| Message ID | 20240403125109.2054881-7-dawei.li@shingroup.cn (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | perf: Avoid placing cpumask var on stack | expand |
On Wed, Apr 03, 2024 at 08:51:05PM +0800, Dawei Li wrote: > For CONFIG_CPUMASK_OFFSTACK=y kernel, explicit allocation of cpumask > variable on stack is not recommended since it can cause potential stack > overflow. > > Instead, kernel code should always use *cpumask_var API(s) to allocate > cpumask var in config-neutral way, leaving allocation strategy to > CONFIG_CPUMASK_OFFSTACK. > > But dynamic allocation in cpuhp's teardown callback is somewhat problematic > for if allocation fails(which is unlikely but still possible): > - If -ENOMEM is returned to caller, kernel crashes for non-bringup > teardown; > - If callback pretends nothing happened and returns 0 to caller, it may > trap system into an in-consisitent/compromised state; > > Use newly-introduced cpumask_any_and_but() to address all issues above. > It eliminates usage of temporary cpumask var in generic way, no matter how > the cpumask var is allocated. > > Suggested-by: Mark Rutland <mark.rutland@arm.com> > Signed-off-by: Dawei Li <dawei.li@shingroup.cn> The logic looks good to me, but I'd like the commit message updated the same as per my comment on patch 2. With that commit message: Reviewed-by: Mark Rutland <mark.rutland@arm.com> Mark. > --- > drivers/perf/dwc_pcie_pmu.c | 10 ++++------ > 1 file changed, 4 insertions(+), 6 deletions(-) > > diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c > index 957058ad0099..c5e328f23841 100644 > --- a/drivers/perf/dwc_pcie_pmu.c > +++ b/drivers/perf/dwc_pcie_pmu.c > @@ -690,9 +690,8 @@ static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_n > { > struct dwc_pcie_pmu *pcie_pmu; > struct pci_dev *pdev; > - int node; > - cpumask_t mask; > unsigned int target; > + int node; > > pcie_pmu = hlist_entry_safe(cpuhp_node, struct dwc_pcie_pmu, cpuhp_node); > /* Nothing to do if this CPU doesn't own the PMU */ > @@ -702,10 +701,9 @@ static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_n > pcie_pmu->on_cpu = -1; > pdev = pcie_pmu->pdev; > node = dev_to_node(&pdev->dev); > - if (cpumask_and(&mask, cpumask_of_node(node), cpu_online_mask) && > - cpumask_andnot(&mask, &mask, cpumask_of(cpu))) > - target = cpumask_any(&mask); > - else > + > + target = cpumask_any_and_but(cpumask_of_node(node), cpu_online_mask, cpu); > + if (target >= nr_cpu_ids) > target = cpumask_any_but(cpu_online_mask, cpu); > > if (target >= nr_cpu_ids) { > -- > 2.27.0 >
diff --git a/drivers/perf/dwc_pcie_pmu.c b/drivers/perf/dwc_pcie_pmu.c index 957058ad0099..c5e328f23841 100644 --- a/drivers/perf/dwc_pcie_pmu.c +++ b/drivers/perf/dwc_pcie_pmu.c @@ -690,9 +690,8 @@ static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_n { struct dwc_pcie_pmu *pcie_pmu; struct pci_dev *pdev; - int node; - cpumask_t mask; unsigned int target; + int node; pcie_pmu = hlist_entry_safe(cpuhp_node, struct dwc_pcie_pmu, cpuhp_node); /* Nothing to do if this CPU doesn't own the PMU */ @@ -702,10 +701,9 @@ static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_n pcie_pmu->on_cpu = -1; pdev = pcie_pmu->pdev; node = dev_to_node(&pdev->dev); - if (cpumask_and(&mask, cpumask_of_node(node), cpu_online_mask) && - cpumask_andnot(&mask, &mask, cpumask_of(cpu))) - target = cpumask_any(&mask); - else + + target = cpumask_any_and_but(cpumask_of_node(node), cpu_online_mask, cpu); + if (target >= nr_cpu_ids) target = cpumask_any_but(cpu_online_mask, cpu); if (target >= nr_cpu_ids) {
For CONFIG_CPUMASK_OFFSTACK=y kernel, explicit allocation of cpumask variable on stack is not recommended since it can cause potential stack overflow. Instead, kernel code should always use *cpumask_var API(s) to allocate cpumask var in config-neutral way, leaving allocation strategy to CONFIG_CPUMASK_OFFSTACK. But dynamic allocation in cpuhp's teardown callback is somewhat problematic for if allocation fails(which is unlikely but still possible): - If -ENOMEM is returned to caller, kernel crashes for non-bringup teardown; - If callback pretends nothing happened and returns 0 to caller, it may trap system into an in-consisitent/compromised state; Use newly-introduced cpumask_any_and_but() to address all issues above. It eliminates usage of temporary cpumask var in generic way, no matter how the cpumask var is allocated. Suggested-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Dawei Li <dawei.li@shingroup.cn> --- drivers/perf/dwc_pcie_pmu.c | 10 ++++------ 1 file changed, 4 insertions(+), 6 deletions(-)