This patch converts the error paths in several functions in ibmphp to use
goto to the end of the function to kill code duplication.

It also kills some memsets of pointer members of a struct where the struct
itself is freed directly after.

This one has 2 modification from the first version: the if is back as
discussed and one extra long line (>> 150 chars) is wrapped now.

These functions are only called from places marked __init or __exit, so
we can mark them also.

John Rose wrote:
> The function release_pcibus_dev() in probe.c defines the release procedure for
> device class pcibus_class.  I want to suggest that this function be scrapped :)
>
> This release function is called in the code path of class_device_unregister().
> The pcibus_class devices aren't currently unregistered anywhere, from what I
> can tell, so this release function is currently unused.  The runtime removal of
> PCI buses from logical partitions on PPC64 requires the unregistration of these
> class devices.  The natural place to do this IMHO is in pci_remove_bus_device()
> in remove.c.

You're right that the class device isn't currently unregistered, and
that was an oversight in the patch I originally sent.  Attatched is a
patch that remedies that situation.  pci_remove_bus_device() *is* the
natural place to unregister the class_dev, and that's just what the
patch does.


> The problem is that this calls pci_destroy_dev(), which calls put() on the same
> "bridge" device that the release function does.  This should only be done once
> in the course of removing a pci_bus, and I doubt that we want to change
> pci_destroy_dev().   The kfree() of the pci_bus struct is also done in both
> pci_remove_bus_device() and release_pcibus_dev().

Yep.  The patch pulls the kfree() out of pci_remove_bus_device() and
calls class_device_unregister() in it's place.  This will put() the
bridge device and free the pci_bus as needed.  put() does need to be
called twice because the bridge device is get()'d twice: once when the
device is registered and once when it's bus device grabs a reference to it.

ibm_unconfigure_device always returns 0, so we do not need to check for a
return value != 0 and can kill the warning. And if the return value is always
the same we don't need a return value. Also some whitespace fixing. And this
time not line wrapped. Really. Absolutely sure. Checked twice.

This patch kills the variable hpcrc from many places. It is used as a
temporary storage for the return code before this is copied to rc which
contains the "real" return code. There are some checks if to copy or not
but at the end they will always have the same value, so we can directly
put the return code in rc and kill hpcrc.

This avoids a return direct before the final return of a function. Better
only modify the return code and fall through to the final return.

Here is a patch for to get cpqphp working with IOAPIC.
My earlier statement that a kernel patch is not needed for 2.6 is
true only when ACPI is enabled.  A similar patch is needed in
pcibios_enable_irq() for 2.4 kernel and I will send it out
later.

The fix is in pirq_enable_irq().  This function is called indirectly
by pci_enable_device().  For device present during boot up, it should
get the proper dev->irq for pcibios_fixup_irqs() has been called to
get the dev->irq from MP table. If the value is still zero, then
this is properly caused by "buggy MP table".  For hot-plug device,
its dev->irq is 0 when the pci_enable_device() is called for it hasn't
gone through the fixup.  Therefore, the code (similiar to the code
in pcibios_fixup_irqs) is needed here.

This kills the explicit setting of rc to -ENODEV in 2 places where it is
not necessary because it will have this value on this path anyway.

Don't use a space directly before '\n' in error/debug messages. This adds
extra code size, causes unneeded line breaks in xterms, noone else does it
and it's just superfluos and ugly.

This also adds a missing ")" in a comment.

This are some whitespace fixes for acpiphp, in most cases killing spaces
before the opening brace of function declarations.

include/linux changes:

- Add dma_pools member to struct device
- Add include/linux/dmapool.h
- Remove pools memober from struct pci_dev
- Replace pci_pool_* functions with macros that map to dma_pool_* functions

- Move drivers/pci/pool.c to drivers/base/pool.c
- Initialize struct device.dma_pools in device_initialize()
- Remove initialization of struct pci_dev.pools from pci_setup_device()

here are some coding style fixes.

The following patch remove irq_resource and dma_resource from pci_dev.  It
appears that the serial pci driver depends on irq_resource, however, it may be
broken portions of an old quirk.  I attempted to maintain the existing behavior
while removing irq_resource.  I changed FL_IRQRESOURCE to FL_NOIRQ.  Russell,
could you provide any comments?  irq_resource and dma_resource are most likely
remnants from when pci_dev was shared with pnp.

This fixes some issues in the dvb subsystem and some nasty things in the
v4l saa7146 driver.

[DVB]
 - dvb-core: aquire -> acquire spelling fix
 - nxt600 frontend: don't send zero-byte messages when probing the PLL
   type
 - Kconfig: add a note that says that the CI of the budget-CI card is
   not actually supported by the budget-CI driver
 - ttusb-dec: Check for presence of crc32 function.  Make unknown types
   of packet less likely to cause packet loss.

[V4L]
 - saa7146: use kernel mint_t()/max_t() instead of homebrewn stuff
 - saa7146: disable video clipping before capturing for sure to prevent
   black pictures
 - saa7146: make sure to disable the right video dma upon device close
 - saa7146: don't free resources if disabling an already disabled video
   overlay

This is a simple fix for some of the problems with bad ACPI frequency values:

  Abort if the frequency field in _PSS is zero, as we're having a
  completely broken ACPI table then.

A more complete overhaul of the acpi-cpufreq driver (where the cause of
the problem lies) is in the latest acpi-test tree, but that's definitely
something to be delayed for 2.6.3 -- and the same is true for the
yet-to-be-written do_div64 conversion.

This fixes my APM problems (without them my laptop, Acer TravelMate
210TEV (Celeron 700, 128 MB RAM), hangs after resuming from APM since
2.6.0-test4).

Modified based on comments from Pavel Machek <pavel@suse.cz>, who
has acked the updated patch.

- Replaces pci.ids with a snapshot from pciids.sf.net from 14 Jan 2004

- Changes gen-devlist.c to truncate long device names rather than reject
  the database
- Changes PCI_NAME_SIZE to 96 (and PCI_NAME_HALF to 43) to allow all
  current pci.ids names to fit
- Modifies gen-devlist.c to truncate at 89 characters rather than 79 -
  allows for two digit instance numbers to be added to the name as well
  while staying within the 96 characters allocated. No names in the
  current pci.ids are any longer than this.
- Modifies names.c to no longer limit device name length when displaying
  both vendor and device name; the truncation is done by gen-devlist.c.

This lets the PPC pci hotplug driver initialize single devices, not just
entire slots.

insert missing iounmap()

This is the pending patch that adds 'address' file to show
PCI-address and a few other minor fixes.
As 2.6.0 is out, I'm resending the patch.
Would you mind taking this?

> > > Thanks.  I had a little time to try your patch today.  Sorry
> > > to report that it isn't working for me.
> > >
> > > I first powered off (successfully the 1st time) a populated slot
> > > and removed and reinserted the card into the same slot.  The slot
> > > powered back up but I was then unable to power it off.  I believe
> > > the following instruction that still exists in power_off_slot()
> > > may be preventing the slot from being powered off more than once.
> > >     func->flags &= (~FUNC_EXISTS);
> > >
> > > I then tried to insert an adapter in an un-populated slot.  For
> > > some reason (which I don't understand yet) there was an enabling
> > > error which I believe caused enable_device() to exit via a path
> > > that bypassed the instruction that sets the FUNC_EXISTS flag.
> > > I was then unable to power off the slot which I believe was due
> > > to the FUNC_EXISTS flag not being set.
> > >
> > > I didn't have time to definitely confirmed the above theories.
> > > I'll take a closer look at this tomorrow unless you are able
> > > to diagnose using my vague clues :)
> >
> > It turns out that both of the above mentioned problems happened
> > because the call to acpiphp_configure_slot() from enable_device()
> > failed after inserting the card.  When this happens enable_device()
> > exits without setting the FUNC_EXISTS flag for any of the slot
> > functions.  Subsequent attempts to power off the same slot fail
> > when power_off_slot() is unable to locate a function with both
> > FUNC_HAS_EJ0 and FUNC_EXISTS flags set.
> >
> > The patch works okay when using a card that allows
> > acpiphp_configure_slot() to succeed but I believe it should
> > be improved to allow the slot to be powered off following
> > device enablement errors.
>
> Thanks for testing and comments.
> I really appreciate it.
>
> This problem turned out to be somewhat fragile state
> transition:
>
> a lifecycle of a slot is (if there's no error)
>
>   function             state
> ----------------------------------------------------
> 0                      nothing
> 1  power_on_slot()  -> SLOT_POWERDON
> 2  enable_device()  -> SLOT_POWEREDON + SLOT_ENABLED
> 3  disable_device() -> SLOT_POWEREDON
> 4  power_off_slot() -> nothing
>
> but if any error occur during enable_device(), slot will remain
> SLOT_POWERDON, but some functions on the card may not have
> FUNC_EXISTS flags, which will eventually prevents powering
> off in power_off_slot(), state transition from 1 to 4 directly.
> I.e, the FUNC_EXISTS flag introduced more states to
> complicate things.
>
> The FUNC_EXISTS flag was introduced after some discussion
> between me and Irene Zubarev, but it has no more meaning
> than that the function has corresponding 'pci_dev' structure.
> So I eliminated the usage of FUNC_EXISTS and the result is
> the patches attached to this mail (for both 2.4 and 2.6.
> I think Greg already applied the 2.4 'cleanup' patch to his tree,
> but it's not in Marcelo's release so I'm re-attaching to
> this mail for anyone interested in this topic.  It's identical
> to the one I posted earlier).
> These patches don't include Gary's patch in his post last week,
> so please apply separately.
>
> Please note that current acpiphp driver cannot handle a
> PCI card that has a PCI-to-PCI bridge on it (support
> for such cards is incomplete).  But if it's treated as
> an error, it should be recoverable anyway.

This just gets rid of a stupid compile warning.

On Wed, Dec 17, 2003 at 04:24:44PM -0800, Greg KH wrote:
> I've applied the pci portions of this patch to my trees and will send it
> on after 2.6.0 is out.

James Bottomley found a bug in it; could you also apply:

This is in case others copy this code (which has already happened...)

tg3.c has a bug where it can find the wrong 5704 peer on a machine with
PCI domains.  The problem is that pci_find_slot() can't distinguish
whether it has the correct domain or not.

This patch fixes that problem by introducing pci_get_slot().

This is needed to show pci bus topology to userspace properly.

When plugging a 33MHz card into a bus that's running at 66MHz, I'd like
to see a better error message than:

acpiphp_glue: notify_handler: unknown event type 0x5 for \_SB_.SBA0.PCI4.S2F0

The following patch would give us:

Device \_SB_.SBA0.PCI4.S2F0 cannot be configured due to a frequency mismatch

which I think is clearer.

The functions are not named *_skel_*, so it seems useful not to call them with
this.

Greg,

As discussed about six or so months ago, we agreed to hold off this
patch until fairly late, due to its ability to catch duplicate PCI
driver names.  Please note that I haven't attempted to reproduce the
problem with recent kernels, and that all ARM kernel patches released
since then have had this patch in.

I'm guessing this will actually be 2.6.1 material since it probably
doesn't show for PCI drivers which are part of the kernel tree.


If pci_register_driver fails, the register the PCI driver structure
will not be registered with the driver model.  pci_register_driver
returns with negative value, and we then attempt to unregister the
driver structure.  This leads to an oops in the driver model.

The driver model does not return the number of devices it successfully
bound the driver to, and neither does pci_register_driver() return
this information.

Therefore, all of the code below is redundant.

(There's a little redundancy left in drivers/pci/pci-driver.c but it
is harmless unlike this block.)