openonload-201205-u1

    8edc22de205ffb4604beb12858da1ec4  openonload-201205-u1.tgz

ChangeLog

@@ -1,4 +1,53 @@
 
+openonload-201205-u1
+--------------------
+
+- Various workarounds for IOMMU-related kernel bugs.
+- Compatibility with RHEL6.3 and linux-3.4.0.
+- Use IP ID of zero with TCP.
+- bug29724: Do not accept unsolicited ARPs that change MAC associated with IP.
+- bug29870: Share iommu domain between VFs in a single Onload stack.
+- Fix previously ignored option EF_TCP_ADV_WIN_SCALE_MAX.
+- bug29996: Do not send packets with IP TTL=0.
+- bug27919: Fix poll() after nonblocking connect on TCP socket.
+- bug27932: Fix RPM dependency bugs.
+- Improve error messages when out of socket buffers.
+- Reduce cost of waking other threads for loopback and pipes.
+- bug30140: Fix bugs in error handling in pipe read/write.
+- bug30140: Fix race with pipes between close() of write-fd and read().
+- Various improvements and minor bug fixes to spin-loops.
+- bug30199: Fix python3 compatibility bug in affinity scripts.
+- bug30214: Return 0 rather than EAGAIN on onload_zc_recv() if called callback.
+- bug23968: Fix warnings from recent gcc versions.
+- bug30218: Fix getsockopt(TCP_INFO) on new socket.
+- bug30219: Do not print scary warning when tcgetpgrp() is called.
+- bug29929: Ensure we always drain OS path in UDP onload_zc_recv().
+- Avoid syscall in ioctl(SIOCATMARK).
+- bug30260: Fix RTNL message length for RTM_GETLINK request.
+- bug30173: Do not wait forever for flush to complete -- it may fail.
+- bug30251: Don't return error with EF_NO_FAIL=0 if socket type not onloadable.
+- bug27042: Reduce delayed-ack time to avoid spurious retransmits.
+- bug26090: Fix send to host with no route and SO_BINDTODEVICE.
+- Fix minor memory leak in cicppl_arp_pkt_tx_queue.
+- bug26089: Fix packets sent via kernel w SO_BINDTODEVICE going to wrong intf.
+- Fix error return code from udp send interrupted by signal.
+- bug30359: Fix error code returned from sigaction and exit_group syscalls.
+- Avoid polling the stack when EF_POLL_ON_DEMAND=0.
+- Improve onload_stackdump's handling of zombie stacks.
+- Add "onload_stackdump -z kill" to terminate stacks lingering after exit.
+- bug30218: Reduce initial RTO to 1s from 3d to match Linux.
+- bug30516: Fix interrupts with scalable packet buffer mode.
+- bug30633: Fix kernel panic on driver reload.
+- License ef_vi samples as BSD so users can cut-and-paste.
+- bug27835: Added extern "C" to ef_vi headers to ease use from C++.
+- bug21349: Keep zwin timer running in FIN-WAIT-1.
+- bug30863: Fix for ARP confirmation when using a gateway.
+- bug30892: Fix inconsistency in locking operations.
+- sfc: Updated sfc net driver to v3_2_1_6099.
+- bug30711: ef_vi: Make SOP/CONT flags available in RX_DISCARD events.
+- bug31026: Fix seg-fault in epoll_ctl(DEL) with EF_UL_EPOLL=0|2.
+
+
 openonload-201205-p4
 --------------------
 

README

@@ -108,6 +108,6 @@ Any questions?
 
 
 ==========
- version: openonload-201205-p4
-revision: 064da35e74f4 (ool201205) openonload-201205-p4
-    date: Thu Jul  5 15:02:49 BST 2012
+ version: openonload-201205-u1
+revision: 71cc77f8f44e (ool201205) oo_32031/openonload-201205-u1
+    date: Wed Jul 18 09:48:13 BST 2012

ReleaseNotes

@@ -1,10 +1,13 @@
 
-OpenOnload 201205 Release Notes
-===============================
+OpenOnload 201205-u1 Release Notes
+==================================
 
- This is a major release of OpenOnload that includes new features and bug
- fixes.  Please see the ChangeLog for the full list of changes.  These
- release notes give further details about the more significant changes.
+ This is an update release of OpenOnload that includes fixes for a number
+ of bugs found during testing.  Please see the ChangeLog for the full list
+ of changes.
+
+ There are no major new features in this release, so the notes below detail
+ features added in the openonload-201205 release.
 
 
 ----------------------------------------
@@ -39,7 +42,7 @@ New configuration options
 
  EF_TCP_RST_DELAYED_CONN
  - Set this option to prevent data being delivered late by TCP due to
-   retransmits.  Useful for trading application where delivering the data
+   retransmits.  Useful for trading applications where delivering the data
    late is worse than not delivering it at all.
 
  EF_DYNAMIC_ACK_THRESH
@@ -124,4 +127,4 @@ Scalable packet buffer mode
 
 
 David Riddoch
-2012/05/02
+2012/07/05

mk/site/ciul.mk

@@ -16,7 +16,7 @@ LINK_CIUL_LIB		:= $(MMakeGenerateLibLink)
 # releases, because the internals of the data structures are exposed.  That
 # means that almost any non-trivial change to ef_vi should cause the MAJOR
 # version number to be incremented.
-lib_maj := 2
+lib_maj := 3
 lib_min := 0
 lib_mic := 0
 CIUL_REALNAME		:= $(MMakeGenerateDllRealname)

scripts/onload_misc/openonload.spec

@@ -38,7 +38,7 @@
 # support older distros) to update this spec to use kernel modules packaging
 # templates.
 
-%define pkgversion 201205-p4
+%define pkgversion 201205-u1
 
 %{!?kernel:  %{expand: %%define kernel %%(uname -r)}}
 %{!?target_cpu:  %{expand: %%define target_cpu %{_host_cpu}}}
@@ -47,7 +47,7 @@
 
 %define have_lsb %( ! which lsb_release > /dev/null 2>&1; echo $? )
 
-%define knownvariants '@(BOOT|PAE|@(big|huge)mem|debug|enterprise|kdump|?(big|large)smp|uml|xen[0U]?(-PAE)|xen|rt|default|big|pae)'
+%define knownvariants '@(BOOT|PAE|@(big|huge)mem|debug|enterprise|kdump|?(big|large)smp|uml|xen[0U]?(-PAE)|xen|rt?(-trace|-vanilla)|default|big|pae|vanilla|trace|timing)'
 %define knownvariants2 '%{knownvariants}'?(_'%{knownvariants}')
 
 # Determine distro
@@ -69,7 +69,7 @@
 
 %define kernel_pkg %( rpm -q --whatprovides /lib/modules/%{kernel} | grep -v source | grep -v base) 
 # form a fake kernel name in known format 2.6.32-0.41-rt
-%define kernel_long %( echo %{kernel_pkg} | sed "s/kernel-\\([a-zA-Z_]*\\)[-\.]\\([2-9].[0-9].*\\)/\\2-\\1/; s/kernel-//")
+%define kernel_long %( echo %{kernel_pkg} | sed "s/kernel-\\([a-zA-Z_-]*\\)[-\.]\\([2-9].[0-9].*\\)/\\2-\\1/; s/kernel-//")
 
 %else
 
@@ -97,6 +97,15 @@
 %define libpcap_devel libpcap-devel
 %endif
 
+# On SLES11 source packages for RT kernels are called kernel-source-rt
+# rather than kernel-source.
+%define rtkernel %( echo %{kernel} | grep -q -- -rt && echo 1 || echo 0 )
+%if 0%{?sles_version} >= 11 && %{rtkernel}
+%define kernelsource  kernel-source-rt
+%else
+%define kernelsource  kernel-source
+%endif
+
 # kmodtool doesn't count 'rt' as a variant.  So I've stolen the bash
 # regexp.  Only faffing with this because rpmbuild BuildRequires doesn't
 # agree that kernel-rt provides 'kernel = blah-rt' !
@@ -128,9 +137,9 @@ Source0		: openonload-%{pkgversion}.tgz
 BuildRoot   	: %{_builddir}/%{name}-root
 AutoReqProv	: no
 %if %{redhat}
-BuildRequires	: gawk gcc sed make bash kernel%{kvariantsuffix_dash} = %{kverrel}, kernel%{kvariantsuffix_dash}-devel = %{kverrel} glibc-common %{libpcap_devel}
+BuildRequires	: gawk gcc sed make bash kernel%{kvariantsuffix_dash} = %{kverrel} kernel%{kvariantsuffix_dash}-devel = %{kverrel} glibc-common %{libpcap_devel}
 %else
-BuildRequires	: gawk gcc sed make bash kernel%{kvariantsuffix_dash} = %{kverrel}, kernel-source = %{kverrel} glibc-devel glibc %{libpcap_devel}
+BuildRequires	: gawk gcc sed make bash kernel%{kvariantsuffix_dash} = %{kverrel} %{kernelsource} = %{kverrel} glibc-devel glibc %{libpcap_devel}
 %ifarch x86_64
 %if %{build32}
 BuildRequires   : glibc-devel-32bit
@@ -160,8 +169,12 @@ Group       	: System Environment/Kernel
 %if %( ! echo %{kvariantsuffix} | grep PAE &> /dev/null; echo $? )
 Requires	: openonload = %{version}-%{release}, kernel-%{target_cpu} = %{kverrel}%{kvariantsuffix}
 %else
+%if %( ! echo %{kvariantsuffix} | grep -E trace\|vanilla &> /dev/null; echo $? )
+Requires	: openonload = %{version}-%{release}, kernel%{kvariantsuffix_dash} = %{kverrel}
+%else
 Requires	: openonload = %{version}-%{release}, kernel%{kvariantsuffix_dash}-%{target_cpu} = %{kverrel}
-%endif       
+%endif
+%endif
 
 %else
 Requires	: openonload = %{version}-%{release}, kernel%{kvariantsuffix_dash} = %{kverrel}
@@ -280,6 +293,9 @@ rm -fR $RPM_BUILD_ROOT
 /usr/libexec/onload/modules
 
 %changelog
+* Mon May 14 2012 Mark Spender <mspender@solarflare.com> 201205
+- Added fix for MRG trace and vanilla kernel variants
+
 * Mon Feb 13 2012 Mark Spender <mspender@solarflare.com> 201202
 - Added depenency fix for MRG rt kernels and RHEL 4 kernel variants  
 

scripts/onload_tool

@@ -126,6 +126,8 @@ do_unload() {
   tryunload sfc_mtd
   tryunload sfc_control
   tryunload sfc
+  grep -q sfc/sfc.ko /lib/modules/$(uname -r)/modules.dep && \
+      log "WARNING: sfc driver may reload due to an in-distro driver"
   return $rc
 }
 

scripts/sfcaffinity.py

@@ -126,20 +126,30 @@ def cmp_irq_name(a, b):
     try:
         a_pre, a_suf = a.rsplit('-', 1)
         b_pre, b_suf = b.rsplit('-', 1)
-        c = cmp(a_pre, b_pre)
+        c = a_pre < b_pre
         if c:
             return c
-        return cmp(int(a_suf), int(b_suf))
+        return int(a_suf) < int(b_suf)
     except:
-        return cmp(a, b)
+        return a < b
+
+
+# Python 3 compat.  They really shouldn't have killed cmp.
+def sort_irq_names(irq_names):
+    import sys
+    if sys.version_info >= (3,0):
+        import functools
+        irq_names.sort(key=functools.cmp_to_key(cmp_irq_name))
+    else:
+        irq_names.sort(cmp=cmp_irq_name)
 
 
 def irq_names_matching_pattern(pattern):
     """Return list of interrupt names that match the given pattern.  The
     list is sorted nicely."""
     n2v = irq_get_name2vec_map()
     irq_names = [n for n in n2v.keys() if re.match(pattern, n)]
-    irq_names.sort(cmp=cmp_irq_name)
+    sort_irq_names(irq_names)
     return irq_names
 
 
@@ -252,12 +262,12 @@ def __repr__(self):
 
 class Topology(object):
     def get_cores(self):
-        return self.id2core.values()
+        return list(self.id2core.values())
     def get_packages(self):
-        return self.id2package.values()
+        return list(self.id2package.values())
     def get_cache_levels(self):
         if self.shared_caches:
-            return self.shared_caches.keys()
+            return list(self.shared_caches.keys())
         else:
             return []
     def get_top_level_shared_caches(self):
@@ -337,14 +347,14 @@ def __build_topology(all_cores):
         cs = core_id_to_core_siblings[core_id] & cores_mask
         top.id2core[core_id] = Core(top, core_id, pkg, ts, cs)
 
-    top.core_ids = top.id2core.keys()
+    top.core_ids = list(top.id2core.keys())
     top.core_ids.sort()
     top.cores_mask = mask.to_int(top.core_ids)
-    top.package_ids = top.id2package.keys()
+    top.package_ids = list(top.id2package.keys())
     top.package_ids.sort()
     top.real_cores = dict([(c.id, c) for c in top.id2core.values()
                            if c.id == c.real_core])
-    top.real_core_ids = top.real_cores.keys()
+    top.real_core_ids = list(top.real_cores.keys())
     top.real_core_ids.sort()
     top.real_cores_mask = mask.to_int(top.real_core_ids)
 
@@ -449,7 +459,7 @@ def dump_irqs():
         outl("%20s: %d" % (irq_name, vector))
     outl("irq_get_vec2names_map():")
     v2n = irq_get_vec2names_map()
-    vectors = v2n.keys()
+    vectors = list(v2n.keys())
     vectors.sort()
     for v in vectors:
         outl("%20d: %s" % (v, v2n[v]))

scripts/sfcaffinity_config

@@ -81,6 +81,16 @@ LL_DEBUG = 10
 
 ######################################################################
 
+def read_file(fname):
+    "Read a small ascii file with a single read() call."
+    # Looks like overkill?  Reason for this obfuscation is to get something
+    # that works on python 2.x and 3.x.
+    fd = os.open(fname, os.O_RDONLY)
+    bytes = os.read(fd, 8192)
+    return str(bytes.decode('ascii'))
+
+######################################################################
+
 class Bucket(object):
     pass
 
@@ -211,19 +221,18 @@ def count_bits(mask):
 
 
 def file_get_int(fname):
-    f = open(fname)
-    return int(f.read())
+    f = read_file(fname)
+    return int(f.strip())
 
 
 def file_get_int_list(fname):
-    f = open(fname)
-    l = f.read().strip()
+    l = read_file(fname).strip()
     return [int(x) for x in l.split()]
 
 
 def file_get_cpumask(fname):
-    cscm = open(fname).read()
-    return mask.comma_sep_hex_to_int(cscm.strip())
+    cscm = read_file(fname).strip()
+    return mask.comma_sep_hex_to_int(cscm)
 
 
 def get_topology():
@@ -238,11 +247,13 @@ def get_topology():
             cores = [c.id for c in topology.get_cores()
                      if c.pkg.id in packages]
             topology = affinity.get_topology(cores=cores)
-    except affinity.NoCores, inst:
+    except affinity.NoCores:
         fail("No cores left after applying filters.")
-    except affinity.BadCoreId, inst:
+    except affinity.BadCoreId:
+        inst = sys.exc_info()[1]
         fail("Bad core id %d" % inst.core_id)
-    except mask.BadMask, inst:
+    except mask.BadMask:
+        inst = sys.exc_info()[1]
         fail("Bad mask in --%s=%s" % (opt, inst.msg))
     return topology
 
@@ -299,7 +310,7 @@ proc_interrupts_rxq_fmts = ['\<%s-rx-', '\<%s-']
 def __interface_n_rxqs(ethname, fmt):
     match = fmt % ethname
     f = do_popen("grep '\<%s[0-9]' /proc/interrupts | wc -l" % match)
-    return int(f.read())
+    return int(f.read().strip())
 
 def interface_n_rxqs(ethname):
     for fmt in proc_interrupts_rxq_fmts:
@@ -313,7 +324,7 @@ def sfc_interface_n_rxqs(ethname):
         dir = "/proc/driver/sfc/nic_%s" % ethname
         cmd = "cd %s 2>/dev/null && /bin/ls -d rxq[0-9]* | wc -w" % dir
         f = do_popen(cmd)
-        return int(f.read())
+        return int(f.read().strip())
     except:
         warn(["%s is missing." % dir,
               "Is debugfs mounted?"])
@@ -324,7 +335,7 @@ def num_cpus():
     # Or alternatively get cpu topology at start of day and use
     # that info here.
     f = do_popen("grep 'core id' /proc/cpuinfo | wc -l")
-    return int(f.read())
+    return int(f.read().strip())
 
 
 def module_loaded(modname):
@@ -423,7 +434,7 @@ def rxq_get_irq(ethx, rxq):
     assert n_rxqs > 0
     match = fmt % ethx + str(rxq) + '\>'
     f = do_popen("grep '%s' /proc/interrupts | awk -F: '{print $1}'" % match)
-    return int(f.read())
+    return int(f.read().strip())
 
 
 def rxq_set_affinity(ethx, rxq, cpumask):

scripts/sfcirqaffinity

@@ -32,7 +32,7 @@ def out(x, file=sys.stdout):
     try:
         file.write(x)
     except:
-        print repr(x)
+        sys.stderr.write("ERROR: repr(x) = %s\n" % repr(x))
         raise
 
 
@@ -105,11 +105,13 @@ def do_irqpattern_corespec(irq_name_pattern, cores_args):
             cores = [c.id for c in topology.get_cores()
                      if c.pkg.id in packages]
             topology = affinity.get_topology(cores=cores)
-    except affinity.NoCores, inst:
+    except affinity.NoCores:
         fail("No cores left after applying filters.")
-    except affinity.BadCoreId, inst:
+    except affinity.BadCoreId:
+        inst = sys.exc_info()[1]
         fail("Bad core id %d" % inst.core_id)
-    except mask.BadMask, inst:
+    except mask.BadMask:
+        inst = sys.exc_info()[1]
         fail("Bad mask in --%s=%s" % (opt, inst.msg))
 
     caches = topology.get_top_level_shared_caches()