Gavin Lambert
2016-08-04 03:53:49 UTC
Another default-branch patchset update is attached. It is still based on
5a70ffc4644b.
For convenience, I've made an online patchset repository
<https://sourceforge.net/u/uecasm/etherlab-patches/ci/default/tree/#readme>
. It contains a README that describes the individual patches (from prior
posts, where I could find them; let me know if something is unclear) and
gives the commands required to clone your own copy of the upstream
repository and apply the patches to it.
Modified patches:
* 0051-fsm_change-external-datagram.patch:
0052-fsm_slave_config-external-datagram.patch:
0053-fsm_slave_scan-external-datagram.patch:
Fixed some spots where it was accessing the wrong datagram.
New patches:
* 0043-ethercat-diag.patch:
Recently posted to the users list by Ralf Roesch, adding a "diag" command to
the ethercat command line tool, to aid in locating lost links and other
comms errors.
Other than rebasing within the patchset, I've made the following tweaks to
the code (otherwise it is largely unchanged):
o Various whitespace fixes.
o Moved a few lines in main.cpp for consistent grouping.
o Changed llc_reset to a bool, since it was being used as a bool.
o Tweaked a few lines from assignment to compound assignment to simplify.
o Removed the data type lookup in EscRegRead and EscRegWrite.
* Since this is all internal and only used to get the size, which is well
known to the caller, the lookup seemed unnecessary.
o Made EscRegRead and EscRegWrite output errors to stderr instead of
stdout.
o Fixed a printf format issue that generated a compiler warning.
o Made EscRegRead and EscRegWrite treat errors as non-fatal.
* Some slaves do not implement all of these registers, and so trying to
read them will produce an "I/O error" exception. In this case it makes more
sense to continue reading the other registers than to abort.
* For example, Beckhoff EL3062 does not implement register 0x030C.
* 0x0308-0x030B can similarly be absent on some older slaves.
I'm a little hesitant about the command name being "diag" - while it's not a
bad name for network diagnostics or error stats it might be confused with
the "Diagnosis History" object as specified in ETG1020, which is an entirely
different thing. (And something that might be useful to add to the tool in
the future.) I'm open to alternative suggestions.
* 0044-diag-readwrite.patch:
This is a further modification on top of the previous patch which replaces
several separate read and write requests with a single read or read+write
request (plus one additional read) per slave. (So naturally it depends on
patch 0026.)
In theory this is more efficient, but most importantly since the reset
occurs using the same datagram as the read, it's now atomic and there's no
risk of losing counts (which could previously happen if the slave
incremented its counter after the read but before the write).
* 0045-slave-config-position.patch:
Adds a "position" field to the structure returned by
ecrt_slave_config_state. This allows you to quickly get the ring position
of a slave from its relative alias:offset address, which in turn allows you
to call other APIs that require this (eg. ecrt_master_get_slave).
Note that the position is only valid if "online" is true, and that it is
possible for the value to be stale (ie. the slave has moved to a different
position) if the network changes and is consequently rescanned after this
call. So use it defensively. (You're reasonably safe in the period between
requesting the master and activating it, as rescans are inhibited during
this time. OTOH, only the application can request the master; an external
tool can't.)
I'm considering whether it would be useful to make a general function
available for this conversion, to avoid duplicating the alias:offset
conversion logic in too many places (eg. the tool requires it as well, but
can't use the slave_config-based conversion since it can't request the
master).
* 0046-e1000e-link-detection.patch:
Fixed link detection in e1000e driver for 3.10-3.16.
This is Christoph Permes
<http://lists.etherlab.org/pipermail/etherlab-dev/2016/000554.html> ' 3.16
patch and 3.14-v2 patch, with the latter backported to 3.12 and 3.10. I
haven't tested these personally.
* 0057-fsm_foe-simplify.patch:
Removes some redundant fields from the FoE FSM; some were unused copy/paste
holdovers from the CoE FSM while others were duplicated unnecessarily
between the read and write operations, which can't be concurrent for a given
slave anyway.
Also fixes the case where the incoming data exceeds the provided buffer to
properly terminate the state machine instead of leaving things dangling.
Although note that this still leaves the FoE conversation itself dangling,
so you'll likely get an error on the next request if this occurs.
* 0058-foe-password.patch:
Adds support for sending an FoE password along with read or write requests.
Also implements the -o option for the foe_read command (which was documented
but not implemented).
Also makes the ioctl behind foe_read actually use the buffer size requested
by the caller (instead of a hard-coded value); though note that foe_read
itself still uses a hard-coded value of its own (but it's larger, so bigger
files should be readable now). It's possible that users on limited-memory
embedded systems might need to reduce this value, but it's still fairly
conservative as RAM sizes go.
* 0059-foe-requests.patch:
Makes FoE transfer requests into public ecrt_* API, similar to SDO requests.
Primarily (following my goal of "parallel all the things"), this allows FoE
transfers to be non-blocking so that transfers to multiple slaves can occur
concurrently from the same requesting thread (previously this was only
possible by using separate threads, since the only API was blocking). Note
that due to patch 0018 you can call ecrt_master_deactivate() to delete these
requests when you're done with them, even if you haven't called
ecrt_master_activate() yet.
It has a possible side benefit that FoE transfers can now be started and
monitored from realtime context, although as FoE is mostly used for firmware
updates this is unlikely to be all that useful in practice.
I considered a few alternative approaches to this (the next leading
contender was to make async versions of the existing FoE ioctls), but this
seemed more consistent with existing APIs. I'm open to suggestions here too
though, since it does feel like a slightly odd fit. (But works quite
nicely.)
* 0060-foe-request-progress.patch:
Adds a way to get a "current progress" value (actually the byte offset) for
async FoE transfers.
5a70ffc4644b.
For convenience, I've made an online patchset repository
<https://sourceforge.net/u/uecasm/etherlab-patches/ci/default/tree/#readme>
. It contains a README that describes the individual patches (from prior
posts, where I could find them; let me know if something is unclear) and
gives the commands required to clone your own copy of the upstream
repository and apply the patches to it.
Modified patches:
* 0051-fsm_change-external-datagram.patch:
0052-fsm_slave_config-external-datagram.patch:
0053-fsm_slave_scan-external-datagram.patch:
Fixed some spots where it was accessing the wrong datagram.
New patches:
* 0043-ethercat-diag.patch:
Recently posted to the users list by Ralf Roesch, adding a "diag" command to
the ethercat command line tool, to aid in locating lost links and other
comms errors.
Other than rebasing within the patchset, I've made the following tweaks to
the code (otherwise it is largely unchanged):
o Various whitespace fixes.
o Moved a few lines in main.cpp for consistent grouping.
o Changed llc_reset to a bool, since it was being used as a bool.
o Tweaked a few lines from assignment to compound assignment to simplify.
o Removed the data type lookup in EscRegRead and EscRegWrite.
* Since this is all internal and only used to get the size, which is well
known to the caller, the lookup seemed unnecessary.
o Made EscRegRead and EscRegWrite output errors to stderr instead of
stdout.
o Fixed a printf format issue that generated a compiler warning.
o Made EscRegRead and EscRegWrite treat errors as non-fatal.
* Some slaves do not implement all of these registers, and so trying to
read them will produce an "I/O error" exception. In this case it makes more
sense to continue reading the other registers than to abort.
* For example, Beckhoff EL3062 does not implement register 0x030C.
* 0x0308-0x030B can similarly be absent on some older slaves.
I'm a little hesitant about the command name being "diag" - while it's not a
bad name for network diagnostics or error stats it might be confused with
the "Diagnosis History" object as specified in ETG1020, which is an entirely
different thing. (And something that might be useful to add to the tool in
the future.) I'm open to alternative suggestions.
* 0044-diag-readwrite.patch:
This is a further modification on top of the previous patch which replaces
several separate read and write requests with a single read or read+write
request (plus one additional read) per slave. (So naturally it depends on
patch 0026.)
In theory this is more efficient, but most importantly since the reset
occurs using the same datagram as the read, it's now atomic and there's no
risk of losing counts (which could previously happen if the slave
incremented its counter after the read but before the write).
* 0045-slave-config-position.patch:
Adds a "position" field to the structure returned by
ecrt_slave_config_state. This allows you to quickly get the ring position
of a slave from its relative alias:offset address, which in turn allows you
to call other APIs that require this (eg. ecrt_master_get_slave).
Note that the position is only valid if "online" is true, and that it is
possible for the value to be stale (ie. the slave has moved to a different
position) if the network changes and is consequently rescanned after this
call. So use it defensively. (You're reasonably safe in the period between
requesting the master and activating it, as rescans are inhibited during
this time. OTOH, only the application can request the master; an external
tool can't.)
I'm considering whether it would be useful to make a general function
available for this conversion, to avoid duplicating the alias:offset
conversion logic in too many places (eg. the tool requires it as well, but
can't use the slave_config-based conversion since it can't request the
master).
* 0046-e1000e-link-detection.patch:
Fixed link detection in e1000e driver for 3.10-3.16.
This is Christoph Permes
<http://lists.etherlab.org/pipermail/etherlab-dev/2016/000554.html> ' 3.16
patch and 3.14-v2 patch, with the latter backported to 3.12 and 3.10. I
haven't tested these personally.
* 0057-fsm_foe-simplify.patch:
Removes some redundant fields from the FoE FSM; some were unused copy/paste
holdovers from the CoE FSM while others were duplicated unnecessarily
between the read and write operations, which can't be concurrent for a given
slave anyway.
Also fixes the case where the incoming data exceeds the provided buffer to
properly terminate the state machine instead of leaving things dangling.
Although note that this still leaves the FoE conversation itself dangling,
so you'll likely get an error on the next request if this occurs.
* 0058-foe-password.patch:
Adds support for sending an FoE password along with read or write requests.
Also implements the -o option for the foe_read command (which was documented
but not implemented).
Also makes the ioctl behind foe_read actually use the buffer size requested
by the caller (instead of a hard-coded value); though note that foe_read
itself still uses a hard-coded value of its own (but it's larger, so bigger
files should be readable now). It's possible that users on limited-memory
embedded systems might need to reduce this value, but it's still fairly
conservative as RAM sizes go.
* 0059-foe-requests.patch:
Makes FoE transfer requests into public ecrt_* API, similar to SDO requests.
Primarily (following my goal of "parallel all the things"), this allows FoE
transfers to be non-blocking so that transfers to multiple slaves can occur
concurrently from the same requesting thread (previously this was only
possible by using separate threads, since the only API was blocking). Note
that due to patch 0018 you can call ecrt_master_deactivate() to delete these
requests when you're done with them, even if you haven't called
ecrt_master_activate() yet.
It has a possible side benefit that FoE transfers can now be started and
monitored from realtime context, although as FoE is mostly used for firmware
updates this is unlikely to be all that useful in practice.
I considered a few alternative approaches to this (the next leading
contender was to make async versions of the existing FoE ioctls), but this
seemed more consistent with existing APIs. I'm open to suggestions here too
though, since it does feel like a slightly odd fit. (But works quite
nicely.)
* 0060-foe-request-progress.patch:
Adds a way to get a "current progress" value (actually the byte offset) for
async FoE transfers.