diff options
author | Bjorn Helgaas <bhelgaas@google.com> | 2015-04-21 14:57:26 -0500 |
---|---|---|
committer | Bjorn Helgaas <bhelgaas@google.com> | 2015-04-21 15:29:39 -0500 |
commit | 9fbbda5c8e0ab9c391e4160a0eb3a06260f9f668 (patch) | |
tree | df1d59127854425e1a7356b416c14f19a205da6e /arch | |
parent | af539e985b3fbe3b602313f317506a2a9e73f5bf (diff) | |
download | linux-9fbbda5c8e0ab9c391e4160a0eb3a06260f9f668.tar.gz linux-9fbbda5c8e0ab9c391e4160a0eb3a06260f9f668.tar.bz2 linux-9fbbda5c8e0ab9c391e4160a0eb3a06260f9f668.zip |
ia64/PCI: Treat all host bridge Address Space Descriptors (even consumers) as windows
Prior to c770cb4cb505 ("PCI: Mark invalid BARs as unassigned"), if we tried
to claim a PCI BAR but could not find an upstream bridge window that
matched it, we complained but still allowed the device to be enabled.
c770cb4cb505 broke devices that previously worked (mptsas and igb in the
case Tony reported, but it could be any devices) because it marks those
BARs as IORESOURCE_UNSET, which makes pci_enable_device() complain and
return failure:
igb 0000:81:00.0: can't enable device: BAR 0 [mem size 0x00020000] not assigned
igb: probe of 0000:81:00.0 failed with error -22
The underlying cause is an ACPI Address Space Descriptor for a PCI host
bridge window that is marked as "consumer". This is a firmware defect:
resources that are produced on the downstream side of a bridge should be
marked "producer". But rejecting these BARs that we previously allowed is
a functionality regression, and firmware has not used the producer/consumer
bit consistently, so we can't rely on it anyway.
Stop checking the producer/consumer bit, and assume all bridge Address
Space Descriptors are for bridge windows.
Note that this change does not affect I/O Port or Fixed Location I/O Port
Descriptors, which are commonly used for the [io 0x0cf8-0x0cff] config
access range. That range is a "consumer" range and should not be treated
as a window.
Fixes: c770cb4cb505 ("PCI: Mark invalid BARs as unassigned")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=96961
Reported-and-tested-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Diffstat (limited to 'arch')
-rw-r--r-- | arch/ia64/pci/pci.c | 5 |
1 files changed, 1 insertions, 4 deletions
diff --git a/arch/ia64/pci/pci.c b/arch/ia64/pci/pci.c index 48cc65705db4..d4e162d35b34 100644 --- a/arch/ia64/pci/pci.c +++ b/arch/ia64/pci/pci.c @@ -240,15 +240,12 @@ static acpi_status resource_to_window(struct acpi_resource *resource, * We're only interested in _CRS descriptors that are * - address space descriptors for memory or I/O space * - non-zero size - * - producers, i.e., the address space is routed downstream, - * not consumed by the bridge itself */ status = acpi_resource_to_address64(resource, addr); if (ACPI_SUCCESS(status) && (addr->resource_type == ACPI_MEMORY_RANGE || addr->resource_type == ACPI_IO_RANGE) && - addr->address.address_length && - addr->producer_consumer == ACPI_PRODUCER) + addr->address.address_length) return AE_OK; return AE_ERROR; |