Currently it only is possible to install one system notify handler
per namespace node, but this is not enough for PCI run-time power
management, because we need to install power management notifiers for
devices that already have hotplug notifiers installed.  While in
principle this could be handled at the PCI level, that would be
suboptimal due to the way in which the ACPI-based PCI hotplug code is
designed.

For this reason, modify ACPICA so that it is possible to install more
than one system notify handler per namespace node.  Namely, make
acpi_install_notify_handler(), acpi_remove_notify_handler() and
acpi_ev_notify_dispatch() use a list of system notify handler objects
associated with a namespace node.

Make acpi_remove_notify_handler() call acpi_os_wait_events_complete()
upfront to avoid a situation in which concurrent instance of
acpi_remove_notify_handler() removes the handler from under us while
we're waiting for the event queues to flush.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Use the run_wake flag to mark all devices for which run-time wake-up
events may be generated by the platform.  Introduce a new wake-up
flag, always_enabled, for marking devices that should be permanently
enabled to generate run-time events.  Also, introduce a reference
counter for run-wake devices and a function that will initialize all
of the run-time wake-up fields for given device.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Acked-by: NLen Brown <len.brown@intel.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

ACPI GPEs may map to multiple devices.  The current GPE interface
only provides a mechanism for enabling and disabling GPEs, making
it difficult to change the state of GPEs at runtime without extensive
cooperation between devices.

Add an API to allow devices to indicate whether or not they want
their device's GPE to be enabled for both runtime and wakeup events.

Remove the old GPE type handling entirely, which gets rid of various
quirks, like the implicit disabling with GPE type setting. This
requires a small amount of rework in order to ensure that non-wake
GPEs are enabled by default to preserve existing behaviour.

Based on patches from Matthew Garrett <mjg@redhat.com>.
Signed-off-by: NMatthew Garrett <mjg@redhat.com>
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Apparently, some machines may have problems with PCI run-time power
management if MSIs are used for the native PCIe PME signaling.  In
particular, on the MSI Wind U-100 PCIe PME interrupts are not
generated by a PCIe root port after a resume from suspend to RAM, if
the system wake-up was triggered by a PME from the device attached to
this port.  [It doesn't help to free the interrupt on suspend and
request it back on resume, even if that is done along with disabling
the MSI and re-enabling it, respectively.]  However, if INTx
interrupts are used for this purpose on the same machine, everything
works just fine.

For this reason, add a kernel command line switch allowing one to
request that MSIs be not used for the native PCIe PME signaling,
introduce a DMI table allowing us to blacklist machines that need
this switch to be set by default and put the MSI Wind U-100 into this
table.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

PCIe native PME detection mechanism is based on interrupts generated
by root ports or event collectors every time a PCIe device sends a
PME message upstream.

Once a PME message has been sent by an endpoint device and received
by its root port (or event collector in the case of root complex
integrated endpoints), the Requester ID from the message header is
registered in the root port's Root Status register.  At the same
time, the PME Status bit of the Root Status register is set to
indicate that there's a PME to handle.  If PCIe PME interrupt is
enabled for the root port, it generates an interrupt once the PME
Status has been set.  After receiving the interrupt, the kernel can
identify the PCIe device that generated the PME using the Requester
ID from the root port's Root Status register. [For details, see PCI
Express Base Specification, Rev. 2.0.]

Implement a driver for the PCIe PME root port service working in
accordance with the above description.

Based on a patch from Shaohua Li <shaohua.li@intel.com>.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Add function pci_check_pme_status() that will check the PME status
bit of given device and clear it along with the PME enable bit.  It
will be necessary for PCI run-time power management.

Based on a patch from Shaohua Li <shaohua.li@intel.com>
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Make pci_bridge_check_ranges() store the PCI_PREF_RANGE_TYPE_64 in
addition to IORESOURCE_MEM_64.  Just like pci_read_bridge_bases().
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Handle the case where the slot bridge that doesn't get a pre-allocated
resource big enough to handle its child resources..  For example pcie
devices need 256M, but the bridge only gets 2M preallocated.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Move bus_size_bridges and assign resources out of pciehp_add_bridge()
and do them all together, one time, including slot bridge, to avoid to
calling assign resources several times when there are several bridges
under the slot bridge.  Using pci_assign_unassigned_bridge_resources.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Reviewed-by: NAlex Chiang <achiang@hp.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

For use by pciehp.

pci_setup_bridge() will not check enabled for the slot bridge, otherwise
update res is not updated to bridge BAR.  That is, bridge is already
enabled for port service.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Reviewed-by: NAlex Chiang <achiang@hp.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

BIOS separates IO ranges between several IOHs, and on some slots, BIOS assigns
resources to a bridge, but stops assigning resources to the device under that
bridge, because the device needs a big resource.

So:
  1. allocate resources and record the failed device resources
  2. clear the BIOS assigned resources of the parent bridge of failing device
  3. go back and call pci assign unassigned
  4. if it still fails, go up the tree, clear more bridges. and try again
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

When clearing leaf bridge resources, trying to get a big enough one, we
could shrink the bridge if there is no resource under it.  Confirm
against the old resource side to make sure we're increasing the
allocation.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

We already track unassigned resources in struct resource, and this
prevents us from overwriting resource flags and info in the unassigned
case.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

This allows us to track failed allocations for later re-trying with
reallocation.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Reviewed-by: NAlex Chiang <achiang@hp.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

We use this in later patches to free resrouce ranges for reassignment in
an effort to support a wider variety of PCI topologies.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Reviewed-by: NAlex Chiang <achiang@hp.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

check ioremap() return value.

Cc: <stable@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Add header file to fix build error:

drivers/pci/hotplug/ibmphp_hpc.c:135: error: implicit declaration of function 'init_MUTEX'
drivers/pci/hotplug/ibmphp_hpc.c:136: error: implicit declaration of function 'init_MUTEX_LOCKED'
drivers/pci/hotplug/ibmphp_hpc.c:797: error: implicit declaration of function 'down'
drivers/pci/hotplug/ibmphp_hpc.c:807: error: implicit declaration of function 'up'
Signed-off-by: NRandy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Yinghai pointed out that the new pci_scan_slot() crashes when called
on an ARI-capable slot that is empty.  Fix this by exiting early from
pci_scan_slot if there is no device in the slot.

Also make next_ari_func() robust against devices not existing in case
the ARI capability is corrupt.  ARI also requires that the devices be
listed in order, so if we find a function listed that is out of order,
stop scanning to prevent loops.
Signed-off-by: NMatthew Wilcox <matthew@wil.cx>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Stanse found an ommitted pci_dev_put on one error path in
cpcihp_generic_init. The path is taken on !dev, but also when
dev->hdr_type != PCI_HEADER_TYPE_BRIDGE. However it omits to
pci_dev_put on the latter.

As it is fine to pass NULL to pci_dev_put, put it in there
uncoditionally.
Signed-off-by: NJiri Slaby <jslaby@suse.cz>
Cc: Scott Murray <scott@spiteful.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

The ISDN4Linux HiSax driver family contains the last remaining users
of the deprecated pci_find_device() function. This patch creates a
private copy of that function in HiSax, and removes the now unused
global function together with its controlling configuration option,
CONFIG_PCI_LEGACY.
Signed-off-by: NTilman Schmidt <tilman@imap.cc>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Don't print out resources without flags to avoid cluttering up the debug
output.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

This is a good cleanup in itself, and makes it easier to modify specific
resource types in later code.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Acked-by: NLinus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

ibmphp driver currently maps only 1KB of ebda memory area into kernel address
space during driver initialization. This causes kernel oops when the driver is
modprobe'd and it accesses memory beyond 1KB within ebda segment. The first
byte of ebda segment actually stores the length of the ebda region in
Kilobytes. Hence make use of the length parameter and map the entire ebda
region.
Signed-off-by: NChandru Siddalingappa <chandru@linux.vnet.ibm.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Stanse found a cut&pasted memory leak in pciehp_queue_pushbutton_work
and shpchp_queue_pushbutton_work. info is not freed/assigned on all
paths. Fix that.
Reviewed-by: NKenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: NJiri Slaby <jslaby@suse.cz>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Now that we return the new resource start position, there is no
need to update "struct resource" inside the align function.
Therefore, mark the struct resource as const.

Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: NDominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

As suggested by Linus, align functions should return the start
of a resource, not void. An update of "res->start" is no longer
necessary.

Cc: Bjorn Helgaas <bjorn.helgaas@hp.com>
Cc: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: NDominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Currently, drivers/pci/quirks.c is built unconditionally, but if
CONFIG_PCI_QUIRKS is unset, the only things actually built in this
file are definitions of global variables and empty functions (due to
the #ifdef CONFIG_PCI_QUIRKS embracing all of the code inside the
file).  This is not particularly nice and if someone overlooks
the #ifdef CONFIG_PCI_QUIRKS, build errors are introduced.

To clean that up, move the definitions of the global variables in
quirks.c that are always built to pci.c, move the definitions of
the empty functions (compiled when CONFIG_PCI_QUIRKS is unset) to
headers (additionally make these functions static inline) and modify
drivers/pci/Makefile so that quirks.c is only built if
CONFIG_PCI_QUIRKS is set.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

No longer needed and causes build breakage.
Reported-by: NStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Add the 8.0 GT/s speed.
Signed-off-by: NMatthew Wilcox <willy@linux.intel.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Take advantage of some gaps in the table to fit in support for AGP speeds.
Signed-off-by: NMatthew Wilcox <willy@linux.intel.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Both PCIe and PCI-X bridges report their secondary bus speed in their
respective capabilities.
Signed-off-by: NMatthew Wilcox <willy@linux.intel.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Move the max_bus_speed and cur_bus_speed into the pci_bus.  Expose the
values through the PCI slot driver instead of the hotplug slot driver.
Update all the hotplug drivers to use the pci_bus instead of their own
data structures.
Signed-off-by: NMatthew Wilcox <willy@linux.intel.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

These enums must not overlap anyway, since we only have a single
pci_bus_speed_strings array.  Use a single enum, and move it to
pci.h.  Add 'SPEED' to the pcie names to make it clear what they are.
Signed-off-by: NMatthew Wilcox <willy@linux.intel.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

The Alternate Routing-ID Interpretation capability allows a single device
to have up to 256 functions.  They can be populated sparsely, so the
current technique of scanning every eighth function is not guaranteed
to find them all.  By introducing a 'next_fn' function pointer, we can
use the linked list of functions in the ARI capability to scan all the
functions which exist.

We can then speed up the pci_scan_slot by skipping the scan of subsequent
devfns for PCIe devices which are the direct children of Root Ports or
Downstream Ports.  These devices are only permitted to implement device
0, unless they are ARI devices, in which case they'll be scanned by the
ARI code above.
Signed-off-by: NMatthew Wilcox <willy@linux.intel.com>
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Commit 84b79f8d (drm/i915: Fix crash
while aborting hibernation) attempted to fix a regression introduced
by commit cbda12d7 (drm/i915:
implement new pm ops for i915), but it went too far trying to split
the freeze/suspend and resume/thaw parts of the code.  As a result,
it introduced another regression, which only is visible on some systems.

Fix the problem by merging i915_drm_suspend() with
i915_drm_freeze() and moving some code from i915_resume()
into i915_drm_thaw(), so that intel_opregion_free() and
intel_opregion_init() are also executed in the freeze and thaw code
paths, respectively.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Reported-and-tested-by: NPedro Ribeiro <pedrib@gmail.com>
Tested-by: NTino Keitel <tino.keitel@tikei.de>
Acked-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Generaltouch protocol allows for coordinates in [0, 0xffff] range and
there are devices reporting coordinates as high as 0x7fff so let's update
the driver to reflect that.
Signed-off-by: NRoy Yin <yhch@generaltouch.com>
Signed-off-by: NDmitry Torokhov <dtor@mail.ru>

Relying on overflow/wrap around isn't exact because if you wrap far
enough, you get back to "valid" values.
Reported-by: NThorsten Pohlmann <pohlmann@tetronik.com>
Signed-off-by: NMike Frysinger <vapier@gentoo.org>
Signed-off-by: NWim Van Sebroeck <wim@iguana.be>
Cc: stable <stable@kernel.org>

When polled input device is opened and closed and there are no other
users of polled device, the workqueue is created and destroyed in
every open / close operation. It is probable that at some point
dynamic allocation of internal parts of the workqueue cause changes to the
workqueue.

When a work is queued to the workqueue the work struct contains pointers
to the workqueue data. If the workqueue has been changed and the work
has never been queued to the new workqueue, work-struct contains pointers
to the non-existing workqueue. This will cause crash at the work
cancellation during device close since cancellation of a work assumes
that the workqueue exists.

To prevent that, work struct is cleaned up at device close. This keeps
work struct clean for the next use.
Signed-off-by: NSamu Onkalo <samu.p.onkalo@nokia.com>
Signed-off-by: NDmitry Torokhov <dtor@mail.ru>

The temperature and voltage limits currently set on these boards are
too conservative and will cause the driver to stop the net device
erroneously in some systems.

Based on a review of the chip datasheets and advice from the designer
of these boards:

- Raise the maximum board temperatures to the specified maximum ambient
  temperatures for their PHYs plus the expected temperature bias of the
  board
- Raise the maximum controller temperature to 90 degrees
- Lower the minimum temperatures to 0 degrees
- Widen the voltage tolerances to at least +/- 10%
Signed-off-by: NBen Hutchings <bhutchings@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

efx_mcdi_poll() uses positive error numbers, matching the MCDI
protocol.  It must negate the result of efx_mcdi_poll_reboot() which
returns the usual negative error numbers.
Signed-off-by: NBen Hutchings <bhutchings@solarflare.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>