Merge branch 'master' of ozlabs.org:ccan

cpuid: Use __asm for MSVC

MSVC only supports inline assembly and does not support the keyword
volatile for assembly.

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: use a sprintf-like function to get cputype as a string

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: avoid unions in parsing data

This is a much more cleaner way to do it and _should_ be easier for people
to use.

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: apply the 32-bit fix

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

io: add io_conn_fd()

Useful for getsockname().

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

io: io_set_alloc()

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: make union more generic.

I really wanted an array of bytes in there, so make it more flexible.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Merge branch 'master' of ozlabs.org:ccan

Merge branch 'io'

ccan/io: add examples.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: io_connect()

Not a perfect solution (we'd ideally want to go to another plan
immediately, but that would involve a malloc).

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: flatten debug callchain further.

Don't call through io_loop, but have it pass the connection back
and call manually.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: flatten debug callchain.

Don't call from the plan-construction function, call after it returns.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: update and improve documentation.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: save errno on io_close, for finish functions.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: handle errors on poll()

Without this, closing an fd results in a spin...

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: io_close_cb()

Overloading io_close() as a callback is ugly: create an explicit
io_close_cb().

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: simplify I/O callbacks.

Use a -1 for error codes.  This makes it easier to write your own io funcs.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: io_set_finish()

Rather than insisting on supplying them on every call to io_new_conn().
Also, this way it can be changed on a connection.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: test custom io functions.

And rename debug_io_plan() to io_plan_debug() so it can be exposed.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: go linear for debugging.

Debugging an async library is a pain: it's nice to force it into a
linear call chain to try to track problems.

Ugly code, though.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: test read after hangup.

In particular, make sure that idle connections don't get closed.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: simplify do_ready.

Have it set the plan itself, rather than passing it back.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: make io functions more generic.

Pass fd and plan explicitly, so they don't need to know the definition
of struct io_conn, and return a bool instead of an enum.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: remove io_state.

Use a NULL next pointer instead to indicate a closing connection.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: remove IO_IDLE state.

Use a NULL io pointer instead to indicate an idle connection.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: remove IO_NEXT state.

It was only used for initial connections which hadn't started I/O.  Now
they are initialized with an io_plan, it can be eliminated.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: initialize connection with an explicit I/O plan.

Rather than going via a callback, which tends to just set up I/O, do
any setup before the call to io_new_conn(), then pass it the io_plan
directly.

The patch shows how much this simplifies our test code.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: generic init function for listening connections.

Instead of assuming they want a connection made from the new fd, hand
the fd to a callback.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: remove conn arg from io_plan constructors.

No longer needed.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: pass struct io_plan explicitly.

This simplifies some things: in particular, we can construct an
io_plan without needing the current io_conn.

On the other hand, we need to expose the structure now.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: replace backend_set_state with backend_wakeup()

We only use it in one place: for wakeup.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: make enum io_state namespace-safe.

Ready for exposure.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: use explicit IO callback functions, instead of io_state values.

Explicit callbacks are slower, but more flexible.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: remove next & finish from union.

Sure, both listener and conn need them, but for different things
(listener uses them simply to set up conn).  Putting them in the
common union was a mistake.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: put explicit poll flags in the plan.

Weaning off enum io_state, to allow custom ones.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: get rid of io_next(), pass callbacks directly.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: rename io_op to io_plan.

This is a better description, since it's I/O we plan to do next.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: timer support.

Upgrade license, since timer is LGPL.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: check for all idle.

It's probably a bug if we're waiting for nothing.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: io_duplex.

Cleaner model for I/O, with cost of complexity if you really want bidir.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: benchmarks.

More stress test than benchmarks, but provides a beginning.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccan/io: new module.

Designed for async I/O.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

cpuid: minor fixes

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: cache processor brand string

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: remove the static func has_feature

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: fix build on 64-bit systems

Some instructions needed to use their correct prefix and correct
registers, such as "eax -> rax" and "pushl -> pushq"

Reported-by: Emilio G. Cota <cota@braap.org>
Tested-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

net: add async operation helpers.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

net: fix leak in test.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

opt: add allocator setting.

Good for tal usage.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

cpuid: ccanlint fixes.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Merge branch 'cpuid' of https://github.com/decltype/ccan

cpuid: parse CPU L1 cache and TLB identifications

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: use uint32_t instead of int/unsigned

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: small fix

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: Minor improvements

Remove the assembly file that checks if the CPUID instruction is
supported and have it in inline assembly as suggested by Rusty.

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

net: Fix potential uninitialized use of variables in net_bind()

If the list of addrinfos give to net_bind() contains only IPv4 or only
IPv6 addresses, then the ipv4 and ipv6 variables may be used uninitialized.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

cpuid: only compile source file if x86 cpu

Suggested-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: prefix every function with cpuid_

Also merge extended features and non-extended in 1 function.

Suggested-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: add 2 new functions + some more tests

The new functions are:
- get_cpu_type
- get_cpu_type_string

Also add more tests on how one would parse the low-level stuff.

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: add proper documentation

Forgot to add this to the main commit, sorry.

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

cpuid: new module

This module parses data provided by the cpuid instruction.
It still needs more work, however, it works for most important
stuff.

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

Merge branch 'master' of ozlabs.org:ccan

Add bitmap module to Makefile.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Merge https://github.com/dgibson/ccan

heap: add example on empty heap check

Suggested-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Emilio G. Cota <cota@braap.org>

heap: new module

Signed-off-by: Emilio G. Cota <cota@braap.org>

ilog: reformat so webpage doesn't get confused.

The extra spaces make my web formatter think they're literal text, which
looks weird.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ilog: relicense to Public Domain.

I noted to Timothy in 2011:
   I finally got around to implementing license checking in ccanlint,
   and it shows that your public domain ccan/isaac module uses your LGPL
   ccan/ilog module, which partially defeats the point.

   You and I are the only ones who did anything significant to ilog; I'm
   happy to relicense to PD or whatever else you want.  Or leave it, in
   which case I'll need to document it clearly.

He finally got back to me, and I sent him this patch, to which he replied:

     Ack.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

tal: add tal_resizez for zero-padded expansion.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

endian: add documentation on endian typedefs.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

short_types: update documentation on be32/le32 etc.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

endian/short_types: sparse support.

If you're doing endian work, sparse's bitwise annotation is extremely useful.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

bitmap: Add helper macro to statically declare bitmaps

For use as direct locals, or when the size is a constant, inside
structure definitions.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

bitmap: Rework to assume always multiple of words storage length

Handling bitmaps which extend some odd number of bits, and assuring they
don't clobber partially overlapped variables is not worth the bother.

Also avoid namespace pollution.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

bitmap: Rework types and sizing macros

Go with "words" rather than explicitly referencing longs in the macros

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

bitmap: Use a typesafe(ish) parameters instead of void *

This also fixes a bug which will hit on platforms where it's not safe to
dereference a (long *) which is not correctly aligned.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

list: add list_next and list_prev helpers.

Some way towards iterating in the middle of the list.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

bitmap: Add first cut at a bitmap module

This adds a module for manipulating bitmaps, based on the API from the
Linux kernel's bitmap.h.  So far it's missing the trickier bits, but is
a little more flexible, allowing bitmaps that don't have to be contained
in an integer number of unsigned longs.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

ccan: Fix calling memset with zero length parameter

Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

tal/path: fix unset vars in error paths.

Compiling with optimization FTW.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

breakpoint: new module.

Thanks to Jeremy Kerr for the idea!

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

endian: remove unnecessary _CONST postfixes.

It makes the names longer, whereas the CAPS is probably sufficient.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

endian: add constant versions.

Sometimes you really need a compile-time constant, but there's no generic way
to create a macro which does this and only evaluates its argument once.

Thus we need specific const macros.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

endian: change license to CC0.

Not worth LGPL for just this header.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

timer: clean up.

Add examples and a documentation fix.
Remove unused cascade function (was used in initial always-step-1-bucket
 version).
Restore timers_dump() to within CCAN_TIMER_DEBUG.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

timer: increase default span to 5.

There's a sweet spot around 5, without going all the way to 13 (which
would make each level 64/128k large).

TIMER_LEVEL_BITS 2: 100000000 in 5.833399-5.884659(5.85912+/-0.015) (18 levels / 32)
TIMER_LEVEL_BITS 3: 100000000 in 5.267610-5.303589(5.29018+/-0.011) (12-13(12.5+/-0.5) levels / 22)
TIMER_LEVEL_BITS 4: 100000000 in 4.215605-4.417297(4.27771+/-0.062) (9-10(9.1+/-0.3) levels / 16)
TIMER_LEVEL_BITS 5: 100000000 in 3.859340-3.972603(3.90149+/-0.034) (7-8(7.1+/-0.3) levels / 13)
TIMER_LEVEL_BITS 6: 100000000 in 3.976157-4.037230(4.00199+/-0.02) (6 levels / 11)
TIMER_LEVEL_BITS 7: 100000000 in 4.014228-4.082031(4.05024+/-0.022) (5-6(5.4+/-0.49) levels / 10)
TIMER_LEVEL_BITS 8: 100000000 in 3.915615-3.978781(3.94972+/-0.02) (5 levels / 8)
TIMER_LEVEL_BITS 9: 100000000 in 3.859413-4.025842(3.89993+/-0.046) (4-5(4.6+/-0.49) levels / 8)
TIMER_LEVEL_BITS 10: 100000000 in 3.983507-4.170152(4.06743+/-0.055) (4 levels / 7)
TIMER_LEVEL_BITS 11: 100000000 in 3.468756-3.610746(3.52843+/-0.04) (4 levels / 6)
TIMER_LEVEL_BITS 12: 100000000 in 3.274397-3.377530(3.3192+/-0.026) (4 levels / 6)
TIMER_LEVEL_BITS 13: 100000000 in 3.116845-3.178162(3.1398+/-0.017) (3 levels / 5)
TIMER_LEVEL_BITS 14: 100000000 in 3.152599-3.264060(3.20733+/-0.043) (3 levels / 5)
TIMER_LEVEL_BITS 15: 100000000 in 3.186556-3.552432(3.25597+/-0.11) (3 levels / 5)
TIMER_LEVEL_BITS 16: 100000000 in 3.139352-3.485653(3.27263+/-0.11) (3 levels / 4)

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

timer: cache the minimal value.

We spend a lot of time searching for the next timer to expire: by caching
the minimum, we can skip most of this work.  Even if timers are deleted,
the minimum will be a starting point for searching.

The expected-usage benchmark has to be increased by a factor of 100,
otherwise it's now too short.

Before:
$ ./expected-usage
1000000 in 12.701647935

After:
$ ./expected-usage  1000000
1000000 in 0.061095153

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

timers: implementation of lazily-ordered timers.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Add Henry Eshbaugh's log module to junkcode for the moment.

It's got some useful stuff in it, but doesn't pass ccanlint here.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Makefile: don't separate all modules with external dependencies.

Using ccanlint's --deps-fail-ignore, we can just test every module
with "make check".  Still exclude them from the 'libccan.a' build
though.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

nfs: add licenses into generated files.

Otherwise ccanlint complains about missing license comments.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccanlint: --deps-fail-ignore.

Useful for bulk testing modules, where some might not have required
external (non-CCAN) dependencies.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ogg_to_pcm: fix compilation of example in _info.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

ccanlint: don't crash if given bad directory name.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Add a set of simple version comparison helpers

These version helpers help to compare major.minor style version numbers,
without the need for open-coded and error-prone bitshifting, multiplication,
and similar.

Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

net: add server support.

Creating a server for IPv4 and IPv6 has similar issues to clients,
with some novel twists.  Slightly different arguments need to be given
to getaddrinfo(), but worse, some platforms (Linux without
/proc/sys/net/ipv6/bindv6only set) automatically bind IPv6 sockets to IPv4
ports as well.

Thus we need a function which can bind (and listen) to one or two fds.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

list: list_pop

list_top + list_del, as suggested by Ben Herrenschmidt.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

list: list_append_list / list_prepend_list

Operations for moving the entire contents of a list.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>