John Baldwin wrote:
On Thursday 02 November 2006 12:45, Andre Oppermann wrote:
andre 2006-11-02 17:45:28 UTC
FreeBSD src repository
Modified files:
sys/kern uipc_socket.c
Log:
Use the improved m_uiotombuf() function instead of home grown sosend_copyin()
to do the userland to kernel copying in sosend_generic() and sosend_dgram().
sosend_copyin() is retained for ZERO_COPY_SOCKETS which are not yet supported
by m_uiotombuf().
Benchmaring shows significant improvements (95% confidence):
66% less cpu (or 2.9 times better) with new sosend vs. old sosend (non-TSO)
65% less cpu (or 2.8 times better) with new sosend vs. old sosend (TSO)
(Sender AMD Opteron 852 (2.6GHz) with em(4) PCI-X-133 interface and receiver
DELL Poweredge SC1425 P-IV Xeon 3.2GHz with em(4) LOM connected back to back
at 1000Base-TX full duplex.)
Sponsored by: TCP/IP Optimization Fundraise 2005
MFC after: 3 month
This breaks writes of 0 bytes (e.g. write(s, NULL, 0)) to sockets because
m_getm2(NULL, 0, ...) returns NULL and thus m_uiotombuf() returns NULL and
sosend_*() now return EFAULT. sosend_copyin() correctly handles this case
since it always allocates at least one mbuf. I'm not sure if m_uiotombuf()
is at fault or if something else is, so I'll let you fix it. This explains
the recent breakage of kcheckpass (KDE screen saver password checker) on
current, and possibly other things as well.
Fix is committed. Thank you for tracking it down.
Also, you've introduced another regression in that if the m_get2m() fails it
should be returning ENOBUFS and not EFAULT to userland. The comments in
sosend_*() about 'EFAULT being the only possible error' are obviously
wrong. :)
We're always calling with M_WAITOK because we're coming from userland and
may sleep forever.
--
Andre
_______________________________________________
[email protected] mailing list
http://lists.freebsd.org/mailman/listinfo/cvs-all
To unsubscribe, send any mail to "[EMAIL PROTECTED]"
