Merge next (v2019.11 release) into master

.version

@@ -1 +1 @@
-2019.01
+2019.11

README.md

@@ -142,5 +142,5 @@ Here are the ways you can get involved:
 - Use the Snabb applications in your network.
 - Create your very own application: [Getting Started](src/doc/getting-started.md).
 - Create Github Issues with your ideas and questions and problems.
-- [Join](https://join.slack.com/t/snabb/shared_invite/enQtMzIyOTIwMTg5ODYyLTMzY2FjMGEzM2QzNDlhMDYxNzU0M2UyNjQ1MDc4MDRjY2Q3MWMwY2Q4YWQ1NDllY2E3NTZkZGUyZTQxNzgyNjc) the [Snabb Slack chat](https://snabb.slack.com/) to hang out and shoot the breeze.
+- [Join](https://join.slack.com/t/snabb/shared_invite/enQtMzIyOTIwMTg5ODYyLWUwYzg2MTIwMjAxZTM1N2RlNGFjMWY1YzRkZGZiN2U5ZGU1NTYwNWJiMmQ3MWQ1MDFjYTg0MjNjNTZjMTFlZGQ) the [Snabb Slack chat](https://snabb.slack.com/) to hang out and shoot the breeze.
 

lib/ljsyscall/syscall/linux/syscalls.lua

@@ -483,33 +483,30 @@ local function get_maxnumnodes()
        return math.floor(((#line+1)/9)*32)
      end
    end
+   -- If we don't know, guess that the system has a max of 1024 nodes.
+   return 1024
+end
+
+local function ensure_bitmask(mask, size)
+  if ffi.istype(t.bitmask, mask) then return mask end
+  return t.bitmask(mask, size or get_maxnumnodes())
 end
 
 function S.get_mempolicy(mode, mask, addr, flags)
   mode = mode or t.int1()
-  local size
-  if ffi.istype(t.bitmask, mask) then
-    -- if mask was provided by the caller, then use its size
-    -- and let the syscall error if it's too small
-    size = ffi.cast("uint64_t", tonumber(mask.size))
-  else
-    local mask_for_size = t.bitmask(mask)
-    -- Size should be at least equals to maxnumnodes.
-    size = ffi.cast("uint64_t", math.max(tonumber(mask_for_size.size), get_maxnumnodes()))
-    mask = t.bitmask(mask, tonumber(size))
-  end
-  local ret, err = C.get_mempolicy(mode, mask.mask, size, addr or 0, c.MPOL_FLAG[flags])
+  mask = ensure_bitmask(mask);
+  local ret, err = C.get_mempolicy(mode, mask.mask, mask.size, addr or 0, c.MPOL_FLAG[flags])
   if ret == -1 then return nil, t.error(err or errno()) end
   return { mode=mode[0], mask=mask }
 end
 function S.set_mempolicy(mode, mask)
-  mask = mktype(t.bitmask, mask)
+  mask = ensure_bitmask(mask);
   return retbool(C.set_mempolicy(c.MPOL_MODE[mode], mask.mask, mask.size))
 end
 
 function S.migrate_pages(pid, from, to)
-  from = mktype(t.bitmask, from)
-  to = mktype(t.bitmask, to)
+  from = ensure_bitmask(from);
+  to = ensure_bitmask(to, from.size)
   assert(from.size == to.size, "incompatible nodemask sizes")
   return retbool(C.migrate_pages(pid or 0, from.size, from.mask, to.mask))
 end

src/Makefile

@@ -45,7 +45,7 @@ LUAJIT_A := ../lib/luajit/src/raptorjit.a
 # for each module that has a top-level selftest () function.
 TESTMODS = $(shell find . -regex '[^\#]*\.\(lua\|dasl\)' -printf '%P ' | \
              xargs grep -s -l '^function selftest *[[:punct:]]' | \
-             sed -e 's_\.lua__' -e 's_\.dasl__' -e 's_/_._g')
+             sed -e 's_\.lua__' -e 's_\.dasl__' -e 's_/_._g' -e 's/-/_/g')
 
 # TESTSCRIPTS expands to:
 #   lib/watchdog/selftest.sh ...

src/README.md

@@ -816,9 +816,12 @@ end
 
 Returns hexadecimal string for bytes in *string*.
 
-— Function **lib.hexundump** *hexstring*
+— Function **lib.hexundump** *hexstring*, *n*, *error* 
 
-Returns byte string for *hexstring*.
+Returns string of *n* bytes for *hexstring*. Throws an error if less than *n*
+hex-encoded bytes could be parsed unless *error* is `false`.
+
+*Error* is optional and can be the error message to throw.
 
 — Function **lib.comma_value** *n*
 
@@ -1016,6 +1019,7 @@ Groups of Snabb processes each have the following special properties:
   mastering (DMA) is disabled upon termination before any DMA memory
   is returned to the kernel. This prevents "dangling" DMA requests
   from corrupting memory that has been freed and reused.
+  See [lib.hardware.pci](#pci-lib.hardware.pci) for details.
 
 The `core.worker` API functions are available in the main process only:
 

src/apps/intel/intel10g.lua

@@ -91,9 +91,10 @@ function new_sf (conf)
 end
 
 function M_sf:open ()
+   self.fd = pci.open_pci_resource_locked(self.pciaddress, 0)
    pci.unbind_device_from_linux(self.pciaddress)
    pci.set_bus_master(self.pciaddress, true)
-   self.base, self.fd = pci.map_pci_memory_locked(self.pciaddress, 0)
+   self.base = pci.map_pci_memory(self.fd)
    register.define(config_registers_desc, self.r, self.base)
    register.define(transmit_registers_desc, self.r, self.base)
    register.define(receive_registers_desc, self.r, self.base)
@@ -497,9 +498,10 @@ function new_pf (conf)
 end
 
 function M_pf:open ()
+   self.fd = pci.open_pci_resource_locked(self.pciaddress, 0)
    pci.unbind_device_from_linux(self.pciaddress)
    pci.set_bus_master(self.pciaddress, true)
-   self.base, self.fd = pci.map_pci_memory_locked(self.pciaddress, 0)
+   self.base = pci.map_pci_memory(self.fd)
    register.define(config_registers_desc, self.r, self.base)
    register.define_array(switch_config_registers_desc, self.r, self.base)
    register.define_array(packet_filter_desc, self.r, self.base)

src/apps/intel_avf/README.md

@@ -0,0 +1,44 @@
+# Intel AVF (adaptive virtual function) app (apps.intel_avf.intel_avf)
+
+The `intel_avf.Intel_avf` app provides drivers for the Virtual Functions exported
+by recent generations of Intel network cards.
+
+The links are named `input` and `output`.
+
+    DIAGRAM: Intel_avf
+                 +-----------+
+                 |           |
+      input ---->* Intel_avf *----> output
+                 |           |
+                 +-----------+
+
+## Configuration
+
+— Key **pciaddr**
+
+*Required*. The PCI address of the NIC as a string.
+
+— Key **ring_buffer_size**
+
+*Optional*. Number of DMA descriptors to use i.e. size of the DMA
+transmit and receive queues. Must be a multiple of 128. Default is not
+specified but assumed to be broadly applicable.
+
+## Supported Hardware
+Ethernet controller [0200]: Intel Corporation Ethernet Virtual Function 700 Series [8086:154c] (rev 02)
+
+## Unsupported features
+* Multiple queues per VF. This driver supports a single queue. The spec allows for up to 4 queues.
+* RSS with only 1 queue RSS doesn't make sense.
+* Multiple vlans are unsupported, `ip link` can be used to map all traffic to a single vlan.
+* Multiple MAC addresses are unsupported, `ip link` can be used to set the mac before snabb startup.
+* All of the advanced offload features are unsupported.
+* 16 byte RX descriptors are unsupported.
+
+## Setting up VFs under linux
+echo 2 > /sys/bus/pci/devices/0000\:03\:00.1/sriov_numvfs
+ip link set up dev enp3s0f1
+ip link set enp3s0f1 vf 0 mac 02:00:00:00:00:01
+ip link set enp3s0f1 vf 0 vlan 10
+
+A more complete example can be found in apps/intel_avf/tests/setup.sh