IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Per the errata of TRM, the RC can't support L1 substate, so remove the L1
substate cap as well as operation for PCIE_RC_CONFIG_L1_SUBSTATE_CTRL2.
Tested-by: Brian Norris <briannorris@chromium.org>
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
rk3399 supports PCIe 2.x link speeds marginally at best, and on some
boards, the link won't train at 5 GT/s at all. Rather than sacrifice 500ms
waiting for training that will never happen, let's use the helper function,
of_pci_get_max_link_speed(), to get the max link speed from DT and specify
link capability.
Signed-off-by: Brian Norris <briannorris@chromium.org>
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The calculation of negotiated lanes is wrong: it should be shifted by
PCIE_CORE_PL_CONF_LANE_SHIFT, but it is shifted by
PCIE_CORE_PL_CONF_LANE_MASK instead. Let's fix it.
Fixes: e77f847df54c ("PCI: rockchip: Add Rockchip PCIe controller support")
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Allow selection of the Rockchip driver for compile testing, even if we
aren't building for ARCH_ROCKCHIP.
[bhelgaas: changelog]
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The default value of common clock configuration is zero indicating
Rockchip's RC is using asynchronous clock architecture but actually we are
using common clock. This will confuse some EP drivers if they need some
different settings referring to this value.
Set the Common Clock Configuration bit in the Link Control Register.
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
If vpcie3v3 is available, we could provide these information via RC's
configure register to make EP able to know the power limit.
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Add Makefile comments to explain the Kconfig and build strategy for ARM64
drivers that work around not-quite-ECAM issues. No functional change
intended.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Use builtin_platform_driver() helper to simplify the code.
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Fix the following warnings:
drivers/pci/host/vmd.c:731:12: warning: ‘vmd_suspend’ defined but not used [-Wunused-function]
static int vmd_suspend(struct device *dev)
^
drivers/pci/host/vmd.c:739:12: warning: ‘vmd_resume’ defined but not used [-Wunused-function]
static int vmd_resume(struct device *dev)
^
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
Reviewed-by: Keith Busch <keith.busch@intel.com>
SRCU lets synchronize_srcu() depend on VMD-local RCU primitives, preventing
long delays from locking up RCU in other systems. VMD performs a
synchronize when removing a device, but will hit all IRQ lists if the
device uses all VMD vectors. This patch will not help VMD's RCU
synchronization, but will isolate the read side delays to the VMD
subsystem. Additionally, the use of SRCU in VMD's ISR will keep it
isolated from any other RCU waiters in the rest of the system.
Tested using concurrent FIO and NVMe resets:
[global]
rw=read
bs=4k
direct=1
ioengine=libaio
iodepth=32
norandommap
timeout=300
runtime=1000000000
[nvme0]
cpus_allowed=0-63
numjobs=8
filename=/dev/nvme0n1
[nvme1]
cpus_allowed=0-63
numjobs=8
filename=/dev/nvme1n1
while (true) do
for i in /sys/class/nvme/nvme*; do
echo "Resetting ${i##*/}"
echo 1 > $i/reset_controller;
sleep 5
done;
done
Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Keith Busch <keith.busch@intel.com>
The Tegra PCI host controller driver no longer relies on any of the 32-bit
ARM glue for PCI, so it can be enabled on 64-bit configurations.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
The PCIe host controller found on Tegra X1 is very similar to its
predecessor on Tegra K1. A bug was introduced in the new revision that
is worked around by always enabling the performance counter, otherwise
accesses to configuration space will block for a number of seconds.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
Tegra210's PCIe controller has a bug that requires the PCA (performance
counter) feature to be enabled. If this isn't done, accesses to device
configuration space will hang the chip for tens of seconds. Implement the
workaround.
Based on commit 514e19138af2 ("pci: tegra: implement PCA enable
workaround") from U-Boot by Stephen Warren <swarren@nvidia.com>.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
Tegra is one of the remaining platforms that still use the traditional
pci_common_init_dev() interface for probing PCI host bridges.
This demonstrates how to convert it to the pci_register_host interface I
just added in a previous patch. This leads to a more linear probe sequence
that can handle errors better because we avoid callbacks into the driver,
and it makes the driver architecture independent.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
Allow PCI host bridge drivers to use the new host bridge interfaces to
register their host bridge.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
Provide a way to allocate driver-specific data along with a PCI host bridge
structure. The bridge's ->private field points to this data.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
Make the existing pci_host_bridge structure a proper device that is usable
by PCI host drivers in a more standard way. In addition to the existing
pci_scan_bus(), pci_scan_root_bus(), pci_scan_root_bus_msi(), and
pci_create_root_bus() interfaces, this unfortunately means having to add
yet another interface doing basically the same thing, and add some extra
code in the initial step.
However, this time it's more likely to be extensible enough that we won't
have to do another one again in the future, and we should be able to reduce
code much more as a result.
The main idea is to pull the allocation of 'struct pci_host_bridge' out of
the registration, and let individual host drivers and architecture code
fill the members before calling the registration function.
There are a number of things we can do based on this:
* Use a single memory allocation for the driver-specific structure
and the generic PCI host bridge
* consolidate the contents of driver-specific structures by moving
them into pci_host_bridge
* Add a consistent interface for removing a PCI host bridge again
when unloading a host driver module
* Replace the architecture specific __weak pcibios_*() functions with
callbacks in a pci_host_bridge device
* Move common boilerplate code from host drivers into the generic
function, based on contents of the structure
* Extend pci_host_bridge with additional members when needed without
having to add arguments to pci_scan_*().
* Move members of struct pci_bus into pci_host_bridge to avoid
having lots of identical copies.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <helgaas@kernel.org>
PCIe controllers in X-Gene SoCs are not ECAM compliant: software needs to
configure additional controller's register to address device at
bus:dev:function.
Add a quirk to discover controller MMIO register space and configure
controller registers to select and address the target secondary device.
The quirk will only be applied for X-Gene PCIe MCFG table with
OEM revison 1, 2, 3 or 4 (PCIe controller v1 and v2 on X-Gene SoCs).
Tested-by: Jon Masters <jcm@redhat.com>
Signed-off-by: Duc Dang <dhdang@apm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
ThunderX pass1.x requires to emulate the EA headers for on-chip devices
hence it has to use custom pci_thunder_ecam_ops for accessing PCI config
space (pci-thunder-ecam.c). Add new entries to MCFG quirk array where it
can be applied while probing ACPI based PCI host controller.
ThunderX pass1.x is using the same way for accessing off-chip devices
(so-called PEM) as silicon pass-2.x so we need to add PEM quirk entries
too.
Quirk is considered for ThunderX silicon pass1.x only which is identified
via MCFG revision 2.
ThunderX pass 1.x requires the following accessors:
NUMA node 0 PCI segments 0- 3: pci_thunder_ecam_ops (MCFG quirk)
NUMA node 0 PCI segments 4- 9: thunder_pem_ecam_ops (MCFG quirk)
NUMA node 1 PCI segments 10-13: pci_thunder_ecam_ops (MCFG quirk)
NUMA node 1 PCI segments 14-19: thunder_pem_ecam_ops (MCFG quirk)
[bhelgaas: change Makefile/ifdefs so quirk doesn't depend on
CONFIG_PCI_HOST_THUNDER_ECAM]
Signed-off-by: Tomasz Nowicki <tn@semihalf.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
ThunderX PCIe controller to off-chip devices (so-called PEM) is not fully
compliant with ECAM standard. It uses non-standard configuration space
accessors (see thunder_pem_ecam_ops) and custom configuration space
granulation (see bus_shift = 24). In order to access configuration space
and probe PEM as ACPI-based PCI host controller we need to add MCFG quirk
infrastructure. This involves:
1. A new thunder_pem_acpi_init() init function to locate PEM-specific
register ranges using ACPI.
2. Export PEM thunder_pem_ecam_ops structure so it is visible to MCFG quirk
code.
3. New quirk entries for each PEM segment. Each contains platform IDs,
mentioned thunder_pem_ecam_ops and CFG resources.
Quirk is considered for ThunderX silicon pass2.x only which is identified
via MCFG revision 1.
ThunderX pass 2.x requires the following accessors:
NUMA Node 0 PCI segments 0- 3: pci_generic_ecam_ops (ECAM-compliant)
NUMA Node 0 PCI segments 4- 9: thunder_pem_ecam_ops (MCFG quirk)
NUMA Node 1 PCI segments 10-13: pci_generic_ecam_ops (ECAM-compliant)
NUMA Node 1 PCI segments 14-19: thunder_pem_ecam_ops (MCFG quirk)
[bhelgaas: adapt to use acpi_get_rc_resources(), update Makefile/ifdefs so
quirk doesn't depend on CONFIG_PCI_HOST_THUNDER_PEM]
Signed-off-by: Tomasz Nowicki <tn@semihalf.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Pull the register resource lookup out of thunder_pem_init() so we can
easily add a corresponding lookup using ACPI. No functional change
intended.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The PCIe controller in Hip05/Hip06/Hip07 SoCs is not completely
ECAM-compliant. It is non-ECAM only for the RC bus config space; for any
other bus underneath the root bus it does support ECAM access.
Add specific quirks for PCI config space accessors. This involves:
1. New initialization call hisi_pcie_init() to obtain RC base
addresses from PNP0C02 at the root of the ACPI namespace (under \_SB).
2. New entry in common quirk array.
[bhelgaas: move to pcie-hisi.c and change Makefile/ifdefs so quirk doesn't
depend on CONFIG_PCI_HISI]
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
Signed-off-by: Gabriele Paoloni <gabriele.paoloni@huawei.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The Qualcomm Technologies QDF2432 SoC does not support accesses smaller
than 32 bits to the PCI configuration space. Register the appropriate
quirk.
[bhelgaas: add QCOM_ECAM32 macro, ifdef for ACPI and PCI_QUIRKS]
Signed-off-by: Christopher Covington <cov@codeaurora.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The acpi_get_rc_resources() is used to get the RC register address that can
not be described in MCFG. It takes the _HID & segment to look for and
outputs the RC address resource. Use PNP0C02 devices to describe such RC
address resource. Use _UID to match segment to tell which root bus the
PNP0C02 resource belongs to.
[bhelgaas: add dev argument, wrap in #ifdef CONFIG_PCI_QUIRKS]
Signed-off-by: Dongdong Liu <liudongdong3@huawei.com>
Signed-off-by: Tomasz Nowicki <tn@semihalf.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
pci_update_resource() updates a hardware BAR so its address matches the
kernel's struct resource UNLESS it's a disabled ROM BAR. We only update
those when we enable the ROM.
It's not obvious from the code why ROM BARs should be handled specially.
Apparently there are Matrox devices with defective ROM BARs that read as
zero when disabled. That means that if pci_enable_rom() reads the disabled
BAR, sets PCI_ROM_ADDRESS_ENABLE (without re-inserting the address), and
writes it back, it would enable the ROM at address zero.
Add comments and references to explain why we can't make the code look more
rational.
The code changes are from 755528c860b0 ("Ignore disabled ROM resources at
setup") and 8085ce084c0f ("[PATCH] Fix PCI ROM mapping").
Link: https://lkml.org/lkml/2005/8/30/138
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Remove the assumption that IORESOURCE_ROM_ENABLE == PCI_ROM_ADDRESS_ENABLE.
PCI_ROM_ADDRESS_ENABLE is the ROM enable bit defined by the PCI spec, so if
we're reading or writing a BAR register value, that's what we should use.
IORESOURCE_ROM_ENABLE is a corresponding bit in struct resource flags.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
pci_std_update_resource() only deals with standard BARs, so we don't have
to worry about the complications of VF BARs in an SR-IOV capability.
Compute the BAR address inline and remove pci_resource_bar(). That makes
pci_iov_resource_bar() unused, so remove that as well.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
If we update a VF BAR while it's enabled, there are two potential problems:
1) Any driver that's using the VF has a cached BAR value that is stale
after the update, and
2) We can't update 64-bit BARs atomically, so the intermediate state
(new lower dword with old upper dword) may conflict with another
device, and an access by a driver unrelated to the VF may cause a bus
error.
Warn about attempts to update VF BARs while they are enabled. This is a
programming error, so use dev_WARN() to get a backtrace.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Previously pci_update_resource() used the same code path for updating
standard BARs and VF BARs in SR-IOV capabilities.
Split the VF BAR update into a new pci_iov_update_resource() internal
interface, which makes it simpler to compute the BAR address (we can get
rid of pci_resource_bar() and pci_iov_resource_bar()).
This patch:
- Renames pci_update_resource() to pci_std_update_resource(),
- Adds pci_iov_update_resource(),
- Makes pci_update_resource() a wrapper that calls the appropriate one,
No functional change intended.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
hv_do_hypercall() assumes that we pass a segment from a physically
contiguous buffer. A buffer allocated on the stack may not work if
CONFIG_VMAP_STACK=y is set.
Use kmalloc() to allocate this buffer.
Reported-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: Long Li <longli@microsoft.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: K. Y. Srinivasan <kys@microsoft.com>
On DT based systems, the of_dma_configure() API implements DMA
configuration for a given device. On ACPI systems an API equivalent to
of_dma_configure() is missing which implies that it is currently not
possible to set-up DMA operations for devices through the ACPI generic
kernel layer.
This patch fills the gap by introducing acpi_dma_configure/deconfigure()
calls that for now are just wrappers around arch_setup_dma_ops() and
arch_teardown_dma_ops() and also updates ACPI and PCI core code to use
the newly introduced acpi_dma_configure/acpi_dma_deconfigure functions.
Since acpi_dma_configure() is used to configure DMA operations, the
function initializes the dma/coherent_dma masks to sane default values
if the current masks are uninitialized (also to keep the default values
consistent with DT systems) to make sure the device has a complete
default DMA set-up.
The DMA range size passed to arch_setup_dma_ops() is sized according
to the device coherent_dma_mask (starting at address 0x0), mirroring the
DT probing path behaviour when a dma-ranges property is not provided
for the device being probed; this changes the current arch_setup_dma_ops()
call parameters in the ACPI probing case, but since arch_setup_dma_ops()
is a NOP on all architectures but ARM/ARM64 this patch does not change
the current kernel behaviour on them.
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Acked-by: Bjorn Helgaas <bhelgaas@google.com> [pci]
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Tomasz Nowicki <tn@semihalf.com>
Tested-by: Hanjun Guo <hanjun.guo@linaro.org>
Tested-by: Tomasz Nowicki <tn@semihalf.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: Tomasz Nowicki <tn@semihalf.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Signed-off-by: Will Deacon <will.deacon@arm.com>
The BAR property bits (0-3 for memory BARs, 0-1 for I/O BARs) are supposed
to be read-only, but we do save them in res->flags and include them when
updating the BAR.
Mask the I/O property bits with ~PCI_BASE_ADDRESS_IO_MASK (0x3) instead of
PCI_REGION_FLAG_MASK (0xf) to make it obvious that we can't corrupt bits
2-3 of I/O addresses.
Use PCI_ROM_ADDRESS_MASK for ROM BARs. This means we'll only check the top
21 bits (instead of the 28 bits we used to check) of a ROM BAR to see if
the update was successful.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
VF BARs are read-only zero, so updating VF BARs will not have any effect.
See the SR-IOV spec r1.1, sec 3.4.1.11.
We already ignore these updates because of 70675e0b6a1a ("PCI: Don't try to
restore VF BARs"); this merely restructures it slightly to make it easier
to split updates for standard and SR-IOV BARs.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Previously we enabled VFs and enable their memory space before calling
pcibios_sriov_enable(). But pcibios_sriov_enable() may update the VF BARs:
for example, on PPC PowerNV we may change them to manage the association of
VFs to PEs.
Because 64-bit BARs cannot be updated atomically, it's unsafe to update
them while they're enabled. The half-updated state may conflict with other
devices in the system.
Call pcibios_sriov_enable() before enabling the VFs so any BAR updates
happen while the VF BARs are disabled.
[bhelgaas: changelog]
Tested-by: Carol Soto <clsoto@us.ibm.com>
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
In the code to handle PAXB v2 based MSI steering, the logic aligns the MSI
register address to the size of supported inbound mapping range. This is
incorrect since it rounds "up" the starting address to the next aligned
address, but what we want is the starting address to be rounded "down" to
the aligned address.
This patch fixes the issue and allows MSI writes to be properly steered to
the GIC.
Fixes: 4b073155fbd3 ("PCI: iproc: Add support for the next-gen PAXB controller")
Signed-off-by: Ray Jui <ray.jui@broadcom.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Add support for the MSM8996/APQ8096 PCIe controller. MSM8996 supports Gen
1/2, one lane, 3 PCIe root complexes with support for MSI and legacy
interrupts, and it conforms to PCI Express Base 2.1 specification.
Add a post_init callback to qcom_pcie_ops, as the PCIe pipe clocks are only
setup after the phy is powered on. It also adds an ltssm_enable callback
as it is very much different from other supported SoCs in the driver.
Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Stanimir Varbanov <svarbanov@mm-sol.com>
Most error branches following the call to pci_enable_device() contain a
call to pci_disable_device(). Add these calls where they are missing.
This issue was found with Hector.
Signed-off-by: Quentin Lambert <lambert.quentin@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Add support for the next generation of the iProc PAXB host controller, used
in Stingray.
Signed-off-by: Oza Oza <oza.oza@broadcom.com>
Signed-off-by: Ray Jui <ray.jui@broadcom.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Scott Branden <scott.branden@broadcom.com>
Per PCIe spec r3.0, sec 2.3.1.1, the Read Completion Boundary (RCB)
determines the naturally aligned address boundaries on which a Read Request
may be serviced with multiple Completions:
- For a Root Complex, RCB is 64 bytes or 128 bytes
This value is reported in the Link Control Register
Note: Bridges and Endpoints may implement a corresponding command bit
which may be set by system software to indicate the RCB value for the
Root Complex, allowing the Bridge/Endpoint to optimize its behavior
when the Root Complex’s RCB is 128 bytes.
- For all other system elements, RCB is 128 bytes
Per sec 7.8.7, if a Root Port only supports a 64-byte RCB, the RCB of all
downstream devices must be clear, indicating an RCB of 64 bytes. If the
Root Port supports a 128-byte RCB, we may optionally set the RCB of
downstream devices so they know they can generate larger Completions.
Some BIOSes supply an _HPX that tells us to set RCB, even though the Root
Port doesn't have RCB set, which may lead to Malformed TLP errors if the
Endpoint generates completions larger than the Root Port can handle.
The IBM x3850 X6 with BIOS version -[A8E120CUS-1.30]- 08/22/2016 supplies
such an _HPX and a Mellanox MT27500 ConnectX-3 device fails to initialize:
mlx4_core 0000:41:00.0: command 0xfff timed out (go bit not cleared)
mlx4_core 0000:41:00.0: device is going to be reset
mlx4_core 0000:41:00.0: Failed to obtain HW semaphore, aborting
mlx4_core 0000:41:00.0: Fail to reset HCA
------------[ cut here ]------------
kernel BUG at drivers/net/ethernet/mellanox/mlx4/catas.c:193!
After 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
and 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices
with a link"), we apply _HPX settings to *all* devices, not just those
hot-added after boot.
Before 7a1562d4f2d0, we didn't touch the Mellanox RCB, and the device
worked. After 7a1562d4f2d0, we set its RCB to 128, and it failed.
Set the RCB to 128 iff the Root Port supports a 128-byte RCB. Otherwise,
set RCB to 64 bytes. This effectively ignores what _HPX tells us about
RCB.
Note that this change only affects _HPX handling. If we have no _HPX, this
does nothing with RCB.
[bhelgaas: changelog, clear RCB if not set for Root Port]
Fixes: 6cd33649fa83 ("PCI: Add pci_configure_device() during enumeration")
Fixes: 7a1562d4f2d0 ("PCI: Apply _HPX Link Control settings to all devices with a link")
Link: https://bugzilla.kernel.org/show_bug.cgi?id=187781
Tested-by: Frank Danapfel <fdanapfe@redhat.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Myron Stowe <myron.stowe@redhat.com>
CC: stable@vger.kernel.org # v3.18+
Export pcie_find_root_port() so we can use it outside of PCIe-AER error
injection.
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Mellanox devices were marked as having INTx masking ability broken. As a
result, the VFIO driver fails to start when more than one device function
is passed-through to a VM if both have the same INTx pin.
Prior to Connect-IB, Mellanox devices exposed to the operating system one
PCI function per all ports. Starting from Connect-IB, the devices are
function-per-port. When passing the second function to a VM, VFIO will
fail to start.
Exclude ConnectX-4, ConnectX4-Lx and Connect-IB from the list of Mellanox
devices marked as having broken INTx masking:
- ConnectX-4 and ConnectX4-LX firmware version is checked. If INTx
masking is supported, we unmark the broken INTx masking.
- Connect-IB does not support INTx currently so will not cause any
problem.
[bhelgaas: call pci_disable_device() always, after iounmap()]
Fixes: 11e42532ada3 ("PCI: Assume all Mellanox devices have broken INTx masking")
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Change Mellanox's broken_intx_masking() quirk from an "all Mellanox
devices" to a quirk for listed devices only.
[bhelgaas: remove #defines, reorder to keep other quirks together]
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Convert all quirk_broken_intx_masking() quirks from HEADER to FINAL.
The quirk sets dev->broken_intx_masking, which is only used by
pci_intx_mask_supported(), which is not needed until after FINAL
quirks have been run.
[bhelgaas: changelog]
Signed-off-by: Noa Osherovich <noaos@mellanox.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Hardware that supports only 32-bit config writes is not spec-compliant.
For example, if software performs a 16-bit write, we must do a 32-bit read,
merge in the 16 bits we intend to write, followed by a 32-bit write. If
the 16 bits we *don't* intend to write happen to have any RW1C (write-one-
to-clear) bits set, we just inadvertently cleared something we shouldn't
have.
Add a rate-limited warning when we do sub-32 bit config writes. Remove
similar probe-time warnings from some of the affected host bridge drivers.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Enthusiastically-Acked-by: Russell King <rmk+kernel@armlinux.org.uk>
Acked-by: Shawn Lin <shawn.lin@rock-chips.com> # rockchip
Acked-by: Thierry Reding <treding@nvidia.com>
Currently the revision isn't available via sysfs/libudev thus if one wants
to know the value one needs to read through the config file, which can be
quite time-consuming because it wakes/powers up the device.
There are at least two userspace components which could make use the new
file: libpciaccess and libdrm. The former wakes up _every_ PCI device,
which can be observed via glxinfo when using Mesa 10.0+ drivers. The
latter, in association with Mesa 13.0, can lead to 2-3 second delays while
starting firefox, thunderbird or chromium.
Link: https://bugs.freedesktop.org/show_bug.cgi?id=98502
Tested-by: Mauro Santos <registo.mailling@gmail.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch
CC: Greg KH <gregkh@linuxfoundation.org>
Linux 4.8 added support for runtime suspending PCIe ports to D3hot with
commit 006d44e49a25 ("PCI: Add runtime PM support for PCIe ports"), but
excluded hotplug ports. Those are now afforded runtime PM by the present
commit.
Hotplug ports require a few extra considerations:
- The configuration space of the port remains accessible in D3hot, so all
the functions to read or modify the Slot Status and Slot Control
registers need not be modified. Even turning on slot power doesn't seem
to require the port to be in D0, at least the PCIe spec doesn't say so
and I confirmed that by testing with a Thunderbolt controller.
- However D0 is required to access devices on the secondary bus. This
happens in pciehp_check_link_status() and pciehp_configure_device() (both
called from board_added()) and in pciehp_unconfigure_device() (called
from remove_board()), so acquire a runtime PM ref for their invocation.
- The hotplug port stays active as long as it has active children. If all
hotplugged devices below the port runtime suspend, the port is allowed to
runtime suspend as well. Plug and unplug detection continues to work in
D3hot.
- Hotplug interrupts are delivered in-band, so while the hotplug port
itself is allowed to go to D3hot, its parent ports must stay in D0 for
interrupts to come through. Add a corresponding restriction to
pci_dev_check_d3cold().
- Runtime PM may only be allowed if the hotplug port is handled natively by
the OS. On ACPI systems, the port may alternatively be handled by the
firmware and things break if the OS puts the port into D3 behind the
firmware's back: E.g. Thunderbolt hotplug ports on non-Macs are handled
by Intel's firmware in System Management Mode and the firmware is known
to access devices on the port's secondary bus without checking first if
the port is in D0: https://bugzilla.kernel.org/show_bug.cgi?id=53811
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CC: Mika Westerberg <mika.westerberg@linux.intel.com>
