094 > ("ARM: 8352/1: perf: Fix the pmu node name in warning message") > introduced a memory leak of irqs on the "Don't bother with PPIs" > return path. This was picked up by static analysis by cppcheck: > > [arch/arm/kernel/perf_event_cpu.c:315]: (error) Memory leak: irqs > > simpele fix is to free irqs when returning. > > Signed-off-by: Colin Ian King <colin.king@canonical.com> > --- > arch/arm/kernel/perf_event_cpu.c | 4 +++- > 1 file changed, 3 insertions(+), 1 deletion(-) > > diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c > index 213919b..9e5b2a5 100644 > --- a/arch/arm/kernel/perf_event_cpu.c > +++ b/arch/arm/kernel/perf_event_cpu.c > @@ -311,8 +311,10 @@ static int of_pmu_irq_cfg(struct platform_device *pdev) > > /* Don't bother with PPIs; they're already affine */ > irq = platform_get_irq(pdev, 0); > - if (irq >= 0 && irq_is_percpu(irq)) > + if (irq >= 0 && irq_is_percpu(irq)) { > + kfree(irqs); > return 0; > + } > > for (i = 0; i < pdev->num_resources; ++i) { > struct device_node *dn; So returning from the middle of a function isn't very clean. Also, why do we return 0 in an error case? Furthermore, this function already has a (partially hidden) error cleanup path: if (i == pdev->num_resources) cpu_pmu->irq_affinity = irqs; else kfree(irqs); So this code should use proper goto driven cleanup. That's faster and cleaner, and is less likely to result in bugs like the above. Thanks, Ingo
On 16/05/15 08:09, Ingo Molnar wrote: > > * Colin King <colin.king@canonical.com> wrote: > >> From: Colin Ian King <colin.king@canonical.com> >> >> Recent commit 3b8786ff7a1b31645ae2c26a2ec32dbd42ac1094 >> ("ARM: 8352/1: perf: Fix the pmu node name in warning message") >> introduced a memory leak of irqs on the "Don't bother with PPIs" >> return path. This was picked up by static analysis by cppcheck: >> >> [arch/arm/kernel/perf_event_cpu.c:315]: (error) Memory leak: irqs >> >> simpele fix is to free irqs when returning. >> >> Signed-off-by: Colin Ian King <colin.king@canonical.com> >> --- >> arch/arm/kernel/perf_event_cpu.c | 4 +++- >> 1 file changed, 3 insertions(+), 1 deletion(-) >> >> diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c >> index 213919b..9e5b2a5 100644 >> --- a/arch/arm/kernel/perf_event_cpu.c >> +++ b/arch/arm/kernel/perf_event_cpu.c >> @@ -311,8 +311,10 @@ static int of_pmu_irq_cfg(struct platform_device *pdev) >> >> /* Don't bother with PPIs; they're already affine */ >> irq = platform_get_irq(pdev, 0); >> - if (irq >= 0 && irq_is_percpu(irq)) >> + if (irq >= 0 && irq_is_percpu(irq)) { >> + kfree(irqs); >> return 0; >> + } >> >> for (i = 0; i < pdev->num_resources; ++i) { >> struct device_node *dn; > > So returning from the middle of a function isn't very clean. > > Also, why do we return 0 in an error case? I believe that's explained in commit 338d9dd3e2aee00a9198e8bf6e7d535d3feeaf32 ("ARM: 8351/1: perf: don't warn about missing interrupt-affinity property for PPIs"): "PPIs are affine by nature, so the interrupt-affinity property is not used and therefore we shouldn't print a warning in its absence." > > Furthermore, this function already has a (partially hidden) error > cleanup path: > > if (i == pdev->num_resources) > cpu_pmu->irq_affinity = irqs; > else > kfree(irqs); > > So this code should use proper goto driven cleanup. That's faster and > cleaner, and is less likely to result in bugs like the above. > > Thanks, > > Ingo >
* Colin Ian King <colin.king@canonical.com> wrote: > On 16/05/15 08:09, Ingo Molnar wrote: > > > > * Colin King <colin.king@canonical.com> wrote: > > > >> From: Colin Ian King <colin.king@canonical.com> > >> > >> Recent commit 3b8786ff7a1b31645ae2c26a2ec32dbd42ac1094 > >> ("ARM: 8352/1: perf: Fix the pmu node name in warning message") > >> introduced a memory leak of irqs on the "Don't bother with PPIs" > >> return path. This was picked up by static analysis by cppcheck: > >> > >> [arch/arm/kernel/perf_event_cpu.c:315]: (error) Memory leak: irqs > >> > >> simpele fix is to free irqs when returning. > >> > >> Signed-off-by: Colin Ian King <colin.king@canonical.com> > >> --- > >> arch/arm/kernel/perf_event_cpu.c | 4 +++- > >> 1 file changed, 3 insertions(+), 1 deletion(-) > >> > >> diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c > >> index 213919b..9e5b2a5 100644 > >> --- a/arch/arm/kernel/perf_event_cpu.c > >> +++ b/arch/arm/kernel/perf_event_cpu.c > >> @@ -311,8 +311,10 @@ static int of_pmu_irq_cfg(struct platform_device *pdev) > >> > >> /* Don't bother with PPIs; they're already affine */ > >> irq = platform_get_irq(pdev, 0); > >> - if (irq >= 0 && irq_is_percpu(irq)) > >> + if (irq >= 0 && irq_is_percpu(irq)) { > >> + kfree(irqs); > >> return 0; > >> + } > >> > >> for (i = 0; i < pdev->num_resources; ++i) { > >> struct device_node *dn; > > > > So returning from the middle of a function isn't very clean. > > > > Also, why do we return 0 in an error case? > > I believe that's explained in commit > 338d9dd3e2aee00a9198e8bf6e7d535d3feeaf32 ("ARM: 8351/1: perf: don't warn > about missing interrupt-affinity property for PPIs"): > > "PPIs are affine by nature, so the interrupt-affinity property is not > used and therefore we shouldn't print a warning in its absence." That should probably be mentioned in the fine code as well, to keep future generations from wondering. Thanks, Ingo
diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c index 213919b..9e5b2a5 100644 --- a/arch/arm/kernel/perf_event_cpu.c +++ b/arch/arm/kernel/perf_event_cpu.c @@ -311,8 +311,10 @@ static int of_pmu_irq_cfg(struct platform_device *pdev) /* Don't bother with PPIs; they're already affine */ irq = platform_get_irq(pdev, 0); - if (irq >= 0 && irq_is_percpu(irq)) + if (irq >= 0 && irq_is_percpu(irq)) { + kfree(irqs); return 0; + } for (i = 0; i < pdev->num_resources; ++i) { struct device_node *dn;