On 3/9/21 8:54 AM, Peter Zijlstra wrote:
> On Mon, Mar 08, 2021 at 09:42:08PM +0100, Erhard F. wrote:
>
>> I can confirm that your patch on top of 5.12-rc2 makes the lockdep
>> splat disappear (Ahmeds' 1st patch not installed).
>
> Excellent, I'll queue the below in locking/urgent then.
>
>
> ---
> Subject: u64_stats,lockdep: Fix u64_stats_init() vs lockdep
> From: Peter Zijlstra <pet...@infradead.org>
> Date: Mon, 8 Mar 2021 09:38:12 +0100
>
> Jakub reported that:
>
> static struct net_device *rtl8139_init_board(struct pci_dev *pdev)
> {
> ...
> u64_stats_init(&tp->rx_stats.syncp);
> u64_stats_init(&tp->tx_stats.syncp);
> ...
> }
>
> results in lockdep getting confused between the RX and TX stats lock.
> This is because u64_stats_init() is an inline calling seqcount_init(),
> which is a macro using a static variable to generate a lockdep class.
>
> By wrapping that in an inline, we negate the effect of the macro and
> fold the static key variable, hence the confusion.
>
> Fix by also making u64_stats_init() a macro for the case where it
> matters, leaving the other case an inline for argument validation
> etc.
>
> Reported-by: Jakub Kicinski <k...@kernel.org>
> Debugged-by: "Ahmed S. Darwish" <a.darw...@linutronix.de>
> Signed-off-by: Peter Zijlstra (Intel) <pet...@infradead.org>
> Tested-by: "Erhard F." <erhar...@mailbox.org>
> Link:
> https://lkml.kernel.org/r/yexicy6+9mksd...@hirez.programming.kicks-ass.net
> ---
> include/linux/u64_stats_sync.h | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> --- a/include/linux/u64_stats_sync.h
> +++ b/include/linux/u64_stats_sync.h
> @@ -115,12 +115,13 @@ static inline void u64_stats_inc(u64_sta
> }
> #endif
>
> +#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
> +#define u64_stats_init(syncp) seqcount_init(&(syncp)->seq)
> +#else
> static inline void u64_stats_init(struct u64_stats_sync *syncp)
> {
> -#if BITS_PER_LONG == 32 && defined(CONFIG_SMP)
> - seqcount_init(&syncp->seq);
> -#endif
> }
> +#endif
>
> static inline void u64_stats_update_begin(struct u64_stats_sync *syncp)
> {
>
Interesting !
It seems seqcount_latch_init() might benefit from something similar.
