Re: [PATCH v3] slub: fix false-positive lockdep warning in free_partial()

[Steven Rostedt <rostedt@goodmis.org>]

On Wed, Feb 05, 2014 at 12:15:33PM +0400, Vladimir Davydov wrote:
> Commit c65c1877bd68 ("slub: use lockdep_assert_held") requires
> remove_partial() to be called with n->list_lock held, but free_partial()
> called from kmem_cache_close() on cache destruction does not follow this
> rule, leading to a warning:
> 
>   WARNING: CPU: 0 PID: 2787 at mm/slub.c:1536 __kmem_cache_shutdown+0x1b2/0x1f0()
>   Modules linked in:
>   CPU: 0 PID: 2787 Comm: modprobe Tainted: G        W    3.14.0-rc1-mm1+ #1
>   Hardware name:
>    0000000000000600 ffff88003ae1dde8 ffffffff816d9583 0000000000000600
>    0000000000000000 ffff88003ae1de28 ffffffff8107c107 0000000000000000
>    ffff880037ab2b00 ffff88007c240d30 ffffea0001ee5280 ffffea0001ee52a0
>   Call Trace:
>    [<ffffffff816d9583>] dump_stack+0x51/0x6e
>    [<ffffffff8107c107>] warn_slowpath_common+0x87/0xb0
>    [<ffffffff8107c145>] warn_slowpath_null+0x15/0x20
>    [<ffffffff811c7fe2>] __kmem_cache_shutdown+0x1b2/0x1f0
>    [<ffffffff811908d3>] kmem_cache_destroy+0x43/0xf0
>    [<ffffffffa013a123>] xfs_destroy_zones+0x103/0x110 [xfs]
>    [<ffffffffa0192b54>] exit_xfs_fs+0x38/0x4e4 [xfs]
>    [<ffffffff811036fa>] SyS_delete_module+0x19a/0x1f0
>    [<ffffffff816dfcd8>] ? retint_swapgs+0x13/0x1b
>    [<ffffffff810d2125>] ? trace_hardirqs_on_caller+0x105/0x1d0
>    [<ffffffff81359efe>] ? trace_hardirqs_on_thunk+0x3a/0x3f
>    [<ffffffff816e8539>] system_call_fastpath+0x16/0x1b
> 
> Although this cannot actually result in a race, because on cache
> destruction there should not be any concurrent frees or allocations from
> the cache, let's add spin_lock/unlock to free_partial() just to keep
> lockdep happy.

Really? We are adding a spin lock for a case where it is not needed just to
quiet lockdep?

Now if it really isn't needed, then why don't we do the following instead of
adding the overhead of taking a lock?

static inline
__remove_partial(struct kmem_cache_node *n, struct page *page)
{
	list_del(&page->lru);
	n->nr_partial--;
}

static inline remove_partial(struct kmem_cache_node *n,
			     struct page *page)
{
	lockdep_assert_held(&n->list_lock);
	__remove_partial(n, page);
}

And then just call __remove_partial() where we don't need to check if the
lock is held or not with a big comment to it.

That, IMNSHO, is a much better solution.

-- Steve