mutexes instead.
This closes the last (known) race issues in ATA which should fix
the various hangs etc seen on heavy loaded systems.
Change from using timeout functions to using callout functions in
the timeout code. This together with above closes the race that could
happen if timeout and device interrupt occured simultaniously.
Also fix the possible recursion in ata_reinit() on very dodgy
devices that could take us down in the probe.
This make "_NEC" devices appear as "NEC" which is more corrent.
The reason is tha NEC originally screwed up on the byteorder in the
model string, so now that they have realized that they prefixed the '_'
so that not every ATA driver on the planet would call them "EN C" :)
In places where we have long delays that doesn't depend on too accurate
timing, use ata_udelay() instead of DELAY() so we dont uselessly spin
the CPU if not nessesary;
Add missing untimeout that would get lost in handling of some
error situations, and caused what looked like random timeouts
afterwards when the timeout fired.
was being unconditionally dereferenced but was NULL for PIO requests.
Check the request flags for a DMA transaction before dereferencing.
Reported by: ceri
Tested by: Radek Kozlowski <radek -at- raadradd.com>
Properly wait for not busy and introduce a timeout for devices not
setting busy (as they should).
Leave a printf in there that states how long the wait was, as I'd like
to get an idea of the variations here. The time needed seems also to be
affected by whether a medium is present or not.
the GEOM topology.
There are still issues with not detaching from cam correctly such that
upon a device detach there's an invalid pointer dereference from the
later call to cam_rescan().
timeout values in the CAM CCBs. Divide by 1000 to get values in seconds
which are what ata(4) timeouts internally use.
This does lose granularity, though, and small values can now round down
to zero. It's probably worth making all ata(4) timeouts in terms of
hz/ticks/milliseconds/something.
channels. This also work when PCI native mode has been selected
(patch for /sys/dev/pci/pci.c needed for that) since pci_get_progif
uses the saved value for progif, not the one stored after we may have
changed from legacy mode to native PCI mode.
controllers (PDC203** PDC206**).
This also adds preliminary support for the Promise SX4/SX4000 but *only*
as a "normal" Promise ATA controller (ATA RAID's are supported though
but only RAID0, RAID1 and RAID0+1).
This cuts off yet another 5-8% of the command overhead on promise controllers,
making them the fastest we have ever had support for.
Work is now continuing to add support for this in ATA RAID, to accellerate
ATA RAID quite a bit on these controllers, and especially the SX4/SX4000
series as they have quite a few tricks in there..
This commit also adds a few fixes to the SATA code needed for proper support.
This adds support for cardbus ATA/SATA controllers. I get roughly the
same transfer speeds as on true PCI controllers. Nice to be able to add
a couble of "real" disks to a laptop :)
Introduce d_version field in struct cdevsw, this must always be
initialized to D_VERSION.
Flip sense of D_NOGIANT flag to D_NEEDGIANT, this involves removing
four D_NOGIANT flags and adding 145 D_NEEDGIANT flags.
Previously the "struct disk" were owned by the device driver and this
gave us problems when the device disappared and the users of that device
were not immediately disappearing.
Now the struct disk is allocate with a new call, disk_alloc() and owned
by geom_disk and just abandonned by the device driver when disk_create()
is called.
Unfortunately, this results in a ton of "s/\./->/" changes to device
drivers.
Since I'm doing the sweep anyway, a couple of other API improvements
have been carried out at the same time:
The Giant awareness flag has been flipped from DISKFLAG_NOGIANT to
DISKFLAG_NEEDSGIANT
A version number have been added to disk_create() so that we can detect,
report and ignore binary drivers with old ABI in the future.
Manual page update to follow shortly.
instead of taskqueue_swi. This shaves from 1 to 10% of the overhead.
Overhaul the locking once more, there was a few possible races that
are now closed.
This gives +10% performance on simple tests, so definitly worth it.
A few percent more could be had by not using M_ZERO'd alloc's, but
we then need to clear fields all over the place to be safe, and
that was deemed not worth the trouble (and it makes life dangerous).
of the leftovers from the old version that really doesn't work anymore.
Add a reset function for host-end of the ATA channel. This is needed
for the SiI3112 in order to whack it back to reality if a device
locks up the SATA interface (thereby preventing that we can reset the
device). The result is that ATA now recovers from the timeouts that
happens with the SiI3112A and more or less all disks based on old
PATA electronics with a Marvell PATA->SATA converter. This includes
lots of the popular SATA dongles and the WDC Raptor disks..
Serialize access to the SATA channels, the chip messes up if
both channels are used at the same time.
The SiI3112 hereby takes the price as the most crappy SATA chip in
existance by a significant amount.
My advise to our userbase is to avoid this chip like the plague...
Setup decent transfer mode defaults as some BIOS's seem to put in
things that it *knows* doesn't work.
(Note to BIOS writers: stop doing that nonsense, we will get things
working with your crappy HW anyways, and then recommend users to buy
someone else's products that "just works", thankyou.. )
Limit the device transfer mode to ATA100/UDMA5 on generic SATA.
Since we dont know if the user is using a pure SATA device or an
old PATA drive with a SATA converter dongle, we need to limit the
speed used here to cover up the problems with Marvell ATA-SATA bridges
used in lots of SATA products.
This workaround is enabled for all detectable SATA controllers as they
seem to have semilar problems here. One notable exception is all the
Promise pdc2037x chips which just always work (cudos to Promise!).
The altio resource magic no longer worked probably due to other changes
in the kernel. Redo that part so it also fits better into ATAng.
Fix detach so it doesn't panic the system when a pccard device is
yanked.
Approved by: re@
TOC's for the same media!! that borks up GEOM.
Although this looks like bad HW the following patch removes the
chance for GEOM panic'ing.
Approved by: re@
Add a sysctl declaration for hw.ata.atapi_dma, which had gone MIA (though
setting it in loader.conf still worked, it was not visible at runtime)
Approved by: sos
GEOM was not designed to handle media that does not have
a size. Blank CD's are of that type, so cheat and set the
media size to -1. This allows burning to work, but makes
GEOM issue outofrange reads that makes the ATAPI subsystem
spew out a few warnings. GEOM should be tought about this.
GEOM was not designed to handle changing the sectorsize
between opens. Writing multitack CD's with both audio and
data tracks needs to change sector size on the fly. We
cheat here and stuff the current sectorsize into GEOM
private internals. GEOM should grow some clean way for this.
in connection with Marvell based SATA->PATA dongles.
The problem was caused by a combination of things working
together to make it hard to spot...
The ATA driver has always started the ATA command, then build
the SG list for DMA and then finally started the DMA engine.
While this is according to specs, it poses a potential
problem as some controllers apparently do not allow for unlimitted
time between starting the ATA command and starting the DMA engine.
At about the same time as ATAng was committed there were lots
of other changes applied, some of which was locking in parts
that causes the busdma load functions to take significantly
longer to load the SG list.
This pushed the time spent between starting the ATA command and
starting the DMA engine over the hill for some controllers
(especially the Silicon Image DS3112a) and caused what looked
like lost interrupts.
The solution is to get all the SG list work or rather all
busdma related stuff done before we even try to start anything.
This has the nice side effect of seperating busdma out the
way it should be, so the working of the ATA machinery is not
cluttered up with busdma droppings, making the code easier
to read and understand.
was that accessing the status reg could occour too fast, confusing
the logic in the flash part. Could not have been located without:
HW donated by: Jonas Bülow <jonas@servicefactory.se>
provide no methods does not make any sense, and is not used by any
driver.
It is a pretty hard to come up with even a theoretical concept of
a device driver which would always fail open and close with ENODEV.
Change the defaults to be nullopen() and nullclose() which simply
does nothing.
Remove explicit initializations to these from the drivers which
already used them.
Fix to the messages output under CAM_DEBUG_CCB: the summary sense
information (error bits and sense key) is in the error field, not
in the result field, of struct ata_request. No other functional change.
This commit puts the relevant code snippets under #ifdef GONE_IN_5
(rather than #ifndef BURN_BRIDGES) thereby disabling the code now.
The code wil be entirely removed before 5.2 unless we find reasons
why this would be a bad idea.
Approach suggested by: imp
