[LTP] [dennis-percpu:for-6.11] [percpu_counter] 05dbad003f: ltp.getrusage03.fail

[Mateusz Guzik]

On Tue, May 28, 2024 at 02:45:35PM +0800, kernel test robot wrote:
> Hello,
> 
> kernel test robot noticed "ltp.getrusage03.fail" on:
> 
> commit: 05dbad003f2b2ececb1cc7428bfa8f470cc34b95 ("percpu_counter: add a cmpxchg-based _add_batch variant")
> https://git.kernel.org/cgit/linux/kernel/git/dennis/percpu.git for-6.11
> 
> in testcase: ltp
> version: ltp-x86_64-14c1f76-1_20240525
> with following parameters:
> 
> 	disk: 1HDD
> 	fs: xfs
> 	test: syscalls-03/getrusage03
> 
> 
> 
> Running tests.......
> <<<test_start>>>
> tag=getrusage03 stime=1716734654
> cmdline="getrusage03"
> contacts=""
> analysis=exit
> <<<test_output>>>
> tst_test.c:1733: TINFO: LTP version: 20240524-4-g22004c7db
> tst_test.c:1617: TINFO: Timeout per run is 0h 02m 30s
> getrusage03.c:43: TPASS: initial.self ~= child.self
> getrusage03.c:59: TFAIL: initial.children = 0, expected 102400
> getrusage03.c:66: TPASS: child.children == 0
> getrusage03.c:86: TFAIL: child.children = 0, expected 307200
> getrusage03.c:104: TPASS: initial.children ~= pre_wait.children
> getrusage03.c:114: TFAIL: post_wait.children = 0, expected 409600
> getrusage03.c:133: TPASS: initial.children ~= after_zombie.children
> getrusage03_child.c:57: TPASS: initial.self ~= exec.self
> getrusage03_child.c:62: TPASS: initial.children ~= exec.children
> 

I confirm this patch broke things, thanks for the report.

Make sure to sit before you take a look at the fix:

commit 6e8b4caf667fb6fad1c63b061e303faab6d917ef
Author: Mateusz Guzik <mjguzik@gmail.com>
Date:   Tue May 28 04:52:10 2024 -0400

    lol

diff --git a/lib/percpu_counter.c b/lib/percpu_counter.c
index c3140276bb36..51bc5246986d 100644
--- a/lib/percpu_counter.c
+++ b/lib/percpu_counter.c
@@ -97,7 +97,7 @@ void percpu_counter_add_batch(struct percpu_counter *fbc, s64 amount, s32 batch)

        count = this_cpu_read(*fbc->counters);
        do {
-               if (unlikely(abs(count + amount)) >= batch) {
+               if (unlikely(abs(count + amount) >= batch)) {
                        raw_spin_lock_irqsave(&fbc->lock, flags);
                        /*
                         * Note: by now we might have migrated to another CPU


Dennis, do I need to submit a v4? Given that this is a trivial one line
fixup perhaps it would be handier if you massaged the stuff on your
branch. I'm fine either way.

That said this really should have been reported by something.