Fix /proc/cpuinfo cpumask warning
v4 cover Fix /proc/cpuinfo cpumask warning
Commit 78e5a3399421 ("cpumask: fix checking valid cpu range") started issuing warnings[1] when cpu indices equal to nr_cpu_ids - 1 were passed to cpumask_next* functions. The commit has since been reverted with commit 80493877d7d0 ("Revert "cpumask: fix checking valid cpu range"."), which raises the question as to how much this proposed patch is needed. Additionally, there's some discussion as to whether or not cpumask_next() should even be validating its inputs[2]. So, with that in mind, I'm fine with the patch being dropped. However, it may still be reasonable to add the checking to /proc/cpuinfo until cpumask_next has made changes and better documented its API.
[1] Warnings will only appear with DEBUG_PER_CPU_MAPS enabled.
[2] https://lore.kernel.org/all/CAHk-=wihz-GXx66MmEyaADgS1fQE_LDcB9wrHAmkvXkd8nx9tA@mail.gmail.com/
This series addresses the issue for x86. riscv has already merged an equivalent patch (v3 of this series). Also, from a quick grep of cpuinfo seq operations, I think at least openrisc, powerpc, and s390 could get an equivalent patch. While the test is simple (see next paragraph) I'm not equipped to test on each architecture.
To test, just build a kernel with DEBUG_PER_CPU_MAPS enabled, boot to a shell, do 'cat /proc/cpuinfo', and look for a kernel warning.
v4:
- The riscv patch has already been merged.
- Mostly rewrote the cover letter as the situation has changed since
78e5a3399421 was reverted.
- Rewrote the commit message in order to try an better clarify things and also to add the reference to the revert commit, which results in the commit no longer claiming its a 'fix' in its summary. [Boris]
v3:
- Change condition from >= to == in order to still get a warning for > as that's unexpected. [Yury]
- Picked up tags on the riscv patch
v2:
- Added all the information I should have in the first place to the commit message [Boris]
- Changed style of fix [Boris]
Andrew Jones (1):
x86: cpuinfo: Ensure inputs to cpumask_next are valid
arch/x86/kernel/cpu/proc.c | 3 +++ 1 file changed, 3 insertions(+)
Signature · 1 line
2.37.3
PATCH v4 1/1 x86: cpuinfo: Ensure inputs to cpumask_next are valid
The valid cpumask range is [0, nr_cpu_ids) and cpumask_next() currently calls find_next_bit() with its input CPU ID number plus one for the bit number, giving cpumask_next() the range [-1, nr_cpu_ids - 1). seq_read_iter() and cpuinfo's start and next seq operations implement a pattern like
n = cpumask_next(n - 1, mask);
show(n);
while (1) {
++n;
n = cpumask_next(n - 1, mask);
if (n >= nr_cpu_ids)
break;
show(n);
}which will eventually result in cpumask_next() being called with nr_cpu_ids - 1. A kernel compiled with commit 78e5a3399421 ("cpumask:
fix checking valid cpu range"), but not its revert, commit 80493877d7d0 ("Revert "cpumask: fix checking valid cpu range"."), will generate a warning when DEBUG_PER_CPU_MAPS is enabled each time /proc/cpuinfo is read. Future-proof cpuinfo by checking its input to cpumask_next() is valid.
--- arch/x86/kernel/cpu/proc.c | 3 +++ 1 file changed, 3 insertions(+)
arch/x86/kernel/cpu/proc.c +3 -0
@@ -153,6 +153,9 @@ static int show_cpuinfo(struct seq_file *m, void *v)static void *c_start (struct seq_file *m,loff_t *pos) { +if (*pos == nr_cpu_ids) +return NULL ; + *pos =cpumask_next (*pos -1 , cpu_online_mask);if ((*pos) < nr_cpu_ids)return &cpu_data (*pos);
Signature · 1 line
2.37.3
Re: [PATCH v4 1/1] x86: cpuinfo: Ensure inputs to cpumask_next are valid
Quoted 24 lines · Andrew Jones
The valid cpumask range is [0,
nr_cpu_ids) andcpumask_next()currently callsfind_next_bit()with its input CPU ID number plus one for the bit number, givingcpumask_next()the range [-1,nr_cpu_ids- 1).seq_read_iter()and cpuinfo's start and next seq operations implement a pattern liken = cpumask_next(n - 1, mask); show(n); while (1) { ++n; n = cpumask_next(n - 1, mask); if (n >= nr_cpu_ids) break; show(n); }which will eventually result in
cpumask_next()being called withnr_cpu_ids- 1. A kernel compiled with commit78e5a3399421("cpumask:
fix checking valid cpu range"), but not its revert, commit80493877d7d0("Revert "cpumask: fix checking valid cpu range"."), will generate a warning whenDEBUG_PER_CPU_MAPSis enabled each time/proc/cpuinfois read. Future-proof cpuinfo by checking its input tocpumask_next()is valid.
Quoted 2 lines
Signed-off-by Andrew Jones <[email protected]>Cc Yury Norov <[email protected]>
Quoted 20 lines
--- arch/x86/kernel/cpu/proc.c | 3 +++ 1 file changed, 3 insertions(+)arch/x86/kernel/cpu/proc.c +3
-0@@ -153,6 +153,9 @@ static int show_cpuinfo(struct seq_file *m, void *v)static void *c_start (struct seq_file *m,loff_t *pos) { +if (*pos == nr_cpu_ids) +return NULL ; + *pos =cpumask_next (*pos -1 , cpu_online_mask);if ((*pos) < nr_cpu_ids)return &cpu_data (*pos);Signature · 1 line
2.37.3