Re: [peterz-queue:locking/core] [futex] 58141fa298: Oops:general_protection_fault,probably_for_non-canonical_address#:#[##]SMP

hi, Prateek,

On Fri, Jul 10, 2026 at 01:36:05PM +0530, K Prateek Nayak wrote:
> Hello Oliver,
> 
> On 7/9/2026 1:47 PM, kernel test robot wrote:
> > [    1.259631][    T1] Running RCU Tasks Trace wait API self tests
> > [    1.261867][    T1] posixtimers hash table entries: 1024 (order: 5, 81920 bytes, linear)
> > [    1.264376][    T1] Oops: general protection fault, probably for non-canonical address 0x123456789abcdef: 0000 [#1] SMP
> 
> Thank you for the report! This is an unpatched address for the runtime
> constant which suggests we are using the address before it is patched.
> Following seems to solve it for me on your config:
> 
>   (Prepared on top of queue:locking/core)
> 
> diff --git a/kernel/futex/core.c b/kernel/futex/core.c
> index c6af6edf5d7ae..0864c6e8cde78 100644
> --- a/kernel/futex/core.c
> +++ b/kernel/futex/core.c
> @@ -2027,6 +2027,8 @@ static int __init futex_init(void)
>  	runtime_const_init(mask,  __futex_mask);
>  	runtime_const_init(ptr,   __futex_queues);
>  
> +	barrier();
> +
>  	BUG_ON(!futex_queues());
>  
>  	for_each_node(n) {
> ---
> 
> Could you please check if this is the same issue you are hitting too?

I applied above patch upon 58141fa298 directly then build kernel with same
config and run same tests, confirmed the issue gone.

Tested-by: kernel test robot <[email protected]>

> 
> My default config seems to have been much less aggressive at
> optimization to have exposed this problem.
> 
> -- 
> Thanks and Regards,
> Prateek
>