sure that it really is by issuing a ISPCTL_ABORT_CMD just on the
off chance the f/w will start it up again and, ha ha, start using
the DMA resources we gave it but are now taking away.
us to not the ints are ok and also to (re)ENABLE isp interrupts. Remove
all splcam()/splx() invocates and replace them with ISP_LOCK/ISP_UNLOCK
macros.
to isp_osinfo substructure (all in prep for SMP). Define MBOX_WAIT_COMPLETE
and MBOX_NOTIFY_COMPLETE macros so that we can now (temp) use tsleep
to wait for mailbox completion. Requires us to guess whether we're
servicing an interrupt or not- will use intr_nesting_level.
Add local strncat function.
define). Fix stupidity wrt checking whether we've gone to
LOOP_PDB_RCVD loopstate- it's okay to be greater than this state.
D'oh! Protect calls to isp_pdb_sync and isp_fclink_state with IS_FC
macros.
Completely redo mailbox command routine (in preparation to make this
possibly wait rather than poll for completion).
Make a major attempt to solve the 'lost interrupt' problem
1. Problem
The Qlogic cards would appear to 'lose' interrupts, i.e., a legitimate
regular SCSI command placed on the request queue would never complete
and the watchdog routine in the driver would eventually wakeup and
catch it. This would typically only happen on Alphas, although a
couple folks with 700MHz Intel platforms have also seen this.
For a long time I thought it was a foulup with f/w negotiations of
SYNC and/or WIDE as it always seemed to happen right after the
platform it was running on had done a SET TARGET PARAMETERS mailbox
command to (re)enable sync && wide (after initially forcing
ASYNC/NARROW at startup). However, occasionally, the same thing
would also occur for the Fibre Channel cards as well (which, ahem,
have no SET TARGET PARAMETERS for transfer mode).
After finally putting in a better set of watchdog routines for the
platforms for this driver, it seemed to be the case that the command
in question (usually a READ CAPACITY) just had up and died- the
watchdog routine would catch it after ~10 seconds. For some platforms
(NetBSD/OpenBSD)- an ABORT COMMAND mailbox command was sent (which
would always fail- indicating that the f/w denied knowledge of this
command, i.e., the f/w thought it was a done command). In any case,
retrying the command worked. But this whole problem needed to be
really fixed.
2. A False Step That Went in The Right Direction
The mailbox code was completely rewritten to no longer try and grab
the mailbox semaphore register and to try and 'by hand' complete
async fast posting completions. It was also rewritten to now have
separate in && out bitpatterns for registers to load to start and
retrieve to complete. This means that isp_intr now handles mailbox
completions.
This substantially simplifies the mailbox handling code, and carries
things 90% toward getting this to be a non-polled routine for this
driver.
This did not solve the problem, though.
3. Register Debouncing
I saw some comments in some errata sheets and some notes in a Qlogic
produced Linux driver (for the Qlogic 2100) that seemed to indicate
that debouncing of reads of the mailbox registers might be needed,
so I added this. This did not affect the problem. In fact, it made
the problem worse for non-2100 cards.
5. Interrupt masking/unmasking
The driver *used* to do a substantial amount of masking/unmasking
of the interrupt control register. This was done to make sure that
the core common code could just assume it would never get pre-empted.
This apparently substantially contributed to the lost interrupt
problem. The rewrite of the ICR (Interrupt Control Register),
which is a separate register from the ISR (Interrupt Status Register)
should not have caused any change to interrupt assertions pending.
The manual does not state that it will, and the register layout
seems to imply that the ICR is just an active route gate. We only
enable PCI Interrupts and RISC Interrupts- this should mean that
when the f/w asserts a RISC interrupt and (and the ICR allows RISC
Interrupts) and we have PCI Interrupts enabled, we should get a
PCI interrupt. Apparently this is a latch- not a signal route.
Removing this got rid of *most* but not all, lost interrupts.
5. Watchdog Smartening
I made sure that the watchdog routine would catch cases where the
Qlogic's ISR showed an interrupt assertion. The watchdog routine
now calls the interrupt service routine if it sees this. Some
additional internal state flags were added so that the watchdog
routine could then know whether the command it was in the middle
of burying (because we had time it out) was in fact completed by
the interrupt service routine.
6. Occasional Constipation Of Commands..
In running some very strenous high IOPs tests (generating about
11000 interrupts/second across one Qlogic 1040, one Qlogic 1080
and one Qlogic 2200 on an Alpha PC164), I found that I would get
occasional but regular 'watchdog timeouts' on both the 1080 and
the 2100 cards. This is under FreeBSD, and the watchdog timeout
routine just marks the command in error and retries it.
Invariably, right after this 'watchdog timeout' error, I'd get a
command completion for the command that I had thought timed out.
That is, I'd get a command completion, but the handle returned by
the firmware mapped to no current command. The frequency of this
problem is low under such a load- it would usually take an 30
minutes per 'lost' interrupt.
I doubled the timeout for commands to see if it just was an edge
case of waiting too short a period. This has no effect.
I gathered and printed out microtimes for the watchdog completed
command and the completion that couldn't find a command- it was
always the case that the order of occurrence was "timeout, completion"
separated by a time on the order of 100 to 150 ms.
This caused me to consider 'firmware constipation' as to be a
possible culprit. That is, resubmission of a command to the device
that had suffered a watchdog timeout seemed to cause the presumed
dead command to show back up.
I added code in the watchdog routine that, when first entered for
the command, marks the command with a flag, reissues a local timeout
call for one second later, but also then issues a MARKER Request
Queue entry to the Qlogic f/w. A MARKER entry is used typically
after a Bus Reset to cause the f/w to get synchronized with respect
to either a Bus, a Nexus or a Target.
Since I've added this code, I always now see the occasional watchdog
timeout, but the command that was about to be terminated always
now seems to be completed after the MARKER entry is issued (and
before the timeout extension fires, which would come back and
*really* terminate the command).
comment. Check against firmware state- not loop state when enabling
target mode. Other changes have to do with no longer enabling/disabling
interrupts at will.
Rearchitect command watchdog timeouts-
First of all, set the timeout period for a command that has a
timeout (in isp_action) to the period of time requested *plus* two
seconds. We don't want the Qlogic firmware and the host system to
race each other to report a dead command (the watchdog is there to
catch dead and/or broken firmware).
Next, make sure that the command being watched isn't done yet. If
it's not done yet, check for INT_PENDING and call isp_intr- if that
said it serviced an interrupt, check to see whether the command is
now done (this is what the "IN WATCHDOG" private flag is for- if
isp_intr completes the command, it won't call xpt_done on it because
isp_watchdog is still looking at the command).
If no interrupt was pending, or the command wasn't completed, check
to see if we've set the private 'grace period' flag. If so, the
command really *is* dead, so report it as dead and complete it with
a CAM_CMD_TIMEOUT value.
If the grace period flag wasn't set, set it and issue a SYNCHRONIZE_ALL
Marker Request Queue entry and re-set the timeout for one second
from now (see Revision 1.45 isp.c notes for more on this) to give
the firmware a final chance to complete this command.
store a bitmask of whether we've set a value into ccb->ccb_h.status,
whether we're in the watchdog routine for this command now, whether
we've set a grace period for this command and whether this command is
actually done.
See comments of rev 1.45 of isp.c for more complete information.
output mailbox values we want to get back out of the chip once a mailbox
command is done. Add storage for the maximum number of output mailbox
registers to the softc.
Roll minor version number.
the handle (i.e., generation number), so we will now need a function that
will take a handle and return a flat index [ 0 .. maxhandles-1 ] for
auxillary routines that need an index to get at buddy store values
(like dma maps or xflist pointers).
Force alphas to prefer mem mapping as the default.
Basically, we have a pointer to a function which we can call which will
return us a pointer to firmware for the card we have. We call this function
(if it's non-NULL) with the address of our mdvec f/w pointer.
The way this works is that if ispfw (as a module or a static) is loaded,
it initializes the pointer in isp_pci, so we can call into to it to fetch
a pointer to a f/w set.
If ispfw is MOD_UNLOADed, it's retained a pointer to our mdvec f/w pointers,
which then get zeroed out so we don't have any references to data that's
now gone from kernel memory. Removing the f/w saves ~360KBytes.
Alas, there is no autounload mechanism that works for is here.
through, establish what our LUN width is. Unfortunately, we can't ask
the f/w. If we loaded the f/w, we'll now assume we have expanded LUNs
(SCCLUN for fibre channel, just plain 32 LUN for SCSI). If we didn't
load firmware, assume 8 LUNs for SCSI and 1 LUN for Fibre Channel. We
have to assume only one LUN for Fibre Channel because the LUN setting
in Request Queue entries is in different places whether we have SCCLUN
firmware or not, so the only LUN guaranteed to work for both is LUN 0.
Clean up the rest of isp.c so that ISP2100_SCCLUN defines aren't used-
instead use run time determinants based upon isp->isp_maxluns.
After starting firmware, delay 500us to give it a chance to get rolling.
Fix the interrupt service routine to check for both isr && sema being zero
before thinking this was a spurious interrupt. Following the manuals,
allow for both Mailbox as well as Queue Reponse type interrupts for regular
SCSI.
(we always support fabric now). Remove SCCLUN definition (we always
support SCCLUN now, if we load the f/w). Add typedef definition of an
external firmware fetch function.
What we'd like to know is whether or not we have a listener
upstream that really hasn't configured yet. If we do, then
we can give a more sensible reply here. If not, then we can
reject this out of hand.
Choices for what to send were
Not Ready, Unit Not Self-Configured Yet
(0x2,0x3e,0x00)
for the former and
Illegal Request, Logical Unit Not Supported
(0x5,0x25,0x00)
for the latter.
We used to decide whether there was at least one listener
based upon whether the black hole driver was configured.
However, recent config(8) changes have made this hard to do
at this time.
Actually, we didn't use the above quite yet, but were sure considering it.
changes: consider a new PDB entry different if Class 3 service parameter
roles change (!!!). Do some checking as we're getting a port database
that traps whether things change while we're doing so. Handle N-port
and F-ports correctly. Fix the fabric login loop to retain a login/binding
if things haven't changed (I mean, why logout a device only to log it back
in). No longer accept, after fabric logins, garbage if we can't get a PDB
entry that matches the device we've just logged into- if it doesn't, log
it out as it is very unlikely to still be what we thought it was. Get rid
of some of the debounce loops because we could get stuck there.
Apparently the f/w has finished the command, but somehow an interrupt is
being lost. So, we just plain wedge when booting alphas.
This is a general routine we've needed for a while.
where we can have targets (based on topology).
Much more importantly, make sure all mods to isp_sendmarker or |= so
we don't lose the marking of a bus that needs to have a marker sent for it.
require full logins after a LIP, which always led to loop resets, and
various other perturbations.
Update 2200 f/w from 2.01.00 release to 2.01.09 release.
seriously- only attempt to logout a previously logged in fabric device.
Fix a longstanding bug for aborting overtime commands- handle halves
have always been reversed.
Clean up some error messages to indicate channel number.
Approved:jkh
Andrew's problems with SCSI on some alphas- do not call isp_update
directly to update parameters- just mark them as being ready to
update for the next command- the system would just hang on a READ
CAPACITY for a drive. Really annoying because it wouldn't even timeout
(and it has a timeout) so either the SET PARAMETERS call was nuking
things or the f/w was really dropping the ball.
approved: jkh
Reviewed by: gallatin@freebsd.org
is gone as a define. We just don't support fast posting for anything less
than the 1240/1080/1280/12160 or Fibre Channel cards.
Put in support for CDB's larger than 12 bytes for parallel SCSI (up to 44
bytes are allowed).
Approved: jkh
