Re: rcu: hlist_bl_set_first_rcu hangs

[Nick Piggin]

On Wed, Jan 12, 2011 at 6:17 PM, Po-Yu Chuang <ratbert.chuang@xxxxxxxxx> wrote:
> Dear Nick,
>
> I am porting arm-based platform from 2.6.35 to current mainline.
> The compiler is gcc-4.4.0.
>
> While booting, the kernel (mainline) hangs after
> "RPC: Registered tcp NFSv4.1 backchannel transport module "
> with similar configurations to 2.6.35.
> Then I found that there is an infinite loop in hlist_bl_set_first_rcu hangs().
>
> The call sequence is:
>
> _d_rehash @fs/dcache.c
> __d_rehash @fs/dcache.c
> hlist_bl_add_head_rcu @include/linux/rculist_bl.h
> hlist_bl_set_first_rcu @include/linux/rculist_bl.h
>
> Current implementation of hlist_bl_set_first_rcu:
>
> static inline void hlist_bl_set_first_rcu(struct hlist_bl_head *h,
>                                        struct hlist_bl_node *n)
> {
>        LIST_BL_BUG_ON((unsigned long)n & LIST_BL_LOCKMASK);
>        LIST_BL_BUG_ON(!((unsigned long)h->first & LIST_BL_LOCKMASK));
>        rcu_assign_pointer(h->first,
>                (struct hlist_bl_node *)((unsigned long)n | LIST_BL_LOCKMASK));
> }
>
> The infinite loop is caused by:
>        LIST_BL_BUG_ON(!((unsigned long)h->first & LIST_BL_LOCKMASK));
> If I comment out this line, the kernel boots well.
>
> The problem looks like that LIST_BL_LOCKMASK is 0 if CONFIG_SMP=n
> and CONFIG_DEBUG_SPINLOCK=n.
>
> #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
> #define LIST_BL_LOCKMASK        1UL
> #else
> #define LIST_BL_LOCKMASK        0UL
> #endif
>
> Therefore !((unsigned long)h->first & LIST_BL_LOCKMASK) is always true.
>
> I also have CONFIG_DEBUG_LIST=y, so boom! Kernel dies.

Ah, that's a silly error! Thanks for the great debugging work, I'll patch it
up.

Thanks,
Nick
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/