Note that asking for "prefer N-port and then Loop" with the 24XX seems to
cause the f/w to crash. So, don't do that.
Firmware crashes are now considered fatal and cause the HBA to be marked dead.
Now that we've avoided the f/w crash in isp_init_24xx, we can process
toplogy options.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@274 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Add some code to do special IOCBs (that can just be ignored if they show
up as status IOCBs)- this is for some performance testing infrastructure.
Add some performance testing stuff.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@273 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Don't use both WWNs in the ICB block unless they're not NAA==2. This makes us
more consistent with QLogic driver setups and thus more likely to fit into the
same zoning setups.
If we're getting a PDB and it's for the Fabric Controller, force a login
to it (just in case it logged out underneath us). Return 'not logged in'
for the 24XX in case the 24XX specific login state isn't in the range of
PLOGI_DONE to LOGGED IN.
Don't do local loop scans if we have more than one channel (NP-IV) and we're
in a FL-port topology- we just end up poking ourselves usually.
Make sure we set loopstate in a couple of places we missed so that we
try and get it again.
Add in the RESET DEVICE and ABORT COMMAND functions for the 24XX cards
(finally).
Fix a longstanding bug where were passing the wrong (off by one) output
pointer to isp_handle_other_response.
Change isp_async...ISPASYNC_CHANGE_NOTIFY to have N-port handle, nlstate and
reason included as we get these for free with PDB Changed stuff for the
24XX card. Eventually we should be able to edit single local PDB entry
states instead of doing a full eval.
Put in the put of a 24XX Task Management Function IOCB. There is no get because
the firmware doesn't return it when done- instead a status iocb is done.
There is no "FAST POST" flag for the 24XX- this is a NOACK flag which means
that there is no interrupt if all goes well.
Fix spelling mistakes.
In isp_add_wwn_entry, ahem, don't allocate PDB entries out of the 'rexerved
for chip' area.
Put more comments in as to why we're terminating an exchange (and make them
all ISP_LOGTINFO).
Add, finally, the correct handling for a QIN_NOTIFY_ACK for a task management
function- this means we construct a ct7_entry_t to get the firmware to send
an FC Response IU for it. So, finally LUN RESETS and TARGET RESETS and
ABORT TASK will actually work and, more importantly, be seen to be working
by initiators.
Fix WWN defaults/active generation and retrieval. It was broken.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@272 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Fix a comment typo. Force on HARD ADDRESS for the 24XX.
Fix retry on error case- most of the time we *don't* want to retry a command
sent back with TDFL_ERROR set- instead we want to set PARITY ERROR as a
CA condition and just FIN the cmd.
Comment for r270:
1. A major botch for local loop handling - we were never getting past loopid zero
in isp_scan_loop. Bad doggie!
2. *Do* scan our loopid- this seems to allow us to find entities which are there
and have a N-port handle that is identical to ours. I have no idea how the QLogic
sorts this out. At least, this is how it seems to work for the 24XX in local loop
mode.
Don't deadlock with the HBA lock when performing a rescan ioctl. Also,
mark the loop state as if we'd just received a PDB event- that way
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@271 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Fix a comment typo. Force on HARD ADDRESS for the 24XX.
Fix retry on error case- most of the time we *don't* want to retry a command
sent back with TDFL_ERROR set- instead we want to set PARITY ERROR as a
CA condition and just FIN the cmd.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@270 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- Add ASYNC_RCV_ERR
- Allow LIP to have a (possible) channel.
- Update for some channel stuff.
- whitespace
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@268 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- Clean up 24XX abort handling.
- Remove the response queue busy stuff and just have an in_intr flag to note
interrupt recursion.
- Add a lower level error flag.
- Clarifiy a comment; make a definition for the 'no task' RX_ID.
- Change severity of f/w timeout message
- Do the wait for mbox0 clearing for 2312 cards > revision 1.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@267 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- Make sure we set/clear TDFL_ERROR
- Make sure we set SENSE VALID when we do a synthetic CTIO that has status.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@266 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Add unit attention injection. Fix some ridiculous Sense things.
Say why we couldn't pci register the driver.
Make isp_thread_event return an int if it fails to send the event. If it
fails to send the event, and this was for the 24XX card finding out the
initiator for a command, return the command with a BUSY status.
An N-port handle of zero is legal for a 24XX.
Add some more special N-Port handle definitions for 2K f/w.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@265 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Warn about bogus sense data in isp_target_start_ctio. Remove some redundant
zero initializations.
Make more common and more useful the debug printouts for the
various target dma mapping routines.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@264 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- Start checking for a couple of things for the type of SAN we're connected to.
- Remove restriction against 2.6.23.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@263 d57e44dd-8a1f-0410-8b47-8ef2f437770f
As of version 4.0.25 the vpindex is available for MID f/w on the 24XX
for ASYNC_CHANGE_NOTIFY in outgoing mailbox 3.
Per QLogic release notes, the immediate notify entry for the 24XX has
various other items of interest in it based upon the ELS code.
Per QLogic release notes, after 4.0.25 f/w the WWPN comes with the
PLOGI/PDISC/ADISC and LOGO ELS codes in the immediate notify structure.
This makes initiator WWN maintenance *way* easier.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@262 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Hmm. Just in case INI_ANY and INI_NONE are not defined, define them here.
This is for the case of external software including isp_tpublic.h only.
Fix some misapplied flags.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@261 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Go for the gusto and merge multichan_development into the head branch. This
will break FreeBSD and NetBSD ports for a while (yawn) and solaris is dubious
anyway.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@260 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Turn off, yet again, fast posting and RIO for 2200 cards. It just seems
to buggy in this driver to be used, and because of the age of the cards
not worth chasing down.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@258 d57e44dd-8a1f-0410-8b47-8ef2f437770f
-revert the part of 1.174 that stopped complaining about not finding
fast posting handles.
-pull REPORT LUNS (and LUN INVENTORY MAY HAVE CHANGE CA condition) from
multi-channel branch
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@256 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- clean up tpublic notes and change tmd_xfr_t to tmd_xact_t
- minor cleanups from mendocino
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@255 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- TM fixes related to new commands serialization
- Other TM improvements
- Minor changes: likely/unlikely and priority of mgmt threads
Work not fully completed, still there are some rare issues.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@240 d57e44dd-8a1f-0410-8b47-8ef2f437770f
1. The kernel's log level is a string, so strcmp needs to be used for
comparisons.
2. Svn HEAD doesn't compile against kernel versions < 2.6.19 because
"bool" had its debut only in 2.6.19. Here's a quick fix.
3. Iscsi-scstd dies with SIGPIPE if run in foreground mode because the reading
end of the init_report_pipe is already closed in this mode.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@238 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- TM fixes and improvements, particularly, TARGET RESET now resets only visible to the initiator devices
- Logs improvements
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@237 d57e44dd-8a1f-0410-8b47-8ef2f437770f
- ABORT_CONSIDER_FINISHED_TASKS_AS_NOT_EXISTING compilation symbol added
- TM fixes and cleanups
- Fixes crash on modprobe with some real devices
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@233 d57e44dd-8a1f-0410-8b47-8ef2f437770f
We currently have multiple scst targets with exported fileio or blockio luns using the same files or block devices.
These fileio or blockio luns end up to have the same lun ID (dev_id_num), linux dm-multipath recognizes them as same lun (because of lun ID) with different paths which is wrong. It should be multiple separated luns from different scst targets.
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@228 d57e44dd-8a1f-0410-8b47-8ef2f437770f
Particularly:
- Added ability for target drivers to set result of command's status delvery to initiator (see scst_set_delivery_status())
- Added ability for user space devices set some mode page parameters to make SCST core behave correspondingly
- TM improvements
- Few possible crashes fixed
- Logging improvements
- Performance improvements
- Docs updates
git-svn-id: http://svn.code.sf.net/p/scst/svn/trunk@227 d57e44dd-8a1f-0410-8b47-8ef2f437770f
