Re: [BUG] msr-trace.h:42 suspicious rcu_dereference_check() usage!

[Peter Zijlstra]

On Mon, Nov 21, 2016 at 01:53:43AM +0100, Jiri Olsa wrote:
> hi,
> Jan hit following output when msr tracepoints are enabled on amd server:
> 
> [   91.585653] ===============================
> [   91.589840] [ INFO: suspicious RCU usage. ]
> [   91.594025] 4.9.0-rc1+ #1 Not tainted
> [   91.597691] -------------------------------
> [   91.601877] ./arch/x86/include/asm/msr-trace.h:42 suspicious 
> rcu_dereference_check() usage!
> [   91.610222] 
> [   91.610222] other info that might help us debug this:
> [   91.610222] 
> [   91.618224] 
> [   91.618224] RCU used illegally from idle CPU!
> [   91.618224] rcu_scheduler_active = 1, debug_locks = 0
> [   91.629081] RCU used illegally from extended quiescent state!
> [   91.634820] no locks held by swapper/1/0.
> [   91.638832] 
> [   91.638832] stack backtrace:
> [   91.643192] CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.9.0-rc1+ #1
> [   91.649457] Hardware name: empty empty/S3992, BIOS 'V2.03   ' 05/09/2008
> [   91.656159]  ffffc900018fbdf8 ffffffff813ed43c ffff88017ede8000 
> 0000000000000001
> [   91.663637]  ffffc900018fbe28 ffffffff810fdcd7 ffff880233f95dd0 
> 00000000c0010055
> [   91.671107]  0000000000000000 0000000000000000 ffffc900018fbe58 
> ffffffff814297ac
> [   91.678560] Call Trace:
> [   91.681022]  [<ffffffff813ed43c>] dump_stack+0x85/0xc9
> [   91.686164]  [<ffffffff810fdcd7>] lockdep_rcu_suspicious+0xe7/0x120
> [   91.692429]  [<ffffffff814297ac>] do_trace_read_msr+0x14c/0x1b0
> [   91.698349]  [<ffffffff8106ddb2>] native_read_msr+0x32/0x40
> [   91.703921]  [<ffffffff8103b2be>] amd_e400_idle+0x7e/0x110
> [   91.709407]  [<ffffffff8103b78f>] arch_cpu_idle+0xf/0x20
> [   91.714720]  [<ffffffff8181cd33>] default_idle_call+0x23/0x40
> [   91.720467]  [<ffffffff810f306a>] cpu_startup_entry+0x1da/0x2b0
> [   91.726387]  [<ffffffff81058b1f>] start_secondary+0x17f/0x1f0
> 
> 
> it got away with attached change.. but this rcu logic
> is far beyond me, so it's just wild guess.. ;-)

I think I prefer something like the below, that only annotates the one
RDMSR in question, instead of all of them.


diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 0888a879120f..d6c6aa80675f 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -357,7 +357,7 @@ static void amd_e400_idle(void)
        if (!amd_e400_c1e_detected) {
                u32 lo, hi;
 
-               rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
+               RCU_NONIDLE(rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi));
 
                if (lo & K8_INTP_C1E_ACTIVE_MASK) {
                        amd_e400_c1e_detected = true;