nfp: bpf: improve handling for disabled BPF syscall
authorArnd Bergmann <arnd@arndb.de>
Fri, 23 Sep 2016 20:23:59 +0000 (22:23 +0200)
committerDavid S. Miller <davem@davemloft.net>
Mon, 26 Sep 2016 06:26:29 +0000 (02:26 -0400)
I stumbled over a new warning during randconfig testing,
with CONFIG_BPF_SYSCALL disabled:

drivers/net/ethernet/netronome/nfp/nfp_net_offload.c: In function 'nfp_net_bpf_offload':
drivers/net/ethernet/netronome/nfp/nfp_net_offload.c:263:3: error: '*((void *)&res+4)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
drivers/net/ethernet/netronome/nfp/nfp_net_offload.c:263:3: error: 'res.n_instr' may be used uninitialized in this function [-Werror=maybe-uninitialized]

As far as I can tell, this is a false positive caused by the compiler
getting confused about a function that is partially inlined, but it's
easy to avoid while improving the code:

The nfp_bpf_jit() stub helper for that configuration is unusual as it
is defined in a header file but not marked 'static inline'. By moving
the compile-time check into the caller using the IS_ENABLED() macro,
we can remove that stub and simplify the nfp_net_bpf_offload_prepare()
function enough to unconfuse the compiler.

Fixes: 7533fdc0f77f ("nfp: bpf: add hardware bpf offload")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/netronome/nfp/nfp_bpf.h
drivers/net/ethernet/netronome/nfp/nfp_net_offload.c

index fc220cd..87aa8a3 100644 (file)
@@ -192,20 +192,10 @@ struct nfp_bpf_result {
        bool dense_mode;
 };
 
-#ifdef CONFIG_BPF_SYSCALL
 int
 nfp_bpf_jit(struct bpf_prog *filter, void *prog, enum nfp_bpf_action_type act,
            unsigned int prog_start, unsigned int prog_done,
            unsigned int prog_sz, struct nfp_bpf_result *res);
-#else
-int
-nfp_bpf_jit(struct bpf_prog *filter, void *prog, enum nfp_bpf_action_type act,
-           unsigned int prog_start, unsigned int prog_done,
-           unsigned int prog_sz, struct nfp_bpf_result *res)
-{
-       return -ENOTSUPP;
-}
-#endif
 
 int nfp_prog_verify(struct nfp_prog *nfp_prog, struct bpf_prog *prog);
 
index 43f42f8..8acfb63 100644 (file)
@@ -148,6 +148,9 @@ nfp_net_bpf_offload_prepare(struct nfp_net *nn,
        unsigned int max_mtu;
        int ret;
 
+       if (!IS_ENABLED(CONFIG_BPF_SYSCALL))
+               return -ENOTSUPP;
+
        ret = nfp_net_bpf_get_act(nn, cls_bpf);
        if (ret < 0)
                return ret;