Re: BUG: sleeping function called from invalid context at kernel/kallsyms.c:305

[Leizhen (ThunderTown)]

On 2022/12/16 18:45, Anders Roxell wrote:
> Hi,
> 
> I'm building an arm64 allmodconfig kernel (see the .config [1]) on
> yesterdays next tag 20221215 with KALLSYMS_SELFTEST enabled and I saw
> the following bug when the selftest ran:
> 
> [29725.015182][   T58] BUG: sleeping function called from invalid
> context at kernel/kallsyms.c:305
> [29725.022953][   T58] in_atomic(): 0, irqs_disabled(): 128,
> non_block: 0, pid: 58, name: kallsyms_test
> [29725.031272][   T58] preempt_count: 0, expected: 0
> [29725.035903][   T58] RCU nest depth: 0, expected: 0
> [29725.040574][   T58] no locks held by kallsyms_test/58.
> [29725.045494][   T58] irq event stamp: 18899904
> [29725.049809][ T58] hardirqs last enabled at (18899903):
> finish_task_switch.isra.0 (core.c:?)
> [29725.059608][ T58] hardirqs last disabled at (18899904):
> test_perf_kallsyms_on_each_symbol (kallsyms_selftest.c:?)
> [29725.069936][ T58] softirqs last enabled at (18899886): __do_softirq (??:?)
> [29725.078670][ T58] softirqs last disabled at (18899879):
> ____do_softirq (irq.c:?)
> [29725.087399][   T58] CPU: 0 PID: 58 Comm: kallsyms_test Tainted: G
>              T  6.1.0-next-20221215 #2
> 0a142be8faea13ac333ed9a1cf4e71b6966ad16e
> [29725.099607][   T58] Hardware name: linux,dummy-virt (DT)
> [29725.104674][   T58] Call trace:
> [29725.107909][ T58] dump_backtrace (??:?)
> [29725.112706][ T58] show_stack (??:?)
> [29725.116883][ T58] dump_stack_lvl (??:?)
> [29725.121666][ T58] dump_stack (??:?)
> [29725.125852][ T58] __might_resched (??:?)
> [29725.130712][ T58] kallsyms_on_each_symbol (??:?)
> [29725.136018][ T58] test_perf_kallsyms_on_each_symbol (kallsyms_selftest.c:?)
> [29725.142008][ T58] test_entry (kallsyms_selftest.c:?)
> [29725.146312][ T58] kthread (kthread.c:?)
> [29725.150567][ T58] ret_from_fork (??:?)
> [29734.975283][   T58] kallsyms_selftest: kallsyms_on_each_symbol()
> traverse all: 5744310840 ns
> [29734.992268][   T58] kallsyms_selftest:
> kallsyms_on_each_match_symbol() traverse all: 1164580 ns
> [29735.049679][   T58] kallsyms_selftest: finish
> 
> looks like the issue is that test_perf_kallsyms_on_each_symbol() does
> these function calls:
> 
>        local_irq_save(flags);
>        t0 = sched_clock();
>        kallsyms_on_each_match_symbol(match_symbol, stat.name, &stat);
>        t1 = sched_clock();
>        local_irq_restore(flags);
> 
> and inside kallsyms_on_each_match_symbol(), cond_resched() is called.
> 
> Any ideas how to solve this?

I was initially hoping that the test process would not be interrupted.
It is not a big problem to remove local_irq_save(). From a probabilistic
statistical point of view: it does not affect the horizontal comparison
between kallsyms_on_each_symbol() and kallsyms_on_each_match_symbol(),
and the vertical comparison before and after optimization is also not
affected.

Or do not call cond_resched() during the test. This method seems to be
a little better.

diff --git a/kernel/kallsyms.c b/kernel/kallsyms.c
index e31b4d87a4e1b71..9e07fabaf08e647 100644
--- a/kernel/kallsyms.c
+++ b/kernel/kallsyms.c
@@ -358,7 +358,8 @@ int kallsyms_on_each_symbol(int (*fn)(void *, const char *, struct module *,
                ret = fn(data, namebuf, NULL, kallsyms_sym_address(i));
                if (ret != 0)
                        return ret;
-               cond_resched();
+               if (!IS_ENABLED(CONFIG_KALLSYMS_SELFTEST))
+                       cond_resched();
        }
        return 0;
 }
@@ -375,7 +376,8 @@ int kallsyms_on_each_match_symbol(int (*fn)(void *, unsigned long),

        for (i = start; !ret && i <= end; i++) {
                ret = fn(data, kallsyms_sym_address(get_symbol_seq(i)));
-               cond_resched();
+               if (!IS_ENABLED(CONFIG_KALLSYMS_SELFTEST))
+                       cond_resched();
        }

        return ret;

> 
> Cheers,
> Anders
> [1] https://people.linaro.org/~anders.roxell/next-20221215.config
> .
> 

-- 
Regards,
  Zhen Lei