Merged Javier's new Intel 82599 driver with VMDq support.

src/apps/intel/intel_app.lua

@@ -13,13 +13,31 @@ local vfio     = require("lib.hardware.vfio")
 local config = require("core.config")
 
 Intel82599 = {}
+Intel82599.__index = Intel82599
+
+-- table pciaddr => {pf, vflist}
+local devices = {}
+
 
 -- Create an Intel82599 App for the device with 'pciaddress'.
-function Intel82599:new (pciaddress)
-   local a = { dev = intel10g.new(pciaddress) }
-   setmetatable(a, {__index = Intel82599 })
-   intel10g.open_for_loopback_test(a.dev)
-   return a
+function Intel82599:new (args)
+   args = config.parse_app_arg(args)
+
+   if args.vmdq then
+      if devices[args.pciaddr] == nil then
+         devices[args.pciaddr] = {pf=intel10g.new_pf(args.pciaddr):open(), vflist={}}
+      end
+      local dev = devices[args.pciaddr]
+      local vf = dev.pf:new_vf(#dev.vflist)
+      dev.vflist[#dev.vflist+1] = vf
+      return setmetatable({dev=vf:open(args)}, Intel82599)
+   else
+      local dev = intel10g.new_sf(args.pciaddr)
+         :open()
+         :autonegotiate_sfi()
+         :wait_linkup()
+      return setmetatable({dev=dev}, Intel82599)
+   end
 end
 
 -- Allocate receive buffers from the given freelist.
@@ -67,34 +85,102 @@ end
 
 -- Report on relevant status and statistics.
 function Intel82599:report ()
-   print("report on intel device", self.dev.pciaddress)
+   print("report on intel device", self.dev.pciaddress or self.dev.pf.pciaddress)
    --register.dump(self.dev.r)
    register.dump(self.dev.s, true)
+   if self.dev.rxstats then
+      for name,v in pairs(self.dev:get_rxstats()) do
+         io.write(string.format('%30s: %d\n', 'rx '..name, v))
+      end
+   end
+   if self.dev.txstats then
+      for name,v in pairs(self.dev:get_txstats()) do
+         io.write(string.format('%30s: %d\n', 'tx '..name, v))
+      end
+   end
+   register.dump({
+      self.dev.r.TDH, self.dev.r.TDT,
+      self.dev.r.RDH, self.dev.r.RDT,
+      self.dev.r.AUTOC or self.dev.pf.r.AUTOC,
+      self.dev.r.AUTOC2 or self.dev.pf.r.AUTOC2,
+      self.dev.r.LINKS or self.dev.pf.r.LINKS,
+      self.dev.r.LINKS2 or self.dev.pf.r.LINKS2,
+   })
 end
 
 function selftest ()
-   -- Create a pieline:
-   --   Source --> Intel82599(loopback) --> Sink
-   -- and push packets through it.
-   vfio.bind_device_to_vfio("0000:01:00.0")
+   buffer.preallocate(100000)
+   sq_sq('0000:05:00.0', '0000:8a:00.0')
+   app.main({duration = 1, report={showlinks=true, showapps=false}})
+
+   mq_sq('0000:05:00.0', '0000:8a:00.0')
+   app.main({duration = 1, report={showlinks=true, showapps=false}})
+end
+
+-- open two singlequeue drivers on both ends of the wire
+function sq_sq(pcidevA, pcidevB)
    local c = config.new()
-   config.app(c, "intel10g", Intel82599, "0000:01:00.0")
-   config.app(c, "source", basic_apps.Source)
-   config.app(c, "sink", basic_apps.Sink)
-   config.link(c, "source.out -> intel10g.rx")
-   config.link(c, "intel10g.tx -> sink.in")
+   config.app(c, 'source1', basic_apps.Source)
+   config.app(c, 'source2', basic_apps.Source)
+   config.app(c, 'nicA', Intel82599, ([[{pciaddr='%s'}]]):format(pcidevA))
+   config.app(c, 'nicB', Intel82599, ([[{pciaddr='%s'}]]):format(pcidevB))
+   config.app(c, 'sink', basic_apps.Sink)
+   config.link(c, 'source1.out -> nicA.rx')
+   config.link(c, 'source2.out -> nicB.rx')
+   config.link(c, 'nicA.tx -> sink.in1')
+   config.link(c, 'nicB.tx -> sink.in2')
    app.configure(c)
---[[
-   app.apps.intel10g = Intel82599:new("0000:01:00.0")
-   app.apps.source = app.new(basic_apps.Source)
-   app.apps.sink   = app.new(basic_apps.Sink)
-   app.connect("source", "out", "intel10g", "rx")
-   app.connect("intel10g", "tx", "sink", "in")
-   app.relink()
---]]
-   buffer.preallocate(100000)
-   app.main({duration = 1})
---   repeat app.breathe() until deadline()
---   app.report()
 end
 
+-- one singlequeue driver and a multiqueue at the other end
+function mq_sq(pcidevA, pcidevB)
+   d1 = lib.hexundump ([[
+      52:54:00:02:02:02 52:54:00:01:01:01 08 00 45 00
+      00 54 c3 cd 40 00 40 01 f3 23 c0 a8 01 66 c0 a8
+      01 01 08 00 57 ea 61 1a 00 06 5c ba 16 53 00 00
+      00 00 04 15 09 00 00 00 00 00 10 11 12 13 14 15
+      16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25
+      26 27 28 29 2a 2b 2c 2d 2e 2f 30 31 32 33 34 35
+      36 37
+   ]], 98)                  -- src: As    dst: Bm0
+   d2 = lib.hexundump ([[
+      52:54:00:03:03:03 52:54:00:01:01:01 08 00 45 00
+      00 54 c3 cd 40 00 40 01 f3 23 c0 a8 01 66 c0 a8
+      01 01 08 00 57 ea 61 1a 00 06 5c ba 16 53 00 00
+      00 00 04 15 09 00 00 00 00 00 10 11 12 13 14 15
+      16 17 18 19 1a 1b 1c 1d 1e 1f 20 21 22 23 24 25
+      26 27 28 29 2a 2b 2c 2d 2e 2f 30 31 32 33 34 35
+      36 37
+   ]], 98)                  -- src: As    dst: Bm1
+   local c = config.new()
+   config.app(c, 'source_ms', basic_apps.Join)
+   config.app(c, 'repeater_ms', basic_apps.Repeater)
+   config.app(c, 'nicAs', Intel82599, ([[{
+   -- Single App on NIC A
+      pciaddr = '%s',
+      macaddr = '52:54:00:01:01:01',
+   }]]):format(pcidevA))
+   config.app(c, 'nicBm0', Intel82599, ([[{
+   -- first VF on NIC B
+      pciaddr = '%s',
+      vmdq = true,
+      macaddr = '52:54:00:02:02:02',
+   }]]):format(pcidevB))
+   config.app(c, 'nicBm1', Intel82599, ([[{
+   -- second VF on NIC B
+      pciaddr = '%s',
+      vmdq = true,
+      macaddr = '52:54:00:03:03:03',
+   }]]):format(pcidevB))
+   print ("Send a bunch of from the SF on NIC A to the VFs on NIC B")
+   print ("half of them go to nicBm0 and nicBm0")
+   config.app(c, 'sink_ms', basic_apps.Sink)
+   config.link(c, 'source_ms.out -> repeater_ms.input')
+   config.link(c, 'repeater_ms.output -> nicAs.rx')
+   config.link(c, 'nicAs.tx -> sink_ms.in1')
+   config.link(c, 'nicBm0.tx -> sink_ms.in2')
+   config.link(c, 'nicBm1.tx -> sink_ms.in3')
+   app.configure(c)
+   link.transmit(app.app_table.source_ms.output.out, packet.from_data(d1))
+   link.transmit(app.app_table.source_ms.output.out, packet.from_data(d2))
+end

src/core/app.lua

@@ -120,7 +120,7 @@ function main (options)
       breathe()
       if not no_timers then timer.run() end
    until done and done()
-   report()
+   report(options.report)
 end
 
 function breathe ()
@@ -147,10 +147,21 @@ function breathe ()
    until not progress  -- Stop after no link had new data
 end
 
-function report ()
-   print("link report")
-   for name, l in pairs(link_table) do
-      print(lib.comma_value(tostring(tonumber(l.stats.txpackets))), "sent on", name)
+function report (options)
+   if not options or options.showlinks then
+      print("link report")
+      for name, l in pairs(link_table) do
+         print(lib.comma_value(tostring(tonumber(l.stats.txpackets))), "sent on", name)
+      end
+   end
+   if options and options.showapps then
+      print ("apps report")
+      for name, a in pairs(app_table) do
+         if a.report then
+            print (name)
+            a:report()
+         end
+      end
    end
 end
 

src/core/lib.lua

@@ -100,6 +100,25 @@ function bitset (value, n)
    return bit.band(value, bit.lshift(1, n)) ~= 0
 end
 
+--- Hex dump and undump functions
+
+function hexdump(s)
+   if #s < 1 then return '' end
+   local frm = ('%02X '):rep(#s-1)..'%02X'
+   return string.format(frm, s:byte(1, #s))
+end
+
+function hexundump(h, n)
+   local buf = ffi.new('char[?]', n)
+   local i = 0
+   for b in h:gmatch('[0-9a-fA-F][0-9a-fA-F]') do
+      buf[i] = tonumber(b, 16)
+      i = i+1
+      if i >= n then break end
+   end
+   return ffi.string(buf, n)
+end
+
 function comma_value(n) -- credit http://richard.warburton.it
    local left,num,right = string.match(n,'^([^%d]*%d)(%d*)(.-)$')
    return left..(num:reverse():gsub('(%d%d%d)','%1,'):reverse())..right
@@ -205,5 +224,31 @@ function selftest ()
 --    assert(readlink('/etc/rc2.d/S99rc.local') == '../init.d/rc.local', "bad readlink")
 --    assert(dirname('/etc/rc2.d/S99rc.local') == '/etc/rc2.d', "wrong dirname")
 --    assert(basename('/etc/rc2.d/S99rc.local') == 'S99rc.local', "wrong basename")
+   assert(hexdump('\x45\x00\xb6\x7d\x00\xFA\x40\x00\x40\x11'):upper()
+         :match('^45.00.B6.7D.00.FA.40.00.40.11$'), "wrong hex dump")
+   assert(hexundump('4500 B67D 00FA400040 11', 10)
+         =='\x45\x00\xb6\x7d\x00\xFA\x40\x00\x40\x11', "wrong hex undump")
+
+   local macA = new_mac('00-01-02-0a-0b-0c')
+   local macB = new_mac('0001020A0B0C')
+   local macC = new_mac('0A:0B:0C:00:01:02')
+   print ('macA', macA)
+   assert (tostring(macA) == '00:01:02:0A:0B:0C', "bad canonical MAC")
+   assert (macA == macB, "macA and macB should be equal")
+   assert (macA ~= macC, "macA and macC should be different")
+   assert (macA:subbits(0,31)==0x0a020100, "low A")
+   assert (macA:subbits(32,48)==0x0c0b, ("hi A (%X)"):format(macA:subbits(32,48)))
+   assert (macC:subbits(0,31)==0x000c0b0a, "low C")
+   assert (macC:subbits(32,48)==0x0201," hi C")
+
+   local ndx_set = new_index_set(4, 'test ndx')
+   assert (string.format('%d/%s', ndx_set:add('a'))=='0/true', "indexes start with 0, and is new")
+   assert (string.format('%d/%s', ndx_set:add('b'))=='1/true', "second new index")
+   assert (string.format('%d/%s', ndx_set:add('c'))=='2/true', "third new")
+   assert (string.format('%d/%s', ndx_set:add('b'))=='1/false', "that's an old one")
+   assert (string.format('%d/%s', ndx_set:add('a'))=='0/false', "the very first one")
+   assert (string.format('%d/%s', ndx_set:add('A'))=='3/true', "almost, but new")
+   assert (string.format('%s/%s', pcall(ndx_set.add, ndx_set,'B'))
+         :match('^false/core/lib.lua:%d+: test ndx overflow'), 'should overflow')
 end
 

src/core/packet.lua

@@ -40,6 +40,24 @@ function add_iovec (p, b, length,  offset)
    p.length = p.length + length
 end
 
+-- fill's an allocated packet with data from a string
+function fill_data (p, d, offset)
+   offset = offset or 0
+   local iovec = p.iovecs[0]
+   assert (offset+#d <= iovec.length, "can't fit on first iovec")       -- TODO: handle more iovecs
+   ffi.copy (iovec.buffer.pointer + iovec.offset + offset, d, #d)
+end
+
+-- creates a packet from a given binary string
+function from_data (d)
+   local p = allocate()
+   local b = buffer.allocate()
+   local size = math.min(#d, b.size)
+   add_iovec(p, b, size)
+   fill_data(p, d)
+   return p
+end
+
 -- Increase the reference count for packet p by n (default n=1).
 function ref (p,  n)
    if p.refcount > 0 then
@@ -78,4 +96,45 @@ function free (p)
    freelist.add(packets_fl, p)
 end
 
+function iovec_dump (iovec)
+   local b = iovec.buffer
+   local l = math.min(iovec.length, b.size-iovec.offset)
+   if l < 1 then return '' end
+   o={[-1]=string.format([[
+         offset: %d
+         length: %d
+         buffer.pointer: %s
+         buffer.physical: %X
+         buffer.size: %d
+      ]], iovec.offset, iovec.length, b.pointer, tonumber(b.physical), b.size)}
+   for i = 0, l-1 do
+      o[i] = bit.tohex(b.pointer[i+iovec.offset], -2)
+   end
+   return table.concat(o, ' ', -1, l-1)
+end
+
+function report (p)
+   print (string.format([[
+         refcount: %d
+         fuel: %d
+         info.flags: %X
+         info.gso_flags: %X
+         info.hdr_len: %d
+         info.gso_size: %d
+         info.csum_start: %d
+         info.csum_offset: %d
+         niovecs: %d
+         length: %d
+      ]],
+      p.refcount, p.fuel, p.info.flags, p.info.gso_flags,
+      p.info.hdr_len, p.info.gso_size, p.info.csum_start,
+      p.info.csum_offset, p.niovecs, p.length
+   ))
+   for i = 0, p.niovecs-1 do
+      print(string.format([[
+            iovec #%d: %s
+         ]], i, iovec_dump(p.iovecs[i])))
+   end
+end
+
 module_init()