arm64: cpuinfo: Fix sysfs cleanup on failure
PATCH 1/1 arm64: cpuinfo: Fix sysfs cleanup on failure
cpuid_cpu_online() creates the CPU registers kobject and the base identification attribute group before optionally merging the SME-specific SMIDR attribute into that group.
If the base group creation fails, return after deleting the kobject. If the SME merge fails, remove the base group before deleting the kobject.
--- arch/arm64/kernel/cpuinfo.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-)
arch/arm64/kernel/cpuinfo.c +12 -2
@@ -380,9 +380,19 @@ static int cpuid_cpu_online(unsigned int cpu) goto out; rc =sysfs_create_group (&info->kobj, &cpuregs_attr_group);if (rc) -kobject_del (&info->kobj); -if (system_supports_sme ()) + goto out_del_kobj; +if (system_supports_sme ()) { rc =sysfs_merge_group (&info->kobj, &sme_cpuregs_attr_group); +if (rc) + goto out_remove_group; + } + +return 0 ; + +out_remove_group: +sysfs_remove_group (&info->kobj, &cpuregs_attr_group); +out_del_kobj: +kobject_del (&info->kobj); out:return rc; }
Signature · 1 line
2.43.0
Will Deacon
maintainerQuoted 6 lines · Yuho Choi
cpuid_cpu_online()creates the CPU registers kobject and the base identification attribute group before optionally merging the SME-specific SMIDR attribute into that group.If the base group creation fails, return after deleting the kobject. If the SME merge fails, remove the base group before deleting the kobject.
Why is it a good idea to remove the entire group if the merge fails? Wouldn't it be more pragmatic to continue with the group that we did manage to create? I don't really get why this proposed behaviour is better.
Will
Quoted 28 lines
arch/arm64/kernel/cpuinfo.c +12
-2@@ -380,9 +380,19 @@ static int cpuid_cpu_online(unsigned int cpu) goto out; rc =sysfs_create_group (&info->kobj, &cpuregs_attr_group);if (rc) -kobject_del (&info->kobj); -if (system_supports_sme ()) + goto out_del_kobj; +if (system_supports_sme ()) { rc =sysfs_merge_group (&info->kobj, &sme_cpuregs_attr_group); +if (rc) + goto out_remove_group; + } + +return 0 ; + +out_remove_group: +sysfs_remove_group (&info->kobj, &cpuregs_attr_group); +out_del_kobj: +kobject_del (&info->kobj); out:return rc; }Signature · 1 line
2.43.0
Quoted 5 lines · Will Deacon
Why is it a good idea to remove the entire group if the merge fails? Wouldn't it be more pragmatic to continue with the group that we did manage to create? I don't really get why this proposed behaviour is better.
You're right. Removing the entire group on an SME merge failure is unncessary. I'll send a v2 incorporating your suggestion.
Thanks for your time reviewing.
Yuho.