Re: [PATCH 4/4] rhashtable: fix vmalloc() deadlock in atomic rehash

[Uladzislau Rezki <urezki@gmail.com>]

On Thu, Nov 05, 2026 at 04:51:39PM +0800, Herbert Xu wrote:
> On Wed, Nov 04, 2026 at 12:11:24PM +0100, Uladzislau Rezki wrote:
> >
> > There are users in kernel. For example:
> 
> Thanks for the list.
> 
> > drivers/GPU
> >     /* Allocate the file in vmalloc memory, it's likely to be big */
> >     iter.start = __vmalloc(file_size, GFP_NOWAIT);
> > 
> > AMD-GPU:
> >     kvzalloc(size, GFP_NOWAIT);
> >     kvzalloc_objs(struct amdgpu_coredump_ring, GFP_NOWAIT);
> >     kvzalloc(ring->ring_size, GFP_NOWAIT);
> 
> These appear to be executing out of process context without spin
> locks so they're not really GFP_ATOMIC.
> 
> > Virtio ring-testing:
> >     (alloc_ring(0)), gfpflags_allow_blocking(0) does not allow blocking.
> 
> I didn't really look into this as it's just test code.
> 
> > net/ipv4/udp.c:
> >     err = bpf_iter_udp_realloc_batch(iter, batch_sks, GFP_NOWAIT);
> 
> This looks buggy in exactly the same way as rhashtable.  It's
> called under spin_lock_bh, and the same lock can be taken on
> the softirq path on another CPU.
> 
> So are there any safe users of kvmalloc with GFP_ATOMIC? If not
> I suggest that we change kvmalloc to disallow vmalloc when called
> with GFP_ATOMIC.
> 
kvmalloc() disallow and vmalloc() allow?

The deadlock looks like from below spin_lock():

<snip>
diff --git a/net/bridge/br_fdb.c b/net/bridge/br_fdb.c
index e4570bbed854..97aac4a4a92b 100644
--- a/net/bridge/br_fdb.c
+++ b/net/bridge/br_fdb.c
@@ -1025,7 +1025,13 @@ void br_fdb_update(struct net_bridge *br, struct net_bridge_port *source,
 			}
 		}
 	} else {
-		spin_lock(&br->hash_lock);
+		if (in_serving_softirq()) {
+			if (!spin_trylock(&br->hash_lock))
+				return;
+		} else {
+			spin_lock(&br->hash_lock);
+		}
+
 		fdb = fdb_create(br, source, addr, vid, flags);
 		if (fdb) {
 			trace_br_fdb_update(br, source, addr, vid, flags);
<snip>

is that applicable for bridge update code?

--
Uladzislau Rezki