Rework the common ACPI device hotplug code so that it is suitable
for PCI host bridge hotplug and switch the PCI host bridge scan
handler to using the common hotplug code.

This allows quite a few lines of code that are not necessary any more
to be dropped from the PCI host bridge scan handler and removes
arbitrary differences in behavior between PCI host bridge hotplug
and ACPI-based hotplug of other components, like CPUs and memory.

Also acpi_device_hotplug() can be static now.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: NMika Westerberg <mika.westerberg@linux.intel.com>

Modify the common ACPI device hotplug code to always queue up the
same function, acpi_device_hotplug(), using acpi_hotplug_execute()
and make the PCI host bridge hotplug code use that function too for
device hot removal.

This allows some code duplication to be reduced and a race condition
where the relevant ACPI handle may become invalid between the
notification handler and the function queued up by it via
acpi_hotplug_execute() to be avoided.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: NMika Westerberg <mika.westerberg@linux.intel.com>

Modify the ACPI namespace scanning code to register a struct
acpi_device object for every namespace node representing a device,
processor and so on, even if the device represented by that namespace
node is reported to be not present and not functional by _STA.

There are multiple reasons to do that.  First of all, it avoids
quite a lot of overhead when struct acpi_device objects are
deleted every time acpi_bus_trim() is run and then added again
by a subsequent acpi_bus_scan() for the same scope, although the
namespace objects they correspond to stay in memory all the time
(which always is the case on a vast majority of systems).

Second, it will allow user space to see that there are namespace
nodes representing devices that are not present at the moment and may
be added to the system.  It will also allow user space to evaluate
_SUN for those nodes to check what physical slots the "missing"
devices may be put into and it will make sense to add a sysfs
attribute for _STA evaluation after this change (that will be
useful for thermal management on some systems).

Next, it will help to consolidate the ACPI hotplug handling among
subsystems by making it possible to store hotplug-related information
in struct acpi_device objects in a standard common way.

Finally, it will help to avoid a race condition related to the
deletion of ACPI namespace nodes.  Namely, namespace nodes may be
deleted as a result of a table unload triggered by _EJ0 or _DCK.
If a hotplug notification for one of those nodes is triggered
right before the deletion and it executes a hotplug callback
via acpi_hotplug_execute(), the ACPI handle passed to that
callback may be stale when the callback actually runs.  One way
to work around that is to always pass struct acpi_device pointers
to hotplug callbacks after doing a get_device() on the objects in
question which eliminates the use-after-free possibility (the ACPI
handles in those objects are invalidated by acpi_scan_drop_device(),
so they will trigger ACPICA errors on attempts to use them).
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: NMika Westerberg <mika.westerberg@linux.intel.com>

The PCI host bridge scan handler installs its own notify handler,
handle_hotplug_event_root(), by itself.  Nevertheless, the ACPI
hotplug framework also installs the common notify handler,
acpi_hotplug_notify_cb(), for PCI root bridges.  This causes
acpi_hotplug_notify_cb() to call _OST method with unsupported
error as hotplug.enabled is not set.

To address this issue, introduce hotplug.ignore flag, which
indicates that the scan handler installs its own notify handler by
itself.  The ACPI hotplug framework does not install the common
notify handler when this flag is set.
Signed-off-by: NToshi Kani <toshi.kani@hp.com>
[rjw: Changed the name of the new flag]
Cc: 3.9+ <stable@vger.kernel.org> # 3.9+
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

If a PCI host bridge cannot be enumerated due to an error in
pci_acpi_scan_root(), its ACPI device object's driver_data field
has to be cleared by acpi_pci_root_add() before freeing the
object pointed to by that field, or some later acpi_pci_find_root()
checks that should fail may succeed and cause quite a bit of
confusion to ensue.

Fix acpi_pci_root_add() to clear device->driver_data before
returning an error code as appropriate.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NToshi Kani <toshi.kani@hp.com>

There are two different interfaces for queuing up work items on the
ACPI hotplug workqueue, alloc_acpi_hp_work() used by PCI and PCI host
bridge hotplug code and acpi_os_hotplug_execute() used by the common
ACPI hotplug code and docking stations.  They both are somewhat
cumbersome to use and work slightly differently.

The users of alloc_acpi_hp_work() have to submit a work function that
will extract the necessary data items from a struct acpi_hp_work
object allocated by alloc_acpi_hp_work() and then will free that
object, while it would be more straightforward to simply use a work
function with one more argument and let the interface take care of
the execution details.

The users of acpi_os_hotplug_execute() also have to deal with the
fact that it takes only one argument in addition to the work function
pointer, although acpi_os_execute_deferred() actually takes care of
the allocation and freeing of memory, so it would have been able to
pass more arguments to the work function if it hadn't been
constrained by the connection with acpi_os_execute().

Moreover, while alloc_acpi_hp_work() makes GFP_KERNEL memory
allocations, which is correct, because hotplug work items are
always queued up from process context, acpi_os_hotplug_execute()
uses GFP_ATOMIC, as that is needed by acpi_os_execute().  Also,
acpi_os_execute_deferred() queued up by it waits for the ACPI event
workqueues to flush before executing the work function, whereas
alloc_acpi_hp_work() can't do anything similar.  That leads to
somewhat arbitrary differences in behavior between various ACPI
hotplug code paths and has to be straightened up.

For this reason, replace both alloc_acpi_hp_work() and
acpi_os_hotplug_execute() with a single interface,
acpi_hotplug_execute(), combining their behavior and being more
friendly to its users than any of the two.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: NMika Westerberg <mika.westerberg@linux.intel.com>

Since _handle_hotplug_event_root() is run from the ACPI hotplug
workqueue, it doesn't need to queue up a work item to eject a PCI
host bridge on the same workqueue.  Instead, it can just carry out
the eject by calling acpi_bus_device_eject() directly, so make that
happen.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Notice that handle_root_bridge_removal() is the only user of
acpi_bus_hot_remove_device(), so it doesn't have to be exported
any more and can be made internal to the ACPI core.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NToshi Kani <toshi.kani@hp.com>

Simplify handle_root_bridge_removal() and acpi_eject_store() by
getting rid of struct acpi_eject_event and passing device objects
directly to async routines executed via acpi_os_hotplug_execute().
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NToshi Kani <toshi.kani@hp.com>

It is required to do get_device() on the struct acpi_device in
question before passing it to acpi_bus_hot_remove_device() through
acpi_os_hotplug_execute(), because acpi_bus_hot_remove_device()
calls acpi_scan_hot_remove() that does put_device() on that
object.

The ACPI PCI root removal routine, handle_root_bridge_removal(),
doesn't do that, which may lead to premature freeing of the
device object or to executing put_device() on an object that
has been freed already.

Fix this problem by making handle_root_bridge_removal() use
get_device() as appropriate.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NToshi Kani <toshi.kani@hp.com>
Cc: All applicable <stable@vger.kernel.org>

This updates _OSC-related messages to be more human-readable.  We now always
show the features we declare support for (this was previously invisible) as
well as the features we are granted control of.

Typical changes:

  -acpi PNP0A08:00: Requesting ACPI _OSC control (0x1d)
  -acpi PNP0A08:00: ACPI _OSC control (0x1d) granted
  +acpi PNP0A08:00: _OSC: OS supports [ExtendedConfig ASPM ClockPM Segments MSI]
  +acpi PNP0A08:00: _OSC: OS now controls [PCIeHotplug PME AER PCIeCapability]
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Test the services we support (extended config space, ASPM, MSI) separately
so we can give a better message.  Previously we said "Unable to request
_OSC control..."; now we'll say "we support %#02x but %#02x are required".
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Test "pcie_ports_disabled" separately so we can give a better message.
Previously we said "Unable to request _OSC control..."; now we'll
say "PCIe port services disabled; not requesting _OSC control".

"pcie_ports_disabled" is true when CONFIG_PCIEPORTBUS=n or we boot
with "pcie_ports=compat".
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

If the _OSC support notification fails, we will never request control
(because "support == OSC_PCI_SEGMENT_GROUPS_SUPPORT", which doesn't include
all the features in ACPI_PCIE_REQ_SUPPORT), so we can return early to
simplify the code.
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Previously, we ran _OSC once to tell the platform that we support
PCI Segment Groups, then we ran it again if we supported any additional
features (ASPM, MSI, or extended config space).  I don't think it's
necessary to run it twice, since we can easily build the complete
mask of features we support before running _OSC the first time.

We run _OSC again later when requesting control of PCIe features;
that's unaffected by this change.
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Previously we used "flags" for both:

  - the bitmask of features we support (segments, ASPM, MSI, etc.), and
  - the bitmask of features we want to control (native hotplug, AER, etc.)

To reduce confusion, this patch splits this into two variables:
"support" is the bitmask of features we support, and "control" is the
bitmask of features we want to control.  No functional change.
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

This doesn't change any of the _OSC code; it just moves it out into
a new function so it doesn't clutter acpi_pci_root_add() so much.  This
also enables future simplifications.
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

There's no need to check whether _OSC exists here; we eventually
call acpi_evaluate_object(..., "_OSC", ...), and that will fail
gracefully if _OSC doesn't exist.
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Make PCI Host Bridge _OSC #defines more consistent.  No functional change.
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

OSC_QUERY_TYPE isn't a "type"; it's an index into the _OSC Capabilities
Buffer of DWORDs.  Rename OSC_QUERY_TYPE, OSC_SUPPORT_TYPE, and
OSC_CONTROL_TYPE to OSC_QUERY_DWORD, etc., to make this clear.
No functional change.
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

This fixes the problem of acpiphp claiming slots that should be managed
by pciehp, which may keep ExpressCard slots from working.

The acpiphp driver claims PCIe slots unless the BIOS has granted us
control of PCIe native hotplug via _OSC.  Prior to v3.10, the acpiphp
.add method (add_bridge()) was always called *after* we had requested
native hotplug control with _OSC.

But after 3b63aaa7 ("PCI: acpiphp: Do not use ACPI PCI subdriver
mechanism"), which appeared in v3.10, acpiphp initialization is done
during the bus scan via the pcibios_add_bus() hook, and this happens
*before* we request native hotplug control.

Therefore, acpiphp doesn't know yet whether the BIOS will grant control,
and it claims slots that we should be handling with native hotplug.

This patch requests native hotplug control earlier, so we know whether
the BIOS granted it to us before we initialize acpiphp.

To avoid reintroducing the ASPM issue fixed by b8178f13 ('Revert
"PCI/ACPI: Request _OSC control before scanning PCI root bus"'), we run
_OSC earlier but defer the actual ASPM calls until after the bus scan is
complete.

Tested successfully by myself.

[bhelgaas: changelog, mark for stable]
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=60736Signed-off-by: NNeil Horman <nhorman@tuxdriver.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NYinghai Lu <yinghai@kernel.org>
CC: stable@vger.kernel.org	# v3.10+
CC: Len Brown <lenb@kernel.org>
CC: "Rafael J. Wysocki" <rjw@sisk.pl>

When hot-adding an ACPI host bridge, use
pci_assign_unassigned_root_bus_resources() instead of
pci_assign_unassigned_bus_resources().

The former is more aggressive and will release and reassign existing
resources if necessary.  This is safe at hot-add time because no drivers
are bound to devices below the new host bridge yet.

[bhelgaas: changelog, split __init changes out for reviewability]
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

We currently enable PCI bridges after scanning a bus and assigning
resources.  This is often done in arch code.

This patch changes this so we don't enable a bridge until necessary, i.e.,
until we enable a PCI device behind the bridge.  We do this in the generic
pci_enable_device() path, so this also removes the arch-specific code to
enable bridges.

[bhelgaas: changelog]
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

Use dev_printk(), acpi_handle_print(), and pr_xxx() to print messages
in pci_root.c.

[bhelgaas: fold in dev_printk() changes, use dev_printk() in
handle_root_bridge_insertion()]
Signed-off-by: NJiang Liu <jiang.liu@huawei.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Len Brown <lenb@kernel.org>

Now the global list acpi_pci_roots pci_root.c is useless, remove it.
Signed-off-by: NJiang Liu <jiang.liu@huawei.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Len Brown <lenb@kernel.org>

[bhelgaas: split out from acpi_handle_printk() changes]
Signed-off-by: NJiang Liu <jiang.liu@huawei.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

[bhelgaas: split out from acpi_pci_roots removal]
Signed-off-by: NJiang Liu <jiang.liu@huawei.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

When a PCI host bridge device receives a Bus Check notification, we
must re-enumerate starting with the bridge to discover changes (devices
that have been added or removed).

Prior to 668192b6 ("PCI: acpiphp: Move host bridge hotplug to
pci_root.c"), this happened in _handle_hotplug_event_bridge().  After that
commit, _handle_hotplug_event_bridge() is not installed for host bridges,
and the host bridge notify handler, _handle_hotplug_event_root() did not
re-enumerate.

This patch adds re-enumeration to _handle_hotplug_event_root().

This fixes cases where we don't notice the addition or removal of
PCI devices, e.g., the PCI-to-USB ExpressCard in the bugzilla below.

[bhelgaas: changelog, references]
Reference: https://lkml.kernel.org/r/CAAh6nkmbKR3HTqm5ommevsBwhL_u0N8Rk7Wsms_LfP=nBgKNew@mail.gmail.com
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=57961Reported-by: NGavin Guo <tuffkidtt@gmail.com>
Tested-by: NGavin Guo <tuffkidtt@gmail.com>
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Reviewed-by: NJiang Liu <jiang.liu@huawei.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
CC: stable@vger.kernel.org	# v3.9+

We should check the acpi_resource_to_address64() return value, which
also removes the need to validate the resource type beforehand.
No functional change.

Found by Coverity (CID 113815).
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Both sub-drivers of the "PCI Root Bridge ("pci_bridge")" driver, "acpiphp"
and "pci_slot", have been converted to hook directly into the PCI core.

With the conversions there are no remaining usages of the 'struct
acpi_pci_driver' list based infrastructure.  This patch removes it.
Signed-off-by: NMyron Stowe <myron.stowe@redhat.com>
Signed-off-by: NJiang Liu <jiang.liu@huawei.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Reviewed-by: NYinghai Lu <yinghai@kernel.org>
Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
Cc: Toshi Kani <toshi.kani@hp.com>

This reverts commit 8c33f51d.

Conflicts:
	drivers/acpi/pci_root.c

This commit broke some pre-1.1 PCIe devices by leaving them with
ASPM enabled.  Previously, we had disabled ASPM on these devices
because many of them don't implement it correctly (per 149e1637).

Requesting _OSC control early means that aspm_disabled may be set
before we scan the PCI bus and configure link ASPM state.  But the
ASPM configuration currently skips the check for pre-PCIe 1.1 devices
when aspm_disabled is set, like this:

    acpi_pci_root_add
      acpi_pci_osc_support
        if (flags != base_flags)
          pcie_no_aspm
            aspm_disabled = 1
      pci_acpi_scan_root
        ...
          pcie_aspm_init_link_state
            pcie_aspm_sanity_check
              if (!aspm_disabled)
                /* check for pre-PCIe 1.1 device */

Therefore, setting aspm_disabled early means that we leave ASPM enabled
on these pre-PCIe 1.1 devices, which is a regression for some devices.

The best fix would be to clean up the ASPM init so we can evaluate
_OSC before scanning the bug (that way boot-time and hot-add discovery
will work the same), but that requires significant rework.

For now, we'll just revert the _OSC change as the lowest-risk fix.

Reference: https://bugzilla.kernel.org/show_bug.cgi?id=55211Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NYinghai Lu <yinghai@kernel.org>
CC: stable@vger.kernel.org	# v3.8+

Found problem on system that firmware that could handle pci aer.
Firmware get error reporting after pci injecting error, before os boots.
But after os boots, firmware can not get report anymore, even pci=noaer
is passed.

Root cause: BIOS _OSC has problem with query bit checking.
It turns out that BIOS vendor is copying example code from ACPI Spec.
In ACPI Spec 5.0, page 290:

	If (Not(And(CDW1,1))) // Query flag clear?
	{	// Disable GPEs for features granted native control.
		If (And(CTRL,0x01)) // Hot plug control granted?
		{
			Store(0,HPCE) // clear the hot plug SCI enable bit
			Store(1,HPCS) // clear the hot plug SCI status bit
		}
	...
	}

When Query flag is set, And(CDW1,1) will be 1, Not(1) will return 0xfffffffe.
So it will get into code path that should be for control set only.
BIOS acpi code should be changed to "If (LEqual(And(CDW1,1), 0)))"

Current kernel code is using _OSC query to notify firmware about support
from OS and then use _OSC to set control bits.
During query support, current code is using all possible controls.
So will execute code that should be only for control set stage.

That will have problem when pci=noaer or aer firmware_first is used.
As firmware have that control set for os aer already in query support stage,
but later will not os aer handling.

We should avoid passing all possible controls, just use osc_control_set
instead.
That should workaround BIOS bugs with affected systems on the field
as more bios vendors are copying sample code from ACPI spec.
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Cc: stable@vger.kernel.org
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

During merging the PCI tree with the PM/ACPI tree, Linus noticed
that we don't use the same lock using patten about ACPI PCI root as
acpiphp.

Here apply the same locking patten, and we need to execute
acpi_bus_hot_remove_device() via acpi_os_hotplug_execute()
as it also holds acpi_scan_lock.

[rjw: Changelog]
Reported-by: NLinus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
No-objection-from: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

Previously, we cached _PRT (PCI routing table, ACPI 5.0 sec 6.2.12)
contents and associated each _PRT entry with a PCI bus number.  The bus
number association means dependencies on PCI device enumeration and bus
number assignment, as well as on the PCI/ACPI binding process.

After 4f535093 ("PCI: Put pci_dev in device tree as early as possible"),
these dependencies caused the IRQ issues reported by Peter:

    pci 0000:00:1e.0: PCI bridge to [bus 09] (subtractive decode)
    pci 0000:00:1e.0: can't derive routing for PCI INT A
    snd_ctxfi 0000:09:02.0: PCI INT A: no GSI - using ISA IRQ 5
    irq 18: nobody cared (try booting with the "irqpoll" option)

This patch removes _PRT caching.  Instead, we evaluate _PRT as needed
in the pci_enable_device() path.  This also removes the dependency on
PCI bus numbers: we can simply look at the _PRT associated with each
bridge as we walk upstream toward the root.

[bhelgaas: changelog]
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=53561Reported-and-tested-by: NPeter Hurley <peter@hurleysoftware.com>
Suggested-by: NBjorn Helgaas <bhelgaas@google.com>
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

Make the ACPI PCI root bridge driver use struct acpi_scan_handler
for representing the object used to enumerate the PCI busses under
PCI host bridges found in the ACPI namespace (and to tear down data
structures representing the bus and devices on it before
unregistering the host bridges' ACPI device nodes).

This simplifies the code slightly and reduces the kernel's memory
footprint by avoiding the registration of a struct device_driver
object with the driver core and creation of its sysfs directory
which is unnecessary.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: NYinghai Lu <yinghai@kernel.org>
Acked-by: NToshi Kani <toshi.kani@hp.com>

The second argument of ACPI driver .remove() operation is only used
by the ACPI processor driver and the value passed to that driver
through it is always available from the given struct acpi_device
object's removal_type field.  For this reason, the second ACPI driver
.remove() argument is in fact useless, so drop it.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: NJiang Liu <jiang.liu@huawei.com>
Acked-by: NToshi Kani <toshi.kani@hp.com>
Acked-by: NYinghai Lu <yinghai@kernel.org>

acpi_install_notify_handler() could fail.  So check the exit status
and give a better debug info.
Signed-off-by: NTang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

The acpiphp driver is confusing because it contains partial support for PCI
host bridge hotplug as well as support for hotplug of PCI devices.

This patch moves the host bridge hot-add support to pci_root.c and adds
hot-remove support in pci_root.c.

How to test it: if sci_emu patch is applied, find out root bus number to
ACPI root name mapping from dmesg or /sys.  To remove root bus:

  echo "\_SB.PCIB 3" > /sys/kernel/debug/acpi/sci_notify

To add back root bus:

  echo "\_SB.PCIB 1" > /sys/kernel/debug/acpi/sci_notify
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>
Acked-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>

The ACPI handles of PCI root bridges need to be known to
acpi_bind_one(), so that it can create the appropriate
"firmware_node" and "physical_node" files for them, but currently
the way it gets to know those handles is not exactly straightforward
(to put it lightly).

This is how it works, roughly:

  1. acpi_bus_scan() finds the handle of a PCI root bridge,
     creates a struct acpi_device object for it and passes that
     object to acpi_pci_root_add().

  2. acpi_pci_root_add() creates a struct acpi_pci_root object,
     populates its "device" field with its argument's address
     (device->handle is the ACPI handle found in step 1).

  3. The struct acpi_pci_root object created in step 2 is passed
     to pci_acpi_scan_root() and used to get resources that are
     passed to pci_create_root_bus().

  4. pci_create_root_bus() creates a struct pci_host_bridge object
     and passes its "dev" member to device_register().

  5. platform_notify(), which for systems with ACPI is set to
     acpi_platform_notify(), is called.

So far, so good.  Now it starts to be "interesting".

  6. acpi_find_bridge_device() is used to find the ACPI handle of
     the given device (which is the PCI root bridge) and executes
     acpi_pci_find_root_bridge(), among other things, for the
     given device object.

  7. acpi_pci_find_root_bridge() uses the name (sic!) of the given
     device object to extract the segment and bus numbers of the PCI
     root bridge and passes them to acpi_get_pci_rootbridge_handle().

  8. acpi_get_pci_rootbridge_handle() browses the list of ACPI PCI
     root bridges and finds the one that matches the given segment
     and bus numbers.  Its handle is then used to initialize the
     ACPI handle of the PCI root bridge's device object by
     acpi_bind_one().  However, this is *exactly* the ACPI handle we
     started with in step 1.

Needless to say, this is quite embarassing, but it may be avoided
thanks to commit f3fd0c8a (ACPI: Allow ACPI handles of devices to be
initialized in advance), which makes it possible to initialize the
ACPI handle of a device before passing it to device_register().

Accordingly, add a new __weak routine, pcibios_root_bridge_prepare(),
defaulting to an empty implementation that can be replaced by the
interested architecutres (x86 and ia64 at the moment) with functions
that will set the root bridge's ACPI handle before its dev member is
passed to device_register().  Make both x86 and ia64 provide such
implementations of pcibios_root_bridge_prepare() and remove
acpi_pci_find_root_bridge() and acpi_get_pci_rootbridge_handle() that
aren't necessary any more.

Included is a fix for breakage on systems with non-ACPI PCI host
bridges from Bjorn Helgaas.
Signed-off-by: NRafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>

Firmware may have assigned PCI BARs for hot-added devices, so reserve
those resources before trying to allocate more.

[bhelgaas: move empty weak definition here]
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Signed-off-by: NBjorn Helgaas <bhelgaas@google.com>