* pci/virtualization:
PCI: Remove __pci_dev_reset() and pci_dev_reset()
PCI: Split ->reset_notify() method into ->reset_prepare() and ->reset_done()
PCI: Protect pci_error_handlers->reset_notify() usage with device_lock()
PCI: Protect pci_driver->sriov_configure() usage with device_lock()
PCI: Mark Intel XXV710 NIC INTx masking as broken
PCI: Restore PRI and PASID state after Function-Level Reset
PCI: Cache PRI and PASID bits in pci_dev
The pci_error_handlers->reset_notify() method had a flag to indicate
whether to prepare for or clean up after a reset. The prepare and done
cases have no shared functionality whatsoever, so split them into separate
methods.
[bhelgaas: changelog, update locking comments]
Link: http://lkml.kernel.org/r/20170601111039.8913-3-hch@lst.de
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
* pci/resource:
PCI: Work around poweroff & suspend-to-RAM issue on Macbook Pro 11
PCI: Do not disregard parent resources starting at 0x0
Conflicts:
arch/x86/pci/fixup.c
* pci/pm:
PCI/PM: Avoid using device_may_wakeup() for runtime PM
x86/PCI: Avoid AMD SB7xx EHCI USB wakeup defect
PCI/PM: Restore the status of PCI devices across hibernation
drm/radeon: make MacBook Pro d3_delay quirk more generic
drm/amdgpu: remove unnecessary save/restore of pdev->d3_delay
PCI/PM: Add needs_resume flag to avoid suspend complete optimization
PCI: imx6: Fix config read timeout handling
switchtec: Fix minor bug with partition ID register
switchtec: Use new cdev_device_add() helper function
PCI: endpoint: Make PCI_ENDPOINT depend on HAS_DMA
pci_target_state() calls device_may_wakeup() which checks whether or not
the device may wake up the system from sleep states, but pci_target_state()
is used for runtime PM too.
Since runtime PM is expected to always enable remote wakeup if possible,
modify pci_target_state() to take additional argument indicating whether or
not it should look for a state from which the device can signal wakeup and
pass either the return value of device_can_wakeup(), or "false" (if the
device itself is not wakeup-capable) to it from the code related to runtime
PM.
While at it, fix the comment in pci_dev_run_wake() which is not about sleep
states.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
The run_wake flag in struct dev_pm_info is used to indicate whether
or not the device is capable of generating remote wakeup signals at
run time (or in the system working state), but the distinction
between runtime remote wakeup and system wakeup signaling has always
been rather artificial. The only practical reason for it to exist
at the core level was that ACPI and PCI treated those two cases
differently, but that's not the case any more after recent changes.
For this reason, get rid of the run_wake flag and, when applicable,
use device_set_wakeup_capable() and device_can_wakeup() instead of
device_set_run_wake() and device_run_wake(), respectively.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
After previous changes it is not necessary to distinguish between
device wakeup for run time and device wakeup from system sleep states
any more, so rework the PCI device wakeup settings code accordingly.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
The test for INTx masking via PCI_COMMAND_INTX_DISABLE performed in
pci_intx_mask_supported() should be done before the device can be used.
This is to avoid writing PCI_COMMAND while the driver owns the device, in
case that has any effect on MSI/MSI-X interrupts.
Move the content of pci_intx_mask_supported() to pci_intx_mask_broken() and
call it from pci_setup_device().
The test result can be queried at any time later using the same
pci_intx_mask_supported() interface as before (though with changed
implementation), so callers (uio, vfio) should be unaffected.
Signed-off-by: Piotr Gregor <piotrgregor@rsyncme.org>
[bhelgaas: changelog, remove quirk check, remove locking, move
dev->broken_intx_masking assignment to caller]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Every method in struct device_driver or structures derived from it like
struct pci_driver MUST provide exclusion vs the driver's ->remove() method,
usually by using device_lock().
Protect use of pci_error_handlers->reset_notify() by holding the device
lock while calling it.
Note:
- pci_dev_lock() calls device_lock() in addition to blocking user-space
config accesses.
- pci_err_handlers->reset_notify() is used inside
pci_dev_save_and_disable() and pci_dev_restore(). We could hold the
device lock directly in pci_reset_notify(), but we expand the region
since we have several calls following each other.
Without this, ->reset_notify() may race with ->remove() calls, which can be
easily triggered in NVMe.
[bhelgaas: changelog, add pci_reset_notify() comment]
[bhelgaas: fold in fix from Dan Carpenter <dan.carpenter@oracle.com>:
http://lkml.kernel.org/r/20170701135323.x5vaj4e2wcs2mcro@mwanda]
Link: http://lkml.kernel.org/r/20170601111039.8913-2-hch@lst.de
Reported-by: Rakesh Pandit <rakesh@tuxera.com>
Tested-by: Rakesh Pandit <rakesh@tuxera.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The wakeup_prepared PCI device flag is used for preventing subsequent
changes of PCI device wakeup settings in the same way (e.g. enabling
device wakeup twice in a row).
However, in some cases PME Enable may be updated by things like PCI
configuration space restoration in the meantime and it may need to be
set again even though the rest of the settings need not change, so
modify __pci_enable_wake() to do that when it is about to return
early.
Also, it is reasonable to expect that __pci_enable_wake() will always
clear PME Status when invoked to disable device wakeup, so make it do
so even if it is going to return early then.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
After a Function-Level Reset, PCI states need to be restored. Save PASID
features and PRI reqs cached.
[bhelgaas: search for capability only if PRI/PASID were enabled]
Signed-off-by: CQ Tang <cq.tang@intel.com>
Signed-off-by: Ashok Raj <ashok.raj@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Jean-Phillipe Brucker <jean-philippe.brucker@arm.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Some drivers - like i915 - may not support the system suspend direct
complete optimization due to differences in their runtime and system
suspend sequence. Add a flag that when set resumes the device before
calling the driver's system suspend handlers which effectively disables
the optimization.
Needed by a future patch fixing suspend/resume on i915.
Suggested by Rafael.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: stable@vger.kernel.org
Commit f44116ae88 ("PCI: Remove pci_find_parent_resource() use for
allocation") updated the logic that iterates over all bus resources and
compares them to a given resource, in order to decide whether one is the
parent of the latter.
This change inadvertently causes pci_find_parent_resource() to disregard
resources starting at address 0x0, resulting in an error such as the one
below on ARM systems whose I/O window starts at 0x0.
pci_bus 0000:00: root bus resource [mem 0x10000000-0x3efeffff window]
pci_bus 0000:00: root bus resource [io 0x0000-0xffff window]
pci_bus 0000:00: root bus resource [mem 0x8000000000-0xffffffffff window]
pci_bus 0000:00: root bus resource [bus 00-0f]
pci 0000:00:01.0: PCI bridge to [bus 01]
pci 0000:00:02.0: PCI bridge to [bus 02]
pci 0000:00:03.0: PCI bridge to [bus 03]
pci 0000:00:03.0: can't claim BAR 13 [io 0x0000-0x0fff]: no compatible bridge window
pci 0000:03:01.0: can't claim BAR 0 [io 0x0000-0x001f]: no compatible bridge window
While this never happens on x86, it is perfectly legal in general for a PCI
MMIO or IO window to start at address 0x0, and it was supported in the code
before commit f44116ae88.
Drop the test for res->start != 0; resource_contains() already checks
whether [start, end) completely covers the resource, and so it should be
redundant.
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
* pci/virtualization:
ixgbe: Use pcie_flr() instead of duplicating it
IB/hfi1: Use pcie_flr() instead of duplicating it
PCI: Call pcie_flr() from reset_chelsio_generic_dev()
PCI: Call pcie_flr() from reset_intel_82599_sfp_virtfn()
PCI: Export pcie_flr()
PCI: Add sysfs sriov_drivers_autoprobe to control VF driver binding
PCI: Avoid FLR for Intel 82579 NICs
Conflicts:
include/linux/pci.h
* pci/resource:
PCI: Don't resize resources when realigning all devices in system
PCI: Don't reassign resources that are already aligned
PCI: Factor pci_reassigndev_resource_alignment()
powerpc/powernv: Override pcibios_default_alignment() to force PCI devices to be page aligned
PCI: Add pcibios_default_alignment() for arch-specific alignment control
PCI: Fix calculation of bridge window's size and alignment
PCI: Ignore requested alignment for IOV BARs
PCI: Make PCI_ROM_ADDRESS_MASK a 32-bit constant
* pci/ioremap:
PCI: versatile: Update PCI config space remap function
PCI: keystone-dw: Update PCI config space remap function
PCI: layerscape: Update PCI config space remap function
PCI: hisi: Update PCI config space remap function
PCI: tegra: Update PCI config space remap function
PCI: xgene: Update PCI config space remap function
PCI: armada8k: Update PCI config space remap function
PCI: designware: Update PCI config space remap function
PCI: iproc-platform: Update PCI config space remap function
PCI: qcom: Update PCI config space remap function
PCI: rockchip: Update PCI config space remap function
PCI: spear13xx: Update PCI config space remap function
PCI: xilinx-nwl: Update PCI config space remap function
PCI: xilinx: Update PCI config space remap function
PCI: ECAM: Map config region with pci_remap_cfgspace()
PCI: Implement devm_pci_remap_cfgspace()
devres: fix devm_ioremap_*() offset parameter kerneldoc description
ARM: Implement pci_remap_cfgspace() interface
ARM64: Implement pci_remap_cfgspace() interface
linux/io.h: Add pci_remap_cfgspace() interface
PCI: Remove __weak tag from pci_remap_iospace()
The introduction of the pci_remap_cfgspace() interface allows PCI host
controller drivers to map PCI config space through a dedicated kernel
interface. Current PCI host controller drivers use the devm_ioremap_*()
devres interfaces to map PCI configuration space regions so in order to
update them to the new pci_remap_cfgspace() mapping interface a new set of
devres interfaces should be implemented so that PCI host controller drivers
can make use of them.
Introduce two new functions in the PCI kernel layer and Devres
documentation:
- devm_pci_remap_cfgspace()
- devm_pci_remap_cfg_resource()
so that PCI host controller drivers can make use of them to map PCI
configuration space regions.
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Cc: Jonathan Corbet <corbet@lwn.net>
These are useful for PCIe host drivers, and those drivers can be modules.
[bhelgaas: don't remove __weak; it's removed elsewhere]
Signed-off-by: Brian Norris <briannorris@chromium.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Shawn Lin <shawn.lin@rock-chips.com>
Currently we opencode the FLR sequence in lots of place; export a core
helper instead. We split out the probing for FLR support as all the
non-core callers already know their hardware.
Note that in the new pci_has_flr() function the quirk check has been moved
before the capability check as there is no point in reading the capability
in this case.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
pci_remap_iospace() is marked as a weak symbol even though no architecture
is currently overriding it; given that its implementation internals have
already code paths that are arch specific (ie PCI_IOBASE and
ioremap_page_range() attributes) there is no need to leave the weak symbol
in the kernel since the same functionality can be achieved by customizing
per-arch the corresponding functionality.
Remove the __weak symbol from pci_remap_iospace().
Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
The "pci=resource_alignment" argument aligns BARs of designated devices by
artificially increasing their size. Increasing the size increases the
alignment and prevents other resources from being assigned in the same
alignment region, e.g., in the same page, but it can break drivers that use
the BAR size to locate things, e.g., ilo_map_device() does this:
off = pci_resource_len(pdev, bar) - 0x2000;
The new pcibios_default_alignment() interface allows an arch to request
that *all* BARs in the system be aligned to a larger size. In this case,
we don't need to artificially increase the resource size because we know
every BAR of every device will be realigned, so nothing will share the same
alignment region.
Use IORESOURCE_STARTALIGN to request realignment of PCI BARs when we know
we're realigning all BARs in the system.
[bhelgaas: comment, changelog]
Signed-off-by: Yongji Xie <elohimes@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
The "pci=resource_alignment=" kernel argument designates devices for which
we want alignment greater than is required by the PCI specs. Previously we
set IORESOURCE_UNSET for every MEM resource of those devices, even if the
resource was *already* sufficiently aligned.
If a resource is already sufficiently aligned, leave it alone and don't try
to reassign it.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Pull the BAR size adjustment out into a new function,
pci_request_resource_alignment(), and add a comment about how and why we
increase the resource size and alignment.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
When VFIO passes through a PCI device to a guest, it does not allow the
guest to mmap BARs that are smaller than PAGE_SIZE unless it can reserve
the rest of the page (see vfio_pci_probe_mmaps()). This is because a page
might contain several small BARs for unrelated devices and a guest should
not be able to access all of them.
VFIO emulates guest accesses to non-mappable BARs, which is functional but
slow. On systems with large page sizes, e.g., PowerNV with 64K pages, BARs
are more likely to share a page and performance is more likely to be a
problem.
Add a weak function to set default alignment for all PCI devices. An arch
can override it to force the PCI core to place memory BARs on their own
pages.
Signed-off-by: Yongji Xie <elohimes@gmail.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Laurent Pinchart reported that the Renesas R-Car H2 Lager board (r8a7790)
crashes during suspend tests. Geert Uytterhoeven managed to reproduce the
issue on an M2-W Koelsch board (r8a7791):
It occurs when the PME scan runs, once per second. During PME scan, the
PCI host bridge (rcar-pci) registers are accessed while its module clock
has already been disabled, leading to the crash.
One reproducer is to configure s2ram to use "s2idle" instead of "deep"
suspend:
# echo 0 > /sys/module/printk/parameters/console_suspend
# echo s2idle > /sys/power/mem_sleep
# echo mem > /sys/power/state
Another reproducer is to write either "platform" or "processors" to
/sys/power/pm_test. It does not (or is less likely) to happen during full
system suspend ("core" or "none") because system suspend also disables
timers, and thus the workqueue handling PME scans no longer runs. Geert
believes the issue may still happen in the small window between disabling
module clocks and disabling timers:
# echo 0 > /sys/module/printk/parameters/console_suspend
# echo platform > /sys/power/pm_test # Or "processors"
# echo mem > /sys/power/state
(Make sure CONFIG_PCI_RCAR_GEN2 and CONFIG_USB_OHCI_HCD_PCI are enabled.)
Rafael Wysocki agrees that PME scans should be suspended before the host
bridge registers become inaccessible. To that end, queue the task on a
workqueue that gets frozen before devices suspend.
Rafael notes however that as a result, some wakeup events may be missed if
they are delivered via PME from a device without working IRQ (which hence
must be polled) and occur after the workqueue has been frozen. If that
turns out to be an issue in practice, it may be possible to solve it by
calling pci_pme_list_scan() once directly from one of the host bridge's
pm_ops callbacks.
Stacktrace for posterity:
PM: Syncing filesystems ... [ 38.566237] done.
PM: Preparing system for sleep (mem)
Freezing user space processes ... [ 38.579813] (elapsed 0.001 seconds) done.
Freezing remaining freezable tasks ... (elapsed 0.001 seconds) done.
PM: Suspending system (mem)
PM: suspend of devices complete after 152.456 msecs
PM: late suspend of devices complete after 2.809 msecs
PM: noirq suspend of devices complete after 29.863 msecs
suspend debug: Waiting for 5 second(s).
Unhandled fault: asynchronous external abort (0x1211) at 0x00000000
pgd = c0003000
[00000000] *pgd=80000040004003, *pmd=00000000
Internal error: : 1211 [#1] SMP ARM
Modules linked in:
CPU: 1 PID: 20 Comm: kworker/1:1 Not tainted
4.9.0-rc1-koelsch-00011-g68db9bc814362e7f #3383
Hardware name: Generic R8A7791 (Flattened Device Tree)
Workqueue: events pci_pme_list_scan
task: eb56e140 task.stack: eb58e000
PC is at pci_generic_config_read+0x64/0x6c
LR is at rcar_pci_cfg_base+0x64/0x84
pc : [<c041d7b4>] lr : [<c04309a0>] psr: 600d0093
sp : eb58fe98 ip : c041d750 fp : 00000008
r10: c0e2283c r9 : 00000000 r8 : 600d0013
r7 : 00000008 r6 : eb58fed6 r5 : 00000002 r4 : eb58feb4
r3 : 00000000 r2 : 00000044 r1 : 00000008 r0 : 00000000
Flags: nZCv IRQs off FIQs on Mode SVC_32 ISA ARM Segment user
Control: 30c5387d Table: 6a9f6c80 DAC: 55555555
Process kworker/1:1 (pid: 20, stack limit = 0xeb58e210)
Stack: (0xeb58fe98 to 0xeb590000)
fe80: 00000002 00000044
fea0: eb6f5800 c041d9b0 eb58feb4 00000008 00000044 00000000 eb78a000 eb78a000
fec0: 00000044 00000000 eb9aff00 c0424bf0 eb78a000 00000000 eb78a000 c0e22830
fee0: ea8a6fc0 c0424c5c eaae79c0 c0424ce0 eb55f380 c0e22838 eb9a9800 c0235fbc
ff00: eb55f380 c0e22838 eb55f380 eb9a9800 eb9a9800 eb58e000 eb9a9824 c0e02100
ff20: eb55f398 c02366c4 eb56e140 eb5631c0 00000000 eb55f380 c023641c 00000000
ff40: 00000000 00000000 00000000 c023a928 cd105598 00000000 40506a34 eb55f380
ff60: 00000000 00000000 dead4ead ffffffff ffffffff eb58ff74 eb58ff74 00000000
ff80: 00000000 dead4ead ffffffff ffffffff eb58ff90 eb58ff90 eb58ffac eb5631c0
ffa0: c023a844 00000000 00000000 c0206d68 00000000 00000000 00000000 00000000
ffc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
ffe0: 00000000 00000000 00000000 00000000 00000013 00000000 3a81336c 10ccd1dd
[<c041d7b4>] (pci_generic_config_read) from [<c041d9b0>]
(pci_bus_read_config_word+0x58/0x80)
[<c041d9b0>] (pci_bus_read_config_word) from [<c0424bf0>]
(pci_check_pme_status+0x34/0x78)
[<c0424bf0>] (pci_check_pme_status) from [<c0424c5c>] (pci_pme_wakeup+0x28/0x54)
[<c0424c5c>] (pci_pme_wakeup) from [<c0424ce0>] (pci_pme_list_scan+0x58/0xb4)
[<c0424ce0>] (pci_pme_list_scan) from [<c0235fbc>]
(process_one_work+0x1bc/0x308)
[<c0235fbc>] (process_one_work) from [<c02366c4>] (worker_thread+0x2a8/0x3e0)
[<c02366c4>] (worker_thread) from [<c023a928>] (kthread+0xe4/0xfc)
[<c023a928>] (kthread) from [<c0206d68>] (ret_from_fork+0x14/0x2c)
Code: ea000000 e5903000 f57ff04f e3a00000 (e5843000)
---[ end trace 667d43ba3aa9e589 ]---
Fixes: df17e62e5b ("PCI: Add support for polling PME state on suspended legacy PCI devices")
Reported-and-tested-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Reported-and-tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: stable@vger.kernel.org # 2.6.37+
Cc: Mika Westerberg <mika.westerberg@linux.intel.com>
Cc: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Cc: Simon Horman <horms+renesas@verge.net.au>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
We would call pci_reassigndev_resource_alignment() before
pci_init_capabilities(). So the requested alignment would never work for
IOV BARs.
Furthermore, it's meaningless to request additional alignment for IOV BARs,
the IOV BAR alignment is only determined by the VF BAR size.
Signed-off-by: Yongji Xie <xyjxie@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Per Intel Specification Update 335553-002 (see link below), some 82579
network adapters advertise a Function Level Reset (FLR) capability, but
they can hang when an FLR is triggered.
To reproduce the problem, attach the device to a VM, then detach and try to
attach again.
Add a quirk to prevent the use of FLR on these devices.
[bhelgaas: changelog, comments]
Link: http://www.intel.com/content/dam/www/public/us/en/documents/specification-updates/82579lm-82579v-gigabit-network-connection-spec-update.pdf
Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
If the PCI device is disconnected, return false immediately from
pci_device_is_present(). pci_device_is_present() uses the bus accessors,
so the early return in the device accessors doesn't help here.
Tested-by: Krishna Dhulipala <krishnad@fb.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Wei Zhang <wzhang@fb.com>
msleep() still sleeps 1 jiffy even when told to sleep for zero
milliseconds. That can end up being 1-2 milliseconds or more. In the
cases of d3_delay and d3cold_delay, that unnecessarily increases suspend
and/or resume latencies.
Do not sleep at all for the respective cases if d3_delay is zero or
d3cold_delay is zero.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
* pci/virtualization:
PCI: Add comments about ROM BAR updating
PCI: Decouple IORESOURCE_ROM_ENABLE and PCI_ROM_ADDRESS_ENABLE
PCI: Remove pci_resource_bar() and pci_iov_resource_bar()
PCI: Don't update VF BARs while VF memory space is enabled
PCI: Separate VF BAR updates from standard BAR updates
PCI: Update BARs using property bits appropriate for type
PCI: Ignore BAR updates on virtual functions
PCI: Do any VF BAR updates before enabling the BARs
PCI: Support INTx masking on ConnectX-4 with firmware x.14.1100+
PCI: Convert Mellanox broken INTx quirks to be for listed devices only
PCI: Convert broken INTx masking quirks from HEADER to FINAL
net/mlx4_core: Use device ID defines
PCI: Add Mellanox device IDs
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>
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 70675e0b6a ("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>
Linux 4.8 added support for runtime suspending PCIe ports to D3hot with
commit 006d44e49a ("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>
The conditions to block D3 on parent ports are currently condensed into a
single expression in pci_dev_check_d3cold(). Upcoming commits will add
further conditions for hotplug ports, making this expression fairly large
and impenetrable. Unfold the conditions to maintain readability when they
are amended.
No functional change intended.
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: Mika Westerberg <mika.westerberg@linux.intel.com>
CC: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The conditions to allow runtime PM on PCIe ports are currently spread
across two different files: The condition relating to hotplug ports is
located in portdrv_pci.c whereas all other conditions are located in pci.c.
Consolidate all conditions in a single place in pci.c, thus making it
easier to follow the logic and amend conditions down the road.
Note that the condition relating to hotplug ports is inserted *before* the
condition relating to the "pcie_port_pm=force" command line option, so
runtime PM is not afforded to hotplug ports even if this option is given.
That's exactly how the code behaved up until now. If this is not desired,
the ordering of the conditions can simply be reversed.
No functional change intended.
Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com>
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>
Currently pcie_portdrv_probe() activates runtime PM on a PCIe port even
if it will never actually suspend because the BIOS is too old or the
"pcie_port_pm=off" option was specified on the kernel command line.
A few CPU cycles can be saved by not activating runtime PM at all in these
cases, because rpm_idle() and rpm_suspend() will bail out right at the
beginning when calling rpm_check_suspend_allowed(), instead of carrying out
various locking and assignments, invoking rpm_callback(), getting back
-EBUSY and rolling everything back.
The conditions checked in pci_bridge_d3_possible() are all static, they
never change during uptime of the system, hence it's safe to call this to
determine if runtime PM should be activated.
No functional change intended.
Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com>
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>
After a device has been added, removed or had its D3cold attributes
changed, we recheck whether its parent bridge may runtime suspend to D3hot
with pci_bridge_d3_update().
The most naive algorithm would be to iterate over the bridge's children and
check if any of them are blocking D3.
The function already tries to be a bit smarter than that by first checking
the device that was changed. If this device already blocks D3 on the
bridge, then walking over all the other children can be skipped. A
drawback of this approach is that if the device is *not* blocking D3, it
will be checked a second time by pci_walk_bus(). But that's cheap and is
outweighed by the performance gain of potentially skipping pci_walk_bus()
altogether.
The algorithm can be optimized further by taking into account if D3 is
currently allowed for the bridge, as shown in the following truth table:
(a) remove && bridge_d3: D3 is currently allowed for the bridge and
removing one of its children won't change
that. No action necessary.
(b) remove && !bridge_d3: D3 may now be allowed for the bridge if the
removed child was the only one blocking it.
Check all its siblings to verify that.
(c) !remove && bridge_d3: D3 may now be disallowed but this can only
be caused by the added/changed child, not
any of its siblings. Check only that single
device.
(d) !remove && !bridge_d3: D3 may now be allowed for the bridge if the
changed child was the only one blocking it.
Check all its siblings to verify that.
By checking beforehand if the changed child
is blocking D3, we may be able to skip
checking its siblings.
Currently we do not special-case option (a) and in case of option (c) we
gratuitously call pci_walk_bus(). Speed up the algorithm by adding these
optimizations. Reword the comments a bit in an attempt to improve clarity.
No functional change intended.
Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com>
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>
The algorithm to update the flag indicating whether a bridge may go to D3
makes a few optimizations based on whether the update was caused by the
removal of a device on the one hand, versus the addition of a device or the
change of its D3cold flags on the other hand.
The information whether the update pertains to a removal is currently
passed in by the caller, but the function may as well determine that itself
by examining the device in question, thereby allowing for a considerable
simplification and reduction of the code.
Out of several options to determine removal, I've chosen the function
device_is_registered() because it's cheap: It merely returns the
dev->kobj.state_in_sysfs flag. That flag is set through device_add() when
the root bus is scanned and cleared through device_remove(). The call to
pci_bridge_d3_update() happens after each of these calls, respectively, so
the ordering is correct.
No functional change intended.
Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com>
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>
This function is always called with an existing pci_dev struct, which
holds a reference on the pci_bus struct it resides on, which in turn
holds a reference on pci_bus->bridge, which is the pci_dev's parent.
Hence there's no need to acquire an additional ref on the parent.
More specifically, the pci_dev exists until pci_destroy_dev() drops the
final reference on it, so all calls to pci_bridge_d3_update() must be
finished before that. It is arguably the caller's responsibility to ensure
that it doesn't call pci_bridge_d3_update() with a pci_dev that might
suddenly disappear, but in any case the existing callers are all safe:
- The call in pci_destroy_dev() happens before the call to put_device().
- The call in pci_bus_add_device() is synchronized with pci_destroy_dev()
using pci_lock_rescan_remove().
- The calls to pci_d3cold_disable() from the xhci and nouveau drivers
are safe because a ref on the pci_dev is held as long as it's bound to
a driver.
- The calls to pci_d3cold_enable() / pci_d3cold_disable() when modifying
the sysfs "d3cold_allowed" entry are also safe because kernfs_drain()
waits for existing sysfs users to finish before removing the entry,
and pci_destroy_dev() is called way after that.
No functional change intended.
Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com>
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>
One some systems, the firmware does not allow certain PCI devices to be put
in deep D-states. This can cause problems for wakeup signalling, if the
device does not support PME# in the deepest allowed suspend state. For
example, Pierre reports that on his system, ACPI does not permit his xHCI
host controller to go into D3 during runtime suspend -- but D3 is the only
state in which the controller can generate PME# signals. As a result, the
controller goes into runtime suspend but never wakes up, so it doesn't work
properly. USB devices plugged into the controller are never detected.
If the device relies on PME# for wakeup signals but is not capable of
generating PME# in the target state, the PCI core should accurately report
that it cannot do wakeup from runtime suspend. This patch modifies the
pci_dev_run_wake() routine to add this check.
Reported-by: Pierre de Villemereuil <flyos@mailoo.org>
Tested-by: Pierre de Villemereuil <flyos@mailoo.org>
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CC: stable@vger.kernel.org
CC: Lukas Wunner <lukas@wunner.de>
Resource allocation for VFs is done via the VF BARx registers in the PF's
SR-IOV Capability, and the BARs in the VFs themselves are read-only zeros
(see SR-IOV spec r1.1, secs 3.3.14 and 3.4.1.11).
Even though the actual VF BARs are read-only zeros, the VF dev->resource[]
structs describe the space allocated for the VF (this is a piece of the
space described by the VF BARx register in the PF's SR-IOV capability).
It's meaningless to request additional alignment for a VF: the VF BAR
alignment is completely determined by the alignment of the VF BARx in the
PF and the size of the VF BAR.
Ignore the user's alignment requests for VF devices.
Signed-off-by: Yongji Xie <xyjxie@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Users may request additional alignment of PCI resources, e.g., to align
BARs on page boundaries so they can be shared with guests via VFIO. This
of course may require reallocation if firmware has already assigned the
BARs with smaller alignments.
If the platform has requested PCI_PROBE_ONLY, we should never change any
PCI BARs, so we can't provide any additional alignment. Also, if a BAR is
marked as IORESOURCE_PCI_FIXED, e.g., for PCI Enhanced Allocation or if the
firmware depends on the current BAR value, we can't change the alignment.
In these cases, log a message and ignore the user's alignment requests.
[bhelgaas: changelog, use goto to simplify PCI_PROBE_ONLY check]
Signed-off-by: Yongji Xie <xyjxie@linux.vnet.ibm.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Whenever a device is resumed or its power state is changed using the
platform, its new power state is read from the PM Control & Status Register
and cached in pci_dev->current_state by calling pci_update_current_state().
If the device is in D3cold, reading from config space typically results in
a fabricated "all ones" response. But if it's in D3hot, the two bits
representing the power state in the PMCSR are *also* set to 1. Thus D3hot
and D3cold are not discernible by just reading the PMCSR.
To account for this, pci_update_current_state() uses two workarounds:
- When transitioning to D3cold using pci_platform_power_transition(), the
new power state is set blindly by pci_update_current_state(), i.e.
without verifying that the device actually *is* in D3cold. This is
achieved by setting the "state" argument to PCI_D3cold. The "state"
argument was originally intended to convey the new state in case the
device doesn't have the PM capability. It is *also* used to convey the
device state if the PM capability is present and the new state is D3cold,
but this was never explained in the kerneldoc.
- Once the current_state is set to D3cold, further invocations of
pci_update_current_state() will blindly assume that the device is still
in D3cold and leave the current_state unmodified. To get out of this
impasse, the current_state has to be set directly, typically by calling
pci_raw_set_power_state() or pci_enable_device().
It would be desirable if pci_update_current_state() could reliably detect
D3cold by itself. That would allow us to do away with these workarounds,
and it would allow for a smarter, more energy conserving runtime resume
strategy after system sleep: Currently devices which utilize
direct_complete are mandatorily runtime resumed in their ->complete stage.
This can be avoided if their power state after system sleep is the same as
before, but it requires a mechanism to detect the power state reliably.
We've just gained the ability to query the platform firmware for its
opinion on the device's power state. On platforms conforming to ACPI 4.0
or newer, this allows recognition of D3cold. Pre-4.0 platforms lack _PR3
and therefore the deepest power state that will ever be reported is D3hot,
even though the device may actually be in D3cold. To detect D3cold in
those cases, accessibility of the vendor ID in config space is probed using
pci_device_is_present(). This also works for devices which are not
platform-power-manageable at all, but can be suspended to D3cold using a
nonstandard mechanism (e.g. some hybrid graphics laptops or Thunderbolt on
the Mac).
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Linus Torvalds