| Message ID | 1402608702-20910-1-git-send-email-steve.mcintyre@linaro.org |
|---|---|
| State | New |
| Headers | show |
what about recvmmsg? On 06/12/2014 02:31 PM, steve.mcintyre@linaro.org wrote: > From: Steve McIntyre <steve.mcintyre@linaro.org> > > sendmmsg() was new in glibc version 2.14. Some platforms may still be > using older versions of glibc; for those platforms, fall back to > multiple sendmsg() calls instead. It may be slightly slower, but will > at least work. > > Signed-off-by: Steve McIntyre <steve.mcintyre@linaro.org> > --- > platform/linux-generic/source/odp_packet_socket.c | 8 ++++++++ > 1 file changed, 8 insertions(+) > > diff --git a/platform/linux-generic/source/odp_packet_socket.c b/platform/linux-generic/source/odp_packet_socket.c > index 8809abc..e1db61f 100644 > --- a/platform/linux-generic/source/odp_packet_socket.c > +++ b/platform/linux-generic/source/odp_packet_socket.c > @@ -361,11 +361,19 @@ int send_pkt_sock_mmsg(pkt_sock_t * const pkt_sock, > } > > flags = MSG_DONTWAIT; > +#if __GLIBC_PREREQ (2,14) > for (i = 0; i < len; i += sent_msgs) { > ret = sendmmsg(sockfd, &msgvec[i], len - i, flags); > sent_msgs = ret > 0 ? (unsigned)ret : 0; > flags = 0; /* blocking for next rounds */ > } > +#else > + for (i = 0; i < len; i += sent_msgs) { > + ret = sendmsg(sockfd, (const struct msghdr *)&msgvec[i], flags); > + sent_msgs = ret > 0 ? 1 : 0; > + flags = 0; /* blocking for next rounds */ > + } ^-- git am warns on white spaces here. Why ifdef is for hole cycle? Not for only 1 line? Best regards, Maxim. > +#endif > > for (i = 0; i < len; i++) > odp_packet_free(pkt_table[i]);
On Thu, Jun 12, 2014 at 02:56:23PM -0700, Maxim Uvarov wrote:
>what about recvmmsg?
recvmmsg has been around much longer - since glibc 2.12. That's way
older than any systems I have. :-)
Cheers,
Any interest in this still, or drop it? On Thu, Jun 12, 2014 at 10:31:42PM +0100, steve.mcintyre@linaro.org wrote: >From: Steve McIntyre <steve.mcintyre@linaro.org> > >sendmmsg() was new in glibc version 2.14. Some platforms may still be >using older versions of glibc; for those platforms, fall back to >multiple sendmsg() calls instead. It may be slightly slower, but will >at least work. > >Signed-off-by: Steve McIntyre <steve.mcintyre@linaro.org> >--- > platform/linux-generic/source/odp_packet_socket.c | 8 ++++++++ > 1 file changed, 8 insertions(+) > >diff --git a/platform/linux-generic/source/odp_packet_socket.c b/platform/linux-generic/source/odp_packet_socket.c >index 8809abc..e1db61f 100644 >--- a/platform/linux-generic/source/odp_packet_socket.c >+++ b/platform/linux-generic/source/odp_packet_socket.c >@@ -361,11 +361,19 @@ int send_pkt_sock_mmsg(pkt_sock_t * const pkt_sock, > } > > flags = MSG_DONTWAIT; >+#if __GLIBC_PREREQ (2,14) > for (i = 0; i < len; i += sent_msgs) { > ret = sendmmsg(sockfd, &msgvec[i], len - i, flags); > sent_msgs = ret > 0 ? (unsigned)ret : 0; > flags = 0; /* blocking for next rounds */ > } >+#else >+ for (i = 0; i < len; i += sent_msgs) { >+ ret = sendmsg(sockfd, (const struct msghdr *)&msgvec[i], flags); >+ sent_msgs = ret > 0 ? 1 : 0; >+ flags = 0; /* blocking for next rounds */ >+ } >+#endif > > for (i = 0; i < len; i++) > odp_packet_free(pkt_table[i]); >-- >1.7.10.4 > > Cheers,
On 07/10/2014 02:46 AM, Steve McIntyre wrote: > Any interest in this still, or drop it? I would like to apply this in case if somebody else use old glibc. Do you think that changing it according to my notes make sense? Best regards, Maxim. > > On Thu, Jun 12, 2014 at 10:31:42PM +0100, steve.mcintyre@linaro.org wrote: >> From: Steve McIntyre <steve.mcintyre@linaro.org> >> >> sendmmsg() was new in glibc version 2.14. Some platforms may still be >> using older versions of glibc; for those platforms, fall back to >> multiple sendmsg() calls instead. It may be slightly slower, but will >> at least work. >> >> Signed-off-by: Steve McIntyre <steve.mcintyre@linaro.org> >> --- >> platform/linux-generic/source/odp_packet_socket.c | 8 ++++++++ >> 1 file changed, 8 insertions(+) >> >> diff --git a/platform/linux-generic/source/odp_packet_socket.c b/platform/linux-generic/source/odp_packet_socket.c >> index 8809abc..e1db61f 100644 >> --- a/platform/linux-generic/source/odp_packet_socket.c >> +++ b/platform/linux-generic/source/odp_packet_socket.c >> @@ -361,11 +361,19 @@ int send_pkt_sock_mmsg(pkt_sock_t * const pkt_sock, >> } >> >> flags = MSG_DONTWAIT; >> +#if __GLIBC_PREREQ (2,14) >> for (i = 0; i < len; i += sent_msgs) { >> ret = sendmmsg(sockfd, &msgvec[i], len - i, flags); >> sent_msgs = ret > 0 ? (unsigned)ret : 0; >> flags = 0; /* blocking for next rounds */ >> } >> +#else >> + for (i = 0; i < len; i += sent_msgs) { >> + ret = sendmsg(sockfd, (const struct msghdr *)&msgvec[i], flags); >> + sent_msgs = ret > 0 ? 1 : 0; >> + flags = 0; /* blocking for next rounds */ >> + } >> +#endif >> >> for (i = 0; i < len; i++) >> odp_packet_free(pkt_table[i]); >> -- >> 1.7.10.4 >> >> > Cheers,
diff --git a/platform/linux-generic/source/odp_packet_socket.c b/platform/linux-generic/source/odp_packet_socket.c index 8809abc..e1db61f 100644 --- a/platform/linux-generic/source/odp_packet_socket.c +++ b/platform/linux-generic/source/odp_packet_socket.c @@ -361,11 +361,19 @@ int send_pkt_sock_mmsg(pkt_sock_t * const pkt_sock, } flags = MSG_DONTWAIT; +#if __GLIBC_PREREQ (2,14) for (i = 0; i < len; i += sent_msgs) { ret = sendmmsg(sockfd, &msgvec[i], len - i, flags); sent_msgs = ret > 0 ? (unsigned)ret : 0; flags = 0; /* blocking for next rounds */ } +#else + for (i = 0; i < len; i += sent_msgs) { + ret = sendmsg(sockfd, (const struct msghdr *)&msgvec[i], flags); + sent_msgs = ret > 0 ? 1 : 0; + flags = 0; /* blocking for next rounds */ + } +#endif for (i = 0; i < len; i++) odp_packet_free(pkt_table[i]);