"warning: context imbalance" in kernel/bpf/hashtab.c

[Song Liu <songliubraving@xxxxxx>]

Hi, 

We are trying to fix some sparse warning in kernel/bpf/hashtab.c (in bpf-next tree). 
The sparse I use was v0.6.3-76-gf680124b. 

These new warnings are introduced by [1]. Before [1], hashtab.c got

[...]
kernel/bpf/hashtab.c:2089:19: error: subtraction of functions? Share your drugs
kernel/bpf/hashtab.c:1421:9: warning: context imbalance in '__htab_map_lookup_and_delete_batch' - different lock contexts for basic block
kernel/bpf/hashtab.c: note: in included file (through include/linux/workqueue.h, include/linux/bpf.h):
./include/linux/rcupdate.h:693:9: warning: context imbalance in 'bpf_hash_map_seq_find_next' - unexpected unlock
./include/linux/rcupdate.h:693:9: warning: context imbalance in 'bpf_hash_map_seq_stop' - unexpected unlock

With [1], we got a few more warnings:

[...]
kernel/bpf/hashtab.c:2141:19: error: subtraction of functions? Share your drugs
kernel/bpf/hashtab.c:1292:27: warning: context imbalance in 'htab_map_delete_elem' - unexpected unlock
kernel/bpf/hashtab.c:1325:27: warning: context imbalance in 'htab_lru_map_delete_elem' - unexpected unlock
kernel/bpf/hashtab.c: note: in included file (through include/linux/workqueue.h, include/linux/bpf.h):
./include/linux/rcupdate.h:693:9: warning: context imbalance in '__htab_map_lookup_and_delete_batch' - unexpected unlock
./include/linux/rcupdate.h:693:9: warning: context imbalance in 'bpf_hash_map_seq_find_next' - unexpected unlock
./include/linux/rcupdate.h:693:9: warning: context imbalance in 'bpf_hash_map_seq_stop' - unexpected unlock

So htab_map_delete_elem() and htab_lru_map_delete_elem() got new warnings, and 
__htab_map_lookup_and_delete_batch() got a slightly different warning. 

After trying different annotations, including the attached foo.diff by Daniel,
we found the simplest fix was something like:

====================== 8< ========================== 
diff --git i/kernel/bpf/hashtab.c w/kernel/bpf/hashtab.c
index 23f73d4649c9c..faad2061f1167 100644
--- i/kernel/bpf/hashtab.c
+++ w/kernel/bpf/hashtab.c
@@ -1279,6 +1279,7 @@ static int htab_map_delete_elem(struct bpf_map *map, void *key)
        ret = htab_lock_bucket(htab, b, hash, &flags);
        if (ret)
                return ret;
+       ret = 0;

        l = lookup_elem_raw(head, hash, key, key_size);

@@ -1314,6 +1315,7 @@ static int htab_lru_map_delete_elem(struct bpf_map *map, void *key)
        ret = htab_lock_bucket(htab, b, hash, &flags);
        if (ret)
                return ret;
+       ret = 0;

        l = lookup_elem_raw(head, hash, key, key_size);

@@ -1476,6 +1478,7 @@ __htab_map_lookup_and_delete_batch(struct bpf_map *map,
                ret = htab_lock_bucket(htab, b, batch, &flags);
                if (ret)
                        goto next_batch;
+               ret = 0;
        }

        bucket_cnt = 0;

====================== 8< ========================== 

These three "ret = 0;" make the sparse warning back to before [1]. However, these
don't really make sense, as we know ret must be zero after the if clause. So it
looks like a bug in sparse. Please help us look into this issue. 

Also, there are a few similar warnings from hashtab.c. Please give us guidance on 
how to fix them properly. 

Thanks in advance,
Song

[1] commit 20b6cc34ea74 ("bpf: Avoid hashtab deadlock with map_locked")

Attachment: foo.diff
Description: foo.diff