arehbein has submitted this change. ( https://gerrit.osmocom.org/c/libosmo-netif/+/34445?usp=email )
Change subject: ipa: Don't break strict aliasing rule ......................................................................
ipa: Don't break strict aliasing rule
Somehow gcc doesn't always warn about this rule being broken. We are breaking the strict aliasing rule here and libosmo-netif currently does not make use of the '-fno-strict-aliasing' flag.
It's possible that this has also been causing nondeterministic timestamps in libosmo-netif stream tests every once in a while.
Related: OS#6164, OS#5753 Change-Id: Ibed543cdfcdda8c0256ce7d8818ff96d6d46e9b0 --- M include/osmocom/netif/ipa.h 1 file changed, 30 insertions(+), 1 deletion(-)
Approvals: pespin: Looks good to me, approved laforge: Looks good to me, but someone else must approve Jenkins Builder: Verified
diff --git a/include/osmocom/netif/ipa.h b/include/osmocom/netif/ipa.h index 1923253..7826895 100644 --- a/include/osmocom/netif/ipa.h +++ b/include/osmocom/netif/ipa.h @@ -23,7 +23,19 @@ uint8_t proto_ext; } __attribute__ ((packed));
-#define OSMO_IPA_MSGB_CB(__msg) ((struct osmo_ipa_msgb_cb *)&((__msg)->cb[0])) + +/* We don't just cast to 'struct osmo_ipa_msgb_cb *', because that would + * break the strict aliasing rule. Casting to a reference to a union with + * a compatible struct member seems to be allowed, though, see: + * N1570 Committee Draft — April 12, 2011 ISO/IEC 9899:201x, + * Section 6.5, §7 */ +#define OSMO_IPA_MSGB_CB(__msg) (&((( \ + union { \ + unsigned long cb; \ + struct osmo_ipa_msgb_cb _cb; \ + } \ + *)&((__msg)->cb[0]))->_cb)) + #define osmo_ipa_msgb_cb_proto(__x) OSMO_IPA_MSGB_CB(__x)->proto #define osmo_ipa_msgb_cb_proto_ext(__x) OSMO_IPA_MSGB_CB(__x)->proto_ext
-
arehbein