Partial implementation of the initialization procedure.

Complete debug messages have become a little overwhelming now that we
are allocating thousands of pages of memory for the adapter. Just for
the moment disabling the hexdumps is the more sensible default.

More fine-grained debug logging is likely needed.

Refactored the error checking to always be done when posting a command
to the command queue. Previously this was a manual step for each command
and that seems more error prone.

Now successfully:

- Providing boot memory to the adapter (6 pages)
- Querying adapter capabilities (current and maximum)
- Setting adapter capabilities (keep current)
- Providing init memory to the adapter (4232 pages !)

The output from the init sequence looks like this:

    TRACE   Read the initialization segment
    TRACE   Write the physical location of the command queues to the init segment.
    TRACE   Wait for the 'initializing' field to clear
    fw_rev                          14      12      1220
    cmd_interface_rev               5
    cmdq_phy_addr                   cdata<void *>: 0x1f000000
    log_cmdq_size                   5
    log_cmdq_stride                 6
    ready                           true
    nic_interface_supported         true
    internal_timer                  2.0108995831647e+14
    health_syndrome                 0
    Command: ENABLE_HCA
    Command: QUERY_ISSI
      cur_issi            =         0
      sup_issi            =
         01
    Command: QUERY_PAGES
    query_pages'boot'               6
    Command: MANAGE_PAGES
    Command: QUERY_HCA_CAP
    Command: QUERY_HCA_CAP
    Capabilities - current and (maximum):
      eth_net_offloads         = 0   (0)
      end_pad                  = 1   (1)
      cq_eq_remap              = 1   (1)
      device_frequency_mhz     = 275 (275)
      log_max_vlan_list        = 12  (12)
      log_min_stride_sz_rq     = 0   (0)
      log_max_klm_list_size    = 16  (16)
      log_max_rqt              = 0   (0)
      log_max_l2_table         = 16  (16)
      log_max_current_uc_list  = 10  (10)
      log_min_stride_sz_sq     = 0   (0)
      log_uar_page_sz          = 0   (8)
      log_max_wq_sz            = 0   (0)
      log_max_current_mc_list  = 14  (14)
      log_max_msg              = 30  (30)
      log_max_stride_sz_rq     = 0   (0)
      max_flow_counter         = 0   (0)
      log_max_eq_sz            = 22  (22)
      log_max_rqt_size         = 0   (0)
      basic_cyclic_rcv_wqe     = 0   (0)
      cache_line_128byte       = 0   (0)
      max_tc                   = 0   (0)
      cmdif_checksum           = 0   (3)
      driver_version           = 0   (0)
      log_max_tis              = 0   (0)
      port_type                = 1   (1)
      wq_signature             = 1   (1)
      log_max_tir              = 0   (0)
      max_indirection          = 4   (4)
      log_max_rq               = 0   (0)
      cq_resize                = 1   (1)
      cq_oi                    = 1   (1)
      cq_moderation            = 1   (1)
      log_max_pd               = 24  (24)
      log_max_mkey             = 24  (24)
      log_max_transport_domain = 0   (0)
      rc                       = 1   (1)
      num_ports                = 1   (1)
      bf                       = 1   (1)
      vport_counters           = 1   (1)
      log_max_eq               = 8   (8)
      pad_tx_eth_packet        = 0   (0)
      log_pg_sz                = 12  (12)
      uar_sz                   = 5   (5)
      cq_period_start_from_cqe = 1   (1)
      uc                       = 1   (1)
      log_max_mrw_sz           = 64  (64)
      log_max_cq               = 24  (24)
      vport_group_manager      = 1   (1)
      log_max_tis_per_sq       = 0   (0)
      start_pad                = 0   (0)
      log_max_cq_sz            = 22  (22)
      nic_flow_table           = 0   (0)
      scqe_break_moderation    = 1   (1)
      ud                       = 1   (1)
      log_max_sq               = 0   (0)
      cqe_version              = 0   (0)
      log_bf_reg_size          = 9   (9)
      sctr_data_cqe            = 1   (1)
      log_max_rmp              = 0   (0)
      cqe_version              = 0   (0)
      log_bf_reg_size          = 9   (9)
      sctr_data_cqe            = 1   (1)
      log_max_rmp              = 0   (0)
      log_max_stride_sz_sq     = 0   (0)
      imaicl                   = 0   (0)
      xrc                      = 1   (1)
    Command: SET_HCA_CAP
    Command: QUERY_PAGES
    query_pages'init'               4232
    Command: MANAGE_PAGES
    Command: INIT_HCA

Pulls in more of the initialization procedure. See especially commit
4575dc7.

Added QUERY_VPORT_STATE, MODIFY_VPORT_STATE, QUERY_NIC_VPORT_CONTEXT.

Note: I am not sure that these commands are actually needed since we are
not using SR-IOV. The PRM mandates using some VPORT commands but I don't
see them in the trace from the Linux mlx5 driver. So we may be able to
remove this code.

Completed Mellanox initialization sequence

Required argument for new code merged from master in v2016.06.

Request exclusive lock on the device.

This commit introduces a clean and working version of the device
initialization.

The Mellanox driver is now split into two apps: 'ConnectX4' to setup the
NIC and 'IO' to attach to a queue pair.

Previously a single IO app instances could be supported with multiple
send and receive queues. The intention was to share traffic amongst
the queues in order to overcome NIC hardware limitations on per-queue
packet rate.

The problem with that approach is that splitting traffic across queues
potentially changes the relative ordering of packets and this
potentially creates problems for the user. Better to avoid "DWIM" and
have the user arrange multiple IO apps in an application-appropriate
way if indeed they do need to overcome the ~16Mpps per-queue limit on
the ConnectX-4 (100G).

Now also explicitly sets promiscuous mode on the ethernet port.

The selftest function, while still very simplistic, now presents its
output more neatly and checks that counter values match expectations.
The test now uses unicast traffic (need add additional tests to check
multicast and broadcast and ...).

Example output:

    selftest: waiting for both links up
    Links up. Sending 10,000,000 packets.

    hardware counter               02:00.0               82:00.0
    ----------------  --------------------  --------------------
    rx_bcast_octets                      0                     0
    rx_bcast_packets                     0                     0
    rx_error_octets                      0                     0
    rx_error_packets                     0                     0
    rx_mcast_octets                      0                     0
    rx_mcast_packets                     0                     0
    rx_ucast_octets          1,000,000,000         1,000,000,000
    rx_ucast_packets            10,000,000            10,000,000
    tx_bcast_octets                      0                     0
    tx_bcast_packets                     0                     0
    tx_error_octets                      0                     0
    tx_error_packets                     0                     0
    tx_mcast_octets                      0                     0
    tx_mcast_packets                     0                     0
    tx_ucast_octets          1,000,000,000         1,000,000,000
    tx_ucast_packets            10,000,000            10,000,000
    selftest: ok

Enabled "overflow ignore" on the receive completion queue so that we do
not need to acknowledge completion entries. This should be safe because
the completion queue and the receive queue are the same size i.e. each
receive queue entry will have a separate completion entry with no
collisions.

Fix the opcode for DISABLE_HCA (currently unused), add assert to
alloc_pages() method.

Add the same set of counters supported by the Intel driver.  Add a
pull() method to the ConnextX4 object to periodically sync the stats
so the "control" app can be run as a regular Snabb app.

Polling for command completion in sync_stats() creates too much
latency and causes packet drops.  This is mitigated by creating
separtate HCAs for each command that reads a statistics register and
perfom polling for completion only in sync_stats().

A HCA is tied to an initialization segment (i.e. device). The previous
commit mishandled the case when multiple devices are run in the same
process.  This is fixed by creating a factory for HCAs per device.

Set the page size parameters for CQs and RQs so that the entire queue
fits in a single page.  The alternative would be to select a fixed
page size and add as many PAS entries as necessary.  It is unclear
whether there is a difference between the methods (e.g. for
perfomance).

e587f8c55 Merge pull request #225 from vavrusa/master
5ea3a881e bpf: add missing constants for linux 4.10 - 4.15
2c691e5a7 Merge pull request #224 from wingo/lseek-syscall-tweak
ae38bdbd7 Make "offset" arg to lseek a signed integer
5cb3b6950 Merge pull request #221 from wingo/util-ls-fix
8e0874609 Promptly close util.ls() dir fd; fix bug with deleted entries
277517436 Merge pull request #220 from sileht/master
8e48fd094 linux/nl.lua: Use ndmsg struct instead of rtmsg for neighbors
3e482bc4e Merge pull request #215 from qsn/gettid
57520cce3 expose gettid
db1a88e94 Merge pull request #214 from jsitnicki/sof-flags-linux
270a6e611 Add missing type for struct scm_timestamping for Linux
e49232047 Add missing SCM_* constants for Linux timestamping API
60fcc6b48 Add constants for SOF_* flags for Linux
26ac34851 Merge pull request #210 from alexandergall/linux-if-ioctls
3e6d3e27c Merge pull request #211 from fperrad/deb
d425a22b2 dummy changelog
b27eca538 update .gitignore
92292aa4b debian files
be257a7e1 debian files generated by lua-create-gitbuildpackage-layout
50a02b94b Add some SIOC{G,S}IF* ioctls for Linux
ee90324 Merge pull request #209 from justincormack/osx_clock
ee17863 Add CLOCK_ constants for OSX
178d244 Merge pull request #208 from justincormack/holes
61450f5 Add SEEK_DATA and SEEK_HOLE constants
9a7b584 Add memfd fnctl sealing support
99beaf5 more test fixes for memfd
cc221e4 fix ctest for new fcntl changes, typowq
56c4c76 fix ctest for new fcntl changes
96073cc fix typo
7a73e8a Add more constants for fcntl, memfd
8d3034c Fix ppc64le syscall numbers for newer calls
62828f6 Merge pull request #206 from johnae/master
c4002b6 The spook project is using ljsyscall.
0b266e8 Merge pull request #205 from lukego/close-fd-safely
ad91aa9 Add more protection against fd double-close
c8baf9e Merge pull request #202 from justincormack/dockerignore
66843c5 Use dockerignore to simplify Dockerfile
7b7211d Merge pull request #200 from justincormack/redo-dockerfile
a779caa Docker Cloud does not start processes at priority 0, remove from test
b85382d Rework Dockerfile and tests
24f7789 Merge pull request #199 from vavrusa/master
2ecf486 linux/constants: added new BPF map and prog types
00c1949 use Alpine 3.4 for docker build
0bcafc6 Merge pull request #196 from kbara/removecunused
a4217a8 Merge pull request #195 from kbara/fixgetcpu
ee67430 Remove unused variables from c.lua
4b2e0b2 Fix getcpu: the cpu argument was incorrectly given the node variable previously
214550a update changelog and rockspec for 0.12 release
21f3fd8 Merge pull request #192 from vavrusa/master
0da437f linux/bpf_prog_load: support custom kernel builds
9981190 fix missing vhangup
f245114 Merge pull request #191 from vavrusa/linux-perf-open
93558c1 linux: added support for tracing/performance counters
0511fb8 Merge pull request #190 from vavrusa/master
1f141ca linux: added new constants (e.g. attach BPF to socket)
36274f3 linux/bpf: added strflag support
4fd3bd6 Merge pull request #189 from vavrusa/master
aaa89cb linux: added support for eBPF
e095295 linux: added new syscall numbers (up to __nr_bpf)
1e079c4 test calls container so just needs file to run
d92625e define a docker compose test
5f14711 Merge pull request #188 from aperezdc/fix-if-nameindex
8915a83 Close socket immediately after error in if_nametoindex()
96c0286 use addons for travis, as learned at fosdem
71241e0 update changelog for unreleased changes
559b499 ignore audit arch constant
5d867d3 new architectures do not have open, will use openat
49d9ff9 test issues with new constants
7cd460e aarch64 audit constants for seccomp
6249e99 more docker examples
dd00af9 rockspec fix
d20033b rockspec for new release
fb17244 update copyright years
9d87597 more Changelog tweaks before release
53856b0 typo
1881526 sometimes winsz is 0, eg if terminal not set up
ab0c08d add error message
a1c207e update Changelog for forthcoming 0.11 release
ce12fb2 addDocker hub to README
602a2b3 fix osx fstatat to use 32 bit stat type, as cannot find how to call to get 64 bit one
8a0a6ad Now have arm machines with working seccomp
d14bd38 Appears that setting maac address on lo often works
6e878a1 remove debug print from test
30e9b5b allow skip on EPERM for adjtimex
bff3e90 Add strace in Docker image for convenient debugging
e67fa31 Use alpine 3.3 for Docker
5148bc3 fix ipv6 tests
185c1a6 more failures with no ipv6
98cc9a2 more fixes for ipv4 only environments for netlink tests
248a935 fix bind errors in environments that do not support ipv6
0eacd64 clean up travis file
5fb71b6 switch to newer Ubuntu in travis
8235724 fix more constant checks not in headers
8dff4ef constants missing on travis
db51b08 update Changelog
7540b04 add new rtnetlink values, so tests work under docker
22604b8 remove test that fails in some environments
d431693 fix waitid test under docker
1595b7d fix swap test under docker
321fdd2 Now an alpine package available
984b533 Add Dockerfile, fix some of the tests that made unreasonable assumptions
9aeff88 recent osx has *at functions
18cd829 better handling for xattr errors
b6bb892 freebsd 11 now has utimensat
7065b0d on freebsd/zfs chflags will fail, skip

git-subtree-dir: lib/ljsyscall
git-subtree-split: e587f8c55aad3955dddab3a4fa6c1968037b5c6e

(random_v4_packets generated wrong sized packets)

This updates the AVF driver to use the new PCI resource acquisition API from
#1436

Fixes the currently broken :new() routine. The breakage happened when the avf
branch was integrated with next, and I failed to test the merge. ^^'

Free PCI resources and remaining packets on rx/tx queues on shutdown. This
fixes a resource leak.

This commit uses BPF object pinning to expose per-interface xskmaps to other
Snabb processes so they can attach to free queues on interfaces initialized
by other Snabb processes.

We reset all hotcounts every second. This is a rough way to establish a
relation with elapsed time so that hotcounts provide a measure of frequency.

The concrete goal is to ensure that the JIT will trace code that becomes hot
over a short duration, but not code that becomes hot over, say, the course
of an hour.

Improve stability of ingress drop monitor and reduce long-tail trace generation

This change makes it so that statistics are correctly core-specific.

Fix snabb top interface view for multiple queues

Add a --profile option to control whether vmprofile is enabled or not.

Disable vmprofile on lwAFTR by default

raptorjit: amend prev. commit, also clear trace exit hotcounts

Add 2019.06.02 changelog entry

Fix side exit count resetting to not reset SNAPCOUNT_DONE

This fixes a bug where packets would not be transmitted unless the tx queue was
full.

This fixes a bug in which data_printer_from_grammar would not print choice case
members when an order is given to body_printer.

XXX: it is not clear to me why some invocations of body_printer take orderings,
since it seems that the pre-ordering always matches the internal ordering logic
of body_printer. \o/

Used to be that configurations read from the command line or stdin bypassed the
consistency checker.

This updates the consistency checkers to expand choices and visit the bodies of
all cases.

This removes the order parameter in the affected functions altogether. Test
suite passes so I am assuming this was dead code from the past.

Default to x86_64 registers, but allow passing of alternative register set
specifications.

GCC 8 added a warning that affects a legitimate use case.

Also add identification of ConnectX type to the driver for
future version-dependent code.

These counters capture the drops due to ingress buffer overflow.  No
other per-queue stats are currently implemented. The counters are
aggregated to a single ingress drop counter.

This fixes a bug in our scalar x86 IP checksum implementation, where some
carries where missed. Really hard to find with random testing, added a
systematic test case for carry propagation.

IO:new() may be called before ConnectX4:new(). The device is unbound
by whichever method runs first (IO:new() waits for ConnectX4:new() to
set up the CXQ in any case).