into mulgrave.(none):/home/jejb/BK/scsi-for-linus-2.5

Hi!

Since the introduction of USER_HZ the SG_[GS]ET_TIMEOUT ioctls may have
a serious BUG as userspace uses a different HZ from the HZ in kernelspace.

In x86 HZ=1000 and USER_HZ=100, resulting in confusing timouts as the
kernel measures time 10 times as fast as userspace.

This patch is an attempt to fix this by transforming USER_HZ based timing to
HZ based timing before storing it in timeout. To make sure that SG_GET_TIMEOUT
and SG_SET_TIMEOUT behave consistently a field timeout_user is added which
stores the exact value that's passed by SG_SET_TIMEOUT and it's returned on
SG_GET_TIMEOUT.

Rolf Fokkens
fokkensr@fokkensr.vertis.nl

P.S. this is the second post of this patch

But it's called from scsi_add_lun()->scsi_alloc_sdev() before the type
is known.  The type is -1 all the time in scsi_initialise_merge_fn()
and scsi always bounces.

This patch makes it do the right thing - just enable block-highmem for
all scsi devices.

Jens had this to say:

"I guess that block-highmem has been around long enough, that I can
 use the term 'historically' at least in the kernel sense :-)

 This extra check was added for IDE because each device type driver
 (ide-disk, ide-cd, etc) needed to be updated to not assume virtual
 mappings of request data was valid.  I only did that for ide-disk,
 since this is the only one where bounce buffering really hurt
 performance wise.  So while ide-cd and ide-tape etc could have been
 updated, I deemed it uninteresting and not worthwhile.

 Now, this was just carried straight into the scsi counter parts,
 conveniently, because of laziness.  A quick glance at sr shows that it
 too can aviod bouncing easily (no changes needed).  st may need some
 changes, though.  So again, for scsi it was a matter of not impacting
 existing code in 2.4 too much.

 So TYPE_DISK check can be killed in 2.5 if someone does the work of
 checking that it is safe.  I'm not so sure it will make eg your SCSI
 CD-ROM that much faster :-)"

In two drivers a wrong size of memory was allocated for cpufreq_driver: as
it must include NR_CPUS times a struct cpufreq_policy (and not struct
cpufreq_freqs). Thanks to Petr Vandrovec for this patch.

QDIO_BH was never actually used anyway, and won't do much good now BHs
are gone.

Beginning of proper refcounting.  New helpers introduced, several drivers
switched to using them for dynamic allocation of gendisks.  Once everything
is switched (and that will be way easier than per-drive gendisks series)
we will be able to add sane reference counts on gendisk, at which point
we can safely put pointer to gendisk in struct block_device / struct request
and we had pretty much won - from that point it's pretty straightforward
crapectomy in drivers.

char *major_name replaced with char disk_name[16];  All uses of ->major_name
replaced with those of ->disk_name and (obviously) simplified big way.  Bunch
of arrays, kmallocs, etc. is gone.

... now it can be done.  We also drop almost all arguments - there is only
one caller and everything is determined by the first argument.

Cleans the handling of partitioning up.  More or less the same story as with
other drivers...

corrected handling of sizes.  Ugh.

get_gendisk() now takes dev_t (instead of kdev_t) and gets an additional
argument - int *part.  Set to 0 for non-partitioned, partition number
for partititoned.  Callers updated.  Yes, I hate passing return values
that way ;-/  We need that since old "minor(dev) - disk->first_minor"
doesn't work for stuff with non-trivial numbers (e.g. floppy) and
get_gendisk() really has to return both gendisk and partition number.
Fortunately, amount of callers of gendisk() is about to drop RSN big way...

The list used to generate /proc/partitions turned into list_head one;
we also restore the old order of elements (originally we added to the end
of list; recent changes had reverted that, now we are back to original
order).

Make IDE set the dma mask to full 32-bit dma.

Just for niceness, loop should clear queue queuedata when it exits.

These days we only require a clear of BIO_UPTODATE on -EIO, we don't set
it on success. This breaks raid5. It appears to clear BIO_UPTODATE fine
but doesn't start out with it set.

This makes loop honor the queue restrictions by basically stacking all
of those, and mirroring the merge_bvec_fn() on the target queue. It also
switches loop to use per-loop device queues, since that is the only sane
way to do this from a performance POV. Also, in principle I find it to
be much nicer if every distinct block device has its own queue.

There's really no reason to BUG() out on a bio that is too big, the
gentleman thing to do would be to print a warning and just end the bio
with -EIO quietly.

Add a function to set queue merge_bvec_fn to mimic the rest of the api,
and also add documentation for that and blk_queue_prep_rq().

The might_sleep() thing caught ide, which calls request_irq() with a
lock held. It can be argued that this is a bad thing, however I think it
can also validly be argued that requesting an irq should not be a
blocking operation. This might even remove some driver bugs where usage
count is not incremented during init...

It can also be argued, that the very first irq requests cannot be
blocking for io anyways, for good reason :-)

Too trivial to see the first time when debugging on weekends ;-))

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

into uml.karaya.com:/home/jdike/linux/2.5/highmem-2.5

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

into home.transmeta.com:/home/torvalds/v2.5/linux

This fixes a procfs crash noticed by Anton Blanchard.

The procfs code can have a reference even to an already exited task, so
it needs to follow special rules accessing p->sig.  The atomic-signals
patch made this bug happen at a much higher frequency, but procfs i
believe was buggy ever since, it potentially used the freed signal
structure - which just did not result in a crash like it does today.

The proper fix is to take the tasklist read-lock in
collect_sigign_sigcatch(), this excludes __exit_sighand() freeing the
signal structure prematurely.

due to synchronize_irq() becoming a no-op when !CONFIG_SMP.

uml.ld.S.

into nuts.ninka.net:/home/davem/src/BK/net-2.5

into nuts.ninka.net:/home/davem/src/BK/net-2.5

With this we avoid doing skb_clone on skbs that will not be kept on
unacked lists.