On Sun, 30 Nov 2003, Matthew Dharm wrote:
> I'm going to pass this one along to Greg, but I think some places in this
> could really use some better comments.  Especially the way you use a single
> buffer inside the loop -- it took me a few minutes to figure out how your
> logic to refresh the buffer with new data worked.
>
> I'm also wondering if the access_xfer_buf() function could use some more
> header comments, stating why this is needed (i.e. spelling out the
> kmap()-isms).

Okay, here it is.  This patch basically just adds comments.  Each routine
that uses the new scatter-gather function gets a brief explanation of
what's going on, and access_xfer_buf() itself gets detailed comments
saying what it's doing and why it's necessary.  You may even want to cut
some of it back; I was pretty verbose.

As a result of the last round of changes, the raw_bulk source files aren't
needed any more.  They can be deleted and the Makefile changed
accordingly.

This patch changes the sddr55 driver to make it use the new scatter-gather
routines.  It has not been tested, but perhaps Andries Brouwer will be
able to try it out.

This patch updates the shuttle_usbat driver to use the new scatter-gather
transfer routines.  The small set of changes needed speaks well for the
original organization of the code.

This has not been tested.

This patch fixes the scatter-gather handling in isd200, replacing an
incorrect routine there with calls to the new routine added in the
previous patch.  It also removes a couple of places where the driver
returned data for commands that shouldn't get any (TEST-UNIT-READY and
START-STOP).

This has not been tested.

This patch adds a small utility routine for storing data in a transfer
buffer.  The next patch uses this routine quite a bit in the isd200
driver.

This patch converts the jumpshot driver to use the new scatter-gather
routines.  It has not been tested.

This patch updates the datafab driver to the new scatter-gather handling,
which makes it safe for systems with >1GByte of memory.
It has been tested by Eduard Hasenleithner.

somebody built his own version of be16_to_cpu(). Such things affect
maintainability.

this fixes the questionable error return codes Paulo noticed
in usblp. I hope I really got all cases now.

This patch updates the sddr09 driver to use the new scatter-gather access
routine.  After installing it, the user who experienced memory access
violations says everything is now working properly.

This patch adds a routine to protocol.c that correctly transfers data to
or from a scatter-gather buffer.  According to Jens Axboe, we've been
using page_address() incorrectly -- it's necessary to use kmap() instead
-- and in fact it doesn't give the desired result when the buffers are
located in high memory.  This could affect anyone using a system with 1 GB
or more of RAM, and one user has already reported such a problem (as you
know).

The three fixup routines in protocol.c and usb.c have been changed to use
the new s-g access routine.  When similar adjustments have been made to
all the subdrivers, we will be able to eliminate the raw_bulk.c source
file entirely.

This patch removes an uncalled subroutine from debug.c.  I only noticed it
when tracking down scatter-gather usage; there didn't seem to be any
reason to repair it since it wasn't being used anywhere.

This patch was written by Andries Brouwer.  It adds to sddr09 the ability
to use 64 MB SmartMedia cards.  I have added a few minor alterations to
make it fit in with my sequence of other patches.

This patch removes some unnecessary scatter-gather code from the sddr09
driver.  In its place a single smaller buffer is re-used each time through
an I/O loop, as opposed to transferring all the data at once.

Andries Brouwer kindly tested this and suggested some improvements to get
it working right.

Added vendor/product ids for Epson, Genius, Microtek, Plustek,
Reflecta, and Visioneer scanners. Removed ids for HP PSC devices as
these are supported by the hpoj userspace driver.

It improves synchronization with hub_irq() and guarantees that the hub
disconnect() routine doesn't exit until the URB's completion routine has
finished.

It changes spin_lock_save() to spin_lock() within the completion routine
and list_del()/INIT_LIST_HEAD() to list_del_init().  It's nothing more
than a minor optimization.

A while back there were some reports of ohci reporting a "bad entry"
diagnostic, mostly with ISO transfers, which were mysterious until
I recently found an easy way to reproduce it.

This patch:

  - Fixes at least one cause of that "bad entry" diagnostic by
    waiting for INTR_WDH before completing ED unlink processing.
    (Else URB unlinking could free TDs on the donelist, so the
    WDH processing would see those entries as "bad".)

  - Merges the patch from Darwin Rambo <drambo@broadcom.com>,
    coping with CPUs that can't do 16 bit accesses (MIPS).

  - Renames a function as start_ed_unlink(), matching its role.

  - Fixes minor debug output issues, including a FIXME to tell
    more info about TDs on the periodic schedule.  And adding
    some missing newlines (makes this patch seem big).

Nobody's complained much about that "bad entry" issue lately, but
if necessary that part would be particularly easy to split out.

Please merge to the next kernel that gets USB patches.

This patch fixes a simple logic error in the routine that copies data from
a driver buffer to a scatter-gather user buffer.

This patch does 2 things (bad, I know -- but they're both pretty small
and pretty obscure).

The CBI specification states in section 2.4.3.1.3 that

	... the host shall also issue Clear Feature for Endpoint Halt
	to the Bulk In pipe if the device reports that the Data In
	command block has Failed.

along with a note in section 2.5.3 that Data Out commands should work
analogously.  This patch does that, along with cleaning up the status
detection logic a little.

For Bulk-only transfers we currently ignore the dResidue field in the CSW,
except for reporting it (without byte-swapping!) in a debug message.  The
patch uses it to compute the residue value returned to the SCSI layer.
Note that the Bulk-only spec allows devices to transfer more data than
they actually use (i.e., they may add padding or ignore stuff) and then
inform the host of this by means of the dResidue value.  The logic used is
simple: our reported residue is the larger of what the device claims and
what we didn't transfer, except that it can't be larger than the total
transfer length.

This patch updates the sdd09 subdriver to make it return Command Failure
with appropriate sense data (rather than Tranport Error) when:

	a MODE-SENSE command requests an unsupported page;

	a CDB includes an unrecognized command code.

This should help prevent confusion and excessive retrying by the SCSI
drivers.

I know Linus is not taking cleanups at this point, but perhaps
you can delete it in your tree. Seems like someone (Oliver?)
fixed all the garbage in old printer.c, so the comment is not
needed anymore.

I reviewed changes, and usblp.c looks correct. I'm doing backport
to 2.4 for Fedora right now.

Just for cdc-acm, it pushes RX URB processing into a tasklet;
and has minor cleanups.

I cc'd Vojtech since he's this driver's maintainer.  If this
checks out, usb-serial will need similar changes.


p.s. the issue is a WARN_ON that tells us:

   >> [<c012046c>] local_bh_enable+0x8c/0x90
   >> [<f8991452>] ppp_asynctty_receive+0x62/0xb0 [ppp_async]
   >> [<c02144f3>] flush_to_ldisc+0xa3/0x120
   >> [<f891f20f>] acm_read_bulk+0xbf/0x140 [cdc_acm]
   >> [<c02684c9>] usb_hcd_giveback_urb+0x29/0x50
   >> [<c027670c>] dl_done_list+0x11c/0x130
   >> [<c0277075>] ohci_irq+0x85/0x170
   >> [<c0268526>] usb_hcd_irq+0x36/0x60
   >> [<c010aeba>] handle_IRQ_event+0x3a/0x70
   >> [<c010b227>] do_IRQ+0x97/0x140
   >> [<c0109624>] common_interrupt+0x18/0x20

into kroah.com:/home/greg/linux/BK/usb-new_drivers-2.6

into kroah.com:/home/greg/linux/BK/usb-new_drivers-2.6

into home.osdl.org:/home/torvalds/v2.5/linux

The pid_alive() check within the loop is incorrect.  If we are within
the tasklist lock and the thread group leader is valid then the thread
chain will be fully intact.

Instead, the check should be _outside_ the loop, since if the group
leader no longer exists, the whole list is gone and we must not try
to access it.

Move the check around, and add comment.

Bug-hunting and fix by Srivatsa Vaddagiri

When doing sync wakeups we must not skip the notification of other cpus
if the task is not on this runqueue.

We need to initialize st->state in tcp_seq_start().  Otherwise
tcp_seq_stop() is run with previous st->state, and it calls the unneeded
unlock etc, causing a kernel crash.

I've got a lot more changes than what's included here.  I've put this
down to the bear minimum to get things working sanely.

Mainly, I just want to get all the people hit by this a chance to use
2.6.0 without having to get our tree. Changes itemized:

 - Fix deadlock possibility in csr.c:read_maps()
 - Fix kmalloc to use ATOMIC in highlevel.c.
 - s/in_interrupt/irqs_disabled/ in ieee1394_transactions.c to fix
   warnings when transactions occured.
 - Introduce a release callback for the host driver and use it correctly.
 - Reorganize the nodemgr probe so we do an initial scan to discover
   devices, check IRM/CycleMaster, then do a final full probe when things
   are kosher. Fixes a problem where device registration and hotplug
   would cause some serious problems when a bus reset was forced in the
   middle of the probe.

into home.osdl.org:/home/torvalds/v2.5/linux