On 09/18/2018 11:52 AM, kernel test robot wrote:
>
> [ 3.265372] BUG: sleeping function called from invalid context at
> mm/util.c:449
> [ 3.288552] in_atomic(): 0, irqs_disabled(): 0, pid: 142, name:
> rhashtable_thra
> [ 3.301548] INFO: lockdep is turned off.
> [ 3.302214] Preemption disabled at:
> [ 3.302221] [<c163e86f>] get_random_u32+0x4f/0x100
> [ 3.327556] CPU: 0 PID: 142 Comm: rhashtable_thra Tainted: G W
> T 4.19.0-rc3-00266-ga79ed8bf #656
> [ 3.328540] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
> 1.10.2-1 04/01/2014
> [ 3.328540] Call Trace:
> [ 3.328540] ? dump_stack+0x55/0x7b
> [ 3.328540] ? get_random_u32+0x4f/0x100
> [ 3.328540] ? ___might_sleep+0x11d/0x170
> [ 3.328540] ? kvfree+0x61/0x70
> [ 3.328540] ? bucket_table_free+0x18/0x80
> [ 3.328540] ? bucket_table_alloc+0x79/0x160
> [ 3.328540] ? rhashtable_insert_slow+0x25d/0x2d0
> [ 3.328540] ? insert_retry+0x1df/0x320
> [ 3.328540] ? threadfunc+0xa3/0x3fe
> [ 3.328540] ? kzalloc+0x14/0x14
> [ 3.328540] ? _raw_spin_unlock_irqrestore+0x30/0x50
> [ 3.328540] ? kthread+0xd1/0x100
> [ 3.328540] ? insert_retry+0x320/0x320
> [ 3.328540] ? kthread_delayed_work_timer_fn+0x80/0x80
> [ 3.328540] ? ret_from_fork+0x2e/0x38
Seems like we need to drop might_sleep_if() from kvfree().
rcu_read_lock()
rhashtable_insert_rehash()
new_tbl = bucket_table_alloc(ht, size, GFP_ATOMIC |
__GFP_NOWARN);
->kvmalloc();
bucket_table_free(new_tbl);
->kvfree()
rcu_read_unlock()
kvmalloc(..., GFP_ATOMIC) simply always kmalloc:
if ((flags & GFP_KERNEL) != GFP_KERNEL)
return kmalloc_node(size, flags, node);
So in the above case, kvfree() always frees kmalloced memory -> and never calls
vfree().
Signed-off-by: Andrey Ryabinin <aryabi...@virtuozzo.com>
---
mm/util.c | 2 --
1 file changed, 2 deletions(-)
diff --git a/mm/util.c b/mm/util.c
index 929ed1795bc1..7f1f165f46af 100644
--- a/mm/util.c
+++ b/mm/util.c
@@ -446,8 +446,6 @@ EXPORT_SYMBOL(kvmalloc_node);
*/
void kvfree(const void *addr)
{
- might_sleep_if(!in_interrupt());
-
if (is_vmalloc_addr(addr))
vfree(addr);
else
--
