On Fri, 2024-07-05 at 19:05 +0200, Markus Elfring wrote: > > External email : Please do not click links or open attachments until > you have verified the sender or the content. > > In 32bit SMP systems, if the system is stressed on the sys node > > by processes, it may cause blkcg_fill_root_iostats to have a > concurrent > > problem on the seqlock in u64_stats_update, which will cause a > deadlock > > on u64_stats_fetch_begin in blkcg_print_one_stat. > > Would you like to mark any references to functions with parentheses? > When deadlock happens, there are 3 CPUs in do_raw_spin_lock blkcg_print_stat seq_read_iter vfs_read ksys_read another 1 CPU is in u64_stats_fetch_begin blkcg_print_one_stat blkcg_print_stat seq_read_iter vfs_read ksys_read > > > To prevent this problem, add spin_locks. > > Another wording suggestion: > Thus use an additional spin lock. > > > How do you think about to use a summary phrase like “Add a spin lock > for stats update > in blkcg_fill_root_iostats()”? > > I can refine commit message as blk-cgroup: Add a spin lock for stats update In 32bit SMP systems, if multiple CPUs call blkcg_print_stat, it may cause blkcg_fill_root_iostats to have a concurrent problem on the seqlock in u64_stats_update, which will cause a deadlock on u64_stats_fetch_begin in blkcg_print_one_stat. Thus use an additional spin lock. > … > > +++ b/block/blk-cgroup.c > > @@ -1134,9 +1134,15 @@ static void blkcg_fill_root_iostats(void) > > cpu_dkstats->sectors[STAT_DISCARD] << 9; > > } > > > > +#if BITS_PER_LONG == 32 > > +spin_lock_irq(&blkg->q->queue_lock); > > +#endif > > flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); > > blkg_iostat_set(&blkg->iostat.cur, &tmp); > > u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); > > +#if BITS_PER_LONG == 32 > > +spin_unlock_irq(&blkg->q->queue_lock); > > +#endif > … > > Under which circumstances would you become interested to apply a > statement > like “guard(spinlock_irq)(&blkg->q->queue_lock);”? > https://elixir.bootlin.com/linux/v6.10-rc6/source/include/linux/spinlock.h#L567 > > Regards, > Markus The spin lock usage is reference from blkcg_print_stat (v6.1.25). I can try guard(spinlock_irq)(&blkg->q->queue_lock); instead. -- boy.wu
