Re: [PATCH] cgroup/cpuset: Adjust exception handling in generate_sched_domains()

From: Waiman Long
Date: Wed Jan 10 2024 - 10:22:05 EST


On 12/31/23 02:28, Markus Elfring wrote:
From: Markus Elfring <elfring@xxxxxxxxxxxxxxxxxxxxx>
Date: Sun, 31 Dec 2023 08:00:31 +0100

Two resource allocation failures triggered further actions
over the label “done” so far.

* Jump to the statement “ndoms = 1;” in three cases directly
by using the label “set_ndoms” instead.

* Delete an assignment for the variable “ndoms” in one if branch.

Signed-off-by: Markus Elfring <elfring@xxxxxxxxxxxxxxxxxxxxx>
---
kernel/cgroup/cpuset.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index ba36c073304a..29ccd52eb45c 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -973,10 +973,9 @@ static int generate_sched_domains(cpumask_var_t **domains,

/* Special case for the 99% of systems with one, full, sched domain */
if (root_load_balance && !top_cpuset.nr_subparts) {
- ndoms = 1;

This statement cannot be removed as it is used in the following statement to size the allocation. This change the behavior to make doms NULL rather than a 1-entry array. This is probably the cause of the kernel test robot failure report.

NAK

doms = alloc_sched_domains(ndoms);
if (!doms)
- goto done;
+ goto set_ndoms;

dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
if (dattr) {
@@ -986,12 +985,12 @@ static int generate_sched_domains(cpumask_var_t **domains,
cpumask_and(doms[0], top_cpuset.effective_cpus,
housekeeping_cpumask(HK_TYPE_DOMAIN));

- goto done;
+ goto set_ndoms;
}

csa = kmalloc_array(nr_cpusets(), sizeof(cp), GFP_KERNEL);
if (!csa)
- goto done;
+ goto set_ndoms;
csn = 0;

rcu_read_lock();
@@ -1123,6 +1122,7 @@ static int generate_sched_domains(cpumask_var_t **domains,
* See comments in partition_sched_domains().
*/
if (doms == NULL)
+set_ndoms:
ndoms = 1;

*domains = doms;
--
2.43.0