<?xml version="1.0" encoding="latin1" ?>
<!DOCTYPE chapter SYSTEM "chapter.dtd">
<chapter>
<header>
<copyright>
<year>2009</year><year>2009</year>
<holder>Ericsson AB. All Rights Reserved.</holder>
</copyright>
<legalnotice>
The contents of this file are subject to the Erlang Public License,
Version 1.1, (the "License"); you may not use this file except in
compliance with the License. You should have received a copy of the
Erlang Public License along with this software. If not, it can be
retrieved online at http://www.erlang.org/.
Software distributed under the License is distributed on an "AS IS"
basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
the License for the specific language governing rights and limitations
under the License.
</legalnotice>
<title>Drivers</title>
<prepared>Bjorn Gustavsson</prepared>
<docno></docno>
<date>2009-11-16</date>
<rev></rev>
<file>drivers.xml</file>
</header>
<p>This chapter provides a (very) brief overview on how to write efficient
drivers. It is assumed that you already have a good understanding of
drivers.</p>
<section>
<title>Drivers and concurrency</title>
<p>The run-time system will always take a lock before running
any code in a driver.</p>
<p>By default, that lock will be at the driver level, meaning that
if several ports have been opened to the same driver, only code for
one port at the same time can be running.</p>
<p>A driver can be configured to instead have one lock for each port.</p>
<p>If a driver is used in a functional way (i.e. it holds no state,
but only does some heavy calculation and returns a result), several
ports with registered names can be opened beforehand and the port to
be used can be chosen based on the scheduler ID like this:</p>
<code type="none">
-define(PORT_NAMES(),
{some_driver_01, some_driver_02, some_driver_03, some_driver_04,
some_driver_05, some_driver_06, some_driver_07, some_driver_08,
some_driver_09, some_driver_10, some_driver_11, some_driver_12,
some_driver_13, some_driver_14, some_driver_15, some_driver_16}).
client_port() ->
element(erlang:system_info(scheduler_id) rem tuple_size(?PORT_NAMES()) + 1,
?PORT_NAMES()).</code>
<p>As long as there are no more than 16 schedulers, there will never
be any lock contention on the port lock for the driver.</p>
</section>
<section>
<title>Avoiding copying of binaries when calling a driver</title>
<p>There are basically two ways to avoid copying a binary that is
sent to a driver.</p>
<p>If the <c>Data</c> argument for
<seealso marker="erts:erlang#port_control/3">port_control/3</seealso>
is a binary, the driver will be passed a pointer to the contents of
the binary and the binary will not be copied.
If the <c>Data</c> argument is an iolist (list of binaries and lists),
all binaries in the iolist will be copied.</p>
<p>Therefore, if you want to send both a pre-existing binary and some
additional data to a driver without copying the binary, you must call
<c>port_control/3</c> twice; once with the binary and once with the
additional data. However, that will only work if there is only one
process communicating with the port (because otherwise another process
could call the driver in-between the calls).</p>
<p>Another way to avoid copying binaries is to implement an <c>outputv</c>
callback (instead of an <c>output</c> callback) in the driver.
If a driver has an <c>outputv</c> callback, refc binaries passed
in an iolist in the <c>Data</c> argument for
<seealso marker="erts:erlang#port_command/2">port_command/2</seealso>
will be passed as references to the driver.</p>
</section>
<section>
<title>Returning small binaries from a driver</title>
<p>The run-time system can represent binaries up to 64 bytes as
heap binaries. They will always be copied when sent in a messages,
but they will require less memory if they are not sent to another
process and garbage collection is cheaper.</p>
<p>If you know that the binaries you return are always small,
you should use driver API calls that do not require a pre-allocated
binary, for instance
<seealso marker="erts:erl_driver#int driver_output-3">driver_output()</seealso>
or
<seealso marker="erts:erl_driver#int driver_output_term-3">driver_output_term()</seealso>
using the <c>ERL_DRV_BUF2BINARY</c> format,
to allow the run-time to construct a heap binary.</p>
</section>
<section>
<title>Returning big binaries without copying from a driver</title>
<p>To avoid copying data when a big binary is sent or returned from
the driver to an Erlang process, the driver must first allocate the
binary and then send it to an Erlang process in some way.</p>
<p>Use <seealso marker="erts:erl_driver#ErlDrvBinary* driver_alloc_binary-1">driver_alloc_binary()</seealso> to allocate a binary.</p>
<p>There are several ways to send a binary created with
<c>driver_alloc_binary()</c>.</p>
<list type="bulleted">
<item><p>From the <c>control</c> callback, a binary can be returned provided
that
<seealso marker="erts:erl_driver#void set_port_control_flags-2">set_port_control()</seealso>
has been called with the flag value <c>PORT_CONTROL_FLAG_BINARY</c>.</p>
</item>
<item><p>A single binary can be sent with
<seealso marker="erts:erl_driver#int driver_output_binary-6">driver_output_binary()</seealso>.</p></item>
<item><p>Using
<seealso marker="erts:erl_driver#int driver_output_term-3">driver_output_term()</seealso>
or
<seealso marker="erts:erl_driver#int driver_send_term-4">driver_send_term()</seealso>,
a binary can be included in an Erlang term.</p>
</item>
</list>
</section>
</chapter>
